IUTAM Symposium on Integrated Modeling of Fully Coupled Fluid Structure

Interactions Using Analysis, Computations and Experiments

FLUID MECHANICS AND ITS APPLICATIONS
Volume 75

Series Editor: R. MOREAU

MADYLAM
Ecole Nationale Suprieure d'Hydraulique de Grenoble
Bote Postale 95
38402 Saint Martin d'Hres Cedex, France

Aims and Scope of the Series

The purpose of this series is to focus on subjects in which fluid mechanics plays a funda-
mental role.
As well as the more traditional applications of aeronautics, hydraulics, heat and mass
transfer etc., books will be published dealing with topics which are currently in a state of
rapid development, such as turbulence, suspensions and multiphase fluids, super and hy-
personic flows and numerical modelling techniques.
It is a widely held view that it is the interdisciplinary subjects that will receive intense
scientific attention, bringing them to the forefront of technological advancement. Fluids
have the ability to transport matter and its properties as well as transmit force, therefore
fluid mechanics is a subject that is particulary open to cross fertilisation with other scien-
ces and disciplines of engineering. The subject of fluid mechanics will be highly relevant
in domains such as chemical, metallurgical, biological and ecological engineering. This
series is particularly open to such new multidisciplinary domains.
The median level of presentation is the first year graduate student. Some texts are mono-
graphs defining the current state of a field; others are accessible to final year undergradu-
ates; but essentially the emphasis is on readability and clarity.

For a list of related mechanics titles, see final pages.

IUTAM Symposium
on Integrated Modeling of
Fully Coupled Fluid Structure
Interactions U sing Analysis,
Computations and Experiments
Proceedings ofthe IUTAM Symposium
held at Rutgers University. New Jersey. U.S.A ..
2-6 June 2003

l:dilL'd hy

HAY\1 BL'\AROYA
f)'I'"rlllldJ! "//Vf,,,./iwlieril & 1\d'<i.\l"iU' Fligill('aiJlg.
RllIgel:1 ['/liI'cl.lin', New 1lI'll/l.l\1'iek, :\'1. 1'. .\.11.

aml

TIMOTHY J. WJ:r
f)e/hfrlmcnf o//vlec)wflic'u! & 11c}"OS/hiC'( Lngi!lccrillg.
RlIlg('f'\ UJlil','r-sill', Nn" !imJlsl1'i"k, NI 1.'. .\',11.

....
"
SPRINGER-SCIENCE-'--BUSINESS MEDIA, s.Y.
A c.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN 978-94-010-3762-4 ISBN 978-94-007-0995-9 (eBook)

DOI 10.1007/978-94-007-0995-9

Cover photo: Wake development behind a cylinder, pinned at one end, and vibrating in response to
its own vortex shedding. This composite overlay of sixteen successive images shows a complex
interaction in the near wake between shear layer vortices, Krrnn vortices and cylinder motion. The
thin strips visible on the left from preceding images highlight the cylinder motion. The energy
transfer between fluid and cylinder motions is discussed in the paper by Benaroya and Wei in these
Proceedings. Photo courtesy of P. Atsavapranee (Naval Surface Warfare Center) and T. Wei (Rutgers
University).

Printed on acid-free paper

AlI Rights Reserved
2003 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2003
Softcover reprint of the hardcover Ist edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
 Table of Contents
Page Number

I Welcome to the IUTAM Symposium ix

Organizers: H. Benaroya, T. Wei, Rutgers University

II Welcome to Rutgers University: J. Flanagan, Vice President for xi

Research, Rutgers University

1. Survey of Naval Computational Needs in Fluid-Structure 1

Interaction (Plenary)
P. Palo, U.S. Navy

2. Novel Use of a Bandwidth Measure for Vortex Induced 27

Vibrations Case Study: The Foinaven Dynamic Umbilical
F. Trarieux, University College London; G. J. Lyons, BPP
Technical Services, Ltd.

3. Chaotic Vibration-Based Damage Detection in Fluid- 43

Structural Systems
B. I. Epureanu, University of Michigan

4. Comparison of Two Seafloor Observatory Mooring Designs 59

S. Han, M. Grosenbaugh, Woods Hole Oceanographic Institute

5. Nonlinear Dynamics of Very High Dimensional Fluid- 73

Structural Systems (Plenary)
E. Dowell, D. M. Tang, Duke University

6. Flapping Membranes for Thrust Production 115

J. M. Jimnez, J. H. J. Buchholz, A. E. Staples, J. J. Allen,
A. J. Smits, Princeton University

7. Experimental Study of a Tethered Cylinder in a Free Stream 125

J. Carberry, K. Ryan, J. Sheridan, Monash University

8. The Effect of Changed Mass Ratio on the Motion of a Tethered 135

Cylinder
K. Ryan, M. C. Thompson, K. Hourigan, Monash University

9. Self-excited Oscillation of Equilateral Triangular Wedge 145

S. Srigrarom, Nanyang Technological University; M. Kurosaka,
University of Washington
vi

Page Number
10. A Critical Review of the Intrinsic Nature of VIV (Plenary) 159
T. Sarpkaya, Naval Postgraduate School

11. A Numerical Study on the Rectilinear Oscillations of a 163

Circular Cylinder
S.Kocabiyik, Q. Al-Mdallal, Memorial University of
Newfoundland

12. Three-Dimensional Modeling of Flow-Induced Vibration for 175

an Elastic Cylinder in a Cross Flow
Y. Liu, R. M. C. So, C. Zhang, The Hong Kong Polytechnic
University

13. Correlation Length and Force Phasing of a Rigid Cylinder 187

Subject to VIV
D. Lucor, J. Foo, G. E. Karniadakis, Brown University

14. Self-Excited Oscillations of Vertical and Horizontal Cylinders 201

in Presence of a Free-Surface
D. Rockwell, M. Ozgoren, N. Saelim, Lehigh University

15. High Mode Number VIV Experiments (Plenary) 211

J. K. Vandiver, H. Marcollo, Massachusetts Institute of
Technology

16. Nonlinear Dynamics of Slender Cylinders Supported at Both 233

Ends and Subjected to Axial Flow
Y. Modarres-Sadeghi, M. P. Padoussis, C. Semler, P. Picot,
McGill University.

17. Some Quandaries and Paradoxes in Fluid-Structure 247

Interactions with Axial Flow (Plenary)
M. P. Padoussis, McGill University

18. Closed-Loop Control of the Resonant Flow-Structure 267

Interaction Using PID Controllers
M. M. Zhang, L. Cheng, Y. Zhou, The Hong Kong Polytechnic
University

19. Vortex-Induced Vibration Structural Response under 279

Parametric Excitation
K.Y. Billah, O. Ahmad, Stevens Institute of Technology
 vii

Page Number
20. Frequency Response and the Existence of a Critical Mass for 289
an Elastically-Mounted Cylinder
R. Govardham, Indian Institute of Science; C. H. K. Williamson,
Cornell University

21. Two-Degree of Freedom VIV of a Circular Cylinder Pinned at 301

One End
C. M. Leong, H. Benaroya, T. Wei, Rutgers University

22. Vortex-Induced Vibrations of Slender Structures in Shear 313

Flow: A Review (Plenary)
M. S. Triantafyllou, F. S. Hover, A. H. Techet, D. K. P. Yue,
Massachusetts Institute of Technology

23. A Tuned Liquid Damper Model 329

J. B. Frandsen, Lousiana State University

24. Unstable Phenomena of a Thin Cylindrical Shell Subjected to 331

Axial Leakage Flow
K. Fujita, A. Shintani, Osaka Prefecture University
25. Acoustic Scattering from a Coated Elastic Shell: Exact Vs. 353
Approximate Theory
A. Norris, Rutgers University

26. A Study of the Vibration of Fluid Coupled Coaxial Cylindrical 363

Shell
E. Gavrilova, St. Ivan Rilski University of Mining & Geology

27. Stability and Non-Linear Dynamics of Clamped Circular 375

Cylindrical Shells in Contact with Flowing Fluid
K. N. Karagiozis, M. P. Padoussis, E. Grinevich, A. K. Misra,
M. Amabili, McGill University

28. Forced Dynamic Uplift of Floating Plates 391

K. Dempsey, I. Vasileva, Clarkson University

29. Low-Dimensional Feedback Control of the von Karman 401

Vortex Street at a Reynolds Number of 100
S. Siegel, K. Cohen, T. McLaughlin, U.S. Air Force Academy

30. Stochastic Bifurcations of the Duffing-Mathieu Equations with 413

Time Delays
M. S. Fofana, Z. Hou, Worcester Polytechnic Institute;
S.T. Ariaratnam, University of Waterloo
viii

Page Number
31. Numerical and Experimental Investigation of a Streamwise 415
Oscillating Cylinder Wake in the Presence of a Downstream
Cylinder
Z. Guo, Huazhong University of Science and Technology;
Y. Zhou, The Hong Kong Polytechnic University

32. Control of Cylinder Drag and Lift Force Amplitude in 425

Turbulent Crossflow
S. A. Isaev, Academy of Civil Aviation; Z. L. Zhdanov, Luikov
Heat and Mass Transfer Institute

33. An Eulerian-Lagrangian Marker and Micro Cell Method for 439

Simulating Fluid Interaction with Solid/Porous Bodies
(Plenary)
R. Bidoae, R. M. Ciobotaru, P. E. Raad, Southern Methodist
University

34. A New Lagrangian-Eulerian Shell-Fluid Coupling Algorithm 453

Based on Level Sets
F. Cirak, California Institute of Technology; R. Radovitzky,
Massachusetts Institute of Technology

35. Application of Higher Order Runge-Kutta Time Integrators in 467

Partitioned Fluid-Structure Interaction Simulations
S. van Zuijlen, H. Bijl, Delft University of Technology

36. Vortex Shedding for Flow Over a Square Cylinder Close to a 479
Moving Ground
S. Bhattacharyya, D. K. Maiti, Indian Institute of Technology

37. Extended Hamiltons Principle for Fluid-Structure Interaction 491

H. Benaroya, T. Wei, Rutgers University

38. Estimation of Pressure Fields Using Velocity Measurements 507

M. Krane, P. Dong, T. Wei, Rutgers University

39. Implicit Multigrid Computations of Unsteady Multiphase 509

Flows in Varying Cross-Sectional Area Channels
S. Gkaltun, H. Saygin, Istanbul Technical University;
M. Muradoo lu, Koc University

40. Dynamic Stability of Stochastic Delay Systems 521

M. S. Fofana, Y. Yong, Worcester Polytechnic Institute
 Welcome to the IUTAM Symposium
Professors Haym Benaroya and Timothy Wei

We are very pleased that the opening day of our Symposium has finally arrived, after three years
of planning, invitations, contracting and fund-raising. We are also happy to welcome every one of
this distinguished group here to Rutgers University in New Brunswick, New Jersey.
We have a wonderful program of talks and plenaries that should give all of us a good idea of
the current thinking in the field of fluid-structure interaction. Speakers will discuss theoretical
methods and experimental techniques, and most importantly what the results tell us about how the
fluid and the structure interact and exchange energies.
Our colleagues are here from many parts of the globe, and from academe, industry and
government. This should be an exciting week for us all as we learn from each other and make new
acquaintances. We have prepared a variety of activities during the week, a dinner at Rutgers
Zimmerli Museum, a trip to Broadway, a tour of the campus and facilities, and enough time for
small groups to hold discussions.
A proceedings volume will be issued that will include all papers submitted that were also
presented at the Symposium. With the proceedings, we believe that a valuable document is added
to the literature. We hope you will agree.
As a final note, it is important to acknowledge all those who in one way or another helped
make the Symposium and the eventual proceedings a possibility. For their financial and moral
support we sincerely thank VP for Academic Affairs Joseph Seneca, Dean of Engineering Michael
Klein, Deputy Dean of Engineering Abdel Zebib, Chair of Mechanical and Aerospace Engineering
Andrew Norris, and Secretary Helene Press also of our department who very ably assisted with
onsite management, all of Rutgers University. We thank Monica Malouf off Meetings & More who
provided the bulk of the organization and fiscal management of the Symposium. She did an
exceptional job that we deeply appreciate. And finally, we very much appreciate all the years of
support of our research by Dr. Thomas Swean of ONR, who is also supporting this Symposium.
Finally we thank one of our graduate students, Pravin Subramanian, for the tremendous job of
pulling together all the papers and putting some uniformity to them for these proceedings.
 Welcome to Rutgers University
Dr. Jim Flanagan

INTERNATIONAL UNION OF THEORETICAL AND APPLIED MECHANICS

SYMPOSIUM OF INTEGRATED MODELING OF FULLY COUPLED

FLUID/STRUCTURE INTERACTIONS

It is a great pleasure for Rutgers the State University of New Jersey to host this timely and
significant convocation devoted to modeling of Fluid and Structure Interactions. The conference
organizers, Professor Haym Benaroya and Professor Tim Wei, have given new visibility to
Rutgers University in bringing together this body of distinguished scientists from around the
world.
Your technical program over 5 days covers a great variety of scientific frontiers in the
interaction of structures and fluid flow. More and more, these issues impact every aspect of our
modern lives ranging from the understanding of circulatory phenomena in our human bodies,
through the launching and retrieval of future space craft, to the design of novel musical
instruments (especially those that are wind powered), and the improved efficiencies of sea craft.
The new science and understanding you are bringing to these questions is conjunctive with our
great progress in computation. No longer do we have to wait hours, or even days, for a super
computer to provide 3-D Navier-Stokes solutions that permit us to alter the parameters of a
structure, and perhaps iterate the computation process. And, new techniques to visualize massive
data sets add impetus to the advances now possible. I congratulate you on your comprehensive
technical program and upon the archival documentation that will result from it.
While you are in our area, and especially since so many of you have come from afar, I hope
you will use the opportunity to get to know our university a bit better. Rutgers is the State
University of New Jersey and we have campuses located in three major urban centers New
Brunswick, Newark and Camden. You presently are located at our New Brunswick Campus
which embodies a number of schools and colleges on both sides of the Raritan River, beginning
with Agriculture and Marine and Coastal Sciences at the very southern extreme, to Liberal Arts
and Urban Policy and Planning in the central location, to Life Sciences, Physics and Engineering
on the northern side. We cover approximately 5,000 acres in this vicinity and our university
curriculum spans virtually every discipline that you would expect to find in a large research
university. Our enrollment is 50,000 and our graduating class a week ago numbered 10,000. So
you can imagine that we provide a significant contribution to the skilled workforce of the
northeast part of our country. We conduct external contract research which this year will exceed
$250M, a large part of this being from federal agencies such as the National Science Foundation
and the National Institutes of Health. Much of this research is conducted across the Raritan River
on Busch Campus in Piscataway, which is our Science Campus, and which we share with the
University of Medicine and Dentistry of New Jersey. I understand you may visit Busch Campus
during the course of your Symposium. We welcome your interest in our school and will be
delighted to show you our activities.
Again, thank you for honoring us by your participation in this distinguished conference. A
warm welcome to Rutgers! I know you will have a productive 5 days of intellectual stimulation
and exchange. And, I wish you a most pleasant visit with us.
 SURVEY OF NAVAL COMPUTATIONAL NEEDS IN
FLUID-STRUCTURE INTERACTION (INVITED)

Dr. PAUL PALO

Naval Facilities Engineering Service Center, Port Hueneme CA U.S.A.

Abstract: This plenary paper presents a broad survey of the computational needs of the U. S. Navy regarding
fluid structure interaction. Problems with the classic periodic/quasi-periodic response are emphasized,
complemented by some transient response problems as well. A collection of Navy problems are presented to
represent (I) operations on the land, the sea surface, below the sea surface, and in the air, (2) spanning the
many existing and emerging hardware classes and types in the U. S. Navy inventory, for (3) a wide range of
operational scenarios that experience fluid-structure interaction. The goals are to credibly illustrate the
incredible diversity of Navy issues in this area, and to provide sufficient descriptions to allow for a
meaningful appreciation of the computational needs.

1. INTRODUCTION

All of the attendees at this conference are well aware of the demanding
phenomenological and computational challenges that are inherent in every fluid-
structure interaction analysis. The problems emphasized at this conference - those with
strongly-coupled interaction - are particularly difficult. While there are encouraging
successes reported with growing frequency in the literature, it is noteworthy to
recognize that they are predominantly for narrowly focused, idealized conditions, with
a significant investment in the pre- and post-processing. Those facts, along with the
very advanced yet primarily still fundamental content of the papers scheduled for
presentation this week, show that fluid-structure interaction is still in its infancy. Each
paper will help advance the patient advance towards generally applicable, reliable, and
computationally tractable analysis tools. In the meantime, those isolated successes for
idealized flows and structures are noteworthy because they demonstrate that this field
may soon be capable of contributing to some real-world interaction problems.

Yet the jump from idealized to real-world scenarios would benefit from some
defmitive directions, or a framework, or perhaps even prioritization. That is the
objective of this opening keynote presentation. The conference organizers, along with
the Office of Naval Research (ONR), felt that the fluid-structure interaction problems

1
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 1-26.
2003 Kluwer Academic Publishers.
2

prevalent in the U. S. Navy would offer a diverse set of representative problems, with
varying degrees of complexity (read, modeling achievabi1ity).

Accomplishing that objective appears relatively straightforward: identify where

fluid-structure interaction interferes with the Navy mission. It is a simple statement, but
it does not have a simple solution. To start with, the scope was immense: the Navy
operates on the land, the sea surface, below the sea surface, and in the air. This meant
that a comprehensive survey needed to encompass all existing and emerging hardware
classes and types in "the U. S. Navy" inventory in all conceivable operational scenarios.
This spans the Fleet and Navy shore resources, the U.S. Marine Corps, and the Military
Sealift Command; there are presently 717,000 total employees, 301 ships in dozens of
classes, and over 4,000 aircraft in at least 50 variations. Because surveying this breadth
is essentially impossible, the flavor of this paper was necessarily shifted to a "sampler"
of Navy problems. Secondly, only "Navy unique", or in some cases "military unique"
applications are presented (for example, more universal problems like galloping
telephone lines and parachutes are not addressed).

The fma1 requirement was to precisely defme "fully-coupled fluid-structure

interaction". Interestingly, this meant quite different things to the people contacted in
this survey. The cases reported here were somewhat subjectively sorted into a
hierarchy of interaction categories. Applications that fit the classic defmition of hydro-
and aero-elasticity ("strongly-coupled exchange of energy between a flow and elastic
structural deformations") are emphasized per the theme of this conference. But other
quite legitimate applications are presented as well, including rigid and plastic body
deformations.

All of the case studies were collected from the Navy Science and Technology
Centers, and all represent actual field experiences. The goal is to convey the general
thrust of each problem, along with the critical constraints and/or conditions integral to
that problem. Evaluating candidate mathematical models is not part of this fie1d-
oriented survey. Describing Navy field problems this way - "as simply as possible but
not too simply" - will allow individual researchers to determine their capabilities and
deficiencies with respect to Navy field problems.

2. STRONGLY-COUPLED, ELASTIC, QUASI-PERIODIC PROBLEMS

2.1. Discussion

This is the type of problem emphasized at this conference. It is often described using
words such as "self-excited" and "unstable", and involves a resonant structural
 3

response. The fact that structural modes must be excited further requires that the flow
instability have some degree of [span wise] coherent spatial structure. Such responses
typically exhibit either periodicity or quasi-periodicity. The classic example of this type
of response is cable strumming, which is extensively reviewed next.

2.2. Underwater Cable Strumming

The Navy uses underwater cables for a wide variety of purposes. Sample applications
include: maintaining acoustic hardware in a specified configuration, searching for
mines, holding buoys in position, transmitting signals, and lowering objects during
construction. Strumming is a serious design concern in a significant number of Navy
cable operations.

Vortex-induced vibration (VIV) can negatively impact cable operations in

numerous ways. Figure 1 summarizes the design spiral for ocean cable structures, and
is formatted to emphasize the importance of VIVo The system parameters and
constraints - spanning cable size to overall system configuration to current profiles to
installation platform characteristics - are categorized in the skewed boxes along the top.
VIV is highlighted in the center. The first decision [diamond] box addresses whether
the design satisfies the operational objectives as determined in five separate analyses
[indicated by the shadowed boxes]. Note how VIVaffects all five analyses. In some
cases the primary concern is the "local" (defmed as on the order of the cable diameter)
effect, such as self-induced noise on hydrophones. In others, it is the overall change in
the "global" deformation of an acoustic array due to the increased drag. While in still
others, it is the extensive additional handling required to attach fairings. If any design
requirement is not satisfied, then some part of the system must be modified, and all five
analyses must be repeated. This spiral can quickly become an intractable computational
burden.

As outlined in the next several Sections, Navy underwater cable configurations and
scenarios range from relatively straightforward to quite complex. Several examples are
given, and each closes with a summary of why it is beyond the idealized strumming
scenano defined by a homogeneous, constant tension cable in orthogonal, uniform
flow.
4

i!
i 'f l~
i j

ill +- l ~ Jl
~ I-
n Ihl
It ..~

4! !
I !
I ,
I
I I
"
I ll! 111 I ~

Itf III I
!
I
I
]

.I
. i
q-
>
> f
I
~

;1 -&
if
II iI

II
II itt .,1
"''''''
@][illB ~I
II
H
 5

2.2.1. Underwater Ranges. The Navy operates numerous underwater training and test
ranges, consisting of a number of cables laid in a predetermined configuration on the
bottom. For example, Figure 2 shows the layout of the many underwater cables that
comprise the SOAR range off San Clemente Island in California. Note the scale in the
lower left comer.

Figure 2. Cable Configuration for the SOAR Range

Figure 2 illustrates the fact that it is often critical during installation to a priori
define the ship path and cable payout speed such that the cable will fall through an
arbitrary (assumed known) current field and touchdown precisely along a desired cable
path. Several of the Navy Science and Technology Centers maintain highly capable and
well-validated numerical structural models for "large displacement cable dynamics"
simulations for a wide range of underwater cable uses. "Large displacement" refers to
responses much larger than the cable diameter - in other words, everything but the
strumming response. By far the largest uncertainty in these models is the uncertainty in
the drag coefficient, because it typically doubles when the cable strums. While some
analytical capabilities exist to partially address this (Triantafyllou, 1990; and many of
6

the plenary talks at this conference), it is always difficult to answer for "real" systems
because of the many combinations of cable configurations, attached bodies, site
characteristics, and installation parameters.
This fundamental uncertainty over strumming led to an ocean experiment in 1990
specifically to collect validation data for planning the SOAR installation. That test
deployed 16 kilometers of 1.7cm diameter cable in 1.1 km of water depth. A 90-degree
"slow tum" was included midway through the deployment. The current profile, ship
track, and payout speed were recorded. Most importantly, numerous pingers were
attached to the cable to track the full descent trajectory. The test data were used to infer
the drag coefficient using the SEADYN numerical cable model (Palo, et. aI, 1992).
Candidate cable trajectories for each of the one-hour pinger descents were recreated for
a range of parametric drag coefficients. A number of numerically-generated descent
trajectories and touchdown positions for pinger #2 are illustrated in Figure 3 versus a
range of candidate drag coefficients.

xl ()4 Run (; 90 deg. Turn @ 1.5 kls w/MTO

drag multiplier SEAJ)YN Data

H=1
(--)=2 Pinger# 2
(-.)=2.4
(-+)=1.8

-7.38
Predicted
Touchdown,
Predicted
Touchdown,
notstrumming ......
.' ....
strumming
-7.42

5.7 5.72 5.74 5.8 5.82 5.84

Easting (ft) XlO4

Figure 3. SOAR pinger trajectory

 7

Comparisons to the measured trajectories (not shown for brevity) confIrmed that the
best overall fIt was with the "drag multiplier = 2" (solid arrow). In other words, the
cable did generally strum. Figure 3 also shows why the drag value is so important: the
incorrect touchdown position assuming a nonstrumming cable is in error by more than
375 meters compared to the correct position. However, the simulations presumed a
uniform drag coeffIcient over the entire wetted cable. But is that correct? Actually, it is
more likely that the degree of strumming (and increased drag) varied versus the cable
length and time. This was in fact confIrmed with additional data from an accompanying
steady-state tow test (included in the presentation).
Thus, for real-world Navy cable installations like this one, a robust "fluid structure
interaction" strumming prediction model must account for: variable tension over the
cable length; complex, 3-dimensional effective flow due to the [sheared] horizontal
current profIle versus depth, the vertical descent velocity, and the longitudinal "slip"
through the water; arbitrary payout rate and ship speed; and mass discontinuities
introduced by the presence of the pingers (and here, one deliberately placed heavy
body).

2.2.2. Towed Acoustic Receive Arrays. These arrays consist of two parts: an acoustic
array consisting of serially-arranged hydrophones, and a "separation" cable often on the
order of one mile long. This two-part cable is towed longitudinally at a specifIed depth
behind a surface ship or submarine. Beamforrning is used to acoustically detect and
locate underwater threats.
In practice, beamforrning is not initiated until the vessel/cable system reaches
steady-state equilibrium. The problem is that if the vessel changes course, it may take
on the order of an hour to reestablish equilibrium - meaning one hour of lost data.
There is presently a growing interest in continuing data collection during transient
maneuvers because beamforrning is possible as long as the array confIguration is
known. There are two candidate solutions for accomplishing this. First, numerical
modeling can be used to simulate the cable response. Strumming complicates this
approach in two ways. First, the simulations suffer from the same drag uncertainty as
the previous section. Plus, it is desirable to know a priori iflhow the cable will strum,
because added flow noise can render the data unusable. The second approach embeds a
fmite number of heading sensors in the array section to discretely defme its overall
confIguration. The danger with this approach is that "noise" from strumming can
overwhelm the steady-state heading "signal" - as illustrated with real data in Figure 4.
8

.'" 110 '_c

"' : . 519 1<1 : 5)

J\.-"
;
~~
f\I U 1,
<ii I I.
._--...

I,f
t
w v' .
w

---_..
II:
C)
w _ , .... w

e-o
I

z
o
I I _ _ .. t

~
J: ~ :).' '
.. I
.---~ I
[ 11 J

T IM E

Figure 4. Strumming as Measured by Heading Sensors

The complications for fluid-structure interaction in this towed receive array

example are (1) the primarily longitudinal flow over most of the cable, (2) the behavior
over the sections that are curved, and (3) the variable tension along the cable length.
 9

Power
Cable
Fairings

Tail for
Towing
LFA Stability
Source
Forward Stern
Suspension Suspension
Chain Chain
10

Lead
Depre

Inboard
Marker Float
Outboard
Inboard Marker
Tow Sweep Wire Inb
with Cutters
oard Outboard
Sweep Wire Outboard
with Cutters Otter
 11

The Mk103 system also exhibits serious strumming effects. The first impact is
logistical: left- and right-hand lay cables are used for the port and starboard sweep
cables to minimize the steady and dynamic drag. The second impact of strumming
remains a problem. The sweep depth is regulated by the position of the lead float on the
tow wire, and by the length of the marker float pendants attaching the marker floats to
the sweep wires. The drag (and degree of strumming) on the various cables varies with
the tow speed (and the unknown current), and as a result the sweep wires remain in a
horizontal plane only at one specific tow speed. Maintaining depth is critical to its
operation and is a continuing concern.
Fairings themselves can introduce new problems. A well-known example is
"kiting", wherein a steady unwanted net transverse lift force appears. This occurred in
the late 1980's in a towed system being developed to map ship wakes. The kiting force
from a small (4 inch wide) fairing tended to fly the towed body with more control than
the fins. The chosen fix was crude but effective: popsicle sticks were glued to the
trailing edge of one side of the fairing - hardly elegant, but it evidently worked.
Lastly, Figure 7 shows two of the lifting bodies included in the Mk103 system, and
these too could be contributing to the strumming behavior (of course they deliberately
kite too). It is quite interesting that past attempts to instrument the arrays and directly
measure the nonplanar "hogging and sagging" local depths along the sweep wires have
all been unsuccessful for the same, simple reason: the strumming was always so violent
it shook the sensors right off the cable!

Figure 7. AIN37U-J Depressor and MkJ03 Outboard Otter

The fluid-structure interaction issues associated with these mine countermeasure

systems echo the previous concerns. The cables have variable tensions and non-
12

orthogonal flow, and they do strum. Fairings provide relief in some cases, but in others
they are unacceptable due to the awkward logistic consequences.

2.2.5. Bottom-mounted Surveillance Arrays. The cable examples presented so far in

this section have emphasized Navy-unique configuration and handling aspects. This last
example returns to the simpler phenomenon of a single cable strumming in a uniform
flow - but with a twist.
Like the acoustic training and test ranges discussed in Section 2.2.1, surveillance
arrays are also bottom-mounted. There are two differences however. First, the sites for
surveillance arrays are dictated by tactical interest rather than convenience. If the
bottom is irregular (often), this can lead to a finite number of "catenary suspensions"
(which can be minimized by controlling the percent slack during payout, but
statistically speaking some suspensions are inevitable). Secondly, these cables are small
- typically fiber optic - to allow for installation from as many types of platforms as
possible. This makes them quite fragile, as well as only slightly negatively buoyant
(thus the tension can be considered relatively constant in analyses).
One of the critical design issues for surveillance arrays has been mechanical failure
due to wear at either end of a suspension. Strumming greatly accelerates failure because
it induces high frequency motions (see also the paper by Grosenbaugh and Han at this
conference). This can be induced by either waves, or steady or tidal currents.
The strumming-induced failure mechanism has been important enough to
surveillance array developers that several past array programs (FDS, Ariadne, and
ADS) sponsored abrasion tests. The latest array programs (SWTRlEC and FDS-C) have
surpassed their predecessors by building a unique test facility (the Controlled
Suspension Test Flume) and conducting a $l.SM test program over S years just to
collect high quality data over a range of conditions. This facility can test SO-ft long
cables in either a O.S knot flow with a 4-ft water depth, or a 1 knot flow with a 2 foot
depth. Fiber-optic cables can be suspended across the flume at varying tensions, with
the two ends drawn over various rock types to measure frictional wear due to
strumming.
 13

2.2.6. Summary, underwater cable fluid-structure interaction. The previous five

sections have illustrated just some of the Navy's applications for undersea cables.
Additional examples could have been presented on steady kiting loads on fairings, and
many other system configurations (test structures, buoys, ROV tethers, etc.). The
common theme in all of them is that strumming is a major problem. Unfortunately, the
state-of-the-art for predicting the occurrence and/or severity of strumming for these
types of complex systems is limited. The following general conclusions can be made:
- The Navy uses cables of all diameters and unit weights. System
configurations range from serially to branch connected.
- Some cables are towed at high speed, some suspended in low currents.
- Cable orientations can range from orthogonal to parallel to the flow.
They can be horizontal or vertical, straight or curved.
- Hydrodynamically speaking, there is a wide range of Reynolds
Numbers.
- Structurally, tensions typically change versus arc length along the
cable, and time. Note also that not all cable modes are "linear,
elastic ". Some are "nonlinear, geometric" wherein the cable exhibits
oscillations without elongation. This affects the strict interpretation of
"fully-coupled, resonant" energy transfer between the fluid and the
structure.
- Cable end conditions include: point masses that mayor may not
transmit vibrations from the adjacent span, lifting bodies, frictional
damping on rocks, and stiffness that varies temporally (say, due to
heave motions of a surface buoy) and/or spatially (due to distributed
drag).
- The primary concern can be either the high frequency/ high modal
dynamic motion (fatigue or sensor degradation), or the overall
increase in the drag.
- When strumming is unavoidable and unacceptable, fairings are used.
At best they add extra handling effort and cost; at worst they decrease
safety, invalidate the use of certain platforms (e.g., UUVs) or
equipment, and introduce a steady kiting force. This leads to two
observations:
o Many cables, typically synthetic lines, use a stranded
construction, so the shape is quite different compared to a
cylinder. Secondly, designing fairings (hard and ribbon) is still
more of an "art" than a science. Numerical modeling has
14

potential to improve our understanding of the steady and

dynamic flow around these shapes.
o More broadly speaking, the two present choices of using either
bare or faired cable are sometimes both unacceptable. So there is
also great interest in any "alternative approach" could allow for
significant improvements in the mission(s).

2.3. Naval Aircraft

Two categories of fluid-structure interaction that are strongly-coupled, elastic, and

periodic/quasi-periodic are: airframe vibration, and turbine blade vibration.

2.3.1. Frame Vibration. Flutter of airplane wings and other appendages is "the classic"
aero elastic phenomenon. It is definitely an ongoing and major problem for the Navy,
given its large number of aircraft classes, the long service life and regular cycle of
major aircraft upgrades, the continual introduction of new ordinance, and the extreme
mission envelopes.
Recent specific problems (and fixes) with one class - the F/A-18 - are presented
next, organized by the section of the aircraft affected. The F/A-18 is shown in Figure 8.
This short list demonstrates how prevalent this problem can be for just one type of
plane. Note M = Mach number.

Figure 8. Navy FlA-18 Hornet

A. Control Surfaces
i. aileron, M > 0.90, below 5K feet (added aileron shroud tripper strip).
ii. rudder (limited M > 1.0 surface deflection in control system).
 15

iii. Wing trailing edge (added control surfaces and wing ribs)
B. Wing
1. 5.6 Hz oscillations with heavy outboard stores (fixed with (a) Active
Oscillation Control (AOC) system plus placards for various stores, and
(b) by increased wing stiffness)
ii. Wingtip launcher/missile buffet vibration.
iii. Outboard wing pylon missile buffet vibration.
C. Empennage
i. Aerodynamic (LEX fence to break vortex) plus structural (vertical tail
root "cleats" and tip rib beef-up) fixes. Required extensive wind tunnel
testing plus scaled flight test data, and vertical tail fatigue tests.

Thus, airframe vibration is obviously a significant problem - not only for the F/A-
18 but throughout the Navy. There is no dependable solution process at this time.
Numerical models cannot practically handle the complexity of these problems
(configuration, number and range of mission envelopes), and data collection is often
prohibitively expensive. In the absence of a reliable solution scheme, most problems are
fixed by ad-hoc adding strakes, etc., until the problem is minimized.

2.3.2. Naval Aircraft Turbine Vibration. High-Cycle Fatigue (HCF) failure of gas
turbine blades is a major problem for Naval aircraft (see Turbine Engine meeting
reference). Higher engine performance requirements, fewer parts, lighter materials,
higher blade loadings and reduced stall margins have all increased HCF susceptibility.
As with the previous section, highlighting the problems with just one aircraft in the
inventory will serve well to illustrate the severity of this Navy-wide problem. In the
years 1982 to 1992, the F402 engine in the AV-8B Harrier experienced numerous
failures. HCF was attributed to 40% of the total failures . 13 different parts experienced
36 operational failures (compressor disks, spacers, stator vanes and blades, turbine
blades and disks). 17 aircraft were lost at a cost of $476 million. With respect to the
replacement F404 engine, 42% of the failures were attributed to HCF. 10 different parts
induced 180 HCF failure incidents. 5 aircraft were lost at a cost of $125 million, and
$60 million was spent on R&D, repair, and retrofit
The currently used HCF design methodology has evolved purely empirically, and
verification techniques have not worked. The empiricism is necessitated by the
enormously complex flow through a multi stage compressor and turbine which consists
of an oscillatory, three-dimensional flow, made more complex by virtue of the fact that
the flow is transonic (hence containing shock waves, shock boundary layer interactions,
16

partly separated flow) with strong blade row interaction effects and excitations caused
by the shedding of wakes from vanes, stator and rotor blades.
Recall that the Navy has over 4,000+ aircraft, many of which perform extreme
maneuvers. Aircraft fluid-structure interaction issues are revisited in Section 4.

2.4. Underwater Propeller Singing

This is yet another "classic" hydroelastic phenomenon. Propeller singing occurs when
trailing edge vortex shedding frequencies coincide with blade structural resonance
frequencies (especially torsional modes). This results in a feedback situation that can
damage the propeller and create significant acoustic noise. Figure 9 illustrates a typical
Navy propeller (for the U.S.S. Belleau Wood) and the phenomenon of trailing vortices.

Figure 9. Navy Propeller and Illustration of Singing

The simple fact is that no one can predict when singing will occur in propellers or
underwater maneuvering fins. Plus, in most cases, the fundamental singing
phenomenon is complicated because the incident flow is disturbed by the presence of
the hull and appendages. To complicate the issue even further, there have been cases of
geometrically identical propellers where one sings and one does not. Situations have
been observed where (1) a mere change of material properties of the structure
essentially eliminated singing, (2) a change of the structural characteristics had no
effect on singing, (3) trailing edge conditioning mayor may not effect singing, and (4)
a leading edge saw tooth vortex decorrelation device eliminated singing. The present
 17

process for dealing this problem is the crudest possible: trial and error experimentation
using techniques that have worked in the past.

2.5. Biomimetrics
This category refers to fluid propulsion systems that mimic nature. One example is
swimming robot fishes that mimic the efficiency of tuna by swirling their tails
synchronously with the alternating vortices shed from the body. The Navy interest here
is to field the most efficient possible UUV s to maximize their range. A second example
is the small craft (on the order of grams of weight) designed to fly undetected and
provide useful battlefield intelligence. While there are as yet no operational
biomimetric systems in the Navy inventory, there is increasing interest in them, and
hence a pending computational need.

3. STRONGLY-COUPLED, ELASTIC/PLASTIC TRANSIENT PROBLEMS

This second category in the survey illustrates Navy problems where fluid-structure
interaction is critical, but not via an exchange of energy through resonant modes; i.e.,
the response is not periodic or quasi-periodic. But they are nonetheless "strongly-
coupled" in terms of their relative interactions.

3.1. Wave-Induced Impact Loads on Surface Vessels

This was a difficult topic to categorize. It is true that this is not a well-understood
phenomenon. However, it is historically such an important topic that the Navy has
made sufficient investments and now has a workable design methodology. Is it
rigorous? No - it is based on extrapolation from experience. Plus, it separates the
problem into two independent parts: (a) estimation of the rigid body dynamics, then (2)
calculation of the resulting fluid pressures, which are then applied to a structural model
of the hull. Is it generally applicable? No - it is directed towards sleek monohulls.
Does it work for the majority of the Navy's vessels? Yes! So, strictly speaking using
that narrow perspective, there is no "Navy computational need".
One class of Navy hulls that is beyond this methodology are the small (35 to 82 foot
length), fast (45+ knots), planing craft used by the Special Forces. In moderate seas the
operators of these vessels can be subjected to accelerations of over 30 g's, which can
and do cause serious injuries. Studies show that the peak accelerations vary greatly
depending on how the hull (dead-rise) and water are modeled, as either: (a) rigid body,
(b) planar deformation of the underside versus impact pressure, or (c) elastic
18

deformation during impact. These observations also apply to the phenomenon of

slamming loads on the wide exposed undersides of catamarans in high seas.
A second category is hull-wave impacts due to encounters with extreme storm
waves (which is not routinely included in Navy design calculations). Large waves do,
on occasion, crush hulls and superstructures, on Navy and other ships. There is
presently no attempt to model or design for these occurrences. But the phenomenon
clearly involves significant (plastic) structural deformation that inarguably alters the
fluid pressures during the impact. One famous Navy example of this is the U.S.S.
Bennington, which suffered the damage shown in Figure 10 in a 1945 typhoon.
Modeling this fluid-structure interaction would seem to offer a fascinating challenge.

88 BEi NINGTON (CV 2~)

Figure 10. Damage to U.S.s. Benningtonfrom one large wave

 19

3.2. Underwater Explosions

This is unquestionably a Navy-unique technical topic. The importance of this aspect of

Navy vessels was made abundantly clear with the bombing of the U.S.S. Cole in
Yemen. All Navy combatants must pass full-scale "live fIre" survivability testing with
actual underwater charges as illustrated in Figure 11.

Figure 11. Representative Navy UNDEX Survivability Test

The physics associated with this phenomenon are far too complex for treatment here
(see Landsberg and Wardlaw references). Generally speaking, the explosion produces
an immediate shock pressure wave that induces [very low modal] hull deformation.
This is followed by a second, smaller pressure wave associated with the collapse of the
cavitation bubble that induces [high modal] hull whipping. This pattern of multiple
pulses (with additional bottom reflections possible as well) from one detonation is
clearly illustrated in Figure 12 from a laboratory test on a fInite length cylinder. The left
fIgure compares experimental versus numerical data. The right fIgure shows how
critical it is to account for the structural deformations.
20

'2000
V\lIOdtyoatJ
. 0000 - - CJ I:ul.1110n VOlOClty - - - - R;oK! Cylo.do,
Cav' ...: n VtlOely - - Oofoml.blo CyI .. do,
PrO$S Uf~ D.ua
0000

1 tF~avltalion Collapse
\ Cavitation C 1011

2EOO
\/: I Second Cavttallon
I Collapse
\ I
I II I
\ If
I '-
"0 2

Figure 12. Underwater Blast: Centerline pressure on cylinder

The fluid-structure survivability interaction issues are largest when the underwater
detonations are close-in rather than stand-off. The foremost problem is handling the
factor of 800 change in density associated with cavitation. There are two instances
where this occurs. The fIrst is the inception, overexpansion, and collapse of the
detonation cavity itself. Secondly, a cavity can instantaneously form along the hull
when it accelerates in the initial pressure shock wave. Additional physical phenomena
that must be treated include (a) bubble jetting, and (b) internal fluid and structural
effects when the body has "double-wetted surfaces" due to internal, liquid-fIlled
cavities (like the double hull on a submarine).
The Navy also has an active development program underway regarding the
clearance of surf zone obstacles prior to an amphibious landing. The fluid-structure
interaction challenges associated with this application include: (a) adequate structural
and energy transmission modeling of partially buried, submerged, and/or partially
submerged obstacles (e.g., I-beam tetrahedrons), and (b) the focused pattern of the
Continuous Rod Warhead (CRW) used. One key computational challenge is having
adequate resources to simulate these fully three-dimensional, time-dependent scenarios
with the necessary large standoff volumes relative to the CRW rod diameter and width
of a tetrahedron leg
 21

A number of industry and Navy numerical models are presently being advanced for
modeling these weapons-related fluid-structure interaction problems. Fundamental
experimental data for simple geometries are also being collected to provide critical
validation. Until these advances are available, shipbuilders will continue to design ships
for survivability based on their "experience and habits".

3.3. Shore Magazine Design

The Navy uses magazines for storing explosives at shore facilities. Because of standoff
requirements for safety, these magazines tie up a lot of expensive land area - for
example 100,000 pounds TNT equivalent of explosives ties up almost 250 acres. A new
Righ Performance Magazine (RPM) structural concept has been advanced which
allows for a 65% reduction in that land area. Properly simulating internal (accidental)
detonations for design requires air-structure interaction starting with the initial
detonation, followed by staggered subsequent sympathetic detonations, and ultimately
fragmentation of the [concrete] structure. Figure 13 shows the aftermath of a magazine
test explosion. Observe how the walls blew out to make an orthogonal pattern.

Figure 13. Aftermath ofNaval Magazine Destructive Test

Promising studies are underway to further improve the performance of the new
RPM design. The improvement comes from the strategic placement of barrels of water
inside the magazine. Tests have demonstrated that the presence of water near an
22

LOAD DISTRIBUTION AND

EXTERNAL FLOW ARRAY/SUPPORT CORE WINDOW
DECOUPLING LAYER

STRUCTURAL MACHINERY NOISE

MITIGATION ELASTOMER
SYNTEC
1-3 COMPOSITE (SINGLE CRYSTAL)
SENSORS/TRANSDUCERS
 23
Note that the structure includes a "decoupling layer" and also a "mitigation
elastomer" that isolate the mass of the sensors and its support frame from the hull. The
fundamental design difficulty is associated with the boundary layer, which includes
both transitional and fully developed turbulent flow conditions. This creates a spatially
and temporally complicated dynamic excitation, which in turn causes the array mass to
vibrate on the rubber "springs". The fluid-structure interaction comes in because the
flow characteristics within the boundary layer - particularly in the transition region -
are quite sensitive to these vibrations.
All of this creates excessive flow-induced acoustic noise. This emerging design
problem has yet to be resolved for real conformal arrays - for either fully-developed or
transition flow conditions.
4.2. Vessel Berthing

This is a transient, rigid body problem, motivated by the forces induced on piers when a
vessel is berthed. In the nominal scenario a vessel is pushed (primarily transversely, but
with some forward speed) into a pier by tugs at low speed. The not-so-rare scenario
addressed here occurs when the vessel accidentally impacts the pier with too much
forward speed, introducing a significant additive structural aspect to the problem. That
scenario has become even more significant as the Navy introduces the new "Modular
Hybrid Floating Pier" concept shown in Figure 15.

Figure 15. Cross-Section o/Navy Modular Hybrid Pier

24

This double-decker pier, made of post-tensioned, high-strength, lightweight,

concrete, floats with the tide on widely-spaced vertical center piles. Because of the
reduced number of piles, the forces associated with vessel berthing/impact have become
more important. The fluid-structure interaction issues are: (a) effect of the large
entrained water mass when the vessel decelerates, (b) blockage effects due to the water
displacing around the hull and under the shallow under keel clearance, and (c) increased
resistance due to the elevated water surface as the vessel-pier separation decreases.

4.3. Two-Body Problems

This category addresses problems where flow disturbances from an upstream body
induce free stream turbulence or perhaps organized vortices onto a downstream body.
In some cases the latter body then experiences fully-coupled fluid-structure interaction.
Examples include:

- Oscillating forces on fighter aircraft rudders due to alternating vortex sheets

shed from the leading section of the fuselage
- Turbulence behind air tankers that induce large dynamic oscillations in the
refueling hose. This is a significant operational problem for all military
aircraft. It is complicated by the fact that even for low speed cases where the
hose is below the aircraft wake, structural oscillations initiate at the zero
tension free end and often migrate up the cable. Drag on the trailing capture
basket does not seem to mitigate this problem. This could be an even larger
obstacle for the refueling of unmanned air vehicles.
- There are significant wake effects behind surface ships that greatly complicate
the deployment and recovery of unmanned underwater vehicles.
- In at least one instance, a Navy gas-fired boiler at a shore facility exhibited
quite severe physical oscillations. This was attributed to a number of
interacting causes: (a) harmonics introduced into the air intake by the fan blade
rotation, (b) pressure amplification by the gas flame, and (c) instabilities due to
an unnatural exit duct pipe configuration necessary to preserve the historic
character of the building. The oscillations were magnified by the supporting
framework of long girders that were mounted on flexible vertical columns.
 25

5. SUMMARY

This plenary paper and the accompanying presentation have highlighted field problems
involving fluid-structure interaction over a wide span of Navy operations. Considering
the vast size and versatility of the Navy's inventory, the cases presented represent
examples of a much larger problem. But even this limited set provides sufficient
evidence that fluid-structure interaction does hinder the Navy's ability to accomplish its
missions.
This survey has also established that there are no accurate and generally-
applicable design tools for addressing these problems. In the majority of cases the
state-of-practice is to either make ad-hoc adjustments and estimates based on historical
evidence, or conduct expensive focused tests directed at each specific problem and/or
candidate solution. Unfortunately, these approaches do not provide insight into the
fundamental problem, and neither can be considered reliable regarding their likelihood
of success.
So the opportunities for applying computational fluid-structure interaction modeling
to Navy problems appear limitless. Scenarios range from the "simple" resonant
strumming of underwater and in-air cables, to the "self-contained" flow field and
vibration of aircraft/ordnance bodies at various Mach numbers, to violent underwater
transient detonations and local hull structural collapse. Generally applicable and
computationally tractable design-oriented models for these phenomena are of course
still far in the future. But the Navy has taken the first steps in that direction by
sponsoring specialized numerical models, validation experiments tailored for specific
applications, and conferences such as this one. The Navy field problems illustrated in
this survey and those specialized models will mutually benefit from each other during
the long march towards truly design-oriented models.

6. ACKNOWLEDGEMENTS

This information was collected primarily from the following Naval organizations:
Office of Naval Research, Ballston VA
Naval Surface Warfare Center, Carderock MD
Naval Sea Systems Command, Explosive Ordnance Disposal Technology
Division, Indian Head MD & Coastal Systems Station, Panama City FL
Naval Underwater Warfare Center, Newport RI and Keyport WA Divisions
Space and Naval Warfare Systems Center, San Diego CA
26

Naval Facilities Engineering Service Center, Port Hueneme, CA

Naval Postgraduate School, Monterey CA
Naval Air Systems Command, Patuxent River MD
This survey would not have been possible without the participation of many Navy
engineers and managers, as well as some academic and industry engineers acquainted
with Navy systems, who took time from their official work to contribute to this study.
Most of the information in this paper was informally received either by email or
telephone.
This survey was sponsored by Dr. Tom Swean, the ONR Science Officer for Ocean
Engineering & Marine Systems. His personal and financial support is gratefully
acknowledged.

7. REFERENCES

Landsberg, M., Keel, A.J., Darel, A., & Ruben, K.M. Parallel, coupled hydrocode simulations of
weapon/target interactions for naval applications, 2nd International Conference on Fluid-
Structure Interaction, June 2003, Cadiz Spain.
Palo, P. A., Teragouchi, L. c., and Smith, M. T., Validation of the SEADYN90 Cable Simulation
Model Using a Three-Dimensional Cable Deployment Data Set, ASCE Civil Engineering in
the Oceans 5, College Station TX, November 1992.
Sarpkaya, T, A Critical Review of the Intrinsic Nature of VI V, IUTAM Symposium on Integrated
Modeling of Fully-Coupled Fluid-Structure Interactions using Analysis, Computations, and
Experiments, New Brunswick NJ June 2003.
Triantafyllou" M., Mooring Line Damping and Current Forces, Joint Industry Study via Noble
Denton and Assoc, Houston TX, 1992.
Wardlaw, A., Jr. and Luton J. A. , "FLUID-STRUCTURE INTERACTION MECHANISMS FOR
CLOSE-IN EXPLOSIONS", Shock and Vibration Journal, 7, (2000), pp.265-275.
2nd National Turbine Engine High Cycle Fatigue (HCF). Coord. meeting, sponsor WP AFB,
Naval Post Graduate School, Monterrey, CA, December 1996.
 NOVEL USE OF A BANDWIDTH MEASURE FOR
VORTEX INDUCED VIBRATIONS

Case Study.' The Foinaven Dynamic Umbilical

FLORENT TRARIEUX* AND GEOFFREY J LYONS**

*Department ofMechanical Engineering, University College London, London
WCIE 7JE, UK

**BPP Technical Services Ltd., Dilke House, 1 Malet Street, London WCIE
7JN, UK

Abstract: An innovative use of a method to detect Vortex Induced Vibration (VIV) bandwidth in the
curvature data of subsea flexible risers and umbilicals is presented. The parameter epsilon gives a valuable
estimate of the bandwidth of signals such as VIV as a single value which may be used to track the behavior
with time and against other measures such as current speed. The method is conveniently based on a peak
counting approach originally formulated by Cartwright and Longuet-Higgins. Low epsilon values close to
zero indicate a narrow band process whereas values near unity indicate a broad-band process. Comparisons
with the alternative kurtosis statistic are also presented. The epsilon measure is shown to have a more useful
linear behavior than kurtosis, and appears to be more sensitive and resilient. Curvature data gathered for
different periods in 2000 and 2001 by the monitoring system installed on Foinaven Petrojarl IV have been
analyzed.

Key words: Vortex Induced Vibration, VIV, FUMS, epsilon, kurtosis, bandwidth

1. INTRODUCTION

Vortex Induced Vibration (VIV) continues to be a source of major concern in the

offshore industry, particularly with risers and umbilicals. The accurate prediction of
response e.g. Lyons (1986b) proves to be rather difficult for realistic applications owing
to the complexities of geometry and environmental loading.
Given this, it is preferable not to rely upon prediction alone, but also to monitor the
behaviour in the installed condition. By this means it is possible potentially to take

27
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 27-42.
2003 Kluwer Academic Publishers.
28

corrective action in advance of likely (fatigue) failure, and also to improve our
understanding of the shortcomings of the predictions made.
Part of the complexity of VIV response is as a consequence of marine riser and
umbilical systems' ability to respond simultaneously to several modes of vibration. The
extent to which this may happen is described as response bandwidth. Plainly it is of
interest to measure this frequency response as well as having basic information
concerning amplitude. Whilst it is possible to perform spectral analyses, this approach
arguably provides too much information for routine consideration. A single-value
measure of bandwidth variation with time (and varying environmental loading) is
highly desirable. Some alternatives have been reported (and adopted by one of these
authors), and used to reasonable effect (System Identification, and kurtosis). This paper
presents another measure for this purpose which has existed for some time. We refer to
this as epsilon, and demonstrate that it has a number of advantages compared with the
alternatives.
Application of this technique is demonstrated with data from the Foinaven
Umbilical Monitoring System (FUMS) for which a wide range of VIV and
environmental loading data are available. This system was developed and installed and
is maintained by BPP Technical Services on the BP Foinaven Petrojarl IV FPSO in the
Atlantic Frontier West of Shetland province of the UK. It was designed to measure (for
the life of the field) the curvature in the umbilical, along with vessel motions and
environmental data. These data combined with current profiles enable effective analysis
of the dynamic response of the umbilical.
In the method presented, curvature time series are processed so as to identify the
occurrence of VIV using the bandwidth parameter epsilon which was originally
developed by Cartwright and Longuet-Higgins for the analysis of wave peaks, wave
pressure, and vessel pitch and roll. Epsilon very usefully has a range between 0 and 1
depending on the frequency content of the signal (narrow or broad-band process).

2. BAND WITH PARAMETER EPSILON

The basis for the technique presented is that developed by Cartwright and Longuet-
Higgins (1956) who carried out substantial work on the distribution of maxima of
random functions to investigate the distribution of wave heights and ship motions.
 29

Extension to consideration for VIV analysis was first reported for experimental and
theoretical data by Lyons (1986a).
The random function may be defmed as the sum of an infmite number of sine
waves:

with:

In : frequency in the interval [0, 00 ]

lPn : random phase in the interval [O,2Jr]

Cn : amplitude such that in any small interval of frequency dl:

f.#
L 1c~ = E(/)dl
fn=f

where E(/) is a continuous function of I called the energy spectrum of I( t)

They derived the following expression based on the proportion of negative maxima to
total maxima for the bandwidth termed epsilon:

(1)

1 N+
with: r = -(1--) (2)
2 N]
30

r : proportion of negative maxima

N; :number of up-crossing periods
N] : total number of maxima

The definitions of positive maxima, positive minima, negative maxima, negative

minima, zero up-crossing, zero down-crossing are shown in Fig 1.

Fmm (l) rulli (2) we obum, & ~ 1- ( ~ )' (3)

positive zero upcrossing

minimum down crossing
minimum

Figure 1. Some signal definitions

Epsilon is conveniently seen to depend only on the number of zero up-crossings and on
the number of maxima in the signal. It is this approach which has been applied as
presented herein.

Based on another approach an alternative (but more computationally intensive)

formulation of epsilon is:
 31

x1 = sin(2pi*0.05*t) x2 = sin(2pi*0.5*t) x3 = sin(2pi*5*t) x1+x2 x1+x2+x3

Sampling Rate = 20 Hz

-1

-2

-3
0 60 120 180 240 300 360 420 480
Time (sec)
32

According to (3), if we consider an infinite number ofsinusoids, the ratio

(~J~O,"dth'n c~l
i.e. Epsilon tends to unity for broad-band signals.

Considering the case of a narrow band signal, for a single sinusoidal signal the ratio

i.e. Epsilon tends to zero for narrow band signals.

3. ALTERNATIVE MEASURES OF BANDWIDTH

3.1. Kurtosis

Kurtosis is a measure of how outlier-prone (from the mean) the data are. It has the
advantage compared with spectral analysis of providing a single value as a measure of
bandwidth. However, unlike epsilon it does not have as useful a restricted range of
values to represent the narrow-band to Gaussian variation.

The use of kurtosis to identify VIV bandwidth was originally proposed by Vandiver
(2000). This has been reported applied to acceleration signals for relatively low
frequency response of tensioned risers compared with the curvature signal that has been
used throughout this study. The technique has also been used to good effect for the
Foinaven curvature data BPP (2000).
 33

For univariate data ~,1; ,... , YN :

N 4

I(Y;-Y)
Kurtosis = ....:;..;;;=1'--_ __
(N -1)s4

N : number of points

Y : mean

s : standard deviation

In the case of a normal (Gaussian) distribution, kurtosis has a value of 3. A

significant benefit of epsilon over kurtosis is that epsilon exhibits a narrower range of
values between 0 and 1 whereas kurtosis values are more widespread. Kurtosis is a
measure of the distribution of peaks of a signal only by inference. It measures the
distribution in terms of range of data points present in the signal (i.e. how outlier-prone
the data points are). Evidently there are more data points clustered at peaks and troughs
than at intermediate positions (since the a sinusoidal signal dwells at peaks and troughs,
and moves most rapidly through the mean position, where the data are relatively
sparse).
If there is a wide range of amplitude (compared to the normal distribution, such as
an exponentially decaying sinusoid) then kurtosis will be very high (substantially
greater than 3), which leads to an untidy variation with time (epsilon is limited to a
maximum of unity and remains tidy).
In the particular case of a single sinusoidal signal, kurtosis will have a value of 1.5 -
rather than 1.8 as had been incorrectly reported elsewhere - Vandiver (2000, 2003).
Kurtosis can produce useful results, but requires a relatively clean signal for best
performance, whereas epsilon is more resilient to noise.
34

3.2. System Identification

No examples are presented herein for System Identification techniques which could be
applied to VIVo Such techniques exist which permit a response spectrum to be
reasonably accurately described by a few values (but not normally a single value),
MandaI et al (1992). This has the advantage of representing a spectrum with low data
storage requirements.

4. THE FOINAVEN UMBILICAL MONITORING SYSTEM (FUMS)

The system was installed in 1996 by BPP Technical Services on the Floating
Production Storage and Offloading vessel (FPSO) 'Petrojarl IV' operated by BP on the
Foinaven Field located in the west of Shetlands in 465 metres water depth, Lyons et al
(1998, 2003). The system uses a specially developed curvature sensor which is
deployed in a spare 5/8 inch hydraulic control hose in the umbilical. The umbilical is of
a lazy-wave configuration, Fig. 3.

Figure 3. Foinaven umbilical configuration

 35

The sensor measures curvature at three locations noted Ll, L2, L3, in two
orthogonal directions in the region of the umbilical bend stiffener, Fig. 4. It is
terminated at the FPSO connector deck level and extends through the vessel depth to
the measurement region below the vesssel keel.
MIB
Curvature is detected using strain gauges Connector

configured in pairs for each location and

direction to ensure tension and temperature
effects are eliminated, and resultant curvature and
plane are calculated. The main processing unit
samples 16 channels of analogue data at 20
samples/sec/channel, for a total of 8192 samples
per channel. Hence each sampling period is just
under 7 minutes. The 20 Hz sampling rate was
chosen to deal with expected 3Hz vortex induced
vibration frequencies. The 7 minute sampling
period being sufficiently adequate to cope with a
few cycles ofFPSO surge and sway.
Signals for environmental (wave statistics) and
vessel motions are also read and stored as part of
the FUMS data set. Additionally detailed current
data through the water column are separately

;'1---
o
available at 10 minute intervals measured using L2

Acoustic Doppler Current Profiling. L3

Subsequent to acquisition the FUMS data are

processed which includes marking for quality,
filtering, calculation of statistical values
Figure 4. Umbilical exist from turret
(minimum, maximum, mean, standard deviation,
showing location of 3 curvature
kurtosis), incrementation of fatigue cycle
sensing locations L1, L2, and L3
counting rainflow tables, and storage of time
series data if defmed criteria have been met.
Where time series data have been stored it has
been possible to carry out further analysis
(epsilon, kurtosis, and spectral).
36

5. ANALYSIS OF CURVATURE DATA

5.1. Raw Data Filtering

The raw curvature data contain all the components of curvature owing to the wave
excitation, surface vessel motion and VIVo
In order to examine only the VIV components in the curvature signal, the raw data
were band-pass filtered between 0.2 and 2 Hz. This was since it is considered that wave
and vessel motion responses would not be significant below 4 seconds period (0.25
Hz), and allowing for imperfect sharpness of the filter. Plainly it is possible that some
VIV response may exist in the region below 0.2 Hz. However, evidence from the
variation in umbilical dynamic response with current indicates that this is probably not
significant, Lyons et al (2003).

5.2. Epsilon Analysis

Whereas kurtosis has been adopted since 2000 as a standard statistical measure as part
of the FUMS, epsilon has yet to be used in this way. Consequently it has only been
possible to perform the epsilon analysis on data which have been retained as time
series. It should be noted that statistical data (measured at 7 minute intervals) are
retained at all times by FUMS, but time series data are generally discarded unless
certain pre-set criteria are met.
The epsilon analysis was implemented in FORTRAN code processing sequentially
the daily records during the period considered. For the year 2000 period analysis
presented herein, 16 records of 7 minutes each were available from 16 June to 12 July
providing a total of nearly 2 hours every day.
 37

5.2.1 Variation of bandwidth with amplitude and current: Fig. 5 shows the results of
the epsilon analysis compared with kurtosis and standard deviation of curvature as well
as current data for the same period. These are shown in accumulated form such that
there is a spread of 16 values clustered about each day for each measurand. Current
profiles were recorded between 64 and 400 metres water depth. The current speeds
presented in Fig. 5 are records at 64 metres depth.

A close match between epsilon, amplitude (standard deviation of curvature), and the
current speed can be clearly seen. For greater clarity, I-epsilon has been plotted in order
to have this and amplitude and current speed varying in the same sense. Evidently there
is a trend towards narrow band response as the amplitude of VIV increases (with
increasing current).
38

-------. ....---- . .--...""""-_....... ----

. _ _.-_............
... _ -___...... .. - ...---

,~~mI~~~u~~~m~~~m~mm~~~~m~~~~~~~~~EI~~~~~!~~~~t~~~~~~~~~m
__
_
__
~_.

~
.....--
_ _-.-.--
_
____ 4 ....------.+_.
_~
_ _ ___ .... .,. ___
_ . _ .1. __
.~

~_.01_ .. _ ..
_.
~. ...
________ ___ __ .............
-----------;.~~----
01 _ . . . . . ______ - " - . _ . ___ ,_
___...--....---..--
..... __ .
_0'-. _ . . . . . . . __ _"'._

11I.-.InJJJ llhlllJ./Ll .!J...UIa !';!..t,n-W ;.-Lr_JJ .IO-AB-J} .t-.UWJ JI.Ln . OO

FOlt~A... eH Del P11 Cun,m Ploflle

b
~ ~ ~ ~ ~ ~., 1., i i I ~ 1'! ,
I ! , , 1 .
I:.I-
.:::
~ ----:,.......,
-';--'-c::
_""",,
";-'.-
" ----:-:'c::----:-,-;-::-----:;...,,----:,,;-,-7-:::---'--,.,;-----:----:~-~-,-:-----:-:-:-'
1. ,

Ia.rt~k (CHi : GUAv!t."Cl11

-1<----...-..-..--..,---......- .-1I( ...,.....-...-- -....--.-- . _ _ . _ _ _ - . . . .-- - - --. - ....,..... -. -. --

~ ~ f:~ f ~~ ~~~ ~~ ~ ~ ~I: -l""i---------

...---..........-y................ -...- . -...-..-. - -. - .... -.....-. - - - - - -.............

c
~~ ~ ~~~ ~ ~~ ~~~~ ~~i~~~~~~~~ m~~~~ ~11--~~~~~~
j --
f\: ~~ ~~f: --I--""--I---------
~ ~ ~ ~ ~ F~ ~ ~~ ~T ~j
_ _____ _ __ >C. __ ~--j--
-- ~ ...
__ .. -I - A--
-- 2--
II-- ___ _ . __- 1 -- I -- ~---'--
. ___ -- - -.
1 - ..
__ 11 ___ -II --
x ------------- . --
____ __ . __ . __ ~ ____ ----- -Jl -- lt - -- -- -- ---~--
1-- -- --1-- ---' --.--
---- --1'- . ----.,------.,---- - --.-- ---- -r -- - ---r------r--- -- - , --- --- ,-- ---- ...-------.-------r----- -
I I I ..., I

sde [CURVI;../LL1j

.....,
<1'::-0.

15-,..,rt-OO 1'1h1uIl'iIO ;:v....,yn-1)1) :n.Jvrt-OO N-.Iull'CIO ~...h,""OI) 1ifrr..r~I;O )~II/lPCIO Z- JII~W 4-,..,).(;0 ~IIJ.(IO 1IhI 1I~1X' 10.,..,10(11) l Z.,M.OO

Figure 5. Comparison of epsilon with standard deviation, kurtosis, and current speed (at
mid-depth). Channel 1 (X component of curvature at Ll), data 16 June to 12 July 2000
(VIV Filter 0.2 to 2 Hz). a: I-epsilon (grouped); b: current speed (grouped); c: kurtosis
(grouped); d: kurtosis (all and grouped - partial period); e: standard deviation (grouped)
 39

The trends are less obvious using kurtosis, although presentation elsewhere in non-
accumulated form does show discemable variation in a more obvious form with current
against a background of generally high values (greater than 3), BPP (2000).

The accumulated form of data presented is perhaps a rather harsh test of the
bandwidth measures, but it does show that epsilon is resilient. Whereas kurtosis is
perhaps best able to show trends for values limited to representing significantly
narrower bandedness it appears to generate less useful results for the broader banded

DIMly Variation of Epsilon

CH1: [CURVE_X_L 1)
06Febo01

(a)

Daily VIIri.tlon of Kurtosis

CH1: [CURVE_X_L1]
06-Feb-Q1

(b)

Figure 6. Epsilon and kurtosis for 64 successive 7 minute intervals. Epsilon shows significant
variation in bandwidth which kurtosis fails to identify.
40

cases, in particular frequently resulting in values greater than 3 (which theoretically

represents a Gaussian process).

A further comparison is made between these two measures of bandwidth in Figs 6 a

,b. These are from 64 consecutive records of 6th February 2001. Epsilon yields a steady
variation with a dip at around record 48 (epsilon approaching 0.5). Kurtosis appears to
remain at a value around 3 in general with no such dip, and a substantial number of
spurious values exist. Evidently in this case epsilon has been able to detect a variation
in bandwidth which has not been identified by using kurtosis.

6. CONCLUSIONS

The bandwidth parameter epsilon was originally proposed by Cartwright and Longuet-
Higgins in 1956 to investigate the distribution of wave heights. The epsilon parameter
based on a peak counting approach has been successfully applied to the curvature
response of a subsea umbilical with particular application for VIVo
Epsilon exhibits a scatter of values between 0 and 1 depending on the frequency
content of the signal and enables rapid assessment of the VIV regime with time
(usefully over periods of hours as well as days). The use of the epsilon method indicates
that vortex induced vibrations may build up gradually (in terms of amplitude) with
associated narrowing of bandwidth as a consequence of varying current.
It is expected that the epsilon measure will provide a useful tool with which to
investigate further effects such as the influence of wave and vessel motion flow
interaction effects on VIV, and other hydrodynamic response.

7. ACKNOWLEDGEMENTS

The authors thank BP Exploration and Production for the publication of the Foinaven
Umbilical Monitoring System data as well as the current profiles supplied by Fugro
Geos.
 41

8. NOMENCLATURE

ADCP : Accoustic Doppler Current Profile

FUMS : Foinaven Umbilical Monitoring System
FPSO : Floating Production Storage and Offloading Unit
VIV : Vortex Induced Vibration

9. REFERENCES

BPP Technical Services Ltd., 'Analysis of Umbilical Performance on the FPSO Foinaven Petrojarl,
Measurement Period 16-Jun-00 to 12-Jul-00', Report ref: TS95/0202 Re: (14) 2/00, 2000.

Cartwright O. E. and Longuet-Higgins M. S., 'The Statistical distribution of the maxima of a random
function, Proceedings of the Royal Society', A, Vol. 237, pp 212-232,1956.

Lyons G J, 'On the Hydrodynamics of Tensioned Slender Flexible Cylinders', PhD Thesis, University of
London, 252 pages, 1986a.

Lyons G J and Patel M H , 'A Prediction Technique for Vortex Induced Transverse Response of Marine
Risers and Tethers', J. Sound and Vibration, Vol 111, pp 467-487, 1986b.

Lyons G J, Brown 0 T, Cook H H, Walls B, Barnay G, 'The Foinaven Umbilical Performance Monitoring
System - A New Approach', Proceedings of the Offshore Technology Conference, OTC 8883, 1998.

Lyons G J, Vandiver J K, Larsen C M, Ashcombe G T, 'Vortex induced vibrations measured in service in the
Foinaven dynamic umbilical, and lessons from prediction', Journal of Fluids and Structures, 17(8),2003.

Mandai S, Witz J A, and Lyons G J, 'Reduced Order ARMA Spectral Estimation of Ocean Waves', Applied
Ocean Research, Vol 14 No 5 pp 303-312, 1992.

Vandiver J. Kim, 'Predicting lock-in on drilling risers in sheared flows, in Flow-Induced Vibration', Ziada
and Staubli Editors, Proceedings of the 7'" International Conference on Flow-Induced Vibration - FIV 2000,
Lucerne, Switzerland, 19-22 June 2000.
42

Vandiver J. Kim, and Lyons G J, personal communication, 18th May 2003, at Fluid Structure Interaction
Symposium, IUTAM, New Brunswick, NJ, USA.
 CHAOTIC VIBRATION-BASED DAMAGE
DETECTION IN FLUID-STRUCTURAL
SYSTEMS

BOGDAN I. EPUREANU
Department of Mechanical Engineering
University of Michigan
3138 G. G. Brown Building
Ann Arbor, MI48109-2125
epureanu @umich.edu

Abstract: A panel forced by unsteady aerodynamic loads is investigated in the

supersonic flow regime. This aeroelastic system exhibits complex dy-
namics including limit cycle oscillations, and chaos. A finite-difference
method and proper orthogonal decomposition are employed to detect
parametric changes. The sensitivity of the chaotic dynamics to para-
metric changes is shown to be an effective tool for detecting structural
modifications, such as variations of the stiffness of the upstream and
downstream mounting points. Most current studies of such problems
are based on linear theories. In contrast, the results presented herein
are obtained using chaotic dynamics, and have the advantage of an in-
creased accuracy in detecting damage and monitoring structural health.

Keywords: Buffeting panels, structural health monitoring, chaotic dynamics

1. INTRODUCTION
Structural health monitoring refers to the use of non-destructive sens-
ing and analysis of system characteristics for the purpose of detecting
structural changes which may indicate damage [ Farrar et al., 2001 ].
Assessing structural integrity provides numerous benefits including in-
creased safety, enhanced asset availability and readiness, reduced logis-
tics burden and cost of operation. Health monitoring addresses commer-
cial needs such as expanding the lifetime of aging assets and extending
operational capabilities.

43
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 43-58.
2003 Kluwer Academic Publishers.
44

Structural monitoring is a pervasive need in air [Roemer et al., 2001],

space [ Kacprzynski et al., 2001b ], sea [ Kacprzynski et al., 2001a ] and
land structures, and may be applied online (during normal operation of
the monitored system) or offline (e.g. during maintenance cycles). On-
line is generally preferred to offline monitoring in most applications (e.g.
monitoring air-frames, unmanned air vehicles, the joint strike fighter,
civil buildings and bridges). However, the technical difficulties associ-
ated with the overlap of normal operating conditions and the measure-
ment requirements have limited the applicability of current techniques
to mostly offline damage detection.
A class of techniques used for structural health monitoring are based
on the vibratory response of a structure. Most such methods moni-
tor frequencies and modes of vibration [ Doebling et al., 1996 ], and
their variation is associated with damage. Several other vibration-based
techniques have been used as well, such as subspace identification and
updating [ Abdalla et al., 1998; Zimmerman, 2000 ], wavelet analyses
[ Amaravadi et al., 2002; Sohn and Farrar, 2001 ], evolutionary algo-
rithms [ Zimmerman et al., 1999 ], Ritz [ Cao and Zimmerman, 1999 ]
or stochastic methods [ Sadeghi and Fassois, 1997 ] and others.
Most of these techniques are designed for linear vibrations while much
fewer apply to nonlinear systems. The early nonlinear methods are based
on system identification [Agbabian et al., 1990; Feeny et al. , 2001; Ljung,
1999; Masri et al., 1987 ], while some recent ones are using neural net-
works [ Zang and Imregun, 2001 ], phase space reconstruction [ Chelidze
et al., 2002; Cusumano et al., 2002; Chatterjee et al., 2002 ] and moni-
toring Lyapunov exponents of the attractor of naturally chaotic systems
[ Chancellor et al., 1996; Wong et al., 2001 ] or linear systems excited
chaotically [ Pecora and Caroll, 1996; Trickey et al., 2002 ]. Despite the
development of these methods, robust and sensitive damage detection
techniques for nonlinear systems continues to be an important challenge.
The main reasons for the development of nonlinear vibration based
damage detection are: (i) nonlinearities are a pervasive presence in many
systems affected by damage, (ii) linear techniques are not accurate when
applied to nonlinear systems because often the presence or level of dam-
age cannot be distinguished (by linear methods) from the nonlinear
healthy system dynamics, (iii) most current damage detection approach-
es do not take advantage of nonlinear dynamics, but rather cope with
their influence or minimize it.
First, nonlinear phenomena are important not only in large but also
in small amplitude vibrations when the strength of nonlinearity is high.
For example, a crack in a plate or a lose joint may lead to changes in
linear coefficients such as stiffness and damping, and can also induce non-
 45

linear dynamics by inducing Coulomb friction and allowing the opening

and closing of a gap (piece-wise linear dynamics with possible impacts
[ Vors et al., 1998]). Second, linear techniques are independent of the
amplitude of vibrations, and thus, the stiffening or softening of a healthy
system due to nonlinearities is often erroneously assumed to be caused by
damage. Third, compared to linear systems, nonlinear systems are much
more sensitive to parametric changes in the neighborhood of bifurcation-
s [ Guckenheimer and Holmes, 1983]. Moreover, nonlinear (limit cycle
oscillations) and chaotic vibrations have been observed in the aerospace
industry (wings [ Dowell, 1989; Kim and Lee, 2000 ], panels forced by
flow-induced loads [ Bolotin et al., 2001; Epureanu et al., 2003a; Epure-
anu et al., 2003b; Niho et al., 2000; Reynolds et al., 1993 ]), the nuclear
industry (heat exchangers [ Copeland and Moon, 1992; Langthjem and
Sugiyama, 2000 ], pipes conveying fluids [ Paidoussis, 1998; Paidous-
sis, 1986 ]), telecommunication and sensor industries (micro-mechanical
systems [ Turner et al., 1998; Miller et al., 1997 ] such as micro-plates
and beams), and the military (spacecraft thermo-shielding panels). De-
spite the growing need of nonlinear methods in these industries, there
are relatively few methods designed to exploit nonlinearities for damage
detection [ Loh and Tou, 1995; Smyth et al., 1999]. Also, most de-
tection methods focus on structures and are not designed to overcome
the complexities encountered in fluid-structural systems, a critical issue
especially for online monitoring.
Detecting parametric variations and damage detection are of great sig-
nificance in various industries, and are a pervasive need in engineering
applications. They are motivated by intellectual reasons and by indus-
trial requirements. For example, advanced methodologies for monitor-
ing structural integrity are needed by the manufacturing and aerospace
industries. A highly publicized case, reiterating the importance of de-
tecting parametric variations in fluid-structural systems, is the recent
catastrophic accident of an airplane whose vertical stabilizer had been
damaged by fatigue and led to a tragic crash. This paper discusses ad-
vanced vibration-based methods designed to predict such failures and
assess structural integrity.
Chaotic dynamics has been observed in a wide variety of systems
[Kantz and Schreiber, 1997; Moon, 1992]. A special class of exciting sys-
tems which exhibit chaotic dynamics are fluid-structural systems where
the interaction between a structure immersed in a fluid and the flow-
induced forces acting on the structure often lead to complex dynamics
[ Copeland and Moon, 1992; Thothandri and Moon, 1999; Wu and Zhu,
1997]. This dynamics has been investigated mostly by using time do-
main methods [Abarbanel, 1996; Abarbanel et al., 1993; Banbrook et al.,
46

1997; Broomhead and King, 1986; Darbyshire and Broomhead, 1996 ]

which are aimed at determining features of chaotic time series [ Fraser
and Swinney, 1986; Nerenberg and Essex, 1990; Schreiber, 1995; Wolf
et al., 1984 ] and characterizing various properties of chaotic systems
[ Grassberger, 1990; Grassberger et al., 1991 ].
In addition to these particularities of chaotic dynamics, the feature
monitored to detect structural changes in this paper is the level of coher-
ence in the dynamics of the system. The approach used for identifying
these coherent structures is proper orthogonal decomposition (POD)
[ Epureanu and Dowell, 2001; Epureanu et al., 2000; Epureanu et al.,
2001]. This approach requires measurements of the dynamics of the
system of interest over a time interval. A model for the spatial coher-
ence of the dynamics is constructed based on these measurements. For
linear systems, the models obtained using POD are similar to models
obtained by modal analyses. However, distinct from linear modal anal-
yses, POD may be used for nonlinear systems. Holmes, Lumley and
Berkooz [ Holmes et al., 1996 ] and Sirovich [ Sirovich, 1987 ] also used
POD in the context of turbulent flows as a technique which allows for
the identification of naturally forming coherent structures from numer-
ical simulations or experiments. These structures contain most of the
energy and are the most important components of the dynamics [ Feeny,
2002; Kappagantu and Feeny, 2000 ].

2. A PARADIGM OF A CHAOTIC
AUTONOMOUS SYSTEM
Exciting results have been obtained by investigating a panel forced
by buffeting aerodynamic loads. This aeroelastic system includes struc-
tural nonlinearities due to the coupling between the bending and the
elongation of the panel (Fig. 1). The aerodynamics is considered linear
and piston theory [ Lighthill, 1953 ] is used. The panel displacements
are considered much smaller than its chord and comparable to the pan-
el thickness. The panel is considered homogeneous, isotropic, and two-
dimensional, and it is modeled using nonlinear von Karman plate theory.
Such panels have been studied extensively by Dowell et al. [Dowell and
Voss, 1965; Dowell, 1966; Dowell and Ventres, 1970; Dowell, 1975 ] in
the supersonic flow regime. They observed that the interaction of dy-
namic (flutter) and static (buckling) instabilities leads to very complex
dynamics, which includes, static deformations, limit cycle oscillations,
and chaos. Most of the previous studies of the dynamics of panels un-
der buffeting aerodynamic loads are based on the Galerkin method for
numerical simulations. Distinct from those methods, a finite-difference
 47
unsteady upper surface pressure
Flow
u.. p~
~
constant lower surface pressure

"/ kupstream

Figure 1. Two-dimensional buffeting panel.

method is used herein, and POD [ Epureanu et al., 2001; Epureanu

et al., 2000; Holmes et al., 1996; Azeez and Vakakis, 2000 ] is used to
detect parametric changes in the aeroelastic system. The sensitivity of
the chaotic dynamics to parametric changes is shown to be an effective
tool in detecting damage, such as loss of stiffness in the upstream and/or
downstream mounting points of the panel.
Fig. 1 shows schematically a panel which is forced by a flow along
its upper surface, while its lower surface is considered at a constant
pressure. In addition, a compression pre-load acts on the panel at its
pinned ends. The flow is considered two dimensional (x, w), and the
panel is assumed one dimensional (x). Although seemingly simple, this
system may exhibit very complex behavior ranging from static buckling
to strongly chaotic (with maximal Lyapunov exponents larger than 4).
Furthermore, for certain cases of panel mounting, the dynamics of the
panel increases its complexity, and exhibits spatio-temporal chaos similar
to a weak turbulence [Reynolds et al., 1993; Reynolds and Dowell, 1993].
The aeroelastic model is obtained by inserting the expression for the
pressure differential across the plate into the structural model. The
aero elastic model thus obtained may be expressed in nondimensional
form as

W'lli

(1)

where tV = 8w/8t, w' = 8w/8x, and tlps is the static/steady pressure

differential across the plate, i. e. the pressure difference at zero flow
velocity, Uoo = o.
Nondimensional coefficients are obtained from Eq. 1 as follows. The
flow velocity is characterized by .x, where .x = (PooU?xP) / (M D). The
in-plane pre-load of the panel is proportional to R x , which is given by
Rx = EhryZ2 / D. The static pressure difference across the panel is char-
acterized by P, where P = - (Z 4 tlps) / (Dh). The panel material is
48
(a) Case R. 0
1 , . ~--'-r-'---~"-'--'" ",' ---.--.-~ .. -- . . (a) Case A _ 0 (b) Case A _ 1.739 x 10'"

~,~!i1\IINlt\IiI\II.(I~A\~llll\!,r'I,N~\~iI~~il~\\~~~/1
2 -'- ~"~-------c

I:'- ;:- ;:- - ;:-;:- - ;:-; - - :' L_

-1 ~-. "' -- .. --.--, .. -- .. -..-......... -.~ ----~~-----"
(b) Case R _ 1.739 x 10...

-:~
' 1----' ' -
(c) Case R ,., 2 x 10-8
---r-- ~~~~~
.r:.
8
--0.2 --0.1 0.0 0.1
(c) Case A =2x 10'"
0.2 --0.2 --0.1 0.0 0.1
(d) Case A '" 2 X10-'
- 0.2

~ 2
O~!~
-1 '
I :5 1

(d) Case R _ 2 x10-'

'0
"g'" 0

('j) -1
>
ON---~--~---_'-'j
I
-1 _ _ _ _ _ __ .L. ______ , ___I
-~.2 -~.~2--O~.1-----C-O.O~0C-C.1-0.2
~ ~ .",_. ~ . __

o 10 20 30 40 50 --0.1 0.0 0\-- 0:2

Nondimensional time Amplitude of the first coherent structure

Figure 2. Left: Time series of w(1/4) and Right: Poincare plots of the first coherent
structure; for four distinct upstream springs.

characterized by G = (Eh 3 ) j (2D) = 6 (1 - l/2). The mass ratio is giv-

en by J-L = (Pool) j (ph). These parameters may be used in Eq. 1 to
obtain

The boundary conditions at the upstream end of the panel may be ex-
pressed as

w"(x = 0, t) 0,
w(x = 0, t) Rupstreamjdownstreamwlll(x = 0, t) 0,

where Rupstreamjdownstream = Dj(kl 3 ) with k being the stiffness of the

upstream or downstream panel attachment spring.

3. RESULTS
Chaotic dynamics is typically more sensitive to parametric variations
than other types of behavior. For example, stable limit cycle oscillation-
s of the panel forced by aerodynamic loads are less influenced by the
changes in the rigidity of the upstream mounting point. To investigate
the sensitivity of the chaotic dynamics to parametric changes, a set of
parameters of the aeroelastic system have been chosen based on a bifur-
cation diagram of the dynamics [ Epureanu et al., 2003b]. For a fixed
A = 150 and a varying R x , the panel is stable and flat for Rx below
the primary flutter boundary which occurs at Rx = -2.587[2. For larger
in-plane compression loads, the amplitude of the period-one limit cycle
 10 ~~

B
l
-------
IO------<:' CaseR=1 .739X10..s
1 ' -- 'CaseR=2X10~
-,- ,,-
/\
'~
-l 25 [ ---- -
:Io--o case-fl = -O-_~
:1 ------CaseR
, I
2x10
: ,
::
49

~ ~ ~ 20
z

R ,1\ i
E i
t.1,
y~ ? ~" ,~ \,~ ~

\
6 ! 1s

~ p' '\ i ~ i

14
2 rI ~' "
I.rl~ '~~'g
5
i"t
W,~ oocl"cr,oCq-.-~A:.=\~
'
'~ -. , '-II"-'CD C

o i.c;jO _ J . _ " a I\-:df):/ ~-",~,,!,_~-._........, ~~..(:I,~

-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15
Amplitude of first coherent structure Amplitude of the first coherent structure

Figure 3, Histogram of point distributions in the Poincare plots for the first domi-
nant mode, Left: (b) and (c), Right: (a) and (d),

oscillations increases until secondary bifurcations take place, and the

complexity of the motion increases. For example, two secondary bifur-
cation points occur for Rx = -2.907[2 and Rx = -3.397[2. Chaotic mo-
tions are observed within the interval Rx = -3.927[2 to Rx = -9.177[2.
Within the chaotic region, there are two windows of periodicity, where
the chaotic dynamics is suppressed and limit cycle oscillations occur in-
stead. These windows of periodicity are observed from Rx = -4.307[2
to Rx = -4.467[2, and from Rx = -5.187[2 to Rx = -5.277[2. For values
of Rx less than -9.177[2, the in-plane load is large enough in magnitude
to suppress the chaotic motions. In the following, a chaotic regime has
been selected by setting the parameters to Rx = _47[2, .A = 150, P = 5,
v = 0.252, and J-l,fM = 0.1.
First, we focus on the detection of loss of stiffness in the upstream
mounting point. Time series obtained for four distinct nondimensional
rigidities: (a) case R = 0; (b) case R = 1.739 x 10- 8 ; (c) case R =
2 x 10- 8 ; and (d) case R = 2 X 10- 7 have been obtained. For simplicity,
only the displacement of the quarter point of the panel is analyzed, i. e.
w at x = 1/4. A clear distinction between cases (a), (c) and (d) has been
observed. This distinction reflects the large differences in the rigidity of
the upstream mounting point (which varies from zero to order 10- 8 , and
then to order 10- 7 ). A much finer distinction is observed between cases
(b) and (c) which correspond to stiffnesses differing by 15%. All cases
exhibit chaotic oscillations, and are characterized by maximal Lyapunov
exponents larger than 2.
The time series indicate that a simple observation of one point of the
panel may reveal large parametric changes. However, small parametric
changes may be more difficult to detect, and a more elaborate analysis
50

is necessary to decompose the dynamics of the plate in components cor-

responding to coherent structures (or dominant modes) present in the
dynamics. The POD method is used to detect coherent structures, and
a correlation matrix is obtained based on 100 snapshots collected over
the time interval t = 0 to t = 50. Similar snapshots are collected for
each value of the upstream end point rigidity. The rapid decrease in the
magnitude of the eigenvalues of the correlation matrix indicates that a
strong spatial correlation exists between the dynamics of various points
on the panel [ Epureanu et al., 2001; Epureanu et al., 2000; Epureanu
and Dowell, 2001; Epureanu et al., 2002; Epureanu, 2003 ]. The spatial
coherence is consistent with the temporal chaotic behavior observed, be-
cause the dynamics of each coherent structure is chaotic in time. It is
interesting to note that the shape of the dominant structures vary with
the parameter R. However, the variation of the dominant modes is small
even for large variations in the value of R.
The dynamics of each of the dominant coherent structures is chaotic.
However, qualitative differences may be observed between the four cases.
Large differences are observed, especially for the first dominant coherent
structure even for small parametric variations. To detect these changes
we computed Poincare sections of the strange at tractors of the dynam-
ics of the quarter point and each of the dominant coherent structures.
The Poincare sections were obtained by sampling in time at every 0.01
nondimensional time units. Similar to the observation of the time series,
the Poincare sections reveal large parametric changes. In contrast, the
dynamics of the coherent structures provides much more detail. Fig. 2
shows a clear distinction between the four cases. Observing these at-
tractors, one may detect changes of as low as 15% in the stiffness of
the upstream end spring. The first dominant coherent structure reveals
most clearly these differences.
The relative distribution of points in the Poincare sections shown in
Fig. 2 may be investigated and quantified using various techniques devel-
oped within the pattern recognition community [ Bow, 2002; Pavlidis,
1982 ] to detect changes in their geometric appearance. One of the
simplest methods is based on histograms. Fig. 3 shows the histogram
obtained for cases (b) and (c), where R = 1.739 X 10- 8 and R = 2 X 10- 8 .
A quantitative difference may be observed, especially around the max-
ima at 0.075. An even more dramatic difference is observed between
cases (a) and (d) where R = 0 and R = 2 X 10- 7 . More detailed analyses
may be performed to quantify the differences in the shape and distribu-
tion of points of the at tractors shown in Fig. 2. Nevertheless, the key
aspect is that decomposing the dynamics of the aero elastic panel along
a basis of the vector space spanned by the dominant structures of the
 51

Figure 4. Histogram for the first dominant mode. Left: Rdownstream = 0 Right:
Rdownstream = 1.75 X 10- 9 The horizontal axes represent the position and velocity
of the dominant structure, while the vertical represents the number of points.

dynamics, and employing Poincare sections of the strange attractors for

each coherent structure is a valid and powerful method for detecting
parametric changes.
Next, we investigate the shape of the attractor of the dynamics of
the dominant coherent structure for various changes in both the up-
stream and downstream mounting points. Fig. 4 shows the histograms
obtained for Rupstream = 1. 75 X 10- 8 and Rdownstream = 0 (on the left);
and Rupstream = 1. 75 X 10- 8 and Rdownstream = 1. 75 X 10- 9 (on the
right). The difference between the level and location of damage may be
observed by the different shape of the histograms not only for the case of
a rigid upstream (as in Fig. 3), but also for a softer upstream end spring
(as in Fig. 4). Furthermore, Fig. 5 shows the histograms obtained for
Rupstream = 1.75 x 10- 8 and Rdownstream = 2. X 10- 9 (on the left); and
Rupstream = 2. X 10- 8 and Rdownstream = 2. X 10- 9 (on the right) when
both the upstream and downstream mounting points are damaged. A
distinction can be made between these two damaged states indicating
that the attractor shape is a feature sensitive to both the location and
the magnitude of the damage.
The effectiveness of the proposed technique may be estimated by com-
paring it to a standard linear analysis where stiffness loss in the structure
is monitored. A linear model may be obtained by setting the parameter
G in Eq. 2 to zero, i.e. linearizing the model. Next, the modes and
frequencies of vibration of the linearized system may be obtained. The
frequencies of vibration have been determined for various stiffnesses of
the upstream spring corresponding to R values from zero to 10- 5 . De-
52

Figure 5. Histogram for the first dominant mode. Left: Rupstream = 1.75 X 10- 8
Right: Rupstream = 2. X 10- 8 , The horizontal axes represent the position and velocity
of the dominant structure. The vertical axis represents the number of points.

spite the large change in the parameter R (100 times larger than the
changes studied by nonlinear means), the variation of the first three
(linear) modes and frequencies has been very small, of a magnitude less
than 0.05%. In contrast the monitored features obtained by the nonlin-
ear analysis are much more sensitive, and undergo changes of more than
25% for a much smaller variation of the parameter R.

4. CONCLUSIONS
The dynamics of an aeroelastic system composed of a panel forced
by unsteady buffeting aerodynamic loads and undergoing limit cycle os-
cillations and chaos has been investigated. The nonlinear von Karman
plate theory has been used to obtain a model for the panel, while a linear
piston theory has been used to model the quasi-steady flow forcing the
panel. The upstream end point of the panel has been considered sup-
ported by a spring of variable stiffness. Changes of 15% in the stiffness
of the spring have been detected by exploring the chaotic dynamics of
the panel. The sensitivity of the chaotic behavior to parametric changes
has been shown to be an effective tool in detecting structural changes.
A finite-difference method has been used to simulate the dynamics of
the aeroelastic system, and coherent structures of the panel dynamics
have been identified using proper orthogonal decomposition. The sensi-
tivity of the dynamics of the coherent structures to parametric changes
has been discussed. The dynamics of the panel and flow have been de-
composed along a basis of the vector space spanned by the dominant
 53

structures of the dynamics. Poincare sections of the strange attractors

for each coherent structure have been used to detect parametric changes.
Finally, the inverse damage detection problem is investigated. The
shape of the attractor of the chaotic dynamics in state space was shown
to provide a wealth of information in regards to the type and magni-
tude of the parametric changes (damage). Separate and/or simultane-
ous changes in the stiffness of the upstream and downstream mounting
points was detected and estimated based on an analysis of the geometric
shape of the attractor of the chaotic dynamics.
The significant advantages of the proposed analysis are accompanied
by a few limitations. The most important limitation is the requirement
that a chaotic oscillation be present. However, this limitation may be
overcome when a chaotic excitation may be provided to the system.
54

References

Abarbanel, H. D. I. (1996). Analysis of observed chaotic data. Springer-Verlag.

Abarbanel, H. D. I., Brown, R., Sidorowich, J. J., and Tsimring, L. S. (1993). The
analysis of observed chaotic data in physical systems. Reviews of Modern Physics,
65( 4):1331- 1392.
Abdalla, M. 0. , Grigoriadis, K. M. , and Zimmerman , D. C .(1998). Enhanced struc-
tural damage detection using alternating projection methods. AIAA Journal, 36(7) :1305-
1311.
Agbabian, M. S., Masri, S. F., Miller, R. F., and Caughey, T. K. (1990). System identi-
fication approach to detection of structural changes. ASCE Journal of Engineering
Mechanics , 117(2):370- 390.
Amaravadi, V. K., Mitchell, K. , Rao, V. S., and Derriso, M. M. (2002). Structural
integrity monitoring of composite patch repairs using wavelet analysis and neural
networks. In Davies, L. P., editor, Proceedings of SPIE: Smart Structures and
Materials 2002: Smart Structures and Integrated Systems, volume 4701, pages 156-
166, San Diego, California.
Azeez, M. F. A. and Vakakis, A. F. (2000). Proper orthogonal decomposition of a
class of vibroimpact oscillations. Journal of Sound and Vibration, 240(5):859-889.
Banbrook, M., Ushaw, G., and McLaughlin, S. (1997) . How to extract Lyapunov ex-
ponents from short and noisy time series. IEEE Transactions on Signal Processing,
45(5):1378- 1382.
Bolotin, V. V., Grishko, A.A., Kounadis, A. N., and Gantes, C. J. (2001). The flutter-
ing panel as a continuous nonlinear nonconservative system. Journal of Vibration
and Control, 7(2):233- 247.
Bow, T . S. (2002). Pattern Recognition and Image Preprocessing. Marcel Dekker, New
York.
Broomhead, D. S. and King, G. P. (1986). Extracting qualitative dynamics from
experimental data. Physica D, 20(1):217-236.
Cao, T. T . and Zimmerman, D. C. (1999). Procedure to extract Ritz vectors from
dynamic testing data. Journal of Structural Engineering, 125(12):1393- 1400.
Chancellor, R. S., Alexander, R. M., and Noah, S. T. (1996). Detecting parameter
changes using experimental nonlinear dynamics and chaos. Journal of Vibration
and Acoustics, 118(3):375-383.
Chatterjee, A., Cusumano, J. P., and Chelidze, D. (2002). Optimal tracking of pa-
rameter drift in a chaotic system: Experiment and theory. Journal of Sound and
Vibration, 250(5):877- 901.
 55
Chelidze, D., Cusumano, J. P., and Chatterjee, A. (2002) . A dynamical systems ap-
proach to damage evolution tracking, part 1: Description and experimental appli-
cation. ASME Journal of Vibration and Acoustics, 124(2):250-257.
Copeland, G. S. and Moon, F. C. (1992). Chaotic flow-induced vibration of a flexible
tube with end mass. Journal of Fluids and Structures, 6(3):705- 718.
Cusumano, J. P., Chelidze, D., and Chatterjee, A. (2002). A dynamical systems ap-
proach to damage evolution tracking, part 2: Model-based validation and physical
interpretation. ASME Journal of Vibration and Acoustics, 124(2):258- 264.
Darbyshire, A. G. and Broomhead, D. S.(1996). Robust estimation of tangent maps
and Lyapunov spectra. Physica D, 89(1):287-307.
Doebling, S. W., Farrar, C. R., Prime, M. B., and Shevitz, D. W. (1996). Damage
identification and health monitoring of structural and mechanical systems from
changes in their vibration characteristics: A literature review. Report LA-13070-
MS, Los Alamos National Laboratories, Los Alamos, NM.
Dowell, E. H. (1966). Nonlinear oscillations of a fluttering plate. AIAA Journal,
4(7):1267- 1275.
Dowell, E. H. (1975). Aeroelasticity of Plates and Shells. Noordhoff International
Publishing, Leyden.
Dowell, E . H.(1989). A Modern Course in Aeroelasticity. Kluwer Academic Publish-
ers, Dordrecht, ii edition.
Dowell, E. H. and Ventres, C. S. (1970). Comparison of theory and experiment for
nonlinear flutter of loaded plates. AIAA Journal, 8(9):2022-2030.
Dowell, E. H. and Voss, H. M. (1965). Theoretical and experimental panel flutter
studies in the mach number range 1.0 to 5.0. AIAA Journal, 3(12) :2292-2304.
Epureanu, B. I. (2003). A parametric analysis ofreduced order models of viscous flows
in turbomachinery. Journal of Fluids and Structures. to appear.
Epureanu, B. I. and Dowell, E. H. (2001). Reduced order system identification of
nonlinear aero elastic systems. In Proceedings of th e First M.I. T. Conference on
Computational Fluid and Solid Mechanics, volume 1, pages 1152- 1160, Cambridge,
Massachusetts.
Epureanu, B. I., Dowell, E. H., and Hall, K. C. (2002). Mach number influence on
reduced order models of inviscid potential flows in turbomachinery. Journal of
Fluids Engineering, 124(4):977-987.
Epureanu, B. I., Hall, K. C. , and Dowell, E. H. (2000). Reduced order models of un-
steady transonic viscous flows in turbomachinery. Journal of Fluids and Structures,
14(8):1215-1235.
Epureanu, B. I., Hall, K. C., and Dowell, E. H. (2001). Reduced order models of
unsteady viscous flows in turbomachinery using viscous-inviscid coupling. Journal
of Fluids and Structures, 15(2):255- 276.
Epureanu, B. I., Tang, L. S. , and Paidoussis, M. P. (2003a). Coherent structures
and their influence on the dynamics of aero elastic panels. International Journal of
Non-Linear Mechanics. to appear.
Epureanu, B. I., Tang, L. S., and Paidoussis, M. P. (2003b). Observations of the
dynamics of panels in supersonic flow. In Proceedings of the Second M.I. T. Con-
ference on Computational Fluid and Solid Mechanics, volume 2, pages 1332-1336,
Cambridge, Massachusetts.
56

Farrar, C. R., Doebling, S. W., and Nix, D. A. (2001). Vibration-based structural

damage identification. Philosophical Transactions of the Royal Society of London:
A - Mathematical, Physical and Engineering Sciences, 359(1778) :131-149.
Feeny, B. F. (2002). On the proper orthogonal modes and normal modes of continuous
vibration systems. Journal of Vibration and Acoustics, 124(1):157-160.
Feeny, B. F., Yuan, C. M., and Cusumano, J. P. (2001). Parametric identification
of an experimental magneto-elastic oscillator. Journal of Sound and Vibration,
247(5):785-806.
Fraser, A. M. and Swinney, H. L. (1986). Independant coordinates for strange attrac-
tors from mutual information. Physical Review A, 33(1):1134-1140.
Grassberger, P. (1990). An optimized box-assisted algorithm for fractal dimensions.
Physics Letters A, 148(1):63-68.
Grassberger, P., Schreiber, T., and Schaffrath, C. (1991). Nonlinear time sequence
analysis. International Journal of Bifurcation and Chaos, 1(3):521-547.
Guckenheimer, J. and Holmes, P. (1983). Nonlinear Oscillations, Dynamical Systems,
and Bifurcations of Vector Fields. Springer-Verlag, New York.
Holmes, P., Lumley, J. L., and Berkooz, G. (1996). Turbulence, Coherent Structures,
Dynamical Systems and Symmetry. University Press, Cambridge, MA.
Kacprzynski, G. J., Gumina, M., Roemer, M. J., Cagulat, D. E., Galie, T. R., and
McGroarty, J. J. (2001a). A prognostic modeling approach for predicting recurring
maintenance for shipboard propulsion systems. In Proceedings of IGTI 2001 Turbo
Expo, volume 1, pages 1-7, New Orlelans, LA. AS ME.
Kacprzynski, G. J., Roemer, M. J., Hess, A. J., and Bladen, K R. (2001b). Extend-
ing fmeca - health management design optimization for aerospace applications. In
Proceedings of IEEE Aerospace Conference, volume 1, pages 1-8, Big Sky, MO.
IEEE.
Kantz, H. and Schreiber, T. (1997). Nonlinear time series analysis. Cambridge, New
York.
Kappagantu, R. V. and Feeny, B. F. (2000). Part I: Dynamical characterization of a
frictionally exited beam. Nonlinear Dynamics, 22(4):317-334.
Kim, D. H. and Lee, 1. (2000). Transonic and low-supersonic aeroelastic analysis of a
two-degree-of-freedom airfoil with a freeplay non-linearity. Journal of Sound and
Vibration, 234(5):859-880.
Langthjem, M. A. and Sugiyama, Y. (2000). Dynamic stability of columns subjected
to follower loads: A survey. Journal of Sound and Vibration, 238(5):809-851.
Lighthill, M. J. (1953). Oscillating airfoils at high Mach number. Journal of the Aero-
nautical Sciences, 20(6):402-406.
Ljung, L. (1999). System Identification - Theory for the User. Prentice Hall, New
York.
Loh, C. H. and Tou, 1. C. (1995). A system identification approach to the detec-
tion of changes in both linear and nonlinear structural parameters. Earthquake
Engineering f3 Structural Dynamics, 24(1):85-97.
Masri, S. F., Miller, R. K, Saud, A. F., and Caughey, T. K (1987). Identification of
nonlinear vibrating structures: Part i-formulation. Journal of Applied Mechanics,
109(54):918-922.
Miller, S. A., Turner, K L., and MacDonald, N. C. (1997). Microelectromechanical
scanning probe instruments for array architectures. Review of Scientific Instru-
ments, 68(11):4155-4162.
 57

Moon, F. C. (1992). Chaotic and Fractal Dynamics: an introduction for applied sci-
entists and engineers. John Wiley & Sons, New York.
Nerenberg, M. A. H. and Essex, C. (1990). Correlation dimension and systematic
geometric effects. Physical Review A, 42(1):7065-7074.
Niho, T., Horie, T., and Tanaka, Y. (2000). Numerical instability of magnetic damping
problem of elastic plate. IEEE Transactions on Magnetics, 36(4, pt. 1):1374-1377.
Paidoussis, M. P. (1986). Stability of a chain of cylinders travelling underwater. In
Proceedings of 5-th International Offshore Mechanics and Arctic Engineering Sym-
posium, volume 1, pages 483-490, New Orleans, LA. ASME.
Paidoussis, M. P. (1998). Fluid-Structure Interactions: Slender Structures and Axial
Flow. Academic Press, San Diego, CA.
Pavlidis, T. (1982). Algorithms for Graphics and Image Processing. Computer Science
Press, Rockville, MD.
Pecora, L. M. and Caroll, T. L. (1996). Discontinuous and nondifferentiable functions
and dimension increase induced by filtering chaotic data. Chaos, 6(3):432-439.
Reynolds, R. R. and Dowell, E. H. (1993). Nonlinear aeroelastic response of pan-
els. In Proceedings of 34-th AIAA/ ASME/ ASCE/ ASC Structures, Structural
Dynamics, and Materials Conference, volume 1, pages 1-13, La Jolla, CA. AIAA.
Reynolds, R. R., Virgin, L. N., and Dowell, E. H. (1993). High-dimensional chaos can
lead to weak turbulence. Nonlinear Dynamics, 4(1):531-546.
Roemer, M. J., Kacprzynski, G. J., Schoeller, M., Howe, R., and Friend, R. (2001).
Advanced test cell diagnostics for gas turbine engines (usaf automated jet engine
test strategy - ajets. In Proceedings of IGTI 2001 Turbo Expo, volume 1, pages
1-10, New Orleans, LA. AS ME.
Sadeghi, M. H. and Fassois, S. D. (1997). Geometric approach to nondestructive
identification of faults in stochastic structural systems. AIAA Journal, 35(4):700-
705.
Schreiber, T. (1995). Efficient neighbor searching in nonlinear time series analysis.
International Journal of Bifurcation and Chaos, 5(2):349-358.
Sirovich, L. (1987). Turbulence and the dynamics of coherent structures, part I: Co-
herent structures. Quarterly of Applied Mathematics, XLV(3):561-57l.
Smyth, A. W., Masri, S. F., Chassiakos, A. G., and Caughey, T. K. (1999). On-line
parametric identification of MDOF nonlinear hysteretic systems. ASCE Journal of
Engineering Mechanics, 125(2): 133-142.
Sohn, H. and Farrar, C. R. (2001). Damage diagnosis using time series analysis of
vibration signals. Smart Materials and Structures, 10(3):446-45l.
Thothandri, M. and Moon, F. C. (1999). An investigation of nonlinear models for a
cylinder row in a cross flow. Journal of Pressure Vessel Technology, 121(1):133-14l.
Trickey, S. T., Todd, M., Seaver, M., and Nichols, J. (2002). Geometric time do-
main methods of vibration based damage detection. In Proceedings of the SPIE
9-th Smart Structures and Materials Conference, volume 1, pages 1-9, San Diego,
California.
Turner, K. L., Miller, S. A., Hartwell, P. G., MacDonald, N. C., Strogatz, S. H.,
and Adams, S. G. (1998). Five parametric resonances in a micro electromechanical
system. Nature, 396(6707):149-152.
Vors, T. E. L., Heertjes, M. F., Campen, D. H., Kraker, A., and Fey, R. H. B. (1998).
Experimental and numerical analysis of the steady state behaviour of a beam
system with impact. Journal of Sound and Vibration, 212(2):321-336.
58

Wolf, A., Swift, J. B., Swinney, H. L., and Vastano, J. A. (1984). Determining Lya-
punov exponents from a time series. Physica D, 16(1):285~317.
Wong, Y. S., Voitcu, 0, and Popescu, C. (2001). A new approach for nonlinear dy-
namic predictions. Advances in Computational Mathematics, pages 1 ~25. preprint.
Wu, S. F. and Zhu, J. (1997). Stability analysis of a nonlinear plate in mean flow.
Journal of Computational Acoustics, 5(3):137~155.
Zang, C. and Imregun, M. (2001). Structural damage detection using artificial neural
networks and measured FRF data reduced via principal component projection.
Journal of Sound and Vibration, 242(5):813~827.
Zimmerman, D. C. (2000). Model validation and verification of large and complex
space structures. Inverse Problems in Engineering, 8(2):93~118.
Zimmerman, D. C., Yap, K., and Hasselman, T. (1999) . Evolutionary approach for
model refinement. Mechanical Systems and Signal Processing, 13(4):609~625.
 COMPARISON OF TWO SEAFLOOR OBSERVATORY
MOORING DESIGNS

SEAN HAN AND MARK GROSENBAUGH

Woods Hole Oceanographic Institution

Abstract: The purpose of this study is to compare two seafloor observatory mooring designs. The first
mooring uses a subsurface float with an "S-tether" positioned above it. The other design uses distributed
buoyancy to create an S-tether in the lower half of the mooring line. The maximum tensions at the anchor and
in the snubber hose, bending fatigue parameter called contact stress-slip parameter at critical locations, and
the dynamic motion in the cable is monitored. The trade-offs between the two designs, and the important
design parameters that affect the tensions, bending fatigue, and the dynamic motion are discussed.

1. INTRODUCTION

Over the next decade, much of oceanographic research will be conducted using long-
tenn measurements made by ocean observatories [1,2,3], which will consist of suites of
instruments mounted at the ocean surface, in the water column, and on the seafloor. A
major component of the observatory system will be moored surface buoys, which will
provide "real-time" two-way communication links via satellite. The moorings that are
proposed for the observatory systems are different from traditional moorings in that
they use electro-optical-mechanical (EOM) cables to deliver power from the buoy to
instruments and data from the instruments to the buoy (from where they are transmitted
ashore). The EOM cable is typically much stiffer than traditional mooring cable
materials, so that the EOM cable will experience higher loads if it is not designed
properly. Also, failures in the communication and power paths can occur due to a
mismatch between the stress-strain characteristics of synthetic fibers, used widely as
strength members, and the internal bundles of copper conductors and optical fibers [4].
Therefore, special care must be taken in designing these systems with EOM cables.

59
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 59-72.
2003 Kluwer Academic Publishers.
60

surface buoy
~

' " rubber hose ~ }

weighted cable {

cable with floats {

(j snubber cable with floats

football floats

~'" subsurface float

AUV docking station
} cable with floats

(a) Benthic design with (b) Benthic design without

a subsurface float a subsurface float

Figure 1. Schematics of two mooring systems.

o
-500

-5
fr -2000
"0

~ -2500 f \'\ . ;~~~~~~~~~~~~~.

~ survival
20mls

0.1 0.2 0.3 0.4 0.5 0.6 0.7

Current speed (mls)

Figure 2. Current profile.

The two single-point mooring designs that are being considered for the cable-linked
mooring system [1]. The first design that is considered in this paper is proposed for
systems that require an AUV docking station in the upper water column. It uses a single
subsurface float (SSF) with an "S-tether" positioned above it (Figure 1). The S-tether is
an EOM cable that is made positively buoyant (by attaching floats) over roughly half of
 61

the cable and made negatively buoyant (by using steel-armored cable or attaching
weights) over the other half of the cable. In light to moderate currents, the tether above
the SSF forms an S-shape. The second design has a more traditional inverse catenary
shape formed by attaching floats near the surface buoy and the anchor. In both designs,
an elastomeric hose (also called snubber) is attached in-line below the buoy to increase
elastic compliance. This is especially helpful in waves and currents where the S-tether
is pulled taut. In this paper, the horizontal offset of the surface buoy, the maximum
tensions at the anchor and in the snubber, the curvature at critical locations, and the
motion of the mooring are compared. The maximum tension is defmed by the sum of
the mean and the four times the standard deviation of the tension.
Responses of the moored buoy systems are obtained using WHOI Cable, a general-
purpose numerical code that was developed at Woods Hole Oceanographic Institution
for calculating the static and dynamic response of moored and towed oceanographic
systems [S]. The fmite difference time domain simulation is built around a
mathematical model of cable dynamics. The cable model is fully three-dimensional and
includes the effects of torsion, bending, and geometric and material nonlinearities.
We assume that the vertical displacement of the surface buoy follows the wave
elevation, which is described by the Modified Pierson-Moskowitz spectrum. This
assumption is valid because the proposed buoy shapes all have large water-plane areas.
The current velocity profile, wind velocities, and the significant wave heights that are
used in the analysis are shown in Figure 2. Values are given for slack water conditions,
design conditions (the conditions that are expected to be exceeded 10% of the time),
and survival conditions (2S-year return storm). These values correspond to conditions
that would be found along the Northern California coast.

2. COMPARISON OF THE TWO DESIGNS

Let us first consider the baseline designs in 1800m deep water. The baseline design
with a SSF consist of an anchor, Sm chain, 1490m Vectran cable. Sm chain, 1800lb
subsurface float, Sm chain, 340m Vectran cable with sixteen 61lb floats at every 20rn,
340 weighted Vectran cable, 13.3Skg coupling, 30m snubber, and the surface buoy. The
baseline design without a SSF consists of an anchor, Sm chain, 260m Vectran cable,
100m Vectran cable with ten 30.llb floats at every lOrn, 1100m Vectran cable, Sm
chain, 800m weighted cable with twenty one S8lb floats at every 40rn, 13.3Skg
coupling, and the surface buoy. he properties of the Vectran cable, snubber hose, and
the surface buoy can be found in [10]. The design with a subsurface float has a scope of
1.16, and the design without a subsurface float has a scope of 1.28.
Figures 3 and 4 show the horizontal offsets of the surface buoy when the wind
velocity is varied from 0 to 20 mis, and the current velocity at the surface is varied from
62

0.05 to 0.07 mls while keeping the same profile as shown in Figure 2. The horizontal
offsets are normalized by the maximum offset achieved when the mooring line
(assuming inextensible) forms a line from the anchor to the surface buoy. For instance,
the maximum offset for the scope s is xmax = depth~ .
For the range of current and wind velocities, the horizontal offset are less than 90%
of the maximum offset when the subsurface float is used. Note that a relatively large
offset can be achieved if the current velocity is large enough. If the current

0.91-----,------,-------,--------r------=::=;;~~..,

0.8

0.7
..........
1;l 1;l 0.6
!:Ei!:Ei
o 0
C; S 0.5
.,...S0.4
N
.1Qmis
.,... ~ /f;--5m1s
~~
__ <

0.3
0.2

0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7

Current velocity at the surface (mls)

Figure 3. Horizontal offet of the surface buoy when a subsurface float is used.
 63

0.9 ...................... ...... swtiVilL

. 20 mls
. 15,hls dehgn
~ ~ 0.8 ..... JOhi/s . .. ........ ... .
1tl0 1tl0 ! : :

. . . . . . .l. . . . . _ . l. . . . . . . l
1.~
0.7
5 m/s
.~ 0.6
::r: ~
. . . .
0
~ wind veloity = Oimls
0.5

0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Current velocity at the surface (mls)

Figure 4. Horizontal offet of the surface buoy when a subsurface float is not used.

velocity at the surface is greater than 0.5m/s, the horizontal offset is greater than 80%
of the maximum horizontal offset even for zero wind velocity. Note that the horizontal
offset under the survival conditions is about 25% greater than that under the design
condition.
When the subsurface float is not used, the horizontal offset of the surface buoy can
reach near the maximum offset if the wind velocity is sufficiently high enough even if
the current velocity is small. IE. addition, the large offset can also be achieved if the
current velocity is large enough even if the wind velocity is small. The horizontal
64

3000 ~===;;:::=~=(a=)===j-----r--l
: survival
----~~-~~~~~~~~
~ 2000
. design
~ 1000 . .......... .. . (
. ............... .. .

-
~

500 1000 1500 2000 2500

(b)
500
..................... It, surviv~l ................ ..
.--. 400
~
300
b"
200
100 0
500 1000 1500 2000 2500

Lagrangian coordinate (m)

Figure 5. Mean and the standard deviation tension along the cable in the design with a SSP.
(a)
3000.-----~
. ----~----~----~----,

.--. 2000 ~ ''\ .. . .... .

~ survival
'--'
~ 1000 ... . . .. ... ,
__ ~ ~e~i~_______________

0
0 500 1000 1500 2000 2500
(b)
700

.--.
~ 500 "'survival ..
'--'
h
b
300 ___ ~l2:s~~ ---------~---- -- .... ---
100
0 500 1000 1500 2000 2500
Lagrangian coordinate (m)

Figure 6. Mean and the standard deviation tension along the cable in the design without a SSP.

offset under the survival condition is less than 4% greater than that under the design
condition. The horizontal offset reaches its maximum quickly with either high winds or
high currents.
 65

Figures 5 and 6 show the mean and the standard deviation of the tension along the
cable (starting from anchor to the surface buoy). Note that both flT andO"T under the
design condition are significantly less than those under the survival condition, and even
more so in the design without a subsurface float than the design with a SSF. For
example, the maximum snubber tension in the design without the subsurface float is
reduced by a factor of 3.4, and the maximum anchor tension by a factor of 2.5, whereas
the numbers are 2.0 and 1.6, respectively, for the mooring with a subsurface float. If the
maximum tensions are comparable in the two designs under the survival condition, we
can expect that the design without a SSF is subjected to smaller loads than the design
with a SSF under less severe conditions.
The horizontal offsets and the maximum tensions at the anchor and the snubber hose
under the survival condition are given in Table 1.

Table 1. Comparison ofthe Two Mooring Designs.

Tmax(snubber) (lb) Tmax(anchor) (lb) Xsurface buov (m)

With, scope=1.21 3321 4525 1192
With, scope 1.52 3113 3978 1746
W/O, scope=I.28 4030 4510 1430
W/O, scope=1.53 3370 3900 2080
W/O, scope=1.78 3040 3530 2640

1.5

1.4 ......... L.. 0 Tmax (snubber) .

: 0 Tmax (anchor)
~ 1.3
.9
rJ)
~
.....Q) 1.2
"0
Q)
.::1 1.1
(;J .....I.. s.~ope ~ i lii ~~~;~ = 1.79
: scope ~ 1.34 :scop'e =~ 1.52 : \
Z
0
"'::'~";~"~'~"~"~~.~.~"~'~'~'~"~":. .. '.'
0.9 .
... . .. . . .. ....... . .... :. . ...... .
~
, , .
_.. - _... -:. . . . . - - - - - --!
.

0.8 L-_---'---_-----'-_ _-'-----_-"--_-----1_ _--'--_----'

400 600 800 1000 1200 1400 1600 1800
S-tether length (m)

Figure 7. Effect of varying S-tether length.

66

From [10], it is found that the length of the S-tether affects the maximum tensions the
most in the design with a SSF, and the length of the mooring in the design without a
SSF. Figures 7 and 8 show the variation in the maximum tensions with the length of S-
tether and the length of the mooring, respectively. The maximum tensions are
normalized by those of the baseline designs. The length of the S-tether must not exceed
three times the SSF depth or 900m in this case so that the S-tether does not interfere
with the surface buoy or the subsurface float. Figure 7 and Table 1 show that the
maximum tensions in the design with a subsurface float are roughly at their minima
when S-tether is 1200m long (scope = 1.52). On the other hand, the maximum tensions
in the design without a subsurface float decrease monotonically

2.----,----.----,---,,---,,---,----.
o Tmax (snubber)
<> Tmax (anchor)

-- _. -..-- ~ ............ ~ .. ......... ; .......... .

0.5 '---------'---------"'--------'-------'------'-------'-------'
1800 2000 2200 2400 2600 2800 3000 3200
Total length (m)

Figure 8. Effect of varying mooring length.

 67

(a)
6r-----~----~------~----~----~

~4
'-'2
~
Of------;--------:------I-----.-;-
-2 0-'-----,:-50...,0,----1:-:0'-=-00-,----1:-:5'-=-00-,----:2-"0-,-00,----:2-"500

0.8
(b)
.-----r------.--'-''---.-------,------,

~0.6
~
,-,0.4
...,
b 0.2

o'-:------~--~----"~~~~~----~
500 1000 1500 2000 2500
Lagrangian coordinate (m)

Figure 9. Contact stress-slip parameter in the design with a SSF.

at least up to the scope of 1.78, and the maximum tensions at this scope are smaller than
the minima achieved by the mooring with a subsurface float.
Figures 9 and 10 show the mean and the standard deviation of Raoofs contact stress
slip parameter. The contact stress slip parameters are calculated based on the five-layer
stand used in [10] and using the methods outlined in [6-9]. The 5m moment releasing
chain between the cable and the SSF is omitted in the design with a SSF (Figure 9).
Raoofs contact stress slip parameter measures the degree of bending fatigue damage of
a cable in tension. In the design with a SSF, the portion of the cable right above the
subsurface float is most susceptible to bending fatigue . In
68

(a)
, ,
0.1.------.-----,.-----,------,------.

~
.. '1
O~~~----~--~~~~~i-"
0.082
.
...
'-'
=L
-0.1 '-------'--------"'--------'--------'-------'
o 500 1000 1500 2000 2500
(b)
0.2 .------.-----,------,------,------.

~
$ 0.1
t3'
OL-~~~----~---- __~----~--~~
o 500 1000 1500 2000 2500
Lagrangian coordinate (m)

Figure 10. Contact stress-slip parameter in the design without a SSF.

(a)
0.15 .------.-----,------,------,-------,

~ 0.1
.. .. ..
................................ .................. .
'-:::, 0.05 ~

::i.
O~----~----~'------~

-0.05 !------:l:-:,------:-:''::-:------,:-:'-:-:-----=-='-::-:---____=_:'
o 2~

0.06 r------.-----,.-~'--,------.----__.

~ 0.04
'-'
b'" 0.02

Lagrangian coordinate (m)

Figure 11. Contact stress-slip parameter in the design with a SSF and with a moment releasing
chain.
 69

the design without a SSF, the lower buoyant section is most susceptible to bending
fatigue. It is also shown that the design with a SSF is more susceptible to bending
fatigue than the design without a SSF. When the 5m-chain is included between the SSF
and the cable (Figure 11), both the mean and the standard deviation of the contact stress
slip parameter decrease considerably. It is interesting to note that the moment is
released properly in the fIrst design, the mean and the standard deviation of the contact
stress slip parameter at the critical locations in the two designs are
E 4.--.---.---,---.---.--.---.---,--,
a~
aJ
:::
: : :

3.5 ..... ~ ...... ' .....

: : : :
::::0
"!" ........... + .. " / ...... .

E
~
3"~~ . . .
,,1
:: I
r.r<JJ 2.5 : : :,...
.................. ::
'~r
I

~
......
2 .......' ... .... .......... l. ........ ;..: ................
:
::.......... ;Or:.........l....... .
'

$3 1.5 ~ ~ ~
. .
........ ......... !......... ......... ........ J":'O ... ......... ; ....... !...... .
~ ~ ~;
~
~

u
-g
~ 05 .........,
. 0" ...,
_ e ~ o()~ ........ , ........ , ...... . :. ................ , ..

z 0'----'----'----'-----'----'----'-----'----"-----'
300 400 500 600 700 800 900 1000 11 00 1200
Length of the buoyant section (m)

Figure 12. Effect of varying buoyant section length on the contact stress-slip parameter.
.... 6
~

@ 5
0..
.S-
'1i
<JJ
<JJ
(1)
b<JJ
3
t)
E20
u
"0
(1)
.S::
0;;

0
E
Z 00

Figure 13. Effect of varying lower buoyant section length on the contact stress-slip parameter.
70

comparable. Therefore, if the moment is released properly in the first design, the
bending fatigue damage can be as low as in the second design.
Figure 12 shows the variation of the contact stress-slip parameter in the design with a
subsurface float when the length of the buoyant portion of the S-tether is varied while
the total buoyancy is kept constant. It is shown that it becomes worse as the buoyant
section lengthens, and no more than 60% of the S-tether should be made buoyant.
Figure 13 shows the variation of the contact stress-slip parameter in the design
without a subsurface float when the length of the lower buoyant portion of the S-tether
is varied while the total buoyancy is kept constant. Unlike in the previous case

~ 1.5 ,---___,_---_r_-----,---_.___--____,
5
.f' I H . ... ...... ...... ... ....... H... .
OJ
o
~ 0.5
r:/)

~ o~~~--~----~--~--~
o 500 1000 1500 2000 2500
Lagrangian coordinate (m)

Figure 14. RMS velocities along the cable in the design with a SSF.

~1.5,------,-----r------,----.---------,
5 survival
.f' 1 ................... ,-.................., . ....__ ._- _.......... .. ..............--~ .. - ......
OJ
o
~ 0.5 . .
r:/) ,I . _~'----~-i---J
~ 0 =--_-=-~-=--=--=-~-_-_-~-'---_-_i\_d_es~~_ ___"_ ______'
o 500 1000 1500 2000 2500
Lagrangian coordinate (m)

Figure 15. RMS velocities along the cable in the design without a S SF.

with a subsurface float, it is better to spread the floats further away from each other
over a longer distance in terms of the bending fatigue.
Figures 14 and 15 show the root mean square velocities. In both cases, the dynamic
velocities are significantly reduced over the length of the snubber. The dynamic
motions of the cable are comparable in both designs, and it is usually smaller under the
design condition than under the survival condition.
 71

3. SUMMARY AND CONCLUSION

It is found that the design with a subsurface float is preferred if a smaller watch circle
and smaller loads on the snubber hose are required. However, the maximum tensions in
the design without a subsurface float can be lowered significantly more if the scope is
increased sufficiently. The load in the lower sections in the latter design is also much
lower than those in the first design.
It was discussed that the length of the mooring affects the maximum tensions at the
anchor and in the snubber the most in both designs. The maximum tensions can be
generally lowered by increasing the length of the S-tether or the length of the mooring.
The bending fatigue damage is affected the most by how the buoyancy floats are
distributed along the cable. Finally, the dynamic motion is affected the most by the
length of the snubber hose.

4. ACKNOWLEDGEMENTS

We gratefully acknowledge the support of the Office of Naval Research (Ocean

Engineering and Marine Systems) under grant N00014-92-J-1269 and the Deep Ocean
Exploration Institute of Woods Hole Oceanographic Institution. Support for the first
author was provided by a Woods Hole Oceanographic Institution postdoctoral
fellowship.

5. REFERENCES

I. R. Detrick, D. Frye, J. Collins, J. Gobat, M. Grosenbaugh, R. Petit, A, Plueddeman, K. Von der Heydt,
B. Wooding, J. Orcutt, 1. Berger R. Harriss, F. Vernon, J. Halkyard, and E. Horton, DEOS moored buoy
observatory design study, National Science Foundation/Ocean Division (2000).
2. National Research Council, Illuminating the Hidden Planet: The Future of Seafloor Observatory
Science, National Academy of Science, 2000.
3. M. Chaffey and E. Mellinger and W. Paul, Communications and power to the seafloor: MBARI's ocean
observing system mooring concept, Proceedings to Oceans (2001) 2473-2481.
4. W. Paul and D. Bently, Conductor survival in lightweight upper ocean working cables, Proceedings to
ISOPE (2001) 708-714.
5. J. Gobal and M. Grosenbaugh, Technical Report WHOI-2000-08, Woods Hole Oceanographic
institution (2000).
6. M. Raoof and R. Hobbs, Analysis of multilayered structural strands, Journal of Engineering Mechanics,
ASCE 114(7) (1988) 1166-1182.
72

7. M. Raoof, Axial fatigue of multilayered strands, Journal of Engineering Mechanics 116(10) (1990)
2083-2099.
8. M. Raoof, Free bending fatigue of multilayered strands, Journal of Engineering Mechanics, ASCE
118(9) (1992) 1747-1764.
9. M. Raoof, Design of steel cables against free-bending fatigue at terminations, The structural engineer
71(10) (1993) 171-178.
10. S. Han and M. Grosenbaugh, Bending analysis of seafloor observatory moorings, Proceedings to
OMAE (2003).
 NONLINEAR DYNAMICS OF VERY HIGH
DIMENSIONAL FLUID-STRUCTURAL SYSTEMS*

EARL H. DOWELL AND D.M. TANG

Duke University

Abstract: The dynamics of mechanical systems of high dimension or many degrees of freedom will be
discussed. The modeling of both fluid and structural systems will be considered with special emphasis given
to fluid-structural or aeroelastic systems. Novel and effective theoretical and computational techniques will
be considered and illustrative correlations with experimental results given. Also briefly considered will be
nanoscale systems that are modeled at the molecular level using similar mathematical techniques. Aeroelastic
systems are those that involved the coupled interaction between a convecting fluid flow and a flexible elastic
structure. The nonlinear dynamical response of such systems is of great current interest. Currently
operational aircraft are known to encounter limit cycle oscillations (LCO) in certain flight regimes and
relatively simple experimental wind tunnel models have been designed to exhibit LCO as well. The LCO
may be either beneficial or dangerous for the safety of the aircraft. The results of several wind tunnel
experiments are discussed and compared to those from mathematical models. The physical models include
(1) an airfoil with control surface freeplay; (2) a delta wing with structural geometrical nonlinearities due to
plate-like deformations; and (3) a very high aspect ratio wing with geometrical structural nonlinearities due to
coupling among torsional twist, transverse bending and fore-and-aft bending. In addition, the theoretical and
computational advantages of modeling the aerodynamic flow field in terms of a set of global modes and also
using a novel form of the harmonic balance method are emphasized. A recent theoretical result for large
shock motions in a viscous transonic flow around an oscillating airfoil undergoing LCO is presented to
illustrate the results to be obtained by such methods.

1. INTRODUCTION

Nonlinear dynamics phenomena in fluid-structural systems have been known for many
years. In the last decade or so, such effects have become of more serious concern to
practitioners. And for that reason, and also because of the advance in theoretical and
experimental methods, a more substantial and concentrated effort has been made by the

This paper is a revision and an extension of an earlier publication by the authors entitled, "Nonlinear
Aeroelasticity and Unsteady Aerodynamics," AIAA Journal, Vol. 40, No.9, pp. 1697-1707, September,
2002.

73
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 73-113.
2003 Kluwer Academic Publishers.
74

research community to understand and pursue how unfavorable nonlinear fluid-

structural of aero elastic effects may be diminished and favorable effects exploited.

Moreover, in the last few years significant advances have been made in constructing
reduced order models for unsteady aerodynamic flows. This work is also discussed
here as it relates to nonlinear aeroelasticity.

There are two possible principal consequences of any nonlinear effect. One is that
the exponentially growing oscillations predicted by an unstable linear model are
attenuated due to the nonlinear effects and a fmite amplitude, steady state oscillation is
obtained. Such limit cycle oscillations (or LeO) have been observed in operational
aerospace craft and wind tunnel models. Leo per se is thus benign in that the
nonlinearity reduces the amplitude of the oscillations. Of course, structural integrity
may still be at issue if the LeO amplitudes are too large. The second principal
consequence is wholly detrimental. In this instance, a system, which may be stable to a
sufficiently small perturbation, can become unstable due to a large disturbance.

The generic possibilities are illustrated in Figure I where the LeO amplitude is
plotted vs. some system parameter, e.g. flight speed. In Figure I a, an aeroelastic
system is depicted which is stable to small or large perturbations below the flutter
(instability) boundary predicted by a linear dynamical model. Beyond the flutter
boundary, Leo arise due to some nonlinear effect and typically the amplitude of LeO
increases as the flight speed increases beyond the flutter speed. In Figure lb, the other
generic possibility is shown. While again Leo exist beyond the flutter boundary now
Leo may also exist below the flutter boundary, if the disturbances to the systems are
sufficiently large. Moreover both stable (solid line) and unstable (dotted line) Leo
now are present. Stable Leo exist when for any small disturbance, the motion returns
to the same Leo at large time. Unstable Leo are those for which any small
perturbation will cause the motion to move to a stable Leo. Theoretically, in the
absence of any disturbance both stable and unstable Leo are possible dynamic, steady
state motion of the system. Information about the size of the disturbance required to
move from one stable LeO to another can also be obtained from data such as shown in
Figure lb. Note also the hysteretic response as flight speed increases and then
decreases.
 75

Benign Nonlinearity Detrimental Nonlinearity

WEAKER NONLINEARITY,
LARGER LCO AMPLITUDE

w
C
::J
I-
::::i
a.
STRONGER NONLINEARITY, :::IE
SMALLER LCO AMPLITUDE <
oo
..J

FLUTTER SPEED
FLUTTER SPEED

FLIGHT SPEED FLIGHT SPEED

(a) Benign nonlinearity (b) Detrimental nonlinearity

Figure 1: (a) Schematic ofLCO response for a benign nonlinearity leading to (only) stable LCO.
(b) Schematic ofLCO response for a detrimental nonlinearity leading to both stable (-) and
unstable (--) LCo. Arrows denote path of system response when flight speed is increasing (-+) or
decreasing (+-).
There are several physical sources of nonlinearity in either the aerodynamic flow or
elastic structure. These are listed and briefly described in Figure 2 and will be further
considered throughout the paper.

This paper summarizes several distinct yet related research thrusts that have proven
particularly fruitful. While the authors draw largely on the experience of the Duke
aeroelasticity team, reference is made to the work of many other investigators for the
benefit of those who may wish to pursue the ever increasing literature on these topics.
However, we do not provide an in depth summary of the total literature, but rather
comment selectively on the efforts of the broader community in the recent and not so
recent past to provide a context for the work discussed here. An electronic
bibliography of over eight hundred items is available from the authors upon request.
76

2. SCOPE OF PAPER

Four specific aeroelastic models are considered in this paper. The first is an airfoil with
a control surface which has freeplay in its attachment to the airfoil. Such a
configuration exhibits LCO due to freeplay well below the classical linear flutter speed
(LFS). The second is a wing with a plate-like structure that undergoes LCO once the
LFS is exceeded, but not usually below the LFS. Here, the nonlinearity is a result of the
tension induced in the plane of the plate-wing when the wing bending is on the order
the wing thickness or greater. The third is a very high aspect ratio wing which exhibits
LCO above the LFS, but also exhibits a sensitivity to initial disturbances below the LFS
that may lead to LCO there as well. Here, the nonlinearity is due to the coupling among
flapwise bending, chordwise (or lag) bending and torsion of the wing structure. This
coupling has long been known to be important for rotorcraft blades that are
cantilevered at the rotor hub (sometimes called hingeless blades). More recently very
high aspect ratio wings have been used for high altitude long endurance (HALE) UAV
(Uninhabited Air Vehicles). Fourthly and fmally, the aero elastic responses due to
nonlinear aerodynamic forces arising from large shock motions in the transonic flow
range are considered.

STRUCTURE:

FREE-PLAY - NONLINEAR STIFFNESS ARISING FROM LOOSELY

CONNECTED STRUCTURAL COMPONENTS
STRAIN-DISPLACEMENT OR GEOMETRIC NONLINEARITY -
NONLINEAR STIFFNESS ARISING FROM LARGE DISPLACEMENT
GRADIENTS
DRY-FRICTION - NONLINEAR DAMPING ARISING FROM
STRUCTURAL COMPONENTS IN SLIDING CONTACT

FLUID:

SHOCK MOTIONS IN TRANSONIC FLOW - PARTICULARLY

IMPORTANT FOR LOW REDUCED FREQUENCIES
SEPARATED FLOW - MOST COMMON IN TRANSONIC FLOW (MAY
BE INDUCED BY SHOCK) AND/OR AT LARGE ANGLE OF ATTACK

Figure 2: Principal nonlinearities that have been identified for aeroelastic systems.
 77

For each of the fIrst three models, experiments (conducted in a low speed wind
tunnel) are also discussed and correlated with theory. For the fourth model, the NASA
Langley Research Center aeroelasticity team has provided valuable benchmark
experiments that are directed toward aerodynamic nonlinearities in the transonic flow
regime. And some encouraging correlations with available theory have been made by
the Langley team as well. It can be expected that our ability to pursue more such
correlations will continue to advance due to the work of the Langley team and the
efforts of the aeroelastic community at large. Also noteworthy, the aeroelasticity group
at DLR, Gottingen has conducted valuable two-dimensional oscillating airfoil flutter
and LCO experimental/theoretical studies.

Cunningham [1] provides an insightful discussion of certain nonlinear

aerodynamics issues based upon his extensive experience with operational aircraft and
wind tunnel models. Friedmann [2] has touched on particular nonlinear aero elasticity
issues in his broad-ranging survey of aeroelasticity as well. Both are recommended to
the reader who wishes further background and a broader context for the present
discussion.

3. STRUCTURAL NONLINEARITIES IN FLUID-STRUCTURAL SYSTEMS

3.1. Airfoil Plus A Control Surface with Freeplay

Many investigators over the last fIfty years have considered the effects of a structural
stiffness nonlinearity on an airfoil with or without a control surface. See [3-31].
Broadly speaking, the literature is characterized by the type of nonlinearity, i.e.
continuous or discontinuous (freeplay), whether the nonlinear spring stiffness is for the
airfoil per se or the control surface, and fmally the nature of the aerodynamic model.
Most of the analysis has been done at low Mach number using classical Theodorsen
aerodynamic theory or approximations thereto or at very high Mach number where
piston theory aerodynamics can be applied. Also, most of the studies have been
theoretical/numerical, but some interesting experimental work has been done as well.
For a more thorough review of the literature, particularly as regards freeplay
nonlinearities, see Conner et al [24]. For continuous spring nonlinearities, the recent
publication by Lin, Wong and Lee [30] has a nice summary.
78

Here, we focus on theoretical/experimental correlation as achieved at low Mach

number, the theoretical effects of transonic Mach number, and the physical and
fimdamental insights that have been gained over the years. For another recent study of
freeplay for an airfoil in the transonic, low supersonic range, see Kim and Lee (2000).
Their results appear comparable to those discussed here, although they use different
solution methods and do not consider a control surface per se.

Figure 3: Sketch of the aeroelastic typical section with control surface.

The configuration considered is the simplest that models the fimdamental physical
phenomena. See Figure 3, which depicts an airfoil with a control surface in two-
dimensional flow. The airfoil structure has three degrees of freedom, plunge (vertical
translation) and pitch (rotation about the spring or elastic axis) and flap rotation
(rotation of the control surface relative to the airfoil per se). The nonlinearity modeled
is the bi-linear torsional stiffness of the attachment of the flap to the airfoil. See Figure
4. For small flap rotation, the bi-linear torsional stiffness is very small (here set to zero
and hence the term, free-play) while for large flap rotation the torsional spring stiffness
approaches a nominal fixed value. Using Lagrange's equations for the structure
including the bi-linear free-play nonlinearity and an appropriate aerodynamic model
one may determine the flutter boundary and the LCO. For the details of the theoretical
model and the wind tunnel experimental model, see Conner et al [24,26,31].
Theodorsen aerodynamics or the corresponding theory of Von Karman and Sears [32]
is used for the low Mach number analyses and results.
 79

Here the focus is on the basic physical phenomena and the correlation of theory and
experiment. In Figure 5 a LeO amplitude (in pitch) is plotted vs. flow speed (velocity)
where the latter is normalized by the flutter speed in the absence of free-play. Beyond
the flutter speed without free-play (here normalized to unity), the LeO amplitude
becomes very large (infmite in the present theoretical model) and no experimental
testing was done for this range of flow speed due to safety concerns. Note that the
effect of free-play is detrimental in that LeO is induced by free-play at flow speeds
substantially below the nominal flutter speed without free-play. No LeO exists above
the flutter boundary predicted by linear theory for this configuration, but rather
exponentially growing oscillations (diverging flutter) are predicted by the theoretical
model.

(,I)

Flap Displacement. 1'1

Figure 4: Elastic restoring moment or torque with asymmetric Jreeplay region about fJ = O.

Now consider the LeO amplitude below the nominal linear flutter boundary.
Shown in Figure 5 is the LeO amplitude in pitch, a, response is normalized by the free-
play angle o. See Figure 3 for the definition of o. The theoretical model predicts that
by using this normalization the normalized LeO amplitude results for ill!Y 0 will lie on
a single, universal curve, see Figure 5. The results from numerical simulation confirm
this and the results from experiment for several 0 over a range from 0.5 0 to 2.20 are also
in close correspondence with each other and the theoretical prediction.
80

0.14
0
V
-l 0. 12
'-
'"cy III
0
S;-
2: 0.()8
..c
-8
'Ei-
0.06
5'"
'c; 0.1.)4
8
<
0.02
z
(l
0.1 0.2 0.3 OA 0.5 0.6 0.7 0.8 0.9
Nondimcnslonal ve!o('lt)

Figure 5: Numerical and experimental normalized steady state RMS amplitude for pitch vs.
normalized flow velocity.

Note that there are both stable and unstable Leo predicted by theory. The unstable
LeO is indicated by the dashed line in Figure 5. This has a simple and important
physical interpretation as follows. In addition to the nominal flutter speed without
freeplay, there is another and different flutter speed predicted by the linear theory when
there is only free-play (formally when 8 ~ 00). For the combination of flow and
structural parameters considered in the present example this normalized flow speed is
about 0.37. See Figure 5. Note however that both a stable and unstable LeO exist at yet
lower flow speeds. The lowest flow speed for an LeO to exist (where the stable and
unstable LeO meet) corresponds to the minimum flow speed for flutter to occur for any
control surface torsional stiffness between zero (oscillations entirely in the free-play
range) and the nominal stiffness in the absence of freeplay. In the present example, this
normalized flow velocity is about 0.18. See Tang, Dowell and Virgin [26] for a more
detailed discussion.
 81

1,0

D 0,9
','
::j'"
:::1.
0,8
~'
"
;>' 0,7

()
:>; C")
'6
0 0,6
c

(jj
>
ill
'5
u::
0,5

6
/~

0.4
0,0 0,2 0.4 0,6 1.0

Mach Number, M

Figure 6: Flutter velocity vs, Mach number, M.

These physical phenomena persist into the transonic flow range although with some
added features. Turning to transonic flow, consider first the flutter boundary in the
absence of free-play. These results were obtained using a reduced order aerodynamic
model based upon proper orthogonal decomposition for the time-linearized Euler
equations of fluid mechanics, See Dowell, Thomas and Hall [31]. The flutter boundary
is shown in Figure 6 in terms of a non-dimensional flutter speed (flutter index) vs.
Mach number. At low Mach number the results are similar to those shown before for
M=O. However for high subsonic/transonic Mach number, there are important
differences. Note the flutter speed tends to a minimum in the transonic Mach number
regime; however also note that the flutter speed rises sharply with Mach number after
the flutter minimum occurs. This is often called the "transonic bucket".
3,0

2.5

6
'~ 2,0
a:
""
u
c::
'"
::>
e
0-

lL
1,5

1.0
....
() c
~
::>
0,5
u::
0.2 0.4 0,6 0,8 1,0

Mach Number, M

Figure 7: Flutter frequency vs. Mach Number, M.

82

The corresponding flutter frequency is shown in Figure 7. This illustrates another

important characteristic of transonic flow, i.e. the change in flutter mode as the
transonic Mach number range is traversed. For low Mach numbers, it is the plunge
mode that dominates the flutter motion, although all modes participate to some degree.
Above M=O.8, the pitch mode becomes dominant and for yet higher M, the flap mode
dominates. To detect these rapid changes in flutter mode, many flutter calculations must
be performed for a range of Mach number and other parameters. Reduced order
aerodynamic modeling is a key enabling methodology for these calculations.

Once the flutter boundary and flutter mode are determined, the harmonic balance
method is a very useful and computationally effective method to compute LCO. The
present configuration provides a good example of how the LCO may be determined
using a harmonic balance approach.

1.0 1.4
~~

..0" ~,.
S 8 1.2
(~ 0.8
~.
0
::!;' ~ 1.0
II 0.6 II:
>~ >-
u 0.8
c
;i- 0.4
Gl
:>
Ti a- 0.6
0 Gl
-a; u::
> 0.2
~::J
.8 0.4
S
u:: 0.0 u::
0.0 0.5 "1 .0 1.5 2.0 2.5 0.5 1.0 1.5 2 .0 2.5
Flap Frequency Ratio, mit'>"~ Flap Frequency Ratio, ('V(l)"

(al Flut.t('r vu lodty

Figure 8: Flutter velocity and frequency vs. flap frequency ratio. M = 0.8.

Consider a fixed M and all other parameters specified except the flap torsional
stiffness (or equivalently, natural frequency). Treating the flap stiffness or flap natural
frequency as a free-parameter, one may determine the flutter speed as a function of,
say, flap natural frequency using a linear aeroe1astic model. See Figure 8. But in reality,
because of free-play, the flap natural frequency is a function of flap rotation amplitude.
See Figure 9. For an assumed harmonic motion of the flap, the equivalent stiffness or
natural frequency of the flap may be determined as a function of flap amplitude from a
 83

(nonlinear) Fourier analysis of the bi-linear torque-flap rotation characteristic, i.e. by a

harmonic balance method. Recall Figure 4. And the result is shown in Figure 9. Cross-
plotting Figure 8 and 9, one may determine the LCO amplitude vs. flow speed or
velocity relationship. In a sense a nonlinear eigenvalue problem has been solved, i.e.
the amplitude (at a given flow speed) for which neutrally stable oscillations (LCO) may
exist has been determined. By considering perturbations in amplitude or flow speed
about these LCO one may determine whether the LCO is stable or unstable. The result
for LCO amplitude vs. flow speed is shown in Figure 10. Such results can be and have
been obtained over a broad range of Mach number. See Dowell, Thomas and Hall [31].

3.2. Low Aspect Ratio, Plate-Like Wing

From one perspective, the literature for this configuration in the context of nonlinear
aero elasticity is very recent and relatively small. See [33-40]. However, if one takes a
broader view and notes that the basic physical mechanism for the nonlinear effect is
that a tension force is induced in the mid-plane of the plate-wing by the out-of-plane
bending when the latter deflection is of the order of the plate thickness, then we recall
that this is indeed the same physical mechanism that leads to nonlinear effects in
general and LCO in particular for "panel flutter" or the flutter of plates and shells. A
panel is a local portion of a wing between pairs of spars and stringers. See the
monograph by Dowell [33] which gives an early account of the fundamentals of the
phenomena and the recent review by Mei [34] that summarizes the recent literature on
panel flutter. It is notable that in much of the recent (and not so recent) panel flutter
research the simple piston theory aerodynamic model is used that is valid only at
sufficiently high supersonic Mach number. However it is the subsonic, transonic and
low supersonic flow regimes that are often most important for applications. Fortunately,
recent theoretical advances make calculations in these Mach number regimes more
feasible and attractive. When such calculations were first done twenty five years ago
they were a feat. See Dowell [33]. Today, although they do require an understanding of
the more sophisticated aerodynamic models, the calculations themselves are no longer
extraordinary in their demand on computer resources [35-37]. The most recent work
for a plate-like wing per se is discussed [38-45].
84

>-.
CI
c:
0 0.9
;::;
0"' 0.8
0
:... 0.7
~

0.. 0.6
;;'j
r;:::: 0.5
..... 0.4
=
0 0.3
'"3 0. 2
>
.; 0.1
0"'
r.r.:: 2 4 6 8 ()I 12 14 16 18 20
Flap rotation, P/8

Figure 9: Non-dimensional equivalent flap frequency,

OJpl OJp nominal, vs. normalized flap rotation, /310.

For a low aspect ratio wing structure which has significant bending in both the
chordwise and spanwise direction, it may within a certain approximation be modeled as
a plate (as distinct from a bending beamltorsional rod model that is often used for high
aspect ratio wing structures). It was shown by Von Karman [46] that if the bending
deflections of a plate are comparable to the plate thickness, then there is a tension
induced into the plate by mid-plane stretching that varies as the square of the plate
deflection (more precisely as the square of the local slopes). This tension, when
appropriately multiplied by the plate curvature to give the relevant transverse force,
gives rise to a cubic stiffuess nonlinearity in the bending plate deflection.

Von Karman and many subsequent investigators were concerned with plate post-
buckling. In the early days of aeronautics and continuing to the present, plate skins on
aircraft between spars and stringers are allowed to buckle. Yet because of the stiffuess
nonlinearity, they retain some useful stiffuess even when buckled. Many years after
Von Karman's original studies, his theoretical model was a key to our understanding of
the flutter of aerospacecraft skin panels. This is because the static pressure and thermal
stress loading of thin skin aircraft panels can often deform them into the nonlinear
regime. And thus to predict the flutter boundary of such panels, not to mention LeO,
requires the nonlinear plate theory of Von Karman. And indeed this theory may be
used to determine the LeO as well. As expected the LeO is almost invariably benign
 85

and the LeO amplitude is typically of the order of the plate thickness. For a further
discussion of these matters, see Dowell [33] and Mei [34].

Now, at least in retrospect, none of the above is surprising perhaps, although for
several years there was distressing disagreement between theory and experiment for
panel flutter until the Von Karman nonlinear model was adopted. After all, when an
elastic plate or beam is fixed at its edges, it will stretch when it bends and a nonlinear
tension force will be induced. However what may be less expected, but nevertheless
also true, is that when a plate is only restrained on a single edge, e.g. alow aspect ratio
wing cantilevered to a wind tunnel wall or an aerospacecraft fuselage, a significant
tension may still be induced as predicted by the Von Karman theory. Typically the
deflections must be larger for a cantilevered plate than for a plate fixed on opposite
edges for the nonlinear effect to be important, however. The LeO amplitude of a
cantilevered plate may be as large as several plate thicknesses, while that for a plate
clamped on all edges is usually less than one plate thickness.

..........-
4.0 1.4 .................... , ..............................................................................' . .,

,
'<'0
<0..
3.5 (>

~..

-
c:'
0
3.0
.Q-
to 1.0 Leo
~ 2.5 CI:
g
a:
0.8

"'rr"
0.. 2.0
u::'" c.

I
0.6

(.~"'"
1.5 ~
0 u...
()
-J 1.0 o 0.4

0.5
'::l
L.............. ..........., ................ L ... , ............. L .., ...., ...., .-.,......~.~.-' 0.2 ':-~:"-::---~---=-=--::-':-~--'
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Reduced Velocity, V=UJ~l' ''(o)"b Reduced Velocity, V=Uj il '''(Il"b

(It) L e o 0 ::'1) rot ation ( b j r~co fre que n cy r at io

Figure 10: Normalized LeO flap rotation, /318. and LeO frequency ratio,
rdOJavs, reduced velocity. M=O,8.

At low Mach numbers good correlation has been demonstrated between theory and
experiment for LeO amplitudes and frequencies . Since these results are well
documented elsewhere see Dowell and Tang [41], here the recent work of Gordnier et
al [42,43] that has extended these correlations into the transonic range is emphasized. In
Figure 11 a cropped delta wing planform is shown. This configuration had been
investigated experimentally by Schairer and Hand [44] and the theoretical calculations
were done by Gordnier et al using both Euler and Navier-Stokes flow models. Initially
86

the theoretical calculations were done using a linear structural model, which gave
predicted LCO amplitudes much greater than those observed experimentally. This led
Gordnier to include nonlinearities in the structural model via Von Karman's nonlinear
plate theory that provided much improved correlation between theory and experiment.
See Figure 12, which shows a plot ofLCO amplitude versus flow dynamic pressure at a
fixed transonic Mach number. Note that the effects of viscosity are modest based upon
the comparisons of results using the Euler vs. Navier-Stokes models. Also the much
improved agreement obtained with the nonlinear structural model suggests that
aerodynamic nonlinearities per se are not as significant for this configuration as are the
structural nonlinearities, as Gordnier notes in his conclusions. Perhaps the most
significant impact of this example is to illustrate that, even for a transonic flow, there
are cases where structural nonlinearities may be dominant.

It is also interesting to note that for this example the most significant aerodynamic
nonlinearity was associated with leading edge vortices rather than shock motion. This
nonlinear aerodynamic flow mechanism has also been studied by Mook and colleagues
Preidikman et al [45] at low Mach numbers in a series of papers. As Gordnier and
Mook note, nonlinear vortex flow phenomena may be important when structural
nonlinearities are weaker and the corresponding structural motions greater. Of course if
the mean angle of attack is sufficiently large (say 10 degrees or more) then vortex
formation may be important for even smaller wing oscillations.

Wingtip

r
Leading-Edge

.5
o
cO
II
..c

.........- - - - c=11 .75 in ------1~

Figure I I: Planform of a cropped Delta Wing

 87

0.45
o E)Cperiment
0.4
o o D. Viscous - Linear
vViscous - Nonlinear
0.35 o Euler - Linear
OEuler - Nonlinear (RC
0.3 t> Euler - Nonlinear LC

.e 0.25
it:
\.Q 0.2
0
0.15 0
<>
V
o. , 0
0 I> <>
I>
<>
0.05 0
<>
4 5
Dynamic Pressure

Figure 12: LCO Response vs. Dynamic Pressure: Correlation ofExperiment with Various
Theoretical Models for Cropped Delta Wing

3.3. High Aspect Ratio, Beam-Like Wing

Again, the available literature is relatively small. See [47-54]. For very high aspect
ratio beams that may bend and twist, it has been known for many years that the flap-
wise bending, chord-wise bending (lag) and torsional deformation (twist) may couple
among themselves to produce a significant structural nonlinearity. See Hodges and
88

Dowell [47]. Also nonlinear aerodynamic stall effects are known to be important and
have been modeled successfully, though semi-empirically, Tran and Petot [48]. Such
issues were first pursued in the context of rotor blades that are often, of course, long
and slender. They have not been important for fixed wing aircraft for the most part.
However, recently and particularly in the context of some Uninhabited Air Vehicles
(UAV), very high aspect ratio fixed wing configurations are of interest. Thus,
researchers have pursued aeroelastic studies of this configuration. Notable work has
been done by Patil, Hodges and Cesnik [49-51] . Their pioneering theoretical studies
have shown a number of interesting nonlinear effects including the presence of LCO
and also the sensitivity of the onset of classical flutter as well as LCO to temporal
disturbances including initial conditions. Their work and that of the Duke team [52] and
the Texas A&M team [53] is discussed here including both theoretical and experimental
studies.

For a very high aspect ratio wing the structural nonlinearity is of a very different
kind than for a low aspect ratio plate-like wing. However it can still be attributed to a
nonlinear relationship between strain and displacement (gradient).

The key physical aspect of the structural nonlinearity arises from a subtle, mutual
coupling among the chordwise bending as well as transverse bending of the wing in
conjunction with twisting about the beam-rod elastic axis of the very high aspect ratio
wing. As in classical high aspect wing structural theory, the chordwise deformation is
treated as a combination of rigid body plunge (translation) and pitch (rotation) with all
elastic deformation occurring along the (one-dimensional) axis of the beam-rod. See
Figure 13. There can be both static as well as dynamic effects of the structural
nonlinearity. For example, static gravity loading or static aerodynamic loading due to a
static angle of attack may change the structural natural frequencies and hence the flutter
speed. The effect of a static angle of attack on flutter speed is shown in Figure 14. The
LCO amplitude vs. flow speed is shown in Figure 15 for a specified angle of attack.
For the latter a semi-empirical model of flow separation or airfoil/wing stall is also
included after the model of Tran and Petot [48].
 89

Figure 13: Photograph ofexperimental model for a high-aspect ratio wing in a wind tunnel
and physical representation ofthe experimental model: NA CA 0012 airfoil. span (L)= 18
inches. chord (2)=2 inches and bendiwz stiffness ratio. ['/1.=44.

From Figure 14, which shows both flutter velocity and flutter frequency vs. static
angle of attack or pitch angle, it is seen that the theoretical model captures well the
results observed experimentally. The LeO results in Figure 15 show hysteretic
response in both the theory and experiment with increasing and decreasing flow
velocity. By studying the nonlinear structural effects with and without flow separation
in the theoretical model, it is found that the hysteretic behavior is due to flow
separation.
38 ,----,----,----,----,----, 25

37 24
'00'
:g \. N
~ 23 nonlinear

.. .. ..-.-.-..-. ........
36 1-~------~-~-=
----1ifleat- .
g 22
~
C \ nonlinear
'u "::>0< _
_
~
0

I-<
35
"
---- ------\---_._---
...
..... .. .. linear. .
"
J:: 21
-
test

15"
I-<
34
15"
/ e /

' . -e/ 20
0:: 0::
test
33 19
32 ' - - - _ - 1 - _ - - 1_ _- ' - _ - - ' - _ - - '
18
o 0.5 1.5 2 2.5 0 0.5 1.5 2 2.5
pitch angle (degrees) pitch angle (degrees)

(a) Flutter velocity (b) Flutter frequency

Figure 14: Perturbation stability solution of the nonlinear aeroelastic system vs. pitch angle.
90

LCO Amplitude LCO Frequency

0.QJ8 24
:: 0.016
~ 0.014
the~IY';' g
23
theory

"""
..:.'.
0.012

..........e-.-
e'
" Q 22

t"
test
0.01
"1:l .... 113113113113

O?
.~
0.008
0.006
..>4.!: .g'"
0
21
-

' U
0.004 ...l

J
c;: 20
0
u 0.002
' f
,
...l , '
0 19
31 32 33 34 35 36 37 38 31 32 33 34 35 36 37 38
flow velocity (mls) flow velocity (mls)

Increasing Velocity Decreasing Velocity

0.02 o.oz
:: S
.,"
0.01 0.01
.g" .0
()
0 ()
0
"
c;: "
c;:

""""
O?
-0.01 """g -0.01
.~ .~

' 0,02 ' 0.02

c;: c;:
0.03 0.03
~ ~
.,; 0.04 .,; 0,04

0.05 -0.05
0 2 4 6 8 10 0 2 4 6 8 10
time (sec.) time (sec,)

Figure 15: Mid-span LeO amplitude, LeO frequency vs. flow velocity, and Leo time history at
mid-soan {or increasinz and decreasinz flow velocitv {or On = 0.25.

The agreement between theory and experiment is encouragingly good for what is
the rather complex behavior of a multi-dimensional nonlinear system. The flutter mode
is dominated by the second spanwise transverse bending and first torsion structural
natural modes. The first chordwise bending mode is also an essential contributor to the
LCO, however. For further discussion of this case, see [52].

Further recent work has been done by Kim and Strganac [53] who used the
structural model of Crespo da Silva [54] to examine store-induced LCOs for a
cantilevered wing-with-store configuration. This model contains structural coupling
terms and quadratic and cubic nonlinearities due to curvature and inertia. Several
 91

possible nonlinearities, including aerodynamic, structural, and store-induced sources,

were considered. Structural nonlinearities were derived from a large deformation
theory. Aerodynamic nonlinearities were introduced through a stall mode1. Store-
induced nonlinearities were introduced by kinematics of a suspended store. All of these
nonlinearities retained cubic nonlinear terms. To examine systematically the response
characteristics, phase plane analysis was performed and the effect of each nonlinearity,
as well as combinations of the nonlinearities were studied. Although various forms of
nonlinear responses were found, of interest was the finding of LCO response at speeds
below the flutter velocity. Furthermore, an unstable boundary was found, above which
responses were attracted to the LCO and below which the responses were attracted to
the nominal static equilibria. Of special importance, such subcritical response was
found for only the case in which complete consideration of structural, aerodynamic, and
store-induced nonlinearities was given. This suggested that studies of nonlinear
aeroelasticity must sometimes consider a full aircraft configuration.

4. EFFICIENT COMPUTATION OF UNSTEADY AERODYNAMIC FORCES

Before turning to the effects of aerodynamic nonlinearities, a review of recently

developed methods for computational fluid dynamics modeling is undertaken.

The literature on unsteady aerodynamic forces alone is quite extensive. A

comprehensive assessment of current practice in industry is given by Yurkovich, Liu
and Chen [55]. An article that focuses on recent developments is that of Dowell and
Hall [56]. They also developed a bibliography of some five hundred items available in
electronic form from the authors. Other recent and notable discussions include those of
Bennett and Edwards [57] and Beran and Silva [58]. Much of the present focus of work
on unsteady aerodynamics is on developing accurate and efficient computational
models. Standard computational fluid dynamic [CFD] models that include the relevant
fluid nonlinearities are simply too expensive now and for some time to come for most
aeroelastic analyses. Thus there has been much interest in reducing computational costs
while retaining the essence of the nonlinear flow phenomena.

There are three basic ideas that are currently being pursued with some success in
retaining the accuracy associated with state-of-the-art CFD models while reducing
aerodynamic model size and computational cost [56,58]. One is to consider a small
(linear) dynamic perturbation about a (nonlinear) mean steady flow. The steady flow
may include both the effects of a shock wave and flow separation, but any shock or
flow separation region motion is considered in the dynamically linear approximation.
92

That is, it is assumed that the shock motion or the separation point motion, for example,
is linearly proportional to the motion of the structure. This is sufficient to assess the
linear stability of the aeroelastic system, but not to determine LCO amplitudes due to
nonlinear aerodynamic effects. Of course in those cases where the structural
nonlinearities are dominant, this simpler aerodynamic model is all one needs to
determine LCO. This approach has enormous computational advantages as the
computational cost is comparable to that of a steady flow CFD model since the
unsteady calculation per se uses a linear model. Although this time linearization
approach can be employed in either the time or frequency domain, usually it is most
efficient to do the calculation in the frequency domain. The time linearization combined
with the frequency domain aeroelastic calculation is several orders of magnitude faster
than a time marching nonlinear CFD method for the determining the flutter boundary
without any loss of accuracy for this purpose. And if a parameter study is conducted
where only the structural parameters are varied, then the additional computational cost
is no more than that using classical aerodynamic methods.

Moreover this approach can be extended to NONLINEAR unsteady flows by

expanding the flow solution in terms of a Fourier series in time. This assumes the flow
motion is periodic in time of course and is most effective if the number of important
harmonics needed in the Fourier Series is small. However this is true of many (but not
all) flows of interest. Here the computational cost is a small multiple (say a factor of
three) of the cost of a steady flow solution. This is the second major idea and again the
harmonic balance method is much faster for determining the LCO than a time
marching of a nonlinear CFD code, typically by one or two orders of magnitude.

The third major idea is to determine the dominant spatial modes of the flow field
and use these, rather than many local grid points, to represent the flow. This is a class of
so called Reduced Order Models. The reduction is from the very large number of flow
variables at local grid points (on the order of a million or more) to a small number of
spatial modes (typically less than one hundred). The reduction in computational cost for
aero elastic analysis is several orders of magnitude, i.e. a factor of one thousand or
more. This approach has been used for potential flow, Euler flow and Navier-Stokes
flows (with a turbulence model) for small dynamic perturbations about a nonlinear
mean steady flow (recall the first major idea discussed above). Current research is
underway to consider nonlinear unsteady flows. Kim and Bussoletti [59] have
discussed how one can construct an optimal reduced-order aero elastic model within the
framework of time linearized CFD models. While in principal fluid eigenmodes can be
used and indeed they provide the underlying framework for reduced order modeling,
the technique known as proper orthogonal decomposition has proven to be the most
computationally attractive method for constructing a set of global modes for the
 93

reduced order model. With a Reduced Order Model (ROM), the aeroelastic
computations are essentially at no additional cost beyond the construction of the ROM
itself. Or to say it another way, the aero elastic computations are no more expensive than
using a classical unsteady aerodynamics. Moreover one can construct a root locus
solution for the true aeroelastic eigenvalues (true damping and frequency of each
aeroelastic mode).

A parallel approach to the last idea is to use the ideas of transfer functions
(sometimes called describing functions in the nonlinear case) in the frequency domain
or Volterra series in the time domain to create small computational models from large
CFD codes. In this approach the form of the transfer function or describing function
(or its time series equivalent) is assumed and the coefficients of the reduced order
model are determined from data generated by the CFD code in a time simulation. A
good discussion of this approach is contained in Beran and Silva [58] and in a series of
papers by Silva [60-65]. Again this approach is most fully developed for the
dynamically linear case and the dynamically nonlinear case is currently a subject of
active research. Raveh, Levy and Karpel [66] have offered a recent and useful
discussion of how these ideas can be implemented within the framework of an Euler
based, CFD model and provided an example of the well studied AGARD 445.6 wing.
Also see Raveh [67].

All of these ideas, individually or in combination, provide the promise of dramatic

reductions in computational costs for unsteady transonic flows including the effects of
shock motion and flow separation. In addition progress continues to be made in
improving the computational efficiency of time marching simulations; see Farhat
[68,69]. And the ideas of dynamic (time) linearization and reduced order modeling can
be used in either the time or frequency domains.

5. NONLINEAR AERODYNAMIC EFFECTS INCLUDING LARGE SHOCK

MOTION AND SEPARATED FLOW

These aerodynamic nonlinearities are both the most difficult to model theoretically and
also to investigate experimentally. Hence it is perhaps not surprising that our
correlations between theory and experiment are not yet what we might like them to be.
As a corollary one might observe that it will in all likelihood be easier to design a
favorable nonlinear structural element to produce a benign LCO, than to assure that
flow nonlinearities will always be beneficial with respect to LCO.
94

5.1. NACA 64AOJOA Conventional Airfoil Models

In Fig. 16 recent results are shown for the LCO of a NACA 64AO 10 airfoil in plunge
and pitch as predicted by an Euler flow model Kholodar et al [70]. Here the LCO
amplitude is plotted versus the flutter speed index (FSI) for a range of Mach number.
As can be seen the LCO is relatively weak (the curves of the figure are nearly vertical)
for most Mach numbers. And for those Mach numbers where the LCO is relatively
strong, it can be either benign (the curves bend to the right) or detrimental (the curves
bend to the left) leading to LCO below the flutter boundary. This example also points
out the substantial amount of data needed to assess LCO under these circumstances. A
Navier-Stokes model has also been used to assess LCO of this configuration for a more
limited range of parameters. The results (not shown) indicate a modest effect of
viscosity provided the mean angle of attack is sufficiently small and no flow separation
occurs.
7
j\ I ~-i ,,'
t I \,84 ,; /

Ji / "
't f
88 f ./'
!
....<78
..i "'"
./'

9i I I/
11
/
I\ f i
( /
f f I 915 !
I I 8 ..III-~'~
I I 1/.
~--"r-I'--
i S/I" __

M=O,75\
t I J. ,. ~
I
./'"
_.IJ'---~

II ,/'
' I
\J
J J ~ ...
.81 . _....
1lil
t
~ Ir-'-
Ii ,.It" ",'
I II /I
.... ~"

tH
/ i I
i I I ... "
of. H f-"''''-
I I II
i .. ,6, P/
(

o iii i 1 i'
3 4 5 6 7 9 10 11

Figure 16: LCO Amplitude vs. Reduced Velocity for NACA 64A010 Airfoil
 95

5.2. NLR 7301 Supercritical Airfoil Models

Another configuration of interest is the supercritical airfoil, NLR 730 I, which has been
studied experimentally by Schewe and his colleagues [71-74]. This has in turn inspired
several theoretical studies using either an Euler or Navier-Stokes flow model. A
correlation among several theoretical models and the result of experiment is shown in
Fig. 17. This figure is drawn from the paper by Thomas, Dowell and Hall [75] who
used a harmonic balance LCO solution method. Results are also shown from Weber et
al [76] and Tang, L. et al [77] both of whom used the more computationally demanding
time marching technique. Note that there is only a single data point from the latter as is
the case from the experiment. However it is clear that to have a more meaningful
correlation it is highly desirable to provide results for LCO amplitude over a range of
FSI and Mach number. Hence it is not yet clear what the conclusion should be with
respect to correlation between theory and experiment. It does appear that the several
theoretical results are in reasonable agreement. More correlations with the
experimental data are needed.

Computational conditions are sensItIve and care must be taken to achieve

reasonable steady initial pressure distributions for this configuration. Also, the LCO
conditions appear to be very sensitive to details of the computations. Tang, L. et al give
results from the CFL3D-NS code illustrating effects of turbulence models, single-block
and multi-block (parallel), multigrid subiterations, and time step. Agreement between
theory and experiment for the LCO motion amplitudes has been difficult to achieve for
this case even including the effects of wind tunnel wall interference. See Castro et al
[78].

5.3. A New Experimental A era elastic Model

Schewe et al [74] have recently further simplified their aeroelastic model by limiting
the airfoil motion to a single degree-of-freedom (DOF) in pitch thus enabling them to
study strictly torsional flutter. According to Schewe et al [74], with their new test
facility and equipment for measuring torsional flutter, they are also able to hold the
mean angle-of-attack constant as flow conditions such as inlet Mach number are varied.
96

-O'l
Q)
'0
~
5.0

4.0
~ Inviscid '" Viscous
I~

Q)
'0 3.0
::J
~
0. Schewe et. al. Expr.
E 2.0 Weber et. al. Compo

-oa::
~ Tang et. al. Compo
..c
o
1.0
i I Flutter
9'L point
()
....J
JKfl
0.60

Figure 17: LCO Response Amplitude vs. (Reduced) Flow Velocity: Comparison of
Inviscid and Viscous Flow Modelsfor NLR 7301 Supercritical Airfoil

Furthennore, they have tested their new single degree-of-freedom aeroelastic model
in a new "adaptive" test section, which has no wind-tunnel wall porosity. With no wall
porosity, modeling wind-tunnel wall boundary conditions in a CFD model is much
simpler and more accurate.

Another valuable aspect of this single structural DOF aeroelastic configuration is

that there are now many more experimental data points, which are shown in Figure 18.
In this case, Schewe et al studied how LCO amplitude response varied with wind-tunnel
Mach number for a fixed wind-tunnel test section inlet total pressure.

In the following we will first consider a range of fixed Mach numbers, and
determine the corresponding total pressure in the case of a very small pitch amplitude in
order to identify the flutter onset total pressure boundary. Once the flutter onset
boundary is established, we will consider a fixed total pressure and then solve for the
corresponding inlet Mach number (Minlt) and frequency (00) for a range of LCO
unsteady pitch amplitudes ((Xl)'
 97

Figure 19a shows the overall computational mesh used in modeling the
configuration. Figure 19b shows the computed Mach number contours for an inlet
Mach number of Minlt = 0.758 and an inlet Reynolds number of Reinlt = 2.2 x 106 .

Next, in order to determine the flutter onset condition for the inlet total pressure
poinlt, we first compute the unsteady, i.e. first harmonic, airfoil moment for a very
small pitch amplitude, in this case UI = 0.001 degree pitch, via the harmonic balance
solver, over a range of frequencies. This is done in order to approximate linear small
disturbance aerodynamics. For the HB flow solver, as the pitch amplitude goes to zero,
one recovers precisely the same result as for a truly linearized CFD model. This is a
nice aspect of the HB method in that it can be used for linear unsteady aerodynamics as
well as nonlinear unsteady flows associated with larger amplitude motions.

Finally, the structural parameters used for the LCO analysis are those reported by
Schewe et. al. [74].

1.2

0) ()
Q) 1.0
: f~'..

'.../

itS 0.8
0
0.)
"0 ()
:8 0.6
a...
E
0.4
.c
~
0::
0 0.2
0
....J
0.0
0.750 0.754 0.756 0.758 0.760
Inlet Mach Number, M,nn

Figure 18: German Aerospace Center DLR NLR 7301 Aeroelastic Model Single Degree-
ofFreedom Experimental LCO Amplitude versus Inlet Mach Number.
98
Computational Mesh

Mach Contours

Figure 19: Complete Tunnel Test Section Computational Mesh and Computed Mach
Number Contours. M inlt = 0.758, Reinlt = 2.2 X 106 .

5
:R' rn
Q) 180
~
J
c
4 ,l/ 120
Q) e>
0
E
3 C
'" 60
~ Q)
>- E
'0 0 0
<IS
Q)
2 ~
'iii
c:
::::l
'"<IS
"0
Q)
-60
'0 'iii
c:
Q) ::::l -120
'0
B '0
'c
Q) 0 '"en
~
<IS 30 40 50 60 70 80
Frequency, f (Hertz)
90 100
'"
.<::;
0..

(a) Magnitude of Unsteady Moment (b) Phase Angle of Unsteady Moment

Figure 20: Magnitude and Phase Angle of Unsteady Moment as Function of

Frequency for Four Different Inlet Mach Numbers.
 99

5.3.1. Unsteady Aerodynamics: Figure 20a shows the magnitude of the computed
unsteady (i.e. fIrst harmonic) moment coeffIcient about the elastic axis (quarter chord)
as a function of unsteady frequency for four different Mach numbers. Figure 20b
shows the computed phase angle for the unsteady moment for the same four Mach
numbers. Readily apparent is a resonant condition, which appears for the higher Mach
numbers near a frequency of 70 Hz. This can be seen from the large increase in the
magnitude of the unsteady moment coeffIcient in addition to the sign change of the
unsteady moment coeffIcient phase angle for the higher Mach numbers near a
frequency of70 Hz.

5.3.2. Mach Number Flutter Onset Trend: Figure 21a shows the computed flutter onset
inlet total pressure as a function of Mach number. Figure 21 b shows the corresponding
flutter onset frequency. The flutter total pressure and frequency are determined by
using the unsteady moment data shown in Figure 20. As can be seen, there is a
dramatic reduction in the predicted flutter onset total pressure over a very small Mach
number and frequency range where the resonance condition is observed for the
unsteady airfoil moment.

5.3.3. Limit Cycle Oscillation Solutions: One of the computational diffIculties for these
wind tunnel test conditions is that we encounter very slow numerical convergence of
the unsteady flow solver in the range of Mach numbers and frequencies associated with
the single degree-of-freedom freedom flutter onset total pressure dip, due to what
appears to be an apparent wind tunnel resonance condition. Also, we have diffIculty
determining a Mach number where flutter onset occurs for the experimentally reported
inlet total pressure of Poinlt = 40K Pa, due to the very slow numerical convergence.
Nevertheless, we decided to attempt to model the nonlinear aeroelastic behavior at a
higher inlet total pressure.
100

100
~ 1400
Cl. N
x
-~ 1200 E- 90
cfE >.
XExperimental ' u
~ 1000 o Computed
<:
::>
(fj
!g 80
0-
(fj
800
1 1
Cl. L1.
iii 600 (i) 70

o'"
'0 c
~
Q) 400
<II Experimental Pc B 60 Natural Frequency, 0),,=44.4 (Hz)
0
C
200
:;
u::
2
:; 0 0.77 0.78
u:: 0.74 0.75 0.76 0.77 0.75 0.76
Inlet Mach Number, Moe" Inlet Mach Number, M,nl!

Figure 21: Computed Mach Number Flutter Onset Trend,

Figure 22 shows the computed limit cycle oscillation pitch amplitude at as a

function of inlet Mach number M inlt based on an inlet total pressure of POinlt = 300K Pa
in the aeroelastic governing equation. The Reynolds number is based on the
experimental total pressure and static temperature. Over the Mach number range
shown in Figure 22, the Reynolds number is approximately Re "" 2.2 x 106 We also
attempted inlet total pressures of POinlt = 80K Pa, and 200K Pa, however due to the very
slow convergence rates near the minimum of the flutter dip, we were simply unable to
achieve convergence in a reasonable computational time.

Even so, it is interesting and perhaps significant to note that the experimental results
of Figure 18 and the theoretical results of Figure 22 are in good qualitative and even
reasonable quantitative agreement.
 101

1.0

-0'>
<l.> 0.8
~
.J
.g 0.6

.=-0..
E 0 .4
<t
..c.
.::
a.. 0 .2.
oo
-l
0.0
0.7630 0 .7640 0.7645 0.7650 0.7655
Inlet Mach Number, M.n "

Figure 22: Computed Limit Cycle Oscillation Behavior as a Function of

Mach Number, pOinlt = 300 K Pa, Re "" 2.2 x 106

5.4. AGARD 445.6 Wing Models

The AGARD 445.6 wing has been discussed by many authors in terms of its flutter
boundary; now we turn to very recent results from Thomas, Dowell and Hall [80] for
LCO. The correlation between theory and experiment for the flutter boundary is shown
in Figure 23 where the Euler flow model is that of Thomas et al. In addition we have
results for LCO amplitude versus FSI for various Mach number. See Figure 24. Note
that a value of first mode non-dimensional modal amplitude of ~=0.012 as shown in this
figure corresponds to a wing tip deflection equal to one fourth of the wing half-span.
Note also that in general the LCO is predicted to be weak and there is no Mach number
for which a benign LCO is predicted. Indeed the strongest LCO is detrimental and
occurs at the low supersonic Mach numbers, i.e. M=1.141 and 1.072. This means that
LCO may occur below the flutter boundary at these two Mach numbers and perhaps
this explains at least in part why flutter (or really an unstable LCO) occurred in the
experiment below the predicted flutter boundary.
102

0.7 ~"""I'"""'I_-"""--"""-""''''''"''T'"''''''-r-.......,
Ya t es E <penmental
Comp ut3tlOn al M ocJe l 49) 33,, 33 ('riC

02~~~~~~~~~~--~~~~
0.4 05 OS 0.7 08 0.9 1.0 1.1 12
Mach Number. M""

Figure 23: Flutter Speed Index vs. Mach Number for AGARD Wing 445.6: Comparison oj
Theory and Experiment

Small amplitude LCO behavior for the AGARD 445.6 wing has also been
calculated by Edwards [81] using the CAP-TSDV computer code which employs a
potential flow model combined with an integral boundary layer representation. The
majority of published calculations for this wing model (actually a series of models with
similar planforms) are for the "weakened model #3" tested in air, since this test covered
the largest transonic Mach number range and showed a significant transonic dip in the
flutter boundary. The focus on this particular configuration may be in some ways
unfortunate, in that the model tested in air resulted in unrealistically large mass ratios
and small reduced frequencies. Weakened models #5 and #6 were tested in heavy gas
and had smaller mass ratios and higher reduced frequencies. Very good agreement was
obtained with experiment for flutter speed index using the CAP-TSDV code over the
Mach number range tested. For the highest Mach number tested, M=O.96, it was noted
that damping levels extracted from the computed transients were amplitude dependent,
an indicator of nonlinear behavior. It was also found that small amplitude divergent (in
 103

time) responses used to infer the flutter boundary would transition to LeO when the
calculation was continued further in time. The wing tip amplitude of the LeO was
approximately 0.12 inches peak-to-peak, a level that is unlikely to be detected in wind
tunnel tests given the levels of model response to normal wind tunnel turbulence.

It is known that viscous flow effects are important at supersonic free-stream Mach
numbers for the flutter boundary of this configuration [81], hence it may not be
surprising to find a significant effect on Leo as well. In that respect, the recent results
of Thomas et. al. [82] are of interest. Figure 25 shows computed Leo solutions for the
viscous and inviscid flow models of the 445.6 wing configuration for the Moo = 1.141
flow condition. Shown is the LeO amplitude of the first structural modal coordinate,
~h versus the reduced velocity V. The beginning of a strongly stable nonlinear LeO
behavior trend is apparent for the viscous flow model for very small amplitudes of
motion. Note the aerodynamic nonlinearity is much weaker and the Leo amplitude
correspondingly greater for the inviscid flow model.

0.012 r--...--...,..........,..--...,r--"'--""'T"""---.---...,r--.,........,
\>J.

'0
M.=O.960
~ 0.010 M.=O.901

0.008

-;; 0.006
.~
u..
'0
-3 0.004
:l
:::
a..
~ 0.002
o
u
-J 0.000 '--......--ti~.......~~I-W-w-L..-......-a.w.......--'
0.2 0.3 0.4 0.5 0.6 0.7
1.12
Reduced Velocity, U .111 wab

Figure 24: LCO Amplitude vs. Flutter Speed Index (Reduced Velocity) for Various Mach
Number for AGARD 445.6 Wing
104

5.5. Benchmark ModeilNA CA 0012

Rivera et. al. [83] provides detailed data from a NACA 0012 Benchmark Model wind
tunnel experiment performed in the NASA Langley Transonic Dynamics Tunnel. In
wind tunnel flutter testing the procedure usually consists of fixing the Mach number
and then decreasing or increasing the density of the flow until the system reaches
flutter. The flutter trajectory for the values of the speed of sound from [83] is marked
on the computed flutter surface computed by Kholodar et.al. [84] using an Euler CFD
model and the efficient aerodynamic modeling methods discussed earlier in this paper;
see Figure 26a. The trajectory has a gap in the Mach number range 0.82 < M < 0.92,
where the compatibility surface linking VIl, Il and M (flutter speed index, mass ratio
and Mach number) passes above the transonic dip "valley". Since the flutter velocity
per se (or Mach number) is taken to be identical for both the computational and
experimental models, it is advantageous for comparison of the computational and
experimental results to consider l/winstead of the flutter speed index VIl. A plot of
1/Ilf versus the Mach number is discussed next, see Figure 26b. In the range 0.30 :s; M
:s; 0.82, there is good agreement between the numerical and experimental results. In the
range 0.82 < M < 0.88, no (experimental) flutter data have been presented in [83].
Computed results of the Euler HB method reveal that in the range 0.82 < M < 0.92
theoretical flutter boundary precipitously drops for the considered values of the mass
ratio and speed of sound. The dotted lines (at M = 0.82 and 0.92) that go down to low
values of l/W are the expected results for values of Il beyond those considered in the
numerical study. It is also possible that at very high mass ratios the value of Vf in the
transonic dip "valley" region (Figure 26a) could rise enough to intersect with the
compatibility surface in the Mach number range making the dotted lines in Figure 26b
connect at very low value of 1fllJ:
 105

Leo Response Amplitude Vs. Reduced Velocity

,';; 0.0015
cD
"C
o
:2
e:;J
t5 0.0010
2
U5
'
u:
'0
(J) 0.0005
"C
:E
0.
E

oU 0.0000
-' OA80,500.520,54 0,56 0,58 0,60 0,62 0.64 o.sn
Reduced Velocity, U../ll'i2Ol"b Duke-Un

Figure 25: Computed AGARD 445.6 Wing Configuration LCO Characteristics.

M~ = 1.141. and Re~ = 860.000 for Viscous Case.

:f:"

M
a) b)

S'
s

e-

M M
c) d)

Figure 26: Comparison of Computational Results with those from Wind Tunnel Testing.
A) Flutter Speed Surface vs. Mach Number and Mass Ratio with the Numerically
Simulated Wind Tunnel Trajectory as Indicated, b) Inverse Flutter Mass Ratio vs. Mach
Number, c) Flutter Reduced Frequency and Frequency vs. Mach Number, d) Flutter Mode
vs. Mach Number.
106

In the range 0.88 g { ::; 0.95, Rivera and his colleagues [83] observed a "plunge
instability", where "the flutter motion consisted of primarily the plunge mode." The
Euler HB method branch in this range of Mach number starts at higher Mach numbers
and the l/ll! values are larger than those of the experiment. The reason for these
disagreements between experiment and theory is believed to be the influence of viscous
effects (which are not accounted for in Euler aerodynamic models): "Flow
Visualization using tufts indicated strong shock-induced separation for this Mach
number range" [83]. Moreover, viscosity increases the effective airfoil thickness as
perceived by the flow. Increased thickness causes a shift of the transonic dip to lower
Mach numbers and also for a given mass ratio it lowers the flutter velocity at Mach
numbers above the transonic dip. See the more extensive discussion of this case in
Kholodar et. al [84].

In the 0.30 ::; M::; 0.82 range, frequency and reduced frequency of flutter results
correlate well between the computational and wind tunnel models, see Figure 26c. (No
experimental flutter frequency or mode data were recorded for high transonic Mach
numbers in Reference [83]). The flutter frequency results show that this is a
coalescence flutter, with a dominating plunge mode. This is also seen in Figure 26d
where the flutter eigenvector is shown. Bendiksen was perhaps the first to note the
precipitous decrease in the flutter boundary that may occur in transonic wing tunnel
testing [85].

6. CONCLUDING REMARKS

Nonlinear effects in aeroelastic systems may be either favorable or unfavorable or a

combination of both. For example, when a structural stiffness nonlinearity is equivalent
to a hardening spring as in the Von Karman model of a plate, no limit cycle oscillations
(LCO) will exist below the flutter boundary determined in the absence of nonlinearity.
And above the flutter velocity the nonlinearity will limit the response, a clearly
favorable outcome. However for other nonlinearities such as structural free-play or
aerodynamic nonlinearities due to flow separation or large shock motion, the effect of
the nonlinearity may be to induce LCO below the nominal flutter velocity, but still limit
the LCO response to a finite amplitude both below and above the nominal flutter
velocity. Whether such nonlinear effects are favorable or not will depend very much on
the particular circumstances and parameters involved. Nonetheless it is clear that
nonlinear effects often lead to LCO and in their absence the alternative would be
catastrophic flutter leading to structural failure. Hence nonlinearities in aeroelastic
systems provide an opportunity for improved safety and performance of modern
 107

aerospacecraft if reliable and computationally effective analysis and design methods

can be developed.

Recent advances in computational models and solution techniques now permit

efficient nonlinear aeroelastic analysis including the determination of limit cycle
oscillations (LCO) for a number of nonlinearities as described in this paper. The most
promising methods are based upon two fundamental ideas. Firstly, time-linearized
models of the fluid (and structure) may be used to construct highly compact, reduced
order aerodynamic (and structural) models. Secondly, with the flutter velocity,
frequency and eigenmode (the aeroelastic motion) determined from such an analysis,
the results may be used as the first step in a rapidly converging iteration process to
solve for the nonlinear LCO using a novel form of the harmonic balance method for
determining the nonlinear aerodynamic forces and the structural deformation.

Nonlinearities that have been successfully modeled theoretically (and the results
confirmed by experiment) include

Structural free-play
Structural geometric (strain-displacement nonlinearities) for low and high
aspect wings of relatively simple construction
separated flow nonlinearities (at low Mach number using a semi-empirical
[ONERA] aerodynamic model)

In addition theoretical fluid models for inviscid or viscous flows with large shock
motion have now been developed that permit computationally efficient flutter and LCO
analysis arising from such aerodynamic nonlinearities. Correlation with experiment
remains an open challenge.

7. REFERENCES

I. Cunningham, A. M., Jr., "A Generic Nonlinear Aeroelastic Method with Semi-Empirical Nonlinear
Unsteady Aerodynamics," Vol. I and 2, AFRL-VA-WP-TR-1999-3014, 1999.

2. Friedmann, P.P., "Renaissance of Aeroelasticity and Its Future," Journal of Aircraft, Vo1.36, No.1, pp.
105-121,1999.
108

3. Woolston, D.S., Runyan, H.L. and Andrews, R.E., "An Investigation of Effects of Certain Types of
Structural Nonlinearities on Wing and Control Surface Flutter," Journal of Aeronautical Sciences, Vol. 24,
1957, pp. 57-63.

4. Shen, S.F., "An Approximate Analysis of Nonlinear Flutter Problems," Journal of Aeronautical Sciences,
Vol. 26,1959, pp. 25.

5. Breitbach, E., "Effects of Structural Nonlinearities on Aircraft Vibration and Flutter," AGARD Technical
Report 665, 1977.

6. Tang, D.M., and Dowell, E.H., "Comparison of Theory and Experiment for Nonlinear Flutter and Stall
Response ofa Helicopter Blade," Journal of Sound and Vibration, Vol. 165(2), 1993, pp. 251-276.

7. Brase, L.O. and Eversman, W., "Application of Transient Aerodynamics to the Structural Nonlinear Flutter
Problem," Journal of Aircraft, Vol. 25, No. II, 1988, pp. 1060-1068.

8. Yang, Z.C. and Zhao, L.C., "Analysis of Limit Cycle Flutter of an Airfoil in incompressible Flow,"
Journal of Sound and Vibration, Vol. 123,1988, pp. 1-13.

9. Yang, Z.e. and Zhao, L.C., "Chaotic Motions of an Airfoil with Nonlinear Stiffness in Incompressible
Flow," Journal of Sound and Vibration, Vol. 138, 1990, pp. 245-254.

10. Hauenstein, A.J., Zara, J.A., Eversman, W. and Qumei, I., "Chaotic and Nonlinear Dynamic Response of
Aerosurfaces with Structural Nonlinearities," AlAA-92-2547-CP, 1992.

II. Liu, J. K. and Zhao, L.e., "Bifurcation Analysis of Airfoil in Incompressible Flow," Journal of Sound
and Vibration, Vol. 154, 1992, pp. 117-124.

12. Tang, D.M., and Dowell, E.H., "Chaotic Stall Response of a Helicopter in Forward Flight," Journal of
Fluids and Structures, Vol. 6, No.3, 1992, pp. 311-335.

13. Tang, D.M., and Dowell, E.H., "Flutter and Stall Response of Helicopter Blade with Structural
Nonlinearity," Journal ofAircraft, Vol. 29, No.5 Sept-Oct, 1992, pp. 953-960.

14. Tang, D.M. and Dowell, E.H., "Nonlinear Aeroelasticity in Rotorcraft," Journal of Mathematical and
Computer Modeling, Vol. 18, No. 3/4,1993, pp. 157-184.

15. Tang, D.M., and Dowell, E.H., "Experimental and Theoretical Study for Nonlinear Aeroelastic Behavior
ofa Flexible Rotor Blade," AJAA Journal, Vol. 31(6) June 1993, pp. 1133-1142.

16. Price, S.J., Alighanbari, H. and Lee, B.H.K., "Post-Instability Behavior of a Two-Dimensional Airfoil
with a Structural Nonlinearities," Journal of Aircraft, Vol. 31, 1994, pp. 1395.

17. Price, S.J., Alighanbari, H. and Lee, B.H.K., "The Aeroelastic Response of a Two-Dimensional Airfoil
with Bilinear and Cubic Structural Nonlinearities," Journal of Fluids and Structures, Vol. 9, 1995, pp. 175-
193.

18. Tang, D.M., and Dowell, E.H., "Response of a Non-rotating Rotor Blade to Lateral Turbulence in
Sinusoidal Pulsating Flow, Part 2: Experiment," Journal of Aircraft, Vol. 32, No.1, Jan-Feb, 1995, pp. 154-
160.
 109

19. Conner, M.D., Virgin, L.N., and Dowell, E.H., "A Note on Accurate Numerical Integration of State-Space
Models for Aeroelastic System with Freeplay," AIAA Journal, Vol. 34, 1996, pp. 2202.

20. Kim, S.H. and Lee, L, "Aeroelastic Analysis of a Flexible Airfoil with a Freeplay Nonlinearity," Journal
of Sound and Vibration, Vol. 195, 1996, pp. 823-846.

21. Lee, B.H.K and Leblanc, P., " Flutter Analysis of a Two-Dimensional Airfoil with Cubic Nonlinear
Restoring Force," National Research Council of Canada, Aeronautical Note NAE-AN-36, NRC No. 25438,
1996.

22. O'Neil, T., Gilliat, H. and Strganac, T., "Investigation of Aeroelastic Response for a System with
Continuous Structural Nonlinearities," AIAA Paper 96-1390, 1996.

23. Tang, D.M., and Dowell, E.H., "Nonlinear Response ofa Non-Rotating Rotor Blade to a Periodic Gust,"
Journal of Fluids and Structures, Vol. 10, No. 7, 1996, pp. 721-742.

24. Conner, M.e., Tang, D.M. , Dowell, E.H. and Virgin, L.N., "Nonlinear Behavior of A Typical Airfoil
Section with Control Surface Freeplay: A Numerical and Experimental Study," Journal of Fluids and
Structures, 11,1997, pp. 89-112.

25. Lee, B.H.K., Gong L. and Wong Y.S., "Analysis and Computation of Nonlinear Dynamic Response of a
Two-Degree-of-Freedom System and Its Application in Aeroelasticity," Journal of Fluids and Structures,
Vol. 11, 1997, pp. 225-2467.

26. Tang, D.M ., Dowell, E.H. and Virgin, L.N., "Limit Cycle Behavior of an Airfoil with a Control Surface,"
Journal of Fluids and Structures, Vol. 12, No.7, 1998, pp. 839-858.

27. Lee, B.H.K, Jiang, L.Y. and Wong Y.S., "Flutter of An Airfoil with a Cubic Nonlinear Restoring Force,"
Journal of Fluids and Structures, Vol. 13, 1999, pp. 75-101.

28. Lee, B.H.K, Price, S.J. and Wong Y.S., "Nonlinear Aeroelastic Analysis of Airfoils: Bifurcation and
Chaos," Progress in Aerospace Sciences, Vol. 35, No. 3, 1999, pp. 205-334.

29. Kim, D.H. and Lee, L, "Transonic and Low-Supersonic Aeroelastic Analysis of a Two-Degree-of-
Freedom Airfoil with a Freeplay Nonlinearity," Journal of Sound and Vibration, Vol. 234, No.5, 2000, pp.
859-880.

30. Liu, L., Wong, Y.S. and Lee, B.H.K., "Application of the Center Manifold Theory in Nonlinear
Aeroelasticity," Journal of Sound and Vibration, Vol. 234, No.4, 2000, pp. 641-659.

31. Dowell, E. H., Thomas, J.P. and Hall, KC ., "Transonic Limit Cycle Oscillation Analysis Using Reduced
Order Modal Aerodynamic Models," AlAA Paper 2001-1212, presented at the 42 nd
AIANASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, Seattle,
Washington, ApriI16-19, 2001.

32. Von Karman, T. and Sears, W.R., "Airfoil Theory for Non-Uniform Motion," Journal of Aeronautical
Sciences, Vol. 5, No. 10, August 1938, pp. 379-390.

33. Dowell, E.H., "Aeroelasticity of Plates and Shells," Kluwer Academic Publishers, 1975.
110

34. Mei, c., Abdel-Motagaly, K. and Chen, R.R., "Review of Nonlinear Panel Flutter at Supersonic and
Hypersonic Speeds," Applied Mechanics Reviews, Vol. 52, No. 10, October 1999, pp. 321-332.

35. Dowell, E.H. and Hall, KC., "Modeling of Fluid-Structure Interaction," Annual Review of Fluid
Mechanics, Vol. 33,2001, pp. 445-490.

36. Hall, K.C., "Eigenanalysis of Unsteady Flows About Airfoils, Cascades, and
Wings," AIAA Journal, Vol. 32, No. 12, 1994, pp. 2426-2432.

37. Dowell, E.H., "Eigenmode Analysis in Unsteady Aerodynamics: Reduced Order Models," AIAA Journal,
Vol. 34, No.8, 1996, pp. 1578-1588.

38. Tang, D.M., Dowell, E.H. and Hall, KC., "Limit Cycle Oscillations of A Cantilevered Wing in Low
Subsonic Flow," AIAA Journal, Vol. 37, No.3, 1999, pp. 364-371.

39. Tang, D.M., Herry, J.K. and Dowell, E.H., "Limit Cycle Oscillations of Delta Wing Models in Low
Subsonic Flow," AIAA Journal, Vol. 37, No. 11, 1999, pp. 1355-1362.

40. Tang, D.M., Herry, lK and Dowell, E.H., "Response of A Delta Wing Model to A Periodic Gust in Low
Subsonic Flow," Journal ofAircraft, Vol. 37, No.1, 2000, pp. 155-164.

41. Dowell, E.H. and Tang, D., "Nonlinear Aeroelasticity and Unsteady Aerodynamics," AIAA Journal, Vol.
40, No. 9, 2002, pp. 1697-1707.

42. Gordnier, R.E. and Melville, R.B., "Physical Mechanisms for Limit-Cycle Oscillations of a Cropped
Delta Wing," AIAA Paper 99-3796, Norfolk, VA, June 1999.

43. Gordnier, R.E. and Melville, R.B., "Numerical Simulation of Limit-Cycle Oscillations ofa Cropped Delta
Wing Using the Full Navier-Stokes Equations," International Journal of Computational Fluid Dynamics,
14(3):211-224,2001.

44. Schairer, E.T. and Hand, L.A., "Measurement of Unsteady Aeroelastic Model Deformation by Stereo
Photogrammetry," AIAA Paper 97-2217, June 1997.

45. Preidikman, S. and Mook, D.T., "Time Domain Simulations of Linear and Nonlinear Aeroelastic
Behavior," Journal of Vibration and Control, Vol. 6, No.8, pp. 1135-1175,2000.

46. Von Karman, T., "Encyklopadie der Mathematischen Wissenschaften," Vol. N, p. 349, 1910.

47. Hodges, D.H., and Dowell, E.H, "Nonlinear Equations of Motion for the Elastic Bending and Torsion of
Twisted Non-Uniform Rotor Blades," NASA TN D-7818, 1974.

48. Tran, C.T. and Petot, D., "Semi-Empirical Model for the Dynamic Stall of Airfoils in View to the
Application to the Calculation of Responses of a Helicopter Blade in Forward Flight," Vertica, Vol. 5, No.2,
1981, pp. 35-53.

49. Pati!, MJ., Hodges, D.H. and Cesnik, C.E.S, "Limit Cycle Oscillations in High-Aspect-Ratio Wings,"
Journal ofFluids and Structures, Vol. 15, No.1, 2001, pp. 107-132.
 111

50. Patil, MJ., Hodges, D.H. and Cesnik, C.E.S, "Nonlinear Aeroelastic Analysis of Complete Aircraft in
Subsonic Flow," Journal of Aircraft, Vol. 37, No.5, Sept-Oct 2000, pp. 753-760.

51. Patil, M.J. and Hodges, D.H., "On the Importance of Aerodynamic and Structural Geometrical
Nonlinearities on Aeroelastic Behavior of High-Aspect-Ratio Wings," AIAA Paper-2000-1448, 2000.

52. Tang, D.M. and Dowell, E.H., "Experimental and Theoretical Study on Flutter and Limit Cycle
Oscillations of High-Aspect Ratio Wings," AIAA Journal, Vol. 39, No.8, August 200 I, pp. 1430-1441.

53. Kim, K. and Strganac, T., "Aeroelastic Studies of a Cantilever Wing with Structural and Aerodynamic
Nonlinearities," AIAA Paper 2002-1412, 43'd A1AAIASME/ASCE/AHS/ACS Structures, Structural
Dynamics and Materials Conference, Denver, CO, April 2002.

54. Crespo da Silva, M.R.M. and Glynn, C.C., "Nonlinear Flexural-Torsional Dynamics of Inextensional
Beams-I: Equations of Motions," Journal of Structural Mechanics, Vol. 6, No.4, pp. 437-448,1978.

55. Yurkovich, R.N., Liu, D.D. and Chen, P.C., "The State-of-the-Art of Unsteady Aerodynamics for High
Performance Aircraft," AIAA Paper 2001-0428, January 2001.

56. Dowell, E.H. and Hall, K.C., "Modeling of Fluid-Structure Interaction," Annual Review of Fluid
Mechanics, 33:445-90, 2001.

57. Bennett, R.M. and Edwards, J.W., "An Overview of Recent Developments in Computational
Aeroelasticity," AIAA Paper No. 98-2421, presented at the AIAA Fluid Dynamics Conference, Albuquerque,
NM, June 1998.

58. Beran, P. and Silva, W., "Reduced-Order Modeling: New Approaches for Computational Physics," AIAA
Paper 2001-0853, 39th Aerospace Sciences Meeting and Exhibit, Reno, NV, January 2001.

59. Kim, T. and Bussoletti, J.E., "An Optimal Reduced Order Aeroelastic Modeling Based on a Response-
Based Modal Analysis of Unsteady CFD Models," AIAA Paper 2001-1525, 42 nd
A1AAIASMEIASCEIAHSIASC Structures, Structural Dynamics and Materials Conference & Exhibit,
Seattle, WA, April 2001.

60. Silva, W.A., "Application of Nonlinear Systems Theory to Transonic Unsteady Aerodynamic
Responses," Journal of Aircraft, Vol. 30, No.5, pp. 660-668, 1993.

61. Silva, W.A., "Extension of a Nonlinear System Theory to General-Frequency Unsteady Transonic
Aerodynamic Responses," 34th AIAA Structures, Structural Dynamics, and Materials Conference, Reston,
VA, pp. 2490-2503, 1993.

62. Silva, W.A., "Extension of a Nonlinear Systems Theory to Transonic Unsteady Aerodynamic Responses,"
AIAA Paper 93-1590, April 1993.

63. Silva, W.A., "Discrete-Time Linear and Nonlinear Aerodynamic Impulse Responses for Efficient (CFD)
Analyses," PhD Thesis, College of William Mary, Williamsburg, VA, October 1997.

64. Silva, W.A., "Identification of Linear and Nonlinear Aerodynamic Impulse Response Using Digital Filter
Techniques," AIAA Atmospheric Flight Mechanics Conference, Reston, VA, pp. 584-597, 1997.
112
65. Silva, W.A., "Reduced-Order Models Based on Linear and Nonlinear Aerodynamic Impulse Response,"
International Forum on Aeroelasticity and Structural Dynamics, NASA Langley Research Center, Hampton,
VA, pp. 369-379, June 1999.

66. Raveh, D., Levy, Y. and Karpel, M., "Aircraft Aeroelastic Analysis and Design Using CFD-Based
Unsteady Loads," AIAA Paper 2000-1325, 41 st AIAAIASMEIASCEIAHSIASC Structures, Structural
Dynamics, and Materials Conference and Exhibit, Atlanta, GA, April 2000.

67. Raveh, D.E., "Reduced-Order Models for Nonlinear Unsteady Aerodynamics," AIAA Journal, Vol. 39,
No.8, pp. 1417-1429, August 2001.

68. Farhat, C., Geuzaine, P., Brown, G. and Harris, c., "Nonlinear Flutter Analysis of an F-16 in Stabilized,
Accelerated, and Increased Angle of Attack Flight Conditions," AIAA Paper 2002-1490, April 2002.

69. Farhat, C., Harris, C. and Rixen, D., "Expanding a Flutter Envelope Using Accelerated Flight Data:
Application to An F-16 Fighter Configuration," AIAA Paper 2000-1702, April 2000.

70. Kholodar, D.B., Thomas, J.P., Dowell, E.H. and Hall, K.c., "A Parametric Study of Transonic Airfoil
Flutter and Limit Cycle Oscillation Behavior," AIAA Paper 2002-1211, presented at the
AIAAlASME/ASCE/AHS SDM Conference, Denver, CO, April 2002.

71. Knipfer, A. and Schewe, G., "Investigations of and Oscillation Supercritical 2-D Wing Section in a
Transonic Flow," AIAA Paper No. 99-0653, 36th Aerospace Sciences Meeting and Exhibit, January 1999.

72. Schewe, G. and Deyhle, H., "Experiments on Transonic Flutter of a Two-Dimensional Supercritical Wing
with Emphasis on the Nonlinear Effects," Proceedings of the Royal Aeronautical Society Conference on
Unsteady Aerodynamics, London, u.K., July 17-18, 1996.

73. Schewe, G., Knipfer, A. and Henke, H., "Experimentelle und numerisch Untersuchung zum transonischen
Fliigelflattern im Hinblick aufnichtlineare Effecte," unpublished manuscript, February 1999.

74. Schewe, G., Knipfer, A., Mai, H. and Dietz, G., "Experimental and Numerical Investigation of Nonlinear
Effects in Transonic Flutter," English Version (Translated by Dr. W.F. King III), German Aerospace Center
DLR Final Report Number DLR IB 232-2002 J 01, Corresponds to Final Report for BMBF: Nichtlineare
Effekte beim transsonischen Flattern (FKZ 13 N 7172), and Internal Report DLR IB 2001 J03, January 25,
2002.

75. Thomas, J.P., Dowell, E.H. and Hall, K.C., "Modeling Viscous Transonic Limit Cycle Oscillation
Behavior Using a Harmonic Balance Approach," AIAA Paper 2002-1414, presented at 43,d
AIAAlASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference and Exhibit,
Denver, CO, April 22-25, 2002.

76. Weber, S., Jones, K.D., Ekaterinaris, J.A. and Platzer, M .F., "Transonic Flutter Computations for a 2-D
Supercritical Wing," AIAA Paper 99-0798, 36 th Aerospace Sciences Meeting and Exhibit, Reno, NY, January
1999.

77. Tang, L., Bartels, R.E., Chen, P.C. and Liu, D.D., "Simulation of Transonic Limit Cycle Oscillations
Using a CFD Time-Marching Method," AIAA Paper 2001-1290, 42 nd AIAAlASME/ASCE/AHS/ASC
Structures, Structural Dynamics and Materials Conference, Seattle. WA, April 2001.
 113
78. Castro, B. M., Ekaterinaris, J. A., and Platzer, M. F., "Navier-Stokes Analysis of Wind-Tunnel
Interference on Transonic Airfoil Flutter," AIAA Journal, Vol. 40, No.7, pp. 1269-1276, July 2002.

79. Thomas, J.P., Dowell, E.H. and Hall, KC., "Modeling Limit Cycle Oscillations for an NLR 7301 Airfoil
Aeroelastic Configuration Including Correlation with Experiment," AIAA Paper 2003-1429, presented at 44th
AIAAlASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference and Exhibit,
April 7-10, 2003.

80. Thomas, J.P., Dowell, E.H. and Hall, K.C ., "A Harmonic Balance Approach for Modeling Three-
Dimensional Nonlinear Unsteady Aerodynamics and Aeroelasticity," IMECE-2002-32532, Presented at the
ASME International Mechanical Engineering Conference and Exposition, November 2002, New Orleans,
Louisiana.

81. Edwards, J.W., "Calculated Viscous and Scale Effects on Transonic Aeroelasticity," AGARD-R-822,
Numerical Unsteady Aerodynamic and Aeroelastic Simulation, pp. I -I - -I,
I March 1998.

82. Thomas, J.P., Hall, KC. and Dowell, E.H., "A Harmonic Balance Approach for Modeling Nonlinear
Aeroelastic Behavior of Wings in Transonic Viscous Flow," AIAA Paper 2003-1984, Presented at 44th
AIAAlASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference and Exhibit,
April 7-10, 2003.

83. Rivera, J.A., et aI., ''NACA 0012 Benchmark Model Experimental Flutter Results with Unsteady Pressure
Distributions," NASA TM 107581 , March 1992.

84. Kholodar, D.B., Dowell, E.H., Thomas, J.P. and Hall, KC., "Improved Understanding of Transonic
Flutter: A Three Parameter Flutter Surface," Submitted for publication to the Journal of Aircraft, 2003.

85. Bendiksen, 0.0., "Improved Similarity Rules for Transonic Flutter," AIAA Paper 99-1350, 40th
AIAAlASME/ASCE/AHS SDM Conference, St. Louis, MO, April 1999.
 FLAPPING MEMBRANES FOR THRUST
PRODUCTION

J.M. JIMENEZ, J.H.J BUCHHOLZ, A.E. STAPLES, J.J. ALLEN,

AND A.J. SMITS

Abstract. The wake of a flexible flapping membrane has been investigated using flow visualization and
particle image velocimetry. The aspect ratio of the flexible membrane was 0.27. The Strouhal number (St)
was varied from 0.44 to 0.16, corresponding to Reynolds numbers between 2.2x 105 and 4.1 xl 05 Under these
conditions, a wake structure was observed that resembles the 2P structure characteristic of flapping foils but
containing at least four vortex pairs per cycle. For St ~ 0.35, vorticity contour plots suggest that each multiple
vortex pair structure has approximately zero net circulation; whereas for lower St, vorticity of one sign is
diminished such that the wake begins to resemble the classical von Karman vortex street. A mean wake
velocity profile measured at the mid-span of the membrane suggests a net drag for all flow conditions
considered, with the minimum drag occurring for St = 0.35. However, preliminary flow visualization
experiments with an increased aspect ratio of 0.45 show a simpler 2S-type structure resembling a reverse von
Karman vortex street for which greater efficiency in thrust production is expected.

1. INTRODUCTION
It has been known for some time (for example, Gray (1936), Triantafyllou and
Triantafyllou (1995)) that the agility of flying and swimming species far exceeds that of
man-made vehicles. To understand the physical mechanisms involved in such
locomotion, researchers have investigated a variety of unsteady devices and analytical
models for the production of thrust and lift (reviews are presented in Triantafyllou et al.
(2000) and Wu (2001)). The present study is an experimental investigation of the thrust
generation and flow field characteristics of a quasi two-dimensional flapping flexible
membrane, fixed at its leading edge. Preliminary experiments have been conducted
using flow visualization and digital particle image velocimetry (DPIV). In this study,
we wish to identify the relevant parameters governing the production of thrust, and
explore the physical mechanisms responsible for optimal operation.

Triantafyllou et al. (1991 and 1993) identified the Strouhal number as the
primary parameter governing the efficiency of thrust production in various oscillating
rigid airfoil experiments, and in a survey of a variety of fish and cetaceans. The
optimal Strouhal number was found to predominantly lie between 0.25 and 0.35, and is
defmed as:

115
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 115-124.
2003 Kluwer Academic Publishers.
116

St= fA (1)
v'
where f is the frequency of oscillation, A is the full width of the wake, often
approximated by the maximum excursion of the trailing edge (peak to peak), and V is
the average swimming speed or relative free-stream velocity. However, this does not
appear to be universal. In a computational study of a swimming tadpole, a geometry
more representative of the present experiments but at a much lower Reynolds number,
Liu et al. (1999) report maximum efficiency at a Strouhal number of 0.72. This is
attributed to the larger excursion of the tadpole's tail; perhaps Reynolds number is also
a factor. Koochesfahani (1987) proposes a similar parameter, the reduced frequency, as
an important parameter governing the structure of the wake generated by a flapping
foil, defmed as:

(2)

where the wake width A has been replaced by the chord length c of the foil. However,
wake structure was also noted to depend on flapping amplitude. Ohmi et al. (1990,
1991), in a similar experiment Koochesfahani (1987), report wake structure to be
dependent upon the product of reduced frequency and flapping amplitude.

Williamson and Roshko (1988) observed the wakes of cylinders freely

vibrating in a cross-flow and identified a number of wake patterns within the parameter
space studied. Most relevant to the present work are wake patterns, which they labeled
'2S' and '2P'. In the first case, two single vortices are shed per cycle, forming a von
Karman vortex street. In the second case, two counter-rotating vortex pairs are shed per
cycle. Similar structures, with the signs of vorticity reversed, have been observed in the
wakes of flapping foils during thrust production (ie. Koochesfahani (1987. We will
therefore use the 2S and 2P naming convention to describe wakes produced in the
present experiment.

2. EXPERIMENT

The present experiment consists of a thin flexible plastic membrane extending

downstream from the trailing edge of a fairing. The membrane has a chord C of 24.5
cm and aspect ratio (span/chord) of 0.27. The fairing is a 20% thick symmetric airfoil
with a chord of 5 cm. A schematic of the experimental setup is shown in Figure 1. A
 117

streamwise traveling wave is approximated on the membrane by periodically imposing

two buckling modes through the displacement of four cables attached to the membrane.
The cables serve as actuators. The two modes are separated by a 90 phase difference
and are depicted separately in Figures 2 and 3. The water channel test section is 0.67 m
wide and 0.33 m deep with a 3.5:1 contraction ratio and flow conditioning consisting of
honeycomb and 6 screens.

Data was acquired using DPIV and flow visualization. The flow was seeded with silver
coated hollow spheres with mean particle diameter of 14 microns and true density of
1.7 glee, and visualized with a 120 mJ/pulse New Wave double-pulse Nd-YAG laser
coupled with a 2048x2048x12-bit resolution pixel resolution Megaplus ES 4.0 CCD
camera. Images were captured using the "frame straddling" technique as fIrst
introduced by Wernet (1991), recorded to RAM, and interrogated using cross-
correlations to produce two-dimensional velocity vector fIelds of the 1.24C x 1.24C
interrogation area. The flow was interrogated using a horizontal laser sheet at the mid-
span level. The latter portion of the membrane is visible in the images, beginning at
approximately 0.8C.
ceo ","Tlenl --

,,,rfoil

Figure 1. Experimental setup.

Figure 2. Flexible membrane actuated by first buckling mode

118

Figure 3. Flexible membrane actuated by second buckling mode.

Hydrodynamic loading on the membrane significantly affected tail amplitude

(and perhaps other aspects of the membrane motion) under variation of forcing
frequency and therefore the Strouhal number was adjusted by changing only the free
stream velocity (to which amplitude was found to be much less sensitive). The
Strouhal numbers studied using PIV were 0.44, 0.35, 0.29, 0.22, and 0.16,
corresponding to Reynolds numbers of2.2 x 105,2.5 X 105,3.1 X 105,4.8 X 105 , and 4.9
x 105. The membrane was forced at a frequency of 0.56 Hz. Nineteen image pairs
were acquired for each Strouhal number at an approximate rate of 6 pairs per second.
The image sequences were used to study the evolution of wake structures and to obtain
a mean wake-velocity profile at the mid-span level.

3. RESULTS AND DISCUSSION

Figures 4, 5, and 7 show velocity vectors and vorticity contours for Strouhal numbers of
0.44,0.35, and 0.22, respectively. Each image has undergone a Galilean transformation
equal to the mean velocity in order to reveal rotation in the vectors that describe the
flow field. Spanwise vorticity (J)z has been nondimensionalized by the free-stream
velocity and membrane chord: (J)z *= (J)z C/U. Qualitatively, the wake structures for all
Strouhal numbers are similar and resemble a reverse 2P structure except that multiple
pairs are produced in each membrane oscillatory cycle. Figure 6 shows a magnified
portion of Figure 5 with the major regions of vorticity indicated. In this case, each half-
cycle structure contains two pairs, which is the pattern most often observed in the
present experiments. In the plane visualized, the net effect of each such structure is to
produce a jet slightly inclined in the downstream direction, implying an inefficient but
existent mechanism of thrust production. Efficiency is further reduced by the
complexity of these structures; some smaller-scale sub-structures induce velocities with
components in the upstream direction.
 119

Latter portion of
flapping membrane
0.5

o
Z*
25.00
21.43
17.86
14.29
10.71
7.14
3.57
0.00
-3.57
-7.14
-10.71
-0.5 -14.29
0.13 U -17.86
-21.43
-25.00
0.5
xlC
Figure 4. Instantaneous velocity field of near wake region of the flappingflexible membrane
with non-dimensionalized spanwise vorticity contour-plot background for aspect ratio sic =
0.27, and St = 0.44. A constant velocity ofUe = 0.87U is subtracted from the velocity in order
to bring out the vortices more clearly (see Zhou et al. 1997).
120

Latter portion of
flapping membrane

Z*

0.11 U
 121

z*
122

Latter portion of
flapping membrane

Z*

0.1 U

U
123
124

6. REFERENCES
Gray, J. (1936). Studies of animal locomotion. J. expo BioI. 16,9-17.
Koochesfahani, M.M. (1987). Vortical patterns in the wake of an oscillating airfoil. AIAA Paper 87-0111.
Liu, H. and Kawachi, K. (1999). A numerical study of undulatory swimming. 1. Compo Phys. 155, 223-247.
Ohmi, K., Coutanceau, M., Loc, T.P., and Dulieu, A. (1990). Vortex formation around an oscillating and
translating airfoil at large incidences. J. Fluid Mech. 211,37-60.
Ohmi, K., Coutanceau, M., Daube, 0., and Loc, T.P. (1991). Further experiments on vortex formation
around an oscillating and translating airfoil at large incidences. J. Fluid Mech. 225,607-30.
Triantafyllou, M.S., Triantafyllou, G.S. and Gopalkrishnan, R. (1991). Wake mechanics for thrust generation
in oscillating foils. Phys. Fluids A. 3, 2835-37.
Triantafyllou, G.S., Triantafyllou, M.S. and Grosenbaugh, M.A.. (1993). Optimal thrust development in
oscillating foils with application to fish propulsion. J. Fluids Struct. 7,205-24.
Triantafyllou, M.S. and Triantafyllou, G.S. (1995). An efficient swimming machine. Sci. Am. 272,64-70.
Triantafyllou, M.S., Triantafyllou, G.S. and Vue, D.K.P. (2000). Hydrodynamics of fishlike swimming.
Annu. Rev. Fluid Mech. 32, 33-53.
Wernet M P (1991). Particle displacement tracking applied to air flows. Fourth Int Conf on Laser
Anemometry, Cleveland - OH, USA, 327-335
Williamson, C.H.K., and Roshko, A. (1988). Vortex formation in the wake of an oscillating cylinder. J.
Fluid. Struct. 2, 355-381.
Wu, T.Y. (2001). On theoretical modeling of aquatic and aerial animal locomotion. Adv. Appl. Mech. 38,
291-353.
Zhou, J., Meinhart, C. D., Adrian, R. J. & Balachandar, S. (1997). Formation of hairpin pockets in wall
turbulence. In Self Sustaining Mechanisms of Wall Turbulence (Computational Mechanics Publications,
Southampton, UK. ed. R. L. Panton), 109-134,

7. AFFILIATION

J.M Jimenez, J.H.J Buchholz, A.E. Staples, J.J. Allen, and A.J. Smits
Princeton University, Department ofMechanical & Aerospace Engineering
Engineering Quadrangle,Olden Street, Princeton, NJ 08540 USA,

Correspondence to:
Juan. M. Jimenez
Princeton University, Department of Mechanical & Aerospace Engineering
Engineering Quadrangle,Olden Street, Princeton, NJ 08540 USA
E-mail: jjimenez@princeton.edu
 EXPERIMENTAL STUDY OF A TETHERED
CYLINDER IN A FREE STREAM

J. CARBERRY*, K. RYAN AND J. SHERIDAN

Department ofMechanical Engineering, Monash University, Melbourne,
Victoria 3800, Australia.
email: josie.carberry@eng.monash.edu.au

Abstract: This paper describes results of an experimental investigation into the motion of a tethered
cylinder. Both the cylinder and the tether are rigid, restricting the cylinder motion to a one-dimensional
arc. The response of the cylinder as the free stream flow is gradually increased is described in terms of
the mean tether angle and the amplitude of the cylinder oscillations. Two distinct states are observed,
where the transition between the states is characterized by an abrupt jump in both the amplitude of
oscillation and the mean tether angle.

Key words: Tethered, cylinder, experimental.

1. INTRODUCTION

The instabilities generated by fluid flow over a body can result in large scale
motion of the body. A number of previous investigations have considered the case
of a cylinder constrained to move transverse to the flow, e.g. Govardhan and
Williamson (2000), or a cylinder with two-dimensional motion in the transverse
and in-line directions, e.g. lauvtis & Williamson (2002). A closely related problem
is that of a tethered rigid cylinder, depicted schematically in Figure 1, where the
cylinder is free to move in an arc about the tether point. Despite the relevance of
this case to the response of tethered bodies submerged in a steady current, this
problem received almost no attention until the recent numerical investigations of
Ryan et al. (2002, 2003) and Pregnalato (2002). In this experimental investigation
the response of a tethered cylinder to a range of flow velocities are analyzed and
compared to the numerical results.

125
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 125-133.
2003 Kluwer Academic Publishers.
126

Ufree stream

..
~
Motion

Figure 1: Schematic of the tethered cylinder in the free stream.

The fluid-structure interaction of a cylinder moving relative to a free-stream results

in significant modification of the forces on the cylinder and the structure of the near
wake. Typically, the cylinder motion is either "free", where the motion occurs in
response to flow-induced forces, or the motion of the cylinder is externally driven.
Until recently the majority of investigations have focused on cylinder motions that
are either transverse or in-line with the free-stream. For a tethered body these two
geometries are limiting cases that are approached for extreme experimental
parameters. One of the interesting features of a tethered body is that the geometry
of the cylinder's motion relative to the free-stream changes with the mean tether
angle.
An elastically mounted cylinder which is constrained to move transversely to
the flow exhibits 2 or 3 different response branches as the free-stream velocity is
varied, Govardhan & Williamson (2000); where the number of branches depends
on the mass-damping of the system. Govardhan & Williamson (2002) also
investigated a lightly damped non-elastically mounted system by removing the
springs, which had previously provided the mechanical restoring force (i.e. the case
where k = 0). The response of the system was found to depend primarily on the
mass ratio m', where m' is the mass of the system normalized by the mass of fluid
displaced by the body. Without the restoring force large oscillations were only
observed for m' less than a critical value, m 'cRIr = 0.54. Essentially the cylinder
displayed two states: for m' > m 'cRIr very small or zero amplitude oscillations were
observed, while for m' < m 'CRIT the cylinder oscillated with large amplitude.
The study of Ongoren & Rockwell (1988) found that a cylinder undergoing
forced oscillations at a range of inclination angles exhibits a number of different
shedding modes. Transverse oscillations (90 inclination) resulted in stable anti-
symmetric wakes, however as the inclination angle was decreased towards 0, or in-
 127

line oscillations, symmetric wake modes were also observed. Interestingly, the
wakes exhibited continual switching between symmetric and anti-symmetric
modes, with the anti-symmetric modes becoming less dominant as the inclination
angle decreased. This high level of mode competition indicates that the forced
inclined oscillations did not produce a single stable wake state.
The majority of previous work on tethered bodies has focused on spheres or
buoys, typically involving interaction with a free-surface. However, in the absence
of a free-surface and surface waves, Govardhan & Williamson (1997a & b) found
that a fully submerged sphere exhibits large-scale oscillations over a wide range of
reduced velocities. Interestingly, at higher reduced velocities the response of the
sphere is not locked-on to the natural structural frequency and is significantly less
then the corresponding Strouhal frequency of a stationary sphere.
To date, the case of a tethered cylinder has received relatively little attention,
with the exception of the two-dimensional numerical investigations of Ryan et at.
(2002, 2003a & b) and Pregnalato et al. (2002). At a mass ratio of m' = 0.833 and
a constant Reynolds number of 200, Ryan et al. (2002, 2003a) observed that the
mean tether angle increases smoothly with increasing reduced free-stream velocity.
The continuous increase in the tether angle was linked to the drag force, which
remained constant and very similar to that of a stationary cylinder, except at high
reduced velocities where there was a slight decrease in the drag coefficient. At low
reduced velocities when the tether was essentially vertical extremely small cylinder
oscillations were observed. However, larger oscillation amplitudes were not
observed until higher reduced velocities, when the tether angle results in motion
that has significant in-line and transverse components.

2. EXPERIMENTAL METHOD

The cylinder, 30 mm in diameter and 594 mm long, is tethered using two 151.5 mm
carbon fiber rods. Both the cylinder and the tethers are rigid and there is no relative
motion between them. The tether is located at the pivot point with precision
bearings such that the tethered cylinder is free to rotate about the pivot point as
shown in Figure 1. The mass ratio of the cylinder was m' = 0.79, resulting in a
positive upwards buoyancy force. The flow velocity, U, is increased from zero to
0.46 ms'\ giving a maximum Reynolds number of 13,700. The reduced velocity is
given by u* = UlIND, where D is the diameter of the cylinder, U is the free-stream
velocity and INis the natural frequency ofthe tethered body in water.
The motion of the cylinder is one-dimensional and is described in terms of the
angle of the tether from the vertical axis, as shown in Figure 2. A 25Hz PAL video
128

camera was used to track the cylinder and the system is calibrated to reduce optical
distortion. For each free-stream velocity 16,384 data points were acquired and the
mean tether angle, Smean, and the normalized oscillation amplitude, S"sld, were
calculated. The standard deviation of the cylinder'S motion is normalized by the
angle subtended by the cylinder diameter, SD.

D
Ufree stream
~

Figure 2: Schematic showing important geometric parameters

By assuming a net zero lift force both the mean tension in the tether and the
mean drag force can be calculated from Smean and the buoyancy force. The
natural structural frequency of the system,IN, is analogous to that of a pendulum
and depends on the tension in the tether. Thus, IN varies with the mean tether
angle :

1 (l-m")g
IN = I Nwater = 2Jr ""
(m + CA)L D COS Bmean
(1)
 129

where L' = LID is the nonnalised tether length and CA (= 1) is the idealized added
mass coefficient. Parallel numerical work described in Ryan et al. (2003b)
indicates that the mean lift force is essentially zero except for cases where large
cylinder oscillations are observed. For large cylinder oscillations, S'std > 0.35, it is
anticipated that there is a mean negative lift force on the cylinder. In these cases
our calculations will overestimate both the tension in the tether and the natural
structural frequency of the system.

3. RESULTS AND DISCUSSION

a)
90r-----~----~----~----~--__.

75

60

emean45 CI
Q
CI
r:P

30
o
o "
,,0
15 "
c
oL-__~c~
C "
____~__--~----~--~
0.0 0.1 0.2 0.3 0.4 0.5
U (m/s)
b)
90,-----------------------------,

75

60

45 c
e
CI
Omean c
o

u*
Figure 3,' Variation of mean tether angle 8mean with a) free-stream velocity and b)
reduced velocity
130

As the free-stream velocity is increased from rest the mean position of the
cylinder changes as shown in Figure 3a. emean increases smoothly with the
exception of a small discontinuity at U = 0.20 mls and a larger discontinuity at U =
0.34 mls. Examination of the oscillation amplitude in Figure 4a shows that the
discontinuities in emean correspond to changes in the motion of the cylinder. At low
a) 0.6

0.5
"be
DO
0.4 CD

e"std o . 3 [J

0.2

0.1 cD
OC
cD ce
0.0
0.0 0.1 0.2 0.3 0.4 0.5
U (m/s)
b) 0.6

0.5
eJ
EI
0.4 ICl

e"stdo.3 [J

0.2

0.1
D ClCc
rP'
c
0.0
0 5 10 15 20 25

U*

Figure 4: Variation of oscillation amplitude (lsld with a) free-stream velocity and b) reduced
velocity

freestream velocities, U < 0.20 mis, the cylinder does not oscillate and esld is zero.
As U is increased through 0.20 mls small periodic cylinder oscillations are
observed and there is a small discontinuity in emean . As the free-stream velocity is
increased further the oscillation amplitude increases slowly but esld remains less
than 0.1.
Between U = 0.33 and 0.35 mls there is a large jump in the oscillation
amplitude and a corresponding change in 8mean The jump in esld appears to
correspond to a transition between two different states: a small amplitude state,
 131

e
which occurs at lower V and corresponds to smaller values of mean , and a large
amplitude state, which occurs at larger emean where the motion of the cylinder is
predominantly transverse to the free-stream.
At V = 0.34 mls the majority of the cylinder oscillations were small in
amplitude. However, there were two distinct time periods where large amplitude
oscillations occurred and during these times there was a corresponding increase in
emean . The switching between states results in the "intermediate" value of esld
shown in Figure 4.
The variation of emean and e sld are plotted against reduced velocity in Figure 3b
and Figure 4b respectively. Historically, u* is the most commonly used
independent variable against which results of flow-induced motion experiments are
plotted. However, for the tethered body u* depends on the natural frequency
e
which, as described in equation (1), is a function of mean . The abrupt increase in
e mean as the cylinder moves from small to large amplitude oscillations causes V to
decrease. Moreover, as V increases the value of V for the large amplitude
oscillations remains almost constant. In fact it can be shown that for a finite drag
force there is a maximum possible value of V , V max , that depends on the physical
properties of the cylinder (m and L) and the drag.
a) 61

49

e47
45

6 10 15 20 26
time (5)
b) 80

75

e 70
85
80

5 10 15 20 25
time (5)

Figure 5 Sample time traces/or a) the small amplitude state (U = 0.31 mls) and b) the large
amplitude state (U = 0.38 mls).

Sample time traces for the small and large amplitude states are shown in Figure
5(a & b) respectively. In both states the motion of the cylinder is periodic in nature
but there is significant modulation of the motion. A beating type of motion is
132

particularly evident for the small amplitude state and a Fourier analysis for these
cases shows two distinct spectral peaks. The variation in the oscillation amplitude
also appears to correspond to changes in the "short term" mean position of the
cylinder. The time traces in Figure 5 show that for both states a decrease in
e
oscillation amplitude corresponds to a decrease in mean This observation further
emphasises the relationship between emean and e"sid which was evident in the
properties of the small and large amplitude states.
The variation in e"sid and emean shown in Figure 5 indicates that we should
expect significant time variation in the structure of the near wake. This is
consistent with the observations of Ongoren & Rockwell (1988) who observed
mode competition for a cylinder undergoing inclined forced oscillations.
Two dimensional numerical experiments at low Reynolds numbers by Ryan et
at. (2003a) show that for similar cylinder properties (m" = 0.833 & L" = 5.05) the
mean tether angle increases smoothly with U, and the transition to the large
amplitude oscillations is not observed. However, as m" is reduced below 0.3 Ryan
et at. (2003b) observed the transition to the high amplitude state, as characterized
by ajump in both e"sidand emean ,

4. CONCLUSION

This investigation considers the motion of a rigid tethered cylinder. Two distinct
e
states were observed: the first state occurs at lower flow velocities where mean is
small and is characterized by small oscillation. The second state is characterized by
higher values of both e"sid and emean and occurs at higher flow velocities. Self-
excited transitions between the two states are observed over a small range of flows.
e
The point mean = 45 defines the boundary where the cylinder motion switches
between predominantly in-line and predominantly transverse. The transition
e
between the small and large amplitude states occurs at a value of mean which is
e
slightly greater than 45. The exact value of mean at which the transition between
the two states occurs may vary with m", L" and Re. However, the point at which
e mean passes through 45 represents a change in the relative geometry of the motion
with respect to the flow, and is therefore expected to be an important factor in
determining when the transition to large oscillation amplitudes occurs.
 133

5. ACKNOWLEDGEMENTS

The authors would like to acknowledge support from the Australian Research
Council. The authors are also indebted to Professor Kerry Hourigan and Associate
Professor Mark Thompson for their insights, provided in many productive
discussions during the course of this research. The second author would like to
acknowledge support provided through a Monash Departmental Scholarship.

6. REFERENCES

JAUVTIS, N. & WILLIAMSON, W. H .K. 2002 Vortex-induced vibration ofa cylinder in two degrees
of freedom, Conference on Bluff Body Wakes and Vortex-Induced Vibrations, Port Douglas, Australia,
53-56.
GOVARDHAN, R & WILLIAMSON, e.H.K. 1997a Dynamics and forcing of a tethered sphere in a
fluid flow, Journal ofFluids and Structures., 11,293-305.
GOVARDHAN, R & WILLIAMSON, e.H.K. 1997b Vortex-induced motions of a tethered sphere,
Journal of Wind Engineering and Industrial Aerodynamics., 69-71, 375-385.
GOVARDHAN, R & WILLIAMSON, e.H.K. 2000 Modes of vortex formation and frequency response
ofa freely vibrating cylinder, J. Fluid Mech., 420, 85-130.
GOV ARDHAN, R& WILLIAMSON, e.H.K. 2002 Resonance forever: existence of a critical mass
and an infinite regime of resonance in vortex-induced vibration, J. Fluid Mech., 473, 147-166.
ON GOREN, A. & ROCKWELL, D., 1988 Flow structure from an oscillating cylinder Part 2. Mode
competition in the near wake, J. Fluid Mech., 191, 225-245.
PREGNALATO, e.J., RYAN, K., M.e. & HOURIGAN, K., 2002 Numerical simulations of the flow-
induced vibrations of tethered bluff bodies. 5'h Int. Symp. FSI, AE, FIV and N. November 17-22, New
Orleans L.A., USA.
RYAN, K, THOMPSON, M.e. & HOURIGAN, K, 2002 Energy transfer in a vortex induced vibrating
tethered cylinder system, Conference on Bluff Body Wakes and Vortex-Induced Vibrations, Port
Douglas, Australia, 57-61.
RYAN, K, PREGNALATO, e.J., THOMPSON, M.e. & HOURIGAN, K. 2003a Flow-Induced
Vibrations of a Tethered Circular Cylinder, Journal of Fluids and Structures, submitted.
RYAN, K., THOMPSON, M.e. & HOURIGAN, K. 2003b Flow-Induced Vibrations of a Tethered
Circular Cylinder, IUTAM Symposium on Integrated Modelling of Fully Coupled Fluid-Structure
Interactions Using Analysis, Computations, and Experiments, Rutgers, NJ., June-6 June 2003.
 THE EFFECT OF CHANGED MASS
RATIO ON THE MOTION OF A
TETHERED CYLINDER

K. RYAN, M. C. THOMPSON AND K. HOURIGAN

Fluids Laboratory for Aeronautical and Industrial Research (FLAIR) ,
Department of Mechanical Engineering,
Monash University, Melbourne, Victoria 3800, Australia

Abstract: The flow past a buoyant tethered cylinder was investigated for a variety
of mass ratios. A critical mass ratio, m* = 0.2 - 0.3 was found for
the uniform flow past a tethered cylinder, below which sustained large
amplitude oscillations were observed up to the highest reduced velocity
simulated in this study. The critical mass ratio was found to coin-
cide closely with that found in previous studies for a hydro-elastically
mounted cylinder using two-dimensional simulations.

1. INTRODUCTION
To date, few studies on the uniform flow past a tethered body ex-
ist, most of these concentrating on the flow past a tethered sphere.
This is despite tethered cylinders having practical applications in sub-
sea pipelines, tethered lighter-than-air-craft, and tethered spars just to
name a few examples.
A significant body of research exists in the related field of freely os-
cillating cylinders, both with high- and low- mass damping. For both
cases, several parameter studies have been performed, including studies
regarding the effect of mass ratio on the modes of oscillation observed.
In particular, Govardhan and Williamson, 2000 observed that for low
mass damped hydro-elastically mounted cylinders three modes of oscil-
lation exist, namely the initial, upper and lower branch, and a critical
mass ratio exists below which high amplitude oscillations continue up
to an indefinite reduced velocity. Recent studies by Govardhan and
Williamson, 2003 have confirmed that this phenomenon occurs up to an
infinite reduced velocity.

135
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 135-144.
2003 Kluwer Academic Publishers.
136

Three phenomena distinguish the tethered cylinder from the hydro-

elastically mounted cylinder. The first is that the tethered cylinder has
a component of motion in both the in-line and transverse directions, as
such variations in both the drag and lift forces directly affect the cylinder
motion.
Secondly, the natural frequency of the cylinder system (expressed in
non-dimensional form as the reduced velocity) is now a function of the
hydro-dynamic loading acting on the cylinder, and varies with lift and
drag throughout the oscillation cycle through the equation:

U*=~= (1)
in D

Where CD is the drag coefficient, CL is the lift coefficient, m* is the

mass ratio (defined as the ratio of body density to fluid density), Fr is
the Froude number, L* is the tether length normalized by the cylinder
diameter, and C A is the added mass coefficient, equal to unity for a
circular cylinder.
At very high Froude numbers (corresponding experimentally to high
velocities) assuming a fixed mean drag and lift, the dependence on the
fluid forces and Froude number impose an upper limit on the possible
maximum value for the reduced velocity. From equation 1, in order to
exceed this maximum value the absolute value of either the mean drag
and/ or the mean lift must decrease.
The third distinguishing feature is that there is no damping (it was
assumed that the tether attachment point was frictionless) with the di-
rect result that, assuming the forcing and resultant cylinder motion are
well approximated by a sinusoidal function, there can be no 'upper'
branch as described by Govardhan and Williamson, 2000. The phase
angle between the total force and the cylinder motion, and the phase
angle between the vortex force and the cylinder motion, must be 180 0
for significant oscillation amplitudes.
There are two parameters of importance as the flow conditions are
varied, namely the mean layover angle, e, and the cylinder oscillations
about this mean layover angle, e;td'
To the authors knowledge, only the two-dimensional studies of Preg-
nalato et al., 2002, Ryan et al., 2002 and Ryan et al., 2003 have reported
research on the flow past a tethered cylinder. In these related paper-
s, only one cylinder mass ratio and tether length ratio were studied
(m* = 0.833, l* = 5.05). However at large layover angles, e (where the
tethered cylinder experiences dominantly transverse oscillations), oscil-
lations similar to that of a freely oscillating cylinder were reported. From
 137

this work it may be assumed that a tethered cylinder may have a critical
mass ratio similar to that found for a freely oscillating cylinder.
In this paper two-dimensional simulations of a tethered cylinder with
various mass ratios (m* = 0.1 - 0.8) and a tether length, l* = 5, are
reported (see Figure 1a.) . The reduced velocity was chosen as the con-
trolling flow parameter. All simulations were performed at a Reynolds
number, Re of 200. Simulations were carried out with a highly accurate
spectral element method coupled with a predictor corrector technique,
details of which may be found in Pregnalato et al., 2002.

2. NUMERICAL METHOD

(a) (b)

Figure 1. Schematic of the tethered cylinder system (a) , and the computational
grid used for the calculations (b) .

The fluid forces acting on the tethered cylinder may be decomposed

into the drag force acting in-line with the fluid flow, the lift force and
the buoyancy force, both acting transverse to the fluid flow, a restoring
tension force (T) in the tether is also present.
The problem is fully described in two dimensions by the coupled sys-
tem of the incompressible N avier-Stokes equations (equations 2a and 2b)
and the equations of motion describing the body acceleration in response
to calculated fluid forces (equations 3a and 3b). The two-dimensional
form of the the Navier Stokes equations is written as:

au
-at + (u \7) u 1 1 2
--\7p + -\7 u, (2a)
PI Re
\7. u = 0, (2b)
138

where u is the velocity field, p is the pressure field, Pi is the density of

the fluid, and Re is the flow Reynolds number.
The equations governing the cylinders motion (here presented in Carte-
sian coordinates) are:

x [2:* [([2 - x2) CD - (Y [CL + ~(1 - m*) F~2 ])]

X (3a)

Y ~ [([2 _ y2) (CL + ~(1 - m*)_l_) - YXCD] (3b)

t2m* 2 Fr2

where [ is the tether length and, = ;r;;.

Note that, is dimensional
and has units of acceleration.
The (two-dimensional) spectral-element mesh used in this study is
shown in Figure lb. It consists of K = 518 macro elements. A com-
prehensive resolution study was performed for a stationary cylinder at
Re = 500 (based on cylinder diameter), and also for a tethered (moving)
cylinder at Re = 200. For each study, the order of the interpolating poly-
nomials was increased from N = 5 to N = 9 to test for grid resolution.
The variation in shedding frequency, lift and drag between the values at
N = 7 and N = 9 are less than 1%. Furthermore, for the fixed cylinder,
the values of all measures for N = 8 (used in all simulations) compare
to within 1% of the numerical values of Blackburn and Henderson, 1999
and Henderson, 1995.

3. RESULTS
The results are divided into the mean results (mean layover angle
and drag coefficient) and the amplitude of oscillations about this mean
layover angle. Variations in the mean results indicate changes in the
wake structure and have been found to be directly linked to changes in
the oscillatory results.
Two oscillation modes have been identified for mass ratios m* < 0.3.
The switch between the two wake types is identified by a discontinuous
jump in the mean layover angle and the mean hydrodynamic forces act-
ing on the cylinder, and a discontinuous jump in the cylinder oscillation
amplitude.

3.1 Mean Layover Results

Figure 2 shows the mean layover angle ((}mn) as a function ofreduced
velocity for each of the mass ratios investigated. In general the rate
of change of mean layover angle (as a function of increasing reduced
velocity) varies inversely with the mass ratio. For mass ratios, m* = 0.3
 139

80
m* =0.1 41-
,
.&. m* =0.2 ,,
70 =0.3
.. m*
.'.'.'''\i'''.'.'.
m* =0.4
60 ---~--. m* =0.5
."."."<>".".,,. m* =0.6
~
50
.ffi!.
m*
m*
=0.7
=0.8
CD
40

30

0~~~~~~~~~1~0~~~~~15~~~~20

U*

Figure 2. Mean layover angle (}mn as a function of the reduced velocity U'

to 0.8, the variation in mean layover angle is a smooth function of the

reduced velocity. However, for m* = 0.1 and 0.2 a distinct jump in
the mean layover angle is observed in the range Bmn = 40 to 50 0 A
very small reduced velocity range (the transition range) is observed for
m* = 0.1 and 0.2 where one of two mean layover angles is possible for a
given value of U* .
The mean layover angle can be determined explicitly by equating the
mean fluid forces acting on the cylinder:

Bmn = Tan -1 [ CD
(1-m* )7r 1 (4)
CL + Fr2

As the Froude number varies smoothly with the reduced velocity (e-
quation 1), equation 4 shows that the discontinuous jump in the mean
140

layover angle must be associated with a discontinuous increase in the

mean drag and/or a discontinuous decrease in the mean lift.
The discontinuous change in the absolute value of the mean drag
and/or lift may account for the reduced velocity transition region (which
from equation 1 is a function of both of these parameters).

3.2 Mean Drag Results

2.5 ~--r.....,..-""-r--'----r-""-""'--r-~r---r--r~-""--r--'-""'I""'-=w
2.4
2.3 = m* 0.1
.&. = m* 0.2
2.2
.'.'.'",?'.'.'. = m* 0.3
2.1 """"."."."ti>'-'-'-"" m* = 0.4
2 ---<}---. m* =0.5
",,,,v,,,,,, m* =0.6
1.9 m*= 0.7
'?
liE 1.8 ofElo m* =0.8

UQ1.7 4.
I

1.6

1.5
1.4
1.3
1.2
1.1
10~~~~-5~~~~-1~O~~~~~~~~~

u*
Figure 3. Mean drag CD(mn) as a function of the reduced velocity U'

The mean drag is plotted as a function of the reduced velocity in figure

3. As anticipated, a significant increase in the mean drag is observed
for mass ratios, m* = 0.1 and 0.2 corresponding to the discontinuous
jump in the mean layover angle. For all the mass ratios investigated
the mean drag reached a maximum value prior to decreasing. In the
case of m* = 0.8 and 0.9 the mean drag is observed to decrease below
that for a fixed cylinder at this Reynolds number (CD < 1.34). This is
 141

a direct result of the limiting reduced velocity which may be predicted

from equation 1.

0.8 J!J. ~_

..
0.6
i
'W
~.., ~
*CD m* = 0.1
0.4
lID
~ 6 m* = 0.2
.,.,.,""',.,.,. m* = 0.3
m*= 0.4
"--"."".".".",.,,.,,""
---<1---. m* = 0.5
,,,,,,0,,,, -,, m* = 0.6
0.2
m* = 0.7
_1llJ. m* = 0.8

I
01
1.5 2 2.5
CD(mn)

Figure 4. Mean drag C D(mn) as a function of the normalized amplitude of oscillation

e;td

Given the dramatic increase in the mean drag observed for m* =

0.1,0.2 it may be assumed that the wake dynamics have significantly
altered, and more specifically that the amplitude of oscillation has also
increased. Figure 4 shows the amplitude of oscillation, e;td as a function
of the mean drag, CD(mn). The oscillation amplitude is defined as the
standard deviation of the cylinder's motion normalized by the angle
subtended by the cylinder diameter, D. (See figure 1a). For m* =
0.1,0.2 the increase in drag is observed to be directly related to an
increase in the amplitude of cylinder oscillation. For higher mass ratios
there is no jump away from a mean drag range, CD = 1.3 - 1.6 and no
jump in the oscillation amplitude.
142

It appears from the above results that there are two distinct states for
the cylinder for m* < 0.3. A state of low amplitude oscillation and low
mean drag, and a state of high amplitude oscillation and high mean drag.
The transition between the two states involves a dramatic increase in the
mean layover angle. The high amplitude state is not observed for higher
mass ratios. The amplitude of oscillation for the high amplitude state
appears remarkably constant for both m* = 0.1 and 0.2 at a value of
e;td ~ 0.8. Note that this does not imply that the cylinder is oscillating
with an amplitude of 0.8 diameters.

3.3 Cylinder Oscillation Results

0.8

~
I

0.6
~

*CD'"
0.4

0.2

10 15 20
u*
Figure 5. Normalized amplitude of oscillation O;td as a function of the reduced
velocity U*

The normalized cylinder amplitude, e;td is plotted against the reduced

velocity in figure 5. Here the constant amplitude of the high amplitude
state (for m* = 0.1,0.2) is clearly observed. A slight flattening of the
 143

O;td/U* curve is noticed for a mass ratio, m* = 0.3, however it is believed

that this deviation is not the high amplitude state observed for lower
mass ratios, as the amplitude of oscillation continues to increase at higher
reduced velocities and the amplitude in this range is not the same as
observed for m* = 0.1 and 0.2.
The range of reduced velocity possible is inversely proportional to the
mass ratio, such that cylinders with a higher mass ratio have a far more
extensive reduced velocity range. With this in mind, the high amplitude
state observed for m* = 0.1 and 0.2 extends over a considerable range
of the possible U* values. The work of Ryan et al., 2003 for a cylinder
mass ratio m* = 0.833 indicates that for high mass ratios the amplitude
of oscillation quickly decreases beyond a maximum value (for U* > 21 in
their study). In our study, for the m* = 0.1 and 0.2 case, the reduced ve-
locity range is restricted such that it is believed that the high amplitude
state will continue up to the highest reduced velocity possible.

4. DISCUSSION - THE EXISTENCE OF A

CRITICAL MASS RATIO
The present findings indicate a critical mass ratio, m~rit between 0.2
and 0.3. For m* < m~rit it is possible for the cylinder to oscillate in the
high amplitude state. This result compares favorably with the findings
of Govardhan and Williamson, 2000 and Govardhan and Williamson,
2003 in their investigations of a low mass-damped cylinder allowed to
oscillate freely in a direction transverse to the free-stream. They found
that for a m * < m~rit high amplitude oscillations resulted. Our present
value of m~rit appears to be considerably less than their critical val-
ue (their findings indicating m~rit = 0.54). However in Govardhan
and Williamson, 2003 the authors reinterpreted the low Reynolds num-
ber, two-dimensional numerical results of Shiels et al., 2001 and found
m~rit = 0.25 for this case. This result appears to compare well with the
present findings, and as such it may be possible to compare the present
results to the freely oscillating cylinder case. However care has to be
taken when comparing our results directly to those of a hydro-elastically
mounted cylinder as, in our case, the cylinder moves in both an in-line
and transverse direction, and the coupling of these motions may yield
different wake structures when compared to the structures in the wake
of a cylinder free to oscillate transverse to the flow field.

5. CONCLUSION
A critical mass ratio, m* = 0.2 - 0.3, has been found for the uniform
flow past a tethered cylinder, below which sustained large amplitude
144

oscillations are observed up to the highest reduced velocity simulated in

this study. The critical mass ratio was found to coincide closely with that
found for a hydro-elastically mounted cylinder using two-dimensional
simulations in previous studies. However, for a tethered cylinder, the
critical mass ratio affects both the mean layover angle as well as the
amplitude of oscillation.

6. ACKNOWLEDGEMENTS
The first author would like to acknowledge support provided through
a Monash Departmental Scholarship, partially funded by an ARC large
Grant AlO017086. The authors would like to acknowledge strong sup-
port from the Victorian Partnership for Advanced Computing (VPAC),
and the Australian Partnership for Advanced Computing (APAC).

References
Blackburn, H. M. and Henderson, R. D. (1999). A study of two-dimensional flow past
an oscillating cylinder. Journal of Fluid Mechanics., 385:255-286.
Govardhan, R. and Williamson, C. (2000). Modes of vortex formation and frequency
response of a freely vibrating cylinder. Journal of Fluid Mechanics., 420:85-130.
Govardhan , R. and Williamson, C. (2003). Resonance forever: existence of a critical
mass and an infinite regime of resonance in votex-induced vibration. Journal of
Fluid Mechanics ., 473:147-166.
Henderson, R. (1995). Details of the drag curve near the onset of vortex shedding.
Physics of Fluids, 7:2102-2104.
Pregnalato, C., Ryan, K., Thompson, M., and Hourigan, K. (2002). Numerical simula-
tions of the flow-induced vibrations of tethered bluff bodies. Proceedings of IMECE
2002: 5th International Symposium on FSI, AE, FIV and N , Article Number 32168.
Ryan, K., Pregnalato, C., Thompson, M., and Hourigan, K. (2003). Flow-induced
vibrations of a tethered circular cylinder. Journal of Fluids and Structures, Ac-
cepted.
Ryan, K., Thompson, M. , and Hourigan, K. (2002). Energy transfer in a vortex in-
duced vibrating tethered cylinder system. Conference on Bluff Body Wakes and
Vortex-Induced vibrations, BBVIV3, pages 57-60.
Shiels, D., Leonard, A., and Roshko, A. (2001). Flow-induced vibration of a circular
cylinder at limiting structural parameters. Journal of Fluids and Structures, 15:3-
21.
 SELF-EXCITED OSCILLATION OF EQUILATERAL
TRIANGULAR WEDGE

SUTTHIPHONG SRIGRAROM
School of Mechanical and Production Engineering
Nanyang Technological University, 50, Nanyang Avenue, Singapore,
639798 E-mail: mssrigrarom@ntu.edu.sg

MITSURU KUROSAKA
Department ofAeronautics and Astronautics
University of Washington, Seattle, WA, 98195-2400, USA
E-mail: kurosaka@aa. washington. edu

Abstract. This paper investigates the characteristics of a particular fluid-structure interaction phenomenon,
i.e. the continuous oscillation in rotational mode of an equilateral triangular wedge in the uniform incoming
flow in the water tunnel. We propose the explanation of this self-excited oscillation. It is the unbalance force
acting on the wedge's side faces that causes such movement. If the wedge is positioned initially
asymmetrically against the freestream, on one side, the flow will be flow-past-flat-plate like, whereas the
other side will be flow-past-sharp-edge like. Due to the unbalanced pressure exerting on the two sides, the
wedge will rotate. When the wedge moves, these mechanisms switch side interchangeably, and bring the
wedge to continuous oscillation. To understand more thoroughly, several experiments were conducted to
investigate such behavior by means of Food coloring dye and Laser Induced Fluorescence flow visualization.
The oscillating frequency is governed by Strouhal numbers, which appears within limited range of 0.12 < SIr
< 0.18. Beyond this range, the wedge is either stationary or rotates only in one direction. Accompanied with,
the simple wedge's dynamic behavior is examined, by considering incoming freestream flow's hydrodynamic
force acting on the wedge.

Keywords: fluid-structure interaction, oscillation, triangular cylinder, vortex shedding, Strouhal number

1. INTRODUCTION

It is well known and in general practice to use the oscillating flaps or duck-fins to
suppress the ocean surface waves approaching the shore. The flaps or duck-fins have
the drawback that it works only on the water surface. Recent studies (Nakashima, 1992)
revealed that if a isosceles triangular wedge is placed in an otherwise uniform flow, it
will be induced to oscillate spontaneously. We foresee that this unique feature, by use
of the equilateral triangular wedge, can be effectively used instead of the flap or duck-

145
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 145-158.
2003 Kluwer Academic Publishers.
146

fin, since, it can not only be used on the water surface, but also submerged in the water.
Furthermore, with proper mechanism, ones can extract energy from the wedge
oscillating and/or spinning motion, of which the energy from the incoming flow or
wave, indeed. The detailed motion of the wedge is, however, not fully understood. This
is due to the unsteady surrounding flow, as well as, continuously moving boundary (the
wedge either oscillates or rotates). To the authors' knowledge, this coupled fluid-
structure interaction has first noticed by Nakashima et al (Nakashima, 1992) and it has
not yet been well studied. The available prior research works were only dealt with the
stationary wedge at different position or oscillation in translational mode (Luo et aI,
1993), or with other geometry (Naudascher & Wang, 1993, Nakamura & Nakashima,
1993, Sakamoto et aI, 2001, Hu et aI, 2002).
In this paper, we propose our observation-based explanation. It is the
unbalance force acting on the wedge's side faces that causes such movement. On one
side, the flow will be flow-past-flat-plate like, whereas the other side will be flow-past-
sharp-edge like. When the wedge rotates, these mechanisms switch side
interchangeably, and bring the wedge to continuous oscillation. Three sets of
experimental results are presented.

2. PROPOSED OSCILLATION MODEL

In the figure 1, we consider the flow pattern around an equilateral triangular wedge, of
which the front face AB inclines, making an angle with the freestream. The flow
separates at both A and B, and the pattern is asymmetrical. The radius of curvature of
the streamline separated at A, rA, is smaller than the radius of curvature at B, rB. As a
result of conservation of angular momentum, the velocity at A is higher than at B;
therefore, the pressure at A is lower than at B (PA < PB ). The wedge rotates clockwise
about the pivot, and the frontal surface AB becomes more inclined to the free stream.
Due to the continuing turning, the triangle is now is a position where A C is
parallel to the free stream. The flow still separates at the upper tip of the wedge (B), but
the flow in the lower part, after turning parallel to the front face, separates only slightly
at the lower tip of the wedge (A), then reattaches to the lower lateral face (A C). The
lower flow, then, changes behaviour to that ofa flat-plate-like flow.
By virtue of the obvious difference between the two flow patterns, the local
pressures differ at the upper and lower parts of the wedge. The upper part, with the
existence of a large eddy, has lower pressure, especially at the core of the eddy,
compared with the free stream; whereas at the lower part the pressure is equal to the
freestream pressure, since the fluid still flows smoothly and parallel there. Therefore,
 147

the flow in the lower part now has greater pressure than the upper part, (PA > Pa ) and
the wedge tends to rotate back to its original position.

Fig. 1 (left): Flow past the equilateral triangular wedge at positive angle of attack (AOA).
Fig. 2 (right): Flow past the equilateral triangular wedge at positive AOA (continued).

As a consequence of the above step, the unbalanced pressure forces the wedge
to rotate counterclockwise back to its original position (under the assumption that the
wedge starts rotating in a clockwise direction, as described in the previous step). Due to
the inertia of the wedge and flow, the motion of the wedge does not stop when it returns
to the symmetric position (AB lies perpendicular to the freestream). Instead, the wedge
continues to swing in the counterclockwise rotation. As a result, the flow pattern
becomes as shown in figures 3 and 4, which correspond to those of figures 1 and 2
flipped upward.
The overall phenomenon can be viewed as the interchange of the flow
patterns, from the flow past the sharp edge to flow over the flat plate, and vice versa as
shown in figure 5.

/'"
.---t.-.-.~

Fig. 3 (left): Flow past the equilateral triangular wedge at negative AOA.
Fig. 4 (right): Flow past the equilateral triangular wedge at negative AOA (continued).
148

Wedge in clockwise direction Wedge in counterclockwise direction

Fig. 5: The oscillating motion ofthe wedge continues with interchangingflow patterns.

3. WEDGE SELECTION

Various shapes of the symmetric triangular wedge can be described and differentiated
by the base angle. In this study 15, 30, 45 and 60 wedges are randomly chosen, as
shown in figure 6 below.

15 base angle 30 base angle 45 base angle 60 base angle

Fig. 6: Selected wedges with different angles at the base.

3.1. Behavior of the 15 ~ 30 0 and 45 wedges

The 15, 30 and 45 wedges exhibit similar behavior when placed in the freestream.
The wedges are initially placed with the base face (the longest side) perpendicular to
the incoming flow. After being set up, the wedges rotate back to the horizontal position,
with the base parallel to the flow (see diagram in figure 7 below). Consequently, the
lower parts of the wedges behave like the flow plates, and the upper parts behave like
converging-diverging channels. The wedges, then, stay still, without further movement.
There is no major oscillation, only fluttering motion, although there are slight
separations at the apexes on the upper parts. This stable behavior happens at all
freestream velocities (Reynolds' number). Therefore, these three wedge angles are not
used in this study.
 149

Fig. 7: Behavior of 15 ~ 30 and 45 wedges. The wedge is in a vertical position, first. Then, it
rotates to horizontal position and becomes stable.

3.2. Behavior of the 60 wedge

The 60 wedge demonstrates unique behavior. When it rotates by 60, either clockwise
or counter-clockwise, its appearance is still similar to the initial 60 wedge, as
implicitly shown in figure 5, in any rolling position. Since, the shape is always
preserved and symmetric, therefore, it does not have the problem in terms of a stable
(or dead) position.

4. EXPERIMENTAL RESULTS

4.1. First set of results conducted at University of Washington (USA)

The study of the oscillation of equilateral triangular wedges were first performed using
Laser Induced Fluorescence (LIF) flow visualization method in the water tunnel at
Department of Aeronautics and Astronautics, University of Washington (UWAA) ,
Seattle, USA. This water tunnel is a recirculating facility with a 3.0m x 0.7m x 0.7m
glass test section and a maximum flow speed of 70 cm/s. A complete description of the
water tunnel can be found in the first author's previous paper (Srigrarom, 1998). The
free-stream turbulence level of approximately 0.3% of the mean flow exists at all tunnel
speed. The incoming flow velocities were varied such that we could observe wide range
of Reynolds and Strouhal numbers.
At UW AA, three wedges of three different sizes were made of Plexiglas. The
wedge dimensions were 18in x 2.5in, 18in x 5in and 8in x 1.0in. Each one had a
symmetric triangle (60 at all comers) cross section. All the wedges were hollow with
plugs for leak water drainage. The thickness of the Plexiglas was \4 inch. The Plexiglas
plates were glued together to form up the wedge. The \4 in holes for supporting hinges
were drilled at the geometrical center on both ends. The effect of unbalanced buoyancy
is assumed negligible. The supporting rig is made up of stainless steel, and was
150

designed to hold the wedge in the water from the wedge's ends; it hangs on the tunnel's
side wall.
We performed two experiments. One was the wedge in uniform flow with
initial perturbation. Another was from the wedge behind the circular cylinder lied
upstream. The latter refers to the wedge under fluctuating incoming flow. Also, it is to
check whether the wedge's oscillation is the same as the natural vortex shedding or not.
The result appeared that their frequencies (from the wedge's self oscillation and from
the cylinder vortex shedding) were not the same, also their associated Strouhal numbers
were different. This means that the wedge's self oscillation is another unique and
distinctive natural phenomenon.

4.1.1. Wedge without cylinder upstream: From the experiments in the water tunnel, of
which the setup has already described earlier, the wedge started to oscillate in rotational
mode by some initial perturbation, then kept oscillating forever. Here, we consider the
oscillating condition, which depends on the incoming flow. In such flow situation, the
wedge oscillates in such a short range of freestream velocity, U" , controlled by
Strouhal number, defined as:
Str = jW (1)
Uoo
where f = oscillating frequency, W = width of the lateral side of the wedge, U00
magnitude of the freestream velocity.
From observation, we found out that the oscillation would take place freely
only when Strouhal number was in the range of 0.12 < Str < 0.18, as shown in table 1
below. This agrees well with other natural oscillating flows (Panton, 1984). In this
specific range, the observed wedge's oscillating frequency is in order of 1 Hz,
depended inversely on the freestream velocity and the size of the wedge as described in
Strouhal number definition. This does not include the rotation of the wedge, which
would occur when the free stream velocity exceeds upper limit, marked by lower
Strouhal number of 0.12. Table 1 summarizes the effect of Strouhal number for the
wedge in water flow:

TABLE 1. Effect ofStrouhal number for the wedge in water flow

Strouhal Freestream velocity Flow situation
number, Str condition
> 0.18 Too low No movement on the wedge, wedge remains stationary
0.12-0.18 Optimum Wedge oscillates in rotational mode continuously
< 0.12 Too high Wedge rotates in one direction at all time

Note that, in our experiments, the way we measured the frequency was the
indirect method. Figure 8 shows the setup of the wedges with the stop watch. Once the
 151

oscillation takes place, we timed the period of each oscillating cycle, the inverse the
number to get the estimate frequency at the matching freestream velocity.

Fig. 8 (left): Experimental setup for frequency response test. The stop watch was located between
the water tunnel test section and the camera. The picture shows the 2.5 inch wedge.
Fig. 9 (right): Experimental setup for frequency response test behind a circular cylinder. Notice
the 3.5 inch diameter circular cylinder on the left and the stop watch at the middle, located
between the water tunnel test section and the camera.

The frequency plot of the oscillating wedge is shown in figure 10 below. Here
we consider the smallest tested wedge of which d = 2.5 inch and has the least polar
moment of inertia (J), and hence, can start oscillating with minimum free stream
velocity (external moment). The wedge triggers vibration at about 5 cmls freestream
velocity. The frequency varies in a small range. The dimensionless parameters
(Strouhal and Reynolds' numbers) plot is shown in the following figure 11.
In figure 11, for the wedge only, Strouhal number is obviously appeared to be
constant at Str ~ 0.14, in considerable range of Reynolds number. This Strouha1 number
appears within limit of the wedge favorable oscillating range, presented in table 1.
152
Wec.fgcayerageoscillatingfrequency

~~--~----~'~5----~20----~~~--~30'
Wilier tlRll'lel speed (cmIs)

Fig. 10 (left): Wedge average oscillating frequency, both wedge only and wedge behind circular
cylinder,
Fig. 11 (right): Wedge average Strouhal number, both wedge only and wedge behind circular
cylinder,

4,1,2. Wedge behind cylinder upstream: Figure 9 shows the experiment of the wedge
behind the cylinder. The objective is to check the dynamical behavior of the wedge
under force excitation (moment), created from the shed vortices from the cylinder. It is
well-known that the cylinder creates the shed vortices at fixed frequency (Str - 0.35).
This implicitly simulates the wavy or fluctuating free stream condition, in which the
wedge placed downstream may be experienced. Note that, the frequency measurement
method is the same as the wedge without the cylinder as shown in figure 8.
In this case, the response frequency by the wedge still varies in a small range,
but at higher values, as also shown in figure 10 The dimensionless parameters (Strouhal
and Reynolds' numbers) plot is also shown in the figure 11. For the wedge behind the
cylinder case, Strouhal number stay approximately at Str -0.3. Therefore, the wedge
responses itself to the cylinder's shed vortices' frequency, compared to the previous
case (wedge without cylinder) where Str - 0.14. In this case, we consider as the forced
oscillation of the wedge by the upstream cylinder. Note that, in both cases, the wedge is
in steady-state.
 153

4.1.3. Laser Induced Fluorescence (LIF) flow visualization images: In this section,
there are images of the oscillation of the smallest wedge with d = 5 inch (3.81 cms).
The oscillating frequency was approximately 1.0 Hz. The free stream velocity was at 24
cm/s. This corresponded to Strouhal number of Str ~ 0.16, and Reynolds number based
on the wedge of the wedge was Rew ~ 9100. The pictures were taken with long shutter
speed. So, Fluorescein appears as the stretched streamlines. The spatial non-uniformity
was the result from the diffusion after released from the upstream injection port. The
surrounding flow pattern appeared as predicted as shown together with the diagram.
The image series in figures 12a-b show the oscillation of the wedge. The
wedge is the white triangle inside the hinge (appeared dark). The camera was aligned
slightly to the left, so that we could see the frontal part of the wedge. The oscillation
appeared stable, alternating back and forth from the positive angle of attack (figure 12a)
about +30 0 to negative angle of attack -30 0 (figure 12b). The agreement to the
proposed mechanism, represented by schematic diagrams, is reasonable. In figure 12a,
at the same wedge position, the upper and lower separation can be traced by the dark
and bright trajectories of Fluorescein streamline. In figure 12b, with the
counterclockwise motion phase, the separation appears clearly as indicated by an arrow.
The wedge tends to swing more beyond the estimated positions about 50, because of its
own inertia. In our analysis scope, we ignore this little overshoot for simplicity.

"./
.L!f.,: !..';rt5....
..... S . ' . .

-. .. c-
<::.:
~- ) .: r l
-~, - -1....>

<:>' -"'; ; _ A'.A

(a)
154

(b)
Fig. 12 (aj and (bj: LIF images o/the wedge. The while triangle is the enhancement o/the wedge
position. The images are presented with the corresponding proposed mechanism, schematic
diagrams. Theflow was at Rew~ 9100, Uoo = 24 cm/s and Str ~ 0.16.

So far, we have discussed about the flow field at the lower part of the wedge,
since we could observe from the flow visualization in the figures. It is imperative that
for the clockwise motion, the same phenomena happens on the upper side . On the
upper side of the wedge, similar mechanisms also appear, but in the opposite sense
alternately. That is, when the lower part behaves like the flat plate, the upper part
behaves like the sharp edge, and vice versa, interchangeably. Although, we could not
see the flow on the upper part from these figures, because of blockage by the hinge and
the supporter, the continuous oscillation supports this assumption.

4.2. Second set of results conducted at Nanyang Technological University (Singapore)

Besides the initial experimental works at University of Washington just presented,

similar experiments had been conducted at Nanyang Technological University (NTU),
Singapore. The objective is to verify the universality, repeatability and uniqueness of
the phenomena. At NTU, a two-dimensional water tunnel, of which the width to height
aspect ratio of 10:1 was used. The details of the water channel can be found in
reference 9 (Yeo, 2002). Here, two equilateral triangular wedges of 1 inch and 3 inches
face width were used. Each wedge had a free-spinning axis at its geometrical center,
and was placed vertically at the middle of the test section of the water channel, as
shown in figure 13. The digital video camera was used to record the tests. The digital
tachometer was used for measure the oscillating frequencies, and placed outside and
above the test section as shown in figure 14.
 155

Fig. 13 (left): The 3-inch tested wedge model, placed inside the two-dimensional water channel
at NTU.
Fig. 14 (right): To measure the frequency, the tachometer was put in place to count the
oscillation.

Fig. 15 (left) and 16 (right) : The I-inch tested wedge model oscillates inside the water channel.
The left figure shows the instant when flow at the lower face becomes flat-plate like, and flow at
the upper face separated. The right figure shows the opposite situation.

Both the 1 inch and 3 inches wedges appeared to be self-excited and oscillate
in rotational mode well as expected, at different conditions (from 700 < Rew < 14000).
Examples of the instant motion images are shown in figures 15 and 16. The plot of the
mean Strouhal number and the Reynolds number based on width for both cases are
shown in figures 17 and 18.
As shown in figures 17 and 18 for both wedges, the triangle oscillated in
almost at the fixed Strouhal number of 0.125, for a wide range of Reynolds number.
This is contributed by the well-designed two-dimensionality and controllability of the
NTU water channel and the test setup. This corresponds to the fmdings earlier that the
156

wedge tends to self-excited oscillate in the range of 0.12 < Str < 0.18 from experiments
atUWAA.

, .
, .
1(112

'"
"'"

~~------~~----~~~-------d, ...

_......
~~bM ... (1j!I~~nPr+Klr .. _

.
f'U'
~
'.I "
01 .. __ ~ _ ___ - --- ---

~ooe

,
" .&7 " ... ,~ , A . 1
R ~AII'Ilbowt.ti41l"", . . . -otfl6llll"..w_
.. " . i1 $
1'(/

Fig. 17 (left); The plot ofStrouhal number and Reynolds number for the wedge oscillation, taken
from the experiments conducted at Nanyang Technological University (NTU) in Singapore. Note
that, the mean value ofStrouhal number is approximately 0.125 (i.e. still within 0.12 < Str < 0.18
oscillation range).
Fig. 18 (right); The plot of Strouhal number and Reynolds number for the wedge oscillation,
taken from the experiments conducted at National Cheng Kung University (NCKU) in Taiwan.
The range of Strouhal numbers are between 0.12 and 0.16 (i.e. still within 0.12 < Str < 0.18
oscillation range).

4.3. Third set o/results conducted at National Cheng Kung University (Taiwan)

Based on the same checking philosophy, another set of experiments were conducted at
Institute of Aeronautics and Astronautics, National Cheng Kung University (NCKU) in
Tainan, Taiwan, to ensure commonality, repeatability and uniqueness of the
phenomena. Here, the smaller triangular wedge was in use and tested in the water table.
 157

The wedge was 6 em wide and 0.5 em thick. The test conditions were narrowly varied
at 39,000 < Rew < 54,000. The wedge appeared to oscillate in wider range at 0.12 < Str
< 0.16, but still within the predicted limit, as shown in figures 17 and 18 above. Some
of the images taken from the experiments are shown in figure 19 below.

w w ~
Fig 19 (a), (b) and (c) : The images from the experiment of the self-excited oscillating triangle in
the water table at NCKU. Figure (a) shows the symmetric case, figure (b) shows the wedge at
negative angle attack, and figure (c) shows the wedge at positive angle attack. Note the different
light intensity, resulting from the different water level (reflection), indicates the incomingflow
and separations at the edges and side faces.

From the experimental results taken at NTU and NCKU, of which the Strouhal
number v.s. Reynolds number graphs in both places are shown in figures 17 and 18
above, it is clear that the self-excited oscillation behavior of the equilateral triangular
wedges in water is always at 0.12 < Str < 0.18. This is similar to the findings from the
initial sets of experiments at University of Washington (UWAA), presented earlier.

5. CONCLUSIONS

The self-excited oscillation in rotational mode behavior of the equilateral triangular

wedge is reasonably explained from the experimental observation, as the consequence
of the alternating flat-plate-like and sharp-edge-like flow patterns. The start-up process
is the unbalance pressure on both lateral sides of the wedge, initiated from the small
fluctuation. The series of flow visualization figures and experimental results from three
different places support the hypothesis. From the frequency observations, we conclude
that this wedge's self-excited oscillation is the unique and distinctive natural
phenomenon.
158

6. ACKNOWLEDGEMENTS

In preparing this paper, the first author (SS) wishes to thank Prof. Fei-Bin Hsiao and the
Institute of Aeronautics and Astronautics, National Cheng Kung University, Tainan,
Taiwan, that allowed him to use their facilities to do the experiments.

7. REFERENCES

The references are listed in alphabetical orders.

\. Hu, C.C, Miau, J.J. and Chou, J.H. (2002), "Instantaneous Vortex-shedding Behaviour in Periodically
Varying Flow", Proc. Roy. Soc. Lond. A, Vo1.458, pp. 911-932.
2. Luo, S.C., Yazdani, M.G., Lee, T.S. & Chew, Y.T. (1993), "Aerodynamic Stability of Square,
Trapezoidal and Triangular Cylinders", Proceedings of the Third International Offshore an Polar
Engineering Conference, pp. 709-714, Singapore, 6-11 June 1993.
3. Nagashima, T. & Hirose, T, "Potential Flow around Two Dimensional Isosceles Triangular Cylinder
SUbjected to Uniform Flow from Base Surface", Journal of Japan National Defense Agency, 1992.
4. Nakamura, Y. & Nakashima, M. (1986), "Vortex Excitation of Prisms with Elongated Rectangular, H
and ~ Cross-sections", J. Fluid Mech., Vol. 163, pp.149-169.
5. Naudascher, E. & Wang, Y. (1993), "Flow-induced Vibrations of Prismatic Bodies and Grids of
Prisms", J. Fluids and Struct., Vol. 7, pp. 341-373.
6. Panton, R. (1984), "Incompressible Flow", John Wiley & Sons, New York.
7. Sakamoto, H.,Takai, K., Alam, M.M. and Moriya, M. (2001), "Suppression and Characteristics of Flow
Induced Vibration of Rectangular Prisms with Various Width-to-height Ratios", in "Fluid Structure
Interaction", (eds.) Chakrabarti, S.K & Brebbia, C.A., pp. 67-76. WIT Press.
8. Srigrarom, S. (1998), "Self-Excited Oscillation of Triangular Wedge", Master Thesis, Department of
Aeronautics and Astronautics, University of Washington, Seattle, WA, USA, June 1998.
9. Yeo, K.H. (2002), "Smart Control of Turbulence by the Use of Drag-Reducing Additives", Batchelor
degree final year project report at Nanyang Technological University, Singapore, April 2002.
 A CRITICAL REVIEW OF THE INTRINSIC NATURE
OFVIV

T.SARPKAYA
Naval Postgraduate School, Monterey, CA

Abstract: This is a concise and comprehensive review of the progress made during the
past two decades on vortex-induced vibration (VIV) of mostly circular cylindrical
structures in uniform flow. The critical elements of the evolution of the ideas,
theoretical insights, experimental methods, and numerical models are traced
systematically; the strengths and weaknesses of the current state of the understanding of
the complex fluid/structure interaction are discussed in some detail. Finally,
suggestions for the future objectives of VIV research are presented.

Von Karrruin once wrote: "Problems never have final and universal solutions, and
only a constant inquisitive attitude toward science and a ceaseless and swift adaptation
to new developments can maintain the security of this nation." During the past century,
a great deal of work has been done on flow-induced vibrations and fluidelastic
instability. The number of contributions and incremental publishing has increased
exponentially. Clearly, to move forward and to shape the art and science of flow-
induced vibrations in the new century, the existing theoretical, experimental, numerical
and empirical evidence must be periodically re-energized and marshaled for new
discoveries and applications. This requires a comprehensive review, at least every 25
years.

The difficulties experienced in describing the nature, identifying the occurrence,

and predicting the characteristics of vortex induced vibrations of bluff bodies and
galloping (to a lesser extent), have been reviewed by Parkinson (1974), Sarpkaya
(1979), Griffin and Ramberg (1982), Bearman (1984), Parkinson (1989), Pantazopoulos
(1994), Sarpkaya (1995), and in books by Chen (1987), Blevins (1990), Naudascher
and Rockwell (1993), Sumer and Fredsoe (1997), and Au-Yang (2001) and, less
formally, in practically every doctoral thesis, as part of the obligatory "previous
studies" section.
Flow-induced vibrations occur in many engineering situations, such as bridges,
stacks, transmission lines, offshore structures, heat exchangers, marine cables, flexible

159
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 159-161.
2003 Kluwer Academic Publishers.
160

risers in petroleum production, and other hydrodynamic and hydroacoustic applications.

This review is not a flat chronology of scientific/engineering developments in VIV and
does not make an effort to refer to everything that has ever been published, but one that
seeks to provide an intimate feeling of physical reality or physical insight. It makes no
promises that are either excessively pessimistic or unreasonably reassuring. It
encourages cross stimulation between relatively idealized physical and numerical
experiments and far more complex technological applications (often found in books and
numerous conference proceedings).

Much progress has been made during the past decade, both numerically and
experimentally, towards the understanding of the kinematics (vice dynamics) of VIV,
albeit in the low-Reynolds number regime. The fundamental reason for the foregoing is
that VIV is not a small perturbation superimposed on a mean steady motion. It is an
inherently nonlinear, self-governed or self-regulated, multi-degree-of-freedom
phenomenon. It presents unsteady flow characteristics manifested by the existence of
large-scale structures, sandwiched between two equally unsteady shear layers.

A phenomenon as robust as the vortex shedding gives rise to forces as

unpredictable as the lift force whose power can be fully appreciated only when one tries
to eliminate VIV without excising the after body. There is much that is known and
understood and much that remains in the empirical/descriptive realm of knowledge.
Industrial applications highlight our inability to predict the dynamic response of fluid-
structure interactions. They continue to require the input of the in-phase and out-of-
phase components of the transverse force, in-line drag, correlation lengths, damping
coefficients, relative roughness, shear, waves, and currents, among other governing and
influencing parameters, and thus the input of relatively large safety factors.

As in the case of many other fluid flows, stability and turbulence, often with large
coherent structures, unknown integral length scales, and motion-dependent coherence
lengths remain as major obstacles to the understanding of the physics and to the
numerical simulation of the dynamics as well as the kinematics of flow structures in the
shear layers and the near-wake of cylinders and cables. There does not appear to exist
one or two parameters into which we can lump our inability to account for the effects of
all the individually non-quantifiable influencing parameters. The most obvious
candidates are those that exhibit large scatter in every experiment, e.g. the fluctuating
lift (its spectra and r.m.s. value), a measure of the turbulence distribution of the ambient
flow (intensity and the integral length scales), and some measure of pressure
fluctuations on the body.
 161

In summary, the discussion of the intrinsic nature of VIV, the role of added mass,
the decomposition of time-dependent force, the linear and non-linear equations of
motion, the free and forced oscillations, the numerical simulations, and the hopes for
suppression are followed by recommendations for future directions. It is concluded that
partly the prediction and thereby the avoidance of VIV and partly the application of
more ingenious means and passive devices may be the road to the future. After all, the
lift will always be there with or without the VIV and the pure circular cylinder will
always be the preferred shape with or without shape modifications.
 A NUMERICAL STUDY ON THE
RECTILINEAR OSCILLATIONS OF A
CIRCULAR CYLINDER

SERPIL KOCABIYIK
Department of Mathematics and Statistics
Memorial University of Newfoundland
St. John's, Newfoundland, Canada Ai C 5S7
serpil@math.mun.ca

QASEM AL-MDALLAL
Department of Mathematics and Statistics
Memorial University of Newfoundland
St. John's, Newfoundland, Canada A1C 5S7
qasem@math.mun.ca

Abstract: The flow structure and loading due to rectilinear oscillations of a circu-
lar cylinder in a steady uniform flow are investigated numerically at a
fixed Reynolds number R = 855. Numerical results are obtained over
broad ranges of two externally specified parameters, i.e. the frequency
of forced oscillation f relative to the natural vortex-shedding frequency
fo (0.5 ~ f / fo ~ 4.0) and the angle of inclination TJ between oscillation
axis and free-stream (TJ = 30, 45, 60, 75). The dimensionless oscil-
lation amplitude is fixed at A = 0.26. The method of solution is based
on the use of truncated Fourier series representations for the stream
function and vorticity in the angular polar coordinate. The N avier-
Stokes equations are solved by using finite difference methods, but with
the boundary vorticity calculated using integral conditions rather than
local finite-difference approximations.

Keywords: Unsteady, incompressible, viscous, rectilinear oscillations, cylinder

1. INTRODUCTION
Most of the previous studies concentrated on understanding the features
of the flow for cylinder subject to controlled forcings. Rectilinear oscil-

163
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 163-173.
2003 Kluwer Academic Publishers.
164

lations form one of the simplest forcing methods on the cylinder e.g.
transverse oscillations (Ongoren and Rockwell (1988a); Williamson and
Roshko (1988); Nguyen and Kocabiyik (1997); Kocabiyik and Nguyen
(1999); Blackburn and Henderson (1999)) and streamwise oscillations
(Ongoren and Rockwell (1988b); Badr, Dennis, Kocabiyik and Nguyen
(1995); Cetiner and Rockwell (2001)). Many properties of controlled
flows are reported in review articles by Berger and Willie (1972), Sarp-
kaya (1979), Bearman (1984), Griffin and Hall (1991), Rockwell (1998)
and also in a book by Sumer and Freds0 (1997).

This present paper gives results for the laminar two-dimensional flow
generated by an infinitely long circular cylinder. Initially the cylinder is
at rest, and then suddenly starts to perform rectilinear oscillations at an
arbitrary angle TJ with respect to the free stream. Only one experimental
study has been made on this problem: Ongoren and Rockwell (1988b).
Visual observations were made in the Reynolds number and frequency
ratio ranges 584 ~ R ~ 1300 and 0.5 ~ f / fo ~ 4.0, respectively, at angle
of inclination TJ = 00 ,45 0 ,60 0 ,90 0 of cylinder oscillation with respect to
free stream. For most experiments, a constant value of dimensionless
amplitude, A = 0.26 , was chosen. They showed that if the cylinder is
excited at an angle other than the cross-stream or streamwise direction,
then there is mixed-mode excitation: the perturbation from the cylinder
motion contains both symmetrical and anti-symmetrical contributions
and the potential for exciting both types of modes. Moreover, these
modes either can be synchronized, i.e. phase-locked, with the cylinder
motion, or they can compete with each other. In their investigation, they
addressed under which conditions these modes occur, for both synchro-
nized and non-synchronized vortex formation. Particular remarkable
finding by Ongoren and Rockwell is that the occurrence of synchronized
vortex formation in the asymmetric mode when the cylinder motion
produces purely symmetrical perturbations of large amplitude. These
experimental findings present a promising ground for extending this s-
tudy into feasibly active control by utilizing the rectilinear oscillation at
an angle other than the cross-stream or streamwise direction. It should
be noted, here that, since excitations at arbitrary angle TJ can be eas-
ily achieved by a simple electronic device and mechanical means, the
present method is practically applicable.

It seems that there are only two numerical studies have been made on
this problem: Kocabiyik and Mahfouz (2002), and Kocabiyik and AI-
Mdallal (2003). These studies have a common focus on when and how
the vortex shedding is synchronized with the cylinder oscillation. In the
 165

first, simulations were made in the range 0.25 ::; f / fo ::; 1 when A = 0.2
and 'T/ = 30, 45 at two values of the Reynolds number R = 500, 104
whereas in the second, numerical calculations were carried out over the
range 0.5 ::; f / fo ::; 4 when A = 0.26 and 'T/ = 45 at the fixed Reynolds
number 855.

Based on the experimental observations by Ongoren and Rockwell (1988b),

finite-difference simulations are made in the present study to portray
the unsteady dynamics of wake flows. The emphasis of this study is
to analyze the vortex formation behind a circular cylinder for different
recti-linear oscillation conditions. Three main parameters are selected
for comparison, i.e., the frequency ratio, f / fo, maximum displacement
amplitude-to-cylinder radius ratio, A = Ymax/a, and angle of inclina-
tion, 'T/, between oscillation axis and the horizontal direction. Here fo,
Ymax and a are the fixed-cylinder (Karman) vortex shedding frequency;
displacement amplitude of the recti-linear cylinder oscillation, and cylin-
der radius, respectively. The Reynolds number based on the diameter,
2a, is fixed at R = 855. Several criteria are satisfied in the selection
of this value for the Reynolds number. First, to make interpretation
of the near-wake dynamics as simple as possible, it is desired to have
a single predominant frequency and purely asymmetrical growth of the
large-scale in the near-wake region. These criteria are ascertained by
cross-spectral measurements of Ongoren and Rockwell between hot-film
probes mounted on either side of the wake. At R = 855, the predicted
value for the Strouhal number of vortex shedding is So = 2afo/U = 0.22.
The primary advantage of the numerical simulation is that wide ranges
of the relevant flow variables can be encompassed. In the present s-
tudy, two parameters are varied in a range 0.5 ::; f / fo ::; 4.0 and
'T/ = 30, 45, 60, 75 and the maximum displacement amplitude-to-
cylinder radius ratio is fixed at 0.26 since preliminary studies showed
that this amplitude was sufficiently large to produce control of the near-
wake structure over a wide frequency range.

The flow configuration of the present recti-linear oscillation is shown in

Figure 1. Since the cylinder is oscillated sinusoidally in time T at a forc-
ing frequency, f, the dimensionless cylinder velocity, V(t), is expressed
by V(t) = excos(nSft) where ex = nASf represents dimensionless veloc-
ity amplitude of oscillation. Here the quantities are made dimensionless
adopting the following relations: t = UT/a and Sf = 2af /U.
166

Figure 1. Physical model and coordinate system

2. METHOD OF SOLUTION SUMMARY

The same basic formulation of the problem described in Kocabiyik and
Mahfouz (2002) is adopted. Modified polar coordinates (~ , e) are used,
where ~ = log(r/a) , with the origin at the centre of the cylinder. The
governing equations are given in the form

(2)

(3)

where 'lj; is the stream function and ( is the (negative) scalar vorticity.
These quantities are all dimensionless and are defined in Kocabiyik and
Mahfouz (2002). The boundary conditions are based on the no-slip
and impermeability conditions on the cylinder surface and free-stream
conditions away from it,

when ~ =0 (4)

and as ~ -+ 00

e -~ ~~ -+ (1 + V (t) cos 1]) sin e - V (t) sin 1] cos e, (5)

 167

e-~~: -T (1 + V(t)COS'T])cosO+ V(t) sin'T] sinO. (6)

The set of conditions (4)-(6) must be satisfied for all t > 0 and for all 0
such that 0 ~ 0 ~ 27r, and moreover, all the dependent variables in the
flow domain must be periodic functions of 0 with period 27r. Thus, in
particular,

'lj;(~,O+27r , t) ='lj;(~,O,t); ((~ , O+27r,t) =((~,O , t). (7)

In the present analysis the calculations are carried out on the basis of
the method of solution adopted by Badr and Dennis (1985) in which the
functions 'lj; and ( were expressed in the form of series

1 N
'lj;(~, 0, T) = 2Fo(~, T) +L (Fn(~, T) cos nO + fn(~ , T) sin nO), (8)
n=l

1 N
(( ~, 0, T) = 2Go (~ , T) + L (Gn (~, T) cos nO + gn (~ , T) sin nO) (9)
n=l

where N is the number of terms in each series. The functions defining

the Fourier coefficients must satisfy the governing equations (2) and (3).
Accordingly, the above approximations are used in these equations in or-
der to obtain a set of differential equation for each of the coefficients in
equations (8) and (9) together with the appropriate boundary condition-
s. The method of solution makes use of boundary-layer structure, but
without any approximations to the Navier-Stokes equations. The only
point worth emphasizing is that the boundary conditions on the func-
tions Fn(~ , t) and fn(~, t) corresponding to equations (2) and (3) are
utilized to deduce set of global conditions, termed integral conditions,
on the functions Gn(~, t) and gn(~ , t). An implicit method of Crank-
Nicolson type is used to integrate the vorticity equation in time and a
special scheme is used for solving stream function equation, following the
work of Badr and Dennis (1985) . The solution procedure requires the
knowledge of surface vorticity distribution. Integral conditions are used
to calculate the surface vorticity cylinder (~ = 0) at every time step.
Following the start of fluid motion, very small time steps (~t = 10- 3 )
are taken since the time variation of the vorticity field is quite fast.
However, as time increases the time step was gradually increased until
reaching l:l.t = 0.025. The number of points in the ~ direction is taken
as 320 with a grid size of l:l.~ = 0.025. This sets the outer boundary of
computational domain at a physical distance of approximately 500 times
the radius of the cylinder for R = 855 and t = 60. Thus, the boundary is
168

sufficiently far away so that the application of the boundary conditions

(5) and (6) does not affect the solution in the viscous region near the
cylinder surface. The numerical solution starts with two terms only in
the Fourier series (8) and (9) and one more term is added when the last
term in the series exceeds 10- 4 The maximum number of terms depends
on the Reynolds number, forcing frequency and amplitude of oscillation.
The number reached a maximum of N = 60 in all cases considered in
this work.

We first performed some accuracy checks on our numerical scheme by

carrying out the computations for the special case when the cylinder
oscillations are in the streamwise ('fJ = 0) direction. In general, for the
cylinder oscillation in the streamwise direction the perturbation pro-
duced by the cylinder will be a symmetrical one, while the naturally
occurring mode of large-scale vortex formation and its upstream influ-
ence is antisymmetrical. Consequently, one expects, in general, com-
petition between the symmetrical and antisymmetrical modes; under
certain conditions, one mode may prevail producing synchronization of
the near-wake flow structure with the cylinder motion. The tests we
carried out in the case of 'fJ = 0 verify the existence of selected modes
of experimental work by Ongoren and Rockwell [6] when the amplitude
ratio is A = 0.26, and f / fo = 0.5, 1.0, 1.8, .0, 2.2, 3, 4 [see Ongoren and
Rockwell (1988b) - figure 2 on page 229].

3. NUMERICAL SIMULATIONS
Runs are carried out for the cases of R = 855 f / fo = 0.5, 1.0, 2.0, 3, 4
when A = 0.26 and 'fJ = 30, 45, 60, 75.

If the cylinder is excited at an angle other than the cross-stream or

streamwise direction, then there is mixed-mode excitation: the pertur-
bation from the cylinder motion contains both symmetrical and anti-
symmetrical contributions and the potential for exciting both types of
modes. It may be possible for one of the modes to emerge as the dom-
inant one, producing a synchronized near-wake structure such that the
instantaneous flow structure locks to the body motion. The results of
this study shows that vortex shedding from the cylinder under consid-
eration is of two types: synchronized and non-synchronized.

For the case of f / fo = 0.5, the large scale vortex formation occurs
in the asymmetric mode for all values of 'fJ. Two opposite vortices are
alternatively shed from the upper and lower surfaces of the cylinder per
 169

half cycle resulting vortex shedding frequency equal to the natural one
i.e. 10; this is the classical mode of vortex shedding leading to forma-
tion of the Karman street. Typical equivorticity and streamline plots for
non-lock regime are shown in Figure 2 for the case of 1/10 = 0.5 when
170

OT

ail o

Tl4

3T/8

~,, '

5T/8

~~ G
3T/4

0
O\J Go

C7l=".
7T/8

Figure 2. Equivorticity lines (left) and sreamlines (right) over one period, T, for
R = 855 and A = 0.26: 1/10 = 0.5 when TJ = 30 0 (T:= 18.18; 6T:S t :S 7T).

77 = 30. Unlike the case of a fixed cylinder in which the shedding vor-
tices are equal in size, the vortices generated in the present case are of
 171

3T/4

c1\ Q o

5T14 Q

oi~ 0
0
0
3T/2
,- Q

oJ
0 0
0 0 OJ
7T/4 ~

Qj

0 0
0 0

o
Figure 3. Equivorticity lines (left) and streamlines (right) over two periods, 2T, for
R = 855 and A = 0.26: f / fo = 2 when TJ = 300 (T ~ 4.55; llT :::; t :::; 13T).

two different size shedding alternately from the upper and lower sides as
shown in Figure 2.
172

Synchronized shedding from the cylinder under consideration can have

either symmetric or asymmetric mode. The results of this study veri-
fy the existence of asymmetric and symmetric synchronized modes de-
scribed by Ongoren and Rockwell (1988b) when the frequency ratio
ranges between 1 and 4 and the amplitude ratio is small but above
the threshold value. Here we present the results only for the cases of
the frequency ratio f / fa = 2.0 and 3.0 when 'rJ = 300 . The case of
f / fa = 2.0 corresponds to a synchronized asymmetric mode: there is
alternate, out-of-phase shedding of vortices from either side of the cylin-
der over two oscillation cycles. Vortices shed at the same frequency of
cylinder oscillation, (f = 2fo) i.e. this asymmetric mode shows period
doubling relative to the classical Karman mode as shown in Figures 3.
This doubling arises from the symmetrical perturbation component in-
duced by the cylinder motion.
The case of f / fa = 3.0 corresponds to a synchronized symmetric mode:
a pair of vortices is shed in phase from both sides of the cylinder dur-
ing one oscillation cycle as shown in Figures 4. The vortices forming at
frequency f preserve their symmetry for a short distance downstream;

OT

ct~~
, ~ @

T/4
)

0 ~ @
a:~~
01 (QJ
T12

a~ 0iY 0 0 G

3T/4

tJ G ~
Cl~
T

~ ~
a:~O
G @

Figure 4. Equivorticity lines (left) and streamlines (right) over one period, T, for
R = 855 and A = 0.26: f / fo = 3 when 'fJ = 30 (T ::: 3.03; 40.4T :s; t :s; 41.4T).
 173

a pair of almost symmetrical vortices are shown at the beginning and at

the end of the cycle.

References
Badr, H.M., and Dennis, S.C.R. (1985). Time-dependent viscous flow past an im-
pulsively started rotating and translating circular cylinder. J. Fluid Mech. 158,
447- 488 .
Bearman, P.W. (1984). Vortex shedding from oscillating bluff bodies. Ann. Rev. Fluid
Mech. 16, 195-222.
Berger, E. and Willie, R. (1972). Periodic flow phenomena. Ann. Rev. Fluid Mech. 4,
313-340.
Blackburn, H.M., and Henderson, R .D. (1999). A study of two-dimensional flow past
an oscillating cylinder. J. Fluid Mech. 385, 255-286.
Cetiner, 0., and Rockwell, D. (2001). Streamwise oscillations of a cylinder in a steady
current. Part 1. Locked-on states of vortex formation and loading. J. Fluid Mech.
427,1- 28.
Griffin, O.M., and Hall, M.S. (1991) . Review - Vortex shedding lock-on and flow
control in bluff body wakes. ASME J. Fluids Engng. 113, 526-537.
Nguyen, P., and Kocabiyik, S. (1997). On a translating and transversely oscillating
cylinder: Part 1: The effect of the Strouhal number on the hydrodynamic forces
and the near-wake structure. Ocean Engng. 24 677-693.
Kocabiyik, S., and Mahfouz, F.M. (2002). Numerical simulation of the flow induced
by a circular cylinder subject to forced oscillations. Proceedings of The 3rd Inter-
national Conference on Engineering Computational Technology, B.H.V. Topping
and Z. Bittnar (Editors), Civil-Comp Press, Stirling, Scotland, paper no. 26 1- 12.
Kocabiyik, S., and Al-Mdallal, Q. (2003). Numerical simulation of laminar flow past
an oscillating circular cylinder. Proceedings of the Eleventh Annual Conference of
the Computational Fluid Dynamics Society of Canada - in press.
Ongoren, A., and Rockwell, D. (1988a). Flow structure from an oscillating cylinder:
Part 1. Mechanisms of phase shift and recovery in the near wake. J. Fluid M echo
191 197- 223.
Ongoren, A. & Rockwell, D. (1988b). Flow structure from an oscillating cylinder: Part
2. Mode competition in the near-wake. J. Fluid Mech. 191 225- 245.
Rockwell, D. (1998) . Vortex-body interactions. Ann. Rev. Fluid Mech. 30 199-229.
Sarpkaya, T. (1979). Vortex-Induced Oscillations. A Selective Review. J. Appl. Mech .
46 241- 256.
Sumer, B.M., and Freds0e, J. (1997). Hydrodynamics Around Cylindrical Structures.
World Scientific.
Williamson, C.H.K., and Roshko, A. (1988). Vortex formation in the wake of an
oscillating cylinder, J. Fluids and Structures 2 355- 381.
 THREE DIMENSIONAL MODELING OF FLOW
INDUCED VIBRATION FOR AN ELASTIC
CYLINDER IN A CROSS FLOW

Y. LIU, R. M. C. SO AND C. H. ZHANG

Abstract: A fully three dimensional flow induced vibration problem for an elastic cylinder in a cross
flow has been calculated. The cylinder is treated as a flexible cable and is allowed to freely vibrate at the
streamwise and transverse directions while the two span wise ends are tightly fixed at two end plates.
The fluid flows are governed by the three dimensional unsteady Navier-Stokes equations that are
numerically solved by a finite volume method on unstructured moving grid. The cable displacement
components at the stream wise and transverse directions are described by linear wave equations of
second order. The coupling between fluid motion and cable response at each time step is treated in an
iterative way so that the fluid and structure coupling can be accounted for properly in affordable PC
resources. Calculations are presented in this work for the first two synchronization states, i.e., St = /0 and
St = 2.0/0 in which St and /0 are the dimensionless shedding frequency of according stationary cylinder
and fundamental natural frequency of cable. Mass ratio m = 10, cylinder aspect ratio a = 16 and
Reynolds number Re = 100 are selected at which the wake flow has been proven to be three dimensional
laminar. Numerical results show that the cross flow locks the St =/0cable motion into the first vibration
mode while locks the St = 2.00/0 cylinder motion into the second vibration mode.

1. INTRODUCTION

Numerous studies have been conducted into the vortex induced vibration of circular
cylinder in cross flow since the early work of Feng (1968). Griffin (1992) compiled
a group of available data from different experimental setups and plotted the cross
flow vibration amplitude vs. mass damping parameter in a single figure . They
found that the vibration amplitude in any real experiment never exceeds 1.5
diameter. It is generally accept now that this self-limiting process of the vibration
amplitude is associated with the phase difference between the fluid force and the
cylinder motion.
In the numerical modeling of fluid and structure interaction problem of circular
cylinder, the coupling between the fluid and cylinder has been modeled by different
investigators. From the view of fluid dynamicist, the structural properties of
cylinder are generally simplified as a mass-spring-damper in which the elasticity
and damping of structure are simplified by a spring and a damper; see the
calculations of Slaouti & Stansby (1994), Zhao et al (1999), Mittal & Kumar
(1999), Mendes & Branco (1999) and So et al (2001). From the other view of
structure analyst, the fluid behaviors are simply denoted by wake oscillator model,
or the unsteady fluid force vs. vibration amplitude relationship is fitted by available

175
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 175-185.
2003 Kluwer Academic Publishers.
176

experimental data; see reviews of Blevins (1990). In fluid and structure interaction
problem, the fluid force on the structure plays a key role in the structure response
property. All of these models, however, have not treated rigorously the coupling
between the fluid and structure. The real three dimensional phenomenon such as
the deformation and bending of flexible cylinder, the variation in fluid load along
the cylinder span due to end flow conditions, and the three dimensional wake have
not taken into accounts.
Newman & Karniadakis (1997) conducted a DNS study of the flow past a freely
vibrating cable. Their work is one of the first studies that deal with the fully
coupled problem of vortex induced vibration of bluff bodies. They have adopted a
spectral element technique in the cross plane and a Fourior expansion in the
spanwise direction. This numerical strategy alleviates the computational burden,
however, entails that the flow and response be periodic in the spanwise direction
and the wavelength be prior known. This implies that their numerical model can be
applicable only for the midspan region of a vibrating cylinder of very long span, or
periodically supported cylinder.
In this paper, a fully three dimensional fluid-structure interaction problem for a
cable in a cross flow has been studied. The fluid flow is governed by 3D unsteady
Navier-Stokes equations under Arbitrary Lagrange - Euler (ALE) frame. The
cylinder response is governed by linear wave equations of second order. The
numerical model utilizes a finite volume method on unstructured moving mesh to
accommodate the motion of cylinder. We report here the simulation results of the
freely vibrating cable in cross flow at two synchronization conditions: St = fa and St
= 2.010. It will be shown that the response mode is highly dependent on the natural
frequency that the shedding frequency is approaching at synchronization.

2. MATHEMATICAL FORMULATION

2.1. Governing Equations

The flow equations are three dimensional, incompressible, unsteady Navier Stokes
equations coupled with a continuum equation. Under ALE frame, they can be
written as:

Veu=O (1)
Dm u 1 2
--+Ve(u-u )u=-Vp+-V U (2)
Dt m Re

in which u (u, v, w) and Urn are the motion velocity of fluid and ALE volume
relative to inertia Cartesian coordinate system (x, y, z), p is the static pressure, Re
D
is Reynolds number and --'!!.- is the time derivative of transport quantity contained
Dt
 177

in the ALE volume. The x, y, z axis is in the streamwise, transverse and spanwise
directions, respectively, and the origin is fixed in space and located at the center of
cylinder when it is stationary.
The cylinder response is governed by linear wave equations of second order:

(3)

in which O(z,t) = (ox,b'y) is the streamwise and transverse displacement

component of cable relative to z axis, a = LID is the aspect ratio, fa
=~T / meD /(2LUiJ is the reduced frequency of cylinder, m = me / pD2 is
the mass ratio. me is the cylinder mass per span length and T is the tension in
cylinder. C F is fluid force coefficients CF(z,t) = (CD' CL ). All ofthe quantities
in equations (1) - (3) have been normalized by cylinder diameter D, fluid density p,
dynamic viscosity Jl and incoming flow velocity Uin . As usual, we still apply the
Strouhal number St = JDIUin to denote the dimensionless vortex shedding frequency
of rigid cylinder. For flexible cylinder, the dimensionless shedding and vibrating
frequency are denoted by Is and Iv.
The freely vibrating cable is bounded by two end plates (Fig. 1). The boundary
conditions are summaried as followings: uniform flow condition is specified at the
inlet; Newman type boundary condition and total mass balance condition are
ensured at the outlet; the flows at transverse boundaries meet symmetry condition;
the flows at two end plates and cylinder surface satisfy non-slip boundary condition.
The initial flow fields are computed from the inlet quantities. For the cylinder
response, we fix tightly the two ends of cable at the end plates and assume the cable
response from stationary state.

2.2. Finite Volume Method

The governing equations (1) - (3) are coupled partial differential equations.
Generally, they have no analytical solutions and can be solved only using
numerical method. We apply an iterative technique at each time step and a finite
volume solver to the equations (1) - (3). For clarity, we employ the superscript n-l,
n, n+ I to denote the quantities at (n-l).1t, n.1t, (n+ 1).1t time step, and * to denote
the intermediate iterative quantities at (n+ I)Llt time step. Suppose we have
obtained the quantities at and prior to n time levels, we attempt to solve the
quantities at n+ 1 time level using the following finite volume method.
178

2.2.1. Finite Volume Methodfor Fluid Flow: First, we recast the equations (1) - (2)
into a general convective - diffusive fonn:

(4)

Second, we adopt finite volume method to discretize the above equation at time
level n + 1about an arbitrary unstructured moving mesh P of volume i1V, surface
A f of velocity Urn:

in which the first term at the left hand side is the nqth backward implicit
differencing fonn of the time derivative. In this work, we apply second order
scheme for all the time and space derivatives, and choose

nq = 2, ao = 1.5, a] = -2.0, a2 = 0.5.

Third, we utilize second order upwind differencing scheme and a flow limiter
for the convective tenn, and second order central differencing scheme for the
diffusive tenn and pressure gradient in equation (5). Being recast into a standard
fonn of finite volume formulation, the final algebraic equations relating ct> at cell
P with its neighboring cell nb at time level n+ 1 are:

A:ct> p = I A!ct> nb + s~ (6)

A<l> =" A<l> + 1.5 pi1V n+1 ,S<l> = S<l> _ 2.0 pi1vn ct>n + 0.5 pi1V n- 1 ct>n-l
PL..nb M T P M i1t

in which Anb is the influential coefficient of neighboring cell nb on cell P and S:

is the sum of the discretized fonn of the source tenn, the higher order convective
tenn and the non-orthogonal diffusive tenn.
Last, we apply SIMPLER algorithm to update the flow velocity u ~ obtained
from equations (6) and pressure P;
equation so that the updated (U~+l ,p;+l)
satisfy simultaneously the discretized momentum and continuity equations.
 179

2.2.2. Finite Volume Method for Cylinder Response: The equations (3) are also
solved by finite volume method at the time level n + 1 and the cylinder slice i using
second order central scheme for the spatial derivative and second order backward
implicit scheme for the temporal derivative. The discretized equations can be recast
into a set oftri- diagonal algebraic equations:

ai = aj-l + aj+l + - -12 ' b C F - - -12 (~n-2

= -- Ui -
2~n-l)
Ui
I:lt 2m I:lt

2.2.3. Fluid Structure Coupling: The fluid and structure coupling is mathematically
formulated by C F in equations (3) and Urn in equations (2). The force coefficients
on each slice of cylinder are calculated by integration the static pressure and the
wall shear stress around cylinder slice circumference:

where the closed integration is performed around the circumference of the cylinder
slice at z, ds and n are the length and outward unit normal of the segmental element.
The mesh moving velocity Urn is governed by Laplacian equations:

(9)

with Urn to be the cylinder vibrating velocity at the cylinder surface and zero else
other boundaries. The Laplacian equations (9) are numerically solved as a special
form of convective and diffusive equation (6).

2.2.4. FVM Solution Procedures for Vortex Induced Vibration: We utilize an

iterative technique at each time step so that the fluid and structure behaviors satisfy
simultaneously the governing equations. The numerical solution procedures are
summarized as follows.
(1) Solve the discretized flow equations (6) to obtain the flow fields (U~+l , p;+l);
(2) Calculate the force coefficients C F using equations (8);
180

(3) Solve the discretized response equation (7) for the cylinder displacement 0;+1 ;
(4) Solve the equation (9) for the mesh moving velocity u:+ 1;

(5) Repeat stages (1) - (4) at time level n+ 1 until convergence criteria are satisfied;
The convergence criterion at each time step is that the sum of absolute mass
imbalance over all control volumes deceases to 0.001 mass inflow rate.

2.3. Numerical Parameters

The flow domain is extended from the coordinate origin to -12.5D upstream and
22.5D downstream along the x axis, -SD and SD along the y axis, and -O.5L and
0.5L along the z axis; see Fig. I. The flow domain is discretized first into 6500
quadrangles in the cross plane using 0 type mesh round the cylinder and H type
mesh in the far region, then slacked non-uniformly 64 layers along the spanwise
direction. The resultant mesh contains totally 416,000 hexahedral elements. The
nondimensional time step is M = 0.02. Generally at least 20,000 time steps, i.e.,
400 time units, are performed consuming about 30 CPU days for a case on a Dell
PC with 1.7G CPU and IG RDRAM memory. A finer mesh of96 layers along the
cylinder span and smaller time step /).t = 0.01 have been tested and the results show
that the difference in calculated 5y' at midspan varies within 5%.

end plate ~=o

On
Ov

::~ - -- _~==oo
IIfvw
w=ollf ~~-++-________________________~On~
end plate

Fig. 1 Problem description

3. NUMERICAL RESULTS
3.1. Flexible Cable

In this section, we shall study the first two multipliers of synchronization: St = /0

and St = 2.0/0, i.e.,/o = 0.142 and 0.071. The other governing parameters are a =
16, Re = 100 and m = 10.0. Thus, the cable of St = 2.0/0 endures less tension than
that of St = /0 from the relationship between tension T and /0.
Let us first consider the statistics of dynamic response and fluid load of freely
vibrating cable vs. cylinder span position z in Fig. 2. The drag force drives the
equilibrium axis of flexible cable to a new position, shifting cable center to 0.178
and 0.703 diameter downstream at St = /0 and at St = 2.0/0, respectively. As
expected, the less tension T endured by the cable of St = 2.0/0 leads to larger
 181

streamwise displacement fix to balance the drag force than the cable of St =fa (Fig.
2a).
The calculations confirm that the streamwise vibration amplitude is much less
than the transverse component (Fig. 2b and 2e). This finding has also been well
validated by the calculations of the mass - spring - damper system, e.g, by So et al.
0.02

A St=1.00fo
A St=1.00fo
0 St=2.00fo
0.75 0.0 15 0 St=2.00fo

8X'0.01

(a) (b)
0.5

A St=1.00fo
0.4 0 St=2.00fo

c~

A St=1.00fo
0 St=2.00fo

0.8 -4 0
Z

(c) (d)
182

St=1.00fo
St=1.00fo o St=2.00fo
0.75
0.75 o St=2.00fo

(e) (f)
Fig. 2 The calculated statistics vs. cylinder span

(2001) and experiments of elastically mounted rigid cylinder, e.g, by Khalak &
Williamson (1996). The maximum &' is about 0.005 and 0.009 diameter at St = 10
and at St = 2.010 while the corresponding maximum ~' is about 0.45 and 0.55
diameter. However, the mean streamwise displacement 8X has the same order of
magnitude as ~' . This means that the streamwise motion of cable also plays
important role in the dynamic response of cable. Both the stream wise and
transverse displacement are non-uniform along the cable span, leading to the three
dimensional deformation and curvature of flexible cable in cross flow.
The free vibrations of cable at the synchronization states not only increase
significantly the fluid load on cable, but also enhance its non-uniformity along the
cable span (Fig. 2c, 2d and 2f). The enhanced fluid load, together with its non-
uniformity along the cable span in turn violate the cable dynamic response at
synchronization and give the risk of excess structural stress or mechanical fatigue.

TI4

(a) St =10 (b) St = 2.010

Fig. 3 y-component offreely vibrating cable profile within halfa vibration period.
 183

The cross flow locks the flexible cable of St = 2.0/0 into the second mode
vibration, while locks that of St =/0 into the first mode vibration, as shown in Fig. 3
for the transverse trajectories of cable axis within half a vibration period. At St = /0
synchronization, the vibrating cable moves as a bow shape, with each cable
segment passing through the equilibrium position or arriving at the maximum
displacement at the same time. At St = 2.0/0 synchronization, the trajectory of
vibrating cable can be fit well by a period of sine function with almost negligible
displacement at the midspan node and maximum displacement amplitude at the z =
-a/4 and a/4 anti-nodes. The maximum peak-to-peak amplitude of the transverse
displacement is 1.27 and 1.50 diameter at St = /0 and St = 2.0/0, respectively. The
large vibration amplitude of flexible cable at St = 2.0/0 is in sharp contrast to the
vanishing vibration amplitude outside St - /0 synchronization region observed in a
mass-spring-damper system or an elastically mounted rigid cylinder in a cross flow.
This shows that the complex three dimensional deformation and curvature of
flexible cable must be considered in the cable dynamics.
Fig. 4 shows the time series of the cable response and fluid load at anti-node.
The dimensionless dominant frequency of CD, CL , OX and & is 0.283, 0.142, 0.283
and 0.142 at St = /0, and 0.279, 0.182, 0.278 and 0.182 at St = 2.0/0. At St = /0
synchronization state, both the frequency of vortex shedding and transverse
displacement approach closely the fundamental natural frequency of cable, and are
approximately half the frequency of drag force and streamwise displacement. At St
= 2.0/0 synchronization, i.e., fa = 0.071, both the vortex shedding and the cable
vibrating frequencies are locked into the frequency of cable-fluid system which
deviates obviously any natural frequency of flexible cable. This shows the complex
fluid and structure interaction of flexible cable at St = 2.0/0 synchronization state
should be considered properly only using fully three dimensional model.

0.5

-0.5

t t
(a) St =10 (b) St = 2.010
Fig. 4 Time history of response andforce at the anti-node offreely vibrating cable
184
5. CONCLUSION

A fully three dimensional flow induced vibration problem for an elastic cable in a
cross flow has been calculated using finite volume method. The fluid flow is
governed by the three dimensional unsteady Navier-Stokes equations and
numerically solved on unstructured moving grid. The cable response is governed by
the wave equations of second order. An iterative technique is used to solve the fluid
and cable equations at each time step. Calculations are carried out for the rigid
cylinder and flexible cable at St = 10 and St = 10 synchronization states while the
mass ratio m = 10.0, the aspect ratio a = 16.0 and the Reynolds number Re = 100.
The numerical results are summarized as follows.
(l). The synchronization between the cross flow and cylinder response
enlarges substantially the fluid loads and their variations, especially at St ~ 2/0
synchronization.
(2). The cross flow locks the vibrating cable of St ~ 10 into the first mode
vibration while locks that of St ~ 210 into the second mode vibration.

6. AFFILIATIONS

Mechanical Engineering Department, The Hong Kong Polytechnic University,

Hung Hom, Kowloon, Hong Kong.

7. ACKNOWLEDGEMENT

Support given by the Research Grants Council of the Government of the HKSAR
under Grant No. PolyU 5166101 E and 5172/02E and by the Hong Kong
Polytechnic University under Central Research Grant No. A-PD75 and A-PE53 is
gratefully acknowledged.

8. REFERENCES
Blevins, R. D. 1990 Flow - induced vibrations. New York: Van Nostrand Reinhold.
Feng, e. c. 1968 The measurement of vortex induced effects in flow past stationary and oscillating
circular and d-section cylinders. M. A. Sc. thesis, University of British Columbia, Vancourver, B.
e., Canada.
Khalak, A. & Williamson, e. H. K. 1996 Dynamics of a hydroelastic cylinder with very low mass and
damping. J. Fluids Struct. 10, 455-472.
Mendes, P. A & Branco, F. A. 1999 Analysis of fluid structure interaction by an arbitrary Langrangian
Eulerian finite element formulation. Int. 1. Num. Meth. Fluids. 30, 897-919.
Mignolet, M. P. & Red-Horse, J. R. 1994 ARMA identification of vibrating structures: Model and
model order estimation. Proceedings of AIAAJASME 35'h Structures, Structural dynamics, and
materials conference, La Jolla, CA, 19-22 April, 3366-3374.
Mittal, S. & Kumar, V. 1999 Finite element study of vortex-induced cross-flow and in-line oscillations
of a circular cylinder at low Reynolds numbers. IntI. 1. Num. Meth. Fluids. 31, 1087-1120.
Newman, D. & Kamiadakis, G. E. 1997 A direct numerical simulation study of flow past a freely
vibrating cable. 1. Fluid Mech. 344, 95-136.
 185

Slaouti, A & Stansby, P. K. 1994 Forced oscillation and dynamics response of a cylinder in a current
investigation by the vortex method. Proc. BOSS'94 Conf MIT, 645-654.
So, R. M. c., Liu, Y., Chan S. T. & Lam, K. 2001 Numerical studies of a freely vibrating cylinder in a
cross - flow. J. Fluids Struct. 15,845-866.
Zhang, C. H., Su, Z. D., Liu, Y., So, R. M. C. 2003 The influence of aspect ratio on three-dimensional
wake of circular cylinders using finite volume method and lattice Boltzmann method. Compo Mech
(accepted).
Zhou, C. Y., So, R. M. C. & Lam, K. 1999 Vortex-induced vibrations of elastic circular cylinders. J
Fluids Struct. 13, 165-189.
 CORRELATION LENGTH AND
FORCE PHASING OF A RIGID
CYLINDER SUBJECT TO VIV
D. LUCOR, J. FOO and G.E. KARNIADAKIS
Division of Applied Mathematics
Brown University
email:gk@cfm.brown.edu

Abstract: We present direct numerical simulations (DN8) of uniform flow

at subcritical Reynolds number past a flexibly-mounted rigid cylinder subject
to vortex-induced vibrations (VIV). We investigate different nominal reduced
velocities near or in the region of maximum amplitude response for a small
mass ratio and zero structural damping. We compute the correlation length
of the flow quantities in the near wake and relate it to the force correlations
along the cylinder. We perform a complex demodulation analysis to quantify
the phase difference between structural displacement and forces. There exists
a reduced velocity region near the 8trouhal frequency, for which a sharp drop
in the spanwise correlation of the flow quantities in the near wake and the
forces is observed. This decrease in the spanwise correlation corresponds to a
poor phasing between displacement and forces but it does not preclude a large
response from the structure.

1 INTRODUCTION
The apparently simple case of VIV of an elastically mounted rigid cylinder
constrained to move tranversely to a uniform flow remains of practical and the-
oretical importance. The distinction between the different types of response as
a function of the nominal reduced velocity depending on whether the cylinder
has a high or low mass-damping parameter has been described in the literature
[1, 2, 3] and shows good agreement. In the case of a low mass-damping, it
seems that there exists three different branches of response: an initial branch
associated with a 28 vortex wake mode, followed by upper and lower branch-
es associated with a 2P vortex wake mode [1, 2]. The complex mechanisms
that induce the mode transitions between the different branches remain to be
explained. In particular, the mechanisms that induce the well-known 'phase-
jump' (jump in the phase between the cylinder displacement and the lift force)
that occurs during the transition from the upper to the lower branch are not
well understood. Other experiments indicate the existence of a reduced velocity
region containing the 8trouhal frequency, for which a sharp drop in the spanwise
correlation ofthe flow quantities in the wake and the forces is observed [3]. This
region stands mainly on the right side end of the upper branch (large amplitude
response) near the transition between the upper and the lower branch (small

187
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 187-199.
2003 Kluwer Academic Publishers.
188

amplitude response). Therefore, this decrease in the spanwise correlation does

not preclude a large response from the structure. The study of the forces and
phasing between cylinder displacement and forces in this region are key to the
understanding of the mode transition. We propose to investigate the existence
of this region of poor correlation using correlation length computations and
phase analysis of DNS results.
The fluctuations of the lift forces acting on a free rigid cylinder subject to
vortex-induced vibrations (VIV) depend on the degree of three dimensionality
present in the near wake of the body. The spanwise correlation length of the
flow gives an accurate measure of this three dimensionality and consequently
gives some indications of the magnitude of the cylinder response amplitude.
However, this information is incomplete as it does not supply a measure of the
phase difference between structure motion and lift forces for instance.
Experimental studies have been limited so far in terms of force measure-
ments. They are limited to measurements at both ends of the cylinder [2, 3] or
on an elemental slice of the cylinder [4, 5] (stationary cylinder). Direct numeri-
cal simulations provide with an alternative tool capable of accurate spatial and
temporal representations of both the pressure and viscous force contributions.
It is therefore possible to compute the correlation length of the forces along the
spanwise direction and to relate it to the correlation length of the flow quantities
in the near wake.
In studies of VIV the value of spanwise correlation length is very impor-
tant as many empirical models rely on it [6]. It is also important in numerical
simulation studies as it provides guidance for the choice of spanwise numerical
resolution; mesh refinements along the spanwise direction are often overlooked
as the strongest flow gradients occur along the streamwise and crossflow direc-
tions. However, related studies and experimental measurements of correlation
length are relatively few in number [7, 8, 9, 10]. They generally agree with the
idea that increasing the amplitude of motion, either through forced or free vi-
brations, increases spanwise coherence. For stationary cylinders, experimental
measurements of correlation length based on the autocorrelation function were
obtained only recently [11, 12]. For moving, rigid, free cylinders, detailed mea-
surements of cross-correlation between lift forces measured at the two ends of
the cylinder can be found in [3]. The influence of nominal reduced velocity on
the correlation length and phasing of the near wake flow remains to be explored.

2 SIMULATION PARAMETERS AND

FORMULATION

Here, we present direct numerical simulations (DNS) results of vortex-induced

vibrations of a smooth rigid cylinder with aspect ratio Lid = 26 and mass ratio
(cylinder mass over displaced fluid mass) m = 2 using spectral/hp elements
method [13]. The Reynolds number is taken to be Re = Udlv = 1000. It is in
the sub critical range resulting in a turbulent near-wake. We set the structural
damping to be zero as we are interested in the maximal response of the system.
We consider that this choice of mass ratio and structural damping puts us
in the low mass-damping parameter range. We only consider the dominant
motion in the cross-flow direction (y-direction) and we preclude any motion of
the structure in the streamwise direction (x-direction). The cylinder is rigid and
thus its motion has no spanwise z-dependence. The governing equations are the
incompressible Navier-Stokes equations coupled with the structural dynamical
 189

equation. The cylinder is represented by a single degree of freedom viscously

damped second-order oscillator subject to the external hydrodynamic forcing,
i.e.

*
ij(t) + ~re i](t) + ~: 7](t) = C;~), 7](0) = 7]0 and i](0) = i]o (1)

where 7] represents the crossflow cylinder response, Vr = is the reduced veloc-

ity based on the free-stream velocity U and the natural frequency f in vacuum of
e
the structure, is the structural damping coefficient, and CL(t) is the spanwise-
averaged local lift coefficient. All of the variables are non-dimensionalized with
the cylinder diameter d and the free-stream inflow velocity U.
The coupled Navier-Stokesjstructure dynamics equations are discretized in
space using a spectraljhp element method that employs an unstructured hy-
brid grid with hp refinement in the (x,y)-plane and Fourier expansions in the
z-direction (cylinder axis) with a dealiasing 3j2 rule. Variable spectral order
up to order p = 11 is used for each element which gives variable p-refinement
in space depending on the regions of different dynamics [14J. The parallel code
N e",/aT F is employed in all simulations [13J. A boundary-fitted coordi-
nate system is employed similar to the laminar flow simulations in Newman
& Karniadakis [15], which has been validated against an Arbitrary Lagrangian
Euler (ALE) formulation [14J that was also developed for moving domains [16].
The computational domain for the (x, y)-plane extends 69d (cylinder diameters)
downstream and 22d in front of the cylinder; it extends 22d above and below
it. A hybrid mesh, refined around the cylinder, with a total of 1,018 elements
is used. Also, 64 z-planes (32 Fourier modes) are used along the spanwise di-
rection. Periodic boundary conditions are imposed at the two ends along the
cylinder axis. This is equivalent to treating the structure as infinitely long, and
then employ (free) periodic boundary conditions on a piece of finite length. An
implicit second-order Newmark integration scheme was used to solve for the
structure [17].

2.1 Correlation Lengths

We test four different reduced velocities for the natural frequency of the oscilla-
tor. We choose our reference reduced velocity Vr = Vref = 4.18 to be based on
the Strouhal frequency of the two-dimensional stationary cylinder wake [14]. We
choose the other three reduced velocities to be above and below the reference
value.
We define our autocorrelation function as follows:

R
uu
(Ix
"Y
)=R (l-x
Uti, ,y,
t)=(-bL~lU(X,y,z,t)U(X,y'Z-I,t)).
1 "\:",,M 2( )
(2)
ML....z=lU x,y,z,t

The autocorrelation function is therefore computed by shifting the signal

to obtain u(x, y, z - I, t), multiplying it by the unshifted sequence u(x, y, z, t)
then summing all the values of the product and normalizing. The bar denotes
the final averaging over time. The signal u(x, y, z, t) is the fluctuation obtained
after we subtract the mean quantity, i.e. averaged value of u(x, y, z, t) in time
for each spanwise (z) location at the (x,y) point. We allow the shift I to be:
M
1= [0, dzjd, 2dzjd, ... , Tdzjdj with dz = LjM. (3)
190

We have M = 64 points in the z-direction. Spatial spanwise periodicity of

the quantity u(x, y, z, t) is used to wrap up the signal in order to keep the
same number of terms in the summation for all shifts. With our definition,
the autocorrelation is symmetric around I = O. We use this autocorrelation
function to compute hydrodynamic force correlations on the structure as well
as velocity correlations at two locations in the near-wake. One point is on the
centerline (x/d = 3.0;y/d = 0) and the other one cylinder diameter above it
(x/d = 3.0;y/d = 1.0).
Taking the time average of the correlation coefficients might mask some of the
fluctuations of the flow quantities in time. Similarly, computing the spectrum
of the velocity components in the wake will not provide a good description of
the signals. In particular, in the case of multi-frequency response or beating
phenomena, short time integration of the correlation coefficient or phase drift
angle analysis might be necessary [10].

2.2 Complex Demodulation Analysis

Multi-frequency responses and beating phenomena demand a time-varying de-

scription of the phase difference between the cylinder displacement and the lift
force . To quantify the phase difference we employ complex demodulation anal-
ysis, which is a more general approach than harmonic analysis in dealing with
non-exact periodic time series [18]. A complex demodulation of a time series
CL(tn, z) (lift coefficient time series at location z in the spanwise direction)
with a dominant frequency component ACL (obtained by taking the FFT of the
span-averaged signal) will give a time varying amplitude RCL (t, z) and phase
<J>CL(t, z) such that:

(4)
Time-dependent amplitude and phase of the signal at time t are determined only
by the signal in the neighborhood of t. The procedure uses a linear filtering that
can be tuned by choosing some free parameters that control the width and the
shape of the filter. This process is repeated for the time-series at each z-location
along the span. Similarly, we would have for the cylinder displacement:

(5)

However, that here we have no z-dependence because the cylinder is rigid. We

define the phase difference II <J> as being:

(6)

3 RESULTS
We investigate four different reduced velocities Vr = [3.76,4.18,4.62,4.99] for
the oscillator. These values should be immediately adjacent (from below) or
included in the region of poor correlation [3]. The idea is to start from the
region of high correlation with Vr = 3.76 and increase the value of Vr toward
Vr = 4.99, referred as CaseIV, and see if we experience a drop in the spanwise
correlation of the flow. These nominal reduced velocities are defined based on
the natural frequency f of the structure in vacuum. The reference reduced
velocity Vr = Vre ! = 4.18, referred as CaseII, was already investigated [14] but
for a somewhat shorter cylinder (L/d = 41T).
 191

3.1 Velocity Correlation Coefficients

The autocorrelation function IRuul (see figure 1, first row) for the centerline
point becomes very small and close to zero around dz/d = 7.0 for all cases.
The function Ruu becomes negative for dz/d > 7.0 for the case of Vr = 3.76
(Casel). For the off-centerline point, only Casel exhibits a larger correlation
than the other cases, with an almost constant value of 0.25. The other cases
drop quickly to small values, and Casell is the only one to present negative
values of Ruu for shifts in the range dz/d E [1.8; 3.5] . This is consistent with
the results by Evangelinos [14].
The autocorrelation function IRvvl (see figure 1, second row) indicates a
high degree of correlation which is expected for a rigid cylinder subject to VIV
with reduced velocities close to the Strouhal frequency. It is very clear for
both the centerline and off-centerline points that an increase of the reduced
velocity, in this case from Vr = 3.76 to Vr = 4.99, drastically decreases the
spanwise correlation of the cross-flow velocity in the wake of the cylinder. For
the centerline point, Casel shows higher correlation than Casell across the entire
domain but the difference is more pronounced for dz / d < 6. The autocorrelation
function IRvv I for CasellI (corresponding to Vr = 4.62) and CaselV show similar
decay for shifts smaller than dz / d ~ 4. Then they deviate considerably and
Rvv for CaselV takes negative values for dz/d E [4.0; 9.0] . Eventually the two
functions reach a similar value of Rvv = 0.1 for dz/d ~ 11.
Finally, the values for IRwwl (see figure 1, last row) are comparatively much
smaller for all reduced velocity cases, especially in the case of the centerline
point where they drop to very small values after dz / d > 1.0. For the case of
the off-centerline point, Rww become negative for all tested reduced velocities
for dz/d E [1; 5] before to go back to very small values around zero. This was
not the case with a shorter cylinder for Casell [14].
Overall, these results are in good qualitative agreement with experimental re-
sults of wake correlation (F. S. Hover, private communication). Also, the results
for Casell are in good agreement with the experimental results for oscillating
rigid cylinders [7] .

3.2 Force Correlation Coefficients

We compute the average one-tenth highest amplitude of response Amax/d for

the different reduced velocities , see Table 1. The response is noticeably larger
for the values of reduced velocities larger than Casel. Surprisingly, the response
remains large for CaselV even if the wake correlation coefficients present a sharp
drop as described in the previous section. Regarding the magnitude of the lift
forces, we notice a decrease of the fluctuations of the span-averaged lift coeffi-
cient as we increase the reduced velocity. The time evolutions of the average lift
force for Casel and Casell are almost equivalent but the forces for CasellI and
particularly CaselV present a clear weakening and exhibit a beating phenomena.
The maximum instantaneous lift amplitude is obtained for Casell and the mini-
mum for Casel. Generally, if the cylinder is long compared to the typical length
over which the correlation coefficients remain large, not all vortices cause forces
in phase with each other, and the net exciting force is smaller. Consequently we
also compute the force correlation coefficients in a similar manner as the wake
correlation coefficients . Figure 2 shows the force correlation coefficients. The
left plot shows the lift correlation coefficients and the right plot shows the drag
correlation coefficients. Forces are integrated quantities of the flow; therefore
192
Canlarlina point: xld .. 3.0, yld .. O lOO .-_~_ __Off,~_'_'._rlin_ .:,.po_;,_t:X_Id_
- 3_.0;.: -__'_
.,,Id .0~_ _~-,
1 0,-_~ __~ _ _ ~_~-'-_~__~""

Canhlrlilla poin!: Xld_3.0, y/d_O

10t--~--~---'~-~-'---~--==~~ lO.-_~ __ __ ~ ~_~ __ __
Off centerline point: xlda3.0, y/d.,.O
~ ~--,

. dud
12

Figure 1: Streamwise, cross-flow and spanwise velocity autocorrelation functions

in the near-wake at a centerline point (x/d = 3;y/d = 0) (left column) and an
off-centerline point (x/d = 3; y/d = 1) (right column), for four different reduced
velocities.
 193

Vr Amax/d
3.76 0.599
4.18 0.736
4.62 0.766
4.99 0.76

Table 1: Average one-tenth highest amplitude of response Amax/d versus nom-

inal reduced velocity Vr .

we expect to have larger and smoother correlation coefficients than in the case
of velocity correlation computed at some pinpoint location in the wake. The
results are very similar to the ones for the correlation in the near wake, in the
sense that there exists a clear drop in the spanwise correlation of both drag and
lift forces as the reduced velocity is increased.
Overall, these results are in good qualitative agreement with experimental
results of cross-correlation coefficients between forces measured at the two ends
of a free rigid cylinder in the same reduced velocity range [3] .

0.3

0.2

0.1

0~--~--~--~6--~~~1~0--~12~ 0~------~---76--~----'~
0 --~12~
dUd dV'

Figure 2: Force correlation coefficients for four different reduced velocities. Lift
coefficient autocorrelation function (left plot). Drag coefficient autocorrelation
function (right plot).

3.3 Phasing Analysis

We perform complex demodulation analysis of the cylinder displacement and

lift force signals for Casell and CaselV. Our goal is to establish a relationship
between force amplitude, cylinder displacement and phase difference between
the two signals. This method turns out to be very useful in particular for multi-
frequency response system presenting a beating phenomena.
Figure 3 shows the results for Casell. Figure 3-(a) and 3-(b) represent the
isocontours of the cylinder cross-flow displacement 7](t, z) and its correspond-
ing demodulated amplitUde signal respectively. The demodulation frequency is
>"'1 = 0.19672. We see that the method isolates the region of larger response but
the fluctuations are small in this case as the beating is not very pronounced.
194

Similarly, Figure 3-(d) and 3-(c) represent the isocontours of the lift coefficient
CL(t, z) and its corresponding demodulated amplitude signal respectively. The
demodulation frequency remains the same with ACL = 0.19672. The correspon-
dence between the two plots is striking and we can see regions of almost zero
amplitude on figure 3-(c) that correspond to regions of very small forcing on
figure 3-(d). For instance, there are two spots with low (blue color) amplitude
values, around (t ~ 400,z/d ~ 9) and around (t ~ 435,z/d ~ 10) that match
the same locations on figure 3-(d). Finally, figure 3-(e) shows the phase dif-
ference in radians mod 271" between the two demodulated signals. For most of
the domain, cylinder displacement and lift forces are in phase and A<I?(t, z) ~ 0,
except at three locations, and in particular at the location of the two spots men-
tioned above. At these locations, the phase difference is smaller than 71" and a
positive value means that the lift force signal is laging the cylinder displacement
signal. The complex demodulation proves that there is globally a good phasing
between displacement and lift force for CaseII.
Figure 4 shows the results for CaseIV. For this reduced velocity, we expect
the three-dimensionality of the flow to be more developed and consequently, the
forces to be less organized along the span. Figure 4-(a) and 4-(b) represent the
isocontours of the cylinder cross-flow displacement 1)(t,z) and its correspond-
ing demodulated amplitude signal respectively. The demodulation frequency
is Ary = 0.1844. The method isolates the alternance of regions of larger and
smaller response, and fluctuations are larger in this case as the beating is more
pronounced. Similarly, Figure 3-(d) and 3-(c) represent the isocontours of the
lift coefficient Cdt,z) and its corresponding demodulated amplitude signal re-
spectively. The demodulation frequency remains the same. This time, we see
that the maximum lift coefficient is subject to very large modulation. We see a
very clear correspondence between alternated streaks of small and large ampli-
tude on figure 3-(c) corresponding to regions of small and large forcing on figure
3-(d). Finally, figure 3-(e) shows the phase difference in radians mod 271" between
the two demodulated signals. Again, regions of high lift amplitude forces corre-
spond to regions where the cylinder displacement and lift forces are in phase and
A<I?(t, z) ~ O. More interestingly, regions of low lift amplitude forces correspond
to regions where the cylinder displacement and lift forces are out of phase and
A<I?(t, z) ~ 71" which explains the overall drop of spanwise correlation. Indeed,
strong three-dimensionality in the wake (see figure 5) influences the topology
of the vortices inducing forces that are not in phase with each other. The net
exciting force becomes smaller as well as the cylinder motion.
The complex demodulation analysis shows that regions of low lift force coin-
cide with regions of poor phasing. Moreover, it establishes a close relationship
between those forces and the cylinder displacement. Regions of large forces oc-
cur at the time or slightly before regions of large cylinder displacement. More
generally, the method gives a striking visual demonstration of the simultaneity
of the loss of force correlation and the phase angle transition and relates it to
the loss of wake correlation.

4 CONCLUSION
We presented DNS of uniform flow at Reynolds number Re = 1000 past a
flexibly-mounted rigid cylinder with low mass-damping and subject to VIVo We
investigated different nominal reduced velocities near or in the region of max-
imum amplitude response for a small mass ratio and zero structural damping.
We focused in particular on the correlation length of the flow quantities in the
 195
(a)

0.5
20

0
10
- 0.5
0
360 380 400 420 440 460
(b)
0.74
20
0.72

10
0.7

0
360 380 400 420 440 460
(e)
3
20
2

10

o~----~------~~----------------==--~----~----
360 380 400 420 440 460
(d)

~. I"
~

I., 2
20 l
oij
I" o
I'

tr
10
,;
lf~11
P'p'
-2

o
360 380 400 420 440 460
(e)
2

20

~
10
o
o 460
360 380 400 420 440
IU/d

Figure 3: Complex demodulation analysis of Case II (Vr = 4.18). Isocontours of

cross-flow displacement 1] (a). Isocontours of amplitude of demodulated cross-
flow displacement R1J (b). Isocontours of amplitude of demodulated Lift coef-
ficient RCL (c). Isocontours of Lift coefficient CL (d). Isocontours of phase
difference .6.<1> between demodulated cross-flow displacement and demodulated
lift coefficient (e).
196
(a)

0_5
20

o
10
-0_5

620 630 640 650 660 670 680 690 700 710 720
(b)
,...".,- - - - ,,.,,
0 _75
20
0.7

10
0_65

o 0 _6
620 630 640 650 660 670 680 690 700 710 720
(e)
3
20
2

10

-2

620 630 640 650 660 670 680 690 700 710 720
(e)

-5
620 630 640 650 660 670 680 690 700 710 720
tU/d

Figure 4: Complex demodulation analysis of Case IV (Vr = 4.99). Isocon-

tours of cross-flow displacement 'T} (a). Isocontours of amplitude of demodulated
cross-flow displacement RTJ (b). Isocontours of amplitude of demodulated Lift
coefficient RCL (c). Isocontours of Lift coefficient CL (d). Isocontours of phase
difference 6.<1> between demodulated cross-flow displacement and demodulated lift
coefficient (e).
 197

10 5 o
xld xld

Figure 5: Pressure isocontour at value -0.15 in the wake of a free rigid cylinder
at Re=1000. View perpendicular to cylinder axis with inflow coming from left to
right (left picture). View perpendicular to cylinder axis with inflow coming from
right to left (right picture). The instantaneous cylinder position is TJ = -0.32.
198

near wake and relate it to the force correlations along the cylinder. We also per-
formed complex demodulation analysis to quantify the phase difference between
structural displacement and forces.
Those results are in good agreement with the experimental results by Hover
[3]. We confirmed that there exists a reduced velocity range, near the mode
transition from the upper to the lower hysteretic branch, and very close to the
Strouhal frequency, for which a severe drop in the spanwise correlation of the
flow quantities in the near wake and the forces is observed. Surprisingly, how-
ever, the cylinder response remains large in comparison to the response in the
initial or lower branches. Both forces and cylinder response demonstrate beat-
ing. There exists a correspondence between a strongly three-dimensional wake
with low correlation and phase angle transition, and a loss of force correlation.
The vortical topoly in the wake is very complex and consists of multiple irreg-
ular cells inducing forces that are not in phase with each other. Large forces
exist where the phase difference between cylinder displacement and lift force is
close to zero and small forces exist where the phase difference is close to 7]".

5 ACKNOWLEDGEMENTS

This work was supported by the Office of Naval Research. Computations were
performed on the ERDC SGr Origin 3800 and on the ARSC SP Power 3 com-
puter.

References
[1] A. Khalak and C.H.K. Williamson. Motions, forces and mode transitions
in vortex-induced vibrations at low mass-damping. 13:813-851, 1999.
[2] R. Govardhan and C.H.K. Williamson. Modes of vortex formation and fre-
quency response of a freely vibrating cylinder. Journal of Fluid Mechanics,
420:85-130, 2000.
[3] F.S. Hover, A.H. Techet, and M.S. Triantafyllou. Forces on oscillating
uniform and tapered cylinders in crossflow. Journal of Fluid Mechanics,
363:97-114,1998.
[4] G.S West and C.J. Apelt. Measurements of fluctuating pressure and forces
on a circular cylinder in the Reynolds number range 104 to 2.5 X 105 7:227-
244,1993.
[5] J.L.D. Ribeiro. Fluctuating lift and its spanwise correlation on a circualr
cylinder in a smooth and in a turbulent flow. J. of Wind Engineering and
Industrial Aerodynamics, 40:179-198, 1992.

[6] R.D. Blevins. Flow Induced Vibration. Van Nostrand Reinhold Company,
New York, New York, 1990.

[7] G.H. Toebes. The unsteady flow and wake near an oscillating cylinder.
ASME Journal of Basic Engineering, 91:493- 502, 1969.
[8] S.E. Ramberg and O.M. Griffin. The effects of vortex coherence, spacing,
and circulation on the flow-induced forces on vibrating cables and bluff
structures. Naval Research Laboratory Report 7945, 1976.
 199

[9] M. Novak and H. Tanaka. Pressure correlations on a vibrating cylinder.

Cambridge University Press, Heathrow, 1977. Fourth Int. Conf. on Wind
Effects on Buildings and Structures, ed. K. Eaton.
[10] Stefan Szepessy. On the spanwise correlation of vortex shedding from a
circular cylinder at high subcritical reynolds number. Physics of Fluids,
6(7):2406- 2416, 1994.

[11] H. Mansy, P.-M. Yang, and D.R. Williams. Quantitative measurements of

three-dimensional structures in the wake of a circular cylinder. Journal of
Fluid Mechanics, 270:277-296, 1994.
[12] C. Norberg. Flow around a circular cylinder: Aspects of fluctuating lift.
Journal of Fluids and Structures.
[13] G.E. Karniadakis and S.J. Sherwin. Spectral/hp Element Methods for CFD.
Oxford University Press, 1999.
[14] C. Evangelinos. Parallel Simulations of VIV in Turbulent Flow: Linear and
Non-Linear Models. PhD thesis, Division of Applied Mathematics, Brown
University, 1999.
[15] D.J. Newman and G.E. Karniadakis. Simulations of flow past a freely
vibrating cable. J. Fluid Mechanics, 344, 1997.
[16] T.C.E. Warburton and G.E. Karniadakis. Spectral simulation of flow past
a cylinder close to a free surface. In FEDSM97-3689, Proc. Fluids Engi-
neering Division Summer Meeting, Vancouver, 1997.
[17] D.J. Newman. A computational study of fluid/structure interactions: flow-
induced vibrations of a flexible cable. PhD thesis, Princeton University,
1996.

[18] P. Bloomfield. Fourier Analysis of Time Series: An Introduction. John

Wiley & Sons, New York, 1976.
 SELF-EXCITED OSCILLATIONS OF VERTICAL AND
HORIZONTAL CYLINDERS IN PRESENCE OF A
FREE-SURFACE

D. ROCKWELL, M. OZGOREN, AND N. SAELIM

Abstract: Self-excited oscillations of elastically-mounted horizontal and vertical cylinders in presence of a

free-surface are addressed for cases of very low mass-damping ratio. Selected modes of vortex formation are
defined in terms of wholefield representations of the flow structure and related to the trajectory of the
cylinder motion.

1. INTRODUCTION

Self-excited oscillation of an elastically-mounted cylinder in absence of a free-surface

has been the focus of a wide range of research over the past few decades, as evident in
the reviews of Sarpkaya (1979), Griffm & Ramberg (1982) and Bearman (1984).
Govardhan & Williamson (2000) provide a detailed overview of recent advances. The
studies of Kha1ak & Williamson (1996, 1997, 1999) show that three different branches
of response can occur on a plane of oscillation amplitude AID versus reduced velocity
Vp They assess the roles of the mass ratio m* and mass-damping parameter m*C;;, and
define the possible modes of vortex formation in the near-wake via quantitative
imaging.
The foregoing investigations, along with those cited in the remainder of this
Introduction, provide a framework for characterizing the self-excited oscillations of
horizontal and vertical cylinders in presence of a free-surface. The following sections
provide brief, and necessarily selective, summaries of investigations related to
oscillations of a cylinder in proximity to a free-surface.

1.1. Horizontal Cylinder Adjacent to a Free-Surface

The nature of the near-wake of a horizontal cylinder immediately adjacent to a free-

surface is indeed distinctive relative to the wake from a fully-submerged cylinder.
Triantafyllou and Dimas (1989a,b) have shown that the wake is convectively unstable,
rather than globally (absolutely) unstable, which would lead to large-scale Karman-like
vortex formation. In addition to this theoretical approach, a number of related

201
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 201-210.
2003 Kluwer Academic Publishers.
202

experiments in recent years have provided insight into the structure of the near-wake of
a stationary cylinder in presence of a free-surface. Miyata, Shikazano & Kanai (1990)
towed a rigidly-mounted cylinder beneath a free-surface and qualitatively visualized the
vortex patterns while measuring the unsteady forces.
Sheridan, Lin & Rockwell (1997) quantitatively characterized the near-wake
patterns of a cylinder located immediately beneath a free-surface at sufficiently high
Froude number, such that free-surface distortions played a role. These patterns are not
necessarily stationary; rather, they can exhibit metastable characteristics, as
demonstrated by Sheridan, Lin & Rockwell (1995).
Controlled oscillations of a cylinder adjacent to a free-surface have provided insight
into the flow structure and loading that potentially may arise for the corresponding case
of an elastically-mounted cylinder. Lin & Rockwell (1999) subjected a cylinder to
controlled oscillations and revealed locked-on patterns of vortex formation, as well as
sharp peaks in the lift signature, at small values of gap distance between the free-
surface and the shoulder of the cylinder. In the event that controlled oscillations are in
the in-line direction, and occur in presence of a steady inflow, it is possible to
fundamentally alter the patterns of vortices and the force characteristics, as
demonstrated by Cetiner & Rockwell (2001). If the free-surface is penetrated during
controlled oscillations of the cylinder in the transverse direction, it is also possible to
generate well-defmed, phase-locked patterns of vorticity, as shown by Lin, Sheridan &
Rockwell (1996). They demonstrated that the streamwise extent of the reattachment
zone behind the cylinder can be altered substantially if the frequency of the forcing
corresponds to an inherent shedding frequency from the cylinder. Finally, rather than
imposed sinusoidal oscillations, Zhu, Lin, Unal & Rockwell (2000) characterized the
phase shift of the vortex shedding from the cylinder in relation to transient motion of a
cylinder starting from rest.
Further details of all of the foregoing investigations are described by Saelim &
Rockwell (2003).
To date, the self-excited vibration of an elastically-mounted cylinder beneath a free-
surface, and its consequence on the patterns of vorticity in the near-wake, in relation to
characteristic amplitudes of vibration, has received very little attention. Saelim &
Rockwell (2003) addressed these features. Representative excerpts from their study are
given herein.

1.2. Oscillations of a Vertical, Elastically-Mounted Cylinder in a Wave

In the event that the inflow is in the form of a deep water wave, rather than a steady,
uniform velocity, and the cylinder is mounted vertically, rather than horizontally, then
the cylinder response and the associated patterns of vortices take on a different
complexity.
 203

If one considers the limiting case of a stationary cylinder in unidirectional

oscillatory flow or, conversely, unidirectional oscillations of a cylinder in quiescent
fluid, it is evident that a rich array of investigations have been carried out; early studies
are summarized by Sarpkaya & Isaacson (1981). The reader is referred to the studies of
Honji (1981), Bearman, Graham, Naylor & Obasaju (1981), Ikeda & Yamamoto
(1981), Iwagaki, Asano & Nagai (1983), Williamson (1985a), Sarpkaya (1986),
Obasaju, Bearman & Graham (1988), Tatsuno & Bearman (1990). These investigations
provide valuable insight into the unsteady loading, and qualitative visualization of
vortex patterns, which are a strong function of the Keulegan-Carpenter number KC and
the Stokes number ~.
In the event that the cylinder is elastically-mounted, in either a simulated or actual
wave, the elastic response characteristics are complex, as evident in the overview of
Sumer & Freds0e (1997). When the wave-like motion is simulated by a unidirectional,
oscillatory inflow, and the cylinder is constrained to vibrate only in the transverse
direction, the occurrence of a locked-on type of response as a function of the ratio of
the natural frequency fn of the elastically-mounted cylinder and the frequency fw of the
simulated wave is described by Sarpkaya & Rajavi (1979) and Bearman & Hall (1987).
Cases where the cylinder is subjected to forced oscillations in quiescent fluid, to
simulate a unidirectional wave, are described by Sumer & Freds0e (1988) and
Kozakiewicz, Sumer & Freds0e (1997). When the cylinder is immersed in an actual
free-surface wave, its transverse response can exhibit certain similarities to those
described in the foregoing, as shown by Angrilli & Cossalter (1982), Kaye & Maull
(1993) and Hayashi & Chaplin (1988).
If one considers in-line, as opposed to transverse, vibrations, then, as demonstrated
by Williamson (1985b), it is possible to predict the resonant in-line vibrations through
consideration of the Morrison equation. Furthermore, Li, Zhan & Lau (1997) address
the case of a horizontal, elastically-mounted cylinder in regular and random waves, and
Anagnostopoulos, Iliadis & Ganoulis (1995) performed a numerical simulation in a
unidirectional oscillatory flow.
The full complexity of this class of flow-structure interactions involves bidirectional
vibrations of a cylinder, which is connected to either a flexible joint or a bidirectional
cantilevered joint, located at the bottom of a wave tank. Sawaragi, Nakamura & Miki
(1987), Isaacson & Maul (1981), Zedan, Yeung, Salone & Fischer (1981), Borthwick &
Herbert (1988) and Kaye & Maul (1993) have addressed various aspects of the cylinder
response, including the effects of frequency ratio fn/fw and, in one of these
investigations, namely that of Isaacson & Maul (1981), variations of additional
parameters, including the mass ratio mipD 2 were accounted for. Rather than a
cantilevered cylinder, as in the foregoing, Lipsett & Williamson (1991, 1994)
characterized the bidirectional response for the configuration of a spring-mounted
cylinder at both ends immersed in a unidirectional oscillatory flow. The response
204

trajectories were found to be a strong function of both the frequency ratio fn/fw and the
Keulegan-Carpenter number KC. Considering this class of bidirectional response as a
whole, it is desirable to know the nature of the vortex patterns in relation to the form of
the cylinder trajectory. Borthwick & Herbert (1988) and Kaye (1989) undertook
qualitative visualization of the free-surface patterns. The quantitative nature of the
vortex patterns for a classical "butterfly" trajectory of the cylinder response are
provided by Downes & Rockwell (2003). They related patterns of instantaneous
vorticity at various phases of the oscillation cycle to the relative velocity vector of the
wave with respect to the cylinder. Further details of related works, briefly summarized
in the foregoing, are also addressed therein.
The emphasis of the present work is characterization of the quantitative flow
patterns as a function of the trajectory of the cylinder motion, which can range from
quasi-unidirectional to various elliptical forms. These features are currently under
investigation by Ozgoren & Rockwell (2000) and an excerpt from that study is
provided herein.

2. EXPERIMENTAL SYSTEM AND TECHNIQUES

The two generic cases considered herein are an oscillating horizontal cylinder adjacent
to a free-surface in a steady uniform inflow and oscillations of a vertical cylinder in a
wave. For the case of a horizontal cylinder adjacent to a quiescent free-surface in
presence of uniform inflow, the cylinder is located within a test-section insert and
supported by a long leaf spring located outside this insert. The cylinder has a diameter
D = 12.7 rnm and a length L = 305 nun, with inflow velocities extending from 25.4 to
152.4 mmlsec. The mass-damping ratio is m'S = 0.000162 and the mass ratio is m' =
2.4. Further details of this arrangement are provided by Saelim & Rockwell (2003).
For the case of the vertical, elastically-mounted cylinder in a deep water wave, the
cylinder has a diameter D = 19.05 nun, a mass-damping ratio m's = 0.00045 and a
mass ratio m' = 2.3. The corresponding Stokes number is fD2jv = 429.6. The Keulegan-
Carpenter number KC is defmed in terms of the peak velocity Uw of the wave. The
natural frequency fn of the elastically-mounted cylinder was matched with the
frequency fw of the incident deep water wave. Additional aspects of the experimental
system, including design of the bidirectional system with circurnferentially invariant
elastic properties, are described by Ozgoren & Rockwell (2003).

3. OSCILLATIONS OF A HORIZONTAL CYLINDER

ADJACENT TO A FREE-SURFACE

Figure 1 shows representative regimes of response of the horizontal circular cylinder.

These regimes, designated as I, II, and IV, correspond to successively larger values of
 205

reduced velocity Vr = U/fnD extending over the range 4.5 :-:; Vr :-:; 12.75. (For purposes
of brevity, regime III is not included herein.) Submergence 110 is defmed as the location
of the cylinder beneath the free-surface in absence of the steady inflow velocity U.
More explicitly, it is the gap between the free-surface and the upper shoulder of the
cylinder. For all images, i.e., regimes, shown in Figure 1, holD = O.
For regime I, the cylinder does not undergo oscillation. Furthermore, as shown by
the pattern of vorticity, no large-scale vortex formation occurs; only small-scale
concentrations are evident in the separating shear layer. In other words, one sees only a
convective-type instability rather than the classical global (absolute) instability that
arises in the near-wake of a bluff-body.

Figure J: Patterns of instantaneous vorticity due to self-excited oscillations of an elastically-

mounted horizontal cylinder in steady inflow, located immediately beneath a free-surface.
(Saelim and Rockwell, 2003)
206

When the reduced velocity Vr is increased, the oscillation amplitude is small but
highly repetitive, as shown in the image of regime II. During the oscillation cycle, the
cylinder pierces the free-surface. The vortex formation predominantly arises from the
lower surface of the cylinder. Two representative concentrations A and B are shown.
Eventually they coalesce to form a single, larger-scale structure, as indicated on the
right-hand side of the image. This type of vortex development and interaction is
characteristic of a classical mixing layer flow. So, in essence, the remarkable feature of
this regime II is that highly coherent oscillations occur with only single-sided vortex
shedding. A still larger value of reduced velocity Vr yields regime IV, which does not
involve piercing of the free-surface. As indicated, relatively large-amplitude
oscillations occur. They are associated with formation of a counter-rotating vortex pair
A-BJ, which represents one-half of the classical 2P mode that occurs in absence of a
free-surface.
Further details of these regimes, as well as additional aspects, are given by Saelim
& Rockwell (2003).

4. OSCILLATIONS OF A VERTICAL, ELASTICALLY-

MOUNTED CYLINDER IN A DEEP WATER WAVE
The response of an elastically-mounted cylinder in a wave is a strong function of the
Keulegan-Carpenter number KC. A representation of the instantaneous flow structure
in presence of oscillation in an elliptical orbit is given in Figure 2. In this case, KC =
2.49, and the instant under consideration corresponds to the location of the black dot on
the schematic of the trajectory.
The image of instantaneous vorticity ill shows layers/clusters of negative (white)
and positive (gray) vorticity. The negative vorticity layer extends about the cylinder
over approximately one-half circumference, as does the positive layer. Previously-
formed clusters of positive vorticity are shown exterior to the negative vorticity layer.
As these exterior clusters of vorticity evolve in time, they decay relatively rapidly, and
outside the field of view, only very low level values of vorticity are detectable.
The corresponding pattern of streamline \{' topology shown in Figure 2 indicates the
occurrence of two saddle points, i.e., apparent intersections of streamlines. Over the
region of the cylinder periphery occupied by the negative (white) vorticity layer, a
separation bubble occurs.
The aforementioned patterns of vorticity ill and streamline \{' topology are
associated with normalized values of streamwise u and transverse v velocity, in which
the normalization is with the peak wave velocity U W ' In both the u and v images, the
white contours represent positive values, and the gray contours negative values of
velocity. At this instant, the value ofu is large and positive on the upper
 207

Figure 2: Patterns of instantaneous vorticity, streamline topology and streamwise and

transverse velocities for a cylinder undergoing orbital oscillations in a deep water wave.
(Ozgoren and Rockwell, 2003)
208

surface of the cylinder, corresponding to generation of the negative (white) layer of

vorticity ro. Furthermore, this vorticity layer generation is also associated with
.
relatively large positive and negative values of transverse (vertical) velocity component
v.

5. CONCLUDING REMARKS

Self-excited oscillation of both vertical and horizontal cylinders in free-surface

configurations can give rise to distinctive trajectories of the cylinder response and, as
demonstrated herein, associated patterns of vorticity and streamline topology. This type
of quantitative imaging can provide a basis for direct interpretation of the loading on
the cylinder, as shown, for example, by Lin & Rockwell (1996), Unal, Lin & Rockwell
(1997), and Zhu, Lin, Unal & Rockwell. Furthermore, such imaging can guide the
development and implementation of new techniques of numerical simulation. In fact, as
shown by Sirisup, Karniadakis, Yang & Rockwell (2003), this type of imaging can
drive low-order simulations of complex flows, such as the interaction of a deep water
wave with a stationary vertical cylinder. This approach involves a low-order model
formulated by the group of Professor George Karniadakis and his group at Brown
University. In essence, for the POD simulation, the eigenmodes are extracted from the
laboratory images. Construction of the low-dimensional model involves use of a
penalty method to address the unsteady boundary conditions. Accurate modeling can be
achieved with only ten degrees of freedom and, furthermore, such a model is
asymptotically stable without the use of artificial dissipation. This general approach has
substantial promise for simulation of the types of limit cycle oscillations described
herein, as well as related classes of flow-structure interaction.

6. AFFILIATIONS

Donald Rockwell, Lehigh University

Muammer Ozgoren, Selcuk University
Nattapong Saelim, Lehigh University

7. REFERENCES
Anagnostopoulos, P., I1iadis, G., & Ganoulis, J. (1995) Flow and response parameters of a circular cylinder
vibrating in-line with the oscillating stream. In P. Bearman (Ed.), Flow-Induced VibrationJpp. 167-179).
Rotterdam: Balkema.
Angrilli, F., & Cossalter, V. (1982) Transverse oscillations of a vertical pile in waves. Journal of Fluids
Engineering, 104,46-53.
 209

Bearman, P. W. (1984) Vortex shedding from oscillating bluff bodies. Annual Review ofFluid Mechanic, i6,
195-222.
Beannan, P. W., Graham, J. M. R., Naylor, P., & Obasaju, E. D. (1981) The role of vortices in oscillatory
flow about bluff bodies. Proceedings of international Symposium on Hydrodynamics and Ocean
Engineering (The Norwegian Institute of Technology), 621-644.
Beannan, P. W., & Hall, P. F. (1987) Dynamic response of circular cylinders in oscillatory flow and waves.
R. King (Ed.), Proceedings of BHRA Conference on Flow-induced Vibrations (pp. 183-190). Cranfield:
Bowness-on -Windemere.
Borthwick, A. G. L., & Herbert, D. M. (1988) Loading and response of a small diameter flexibly mounted
cylinder in waves. Journal of Fluids and Structures, 2,479-501.
Cetiner, 0., & Rockwell, D. 2001 Streamwise oscillations of a cylinder in steady current. Part II: free-surface
effects on vortex fonnation and loading. Journal of Fluid Mechanics, 427,29-59.
Downes, K., & Rockwell, D. (2003) Oscillations of a vertical elastically-mounted cylinder in a wave:
Imaging of vortex patterns. Journal of Fluids and Structures (in press).
Govardhan, R., & Williamson, C. H. K. (2000) Modes of vortex formation and frequency response of a
freely-vibrating cylinder. Journal of Fluid Mechanics, 420, 85-130.
Honji, H. (1981) Streaked flow around an oscillating circular cylinder. Journal of Fluid Mechanics, 1077,
509-520.
Iwagaki, Y., Asano, T., & Nagai, F. (1983) Hydrodynamic forces on a circular cylinder placed in wave-
current co-existing fields. Memo Faculty of Engineering (Kyoto University, Japan), 45, 11-23.
Griffin, O. M., & Ramberg, S. E. (1982) Some recent studies of vortex shedding with application to marine
tubulars and risers. Journal of Energy Research and Technology, 104, 2-13.
Hayashi, K., & Chaplin, J. R. (1998) Vortex-excited vibration of a vertical circular cylinder in waves.
International Journal of Offshore and Polar Engineering, 8, 66-73 .
Ikeda, S. & Yamamoto, Y. (1981) Lift force on cylinders in oscillatory flows. Report of Department of
Foundation Engineering and Construction Engineering (Saitama University), No.IO.
Isaacson, M. De St. Q. & Maull, D. J. (1981) Dynamic response of vertical piles in waves. Proceedings of
International Symposium on Hydrodynamics and Ocean Engineering (The Norwegian Institute of
Technology), 887-904.
Kaye, D. (1989) Oscillation of a vertical cylinder in waves. Ph.D. Dissertation, University of Cambridge.
Khalak, A., & Williamson, C. H. K. (1996) Dynamics of a hydroelastic cylinder with very low mass and
damping. Journal of Fluids and Structures, iO, 455-472.
Khalak, A., & Williamson, C. H. K. (1997) Fluid forces and dynamics of a hydroelastic structure with very
low mass and damping. Journal ofFluids and Structures, II, 973-982.
Khalak, A., & Williamson, C. H. K.(1999) Motions, forces and mode transitions in vortex-induced vibrations
at low mass-damping. Journal ofFluids and Structures, i3, 813-851.
Kozakiewicz, A., Sumer, B. M., Freds0e, J. & Hansen, E. A. (1997) Vortex Regimes Around a Freely
Vibrating Cylinder in Oscillatory Flow. International Journal of Offshore and Polar Engineering 7, 94-
103.
Lin, J .-c., Sheridan, J., & Rockwell, D. (1996) Near-wake of a perturbed, horizontal cylinder at a free-
surface. Physics of Fluids, 8, 2107-2116.
Li, Y. S., Zhan, S., & Lau, S. L. (1997) In-Iine response of a horizontal cylinder in regular and random waves.
Journal of Fluids and Structures II, 73-87.
Lin, J .-c., & Rockwell, D. (1996) Force identification by vorticity fields: Techniques based on flow imaging.
Journal ofFluids and Structures, iO, 663-668 .
Lin, J.-C., & Rockwell, D. (1999) Horizontal oscillations of a cylinder beneath a free-surface: Vortex
fonnation and loading. Journal ofFluid Mechanics, 389, 1-26.
Lipsett, A. W., & Williamson, I. D. (1991) Modelling the response of flexibly mounted cylinders in
oscillatory flow. Proceedings of the First International Offshore and Polar Engineering Conference (pp.
370-377) Edinburgh, UK.
210

Lipsett, A. W., & Williamson, l. D. (1994) Response of a cylinder in oscillatory flow. Journal of Fluids and
Structures, 8,681-709.
Miyata, H., Shikazono, N., & Kanai, M. (1990) Forces on a circular cylinder advancing steadily beneath the
free surface. Journal of Ocean Engineering, 17, 81-104.
Obasaju, E. D., Bearman, P. W., & Graham, J. M. R. (1988) A study of forces, circulation and vortex patterns
around a circular cylinder in oscillating flow.Journal ofFluid Mechanics, 196, 467-494.
Ozgoren, M., & Rockwell, D. (2003) Interaction of a deep water wave with a stationary and bidirectionally
oscillating vertical cylinder (in preparation.).
Saelim, N., & Rockwell, D. (2003) Self-excited oscillations of a horizontal adjacent to a free-surface (in
preparation ).
Sarpkaya, T. (1979) Vortex-induced oscillations. Journal of Applied Mechanics, 46,241-258.
Sarpkaya, T. (1986) Force on a circular cylinder in viscous oscillatory flow at low Keulegan-Carpenter
numbers. Journal of Fluid Mechanics, 165,61-71.
Sarpkaya, T., & Isaacson, M. De St. Q. (1981) Mechanics of Wave Forces on Offshore Structures. New York:
Van Nostrand Reinhold.
Sarpkaya, T., & Rajavi, F. (1979) Dynamic response of piles to vortex shedding in oscillatory flows.
Proceedings of the 11th Offshore Technology Conference, OTC Paper 3647, 2523-2528.
Sawaragi, T., Nakamura, T., & Miki, H. (1977) Dynamic behavior of a circular pile due to eddy shedding in
waves. Coastal Engineering in Japan 20,109-120.
Sheridan, J., Lin, J.-C., & Rockwell, D. (1997) Flow past a cylinder close to a free surface. Journal of Fluid
Mechanics, 330, 1-30.
Sheridan, J., Lin, J .-c. & Rockwell, D. (1995) Metastable states of a cylinder wake adjacent to a free surface.
Physics of Fluids, 7, 2099-2101.
Sirisup, S., Karniadakis, G., Yang, Y. and Rockwell, D. (2003) Wave-structure interaction: Simulation driven
by quantitative imaging. Proceedings of the Royal Society ofLondon (in press).
Surner, B. M., & Fredsoe, J. (1997) Hydrodynamics around cylindrical structures. World Scientific, London.
Surner, B. M., & Fredsoe, J. (1988) Transverse vibrations of an elastically mounted cylinder exposed to an
oscillating flow. Journal of Offshore Mechanics and Arctic Engineering, 110,387-394.
Tatsuno, M., & Bearman, P. W. (1990) A visual study of the flow around an oscillating circular cylinder at
low Keulegan-Carpenter numbers and low stokes numbers. Journal of Fluid Mechanics, 211, 157-182.
Triantafyllou, G. S., & Dimas, A. A. (1989a) The low Froude number wake of floating bluff objects. Internal
Report MITSG89-5, Massachusetts Institute of Technology, Cambridge, MA.
Triantafyllou, G. S., & Dimas, A. A. (1989b) Interaction of two-dimensional separated flows with a free
surface at low Froude numbers. Physics of Fluids A, 11, 1813-1821.
Unal, M. F., Lin, J.-C., & Rockwell, D. 1997 Force prediction by piv imaging: A momentum-based approach.
Journal of Fluids and Structures, 11, 965-971.
Williamson, C. H. K. (1985a) Sinusoidal flow relative to circular cylinders. Journal of Fluid Mechanics, /55,
141-174.
Williamson, C. H. K. (1985b) In-line response of a cylinder in oscillatory flow. Applied Ocean Research, 7,
97-106.
Zedan, M. F., Yeung, J. Y, Salane, H. J., & Fischer, F. J. (1981) Dynamic response of a cantilever pile to
vortex shedding in regular waves. ASME Journal of Energy Resources Technology, 103,32-40.
Zhu, Q., Lin, J.-C., Unal, F., & Rockwell, D. (2000) Motion of a cylinder adjacent to a free-surface: Flow
patterns and loading. Experiments in Fluids, 28,559-575.
 HIGH MODE NUMBER VIV EXPERIMENTS

J. KIM VANDIVER, HAYDEN MARCOLLO

Abstract: A simple equation is presented which provides the maximum achievable mode number for a
flexible cylinder, towed by the top end with a weight at the bottom end. The maximum achievable mode
number, while towing in still water, is shown to depend primarily on mass ratio, length to diameter ratio and
the maximum allowable angle of departure from vertical at the top end. Modal overlap in lock-in regions is
shown to depend strongly on mass ratio in uniform flow, but not in sheared flow. The reduced velocity
bandwidth parameter is introduced to quantify the extent oflock-in regions in sheared flow. Two shear
parameters are shown to be useful in characterizing low and very high mode number response. Finite length
lock-in regions are described for cylinders with infinite length dynamic properties.

1. INTRODUCTION

As offshore oil and gas production pushes into deeper water, drilling, production and
export risers as well as TLP tendons have become critical elements in the design of
these facilities. All of these long cylinders are susceptible to significant vortex-induced
vibration. Gulf of Mexico loop currents and detached eddies are the source of
considerable concern to engineers having to design for fatigue resistance to VIVo
The prediction of VIV is currently based on data from relatively short cylinders
tested in laboratory environments. The most successful VIV prediction programs for
long cylindrical marine structures are empirically based and make use of the data from
these laboratory models. Many assumptions are required to extend the simplified
experimental results to the prediction of long cylinders in ocean currents. For the most
part the prediction programs are intentionally quite conservative, sometimes resulting in
the unnecessary use of VIV suppression.
Model tests with long flexible cylinders, responding at high mode number are highly
desirable and are needed to help validate and fme-tune the prediction programs. There
are many difficult challenges faced in conducting experiments at high mode numbers.
The challenges range from the design of the experimental apparatus, to the
identification of the key fluid and structural parameters essential to the understanding of
the model test results.
This paper deals with some of the challenges posed in the experimental design and
then addresses the important physical parameters that help one to understand the
response of a long cylindrical structure to VIVoBy carefully examining the dynamics of

211
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 211-231.
2003 Kluwer Academic Publishers.
212

the expected behavior, several new parameters have been formulated. These parameters
make it easier to understand the behavior oflong cylinders in shear flows.

2. EXPERIMENT DESIGN

2.1 Objectives

High mode number experiments are expected to address a variety of questions,

including:
Does lock-in occur at high mode number?
How much VIV suppression coverage is needed over the length of a riser to
prevent high damage rate VIV events?
What is the probability of single versus multi-mode response?
What is the relationship between in-line and cross-flow response?
A Phase I testing program, sponsored by an industry consortium of companies,
known as Deepstar, is being planned for the deep waters of Lake Seneca in Upper New
York State. These tests will serve as a 'proof of concept' opportunity to test the sensors,
the data acquisition system and the testing methods. The model will be towed behind a
boat in a near vertical configuration with a large weight hanging off the bottom. The
test matrix will address some of the issues mentioned in the questions above. Both
uniform and sheared flow tests will be attempted.
Phase II testing is being discussed for a real, sheared current profile in the offshore
environment, such as the Loop or the Gulfstream. The offshore environment introduces
challenges and conditions beyond our control, such as the presence of surface waves.

2.2 Maximum mode number

In order to investigate the effects of high mode number VIV with model testing it is
desired to design the model so that relatively high mode numbers are achievable. By
performing some simple rearrangements of formula with approximations, the mode
number, n, can be estimated giving insights into the important physical parameters
required for the model.
For long cylinders where the tension dominates the bending stiffness effects in
determining the natural frequencies, the natural frequencies may be approximated by

(1)
 213

where Land T are the length and average tension, while mr is the mass per unit length
including added mass. For the lake experiments the tension is nearly constant along the
length.
The highest shedding frequency, /v,max is related to the highest velocity, Umax , the
diameter, D and the Strouhal number, St as follows:
{'
= StD
U max
lv,max (2)

The frequency of the highest mode number possibly excited will be approximately the
same as the highest excitation frequency. Equating the frequencies and rearranging
gives:

(3)

In the case of a towed riser model, the horizontal component of the towing force
required at the top end of the riser is equal to the total drag force, and is given by

(4)

The total drag force, F D, is the accumulation of local drag forces integrated over the
length of the cylinder. Solving for Tmax yields

T = FD z=o
(5)
max sinrjJ sin rjJ

Assuming the tension is approximately constant requires that rjJ be small. By squaring
expression (3) and substituting for Tmax from (5) yields an expression for n~, the
highest achievable mode number for the specified conditions.
214

(6)

If T is not approximately constant, then Equation (6) is a lower bound. The expression
(U 2) is the spatially averaged square of the velocity, U 2 (z), over the length of the

cable. Ps and Pf are the densities of the structure and fluid, while Ca is the added
mass coefficient.
Expression (6) has several different parameter groups. For uniform flows the ratio of
the maximum velocity to the average velocity equals one, for sheared flows the ratio is
higher and thus the maximum mode number achievable is higher. The equation also
reveals that the maximum mode number increases with top angle, mass ratio and aspect
ratio. It may be desirable to limit the top angle so that the incident inflow remains close
to perpendicular to the pipe axis. The mass ratio should be near to realistic prototype
risers. The easiest parameter to systematically vary is limited to the aspect ratio. Larger
aspect ratios lead to higher responding mode numbers.
For the proposed experimental model the aspect ratio is about 4100. By limiting the
top angle to 20 degrees, the achievable cross-flow mode number, as estimated from
Equation (6) is 18. This is for a cylinder length of approximately 427 feet (130m), a
diameter of 1.25 inches (0.03 18m), a specific gravity of 1.5, an St of 0.17, a CD of 2.0,
and for a Ca of 1.0. In-line frequency content is approximately a factor of two greater
than the dominant cross-flow frequencies. For a tension dominated system this means
that if the cross-flow maximum mode number is 18 then the maximum in-line mode
will be approximately Mode 36.

2.3 Resolving in-line and cross-flow components

Maximizing the aspect ratio naturally results in choosing as small a diameter as

practical. With a small diameter, the ability to install instruments becomes more
difficult. Additionally smaller diameters have lower torsional stiffness. One of the
objectives of the testing program is to be able to resolve both in-line and cross-flow
motion components at every location that there is a sensor. At each sensor one must
also know the orientation. This requires that the torsional stiffness be kept reasonably
high. The torsional stiffness K(J is given by
 215

K 0 -_ GJ (7)
L

where G is the shear modulus and J is the polar area moment of inertia.
The proposed model is Aluminum tubing (6061) with a 1.25 inch O.D. x 0.120 inch
wall thickness. The length will be 130 m for the Phase I tests and approximately 400 m
for the Phase II tests. The respective values of torsional stiffness are 9.1 and 3.0 N-m
per radian. This is adequate for the shorter model, but might be a problem at longer
lengths.

2.4 The Number ofrequired Sensors

A current topic under evaluation is the number of sensors required to describe the
vibration in space and time. If the mode shapes are approximately sinusoidal in shape,
then a spatial Nyquist criterion would suggest a minimum of two measurement points
per wavelength. The number of wavelengths in each mode shape is nl2. Therefore, a
rough estimate of the number of required sensors is the maximum mode number. In the
example computation in Section 2.2 that number is 36 biaxial sensors for the 130-meter
long model. For a fixed maximum top towing angle the number of required sensors
increases approximately in proportion to the square root of the length.
The large number of sensors required to resolve the response at high mode numbers
is both a financial and technical challenge. Current technology provides a choice among
strain sensors, accelerometers and angular rate sensors. The installation of a large
number of sensors requires a serial, digital, data transfer. Otherwise there would be too
many conductors running down the center of the pipe. Additionally if there is too much
information to be passed along the serial connection, as would happen with a high
sampling rate and high number of sensors, then the information must be stored at each
sensor location. Synchronous data acquisition among all sensors is extremely important
if modal analysis is to be performed. For this reason a common trigger has to be
designed into the acquisition system so that the sensors sample together.

3. DIMENSIONLESS PARAMETERS IMPORTANT TO THE PREDICTION OF

VORTEX-INDUCED VIBRATION OF LONG, FLEXIBLE CYLINDERS IN
OCEAN CURRENTS

The heading above is the title from a paper (Vandiver, 1993) which discusses several
important parameters including the reduced damping or mass-damping parameter, mass
ratio, shear fraction, (L1 V / Vmax ) , the number of modes within the shear bandwidth
216

(Ns ), and the wave propagation parameter (nt;n). Some additional discussion and
some new parameters follow.

3.1 The Reduced Velocity Bandwidth

Also introduced in that paper is the concept of "the lock-in bandwidth of the wake".
This is a measure of the ability of the wake to synchronize with the motion of a
vibrating cylinder in a sheared flow. It is based on the concept that at a specific
vibration frequency and amplitude there is a flow velocity, v;, ,
which is ideally suited
to supporting lock-in. Furthermore, this ideal flow velocity is centered in a range of
flow velocities which make up the lock-in region. At this center velocity an ideal
reduced velocity may be defmed as

v =~
RcfvD (8)

It is important to note that this reduced velocity is defined in terms of the vibration
frequency, and not any fixed natural frequency.

In the presence of a shear, the velocity varies in the axial direction along the
cylinder in both directions from this most favorable position. Lock-in or wake
synchronization is able to persist over a limited range of velocity values, which is
defmed as ~ V . This variation when divided by the center velocity provides a
defmition of the lock-in bandwidth, dVR

VR,u - VR,L
dV = -~V = ---'-'----'=- (9)
R Vc VRc

The second part of this equation is in terms of the upper and lower reduced
velocities corresponding to the limits of wake synchronization. This lock-in bandwidth
parameter has also been referred to as the "reduced velocity bandwidth" in the user
guide to the response prediction program SHEAR7 (Vandiver et aI, 2002). This is a
parameter used in design computations to predict the extent of a potential lock-in
region. A commonly prescribed design value for this number is 0.4. This specific
value is to be interpreted as meaning lock-in may exist in a sheared flow over a range of
flow velocities, which may vary 20% around the most favorable velocity.
 217

Figure 1 is a conceptual sketch borrowed with permission from Prof. Michael

Triantafyllou of MIT (2003). It shows how the lock-in bandwidth and the strength of
the excitation might vary with reduced velocity, defmed in terms of the measured
vibration frequency and cylinder vibration amplitude. Outside of the positive energy
region the lift coefficient will be negative and will produce hydrodynamic damping.
Outside of the wake capture region, the wake will not be correlated to the motion of the
cylinder. Experimentally such information has been obtained by driving 2D rigid
cylinders, at constant frequency and amplitude in a constant speed uniform flow.
A very useful dimensionless frequency has been described by (Govardhan and
Williamson, 2000). It is Iv/ Ivowhere Iv
is the vibration frequency and Ivois the
stationary cylinder Strouhal frequency associated with the ideal center velocity, Vc ,and
is given by
{" = SYc (10)
Jvo D

If Iv and IL are the upper and lower bounds of the lock-in range, then an
experimentally determined estimate of the lock-in bandwidth is given by

(11)

AJd

Wake
Capture
Region

Positive
Energy
Region

Reduced Velocity. UIfD

Figure 1. Wake capture range in terms ofAID and reduced velocity. From Triantafyllou, 2003.
218

This way of expressing the bandwidth has the advantage that Sf and therefore Iva may
be allowed to be a function of Reynolds number. Equations 9 and 11 are equivalent if
within a lock-in region the spatial velocity variation along a vibrating cylinder in a
sheared flow is equivalent to a slowly changing frequency variation over time in a
uniform flow. This cannot be strictly true, but there is considerable experimental
evidence to suggest that, as an engineering model of the real world, it is adequate for
use in designing risers to resist VIVo
In section 3.3 the role of reduced velocity bandwidth in determining response of a
flexible cylinder in sheared flow is addressed. Before engaging that discussion it is
important to address the role of added mass, because it is often not correctly
understood.

3.2 The Role ofAdded Mass in Uniform Flow

The reduced velocity bandwidth discussion above made no reference to added mass.
The effect of added mass on lock-in of spring-mounted cylinders in uniform flow is
well understood. (Vandiver, 1993) describes it quite carefully. In brief, the most
significant aspect is that in a uniform flow added mass decreases dramatically as the
reduced velocity is increased through the lock-in range. A decrease in added mass
results in an increase in the natural frequency of the cylinder. A very common form of
the reduced velocity is one defined in terms of a fixed frequency, such as the natural
frequency in vacuo or in still water. The symbol used for this form in this paper is VRn
A plot of response AID versus VRn, for a flexible cylinder vibrating in first mode
with pinned ends is shown in Figure 2a.
 219

1.901~------------------------------------------~

a.8a

0.90~~~~~~~~~~~~~~~~~~~
0.0 2.0 4.0 6.0 8 .0 10.0 12.0
REDUCED VELOCITY

Figure 2a. RMS response/D versus reduced velocity. VRn based on the natural frequency in air.
First and second mode cross-jlow vibration of a flexible cylinder From a 1981 experiment
conducted by Exxon Production Research. Published with permission from ExxonMobil.

2.9~----------------------------------------~

x C2-V)

9.0~~~~~~~~~,-,-,-,-,-,-,-,-,-~,-,-~~~
9.9 2.a 4.9 6.0 8.0 10.0 12.0
REDUCED VELOCITY

Figure 2b. Vibration frequency versus reduced for the 1981 EPR experiment on aflexible
cylinder in water. VRn defined using the In in air. Published with permission from ExxonMobil.
220

This is data from an experiment conducted by Matt Greer of Exxon Production

Research in 1981 at the Skibstekmisk Laboratorium in Denmark. Ten hollow
aluminum segments (outside diameter=0.12m, inside diameter=0.11m) were mounted
on a central tensioned steel rod. Plastic spheres were placed inside the segments to
achieve the desired mass ratio, when flooded. The model was 9.93 meters in length and
had an external diameter of 0.120 meters. The composite cylinder had a specific gravity
of 1.0 and a mass per unit length, including trapped water, but not including added mass
of 11.3 kg/m. The cylinder was tensioned between struts beneath a towing tank
carriage. The damping ratio ofthe first mode in air was approximately 1.0 to 1.5% of
critical and the natural frequency in air was 1.936 Hz. The tension was 16.7 kN. The
tension increased by as much as 8% due to drag force during the tests. The value of fn
used to compute V Rn in Figures 2a and 2b included the variations caused by tension
variation, but does not include the variation in natural frequency caused by added mass
variation.
When reduced velocity is defmed in terms of a fixed natural frequency such as the
natural frequency in vacuum, air or still water, the extent of the lock-in range depends
on the mass ratio. When this definition of reduced velocity is used, low mass ratio
cylinders have a much greater value of the upper limit of the reduced velocity range
than high mass ratio cylinders. This is quite evident in Figure 2a. The upper end of the
lock-in range is 9.5. Williamson has shown that the maximum upper limit value of
VRn is given by

VRn,u =9.25
~m*+C/
Im*-.54 (12)

where the mass ratio, m*, is defmed as

(13)

The upper limit of the reduced velocity in Figure 2, for a Ca of 0.0, is predicted by
equation (12) to be 13.6. The actual upper bound shown in the figure is approximately
9.5. The reason the upper bound is not achieved is that the second mode of vibration
asserts control of the wake synchronization. The second mode amplitude is also shown
in Figure 2a. The second mode's potential lock-in range overlaps that of the first, and
for reasons not yet fully comprehended, the second mode becomes dominant at this
particular reduced velocity. This phenomenon is central to the behavior of high mode
 221

number response. For typical risers with mass ratios of 1.0 to 3.0, lock-in regions will
always overlap at high mode number.
Describing high mode number response under controlled uniform flow conditions is
a high priority objective for the experiments and understanding mode to mode
transitions is a particular focus .
Thus far two kinds of reduced velocity bandwidth have been described. VRn is
based on a fixed natural frequency and is relevant to determining the lock-in bandwidth
and extent of modal overlap for uniform flow conditions only. As the flow speed
changes with time the natural frequency changes with the change in added mass.
Equation (12) reveals a significant disadvantage of using VR,n' The values depend on
the C. value the user prescribes. Using an added mass coefficient of 0 or 1 to defme
the natural frequency used to compute VR,nshifts the value.
VR, which depends on the actual vibration frequency, has a bandwidth defined as
dV R. This bandwidth will be observed in a uniform flow experiment in which the
cylinder is forced to vibrate over a range of frequencies and it will also determine the
range of the lock-in region on a cylinder in a sheared flow. In uniform or sheared flow
reduced velocity defmed in terms of the actual vibration frequency does not change
with added mass, added mass coefficient or natural frequency. This topic is expanded
upon in the next section.

3.3 The Role of Reduced Velocity Bandwidth

The reason that low mass ratio cylinders have such a broad lock-in range in uniform
flows is that the natural frequency increases with flow speed. This is because the fluid
added mass decreases as the flow speed and reduced velocity increase. If the mass ratio
is less than 0.54, there is no upper bound of lock-in-reduced velocity (Govardhan and
Williamson, 2000). This is also evident from Equation (12).
If the reduced velocity is defined in terms of the actual observed vibration
frequency, lv, then the dependence on mass ratio is removed.

(14)

When plotted this way the bandwidth of the lock-in range is a function only of the
ability of the wake to synchronize with the motion of the cylinder, and is not a function
of added mass. Plotted in Figure 2b is VRn versus the ratio of the actual vibration
frequency to the natural frequency in air for the same experiment that produced the data
in Figure 2a. In Figure 2a the upper and lower bounds of the lock-in range in terms of
VRn for Mode 1 are 3 and 9.5. The same range in terms of VR defined in Eq. (14) is 5.2
222

and 9.5. This is a variation of 30% around a center value of 7.35. Alternatively it
could be said that the reduced velocity bandwidth, dVR, was 0.6.
If we defme the ideal reduced velocity for the occurrence oflock-in as the inverse of
the Strouhal number,

(15)

then a reduced velocity lock-in bandwidth may be defmed as

dVR -- VR ,v -VR,L _- I v - IL (16)

VR,i Iva
which is another way of arriving at Equation (11).
Two phenomena influence the range of flow speeds over which a natural mode of
response of a flexible cylinder may exhibit lock-in. One is added mass variation and
the other is the ability of the wake formation process to synchronize with the motion of
the cylinder. Both may affect the range of flow speeds that defme the lock-in region,
but at high mode number in sheared flow, it is only the ability of the wake to
synchronize with the motion of the cylinder that is important.
The principle focus of this paper is to understand the vibration of flexible cylinders
in sheared flow. It is helpful to pose the problem as a question. Over what spatial
variation in flow speed is the wake able to synchronize with the vibration of a cylinder
at one of its natural frequencies?
Figure 3 shows a linear sheared flow. Mode 'n ' of a riser exposed to the flow is
likely to experience flow-induced excitation at its natural frequency in regions where
the flow speed will give rise to cross-flow lift forces whose frequencies are at or near
the natural frequency of the mode. Such a region is shown in the figure . The region is
characterized by upper and lower velocities which define the limits of the wake
synchronization region. Between these limits it is assumed that the wake is
synchronized with the cylinder motion. The length of this region can be described in
terms of a reduced velocity bandwidth as given in Equation (9).

(17)

The extent of the lock-in region in a sheared flow is not dependent on mass ratio or
added mass. It is principally dependent on the ability of the wake to synchronize with
 223

the cylinder vibration. It is not dependent on added mass variation because that effect
would require the natural frequency to vary with position along the riser, whereas there
can be only one natural frequency for each mode in a given sheared profile. In a shear
the added mass does vary with reduced velocity all along the riser, but at steady state
the modal mass including added mass is a constant, given by

Mn = J(m(x)+ma(x));t~(x)dx (18)
o

where m(x) is the riser material mass per unit length including contents and ma (x) is
the added mass/length. The added mass distribution along the riser for a given shear
profile and vibration frequency is fixed and so also is the modal mass and the
associated natural frequency.
If the sheared flow profile were to change, for example as with a tidal variation, the
position of the power in region would move for each mode. This would cause the
added mass distribution for each mode on the riser to vary slowly because outside of
the lock-in region the added mass at high and low reduced velocities
224

u =/n D
C St
L. In
L

Figure 3. Power-in region/or mode 'n' in a linear sheared flow.

changes slowly. Therefore the modal mass and the natural frequency of each mode
changes slowly with incremental changes in the sheared profile.
Added mass effects on the natural frequencies and therefore on lock-in behavior are
only associated with actual changes in the flow profile. For a steady state sheared
profile the extent of power-in regions is governed only by the wake synchronization
bandwidth as quantified in the parameter dVR When reviewing drilling riser data, this
author has found that as the velocity profile changes slowly with time, only the first two
or three modes show significant variation in natural frequency due to added mass
variation. The first mode shows the most variation, followed by the second and then
others in descending order. It is predicted that higher mode number response in sheared
flows will show little variation in natural frequency due to variation in added mass.
 225

3.4 Shear Parameters and Correlation Length

Another way to think about the power-in region defined by the bandwidth dVR is as the
length over which vortex-induced lift forces are correlated with the vibration of the
cylinder at its natural frequency. This length is assigned the symbol Lin. In a linear
shear the change in velocity over a length M is given by

(19)

When M is the length Lin' then 11 V is the variation of velocity over the wake-
synchronized power-in region. Dividing both sides of Equation (19) by v;, , the center
velocity for the lock-in region, provides a new expression for dVR in terms of the
shear gradient.

dV = I1V (20)
R Vc

Solving this equation for Lin yields

(21)

A useful expression for vibration of a riser, which exhibits standing wave vibration
over its entire length, is the ratio of the power-in length to the total length. Equation
(21) is easily modified to yield such an expression.

Lin
(22)
L
[ L dV]
Vc dx
226

The quantity in square brackets is a shear parameter, which is conceptually useful when
considering the structural dynamic response prediction problem. For example, consider
a linear velocity profile specified by

V(x)=~x
v (23)
L
then

dV Vrnax
(24)
dx L
and from Equation (22)

(25)

This tells us that the fraction of the length occupied by the power-in region is larger at
higher flow velocities. This expression fails as the center velocity approaches Vmax
because the lock-in region cannot extend above Vmax and Lin will be only half as long as
predicted by Equation (25) when Vc = Vmax .
In summary Equation (25) is a conceptually useful tool in understanding the
relationship between shear and the correlation length which is available to excite each
mode of vibration. It is probably true that the reduced velocity lock-in bandwidth,
dVR ' is itself a function of the shear gradient, but that level of understanding must
await future experimental research.

3.5 Reduced Damping Parameter

The mass-damping or reduced damping parameter as defined for uniform flows is given
by
 227

(26)

The expression on the right is from Griffin (1998) and that on the left is from Vandiver
(1993). r is the structural damping constant per unit length, OJ is the vibration
frequency, U is the flow velocity, Ss is the structural damping ratio, and Pf is the
fluid density. The expression on the right is shown in Vandiver (1993) to reduce to that
on the left. It is well known that this parameter may be used to predict resonant
response AID. Response decreases with increasing values of Sg. At values less than
about 0.1, cross-flow VIV reaches limit cycle amplitudes of one to two diameters
depending on the mode shape. This parameter has been extended by Vandiver (2002)
to the sheared flow case, and may be written as

S = RnOJn (27)
u PfU2Lin
where Rn is the modal damping, OJn is the natural frequency, and Lin is the power-in
length as defined earlier. Vandiver (2002) may be checked for further details of the
derivation and examples of its use.

3.6 The Response of Dynamically Infinite Cylinders

Infinite cylinders are defined in Vandiver (1993) as ones in which vibration waves
excited in one region die out due to structural and hydrodynamic damping before
reaching boundaries. Thus, standing waves do not dominate the response as with low
mode number short cylinders. Vandiver (1993) provides a parameter, nSn' for
assessing if a cylinder is likely to behave as of infinite length with travelling waves or
of short length with standing wave mode shapes. n is the mode number and Sn is the
modal damping ratio. Above nSn:::: 1 the vibration approaches that of an infinite
cylinder.
In a recent doctoral dissertation by Jung Chi Liao (2001), a theoretical solution is
presented for an infinite cylinder excited over a finite region by an harmonic force
applied in a standing wave pattern. This is shown in Figure 4. The excitation is of the
228

form f (X, t) = Fo sin ( 2~X ) cos (mt) and is applied over a finite length, Lin. If it

is assumed that the force is due to cross-flow VIV, ~ may be expressed as

(28

CL is the local lift coefficient, which is AID and reduced velocity dependent. Liao's
solution is shown in Figure 4. Within the power-in region a standing wave appears.
Outside of the power-in region the solutions are traveling waves, which decay due to
structural and hydrodynamic damping as they travel away from the power-in region.
The maximum amplitude of the standing wave at the center of the excitation region
is given by Liao as

(29)

where N in = Lin / A and A is the wave length of vibration waves in the cylinder with
frequency m. The damping, S ' is the structural damping only inside of the power-in
region and is assumed to be reasonably small, less than 0.15. This equation tells us that
given a sufficiently long power-in region (NinS > 2) maximum amplitude standing
waves may be achieved.

t I I I
o
Axial coordinate, X

Figure 4. Harmonic excitation in a finite length power-in region on an infinitely long cylinder.
The wavelength of the excitation matches that of the standing waves in the cylinder at that
frequency.
 229

CI)
"C
::::I
Sandin ~ wavis
:!::
Q. n
E
III
Tra eling w ve V~ IV \ Tra eling w ve
/
CI)
UI
C regi ~n ~ regi n
oQ.
UI / ~
~
----- ~ o
~
-------
Axial coordinate, x

Figure 5. Magnitude of the response of the infinitely long cylinder. given the excitation shown in
Figure 4.

The lift coefficient is of course dependent on many factors including reduced velocity
and AI D. If the power-in region is not long enough for the waves to reach the
maximum value, Equation 7(29) tells us that the maximum amplitude response in the
power-in region will be smaller.
Such behavior is unlikely to be seen on typical offshore petroleum industry risers.
However, it is very likely to occur on cables used to tow heavily weighted deep survey
vehicles behind ocean vessels. The tow cable typically enters the water almost
horizontally due to total system drag as given in Equation (4) but then curves
downward, eventually becoming nearly vertical at the survey vehicle.
The survey vehicle serves as a reflecting boundary condition. The near vertical
region of the cable near the vehicle will have passing by it a nearly uniform flow
crossing the cable perpendicular to its axis. This flow will excite standing wave VIV
near the boundary. This is a semi-infInite cable problem. The waves generated in a
power-in region of length Li)2, starting at the attachment point on the vehicle are
equivalent to those that would be generated in a power-in region twice as long on a
cable extending to infInity in both directions. Half of the wave energy travels in each
direction for the infmite cable. By symmetry arguments only half the power-in length
is needed to supply the wave energy traveling in only one direction in the semi-infmite
case.
Waves generated in the power-in regions propagate along the cable never to return.
At any equilibrium amplitude the power going into creating waves in the power-in
230

region is equal to the sum of the power lost to damping in the power-in region and the
power radiated away from the power-in region as traveling waves.
There is a convenient expression for N in in terms of the local velocity gradient.
As before Lin may be expressed as

L = dVR (30)
in [~ ~]

. N.
smce L. , then
=---'!!...
In A

N = d~ (31)
in [~ :]

From this expression it is easy to see that the stronger the shear, the shorter the power-
in region as measured in wave lengths. From Equation (29) it is easy to see that as N in
increases so does the response amplitude in the power-in region.
~ dV is a very useful dimensionless parameter which expresses the shear gradient in a
Vdx
way which is significant to structural dynamic response on very long cylinders.
There is one situation applicable to offshore risers where the previous results may be
appropriate. Consider the example of a very long riser which is covered by fairings
over most of its length, except for one bare section. How long must the bare section be
in order for VIV to result in significant response amplitudes in the bare region? An
approximate answer is given by Equation (29). The faired region of the riser will have
significant hydrodynamic damping for any traveling waves created in the region
without fairings. Due to the large damping in the faired region the riser will behave as
if it were of infinite length. No standing waves will occur outside of the bare region.
As before a bare region at one end of a riser of length Lin /2 is equivalent to a region
of length Lin found in the middle of an infinite cable.
 231

4. CONCLUSIONS

The purpose of this paper was to

identify areas of weakness in our understanding of high mode number VIV,
describe experiments that might help resolve some of the questions, and
identify dimensionless parameters which might help in the planning of experiments
and help us to better understand the results.
Four particularly useful parameters have been discussed: a reduced velocity
bandwidth parameter, a reduced damping parameter appropriate for sheared flows and,
finally, two shear parameters which have significance in structural dynamic response
prediction. A simple formula has been provided for estimating the maximum
achievable mode number. It shows that LID, mass ratio and top towing angle are the
key parameters that control maximum mode number.

5. REFERENCES

Govardhan, R. and Williamson, C.H.K. (2000). Modes of vortex formation and frequency response for a
freely-vibrating cylinder. Journal ofFluid Mechanics, 420, 85-130.

Griffin, O. M. and Ramberg, S. E. (1982). Some recent studies of vortex shedding with applications to marine
tubulars and risers. ASME Journal of Energy Resources Technology, Vol. 104, pp. 2-13.

Liao, J. C., "Vortex-Induced Vibration of Slender Structures in Unsteady Flow", Doctoral dissertation, MIT
Department of Mechanical Engineering, February 2002, J.K. Vandiver, Supervisor.

Triantafyllou, M.S. , "VlV of Slender Structures in Shear Flow", nJTAM Symposium on Flow-Structure
Interactions, June 2003, New Brunswick, NJ

Vandiver, J. K.. (1993). Dimensionless Parameters Important to the Prediction of Vortex-Induced Vibration of
Long, Flexible Cylinders in Ocean Currents. Journal ofFluids and Structures, Vol. 7, No.5., pp. 423-455.

Vandiver, J. K. (2002). A Universal Reduced Damping Parameter for Prediction of Vortex-Induced Vibration.
Proceedings of the 2(jh International Conference on Offshore Mechanics and Arctic Engineering, OMAE 2002,
Oslo, Norway.

Vandiver, J. K. et aI, (2003). SHEAR7 User Guide, MIT Department of Ocean Engineering.
 NONLINEAR DYNAMICS OF SLENDER CYLINDERS
SUPPORTED AT BOTH ENDS AND SUBJECTED TO
AXIAL FLOW

Y. MODARRES-SADEGHI, M.P. P AlDOUS SIS,

C.SEMLER & P. PICOT

Abstract. In this paper the weakly nonlinear equations of motion, correct to third order of magnitude, are
presented for a slender cylinder subjected to axial flow. The cylinder is considered to be extensible and two
coupled nonlinear equations describe its motions, involving both longitudinal and transverse displacements.
The inviscid component of the fluid force is modeled by an extension of Lighthill's slender-body work, and
viscous, hydrostatic, gravity and pressure-loss forces are added in a similar manner as for cantilevered
inextensible cylinders. However, both the derivation and the final equations have many different and
distinctive features. The equations are discretized via Galerkin's method and solved by Houbolt's finite
difference method. Bifurcation diagrams with flow velocity as the independent variable, supported by phase-
plane plots, show that the system loses stability via a supercritical pitchfork bifurcation leading to divergence.
At higher flow velocities, a secondary Hopfbifurcation leads to flutter and, at higher flow velocities, the limit
cycle evolves into chaotic oscillation. In some cases, an oscillatory large-amplitude limit cycle is found with
no clear physical origination.

1. INTRODUCTION
Historically, the first specific study on the dynamics of a slender cylinder subjected to
axial flow was by Hawthorne (1961) and was concerned with the stability of the
Dracone barge. The Dracone is a long flexible towed tube with tapering ends, which
had been designed to carry oil and other liquids lighter than sea-water. This analysis
was extended and generalized for cylinders with any boundary conditions and was
supported by experiments, by Pa"idoussis (1966a,b). Finally a more general, corrected
linear equation of motion was derived by Pa"idoussis (1973), and the theory was further
extended to deal with cases of confmed flow. The dynamics of long, very slender
cylinders - modeled as strings, rather than beams - has been studied by Triantafyllou &
Chryssostomidis (1985). Also the dynamics of clustered cylinders in axial flow has
been extensively studied [by Chen (1975) among others; see Pa"idoussis (2003)], both
because of its inherent interest and for applications to tube-in-shell type heat
exchangers, and so on. Also, additional extensions have been made to deal with the
dynamics in highly confmed annular flow (e.g. Pa"idoussis et al. 1990). The dynamics of
towed cylinders is of interest, not only for the Dracone problem, but also for the towed

233
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 233-246.
2003 Kluwer Academic Publishers.
234

arrays used in oil exploration. Apart from Hawthorne's original work, studies on towed
systems were made by Dowling (1988a, b) and many others; see, e.g. PaYdoussis
(2003).
The dynamics of cylinders in axial flow is dynamically similar to that of
axially moving one-dimensional structures in quiescent fluid, such as paper web in
paper-making, and traveling chains, bands and tapes (Mote 1968; Parrnila 1986). The
dynamics of this system is also closely related to that of pipes conveying fluid
(PaYdoussis 1998).
Recently, the dynamics of cantilevered cylinders in axial flow has been re-
examined via a nonlinear equation of motion for the fIrst time (PaYdoussis et al. 2002;
Lopes et al. 2002; Semler et al. 2002), but using the inextensibility assumption, which
is reasonable only for cantilevered cylinders. In this three-part paper, the physical
dynamics, the derivation of the equation of motion and the nonlinear dynamics of
cantilevered cylinders in axial flow have been discussed in detail.

2. DESCRIPTION OF THIS STUDY

In this paper, the nonlinear dynamics of slender cylinders supported at both ends and
subjected to axial flow is studied. The cylinder is considered to be extensible, and two
coupled nonlinear equations describe its motion, involving both longitudinal and
transverse displacements. The slender cylinder is of length L, diameter D, cross-
sectional area A, mass per unit length m, and flexural rigidity 1, centrally located in a
rigid channel within which a fluid flows parallel to the channel centreline. The
undeformed cylinder axis is coincident with the X-axis (in the direction of gravity), and
the cylinder is assumed to oscillate in the (X, Y) plane. Here, (X, Y,Z) is a Lagrangian
coordinate system, while (x,y,z) is the corresponding Eulerian one.
The basic assumptions made for the cylinder and the fluid are as follows: (i)
the fluid is incompressible; (ii) the mean flow velocity is constant; (iii) the cylinder is
slender, so that the Euler-Bernoulli beam theory is applicable; (iv) although the
deflections of the cylinder may be large, the strains are small.

3. NONLINEAR EQUATIONS OF MOTION

The weakly nonlinear equations of motion, correct to third-order of magnitude, are
derived using Hamilton's principle. Here, lateral deflections are assumed to be of fIrst-
order magnitude, 0 ( E) , while axial ones of second-order. For convenience, inviscid,
hydrostatic and viscous forces are determined separately, not together, say by direct
application of the Navier-Stokes equations.
 235

The inviscid component of the fluid force acting on the cylinder, F 8x, is A

modeled by an extension of Lighthill's slender-body work to third order of magnitude

(Lopes et al. 2002). The hydrostatic forces, ~<8x and F"ox, the resultants of the
steady-state pressure, p, acting on the cylinder, are derived following Paidoussis
(1973), and the viscous frictional forces in the normal and tangential directions, FN8x
and p;8x, are formulated on the basis of semi-empirical expressions proposed by
Taylor (1952). All of the forces are expressed correct to third order of magnitude. Space
limitations preclude presenting the detailed derivation here; this will be done in a
forthcoming paper. The resulting nonlinear equations of motion, in dimensionless form
are
(1 - fJ) U" - X (fJ w.. , + 2U " f:fJ'"
pv v + U 2
v"v' ) - II ("
0 u + v'")
v - ("",
v v + v '",)
v + -:;1 C, fJ2
V

1 U
-2 '( c" -c, )(fi.'''J
U- w +v +2
1 U 2 c, ( l+h ) (u, -21v,,( 1) vv, ")
+ 1-';-20

- -1
2
u' cv'('1'1
d
fi('I'I'I'I)
v v +--
U U
fJ II) +r (,
v v+vv +-vv u -
2
- v + ( 1-.;'- -1 0 ) 'vv
I"
2 2
")

+ [; U 'Cb ( r
1- 0) + 0 + (1 - 2v) Do ] + 0 ( e' ) = 0,
(1)
236

(I +(x-I) p)ii+2xU fP v'( I-~ v'') +xU ' v'( 1-'; v" ) - XV'(Pii+3fP U Ii' +2U V)

-X(4U u'+2fPu+;fPVv')(fPv'+Uv')+~&U ' Cd (v'lv'I+: (v'1~+vlv'D+ : ' vl~)

{~u 'C (1-0)+ ro+(1-2v)IToJ( -v" +v'u" +v"u' +;v"v") -~( v'u" +v'u' +;v"v")
b

+V,4, - (8'"
VV vIII +Vu
" ., +2"
v v(" v +2u ' v(4' +4u"vI I +3UIII")
+2" ~rl""" +-&
V -,{IJW 1 U ' ( C+C h) v,
2 "'

+;8U
1 1 13 +[U-U ( 1) ( 1-0)]"
'c, ( I+h ) ( -;V v +( 1-~-;1 O ) ( -v" +VU
' " +VU
" ' +;V
3 " V "))

t 2 f3 ' , VV
" 1 V. 2 V' ) 1 2 "P. '\j P PR
- - 8U (C -C) ( v +VU +P-+ - P - + - 8 U C -V+-UV+-UV- - - V
'. [fa fa
d " .J
1 jJ fo J
2 "' U 2 U' 2 " U U U ' 2 U'

+y ( v, -;1 V13 +[U-U ( 1) ( 1-0)]"

v +( 1-~-;O
1 ) ( -V" +VU
' " +VU ")) +0 ( E5) =0.
" ' +;3V" V (2)

In these equations, U and v are nonlinear displacements in the longitudinal and

transverse direction, respectively; U is dimensionless flow velocity, used extensively
as an independent parameter in studying the dynamics of the system; {3= pA I{m + pA )
is a mass ratio; U. = EAL' I El, n = PAL'l El and r = rL' I El are dimensionless axial
flexibility, pressurization and externally imposed uniform tension, respectively; c n and
C
I
are the coefficients of frictional forces in the normal and tangential (longitudinal)
directions, respectively; Cd is the coefficient of transverse form drag; 8 = 0 signifies
that the downstream end is free to slide axially (or wholly free), while 8 = 1 if the
supports do not allow axial sliding; v is the Poisson ratio; y= (m - pA) gL3 is a lEI
gravity coefficient; cb is the base-drag coefficient acting in the longitudinal direction at
the free end of the cylinder when 8 = 0; & = LI D is the slenderness ratio; h = D/ Db
is a hydraulic coefficient, Dh being the hydraulic diameter; and X = M / pA is an added
mass coefficient which increases with increasing confmement.
These equations of motion are valid for all boundary conditions of a slender
cylinder subjected to axial flow, supported at both ends or free at the downstream end.
 237

4. METHOD OF SOLUTION

The equations of motion for this system are nonlinear and there is no analytic solution
for them. Therefore, as a fIrst simplifIcation, these equations are transformed into a set
of second-order ordinary differential equations by Galerkin's technique; the
eigenfunctions of a bar in axial vibration, If/j ( ; ) , and those of a beam with the same
boundary conditions as the cylinder, <Pj are used as a suitable set of comparison
I:
( ; ) ,

functions . Thus, u(~,,) = If/ j (~) P j (,) and v(~,,) = I ;~' j (~)qj (,) , where
Pj ( r) and q j ( r) are the corresponding generalized coordinates, and NU and NV
represent the number of modes in the longitudinal and the lateral direction, respectively.
Substituting these expressions into the dimensionless equations of motion leads to the
following set of ordinary differential equations:

+E'/j, qj Iq,1+ E'ij, qj

. Iq, +
1Fij.' qjq,q, + Fij.' qjq,q,. + F".' qjq,. .q,+ F' . . + Fij.' qjq,q,
ij. qjq,q, .. = o. (4)
In these equations, the internal dissipation of the cylinder is assumed to be viscoelastic
and of the Kelvin-Voigt type. The indices i, j, k, I are integers ranging either from 1 to
NU or from 1 to NV. The terms M; , C ; , K ;and M ; , c;,
K; correspond to linear mass,
damping and stiffness matrices in the longitudinal and transverse directions,
respectively, while the other terms are nonlinear.
In order to analyze this set of equations, a fmite difference method is used; see
Semler et al. (1996). This is an initial-value problem solver in which the system of
equations is integrated numerically for one initial condition at a time, and then the state
of the system thereafter at any time r can be reproduced.

5. NONLINEAR ANALYSIS FOR A SIMPLY-SUPPORTED SYLINDER

Linear analysis of this system (PaYdoussis 1973), based on small motions, can only
predict the point of fIrst loss of stability, without any defmitive prediction of its post-
critical behaviour. Linear theory predicts that, in general, at low flow velocities, the
cylinder is stable; then, as the flow velocity increases, the cylinder is subjected to
238

divergence (buckling) in its first mode. Linear theory also predicts the occurrence of
second-mode buckling of the system, and at higher flow velocities the existence of
coupled-mode flutter (so-called "PaYdoussis flutter") in some cases. In order to
investigate the validity of the post-critical behaviour of the system predicted by linear
theory, and also to find the amplitude of buckling as well as the amplitude and
frequency of oscillatory motion if flutter does exist, one needs to perform a nonlinear
analysis of the dynamics. The information gained from a nonlinear analysis of the
system is often summarized in a bifurcation diagram in which, typically, the amplitude
of motion is plotted as a function of one parameter of the system; in this paper, the
amplitude of the midpoint of a simply supported cylinder is plotted as a function of the
dimensionless flow velocity.
In the following sections the influence of different parameters on
the behaviour of the system is examined for the following physical
parameters: D = 0.0254 m, p = 1000 kgl m 3 , m = 0.5817 kg/ m ,
M = 0.507 kg/m and E = 2.76 X 10 Pa; leading to the following
6

dimensionless values: U = 3 UL, f3 = 0.47, r = 12.996t and r = 0.27 t , tl

where U is in m/ s, L in m and t in s. Here U ,r and r are in terms of L,
while U will be used as an independent variable. It is assumed that X = 1
and h = 0, which corresponds to a cylinder in unconfined flow, and that
the coefficient of form drag is zero, Cd = o. The cylinder is not allowed to
slide at the downstream end; hence, 8 = 1 .

Dimensionless flow velocity, U

Figure 1. Bifurcation diagrams of a simply-supported cylinder with

f3 = 0.47,
Y = 3.4, 0 = 1, v = 0.47, X = 1, II o = 10 000, IT = f = cb = Cd = 0
and different values of BC" = BC, ' ranging from 5 to 40.
 239

5.1. Influence of Frictional Coefficients (c n and c,)

In this study, frictional coefficients in the normal and tangential directions are assumed
to be the same cn = c, (Paldoussis 2003, Appendix Q). Also, since in the equations of
motion they always appear as &en and BC,' = BC, is the same
the effect of varying BC n

as varying c n = c,' Figure 1 shows the bifurcation diagram of the system for different
values of BC n = BC, . With increasing frictional coefficients, the first bifurcation point
(divergence) occurs at progressively lower flow velocities. Also at a fixed flow
velocity, the amplitude of buckling is increased with larger frictional coefficients.
These two effects could be explained by noting that increasing the frictional
coefficients results in half the cylinder being subjected to an increasing compressive
load.

5.2. Influence of Coefficient of Base Drag (cb ) , Externally Imposed Uniform Tension
(r) and pressurization (IT)

Both the coefficient of base drag (c b ) and externally imposed uniform tension
(r = fL' / E1) represent a pre-strain in the longitudinal direction. Therefore, varying Cb

or f should have a similar effect on the dynamical behaviour of the system.

0,07 .---~-~-~-~--~-~-~-_-------,

.-,.0
&"

Dimensionless flow velocity,u

Figure 2. Bifurcation diagrams of a simply-supported cylinder with

f3 = 0.47, r = 3.4, c" = c, = 0.025, & = 20, 8 = 1, v = 0.47, X = 1, ITo = 10 000, IT = c, = Cd =0
and different values of f, ranging from -5 to 20.
240

Figure 2 shows the bifurcation diagrams of the system with different values of
r. Bifurcation diagrams for varying cb (not shown here) are as expected. When the
cylinder is placed in pressurized fluid (this represents a restoring force like an increased
tension), higher flow velocities are needed to make the cylinder unstable. Therefore, the
first critical flow velocity (for divergence) increases. With increasing cb and at a r
fixed flow velocity, the amplitude of bucking decreases. This is because in a cylinder,
pressurized from the sides or stretched from the free end, the lateral displacement will
be reduced.

5.3. Influence ofDimensionless Axial Flexibility (II.)

The dimensionless axial flexibility II. = EAr' / El is a measure of the axial rigidity as
compared with the transverse rigidity of the system. For a full cylinder (not hollow),
II. = ( 4L/ Dr; this implies that, for a fixed D, increasing the value of II. means a
larger L; therefore, a larger amplitude of buckling should be expected. Figure 3 shows
the bifurcation diagram of the system for different values of II. in which, the larger the
value of II. is, the smaller the amplitude of buckling becomes! The answer to this
paradox lies in the fact that the dimensionless flow velocity and the dimensionless
transverse displacement, both depend on the length of the cylinder. Once this is taken
into account, the results are as one would expect.

:.. .. ....J. ...!~:~~~~t::

--- --- ~ ~ .~ ~ -
: ... ----~-- - ~ -...~:...... -------- - - .u..u.i.a.-...&-.a.
- ~ ------- c ~-- "'."'AA"'... ------
________ [,!!1:"'''''''ll~~~U
c -------
,__________
.......
_ __
Dimensionless flow velocity, U

Figure 3. Bifurcation diagrams of a simply-supported cylinder with

f3 = 0.47, r = 3.4, c, = c, = 0.025, e = 20, 8 = 1, v = 0.47, X = 1, IT = = cb = Cd r = 0
and different values of ITo' rangingfrom 1000 to 10 000.
 241

In Figure 3, showing the effect of II, , the flow velocities are in the range
0< U :s; 10, sufficient for the comparisons made in the previous paragraph. Figure 4
shows the bifurcation diagrams of the system (obtained with NU=NV=2) with different
values of II, for a very wide range of dimensionless flow velocity. Here, for the first
time we see that there is a post-divergence secondary bifurcation leading to flutter. As
an example, for II, = 2000, the system is stable at very low flow velocities, up to
where it loses stability via a pitchfork bifurcation. At higher flow velocities, a Hopf
bifurcation leads to flutter, and this evolves into chaotic oscillations. Figure 5 shows the
phase plane plots of this system. At U =18, the response is periodic (Figure 5a), and
later a period-doubling bifurcation occurs (Figure 5b). This new stable solution again
loses its stability through another period-doubling bifurcation, leading to period-four
oscillation (Figure 5c). Figure 5d shows that the motion for U =20.4 is chaotic. At
even higher flow velocities, a period-bubbling sequence leads back to period-four
(Figure 5e) and period-two oscillations (Figure 5f). Note that the results for
U >20 or 25 are quantitatively not very reliable. Calculations with NU=2 and NV=4
show a qualitatively similar behaviour, but with some quantitative differences.

0 . 4 . , - - _- _ -_ - - - -- _-_-_--_--::::10

-l--~-----
0.3

0.2
-- -----'
, -
0.1
_200
.1000
.. 2000
-10000
-0.1 ----- --,

-0.2

_____ __
-0.3

-0.4 .l..-_~_~_~ ~_~_~ ~__'

Dimensionless flow velocity, U

Figure 4. Bifurcation diagrams of a simply-supported cylinder with

fJ = 0.47, r = 3.4, c, = c, = 0.025, G = 20, 5 = 1, v = 0.47, X = 1, IT = f = cb = Cd = 0

and different values of Il" for a very wide range of dimensionless flow velocity.
242

5.4. Effect of Large Initial Values

For the system under consideration, when one uses a very large initial value for the first
generalized coordinate compared with the length of the cylinder (e.g.
q l (Initial) = 0.2, 0.3, ... ) in the FDM program, especially for large II. , a large-amplitude

flutter-like motion over a range of flow velocities appears, which is unreasonably large.
Also the frequency of the resultant motion is about 30 Hz, which is very high compared
with the first-mode frequency of the system (:::: 3 Hz) . Figure 6 shows that for a large
value of q l(Initi,l) (e.g. 0.2), the large-amplitude motion exists over a wide range of flow
velocities; as ql (lnitial) is reduced, however, the range of existence of this large-amplitude
motion is diminished, and eventually (e.g., for ql(Ini"-I) = 0.06) it disappears, as in Figure
2.
 243

- - -,- - -- .. - -- ~

_ _ ~L- ___ ~ _
6 ,
___ .. ___ ...." __ __ 1_ ______ __ _ -<,_ __ _
, , , 1
1 , 1 ,

,
___ l ___ ...1 ____ '- ___ L_ ___ .1. ___ ..1 __ _ _
, _ _ _ _ _ _ _ _ _ _ , ___ _
, , , , ,
_ __ l ___ J ____ ,____ !. _ _ 1 _ ___' ___ _ _ __ 1 __ _
, , I ,--

,
0.05 011
, 0.15
, ,
0'2 ~ 0 0 0. 5 O!l
---T---
, " -,.-
,, --- T---..,--- -r- - --
, -4 ___ .
1
,
"
, , ,
___ .., __ __ .... _ __ .. ___ .. _ __ .., __ _ _ ,
,

, , , , , ,
~

- - - - - - ; - - - -0- - - - .. - -
,,
~ ~

"
" , " ,
.e - - - .. - - - . . - - -:- - - - ~ - - - ~ - - - i - - --
.a ___ .I. _ _ _ ..I ____ ' ____ L _ .e - - _ .. - - _ ..1 _ _1 ____ L ___ .. __ __ 1 _ __ _
~
1 , , ,

1 , , , 1 , , ,

.a __ _ , __ . . 1____ ,1_ ___ , ___ , ___.... __

, _ -

1 L 1 1 , ,
___ .I. ___ ...1 ___ _ ' _ ___ L _ __ l ___ -' __ _ _
1 , , 1 , ,

" "

" r-~~--~~~--~...,

f)-L-~--:----.--
,
. ___ ~---- "
, "

~ t--O.~r-~011--o~tr~~~H
- --T---., -- - -

Figure 5. Phase plane plots of a simply-supported cylinder with

fJ = 0.47, r = 3.4, c" = c, = 0.025, & = 20, 0 = 1, v = 0.47, X = 1 , ITo = 2000, IT = r =c
b = Cd =0
for (a)U =18, (b)U =19, (c)U =20.1, (d)U =20.4, (e)U =21.8 and U =22.
244

All the results up to this point have been based on calculations with 2
comparison functions (modes) in the longitudinal direction and 2 in the transverse
direction. If one increases the number of modes, which corresponds to a more exact
solution, the range of existence of the large-amplitude motion decreases. The larger the
number of modes in the transverse direction (NV), the smaller is the range of existence
of the unrealistic large-amplitude motion. For example in the bifurcation diagram of the
simply supported cylinder subjected to axial flow for NV=8 and NU=2 (not shown here)
there is no sign of the large-amplitude motion, even for large values of ql(Jnitial)

..,.......
.....

.... .: ..,
0.2

. ' !
, ,

-- -
, . ,
------------------------- T -------,--

.() . .......~u ........... ~

.(),-'----'-----'-------'----'-----'-----'------'
Dimensionless flow velocity, U

Figure 6. The appearance of a large-amplitude flutter-like motion over a range offlow velocities
in the bifurcation diagram of a simply supported cylinder for big initial values ofq, and
ITo = 10 000.

6. CONCLUSION

In this paper, a nonlinear equation of motion, correct to third order of magnitude, has
been presented for the dynamics of a slender cylinder subjected to axial flow. This
equation is probably not the defmitive nonlinear equation of motion for this system,
since it was not obtained by a unified nonlinear treatment of the fluid mechanics. Based
on the derived nonlinear equations of motion, the dynamics of the system was studied
from a nonlinear point of view, and the existence of post-divergence instabilities of the
cylinder was proved. Also, the effect of some of the key parameters affecting the
critical flow velocities and the amplitude of the resultant motions were explored. If one
imposes a very large initial value for the first generalized coordinate, a large-amplitude
high-frequency flutter-like motion of the cylinder is predicted; this unrealistic motion
 245

disappears when the number of transverse-direction modes used in the solution is

increased sufficiently.

7. AFFILIATIONS

Department of Mechanical Engineering, McGill University, Montreal, QC, Canada

8. REFERENCES
Chen, S. S. 1975 Vibration of nuclear fuel bundles. Nuclear Engineering and Design
35, 399-422.
Dowling, A. P. 1988a,b The dynamics of the towed flexible cylinders. Part 1: Neutrally
buoyant elements, Part 2: Negatively buoyant elements, Journal of Fluid
Mechanics 187, 507-532 and 533-571.
Hawthorne, W. R. 1961 The early development of the Dracone flexible barge.
Proceedings of the Institution of Mechanical Engineers 175, 52-83.
Lighthill, M. I. 1960 Note on the swimming of slender fish. Journal of Fluid Mechanics
9,305-317.
Lopes, J. L., Pai"doussis, M. P. & Semler, C. 2002 Linear and nonlinear dynamics of
cantilevered cylinders in axial flow. Part 2: the equation of motion. Journal of
Fluids and Structures 16, 715-737.
Mote, Ir, C. D. 1968 Dynamic stability of an axially moving band. Journal of the
Franklin Institute 285, 329-346.
Pai"doussis, M. P. 1966a,b Dynamics of flexible slender cylinders in axial flow. Part 1:
Theory, Part 2: Experiments. Journal of Fluid Mechanics 26, 717-736 and
737-751.
Pai"doussis, M. P. 1973 Dynamics of cylindrical structures subjected to axial flow.
Journal of Sound and Vibration 29, 365-385 .
Pai"doussis, M. P. 1998 Fluid-Structure Interaction: Slender Structures and Axial Flow,
Vol. 1. London: Academic Press.
Pai"doussis, M. P. 2003 Fluid-Structure Interaction: Slender Structures and Axial Flow,
Vol. 2. London: Academic Press / Elsevier.
Pai"doussis, M. P., Mateescu, D. and Sim, W.-G. 1990 Dynamics and stability of a
flexible cylinder in a narrow coaxial cylindrical duct subjected to annular flow.
Journal ofApplied Mechanics 57, 232-240.
Pai"doussis, M. P., Grinevich E., Adamovic, D. & Semler, C. 2002 Linear and nonlinear
dynamics of cantilevered cylinders in axial flow. Part 1: Physical Dynamics.
Journal of Fluids and Structures 16, 691-713.
246

Pramila, A. 1987 Natural frequencies of a submerged axially moving band. Journal of

Sound and Vibration 113, 198-203.
Semler, c., Gentleman, W. C. & Pa"idoussis, M. P. 1996 Numerical solutions of second-
order implicit ordinary differential equations. Journal of Sound and Vibration
195,553-574
Semler, c., Lopes, J.-L, Augu, N. & Pa"idoussis, M. P. 2002 Linear and nonlinear
dynamics of cantilevered cylinders in axial flow. Part 3: nonlinear dynamics.
Journal ofFlu ids and Structures 16, 739-759.
Taylor, G. I. 1952 Analysis of the swimming oflong and narrow animals. Proceedings
of the Royal Society (London) A214, 158-183.
Triantafyllou, G. S. & Chryssostomidis, C. 1985 Stability of a string in axial flow.
ASME Journal of Energy Resources Technology 107, 421-425.
 SOME QUANDARIES AND PARADOXES IN
FLUID-STRUCTURE INTERACTIONS WITH AXIAL FLOW

MICHAEL P. PAIDOUSSIS

Abstract: . A number of interesting quandaries and paradoxes, some resolved and some not,
are reviewed briefly; they all involve axial flow over generally slender structures, as follows.
(i) For pipes and shells conveying fluid, does post-divergence flutter occur or not? (ii) Do
aspirating pipes and shells lose stability at infinitesimally small flow rates? (iii) Do pipes and
shells with different kinds of end-support (e.g. , clamped at one end and simply supported at
the other) experience flow-induced damping , positive or negative, at arbitrarily small flows?
(iv) In view of the similarity in the equations of motion of cylindrical structures subjected
to (a) internal and (b) external axial flow , in their simplest form, how far can the analogy
between internal and external flow be carried in assessing the dynamical behaviour of such
systems, and what real physical effects intervene to break it down? (v) Should convective
fluid-acceleration terms be taken into account in the analysis of the dynamics of strings or
beams undergoing extrusion or deployment in dense fluid, and similarly for traveling web?
(vi) Taking annular fluidelastic systems as an example, is the precise specification of boundary
conditions on the fluid at the upstream and downstream ends essential in the determination
of system stability? These are described in greater detail and clarified to the extent possible
in the body of the paper.

1. INTRODUCTION

A number of quandaries and paradoxes have come to the fore while writing
Fluid-Structure Interactions: Slender Structures and Axial Flow (Paidoussis
1998, 2003). Some have been extant for some time, and indeed one had been
resolved prior to this; others emerged while bringing information together from
various sources. Some were resolved thanks to the effort of formulating them
more clearly and comprehensibly than theretofore, and therefore studying them
more deeply, while some remain unresolved to this day.
In what follows, an account of each quandary or paradox is given, together
with a historical perspective (which helps to animate the discussion) and an
exposition of the current state of understanding.

2. POST-DIVERGENCE FLUTTER OF PIPES AND SHELLS

The question of existence of post-divergence flutter of pipes and shells conveying

fluid is posed exclusively for pipes or shells with supported ends, i.e. for

247
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 247-266.
2003 Kluwer Academic Publishers.
248

inherently conservative systems. Indeed, cantilevered systems may actually

lose stability by flutter, and they are excluded from this discussion.
We start with a pipe with clamped or simply supported ends conveying
fluid . It has been known for a long time - see, e.g., Feodos'ev (1951), Housner
(1952) - that, for sufficiently high flow, the pipe loses stability by divergence,
i.e. it buckles. This has been confirmed by experiments, by Dodds & Runyan
(1965) and others. Then, in the 1970s, Pai:doussis & Issid (1974) found that
linear theory predicts the existence of post-divergence flutter, at higher flow
velocities. This is problematic, because the work done by the fluid on the pipe
in a presumed cycle of oscillation of period T is found to be

where M is the mass of the fluid per unit length, U the flow velocity, w(x, t)
the lateral displacement of the pipe, L the pipe length, and t is time; hence,
since 8w/8t = 0 at both x = 0 and x = L,
~W=O. (2)
This, then, constitutes a paradox: for how is it possible for the pipe to flutter
if (i) the system is conservative and (ii) no energy is supplied to sustain the
oscillation? Here, it is noted that frictional effects are exactly cancelled out by
pressure loss effects, so that this is a result applicable to both inviscid and (real)
viscous flows. Also, theory predicts coupled-mode flutter, even if dissipation in
the pipe material is accounted for.
An ingenious attempt to resolve the paradox was made by Done & Simpson
(1977), by considering the downstream end to be free to slide axially, though
prevented from moving laterally. Without reproducing this work here, the
crux of the matter is that the momentum flux of the fluid issuing from
the downstream end, M U 2 , causes a mean contraction c of the pipe; the
system then oscillates with contraction amplitude c, and hence with a lateral
oscillation, without any net work being required to maintain the oscillation.
In this, it should be remembered that, in the equation of motion, axial
displacement does not appear explicitly - any more than it does in the
Euler-Bernoulli equation for vibration of a beam with supported ends; but
its existence is nevertheless implied.
However, the definitive work on this was done by Holmes (1977, 1978) who
undertook a nonlinear analysis of the problem. Indeed, this is a must, since
linear theory is strictly applicable only up to the first loss of stability. First,
a finite-dimensional analysis was undertaken (Holmes 1977). As predicted by
linear theory also, there is no restabilization after divergence if dissipation is
taken onto account. The secondary bifurcation corresponding to the coupled-
mode linear flutter threshold is unstable, and hence does not materialize,
 249

(a)

Figure 1. Qualitative pictures of the vector field for (a) u < 7r ; (b) 7r < U < 27r,
i.e. after the pitchfork bifurcation and divergence; (c) u > 27r, just after
where linear theory predicts coupled-mode flutter. (d) Schematic of transient
oscillation just after u::::: 27r [after Holmes (1977, 1975) and Pai"doussis (1g9S)}.

as shown in Fig. l(c) . The only stable attractors are the two sinks, as
shown. Although there could be a local oscillation about the origin, as shown
diagrammatically in Fig. l(d), eventually the system is captured by one of
the sinks. For a Hamiltonian system (zero dissipation), this is confirmed by
computations, as shown in Fig. 2. However, the oscillation observed in Fig. 2
is pathologically nonrobust: the slightest amount of damping kills it; hence,
the coupled-mode flutter is only transient. Indeed, post-divergence flutter has
never been observed experimentally.
Next, Holmes (1978) undertook an infinite dimensional analysis, assessing
stability via the Lyapunov second (direct) method, proving unequivocally
that post-divergence flutter is impossible. The title of the paper, "Pipes
supported at both ends cannot flutter", started a remarkable trend: to state the
main conclusion of the paper in its title ~ particularly useful for overworked
executives, who could thus digest the gist of a paper between the de rigueur
250
0.6
q,
0.4

.
O~

'1
.:
0.0

~.2

" ~.4

(aJ (b)
~.6
~.06 ~.04 ~.02
.,
0.00 0.02 0.04 0.06

Figure 2. (a) "Limit cycle" at u 2.025n and (b) phase-plane plot at

u = 2.02l7r (Pai'doussis 1998).

Im(n)

Figure 3. Argand diagram for a clamped-clamped shell conveying fluid for

circumferential mode number n = 2: - - - , m = 1; - - - , m = 2;
- , combined m = 1 and m = 2; - - , m = 3, where m is the axial
wavenumber (Pai'doussis fj Denise 1972; Pai'doussis 2003).

champagne and the hors d'oeuvres on a first-class flight! Detailed information

 251

on all the foregoing is provided by Pa'idoussis (1998).

20 (a)

---- ---- -
10
...c:!
...
-
.......... 3
2
0
~
-10

-20

10 (b) 2

- - ...
-.. --
... - -
5
....
--- --....-.
...c:!

-
..........
N

.. - -
c: 0
-.:t:
-5

-10

0 2 3 4 S 6
V

Figure 4. Nonoscillatory dimensionless modal amplitudes Anm versus the

dimensionless flow velocity V for a typical shell conveying fluid , with small
structural damping (( = 0.01) and without comparison mode participation:
(a) And h and (b) An2 / h; - , stable branches; - - - , unstable branches
(Amabili et al. 1998; Paidoussis 2003).

The next question is whether all of this also applies to shells conveying
fluid. Linear theory predicts divergence, followed by coupled-mode flutter, as
shown for instance in Fig. 3; for water-flow (Weaver & Unny 1973) the two
critical flow velocities are more widely separated. In experiments with air-flow,
the system was found to lose stability by flutter directly (Pa'idoussis & Denise
1972), and it was reasoned that, since divergence and flutter were theoretically
252

so close, the onset of divergence entrained flutter immediately.

It was a considerable time later that a nonlinear model became available
(Amabili et al. 1999). This model is based on the Donnell nonlinear shell
equations and the linear fluid-structure interaction model of Paldoussis &
Denise. Typical results are shown in Fig. 4. The system loses stability by a
strongly sub critical pitchfork bifurcation at dimensionless flow velocity V ~ 3.3;
all subsequent bifurcations lead to static solutions, both stable and unstable.
Hence, the conclusion is that there is no post-divergence flutter in this case
either. In this light, the observed flutter must be an oscillatory divergence: as
the shell collapses inwards, the flow is constricted, and the build-up of pressure
forces the shell to re-open with enough inertia to be flung into the azimuthally
antiphase shape; and so on. The results in Fig. 4 are without comparison mode
participation. With it, the dynamics is more interesting, but the foregoing
general conclusion is not altered.
More recent experiments, some with metal shells conveying water, are
described by Karagiozis et al. (2003) at this very Symposium; the whole issue
of post-divergence flutter is discussed in Paldoussis (2003).
One weakness in the conclusion reached above, so far as shells are concerned,
is that unsteady viscous effects are not accounted for in the extant theory, while
it has been shown, for shells in particular (Nguyen et al. 1994) and fluidelastic
systems in general (Paldoussis 2003), that unsteady viscous effects can affect
the dynamics significantly, e.g. causing an inherently conservative system to
lose stability by flutter.

3. DISCHARGING VERSUS ASPIRATING CANTILEVERS

For a cantilevered pipe, equation (1) leads to

LlW = -MU iT
o
[(BW)2
Bt L
+ U (BW) (BW)
Bt L Bx L
1dt # O. (3)

If (Bw/Bx)L(Bw/Bt)L < 0, where the overbar indicates the long-time average,

Ll W > 0 is obtained, indicating that the pipe can gain energy from the fluid,
and indeed this is the mechanism of loss of stability by flutter at sufficiently high
U - say at Uc /. This also agrees with experimental observation (Bourrieres
1939; Benjamin 1961; Gregory & Paldoussis 1966).
It is interesting to remark that, if the fluid velocity is reversed, so that the
fluid is not discharged at the free end but it is aspirated instead, i.e. if U is
replaced by -U, then the reverse dynamical behaviour is predicted (Paldoussis
& Luu 1985): the system is initially unstable (0 < U < Uc /), and at Uc / it
regains stability! If dissipation is taken into account, this perplexing finding is
 253

Figure 5. Schematic of apparatus to test nonoccurrence of flutter in a pipe

aspirating fluid (Paidoussis 1999).

modified, but not radically. Apart from its inherent fundamental interest, this
has repercussions on ocean mining, e.g. of manganese nodules, where essentially
a vacuum-cleaner tube sucks water from the bottom of the sea, together with
the minerals. Is this system unstable? From the above, it would appear so.
Thanks to the late David J. Maull of Cambridge University, this paradox
was linked to the quandary that perplexed Richard Feynman in the late 1930s:
does a rotary lawn sprinkler rotate backwards if the water were sucked rather
than discharged?
The matter was finally resolved via a more careful assessment of the
problem, based on the "obvious" realization that the flows exiting and entering
the free end of the pipe are not at all similar - obvious, that is, after the
fact! So, replacing U by -U is not the total picture. Indeed, it was shown
that for the aspirated flow there is a depressurization at the pipe inlet and
throughout, equal to p = - M U 2 / A, A being the flow cross-sectional area.
This effectively counteracts the M U 2 term due to the momentum flux, which
is of cardinal importance in determining the dynamics - d. the second term in
equation (3). Hence, no instability at all is possible for the aspirating cantilever,
and similarly the sucking lawn sprinkler rotates neither backward nor forward!

An experiment was conducted, as shown in Fig. 5. After the pump was

switched on, the discharging pipe became bent as a result of the centrifugal force
M U 2 / R at the lower 90 elbow, R being the radius of curvature. The aspirating
254

pipe, however, after a brief transient, remained straight, since M U 2 IRis

exactly cancelled out by the opposing pAl R term, with pA = - M U 2 . So, this
paradox was totally resolved (Paidoussis 1999), and ocean mining may happily
carryon! Ever mindful of the readers' precious time, the conclusion was put in
the title a la Holmes: "Aspirating pipes do not flutter at infinitesimally small
flow". A fuller account may be found in Paidoussis (1998, 1999).

4. PIPES AND SHELLS WITH DIFFERENT END-SUPPORTS

In the 1980s a series of papers by Horacek & Zolotarev treated the dynamics of
fluid-conveying shells with different, or "asymmetric", end-support conditions;
e.g., shells simply supported at the upstream end and clamped at the
downstream one (ss-c, for short). It was found [see, e.g., Horacek & Zolotarev
(1984)] that ss-c shells conveying inviscid fluid exhibit flow-induced damping
for all U > 0 (see Fig. 6), before eventually losing stability by divergence. Even
more radical was the prediction that c-ss shells exhibit negative flow-induced
damping, i.e. they are unstable, again for all U > O!
On the other hand, calculations by Paidoussis et al. (1993) showed that

1------ 6
.,.
0
~

X
3
r
I
t
If 4
..,
0
~

x
2
: ".
./ I
I 6'
~
'-"
I 2
I ~

I
0 0
4 5
3 Uc/
U x 102 -2
I
I
I
I II
U I "
~l--~ ~t - - - - - - - .~ft

Figure 6. Argand diagram for a pinned-clamped system obtained by Honicek f3

Zolotarev (1983); (Pai"doussis 2003).
 255

both ss-c and c-ss systems behave as conservative systems, i.e., zero positive

r
or negative flow-induced damping up to the onset of divergence. Indeed, the
work done in a presumed period of oscillation is

~w ~ -p<a' I(n,A)U 1'[(:)' + U ( : ) ( : ) dt, (4)

where p is the fluid density, U the flow velocity, a and L the shell radius and
length, respectively, I(n,).) a functional of Bessel functions dependent on the
circumferential and axial wavenumbers, and w the radial shell displacement
corresponding to w(x, B, t) = w(x, t) cosnB. Clearly, in the case of supported
ends, whether c-ss or ss-c, one must have

LlW = O. (5)
So, again, the quandary arises of how can that be so, yet at the same time the
system be subject to flow-induced damping, positive or negative?
At first it was thought that Vol'mir's semi-membrane shell theory initially
used by Horacek & Zolotarev (H & Z from now on) was the culprit, but H & Z
later found the same kind of results with the fuller Goldenveizer-Novozhilov
shell theory. It was then thought that the problem must be numerical. [It must
be stressed here that the predicted damping is rather small; the damping ratio
is ( = Im(n)/Re(n) , so that ( ,...., 0(10- 3 ) in Fig. 6]. To this end, an extensive
study was undertaken by Misra, Wong & Paidoussis (2001) - M,W & P, for
short.
However, before that, it occurred to the author that, if this happens
with shells, then why not with pipes (governed by the Euler-Bernoulli beam
equation)? The initial set of calculations produced results similar to H & Z's
for shells! Thus, the c-ss system behaves as a nonconservative one, with
Im(w) < 0 for all U < Ucd, suggesting flutter from arbitrarily small U up
to the onset of divergence. Yet, the c-ss and ss-c systems are conservative
by conventional wisdom, as much as ss-ss and c-c systems are! This paradox
was resolved via more careful calculations. The comparison functions used in
the Galerkin-type solution were the beam eigenfunctions, the eigenvalues of
which had been computed to "only" 6 significant figures. Calculations with 9
significant-figure accuracy showed the c-ss to be totally conservative and stable
up to U = Ucd . (Interestingly, the ss-c system was found to be conservative
even with 4 significant-figure accuracy, indicating that the numerics in this
case are much less sensitive to imprecision.) So, elementary stability theory
was not turned on its head, and the virtuous could continue sleeping peacefully
at night! This also gave confidence to M, W & P that the cause of the H & Z
"rogue" findings was numerical inaccuracy.
Typical results for a c-ss shell conveying fluid are shown in Fig. 7. The
results in Fig. 7(a) suggest the same behaviour as found for the pipe; as the
256

precision is increased, the values of Im(O) are diminished, implying they would
vanish if precision were increased sufficiently; moreover, for the ss-c system with
U replaced by -U one obtains Im(O) ,..., 0(10- 7 ). Thus, between these findings
and equations (4) and (5), M, W & P (2001) concluded that the problem had
been resolved: the c-ss and ss-c systems are indeed conservative.

However, the paradox was resurrected by newer calculations by H & Z

(2002): with very small L/a, the values of Im(O) are significantly larger, too
large to be explained as due to numerical imprecision. Moreover, H & Z claim
to have proven in a soon-to-be published paper that the c-ss and ss-c systems
are nonconservative. Hence, the resolution of this paradox has proven to be
singularly slippery. Anyone who has tried to catch a fish with his bare hands
knows the feeling. The author believes that, ultimately, the system will be
shown to be conservative. The Im(O) f= 0 for arbitrarily small U found by
H & Z may perhaps be associated with incompletely defined fluid boundary
conditions (see Section 7). Clearly, further work is needed to clarify this issue .

....0
~
0 . -*-:-=i--"'~. ......... ... . . ........... .
x -0.2 .... - .. _.j. ~- ..... ~

e: -0.4 -x- .... _. ._ .,,- .:;;:----..

.""-.-.c
-0.6
~ -0.8 (a)
-1.0

-~~ o 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Dimensionless flow velocity, U

Figure 7. (a) Variation of Im(O) with U for a clamped-pinned system and

n = m = 2 for different accuracies in the axial wavenumbers and program
tolerances from simple double precision (x) to 10- 8 tolerance (0) . (b) The
same for a pinned-clamped system and reversed flow direction (Wong 2000;
Paidoussis 2003).
 257

5. TUBULAR BEAMS WITH INTERNAL AND EXTERNAL

FLOW: SIMILARITIES AND DIFFERENCES

In the simplest possible case of a pipe (tubular beam) containing flow, the
equation of motion may be written as

(6)

in which M and m are the mass of the conveyed fluid and of the pipe,
respectively, per unit length, and EI is the flexural rigidity. Similarly, the
simplest possible form of the equation of motion of a cylindrical beam immersed
in axial flow may be written as

(7)

where M here is the fluid added mass per unit length, and Fv represents the
viscous terms associated with longitudinal and transverse viscous forces along
the beam and a base drag at the downstream end,

(8)

i~.-.-.-.~.. __ ..~ / ' (al

<I..............~U~ ... ~~ (b)

~..-.-.-.-.. -.~
---+
.U

Figure 8. (a) Cantilevered pipe conveying fluid and (b) cantilevered cylinder
in axial flow, undergoing lateral motion (Pai'doussis 1998, 2003).
258

CT and CN are the longitudinal and tangential viscous force coefficients, CD

a zero-flow viscous damping coefficient, C b the base drag coefficient, D the
cylinder diameter, and p the fluid density; 15 = 0 if the beam is free (or free to
slide axially) at the downstream end, and 15 = 1 if it is positively fixed.
In the case of a beam with supported ends, the standard beam boundary
conditions apply, irrespective of internal or external flow. The same applies for
a cantilevered beam with internal flow, as shown in Fig. 8(a); in the case of
external flow, however, the system conventionally studied is usually terminated
by a short tapered end, as shown in Fig. 8(b). For this reason, the boundary
conditions are approximately given by

EI 81 y2 I
2
= 0,
uX x=L
(9)

where Xe is the length of the tapered end (xe / L < < 1), and f is a slenderness
parameter for the tapered end. For a well-streamlined tapered end, f = 1;
generally, however, f < 1 because of 3-D effects and flow separation. For a
blunt end, f = o. In general, therefore, 0 :::; f :::; 1. It is remarked that if f = 0
and Xe = 0 (blunt end) and furthermore Fv = 0, the equations of motion and
boundary conditions of the internal and external flow systems become identical.
Let us now compare the experimentally observed dynamical behaviour of
the two systems, Tables 1 and 2. The discrepancies in dynamical behaviour
are marked by the superscripts 1, 2, 3, and they are discussed in what follows.

Table 1. Observed dynamics of supported-end systems

Instability Internal flow External flow

Divergence Yes Yes
Coupled-mode flutter No Yes l

Table 2. Observed dynamics of cantilevered systems;

b.e.: blunt free end; t.e.: tapered free end

Instability Internal flow External flow b.e. External flow t.e.

Divergence No Yes 2
Flutter Yes Yes
 259

5.1. Divergence in supported-end systems

In the case of divergence in the external flow system, deleting all time-
dependent terms from (7) and (8) and taking J = 1 leads to

EI dx
4
d Y + M U2 [1 - 1
'2CT d2 Y
(-'21 L - x)/ D ] dx + '21 ( MUD
2/)
CN dy
dx = 0, (10)
4 2

where we have replaced pD CT by (pA/ D)CT == (M/ D)CT, with CT = (4/7r)CT ,

and similarly for the CN term.
For external flow and sufficiently small CT it is seen that the second term
acts as a compressive load. Hence, for a high enough flow velocity U, the system
loses stability by divergence, just like a column subjected to a compressive load.
For internal flow, CT = CN = 0, and exactly the same mechanism applies.
As seen in Table 1, the results for the two cases agree.

5.2. Divergence in cantilevered systems

It is well known that a column subjected to a follower load does not lose stability
by divergence, and the same clearly applies to the case of a beam with internal
flow.
We next consider the case of external flow and a blunt free end. For f = 0,
the boundary conditions become identical to those for internal flow and, since
J = 0, equation (10) is replaced by

d4 y 1 ] d2 y
E I dx 4 + M U2 [ 1
1 - '2 CT (L - x) / D - '2 Cb dx 2 + '21 ( M U 2 / D ) CN dy
dx = 0,
(11)
where Cb = (4/7r)Cb; hence, again, for sufficiently small CN and CT , and since
Cb '" 0(1), this system is subject to a follower force, and loss of stability by
divergence is not possible, hence agreeing with the observations of Table 2.
For the case of a tapered free end, equation (11) still holds, but the second
boundary condition (9) is

(12)

Indeed, it is the second term in this equation that is responsible for the
divergence, as per Table 2 (superscript 2). For those with an aeronautical
background, it may be interesting to note that the mechanism is very similar
to that giving rise to lift in delta wings (Triangafyllou 1998). The pertinent,
second term in the second equation (9) may be written as f MUv . The lift on a
low aspect-ratio wing (AR = span/chord = s/c 1) is L = ~pU2 ACL, with
A = cs and C L the lift coefficient. Taking CL ~ (dCL/d'ljJ)'ljJ ~ (dCL/d'ljJ)(v/U)
and recalling from slender wing theory (Katz & Plotkin 1991) that dCL/d'ljJ ~
~7rAR, it is seen that L = ~pS2 Uv = M*Uv, where M* is the added mass
260

for the wing. Thus, just as slender wing theory predicts, the taper is essential
for generating lift; and for f =I- it is this lift that causes the divergence. This
explains the discrepancy (superscript 2) in Table 2.

5.3. Coupled-mode flutter in end-supported systems

The work done by all the forces on the system in a period of oscillation is found
to be

,6.W = - ~cN(MU/D) loTloL (ii + Uyy/)dxdt

- ~(M/D) loTlo\C*y2 - CT U 2 yy/)dxdt, (13)

in which c* is related to CD, and ( )' = a/ax, (.) = a/at. Hence, for
instability,

For internal flow, CN, CT and c* are all zero, and hence ,6. W = 0, which
is the case discussed in Section 2, concluding that coupled-mode flutter is not

possible, as in Table 1.
For external flow, however, ,6.W > and hence coupled-mode flutter are
in principle possible, as per equation (14), and in fact flutter does arise. This
explains the discrepancy between internal and external flow dynamics, marked
by superscript 1 in Table 1.

5.4. Flutter in cantilevered systems

In this case, we have

L {l
- 10rT 10r MU (.2
"2 cw-[) y + Uyy I) +"21M
D
( *
C
2
Y - CT
U 2 . I)} d d
yy x t, (15)

where, the first two terms are the most important.

For internal flow, f = 0, Cb = CN = CT = c* = 0, and we recuperate
equation (3) and reach the same conclusion: if JOT yyl dt < 0, then ,6.W > 0,
and hence flutter is possible.
For external flow and a blunt end (f = 0), a large contribution is possible
from the inviscid forces (the first term). There is, however, an opposing and
 261

possibly large contribution from the base drag (second term). Evidently, the
two together result in no flutter in the experiments. Indeed, if calculations are
performed with Cb = 0 artificially, then flutter becomes possible. Hence, it is
the presence of the base drag that is responsible for the discrepancy between
internal and external flow, superscript 3 in Table 2.
For external flow and a streamlined end (f :j:. 0), some work is done by
the first term, again opposed by a diminished second term, but evidently the
balance is such as to render flutter possible. It is interesting that for f = 1 and
Cb = 0, the system becomes quasi-conservative, and only coupled-mode flutter
is possible, as shown by calculation.
Thus, although the system is most strongly nonconservative in the case of
f = 0 in terms of inviscid forces, the viscous forces (particularly the base drag)
produce anomalies in the internal/external flow analogy.

6. TRAVELING WEB AND EXTRUSION F.S.I.

In extrusion systems, one usually ignores fluid-structure interaction if the

ambient fluid is air, because the fluid added mass per unit length is so much
smaller than that of the object being extruded. This also applies, a jorteriori,
to deployment of appendages in space. The same, however, does not hold true
for extrusion or deployment in dense fluid, e.g. in oil in the first case or water
in the latter. Also, fluid-structure interaction for traveling web, albeit in air, is
generally nonnegligible, for the fluid added mass for sufficiently wide and thin
web may not be negligible as compared to its own mass per unit length.
In the early work on the dynamics of traveling web (Pramila 1986), modeled
as a tensioned strip, the equation of motion was taken to be

(16)

where V is the forward speed of the web, of mass per unit lenght m, and
M = 0: [p(7r/4)b 2 ] is the fluid added mass, b being the web width and p the
fluid density; generally, 0: :s: 1. Similarly, for a beam being deployed in dense
fluid, Taleb & Misra (1981) have used the following equation (simplified here,
with no loss of generality):

(17)

where i = aL/at.
262

Later, Pramila (1987) corrected the equation for the web, noting that apart
from entrained fluid (i.e. apart from boundary-layer effects) the ambient fluid
does not have a mean axial velocity, but a mainly lateral one due to transverse
web motions. Hence, equation (17) was replaced by

02w ( 0 0 ) 2 o2y
- T ox 2 + m ot + V ox y + M ot2 = 0 ; (18)

thus, effectively, the surrounding fluid contributes only an added mass.

Experimental data tend to support this. The question is whether the same
applies to the extrusion system. This turns out to be a rather subtle issue.
An appealing conceptual device for the web problem is the following. The
web is imagined to be sandwiched within a massless and frictionless tensioned
sheath. Thus, the system is similar to a pipe of vanishing mass containing an
ideal fluid of mass m per unit length traveling within the pipe/sheath with
velocity V, and immersed in a stagnant fluid with added mass M per unit
length. Hence, equation (18) is justified without difficulty.
It is tempting then to think of the extrusion problem in a similar fashion,
but (i) to consider that now the sheath is stretching and advancing relative
to the fluid, and (ii) to recast this flow problem into one involving the fluid
traveling relative to the sheath rather than vice versa, thus justifying the term
M[(%t) + L(O/OX)]2 y in equation (17). In this, the analogy to the swimming
of a slender fish may be invoked (Lighthill 1969, 1975) , in which the term
M[(%t) + V(%x)J2y is the dominant fluid-dynamical term in the equation
of motion, V being the fish forward speed.
There is, however, a subtle fallacy in this, as uncovered with the help of
C.D. Mote (2002). In fact, the sheath in the above conceptual model travels at
the same speed as the deploying beam within, and hence is part of the beam;
alternatively, one might consider the sheath to be immobile and infinitely long
and the beam to be deploying within it. Furthermore, relative to the fixed root
of the beam, there is no axial flow velocity, and hence Coriolis and centrifugal
forces ML(02 y/ oxot) and ML 2(fJ2y/8x 2) do not arise. Indeed, the axial fluid
momentum M L = 0, and the correct equation of motion in this case is

(19)

The fish problem, however, remains intact, for in this case there is no fixed
portion of the fish, and one is justified in transforming the problem from one
in which the fish moves forward in quiescent fluid to one of a fish undulating
in place in a backward flowing fluid. For further discussion, refer to Pai:doussis
(2003). Currently, work is under way to see how significantly different is the
dynamics of beams under deployment (otherwise appealingly referred to as the
inverse spaghetti problem) as predicted by equation (19) instead of (17).
 263

7. THE IMPORTANCE OF FLUID BOUNDARY CONDITIONS

Simple analyses of the dynamics of pipes conveying fluid, e.g. via equation
(6), can be carried out without worrying about what happens to the fluid
beyond the downstream end, x = L, or upstream. The same applies to shells
conveying fluid. E.g., for a cantilevered shell one needs not necessarily specify
how pressure perturbations decay after the fluid leaves the free end. This
is strictly so if one uses a "point- formulation" for the perturbation field, in
which, effectively, each segment is treated as if it were infinitely long; hence
the pressure perturbation at that segment may be formulated irrespective ofthe
overall perturbation pressure field. On the other hand, if a Fourier-transform
solution is used, the method itself demands the specification of this pressure
from -00 to 00.
For the pipe problem, analyses ignoring the fluid dynamics beyond the pipe
extremities are remarkably successful. To test that this is also the case for the
shell problem, Shayo & Ellen (1974) first, and others since, have constructed
so-called outflow models for the flow beyond the downstream end of the shell. It
was found that, provided the length/radius ratio L/a is large, in this case also
the outflow dynamics is not of crucial importance in determining the dynamics.
On the other hand, as first shown by Miller (1970) for a blade in narrow
channel flow, if the system is stable in un constricted flow, (i) a localized
constriction upstream can generate a negative fluid dynamic damping, and
hence result in flutter, and (ii) a localized downstream constriction can produce
a positive damping, thus enhancing stability. This also holds true for flexible
or flexibly mounted cylinders in annular flow, as first shown by Hobson (1982)
and Spurr & Hobson (1984) and confirmed experimentally. The situation is
more subtle if the annulus is terminated by a diffuser. It is shown that for a
diffuser with sufficient pressure recovery, it is possible to induce negative fluid-
dynamic damping and hence instability. Therefore, a thorough understanding
and specification of the flow field, at the extremities of, and even beyond, the
domain of the structural problem (0, L), is necessary for proper prediction of the
dynamical behaviour. These questions are thoroughly discussed in Paidoussis
(2003; Chapter 11).
Another area in which upstream and downstream effects, e.g. flow
resistance, are crucial in determining the dynamics is related to flow in pliable
shells or collapsible tubes, modeling venal or pulmonary dynamics; see, e.g.,
Bertram (1995) and Paidoussis (2003; Chapter 7).
Sample results for a flexibly supported rigid cylinder in annular flow are
shown in Table 3, where two effects are simultaneously investigated: (i)
specification of end conditions, i.e. that the perturbation pressure vanishes
beyond the problem domain via Heaviside functions, and (ii) for potential flow,
whether the nonpenetration condition is specified on the moving boundary or
264

Table 3. The effects (i) and (ii), defined in the text, on stability of a centrebody
in annular flow; Ka: added stiffness; C a : added damping

Motion Flow Effects (i) & (ii) Ka CA Instability

Rocking Potential No =0 =1=0 Flutter
Potential Yes =1=0 =0 Divergence
Viscous No =1=0 =1=0 Div /flutter
Viscous Yes =1=0 =1=0 Div /flutter
Lateral Potential No =0 =0
Potential Yes =1=0 =0 Divergence
Viscous Yes =1=0 =1=0 Div /flutter

at its mean position. It is clear that the effect is profound.

The same sensitivity to these effects applies to numerical solutions, which
require a solution domain larger than the (0, L) problem domain: specifically
what does one specify for the transition between one set of mesh points on the
moving body and a rigid extension beyond? If the transition is smooth, this
implies the existence of a gossamer physical connection, which is enough to
render conservative a system that may in fact not be. However, this discussion
is beyond the scope of this paper.

8. CONCLUSION

A number of interesting quadaries or paradoxes have been discussed, all

involving fluid-structure interactions, drawn mainly from PaYdoussis (1998,
2003), wherein the interested reader may find a fuller discussion, as well as
other cases.
The author gratefully acknowledges the support of NSERC of Canada and
FCAR of Quebec for this research.

9. AFFILIATION

Thomas Workman Emeritus Chair, Department of Mechanical Engineering,

McGill University, 817 Sherbrooke Street W., Montreal, QC, Canada H3A
2K6. E-mail: mary.fiorilli@mcgill.caandmichael.paidoussis@mcgill.ca.
 265

REFERENCES

Amabili, M., Pellicano, F. & PaYdoussis, M .P. 1999 Nonlinear dynamics and stability of
circular cylindrical shells containing flowing fluid. 1. Stability. Journal of Sound and Vibration
225, 655-699.
Benjamin, T.B. 1961 Dynamics of a system of articulated pipes conveying fluid. II. Experi-
ments. Proceedings of the Royal Society (London) A 261, 487-499.
Bertram, C.D. 1995 The dynamics of collapsible tubes. In Biological Fluid Dynamics
(eds C.P. Ellington & T.J. Pedley), pp. 253-264.
Bourrieres, F.-J. 1939 Sur un phenomene d'oscillation auto-entretenue en mecanique des
flu ides reels. Publications Scientifiques et Techniques du Ministre de l'Air, No. 147.
Dodds, H.L. Jr. & Runyan, H.L. 1965 Effect of high velocity fluid flow on the bending
vibrations and static divergence of a simply supported pipe. NASA Technical Note D-2870.
Done, G.T.S. & Simpson, A. 1977 Dynamic stability of certain conservative and non-
conservative systems. I.Mech.E. Journal of Mechanical Engineering Science 19, 251-263.
Feodos'ev, V.P. 1951 Vibrations and stability of a pipe when liquid flows through it.
Inzhenernyi Sbornik 10, 169-170.
Gregory, R.W. & Pai'doussis, M.P. 1966 Unstable oscillation of tubular cantilevers conveying
fluid. II. Experiments. Proceedings of the Royal Society (London) A 293, 528-542.
Hobson, D.E. 1982 Fluid-elastic instabilities caused by flow in an annulus. Proceedings of
3rd International Conference on Vibration in Nuclear Plant, Keswick, pp. 440-463. London:
BNES.
Holmes, P .J . 1977 Bifurcations to divergence and flutter in flow-induced oscillations: a finite-
dimensional analysis. Journal of Sound and Vibration 53, 471-503.
Holmes, P.J. 1978 Pipes supported at both ends cannot flutter. Journal of Applied Mechanics
45, 619-622.
Horacek, J. & Zolotarev, 1. 1984 Influence of fixing the edges of a shell conveying fluid on its
dynamic characteristics. Soviet Applied Mechanics 20, 756-765.
Horacek, J. & Zolotarev, 1. 2002 Free vibration and stability of cylindrical shells in interaction
with flowing fluid. In NATO CLG Grant Report No. PST.CLF.977350 (eds F. Pellicano,
Y. Mikhlin & 1. Zolotarev), pp. 45-82. Prague: Institute of Thermomechanics (ISBN 80-
85918-76-5).
Housner, G.W. 1952 Bending vibrations of a pipe line containing flowing fluid. Journal of
Applied Mechanics 19, 205-208.
Karagiozis, K.N., Grinevich, E., Pai'doussis, M.P., Misra, A.K. & Amabili, M. 2003 Stability
and non-linear dynamics of clamped circular cylindrical shells containing flowing fluid.
Proceedings IUTAM Symposium on Integrated Modeling of fully Coupled Fluid-Structure
Interactions (eds H. Benaroya & T. Wei), Rutgers University, U.S.A.
Katz, J. & Plotkin, A. 1991 Low-Speed Aerodynamics: from Wing Theory to Panel Methods.
New York: McGraw-Hill.
Lighthill, M.J. 1969 Hydromechanics of aquatic animal propulsion. Annual Review of Fluid
Mechanics 1, 413-446.
Lighthill, M.J. 1975 Mathematical Biofluiddynamics. Philadelphia, PA: SIAM.
Miller, D.R. 1970 Generation of positive and negative damping with a flow restrictor in
axial flow. In Proceedings of the Conference on Flow-Induced Vibrations in Reactor System
Components (eds G.S. Rosenberg, M.W. Wambsganss & R.P. Carter), Report ANL-7685,
pp. 304-311 , Argonne National Laboratory, Argonne, IL, U.S .A.
Misra, A.K., Wong, S.S.T. & Pai'doussis, M.P. 2001 Dynamics and stability of pinned-clamped
and clamped-pinned cylindrical shells conveying fluid. Journal of Fluids and Structures 15,
1153-1166.
Mote, C.D. Jr. 2002 Private communication (15 Mar 2002).
266

Nguyen, V.B., Paldoussis, M.P. & Misra, A.K. 1994 A CFD model for the study of the
stability of cantilevered coaxial cylindrical shells conveying viscous fluid. Journal of Sound
and Vibration 176, 105-125.
Paldoussis, M.P. 1998 Fluid-Structure Interations: Slender Structures and Axial Flow.
Volume 1. London: Academic Press.
Paldoussis, M.P. 1999 Aspirating pipes do not flutter at infinitesimally small flow. Journal
of Fluids and Structures 13, 419-425.
Paldoussis, M.P. 2003 Fluid-Structure Interations: Slender Structures and Axial Flow.
Volume 2. Oxford: Elsevier.
Paldoussis, M.P. & Denise, J .-P. 1972 Flutter of thin cylindrical shells conveying fluid.
Journal of Sound and Vibration 20, 9-26.
Paldoussis, M.P. & Issid, N.T. 1974 Dynamic stability of pipes conveying fluid. Journal of
Sound and Vibration 33, 267-294.
Paldoussis, M.P. & Luu, T.P. 1985 Dynamics of a pipe aspirating fluid, such as might be
used in ocean mining. ASME Journal of Energy Resources Technology 107, 250-255.
Paldoussis, M.P. & Semler, C. 1993 Nonlinear dynamics of a fluid-conveying cantilevered
pipe with an intermediate support. Journal of Fluids and Structures 7, 269-298.
Pramila, A. 1986 Sheet flutter and the interaction between sheet and air. TAPPI Journal
69,70-74.
Pramila, A. 1987 Natural frequencies of a submerged axially moving band. Journal of Sound
and Vibration 113, 198-203.
Shayo, L.K. & Ellen, C.H. 1974 The stability of finite length circular cross-section pipes
conveying inviscid fluid. Journal of Sound and Vibration 37, 535-545.
Spurr, A. & Hobson, D.E. 1984 Forces on the vibrating centrebody of an annular diffuser.
Proceedings Symposium on Flow-Induced Vibrations, Vol. 4 (eds M.P. Paldoussis & M.K.
Au-Yang), pp. 41-52. New York: ASME.
Taleb, LA. & Misra, A.K. 1981 Dynamics of axially moving beam submerged in a fluid.
AIAA Journal of Hydronautics 15, 62-66.
Triantafyllou, M.S. 1998 Private communication (3 Nov 1998).
Weaver, D.S. & Unny, T.E. 1973 On the dynamic stability of fluid-conveying pipes. Journal
of Applied Mechanics 40, 48-52.
Wong, S.T.-M. 2000 Dynamics of fluid-conveying shells with symmetric and asymmetric
end-supports. M.Eng. Thesis, Department of Mechanical Engineering, McGill University,
Montreal, Quebec, Canada.
 CLOSED-LOOP CONTROL OF THE RESONANT
FLOW-STRUCTURE INTERACTION
USING PID CONTROLLERS

M. M. ZHANG, L. CHENG AND Y. ZHOU

Abstract: Closed-loop control of vortex shedding and flow-induced vibration of a spring-supported

square cylinder in cross flow is experimentally investigated. The fluid field is perturbed through the
cylinder surface oscillation generated by three piezoelectric ceramic actuators, embedded underneath the
surface and controlled by a Proportional-Integral-Derivative (PID) controller. Three control strategies
were investigated using different feedback signals, including the turbulent flow signal measured by a
hot-wire, structural oscillating signal obtained by a laser vibrometer and a combination of both signals.
Investigation was conducted at the resonance condition, when the vortex-shedding frequency coincided
with the natural frequency of the fluid-structure system. The flow and structural vibration were
measured using particle image velocimetry, laser-induced fluorescence flow visualization, laser Doppler
anemometer and laser vibrometer. It is observed that the closed-loop perturbation may lead to an almost
complete destruction of the Karman vortex street and a reduction in the structural vibration, vortex
shedding strength and drag coefficient by 82%, 65% and 35%, respectively, outperforming by far an
open-loop control.

1. INTRODUCTION
Control of flow-induced vibration on a bluff body due to periodic vortex shedding
is of fundamental interest in many engineering applications. A variety of control
techniques have been developed in the past, which can be roughly classified as
passive and active controls. Passive techniques rely on modifying geometry, adding
vortex generators, grooves or riblets to bluff bodies to affect the formation of the
vortex shedding (Zdravkovich, 1981); they require no external energy input to the
system. In contrast, active methods involve the input of energies via actuators to
bring desirable changes to the flow-structure system, using either independent
external disturbance or feedback-signal controlled system. The former is often
referred to as open-loop control, whereas the latter is called the closed-loop control.
In both cases, the control performance strongly depends on activating mechanisms
and, in latter case, also on the control strategy used.
Typical examples of the open-loop control include acoustic excitation (Hsiao,
1991), oscillating cylinder (Bearman, 1984) and surface bleeding (Williams, 1992).
Recently, Cheng et al. (2003) investigated a novel open-loop technique using
curved piezoceramic actuators embedded underneath the surface of a square
cylinder to alter interactions between a flexibly supported cylinder and cross flow.
Provided the perturbation frequency was properly set, both vortex shedding and
induced vibration were significantly reduced as a result from actuator-generated
surface perturbation. This open-loop technique, however, suffers from two major

267
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 267-277.
2003 Kluwer Academic Publishers.
268

drawbacks. First, the perturbation frequency range to achieve desired performance

is relatively narrow. Second, the required perturbation amplitude was rather large,
about 2.8% of the cylinder height or 25% ofthe vibration amplitude ofthe cylinder.
These problems might be resolved if a closed-loop system is developed.
The choice of the feedback signal is crucial for the performance of a closed-
loop system. Most previous closed-loop techniques involving flow or flow-induced
vibration control have their feedback signals from flow (hot-wire signals). See
Ffowcs Williams et al. (1989), Huang (1996), Berger (1967) and Filler (1991) for
examples. This strategy should work quite well provided flow is to be controlled.
For the same token, one may consider the structural vibration signal to be ideal for
the control of structural vibration. One example is Baz et al. (1991). Zhang et al.
(2003) investigated the closed-loop control of vortex shedding and flow-induced
vibration of a flexibly supported square cylinder in cross flow based on a
Proportional-Integral-Derivative (PID) controller. Their feedback signal was
provided by the streamwise fluctuating velocity measured by a hot-wire. They have
achieved an effective control of both vortex shedding and flow-induced vibration.
However, the performance of their system was not significantly superior to an
open-loop system. One may surmise that their feedback signal was from flow only,
containing no information on structural vibration or flow-structure interactions, and
might not provide the optimum feedback signal to control a fluid-structure
interaction system. This begs the question: which is the best feedback signal, flow
or structural vibration or something else?
The present investigation pursues two objectives: (1) to improve the control
systems developed by Cheng et al. (2003) and Zhang et al. (2003) and find an
optimum strategy to control fluid-structure interactions, (2) to shed light upon the
underlying physics of flow-structure interaction under external perturbation. Three
control strategies, utilizing feedback signals from either turbulence flow, structural
vibration or the combination of both, are considered and compared.

2. EXPERIMENTAL DETAILS

All experiments were performed in a closed circuit wind tunnel with a square
working section of 0.6 m x 0.6 m and 2.4 m long providing a uniform flow velocity
50 ms l ). More details of the tunnel were given in Zhou, et al. (2002). A square
cylinder of height h = 15.2 mm, flexibly supported on springs at both ends, was
placed 0.2 m downstream of the exit plane of the tunnel contraction and allowed to
vibrate laterally, as shown in Fig. l. The free-stream velocity (U",) was adjusted to
be about 3.58 ms l , corresponding to a Reynolds number, Re, of 3500. At this Re,
resonance occurred, that is, the vortex shedding frequency Is coincided with the
natural frequency f~ (= 30 Hz) of the fluid-cylinder system, the maximum
displacement of cylinder, Ymax, being about l.2 mm or 0.08h. Details about the
installation of the cylinder and characteristics of the actuators can be found in
Cheng et al. (2003). As shown in Fig. 1, the upper side of the cylinder, parallel to
the flow, was made ofa thin plastic plate (l3.8 mm x 493 mm, 2/3 of the cylinder
 269

length) of 3 mm thick, which was installed symmetrically about the

mid span

-
Flow

9 1
A-A

Plastic
Cylinder Plate

Figure 1. Experimental Setup. (f) Monitor hot-wire was located at x = 2h, Y = 1.5h, Z = 0; (?)
Feedback hot-wire was located at x = 1.6h, Y = -2.5h, z = O.

of the cylinder and flush with the rest of the cylinder surface. Three curved
piezoelectric ceramic actuators were embedded in series in a slot underneath the
plate. When placed within an electric field, the piezoelectric effect resulted in a
strain in material. Under an applied voltage, the actuator deformed out of plane,
driving the thin plate up and down and generating the desired surface perturbation.
The lateral structural displacement (Y) was measured by a laser vibrometer, and
the streamwise fluctuating velocity (u) by a 5 f.lIT1 tungsten hot-wire, placed at xlh =
1.6, ylh = -2.5 and zlh = 0 (hot-wire <6l in Fig.I). Both signals were used either
separately or simultaneously as feedback signals. After conditioning, the feedback
signals were low-pass filtered at a cut-off frequency of 200 Hz and then sent to a
proportional-integral-derivative (PID) controller built in with a 16 bit AD and DA
converter. The control signal was amplified by two dual channel piezo driver
amplifiers (Trek PZD 700) in order to drive the piezoceramic actuators.
The LIF flow visualization and PlV measurements were conducted using a
Dantec standard PlV21 00 system. The digital particle images were taken by a CCD
camera (HiSense type 13, gain x 4, single for LIF or double frames for PlY, 1280 x
1024 pixels) and the illumination was given by two New wave standard pulsed
laser sources of a wavelength of 532 nm, each having a maximum energy output of
270

120 mJ. A Dantec FlowMap Processor (PIV2100 type) was used to synchronize
image taking and illumination. Flow velocities, u and v, along the x and y direction,
respectively, in the wake (xlh = 3) were measured using a two-component LDA
system (Dantec Model 58N40).
To monitor the control performance, the structural vibration and fluctuating
flow velocity were measured simultaneously by the laser vibrometer and another
hot-wire (hot-wire CD in Fig. 1). The latter was located at xlh = 2, ylh = 1.5 and zlh
= o. Both signals were conditioned and digitized using a 12-bit AD board at a
sampling frequency of 3.5 kHz per channel. The duration of each record was about
20 s.

3. CONTROL STRATEGIES AND CONTROLLER DESIGN

Three control strategies were considered depending on feedback signals used,
namely, PID-Y, PID-u and PID-Yu, referring to PID control using the Y signal
measured by laser vibrometer, the u signal measured by hot-wire (?) and the
combination of the two signals, respectively. The controller was developed and
implemented based on dSPACE system, which had a real-time system for rapid
control prototyping, production code generation, and hardware-in-the-Ioop tests. A
digital signal processor (DSP) with SIMULINK function of MATLAB and
software (ControlDesk 2.0) was used for sampling and processing feedback signals.
The output of a PID controller is proportional to the sum of the input signal, its
integral and derivative. The proportional gain (P), integral gain (I) and differential
gain (D) of a PID controller can be individually or simultaneously adjusted. For
each control strategy (PID-Y, PID-u or PID-Yu), gain coefficents should be adjusted
during experiment to achieve a maximum reduction in the cylinder vibration
amplitude Yand flow fluctuation velocity u. The tuning procedure was to first keep
1= D = 0 and vary P until the root mean square (rms) values, Ynns and unns , of Yand
u reached the minimum. Then I and D were successively added and adjusted until
the optimal performance was achieved. The same procedure was followed
separately for PID-Y or PID-u, PID-Yu. The tuning process led to the following
optimal control parameters: P y = 1.2,!y= -0.3, D y= -0.0004 for PID-Y, P u = 3.5, Iu
= 0.2, Du = 0.0001 for PID-u and P y = 1.2, Pu = 004, !y = 0.2, Iu = 0.2, Dy = 0.001,
Du = 0.0001 for PID-Yu. Unless otherwise stated, the parameters were used in
experiments.
It was found during the tuning process that, irrespective of control strategies, P
was much more effective than lorD in controlling vortex shedding and flow-
induced vibration. P control generates a control action that is proportional to
structural oscillation velocity (Y) and flow velocity, thus physically causing a
change in the system damping. Theoretically, the resonant flow-structure system
was surely very sensitive to any damping variations. On the other hand, I and D
controls are physically linked to displacement and acceleration feedback,
respectively. The former has impact upon the system stiffness, whereas the latter
influences the effective mass. Both may, in principle, alter f~ to some extent.
 271

However, this slight change in f~ IS probably not enough to cause any

considerable effect on the strongly coupled vortex and structure synchronization,
which spans over a certain lock-on frequency range (Gowda 1975).
0.06 0.4
0.0 5 5 0.36
r""
0.3 0

D
Y ' ''' ' I h
U ,m , I U '"

o .3
0.26
0.04
0.033
Y rms I h 0 .2 U rm . ! U '"
0.0 25

0 . 02

'~'
0.0 10
O. I

U nperturbed P IDY P ID -u P I D -Y u
C on Ir o 1 C ontrol C ontrol

Figure 2. Comparison in Urms and Yrms among various control strategies.

4. PERFORMANCE OF VARIOUS CONTROL STRATEGIES

Figure 2 compares control performances of three control strategies in terms of Yr: s

and U ;rns , which was measured by the monitor hot-wire CD. Unless otherwise stated,
the asterisk denotes the normalization by hand U in this paper. Compared to the 00

unperturbed case, Yr: s and u;rns were reduced respectively by 40% and 17% using
PID-Y, 53% and 32% using PID-u and 82% and 70% using PID-Yu. Evidently,
PID- Yu overwhelms the other two strategies in performance, and PID-u
considerably exceeds PID- Y. It seems that in all cases synchronizing vortex
shedding and cylinder oscillation was greatly weakened. Cheng et at. (2003)
attempted to manipulate the same fluid-structure system using an open-loop control
system. In their case, whether the flow or structural vibration was enhanced or
impaired depended on the perturbation frequency (f,,). Both Yrrns1h and urmsiUro were
reduced outside the synchronization range, i.e. J"='p J p hlU =0.11 - 0.26, but 00

increased within the range, doubling at f; = fph / U = 0.13. Furthermore, the 00

maximum reduction was 75% in Yrmslh and 68% in urrnslU,," considerably less than
that achieved by PID- Yu.
Typical photographs from flow visualization are presented in Fig. 3. The solid
square in the figure denotes the cylinder position. The case without any external
perturbation is given in Fig. 3(a) as a baseline for comparison. The best performed
open-loop controlled case when f; = 0.1 "* is" = f~* = 0.13 (Cheng et at. 2003) is
also included. The iso-contours of the normalized spanwise vorticity (not shown),
lU; = lUzh I Uoc; were calculated, from the PIV measurement, which provide
272

(a) (b)

Flow Flow

(c) (d) (e)

Flow Flow Flow

 273

Yu requires a lower actuating voltage and therefore smaller perturbation amplitude

than the other two control strategies but achieves a much better performance.
The overall performances of the three closed-loop control strategies, together
with open-loop control case (f; = 0.1) are summarized in Table I for comparison.
The circulation (r) around a vortex is estimated by numerical integration r*

r t.A
U h = I,I(coz )ij -h2 (Brain et al., 1983), where (co z )ij IS PlY-measured
00 )

spanwise
0.' 0.' .;" .,' " ", . - - -- - -.. - -,
.d 0.05 .s= O.O j
~ 0 o ::: 0
>- -OM >- >- .O. O ~

.0.' ~=~=====::::;"
, 0 .'
0.4 0 .4
0.'

2
g

'";" 0
-0 .4 ;:S
0
. 0 .4
-0. 8
:=:~:::'::::::~=~=~
I .:1 tl!lliillllilli!!--U-H!!!!u!!!!i 1"'
0.'

! :: ~
~ . --.-. . ,i
>~ -60 ~m~~.m n_~,~_m n,n n~n~~m., >
Q. -30
-60
.90 ........~~~~~~~.....J 90 . . . .~~~~~~~-"
o 0.2 0.4 0.6 0 .8 1.0 1.2 1.4 o 0.2 0.4 0.6 0 .8 1.0 1..2 1. 4 o 0.2 0.4 0.6 0.8 1.0 1.2 1.4
t (secs) t (sees) t(secs)

(a) (b) (c)

Figure 4. Typical transition of structural vibration flow velocity (u) and p erturbation ro,
voltage (VpJ signals when PID controller was switched on. (a) PID- Y; (b) PID-u; (c) PID- Yu.

Table 1. Comparison among various control strategies

C ontrol Open-loo p
Strategy f;
= 0. 1 PID- Y PID-u PID-Yu

75% .!- 40 % .!- 53% .!- 8 2%.!-

u"". 68 % .!- 17 % .!- 32 % .!-

5 0% .!- 22% .!- 34% .!- 65 % .!-

141.4 volts 83 .7 volts 4 7 .4 volts 27.1 volts

35.1 % 20 .5 % 14.7 % 7 .0%

21.0%.!- 35 . 1%'!-

vorticity over area t.A = AxL'ly, Ax and L'ly being the integral step along x and y
directions, respectively. Integration was conducted over an area enclosed by the
cutoff level lco;c I= 0.3, about 7% of lco;max I' as used by Brian et al. Errors
274

associated with the r estimate was about 15%. The drag coefficient, CD, was
calculated based on the cross-flow distributions of mean velocity U and Reynolds
-* -*
stresses u 2 ,v 2 and uv measured by LDA at xlh = 3 of the flows with and
without perturbation (not shown).

C =2f'" U
D - 00 U
tooU-U)d(~)+2foo
h - 00
(v2 _u 2 )d(L)
U2 h
00 00 '"

(1)
(Antonia and Rajagopalan, 1990). It can be seen that PID-u control outperforms
PID- Y control in every category, resulting in a higher reduction percentage in Ynns/ h,
urmslUoo and [' despite smaller perturbation voltage amplitude. The PID-u strategy
has its control (feedback) signal from flow and therefore the effect of the control
action is to suppress directly the flow excitation and subsequently (or indirectly) the
structural vibration. On the other hand, the PID-Y strategy tackles directly the
structural vibration, which provides control information. Naturally, this strategy
may not be so effective in suppressing flow excitation, the corresponding decrease
in Urms being only 17%. As a matter of fact, even the decrease in Y rms is less than
that of PID-u. The result conforms to the fact that the excitation of structural
vibration originates from the fluid forces and the structural vibration is a passive
response of fluid excitation. It is surely a more effective control strategy if the
excitation origin is tackled, rather than the passive response.
Among all control methods listed in Table 1, PID-Yu has unequivocally the best
performance in minimizing Yr~s , u;ms and T. Its required actuating voltage Vp or
the perturbation amplitude Yp is only 50%, 30% and 20% of that used for PID- Y,
PID-u and the open-loop control (f; = O.l), respectively. The observation suggests
the possibility to develop a more compact, self-contained and low energy control
system if the PID-Yu strategy is applied. Additionally, without perturbation, CD was
1.88. CD drops by 21.0% for the open-loop control and by 35.l% for the PID-Yu
control. Hsiao (1991) observed a reduced CD in an acoustically excited circular-
cylinder wake. The observation was linked with a narrower wake and the smaller
defect of mean velocity profile. The cross-flow distribution of U suggests an
increasing wake width due to the perturbation on the cylinder. However, the
-,
maximum u 2 and v 2 are reduced because of the perturbation. It is therefore
proposed that the perturbation leads to greatly weaken flow separation or vortex
shedding and subsequently an increased backpressure. As a result, CD decreases.

5. DISCUSSIONS

To understand the physics behind weakened vortex shedding and structural

vibration, the phase shift ( I/Jyu ) between vortex shedding and structural vibration is
 275

calculated from simultaneously measured Yand u using tPYu == tan -1 (Qyu / CO Yu ) ,

where CO Yu and Qyu represent the cospectrum and quadrature spectrum of Yand u,
respectively. The spectra were calculated from the FIT of the correlation
Y(t + .)u(t) . Cheng et al. 's (2003) analysis indicated that the phase relation
between the lateral velocity, v, of the flow around the cylinder and the lateral
structural oscillating velocity, Y, could be approximately described by tPYu; rPYu =
o corresponded to synchronization between v and Y, and tPYu = 7r the opposite
movement between Y and v. Without
3~,j,f.'
= O'13
o W
~L:J0~ I
\l
(a)
.3 ~~~~~~~~~~

.: ~~) o:
<
Q>a>loop: =0 \
i A 0.15
j
=
(b)

"
~.:~(C)
3 [nrv.!!!(1f:\A A hJM iU" \ ~PID-Y 0.42 iCC)

.: t~
o '

tf. ,\~(0+d) ~ E::~P;:;;ID

;:; ';;::;U;f;;G~~~~0.2~4~~~1
.: t\!J,
(d)

tf: ~ (e) ~ ~..

[ PID - Yu ;, I
!. 0.11 (e)

0.05 0.\ 0.\5 0.2 0.25 0.3 10-2 10-\ 10

( = jh / U~ t' = fh / V",

Figure 6. Phase shift thu between structural Figure 7. Spectral coherence Coh yu
displacement Y and fluctuating streamwise between structural displacement Y and
flow velocity u with and without control: (a) fluctuating streamwise flow velocity u
unperturbed; (b) open-loop control, with and without control.
I; = 0.1; (c) PID-Y; (d) PID-u; (e) P/D-Yu.

perturbation (Fig. 6a), tPYu is zero near I: = 0.13, consistent with the
synchronizing vortex shedding and structural vibration. The plateau about I:
indicates the synchronizing Y and u signals over a range of frequencies. Once
perturbed using the open-loop control (I; = 0.1), tPYu was changed from 0 to 7r in
a narrow frequency range about I: (Fig. 6b). This implies a change in the nature
of the fluid-structure interaction, the synchronizing v and Y turning into anti-
phased interactions against each other. As a result, the spectral coherence between
Y and u, defined by Cohyu = (Co~u + Q~u)/ EyEu' at I:
reduced from 0.65 to 0.15
(Fig. 7b). This quantity provides a measure of the degree of correlation between the
Fourier components of Yand u. The drastic reduction in Coh yu means a decoupled
correlation between vortex shedding and structural vibration. When closed-loop
276

controls were deployed, tPYu also shifted from 0 to 1C, as evident in Figs. 7c-e. It is
noteworthy that the frequency range over which tPYu = 1C exceeds significantly
that in the open-loop control. This frequency range is largest for PID-Yu, covering
the 0.11 to 0.27 (Fig. 6e), and is essentially the same as the synchronization range
(0.11 to 0.26) between vortex shedding and induced vibration for bluff bodies with
fixed separation points (Gowda 1975). The observation suggests that PID-Yu has
successfully changed the relationship between v and Y from in-phase to anti-phase
over the entire synchronization range. Meanwhile, the peak at I:in Coh yu (Fig.
7c-e) recedes, compared with the unperturbed flow, and in effect vanishes when
PID- Yu is deployed.

6. CONCLUSIONS
The presently developed flow control effectively turns the in-phased
vortex shedding and structural vibration into the anti-phased, implying a
change in the nature of fluid-structure interaction. As a result, both vortex
strength and structural vibration are remarkably impaired, and their
correlation appears diminishing. The drag coefficient is also greatly
reduced.
2 The PID-u control strategy outperforms the PID- Y strategy. While the
latter targets the control of passive structural vibration, the former directly
tackles the fluctuating flow, which is the origin of the fluid excitation
forces and vortex-induced vibration, thus being more effective. Among the
three strategies, the performance of PID- Yu is the best; the visualized
wake appears radish-like, suggesting an almost complete destruction of the
Karman vortex street. Such a performance is attributed to its control signal,
which is the combination of flow excitation and structural vibration, thus
reflecting the non-linear interactions between fluid and structure.
3 The closed-loop control has many advantages over an open-loop system.
While the open-loop control depends on the perturbation frequency I;,
the closed-loop control can always suppress vortex and its induced
vibration. Furthermore, with the deployment of a closed-loop control, the
required perturbation amplitude or voltage can be greatly reduced,
pointing to the possibility of developing a more compact and self-
contained control system.

7. AFFILIATIONS
Department 01 Mechanical Engineering, The Hong Kong Polytechnic University,
Hung Hom, Kowloon, Hong Kong
 277

8. REFERENCES
Antonia, R. A., Rajagopalan, S., 1990. AIAA Journal 28, 1833-1835 .
Baz, A., Ro, J., 1991. Journal of Sound and Vibration 146,33-45.
Beannan, P. W., 1984. Annual Review of Fluid Mechanics 16, 195-222.
Berger, E., 1967. Physics of Fluids 10, 191-193.
Brian, C, Donald, C, 1983. Journal of Fluid Mechanics 136,321-374.
Cheng, L., Zhou, Y., Zhang, M. M., 2003. Journal of Fluids and Structures 17(7) (in press).
David R., Williams, Hussein, Mansy, Cliff, Amato, 1992. Journal of Fluid Mechanics 234, 71-96.
Ffowcs Williams, J.E., Zhao, B. C , 1989. Journal of Fluids and Structures 3, 115-122.
Filler, J. R., Marston, P. L., Mih, W. C , 1991. Journal of Fluid Mechanics 231, 481-499.
Gowda, B. H. L., 1975. Deutsche Lufl- und Raumfahrt Forschungsbericht, No. 75-01.
Hsiao, F. B., Shyu, J. Y., 1991. Journal of Fluids and Structures 5, 427-442.
Huang, X. Y., 1996. Experiments in Fluids 20, 218-224.
Zdravkovich, M. M., 1981. Journal of Wind Engineering and Industrial Aerodynamics 7,145-189.
Zhang M. M., Cheng L., Zhou Y., Li P. 2003 3rd Int. Sym. on TSFP, 25-27 June 2003, Sendai, Japan.
Zhou, Y., Zhang, H. J., Liu, M. W., 2002. Journal of Fluid Mechanics 458, 303-332.
 VORTEX-INDUCED VIBRATION STRUCTURAL
RESPONSE UNDER PARAMETRIC EXCITATION

K. Y. BILLAH AND OWAIS AHMAD

Department of Civil, Environmental and Ocean Engineering
Stevens Institute Of Technology, Castle Point on Hudson; Hoboken
NJ-07030, U.S.A
E-mail: kbillah@Stevens-tech.edu E-mail: oahmad@Stevens-tech.edu

Abstract: A model for calculating response associated with vortex-induced vibrations of a cylinder is
proposed. The model considers the periodic wake as a nonlinear oscillator interacting nonlinearly with the
body oscillator. For studying structural response, parametric excitation is taken as the driving mechanism.
The nonlinear stiffness considered in the model arises out of the vortex-structure interaction and the nonlinear
fluid damping is due to higher-order fluctuations. This model represents the mechanism of vortex-induced
vibration that includes lock-in motion over a bandwidth, zones of instability for various frequency ratios,
phase jump and hysteresis in structural response. Mathematical analysis of the model shows that there are two
distinct levels of response. A significant phase jump during the mode transition is also observed. The study
for structural response shows a hysteretic behavior at both ends of the lock-in zone due to parametric
excitation and high amplitude nonlinear damping resembling experimentally observed free vortex-induced
vibrations of cylinders. This study suggests that the structural response due to vortex-induced vibration is
nonlinear and parametric.

1. INTRODUCTION

As the mechanism of VIV is not understood very well, continuous efforts are being
made by researchers to improve the understanding of the relation between the unsteady
forces, near-wake vorticity field and the motion of a cylinder to develop a model to
describe a mechanism involving vortex-induced vibration. Most of the works done in
this field were aimed at matching the structural response without understanding the
complex nature of the problem of VIVo Experimental studies made so far shows that
the [1,2]:
Response in water is different from that in wind.
Forced VIV experiments are different from that of free vibration.
Responses have jump and hysteresis.

279
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 279-288.
2003 Kluwer Academic Publishers.
280

Variation of mass ratio affects the response even if the mass-damping ratio is
kept constant.
At low mass ratio, the synchronization range is larger.
For very light structures resonant vibration has no upper limit in flow speed.

VIV occurs due to non-linear interaction between the body and the periodic wake. It is
a near wake phenomenon, naturally existing dynamic instability problem of the body
[3]. In this an energy transfer process is initiated in the presence of a flexible body and
the natural wake interacts with the motion-induced wake. In VIV the fluctuating lift
force is due to vortex shedding and the structural frequency is close to the natural
frequency of the system. This study focuses primarily on two efforts, first is an
attempt at a comprehensive understanding of the problem of VIV and second is the
formulation of an improved mathematical model for realistic structural response.

2. APPROACH

Models that capture the essential features of the problem are useful in providing insight
into the fluid-structure interaction, the influence of the parameters, and as a convenient
mechanism for generating appropriate loads in structural analysis and design.
For explaining the phenomenon of VIV various models have been suggested,
considering the system as single degree of freedom or coupled body-wake system. Here
we have considered a single degree of freedom non-linear model for getting the realistic
structural response. Considering the nonlinear stiffness arising out of the vortex-
structure interaction, non-linear damping resulting from high amplitude oscillations and
vortex shedding is assumed as periodic. This study shows that the system is: bistable,
hysteresis & phase jump is observed.

3. MECHANISM OF BODY VIBRATION

The body vibrates as a result of parametric instability, defined as the process in which
exponential gain is obtained by adding energy to a system via a time-varying system.
Thus for the present model:
Parametric instability, inherently nonlinear phenomenon, plays the key role in
transforming energy from the periodic wake to the body that causes it to vibrate. For a
flexible body this mechanism instigates the body into an oscillator.
 281

4. COUPLED EQUATIONS OF MOTION -- ASSUMPTION OF

PARAMETRIC EXCITATION

For Modeling VIV, We have assumed that the wake acts as self-excited oscillator for a
stationary cylinder and is modeled as Vander Pol oscillator; non-linear coupling terms
are then intervened to take into consideration the non-linear interaction between the
body and the periodic wake.
We have used numerical and analytical approaches to solve the coupled nonlinear
differential equations resulting in an algebraic equation for the numerical solution of the
amplitude versus frequency ratio. Then we have discussed our results in light of
available experimental data.
The equations of motion for the coupled body-wake system are:
X + 2 ~(O n X+ (0 n 2 X + 2 a xy + 4 fJx 3 Y = 0
Y+ f(y,y) + (2(Os)2 y + ax 2 + fJX4 = 0
[l(a, b)]
where, 'x' is the body and 'y' the wake d.o.f.
Use is made of a 4th order Runge-Kutta algorithm to integrate the coupled equations
numerically and the results indicate: Transformation of energy, Phase jump, Negative
damping, Formation length reduction. These results are in agreement with experimental
observations.
The auto-parametric coupling between two degrees of freedom 'x' and 'y' actually
involves nonlinear coupling between the coordinates. When such a coupled system is
decoupled, assuming a d.o.f to be periodic, the equation of motion for the second d.o.f.
becomes one with periodic coefficients. This assumption is often used in nonlinear
mode-coupling problems for introducing simplification; it generates the condition for
parametric excitation effect for the other d.o.f.

5. SINGLE-DEGREE-OF-FREEDOM MODEL EQUATION FOR BODY

VIBRATION

An equation for structural motion can be derived from decoupling Eqs.l (a, b) on the
assumption that vortex shedding is periodic; this equation is then used to study the
relation of amplitude and frequency ratio.

x+ 2<7 + (1 + &)x + f(x, x) + (a + bx 2 )xCos(2nt) = 0

[2]
Where,
282

f (x, i) = (ax + fJx 2 + yx 4 ) i

x = Displacement of the structure

e = Frequency shift variable
a, b = Controls the force magnitude
~ = Mechanical damping
a = Mechanical damping parameter
~ = Nonlinear damping variable
y = Nonlinear stiffness "Stiffness saturation" variable
n = Frequency ratio
Due to coupling with periodic wake three different effects are introduced to the s.d.o.f.
linear body oscillator. These are: Linear frequency shift, Fundamental instability
mechanism and Instability saturation. If the interaction is nonlinear between the vortex
street and the flexible structure then for the freely-vibrating cylinder the lock-in region
should be function of not only the frequency ratio but also of mass, stiffness and
damping.

6. MATHEMATICAL ANALYSIS OF THE MODEL SHOWS THAT FOR

VIV

Energy transfer mechanism that initiates large amplitude vibration is induced

by a parametric instability condition and energy transfers from the vortices to
the structure, not directly from the uniform flow.
Bistable Structural Response.

7. APPROXIMATE SOLUTION OF THE S.D.O.F. MODEL

Parametric excitation is amenable to solution by the known properties of the Mathieu

equation if only linear excitation is considered. But when nonlinear terms are included
it is necessary to use the theory of approximation. Using Method of Averaging:
For the resonant case let the solution be

x(t) = A(t)Cos If/ (t) = A(t)Cos [Ot + B (t)]

where,
O=l-L\
 283

L1: small detuning from unity

Amplitude A (t) and phase 8 (t) are obtained as

( ~)
&
= (~+
4D
~JSin
8D
2B - (~+
2
~+
8 ~J
16

( -d () ) = (a
- bA -
+- 2 ) 3 dA-
Cos 2() - (0 - 1 - e)+ - 2

dt 40 40 80

8. STABILITY OF THE STEADY-STATE SOLUTION

When a non-linear system is excited parametrically the stability of the equilibrium

position may be decided on the basis of linearized approximation of the equation of
motion.
FOR ZERO MECHANICAL DAMPING: Shaded region is unstable; any deviation
from the stable, however small is sufficient to cause a rapidly increasing displacement.
Zone of instability is for 0 around unity is given by

(1+8- :)<01+8+ :)

L+-____
FOR NON-ZERO MECHANICAL DAMPING: Zone of instability IS reduced

I F_o_rt: = 0

1 Q--

proportionally (non-linearly) to mechanical damping but is contained within the range

of instability region given as above. Thus the body which was unstable for a=O at a
location on the frequency scale may become stable for 0>0. Corresponding zone of
instability is given by

a -a- } <0<1+8+
1+8- { -
2 2

16 4
 L
284

For 0;=0 For 0;= 0, U. 0;= 0

I +_
1 0--
lL+~ 1 0--

Due to mechanical damping features of unlimited growth remains on a reduced zone.

When damping is present there is a destabilizing effect above a certain excitation level.
The stability studies of rest position provided the following characteristics of the single
d.o.f. model: Bandwidth zone of instability around fs/fn=l, the zone of instability is
asymmetric due to the presence of the frequency shift, the width of the zone of
instability is a function (nonlinear) of mechanical damping, the width of this zone is
also a function of the coupling coefficient.

9. STRUCTURAL RESPONSE

VIV structural response under parametric excitation shows bistability and hysteresis [4].
Structural response is given by

x+ 2$ + (1 + &)x + f(x, X) + (a + bx 2 )xCos(2nt) = 0

10. NONLINEAR STIFFNESS ARISING OUT OF VORTEX-STRUCTURE

INTERACTION

Once higher amplitudes are reached, the nonlinear stiffness takes over, reducing the
response amplitudes. The response here is dictated by two mechanisms: parametric
excitations and nonlinear stiffness. The interaction produces the hysteresis effect on the
upper bound of the lock-in condition. This leads to an important conclusion that by
continuously pumping energy into the system is continually affecting the stiffness of
the structure.

x + 2y X + w: (1 + a Cos 2w,!)x + A,X' + A,X' = 0

 285

On Using Method of Averaging

dA
dt
= -r[l- ~ aT
Sin2]

d =-&+~au COS2"'+~AA2 +~A A4

dt 4 n If' 8 1 16 2

The steady state solutions is given by

Cos2st = ~~ - Sin 2 (2st)}

Existence ofbistability; corresponding steady state is given by

5 4 3
8 n
2 1
-A A -U A -lV
16 2 4 n [a- [- r J2] -&=0
un
286

BISTABLE SYSTEM

1.4
i
G>
1.2
1
" 0.8
::J
~
Q.0.6
E
0.4 ST ABr.;.--
0.2 STABLE
NSTABLE .~.
.-- STABLE
0 <
-E < 0.9
0 0.3 0.6 1.5
fslfn IJIo

11. NONLINEAR FLUID DAMPING AND PARAMETRIC EXITATION

When the damping parameter was decreased the structure keeps oscillating between the
peaks and does not damp out to the rest position, even though the damping parameters
were set to zero, the structure did not show signs of instability. This behavior is due to
the nonlinear stiffuess of the structure that makes the structure act as a self-limited
oscillator.
For flow-induced vibration problems higher harmonic, higher-order fluctuations arising
out of the interactions of various frequency components can be significant source of
high amplitude excitation in addition to being a source of dissipation. The form of
(x , X ) for excitation and damping at higher amplitude will be taken as
f (X , X) = (ax + fix 2 + rx 4 ) X
Where y>O and 13 can be positive or negative. When 13 is negative the term (-p X2 X)
causes a negative damping effect at higher amplitude but its effects are curbed at still
higher amplitude by the fourth order term.
On using the nonlinear fluid damping and parametric excitation for the steady state
response, Method of Averaging gives
 287

Jump Phenomenon

3
2.5
III 2
'C
:: 1.5
::l

Q. 1
E
< 0.5
0
----.
STABLE

-0.5 1.3
fs/fn

~=0.1, y=0.005, a= 0.3, Q=f.lfo =0.67 - 1.19, a=1.1, E=0.05

12. CONCLUSIONS

Mathematical analysis shows that energy transfer that initiates large amplitude VIV is
induced by a parametric instability condition
The origin of the hysteresis is attributed to the basic bistable nature of nonlinear
interaction
Responses are complex since parametric excitation can be modified by the presence of
nonlinear stiffness and/or nonlinear damping
Detailed investigation of the model shows that the final part of the two branches is
unstable, and this instability leads to transition.
288

13. REFERENCES

I.Khalak, A., & Williamson, C. H. K., "Fluid forces and dynamics of a hydroelastic structure with very low
mass and damping," J. of Fluids & Structures, 11, pp. 973-982, 1997.
2. Ocean Engineering & Marine Systems, "FYY99 Annual Report: Marine Platform Systems,"
www.Onr.navy.mil.
3. Billah, K. Y. R., "A study of vortex-induced vibration", Ph. D Dissertation, Princeton University, 1989.
4. Billah, K. Y. R., "Nonlinear Interaction of Absolute Instability and Structural Motion in Vortex-induced
Vibration", BBVIV3 Conference, Port. Douglas, Australia, 2002.
5.Williamson, C. H. K., & Roshko, A., "Vortex formation in the wake of an oscillating cylinder, " J. ofFluids
& Structure, Vol. 2, pp. 355-381, 1988.
6. Olinger, D. l., & Sreenivasan, K. R., "Nonlinear dynamics of the wake of an oscillating cylinder, " Physical
Review Letters, 60, pp. 797-800, 1988.
7. Monkewitz, P. A., "The absolute and convective nature of instability in two-dimensional wakes at low
Reynolds numbers, " Physics of Fluids, Vol. 31(5), pp. 999-1006, 1988.
8. Bishop, R. E. D., & Hassan, A. Y., "The lift and drag forces on a circular cylinder oscillating in a flowing
fluid," Proc. Royal Society, Vol. 277A, pp. 51-75, 1964.
9. Constantionos, E., Kamiadakis, G. M., "Dynamics and flow in the turbulent wake of rigid and flexible
cylinders subject to vortex-induced vibrations," l. Fluid Mechanics, Vol. 400, pp. 91-124, 1999.
10. Zdravkovich, M. M., "Modification of vortex shedding in the synchronization range," J. Fluid
Engineering, Vol. 104, pp. 344-349,1991.
11. Skop, R. A., & Balsubramanium, S., "A new twist on an old model for vortex-excited vibrations," J. of
Fluids & Structures, 11, pp. 395-412, 1997.
12. Gammaitoni, L., Hanggi, P., lung, P., Marchesesoni, F., " Stochastic resonance," Reviews of Modern
Physics, Vol. 70, 223-287, 1998.
 FREQUENCY RESPONSE AND THE
EXISTENCE OF A CRITICAL MASS FOR
AN ELASTICALLY-MOUNTED CYLINDER

R. GOVARDHAM
Mechanical Engineering, Indian Institute of Science
Bangalore - 560012, INDIA
raghu@meeheng.iise.ernet.in

C.H.K. WILLIAMSON
Mechanical & Aerospace Engineering, Cornell University
Ithaca, NY 14850, USA
ew26@eornell.edu

Abstract: In this paper, we study the transverse vortex-induced vibrations of a

cylinder at low mass-damping values. The response in this case consist-
s of three distinct branches; namely the initial, upper and lower. For
an elastically-mounted cylinder, the oscillation frequency can be shown
to be primarily dependent on the mass ratio [m*=mass/displaced fluid
mass]. For large mass ratios, m* = 0(100), the vibration frequency for
synchronization lies close to the natural frequency (f* = f / IN ,..., 1.0),
but as mass is reduced to m* = 0(1), r
can reach remarkably large
values. We deduce an expression for the frequency of the lower-branch
vibration, as follows: !tower = J
m':'~t.~4' which agrees very well with
a wide set of experimental data. This frequency equation indicates the
existence of a critical mass ratio, where the frequency r
becomes large:
m~rit = 0.54. When m* < m~rit, it can be shown that the lower branch
can never be reached and ceases to exist. In this case, the upper branch
regime of synchronisation is predicted to continue to infinite normalized
flow speed, U*, where U* = U/ IN D is the conventionally used nor-
malized flow speed. Experiments with a cylinder having no structural
restoring force i.e. IN = 0, corresponding to U* = 00, show that this
is indeed the case. A critical mass, m~rit = 0.542, is obtained from
these no spring experiments, which is in remarkable agreement with the
earlier predictions from the elastically-mounted cylinder experiments.

289
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 289-300.
2003 Kluwer Academic Publishers.
290
Keywords: frequency response, cylinder, vortex induced vibrations, critical mass

1. INTRODUCTION
Vortex-induced vibrations of cylinders has been studied extensively,
as may be seen from the reviews of Sarpkaya (1978), Bearman (1984)
and Parkinson (1989). In the present work, we study the effect of mass
ratio (m* = mass/displaced mass) on the response of such an oscillating
cylinder.
The response of a elastically-mounted cylinder free to vibrate trans-
verse to the flow has been studied extensively since the early work of Feng
(1968). For low values of mass-damping (m*(), recent studies by Khalak
& Williamson (1999) and Govardhan & Williamson (2000) indicate that
there exist three response branches; namely the initial, upper and lower,
as shown in figure 1. [The mass ratio, m*=(oscillating mass)/(displaced
fluid mass); and the damping ratio, ( = structural damping/critical
damping]. The relationship between the response, forces and wake dy-
namics for such an oscillating cylinder is studied in detail in Govardhan
& Williamson (2000). Vorticity contours corresponding to the different
1.0

UPPER
0
0.8 8>0

A*
~WWER
,..
0.4

INITIAL ... :
0.2 .t

0.0
0.0
..' l2.0
U*

Figure 1. Low mass-damping case exhibiting 3-modes. Amplitude response as a

function of flow speed, showing three response branches; namely the Initial, Upper
and Lower. [m* = 8.63, em*
+ OAK =0.0185]. ~, Initial; 0, Upper; ., Lower; 0,
desynchronized regime.

response branches shown in figure 2, indicate that the Initial branch is

associated with the 2S wake mode, while both the Upper and Lower
branches correspond to the 2P-mode; '2S' indicating 2 Single vortices
formed per cycle, and '2P' meaning 2 Pairs of vortices formed per cy-
cle, as defined by Williamson & Roshko (1988) based on their forced
oscillation experiments. In the case of forced vibration, the 2S and 2P
modes have also been shown using PIV by Carberry, Sheridan & Rock-
well (2001) for the forced transverse vibration of a circular cylinder,
and these modes are also observed from forced oscillations of a tapered
cylinder by Techet et al. (1998).
 291

LOWER

Figure 2. Vortex formation modes in the different response branches. (a) Initial
branch - 28; (b) Upper branch - 2P; (c) Lower branch - 2P. In the Upper branch case,
there are two vortex pairs formed per cycle, although the second vortex of each pair
is much weaker than the first vortex, and decays rapidly. Vorticity contours levels
shown are separated by (wD/U) = 0.4. 28, 2P modes are as defined by Williamson
& Roshko (1988).

In the present paper, we shall study the effect of mass ratio on the
response of a cylinder free to vibrate transverse to the flow at low values
of mass-damping.

2. EXPERIMENTAL DETAILS
The present experiments were conducted using a hydro elastic facility,
which is described in Khalak & Williamson (1999), in conjunction with
the Cornell-ONR Water Channel. The hydro-elastic facility comprises a
carriage mounted on air-bearings situated above the channel test-section,
which allows a vertical cylinder in the fluid to move transverse to the
free-stream. As there is no metal-metal contact, the damping associated
with the air-bearing system is extremely small, and in all cases reported
here the mass-damping parameter is very small [(m* + CA)( < 0.05)].
The turbulence level in the test section of the Water Channel was less
than 0.9%, in the 0.381 m x 0.508 m cross section, over the range of
free stream velocities, U (0.04 - 0.32 ms- 1 ), used in this study. Test
cylinders of diameters 0.0381m and 0.0794m were used for the elastically-
mounted case and the zero structural restoring force case, respectively.
The corresponding length-diameter ratios were 10 and 6, respectively.
For the purpose of employing DPIV, the flow was seeded with 14-
micron silver-coated glass spheres, which were illuminated by a sheet of
laser light from a 5W Argon ion laser. Pairs of particle images were
captured using a high-resolution CCD Kodak Megaplus (1008 x 1018
pixels) camera, and analyzed using cross-correlation of sub-images, our
implementation of which is described in more detail in Govardhan &
Williamson (2000).
292

3. FREQUENCY RESPONSE AND THE

EXISTENCE OF A CRITICAL MASS
The oscillation frequency (f) of an elastically-mounted cylinder de-
pends primarily on the mass ratio (m*), as discussed for example in
Bearman (1984). This may be seen clearly from the equation for the
cylinder frequency response (f* = i / iN) shown below, which is formu-
lated here along the lines of Khalak and Williamson (1999), as follows:

1*= /m*+GA (1)

Vm* + GEA
where GEA is the effective added mass due to wake vortex dynamics,
and GA is the potential added mass (GA=1.0 for a circular cylinder).
At high mass ratios, the large m* values overwhelm the effect of the
(a) m* ~ 8.6 (b) m* ~ 1.2

1.2 ' .2
Upper

Upper

08 i, 0.8
A*
~~wer .......... Lower

.... .
~
0.4 0.4 Initial.

Initial!
f~

..........-
f*
.~ -.-- -- -

0~-7~~~12~-J~6~~20--~24 O~-7~~~12~-J~6~~,O--~24
U* U*

Figure 3. Amplitude - frequency plots for (a) m* = 8.6, and (b) m* = 1.2, plotted
to the same scale. The mass-damping parameter is small in both cases; (a) (m* +CA)
=0.02, (b) (m* + CA) =0.01. iVD is the stationary body vortex shedding frequency
and iN is the structural natural frequency . .6., initial; 0, upper; ., lower; \l, desyn-
chronized.

effective added mass (GEA), thus yielding f* = fliN""" 1.0. Hence,

at high mass ratios, as in Feng (1968), the response frequency (f) is
close to the natural frequency (f N) in the synchronization regime. On
the other hand, at low m* the effective added mass (GEA) influences f*
significantly, and results in a marked deviation of f* from unity. This has
been seen for example in more recent experiments of Moe & Wu (1990),
Khalak & Williamson (1999) and Govardhan & Williamson (2000).
 293

"', Noobserved
/\ synchronized
/
\ pattern

P+S
1\
1.5

2P+2S

<f,..' 1)
Figure 4. Amplitude response for two different mass ratios (m*) plotted versus the
frequency ratio (tva/ f), where Iva is the stationary body vortex shedding frequency
and I is the body oscillation frequency. 0, m* = 1.19 and (m* + CA) = 0.0110; ~,
m* = 8.63 and (m* +CA) = 0.0185. Solid symbols indicate the Lower branch regimes .
... Williamson & Roshko (1988) map of wake modes.

The effect of mass ratio on the amplitude response is illustrated by

the two example cases shown in figure 3. As observed by Griffin &
Ramberg (1982), the amplitude response for the lower mass ratio case
indicates a widening of the region of vigorous synchronized oscillations
(in U* = UI JND) compared to the larger m* case. However, as shown
by Khalak & Williamson (1999), when the two amplitude responses are
plotted versus the parameter (U* 11*)8, which is equivalent to (Ivai f),
then the data sets collapse very well especially in the lower branch, as
shown in figure 4 [Jva= non-oscillating body vortex shedding frequency;
J = body oscillation frequency; 8 = Strouhal number].
A striking aspect of the frequency response is the almost constant
value for the lower branch frequency over the complete response branch.
This constant level of frequency is observed in both cases shown in fig-
ure 3 and at every mass ratio investigated (m*=0.8 to 20), and seems to
be a general characteristic of this lower branch. This observation is also
supported by the frequency data of Hover et al. (1998) and Khalak &
Williamson (1999), both at similar low mass-damping values.
A large set of data for the Lower branch frequency (I1~weT) plotted
versus m *, is shown in figure 5. This data is from our own experi-
ments, and from Hover et al. (1998), Khalak & Williamson (1999), and
Anand (1985). The data collapse very well onto a single curve. Since
the response in the lower branch is remarkably sinusoidal and periodic,
equation (1) is a very good representation of the frequency response.
The effective added mass (CEA) in equation (1) is itself a function of
294

2.5 I

f*LOWER
1.5

I +- - - - - _.. _. _.. _.. _.. _ .._.. - - - ._.. _ .._.. _. . - - --.

I

0.5 0 \
10 15
m*
m* . ~ 0.04
an

Figure 5. Variation of the lower-branch frequency (Hower) as a function of the

mass ratio m*. The equation for Hower fits the data remarkably well, and indicates
a dramatic increase in Hower as we approach the critical mass ratio, m~rit = 0.54 .
, Present data; .0., Khalak & Williamson (1999); 0, Hover et al. (1998); 0, Anand
(1985) .

{ (Jva / j), A *}, and will have a unique value at each point along the low-
er branch, when plotted in the plane {(Jva/j),A*}, as in figure 4. Since
at low values of mass-damping, (m* +CA)( < 0.05, all lower-branch data
sets lie nearly along the same line in this plane, almost independently of
mass ratio, m*, the value of CEA along this line will be independent of
m *. Further, in the lower branch, the frequency response, 1*, is nearly a
constant for any given mass ratio, and equation (1) indicates that CEA
is a constant along the entire collapsed lower branch data in figure 4.
This implies that there is a single value of CEA that represents all the
lower branch frequency data, which may be found as the best fit of CEA
in equation (1) that represents the experimental data of figure 5. From
this analysis, we find CEA = -0.54 0.02, and we thereby deduce the
following Lower-branch frequency equation as:

m* + 1.0
fz~wer = m* - 0.54
(2)

This curve is drawn through the data in figure 5, and it represents

the data very well.
An important consequence of equation (2) is that the frequency be-
comes infinite as the mass ratio reduces to the limiting value of 0.54.
Therefore we conclude that there exists a critical mass ratio:

Critical mass ratio, m~rit = 0.54 (3)

 295

The existence of a critical mass is, for us, a surprising and interesting
result.
As the mass ratio decreases, the normalized velocity U* for the start
of the lower branch varies, maintaining the condition seen in figure 4:

(Jval !)start = 1.15 (4)

Since the frequency ratio (Jval j) may also be written in terms of the
normalized flow speed, U*, as (Jval j) = (U* 1]*)8, we can rewrite the
above condition in terms of U* for the start of the lower branch (U;tart)
by using equation (2) and assuming 8 = 0.2 as:

*
Ustart = 5.75 V m* + 1
m* _ 0.54 (5)

This shows that the velocity for the start of the Lower branch:

U;tart -+ 00 as m* -+ m~rit

4. RESPONSE AT U* = 00: CYLINDER WITH

NO STRUCTURAL RESTORING FORCE
The prediction of large-amplitude response at infinite U* suggested
by the elastically-mounted cylinder experiments in the previous section,
may be directly verified by experiments with a cylinder having zero
structural restoring force i.e. a cylinder with no springs attached. In this
case, the natural frequency, ! N, is zero, and therefore the conventional
normalized velocity, U* = UI!ND = 00.

(a) .-1.67 (b) -'.45

yID o~----------------~ o

.IO!-------:I";;-O----~20 .IO!--------:'I";;-O-----~

tiT,. tiT..

Figure 6. Time trace of response for a cylinder with no springs for (a) m* = 1.67,
and (b) m* = 0.45. In (a), no significant oscillations are observed, while in (b)
large-amplitude oscillations are seen. Re ~ 5100 in both cases.

The response time trace for a very low mass ratio, m* = 1.67, cylinder
is shown in figure 6(a) and does indeed seem intriguing. Despite the fact
that one could move the cylinder, mounted on air-bearings, simply by
296

brushing it with a feather, we found that the body remained stationary.

This was surprising to us given that the cylinder is completely free to
move transverse to the flow, and is subjected to large transverse (lift)
forces, as may be inferred from the strong periodic shedding of vortices
seen in the experiments. However, simply by reducing the mass ratio to
m* = 0.45, we discovered large-amplitude oscillations to occur. These
vigorous oscillations (in figure 6b) were at amplitudes of the order of
A* = AID ~ 0.80.

m* =0.45
A*

m* =1.67

Figure 7. Response data for experiments with no springs, over a range of flow speed-
s, on the frequency-amplitude plane { Uva/ f) , A * }, indicating that the oscillations
are essentially independent of flow speed, in this plane. Iva is the vortex shedding
frequency in the absence of body oscillations, and I is the actual cylinder oscillation
frequency. 0, m*=1.67; 0, m*=0.45; .. " Williamson & Roshko (1988) map of wake
modes.

It should be noted that the oscillation amplitudes and also the normal-
ized oscillation frequency (f I ivo) remained nearly independent of flow
speed over a wide range of flow velocities (or Reynolds numbers) investi-
gated, at a fixed mass ratio. This leads to a collapse of the response data
(over all the flow speeds) to a single point on the {(fvol 1), A*} plane,
as shown in figure 7. We refer to this point as the 'Operating Point' in
accordance with Govardhan & Williamson (2002). This operating point
can be shown to be primarily a function of the mass ratio, at small val-
ues of mass-damping, as discussed in detail in Govardhan & Williamson
(2002).
The huge change in the dynamics of the cylinder, from essentially
no oscillations to large-amplitude vibrations, triggered by only a small
change in mass ratio from m* = 1.67 to m* = 0.45, suggests that there
does indeed exist a critical mass ratio at which there is a sudden 'catas-
trophic' change in the body dynamics. Keeping this in mind, the mass
ratio was decreased in small steps from the larger m* value, and the re-
 297

suIt is shown in figure 8. It is clear that the cylinder suddenly starts to

undergo large-amplitude vigorous oscillations as the mass ratio crosses
a critical mass ratio:
Critical mass ratio, m~rit = 0.542 0.01 (6)
For m* > m~rit' the cylinder is almost stationary, while for m* <
m~rit' the cylinder exhibits large-amplitude oscillations.

m':,..=O.542

A* ~ I
I
I
0.5 I
I

m*

Figure 8. Existence of a critical mass for a cylinder with no springs. Amplitude

response (A*) as mass ratio (m*) is decreased, showing the sudden appearance of
large-amplitude oscillations for m* less than a critical value (m~rit) of 0.542 . , m* <
m~rit; 0, m* > m~rit.

5. CRITICAL MASS AND INFINITE

REGIME OF SYNCHRONIZATION
The critical mass results from the cylinder with no structural restor-
ing force case, is in remarkable agreement, both conceptually and nu-
merically, with the predictions from the elastically-mounted cylinder ex-
periments of section 3. There are indeed large amplitude oscillations
even at infinite U*, when m * < m~rit = 0.54. This implies that when
m* < m~rit' there will exist an infinitely wide range of flow speeds where
vigorous oscillations occur or in other words the regime of synchroniza-
tion is infinitely wide. This result is distinctly different from the classical
case, where significant oscillations occur only over a narrow range of flow
speeds where the vortex shedding frequency, Iv, is of the order of the
mechanical natural frequency, IN, i.e. Iv rv IN. The infinite regime
of synchronization is illustrated in figure 9 for m* = 0.52, and includes
data both from elastically-mounted cylinder studies and also from cylin-
ders with no springs corresponding to U* = 00 . As may be seen the
amplitude remains large (A* :::::: 0.7) at U* = 00.
298

A*

U*

Figure g. Response for m* < m~rit> showing an infinite regime of 'resonance'. The
response amplitude at U* = 00, which corresponds to experiments without springs,
remains large (A* ~ 0.7). In this case, m* = 0.52. 0, initial branch; ., upper branch;
0, experiments with no springs (U* = 00).

6. CONCLUSIONS
In this paper, we study the transverse vortex-induced vibrations of a
cylinder at low mass-damping values. The response in this case consists
of three distinct branches; namely the initial, upper and lower.
In the elastically-mounted cylinder case, the oscillation frequency can
be shown to be primarily dependent on the mass ratio [m * =mass / displaced
fluid mass]. For large mass ratios, m* = 0(100), one expects from classi-
cal work that the vibration frequency will lie close to the natural frequen-
cy (f* = i / iN '" 1.0). However, for very low mass ratios, m* = 0(1),
1* can reach remarkably large values. We deduce an expression for the
frequency of the lower-branch vibration, as follows:

m* + 1
m* - 0.54
This frequency equation indicates the existence of a critical mass ratio,
where the frequency 1* becomes large:

Critical mass ratio, m~rit = 0.54 (Elastically - mounted cylinder)

Also, we deduce an equation for the start of the lower branch: U:tart =
5.75) m~~t.~4' which indicates that when m* < m~rit' the lower branch
can never be reached and ceases to exist. In this case, the prediction
is that the upper branch regime of synchronization continues to infinite
normalized flow speed.
In the case of a cylinder with no structural restoring force i.e. no
attached springs, the natural frequency, iN, is zero, and therefore the
conventionally defined U* is infinite. Experiments under these condition-
s indicate that there are negligible oscillations as mass ratio is reduced
 299

from large values to m* of the order of unity. However, a further re-

duction in mass exhibits a surprising result; large-amplitude oscillations
suddenly appear for values of mass less than a critical mass ratio of
0.542:

Critical mass ratio, = 0.542 (Cylinder with no springs)

This result for the critical mass from experiments with a cylinder
having no structural restoring force is in remarkable agreement with the
earlier predictions from the elastically-mounted cylinder experiments.

7. ACKNOWLEDGEMENTS
The support from the Ocean Engineering Division of O.N.R., moni-
tored by Dr. Tom Swean, is gratefully acknowledged (O.N.R. Contract
No. N00014-95-1-0332) .

References
Anand N.M. (1985) Free span vibrations of submarine pipelines in steady and wave
flows. Ph.D. thesis Norwegian Institue of Technology, Trondheim, Norway.
Bearman P.W. (1984) Vortex shedding from oscillating bluff bodies. Annual Review
of Fluid Mechanics 16, 195-222.
Carberry J., Sheridan J . & Rockwell D. (2001) Forces and wake modes of an oscillating
cylinder. J. Fluids Struct. 15, 523-532.
Feng C.C. (1968) The measurements of vortex-induced effects in flow past a station-
ary and oscillating circular and D-section cylinders. Master's thesis, University of
British Columbia, Vancouver, B.C., Canada.
Govardhan R. & Williamson C.H.K. (2000) Modes of vortex formation and frequency
response of a freely vibrating cylinder. J. Fluid Mech. 420, 85-130.
Govardhan R. & Williamson C.H.K. (2002) Resonance forever: existence of a critical
mass and an infinite regime of synchronization in vortex-induced vibration. J. Fluid
Mech. 473, 147-166.
Griffin O.M . & Ramberg S.E. (1982) Some recent studies of vortex shedding with
application to marine tubulars and risers. ASME J. of Energy Resources Technology
104, 2-13.
Hover F.S., Techet A.H. & Triantafyllou M.S. (1998) Forces on oscillating uniform
and tapered cylinders in crossflow. J. Fluid Mech. 363, 97-114.
Khalak A. & Williamson C.H.K. (1999) Motions, forces and mode transitions in
vortex-induced vibrations at low mass-damping. J. Fluids Struct. 13, 813-85l.
Moe G. & Wu Z-J. (1990) The lift force on a cylinder vibrating in a current. ASME
J. of Offshore Mech. and Arctic Engng. 112, 297-303.
Parkinson G. (1989) Phenomena and modelling of flow-induced vibrations of bluff
bodies. Progress in Aerospace Sciences 26, 169-224.
Sarpkaya T. (1978) Fluid forces on oscillating cylinders. ASCE J. of Waterway, Port,
Coastal, and Ocean Division 104, 275-290.
Techet A.H., Hover F.S. & Triantafyllou M.S. (1998) Vortical patterns behind a ta-
pered cylinder oscillating transversely to a uniform flow. J. Fluid Mech. 363, 79-96.
300

Williamson C.H.K. & Roshko A. (1988) Vortex formation in the wake of an oscillating
cylinder. J. Fluids Struct. 2, 355-381.
 lWO-DEGREE OFFREEDOMVIV OF A
CIRCULAR CYLINDER PINNED AT ONE END

C.M. LEONG, H. BENAROYA, & T.WEI

Department ofMechanical & Aerospace Engineering
Rutgers, The State University of New Jersey
Piscataway, New Jersey 08854-8058
USA

Abstract: This paper is a report on a study of vortex-induced-vibrations of a flexibly-mounted rigid circular

cylinder immersed in a unifonn free stream. This fluid-structure interaction experiment was carried out in the
Rutgers Free-Surface Water Tunnel facility using a thin wall, 2.54 cm diameter, 109 cm long circular cylinder
with a low mass-ratio. The cylinder was elastically mounted at its bottom end to the text section floor by a
small diameter stainless steel pin. The top end of the cylinder was free to oscillate in both the cross-stream
and streamwise directions. Extensive laser induced fluorescence flow visualization studies were carried out
at the cylinder mid-height to investigate the response of the cylinder as well as the vortex shedding patterns.
Flow visualization experiments were carried out over a Reynolds number range, based on cylinder diameter
and free-stream speed, ranging from 820 to 6800. A range of vibration and vortex shedding modes were
observed from these flow visualization studies. Symmetric vortex shedding was observed at low reduced
velocities which evolved to 'C' shaped and 'figure eight' oscillations with increasing reduced velocity.

1. INTRODUCTION

In this paper, preliminary flow visualization studies of fully-coupled, two degree-of-

freedom vortex-induced-vibrations (VIV) of a circular cylinder, pinned at one end, are
presented. Fully-coupled VIV of cylinders in cross-flow have received considerable
attention over the past decade. The papers contained in this volume represent an
excellent sampling of some of the best research on the topic. The challenge of the
fully-coupled problem is that the interactions between fluid dynamics and structural
motion, which are frequently non-linear, are inherently included in the problem. More
traditional decoupled studies are done either by forcing the structural motion and
examining the vortex dynamics, or mathematically assuming some form of periodic
fluid forcing function(s) and computing the structural response. Indeed, it is likely that
the majority of research into fluid-structure interactions is still done in the traditional,
i.e. decoupled way.

301
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 301-311.
2003 Kluwer Academic Publishers.
302

The classic two-dimensional VIV experiment is commonly referred to as the

elastically mounted cylinder. Under the right conditions, Karman vortices shed from
the cylinder induce fluctuating lift forces which excite transverse oscillations. When
VIV occurs, the frequency of vortex shedding modes, i.e. repeatable vortex patterns, is
approximately equal to the oscillation frequency of the cylinder -- this is defined as
synchronization and the range of flow speeds for which this occurs is called the
synchronization regime. For extensive reviews on this subject, please see Bearman
(1984), Chen (1987), and Williamson (1996).
Recently, Benaroya & Wei (2000) examined the fully-coupled VIV of a circular
cylinder pinned at one end and free to oscillate like an inverted pendulum. The cylinder
was constrained to move only in the cross-stream plane by a cart and track assembly
mounted to the top, free-end of the cylinder. The purpose of their research was to
introduce a new paradigm for reduced-order modeling of VIV in which temporally and
spatially resolved measurements would be used to 'derive' mathematical expressions
for the fluid forcing.
Proof-of-concept experiments and analysis for this modeling approach was
conducted by Dong (2002) and will appear as a journal article in Dong, et al. (2003)
shortly after these proceedings. In that study, a single degree-of-freedom model was
developed in the form of a mechanical energy equation for the cylinder. Fluid 'forcing'
appeared in the form of fluid kinetic energy fluxes, time derivatives and as mechanical
work rate terms. It was shown that complex phenomena, such as quasi-periodic
beating, could be examined and understood using this approach.
A limitation of the work reported in Dong, et al. (2003) is that the mechanical
energy equation used to model structural response does not permit more detailed
examination of higher degree-of-freedom fluid-structure interaction problems. The
mechanical energy equation is a scalar equation and can only be used for single degree-
of-freedom problems. For higher order problems, additional equations would be
necessary.
Work toward generalizing the model using variational methods in currently in
progress. In order to facilitate this effort, experiments have been initiated to understand
and quantify two degree-of-freedom VIV of pinned cylinder problem. This paper
contains flow visualization and preliminary frequency-amplitude response
characteristics for this flow.
Two degree-of-freedom VIV of an elastically mounted cylinder was reported by
Jauvtis & Williamson (2002). While reporting on the existence of different oscillation
regimes, the principal focus of their work was on comparing the transverse motions in
the two degree-of-freedom experiments to the one degree-of-freedom case where the
structure was constrained to move in the transverse direction. In the oral presentation
of their paper, Jauvtis presented amplitude response characteristics along with
schematic drawing of the two-dimensional cylinder trajectories at different reduced
 303

velocities; this does not appear in their written paper. The various types of trajectories
observed included purely streamwise oscillations as well as 'C' shaped and 'figure
eight' patterns.
The focus of this investigation was further characterization of two degree-of-
freedom VIV in the wake of a surface-piercing, rigid circular-cylinder mounted as an
inverted pendulum. Of particular interest was the identification of different cylinder
trajectories as a function of reduced velocity and the vortex patterns associated with
these motions. It is important to note that there are significant differences between the
present study and that of Jauvtis & Williamson (2002). Firstly, in their study, the mass
ratio was significantly higher than in the present work, m* = 6.9 for Jauvtis &
Williamson (2002) versus 0.42 in the present investigation. More importantly, perhaps,
is the fact that for the inverted pendulum, the cylinder can lean in the flow direction at
fairly large angles, up to 10. Complex vortex shedding phenomena including cellular
shedding and axial flows along the Karrruin vortex cores may occur under these
conditions. An examination of three-dimensionality and free-surface effects for cross-
stream oscillations of the inverted pendulum was conducted by Voorhees & Wei
(2002).

2. EXPERIMENTAL APPARATUS

In this study, two degree-of-freedom vortex-induced-vibration of a low mass-damping-

ratio circular cylinder mounted as an inverted pendulum was examined. The cylinder
was 2.54 cm in diameter (D) and was constructed from 0.16 cm thick wall acrylic tube.
It was 128 cm long and immersed in a uniform flow of water, ~107 cm deep. It was
restrained at its bottom end by a stainless steel pin. Unlike Voorhees (2002), the
cylinder was free to oscillate both in the streamwise as well as the cross-stream
directions. A schematic drawing of this model problem is shown in Figure 1.
To permit two-color laser induced fluorescence (LIF) flow visualization and to
minimize refraction effects as laser light passed through the cylinder, a 10.16 cm long x
2.54 cm diameter dye injection module was installed close to the cylinder mid-height.
The module was simply a 7.62 em length of the 0.16 em wall acrylic tube capped at
both ends by 2.54 cm lengths of 2.22 em diameter circular acrylic bar. The volume
between the two end plugs was divided into two equal halves using a 5 0. 8 em length of
acrylic flat stock. Two axial dye injection slots were milled into the tube wall 60 apart.
These would be aligned such that each injection slot would be 30 from the nominal
forward stagnation line of the cylinder. The purpose of the divider, then, was to permit
injection of two different dyes, one from either side of the cylinder. The end plugs were
inserted halfway into the dye injection module so that a 1.27 cm length extended from
304

I" (Oi.rnmr)

t
sO" (Lcngth)

Figure J.' Schematic drawing oflhe

inverted pendldllm experiment. Note thaI.
iI/lieu of a fea/spring. the cylinder was
mounted /0 the water tunnel floor using II
3.2 mm diameter stainless steel pin.

both ends and could be mated with the top and bottom segments of the cylinder. The
upper end plug was drilled with two small diameter holes to which dye injection lines
were attached. The dye injection module was positioned -70 em from the boltom of the
cylinder. Two different color laser fluorescing dyes, rhodamine (orange-red) and
fluorescein (yellow-green) were supplied from constant head reselVoirs through 0.32 em
diameter flexible plastic tubing.
As noted earlier, the cylinder was mounled to the floor of the Rutgers Free Surface
Water Tunnel facility using a 0.32 em diameter stainless steel pin; both ends of the pin
were threaded. The top end of the pin was secured to the bottom of the cylinder using a
second plug, 2.22 em in diameter and 3.81 cm in length. The bottom end of the pin was
screwed to a 1.27 cm thick base plate. The effective length of the pin after installation
 305

was 5.08 cm. In order to minimize end effects, another 2.22 cm diameter plug was
inserted up inside the cylinder with its bottom face ~2.54 cm above the bottom of the
cylinder. A ~2.54 cm clearance between the bottom of the cylinder and the base plate
was left in order to prevent contact as the cylinder oscillated. The mass-ratio, m*, of the
cylinder was 0.42 where m* is defmed as the ratio of the structure's total mass to the
mass of water displaced by the structure. The cylinder's damping ratio, S, was 0.058.
Its natural frequency, fno was 1.14 Hz in water.

3. RESULTS AND DISCUSSION

Two-color LIF flow visualization studies were conducted at the mid-height of the
cylinder over a range of reduced velocities t, 1.1 :::;; U* :::;; 8.3, corresponding to a
Reynolds number range, based on cylinder diameter and free-stream velocity, from 820
to 6050. Two different color dyes were injected from each of the two dye injection
slots machined into the injection module described previously. Fifteen minute long
SVHS video records were made for a number of different flow speeds across the speed
range. In order to identify hysteresis effects, visualization studies were conducted by
both incrementing up and decrementing down across the speed range. Each time the
speed was changed, a minimum of thirty minutes were permitted to elapse before video
recording commenced. This ensured that transients associated with the speed changes
were no longer present.
A set of five still photographs taken from five different LIF video sequences is
shown in Figure 2. Each successive image shows the flow behind the cylinder at
increasingly higher speeds. In each picture, flow is left-to-right with the cylinder
appearing on the left side of the photograph. The illuminating laser sheet was
horizontal and passed through the cylinder mid-height. The video camera was
positioned to view the flow from below the water tunnel test section.
The photograph in Figure 2a shows the classic Kil1l1lin vortex street behind a
stationary cylinder. The reduced velocity in this case, U* = 1.12 (Re = 823), was
sufficiently low that VIV did not occur.
At slightly higher speeds, 1.74 :::;;U* :::;;2.37, the cylinder would oscillate purely in
the stream direction. Figure 2b shows an example of the instantaneous flow behind the
streamwise oscillating cylinder at U* = 1.74 (Re = 1278). In this speed range,
symmetric vortices were shed from the cylinder in a manner similar to an impulsively
started cylinder or a cylinder at low Reynolds numbers. Evidence of the competition
between symmetric vortices may be detected in the dye patterns more than one cylinder

t The reduced velocity, U*, is defined as U* = U / fnD, where U, D and fn are the free-stream speed, cylinder
diameter and cylinder natural frequency, respectively.
306

diameter downstream of the cylinder. While two equal strength, opposite sign vortices
are rolling up immediately behind the cylinder, symmetry is immediately broken in the
next two vortices. The reader is reminded here that the two different dyes fluoresce
with different intensities. Consequently, the symmetry between the upper and lower
vortices still attached to the cylinder in Figure 2b may not be visually as convincing as
watching the video recordings or live experiment.
With increasing flow speed, the competition between vortices shed from opposite
sides of the cylinder becomes stronger, leading to alternate-side shedding and the
introduction of spanwise oscillations along with the streamwise motions. The fIrst mode
of fully two degree-of-freedom cylinder oscillations was a 'C-shape' trajectory with the
open side of the 'C' facing downstream. Pure 'C-shape' oscillations were observed
from U* = 2.78 (Re = 2037) to U* = 4.39 (Re = 3221). Over the range, 2.37 ~U* ~
2.78 (1734 ~Re ~2037), there appeared to be a transitional regime in which the
trajectories would switch back and forth between streamwise and 'C-shape'. An
example of flow associated with the transitional and 'C-shape' regimes are shown in
Figures 2c and 2d, respectively.
The fmal oscillation regime observed in this study was the 'fIgure eight' occurring
for U* > 4.39. Because of the strength of the stainless steel pin coupled with
limitations in the quality of the flow visualization at high tunnel speeds, the upper limit
on this trajectory mode could not be determined. The degradation of visualization
quality can be seen in the photograph for this mode, Figure 2e.
Further details of the VIV characteristics can be seen in the amplitude response
plots shown in Figure 3. Nondimensional streamwise and cross-stream oscillation
amplitudes are plotted in Figures 3a and 3b, respectively, as functions of reduced
velocity. Note that the amplitude data at each value of U* represent the largest
amplitude observed in the corresponding fIfteen minute video record. Thus, if beating
phenomena were to occur, such as that described in Dong, et al. (2003) at a particular
U*, the data points in Figures 3a and 3b would represent the largest amplitude of the
beat cycle.
One of the shortcomings of the amplitude response plots as shown in Figure 3 is
the lack of phase information between the two components of vibration. To help the
reader better imagine the modes of vibration, schematic drawings of the streamwise,
'C-shape' and 'fIgure eight' trajectories are indicated along the top of Figure 3a. In
addition, vertical dotted grey lines have been superimposed on the plots to
approximately identify the U* boundaries between oscillation regimes.
In closing, it is perhaps worth noting that there does not appear to be an upper and
lower branch in the cross-stream amplitude response plot shown in Figure 3b. This is
in contrast to the fmdings reported in Jauvtis & Williamson (2002) that the transverse
motions in the two degree-of-freedom elastically-mounted cylinder were very similar to
that of the single degree-of-freedom case. It is quite likely that the present experiments
 307

were conducted for such a low mass-ratio cylinder that different branches do not exist.
This is discussed in Khalak & Williamson (1999) and is a topic for further
investigation.

(a) (b)

(c) (d)
308

(e)

Figure 2: Single still photographs taken from LIF video sequences offlow past the cylinder mounted as an
inverted pendulum at (a) U* = 1.12, Re = 823; (b) U* = 1.74, Re = 1278; (c) U* = 2.36, Re =
1734; (d) U* = 2.97, Re = 2176; (e) U* = 4.39, Re = 3221. In each photograph, flow is left-to-
right with the cylinder on the left side of the image. Note the significant differences between dye
patterns at each of the different flow speeds.
 309

1.2

0.8
8 o
o

o
o

~ 0.6 o o
o
o
0.4 o

0.2

o
o 5 6 7 8 9 10
U*
1.2

o o

8 0

0.8 o
o
Cl
>= 0.6

0.4

0.2

o
o 2 3 4 5 6 7 8 9 10
U*

Figure 3: Amplitude response plots for both (a) streamwise and (b) cross-stream directions. Maximum
observed amplitudes are plotted versus dimensionless reduced velocity. Circles indicate data
taken while incrementing up through the speed range. Squares denote data taken while
decrementing down through the speed range. Schematic drawings of the different cylinder
trajectories have been included at the top of the figure. In addition, grey dotted lines demarcate
the boundaries where trajectories were observed to change.
310

4. CONCLUSIONS
Two-color LIF studies were conducted to characterize the two degree-of-freedom VIV
of a circular cylinder mounted as an inverted pendulum. The pendulum had a low
mass-ratio of m* = 0.42. Three distinct oscillation modes were detected, pure
streamwise, 'C-shape' and 'figure eight'. In addition, a transition mode between the
streamwise and 'C-shape' oscillations were observed. The existence of combined 'C-
shape' and 'figure eight' (effectively a Lissajous pattern) needs to be examined more
closely. In addition, a pure cross-stream oscillation was observed at higher reduced
velocities in a preliminary experiment, but was not reproduced in these later studies.
This is also a point of further research. Finally, quantification of the flowfield using
DPIV, similar to that reported in Dong, et al (2003) will be the focus of continuing
experimentation.

5. ACKNOWLEDGEMENTS
Support from the Office of Naval Research through Dr. Thomas Swean is gratefully
acknowledged.

6. REFERENCES
Bearman, P.W. 1984 Vortex shedding from oscillating bluff bodies. Ann. Rev. Fluid Mech. 16,195-222.
Benaroya, H. & Wei, T. 2000 Hamilton's principle for external viscous fluid-structure interaction. J Sound
& Vibration, 238, 113-145.

Chen, S.S. 1987 Flow-induced vibration Q[ cylindrical structures. Hemisphere Publishing, New York.
Dong, P. 2002 Phase averaged transport in the vortex-induced oscillation of a cylinder: Experiment and
modeling. Ph.D. Dissertation; Dept. ofMech. & Aero. Eng'g.; Rutgers, The State University.
Dong, P., Benaroya, H. & Wei, T. 2003 Integrating experiments into an energy-based reduced-order model
for vortex-induced-vibrations of a cylinder mounted as an inverted pendulum. J Sound & Vibration (to
appear).
Jauvtis, N. & Williamson, C.H.K. 2002 Vortex-induced vibration ofa cylinder in two degrees of freedom.
proceedings of the 3'd Bluff Body Vortex Induced Vibration Conference, Queensland, Australia, Dec.
2002.
Khalak, A. & Williamson, C.H.K. 1999 Motions, forces and mode transitions in vortex-induced vibrations at
low mass-damping. J. Fluids and Structures, 13, 813-851.
Voorhees, A. 2002 Three-dimensionality in the wake of a surface piercing cylinder oscillating as an inverted
pendulum. M.S. Thesis; Dept. of Mech. & Aero. Eng'g.; Rutgers, The State University.
 311

Voorhees, A. & Wei, T. 2002 Three-dimensionality in the wake of a surface piercing cylinder mounted as an
inverted pendulum. proceedings of the 3'd Bluff Body Vortex Induced Vibration Conference,
Queensland, Australia, Dec. 2002.
Williamson, C.H.K. 1996 Three dimensional wake transition. J. Fluid Mech. 328, 345-407.
 VORTEX-INDUCED VIBRATIONS OF SLENDER
STRUCTURES IN SHEAR FLOW

M.S. TRIANTAFYLLOU, F.S. HOVER, A.H. TECHET,

D.K.P. YUE
Department of Ocean Engineering, Massachusetts Institute of Technology,
Cambridge, Massachusetts, 02139, USA, mistetri@mit.edu

Abstract: Long slender structures in shear flows possess a multi-frequency response, because the
various parts of the structure are subject to excitations at different frequencies. We review the basic
mechanisms of response in shear flow and outline the outstanding problems, whose solution is needed to
obtain accurate assessment of the VIV problem.

Key words: vortex-induced vibrations, shear flow, multi-frequency response.

1. INTRODUCTION

Long cables, hawsers are used extensively today for naval, oceanographic and
industrial applications in the ocean, because they can reach the deepest parts of the
ocean. Applications include towing and anchoring of ships, various structures and
buoys, acoustic towed arrays, remotely operated vehicles, cable-laying, and drilling
for production of oil and gas.
In towed cables and hawsers, the drag coefficient of the cable can be
amplified due to Vortex Induced Vibrations (VIV) from a value near 1 for
stationary cylinders to a value of 3 or higher, depending on the amplitude of VIV.
The drag coefficient affects the static shape, and unsteady loads can cause violent
vibrations and fatigue (Alexander 1981, Chapman 1984, Bourget and Marichale
1990, Yoerger et a1.1991, Grosenbaugh et a1.1991, Welch & Tulin 1993).
Similarly, in moored ships and structures, especially in open waters, the drag
coefficient may be amplified due to VIV, causing the damping of the overall

313
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 313-327.
2003 Kluwer Academic Publishers.
314

system, which depends on the drag force acting on the mooring lines, to rise
substantially. Since moored systems at sea have long-period natural frequencies,
on the order of one hundred seconds, which are underdamped, a good estimate of
the total damping is crucially important. The damping due to the presence of
mooring lines can be from 30% up to 85% of the total damping (Huse &
Matsumoto 1989, Koterayama 1989, Triantafyllou et al.1994, Triantafyllou 1999).
Another application where drag coefficient amplification due to VIV has significant
effects is cable-laying, because the precise location of where the cable will
eventually land, and the amount of tension needed to avoid buckling, depend on the
drag force. For oceanographic moorings in the open ocean the most crucial loads
are those due to VIV, which seriously limit their life at sea (Grosenbaugh 1991).
Offshore moorings and risers are used routinely today up to depths of 2,500 m. It is
widely recognized that, particularly for risers, VIV is the most important problem
leading to large static loads and fatigue due to dynamic loads (Brooks 1987, Allen
1999, Allen & Henning 2001). Finally, very large structures in the shape of spar
buoys are used in the ocean today, such as the Genesis platform with a diameter of
37m and a vertical height of 198m, fitted with strakes to reduce VIV. These
platforms, operating at Reynolds number around 108 exhibit VIV, which cannot be
predicted on the basis of existing understanding of low-Reynolds number VIVo

The hydrodynamic problem of VIV, which involves interaction of a flexible

structure with vortex shedding in a turbulent wake, has been and remains an area of
intense research effort (Sarpkaya 1976, Williamson & Roshko 1988, Ongoren &
Rockwell 1998, Zdravkovich 1997, Evangelinos et at. 2000, Govardhan &
Williamson 2000). The viscous flow behind vibrating cables and risers is
dominated by complex flow mechanisms, including changing patterns of vortex
formation, formation of secondary vortical structures, turbulence in the wake, and
correlation length effects. A series of at-sea experiments has demonstrated that the
behaviour of cables at sea is strongly influenced by shear and can result in
substantial changes in the drag coefficient (Y oerger et at. 1991, Grosenbaugh et al.
1991). In the offshore industry, similar VIV problems apply for risers. At shallow
depths, risers behave principally like beams, hence exciting very few modes.
Prediction of VIV is somewhat easier because single mode prediction procedures,
based on empirical models (Brooks 1987, Allen 1999) or laboratory experiments,
 315

can lead to realistic predictions. As the water depth increases, however, risers tend
to behave partly as cables and partly as beams, with the possibility of having multi-
modal response in a sheared current. Intense activity in the offshore industry has
led to the development of experiment-based predictive programs for risers whose
results are compared successfully against full-scale experiments, although
predicting multi-modal response and high Reynolds number behavior, remain
outstanding problems. A widely used programs in the offshore industry is the
program VIVA (Triantafyllou et al. 1999) which is based on

Results from basic research, such as fundamental qualitative properties of

the added mass, lift and drag coefficient of a vibrating cylinder, and the
correlation length oflong structures.
An extensive hydrodynamic database specific for riser shapes (bare
cylinder, straked cylinder, cylinder with attached control pipes, etc.).
Correlation length information based on theoretical and experimental
considerations.

One of the basic outstanding problems today is the lack of data in the open
literature for VIV at high Reynolds numbers (Re). While for a stationary cylinder in
cross-flow it is well known that significant transitions occur with dramatic effects
on the loads and frequency of vortex formation (Shih et at. 1993), very little
information is available for vibrating cylinders.

2. VIV PREDICTION OF RIGID CYLINDER IN UNIFORM FLOW

We start by outlining the basic concepts for VIV prediction as applied to a uniform
cylinder in uniform flow, because several of these concepts can be extended to
shear flow.
First, we have to consider for simplicity a section of a circular rigid
cylinder of constant diameter D within a uniform cross flow of velocity U. If the
cylinder is allowed to oscillate transversely and is mounted on springs of constant k
and damping constant b, vortex-induced vibrations follow, because the Karman
vortices exert unsteady loads, and, under synchronization conditions, the mass-
316

damper-spring system will oscillate with frequency close to its natural frequency.
Figure 1 shows the amplitude of the transverse response, non-dimensionalized by
the diameter, AID, as a function of the reduced velocity Vrn = U I In D, where In
denotes the natural frequency ofthe system in quiescent water, i.e. with added mass
coefficient equal to 1. The response, depending on the ratio of cylinder mass m to
nominal added mass ratio, m*=mlma, where ma=P7!SD214, P is the water density,
and s the span, can be wide-band, in contrast with a classical mass-damping-spring
system, which exhibits a very narrow-band response. The reason for this difference
is the fact that the added mass of a vibrating cylinder in cross-flow is a strong
function of the reduced velocity- in fact it can take on even negative values.
Forced vibrations of the cylinder can provide valuable information on the
fluid forces. In this set of experiments, the cylinder is forced to vibrate transversely
to an oncoming stream U with a harmonic motion of amplitude A and frequency f
The transverse and in-line loads are measured as function of time; The average of
the inline force provides the steady drag force, while the transverse force is
decomposed into a component in-phase with acceleration and a component in-
phase with velocity. The forces, properly non-dimensionalized, provide the drag
coefficient CD, lift coefficient in phase with velocity Clv , and lift coefficient in
phase with acceleration Cia, which can be reformulated to provide the added mass
coefficient Cm (Hover et al. 1998). These coefficients can be plotted as a function
of the amplitude and frequency of response as shown in Figure 2. When the free
vibrations are harmonic, then there is direct correspondence between free vibration
and forced vibration experiments. Indeed, as shown in Figure 2, the free vibration
results follow closely the isocontours of constant qv.
The hydrodynamic coefficients, then, are the building blocks for
predicting VIVo One can write down an equation of motion for the vibrations of
the flexibly mounted cylinder y(t):

my"(t) + by'(t) + ky(t) = f(t) (1)

 317

Figure 1: Response of a flexibly mounted cylinder versus reduced velocity. Upper graph is
AID. second graph is lift coefficient. third graph is lift force phase angle with respect to the
motion. lower graph is end lift force correlation coefficient. Reynolds number is 30.000.
aspect ratio 26. mass ratio 3. damping ratio 3.5% (Hover et al. 2002).

where f(t) denotes the fluid force, which we decompose into a part in phase with
velocity and one in phase with acceleration. By assuming harmonic motion with
complex amplitude Y and frequency w, i.e., y(t) = Re [Ye iWl), we write:

where Re[x} stands for real part of x, and q=O.5psDU2 By substituting into
equation (1), and resolving real vs. imaginary part, we find:

bwA = qC1v (3)

MaJ =k,
318

where M is the mass plus added mass, and A is the absolute value of Y. Equations
(3) can be solved to provide the amplitude and frequency of response. Their
solution must be iterative because the added mass and damping coefficients are
functions of the amplitude and frequency .

~ 0.05
o ~. '* 0.02
c ~s '" 0.01

6.1 6.2

Figure 2: Lift coefficient in phase with velocity Clv (left) and lift coefficient in phase with
acceleration Cia as jUnction ofAID and reduced velocity. Reynolds number 20,000.
Superimposed are free vibration tests for three different values of the damping coeffiCient
(Smogeli 2002).

The first equation expresses the fact that the energy from the fluid is balanced by
the damper; hence if the damping is negligible, free vibrations will take place at
values of C'v=O. The second of equations (3) simply expresses the fact that
vibrations occur at resonance; except that the natural frequency varies because of
the variability of the added mass coefficient. Indeed this variability, which is very
wide, is the root cause of the noted wide-band response of VIV, and is a crucially
important concept for accurate prediction. The variation in added mass can be
easily explained on the basis of the vortices shed, which are of the same frequency
as the motion of the cylinder. One can envision the vortices as low pressure areas
imposing lift forces; when the lift forces synchronize with acceleration they can
reduce the apparent fluid mass to even negative values, or - in the opposite case -
increase the apparent mass.
Two basic concepts that result from forced vibrations are those of wake
capture and wake excitation regions. Wake capture is the gray-shaded region in
Figure 3, wherein the wake of a vibrating cylinder forms vortices at exactly the
frequency of cylinder vibration, rather than at the Strouhal frequency. Outside this
region the loads are found to contain two frequencies, one at the cylinder vibration
frequency and another close to the Strouhal frequency. The second shown region,
 319

denoted as positive energy region in the graph, is the region where the lift
coefficient in phase with velocity is positive, hence providing positive energy to the
system. Free vibrations will take place only within this region, since the fluid
provides the needed energy for vibration, which is dissipated through the structural
damping. Vibrations in the overlap region between these two regions are
monochromatic free vibrations; hence the overlap region is what is usually referred
to as the lock-in region.

AJd

Wake
Capture
Region

Positive
Energy
Region

fdJU
Wake Lock-in Region

figure 3: Wake capture vs. wake pxritntinn (nnsitiw>!'n!'r!7Vl r!'"inns.

1.5 - - -r -- ---Y- ... - : -- - -r
a ! !" I ! !

i....
! ! ! !

i.....
! .t ,! !
-----~~:.j.-- ---:t-----l

~ 0.5
!
i
I
I
I '.'
!
i
I
I
.' oo~b----t----!
t 00 i !CXXlo"" I I
! vv~!

Figure 4: Response of a flexible cylinder (in-line: circles, transverse: dark circles) as

jUnction of reduced velocity. Upper graph shows corresponding trajectories (Davis et al.
2000).
320

3. VIV PREDICTION OF A FLEXIBLE CYLINDER IN UNIFORM FLOW

Moving to the case of a flexible cylinder in uniform flow, one must reformulate the
governing equations to account for the fact that the vibration is a function of span
as well as time. This is not particularly difficult, except now that equations (3)
become integral relations (integrating across the span) rather than algebraic
relations. The details, including numerical implementation can be found in
Triantafyllou (1998). It is to be expected that the response of a flexible cylinder
will be larger at the antinodes than the response of a rigid cylinder, because there is
continuous energy input at the nodes, which is dissipated at the anti nodes.
A more difficult issue is the appearance of in-line as well as transverse
oscillations. As shown in Figure 4, C-shaped ("cresecent") or figure-8 motions are
characteristic of this vibration, in this case for a flexible cylinder (Davis et al.
2000). There can be significant effect from in-line on transverse oscillations
(Sarpkaya 1995)

4. CORRELATION LENGTH

In order to experience large VIV, the vortex shedding, and hence the excitation
force must be well correlated along the length of the structure. Studies show that
the motion of the cylinder reinforces correlation; a three-dimensional instability
that causes the appearance of streamwise, intense-vorticity, low circulation
(compared to the Karman vortices) rib vortices, is the cause of an apparent drop in
correlation; nonetheless, proper filtering shows that correlation length is large in
vibrating cylinders.
It has been found, however, that for values of the reduced velocity near
peak response, correlation between the lift forces at the ends of a cylinder section
drops to very low values (Hover et al. 1998). For a tapered cylinder in uniform
flow, this lack of correlation has been associated with the appearance of a "hybrid
mode" of vortex shedding (Techet et al. 1998). Numerical simulations (Lucor et
al. 2001) have shown the details of such hybrid mode formation in uniform
cylinders in linearly sheared flow.
 321

The drop in correlation has been found in a wide range of Reynolds

numbers, showing that the basic phenomenon is controlled by large scale vortex
dynamics. In Figure 5 we show a comparison of the response and correlation
length in a smooth cylinder to that of a cylinder fitted with tripping wires, to force
earlier transition to turbulence in the boundary layer of the cylinder, following the
procedure of Fage & Warsap (1929), James & Truong (1972), Igarachi (1986), and
others. As seen, there is a decrease in the reduced velocity region where the
correlation drops for the cylinder with wires, but the qualitative features remain the
same. Also, one should note that the region of low correlation is associated with a
region of very high amplitude response. Detailed studies in the wake of a cylinder
(Hover et al. 2002), using hot wire anemometry, show that the lack correlation
manisfests itself in the velocity fluctuations at various locations downstream and
along the cylinder span.

cP8
...... :
+
s~ooth
wires
I.
'" ~ ...
fo.6 .;1;:

0.4
..t.. ~
,it.. .
0.2. _._I,If .~ .:
~, .~
.,~ f--'''''----I--c-----+----+---'-.-."'- ~.-'-.
-'-.,-"-.:'-.~'-.;:-'-.: - :~"7.:;+.
-'--'..
'-+ --j

~ .. ~ ...,.:'.~.;:.:~.;
1.5
1 ~.~"'::':: =..=.::':: - - --
0.5'---'-=:.~---~---~---~-----'--'
4 6 8 10 12

~111111111~; ..~:r+.. ...... .

+:.p-" .-'-; ' "~'..J,*+...., +/~-I+-f.
u. u 0 +,;+
i=FR-
.
.
. ... -If"!'I-+"'+
.
-1~~---r--~----~---r---+----b-~

Figure 5: Response of a flexibly mounted smooth cylinder versus the response of a flexibly
mounted cylinder with trip wires. Upper plot AID; middle plot frequency of response; lower
plot cross-correlation between end liftforces (Hover et al. 2001).
322

The fact that low correlation is associated with high amplitude vibrations is not a
paradox. In the case of the formation of hybrid modes, these are structurally stable
patterns, not randomly varying excitation. Also, it is important to note that the
magnitude of the lift coefficent in phase with velocity is not what controls how high
the response will be, at least for very low damping systems: The controlling
parameter is the value of AID at which the C/v coefficient becomes zero. In Figure
6 we show a typical C/v versus AID curve for constant reduced velocity, as well as
the related lift coefficient in phase with acceleration. The slope at the zero down-
crossing is equal to the hydrodynamic damping (Triantafyllou et al. 1995), while
the maximum value of lift is not a crucial quantity as far as maximum amplitude is
concerned.

5. EFFECT OF SHEAR AND MULTI-MODAL RESPONSE

Shear oncoming flow is known to reduce the amplitude of vibration in a uniform

cylinder (Stansby 1976, Humpreys et at. 1986). An important effect of shear,
however, in long tethers and risers is to cause the appearance of multi-frequency
response.

V 600
, Vr 6,00
r

0,2 0,4 0,6 0,8 1,2 1,4 02 0,4 0,6 0,8 1,2 1,4
NO AID

Figure 6: C'v (left) and Cia (right) versus AID for constant reduced velocity Vr (Smoge/i
2002),
 323

Indeed, various parts of the structures are subject to excitation at different

frequencies, since the oncoming velocity varies. Energy is input in a certain
location - at or near the local Strouhal frequency -- travels along the structure, and
then is dissipated at another location. This necessitates the introduction of complex
modes in the numerical prediction of VIV (Triantafyllou 1998). Indeed, in a
classical standing wave analysis, energy can not travel beyond nodes, and hence a
different methodology is needed, one that allows standing waves, or traveling
waves, or a combination of the two. This is achieved by assuming the amplitude to
be complex, which introduces an amplitude and a constant phase at each location of
the structure. An outstanding problem today is the partitioning of energy among
the multiple frequencies existing in a long structure in shear flow. Gopalkrishnan et
al. (1992) treated the problem of a cylinder section undergoing multi-frequency
vibration (two or three simultaneous frequencies). The lift coefficient in these
conditions is divided among the participating frequencies; also, the lift and the
drag coefficients are affected by the separation between the frequencies. Still, there
is no general theory available to model the partition of energy among the dozen of
frequencies that a very long cable or riser may be subjected to in a sheared flow.
This is a problem requiring immediate attention.

6. CONCLUSIONS

Significant progrees has been made in the predicition of VIV of hawsers and risers.
Experiments derived from short span cylinders are used, together with a strip
theory approach and a correlation length, to derive the vibrations of long structures
in a sheared flow. When the shear is mild and few frequencies are excited,
response is reliable; for multi-frequency vibrations the underlying theory is
missing, while the experimental database is not available yet.

7. ACKNOWLEDGEMENT

Support by ONR, grant number N00014-95-1-1106, monitored by Dr. Thomas

Swean, Jr., is gratefully acknowledged.
324

8. REFERENCES

Alexander C.M, 1981, "The complex vibrations and implied drag of a long oceanographic wire in cross-
flow", Ocean Engng., 8,379406.

Allen D.W., & D.L. Henning, 2001, "Prototype VIV Tests for Production Risers", OTC 13114,
Houston, Texas.

Allen D.W., 1999, ''Vortex-induced vibration analysis of the Auger TLP production and steel catenary
export risers", OTC 7821, Houston, Texas.

Bourget PL., & D. Marichal, 1990, "Remarks about variations in the drag coefficient of circular
cylinders moving through water", Ocean Engng., 17 (6), 569-585.

Brika D., & Laneville A., 1993, "Vortex-induced vibrations of a long flexible circular cylinder", J.
Fluid Mech., 250,481-508.

Brooks I.H., 1987, "A pragmatic approach to vortex-induced vibrations of a drilling riser", OTC Paper
5522, Houston, Texas.

Chapman D.A., 1984, "Towed cable behaviour during ship turning manoeuvres", Ocean Engng., II,
327-361.

Davis, J.T., F.S. Hover, A. Landolt, & M.S. Triantafyllou, 2000, "Vortex Induced Vibrations of Rigid
and Flexible Cylinders", Proc. Workshop on Vortex-Induced Vibrations of Offshore Structures.

Fage A., & J.H. Warsap, 1929, "The effects of turbulence and surface roughness on the drag of a
circular cylinder", Reports and Memoranda of the Aeronautical Research Committee No. 1283.

Grosenbaugh M.A., 1996, "On the dynamics of oceanographic moorings", Ocean Engng., 23,7-25.

Gopalkrishnan R., M.A. Grosenbaugh, & M.S. Triantafyllou, 1992, "Amplitude modulated cylinders in
constant flow: Fundamental experiments to predict response in shear flow. Proc. Third Intern. Symp.
Flow Induced Vibrations and Noise ASME, Anaheim, California.
 325

Govardhan R., & C.H.K. Williamson, 2000, " Modes of vortex formation and frequency response of a
freely vibrating cylinder" , J. Fluid Mech., 420, 85-130.

Grosenbaugh M.A., D.R. Yoerger, M.S. TriantafYllou & F.S. Hover, 1991, "Drag Forces and Flow-
Induced Vibrations of a Long Vertical Tow Cable - Part II: Unsteady Towing Conditions", J. Offshore
Mech. Arctic Engng. (ASME), 1\3 (3), 199-204.

Hover F.S. , A.H. Techet, & M.S. TriantafYllou, 1998, "Forces on oscillating uniform and tapered
cylinders in cross flow", J. Fluid Mech., 363,97-114.

Hover, F.S. J.T. Davis, & M.S. TriantafYllou, 2002, "Three-Dimensionality of Mode Transition in
Vortex-Induced Vibrations ofa Circular Cylinder," European Journal of Mechanics B, to appear.

Hover, F.S. H. Tvedt, & M.S. TriantafYllou, 2001 , "Vortex-induced vibrations of a cylinder with
tripping wires, " J. Fluid Mech., 448, 175-195.

Humphreys lA., & D.H. Walker, 1987, "Vortex excited response of large scale cylinders in sheared
flow", Proc. Sixth In!. Offshore Mech. Arctic Engng Symp. (ASME), Houston, Texas, 139-147.

Huse E. , & K. Matsumoto, 1989, Mooring Line Damping due to First and Second Order Vessel Motion,
OTC# 5676.

Igarashi T. , 1986, "Effect of tripping wires on the flow around a circular cylinder normal to an
airstream", Bull. Jap. Soc. Mech. Eng., 29, 2917-2924.

James, D.F., & Q.S. Truong, 1972, "Wind load on cylinder with spanwise protrusion", Proc. ASCE, 1
Eng. Mech. Div., 98, 1573-1589.

Koterayama W., 1989, " Viscous damping forces for slow drift oscillation of the floating structure acting
on the hull and the mooring lines", Proc. Eighth Intern. OMAE, The Hague, The Netherlands.

Lucor D., L. Imas & G.Em. Karniadakis, 2001, "Vortex displocations and force distribution of long
flexible cylinders subjected to sheared flows", J. Fluids Struct., 15,
641-650.
326

Ongoren A., & D. Rockwell, 1988, "Flow structure from an oscillating cylinder part I. Mechanisms of
phase shift and recovery in the near wake", J. Fluid Mech. 191,197-223.

Sarpkaya T., 1979, "Vortex-induced oscillations", J. Applied Mech., 46,241-257.

Sapkaya T., 1995, "Hydrodynamic damping, flow induced oscillations and biharmonic response",
ASME J. Offshore Mech. Arctic Engng., 177,232-238.

Shih W.c.L., C. Wang, D. Coles & A. Roshko, 1993, "Experiments on Flow Past Rough Circular
Cylinders at Large Reynolds Numbers", J. Wind Engng. Industrial Aero., 49,351-368.

Smogeli 0., 2002, "Design and evaluation of a force feedback control system for VN experiments",
Thesis, Marine Technology NTNU and Ocean Engineering MIT, Cambridge, MA.

Stansby P.K., 1976, "The locking-on of vortex shedding due to the cross-stream vibration of circular
cylinders in uniform and shear flows", J. Fluid Mech. 74, 641-665.

Techet A.H., F.S. Hover, & M.S. Triantafyllou, 1998, "Vortical patterns behind tapered cylinders
oscillating transversely to a uniform flow", J. Fluid Mech., 363, 79-96.

Triantafyllou M.S., D.K.P. Vue, & D.Y.S. Tein, 1994, "Damping of Moored Floating Structures", OTC
7489, Houston Texas.

Triantafyllou M.S., & M.A. Grosenbaugh, 1994, "Prediction of flow-induced vibrations in sheared
flows", Proc. 6th Intern. Conference on Flow-induced Vibrations, London, U.K.

Triantafyllou G.S., 1998, "Vortex Induced Vibrations of Flexible Structures", Summer Meeting of the
American Society of Mechanical Engineers, Washington, D.C.

Triantafyllou M.S., 1999, '"Cable Dynamics for Offshore Applications", in Developments in Offshore
Engineering: Wave Phenomena and Offshore Topics, editor J.B. Herbich, 256-294, Gulf Publishing
Company, Houston, Texas.

Triantafyllou M.S., D.K.P. Vue, & Y.S.D. Tein, 1993, "Damping due to Mooring Lines", Proc.
Offshore Technology Conference, Paper No. 7891 , Houston, Texas.
 327

Triantafyllou M.S., G.S. Triantafyllou, Y.S.D. Tein & B.D. Ambrose, 1999, "Pragmatic Riser VIV
Analysis", Proc. Offshore Technology Conference, Paper No. 10931, Houston, Texas.

Welch S., & M.P. Tulin, 1993, "An experimental investigation of the mean and dynamic tensions in
towed strumming cables", J. Offshore Polar Engng., 3 (3), 213-218

Williamson C.H.K. , & A. Roshko, 1988, " Vortex formation in the wake of an oscillating cylinder", J.
Fluids Struct., 2,355-381.

Yoerger D.R., M.A. Grosenbaugh, M.S. Triantafyllou, & J.J. Burgess, 1991, "Drag Forces and Flow-
Induced Vibrations of a Long Vertical Tow Cable - Part I: Steady-State Towing Conditions", J.
Offshore Mech. Arctic Engng. (ASME), 113,117-127.

Zdravkovich M.M., 1997, Flow around circular cylinders, Vols. I and 2, Oxford University Press:
Oxford.
 A TUNED LIQUID DAMPER MODEL

JANNETTE B. FRANDSEN
Department of Civil and Environmental Engineering,
Louisiana State University, Baton Rouge, Louisiana 70803, USA.
Fax: 2255788652, e-mail: frandsen@lsu.edu

Abstract: For some large structures, structural damping alone has been insufficient
to limit the dynamic motions to acceptable levels for serviceability considerations
(e.g. the review paper by Kareem et al. 1999). Thus, auxiliary dampers have been
added. Designers are faced with the task of understanding complex fluid-structure
interactions when attempting to estimate energy dissipation performance of, for
example, tuned sloshing dampers. To this end, a numerical wave tank can provide
useful information on the free surface motions, resonant frequencies, etc.

In the present paper, the nonlinear effects of sloshing motions in moving

liquid tanks (Frandsen 2002) are investigated in connection with using tuned liquid
dampers (TLD) to suppress structural vibrations in tall buildings or other structures
to along-wind or cross-wind forces. The current, fully nonlinear model simulates
two-dimensional, non-overturning waves during structural vibrations. Numerical
solutions of the governing nonlinear potential flow equations are obtained using a
finite-difference time-stepping scheme on adaptively mapped grids.

Simulations of sloshing motions in forced excited tanks are carried out in

which tank size and fluid-to-structure mass ratio is varied. Natural frequencies of
the structure-TLD system are extracted for small to steep amplitude waves. All
results obtained show that the coupling of a liquid storage tank to a structure can
change the behaviour of the entire system considerably. An optimum TLD-
structure system is identified relative to shift in resonance frequency. Good
agreement for small horizontal forcing amplitude is achieved between the
numerical model and first order small perturbation theory. The dependence of the
nonlinear behaviour of the solution on the wave steepness is discussed. It is found
that nonlinear effects become important when the steepness reaches about 0.1.

REFERENCES
Frandsen, J.B. (2002). "Sloshing effects in periodically and seismically excited tanks." In Proceedings
of the fifth World Congress on Computational Mechanics, Vienna, Austria. ISBN 3-9501554-0-6.

Kareem, A., Kijewski, T. and Tamura, Y. (1999). "Mitigation of motions of tall buildings with specific
examples of recent applications." In Journal of Wind and Structures, Vol. 2, no. 3, pp. 132-184.

329
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 329.
2003 Kluwer Academic Publishers.
 UNSTABLE PHENOMENA OF A THIN
CYLINDRICAL SHELL SUBJECTED TO AXIAL
LEAKAGE FLOW

K. FUJITA AND A. SHINTANI

Abstract: The unstable phenomena of a thin cylindrical shell subjected to an axial leakage flow
are investigated in this paper. The analytical model is composed of an elastic shell and a rigid
one which are arranged co-axially. Considering the fluid structure interaction between a shell
and a fluid flowing through a narrow passage, the coupled equation of motion between a shell
and a flowing fluid is derived using the Donnell's shell theory and the Navier-Stokes equation.
Focusing on the vibrational characteristics of a thin cylindrical shell, that is, the axisymmetric
(ring-like) vibration, the lateral (beam-like) vibration and the circumferential (ovaling) vibration,
the unstable phenomena of a thin cylindrical shell are clarified by using the root locus (Argand
diagram) based on the complex eigenvalue analysis. The critical fluid velocities for the divergence
and flutter phenomena of a thin cylindrical shell subjected to an axial leakage flow are analyzed
and defined. Moreover, comparing with the uncoupled vibration modes, namely dry modes, the
coupled vibration modes between a shell and a flowing fluid, namely wet modes, are shown to be
considerably changed between before and after the unstable phenomena. The numerical parameter
studies on the shell with both simply-supported ends are performed taking the dimensions of a shell,
the characteristics of a flowing fluid and the clearance of an annular gap between an elastic and a
rigid shells as parameters. The influences of these parameters on the threshold of the instability
of the coupled vibration between a shell and a flowing fluid are investigated and discussed.

1. INTRODUCTION

A fluid is often conducted as the energy transfer medium in many industri-

al machineries. On the other hand, many machines and structures must be
constructed and operated in flowing fluid.
The velocity of fluid becomes faster in order to increase the efficiency and
make the size more compact. In this paper, the fluid-elastic instability which
occurs in a coaxial cylindrical shell structures subjected to an axial leakage flow
is investigated. The machines and structures which are composed of coaxial
cylindrical shells are recognized in the combustors of gas turbines, the core in-
ternals of nuclear power plants, the annular structures in large diameter pipings
and various large scale seals and valves.
The vibrational mechanisms of machines and structures subjected to an
axial leakage flow have been studied by many researchers [1]-[4]. However, the
studies conducted to be based on a shell theory have not reported so many. As
representatives, M. P. Pa"idoussis et al. [5] studied the dynamics and stability
of coaxial cylindrical shells containing flowing fluid using the Fliigge's equation
and the velocity potential theory. S. Ziada, E. T. Biihlman, and U. Bolleter
[6] reported the instability of the heat-shielding shell of an after-burner due to

331
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 331-351.
2003 Kluwer Academic Publishers.
332

a cooling flow. One of authors [7] has also studied the instability phenomena
of a coaxial cylindrical shell due to an annular leakage flow experimentally
and analytically. Accordingly, the further remaining study can be considered
to be still necessary for developing the improved coupled vibrational analysis
methodology.
In this paper, considering the fluid structure interaction between a shell
and a fluid flowing through a narrow passage, the coupled equation of motion
between a shell and a flowing fluid is derived using the Donnell's shell theory
and the Navier-Stokes equation which is applied to an annular leakage flow
in accompanying with some approximated assumptions. The vibrational char-
acteristics of a thin cylindrical shell can be classified into the axisymmetric
(ring-like) vibration, the lateral (beam-like) vibration and the circumferential
(ovaling) vibration as shown in Fig.I. Focusing on those three types of vi-

Eigenmode in Circumferential Direction

0. 0
Axisymmetric lateral
(Ring-like)
Vibration
.', , (beam-like)
Vibration
" ..... -~ . ..' ",,,, .. ,,"

u=o u=l
Circumferential (Ovaling) Vibration

.
,
0 0
,

lF2
... __
,,
I
.
.'

11=3
.,
0
/'
,
.

IF'!
\
,

Eigenmode in Radial Direction

~ 5-:53- t:~~
m=1 m=2 111=3

Figure 1. Vibrational characteristics of a thin cylindrical shell

brations, the unstable phenomena of a thin cylindrical shell are clarified by

using the root locus (Argand diagram) due to the complex eigenvalue analysis.
The critical fluid velocities for the divergence and flutter phenomena of a thin
cylindrical shell subjected to an axial leakage flow are analyzed and defined.
Moreover, comparing with the uncoupled vibration modes which are namely
dry modes, the coupled vibration modes, namely wet modes, are shown to
be considerably changed between before and after the unstable phenomena.
Finally, regarding the dimensions of a thin cylindrical shell, the velocity of
an axial leakage flow, the characteristics of flowing fluid and the clearance of
 333

an annular gap as the parameters, the physical influences of parameters on

the threshold of the instability of the coupled vibration are considered and
discussed.

2. EQUATION OF MOTION

Let us consider the thin cylindrical shells subjected to the axial leakage flow
shown in Fig.2. The inner shell is set to elastic, on the other hand, the outer

displacement of shell
w

elastic shell
rigid shell
fluid fluid velocity
(axial leakage flow)

Figure 2. A nalytical model of thin cylindrical shell

shell is set to rigid, and both are set coaxially. The dimensions of the inner
elastic shell are length: e, radius: a, thickness: h, the Young's modulus: E,
the Poisson ratio: v, density: p. The leakage flow passes through the annulus
between two shells. The inner elastic shell vibrates due to the fluid force of the
axial leakage flow. In this paper, the annular gap is considered to be sufficiently
small compared with the radius of shell. On the other hand, even if the inner
structure is a rigid cylinder and the outer structure is an elastic shell, our
theory can be similarly applied by changing the acting direction of the fluid
pressure.
The equations of motion of an incompressible fluid based on the Navier-
Stokes equations are given by equations of continuity and momentum:

aU + aV + oW = 0 (1)
ax ay azI
aU aU aU aU 1 aP a2 u (2)
at + U ax + v ay + WaZI = - PI ax + vI aZJ
aV aV aV aV 1aP a2 V
-+U-+V-+W- =---+1/1--, (3)
at ax ay azI Play aZJ
334

where U, V, W: flow velocity ofthe X-, ()-, zr directions, respectively (y = a()),

P: pressure, pr density of a fluid and Vj: kinematic viscosity of a fluid as
shown in Fig.2.
The equations of a thin cylindrical shell are given by the Donnell's equa-
tions:

(4)

(5)

(6)

where J-t = ph, and Px, Po, pz: external fluid forces acting on a shell, m x ,
mo: moments due to the fluid pressure acting on a shell, and the extensional
rigidity and the flexural rigidity of a shell C, Dare C = Eh/(l - v 2 ), D =
Eh 3 /12(1 - v 2 ).
Now, introducing the axial and the circumferential flow rate Qx = foH U dz j ,
Qy = fOH Vdzj, where H is the gap, and integrating the equations (1)-(3), the
flow velocity W disappears by using boundary conditions between a shell and
a fluid and the following equations which describe the relation among the flow
rate, pressure and gap are obtained:

oQx
ax +
oQy
oy +
oH _
at -
(7)

~ oP = _~ {oQx + ~ (Q;) +~ (QxQy) + 12vjQx } (8)

pj ax H at ax H oy H H2

~ oP = _~ {oQy + ~ (QxQy) +~ (Q;) + 12VjQy}. (9)

Ploy H at ax H oy H H2

The boundary conditions of the pressure at the inlet and outlet are

pea, y, t) p. _ (1 + ( ) PiQ;(O, y, t) (10)

m m 2H2(0, y, t)

P. I" PIQ;(,y,t)
pel, y, t) = out + .,out 2H2(, y, t) , (11)

where Pin and Pout are the pressures just in front of the inlet and just behind
the outlet in an axial leakage flow, and (in, (out are the pressure loss coefficients
at the inlet and outlet.
 335

3. UNSTEADY PRESSURE

The gap, pressure and flow rate can be divided into the steady and the unsteady
parts as H = fI +6.H(x, y, t), P = P(x) +6.P(x, y, t), Qx = Qx +6.Qx(x, y, t),
Qy = 6.Qy(x, y, t), where the bar and 6. denote the steady and the unsteady
values. The steady circumferential flow rate is neglected because it is regarded
smaller than the axial flow rate. Moreover, fI and Qy are constant. By applying
the perturbation method and neglecting the higher order terms of the unsteady
components, the equations (7)-(11) can be linearlized. Eliminating the flow
rates from these equations, we can derive the relation between the unsteady
pressure and the unsteady gap
1 ({)2 {)2 ) 1 {)2 t:..H 2Qx {)2 t:..H
pj {)x2 + {)y2 t:..P = fI fji2 + fI2 {)x{)t
1211j {)t:..H Q~ {)2 t:..H 3611jQx {)t:..H
+ fI3 ---at + fI3 [f;2 + --yj4 --a;-. (12)

4. FLUID-SHELL COUPLED VIBRATION DISPLACEMENTS

The vibrational characteristics of a thin cylindrical shell can be classified rough-

ly into three types. The first is an axisymmetric vibration called as a ring-like
vibration, which is the vibration mainly due to elongation and shrinkage in
circumferential direction depending on the membrane stiffness of a cylindrical
shell. The second is a lateral vibration, which is also called as a beam-like
vibration. This is the bending vibration in the lateral direction of a cylindrical
shell, similar to a beam vibration. The third is a circumferential vibration or
an ovaling vibration, which is mainly due to the bending in the circumferential
direction. Due to this vibration, the cross area of the cylindrical shell becomes
non-circle. In this paper, classifying the vibrations into three types of vibra-
tions, the fluid-structure coupling and the unstable behavior are considered
and discussed.
The displacements of a shell subjected to an axial leakage flow are assumed
to be approximated by superposing the eigenmode functions in a vacuum (dry
modes). Employing the eigenmodes in the axial, circumferential and radial
directions, umn(x), vmn(x) and wmn(x), respectively, the fluid-coupled dis-
placements u(x,B,t), v(x,B,t) and w(x,B,t) are expanded as follows:

U(X,O,t))
( v(x,O,t) = LL (Umn(x)cosno)
00 00

Vmn(X) sin nO emn(t), (13)

w(x, 0, t) m=l n=O wmn (x) cos nO

where ~mn(t) is the function of time. The numerical calculation is done un-
der the both end conditions which are simply supported in the radial and
circumferential directions, not supported in the axial direction. The eigen-
mode functions in a vacuum are as follows: umn(x) = C1mncos(mrrx/C),
336

Vmn(X) = C2mn sin(m7rxj), wmn(X) = C3mn sin(m7rxj). Here, these equa-

tions show the mode functions of the axisymmetric, lateral and circumferential
vibrations for n = 0, n = 1, n ~ 2, respectively.

5. COUPLED FLUID FORCE

Here, in this section, the unsteady pressure 6..P is derived and their fluid forces
and moments are calculated more specifically. Since the unsteady gap width
6..H coincides with the radial displacement of a shell, the following relation can
be obtained: 6..H = w. Using this relation, Eq.(12) is rewritten as

pI!
EP
( 8x 82 )
2 + 8y2 /:}'P(x, e, t) = L L cos ne {Amn(x)tmn(t)
"'" "'"

m=l n=O

+Bmn(x)~mn(t) + Cmn(X)~mn(t)}, (14)

where the coefficients Amn(x), Bmn(x), Cmn(x) are the functions given by the
modal function wmn(x) and its derivatives. Since y is described as y=a(),
the left hand side of Eq.(14) is the function of x, () and t. Similarly to
Eq.(13), the unsteady pressure 6..P is assumed to be written as 6..P(x, (), t) =
2:~=o 6..Pn (x, t) cos n(), where 6..Pn means the unsteady pressure at () = O. Sub-
stituting this into Eq.(14), the ordinary differential equation is obtained. Solv-
ing the obtained ordinary differential equation under the linearlized boundary
conditions based on Eqs.(lO), (11) , the relation between the unsteady pressure
6..P(x, (), t) and the modal displacement is obtained:

/:}'P(x, e, t) = ff
m=l n=O
cosne{ inmn (X)tmn (t) + Cmn~mn(t) + kmn~mn(t)}. (15)

The functions mmn(x), cmn(x) and kmn(x) are coefficient functions. Their
concrete forms are omitted.

6. FLUID-SHELL COUPLED EQUATION AND STABILITY ANALYSIS

The fluid forces acting on a shell are given by Px = -J.Lt86..U j8z t lz,=o , mx =
J.Lt86..U j8z t lz/=o(hj2), Pe = -J.Lt86.. V j8z t lz,=o, me = J.Lt86.. V j8z t lz/=o(hj2),
pz = 6..P(x, (), t)lz/=o = 6..P(x, (), t). Assuming the velocity distribution as
6..U = (lj2J.Lt)86..Pj8xzt(zt-H), 6..V = (lj2J.Lt)86..Pj8()zt(zt-H), the fluid
forces become Px = -(Hj2)86..Pj8x, Pe = (Hj2)86..Pj8(), pz = 6..P(x,(),t),
mx = (hHj4)86..Pj8x , me = (hHj4)86..Pj8().
Substituting these and the equations of eigenmodes in a vacuum into the
equations of a thin cylindrical shell (4)-(6) and utilizing the orthogonality of
the mode of the axial and circumferential directions, the following fluid-shell
coupled equation of motion is obtained, that is, multiplying Eqs.(4)-(6) by the
eigenfunction of the p-th in the axial direction and the q-th in the circumferen-
tial direction and integrating the resultant equation along the axial direction
 337

(0 to C) and the circumferential direction (0 to 27r);

L
00

tpq(t) + W;q~pq(t) = - {mpmqtmq(t) + Cpmqemq(t) + kpmq~mq(t)}, (16)

m=l

where Wpq is the eigenfrequency of the shell in a vacuum. The left hand side
of the above equation is described on only the shell, so it can be understood
that the left hand side is orthogonalized with the p-th and q-th mode. On
the other hand, the fluid forces in the right hand side in Eq.(16) can not be
orthogonalized. That is, the combination of the function of time with m =
1,2, ... ,00 is left. By this equation, even in the fluid-shell coupled vibration
subjected to the axial leakage flow, the vibration modes in the axial direction
and the circumferential one can be transformed to the product of the function
of each variable when a coupled eigenvalue analysis is performed for Eq.(16).
And the orthogonality condition among the coupled modes can be applicable.
Therefore, it is seen that similarly to the modes in a vacuum, the axisymmetric,
lateral and circumferential vibrations are considered separately, even in a fluid-
shell coupling. Here the concrete form of the coefficient m pmq , cpmq , kpmq
are omitted. Truncating the number of mode as M, employing vectors and
matrices form, and changing q to n, Eq.(16) can be rewritten as follows;

where {~n(t)} = {6n(t),6n(t)""'~Mn(t)V, [Ms,nl = I MxM , [Ks,nl = diag

{wrn,w~n"" ,w~n}, [Ma,nl = [mmpnh<m,p<M, [Ca,nl = [cmpnh<m,p<M, [Ka,nl
= [kmpnh::;m,p::;M' The matrices [Ms,nl~ [Ks~nl are the modal ma;s and stiffness
of only a shell, and [Ma,n]' [Ca,n]' [Ka,nl are the added mass, added damping,
added stiffness matrices of a fluid in a gap, and n=O, n=l and n ~2 denote the
axisymmetric, lateral and circumferential vibrations respectively. By calculat-
ing the complex eigenvalue s and plotting the root locus (Argand diagram),
the stability of a system can be evaluated.
Here, there are 2M complex eigenmodes of Eq.(17) for fixed n. M pairs of
eigenvalues exist. Each pair is a set of complex conjugate numbers or a pair of
two real numbers. Let us call the pairs the 1,2,3,'" M-th eigenvalues. When
the imaginary part of a complex eigenvalue is not zero, we adopt the complex
eigenvalue with a positive imaginary part in a pair of complex conjugate eigen-
values. On the other hand, when the imaginary part is zero, we adopt the
eigenvalue with a larger real part in a pair of eigenvalues. Thus, the focused
M complex eigenvalues are written as {Al,n, A2,n,"', AM,n}' Let the complex
eigenvectors corresponding to Ak,n be {Wk,n}, and m-th element of {Wk,n} be
Wk,n,m' Moreover, let rJk,n(t) be the coupled function of time corresponding
to {Wk,n}' Therefore, the coupled k-th complex eigenmodes are expressed as
338

follows:
M

u'k,n(x,8) =L umn(x)cos(n8)Wk,n,m
m=l
M

v'k,n(x,8) =L Vmn(X) sin(n8)Wk,n,m (18)

m=l
M

w'k,n(x,8) =L Wmn(X) cos(n8)Wk,n,m.

m=l

Here, note the coupled complex eigenvalues and eigenmodes vary when the flow
velocity varies. By employing these equations, the displacement of a cylindrical
shell can be expressed as follows:

u(x, 8,
( vex, 8, t)
t)) = L L
M N ( u'k,n(x,8) )
v~,n (x, 8) 1]k,n (t) (19)
w(x, 8, t) k=l n=l Wk,n (x, 8)

The fact that the imaginary part of complex eigenvalue s satisfies Im(s) -
o and the real part Re(s) > 0 implies dynamically unstable (flutter), and
Im(s) = 0 and Re(s) > 0 implies statically unstable (divergence). According
to these procedures, the critical flow velocity and complex eigenmode shape for
the leakage-flow induced vibration can be evaluated.

7. NUMERICAL ANALYSIS

7.1. Dimensions for Numerical Analysis Studies

As numerical analysis examples, let us consider the inner shell is elastic, and
the outer shell is rigid like a hollow rigid cylinder as shown in Fig.2. Two types
of shell models are considered here: One is the shorter elastic shell, the other is
the longer one comparing with their radii shown in Table 1. The elastic shell is

Table 1. Dimensions of numerical models

length [m] radius [m] thickness [m] gap width [m]
Case 1 0.5 1 0.002 0.003
Case 2 2 1 0.002 0.003

assumed to be simply supported at both ends in the radial and circumferential

directions as explained above. The fluid is water, the elastic shell is considered
to be made of steel. Hereafter, let us call these models Case 1 and Case 2.
 339

7.2. Eigenvalue Analysis of Numerical Models in Dry and Wet Conditions

Figures 3 and 4 show the eigenfrequencies in dry and wet conditions for Case
1 and Case 2 respectively. The dry condition means the eigenvalues in a

7000 F=====================~~ '02000

Q)

O~O [~__________________~~
V>
"-
"0
~
~
:;, '"
~1500
~ 5000 >.
<)
~
<::
Q)
>.
(,) 4000 ::J
<:: ~ 1000 4
....<::
Q)
::J

~3000
....<:: Q)

""
-ii) 500
3

gj,2000 2
x
Q)
Q) 0=1 ~
m=1 __
1~0L-----~5----~
10~~~15~==~20
o
ci rcumferential mode number o 5 10 15 20
C i rcumferent i a I mode number

(a) Dry mode (b) Wet mode

Figure 3. Eigenvalue analysis (Case 1)

0000 ,-----~----~----------~
o350 ,-----~----------~----_,
Q)
V>

~5000 ti 300
"-
"0 ~
~ 4000 >- 250
<)
<::
>.
g3000
Q)
~ 200
C'
Q)
::J
C' ~ 150
<::
....<::
~ 2000 Q)

.~ 100
Q)
Q)
_~ 1000
Q)
~ 50 m=1

0L-----~5----~10----~15~--~20 ~ . L -_ _ _ _ ~_ _ _ _ _ __ _ _ _~_ _ _ _~
<) 00 5 10 15 20
C i rcumferent i a I mode number
C i rcumferent i a I mode number

(a) Dry mode (b) Wet mode

Figure 4. Eigenvalue analysis (Case 2)

vacuum, and the wet means those in a still fluid. Hereafter let's call them dry
modes, and wet modes respectively. When a fluid is air, wet modes almost
equal the dry ones. In these figures, the axial mode number m is taken as
a parameter. As a fluid is water, the eigenfrequencies of wet modes become
much lower due to the added mass effect of the fluid . Especially in Case 2, the
eigenfrequencies of dry modes show the minimum at the mode (m, n) = (1,8).
340

On the other hand, those of wet modes show that at the mode (m,n) = (1,7).
Hereafter, let us consider the circumferential mode n = 7 as the representative
example for the circumferential vibration.

7.3. Stability on Axisymmetric, Lateral and Circumferential Vibrations

Concerning the axisymmetric, lateral and circumferential vibrations of a thin
cylindrical shell subjected to an axial leakage flow, their dynamic stability is
investigated in this subsection. Figures 5 and 6 show the root loci (that is
Argand diagrams) of three types of vibrations for Case 1 and Case 2 respec-
tively. Regarding the axisymmetric vibration in Case 1, the 1st axial mode

200
800

3rd mo e 700
3rd mode
150 600
t: t:
'"c. 2nd mod ~500
~100 0
;:-400
'"
C to
C
OIl "bi,300
to to
E E
-50 1st fOOd -200

...
10
flu ter divergenc 100
35 (1st. divergence)

35.5 (2nd.
9.0 9. 06 35 f lutter )
0 0
-3 -2 -1 -3 -2 -1 0
Real part Real part

(a) Axisymmetric (b) Lateral

700
0 3 d mode
600

0
Ir 2 d mode

30

ii= [mI,)

100 30 ~1 t mode 2nd 32-+

20 1st 24 -+
o
-5 o
Rea I part

(c) Circumferential (n = 7)

Figure 5. Root locus (Argand diagram), Case 1

is stable, and the 2nd mode goes into flutter phenomenon at (J = 9.03m/s,
and into divergence phenomenon at (J = 9.06m/s increasing the fluid velocity.
For the lateral vibration in Case 1, the 1st axial mode becomes divergence at
 341
50 200

3rd mode o 3rd rood

oW
150
t t
1!iw 0
'""-
~ ;::;~~. ~100 2nd rood

.,'"
.=20
.,'"
t::

'"E '"E
:V':!:'::"
U=O[m/ s)
- 50 1st rood
10
2nd 34-+
8 1st 32-+
0 0
-2 -1. 5 -1 -{).5 0.5 -3 -2 -1
Rea I part Rea I part

(a) Axisymmetric (b) Lateral

100
3rd mode 0

80
...
~ 60
>-

.=.,'"
~
2nd mode 0

'"
E
40
~ 20.5

20 14~ ------
.5
15
1st mOld,! _
U=O m/s]
8 10
-2 0 2
Real part
(C) Circumferential (n = 7)

Figure 6. Root locus (Argand diagram), Case 2

(J = 35.0m/s, and the 2nd mode flutter at (J = 35.5m/s. Moreover, for the
circumferential vibration in Case 1, the 1st axial mode becomes divergence at
(J = 24m/s, and the 2nd mode flutter at (J = 32m/s. The axisymmetric vibra-
tion can be seen to become unstable first among them. That is, for the smaller
ratio of length to radius (R/a), the axisymmetric vibration can be considered
to be apt to go into unstable phenomenon among them. On the other hand, all
modes of the axisymmetric vibration in Case 2 are stable increasing the fluid
velocity. For the lateral vibration in Case 2, the 1st and 2nd axial modes go
into the divergence phenomena at (J = 32m/s and (J = 34m/s respectively.
Further, for the circumferential vibration in Case 2, the 1st axial mode be-
comes divergence at (J = lOm/s, and the 2nd mode flutter at (J = 15m/s. The
circumferential vibration can be seen to become unstable first among them for
the larger ratio of Cia. From both numerical studies, the determination that
which vibration mode among three types of vibrations of a thin cylindrical shell
342

subjected to an axial leakage flow goes first into the unstable phenomenon is
found to be due to the dimensions of a shell.

7.4. Complex Eigenmode

Figure 7 shows the 2nd complex eigenmode of the axisymmetric vibration in
Case 1 which becomes unstable first. Figure 7(a) indicates that in the still

t=O t = (1/8)T t = (2/8)T t=O t = (1/8)T

t = (3/8)T t = (4/8)T t = (5/8)T t = (3/8)T t = (4/8)T t = (5/8)T

t = (6/8)T t = (7/8)T t=T

t = (7/8)T t=T
upstream -+ downstream
upstream -+ downstream
(b) tJ = 9.03 [m/s], Flutter, 2nd
(a) tJ = 0, 2nd mode mode

q
M
. .... . .

upstream -+ downstream
(c) tJ = 9.06 [m/s], Divergence,
2nd mode
Figure 7. Complex eigenvalues of the axisymmertic vibration (Case 1)

fluid at tJ = 0, Fig.7(b) the flutter mode at tJ = 9.03m/s, and Fig.7(c) the

 343

divergence mode at (j = 9.06m/s. These figures show the mode shapes at

t =0 t = (1/8)T t = (2/8)T

t = (3/8)T t = (4/8)T t = (5/8)T

upstream --+ downstream
(b) (j = 32 [m/s], Divergence, 1st
mode
t = (6/8)T t = (7/8)T t =T
upstream --+ downstream
(a) (j = 0, 1st mode

Figure 8. Complex eigenvalues of the lateral vibration (Case 2)

t= 0 t = (1/8)T t = (2/8)T

t = (3/8)T t = (4/8)T t = (5/8)T

upstream --+ downstream
(b) (j = 10 [m/s], Divergence, 1st
mode
t = (6/8)T t = (7/8)T t =T
upstream --+ downstream
(a) (j = 0, 1st mode

Figure 9: Complex eigenmode of the circumferential vibration (1st mode, Case

2)

every instances, t = 0, T /8, 2T /8, .. , 7T /8, T over one period T. The complex
mode at (j = Om/s is almost the same as the dry mode. However, the flutter
mode shape at (j = 9.03m/s is fairly different from that at (j = Om/s, and
344

the similar shape to the divergence mode shape at [; = 9.06m/s can be found
among the flutter mode shapes of [; = 9.03m/s at every instances of time cut.
Further, the just side view of the divergence mode is also added to the right
hand side of Fig.7(c). Similarly, in the following divergence mode shapes, the
subfigures in the right side denote the just side views.
Figure 8 shows the 1st complex eigenmode of the lateral vibration in Case
2 which goes into the divergence phenomenon. Figure 8(a) indicates that in
the still fluid at [; = Om/s which is almost the same as the dry mode, and
Fig.8(b) the divergence mode at [; = 32m/s which is fairly different from that
at [; = Om/so Especially, the symmetry of a mode shape in the axial direction
is found not to be kept as for the divergence mode at [; = 32m/s.
Figures 9 and 10 show the 1st and 2nd complex eigenmodes of the cir-
cumferential vibration in Case 2 which go into the divergence and the flutter
phenomena respectively. Figure 9(a) indicates the 1st eigenmode in the still

t=O t = (1/8)T t = (2/8)T t=O t = (1/8)T t = (2/8)T

t = (4/8)T t = (5/8)T t = (3/8)T t = (4/8)T t = (5/8)T

t = (6/8)T t = (7/8)T t =T
t = (6/8)T t = (7/8)T t=T

upstream -+ downstream upstream -+ downstream

(a) [; = 0, 2nd mode (b) [; = 15 [m/s], Flutter 2nd

mode
Figure 10: Complex eigenmode of the circumferential vibration (2nd mode,
Case 2)

fluid at [; = Om/s which is also almost the same as the dry mode, and Fig.9(b)
the divergence mode at [; = 10m/so Moreover, Fig.lO(a) indicates the 2nd
eigenmode in the still fluid at [; = Om/s which is also almost the same as the
dry mode, and Fig.lO(b) the flutter mode at [; = 15m/s which is considerably
different from that at [; = Om/so Especially, the location of the maximum am-
plitude of the flutter mode over one period is seen not to be fixed axially and
to move along the axial direction. This behavior can be considered to mean
the generation of a traveling wave axially.
 345

7. 5. Influence of Gap Width on Vibration Behavior

~20 ~
I --------~

~151
o
Q)
>10

'"
(,)

.~ SL---___ _ _ _ _-----.J
(,) 4 6 8 10
gap width [ml , 10'

(a) Flutter (b) Divergence

Figure 11: Influence of the gap width on the critical velocity (axisymmetric,
Case 1)

~w ,---------,
'-
.5 50
>-
:~40
(,)
~30
Q)
>20

'" [L-______
~I: -----.J
(,) 4 6 8 10
gap width [ml ,10'

Divergence

Figure 12. Influence of the gap width on the critical velocity (lateral, Case 2)

~60 ri- - - - - - - -- , ~60 , -- -- - -- - - ---,

'-
:so l .550
>- i >-
"~4o l :~40

830 l j
(,)
~30
Qi I I Q)

;20 1~
. :
>20

.~10
.... '"
. ~10
....
"5 0 4 6 8 10 5 0 ' - - -4 - - 6- -8------"
10

gap width [ml ,10' gap width [ml M'

(a) Divergence (b) Flutter

Figure 13: Influence of the gap width on the critical velocity (circumferential,
n = 7, Case 2)
346

In the following, the influence of the gap width on the stability of three type-
s of shell vibrations (axisymmetric, lateral and circumferential) is examined.
Figure ll(a), (b) show the influences of the gap width on the critical velocity
of the flutter and the divergence phenomena of the axisymmetric vibration in
Case 1 respectively. Figure 12 shows the influence of the gap width on the
critical velocity of the divergence phenomenon of the lateral vibration in Case
2. Further, Fig.13(a), (b) show those of the circumferential vibration in Case 2.
The smaller gap width can be seen to make all of vibrations of a thin cylindrical
shell subjected to an axial leakage flow more unstable.

7.6. Influence of Thickness on the Vibration Behavior

Let us consider the influence of the thickness of an elastic shell on the stability
using the dimensions shown in Table 1 except the thickness. The relations

co
.~1O ~
~oLI______________
2 3 4 5 ~ 1 2 3 4 5
th i ckness [ml ,10-3 th i ckness [ml ,10-3

(c) Circumferential
(a) Axisymmetric (b) Lateral (Case 2,
(n = 7), (Case 2,
(Case 1, m = 2) m = 1)
m = 1)

Figure 14. Influence of thickness on the critical velocity (divergence)

between the thickness of an elastic shell and the critical velocity for the 2nd
mode of the axisymmetric vibration in Case 1, the 1st mode of the lateral
vibration in Case 2, and the 1st mode of the circumferential vibration in Case
2, are illustrated in Fig.14(a), (b), (c) respectively. The critical velocities are
almost constant for the axisymmetric vibration because the eigenvalues of the
shell in a vacuum are almost same even if the thickness of a shell is changed.
That is, the axisymmetric vibration can be considered to be affected mainly by
the extensional rigidity, and a little by the flexural rigidity. On the contrary,
the critical velocities for the lateral and the circumferential vibrations can be
considered to become higher due to the increase of the bending stiffness of a
shell.
 347

7.7. Influence of a Fluid (Liquid and Gas)

Figure 15 shows the influence of a sort of fluid on the root locus (argand
diagram). Water is adopted as a liquid, and air is adopted as a gas in this

100
1~ ,-------~~[~m
/ s~
l ,-----~--~
3rd mode 0
3rd mode

80

-:: ~ 1000
~ 60 '"
Co 400
>- ;:- 500
.='"
~
2nd mode 0 2nd roode 0
'"
~
c:
40 300
'"~ 20. 5 ~ 500
E

20 -----
14'-.!. .5 15
1st mode
o
1st mOld!-
U=O m/ s] 200
8 10 0~_ _~____~____~~3OO~~~
0 -10 -5 0 10
-2 o 2
Rea I part Rea I part

(a) Liquid (water) (b) Gas (air)

Figure 15: Influence of fluid (liquid and gas) on the root locus of the circum-
ferential vibration (Case 2)

figure. The critical velocity is found to become much higher when a fluid is
changed from water to air. This can be considered mainly to depend on the
decrease of the fluid force due to the density of air. Note these results are
simply due to the parameter studies, although they are not practical as the
critical velocities exceed the speed of sound.

7.8. Influence of Kinematic Viscosity on Vibration Behavior

Figures 16 and 17 show the root locis (Argand diagrams) of the axisymmetric
vibration in Case 1 and of the circumferential vibration in Case 2 respectively
in order to investigate the influence of the kinematic viscosity of a fluid. Figure
16(a) indicates the case of 1// = 1.004 x 10- 6 m 2 /s corresponding to a water at
the room temperature To = 20C, and Fig.16(b) the case of 1// = 0.295 x 10- 6
m 2 /s corresponding to a water at To = 100C. The reduction of the real
part which means the damping term in the equation of motion is observed
and the critical velocity at which the real part equals zero is hardly changed
when the kinematic viscosity decreases in these figures. From Fig.17 for the
circumferential vibration, the same tendency is also found to be observed as
that for the axisymmetric vibration.
348

200 ~--------~--~
200 , - - - - - - - - - - - , - - - - - - - - ,

150
t:: "t
150

'"c. 2nd mod '"c.

o

..'"'"
~100
~100
c:
'"c:
a~[IIV'" bO

-'"
E E
- 50 1st rood 50
10
~ 10~
flu ter di vergenc
0~==~~~~~==~9.~0~~9.0~6~
-3 -2 -1 -2 -1
Real part Real part

(a) Vf = 1.004 x 1O- 6 [m 2 Is] (b) Vf = 0.295 x 1O- 6 [m 2 /s]

(Room Temperature, To = 20C) (To = 100C)
Figure 16: Influence of the kinematic viscosity on the root locus of the axisym-
metric vibration (Case 1)

100
100
3rd mode 0
0 3rd mode

80 80
+'
"t
~ 60 ~ 60
>-
~
2nd mode 0
'" 2nd mode 0
.~ 40
bO !~ 1'--.. 20.5
20
'"E 14 14.5
20.5

20 14\.! .5 15 20
15
l'tmod i -
1
a~[m/:JI
lstrnode
U"O m/s]
8 10 8 10
0 0
-2 o 2 -2 o 2
Rea I part Rea I part

(a) Vf = 1.004 x 1O- 6 [m 2 Is] (b) Vf = 0.295 x 1O- 6 [m 2 Is]

(Room Temperature, To = 20C) (To = 100C)
Figure 17: Influence of the kinematic viscosity on the root locus of the circum-
ferential vibration (Case 2)

7.9. Influence of Structural Damping on Vibration Behavior

In this subsection, let us consider the influence of the structural damping of an
elastic shell on the vibration behavior. For the above numerical calculations
and discussions, the structural damping of a shell is not taken into considera-
 349
100 100
0 3rd d. 0 d mod.

80 80

t' t'
~80 ~ 60

.,~ .,~ 0
.~ 40 2nd mode o~ .~ 40 2nd mode
bD
20
20. 5 .,
bD 0

i~
111 E
14 ~
I .5
20
U~[:.( IS 20

U~[m/'ll'l~
1 .5
1st mode
8 1st mode 8
10 10
0 0
-2 o 2 -2 o 2
Rea I part Rea I part

(a) Vf = 1.004 x 1O-6[m 2 Is] (b) vf = 0.295 x 1O-6[m 2 Is]

(Room Temperature, To = 20C) (To = 100C)

Figure 18: Root locus (Argand diagram) of the circumferential vibration (Case
2) when the structural damping ratio ( = 0.05 is given in numerical analysis

tion in order to make the unstable phenomena more notable. Figure 18 shows
the root loci (Argand diagrams) of the circumferential vibration in Case 2, in
which the structural damping ratio of a shell ( = 0.05 is given in stead of that
in Fig.17 without structural damping. As the dimensions and calculation con-
ditions except the structural damping in Fig.18 are the same as in Fig.17, the
comparison between both figures shows the influence of the structural damping
of an elastic shell on vibration behavior. The curves of root locus in Fig.18(a)
can be seen to be shifted to the left hand side comparing with Fig.17(a). The
structural damping is found to make the vibration more stable. The same ten-
dency is also observed in the comparison between Fig.18(b) and 17(b) when the
kinematic viscosity vf is changed from 1.004 x 1O-6m 2 Is to 0.295 x 1O-6m 2 Is .

8. CONCLUSIONS

The following conclusions are obtained:

(1) The coupling vibrational analysis method of a thin cylindrical shell sub-
jected to an axial leakage flow is proposed by using the Donnell's shell
theory and the Navier-Stokes equation.
(2) Focusing on the three types of vibrations, that is the axisymmetric, lat-
eral, and circumferential vibrations of a thin cylindrical shell subjected
to an axial leakage flow, the unstable phenomena are clarified by using
the root locus based on the complex eigenvalue analysis.
(3) From numerical parameter studies, the dimensions of a shell are found
350

to have much influence on the determination that which vibration mode

among three types of vibrations of a thin cylindrical shell subjected to
an axial leakage flow goes first into the unstable phenomenon, that is,
divergence or flutter.

(4) The coupled vibration modes between a shell and a flowing fluid are
shown to be considerably changed between before and after the unstable
phenomena. Especially the location of maximum amplitude of the flutter
mode over one period is seen not to be fixed and to move along the axial
direction. A traveling wave can be found to generate axially.

(5) The smaller gap width is found to make all of vibrations of a thin cylin-
drical shell subjected to an axial leakage flow more unstable.

(6) When a fluid is changed from water to air, the critical velocity is found to
be higher mainly due to the density of air. Further, when the kinematic
viscosity of water decreases, the reduction of the real part in the root
locus is observed, but the critical velocity is hardly changed.

(7) When the structural damping increases, the curves of root locus are shift-
ed to the left hand side slightly. The structural damping is found to make
the vibration more stable.

AFFILIATIONS
Katsuhisa Fujita Atsuhiko Shintani
Professor Research Associate
Mechanical Systems Engineering Mechanical Systems Engineering
Graduate School of Engineering Graduate School of Engineering
Osaka Prefecture University Osaka Prefecture University
Japan Japan

9. REFERENCES

[1] Hobson, D. E. (1982). Fluid-elastic instabilities caused by flow in an an-

nulus, Proceedings of BNES 3rd Int. Conf. in Vibration in Nuclear Plant,
440-461.

[2] Paldoussis, M. P. (1966). Dynamics of flexible slender cylinders in axial

flow, part 1. theory, part 2. experiment, Journal of Fluid Mechanics, 26,
717-751.

[3] Chen, S. S., & Wambsganss, M. W . (1972). Parallel-Flow-Induced Vibra-

tion of Fuel Rods, Nuclear Eng. and Design, 18, 253-278.
 351

[4] Fujita, K., & Shintani, A. (2001) . Axial leakage flow-induced vibration of
the elastic rod as the axisymmetric continuous flexible beam, Journal of
Pressure Vessel Technology, ASME, 123, 4,421-428.

[5] Paldoussis, M. P., et. al. (1984). Dynamics and stability of coaxial cylin-
drical shell containing flowing fluid, Journal of Sound and Vibration, 97,
2, 201-235.
[6] Ziada, S., & Biihlman, E. T., Bolleter, U. (1987) . Model tests on shell flut-
ter due to a leakage flow Int. Conf. on Flow-Induced Vibrations Bowness-
on Windermere, England, M2, 521-529.

[7] Fujita, K. , Ito, T. & Baba, K. (1993). Study on the unstable vibration
of cylindrical shells due to leakage flow, Proc. of the Int. Conf. on Aero-
Hydroelasticity, 286-295.
 ACOUSTIC SCATTERING FROM A
COATED ELASTIC SHELL: EXACT VS.
APPROXIMATE THEORY

ANDREW N. NORRIS
Department of Mechanical and Aerospace Engineering
Rutgers University, Piscataway, NJ 08854-8058
norris@rutgers.edu

1. INTRODUCTION
The interaction of acoustic waves with fluid-loaded shell structures is
an ongoing area of interest in underwater acoustics, ship construction
and detection. The elasticity of the structure permits strong surface-
borne wave effects, which can be significant or dominant in large struc-
tures (many wavelengths in size). The significance of surface borne waves
is compounded by low loss factors, enabling them to propagate large
distances on the structure, including multiple circumnavigations, on e.g.
cylindrical structures, which can lead to strong resonance response.
Acoustic scattering from fluid-loaded thin elastic shells can be deter-
mined exactly for two canonical geometries only, i.e., circular cylindri-
cal and spherical shells [1], both using normal mode expansions. Other
methods must be employed for other geometries. However, the large
size and the mutiple wave effects makes exact numerical solution of re-
alistic acoustic scattering problems untenable for the foreseeable future.
Although exact computation remains prohibitive, there is a robust alter-
native technique that is based on asymptotic approximations: the ray
method. Being based upon asymptotic approximations, the ray tech-
nique includes important physical effects such as energy conservation
along rays and phase matching.
The purpose of this paper is two-fold. First to develop and apply
the ray technique to coated shells, comprising stell and an outer softer
coating. The presence of the coating introduces significant damping and
other effects which have not been previously considered. At the same
time, the approximate colated shell theory is bench-marked against an
exact code for cylindrical multi-layered structures, using the full elas-

353
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 353-362.
2003 Kluwer Academic Publishers.
354

ticity theory. This allows us to assess for the first time the accuracy
of the reduced order modeling for a coated structure. In particular, we
consider the application of high-frequency ray and coupling theory to
predicting the scattered response for oblique plane wave incidence from
coated cylindrical shells shell, Fig. 1. We utilize the recently developed

Coating

~-=:~
~
Figure 1. Cylindrical coordinates and coating configuration in a 2-layer shell.

ray theory description of scattering from a shell of arbitrary shape [2].

The key ingredients are membrane waves of longitudinal or shear nature,
as distinct from flexural waves. The formulation is entirely in physical
space and does not require transforms or spectral integrals. The interac-
tion between the fluid (acoustic) wave and the structural waves are via
ray-based coupling coefficients, that are explicitly determined by physi-
cal quantities which describe the thin elastic shell and the surrounding
fluid. Knowledge of the launching and detachment coefficients at any
point on the shell's surface allows one to determine the scattered pres-
sure wave field shed into the surrounding fluid for any smooth convex
shell. Previous work compared the far-field form function predictions of
the asymptotic theory with the normal mode series for uniform spherical
and cylindrical shells [2]. The accuracy was remarkable considering the
simplicity of the approximate model. Here we generalize the theory to
layered shells and include damping in the exterior coating.
While approximate models such as the one discussed are extremely
useful in understanding complex solutions, and more generally in inverse
problems, they need to be properly bench-marked with exact code for
canonical structures. In applying the theory to coated cylindrical shell-
s, it was surprising to learn that although there is large literature on
acoustic scattering from shells and coated shells, no specific work has
 355

appeared on the topic of oblique scattering from a coated clyindrical

shell.
The following Section outlines the asymptotic and benchmark codes,
followed by some numerical comparisons, and discussion.

2. THEORY: ASYMPTOTIC AND EXACT

MODELS
2.1 Ray theory for a coated shell
As in most techniques for analyzing scattering from large structures,
the ray method employs thin shell, reduced order models for the struc-
ture (full elastic modeling ofrealistic structures remains a distant dream).
A set of governing effective shell equations, derived in the Appendix, are
the starting point for our analysis. The key to the ray approach is to
identify wave solutions that exist in the presence of fluid loading, which
are determined by first assuming the displacements as time harmonic
traveling waves (see the Appendix for notation):
(u, v, w) = (U, V, W) ei(kxx+kya</>-wt) . (1)
The system of equations (A.2) then becomes
M,+ WTJ 0, (2)
TJT, + ~ (1 + 2aAk~ + 2aak; + a2f3 2k 4 - k~a2) ~,(3)
where, = (U, V)T, TJ = (ikxa-1(v + aak 2), ik ya- 1(1 + aak2))T,

(4)

kp = w/cp is the plate wavenumber, and k 2 = k~ + k;. The term a2f3 2k 4

can be ignored as being of smaller order, (ka)2(kh)2, in comparison with
the other terms in (1 + 2aAk~ + 2aak~ + a 2f3 2k 4 - k~a2), which are of
order unity. We note that h a, and kh is considered small, as is
normal in thin shell theories.
Progress is made by explicitly assuming a local impedance approxi-
mation that relates the acoustic surface pressure to the normal velocity:
p = -iwZsw. (5)
This enables us to eliminate the pressure and obtain a system of struc-
tural equations. Thus,
(6)
356
Form functi o n : E xsct snd rsy th eo ry

0 .5

0~---7----~'~0----~----2~0----~2~5----~'0

k, a

Figure 2. The far-field scattered amplitude for a uniform cylindrical steel shell in
water. The incident wave is broadside (8 = 0) and the observation is at azimuth
= 60 0 Ray theory is black and the exact result is the blue curve [13].

where the non-dimensional quantity E is

C [
= - .- 2 Zs
2C-(>.k 2 + aky)]
+ Zm + -. 2-1
, (7)
E x
-zwa -zwa
and Zm and Zs are impedances

Zm = -iwph, Zs = Pfcf sec(}o . (8)

Here Pf and cf are the fluid density and sound speed, respectively. The
choice of the fluid surface impedance Zs is motivated by the continuity
condition between the normal velocity and the normal component of the
pressure gradient (i.e., the radial compnenent of the fluid equilibrium
equation). The angle (}o, yet to be defined (it may be complex-valued)
will be specified by the wave type being sought through standard phase
continuity arguments (phase matching). Eliminating W gives two ho-
mogeneous equations for the in-surface components. Setting the deter-
minant to zero and expanding in E, yields a quadratic in k 2 :

0= (k 2 - k;)[~(1 - v)k 2 - k;] + E {[k; - k2 + ~(1 + v)k;](v + aak2)2k;+

[k; - k 2 + ~(1 + v)k;](1 + aak2)2k~ + (1 + v)k; k~(v + aak2)(1 + aak2)} ,

correct to 0(E2). The two roots are, again expanding in E,

(9)
(10)

where k; = 1~l/k~. The pair of waves in (9) and (10) correspond to the
extensional (or quasi-extensional) wave and the shear (or quasi-shear)
 357

wave, respectively. The dependence on the directional cosines nx =

kx/k, ny = ky/k, which indicate different waves speeds as the orientation
of the wave normal (nx, ny) changes. Equations (9) and (10) are the
starting point for an analysis that mirrors that described in [3]. Further
discussion on the asymptotic properties and accuracy of these dispersion
relations is in [4], and [5] describes a general approach for applying
the ray theory to arbitrary smooth structures, with applications. We
skip the details here, but note only that the final prediction for the
scattering amplitude of an incident plane wave can be reduced to an
explicit summation of helical ray terms, permitting rapid computation.
This is in sharp contrast to the exact numerical code discussed next.

Exact far-fie ld amplitude \/S. ka and az imuthal angle

30.---------_ _- - - - - -- -- - - --------.
~ ------~--~=---~--~--

25

20

.. 15
-'"

'0

-
150 ,00 -50 ;0 '00 ,;0 -100 -50

Figure 3. For the same configuration as in Fig. 2, the left and right pictures show
the exact and ray theory predictions for the far-field scattered amplitude as functions
of frequency, ka, and observation angle, 1> [13].

2.2 Benchmark cylinder code

Oblique incidence has been considered for solid cylinders [6] and hol-
low cylindrical shells [7, 8], but not for coated cylindrical shells. In
their paper on oblique incidence on a hollow cylindrical shell, Leon et
al. [8] gave explicit formulae for computing the scattered response, and
also provided comparisons with experimental measurements for an alu-
minum shell in water. The present configuration generalizes that of of
Leon et al. [8] and of Honarvar and Sinclair [9] to the case of a hollow
inhomogeneous cylindrical shell comprising two distinct layers. Oblique
incidence on solid cylinders and hollow cyindrical shells is also discussed
in detail in the monograph by Veksler [10]. Ricks and Schmidt [11] pro-
vided a general formalism for generating the solution of multi-layered
358
Fo rm function: Exact tilnd raytneory Form function: Exact end ray tneory
1 .5 , - - - - , - - - - , - - - - , - - - - , - - - - , - - - - - - - - ,

Nrill vv~
(:\

0 .5

e =20

0L-~~~170-~175-~270-~275-~30 o
o
k, a
10 15

k, a
20
" 30

Figure 4. The backscattered (4) = 00) amplitude for broadside (8 = 00) and oblique
(8 = 20 0 ) incidence on a coated shell. The steel structure is as before, plus a "rubber"
layer of the same thickness. Ray theory is black and the exact result is blue [13].

cylindrical structures under 3-dimensional excitation, using the global

matrix formalism. In this procedure, the full set of continuity equations
at the multiple interfaces are solved simultaneously, resulting in a system
of numerical equations which is large but stable. The alternative propa-
gator matrix approach, on the other hand, results in a smaller system of
equations but can lead to numerical difficulties if not properly addressed.
We adopt the approach of Ricks and Schmidt [11] in order to guarantee
the numerical stability of the resulting system of linear equations. Space
limitations preclude further details of the code.

3. RESULTS AND DISCUSSION

All computations were performed for the same underlying structure,
a cylindrical steel shell in water, thickness to radius h = a/gO. Figure 2
compares predictions for the far-field amplitude, 111, as a function of the
non-dimensional frequency kfa = wa/ef. The form function 1 is defined
by the limiting behavior of cylindrically spreading scattered wave (the
total acoustic pressure is the incident plus the scattered)
ei(kr-wt)
PSC = 1 Kr + o((kr)-1/2), kr --+ 00. (11)
kr
Regarding Fig. 2, we first note that the curves start at a non-zero fre-
quency, as the ray theory does not hold in the static limit . The overall
agreement is excellent, especially at the resonances, which in this case
are due to circumnavigating longitudinal waves (no transverse rays are
 359

excited for broadside incidence () = 0). The curves in Fig. 2 are for a
particular observation direction, while Fig. 3 covers all azimuthal look
directions -1f ::; ::; 1f. The resonances are evident in this picture,
and the overall amplitudes are in very good agreement. A breakdown in
the forward direction II : : : : 1f is expected, due to the geometrical shad-
ow there. The first example with a coating is in Fig. 4. Here and in
Form junction: E Xlll ct IiI n d ra yl n e ory Form fun ction : Exacll'n d ray t h eor y
1.' , - - - , - - - , - - - , - - - , - - - , - - - - - - ,

0.' 0.'

OL-~~~,~
O --,~
'-~'~
O -~,~,-~
JO OL-~~~,~O-~,~
' -~'~
O -~,~,-~
,O
kf a kf a

Figure 5. Backscattered (> = 0) amplitude for broadside (8 = 0) and oblique

(8 = 20) incidence on a coated shell with light damping TJ = 0.1. The left hand curve
is the "raw" ray theory, and the right hand curve uses the dynamic damping (12) .

subsequent figures, the thickness of the outer coating is the same as the
steel, h, and the properties are p = 2,000 kg/m 3 , extensional wave speed
cp = 1,000 mis, and 1/ = 0.28. The coating in Fig. 4 is undamped, and
in general we find that the ray theory does very well, although some
discrepancies are introduced. First, there is a variation in the ampli-
tude between resonances, which can be attributed to the very simple
impedance model used for the specularly reflected field [3, 5]. More sig-
nificant is the appearance of the misplaced resonances, particularly in
the () = 20 case. At this angle of incidence, which exceeds the longitu-
dinal critical angle, the resonances are due entirely to quasi-shear waves.
The shift of the resonances at the higher frequency indicates the effective
shear waves speed is increasingly in error. Damping is introduced in the
coating layer through a complex stiffness, characterized by TJ, such that
the coating Young's modulus is E = Eo(1 + iTJ). A case of light damp-
ing is illustrated in Fig. 5, where the ray theory begins to exhibit large
resonances at higher frequencies. The quasistatic effective shell model
is apparently inadequate in capturing the dynamic loss mechanism of
the structural wave. The right hand curve in Fig. 5 utilizes an ad-hoc
dynamic loss factor, derived by averaging the wave impedances in the
360
Form functkJn: EXict 100 ray thaory Form lunctkJn: EXict 100 ray thaory
l.sr--~-~-~--'-~-~-----, l.S,--,--,--~-~-~-----,

0.5 0.5

0~~~~1~0-~'~S-~2~0-~2~S-~30 oL--~-~-~-~-~~
o 10 15 20 25 30
k, a k, a

Figure 6. The backscattered (</J = 0) amplitude for broadside (8 = 0) and oblique

(8 = 20) incidence on a coated shell with heavy damping of 'fJ = 0.5.

two layers, thus:

N N
cp = (1 + i,)( ~)1/2, with, = Im( L pjCpjhj)/1 L PkCpkhkl. (12)
P j=l k=l

The motivation for this crude approach is the realization that the qua-
sistatic averaging ignores the dual wave nature of the layered structure.
The efficacy is not unreasonable even at relatively large damping, as in
Fig. 6, where again the prediction according to (12) is on the right.
However, the discrepancy between the amplitudes in Fig. 6 points to
the inadequacy of the simple specular reflection approximation used.
In summary, we have demonstrated that the modified Donnell shell
theory combined with ray theory provides a good model for the response
of structures with thick coating and low damping (0.00.1). Comparison-
s with the benchmark solution show reasonable agreement for coatings
with similar thickness to the substrate, and moderate damping. Dis-
crepancies emerge afor coatings with high damping, and these can be
attributed to inadequate modeling of the structural wave damping (P
and S), and increased error in the background magnitudes. The low-
ered amplitude at high k fa suggests using a modified specular reflection,
which should not be too difficult to implement. The discrepancy in the
wave damping presents a larger issue, and will require a new type of ef-
fective medium approximation that goes beyond the quasistatic models
normally used. A good start in this direction would be a reliable theory
for wave damping in a bilayer.
 361

Appendix: Effective Shell Equations

Our objective is a set of effective shell equations governing the composite layered
cylindrical shell comprising layers j = 1,2, ... , N (in practice N = 2) with elastic
properties E j and Vj in layer j. These will serve as the boundary conditions for the
exterior acoustic problem of ultimate interest. Stresses acting in the lh layer are
derived using the Kirchhoff assumption:
E 1 v
ax J 2 [u ,x
-1
- Vj
+ Vj( -U,
a
+ w) - z(w ,xx + -tw,
a
)] ,

E l l
+ -v, + W - z(VjW,xx + 2w,)],
~ [VjU,X (A.I)
-~ a a
E- 1 Z
2(1: Vj) [;;;u , + v,x - 2;;;w ,x].
Here x is the axial coordinate, the azimuth, and Z the thickness coordinate, which is
removed in the averaging process below. The axial, zimuthal and radial displacements
are u, v and w, respectively. A comma suffix indicates the partial derivative by the
subsequent parameter. The radius of the shell, a is taken as fixed for all layers,
consistent with the thin shell approximation. A veraging through the thickness Z
leads to the effective equations of motion,
1
phu,tt = Nx,x + -Nx,,
a
1
phv,tt -N,
a
+ Nx ,x , (A.2)
2 1
phw,tt = Mxx,xx + -Mx,x
a
+ 2M,
a
- N + p,
where p is the averaged mass density, h the total thickness, p is the pressure acting
from the fluid medium, and the forces and moments follow from eq. (A.I). Sub-
stitution into the effective equations yields the desired boundary conditions for an
inhomogeneous thin circular cylindrical shell
I-v I+v v 2 1
u,x x + - 2 2 U, + -2- v ,x + -w,x - QV W,X - 2 U,tt 0,
a a a cp
I+v I-v 1 1 2 1
--u
2a'x
+- -v
2 ,xx
+ -v
a2 '
+ -w
a2 '
Q
- -V w - -Vtt
a ' c~'
0, (A.3)
vII 2 4 >. 1
-u,x
a
+ 2V,
a
+ 2a W + f3 V W - 2-w,xx
a
+ c2W,tt
p

-Q (V u ,x + ~ Vv, + :3 W, )
2 2
p

where V 2 w = W,XX + w,/a 2 , and the various constants are

Cp -
_ (Q.)1/2 .
ph
362

Here Zj is the z-coordinate of the inner boundary of the /h layer, with z = 0 at the
geometric mean of the shell thickness. Note that the Ct and A vanish for symmetric
configurations, that is, these oefficients are absent from the Donnell equations. For
the single layer (N = 1) we have Ct = A = 0 and (32 = h 2 /12, and the equations
reduce to the Donnell equations for uniform thin shells [12] . For the two-layer of
interest here,
1 (E2 El) 1 ( E2V2 Elvl )
Ct = 2C hlh2 1 _ v2 - 1 _ v 2 '
2 1
A = 2C hlh2 1 _ v 2 - 1 _ v 2 '
2 1

(32 = _1_ [ Elhl (h 2 + 3h 2 ) + E2h2 (h 2 + 3h 2 )] . (A.5)

12C 1 - vf 1 2 1_ vi 2 1

References
[1] G. C. Gaunaurd and M. F. Werby, "Acoustic resonance scatter-
ing by submerged elastic shells", Appl. Mech. Rev. , 43, 171-207
(1990).
[2] A. Norris and N. Vasudevan, "Acoustic wave scattering from thin
shell structures," J. Acoust. Soc. Am., 92, 3320-3336 (1992).
[3] A.N. Norris and D. A. Rebinsky, "Acoustic coupling to mem-
brane waves on elastic shells," J. Acoust. Soc. Am. 95, 1809-1829
(1994) .
[4] A.N. Norris and D. A. Rebinsky, "Membrane and flexural waves
on thin shells," ASME J. Vib. Acoust. 116 , 457-467 (1994) .
[5] Y. Yang, A.N. Norris and L. C. Couchman, "Ray tracing over
smooth elastic shells of arbitrary shape," J. Acoust. Soc. Am.
99, 55-64 (1996).
[6] L. Flax, V. K. Varadan and V. V. Varadan, "Scattering of an
obliquely incident acoustic wave by an infinite cylinder", J. A-
coust. Soc. Am. 68, 1832-1835 (1980).
[7] N. D . Veksler, V. M. Korunskii and S. A. Rybak, "Scattering of
an obliquely incident plane wave by a circular cylindrical shell" ,
Sov. Phys. Acoust. 36,5-7 (1990).
[8] F. Leon, F. Lecroq, D. Deculton and G. Maze, "Scattering of an
obliquely incident acoustic wave by an infinite hollow cylindrical
shell", J. Acoust. Soc . Am. 91(3), 1388-1397 (1992).
[9] F. Honarvar and A. N. Sinclair, "Scattering of an obliquely inci-
dent plane wave from a circular clad rod", J. Acoust. Soc. Am.
102(1), 41-48 (1997).
[10] N. D. Veksler, Resonance Acoustic Spectroscopy (Springer-
Verlag, New York, 1993).
[11] D. Ricks and H. Schmidt, "A numerically stable global matrix
method for cylindrically layered shells excited by ring forces", J.
Acoust. Soc. Am. 95(6), 3339-3349 (1994).
[12] M. C. Junger and D. Feit, Sound, Structures, and Their Interac-
tion (MIT, Cambridge, MA, 1986).
[13] A color version of this paper is available in pdf format at
http://rci.rutgers.edu/ ,,-,norris/papers
 A STUDY OF THE VIBRATION
OF FLUID COUPLED COAXIAL CYLINDRICAL
SHELL

ELENA GAVRILOVA

Abstract: Some study of the vibrations of fluid coupled coaxial cylindrical shell is presented. A f1uid-
.tructure interaction .Ystem (FSIS) of two coaxial circular cylindrical shells with the same height partially
filled with an incompressible inviscid fluid to the same depth in the inner shell and in the outer annular gap
between them is under consideration. The outer shell is rigid and the inner is elastic. The shells' bottom is an
absolutely rigid plate. Objectives of the study include: determination of the natural frequencies and forms of
the free vibrations of the considered FSIS; reflection of running waves in the elastic shell of FSIS;
examination of nonlinear vibrations of FSIS; investigation of a nonstationary vibration response of FSIS
under a horizontal excitation; preliminary steps in numerical computer simulation of the above problems. The
method of Bubnov-Galerkin is applied to solve the considered problems. The investigation can be applied as
a base for further study of nonlinear vibrations and considering FSISs where the elastic shell does not have an
ideal shape and possesses some initial deformations.

Key words: fluid-structure interaction, fluid-coupled coaxial cylindrical shell, the Bubnov-Galerkin
method

1. INTRODUCTION

Fluid-structure interaction provides accurate solutions and plays an important role in

many different types of real-world situations and industrial applications such as
biomedical, material processing, automotive, aeronautical and civil engineering.
Circular cylindrical tanks are extensively used for storage of fluids in the industry as
well as in municipal water supply systems. Usually such constructions are under action
of intensive dynamic loads, and as a result it is possible appearing of large cyclic
tensions, stability loss, etc. That is why the study of the dynamic behavior of fluids in
tanks is important in the earthquake-proof design and reliability assessment of fluid
storage tanks. There have been many research papers [5], concerning the vibration
response characteristics of a thin cylindrical shell containing fluid under external
sinusoidal and/or seismic excitations. Liu and Uras [3] propose a method for the

363
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 363-374.
2003 Kluwer Academic Publishers.
364

investigation of the dynamic buckling of a fluid-structure system under a horizontal

excitation and a good agreement has been obtained in comparison to some experimental
results. Okazaki et all [4] present seismic response analysis of a multi-walled coaxial
cylindrical tank under a horizontal excitation using the fInite element method. Kubenko
et all [2] consider the nonstationary response of a cylindrical shell partially fIlled with
fluid in the cases of its free vibrations and under some periodic excitation.

. . "'z
- ~
=

rI
0
~& 11
2R j

2R2
~

Figure 1. Model a/the shells and co-ordinate system.

 365

A fluid-~tructure interaction Ystem (FSIS) of two coaxial circular cylindrical shells

with the same height L that are partially filled with an incompressible inviscid fluid to
the same depth H in the inner shell and in the outer annular gap between both shells is
under consideration - see Figure 1. The outer shell is rigid, the inner is elastic with
thickness h. The shells' bottom is an absolutely rigid plate. The elastic shell's radial
displacement is w, the function of the tensions in its middle surface is (/J and the elastic
shell's material properties are: Young's modulus E, flexural rigidity D and density p;
the fluid mass density is P F and the gravitational acceleration is g. The radius of the
inner shell is R1 and the radius of the outer shell is R2, t denotes time. Let (/J1 is the
velocity potential of the fluid in the inner shell, (/J2 - the velocity potential of the fluid
in the annular gap between the shells, Pf - the fluid-dynamic pressure and PL - the
fluid-static pressure.

2. DETERMINA nON OF TIlE NATURAL FREQUENCIES AND FORMS OF

TIlE FREE VIBRATIONS OF FSIS

The governing equations of incompressible inviscid fluids are:

V 2 (/J1 =0 0:5.r:5.R1,0:5.0:5.27r, O:5.z:5.H, (Ia)

V 2 (/J2=0 R1:5.r:5.R2,0:5.0:5.2n, O:5.z:5.H, (Ib)

where

Both functions w(0, z; t) and (/J(O, z; t) satisfy the system of differential equations [2]:
366

as the last one can be reduced to an equation of eighth order in respect with the function
w [2]:

(2)

where

p(e, z; t) = a(z )p(e, z; t), a(z) = {~

050z50H;
P=Pj +PL;
H <z;

The following boundary conditions are used:

(i) Free surface boundary conditions:

IJ 2 r[J j 1Jr[J j
- - + g - - = O at z =H, 050r50Rj,
1Jt2 &- (3.a)

1J2r[J2 +g1Jr[J2 =0
at z = H, R j 50 r 50 R 2 .
(3.b)
it 2 &-

(ii) Cylindrical symmetry condition: r[J j is finite at r = O.

(iii) Fluid-base plate velocity compatibility:
 367

(iv) Boundary conditions describing the supporting way of the elastic shell-
fixed on one end (bottom) and free on the other (top):

a.v
w(o,z;t} = a;{o,z;t} = 0 at z = 0; (S.a)

02w o3 w
- 2 {0,z;t}=-3 {O,z;t}=O at z=H. (S.b)
Oz Oz

(v) Fluid-shells interface normal velocity compatibility:

o([J2 o([J i __ o([J 2 __ aw

--=0 at r=R2
a- ' a- a- at at r = Ri (6)

The problem is to determine the wave forms of the motion considering the dynamic
fluid-structure interaction of this system.
The shell displacement w(O,z;t} is:

I I I Ih n,m{t}sinn6Zm {z},
00 00 00 00

w(O,z;t}= fJn,m {t}cosn6Z m {z}+ (7)

n=O m=i n=im=i

where

k ) -cos""!!!:"'z
2m (z ) =cosh( ""!!!:"'z ( k ) -coshk
. . m [.
m +cosk (k
smh""!!!:"'z ) -sm....!!!:...z
. (k )] ,
H H smhkm +smkm H H
368

and k m are roots of equation: cos k m cosh k m = -1 (m = 1,2,... ). The potential

functions are obtained, using equations (la)-(lb), (4) and (6). Let ~ and 1] measure the
superelevation of the fluid surface displacements in the inner shell and in the annular
gap between the shells. Using equations (3a)-(3b), it is easy to fmd the expressions [2]:

"f: = _g-] o([J 1 at z = H, 0 ~ r ~ R], 1] = -g

-] -
0t1J2
- at z = H, R] ~ r ~ R2 .
a a
If

00 00 00 00

~= L Ldt (t)cosn8Jn(Pn,F)+ L Ld2 (t)sinn8J n (Pn,F), i (8a)

n=O j=] n=1j=]

(8b)

the following two independent systems of equations are obtained using the shell
equation (2) and the method of Bubnov-Galerkin:
 369

lJn,m{t}+( mJ,m f fJn,m{t}+ mf=l q':t;jm lJn,m jv}

j

mj*m (9a)
OCJ OCJ
+ Lgl}'md7,j {t}+ Lgl}'me7,j (t}=O
j=l j=l

OCJ
d7,j(t}+m~,jd7,j(t)+ L8::,;1 Jl n,m(t)=0, (9b)
m=l

OCJ
e7,j (t)+ m/z, j e7,j (t)+ L 5~,j lJn,m (t) = 0; ( 9c)
m=l

2
J;,m (t)+ (m~,m) f;,m (t)+ L q':t;jm J;m j (t)
OCJ

m =l j

mj*m (lOa)
OCJ OCJ
+ Lgj,md~,j(t)+ Lgj,me;,j(t}=O
j=l j=l

J;,j (t)+ m;,jd;,j (t)+ L5~,j /2 n,m (t) = 0,

00

(lOb)
m=!
370

I
00

e;,j (t) + m~je;,j (t) + ;S::Z,j hn,m (t) = O. (lOe)

m=l

Both systems of equations (9a)-(9c) and (lOa)-(lOc) have the same solutions at one
and the same initial conditions and that is why they can be considered in one and the
same way. Let m=1,2, .. .,N1; j=1,2, ... ,N2 and n=n(fixed). The system of
equations (9a)-(9c) can be rewritten as:

N N
Xi (t)+ wi Xi (t)+ I abxc (t)+ I f3~xc (t) = 0 (i = 1,2, .. ., N) .
c=l c=l

The last equation can be transformed in a normal form. If the roots of its characteristic
equation are ill ,il2 ,.. .,ilN, the following substitution is made:

N
Xi = IZk ,i';k (i = 1,2, .. .,N),
k=l

where Z k,i are the components of the natural vectors of the system, corresponding to
the every one of the roots ili. Then it is performed as

.. 2
';k+ilk';k=O (k=1,2, ... ,N}

The generalized coordinates of the shell and of the fluid free surfaces in the inner
shell and in the annular gap between the shells are correspondingly

_ N _. N _ . N
ft,m = IZk,m';k; d;'} = IZk,N1+j';k; e;'} = IZk,N1+N, +j';k.
k=l k=l k=l
 371

The system of equations (lOa)-(10c) can be transformed in the same way. Then the
function W can be represented as:

NooN
I I
00

W= +I~kW7,k
I1JkW~,k,
n=O k=l n=lk=l

where the natural forms of vibration of the elastic shell at the accepted conditions are:

NJ N/
w7,k (z ,o)= C1 cos nO IZk,mZm (z), w~,k (z,o)= C2 sin nO IZk,mZm (z).
m=l m=l

The equations of the free surface displacements of the fluids in the inner shell and in the
annular gap between the shells are received analogously.

3. RUNNING WAVES IN TIlE ELASTIC SHELL OF FSIS

Further some method will be used that is more accommodated for the research of
nonstationary processes - it is the wave method, appearing as an analogy of the method
ofD' Lambert in the rod theory [2].
The shell displacement function w is expressed in a series of running waves:

I I
00 00

W= Anm (t)cos[nO-lfInm(t)]zm (z),

n=Om=1

where Anm (t) and IfI nm (t) are unknown functions of time t, and they can be
considered as generalized coordinates of the elastic shell. It is easy to see that this
expression can be obtained from equation (2), using that
cos{n 0 -Ifl nm ) = cos nO cos IfI nm + sin nO sin IfI nm and substituting
372

The following expressions of functions ~ and 17 are taken from (8a), (8b):

~ = I IFn,j(tYn(Pn'F)cos(no-o~~~ J
n=Oj=1

_~ ~ (t ( .) J~~n,jR1)
n=O j=l n n,j 1
{ )1 (
(2))
17- L...J L...JHn,j t J n l1n,jr - Y'~ .R )Ynl1n,jr cos nO-On,j .

After long calculations a system of six nonlinear equations with unknown

1)(1) A2).
A nm' If/nm' E n,j' H n,j' Un,j' Un,j IS 0 tame. orne speCIa case IS conslderedm
b . d S . 1 .

details when the elastic shell displacement and the displacements of the fluid free
surfaces in the inner shell and in the annular gap between the shells can be described
with the help of monomial approximation. The frequency equation is obtained. Some
calculations are made with corresponding graphics. It is seen that on the fluid free
surfaces the running wave in the elastic shell generates again running waves in the
fluids with increasing average velocities and decreasing amplitudes.

4. NONLINEAR PROBLEM OF THE FREE VIBRATIONS OF FSIS

A nonlinear problem of the free vibrations of the given FSIS is considered taking
that its nonlinearity is as a result of big gradients of the shell vibrations. The vibrations
of the fluid free surfaces in the inner shell and in the outer annular gap between the
shells are supposed to be corresponding to the linear theory. The fluid boundary
conditions are formulated in the limits of linearized performances. It is made an
analysis of the FSIS nonlinear vibrations which correspond to some of its natural
frequencies . Let this frequency is Q k - it is a root of the characteristic equation
compound for limited number of terms in the expanding of functions w, ~ and 17.
Usually for the solution of nonlinear problems of FSIS dynamics some simplifying
method is used - in it the forms of the "dry" natural vibrations of the shell (without
fluid) are taken as basis functions for the elastic shell with fluid and only one or two
 373

forms are considered. Such method could lead to essential mistakes in many cases of
the determination of the nonlinear dynamic characteristics of FSIS.
The elastic shell radial displacement is approximated by

where ~k (t)' lJdt) are generalized displacements of the elastic shell due to the taken
frequency ilk and w?,k (i = 1,2) are its corresponding natural forms. In such way the
displacement of the elastic shell w is expressed directly through the forms of its
vibrations, which were found taking into attention its interaction with fluids.
Analogously the deformations of the free surfaces of fluids in the inner shell and in the
annular gap between the shells can be presented as

using their vibration forms, which were found taking into attention the interaction
between fluids and elastic shell. Further the following presentations are used:

N j

w= A{t)cos[nB-f//{t)] LZk,mZm{z),
m=l

N,
~ = A{t )cos[nB - f//{t )]2: Zk,N + jJ n (f3n,jr),
j

j=l
374

where A{t) = ~f (t)+ 'If (t), tgf//{t) = '1k ((t)) . Applying the Bubnov-Galerkin method
~k t
to the shell equation (2), a nonlinear system for the determination of the unknown
functions A and f// is obtained. Its consideration is made following [2].

The study of the nonstationary vibrational response of this FSIS under a horizontal
excitation is made using essentially the results from [1].

Department of Mathematics
St. Ivan Rilski University of Mining & Geology
Sofia 1700, Bulgaria
e-mail: elena@netissat.bg

5. REFERENCES

I. Gavrilova, E. G. (1999). Theoretical detennination of the vibration offluid-coupled coaxial cylindrical

shell under horizontal excitation. Proceedings of Euromech Colloquium: Vibro-Acoustics and Flow-
Structure interactions, 19-23 April 1999, Cargese (Corsica, France), 179-186.
2. Kubenko, V. D., Kova1chuk, P. S. & Podchasov, N. P. (1989). Nonlinear Vibrations of Cylindrical
Shells. Kiev: Visha skola.
3. Liu, W. K., & Uras, R. A. (1990). Dynamic buckling of liquid-filled shells under horizontal excitation.
J. of Sound and Vibration, 141(3),389-408.
4. Okazaki K., Nozaki, Y., Tani, J. & Shimizu, N. (1990). Seismic response analysis of a multi-walled
coaxial cylindrical tank under horizontal excitation. ASME PVP-191. Flow-Structure Vibration and
Sloshing - 1990, 55-61.
5. Rammerstorfer, F. G., Scharf, K. & Fisher, F. D. (1990). Storage tanks under earthquake loading. Appl.
Mech. Rev., 43(11), 261-281.
 STABILITY AND NON-LINEAR DYNAMICS OF
CLAMPED CIRCULAR CYLINDRICAL SHELLS
IN CONTACT WITH FLOWING FLUID

K. N. KARAGIOZIS, M. P. PAinouSSIS, E. GRINEVICH,

A.K. MISRA, M. AMABILI

Abstract: This study presents experimental results on the non-linear dynamics and stability
characteristics of a thin-walled clamped-clamped circular cylindrical shell in contact with fluid. It
also discusses theoretical results for simply-supported shells conveying inviscid and incompressible
fluid . The non-linear Donnell shallow shell theory, with structural damping, is used to describe the
large-amplitude shell vibrations. The interaction between the flowing fluid and the shell structure is
formulated with linear potential flow theory. The aim of the experimental study was to gather for
the first time important data points of the critical flow velocity for instability and maximum flexural
displacement, and to analyze the experimental results to validate the theoretical model. The
experimental study involved two set-ups: one containing a clamped-clamped silicone rubber shell
and flowing air in internal and external flow configurations, and the second an aluminum shell and
water as the flowing fluid. The interaction between the shell and the fully developed flow, in both
cases, gives instabilities in the form of divergence at sufficiently high flow velocities. The
experimental results show a softening type nonlinear behaviour with a large hysteresis in the velocity
for the onset and cessation of divergence.

1. INTRODUCTION

The aeroelasticity of shells conveying, or immersed in, fluid is a well researched

subject area; see, e.g., Dowell's (1975) monograph. However, most of the work
has been on compressible and particularly on supersonic flow, reflecting the
great interest there is on the effect of high-speed flow on the outer-skin panels of
aircraft, missiles and aerospace vehicles. However, there are many other
engineering applications in which shells are subjected to subsonic or
incompressible flows, e.g. thermal shields of nuclear reactors, jet pumps, heat
shields in aircraft engines, heat exchangers, storage tanks, and so on.
Furthermore, veins and pulmonary passages in biomechanics are usefully
modeled as shells conveying fluid.
The subsoniclincompressible applications were for a long time thought
to be uninteresting, since panels or shells with such flows were supposed to be
subject to only a mild divergence (buckling). However, the observation of
flutter in such systems (Paidoussis & Denise 1972) provided a new impetus for
such studies. Paidoussis and Denise, in addition to their experiments with
cantilevered and clamped-clamped shells, also provided the first analytical

375
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 375-390.
2003 Kluwer Academic Publishers.
376

model: a linear model using F1iigge's shell equations and assuming the flow to
be inviscid. Additional work was conducted by Weaver & Unny (1973), Shayo
& Ellen (1974), and many others. The effect of viscosity, which can be very
important, was examined by PaYdoussis et al. (1985), Nguyen et al. (1994), see
PaYdoussis (2003a, Chapter 7).
The physical behaviour observed in experiments with shells conveyhlg
fluid, considering here only shells with clamped ends since this is the topic of
this paper, is that at sufficiently high flow velocity the system develops flutter
[see PaYdoussis & Denise (1972)]. Linear theory, on the other hand, predicts
that the system should lose stability by divergence, and then at slightly higher
flow by coupled-mode flutter. The interval between the two is generally small,
especially if the fluid conveyed is air. Hence, it was reasoned that in the
experiments of PaYdoussis & Denise (1972) flutter was entrained by the
divergence, and that was the reason why divergence by itself was not observed.
In the case of external (annular) flow, however, divergence was observed but not
flutter (though the maximum flow velocities available were probably
insufficient for that).
Mindful of the controversy on the existence of post-divergence
coupled-mode flutter in pipes conveying fluid, which was only resolved via
nonlinear theory [see PaYdoussis (2003b)], it is obvious that its existence for the
shell problem would have to be decided by nonlinear theory in this case also -
thereby clarifying also the nonexistence of divergence in the internal flow
experiments.
Geometric nonlinear effects on the dynamics of shells are unusually
strong. Shell motions of the order of the shell thickness are sufficient to induce
important nonlinear behaviour. The reader is referred to the review by Amabili
& PaYdoussis (2003). The effect of geometric nonlinearities on the dynamics of
shells conveying fluid has been studied in a series of papers by Amabili et al.
[see Amabili et al. (1999)] developing a nonlinear analytical model for shells
with simply supported ends, using the Donnell nonlinear shallow shell equations
and the PaYdoussis & Denise linear inviscid fluid-structure interaction model.
Various variants of this analysis consider (i) different modal expansions,
showing that axisymmetric modes are essential for capturing the dynamics
properly, (ii) the contribution of the companion modes, etc. The most
significant findings of this research are two: (a) the system loses stability by a
strongly subcritical pitchfork bifurcation, which means that, given enough
perturbation, divergence may occur at much lower flows than the linear
threshold; (b) only static solutions are found to exist beyond the first bifurcation,
some stable and some unstable, but no oscillatory solutions are predicted. This
 377

would suggest that the observed flutter could well be a dynamic divergence.
Similar results were obtained with external flow by Amabili et al. (2001).
The foregoing provide the motivation for the experiments to be
described: namely to investigate whether (a) and (b) are true - for shells that
have clamped ends, for experimental convenience. In this, it should be
mentioned that the effect of fluid viscosity, not taken into account, could change
the theoretical findings, e.g. resulting in coupled-mode flutter or even making
the system lose stability by flutter - see, e.g., Paldoussis (2003a); however the
difference between clamped and simply supported ends is expected to affect the
dynamics only quantitatively.

2. THEORETICAL MODEL AND TYPICAL RESULTS

A complete theoretical analysis of nonlinear vibrations of empty, fluid-filled and

fluid-conveying simply-supported circular cylindrical shells has been presented
in a series of papers by Amabili, Pellicano and Paldoussis (1999). In that
analysis, the nonlinear Donnell shallow-shell theory was used to defme the
structural behaviour of the shell. A cylindrical coordinate system (O;x,r,O) is
chosen, with the origin 0 located in the centre at one end of the shell. The
displacements of the middle surface of the shell are denoted by u, v, w in the
axial, circumferential and radial direction, respectively. The equation of motion
for large-amplitude transverse vibrations of a circular cylindrical shell is given
by
1 a2 F(&F a w aF aw &F &w)
I
DV w+chw+ phw= f +q+--+- - - - 2 - - - - + - -
4
2 2 2
(1)
Rax R2 ao ax axao axao ax ao
2 2 2 2 2

where D = Eh/[12(1- v 2 )] is the flexural rigidity, E is Young's modulus, v

Poisson's ratio, h the shell thickness, R the mean shell radius, p the mass density
of the shell, c (kg m3 s- 1 ) the damping coefficient, and f and q are the radial
pressures applied to the surface of the shell due to external forces and the fluid,
respectively. Specifically, the term q involves the pressures exerted by both the
internal and external fluid on the shell surface. The radial deformation w is
positive inward, with w= (awjat), and w= 2 (a w/at
2 ). The in-plane stress
function F is given by the following expression:

1 V4F _ 1 a w 1 [( a w)2 a wa 2w]

2 2 2
(2)
Eh - - R ax2 + Ii! axao - ax ao 2 2
378

where V4 = [a 2 /ax 2 +a2 /(R 2 a0 2 )f in both (1) and (2). For the solution of the
problem, the flexural deformation w was expanded in the form
~ ~
w(x,O,t) = I[ A",.n (t)cos( nO) + Bm,n (t)sin( nO) ] sin ( Amx) + IA""o (t)sin( Amx) (3)
m=l m=l

where for each mode, n is the number of nodal diameters and m is the number of
longitudinal half-waves, Am = mTC/ L, and t is the time; Am,n (t), Bmn(t) , and
Am ,o (t) are unknown functions of time. This expansion for w involves the
driven mode and its companion, along with additional axisymmetric modes.
Neglecting the prestress due to fluid weight and restricting the fluid
volume to be delimited by the length of the shell, the fluid motion, described by
the velocity potential <D, satisfies the Laplace equation. For small perturbations
of the flow and assuming a steady mean flow velocity, the perturbation pressure
is given by P=PF(a<D/at+ua<D/ax), where PF is the density of the fluid.
Defining w, <D and P as periodic functions with main period 2L, one can
employ the method of separation of variables to solve for the potential <D and
subsequently for the perturbation pressure of the flow. Assuming no cavitation
at the fluid-shell interface and zero pressure perturbation at both ends of the
shell, the solution for the perturbation pressure on the shell is given by
L In ( mTC R/ L ) ( a a )2
Pr=R =PF mTC I: (mTCR/ L) at +u ax w (4)

where In ( mTC R/ L) is the modified Bessel function of the first kind and order n,
and I: = (mTC R/ L) is its derivative with respect to the argument. In the
aforementioned theoretical analysis of Amabili et al. the flexural mode was
expanded for N j = 2 and N2 = 3, giving a model with a total of seven degrees
of freedom (dot). Subsequent work included additional modes for the flexural
deformation, allowing comparisons to the simpler 7 dof. The Galerkin method
was used to discretize the equation of motion, and the seven ordinary differential
equations were solved for their unknown time functions, Am,n (t), etc., using an
arc-length continuation numerical scheme embedded in the AUTO software
(Doedel et al. 1998).
 379

l
~',
20

. .... ....
10

.
""..( 0
- ......... 2 .. _ _ ..

...

-10

-20

IO

.......... .. .. ...
-5
0

"",-
..... ... -"'"
""~1
.\0

0 2 3 4 6
V

Figure]: Theoretical results for simply-supported shells conveyingfluid.

Amplitude of non-oscillatory solutions versus the non-dimensional flow velocity; fluid
model ofPardoussis & Denise (viscous damping is 0.0]). Amplitude of the first and
second longitudinal modes: AI,nlh and A2,nlh ( ___ , stable branches;,
unstable branches).

A typical result for a simply-supported shell conveying inviscid,

incompressible fluid is shown in Figure 1. As seen in the figure, the system
loses stability by a strongly subcritical pitchfork bifurcation. Indeed, the range
of the nondimensional velocity for which the system might diverge, given
enough disturbance, extends over 1.31<V<3.33. At V=1.91, there appears a
second stable solution (divergence) associated with the second longitudinal
mode (branch 2 in the figure). After this value of V, the system could easily
undergo jumps, as a consequence of the coexistence of four stable solutions, as
well as some unstable ones. It is significant that after the threshold of the first
divergence, all stable solution branches correspond to static solutions; no post-
critical flutter is predicted. This suggests that the observed flutter in the
Paidoussis & Denise (1972) experiments may well have been a dynamic
divergence - see Section 4.2 and 4.3. The same qualitative features seen here
380

for the simply supported shell, namely subcritical divergence and no post-
divergence flutter, are expected also to arise for clamped shells.
So far, the model has been compared with experiments for simply
supported shells, empty or filled with stationary fluid. Agreement was found to
be good. So, this makes this model a good candidate for a modified model,
which can be used for clamped-clamped shells, allowing comparison with the
experiments described here. Preliminary results for the non-linear dynamics of
clamped shells with point-force excitation showed softening behaviour for the
frequency-amplitude curve.

3. EXPERIMENTAL APPARATUS

The experiments were conducted in two different experimental apparatuses.

The first involved elastomer shells subjected to air-flow, either internal or
external, and the second involved aluminum shells with internal water flow.
Figure 2 shows variants of the first apparatus used (a) for annular and
(b) for internal flow. The lower part is a wooden chamber providing a large area
contraction; this, together, with screens and a honeycomb improve the
uniformity of the flow . The air is supplied, at the bottom of the chamber, from a
bank of compressors interconnected to a surge tank, providing air at about 690
kPa (100 psi). Depending on the application, the pressure of the air entering the
shell varied from 476 to 627 kPa.
For the external flow experiments, Figure 2(a), a long solid aluminum
cylinder with an ogival upstream end provides the lower support for the shell. A
similar cylinder provides the upper support, as shown. The air flow is bounded
on the outside by a Plexiglas tube (thus creating an annular flow). Different
inside diameters of that tube permit testing with different annular gaps, ranging
from 10.5 to 22.5 mm, while the shell mean radius is 26.25 mm. Also,
connecting the inner air volume of the shell to different points of the flowing
stream allows experiments in which the mean differential pressure across the
shell is varied.
Figure 2(b) shows the set-up used for internal air-flow experiments.
The shell was initially 100.4 mm long. It was progressively shortened to
examine the effect of reducing the length-to-radius ratio on the dynamics. To
ensure that a high enough flow velocity could be achieved for instability, an
additional solid aluminum cylinder, with a radius of 11.5 mm, could be installed
within the shell, supported on the upper part of the Plexiglas tube.
The elastomer shells used in these experiments were made of silastic
(RTV silicone rubber), cast by ourselves in a special mould; see Paldoussis
(1998, Appendix D). For the shells used in these experiments, the average
 381

values of the material properties are: E = 2.82 X 105 N m-2 ,

Ps = 1.16 X 103 kg m-3 , v=0.47, and .u=0.0195 , where E is Young's modulus,
Ps is the mass density of the shell, v is Poisson's ratio and .u is a hysteretic
damping coefficient.
Air out

(a)

AiTin

(b)

Figure 2. (a) Set-up for annular and (b) internal air-jlow experiments.

The second apparatus involves a modification of the water tunnel

shown in Figure 3(a); the modified test-section is shown in Figure 3(b). The
shell is mounted in a specially designed contraction of the test-section; the test-
section is made of Plexiglas, allowing visual observation. The space between
the outer surface of the shell and the Plexiglas test-section is filled with
382

quiescent water, which could be pressurized via an external pressure line. The
water tunnel provides water velocities up to 6 mls at the entrance of the
Plexiglas test-section. The water is de aerated on a continuous basis via a special
deaeration system using partial vacuum. As seen in Figure 3(a), a honeycomb,
screens and an area-reduction ensure proper mixing and uniformity of the flow.
The following material properties pertain for the aluminum shells:
E=70xlO'Nm-2 , p s =2.7xl0 3 kgm- 3 , v=0.33; where, E is Young's
modulus, Ps is the density of the material, and v is Poisson's ratio of the shell.
The flow velocity in the air-flow experiments was measured with an
orifice plate upstream of the apparatus, and then corrections were made for the
static pressure, hence for the density, at the location of the shell as compared to
the orifice plate. The flow velocity in the water tunnel was measured by Venturi
meters, and the flow velocity in the shell was calculated by continuity.
In all experiments it was necessary to monitor for the flow velocity and
the amplitude of the shell static or dynamic deformation. In all experiments,
divergence (buckling) developed abruptly, and hence it was necessary to capture
its development carefully. For this, two analog non-contacting laser sensors and
an analog video camera were used. The first laser sensor (Matsushita ANR
1282) has a resolution of 4/13/40 ~m. The second laser sensor was a Micro-
Epsilon optoNCDT 1400-200 system with a static resolution of 40 ~m and a
dynamic resolution of 200 ~m at 1 kHz. In the air-flow experiments the first
sensor was used, while in the water-flow experiments both sensors were needed.
 383

To de-aerator _
Vent To drain
+t +
Heat exchanger
t Water supply

t
Water inlet
ff
Q.jd,7 j
Turning vanes
to tunnel via
heat exchanger

Pump
Test section ----l

(a)
From
de-aerator

Plexiglass
cylinder

Quiescent
water _ _-...lJ-LL'-"t-_+-"'''IH
Pressurization
line

(b)

Figure3. Water tunnel experiments. (a) Water tunnel; (b) test-section and aluminum
shell.
384

In all cases, it was attempted to train the laser beam on a point of the
shell displaying maximum amplitude. The video camera was also set up to
record deformation of the shell on VHS tape as buckling developed.

4. CLAMPED-CLAMPED RESULTS

4.1. Air-flow experiments with annular flow

This section presents experimental results for the dynamical behaviour of the
shell system in axial external (annular) flow. The shell dimensions used in these
experiments are given in Table 1.

Table 1. Shell dimensions for the external airflow configuration

Shell system Length Diameter Thickness Length/radius

Air-rubber 40 mm 52.5 mm 1.5mm 1.5

Figure 4 shows the results of a typical experiment as a plot of the shell

amplitude versus the flow velocity. The displacement measured is the
maximum amplitude of the shell (or in the neighbourhood of that point). It is
seen that at small flow velocities the shell remains stable with near-zero initial
deformation. When the flow velocity reached a value of 45.2 mls the shell lost
stability by divergence. The shell walls moved inward, and a new stable static
equilibrium was obtained with a total radial displacement of 5.03 mm, which
corresponds of 3 times the shell thickness, h. At this new equilibrium the
circumferential wave-number jumped from zero to four. The flow velocity was
then decreased slowly and more data points were recorded and are shown on the
plot. When the flow velocity reached 40 mis, the shell returned to its original
circular shape of zero wall displacement. This flow-dependent hysteresis shows
that the divergence instability is subcritical; i.e., the nonlinear behaviour is of
the softening type.
 385

5
Shell Amplitude [mm]

0
0 10
0 20 30 40 50

-1

Flow Velocity [m/s]

386

Table 2. Pressurization effect for the air-rubber system with annular flow
(lengthlradius= 1. 5), U c is the critical flow velocity for buckling, URis the
restabilization flow velocity, n is the circumferential wave number, A is the maximum
shell amplitude

Mean inwards Instability Uc UR n A Stability

pressurization (mls) (mls) (mm)
No Divergence 45.2 40.0 4 5.03 Subcritical-
softening
Yes Divergence 22.2 18.9 4 1.74 Subcritical-
softening

4.2. Air-rubber experiments for internal flow

In all internal flow cases the shell system lost stability by divergence leading to
dynamic buckling. The mechanism of dynamic buckling is as follows: (i) the
shell walls collapse inwards because of the divergence, with sufficient amplitude
because of the high flexibility of the shell for the flow to be considerably
constricted; (ii) the resultant build-up of upstream pressure forces the shell to re-
open; (iii) because of the forces leading to divergence, the re-opened shell does
not return to its circular shape, but rather to its antiphase buckled shape; (iv) this
is repeated dynamically, so that the observed alternation of buckled shapes
produces a "breathing" oscillation, indistinguishable from flutter. Because of
the violent nature of the dynamic buckling phenomenon, it was difficult to
obtain exact values for the shell amplitude at buckling and almost impossible to
obtain exact values for the re-stabilization velocity. This phenomenon repeats
itself for as long as the air-flow velocity is kept at values around the critical air-
flow velocity for divergence. Hence, a more gentle approach was used to obtain
some useful data points. The air-flow was kept as low as possible, and strong
flow perturbations were induced downstream of the shell. Because the flow was
smaller, the shell buckled and vibrated less violently, giving some useful
information. For example, for LlR close to 4, in one such case buckling
occurred at 22.6 m1s with n=3. With no perturbation, buckling occurred at about
35.9 m1s.
Further tests indicated that a kind of circumferential traveling wave
might exist during the violent vibration of the shell. The fact that divergence
could be induced at smaller flow velocities than the critical flow suggests that in
this case also, the divergence is subcritical.
 387

4.3. Water-flowlaluminum-shell experiments

In the water flow experiments the aluminum shell used had the dimensions
given in Table 3. The flow velocity (always internal) was increased until
buckling occurred. A representative result of these experiments is shown in
Figure 5.

6 I I I

I~
E5
.. r~
c:
.9 4
~
<2 I /
J
3
0)
"0
Q)
r5i 2
~
I ~..----

o
I I
o 2 4 6 8 10 12 14 16 18

Flow Velocity [mls]

Figure 5: Internal axial flow for water-aluminum system with pressurization.

Figure 5 shows that the shell loses stability by divergence at a critical

flow velocity of 16.1 mls. In this new stable equilibrium the shell has a
circumferential wave number of n=6. The radial displacement of the shell at the
first point of buckling is 5.98 mm, about 44 times the shell thickness. By slowly
decreasing the flow velocity, the shell amplitude is also decreased until the
original shape of n=O has been restored at a flow velocity of 9.09 mls. In
contrast to the rubber shells, the flexural rigidity here is sufficiently large to
result in relatively small amplitude at divergence, hence no appreciable
restriction of the flow. This is believed to be the reason for the non-occurrence
of dynamic divergence.
388

Table 3. Water-aluminum shell system/or internaljlow experiments; shell dimensions

Shell system Length Diameter Thickness Length/radius

Water- 122.5 nun 82.25 nun 0.137 nun 3
aluminum

Once more, the subcritical and softening non-linear response of the

shell is clearly shown in Figure 5 (compare the critical velocity 16.1 m/s to the
re-stabilization velocity of 9.09 m/s of the return point).
The pressurization effect was also studied in these experiments. The
differential pressure (inwards) in the experimental results in Figure 5 was set to
5.86 kPa. Additional experiments were conducted for a range of differential
pressures, and data points for the critical flow velocity, and shell displacement
were recorded. In all cases the shell has displayed a softening type of non-linear
response.

5. CONCLUSIONS

Analysis of the experimental data shows that thin circular cylindrical shells
conveying fluid lose stability by divergence. For shells that are not too flexible
(e.g., the aluminum shells in the water experiments) in internal flow, and also
for the external flow experiments, a clear static divergence was observed. In the
internal flow experiments with rubber shells, the large-amplitude-related
constriction at divergence resulted in a dynamic buckling phenomenon. The
circumferential wave number, n, associated with the observed divergence
depends on physical characteristics, geometry and boundary conditions.
By decreasing the flow velocity, after buckling, it was observed that the
shell system returns to its original stable shape at much smaller velocities
compared to the critical buckling velocity. This is clearly a subcritical non-
linear behaviour of the shell system (softening type nonlinear response) present
for both internal and external (annular) flow. The magnitude of the critical
velocity (velocity value at buckling) is a function of physical properties and
geometrical constraints of the shell, boundary conditions, flow characteristics
and differential pressure between the internal and external fluids; e.g., increasing
the inwards differential pressure on the shell wall, divergence occurs earlier as a
combined effect of fluid structure interaction and pressure difference. Another
important observation is that, other than the dynamic buckling discussed above
(which could be mistaken for flutter), no true flutter was ever observed.
 389

The theoretical study of Amabili et al. has shown that a simply-

supported circular cylindrical shell conveying incompressible inviscid fluid
loses stability by divergence. It was shown that the shell system undergoes a
subcritical instability, characterized by a strong nonlinear softening-type
response. No flutter is predicted. The theoretical results for simply-supported
shells are in excellent qualitative agreement with the experimental results of
clamped shells presented in this study. A new theoretical study has been
undertaken for shells with clamped ends with either internal or external flow, for
direct quantitative comparison with the experiments. Preliminary results from
this theory (frequency-amplitude results for forced vibrations of shells without
flow) are in excellent agreement with experiments.

6. ACKNOWLDGEMENTS

The authors would like to thank NSERC of Canada and FCAR of Quebec for
their financial support. We would like to thank Mr R. Burt of Ball Packaging
Corporation for kindly supplying the aluminum shells.

7. AFFILIATIONS

Karagiozis, Pai'doussis, Grinevich and Misra: Department of Mechanical

Engineering, McGill University, 817 Sherbrooke St. West, Montreal, QC,
Canada H3A 2K6. E-mail: kkarag@po-box.mcgill.ca.mary.fiorilli@mcgill.ca
Amabili: Dipartmento di Ingegneria Industriale, Parco Area delle Scienze
181/A, Universita, di Parma, 43100, Parma, Italy. E-mail: marco@me.unipr.it

8. REFERENCES
Amabili,M. & Pai'doussis,M.P. 2003 . Review of studies on geometrically nonlinear vibrations and
dynamics of circular cylindrical shells and panels, with and without fluid structure interaction.
Applied Mechanics Reviews, 56 (in press).
Amabili,M., Pellicano,P. & Pai'doussis,M.P. 1999. Nonlinear dynamics and stability of circular
cylindrical shells containing flowing fluid . Part I: Stability. Journal of Sound and Vibration, 225 ,
655-699; see also Parts II-IV, published in 1999 and 2000.
Amabili ,M. , Pellicano,F. & Pai'doussis,M.P. 2001. Nonlinear stability of circular cylindrical shells in
annular and unbounded axial flow . Journal of Applied Mechanics, 68, 827-834.
Amabili,M. , Pellicano,F. & Pai'doussis, M.P. 2002. Nonlinear dynamics and stability of circular
cylindrical shells conveying flowing fluid . Computers and Structures, 80, 899-906.
Doedel,EJ., Champneys,A.R., Fairgrieve,T.F., Kuznetsov,Y.A, Sandstede,B. & Wand,X. 1998.
AUTO 97: Continuation and Bifurcation Software for Ordinary Differential Equations (with
HomCont). Concordia University, Montreal, Canada.
Dowell, E.H. 1975. Aeroelasticity of Plates and Shells . Leyden: Noordhoff International
390

Nguyen,V.8., Paldoussis,M.P. & Misra, A.K. 1994. A CFD-based model for the study of the
stability of cantilevered coaxial cylindrical shells conveying viscous fluid. Journal of Sound and
Vibration, 176, 105-125.
Paldoussis,M.P. 1998. Fluid-Structure Interaction: Slender Structures and Axial Flow, Vol. 1.
London: Academic Press.
Paldoussis,M.P. 2003a. Fluid Structure Interaction: Slender Structures and Axial Flow, Vol. 2.
Oxford: Elsevier Science.
Paldoussis,M.P. 2003b. Some quandaries and paradoxes in fluid-structure interactions with axial
flow. In Proceedings IUTAM Symposium on Fluid-Structure Interaction, New Brunswick, NJ,
U.S.A.
Paldoussis,M .P. & Denise,J.P. 1972. Flutter of thin cylindrical shells conveying fluid. Journal of
Sound and Vibration, 20, 9-26.
Pa'idoussis,M.P., Misra,A.K. & Chan,S.P. 1985. Dynamics and stability of coaxial cylindrical shells
conveying viscous fluid. Journal of Applied Mechanics, 52, 389-396.
Shayo,L.K. & Ellen,C.H. 1974. The stability of finite length circular cross-section pipes conveying
inviscid fluid. Journal of Sound and Vibration, 37, 535-545.
Weaver, D.S. & Unny, T.E. 1973. On the dynamic stability of fluid-conveying pipes. Journal of
Applied Mechanics, 40,48-52.
 FORCED DYNAMIC UPLIFT OF FLOATING
PLATES

KEVIN DEMPSEY AND IRENA VASILEVA

Department of Mathematics and Computer Science
Clarkson University, Potsdam, New York 13699-5815
demps ey@clarkson. edu, vasilevi @clarkson.edu

Abstract: A thin plate of infinite extent floating on a perfect fluid of finite depth is
subjected to a central concentrated dynamic uplift force. For a specified
upward displacement time-history at the center of the plate, the time-
history of the uplift force potential is the solution of a Volterra integral
equation of the first kind. A model problem is used to shed light on the
ill-posed nature of the forced dynamic uplift problem and two canonical
problems are discussed.

Keywords: Volterra integral equation, ill-posed.

1. INTRODUCTION
When a floating plate is subjected to forced uplift, the dynamics of
the plate and fluid are intertwined in a fully-coupled manner. A fully
nonlinear analysis requires large-displacement theory for the plate and
fully viscous theory for the fluid. Because of the fully coupled nature of
the problem, and the inherent level of nonlinearity, an analytical formu-
lation is unavoidably followed by numerical computation. A significant
component of the latter involves code validation. That is where the
present study comes in. In this paper, a linearized mathematical model
of the problem is analyzed. The plate is taken to be thin and linearly
elastic, the fluid is taken to be incompressible, irrotational and nonvis-
cous, and the upward displacement of the plate-fluid interface is assumed
to be small. Though only weakly coupled, the linearized model identifies
the nondimensional parameters of the problem and provides useful in-
formation for both experimentalists and numerical simulators alike. The
linearized model is used to study two canonical floating plate problems.

391
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 391-400.
2003 Kluwer Academic Publishers.
392

In the linearized model, a characteristic length scale that emerges is

= (~) 1/4, (1)

where D is the flexural rigidity of the plate, {! is the density of the fluid,
and 9 is the acceleration due to gravity. The nondimensional parameters
of the problem are Hand /-l, where H is the depth of the fluid and
ph
/-l = - (2)
{!

is a plate to fluid weight ratio; p and h are the density and thickness of
the plate, respectively. A schematic of a floating plate with a concentrat-
ed point load at the origin is shown in Figure 1. In this figure, rand z
are dimensionless radial and vertical coordinates; the dimensional equiv-
alents are r and zh. Before uplift, the plate is at rest and is centered at
r = 0 with its mid surface coincident with the z = 0 plane. At nondimen-
sional time T ~ 0, the uplift force is (!gh2 P(T)8(r)jr, with P(O) = 0,
and the uplift displacement of the plate mid surface is hw(r, T). Both the
forcing and response are axisymmetric. True time is t = T v'i79. The
fluid is assumed to be ideal and irrotational so that the fluid pressure at
the plate-fluid interface obeys Bernoulli's law and the velocity field in
the fluid is the gradient of a dimensionless harmonic potential (r, z, t).
The radially outward and vertically upward velocities in the fluid are

jgo
ur(r, z, t) = h y Por' (3)

In what amounts to the incompressibility constraint of an ideal fluid,

the potential (r, z, t) satisfies the cylindrical Laplace's equation

02 h2 2
oz2 + 2 'V = 0, (4)

in the region r >

- 0 , _1
2 - hH <
1:.. - z < 2' T >
- _1 - O. Here

(5)

and the linearized boundary conditions on (r, z, t) are

(6)

Physically, these boundary conditions ensure that the vertical velocity of

the fluid and plate are equal at the interface z = w - ~ (the w disappears
 393

after linearization) and that there is no flow through the bottom surface
Z = - ~ - ~ H. With the interface pressure incorporated from a linearized
Bernoulli equation, the dimenionless equation governing the dynamics
of the plate is

a2 w a 8(r)
aT + -aT (r, -2' T) + V' w + w = P(T)-
1 4
J-L-
2 . (7)
r

Because (4)-(6) are able to be solved for (r, z, T) in terms of w(r, T)

independently of (7), the problem is weakly coupled. The system of
equations (4)-(7) are solvable "in closed form" using the zeroth-order
Hankel transform (see Sneddon, 1995). From hereon in,

(8)
denotes the (dimensionless) uplift displacement profile at the center of
the plate. If the central uplift displacement hro(T) is specified, which
is the case considered here, the nondimensional uplift force potential
'f/( T) = pI (T) is the solution of a Volterra integral equation of the first
kind and of convolution type:

J
T

ro(T) = k(T - u)'f/(u) du, ro(O) = ro/(O) = 0. (9)

o
It follows from Dempsey and Zhao, 1993 that

k( ) = ~ -
T 4 Joroo X 1COS(f3T)
+ x4 x,
d k(O) = 0, (10)

where

1 +x4
f3(x) = 1 (J-L > O,H > 0). (11)
J-L + xtanh(Hx)
When J-L = 0, which corresponds to assuming that the density of the
plate is negligible, f3(x) '" x 5/ 2 as x -+ 00. In this paper, the ill-posed
nature of the problem is elucidated by taking, for simplicity, f3(x) = x 2
as a model.
It turns out that many of the practical situations surveyed reduce to
one of two canonical problems. In one, the upward velocity is constan-
t, and, in the other, the upward acceleration is constant. This paper
will discuss these two canonical problems with a view to determining P
to a known level of accuracy. Accurate numerical results are useful for
benchmark purposes in a fully coupled nonlinear simulation. Accurate
394

numerical results are also useful in another way. Designers and exper-
imentalists need to know how P depends on the parameters J.L and H.
This parameter dependence is not immediately evinced by the integral
equation.

Zhj
I}ghf2 P( t)8 (r) / r
t
h~l====~-::========---+-. rl!
1
HI!

~77777777
Figure 1. An infinite plate of thickness h, floating on a fluid of finite depth H f ~ h,
is subjected to a central concentrated uplift force {]ghf2 P( r)b(r)/r.

2. CANONICAL PROBLEMS
In many practical cases, a useful mathematical model of the forced
uplift of a floating plate is a variant of one of two canonical problems.

2.1 "Constant" Velocity

In the first canonical problem, the center of the plate is forced upwards
at a "constant" velocity so that

W(T) = V ( T -
1
- :
-aT) ' (12)

(13)

The exponential term is required in order that w'(O) = 0, otherwise

w' (T) ~ V. In this paper, a = 50, a value based on experimantal data.
 395

2.2 Constant Acceleration

In the second canonical problem, the center of the plate is forced
upwards at a constant acceleration so that

(14)

w'(r) = Ar, w"(r) = A. (15)

3. MODEL PROBLEM
When j), = 0, j3 in (11) is asymptotic to x 5 / 2 as x -+ 00. With j3 set
equal to X 5/ 2 , (10) gives (Glasser, 2003)

7r 1 95 r 8 1 11031012"7109 10 )
k(r) = 4" - 327r7/2 G 5 :13 16777216 0 ~ !. l... !. !. l
(
~ ~ !. ~ . 1
10 4 10 2 2 10 4 10 8 8 8 8
(16)
where G is the Meijer G function (Prudnikov et al., 1990). Unfortunately
this result is numerically more useful than it is analytically. In order to
arrive at a model problem that is analytically tractable but still capable
of revealing the ill-posed character of the problem, j3 is set equal to x 2 ,
in which case (10) gives

k(r) = ~ - ~ (Xl cos(xr) dx =~ (1 _ e-1T1) . (17)

4 2 Jo 1 + x2 4
Equations (9) and (17) constitute the model problem.

3.1 Ill-Posedness
An analysis of the Volterra integral equation of the first kind (9) with
the kernel k(r) defined by (17) points to the model problem being mildly
ill-posed. This is because the singular values {j),n}~=l (see Kress, 1999)
associated with the integral operator in (9) display a O(I/n2) decay as
n -+ 00.
Unlike the actual floating plate problem (9)- (11), the model problem
(9),(17) is able to be solved in closed form. For the model problem, tp(r)
and P(r) are given by

tp(r) = -7r4 (w' + w ") , P(r) = ~ (w+w'). (18)

7r
From these equations one is able to see that the ill-posedness is due to
differentiation. For a given central uplift displacement profile hw(r), let
396

o< 8 1, and let

W~(T) = W(T) + 8Tsin n8T ~ W(T). (19)

Then w~ (0) = (W~)' (0) = 0, and

(20)

(21)

For an arbitrarily small amplitude but high frequency change in the

uplift displacement profile, the amplitude of the same high frequency
change in the uplift forcing can be made arbitrarily large, simply by
changing n. Thus a small change in W(T) does not necessarily mean a
small change in P(T). Since the forcing P does not depend continuously
on the data w , the problem is ill-posed (see Zauderer, 1998, p.138).

3.2 Canonical Problems

For the Constant Velocity problem (see 2.1), the model problem so-
lutions for cp and P are, from (18),

(22)

(23)

On the other hand, for the Constant Acceleration problem (see 2.2),
the model problem solutions are

(24)

3.3 Numerical Results

If the dimensionless time interval 0 ::; T ::; 1 is discretized according to
Ti = i/n, i = 0,1, ... ,n, and if the midpoint rule is used to approximate
(9) sequentially at each value of Ti in this uniform mesh, the midpoint
values of cp are able to be determined directly from
 397

whereupon the mesh point values of P follow from

(26)

where i = 1,2, ... ,n for both equations.

The numerical results obtained for the model problem from (17), (25)
and (26) for the canonical uplift displacement profiles in (12) and (14)
are shown in Figures 2 and 3, respectively. In Figure 2, for the Constant
Velocity case, the jagged lines are the numerical solutions while the s-
mooth lines are the exact solutions given by (22) and (23). In Figure 3,
for the Constant Acceleration case, the numerical and exact solutions
for the model problem are virtually coincident. Here the exact solutions
are listed in (24). The numerical solutions were obtained using Maple
8, with the default value of Digits, namely, 10. In the Constant Ve-
locity case, the maximum absolute errors in cp and P on the interval
[0,1] are 9.738 (at T = 0.04) and 0.127 (at T = 0.02), respectively. The
corresponding errors for the Constant Acceleration case are 0.00636 and
0.000127 (both at T = 0.02). It is worth noting that the model problem
is not solvable for w( T) = T, since the initial velocity has to be zero,
and that the Constant Velocity w profile is very close to T whereas the
Constant Acceleration w profile is considerably different from T. Ac-
cordingly, there is considerably more numerical noise in the Constant
Velocity case than in the Constant Acceleration case. Whereas the mid-
point rule yields an acceptable solution in the Constant Acceleration
case, the noise in the Constant Velocity case is excessive and calls for
regularization. Notably, because P is obtained from cp by integration,
the noise in P is less than in cp.

3.3.1 Tikhonov Regularization. A common method of regulariza-

tion is Tikhonov regularization (see Engl et al., 2000). If the uniform
midpoint rule (25) is written in the form

Tcp = w, (27)

where the n x n matrix T is lower triangular and Toeplitz, a rudimentary

form of Tikhonov regularization involves solving for cp from

(28)

for a suitably determined small value of a. Figures 4 and 5 show the

numerical results obtained for the model problem using Tikhonov regu-
larization (28) and (26) with a = 2 x 10- 8 . In Figure 4, for the Constant
398

Velocity case, the maximum absolute errors in 'P and Pare 2.333 (at
T = 0.02) and 0.0458 (at T = 0.04), respectively. In Figure 5, for the
Constant Acceleration case, the values of'P at the end of the interval
diverge. Tikhonov regularization is known to do this. On the inter-
val [0,0.8], the maximum absolute errors in 'P and P for the Constant
Acceleration case are 0.00418 and 0.0000837 (both at T = 0.02), respec-
tively. For both canonical problems, Tikhonov regularization improves
the computational accuracy away from the end of the interval.

4. CONCLUSIONS
Based on the model problem (9),(17), the axisymmetric forced dynam-
ic uplift of a floating plate (9)-(11) is mildly ill-posed. For computational
accuracy, the numerical results for the model problem point to the need
for some form of regularization. The "best" regularization technique
has yet to be determined. A major breakthrough in this work was the
identification of two canonical problems: Constant Velocity uplift and
Constant Acceleration uplift. Regularization improves the computation-
al accuracy for both problems. From an engineering viewpoint, the two
canonical problems are markedly different in the severity of the uplift
forcing that is required. As might be expected, the Constant Velocity
case requires a very rapid initial increase in the uplift force, whereas the
Constant Acceleration case requires a more gradual increase.

References
Dempsey, J. P. and Zhao, Z. G. (1993). Elastohydrodynamic response of an ice sheet
to forced sub-surface uplift. J. Mech. Phys. Solids, 41(3):487-506.
Engl, H. W., Hanke, M., and Neubauer, A. (2000). Regularization of Inverse Problems.
Kluwer Academic Publishers.
Glasser, L. M. (2003). Personal communication.
Kress, R. (1999). Linear Integral Equations. Applied Mathematical Sciences, 82. Springer-
Verlag, New York, 2nd edition.
Maple 8. Waterloo Maple Inc. http://www.maplesoft.com
Prudnikov, A. P., Brychkov, Y., and Marichev, O. (1990). Integrals and Series, Volume
3: More Special Functions. Gordon and Breach Science Publishers.
Sneddon, I. N. (1995) . Fourier Transforms. Dover Publications.
Zauderer, E. (1998). Partial Differential Equations of Applied Mathematics. John
Wiley & Sons, NY, 2nd edition.
 399

r.pIV Plv
50
2.5

40

30

1.5

20

'0

0.5

-'0 0 .2 0.4 0 . 0." , T

Figure 2. Graphs of r.pIV and PIV versus T for the Constant Velocity case using
the uniform midpoint rule (25) and (26) with n = 50.

r.pIA PIA
2.5

1.5

0.5

L---~0~.2----~0.~4----~0.
~--~0~."~--~' T

Figure 3. Graphs of r.pIA and PIA versus T for the Constant Acceleration case using
the uniform midpoint rule (25) and (26) with n = 50.
400

'PIV PIV
50
2.5

40

30

1.5

20

10

0.5
+-~~0~.2====;0.4~===0~.6~==~0~.====~' T

-10 .~--~1 T
L---~0~2----~0.~4--~0~.6~--~0~

Figure 4- Graphs of 'PIV and PIV versus T for the Constant Velocity case using
Tikhonov regularization (28) and (26) with n = 50 and a = 2 x 10- 8 .

PIV

1.8

1.6

1.4

1.2

0 .

0.6

0.4
0.5

0.2

L---~0~.2----~0.74----07.6~--~0~.8~--~' T ~--~~--~0.~4----~0.6~--~0~.~---C1 T

Figure 5. Graphs of 'PIV and PIV versus T for the Constant Acceleration case
using Tikhonov regularization (28) and (26) with n = 50 and a = 2 x 10- 8 .
 LOW-DIMENSIONAL FEEDBACK CONTROL OF
THE VON KARMAN VORTEX STREET AT A
REYNOLDS NUMBER OF 100

STEFAN SIEGEL, KELLY COHEN, THOMAS McLAUGHLIN

US Air Force Academy

Abstract: The effect of feedback flow control on the wake of a circular cylinder at a Reynolds number of 100
is investigated in direct numerical simulation. Our control approach uses a low dimensional model based on
proper orthogonal decomposition (POD). The controller applies linear proportional and differential feedback
to the estimate of the first POD mode. Actuation is implemented by displacing the cylinder normal to the
flow. The closed loop feedback simulations explore the effect of both fixed phase and variable phase
feedback on the wake. While fixed phase feedback is effective in reducing drag and unsteady lift, it fails to
stabilize once the low drag state has been reached. Variable phase feedback, however, achieves the same drag
and unsteady lift reductions while being able to stabilize the flow in the low drag state. In the low drag state,
the near wake is entirely steady, while the far wake exhibits vortex shedding at a reduced intensity. We
achieved a drag reduction of close to 90% ofthe vortex-induced drag, and also lowered the unsteady lift force
by an order of magnitude.

Key words: Feedback Flow Control, Circular Cylinder, Wake, Vortex Shedding, Drag Reduction

1. INTRODUCTION

Two-dimensional bluff body wakes have been investigated for quite some time. In a
two-dimensional cylinder wake, self-excited oscillations in the form of periodic
shedding of vortices are observed above a critical Reynolds number of approximately
47 (Williamson (1996)). This behavior is referred to as the von Karman Vortex Street.
According to Williamson (1996), the regime of laminar vortex shedding extends to a
Reynolds number of approximately 180, before three-dimensional instabilities occur.
This is the Reynolds number regime that we target in this investigation. However, the
Karman vortex street as the fundamental feature of this type of wake flow is sustained
to very large Reynolds numbers (on the order of millions). Therefore the lessons
learned at low Reynolds numbers will still be applicable to applications of practical
interest at much higher Reynolds numbers. Conversely, it would be impossible to

401
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 401-412.
2003 Kluwer Academic Publishers.
402

control the flow at high Reynolds numbers without being able to successfully do so at
low Reynolds numbers.
The non-linear oscillations of the vortex street lead to some undesirable effects
associated with unsteady pressures such as fluid-structure interactions
(Roussopulos(1993 and lift/drag fluctuations (Park et al. (1994. Examples of real
life applications subject to these are wakes of submarine turrets, ocean drilling rigs, and
flame holders in turbine engines. Also, the vortices themselves greatly increase the drag
of the bluff body, compared to the steady wake that can be observed at lower Reynolds
numbers. Monkewitz (1996) showed that the von Karman Vortex Street is the result of
an absolute, global instability in the near wake of the cylinder. Further downstream the
flow is convectively unstable. This absolute instability is causing the flow to behave as
a self sustained oscillator, with internal positive feedback leading to temporal
amplification of the oscillation by the recirculation region downstream of the cylinder.

2. NUMERICAL METHODS

The goal of this paper is to apply the approach developed by Cohen et al. (2003) to a
full Navier Stokes simulation of the flow field. A schematic of this approach is shown
in Figure 1.

o o

IActuation I ILC_ _----'

Figure 1. Setup of Simulation and Sensors

A limited number of sensors (n > 1) are used to estimate the state of the flow which
is characterized using a low dimensional model. The controller then acts on the flow
state estimates in order to determine the actuator displacement (Figure 1 shows the
overall setup of this experiment).
 403

2.1. CFD Model

For the numerical simulations, Cobalt Solutions' Cobalt solver V.2.02 running on a
Beowulf cluster was used. This code can operate in many different modes using various
turbulence models, however, for the present investigation it was used as a direct Navier
Stokes solver with second order accuracy in time and space. Cobalt operates on
unstructured grids, for all investigations an unstructured two-dimensional grid with
63700 Nodes / 31752 Elements was used. The grid extended from -16.9 cylinder
diameters to 21.1 cylinder diameters in the x (streamwise) direction, and 19.4 cylinder
diameters in y (flow normal) direction.
The Strouhal number obtained from the COBALT CFD model used in this effort is
St = 0 .163 at Re = 100 which compares well with the reported literature (Williamson
(1996.

2.2. POD Modeling and Estimation

Feasible real time estimation and control of the cylinder wake may be effectively
realized by reducing the model complexity of the cylinder wake as described by the
Navier-Stokes equations, using POD techniques. POD, a non-linear model reduction
approach is also referred to in the literature as the Karhunen-Loeve expansion (Holmes
et al. 1996). The desired POD model contains an adequate number of modes to enable
modeling of the temporal and spatial characteristics of the large-scale coherent
structures inherent in the flow.
In this effort, the method of "snapshots" introduced by Sirovich (1987) is employed
to generate the basis functions of the POD spatial modes from the numerical solution of
the Navier-Stokes equations obtained using COBALT. In all 200 snap-shots were used
equally spaced at 0.00735 seconds apart. The time between snapshots is five times the
simulation time step. The snap-shots were taken after ensuring that the cylinder wake
reached steady state. This decision was made in order to be able to estimate the mode
amplitudes based on sensor information, which will yield the U and V component of
velocity. Since the change in mean flow distribution is an important quantity, we chose
the U velocity component over the V velocity component. We found that more than
99.98% of the kinetic energy of the flow lies in the first eight modes, with more than
90% in the first four modes. The POD algorithm was applied to the fluctuating velocity
component in the direction of the flow as described in Equation (1). The decomposition
of this component of the velocity field is as follows:

U(X,y,t)= U(x,y)+u(x,y,t)
(1)
404

where U[m/s] denotes the mean flow velocity and u[m/s] IS the fluctuating
component that may be expanded as:

= L ak (t)~i
n (k)
u(x, y, t) (x, y)
k=l (2)

where ak(t) denotes the time-dependent coefficients having units ofm/s and <I>(x, y)
represent the non-dimensional spatial Eigenfunctions determined from the POD
procedure. Once the spatial POD Eigenfunctions have been derived, the corresponding
time-dependent coefficients ak(t) , or mode amplitudes, need to be calculated. For this,
two different schemes are reported in literature. Most often a Galerkin projection is
used, which involves projecting the spatial Eigenfunctions onto the Navier Stokes
equations. This process involves spatial derivatives of the snapshots, which are,
particularly in the case of experimental data, inherently sensitive to noise. Gillies
(1995) used a simple least squares fit, which we found to be much more effective.

The sensor grid employed for all simulations employs a total of 35 sensors in the
near wake of the cylinder. The main advantage of this sensor grid over others
investigated is in its ability to provide a global estimate of the mode amplitudes that
shows little error compared to using all grid points. This holds true both for the
unforced case as well as the feedback controlled cases. Typical errors of mode
estimates are negligible in phase and less than 5 % in amplitude.

2.3. Controller

The Cobalt CFD solver has the ability to perform rigid body motion of a given grid.
This feature was used to perform both time periodically forced and feedback controlled
simulations with a single actuator. For all investigations, only displacement of the
cylinder in flow normal (y) direction was employed for forcing the flow. Following
earlier developments the control algorithm acts on the estimate of the Mode 1
amplitude only (Cohen et al. (2003)). This design decision was made based on our
earlier investigations controlling a low dimensional model of the flow. For the low
dimensional model, proportional gain applied to Mode 1 only was sufficient to suppress
vortex shedding. Our CFD simulations require a filter to avoid feeding back of small
amounts of noise. Furthermore, we employed a Proportional and Differential (PD)
feedback control strategy (Figure 2).
 405

Sensors

Low Pass PD Mode

Filter Controller Estimator
fc = 4*fn (Mode 1)

Actuator Command Modes Sensor Information

(Displacement) (POD Amplitudes) (U Velocity)
406

3.1. Unforced and open loop results

In a CFD simulation, the flow field is started abruptly at time zero. The flow evolves
from a Stokes-type streamline pattern at the start of the simulation through a steady
wake with two closed recirculation bubbles into the unsteady von KaIllliin Vortex
Street. During this startup, the flow reaches a state of minimum drag. It is worth noting
that the minimum drag does not coincide in time with the steady wake as one might
expect, but rather with a vortex shedding pattern with a very large wavelength. The
total drag in this situation is about 16 % less than in the steady state vortex shedding
situation. Thus one may argue that a feedback control scheme aiming to suppress the
vortex shedding may be able to recover up to this portion of the total drag, at best. We
refer to this portion of the overall drag force as the vortex induced drag, since it is
caused by the vortex shedding in the unsteady wake flow. It is a portion of the pressure
drag. After about 10 shedding periods after the startup of the simulation, the wake
approaches a time periodic vortex shedding state. The mean recirculation zone ends at
x!D = 1.9 in this flow state.
The cylinder wake flow can be forced in an open loop fashion using sinusoidal
displacement of the cylinder with a given amplitude and frequency. Koopman (1967)
investigated the response of the flow to this type of forcing in a wind tunnel
experiment. He found a region around the natural vortex shedding frequency where he
could achieve "lock-in", which is characterized by the wake responding to the forcing
by establishing a fixed phase relationship with respect to the forcing. The frequency
band around the natural vortex shedding frequency for which lock-in may be achieved
is amplitude dependent. In general, the larger the amplitude, the larger the frequency
band for which lock-in is possible. However, a minimum threshold amplitude exists
below which the flow will no longer respond to the forcing any more. In Koopman' s
experiment, this amplitude was at 10% peak displacement of the cylinder.
We resampled the lock-in region in the CFD simulation at select amplitude and
frequency pairs. The simulations activated the forcing always at the same time, 3.3
seconds after the start of the simulation, which resulted in the forcing being 180 degrees
out of phase with the vortex shedding. The flow field goes through a transient phase
before lock-in is achieved after a certain number of shedding cycles. We refer to the
time during which the flow adjusts to the forcing as the settling time. A scan through
different forcing amplitudes was performed at the natural shedding frequency with
amplitudes ranging from 1 to 30 % of the cylinder diameter. The settling times
observed in these cases are shown in Figure 3. While the settling times are roughly
constant down to a forcing amplitude of 5%, for smaller amplitudes a drastic increase in
settling times can be observed. This manifests the behavior observed by Koopman
around 10% forcing amplitude, albeit shifted towards somewhat smaller amplitudes.
There are two possible explanations for this. Koopman used spanwise coherence as an
 407

indicator for lock-in, which may occur at larger amplitudes than the local lock-in
observed in our two-dimensional simulations. Additionally, his experiment was
conducted in a wind tunnel environment which features more mean flow turbulence
than the CFD simulations. This would also tend to increase the amount of forcing
needed to overcome the turbulence and achieve lock-in.

~,----,----,----,----~----,---~----,

30

10

~L---~0~.~----~01----~O.I~6----0~.2----0~~----~O.'--~O.M
NDI-I

Figure 3. Open loop forcing settling time

Shifting the forcing frequency away from the natural shedding frequency yields a
qualitatively different behavior, ultimately yielding a chaotic flow behavior at and
beyond the lock-in limit according to Koopman. We were able to verify this behavior in
the simulation.
The open loop forcing results have important implications for the closed loop
feedback control runs. Since our POD model is based on unforced flow field data, it can
only capture flow behavior that is phenomenologically similar to the unforced wake. In
terms of the lock-in region, this flow behavior is encountered as long as the controller
keeps the flow within the lock-in region. The chaotic behavior at off-natural
frequencies is clearly not modeled in the POD modes. Also, more importantly, if the
displacement of the cylinder becomes smaller than about 5% of the cylinder diameter,
the flow will no longer be responsive to the forcing.
408

3.2. Variable Phase Feedback

During an investigation into different sensor configurations, we used a sensor field with
35 sensors localized between x!D = 0.75 and xlD 1.75. As was later discovered, this
sensor field developed a large estimation error with respect to the phase error of the
Mode 1 estimate, when compared to an estimate based on the entire flow field, as
shown in Figure 4.

..
:r-'
1-1
...

~'~~'~~--~~~~'--~-'~'~~~.~'--~~7~~ .
TmII(a]

Figure 4. Feedback phase during feedback controlled run.

Nonetheless, this phase error led to a stabilization of the wake at a drag reduction of
about 15% with an unsteady lift amplitude reduction of 90%, compared to the unforced
flow field (Figure 5). Inspecting the phase error, one can see that due to the effects of
the local sensor field the phase advance is reduced to almost zero in the steady state
case. This phase advance angle stabilizes the flow field at a low level of vortex
shedding, with the recirculation length extended to x!D = 3.95, or more than twice the
unforced length.
 409

..

.(Ulal---;----;-----;-.------;-------;---~
-00

Figure 5. Lift and Drag using variable phase feedback control. The controller is acitvated at 3.01
s and deactivated at 8 s

i
..,
~ 0

~ 0
.,.
~
.,

Figure 6. Mean flow (top) and RMS velocity distributions (bottom). Left, uncontrolled, right,
controlled case. The cylinder is centered at (0,0) and of diameter 1, flow from left to right.
Negative isocontours are dashed, positive isocontours are solid lines.
410

While in the simulation shown in Figures 3-6 the phase advance was a result of the
sensor placement, the same effect can be achieved using a global sensor field in
combination with a variable phase advance based on the non - fluctuating mean flow
mode. Thus we fmd that a variable gain strategy that adjusts the feedback gains
according to the change in the mean flow achieves better results than a fixed gain
control approach.
The drag and unsteady lift force reduction manifests itself in a change in the mean
flow, as well as the RMS distribution. Figure 5 compares the unforced mean flow and
RMS distributions to those encountered in the stabilized state, between 6 and 8 seconds,
in the feedback controlled run. The recirculation zone length has almost doubled in
length, and the peak in the RMS distribution is shifted from xJD=2.5 to xJD = 5. Also, it
can be seen that the wake up to 3 diameters downstream of the cylinder is entirely
steady.

3.3. Stability Analysis of Feedback Controlled Flow

When applying feedback control, significant changes in the mean flow field occur, as
shown in the previous section. It is therefore of interest to investigate how the stability
characteristics of the mean flow are modified as a result of the mean flow changes.
Linear stability analysis based on numerical solution of the Orr-Sommerfeld equations
using spectral methods (Trefelthen (2000)) was used to analyze these changes.

,r
I
..

Figure 7. Linear stability analysis of unforced and feedback controlled flow fields

Figure 7 compares the maximum growth rate of the unforced flow field at a
Reynolds number of 100 to the steady state feedback controlled flow field (Run 122).
Despite the fact that the near wake fluctuations are suppressed by the feedback as
 411

shown in the previous section, the flow field has become more unstable beyond two
diameters downstream of the cylinder. Comparing the unforced flow to a stable flow
field at a subcritical Reynolds number of 40, one can see that the Karman vortex street
at Re = 100 leads to a more stable flow field beyond xJD = 3.

4. DISCUSSION

We used Proper Orthogonal Decomposition (POD) as a tool to process multiple sensor

signals into a global estimate of the flow state. POD allows for stable global wake state
estimates, enables multi sensor evaluation and eliminates artifacts of local sensing, i.e.
sensing at nodes of the vortex street. It also allows for an accurate state estimate when
the effect of the controller causes major changes both in the mean flow and the rms
amplitudes of the fluctuating velocity components. However, we fmd it necessary to
account for the changes in the mean flow by adding a mean flow mode to the model.
While we used only Mode 1 for closing the feedback loop, all the higher order POD
modes experienced proportional reductions in mode amplitude. This suggests a strong
coupling between all modes, and implies that the existence of the higher order modes is
conditional on the presence of the fundamental modes. This confirms the results of our
previous work (Cohen et al. (2003)).
While feedback control was able to stabilize the near wake of the cylinder, vortex
formation still occurred further downstream. While the reasons for this are not entirely
clear, we suggest several possible causes. The change in the mean flow caused by the
controller lengthens the recirculation zone. This moves the vortex formation location
further downstream and causes a reduction in both drag and rms lift force. While both
of these effects are desired, the downstream shift in vortex formation location causes a
larger spatial separation between the actuation, which remains at the cylinder, and the
oscillations the actuator attempts to cancel. This requires both more actuation input, and
also an adjustment in the actuation phase in order to account for the time a given
disturbance takes to travel from the actuator to the vortex formation location. At the
same time the disturbances caused by the actuator need to travel through a region of the
flow which, while stabilized, is only stabilized within a narrow range of phase angles. If
the far wake requires a phase angle for stabilization that at the same time destabilizes
the near wake, a physicallirnit has been reached in terms of what can be achieved given
the actuator location. This effect may limit the spatial range for which stabilization can
be achieved with the current actuator setup.
Despite all these problems, we were able to suppress the oscillations in the near
wake without actively modifying the mean flow or changing the separation point using
for example momentum injection. Thus this effort shows that the cylinder wake flow
can be improved in terms of drag and unsteady lift by feedback control. For this reason,
412

one would expect the current control approach to be applicable to wake flows with
fixed separation points, like the flow around a D- shaped cylinder. The same cannot be
said for approaches that aim at moving the separation point aft, for example by tripping
the boundary layer or using blowing and suction upstream of the separation point to
delay separation.
Overall, we were able to reduce the effect of vortex shedding on both the unsteady
lift and the vortex induced (pressure) drag by about an order of magnitude.

5. ACKNOWLEDGMENTS

The authors would like to acknowledge funding for this research from the Air Force
Office of Scientific Research, program monitor Dr. Belinda King. We would also like
to acknowledge the fruitful discussions and information exchange with Gilead Tadmor
and Bernd Noack.

6. REFERENCES

Williamson, C.H.K., "Vortex Dynamics in the Cylinder Wake", Ann. Rev. Fluid Mech., 1996,28:477-539
Roussopoulos, K., "Feedback control of vortex shedding at low Reynolds numbers", Journal of Fluid
Mechanics, Vol. 248, 1993, pp. 267-296.
Park, D. S., ladd, D. M., and Hendricks, E. W., "Feedback control of von Karman vortex shedding behind a
cylinder at low Reynolds numbers", Phys. Fluids, Vo. 6, No.7, 1994, pp. 2390-2405.
Monkewitz, P. A., "Modeling of self-excited wake oscillations by amplitude equations", Experimental
Thermal and Fluid Science, Vol. 12, 1996, pp. 175-183.
Koopmann, G., "The Vortex Wakes of Vibrating Cylinders at Low Reynolds Numbers", Journal of Fluid
Mechanics, Vol. 28 Part 3,1967, pp. 501-512.
Gillies, E. A., "Low-dimensional characterization and control of non-linear wake flows", PhD. Dissertation,
University of Glasgow, Scotland, June 1995.
Cohen, K., Siegel, S., Mclaughlin, T., Gillies, E., "Feedback Control ofa Cylinder Wake Low Dimensional
Model", AIAA Journal, Vo141, No.8, August 2003 (Tentative).
Holmes, P., Lumley, J. L., and Berkooz, G., "Turbulence, Coherent Structures, Dynamical Systems and
Symmetry", Cambridge University Press, Cambridge, Great Britain, 1996, pp. 86-127.
Sirovich, L., "Turbulence and the Dynamics of Coherent Structures Part I: Coherent Structures", Quarterly of
Applied Mathematics, Vol. 45, No.3, 1987, pp. 561-571.
Siegel, S, Cohen, K., Mclaughlin, T. "Feedback Control of a Circular Cylinder Wake in Experiment and
Simulation" AIAA Paper 2003-3569, June 2003.
Trefelthen, L.N., "Spectral Methods in MA TLAB", SIAM, Philadelphia, PA, 2000, pp. 145-152
 STOCHASTIC BIFURCATIONS OF THE DUFFING-
MATHIEU
EQUATIONS WITH TIME DELAYS

M. S. FOFANA
Dynamics and Control ofManufacturing Systems
Manufacturing Engineering Program, Worcester Polytechnic Institute
Worcester, MA 01609, USA

S. T. ARIARATNAM
Solid Mechanics Division, Faculty of Engineering
University of Waterloo, Waterloo, Waterloo
Ontario N2L 3G1, Canada

ZHIKUNHOU
Mechanical Engineering Department
Worcester Polytechnic Institute, Worcester, MA 01609, USA

Abstract: In the past three decades, the study of stability of buildings, bridges, beams,
columns or shells under the influence of parametric and stochastic excitations has
traditionally modelled central deflections by the Duffing and Mathieu equations. The
study of stability of solutions of these equations is well established in the literature.
Time delays appear to be a natural occurrence in structural systems as a result of the
excitations of the feedback mechanism. However, less attention is being paid on the
instability induced by the time delays. In this paper, the influence of the stochastic
delay version of the Duffmg-Mathieu equations is studied. A linearized stability
analysis of a transcendental characteristic equation of the nonlinear equations is
analyzed. As stability is lost, two types of bifurcations, namely, subcritical and
supercritical bifurcations are discussed. Then, conditions ensuring stable and unstable
bifurcations due to linearized stochastic perturbation are derived. The Markovian
diffusion approximation according to the integral averaging method and Lyapunov
exponents are employed to obtain explicit analytical results relating to the stability
conditions in the stochastic sense.

413
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 413.
2003 Kluwer Academic Publishers.
 NUMERICAL AND EXPERIMENTAL
INVESTIGATION OF A STREAMWISE
OSCILLATING CYLINDER WAKE IN THE
PRESENCE OF A DOWNSTREAM CYLINDER

ZHAOLI GUO AND YU ZHOU

Abstract: In this paper we use a newly developed lattice Boltzmann technique to simulate the wake of a
streamwise oscillating cylinder in the presence of a downstream stationary cylinder. The oscillating
frequency ratio f,/t" varies between 0 and 1.8, where Ie is the oscillating frequency of the upstream
cylinder and t, is the natural vortex shedding frequency of an isolated stationary cylinder, and the
oscillating amplitude A is fixed at 0.5 cylinder diameter, D. Three typical flow structures, depending on
f,/t, and AID, have been identified at the cylinder center-to-center spacing LID = 3.5, which are in
excellent agreement with experimental data. The lift and drag coefficients on the two cylinders are also
examined for each flow structure.

1. INTRODUCTION

Structural failure may result from synchronization between the fluid excitation
force and the system natural frequency in the streamwise as well as in the lateral
direction. In the past, most studies focused on the transverse oscillation of one
single or two side-by-side cylinders (e.g. Lai et al. 2003), perhaps because the lift
force is frequently predominant over the drag force. However, the drag force can be
important and even exceed the lift, e.g. in the case of a lightly damped structure in a
water cross-flow. Previous studies have uncovered many important aspects of
physics associated with the wake of an isolated streamwise oscillating cylinder.
Five modes (Ongoren & Rockwell 1988; Xu et al. 2002) of the flow structure have
been identified, depending on the combination of Aid and J/fs, where Ie is the
excitation frequency andls is the natural vortex shedding frequency of a stationary
cylinder, A and d are the oscillation amplitude and the diameter of cylinder,
respectively. However, flow-induced vibration often involves multiple structures in
engineering. It is of both fundamental and practical significance to understand how
a neighbouring cylinder would affect interactions between flow and an oscillating
cylinder. The experimental or numerical data of the effects of a streamwise
oscillating cylinder on a downstream cylinder wake have been scarce.
This work aims to investigate numerically and experimentally the flow around a
streamwise oscillating cylinder in the presence of a downstream stationary cylinder.
The numerical technique employed here is the lattice Boltzmann method (LBM), a
newly developed numerical method for incompressible flows (Chen & Doolen
1998). Experimental investigation was conducted in a water tunnel using laser-

415
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 415-424.
2003 Kluwer Academic Publishers.
416

induced fluorescence (LIF) flow visualization technique. The wake was generated
by two tandem circular tubes of identical diameter, the upstream cylinder being
forced to oscillate harmonically in the streamwise direction at Aid = 0.2 - 0.67 and
fils = 0 - 1.8. The Reynolds number, Re, and the cylinder center-to-centre spacing,
Lid, were 150 - 300 and 2.5 - 4.5, respectively.

2. NUMERICAL DETAILS

2.1 Lattice Boltzmann method

Unlike conventional direct numerical simulations (DNS) based on the discretization

of the Navier-Stokes equations, LBM is based on microscopic model or kinetic
equation for a fluid system. LBM is characterized by a clear picture of the physics
of fluids, the natural parallelism, and ease to handle interactions between fluids or
phases. The reliability and efficiency of LBM have been well demonstrated by a
number of studies in various fields (Chen and Doolen, 1998). Briefly, in LBM one
considers a fully discrete space-time kinetics of imagined fluid particles moving
along a regular lattice and colliding at the lattice nodes, following pre-specified
local "collision rules":

f(X + e;l'y,.t,t + l1t) - f(x,t) = Q i (f(x,t)), (1)

where fi (x,t) (i=I, ... ,b) is the distribution function (DF) of fluid particles with
velocity ei at position x at time t, and n i is the collision operator, which determines
the scatter rate of fi. The collision is such that mass and momentum are
conservative, viz.

(2)

One simple collision operator is the so-called BGK operator (Qian et ai, 1991):

(3)

where r is the nondimensional relaxation time and f(e q ) is the equilibrium

distribution function (EDF). The density and velocity of the fluid are defined in
terms of the DFs,

p= If, pu= Ie;!;. (4)

 417

With an appropriate lattice and EDF, the incompressible Navier-Stokes

equations can be derived from the lattice Boltzmann equation (1) in the
macroscopic time and space scales through the multi-scale Chapman-Enskog
expansion in the low Mach number limit,

V'u=O
au 1 2
-+(uV')u=--V'p+vV' u
(5)
at p

In Eq. (5), p = c; P is the pressure and v = c; ( r - 0.5) I1t is the shear

viscosity, and Cs is a model-dependent sound speed.
In the present work, we use a 2D square model with 9 discrete velocities (Qian
et al 1991), given by

(O,O) i =
e i = { [cos(i-l)Jr,sin(i-l)Jr]c i=1-4 (6)

.J2[cosi - 5)Jr + Jr / 4),sini - 5)Jr + Jr / 4)]c i = 5 - 8

where c = /'"xl!'!.t, /'"x and I'!.t are the lattice spacing and time step, respectively. The
EDF of the model is given by

r(eq)
Ji
= .[1 + eiu + (e
W, 2
i 'U)2 _
4
u.u: 2 (7)
Cs 2c s 2c s

where %=4/9, lUF4/9 (i=1-4), lUF4/9 (i=5-8), and Cs = c / Jj .

2.2 Fluid-Structure coupling

Several basic schemes have been used in CFD to simulate coupled fluid-structure
problems. In LBM, since a fixed uniform lattice is used in the computation, a node
belonging to a solid region may be occupied by fluid after a time step and vice
versa. Therefore, one must reconstruct the information of the new fluid-node in the
next step.
Ladd (1994) was perhaps the first to apply the LBM to a system involving
moving structures in the study of particle suspensions. Aidun et al (1998) and also
Qi (1999) improved Ladd's method later. In Ladd's method, including the two
improved versions by Aidun et al and Qi, the boundary of a solid body is
418

represented by mid-point lying between two adjacent nodes. The distribution

functions associated with the nodes adjacent to the boundary is updated according
to the bounce-back rule:

(8)

where xI is the fluid node adjacent to the boundary point Xb along the direction ej,
i.e., X f = Xb + Me j /2. In Eq. (8), J/ (X f' t) is the post-collision DF at time t
of the DF f' for particles with velocity ej' (= - eD, viz.

(9)

Therefore, the force exerted on the boundary point by the fluid particles with
velocity ej' at xI is

and the total force exerted on the body by the fluid is

F = IFj(x b ). (11)
b,i

In the above method, the representation of the object boundary by some discrete
points may result in a jagged boundary even for a physically smooth surface. Guo
et al. (2002) recently proposed a new boundary treatment for a curved boundary.
This treatment applies extrapolation to the non-equilibrium part in DF and
preserves the accuracy of the physical boundary without generating a jagged
boundary. This treatment is used in the present study.
In Ladd's and Qi's methods, the interior of a solid object is filled with a fluid,
which may produce physically unreasonable results as the fluid particle density is
close to the fluid density. In Aidun et al.'s method, the internal fluid is removed,
and the physical variables associated with a new fluid node are approximated by
those of its nearest neighbouring fluid nodes. However, this approach is only
applicable to the case when an object moves slowly so that at least one of its
nearest neighbours is a fluid node before and after one time step. For a fast moving
object, a cluster of new fluid nodes may appear at one time, and perhaps some of
them are surrounded by these new fluid nodes, and therefore the unknown variables
of such nodes cannot be determined by interpolation method. In the present case,
we directly assign the object velocity to the new fluid nodes, and the DFs are set to
 419

be its EDFs with the constant density. This treatment can produce reasonable
results, which are very similar to those produced using Aidun's method ifthe object
moves slowly.

2.3 Computation setup

Numerical simulations are carried out in a two-dimensional space for the same flow
configuration as experimental investigation (Section 3). Aid, Re and f.J1s
investigated were 0.2 - O.S, ISO - 300 and 0.2 - 1.8, respectively.
The computational domain was a 40Dx20D rectangular area (Fig. 1). Two
circular cylinders of an identical diameter, D, in a cross flow have a center-to-
center spacing TID = 2 - 6, arranged in tandem. The upstream cylinder oscillates
harmonically in the streamwise direction at a fixed amplitude, A. The Reynolds
number, Re, based on D and the free-stream velocity Uoo , is ISO - 300 (the flow is
essentially laminar) andlells ranges between O.S and 1.8.
In the computation for each case, the amplitude of the time step, I'lt, is made
equal to that of the lattice spacing Ax, which results to c =1.0. The free-stream
velocity is fixed at Uoo = O.OSm/s, which ensures a small Mach number (~0.087) for
the flow. The computational mesh is a uniform 640x320 lattice, which gives I'1t
=0 .062Ss, Ax = 0.062Sm. Therefore, the relaxation time is determined from

30D )

Figurel. Computational Domain

Re = DU", / c~(r -O.5)~t.

(12)

Initially, the velocity in the entire domain is set to be zero, and the density is set
to be unity. The DFs at each node in the whole domain are initialised to be the
corresponding EDFs. Out-flow conditions are applied to the bottom, top, and right
boundaries. The drag and lift forces on both cylinders are calculated based on Eq.
(11 ).
420

3. EXPERIMENTAL DETAILS

The LIF measurements were carried out in a water tunnel with a square working
section (0.15mxO.15m) of 0.5m long. The working section is made up of four
0.02m thick clear perspex panels. A regulator valve controls the flow speed and the
maximum velocity attained in the working section is about 0.32m/s. More details of
the tunnel can be found in Zhou et al. (2001).
Two acrylic circular tubes of an identical diameter d = 0.01 m were horizontally
mounted in tandem at the mid-plane of the working section. They were cantilever-
supported; the gap between the cylinder tip and the working section wall was about
0.5 mm, thus resulting in a blockage of about 7%. The upstream cylinder, driven by
a D.C. motor through a linkage system, oscillated harmonically in time in the
streamwise direction. The D. C. motor was controlled by a microcomputer so that
the oscillating frequency ofthe cylinder could be precisely obtained. The structural
oscillation amplitude was fixed at Aid = 0.5, whilefewas varied so thatfeifsranged
from 0 to 2. The first-mode natural frequency of the fluid-cylinder system was
estimated to be about 32 Hz, one order of magnitude greater than the maximum fe
(= 2 Hz), that is, the imposed oscillation was far away from resonance.
Dye (Rhodamine 6G 99%), which had a faint red colour and will become
metallic green when excited by laser, was introduced through one injection pinhole
located at the mid-span of the cylinder at 90, clockwise and anticlockwise,
respectively, from the leading stagnation point. A thin laser sheet, which was
generated by laser beam sweeping, provided illumination vertically at the mid-
plane of the working section. A Spectra-Physics Stabilite 2017 Argon Ion laser
with a maximum power output of 4 watts was used to generate the laser beam and a
professional digital video camcorder (JVC GY-DV500E) was used to record the
dye-marked vortex streets at a framing rate of 25 frames per second. Measurements
were carried out for Lid = 2.5, 3.5 and 4.5 and Re = 150 and 300.

4. PRESENTATION OF RESULTS

Both numerical simulation and experimental data indicate that the flow structure is
largely dependent upon the combination of Aid and fe/Is, not so much on Lid and
Re. This is consistent with the results for the flow around an oscillating cylinder
(Kamiadakis & Triantafyllou 1989). Xu et al. (2002) measured the flow around a
streamwise oscillating cylinder and noted that the various modes of the flow
structure occurred at a fixed Aid asfe/Is varied. As such, the present investigation is
mainly focused on the dependence of the flow structure onfe/Is at Aid = 0.5.
Three distinctive flow structures have been identified. The vorticity contours
(Fig 2) from the numerical simulation exhibit a flow structure in excellent
agreement with that measured using the LIF flow visualization, indicating that the
LBM technique can be used to calculate reliably the flow field around two inline
cylinders, one of which vibrates. It has been confirmed based on the power spectral
density function that, in all cases, vortex shedding from the upstream cylinder is
 421

*min= -21.8, *max= 21.8, * = 0.56

*min= -7.0, *max= 6.0, * = 0.5

*min= -4.6, *max= 5.0, * = 0.32

422

~l ~ ~, I. " I~ ~ I~ ,~
:: :~ :: :: :~ :: :; '! :~ ::
'. ' " "
:~ :~ :~
"" ' \ ' , II " II " " 1\" I, II I 0.1
" 1\ ' I ' , II ,I II I,' ,\'," II I
,I II 1 \ ' , I I , ' 1\ 1\ " I, ' , I I II I
(a) 11111,',11 J I II " ,III ',11111
1111111111,'"'1,',, ',11 11 1
U
c '
I :111,: ,11\: I'::':' ': ,III: I
, , I I ,I I I

"" "" " " I'

" -0.1

" '":
,I " " II
" "
-, "" " " ,
,I I' "
I " " "I'
I' I'
"I' I'
I
I'
I ~ ~ l I I:
~
I'
1 -0.2

120 530 550 560 500 525 5,.

tU./D tU./D

-,
400 ...
tU./D tU./D

Figure 3 Time histories of the drag and lift coefficients of the upstream and downstream
cylinder with AID = 0.5, liD = 3.5. (a) fils = 1.8, (b) 1.08, (c) 0.5. Dashed line: upstream
cylinder; Solid line: downstream cylinder.

coefficient, CL , of no more than 0.2 on either cylinder at!elfs = 1.8 (Fig 3a). The
drag coefficient, CD, on the downstream cylinder is small but very large on the
upstream cylinder. Correspondingly, the time-averaged lift coefficient, CL , and
root mean square value, C~, are small on both cylinders (Table 1). On the other
 423

hand, the time-averaged drag coefficient, CD' and root mean square value, C~, are
both very large on the upstream cylinder. Although CD on the downstream
cylinder is 0.46, smaller than that (about 1) on an isolated circular cylinder, the
corresponding C~ reaches 0.48, one order of magnitude larger than its counterpart
of a single cylinder (e.g. Chen 1987).
In comparison, for 0.8::; !ells < 1.6, alternative vortex shedding occurs from both
cylinders (Fig 2). The flow structure behind the downstream cylinder is
characterized by two rows of vortices: one consists of single vortices, and the other
consisting of counter-rotating vortex pairs. This would be clearer if the flow field is
extended further downstream (not shown here). At!elfs = 1.08, the maximum
vorticity level is 0.35, about one third ofthat at!elfs = 1.8. The maximum CL and CD
for!elfs = 1.08 (Fig 3b) on the downstream cylinder increase significantly,
compared with the case of!elfs = 1.8. While the corresponding C L (Table 1) is
small, C~ increases greatly, exceeding that (0.45 - 0.75 for Re = 200 - 700, Chen
1978) on a single cylinder, as a result from alternating vortex shedding associated
with both cylinders. Interestingly, CD as well as C~ climb considerably.

Table 1. Drag and lift coefficients for differentflls (AID = 0.5, LID =3.5).

Upstream cylinder Downstream cylinder

!elf - - - -
CD CL C'D C'L CD CL C'D C'L
0.5 1.37 0.00 1.43 0.597 0.77 0.00 0.93 1.003
1.08 1.99 -0.027 2.41 1.499 1.08 -0.095 1.22 0.961
1.8 1.62 0.000 3.71 0.084 0.46 0.000 0.48 0.091

As!elfs reduces to 0 - 0.8, the successive vortices shed from the upstream
cylinders hit the downstream cylinder alternatively, forming a single staggered
street downstream (Fig 2c). The maximum vorticity level atf,!1s = 0.5 is further
reduced, compared with that at!elfs = 1.08. The maximum CD on the downstream
cylinder at!elfs = 0.5 is smaller than that atf'!ls = 1.08, the maximum CL increases
marginally. Accordingly, CD and C~ reduce appreciably, and C~ increases
slightly.

5. CONCLUSIONS
The numerical and experimental investigations indicate that the flow structure
behind a streamwise oscillating cylinder in the presence of a downstream cylinder
can change drastically as!elfs differs. Three different flow structures have been
identified. These flow structures are distinctive from those behind an isolated
424

streamwise oscillating cylinder (Xu et al. 2002) at approximately the same

combination of AID andf,)fs.
The drastic variation in the flow structure has a profound effect on the mean and
fluctuating drag and lift coefficients on both cylinders and, expectedly, both
coefficients are significantly different from those on an isolated streamwise
oscillating cylinder.
The good agreement between measurements and numerical calculation indicates
that LBM is an efficient and reliable numerical tool for simulating flow around an
oscillating cylinder and associated fluid-structure interactions.

6. AFFILIATIONS

Department ofMechanical Engineering, The Hong Kong Polytechnic University,

Hung Hom, Kowloon, Hong Kong

7. REFERENCES

Aidun e.K., Lu Y.N., and Ding E. 1998 Direct analysis of particulate suspensions with
inertia using the discrete Boltzmann equation. J. Fluid Mech. 373, 287.
Chen S. & Doolen G., 1998 Lattice Boltzmann method for fluid flows. Ann. Rev. Fluid.
Mech, 30,329.
Chen, S. S. 1978 Flow-induced vibration of circular cylindrical structures (p.260).
Hemisphere Publishing Corporation, New York.
Guo Z., Zheng e., & Shi B., 2002 An extrapolation method for boundary conditions in
lattice Boltzmann method. Phys. Fluids, 16,2007.
Kamiadakis G. E. & Triantafyllou G., 1989 Frequency selection and asymptotic states in
lamina wakes. J. Fluid. Mech. 199,441.
Ladd A.J.e. 1994 Numerical simulations of particulate suspensions via a discretized
Boltzmann equation Part I. Theoretical foundation. J. Fluid. Mech. 271, 285.
Lai, W.C., Zhou, Y. & So, R.M.C. 2003 Interference between stationary and vibrating
cylinder wakes, Physics o/Fluids, 15,1687-1695.
Ongoren, A. and Rockwell, D. 1988 Flow structure from an oscillating cylinder. Part II.
Mode competition in the near wake. J. Fluid Mech. 191, 225.
Qian Y. d'Humires D. and Lallemand P. 1991 Lattice BGK models for Navier-Stokes
equation. Europhys. Lett. 17,479.
Qi D.W. 1999 Lattice Boltzmann simulations of particles in nonzero Reynolds number
flows. J. Fluid Mech, 385, 41.
Xu, S J, Zhou Y & So R M C 2002 Proceedings of Conference on Bluff Body Wakes and
Vortex-Induced Vibrations, pp. 183-186, 17-20 December 2002, Port Douglas,
Queensland, Australia.
Zdravkovich, M.M. 1987. The effects of interference between circular cylinders in cross
flow. ASME Journal 0/ Fluids Engineering, 1, 239.
Zhou, Y., Wang, Z. J., So, R. M. e., Xu, S.J. and Jin, W. 2001 Free vibrations of two side-
by-side cylinders in a cross flow. Journal o/Fluid Mechanics, 443,197.
 425
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 425-437.
2003 Kluwer Academic Publishers.
426
427
428
429
430
431
432
 433

a) g) m)

b) h) n)

c) i) o)

d) j) p)

e) k) q)

f) l) r)
434

a) b)

c) d)

e) f)
 435

a) b)

c) d)

e) f)
436
437
 AN EULERIAN-LAGRANGIAN MARKER AND
MICRO CELL METHOD FOR SIMULATING FLUID
INTERACTION WITH SOLID/POROUS BODIES

RAZVAN BIDOAE, REMUS M. CIOBOTARU AND

PETER E. RAAD
Mechanical Engineering Department
Southern Methodist University
Dallas, Texas, 75275-0337

Abstract: This paper presents an Eulerian-Lagrangian Marker and Micro Cell (ELMMC) method, developed
to numerically simulate the interaction between clear fluid flow and solid or porous obstacles. The extended
method can simulate both transient and pseudo-steady state problems that involve wave breaking, impact
between fluid fronts, and impact between free surfaces and solid or porous boundaries. The capabilities of the
new method are demonstrated by simulating a solid/porous sphere falling in a pool and a large wave
impacting an unrestrained solid/porous structure. Also, the paper presents a numerical study ofthe interaction
of a large wave with three different types of a tall fixed structure. The numerical results are compared with
experimental data available from experiments performed at the University of Washington.

Key words: free surface, fluid-structure interaction, wave, porous media, impact force.

1. INTRODUCTION

Numerical models used in the simulation of tsunami waves are very complex because
the flow near the run-up front displays strong nonlinearity in comparison to the motion
away from the front, and because of wave breaking near the shore under the influence
of the sea bottom and shoreline structures. In addition, tsunamis cause substantial
erosion and scouring of the shoreline, as well as transport of debris, all requiring a
complex numerical modeling capability. Numerical modeling have greatly evolved to
the point that it has become possible to not only use them for simulating waves in the
open seas, but also for estimating the effects of large waves (tsunami waves) on the
shoreline and even farther inland. Large waves impacting solid or porous unrestrained
obstacles are common phenomena in nature. An accurate treatment of impact between a
fluid front and unrestrained solid/porous obstacles becomes especially useful near the
shoreline, where the transport of natural obstacles may be simulated by unrestrained
porous or solid obstacles. In the case of tsunami waves, the ability to correctly estimate

439
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 439-452.
2003 Kluwer Academic Publishers.
440

the impact forces between the wave and shoreline structures and transport phenomena
of solid and/or porous obstacles (cars, houses, debris) is critical to the development of
mitigation measures. Increasing the accuracy of numerical simulations of tsunami
events can help designers and planners to provide adequate level of protection to
humans and shoreline structures.
The main challenge in solving numerically the full Navier-Stokes equations is
posed by the moving boundaries, thus the free surface and moving obstacles is a serious
test modeling for any programmer. In the same time the free surface can advance,
break, and coalesce in time; thus, the algorithm that tracks it must be very accurate, and
must also allow the accurate imposition of boundary conditions for the flow. Many
methods that track the free surface have been proposed in the past, each of which has
some advantages as well as some disadvantages. The volume of fluid method, level set
function, and marker and cell method are the main algorithms for tracking in time the
free surface over a fixed grid. The differences between algorithms for tracking the free
surface and interfacial flows are described in detail by Floryan and Rasmussen [1], and
Scardovelli and Zaleski [2].
The first numerical method capable of treating multi-valued two-dimensional clear
fluid flow with a free surface was the marker and cell (MAC) method, proposed in 1965
by Harlow and Welsh [3]. Chan and Street [4] proposed the SUMMAC method that
introduced the application of the pressure boundary condition directly at the free
surface and the extrapolation of velocity components from the fluid side to obtain the
velocity boundary condition. Other notable contributions include the work of Miyata
and Nishimura [5] who developed the TUMMAC and the work of Unverdi and
Tryggvason [6] who introduced an independent, unstructured grid to describe the
interface between immiscible fluids. The method was later extended by Juric and
Tryggvason [7] to track the interface resulting from material phase change.
Wave-structure interaction has been the focus of studies in naval hydrodynamics,
coastal and offshore engineering for many years. Various models that try to simulate
wave-structure interaction have been developed in the past few years. The linear
diffraction theory for deep water was first developed by Havelock [8], and was later
extended by MacCamy and Fuchs [9] to general water depths. Wang et al. [10]
investigated the three-dimensional scattering and diffraction of a solitary wave by a
surface-piercing vertical cylinder held fixed in shallow water. The Boundary Element
Method was employed by Skouroup et al. [11] to model the three-dimensional wave-
structure interaction. The temporal updating of the boundary was achieved with a time-
stepping method based on the fully non-linear free surface conditions. Cheung et al.
[12] used a similar approach to study the effects of a collinear current on the diffraction
of regular waves around three-dimensional surface-piercing bodies. The boundary
conditions of the wave propagation problem were satisfied by a time-stepping
procedure and the field solution was obtained by an integral equation method.
 441

Independently from the development of numerical methods for free surface flows,
major efforts have been expended on the development of numerical and theoretical
methods for the simulation of flow inside porous media. The flow through a saturated
solid matrix has been modeled by Vafai and Tien [13], and Lage et al. [14]. Also, Yu
and Chwang [15], and Losada and Silva [16] have made theoretical contributions of
flow through unsaturated porous media. But, to the best knowledge of the authors, this
is the fIrst paper that treats impact between a fluid front and a dry porous medium.
The method described in this paper is an extension of the three-dimensional
Eulerian-Lagrangian Marker and Micro Cell (ELMMC) technique (Bidoae and Raad,
[17], Bidoae [18]). In the ELMMC method, the free surface is tracked by the use of
unconnected, massless, "floating" Lagrangian markers, named surface markers, while
the flow fIeld is calculated in a fIxed, Eulerian system discretized with rectangular
computational cells. The primitive variables are defmed on a "staggered grid," in which
the velocity components are defIned on the cell faces and the scalar variables (i.e.,
pressure, velocity divergence, kinetic energy, rate of dissipation) are defIned on cell
centers. The surface markers delineate the full and empty parts of the computational
domain and thus make it possible to accordingly flag the computational cells as full,
empty, or suiface. Surface cells and their neighboring full cells are subdivided into
smaller cells, named micro cells. These micro cells, in conjunction with the surface
markers, make it possible to prescribe free surface boundary conditions right on the free
surface as opposed to at the centers of surface cells. In addition, computation is carried
out only in those cells that are flagged as full or surface. The pressure Poisson equation
is discretized by the use of a fmite difference approach and solved with a
preconditioned conjugate gradient method. Global mass conservation is enforced by the
use of a mass-imbalance correction scheme. The solid/porous obstacles are tracked in
the Lagrangian domain in a similar manner as the free surface, with the velocity and
position of the obstacles being computed by solving the equation of motion with a
fourth order Adams-Bashforth predictor-corrector method.

2. BASIC METHODOLOGY

The flow at time t is obtained by numerically integrating the equations proposed by Hsu
and Cheng [19]:
442

p [~av
f
+vv,(v)l=-~V(lPp)+EV2V_V_CFPflvlv+p
IP at IP IP IP IP K.JK f
g
(1)

VV=O
The governing equations are integrated in time using a projection method developed
by Chorin [20]. The solution cycle begins with the computation of tentative velocities
based on the final velocity field computed in the previous time step. Eliminating the
pressure term from Eq. (1) yields the differential equation for the tentative acceleration
in the x direction:

Then, the movement of markers from their current to their new locations is
determined from:

(3)

where nand n+ 1 represent the current and new time levels, respectively; k denotes the
marker number; and Vk represents the Darcy velocity for marker k and is calculated by
the use of the classical volume-weighting scheme involving eight surrounding
neighbors in each direction.
After advecting the free surface, an assignment of the external tentative velocity
field is made based on the new location of the free surface interface. At the same time,
the computational domain is reflagged for the fluid flag. Up to this point, the tentative
velocity field has been set and the new configuration of the free surface has been
determined.
The next step in the computational cycle is the calculation of the incompressibility
deviation, D, based on the tentative velocity field, the surface tension, and the impact
condition wherever it occurs. The general form of the Poisson equation used is:

(4)
 443

where the incompressibility deviation D is estimated as D

au ov a;;
aw
= ax + 0' + and f is

the local surface tension, is estimated based on the local curvature radius of the fluid
free surface. The Poisson equation is solved only in those macro cells whose center is
wet. When impact with a solid/porous boundary occurs, the Poisson equation is
corrected according to the impact boundary condition:

(5)

After solving a Poisson equation for the pressure field, the [mal velocity field is
calculated by correcting the tentative velocity field such as:

_ ap Of
U=U+-- (6)
ax PI
In the next step the hydrodynamic forces that act on the solid and porous obstacles
are computed. In the case of a solid obstacle, the hydrodynamic forces are computed in
each direction by integrating the fluid pressure around the solid body, such as:

F = Hn.pdA (7)

For those cells that are partially filled with fluid and have a dry or a solid center (the
Poisson equation is not solved for these cells), the pressure field is extrapolated based
on a weight average scheme.
The hydrodynamic forces on porous obstacles are computed by integrating the
Darcy and Forchheimer contribution inside the porous volume with:

(8)

where V is relative velocity between the fluid and the porous obstacle and \j is the
volume. Also, to determine the Archimedes forces that act on a porous obstacle, the
volume of the solid matrix located inside the fluid is computed at each time level.
444

Based on the hydrodynamic forces, the governing equation of motion is solved with the
Adams-Bashforth predictor-corrector method. The computed velocity of the porous
obstacle usually is not identical with the velocity used in Eq. (8) to predict the
hydrodynamic forces. Based on the latest velocity of the porous obstacle, the
hydrodynamic forces are computed again and the equation of motion is solved. This
iterative process continues until the two velocities match to within a very small
tolerance.
Using the same Adams-Bashforth predictor-corrector method, the new position of
an obstacle is determined, followed by a reflagging process of the computational
domain, but only in the vicinity of the obstacle. At the same time, the new position of
the obstacle affects the local velocity field. So, new velocity types are computed based
on the new obstacle position. The velocity information near the solid surface is
corrected such that the no-slip boundary condition is valid everywhere on the solid
obstacle.
The last steps in the computational cycle are the assignment of the fmal surface
velocity, according to the internal velocity field, the calculation of the position and
velocity of the porous/solid obstacles, and the determination of the position of the free
surface.

3. RESULTS

To demonstrate the capabilities of this new method, two simulations of complex free
surface flow problems are presented. The first problem is a numerical study (method
validation) that analyzes the impact forces between a single large wave and tall column.
In the second problem, water from behind a broken dam is released and impacts a large
unrestrained obstacle made from both porous and solid materials. The free surface is
shown in the form of panels, created from triangulating the surface markers.
 445

000
~
.---
=
0.41 m
water
H
x
0
5.90m 5.26m I
16.62m

Figure 1. Experimental setup (side view)

For the fIrst problem the experiments were performed at the University of
Washington at Seattle (UW). With reference to Figure 1, the experimental setup
consists of a rectangular tank with dimensions 16.62 m long, 0.61 m wide, and 0.45 m
tall. The volume of water initially contained behind an infInitesimally thin gate is 5.9 m
x 0.61 m x H m, where H (the water depth) was varied between 0.1 m and 0.3 m. The
center of the structure cross-section is placed at Xc = 5.26 m downstream of the gate and
Yc = 0.3 m from the near sidewall of the tank. The three cross-sectional shapes of the
vertical structure were a square with the side of 0.12 m, a rhombus with the side of 0.12
m, and a circle with diameter D = 0.14 m. Collected measurements included the time
history of the net force on the structure, the time history of the fluid velocity at different
locations upstream and around the structure as well as free surface profIle along the
mid-channel plane.

z Region 0 interest

~
9.lOm

-
inlet~ V-0utlet
H
x
0 5.90m 350m
16.62m

Figure 2. Numerical domain for 2D simulation without structure in place

Due to the great length of the experimental tank, a fully three-dimensional

numerical simulation of the entire length of the tank could not be performed in one step.
446

Instead, we chose to simulate the wave-structure interaction in a two-step approach.

This approach allows us to overcome the difficulty of performing a full 3D simulation
without significant loss of physics. The first step consists of a two-dimensional viscous
simulation of the water flow without the structure in place (figure 2). The domain is
discretized with rectangular, uniform grid of 0.01 x 0.01m.
In the physical experiment, since it is impossible to completely drain the tank
downstream of the gate, a layer of water approximately 0.02 m deep remains always on
the bottom of the tank. The right boundary of the domain is an outflow boundary so the
water leaves the domain. This is not the case in the experimental tank that has a solid
wall as the right boundary. This step provides the input data for the second step that is a
fully three-dimensional simulation of a section of the tank with the fixed structure in
place. The domain is discretized with macro cells of dimensions 0.01 x 0.01 x 0.01 m.

2.
21
18
15
u. 3 u.12
(a) H=O.15m 9 (b) H =O.3m ......-::: ... _
_._._.- _._._ .. experimental 6 - - - - - - - - experimental
- - numerical - - numerical
O2
12 10 11 12

Figure 3. Longitudinal force on the square structure

Square structure. The first case investigated was that of the interaction between
a large wave and a vertical structure with a square cross-section. Figure 3 presents the
numerical and corresponding experimental results of the time histories of the
longitudinal hydrodynamic force. The force value rises from zero to a maximum value
immediately after the impact takes place. The profiles of both the experimentally and
numerically obtained forces exhibit very nice correlation. In the majority of the cases,
the numerical impact force slightly overestimates the experimental values. After the
impact occurs, the force profile exhibits a region where the force magnitude is almost
constant in time. The "length" of this relatively flat zone becomes shorter as the H
value increases, from about 5 s (figure 3(a to only about 2 s (figure 3(b. This is
explained by the water level in the vicinity of the structure varies little after the moment
of impact. The third zone that can be distinguished in figure 3 is represented by a
decrease in the magnitude of the force in time. The time rate of force decrease is higher
for higher initial water impoundment. All cases exhibit a delay between the time of the
impact recorded in the experiment and the time of the impact predicted by numerical
simulations. Some of the possible causes are: (i) the first step of the computation is a
two-dimensional simulation that neglects wall effects, (ii) imperfect coupling at the
 447

interface between the two- and the three-dimensional domain and (iii) errors introduced
by the force transducer.

,.
16~-,~~,-~~.-~-,,-~,-~,,

12
10
u. 8

_._._.- _._._.- experimental

- - numerical

Figure 4. Longitudinal force on the cylindrical structure

Circular structure. Figure 4 displays the comparison between the experimental

and numerical results for the x-component forces. The first case (figure 4(a))
corresponds to a water impoundment behind the gate, H = 0.225 m. The force profile
exhibits two distinct zones, the impact zone and the constant magnitude zone. The force
profile displays a different behavior after the impact moment. The magnitude of the
force increases almost linearly with respect to time, the force amplitude reaches a
maximum value after about 3 s from the moment of impact. As for the corresponding
square structure simulation (figure 3), the only case in which the force profile does not
exhibit the quasi-linear drop in the magnitude in the last part of the simulation is for H
= 0.15 m (figure 4(a)). It is worth noting that for H= 0.3 m (figure 4(b)) the magnitude
of the numerically computed impact force overestimates the experimental value by
about 30%. Otherwise, the agreement of the two sets of results is favorable for all of the
above-discussed cases.

16 r--r----r--'c--r~:;.::::::::r:;;:::;::::.;:;;;~-:--'---rj 27 ~-''-r--r--r~~'-~-',-~'-~~
1. l 2'
12 21
10
u. 8
6 (a) H = 0.225 m (b) H =O.Sm
-"_._-- -"_._-- experimental - ---.- _._._.- experimental
- - numerical - - numerical

10 11 12

Figure 5. Longitudinal force on the rhomboidal structure

Rhombus structure. The comparison of the experimental and numerical results

of the longitudinal force on a rhomboidal structure is shown in figure 5. The numerical
force data shows a close agreement with the experimental force data. One can observe a
448

similar pattern for the force profile to the one discussed for the square shape, i.e. the
impact region, the quasi-constant magnitude region and the drop region. For the lowest
H value, the force profile after the impact is relatively flat for the entire length of the
simulation. As H increases, the flat region of the force profile becomes shorter in time
while growing in amplitude and almost disappears for the highest H case (figure 5(b)).
Comparing the force profiles for the square shape (figure 3) with those for the rhombus
shape (figure 5) one can observe that the latter exhibits a rounder, less abrupt pattern
right after impact.
The "flat" region is very obvious for the square and rhombus cases while for the
circular case the force exhibits oscillations. It is shown that the circular shape allows for
a slower (linear) buildup in time of the hydrodynamic force as opposed to the sudden
rise that occurs for the square and rhombus. The "drop" region indicates that the main
body of water traveling downstream the channel has passed the obstacle region and
consequently the force exerted on the structure decays in time. Also was shown that the
best shape from impact mitigation is the circular shape, given the fact that the structure
is subjected to the minimum amount of force per unit length. This study is useful in
providing both insight into this problem and guidance for designing future experiments.
However, there are several areas of uncertainty that need to be addressed in a future
work. The numerical uncertainties are relatively small in comparison with the modeling
uncertainties, which future work should focus on. First, more additional work is needed
to determine the influence of the boundary condition used in this approach for the inlet
of the three-dimensional channel and to devise more accurate treatments.

1.4 ,.,...,.,--,....,~-,---,-,-...,...,,.,...--,--,....,~.,..,.---,-...,...., 1.4 rr-~'-~'-~'-'-'-'-'-'-'-"---'

1.2 1.2 , . . . d)
1.0 (a) )
1.0 _._._. - _._._.- experimental
0.8 _._._.- _._._.- experimental
0.8 - - numerical
- - numerical
0.6
0.4
0.2
0.0 3!:--~!:--'---:!:--'---:!:--'---:!:--'---:!:--'---:!:--'--:"0

Figure 6. Longitudinal velocity in front ofthe structure (square and cylindrical)

In figure 6 we present a comparison between experimental and numerical results of

the longitudinal velocity. The measurement point is located in the center of the tank,
0.06 m in front of the obstacle and 0.03 m above the bottom of the tank. The
experimental values for the square structure (figure 6(a)) obtained with an LDV
technique while those for the cylindrical structure were obtained with a DPIV method.
Both scenarios are for an initial H = 0.25 ill. Just before the impact, the water front has
a velocity of 1.4 mls for both cases. For the square structure, after the initial impact, the
longitudinal velocity decreases abruptly yielding a bigger impact force of 15.3 N versus
 449

12.75 N obtained for the cylindrical structure. Another notable difference is observed 7s
after impact, where the fluid velocity for the cylindrical structure (figure 6(b)) is 0.4
mls as compared to 0.2 mls in the case of the square structure.

Figure 7. Interaction offree surface flows with unrestrained solid/porous dike

In the second example, the computational domain is 1.2 m long, 0.6 m wide, and
0.8 m high (figure 7). The domain is discretized with 120x60x80 cubical macro cells,
each 0.01 m in size. The structure is composed of a solid obstacle sandwiched between
two porous obstacles on its sides; all three parts have a parallelepiped shape with
identical dimensions. The porous layers are made from aluminum foam with porosity (jJ
= 0.912, permeability K = 2.883xl- 7 m2, and Forchheimer coefficient CF= 0.316. The
structure, measuring 0.2 x 0.24 x 0.2 m, is positioned 0.3 m downstream of the dam.
The initial condition is shown in the top left frame of figure 7. For consistency in the
description, the wall at x = 0 m will be referred to as the upstream wall, and the
opposing wall at x = 1.2 m will be referred to as the downstream wall.
450

1 .2 1 .2
11
1.1
11
1.1 1
0.9
1 0.8
0.7

Horizontal velocity [m/s]

0.9 0.6
[ ]
o [m]

0.5
osition

0.8 0.4
o ta Posi
Pos

0.3
0.7 0.2
o izontal

0.1
0.6 0
Hori

-0.1
Hor

0.5 -0.2
-0.3
0.4 -0.4
-0.5
0.3
Impact -0.6
-0.7
0.2 -0.8
08
0 0.5 1 1.5 2 2.5
5
T ime [[s]]
 451

4. ACKNOWLEDGEMENTS

This material herein is based upon work supported by the National Science
Foundation under Grants No. CTS-9422762, CMS-9503604, and CTS-9978680. The
authors would like to thank Dr. Catherine Petroff and Mr. Halld6r Amason of the
University of Washington, Seattle, for providing the experimental data and for their
valuable help.

5. REFERENCES

I. J. M. Floryan and H. Rasmussen, 1989, ''Numerical Methods for Viscous Flow with Moving
Boundarie," Appl. Mech. Rev., 42, pp. 322.
2. R. Scardovelli and S. Zaleski, 1999, "Direct Numerical Simulation of Free Surface and Interaction
Flow," Ann. Rev. Fluid Mech., 31, pp. 567.
3. F. H. Harlow and J. E. Welch, 1965, "Numerical Calculation of Time-Dependent Viscous
Incompressible Flow of Fluid with Free Surface," Phys. Fluids, 8, pp. 2182.
4. R. K. C. Chan and R. L. Street, 1970, "A Computer Study of Finite-Amplitude Water Waves," J.
Compo Phys., 6, pp. 68.
5. H. Miyata and J. Nishimura, 1985, "Finite-Difference Simulation of Nonlinear Ship Waves," J.
Fluid Mech., 157, pp. 327.
6. S. Unverdi and G. Tryggvason, 1992, "A Front-Tracking Method for Viscous, Incompressible
Multi-Fluid Flows," J. Compo Phys., 100, pp. 25.
7. D. Juric and G. Tryggvason, 1996, "A Front-Tracking Method for Dendritic Solidification," J.
Compo Phys., 123, pp. 127.
8. T. H. Havelock, 1940, "The pressure of water waves upon a fixed obstacle," Proceedings of Royal
Society London, Series A, 175, pp. 409.
9. R. C. MacCamy, and R. A. Fuchs, 1954, "Wave forces on piles: a diffraction theory," Technical
Memo 69, Beach Erosion Board, U.S. Army Corps of Engineers.
10. K.-H. Wang, T. Y. Wu, and G. T. Yates, 1992, "Three-dimensional scattering of solitary waves by
vertical cylinder," J. Waterway, Port, Coastal, and Ocean Eng., 118, pp. 551.
11. J. Skouroup, M. J. Stemdorff, and E. A. Hansen, 1992, "Numerical modeling of wave-structure
interaction by a three-dimensional non-linear boundary element method: a step towards the
numerical wave tank," Ocean Engineering, 19, pp. 437.
12. K. F. Cheung, M. Isaacson, and J. W. Lee, 1996, "Wave diffraction around three-dimensional
bodies in a current," J. Offshore Mech. and Arctic Eng., 118, pp. 247.
13. K. Vafai, and C. L. Tien, 1991, "Boundary and Inertia Effects on Flow and Heat Transfer in
Porous Media," Int. J. of Heat and Mass Transfer, 24, pp. 195.
14. J. L. Lage, B. V. Antohe, and D. A. Nield, 1987, "Two-Types of Nonlinear Pressure-Drop Versus
Flow-Rate Relation Observed for Saturated Porous Media," J. F1. Eng. 119, pp. 700.
15. X. Yu, and A. T. Chwang, 1994, "Wave-Induced Oscillation in Harbor with Porous Breakwaters,"
J. Waterway, Port, Coastal and Ocean Eng., 120, 125.
16. 1. J. Losada, R. Silva and M. A. Losada, 1997, "Effects of Reflective Vertical Structures
Permeability on Random Wave Kinematics," J. Waterway, Port, Coastal and Ocean Eng., 123, pp.
347.
452

17. R. Bidoae and P. E. Raad, 1998, "Mitigation of Large Wave Damage on Shoreline Structures,"
Proceedings of ASME -FD, CD-ROM, Forum 190, Paper No. 5222.
18. R. Bidoae, 2000, "Interaction of Free Surface Flow with Solid Obstacles and Porous Media", PhD
Thesis, ME Dept., Southern Methodist University.
19. C. T. Hsu, and P. Cheng, 1990, "Thermal Dispersion in a Porous Medium," Int. J. Heat and Mass
Transfer, 33, pp. 1587.
20. Chorin, A. J., 1968, "Numerical Solution of the Navier-Stokes Equations," Mathematics of
Computation, 22, pp. 745.
 A NEW LAGRANGIAN-EULERIAN SHELL-FLUID
COUPLING ALGORITHM BASED ON LEVEL SETS

FERMI CIRAK
Center for Advanced Computing Research
California Institute of Technology
Pasadena, CA 91125
cirak@cacr. caltech. edu
RA UL RADOVITZKY
Department of Aeronautics and Astronautics
Massachusetts Institute of Technology
Cambridge, MA 02139
rapa@mit.edu

Abstract. We propose a computational method for the coupled simulation of

a compressible flow interacting with a thin-shell structure undergoing large de-
formations. An Eulerian finite volume formulation is adopted for the fluid and
a Lagrangian formulation based on subdivision finite elements is adopted for the
shell response. The coupling between the fluid and the solid response is achieved
via a novel approach based on level sets. The basic approach furnishes a general
algorithm for coupling Lagrangian shell solvers with Cartesian grid based Eulerian
fluid solvers. The efficiency and robustness of the proposed approach is demonstrat-
ed with an airbag deployment simulation. It bears emphasis that in the proposed
approach the solid and the fluid components as well as their coupled interaction
are considered in full detail and modeled with an equivalent level of fidelity without
any oversimplifying assumptions or bias towards a particular physical aspect of the
problem.

1. INTRODUCTION

A large class of fluid-shell interaction problems requires a combined La-

grangian/Eulerian description of the governing dynamics. Lagrangian
formulations are inadequate for describing high-speed flows and/or
flows with significant vorticity because the mesh inevitably incurs de-
formation induced distortions, which breaks the numerical method.
This problem can be alleviated by recourse to continuous adaptive
remeshing (Radovitzky and Ortiz, 1998), but not cured in general, espe-
cially in three dimensions. Eulerian approaches with the field equations
formulated in terms of spatial variables and fixed meshes are better
suited for most fluid flows. By contrast, large deformations of solids are
more adequately described in a Lagrangian framework. The principal
advantage of the Lagrangian approach for solids lies in its ability to
453
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 453-466.
2003 Kluwer Academic Publishers.
454

naturally track the evolution of material properties associated with

the material points as well as in the treatment of boundary conditions
at material surfaces such as free boundaries or fluid-solid interfaces.
In contrast to Eulerian approaches, boundary conditions are enforced
at material surfaces ab initio and require no special attention. In this
work, we adopt an Eulerian finite volume formulation for the fluid and
a Lagrangian formulation based on subdivision finite elements for the
shell.
A number of different basic strategies have been proposed for cou-
pling the response of interacting solids and fluids. In the convention-
al Arbitrary-Lagrangian-Eulerian (ALE) approach (Donea, 1982), the
coupling of the fixed Eulerian and moving Lagrangian meshes is ac-
complished through an intermediate region in which the mesh moves
with a prescribed velocity. The specification of the mesh evolution is
key to the success of ALE methods, which unfortunately requires a
priori knowledge of the solution. In particular, for structures with very
large deformations the success of ALE methods is not always assured.
The ALE method also requires frequent mesh updating as the solid
boundary undergoes large deformations.
Our approach has its origin in the Ghost Fluid Method (Fedkiw et aI,
1999) and is an extension of the general Eulerian-Lagrangian coupling
strategy for bulk solids with compressible flows presented previously
(Cummings et aI, 2002; Meiron et aI, 2001). The dynamic deformations
of the solid shell are formulated within a large-deformation Lagrangian
finite element framework and an Eulerian finite volume formulation
is adopted for solving the compressible flow equations. The coupling
between the Eulerian fluid solver and the Lagrangian shell solver is
accomplished via a novel technique based on level sets. At each time
step, the signed distance function from the solid boundary is computed
on the Eulerian grid by a demonstrably optimal algorithm proposed
by Mauch (Mauch, 2001). The resulting implicit representation of the
fluid-shell boundary in the deformed configuration is used to enforce
the conservation laws at the boundary between the fluid and the sol-
id. It bears emphasis that the ability to compute the level set with
optimal complexity prevents this step from becoming a computational
bottleneck in the simulations.
In the following sections, we first describe the numerical formulations
employed for the shell and the fluid. Subsequently, we describe the fluid-
shell coupling algorithm. Finally, we present an example of application
of the overall numerical method to the simulation of the deployment of
an airbag. This example demonstrates the feasibility and power of the
proposed modeling strategy in capturing the complex features of the
interaction between the flow and the dynamic deformation of the shell.
 455

2. LAGRANGIAN THIN-SHELL SOLVER

The mechanical shell response is computed with the recently introduced
subdivision finite elements (Cirak and Ortiz, 2001; Cirak et aI, 2000) .
We employ a Kirchhoff-Love type kinematic assumption, which allows
for arbitrary large displacements and rotations of the shell. As it is well
known, the Kirchhoff-Love energy functional of the thin-shell depends
on the first and second order derivatives of displacements. It is known
from approximation theory that the convergence of the related finite-
element procedure requires smooth Cl-continuous shape functions. On
unstructured meshes it is not possible to ensure strict slope continuity
across finite elements when the elements are endowed with purely local
polynomial shape functions and the nodal degrees of freedom consist
of displacements and slopes only. Inclusion of higher derivatives among
the nodal variables lead to several well known difficulties, e.g. spurious
oscillations in the solution, non-physical higher order derivatives at
the boundary vertices, or complex schemes for nonsingular parame-
terization of the derivatives for large rotations. These difficulties can
be avoided by using "nonlocal" subdivision shape functions for the
discretization of Kirchhoff-Love type shell theories on unstructured
meshes. The displacement field within one finite element is interpolat-
ed through the displacements of the vertices attached to the element
and the immediately adjacent vertices in the mesh. In the resulting
shell discretization, the nodal displacements of the subdivision finite
elements constitute the only unknowns of the shell problem.
In the following we give a brief summary of the assumed shell kine-
matics. A class of finite-deformation Kirchhoff-Love shell theories may
be obtained from the ansatz:

(1)

(2)

with
Ii
--<() 3
<-Ii
2 - - 2

where r.p(()\ ()2, ()3) is the position vector of a material point associated
with the convective coordinates {()\ ()2, ()3} within the shell body in its
undeformed configuration. Similarly, r.p(()l, ()2, ()3) with respect to the
deformed configuration of the shell. The pair {()l, ()2} defines a system
of surface curvilinear coordinates, and the functions x and x furnish a
parametric representation of the undeformed and deformed shell middle
surfaces, respectively. The thickness of the undeformed shell is Ii and
456

the parameter ()3 determines the position of a material point on the

normal to the undeformed middle surface.
The unit normals to the undeformed and deformed shell middle
surface are:
(3)
Here and henceforth comma is used to denote partial differentiation.
With the aid of these kinematic assumptions, the deformation gradient
F for the shell body may be expressed in the form (Marsden and
Hughes, 1983):
(4)
In this derivation the summation convention is assumed to be in force.
The potential energy of the shell body takes the form

(5)

where, for an elastic material, W is the strain-energy density per unit

undeformed volume and
I(fi! x 92) 931 (6)
p, = ICal x (2) . a31
accounts for the curvature of the shell in the computation of the ele-
ment of volume. We consider a Neo-Hookean material extended to the
compressible range.
AO p,o
- P,ologJ + 2(trC - 3)
2
W(C) = 2(logJ) (7)

where AO and P,o are material parameters,

J = det(F) (8)
is the determinant of the deformation and
(9)
is the right Cauchy-Green deformation tensor.
In our subdivision scheme based finite element method, the dis-
cretized shell surface is entirely determined by the vertex positions of
the triangulation, or control mesh. This results in a particularly simple
discretization of the undeformed and deformed shell surfaces within
one finite element (Cirak and Ortiz, 2001):
NP
-Xh = '"'
L..t N I -XI (10)
1=1
 457

Figure 1. Triangular element and its control nodes.

where XI and XI are the vertex coordinates, and N P is the number

of the vertices in the one neighborhood of the element (Fig. 1). The
subdivision method guarantees that all such element patches exactly
match at their boundaries at least in C 1 sense. Introduction of the dis-
cretization into the weak form yields a semi-discrete system of equations
of the form:
(11)

where M h is the mass matrix, f~nt(Xh) is the internal force array, and
fhxt(t) is the external force array.
The equations of motion are integrated in time with the explicit
Newmark algorithm:

xn + .6.. 1.6. 2 .'

tXn +"2 t Xn (12)
(1 -1').6.tx n + xn (13)

in which x pre are the predictor velocities and the final velocities follow
from:

M-l(f~xt _ fint(x n ) (14)

xpre + l'.6.tx n +l (15)

where I' is the Newmark damping parameter.

458

3. EULERIAN COMPRESSIBLE FLUID SOLVER

In this section we summarize the details of the fluid flow formulation.

The algorithm employed (Samtaney and Zabusky, 1994; Samtaneyand
Meiron, 1997) solves the equations of inviscid compressible flow written
in strong conservation form:

(16)

U {p, pu, pv, pw, E, p(}T (17)

F(U) {pu, pu 2 + p, puv, puw, (E + p )u, p(u}T (18)
9(U) {pv, puv, pv 2 + p, pvw, (E + p)v, p(v}T (19)
ll(U) {pw, puw, pvw, pw 2 + p, (E + p)w, p(w}T (20)
where p is the density, u is the velocity vector, u, v, and ware its
Cartesian components, p is the pressure, e is the specific internal energy
and E = e + ~ II U 112 is the specific total energy. In the case of a perfect
gas the equation of state takes the form:
p = (r - l)pe (21)
in which I is the specific heat ratio.
The spatial discretization scheme corresponds to a finite volume
formulation on a Cartesian grid. The spatially-discretized equations
are integrated in time by recourse to the second-order Runge-Kutta
algorithm. The fluxes at the cell interfaces may be calculated either by
the Equilibrium Flux Method (EFM) (a kinetic flux vector splitting
scheme) (Pullin, 1980), or the Godunov or Roe method (a flux dif-
ference splitting scheme). Second order accuracy is achieved via linear
reconstruction with Van Leer type slope limiting applied to projec-
tions in characteristic state space (Leveque, 2002). More details of this
formulation and its parallel implementation including adaptive mesh
refinement capability can be found in Ref. (Asci Annual Report, 2000).

4. EULERIAN-LAGRANGIAN FLUID-SHELL
COUPLING

A critical aspect of the proposed numerical strategy is to devise a

theoretically sound and computationally effective means of algorith-
mically coupling the Lagrangian shell solver and the Eulerian fluid
solver. Towards this end we extend the formulation presented in pre-
vious work for coupling compressible flows interacting with bulk solid
 459

materials (Meiron et aI, 2001; Asci Annual Report, 2000). We adopt

a loosely-coupled approach in which the boundary conditions in each
solver are applied at the beginning of the time step. The appropriate
boundary conditions in each solver are determined in a way to satisfy
the conservation laws at the boundary taking into consideration the
boundary motion.
The zero-mass-flux and free-tangential-flow (slip) boundary condi-
tions appropriate for the inviscid Euler model we use are enforced in
the fluid solver through the common approach of extrapolating fluid
velocities and two thermodynamic variables (pressure and density) into
so-called ghost cells, Fig.2. Ghost cells lie outside the physical boundary
and the value of the solution vector within the ghost cells is chosen so
that the boundary conditions are enforced when the numerical flux is
computed across a cell-interface. The solution of the Riemann problem
at a moving interface results in a modified reflective boundary condition
in which the contact velocity has to match the velocity of the fluid-
solid interface in order to satisfy the zero mass flux. To this end, the
cell averages in the active fluid domain are initially extrapolated to the
adjacent inactive cells. Subsequently, the velocities in the inactive cells
are modified so that the fluid velocity normal to the fluid-shell interface
is zero:

(22)

where UF is the fluid velocity extrapolated from the active fluid cells

/
V
./
-- ShellI

V Fluid

1\
Figure 2. Identification of the fluid and the ghost cells given the instant location of
the shell
460

and Us is the shell velocity. More details on the extrapolation algorithm

may be found in the cited references.
The location of the fluid-shell interface, its normal n and its tangent
t are conveniently obtained at each time step on the fluid grid by
recourse to a level set approach. The level set function ip( x) is the
signed distance from the fluid cells to the fluid-shell interface. The
location of the fluid-shell interface is then implicitly represented in the
fluid grid by the zeroth level set:

(x) = 0 (23)
The sign of ip determines whether the computational cell is interior (ip <
0) or exterior (0 :s; ip :s; ips) to the real fluid. A key ingredient for the
success of the proposed strategy is an efficient algorithm for the com-
putation of the signed distance function presented by Mauch(Mauch,
2001). The location of the boundary and the boundary velocity are
supplied by the Lagrangian solver to the Eulerian solver.
The transfer of momentum from the fluid to the shells is accom-
plished via a consistent integration of the fluid pressure as traction
boundary conditions on the shell finite element mesh. The level set in-
formation is used again for effecting the~bilinear~interpolation of the
fluid cell pressures onto the precise location of the shell in an efficient
manner.
The proposed method avoids the robustness issues associated with
cut-cell approaches- especially in three dimensions- and furnishes, in
effect, a general coupling method for the interaction of high-speed
flows with highly-deformable solids. The introduced algorithm's overall
computational complexity is determined by the level set computation
and is of order O(m + n), where m is the number of grid points in
the subset of the fluid grid where the level set is required and n is the
number of elements in the shell mesh.
The resulting algorithms are implemented on the Virtual Test Facil-
ity for Simulating the Dynamic Response of Materials (Cummings et
aI, 2002). Details of its software architecture as well as of its implemen-
tation and scalability properties on large numbers of processors may
be found in the cited reference.

5. EXAMPLE

We consider the simulation of air bag deployment for demonstrating

the feasibility and power of the developed method in computing the
complex interactions of high speed flows with highly-deformable thin-
shell structures.
 461

The simulation corresponds to an initially-flat airbag made of an

elastic fabric with a Young's Modulus of E = 6.0 . 109 Pa , Poisson's
ratio of v = 0.3, and mass density of p = 1000.0~. The thickness of
the airbag is 7.3 . 1O- 4 m and the diameter in its flat initial configura-
tion is D = 0.74m. The discretization of the air bag consists of 10176
subdivision thin-shell elements and 5101 vertices.
The gas enters the airbag with a pressure of p = 12.0atm, mass den-
sity of p = 16.0~, and velocity of U z = 73.0~. The initial properties
of the gas are: p = 1.0atm, p = 13.0~, and'Y = 1.4. The expanding
gas generates a weak shock wave that propagates inside the airbag and
interacts with the air bag walls. The inlet conditions of the gas are kept
constant until 9,71ms and later changed to reflecting type boundary
conditions. The fluid domain of 0.86m x 0.86m x 0.49m is discretized
with 48 x 48 x 62 fluid cells.
Figures 3-6 show a sequence of snapshots of the simulation. The
deformed airbag meshes are shown on the left of each figure and the
pressure isocontours on the center-plane are shown on the right of each
figure. The portions of the fluid grid that are external to the airbag
have been left out for visual clarity by making use of the level set
function one more time at postprocessing. Important features of the
mechanics of the air bag deployment process can be observed in these
figures, including the high-frequency wrinkling modes of the airbag
fabric and the shock reflections of the gas on the deforming air bag
walls. The ability of the presented method to capture these complex
features of the coupled interaction between the flow and the air bag
fabric is particularly noteworthy.
The simulations were performed using 58 Intel Pentium III proces-
sors on a Beowulf cluster with a one Gbps fast ethernet switch. The
computation time for the whole simulation was approx. 15 hours.

6. CONCLUSIONS

We have developed a robust numerical approach for the simulation

of compressible flows interacting with highly-deformable shells. The
method furnishes an effective means of coupling Eulerian fluid solver-
s and Lagrangian large-deformation shell solvers, thus exploiting the
power of these well established formulations in each component. We
have demonstrated the efficiency and versatility of the proposed strat-
egy in a simulation of an air bag deployment process. It bears emphasis
that in the proposed approach both the solid and the fluid components,
as well as their coupled interaction, are considered in full detail and
modeled with an equivalent level of fidelity without any oversimplifying
462

Figure 3. Simulation of airbag deployment: Deformed finite element mesh and slice
of pressure isosurfaces at step 2500 after 4.25ms

assumption or bias towards a particular physical aspect of the problem,

as is common in most fluid-solid interaction approaches.

7. ACKNOWLEDGEMENTS

The partial support of DoE through Caltech's ASCI Center for the
Simulation of the Dynamic Response of Materials (DOE W-7405-ENG-
48, B523297) is gratefully acknowledged. We are grateful to Michael
Aivazis, Dan Meiron, and Julian Cummings from the Caltech ASCI
Center for providing the Virtual Test Facility for Simulating the Dy-
namic Response of Materials. The algorithms presented in this work
have been implemented on this computational framework.

References

Radovitzky, R. and Ortiz, M., "Lagrangian finite element analysis of Newtonian fluid
flows," International Journal For Numerical Methods In Engineering, Vol. 43,
No.4, 1998, pp. 607-617.
 463

Figure 4. Simulation of airbag deployment: Deformed finite element mesh and slice
of pressure isosurfaces at step 5000 after 8.16ms

Donea, J., "An arbitrary Lagrangian-Eulerian finite element method for transien-
t fluid-structure interactions," Computer Methods in Applied Mechanics and
Engineering, Vol. 33, 1982, pp. 689-723.
Fedkiw, R., Aslam, T., Merriman, B., and Osher, S., "A Non-Oscillatory Eulerian
Approach to Interfaces in Multimaterial Flows (The Ghost Fluid Method)," J.
Comput. Physics, Vol. 152, 1999, pp. 457-492.
Cummings, J., Aivazis, M., Samtaney, R., Radovitzky, R., Mauch, S., and Meiron,
D., "A virtual test facility for the simulation of dynamic response in materials,"
Journal Of Supercomputing, Vol. 23, No.1, 2002, pp. 39-50.
Meiron, D., Radovitzky, R., and Samtaney, R., "The Virtual Test Facility: An En-
vironment For Simulating The Nonlinear Dynamic Response Of Solids Under
Shock And Detonation Wave Loading," Proceedings of the Sixth U.S. Nation-
al Congress on Computational Mechanics, U.S. Association for Computational
Mechanics, Dearborn, MI, 200l.
Mauch, S., "A Fast Algorithm for Computing the Closest Point and Distance Trans-
form," Preprint, http://www.acm.caltech.edu/-seanm/sojtware/cpt/cpt.html,
200l.
Cirak, F. and Ortiz, M., "Fully CI-Conforming Subdivision Elements for Finite
Deformation Thin-Shell Analysis," Internat. J. Numer. Methods Engrg., Vol. 51,
2001, pp. 813-833.
464

Figure 5. Simulation of airbag deployment: Deformed finite element mesh and slice
of pressure isosurfaces at step 7500 after 12.13ms

Cirak, F., Ortiz, M., and Schroder, P., "Subdivision Surfaces: A New Paradigm
for Thin-Shell Finite-Element Analysis," Internat. J. Numer. Methods Engrg.,
Vol. 47, No. 12, 2000, pp. 2039-2072.
Marsden , J. E. and Hughes, T. J. R., Mathematical foundations of elasticity,
Prentice-Hall, Englewood Cliffs, N.J. , 1983.
Samtaney, R. and Zabusky, N. J ., "Circulation deposition on shock-accelerated pla-
nar and curved density-stratified interfaces: models and scaling laws," J. Fluid
Mech., Vol. 269, 1994, pp. 45-78.
Samtaney, R. and Meiron, D. I., "Hypervelocity Richtmyer-Meshkov instability,"
Phys. Fluids, Vol. 9, No.6, 1997, pp. 1783-1803.
 465

Figure 6. Simulation of airbag deployment: Deformed finite element mesh and slice
of pressure isosurfaces at step 10000 after 18.02ms
466

Pullin, D. I., "Direct simulation methods for compressible ideal gas flow." J. Comput.
Phys. , Vol. 34, 1980, pp. 231-244.
LevVeque, R. J. , Finite Volume Methods for Hyperbolic Problems, Cambridge
University Press, 2002.
"ASCI Alliance Center for the Simulation of Dynamic Response of Ma-
terials, FYOO Annual Report," URL: http:/ / www.cacr.caltech.edu/ AS-
AP / onlineresources/publications/, 2000.
 APPLICATION OF HIGHER ORDER
RUNGE-KUTTA TIME INTEGRATORS IN
PARTITIONED FLUID-STRUCTURE
INTERACTION SIMULATIONS

SANDER VAN ZUULEN* AND HESTER BUL

Faculty ofAerospace Engineering
Delft University of Technology
P.O. Box 5058,2600 GB Delft
The Netherlands
A.H. vanZuijlen@lr.tudelft.nl

Abstract: In the present study we focus on higher order time integration methods applied
to fluid-structure interaction (FSI) simulations. It is our opinion that efficiency
can be gained by application of higher order Runge-Kutta time integrators even
when only engineering levels of accuracy are required.
The fluid and the structure are integrated using implicit, third to fifth order
ARK (Additive Runge-Kutta) schemes. In the partitioned simulations, the struc-
ture is integrated first and the fluid sequentially. An explicit ARK scheme is used
to treat the coupling in a consistent way. The resulting implicit/explicit scheme
(IMEX) is tested on a linear piston problem. The IMEX scheme has design or-
der, but may need sub-iterating to obtain a stable solution for large time steps.
Results showed that large efficiency gains can be made compared to the popular
second order Backward Differentiation Formula (BDF) scheme.

Keywords: fluid-structure interaction, Additive Runge-Kutta, higher order time integration

1. INTRODUCTION
Over the last decades, numerical simulations have been increasingly used
in the design process to obtain an ever higher efficiency. The Computation-
al Fluid Dynamics (CFD) community has developed advanced and efficient
solvers to simulate flows around complex geometries. On the other hand the
Computational Structural Mechanics (CSM) community has introduced struc-

* Funded by the Dutch National Research Foundation (NWO)

467
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 467-477.
2003 Kluwer Academic Publishers.
468

fluid m

k
L x, q
0
 469

wherein U = (:u) the fluid state vector, 8K(t) is the boundary of K, F(U) =
(p~~p) is the flux vector, is the velocity of the boundary 8K and n is the
/'i,

unit normal vector perpendicular to 8K pointing outwards. The governing

equation (1) is given in the Arbitrary Lagrange Euler (ALE) formulation to
cope with the moving domain [7]. The pressure is given by the isentropic
relation pi p'Y is constant. The structural system is modelled as a single degree
of freedom mass-spring system, with a mass m and a spring stiffness k. The
system is governed by the undamped one degree of freedom model

mij + kq = fext, (2)

wherein ij denotes the structural acceleration ~q and fext is the external force
acting on the piston.

Coupling between fluid and structure

The coupling between fluid and structure is realized through the external force
in (2) and posing boundary conditions on the fluid. The external force is the
pressure force originating from the pressure difference between the ambient
pressure Po and the pressure in the fluid domain at the interface

fext = p(x = L + q) - Po. (3)

For the flow we prescribe two boundary conditions: one for the solid wall at
x = 0 and one for the interface with the piston at x = L + q

u(x = 0) = 0, u(x = L + q) = q. (4)

Linearization
We linearize the governing equations by looking at small perturbations (denot-
ed by') from an equilibrium. By satisfying the Geometric Conservation Law
[8], the ALE terms drop out of the linearized system and we obtain

f_ (;' dx + f _ 10 (;'. nds

dd-t}k = 0, (5)
o }aKo

with the Jacobian 10 = (~6)' In (5) we non-dimensionalized the variables by

using length L, speed-of-sound Co and density Po, all based on the values for
the equilibrium solution:
- Co - p _ pu
t = L t, p=-, pu= - - . (6)
Po Poco
470

The structural equation yields

(7)

wherein in = ml(poL) the mass ratio between structural mass and fluid mass
and k = kL I (poc~), the ratio between the structural spring and the fluid stiff-
ness. In the linear, non-dimensional form, the boundary conditions yield
fj'(x=O) =0, fj'(x=l)=q. (8)

2.1 Space discretization

The flow equations (5) will be discretized using a cell-centered Finite Vol-
ume (FV) method on a uniform mesh of N cells. We use a standard second or-
der central scheme [9] without additional stabilization. This way we avoid sta-
bilization of the partitioned scheme by over-dissipation in the flow. In the fluid
domain we introduce two ghost cells Ko and Kn+l to cope with the boundary
conditions in (8). By discretizing the boundary conditions and the governing
equations at the boundary we obtain
(9)

wherein box = liN is the cell size of the uniform mesh. The mass flow is
obtained by extrapolation, hence
(10)

The structural equation (7) is transformed to a system of two ordinary dif-

ferential equations

~
d[Q
-, + (0-1 0w ) Q-, -_
2 (
0
15'N::l/2 )
' (11)

wherein {J' = (r,), the structural state vector and w = Jk I in the structural
radial frequency. For the external force on the structure we need P~ +1 /2' which
is obtained from a linear interpolation between the last fluid cells K N and the
ghost cell K N +1 .

2.2 Time integration

When we denote all discrete state variables in a single vector w, the semi-
discrete system is obtained

tV = Aw = (As ASf) ( Ws ) , (12)

Afs Af Wf
 471

wherein Ws = Q' the structural state vector, Wf = [[I the fluid state vector. The
matrix Af is a 2N x 2N band-matrix which contains the discretization of the
fluid domain and As is a 2 x 2 matrix as in (11). The coupling matrices Afs
(2N x 2) and Asf (2 x 2N) will generally only have a relatively small amount
of non-zero entries, since the coupling only takes place at the boundary of the
domain.
In order to simulate the fluid-structure interaction, (12) needs to be inte-
grated. Since future applications involve engineering problems, it is envisaged
that grid clustering (such as in boundary layers), will introduce a large range
of eigenvalues, giving rise to the stiffness of the system. Stiffness can cause
the time step to be limited by stability rather than accuracy considerations.
Hence, we only consider A-stable, implicit time integration methods, which
can cope with stiffness in a robust fashion. We consider the first and second
order Backward Differentiation Formula (BDF) schemes [10] and third to fifth
order Additive Runge-Kutta (ARK) schemes [11].

Backward Differentiation Formula

The BDF scheme is a multi-step scheme, which uses the solution at previous
time steps to obtain a higher order extrapolation in time:
(3 n+1 + (3 n + (3 n-1
1W O~t -lW = R(wn+l), (13)

which is second order accurate when (31 = 3/2, (30 = -2, (3-1 1/2 and
first order accurate (backward Euler) when (31 = 1, (30 = -1, (3-1 = O. Multi
step methods have the disadvantage that they are not self-starting and accuracy
and stability can be lost when the time step is changed during the simulation.
However, they only need to solve one implicit system for every time step. Third
and higher order BDF methods are rarely used in engineering applications,
since they are not A-stable and can therefore result in instability.

Additive Runge-Kutta
ARK schemes are multi-stage methods, for which the solution at t n + 1 can be
made of arbitrary high order by a clever combination of the stage values, while
ensuring A-stability. For every stage k we solve
k
w(k) = w n + b..t L akiF(i), k = 1 ... S, (14)
i=l

wherein F(i) = F(w(i)), the flux at stage i. After computing s stages, the
solution at the next time level is
s
w n +1 = w n + b..t L biF(i). (15)
i=l
472

In this paper we consider third to fifth order ARK2 methods, which combine an
ESDIRK (Explicit first stage, Single diagonal coefficient, Diagonally Implic-
it Runge-Kutta) and ERK (Explicit Runge-Kutta) scheme. The third, fourth
and fifth order schemes consist of a 4, 6 and 8-stage algorithm respectively.
The coefficients aki and bi are usually presented in a Butcher tableau (here
represented with s = 5):
Cl 0 0 0 0 0 Cl 0 0 0 0 0
C2 a21 a22 0 0 0 C2 ii21 0 0 0 0
C3 a31 a32 a33 0 0 C3 ii31 ii32 0 0 0
C4 a41 a42 a43 a44 0 C4 ii41 ii42 ii43 0 0
C5 a51 a52 a53 a54 a55 C5 ii51 ii52 ii53 ii54 0
b1 b2 b3 b4 b5 b1 b2 b3 b4 b5
ESDIRK ERK

The time level of the stage is t(i) = t n + Cillt and the coefficients Ci simply
follow from Ci = 2: j aij. For the ARK methods the Ci and bi coefficients are
identical for the ESDIRK and ERK schemes. The Butcher tableaus for the
methods used in this paper can be found in [11].

2.3 Coupled simulations

A coupled fluid-structure system can be solved in a monolithic or partitioned
manner. In the monolithic approach we can integrate (12) with any (implicit)
time integrator. For this academic problem we do not need iterative methods to
solve the implicit system, hence the monolithic system can be solved directly.
In the partitioned approach, the fluid and structural systems are solved sepa-
rately and the coupling is treated as some (explicit) force/boundary condition.

Partitioning Algorithm
We apply a Gauss-Seidel type of partitioning, integrating the structure first ac-
cording to
(16)
wherein the coupling flux FS*) = AsfW; needs to be predicted. Next the fluid
is integrated
(17)
wherein the coupling flux Ff~*) Afs w~ k). In (16) and (17) the explicitly
known contributions are
k-l
E(k)
s w~ + llt L aki(Fs(i) + FS\ (18a)
i=l

+ IIt L
k-l
E(k)
f wfn aki
(F,(i)
f + p(i))
fs , (18b)
i=l
 473

with the fluxes defined by Fs(i) = Asw~i), FS) = AsfW?i), Ff(i) = Afw?) and
Ff~i) = AfsW~i). This completes one partitioned integration and we can start
sub-iterating by defining Fs\*) = AsfW?k) and repeat the integrations from (16).

Predictors
Choosing a predictor for Fs\*) (or wj) is not straight-forward. Some predic-
tors were designed in combination with specific time integration methods to
obtain stable schemes [2, 12], but only for combinations of second order time
integrators at most. For higher order ARK schemes we propose to use

(19)

as a predictor for the coupling flux at stage k = 2 ... s. In the present case,
the flux predictor is equal to a state predictor for Wf, since Asf is constant.
Substitution of the predictor (19) into (16) shows that in fact we solve
k-l k-l
(I - akk~tAs)w~k) = w~ + ~t L akiFs(i) + ~t L O,kjFsY). (20)
i=l j=l

In (20) the coupling fluxes are treated in a consistent, explicit way in order to
retain the higher order of the scheme. Due to the combined implicit/explicit
nature of the scheme, we refer to it as IMEX.

3. RESULTS AND DISCUSSION

In the computations the structural mass is set to in = 2 and the spring s-
tiffness to k = 1.429. With these settings the fluid has a strong influence on
the structural motion without dominating the structural dynamics. We feel that
this is representative for a large range of aeroelastic problems such as flutter.
The system has a coupled radial frequency of We = 1.01 and a coupled period
of P = 6.19. The fluid domain is discretized into 64 FV cells. The exact
solution [5, 6] is used as the initial condition. At the end of a simulation over
five periods we measure the error in the structural velocity and displacement,
in the fluid density field and in the system energy. The error is determined with
respect to a temporally exact solution (the numerical solution of a monolithic
simulation with a fifth order Runge-Kutta method and a very small time step
~t = P/5120). The implicit systems are solved with a direct solver and the
amount of work is measured as the total number of implicit system solves.

Computational efficiency of higher order partitioned IMEX schemes

We investigate the computational efficiency of the higher order partitioned
474

64 FVC, k=1.429, m=2, t=5P 64 FVC, k=1.429, m=2, t=5P

0 0
IMEX-ARK3
IMEX-ARK4
IMEX-ARK5
-2 -2
mono. BDF2

log |(E-Eex)/ E(t=0)|

log |(E-Eex)/E(t=0)|

-4 -4

-6 -6

10
10

-8 -8
IMEX-ARK3: 3.0
IMEX-ARK4: 4.6 1 sub-iteration
-10 IMEX-ARK5: 5.0 -10
mono. BDF2: 3.0 2 sub-iterations
Partitioned ESDIRK5: 3.0
-12 -12
2 2.5 3 3.5 4 4.5 2 2.5 3 3.5 4 4.5
10 10
log (work) [work in # of implicit stage evaluations] log (work) [work in # of implicit stage evaluations]
 475

64 FVC, k=1.429, m=2, tend=400P

1
64 FVC, k=1.429, m=2, Pexact=6.1916

0
0.5

-0.2
| |
q [-]

E-E(t=0)
E(t=0)

0
-0.4

-0.6
-0.5
exact
exact
mono. BDF2, t = P/50 mono. BDF2, t = P/50
-0.8
IMEX-ARK4, t = P/10, 1 iter. IMEX-ARK4, t = P/10, 1 iter.
-1 IMEX-ARK4, t = P/5, 2 iter. IMEX-ARK4, t = P/5, 2 iter.

2450 2460 2470 -1

0 500 1000 1500 2000 2500
t [-] t [-]
476

Stability
Partitioned schemes are known to cause stability problems and apart from some
first or second order schemes, we did not find any higher order unconditionally
stable scheme. The partitioned schemes generate energy at the interface, but
since the ARK schemes are dissipative, the resulting scheme can be stable. For
coarse time steps (say Llt2:: P/4, or CFL 2:: 100) the IMEX schemes generally
suffered from instability. However, since the temporal resolution at these time
steps is poor, one does not expect to run computations with such coarse time
steps. For smaller time steps (say Llt2:: P/10) we obtained that generally one
partitioned IMEX solve adds energy to the system compared to the monolithic
solution. Whether this amount is larger than what is dissipated by the mono-
lithic scheme depends on the integration method and the time step. However,
even when energy is added, it tends to be only a small amount and can easily
be dissipated by artificial/physical dissipation in fluid or structure. This can
indicate that when physical instabilities are sought, the numerical scheme will
provide a conservative answer (system will become unstable before it physi-
cally would).

4. CONCLUSIONS
In this paper we considered third to fifth order IMEX-ARK schemes for the
partitioned simulation of fluid-structure interaction. The IMEX-ARK schemes
showed to be computationally more efficient than the monolithic BDF2 scheme.
When using a non-consistent predictor the order of the scheme reduces, dimin-
ishing computational efficiency. We observed that additional sub-iterations
increase the stability of the scheme. For the fourth and fifth order schemes we
found that a second sub-iteration does not affect the efficiency severely. The
excellence of higher order schemes over the popular second order BDF has
been demonstrated for a long-term simulation.

References
[1] C. Farhat, and M. Lesoinne, "Two efficient staggered algorithms for the serial and parallel
solution of three-dimensional nonlinear aeroelastic problems", Camp. Meth. Appl. Mech.
Engrg., vol. 182, pp. 13-22,2000.
[2] S. Piperno, and C. Farhat, "Partitioned procedures for the transient solution of coupled
aeroelastic problems - Part II: energy transfer analysis and three-dimensional application-
s", Camp. Meth. Appl. Mech. Engrg., vol. 190, pp. 3147-3170, 2001.
[3] P. Geuzaine, G. Brown, C. Harris, and C. Farhat, "Aeroelastic dynamic analysis of a full F-
16 configuration for various flight conditions", AIAA Journal, vol. 41, No.3, pp. 363-371,
2003
[4] H. Bijl, M.H. Carpenter, V.N. Vatsa, and C.A. Kennedy, Implicit Time integration schemes
for the unsteady compressible Navier-Stokes equations: laminar flow, Journal of Compu-
tational Physics, 179, (2002),1-17.
 477

[5] S. Piperno, C. Farhat, and B. Larrouturou, "Partitioned procedures for the transient solu-
tion of coupled aeroelastic problems - Part I: model problem, theory and two-dimensional
application", Compo Meth. Appl. Mech. Engrg., vol. 124, pp. 79-112, 1995.
[6] EJ. Blom, A monolithicalfluid-structure interaction algorithm applied to the piston prob-
lem, Compo Meth. Appl. Mech. Engrg., 167, (1998), 369-391.
[7] M. Lesoinne, and C. Farhat, "Geometric conservation laws for flow problems with moving
boundaries and deformable meshes, and their impact on aeroelastic computations", Compo
Meth. Appl. Mech. Engrg., vol. 134, pp. 71-90, 1996.
[8] H. Guillard, and C. Farhat, "On the significance of the geometric conservation law for
flow computations on moving meshes", Compo Meth. Appl. Mech. Engrg., vol. 190, pp.
1467-1482,2000.
[9] C. Hirsch, "Numerical Computation of internal and external flows. Vol. 1: Fundamentals
of numerical discretization". John Wiley & Sons Ltd. 1994.
[10] E. Hairer, S.P. Norsett, and G. Wanner, "Solving ordinary differential equations I, nonstiff
problems". Springer Verlag 2000.
[11] C.A. Kennedy and M.H. Carpenter, "Additive Runge-Kutta Schemes for Convection-
Diffusion-Reaction Equations", Appl. Num. Math., vol. 44(1-2), pp. 139-181,2002.
[12] C. Farhat, P. Geuzaine, and Celine Grandmont, "The discrete geometric conservation law
and the nonlinear stability of ALE schemes for the solution of flow problems on moving
grids", J. Comput. Phys., vol 174, pp 669-694,2001.
 VORTEX SHEDDING FOR FLOW OVER A SQUARE
CYLINDER CLOSE TO A MOVING GROUND

s. BHATTACHARYYA
D. K. MAITI
Department of Mathematics, Indian Institute of Technology
Kharagpur 721302, India
email: somnath@maths.iitkgp.ernet.in

Abstract:
Flow past a cylinder of square cross-section placed near a plane sliding wall has been inves-
tigated. This work aims to address questions regarding the characteristics of the vortex shedding
regime and its modifications with the variation of wall to cylinder gap height and Reynolds num-
ber. The governing unsteady Navier-Stokes equations are discretised through the finite volume
method. A SIMPLER algorithm has been used to compute the discretised equations iterative-
ly. A uniform velocity profile equal in speed of the wall speed impinges on the cylinder. An
alternate vortex shedding is found in the cylinder near wake for Reynolds number ( based on
cylinder height) greater than 80. The boundary layer along the moving wall separates and the
secondary vortex forms in the region downstream of the cylinder. Unlike the stationary ground
case where the vortex shedding suppression occurs beyond a critical value of gap length, here
the vortex shedding takes place even at low gap length ratio 0.1. We found that the cylinder
experiences a upward force at higher values of Reynolds number. The gap flow is strong and
the velocity profiles overshoots its free stream value within this region.

1 INTRODUCTION
An understanding of the flow past a bluff body close to a moving ground is important
in automobile and aeronautical industries. The impermeability of the wall gives an
irotational constraints to the cylinder wake which canot spread without limit. This
can have distinct influence on the vortex shedding behind a bluff body. Bluff body
flow in close proximity of a stationary wall has drawn interest of several authors in
recent years, namely, Bearman and Zdravkovich [1], Bosch et al. [2], Lei et al.[3] and
Zovatto and Pedrizzetti [4]. The stationary wall induces a shear in the incoming veloc-
ity profile and the flow differs from its unbounded counterpart. Along the stationary
wall a shearlayer with negative vorticity forms. This negative vorticity interacts with
the vorticity originates from the lower side of the bluff body. It has been found that
beyond a critical gap length between the bluff body and stationary wall, the vortex
shedding suppression occurs (Bearman and Zdravokovich [1] ). For smaller values of
the gap length the vortex shedding process from the bluff body is different from that
of cylinder placed in a free stream ( Zovatto and Pedrizzetti [4] ).

479
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 479-489.
2003 Kluwer Academic Publishers.
480

When the wall is considered to be moving horizontally, the classical boundary lay-
er on the wall is absent and the flow mechanism is different from that of the stationary
wall. There are only a small number of theoretical studies published in open liter-
ature. Bearman [5] reviewed earlier studies on bluff body flows that are relevant to
the understanding of vehicle flows. There they predicted that the velocity profile may
overshoots in the gap flow between the lower surface of the cylinder and the moving
wall. The bluff body close to the moving wall may experience a upward force. Arnal
et al. [6] studied numerically the flow around a rib in contact with a sliding wall. Ar-
nal et al.[6] pointed out that the positive vortex which forms in the downstream lower
corner of the rib is much weaker when the downstream wall is fixed than when it is
moving. Kumarasamy and Barlow [7] studied the flow over a half-cylinder close to
a moving wall. Senior and Zhang [8] studied experimentally the force and pressure
behavior of diffuser-equipped bluff body in ground effect. Recently Jones and Smith
[9] made a study on the gap flow between car underbody and moving ground.
In this paper, we studied numerically the flow past a square cylinder which is
placed close to a moving wall. The wall travels at a uniform speed, namely the speed
of the oncoming free stream of fluid. Flow has been considered to be two-dimensional
and laminar. The influence of the moving wall on vortex shedding process behind the
cylinder and the gap flow under the cylinder has been investigated. Our results show
that an alternate vortex shedding occurs for all values of the gap length. The cylinder
experiences a upward force at higher values of Reynolds number and cylinder to wall
gap height ratios.

2 STATEMENT OF PROBLEM

We consider the wall lying along the x-axis and a square cylinder oflength A is placed
parallel to the wall at a height H from the wall. The far-field and the wall is considered
to be moving along x-axis with a constant speed U and the cylinder is stationary. The
velocity field is non-dimension ali sed by U and the height of the cylinder A is taken to
be the characteristics length scale. Thus the body at height y = HI A = L occupies
the range of x between zero and unity in the frame moving with the body (Fig. I). The
pressure is non-dimensionalised by pU 2 , where p is the density of the fluid. Reynolds
number Re is defined as Re = U AI v and L = HI A is the non-dimensional gap
height between the cylinder and the plane wall.
The flow is governed by the Navier-Stokes equations for unsteady two-dimensional
laminar motion, subject to boundary conditions of no slip along the cylinder surface.
At the far upstream and on the wall the Cartesian components of velocity (u, v) are tak-
en to be (l, 0). A symmetry boundary condition is imposed on the top lateral boundary,
and a vanishing -gradient boundary condition is applied at the outflow boundary.

3 NUMERICAL METHODS
We employed a finite-volume method to compute the governing equations on a
staggered grid system. An Euler implicit scheme was used to discretize the time
derivatives present. The resulting discrete equations are solved by using a pressure
 481

t
--
u~~ Y

-- D
- Ground P~ane, u=l

Figure 1: The flow configuration of a square cylinder at a distance L above horizontal

moving ground.

correction based iteration algorithm SIMPLER ( Patankar [l0]). The solution pro-
cedure starts by supplying initial estimates for the velocity and pressure fields and a
converged solution is obtained by iteration. We have presented the results when the
periodic vortex shedding is achieved. At the initial stage of the motion the time step
is taken to be 8 = 0.001 which has been subsequently increased to 0.005 after the
transient state. Flow is assumed to start from rest impulsively.

1IIIIIIIIi 4 y
3
2
1
o
-5 o 5
x
10 15 20

Figure 2: The arrangement of the computational grid in the computational domain.

A non-uniform grid distribution in the computational domain is incorporated (see

Figure 2). The grid is finer near the surfaces of the square cylinder and bottom wall
to better resolve the gradients near the solid surfaces. The first point at a distance
from each wall is O.01A. We perform the computation on a 185 x 170 grids with the
first and second number being the number of mesh points in the x-direction and in the
y-direction, respectively. Additional runs for the coarser (120 x 140) grids and finer
grids (235 x 200) were undertaken for a check of grid independence. Further reduction
on step sizes close to the solid surfaces does not produce any signifiacnt change. The
effect of grid size on St and C Dav at various values of Reynolds number for uniform
flow past a square cylinder is presented in Figure 3.
In order to assess the accuracy of our numerical method, we have computed the
482

Freestream:
0.35 o Frankeelal. [19901
-Amal el al. [1991]
0 .3 x Hwang et al. {1997]
+ Present result. 120xl40
---_ Moving ground
0.25 - Present result. 185x 170
Ii) Present result. 235x200
0.2

0.15

0.1 ------------------------ -----'----.. --.,:!I

200 400 600 800 1000 1200 1400

Re

(a)

4 .5 Freestream: Freestream:
-Present result, 120xl40 ---- Moving ground + Franke et al. [19901
4 - Present result, 185x170 ... Treid1er [1991J

~: ji~~~~~-_-_-~_~-:_~:_I~ t_~-~- -~-~:-:~ :t--~-~-i:~: :~1!~-t:-1~-!~-5 ~2-~ ~1

1.5

200 400 600 800 1000 1200 1400 1600

Re

(b)

Figure 3: Comparison among (a) Strouhal number (St) and (b) average drag coeffi-
cient (CDav) results for the flow past a square cylinder in Jreestream at different grid
sizes.
 483

Strouhal number(St) and CDav for uniform flow past a square cylinder without any
ground effect at different values of Re and compared with the results due to Davis
andMoore [11], Franke and Rodi [12], Treidler [13], Arnal et al.[6], Li et al.[I4] and
Hwang and Yao [15]. We find that our computed results presented in Figure 2 are in
good agreement with those of above mentioned published papers.

4 RESULTS AND DISCUSSIONS

The present flow field is governed by two parameters, namely, the Reynolds num-
ber Re and the ground to cylinder gap height L. We have presented the numerical
solutions at L = 0.5,0.25,0.1 for different values of Reynolds number.
The variation of Strouhal number ( St) with Reynolds number at the various values
of cylinder to wall gap length L = 0.5,0.25,0.1 is presented in Figure 3a. A periodic
flow field is observed after a short transient flow. The vortex-shedding frequency is
determined from the time evolution plot of the lift coefficient distribution. The time
period T is the difference between the two successive non-dimensional time at which
the lift coefficient crosses its mean value. The behavior of the Strouhal number with
Reynolds number is not uniform. The periodicity in the frequency of vortex shedding
decreases with the decrease of gap height ratio of the cylinder to the wall. The decrease
becomes more significant for higher values of Reynolds number flows. The variation
of St with Reynolds number at gap length L = 0.1 is not significant. It has been found
that for the present case the transition from steady flow to a periodic vortex shedding
regime occurs at larger Reynolds number compare to the case where cylinder is in
free stream. The periodic shedding is delayed because the vorticity shed from the
cylinder's wall side couples with the wall vorticity which arrests its evolution. In the
unsteady regimes, the vortex shedding pattern is similar to the von Karman vortex
street, even if the sliding wall produces a cancellation on the positive vortex emerges
from the lower shear layer.
We have seen ( refer Fig.3a) that vortex shedding occurs even at smaller gap length
L = 0.1. The streamlines exhibit oscillation and the flow is periodic up to a Reynolds
number 1000.0, while in the case of flow past a bluff body in proximity of a stationary
wall the vortex shedding supression occurs beyond a critical value of cylinder to wall
separation length (Bearman [1]). For the case of stationary wall the jet through the gap
between the cylinder and the wall separates near the front lower comer of the cylinder
attaches behind the cylinder surface and does not roll-up to form vortex shedding.
The separated upper shearlayer of negative vorticity is almost horizontal for the case
of stationary wall beyond the critical gap length. But in the case of moving wall, the
wall shear layer is not strong enough to interact with the lower shear layer positive
vortex and prevents roll-up.
Figure 3b show the average drag coefficient C Dav at different Re for L = 0.5,
0.25,0.1. The instantaneous drag coefficient is averaged over shedding cycle to ob-
tain the CDav. In the case of cylinder placed in a free stream the average drag co-
efficient increases monotonically with the increase of Reynolds number. In this case
for flow past a cylinder close to a moving wall, the monotonicity of drag coefficient
with Reynolds number is absent. Due to the presence of the wall shear layer the drag
484

coefficient is changed greatly. The effect due to the variation of the gap length on the
average drag coefficient is small.
Figure 4 display the evolution of vorticity during a cycle of vortex shedding when
Re = 500.0 and L = 0.25. The shedding cycle can be identified with one period
in drag cycle, starting and ending at the point of maximum drag. In the vicinity of
the cylinder and the moving wall, there are three shear layers, the two shear layer that
develop along the surface of the cylinder on the top and bottom respectively, and the
shear layer that develops along the wall. The upper shear layer on the cylinder has
negative vorticity, while the lower shear layer on the cylinder has positive vorticity.
These vortices attain the maximum strength before being shed. A shear layer of neg-
ative vorticity forms along the moving wall in the downstream of the cylinder. This
boundary layer along the moving wall separtes further downstream of the cylinder.
Due to the absence of the classical boundary layer on the moving wall, the secondary
vortex strength is greatly reduced compare to the case of a stationary wall. Figure
4 shows that at the starting of the shedding cycle where the cylinder experiences the
maximum drag, the negative vortex induced at the upper shear layer grows rapidly
before being shed. The negative vortex shedding occurs at t = T + T / 4 and positive
vortex induced in the lower shear layer grows in size and elongates at t=T + T /2. At
t=T + 3T / 4 the positive vortex is shed in the wake. Results show that the the positive
vortex induced at the lower side of the cylinder is quite strong and size is much big-
ger than the case when the cylinder is in the vicinity of a stationary wall. The strong
positive vortex induces unsteady separation and secondary vortex in the moving wall
boundary layer. The positive vortex shed from the lower shear layer pushes upwards
the upper shear layer induced negative vortex in the wake of the cylinder.
The time average mean pressure distribution Cp on surface of the cylinder at var-
ious gap length L = 0.1,0.25,0.5 with Re = 500.0, 1000.0 is described in Figure 5.
The pressure is negative on the lower face of the cylinder with a suction (-Cp ) peak
to develop at the front of the undersurface. The position of the suction peak shifted
slightly with the increase of Re.
The average lift experienced by the cylinder at different Reynolds number is il-
lustrated in Figure 6 for L = 0.1,0.25,0.5. The lift takes positive or negative values
depending on the values of Reynolds number and gap length. The lift experienced by
the cylinder decreases as the Reynolds number increases. Our results show that the
lift is positive for lower values of Reynolds number. At a lower value of gap length (
L=O.l) the lift is positive for the entire range of Reynolds number. For smaller values
of gap length the lubrication effect due to viscosity plays a major role.
The flow under the cylinder is described in Figure 7 at different values of Reynolds
number and cylinder to plane gap height ratio. The horizontal velocity u is positive
and overshoots near the moving wall. It may be noted that in a frame fixed in the
ground, the local fluid motion is temporaily in the opposite direction ( rightwards)
from that of the ( leftwards moving) body. The vertical velocity is downward near the
entry region of the gap and become upwards as we move away from the entry region.
Our results show that the fluid is slowing down relative to ground velocity as the gap
is narrowing. The fluid velocity increases with the rise of Reynolds number.
 485

~--~--~--~--~--~---r---r---n 3

-'~});)
2.5
2
..... " .' 1.5 y

o -7 2 6.5 3 4 5 6 7 8
x

(a)

./;r~ff(~;~'
' .. :-' -,,"

o 3 4 5 6 7 8
x

(b)
486

". 0< 3.5

3
2.5
2
1.5 y

l:J::~=~~~~~~.5~~..c.:'~'S:I--'':;4~~~:SJ:- ~.5
o 5 6 7 8
x

(c)

,- 3.5
3
2.5

<::~(~~1j;-7 2
1.5 y
1
6.5 0.5
0
6 7 8
x

(d)

Figure 4: Iso-vorticity contours during one shedding cycle for gap length L = 0.25 at
Re = 500.0. (a) t = 60.49(= T); (b) t = 61.748(= T + T/4); (c) t = 63.005(=
T + T/2); (d) t = 64.263(= T + 3T/4).

0.5

0.5 :

-1.5

-2LA-------B~------C~-------D------~A

Figure 5: Time-average pressure coefficient (Cp ) over the cylinder for gap length
L = 0.5,0.25,0.1 at Re = 500.0,1000.0
 487

Figure 6: Average lift coefficient (C Lav ) asfunction of Reynolds number (Re)for gap
length L = 0.5,0.25,0.1.

2.5

1.5
\~\

:: '- -'_- '-_ '-_ - '- _- '- _-' - _-' x_=O-': . ~"'_~:. ;~_\. l\_-:_ : _ '/
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

(a)

-0.4 ' - - - - ' - - - ' - - - ' - - - ' - - - - ' - - - - ' - - - ' - - - ' - - - - ' - - - '
o 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Y

(b)

Figure 7: Velocity profile in the gap flow for gap height L=0.5, 0.25, 0.1 at Re=500,
1000 at different x-station. (a) u; (b) v.
488

5 CONCLUSIONS
The flow past a square cylinder in the vicinity of a plane moving wall is considered.
The two dimensional laminar flow is studied for different values of cylinder to wall
gap length and the Reynolds number. A negative vortex is generated at the front top
comer of the cylinder and moves behind the cylinder as it grows. The lower down-
stream comer of the cylinder induces a strong positive vortex. Negative vortex forms
along the sliding wall below the positive vortex close to the cylinder. The boundary
layer along the moving wall separtes further downstream of the cylinder and a sec-
ondary vortex appears colse to the wall. The vortex shedding frequency decreases as
the gap height between the cylinder and the wall reduces. Unlike the stationary wall
case, here the wall shear layer is not strong enough to prevent roll-up in the lower
shear layer. An alternate vortex shedding process takes place even at the cylinder to
wall separation height L = 0.1. The positive vortices which originates from the cylin-
der is stronger than the case of flow past a cylinder in presence of the stationary wall.
Velocity profile overshoots in the gap flow region between the cylinder lower surface
and the wall. The average drag experienced by the cylinder is higher than the corre-
sponding unbounded case. Cylinder experiences a downward force for higher values
of gap height ratio at large Reynolds number.

6 ACKNOWLEDGEMENTS

The authors wish to express their thanks to the CSIR, India for a grant to support this
study.

7 REFERENCES
1. Bearmen PW and Zdravkovich MM. Flow around a circular cylinder near a plane bound-
ary. J. Fluid Mech. 1978; 89: 33-47.

2. Bosch G and Rodi W. Simulation of vortex shedding past a square cylinder near a wall.
Int. J. Heat Fluid Flow. 1996; 17: 267-275.

3. Lei C, Cheng L and Kavanagh K. Re-examination of the effect of a plane boundary on

force and vortex shedding of a circular cylinder. 1. Wind Eng. Ind. Aerodyn. 1999; 80:
263-286.

4. Zavatto L and Pedrizzetti G. Flow about a circular cylinder between parallel walls. J.
FluidMech. 2001; 440: 1-25.

5. Bearmen pw. Review-bluff body flows applicable to vehicle aerodynamics. 1. Fluid

Eng. 1980; 102: 265-274.

6. Arnal MP, Georing DJ and Humphrey JAC. Vortex shedding from a bluff body adjacent
to a plane sliding wall. J. Fluids Eng. 1991; 113: 384-398.
 489

7. Kumarasamy S and Barlow JB. Computation of unsteady flow over a half-cylinder close
to a moving wall. J. Wind Eng. Ind. Aerodyn. 1997; 6971: 239-248.

8. Senior AE and Zhang X. The force and pressure of a diffuser-equipped bluff body in
ground effct. J. Fluids Eng. 2001; 123: 105-111.

9. Jones MA and Smith Fr. Fluid motion for car undertrays in ground effect. J. Eng. Maths.
2003; 45: 309-334.

10. Patanka Sv. Numerical heat transfer and fluid flow. 1980; Hemisphere, Washington DC.

11. Davis RW and Moore EF. A numerical study of vortex shedding from rectangles. 1.
Fluid Mech. 1982; 116: 475-506.

12. Franke R and Rodi W. Numerical calculation of laminar vortex shedding flow past cylin-
ders. J. Wind Engg. Ind. Aerodyn. 1990; 35: 237-257.

13. Treidler EB. An experimental and numerical investigation of flow past ribs in a channel.
Ph.D. Thesis. 1991; University of California at Berkeley, Berkeley, CA.

14. Li G and Humphrey JAC. Numerial modelling of confined flow past a cylinder of a
square cross-section at various orientiation. Int. J. Num. Meth. Fluids 1995; 20: 1215-
1236.

15. Hwang RR and Yao C. A numerical study of vortex shedding from a square cylinder
with ground effect. J. Fluid Engg. 1997; 119: 512-518.
 EXTENDED HAMILTON'S PRINCIPLE FOR FLUID-
STRUCTURE INTERACTION

HAYM BENAROYA, TIMOTHY WEI

Department ofMechanical and Aerospace Engineering,
Rutgers University, Piscataway, New Jersey, USA

Abstract: The problem of vortex-shedding from bluff bodies has been examined for over a century, as
reflected by the extensive literature on the subject. The focus of these foregoing researches can be split into
two broad categories: investigations into the flow characteristics around a body in a flow, and studies of the
response of a bluff body to the forces from the flow.
The approach sought here, to derive a set of equations of motion for a structure subjected to
vortex-shedding loads from first principles, represents a novel approach to a long-studied problem. The work
at hand also embraces two disciplines: vortex-shedding from bluff bodies, and the dynamics of a compliant
offshore structure.

1. COMPLIANT OFFSHORE STRUCTURES

Compliant structures provide an attractive alternative to traditional offshore platforms.

Traditional platforms resist forces due to current, waves, and wind. Thus, the structure
is assumed to undergo displacements small enough to allow linear dynamic methods to
be used to solve for the response. Compliant structures also undergo small
displacements, but these displacements are large enough to necessitate the introduction
of nonlinear methods to solve for the structural response. An extensive review of the
nonlinear dynamics of compliant structures has been presented by Adrezin et al. (1996)
[1]. Traditional platforms, since they resist the forces of the ocean environment, are ill-
suited to deep water applications, as the cost to make a structure strong enough to
completely withstand wave, current, and wind forces becomes prohibitive with
increasing depth. Compliant structures are better suited to deeper-water applications,
since such facilities need not be as reinforced against the ocean environment as the
more traditional structure.
The articulated tower is attached to the ocean via a universal joint. The tower

491
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 491-506.
2003 Kluwer Academic Publishers.
492

includes a ballast chamber near the base, and a buoyancy chamber nearer the surface.
The universal joint allows tower motion to occur in three dimensions.
The tension leg platform (TLP) design takes a different approach to surviving
the ocean environment. The TLP consists of a platform connected to a submerged
pontoon (to provide buoyancy). The platform system is in turn moored to the ocean
floor via several slender, flexible cables attached to the comers of the platform.
The articulated tower and a single mooring cable of the TLP may both be
modeled as a partially submerged beam. The methods used here can then be used as a
basis for understanding and modeling the full-articulated tower or tension leg platform.

i.i. Vortex-induced Vibration

The phenomenon of vortex-induced vibration has been investigated for many years.
Previous reviews of the subject matter have been performed by Marris (1964) [10],
Berger and Wille (1972) [5], King (1977) [9], and Sarpkaya (1979) [12]. More recently,
these earlier reviews have been updated by Griffm (1982) [7], Bearman (1984) [3], and
Billah (1989) [6]. Also, Zdrakovich (1996) [13] provides an overview of different
modes of vortex shedding. Vortex-induced vibration is based on the von Karman vortex
street.
An important aspect of vortex-induced vibration is lock-in or synchronization.
As the flow velocity past a responding bluff body increases, the frequency at which
vortices are shed from the body increases almost linearly with flow velocity. However,
when the vortex-shedding frequency reaches the natural frequency of the structure, the
vortex-shedding frequency does not further increase with flow velocity. Rather, the
shedding frequency remains locked-in to the natural frequency of the structure. At a
higher flow velocity, however, the linear dependence of shedding frequency upon flow
velocity resumes. Within the synchronization region, large body motions are observed
(the structure undergoes near-resonant vibration). The lock-in phenomenon has been of
great interest to many researchers, both for description of the underlying fluid
dynamical mechanisms causing synchronization and for prediction of structural
responses.
Work with structures undergoing vortex-induced vibration can be classified
into three main types according to Gupta et al. (1996) [8]. The first class consists of
wake-body coupled models, in which the body and wake oscillations are coupled
through common terms in equations for both. The second class relies upon
measurement of force coefficient data from experiments. The third class uses a single
dynamic equation, but includes aeroelastic forcing terms.
 493

2. McIVER'S EXTENSION OF HAMILTON'S PRINCIPLE

Here we introduce the main thrust of our work, as it currently stands. We use a slightly
different notation in order to conform to the literature of the application at hand. This
work follows that of McIver [11], and Benaroya and Wei [4]. It is an extension of
variational mechanics. In 1973, McIver [11] published a work with broad implications
for modeling complex fluid-structure interactions. The central feature of his work was
the broadening of Hamilton's principle to include integral control volume concepts from
fluid mechanics. In this section, we summarize the key developments by McIver.
McIver uses Reynolds transport theorem to extend Hamilton's principle.
At some time, to, a collection of fluid particles comprising a system occupies the same
space as a control volume. It is possible to write the rate of change of any property of
that system in terms of control volume parameters. This is the classic Reynolds
transport theorem, which may be written as

~
Dt lystem
rontrol~(Ap
r (Ap )dv = ~ol at
)dv + !ontrol (Ap)U . nds.
surface
(1)

In this form, A represents the property of interest per unit mass (intensive property),
p is the fluid density, dv and ds are differential volumes and control surface area
elements, respectively, and U = U(x, t) is the fluid velocity at any point on the
control surface.
The strength of McIver's work was in identifying an approach for analyzing
complex interactions where the system boundaries are not necessarily well defmed or
where the system configuration at two distinct times may not be readily prescribed. In
the classical Hamilton's principle approach, the system contains one or more solid
objects whose positions may be prescribed at specific times. That is, the system is of
fixed mass containing the same material elements at all times. By introducing Reynolds
transport theorem, McIver generalized the analysis to include control volumes where
material is permitted to cross the boundaries. Specifically, by applying Reynolds
transport theorem, we obtain

8Lsystem + 8W - rontrol~(PU).
~ol at
or dv - !ontrol
surface
(pU) &(U - Vcontrol) nds = 0,
494

(2)
where, in the Lagrangian of the open control volume, L system' the mass is not fixed.
We have retained the possibility of a moving control surface by including V control'
which may have a different value in different regions of the control surface. The control
surface here implies an open region since at closed
portions flow velocity U = Vstructure . Equation 2 is a statement of the principle of
virtual work.
Now integrate with respect to time over the interval t1 to t2, and, again requiring the
system configurations at t 1 and t 2 to be prescribed, the extended form of Hamilton's
principle for a system of changing mass can be expressed as

8 fI
L system dt + f 8W dt - f dt
I I
!ontrol
surface
(pU) 8r(U - V controJ n ds = 0,
(3)
where 8W is the virtual work performed by the non-potential forces acting on the
same system. If the control surface (CS) does not move, then V control = 0.
If the system configuration is not prescribed at the end times, we must proceed
differently after Equation 2. If the system is not prescribed at t) and t2 ,the variation
of the displacement 8r' 0, rather, we have the following relation 8r = Udt . We
have replaced the arbitrary variation 8r by the actual dr = fdt. From this, we can
see that the variational operator is related to the time differential operator by
8(-) = dt d() / dt. Begin with Equation 2, interchange the partial derivative with the
integration over the control volume, and replace the variation as noted above, and,

d(r +
.!.
n)control [ ]
___ vo_l_ = W- rontrol P U2 e + (U - VcontroJ n ds.
dt tiface 2
(4)

This equation states that the change in energy of a system equals the rate at
which nonconservative work is done on the system plus the rate of gain of energy by
virtue of the fluid flowing through the control surface and the advancing control surface
engulfing particles.
 495

3. THE EXTENSION FOR EXTERNAL VISCOUS FLOWS

McIver derived his extension for applications where the fluid is encased in the
structure. The equations derived above assume a steady frictionless flow. Examples he
studied include the rocket, and flow in a pipe. The application of interest here has the
structure within the fluid. In particular, we are interested in generalizing the McIver
extension of Hamilton's principle so that we can model the vortex-induced oscillation
of a structure. This is a viscous external fluid-structure interaction. McIver's extension
utilizes the control volume concept to account for fluid mass that enters and leaves the
structure. This same idea can be applied to a control volume around a fluid that has a
structure internally.
Modeling of the internal flow problem has the advantage that, assuming no
cavitation, the fluid is bound by the structure. With external flows, the fluid is
unbounded and the modeling becomes more challenging.
In this development it is useful to think of the system, comprising a structure
surrounded by a moving fluid, as one that is defmed using two control surfaces. The
fIrst control surface is at the structure surface. It is a closed control volume. The second
control surface is at some distance from the structure, as shown in the fIgures below.
This control surface may be partially closed and partially open, or all open, depending
on the application. It is important to keep track of the various portions of the control
surface so that the parameters are appropriately prescribed.
For such a control volume, (i) there is a time rate change of momentum
within the control volume due to the unsteady character of the flow, (ii) there is a net
momentum flux across the boundaries of the control surface, (iii) there IS an
instantaneous pressure p acting on the control surface, and (iv) there is an
instantaneous shear stress r acting on the control surface.
The integral over the control volume needs to be interpreted here to include
the inner (structural) as well as outer (fluid) control volumes. We fmallyobtain
496

d (I:tructure + II structurJ~~7trol ( .)
- - - - - - - - - - - + m jIuidUU control
dt vol

=- fpen PU 2 (U - Vcontrol) n ds
~s 2
+ ~~sed(-pn+'tJ . Uds+ ~;n(-pn+'tJ.Uds.
(5)

The following procedure will be used for working with Equation 5. We will substitute
expressions for the kinetic energies on the left hand side. On the right hand side we will
have experimentally based analytical expressions for the flow velocities, pressures and
stresses. This relation will allow the derivation of an expression for the acceleration of
the structure. This will be integrated twice to find the expression for the structural
displacement as a function of time and the system parameters. This result will then be
compared to the experimentally derived structural displacement as a function of time.
The two functions will be compared, permitting an evaluation of the analytical
framework and its components.
There is no possibility of analytically arriving at expressions for each term and
each function in Equation 5. Therefore, it is necessary that an experimental program be
run in parallel for particular applications. The power of this energy-based approach is
twofold. The first is that an analytical framework is created to organize our
understanding of a complex nonlinear and interactive phenomenon. But second and
equally important is that the experimental program provides us with invaluable
information about some of the components of these equations, and this permits us to
utilize the variational tools in the derivation of the equations of motion.

3.1. Control Volume Definition

First let us describe the control volume of interest here. Consider a top view of a
circular cylinder with two control surfaces, one at the surface of the cylinder and the
other some concentric distance out in the surrounding water. One question that arises
when considering various possible control volumes is whether a particular control
volume has significant advantages either for the analytical formulation or for the
experimental procedures, the results of which are required as input to the analytical
model. This will have to be considered as part of an examination of the proposed
methodology.
 497

4. STATIONARY OUTER CONTROL VOLUME: CYLINDER OSCILLAING

ABOUT CONTACT AT BASE

Here we take the cylinder to be connected only at its base via a leaf spring. It behaves
like a column supported only at its base. For purposes of this example we assume that
the cylinder is rigid, as above, and that three-dimensional effects can be ignored. A top
view of the cylinder is shown in Figure 1. There are a number of possible control
volumes. Here we use control volume 4, which extends from O.4X / D from the left
edge of the cylinder to 1.6X / D from the right edge of the cylinder.
498

5 r-
r-----------------------------------------~

1 2 3 4 5 6 7

4 r-

o
3

-
Q

2 -

I I I I I
1 2 3 4 5
XID
Figure 1: Top View O/Cylinder in a Variety a/Possible Control Volumes

This control volume appeared to capture the essential dynamics. The single generalized
coordinate that defmes the cylinder location is the angle of rotation Brad. We have an
additional term in the potential of the structure due to the difference between the
buoyancy force and the weight. We assume that the resultants of these distributed
forces act at the center of geometry of the circular cylinder. Then, for some rotation B,
this additional potential results in the moment (mg - B)1'-sinB, where mg is the
 499

weight of the cylinder, B is the total buoyancy force (which equals the weight of the
displaced fluid) and L is the length of the cylinder. Let lobe the mass moment of
inertia for the circular cylinder about its base, kT be the torsional spring constant at the
base, then the governing equation is

O(I)j +kTO-(mg-B) ~ sinO )+m jlUidUO = t~sed(- pn+'CJ. U ds

+ ~;n[ -~ U 2
Un +(- pn+'CJ. U JdS.
(6)

This equation, as well as the other cases above, can also be evaluated numerically if
written in the form

where F (t) is the sum of all the remaining terms,

Then, we solve for 0 by integrating both sides of Equation 7, and then

integrating again with respect to time. There are numerical issues to be resolved due to
the complexities of the functions on both sides of the equals sign. A model problem is
presented subsequently. We fIrst present a brief discussion on the experimental
apparatus used and the kinds of data that are obtained.

5. MODEL PROBLEM

A key objective of the present work is to prove the concept of integrating detailed
experiments with reduced order analytical modeling. In this context, we chose a
500

geometrically simple model problem in which the fluid-structure interactions were fully
coupled. That is, flow excited structural motions that, in turn, modulated the flow.
Mathematically or experimentally complexities, such as strong three-dimensionality,
were deferred for future development.

1 (Diameter)
00

60 (Length)
00

~
~~'::ctiO~

Figure 2: Experimental Setup.

The model problem addressed in this study was the vortex-induced motion of a
low mass-ratio circular cylinder. The cylinder was restrained at its bottom end by a leaf
spring with freedom to move in the cross-stream plane only. A schematic drawing of
this model problem is shown in Figure 2. One can think of it as an inverted pendulum
 501

excited by its own periodic vortex shedding. As will be described in greater detail, the
amplitude of motion of the free, upper end was sufficiently small that the flow could be
considered to be nominally two-dimensional.
The physical model used in this study was a 2.54 cm diameter (D) cylinder
constructed of thin wall aluminum tube. It was 128 cm long and immersed in a uniform
flow of water, approximately 107 cm deep. The mass ratio was 1.53, the damping ratio
was 0.054, and the cylinder natural frequency, in' was 1.25 Hz. For a detailed
description of the cylinder and preliminary observations of the associated flow
dynarnics, the reader is referred to Atsavapranee, et al. [2].

6. APPLYING EXPERIMENTS TO THE REDUCED ORDER ANALYTICAL

MODEL

The flow measurements from an inverted oscillating pendulum experiment were

analyzed and a set of time traces were developed of three key fluid kinetic energy
transport terms, net kinetic energy flux, time rate of change of fluid kinetic energy, and
rate of work done by viscous forces. The plots shown in Figure 3 are precisely the fluid
'forcing' functions needed to analytically determine the motion of the cylinder. In this
section, we show how these data were applied to the governing equation and compare
the theoretical prediction of cylinder motion with the actual, experimentally measured
oscillations.
The precise form of the equation of motion used in this analysis is Equation 6.
The equation was simulated using MA TLAB in which the fluid forcing terms,
appearing on the right hand side, were the experimentally determined functions
presented in Figure 3. The traces are labeled as follows: d (KE) fluid / dt = time rate
of change of fluid kinetic energy in the control volume, (KE) fluid flux = kinetic
energy flux across control surface, viscous work = rate of work done by viscous
forces, d (KE + P E) cylinder / dt = time rate of change of cylinder kinetic plus
potential energies, and pressure work = rate of work done by pressure forces, as
computed from the total energy. The system includes the interior structure and
surrounding fluid. Since the experimental data were necessarily provided in the form of
a discrete dataset with sampling points every fifteenth of a second, a fast Fourier
transform was performed on the data within MA TLAB. For this initial calculation, one
hundred terms in the Fourier transformed signals were retained. Subsequent detailed
502

analysis will be conducted to determine the minimum number of terms necessary to

accurately model the cylinder dynamics. The experimentally determined structural
displacement is given in Figure 4.
The experimental data was phase-averaged over many experimental trials to
remove as much of the noise as possible. The resulting dataset was the used as input to
the MATLAB program. The analytical output is given in Figure 5. This plot still
exhibits the flattened plateaus signifying that the numerical method is having some
difficulties in tracking the dynamics. The maximum response amplitude is within 20%
of the experimental results, and the response frequency compares favorably to the
observed frequency. There is also some evidence of the observed beating behavior, but
some of the characteristics are obscured by the artificial plateaus.
The singularity observed in the model results not from a problem of existence,
but from one of uniqueness. That is, the equations of motion for this system have a
constant solution when the velocity of the tower is zero. This solution is not physical,
but the numerical solvers within MA TLAB seem to favor the constant solution. The
possibility of including an artificial kick from the nonphysical constant solution needs
to be investigated.
 503

d(KE).,..idt
(KE~wl tlux
1500 viscous work

d(KE+PE)cylmd./dt

pressure work

1000

t::
0
0..
en 500
=:
~
E-<

0/}
....
~ 0
=:
~

-500 \ '
\,1 \ i
'/
ii
, 1
\,
'J
i,i \
'/
'
\
\1
,
V ' /
\ '
'/ \
I' t
"
-1000
0 1 2 3 4 5 6 7 8 9 10
t (sec)

Figure 3: Phased-Averaged Energy Terms in the Integral Energy Transport Equation/or the
Given Control Volume
504

0.8,-----,-----,-----,-----,-----,-----,-----,-----,------,----,

0.6

0.4

0.2

E
'-'
"<D 0
E
~
0.-0.2
'"
i5

-0.4

-0.6

-0.8

2 3 4 5 6 7 8 9
Time. sec

Figure 4: Experimental Cylinder Displacement

 505

0.8 r----,-----,----~----r_--_,----_r----,_----._--_,----~

0.6

E
<.>
co

'"~
iil
a.
'"
i5
-0.2

-0.4

-0.6

-0.8 '---__---'-____-L.._ _ _ _- ' - -_ __ _-'---_ _----'_ _ _ _---'-_ _ _ _...L.._ _ _ _- ' - -_ _ ---'~ _ ____'

12 13 14 15 16 17 18 19
Time. seconds

Figure 5: Analytically Calculated Response, After Transients Have Disappeared

7. DISCUSSION

The work of McIver has been extended to model the oscillation of a structure in a fluid
flow. In the work presented here, it is assumed that the system configuration is not
prescribed at the end times. This led to a single equation governing the motion of the
structure as it is coupled to the fluid system. This equation has the units of power, and
is therefore a power balance between fluid and structure. Other, more general
approaches are under development.
The above theoretical developments rest heavily in a practical and literal sense
upon experimental input. The derived governing equations are semi-empirical,
requiring experimentally developed functions. However, this is certainly an explicit
trademark of all fluid mechanics, and is implicit in all of science and engineering. We
view it as a positive aspect of the model, that it is inexorably linked to physical data.
506

8. ACKNOWLEDGEMENTS

This work is supported by the Office of Naval Research Grant No.N00014-97-1-0017.

We would like to thank our program manager Dr. Thomas Swean for his interest and
fmancial support. We also thank Dr. Seon Han for her contribution of some of the
schematic figures.

9. REFERENCES

[1] R. Adrezin, P. Bar-Avi, and H. Benaroya. Dynamic response of compliant offshore structures - review.
Journal of Aerospace Engineering, 9(4):114-l31, 1996.
[2] P. Atsavapranee, A. Voorhees, H. Benaroya, and T. Wei. Lock-in regimes and vortex shedding modes on
a cantilevered cylinder. ASCE Eng'g Mechanics Division Conference, Baltimore, 1999.
[3] P. Bearman. Vortex shedding from oscillating bluff bodies. Annual Review of Fluid Mechanics,
16:195-222,1984.
[4] H. Benaroya and T. Wei. Hamilton's principle for external viscous fluid structure interaction. Journal of
Sound and Vibration, 238(1): 113-145, 2000.
[5] E. Berger and R. Wille. Periodic flow phenomena. Annual Review of Fluid Mechanics, 4:3l3-340,
1972.
[6] K. Billah. A Study of Vortex-Induced Vibration. PhD thesis, Princeton University, 1989.
[7] O. Griffin. Vortex streets and patterns. Mechanical Engineering, March:56-61, 1982.
[8] H. Gupta, P. Sarkar, and K. Mehta. Identification of vortex-induced response parameters in time domain.
Journal of Engineering Mechanics, 122(11):lO31-1037, 1996.
[9] R. King. A review of vortex shedding research and its application. Ocean Engineering, 4:141-172,1977.
[lO] A. Marris. A review of vortex streets, periodic wakes, and induced vibration phenomena. Journal of
Basic Engineering, pages 185-196, 1964.
[l1] D. McIver. Hamilton's principle for systems of changing mass. Journal of Engineering Mechanics,
7(3):249-261,1973.
[12] T. Sarpkaya. Vortex-induced oscillations: A selective review. Journal of Applied Mechanics, 46:241-
256,1979.
[l3] M. Zdrakovich. Different modes of vortex shedding: An overview. Journal of Fluids and Structures,
10:427--437,1996.
 ESTIMATON OF PRESSURE FIELDS USING
VELOCITY MEASUREMENTS

M. KRANE, P. DONG, T. WEI

Rutgers University, Piscataway, NJ

Abstract: An approach to estimating pressure fields due to unsteady flows associated

with flow- induced vibration is presented. Pressure estimates of this type are useful for
modeling the hydrodynamic forces on structures undergoing flow-induced vibration
(Paper F02). The approach is based on a boundary-element solution to the Poisson
equation for total pressure, using measurements of both the velocity field and the
motion of the structure as input. An outline of the problem formulation and numerical
issues will be presented. An application of the method to the case of flow-induced
vibration of a cantilevered cylinder will be presented. Measurements of the velocity
field in a finite control volume containing the vibrating cylinder are used to estimate the
pressure and pressure work both on the outer boundary of the control volume, as well as
on the cylinder surface. Estimates of the hydrodynamic energy input to the cylinder are
presented, with emphasis given to the separate contributions of cylinder motion and
fluctuations of the wake.

507
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 507.
2003 Kluwer Academic Publishers.
 IMPLICIT MULTIGRID COMPUTATIONS OF
UNSTEADY MULTIPHASE FLOWS IN VARYING
CROSS-SECTIONAL AREA CHANNELS

SECKIN GOKALTUN AND HASAN SAYGIN

Computational Science and Engineering Programme, Informatics Institute,
Istanbul Technical University, Maslak ,Istanbul, 34469
gokaltunse@itu.edu.tr, sayginh@itu.edu.tr

METIN MURADOGLU
Department of Mechanical Engineering
Ko; University, Rumelifeneri Yolu, Sanyer, Istanbul, 34450
mmuradoglu@ku.edu.tr

Abstract: Computations of two-dimensional multiphase flows involving complex

geometries are presented. The numerical method is based on the front-tracking
method in which one set of governing equations is written for the whole computa-
tional domain and different phases are treated as a single fluid with variable material
properties. The front-tracking (FT) methodology is combined with a newly devel-
oped finite volume (FV) solver based on dual-time stepping, alternating direction
implicit (ADI) multigrid method. First the FV solver is tested for a single phase flow
in a constricted channel and then the FV1FT method is used to compute a free rising
bubble in a straight channel and in channels with symmetric and antisymmetric
constrictions. Finally, some preliminary results are also presented for many bubbles
freely rising in the constricted channel.

Keywords: multiphase flow, front-tracking, dual-time stepping, complex geometries

1. INTRODUCTION

Accurate modeling of multiphase flows in complex geometries is of great

importance in many engineering applications and natural processes. In
particular, dynamics of bubbles or drops in capillary flows involving
complex geometries find wide range of applications in Micro Electro
Mechanical Systems (MEMS) devices and biological systems.
The presence of deforming phases makes the multiphase flow com-
putations a challenging problem and strong interactions between the
phases and complex boundaries add further complexity to the prob-
lem. Therefore the progress was rather slow and the computations
of multiphase flows have been usually restricted to simple geometries
and boundary conditions. Since nearly all multiphase flows of practical
interest involve complex geometries and boundary conditions, it is of

509
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 509-519.
2003 Kluwer Academic Publishers.
510

obvious interest to extend the modeling and computational techniques

to treat multiphase flows in complex geometries.
The front tracking method has been developed for simulations of
dispersed multiphase flows and is based on writing one set of equations
for the whole computational domain and treating different phases as
single fluid with variable material properties. In this method, the fronts
are explicitly tracked in a Lagrangian frame and the effects of surface
tension forces are accounted for by treating them as body forces to be
added to the momentum equations. Detailed description of the front
tracking method can be found in Trygvasson et al. (2001).
In the present work, the front tracking (FT) method developed by
Unverdi and Trygvasson (1992) is incorporated into a newly developed
finite volume (FV) method in order to facilitate efficient and accurate
modeling of multiphase flows involving complex geometries. The FV
method is based on the concept of dual (or pseudo) time stepping
(see Caughey, 2001) and is developed for unsteady computations of
incompressible viscous flows. The dual time-stepping method is based
on subiterations in pseudo time and has a number of advantages such
as direct coupling of the continuity and momentum equations in in-
compressible flow equations, the elimination of factorization error in
factored implicit schemes, the elimination of errors due to approxima-
tions made in the implicit operator to improve numerical efficiency,
the elimination of errors due to lagged boundary conditions at both
solid and internal fluid boundaries, and ability to use non-physical,
preconditioned iterative method for more efficient convergence of the
subiterations (see Caughey, 2001). The details of the present FV /FT
method are given by Gokaltun (2003).
First the accuracy of the base FV solver is verified for a single
phase, incompressible, unsteady flow in a converging-diverging chan-
nel and the results are compared with the solution obtained with the
Fluent 1 segregated solver. Then the FV /FT method is applied to a
single free rising bubble in a straight channel and the results are com-
pared with the finite-difference/front tracking method implemented in
FTC2D code of Traggvason et al. (2001). The method is then applied
to a single free rising bubble in channels with symmetric and antisym-
metric constrictions and to many bubbles rising in the channel with
symmetric restrictions under the sole action of buoyancy.

1 Fluent Inc., Lebanon, NH, USA

 511

2. MATHEMATICAL FORMULATION

Following Unverdi and Tryggvason (1992), the Navier-Stokes equations

are written for the whole flow field and different phases are treated
with variable material properties. The effects of surface tension are
modeled as body forces and are included in the momentum equations as
8-functions at the phase boundaries. In the Cartesian coordinates, two
dimensional time dependent Navier-Stokes equations for incompressible
flow can be written in conservative form as

-8q
8t
+ 8f
-
8x
+ 8g
-
8y
= -8fv + -8gv
8x 8y
-llpG + J
8(x - xf )a/'i,nds (1)

where

q = (p~) upu2 ) ,g = ( p+~vpv 2),

pv ,f = (p +puv (2)

fv = (T~X) ,gv = (T~Y) . (3)

Txy Tyy

and the viscous stresses are given for a Newtonian fluid as

8u 8V)
Txy = J-l ( 8y + 8x ' (4)

In Eqs.(2)-(4), u, v, p, p and J-l denote the velocity components in x

and y directions, the pressure, the density and the dynamic viscosity,
respectively. The third term on the right hand side of Eq.(l) repre-
sents the body force due to buoyancy with G being the gravitational
acceleration and IIp = Po - p, where Po is the density of the ambient
fluid. The last term represents the effects of the surface tension and
8, Xf, a, /'i" nand ds denote the Dirac delta function, the location of
the front, surface tension, twice of mean curvature, the outward unit
normal vector on the front and the front element, respectively.
The fluids in and out of the bubble are assumed to be incompressible
and the effects of heat transfer are neglected. Therefore the viscosity
and density remain constant in each fluid particle, i.e.,

Dp =0. DJ-l =0. (5)

Dt ' Dt
The flow regime of bubbly flows are characterized by four nondimen-
sional parameters as discussed by Clift et al. (1978). These are Morton
number, M= poU gJ.t~3' the Eotvos number, Eo = Po9d~,
U
the density ratio
'Y = E!s!.b.po and the viscosity ratio ( = I!P..,
J.tb
where de is the equivalent
512

bubble diameter and the subscripts 0 and b refer to the ambient and the
bubble fluids , respectively. The Reynolds number based on the bubble
rise velocity (terminal velocity) is defined as Re = p~de, where U is the
rise velocity.

3. NUMERICAL PROCEDURE

As can be seen in Eq.(1), the continuity equation is decoupled from

the momentum equations since it does not have any time derivatives
in incompressible flows . To circumvent this difficulty and to be able to
use time marching algorithms, pseudo time derivative terms augmented
with a preconditioning matrix is added to Eq.(1) as

aw aq af ag = -+--(po-p)G+
r -1 -+-+-+- afv agv
aT at ax ay ax ay
f 8(x-xf)0"/'i,nds
(6)
with
~ 00)
r- 1 = (~ pO, (7)
~ 0 p
where f3 is a preconditioning parameter with dimensions of velocity
and a is a dimensionless parameter to be determined. In Eq.(6), T
denotes the pseudo time and the dual-time stepping method is based
on marching in pseudo time until convergence for each physical time
step. Since the transient solution in pseudo time is not of interest,
we are free to use any non-physical convergence acceleration technique
such as preconditioning, local time-stepping and multigrid methods. To
facilitate treatment of complex geometries, Eq.(6) is transformed into
a curvilinear coordinates defined by

~ = ~(x, y); 'f} = 'f}(x, y). (8)

In the solution process, the physical time derivatives are approximat-

ed by a three point second-order backward implicit method and the
spatial derivatives are approximated by a cell-centered finite volume
method that is equivalent to a second order central differences on a
regular Cartesian grid. Time integration in pseudo time is achieved by
a diagonalized alternating direction implicit (DADI) method. A front
tracking method similar to that of Unverdi and Tryggvason (1992) is
developed for treatment of different phases and the surface tension.
The details of the numerical method can be found in Gokaltun (2003).
The complete solution procedure can be summarized as follows: In
 513

advancing solutions from physical time level n (t n = n .b.t) to level

n + 1, the location of front points at the new time level n + 1 are first
predicted using the explicit Euler method

(9)
where X f and Vf denote the position of front marker points and the
flow velocity interpolated from the neighboring fixed grid points onto
front marker points Xf, respectively. Then the material properties and
surface tension forces are evaluated using the predicted front position
X f , i.e.,

pn+l = p(Xfn+l); p,n+l = p,(Xfn+l); fb n+ 1 = fb(Xfn+l). (10)

The velocity and pressure fields at new physical time level (Un+l and
pn+l) are then computed by solving the flow equations by the FV
method and finally the front position is corrected as

X f n+l =Xf n + 2.b.t (Vf n + V f

n+l)
(11)

After this step, the material properties and the body forces are also
re-evaluated using the the corrected front position. Cubic B-splines are
used for all the interpolations from the fixed grid onto front points
and from the front points onto fixed grid as well as for distributing
surface tension forces onto fixed grid. The overall method is second
order accurate both in time and space. It is emphasized that the method
is implicit in physical time and the physical time step .b.t is solely
determined by accuracy considerations.
An auxiliary regular Cartesian grid is utilized for tracking of the
front positions in the curvilinear grids and is found to be very robust
and efficient. Details of the tracking algorithm can also be found in
Gokaltun (2003).
We note that, in addition to the preconditioning method, a multigrid
method similar to that of Caughey (1988) and a local time stepping
method are used to further accelerate convergence rate in pseudo-time
stepping.

4. RESULTS AND DISCUSSION

4.1. SINGLEPHASE INCOMPRESSIBLE CHANNEL FLOW

The accuracy of the base finite volume solver is first verified for a single-
phase incompressible flow in a channel with a symmetric constriction.
514

1.5

Figure 1. A portion of a 48 x 240 grid used for the constricted channel.

The channel dimension is 10m x 2m with the constriction at x = 5m.

The fluid in the channel is initially at rest and a uniform flow of 10mjs
is applied instantenously at the inlet section (x = 0). No-slip boundary
conditions are used at walls, pressure is fixed to Po = 100 Pa at outflow
and velocity is fixed to the uniform inlet profile of Uin = 10 mjs at the
inlet. Reynolds number based on the inlet velocity and channel height
is 50. The constricted portion of the channel and the computational
grid is shown in Figure 1. The dashed vertical lines shown in Figure
1 are used to mark the locations where the present calculations are
compared with the solutions obtained with Fluent's segregated solver.
The velocity profiles at the neck (x = 5 m) and at the downstream of
the constriction (x = 6 m) are plotted in Figure 2 together with the
Fluent results. As can be seen in this figure, the present results are in
a very good agreement with the Fluent results confirming the accuracy
of the base FV solver.

2 ..,..

1.75 1.75

'.5 '.5
o o. 0 o. 0 .0 00

'.25 o. 0 ~~l 1.25

H053inc 96byl92dt=O.OO5
g,
fluent SOby200 dt::O.OOI
...
0.75 0. 0 .0 II) 0.75
o. 0 o. 0 00 00

0.5 0.5

0.25 0.25

10 15 20 25 10 15 20 25
U(mI.) U(mlS)

Figure 2. Axial velocity profiles at axial locations. x = 5m (left plot) and x = 6m

(right plot). Empty circles denote the Fluent results and the filled squares denote
present results.
 515

Figure 3. Surface tension forces as distributed onto neighboring grid cell centers
near the front.(left plot) Velocity field in and around the rising bubble.(right plot)
Grid: 96 x 480; dt = 0.05s.

4.2. SINGLE BUBBLE FREELY RISING IN A STRAIGHT CHANNEL

The FV 1FT algorithm is first tested for a two-dimensional free rising

bubble in a straight periodic box of the size of 2de x lOde, where de
is the initial bubble diameter. No-slip boundary conditions are applied
at the side walls and periodic boundary conditions are used in the
vertical direction. The bubble is initially at rest at y = 2de and starts
rising due to buoyancy. The Eotvos and Morton numbers are specified
as Eo = 8.83 and M = 6 X 10- 2 , respectively. The surface tension
forces as distributed on the neighboring fixed grid cell centers and the
velocity vectors in and around the bubble are shown in Figure 3 at time
t = 0.5s.
The computation is performed on a 96 x 480 grid and the physical
time step is taken as Llt = 0.05s. The bubble is observed to be in the
ellipsoidal shape for this test case. The position of the bubble centroid
and the terminal velocity are plotted in Figure 4 together with the
results obtained with the FTC2D code of Thyggvason et al. (2001).
As can be seen in this figure, the present results are in a very good
agreement with the FTC2D results for this test case indicating the
accuracy of the present FV1FT method. Note that the same grid is
used both in the FTC2D code and in the present solutions.

4.3. SINGLE BUBBLE FREELY RISING IN VARIABLE CROSS SECTIONAL

CHANNEL

The results presented so far confirm the accuracy of the FV 1FT method
but the main focus of the present work is to show the ability of the
method for treating bubbly flows in complex geometries where the
bubbles strongly interact with the solid boundaries. The first example
516

Figure 4. The position ofthe centroid of the bubble (left plot). The terminal velocity
of the bubble (right plot). Symbols denote FTC2D results and solid lines denote the
present calculations. Grid: 96 x 480; dt = 0.05s.

concerns with the free rising bubble in a symmetrically constricted

channel as shown in Figure 5. In this figure, snapshots of the free
rising bubble in the channel are plotted at selected time frames of t =
O.ls, 2.0s, 4.0s, 5.0s, 6.0s, 8.0s, 9.0s. Here the governing non-dimensional
numbers are Eo = 9.81, M = 1.57, 'Y = 0.8 and ( = 1. The initial
and boundary conditions are the same as the free rising bubble in the
straight channel case. The ellipsoidal shape is seen prior and after the
constriction as in the straight channel case but large deformation of
bubble is observed while it passes through the constriction.

o 0 0 0 0 o~~~
o 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5

Figure 5. Motion of a free rising bubble through a symmetrically constricted chan-

nel. Eo = 9.81, Mo = 1.57, ,=0.8, (=1, t = O.ls, 2s, 4s, 5s, 6s, 8s, 9s. Grid: 96 x 480;
dt = 0.05s.
 517

~
-................ .....
"'", '''''''...

-
-0.5 0.9
....,..
~'"
..
... ,-"
-1 .- ........... ;,.
0.8

" ,, ......... gO.7

~-1.5 , ,, 0;
<D
c -2 .......
\ ' , ~O.6
:g
"g>-2.5 '. 'E 0.5
'. iD
13'" I-, ..... 48)(240 Grid 1 " *; 0.4
'"
-3 ,.. 96X480 Grid

"'" 3 .5 - 128X640 Grid , J5

.g 0.3
-E- '.",
~ -,
<D
0.2
<L
'.
-4.5 .......... " 0.1

-50 0
6 10 12 0 6 10 12
Time Time

Figure 6. Grid convergence analysis. Change in bubble area (left plot). The terminal
velocity of the bubble (right plot). dt = 0.05s.

...... dt - 0.1
.... , dt",O.05 """ dl _ 0.1

.
I ,..... cit _ 0.025 0.9 --, dt _ O.05
,.. ,.. dl _ 0.025
- dt _ O.012S
0.8 .#'-.... -
~, dl _ 0.0125

,l'
;,.
'g07 i ............... .
.,.,.....I
0;
,/ ~O.6

/' ~O.5

'/
iD
i.4
.gO.3
<D
0.2

0.1

~~~~--~---6~--~---'10~~'2 ~~--~--~--~6--~~~1~0--~,2
Time Time

Figure 7. Time stepping error convergence analysis. The position of bubble centroid
(left plot). The terminal velocity of the bubble (right plot). Grid: 96 x 480.

Although the flow is incompressible, the bubble area changes due

to numerical errors and this is shown in Figure 6a for different grids.
It is clearly seen that the change in bubble area gets smaller as grid is
refined. In Figure 6b, the terminal velocity of the bubble is plotted as
function of time for different grid sizes to show the grid convergence.
The position of the bubble centroid and the terminal velocity are plot-
ted in Figure 7 as function of time for various physical time step to
demonstrate the time stepping error convergence on a 96 x 480 grid.
Figure 8 illustrates the bubble-wall interactions more dramatically
as the free rising bubble passes through the passage formed by circular
blocks located anti-symmetrically along the channel. The boundary and
initial conditions and the material properties are the same as in the
symmetrically constricted channel case.
Finally snapshots of three identical bubbles rising in a symmetrical-
ly constricted channel are plotted in Figure 9 at various time frames
showing the strong interaction of bubbles with each other and with the
solid wall. It is noted that coalescence and break up are not allowed
518

18 18 18 18 18 18 18 18 18

16 16 16 16 16 16 16 16 16

14 14 14 14 14 14 14 14 14 a
12 12 12 12 12 12 12 0 12

Figure 8. Snapshots of the motion of a free rising bubble m an an-

ti-symmetrically constricted channel. Eo 9.81, Mo 1.57 and
t = Os, 4s, 8s, 12s, 14s, 16s, 18s, 20s, 22s. Grid: 128 x 1280, dt = 0.05s.

in the current implementation. Figure 9 shows that the grid should be

refined to better resolve the bubble-bubble and bubble-wall interactions
for this test case.

0
0
0
0
6
cD

0 4 00
00
3 000
0

a a a a a a a
o 0.5 1 1.5 o 0.5 1 1.5 o 0.5 1 1.5 o 0.5 1 1.5 o 0.5 1 1.5 o 0.5 1 1.5 o 0.5 1 1.5

Figure 9. Snapshots of the motion of three identical bubbles rising freely through
a symmetrically constricted channel. Eo = 9.81, Mo = 1.57, ,),=0.8, (=1 and
t = Os, 6s, lOs, 12s, 14s, 15s, 16s. Grid: 96 x 480, dt = 0.05s
 519

5. CONCLUSION

The computations of buoyant bubbles in variable cross sectional area

channels are reported. The newly developed FV method and the FV 1FT
method have been tested for standard single and multiphase flow cases
and the accuracy and efficiency has been demonstrated. It has been
shown that the present method can be used for computations of dis-
persed multiphase flows involving complex geometries. The future work
includes the generalization of the present method to axisymmetric and
three dimensional multiphase flows.

6. ACKNOWLEDGEMENTS

This work is supported by the State Planning Office of Turkish Gov-

ernment under the Contract Number 146. Computers at the Center of
Excellence of Istanbul Technical University were used for most of the
computations.

References

Caughey, D. A. Diagonal Implicit Multigrid Algorithm for the Euler Equations.

AIAA Journal, 26:841- 851, 1988.
Caughey, D. A. Unsteady Flows Past Cylinders of Square Cross-Section. Computers
and Fluids, 30:939- 960, 200l.
Clift, R. and Grace J . R, Weber M. E. Bubbles, Drops, and Particles. Academic
Press, 1978.
Gokaltun, S. Numerical Solution of Unsteady Incompressible Flows with Applica-
tions to Multiphase Flows. M.Sc. Thesis, ITU, Istanbul, Turkey, 2003.
Trygvasson, G., Bunner, B., Esmaeeli, A., Juric, D. , AI-Rawahi, N. , Tauber, W.,
Han, J., Nas, S. and -J. Jan, Y. A Front Tracking Method for the Computations
of Multiphase Flows. Journal of Computational Physics, 169:708- 759, 200l.
Unverdi, S. O. and Tryggvason, G. A Front-Tracking Method for Viscous,
Incompressible, Multi-fluid Flows. Journal of Computational Physics, 100, 1992.
 DYNAMIC STABILITY OF STOCHASTIC DELAY
SYSTEMS

M. S. FOFANA AND Y. YONG

Dynamics and Control ofManufacturing Systems
Manufacturing Engineering Program, Worcester Polytechnic Institute
Worcester, MA 01609, USA

Abstract: This paper is aimed at exannmng the influence of time delays and
multiplicative white noise on nonlinear dynamical systems that exhibit degenerate Hopf
interactions. Our desire for this study stems from the continued interest to understand
the stability bifurcations of machining systems with fluctuating regenerative
excitations. The excitations introduce time delays in the restoring and damping forces,
and may produce persistent modes of machining failures even at incipient time delays
prior to the fully growth rates of the excitations. Generally, systems involving time
delays are best described by delay differential equations, and the analysis of such
equations are accompanied with complications even when nonlinearities are omitted.
The consideration of multiplicative white noise adds a greater impediment to the
analysis. Frequently, the assumption of small delays as compared to unity is imposed in
order to reduce the infInite-dimensional character of the time delay problem to a fInite-
dimensional problem using conventional asymptotic techniques such as (i) the Taylor
series expansion, (ii) the integral averaging method, (iii) Fourier series, (iv) multiple
scale and (v) harmonic balancing. It is well known that the use of these asymptotic
techniques with the assumption of small time delays usually leads itself to results,
which do not reflect the long term stability bifurcations of the original time delay
system. Alternatively, we focus on the use of the centre manifold theorem and classical
theorem of Hopf bifurcation for the study of periodic solutions of dynamical systems,
with special attention that the long term stability bifurcations of the original nonlinear
stochastic delay systems is preserved, and moreover the time delays are not small. The
machining system considered is modeled as single degree of freedom and the equations
governing the motion contain multiplicative white noise, multiple time delays and
nonlinearity. The computation of the Poincare-Lyapunov coeffIcients, Floquet
exponents and moment Lyapunov exponents of the reduced delay systems will enable
us to determine suffIcient conditions for the possible Hopf interactions. The dependence
of the interactions on parameter variations will be captured qualitatively.

521
H. Benaroya and T. Wei (eds.), IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments, 521.
2003 Kluwer Academic Publishers.
 Mechanics
FLUID MECHANICS AND ITS APPLICATIONS
Series Editor: R. Moreau
46. U. Frisch (ed.): Advances in Turbulence VII. Proceedings of the Seventh European Turbulence
Conference, held in Saint-Jean Cap Ferrat, 30 June3 July 1998. 1998 ISBN 0-7923-5115-0
47. E.F. Toro and J.F. Clarke: Numerical Methods for Wave Propagation. Selected Contributions
from the Workshop held in Manchester, UK. 1998 ISBN 0-7923-5125-8
48. A. Yoshizawa: Hydrodynamic and Magnetohydrodynamic Turbulent Flows. Modelling and
Statistical Theory. 1998 ISBN 0-7923-5225-4
49. T.L. Geers (ed.): IUTAM Symposium on Computational Methods for Unbounded Domains.
1998 ISBN 0-7923-5266-1
50. Z. Zapryanov and S. Tabakova: Dynamics of Bubbles, Drops and Rigid Particles. 1999
ISBN 0-7923-5347-1
51. A. Alemany, Ph. Marty and J.P. Thibault (eds.): Transfer Phenomena in Magnetohydrodynamic
and Electroconducting Flows. 1999 ISBN 0-7923-5532-6
52. J.N. Srensen, E.J. Hopnger and N. Aubry (eds.): IUTAM Symposium on Simulation and
Identication of Organized Structures in Flows. 1999 ISBN 0-7923-5603-9
53. G.E.A. Meier and P.R. Viswanath (eds.): IUTAM Symposium on Mechanics of Passive and
Active Flow Control. 1999 ISBN 0-7923-5928-3
54. D. Knight and L. Sakell (eds.): Recent Advances in DNS and LES. 1999 ISBN 0-7923-6004-4
55. P. Orlandi: Fluid Flow Phenomena. A Numerical Toolkit. 2000 ISBN 0-7923-6095-8
56. M. Stanislas, J. Kompenhans and J. Westerveel (eds.): Particle Image Velocimetry. Progress
towards Industrial Application. 2000 ISBN 0-7923-6160-1
57. H.-C. Chang (ed.): IUTAM Symposium on Nonlinear Waves in Multi-Phase Flow. 2000
ISBN 0-7923-6454-6
58. R.M. Kerr and Y. Kimura (eds.): IUTAM Symposium on Developments in Geophysical Turbu-
lence held at the National Center for Atmospheric Research, (Boulder, CO, June 1619, 1998)
2000 ISBN 0-7923-6673-5
59. T. Kambe, T. Nakano and T. Miyauchi (eds.): IUTAM Symposium on Geometry and Statistics
of Turbulence. Proceedings of the IUTAM Symposium held at the Shonan International Village
Center, Hayama (Kanagawa-ken, Japan November 25, 1999). 2001 ISBN 0-7923-6711-1
60. V.V. Aristov: Direct Methods for Solving the Boltzmann Equation and Study of Nonequilibrium
Flows. 2001 ISBN 0-7923-6831-2
61. P.F. Hodnett (ed.): IUTAM Symposium on Advances in Mathematical Modelling of Atmosphere
and Ocean Dynamics. Proceedings of the IUTAM Symposium held in Limerick, Ireland, 27
July 2000. 2001 ISBN 0-7923-7075-9
62. A.C. King and Y.D. Shikhmurzaev (eds.): IUTAM Symposium on Free Surface Flows. Pro-
ceedings of the IUTAM Symposium held in Birmingham, United Kingdom, 1014 July 2000.
2001 ISBN 0-7923-7085-6
63. A. Tsinober: An Informal Introduction to Turbulence. 2001
ISBN 1-4020-0110-X; Pb: 1-4020-0166-5
64. R.Kh. Zeytounian: Asymptotic Modelling of Fluid Flow Phenomena. 2002
ISBN 1-4020-0432-X
65. R. Friedrich and W. Rodi (eds.): Advances in LES of Complex Flows. Prodeedings of the
EUROMECH Colloquium 412, held in Munich, Germany, 4-6 October 2000. 2002
ISBN 1-4020-0486-9
66. D. Drikakis and B.J. Geurts (eds.): Turbulent Flow Computation. 2002 ISBN 1-4020-0523-7
67. B.O. Eno and C.M. Hedberg: Theory of Nonlinear Acoustics in Fluids. 2002
ISBN 1-4020-0572-5
 Mechanics
FLUID MECHANICS AND ITS APPLICATIONS
Series Editor: R. Moreau
68. I.D. Abrahams, P.A. Martin and M.J. Simon (eds.): IUTAM Symposium on Diffraction and
Scattering in Fluid Mechanics and Elasticity. Proceedings of the IUTAM Symposium held in
Manchester, (UK, 16-20 July 2000). 2002 ISBN 1-4020-0590-3
69. P. Chassaing, R.A. Antonia, F. Anselmet, L. Joly and S. Sarkar: Variable Density Fluid
Turbulence. 2002 ISBN 1-4020-0671-3
70. A. Pollard and S. Candel (eds.): IUTAM Symposium on Turbulent Mixing and Combustion.
Proceedings of the IUTAM Symposium held in Kingston, Ontario, Canada, June 3-6, 2001.
2002 ISBN 1-4020-0747-7
71. K. Bajer and H.K. Moffatt (eds.): Tubes, Sheets and Singularities in Fluid Dynamics. 2002
ISBN 1-4020-0980-1
72. P.W. Carpenter and T.J. Pedley (eds.): Flow Past Highly Compliant Boundaries and in Col-
lapsible Tubes. IUTAM Symposium held at the Univerity of Warwick, Coventry, United
Kingdom, 26-30 March 2001. 2003 ISBN 1-4020-1161-X
73. H. Sobieczky (ed.): IUTAM Symposium Transsonicum IV. Proceedings of the IUTAM Sym-

posium held in Gottingen, Germany, 2-6 September 2002. 2003 ISBN 1-4020-1608-5
74. A.J. Smits (ed.): IUTAM Symposium on Reynolds Number Scaling in Turbulent Flow. Proceed-
ings of the IUTAM Symposium held in Princeton, NJ, U.S.A., September 11-13, 2002. 2003
ISBN 1-4020-1775-8
75. H. Benaroya and T. Wei (eds.): IUTAM Symposium on Integrated Modeling of Fully Coupled
Fluid Structure Interactions Using Analysis, Computations and Experiments. Proceedings of
the IUTAM Symposium held in New Jersey, U.S.A., 2-6 June 2003. 2003 ISBN 1-4020-1806-1

Kluwer Academic Publishers Dordrecht / Boston / London