Applied Mathematical Modelling 35 (2011) 1036–1053

Contents lists available at ScienceDirect

Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm

One-way and two-way couplings of CFD and structural models

and application to the wake-body interaction
Osama A. Marzouk *
Department of Engineering Science and Mechanics (Mail Code 0219), Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, USA

a r t i c l e i n f o a b s t r a c t

Article history: The current study focuses on the wake-body interaction of a circular cylinder, whose trans-
Received 20 May 2009 verse free vibration is modeled by a mass-spring-damper system coupled to a computa-
Received in revised form 22 July 2010 tional fluid dynamics (CFD) model for the flow and wake. We first simulate the free
Accepted 29 July 2010
vibration of the elastically-mounted cylinder and the wake, and analyze the transverse load
Available online 7 August 2010
it exerts on the cylinder and its phase with the vibration. We vary the damping by three
orders of magnitude and examine the difference in the wake-body interaction for
Keywords:
slightly-damped and highly-damped systems. We then use the spectral properties of the
Model coupling
Free vibration
free vibration and use them to construct two different types of forced vibrations: one con-
Forced oscillation sists only of the fundamental component of the free vibration, and the other accounts for
Circular cylinder all spectral properties of it. We compare the wake load for each type to that corresponding
Wake to the free vibration. The forced vibrations correspond to a one-way coupling and the
Finite difference information is communicated from the CFD model to the structural model, whereas the
free vibration corresponds to a two-way coupling of the models. By comparing the spectral
properties of the wake load, including the phase relation of its components with the vibra-
tion, which we obtained for the free vibration and for the equivalent forced vibration, we
identify the effects of the wake feedback. The findings show that a forced vibration does
not reproduce exactly the wake load at small and intermediate levels of structural damp-
ing. As the damping increases, the vibration changes from being in-phase with the wake
load to being 90° out-of-phase with it, corresponding to two different wake states, and
the forced vibration gives wake load that is very close to the one occurring in the case of
full wake-body interaction.
Ó 2010 Elsevier Inc. All rights reserved.

1. Introduction

A fluid stream over a streamlined body generates a wake region with periodic variations in the fluid motion and pressure
field. This quasi-steady state induces alternating loads on the body, causing it to vibrate if it is elastically mounted. This
vibration, in turn, affects the wake dynamics and re-shapes the wake loads. This wake-body interaction is a complex problem
where the body dynamics and wake dynamics instantaneously interact with each other. A circular cylinder is special case of
streamlined bodies because of its simple geometry, described by a single parameter, namely the diameter, thereby eliminat-
ing necessary parametric studies on the individual effects of multiple geometric variables. The galloping phenomenon does
not occur for a circular cylinder [1,2], unlike square and semicircular cylinders. Therefore, we can focus on one mode of linear
vibration, which is not interfered with other vibration modes. In addition, cylinder-like elements are used in many engineer-
ing applications, such as offshore structures and flue stack gases. For these reasons, we will consider an infinitely-long

* Fax: +1 801 515 0366.

E-mail address: omarzouk@vt.edu

0307-904X/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2010.07.049
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1037

circular cylinder as the body which is interacting with its wake, undergoing a transverse free vibration in the direction
orthogonal to both the free-stream and the span, as illustrated in Fig. 1. To model and simulate this problem, one needs
to couple a fluid model to a structural model and allow two-way communication between them through appropriate math-
ematical formulation and numerical implementation so that the free vibration of the cylinder affects the flow field through
the boundary conditions at the moving interface in addition to updating the grid details, and also the instantaneous wake
load on the cylinder is applied in the structural model. On the other hand, if the vibration is forced and prescribed in an ex-
plicit manner, such as being controlled by an external mechanism, this vibration is not affected by the wake load on it, but
the fluid model is still influenced by this forced vibration. Thus, the two-way coupling is reduced to a one-way coupling.
The two problems, free and forced vibrations of a circular cylinder placed in a free-stream, were studied computationally
and experimentally at different conditions of the free-stream and the elastically-mounted cylinder or the forced vibration.
When the forced vibration is considered, it is usually described as a simple sine function. Thus, it is described by a single
amplitude and a single frequency [3–10]. On the other hand, the elastically-mounted cylinder is attached, either physically
in experimental studies or virtually in computational ones, to a spring and a damper [11–16], and the free vibration is deter-
mined by the wake-body interaction. The mass, damping coefficient, and spring constant are the parameters of the structural
system to govern the free vibration, either in a dimensional or a non-dimensional form.
The majority of these studies focus on either a forced vibration or a free one. There are few studies that analyzed both
problems, but in an isolated manner, without attempting to establish a relation between them. For example, the authors
in Ref. [16], who simulated both problems in their study of transverse vibration, refer to each group of simulations as a
‘batch’, reflecting the lack of systematically linking the results of the two problems. The authors commented on a difference
between the two problems, which is that the free vibration is not always purely sinusoidal, whereas the forced one is. Also,
they indicated the existence of lock-in zone in the frequency of the forced vibration and a lock-in zone in the spring of the
elastically-mounted cylinder. However, this was conducted in a qualitative manner, and the relation between the two zones
was not studied.
An improved better step was taken by the authors in Ref. [17], who also studied computationally both free and forced
vibrations of a cylinder, because they considered a case of free vibration and applied a similar forced vibration and compared
the wake loads in the time domain for both problems. The loads ‘looked’ similar based on their figures. The drawback of this
analysis is that the time domain is not an efficient tool to analyze the similarities or differences in the wake loads, and one
needs to perform this analysis in the spectral domain because many differences in the spectral properties between two sets
of wake loads can be concealed in the time domain. Other than commenting that two sets of wake loads are similar, the
authors did not examine the level of similarity between them, by computing and comparing the frequency or the standard
deviations for example. In addition, the authors followed this procedure for a single case of free vibration and it is not clear
how the damping affects this ‘time domain’ similarity.
The computational brief study in Ref. [18] is even a better step toward analyzing the difference between free and forced
vibrations, in terms of the influence on the wake. In addition to constructing the forced vibration as a simple sine wave,
which was referred to as ‘pure-tone’, that is similar to the free vibration, the authors considered a multi-frequency forced
vibration that consists of a primary component added to one or two smaller components at other frequencies. Their objective
was to generate a transverse wake load that is similar to the one obtained in their simulation for a free vibration. The damp-
ing of the elastically-mounted cylinder was constant, but was not specified. The capability of the pure-tone forced vibration
to reproduce a similar wake to the one for the free vibration was judged based on a single variable only, which is the phase-
shift between the fundamental components in the wake load and the displacement of the moving cylinder. This is a serious
limitation in this study because it ignored many properties of the wake load, such as its magnitude and frequency. The pure-

Fig. 1. The mass-spring-damper system.

1038 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

tone forced vibration did not satisfactorily reproduce the phase-shift obtained for free vibrations; when the natural fre-
quency of cylinder was equal to the natural frequency of the wake, the difference was about 20°. The authors then applied
multi-frequency forced vibration, trying to achieve a better matching with the wake load in the case of free vibrations. They
considered three cases of multi-frequency forced vibration: one with three components and two with two components. For
the multi-frequency forced vibrations, the matching with the free-vibration wake load was judged based on its peak value in
the time domain and its phase-shift. A major concern in the analysis is that the wake load shown by the authors was very
irregular, thus one cannot rely on the peak when analyzing the matching of these loads; two equal peaks can correspond to
very dissimilar sets of wake load. None of the three multi-frequency vibrations was able to match the free-vibration wake
load. This could be because the free vibration the authors considered did not have a simple spectral composition but had
many components. The authors ignored all the components other than they described as being ‘of interest’, filtering them
out in the spectral domain. The authors did not explain why particular components were kept and others were ignored.
But this simply states that the exact free vibration was not reconstructed in the forced vibration, which can justify the
dissimilar results in the wake load. Based on our results in the current study, this alteration in the vibration can induced
serious effects in the resulting wake load, even if the eliminated components have small magnitude. A better approach
would be to consider free vibrations that have simple spectral composition and periodic form in the time domain, thus
can be accurately reconstructed as forced vibrations. We pay attention to this issue in the current study, so that dissimilar
wake-load results can be interpreted correctly. Another concern about the analysis of the authors is that when the
component ‘of interest’ in the free vibration was very small, their spectral analysis was unable to capture it. In this case, they
‘estimated’ the free-vibration component by solving the decoupled linear system of the elastically-mounted cylinder while
applying the corresponding component in the wake load as an explicit forcing term. This approach simply neglects the
two-way coupling between the wake and the vibration; it also does not automatically account for the phase relation of
the ‘estimated’ component with the other directly-resolved components. Adding this to the other concerns we mentioned
about the analysis of the authors and conditions of the vibration, this study does not reasonably help understand the
difference between forced and free vibrations in terms of the effect of the wake.
In the study in Ref. [19], it was shown that when the transverse free vibration is irregular, there is no robust phase-shift
between the transverse wake load and the vibration because this will be continuously changing in time. This was related to
the wake state by indicating that the unrepeatable vibration does not allow the wake to establish a consistent response with
robust phase-shift with the vibration. This phase-shift is one of the important variables that should be used to study the
wake-body interaction and the capability of an equivalent forced vibration to reproduce it. Therefore, we only consider peri-
odic free vibrations.
In the current computational study, we take advantage of the preceding observations and comments by other authors,
trying to avoid the weaknesses but combine and improved the strengths in the individual works, and apply this to examine
systematically and accurately the difference between the transverse wake load obtained for transverse periodic free vibra-
tion of an elastically-mounted cylinder in a free-stream and the wake load obtained for fully- and partially-equivalent forced
vibrations. We use several spectral quantities to measure the level of similarity. We accurately account for all the spectral
properties in the free vibration when constructing the equivalent forced one, so that we can establish meaningful conclu-
sions about the importance of the two-way coupling and to what level a one-way coupling can be equivalent. We intention-
ally, however, eliminate some of the spectral properties of the free vibration as another type of single-frequency forced
vibration, to examine the influence of the small, almost negligible, component in the free vibration. We initially paid atten-
tion to a slightly-damped system and a highly-damped system because the wake-body interaction are very different in these
cases, but we extended the study to other systems, with ten different damping values, to examine the continuous effect of
the damping on the wake-body interaction, and the performance of the partially- and fully-equivalent forced vibrations.

2. CFD model

We simulate the fluid motion and wake by integrating the nonlinear Navier–Stokes equations using the finite difference
scheme. The fluid is assumed to be incompressible and two-dimensional. Its velocity field, ~
u, and pressure field, p, are gov-
erned by the following divergence-free continuity equation and momentum equation in Euler description:
r ~
u ¼ 0; ð1Þ
@ 1
~
u þ~ u ¼  rp þ mr2~
u  r~ u; ð2Þ
@t q
where the density q and the kinematic viscosity m are constants. The continuity equation, Eq. (1), does not include a pressure
term. To avoid the computational difficulties resulting from the poor coupling between the continuity and momentum equa-
tions, Eqs. (1) and (2), we use the artificial-compressibility method [20–23] and improve the coupling between the pressure
and velocity fields by adding pseudo local time-derivative terms to both the continuity and momentum equations. In flux
vector form, and with arbitrary Lagrange–Euler description, the governing equations become
@~ @ @ ! ! @ ! !
^þ ~
q q þ ðE  Ev Þ þ ðF  F v Þ ¼ 0: ð3Þ
@t @s @x @y
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1039

! ! ! !
The vectors of unknown flow variable, ~
^ and ~
q q, the inviscid flux vectors, E and F , and the viscous flux vectors, Ev and F v are
0 1 0 1 0 1 0 1
0 p
!
^
qbu !
qb^v
^¼B
~
q
C
@ u A; ~ B C
q ¼ @ u A;
B C
E ¼ @ uðu  ug Þ þ p A;
B
F ¼@ uv
C
A;
v v uv v ðv  v g Þ þ p
0 1 0 1
0 0
! !
B C B @u @ v C
Ev ¼ m@ 2 @u@x A; F v ¼ m@ @y þ @x A;
@v
2 @@xv
@u
@y
þ @x
where ug and vg are the velocity components of the grid points in Cartesian x and y coordinates, respectively. The gravity acts
in the z direction, thus it does not appear in Eq. (2). The original continuity equation, Eq. (1), is elliptic in space, whereas the
modified continuity equation, first row in Eq. (3), is elliptic in space but hyperbolic in pseudo-time. The pseudo wave speed
qffiffiffi
^ provides a mechanism for propagating information throughout the domain and drives the divergence of the velocity to-
b
ward zero at each time step while solving the equations in the pseudo time s in an iterative manner.
The semi-implicit method for pressure-linked equations (SIMPLE) [24], the auxiliary potential function [25], the frac-
tional-step method [26], and the pressure-implicit with splitting of operators (PISO) [27]) follow a segregated approach in
which a continuity-based and a momentum-based equation are solved in sequence, which imposes lagging in the coupling
between the pressure, or the auxiliary potential function, and the velocity. Following the instantaneous approach, the system
of coupled differential equations in Eq. (3) is solved instantaneously, leading to a block tri-diagonal linear system to be
solved for all three unknown variables after discretizing the differential terms using the finite difference scheme [28]. This
better direct treatment of the inter-equation coupling is faced with increased complexity in the algebraic system to be solved
[29]. For two-dimensional problems, the cost of simulation is, however, not very challenging. We needed 2.23 s per time
step, when the simulation is performed on a single node of an SGI ALTIX 3700 supercluster, with 1.6 GHz CPU speed. We
typically run each simulation for 50000 time steps to obtain a sufficiently long interval of transient-free results that are suit-
able for accurate post-processing.
The value of b^ controls how fast the pseudo terms vanish and the original equations, Eqs. (1) and (2), are restored. Exam-
ining numerical studies that have employed the artificial-compressibility method shows that b ^ can vary by orders of mag-
qffiffiffi
nitude [30]. Some studies chose b ^ to be close to a dominant convective velocity in the problem [31]. However, there is no
^
universal optimum value for b. Rather, it depends on the problem configuration, grid resolution, and time step. We fix b ^ at
4U2, where U the free-stream velocity (Fig. 1), in all of our simulations. We examined different (smaller and larger) values of
^ and found that this value is very reasonable in terms of the rate of convergence. The convergence is considered to be
b
achieved if the divergence-free condition in the original continuity equation, Eq. (1), is met within a tolerance of
1.0E4 U/D, where D is the cylinder diameter. This is also an implicit condition on the vanishing of the pseudo pressure term
in the modified continuity equation. Whereas no explicit convergence criterion was set based on the vanishing of the pseudo
terms in the modified momentum equation, second and third rows in Eq. (3), the L1-norm of the vector pseudo term, @~ q=@ s,
was monitored and we found it decreases in a proportional manner with the numerical divergence. Therefore, it is sufficient
to set a convergence criterion based on numerical divergence.
The system of governing equations, Eq. (3), is first non-dimensionalized using the cylinder diameter D as a reference
length, the free-stream velocity U as a reference velocity, and qU2 as a reference pressure. It is then transformed into cur-
vilinear body-fitted coordinates and integrated over an O-type grid. The grid transformation and boundary conditions are

Fig. 2. The graded grid and boundary conditions in the physical domain (quarter domain is shown) and the corresponding uniform grid and boundary
conditions in the computational domain (entire domain is shown). Both grids are coarsened for clarity of their structure.
1040 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

Fig. 3. Radial grading (in terms of the non-dimensional spatial step in the radial direction) of the grid under three values of the grading parameter.

illustrated in Fig. 2. We show a quarter of the physical domain shown, which lies in the first x  y quadrant, and coarsened
form of the grids (45 radial lines and 40 rings) to avoid the blurred view of the full grid, where grid lines and structure are
indiscernible. In the physical domain, the grid has 200 points along the radial lines (the radial grid index, i, varies from 1 to
im = 200) and 225 points along the rings (the angular grid index, j, varies from 1 to jm = 225). The surface of the moving cyl-
inder is mapped onto the i = 1 ring, and the no-slip boundary condition is applied there using the information communicated
from the structural model. The outer boundary is mapped onto the i = 200 ring, where either an inflow or outflow boundary
condition is applied, depending on the currently-known direction of the fluid at this boundary. Over the inflow portion of the
outer boundary, the velocities are specified, u = U and v = 0, and the pressure is extrapolated along each j= constant line. Over
the outflow portion of the outer boundary, the velocities are extrapolated along each j = constant line and the pressure is set
to the free-stream value, p = P. Periodic boundary conditions are applied between the first and last radial lines, j = 1 and j = jm,
which are treated as contiguous to each other. As shown in Fig. 2, we use radial grading and the step size is nonuniform, with
D
r¼ þ aD½ekði1Þ  1; ð4Þ
2
where r is the radial coordinate and k is a constant factor that depends on three grid properties: the maximum radius, rm, the
grading parameter, a, and the number of radial grid points, im. Once these properties are chosen, k is computed from
   
r m =D  0:5
k ¼ exp ln þ 1 =ðim  1Þ : ð5Þ
a
The effect of the a on the radial grading is illustrated in Fig. 3, where we compare the change in the radial step size, Dr over
the i index, for three values of a, namely 0.1, 0.3, and 0.5. As a decreases, the radial resolution near the cylinder is improved,
allowing for good capturing of the steep near-wall gradients. However, very small a, such as 0.1, results in coarse resolution
at the outer boundary, which can affect the accuracy in applying the outflow boundary conditions. Taking both effects into
consideration, we set a = 0.3. For the chosen rm = 25D and im = 200, Eq. (5) gives k = 0.02219. We set jm = 225, as mentioned
before, to keep the cells faces nearly equal and avoid highly stretched cells in the entire physical domain, which can be no-
ticed in Fig. 2. Due to the free or forced vibration of the cylinder, all grid lines move accordingly except the i = im ring, which is
attached to the inertial frame. All other grid points will exhibit continuous motion in both the physical and computational
domains. In that case, the rings of the grid are no longer exact circles except the cylinder surface, i = 1, and the outer circle,
i = im. The grading formula in Eq. (4) is adjusted to adapt to these deformations, and a new grid is generated every time step,
using the vibration information communicated from the structural model. Under these conditions, rm and k need to be ad-
justed and they take different values for the radial lines, instead of being universal constants.
We use second-order central differencing in space for the viscous terms and an upwind scheme for the inviscid terms. We
use a second-order backward difference for the temporal derivatives, and implicit Euler differencing for the pseudo-temporal
derivatives. The algebraic system of equations is solved using the line-relaxation scheme. A constant time step, 0.02D/U, is
used; it yielded stable solutions for all simulations.

3. Structural model

3.1. Free vibration

Referring to Fig. 1, the vibration of the elastically-mounted cylinder in the y direction is subject to four different forces:
inertia force, damping force, restoring force, and wake lift force acting in the y direction per unit span, LW. Therefore, the
forced linear system governing the displacement Y is
d2 d
m 2
Y þb Y þ kY ¼ LW ; ð6Þ
dt dt
In Eq. (6), m is the mass per unit span, accounting for the apparent-mass effect, and b is the damping coefficient, accounting
for any sources of damping affecting the system. We indicated before that information from the structural model, in terms of
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1041

the displacement and velocity of the cylinder, needs to be communicated every time step to the CFD model in order to apply
the no-slip boundary condition and to reconstruct the grid and update the velocities of its points. On the other hand, the
wake lift force per unit span, LW, needs to be communicated at every time step to the structural model as indicated in Eq.
(6). This establishes the two-way coupling between the two models. The sketched spring and damper in Fig. 1 are virtual
and not modeled explicitly in the CFD model. In other words, they are transparent to the fluid, but their effects on the wake
are accounted for implicitly through the free vibration of the cylinder.
In non-dimensional form, Eq. (6) can be written as
2
d 4pf d  4p2 
Y þ Y þ 2 Y ¼ nC L : ð7Þ
dt
2 U r dt Ur

All variables and system parameters that appear in Eq. (7) are non-dimensional. Y* is the non-dimensional displacement of
the cylinder, t* is the non-dimensional time, f is the damping ratio, Ur is the reduced velocity, n is the mass parameter, and CL
is the lift coefficient. These variables and parameters are defined as

Y   Y=D; t   tU=D;
b U
f  pffiffiffiffiffiffiffi ; U r  pffiffiffiffiffiffiffiffiffiffi ;
2 km D k=m
qD2 LW
n ; CL  :
2m 0:5qU 2 D2

The system in Eq. (7) is integrated in the non-dimensional time using the fourth-order Runge–Kutta scheme.

3.2. Forced vibration

In the forced vibration, the structural model is no longer a dynamical system that communicates information from and to
the CFD model. Rather, the ordinary differential equation, Eq. (7), is replaced by an algebraic equation that prescribes the
vibration information, in terms of the displacement and velocity of the cylinder at every time step. Whereas this information
is still communicated to the CFD model, the structural model does not need the wake load to be communicated from the CFD
model. The functional form of the forced vibration is derived such that it reconstructs the free vibration, for a certain set of
parameters of the dynamical system. We consider two types of forced vibrations: one which is partially equivalent to the free
vibration by accounting only for its fundamental component, and another which is fully equivalent to the free vibration. The
equivalence is based on matching the free and forced vibration in the spectral domain. Therefore, the specification of the
forced vibration comes after the free vibration is computed and analyzed spectrally. Although the results have not been
shown yet at this point, they will show that the displacement of the elastically-mounted cylinder, can be described as a
two-term Fourier series as follows:

Y ¼ Y1 sinðfY t þ b1 Þ þ Y3 sinð3f Y t þ b3 Þ: ð8Þ

The first term is the fundamental component, and the second term is its third-superharmonic. There are five spectral prop-
erties in Eq. (8). The first three are associated with the fundamental component: the magnitude, Y1, the frequency, fY, and the
absolute phase, b1. The fourth and fifth spectral properties in Eq. (8) are associated with the third-superharmonic compo-
nent: the magnitude, Y3, and the absolute phase, b3.
The absolute phases of the free-vibration components, b1 and b3, are not of special interest in our analysis; what is impor-
tant is the phase-shift of their components, designated here by /Y3/Y1. We also would like to express Y3 as a fraction of Y1.
However, because we found that the magnitude ratio of the two components, Y3/Y1, is in the order of 0.0025, we use Y1/Y3
as a relative representation of Y3 when examining the sensitivity of the third-superharmonic of the free vibration to f. More-
over, we express fY as a fraction of the natural wake frequency, fWo, which is the computed shedding frequency in the wake of
the fixed cylinder when the wake is not affected by the vibration. Knowing that the wake frequency coincides with the fre-
quency of the wake load, we spectrally analyzed this load when no structural modeling is applied and computed the fre-
quency of the wake load, thus the natural wake frequency. With these steps, the free vibration is expressed as
"
     1    #
fY Y1 fY
Free Vibration : Y ¼ Y1 sin fWo t þ sin 3f Wo t þ /Y3=Y1 : ð9Þ
fWo Y3 fWo

Eq. (9) includes now four spectral properties, three of which are relative: fY/fWo, Y1/Y3, and /Y3/Y1. Relative spectral properties
are advantageous because they give insight into the relationship between the two components of the vibrations, and their
relation to the natural wake. There is still one absolute property, Y1.
Based on the free-vibration expression in Eq. (9), we apply two types of forced vibrations, corresponding to two levels of
reconstructing the free vibration. The first type of forced vibration contains only the fundamental component of the free
vibration. Therefore, we refer to this type as single-frequency forced vibration.
1042 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

    
fY
Forced Vibration; Single Frequency : Y ¼ Y1 sin fWo t : ð10Þ
fWo
The second type of forced vibration contains both components of the free vibration, thus we refer to it as two-frequency
forced vibration. It is described exactly by Eq. (9). The difference is that the four spectral properties are now input values
used to reconstruct the free vibration.
"
     1    #
fY Y1 fY
Forced Vibration; Two Frequencies : Y ¼ Y1 sin fWo t þ sin 3f Wo t þ /Y3=Y1 : ð11Þ
fWo Y3 fWo

The differences in the wake load between the case when the free vibration is taking place and the case when the equivalent
two-frequency forced vibration are ascribed to the change of the coupling approach from two-way to one-way. Moreover,
the differences in the wake load between the two types of forced vibration is ascribed to the influence of the third-super-
harmonic component in the vibration.
Because the accuracy of the spectral analysis is critical in the current study, we carefully select the time domain data to be
spectrally analyzed, by having a long interval of time so that the spectral resolution is high, Df/fY  0.018. We also adjust the
interval length so the implied periodicity in the Fourier analysis is satisfied and we avoid spectral leakage, which is reflected
in the very sharp spikes in spectral-domain results in Section 4.3.

4. Results

4.1. Specific models parameters

In the preceding analysis, several parameters were introduced. Before presenting the modeling results, we discuss here
the values or ranges assigned to these parameters. In this study, we set n = 1, so that CL acts directly as an excitation in the
dynamical system for the free vibration. We set Ur = 5.096, which is equal to U/(D fWo). We apply 10 values of f: 0.001, 0.003,
0.1, 0.03, 0.1, 0.18, 0.3, 0.42, 0.65, and 1. The minimum and maximum f correspond to a slightly-damped and a highly-
damped systems with different wake-body interaction as will be discussed in Section 4.3. The f values correspond to a nearly
equal spacing in the log scale, with the exception of 0.18, 0.42, and 0.65, which were added when we found steep changes in
one or more spectral properties of the free vibration. It should be mentioned that for an uncoupled forced damped system, it
is well-known that f = 1 corresponds to the critically-damped system, and the vibration is inhibited. However, this does not
apply strictly to the free-vibration system in Eq. (7) because it is not uncoupled, and vibration is sustained at f = 1, although
its amplitude is considerably reduced. The Reynolds number, Re  UD/m, is set to 200. At this value, a two-dimensional (2D)
CFD model was found to be in agreement to a three-dimensional (3D) one [32]. We note that 3D effects develop near
Re = 190 [33], but the alteration in the amplitude of the wake load, for a fixed cylinder at Re = 200 is small [34]. In addition,
the motion of the cylinder acts toward enhancing the two-dimensionality in the near wake [35]. Therefore, our CFD modeling
with two-dimensionality assumption is able to capture the actual wake states, yielding correct and accurate results. More-
over, because the wake and its load at this Re is very coherent, the resultant free vibration has simple response and the two-
component description in Eq. (9) is valid. In fact, 2D simulations have been used at even higher Re [36–39,15]. There are
other values of some parameters that will be used when performing test cases to validate the models and their coupling (Sec-
tion 4.2). These parameters will be stated in the relevant parts of the text.

4.2. Models validation

We present here three levels of validations: the first is for the uncoupled CFD model, the second is for the one-way cou-
pling of the structural and CFD models, and the third is for the two-way coupling of them. For the uncoupled CFD model, we
tested its prediction of the natural wake frequency at Re = 200, expressed in its non-dimensional value: the Strouhal number,

Table 1
Measured and computed St°.

Reference St°
Current, 2D 0.19625
Ref. [32], 2D 0.1957
Ref. [32], 3D 0.1936
Ref. [34], 2D 0.198
Ref. [40], 2D 0.195
Ref. [41], 2D 0.1944
Ref. [42], 2D 0.196
Ref. [43], exp. fit 0.195
Ref. [44]a, exp. 0.189
a
At Re = 215.
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1043

St°  fWoD/U. Table 1 is a listing of other predicted values at the same Re from independent 2D and 3D simulations and mea-
surements [32,34,40–44]. The current predicted value is in good agreement with the other listed ones.
To ensure that the one-way coupling is implemented correctly, we tested it for a case of forced vibration of the first type,
Eq. (10), at the same Re. The spectral properties of this test-case forced vibration are: Y1 = 0.15D and fYD/U = 0.3. The com-
puted lift coefficient, CL, is compared in Fig. 4 with the one computed by an independent 2D simulation [45] using the finite
volume scheme. The agreement between the two simulations indicates that we implement the one-way coupling success-

Fig. 4. Comparison of the lift coefficient, corresponding to a test-case forced vibration, from the current simulation using the finite difference scheme and
from an independent simulation [45] using the finite volume scheme.

Fig. 5. Comparison of a non-dimensional test-case vibration (in two-dimensions) from the current simulation using the finite difference scheme and from
an independent simulation [17] using the spectral element scheme.

Fig. 6. Instants of the velocity magnitude and direction in the near-wake region when the cylinder is undergoing free vibration with low damping
(f = 0.001) and high-damping (f = 1). Both instants correspond to the cylinder being at the zero-displacement position, moving upward.
1044 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

fully. We finally tested the two-way coupling for a case of free vibration at the same Re. For this test case, the system in Eq.
(7) has the following parameters: n = 0.05, f = 0.01, and Ur = 1/St°. The free vibration is in two dimensions, where another
dynamical system for the horizontal displacement, X, is coupled to the CFD model. This system has the same form of Eq.
(7), except that the drag coefficient, CD, replaces the lift coefficient, CL, in the coupling term on the right-hand side. The
non-dimensional free vibration pattern in the X/D  Y/D plane is compared in Fig. 5 with the one obtained from an indepen-
dent 2D simulation [17] using the spectral element scheme. The good matching is a testimony to the good performance of
the two-way coupling, even when the free vibration is not limited to one dimension.

4.3. Strong versus weak interaction

Depending on the damping, we found that the wake-body interaction will vary between two situations. In the first
situation, corresponding to a slightly-damped system in Eq. (7) and large-amplitude free vibration, the region of vortex
formation or the near-wake region is displaced away from the cylinder, and therefore the wake load to decrease in strength.
This scenario is reversed for a highly-damped system; as the vibration amplitude decreases, the displacement of the vortex-
formation region decreases, which intensifies the influence of the wake and its load on the cylinder. In the limiting case of
infinite damping and vanished vibration, the natural wake is restored and there is no longer wake-body interaction. There-
fore, we will refer to the first situation by strong wake-body interaction and to the second one as weak wake-body interaction.

D D

D D

Fig. 7. Comparison of the lift coefficient and the scaled non-dimensional displacement, in the time domain, between free vibrations with five levels of
damping (f from 0.001 to 1).
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1045

As a representative case of the strong interaction, we consider a low damping with f = 0.001. As a representative case of the
weak interaction, we consider a high damping with f = 1. Two instants of the velocity distributions in the near-wake for
these two situations are contrasted in Fig. 6. The vibration phase is the same in both wakes, where the cylinder is acceler-
ating upward from the position of zero displacement. The two contra-rotating vortices in both situations are located at the
small areas of minimum velocity magnitudes. As indicated in the figures, the location of these vortices is different for the
wake states.
We go deeper in contrasting the two situations of wake-body interaction through examining the wake load expressed in
terms of the lift coefficient, CL, the free vibration expressed in terms of the non-dimensional displacement, Y/D, and their
phase correlation. In the time domain, the superimposed CL and a scaled Y/D for five different values of f spanning the range
from 0.001 to 1 are compared in Fig. 7, for f = 0.001, 0.03, 0.01, 0.1, and 1. The scaling of Y/D was made such that its peaks and
valleys have comparable values of those of CL, isolating the effects of the changed strength of CL, which will be studied sep-
arately. The scaling factor increases with f due to the combined effect of having more-damped vibration and stronger wake
activity driving the cylinder. There are two noticeable differences between CL in the two interaction situations, correspond-
ing to f = 0.001 and f = 1. The first is related to the modulation in CL due to increased contribution of superharmonic com-
ponents, causing it to depart from a simple sine function. We will study this in more detail when we discuss the forced

Fig. 8. Comparison of the lift coefficient and the scaled non-dimensional displacement, in the time domain, between low-damping free vibration (f = 0.001)
and corresponding forced vibrations with one and two frequencies.
1046 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

vibrations. The second is related to the phase-shift between CL and Y/D, which can be crudely estimated from the temporal
shift in the peaks or valleys because both CL and Y/D have the same frequency due to the lock-in. Both differences decay con-
tinuously as the damping increases. From our spectral analysis, we found that the phase-shift at f = 0.001 is 0.7°, which in-
creases to 89.8° at f = 1. Fig. 7 also shows that, regardless of the damping, CL and Y/D are periodic, and the fundamental
component in CL leads the fundamental component in Y/D. Although Fig. 6 and visually inspecting the wake, are helpful
in monitoring the observable changes in the wake and flow motion in the two situations of body-wake interaction, we can-
not extract details about the lift force and its differences between the two situations from them. This force, and consequently
CL, comes from the integral of the pressure and shear stresses at the surface of the cylinder, thus it is determined by the state
of the flow motion very near from the cylinder before they are convected downstream. As an example, the two mentioned
differences in CL cannot be identified from Fig. 6. For this reason, we will limit the forthcoming discussion and results to CL
and its spectral properties, which can be computed accurately and used as quantitative measures of the wake equivalence
and for the free and forced vibrations. Besides, it is one of the objectives of the current study to investigate the changes in the
spectral properties of CL as the free vibration is damped and the wake changes between its two states.
After analyzing the free vibration in the spectral domain and extracting its spectral properties, as discussed in Section 3.2,
we construct and apply the equivalent two-frequency forced vibration, and the partially-equivalent single-frequency forced
vibration. These two types of one-way coupling will be compared here to the case of free vibration, for the two representa-
tive situations of wake-body interaction, by comparing the spectral properties of CL. Starting with the slightly-damped sys-
tem with f = 0.001, the time-domain results for CL superimposed on the scaled Y/D for the free and the two forced vibrations

Fig. 9. Comparison of the lift coefficient and the non-dimensional displacement, in the spectral domain, between low-damping free vibration (f = 0.001)
and corresponding forced vibrations with one and two frequencies.
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1047

are compared in Fig. 8. The peak values of CL for both types of forced vibration match those for the free vibration. However,
there is noticeably stronger CL modulation in the case of single-frequency forced vibration. Thus, the single-frequency forced
vibrations fails to reproduce the correct superharmonic components in CL. The same figure also indicates, although qualita-
tively, that the phase-shift between the fundamental components in CL and Y/D is reproduced well by both types of forced
vibration. In the spectral domain, the corresponding results for CL and Y/D are plotted in Fig. 9. This figure justifies the two-
component description of the free vibration in Eq. (9). It also shows the difference between the two types of the forced vibra-
tion, which lies in the existence of the third-superharmonic in Y. As suggested from the time-domain results in Fig. 8, the
magnitude of the third-superharmonic in CL, designated by L3, is much stronger in the case of single-frequency forced vibra-
tion as compared to the other vibrations. For this particular case, the ratio L3/L1 = 80.5% for the single-frequency forced
vibration, which is significantly larger than the 25% for both the two-frequency forced vibration and the free vibration. How-
ever, both types of forced vibrations share a common difference with the case of free vibration, which is the stronger com-
ponents in CL at 2fY, and the appearance of a component at 5fY. Noticing that the magnitudes of these components are nearly
the same for both types of forced vibration, despite the difference in their spectral composition, indicates that this difference
is due to the one-way coupling. In other words, this difference is directly related to inhibiting the wake to feedback to the
vibration and communicate mutually with it.

Fig. 10. Comparison of the lift coefficient and the scaled non-dimensional displacement, in the time domain, between high-damping free vibration (f = 1)
and corresponding forced vibrations with one and two frequencies.
1048 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

We now move to the highly-damped system with f = 1. CL and superimposed scaled Y/D in the time domain are compared
for the free and the two forced vibrations in Fig. 10. Unlike the case of low damping in Fig. 8, there is no identifiable differ-
ence in CL among the cases, which resembles a sine function for this situation of wake-body interaction. In the spectral do-
main, Fig. 9 shows that CL for the free vibration has only two components at fY and 3fY, and the extra smaller component at
2fY in Fig. 9 does not exist, and also L3 is much smaller now relative to L1. The ratio L3/L1 for the free vibration is only 1.3%, as
compared to 25% in Fig. 9. Looking at CL for the constructed forced vibrations, we find that L3 for the single-frequency forced
vibration is larger, L3/L1 = 3.7%, than the ones for the two-frequency forced vibration and the free vibration. This disparity in
CL with single-frequency forced vibration is similar to what we found for the low-damping case, but it is now mitigated.
There is an extra component in CL at 2fY for both forced vibrations, which does not exist in the free vibration. This component
has a trivial magnitude, but it is nearly equal for both types of forced vibration, which is a similar behavior to what we found
in the low-damping case.

4.4. Effect of damping

After contrasting the wake state, the lift coefficient, the free vibration, and the behavior of the two types of forced vibra-
tion for the two different situations of wake-body interaction with two values f, we consider here the range of f between the

Fig. 11. Comparison of the lift coefficient and the non-dimensional displacement, in the spectral domain, between high-damping free vibration (f = 1) and
corresponding forced vibrations with one and two frequencies.
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1049

two values, and how CL and Y and their phase-shift behave as f increases from 0.001 to 1. We start with the three relative
spectral properties of the free vibration, which appear in Eq. (9): fY/fWo, Y1/Y3, and /Y3/Y1. These properties are plotted in
Fig. 12 against f. Both Y1/Y3 and /Y3/Y1 have non-monotonous behavior and exhibit maximum values near f = 1. The smallest
value of Y1/Y3 is 379.2, at f = 0.001. Therefore, the third-superharmonic in the free vibration, and consequently in the two-
frequency forced vibration, is ‘negligible’ relative to the fundamental component. However, we showed that filtering out this
component, as in the single-frequency forced vibration, resulted in remarkable disparity in the corresponding CL at f = 0.001.
We now examine how this disparity changes with f. /Y3/Y1 is bounded by 77.2° and 91.7°. The remaining spectral property of
the free vibration, Y1, is plotted against f in Fig. 13 in a non-dimensional form, Y1/D. It decreases continuously with f, from
0.4922 at f = 0.001 to 0.1948 at f = 1.
The spectral properties of CL, plotted against f, for the free and two forced vibrations will be examined in the remaining
figures. The magnitude of the fundamental component, L1, is plotted in Fig. 14. Both types of forced vibration accurately
reproduce L1 values corresponding to those obtained for the free vibration. This accuracy is independent of f. The phase-shift
between the fundamental component in CL and the fundamental component in Y is designated by /L1/Y1, and similarly the
phase-shift of the third-superharmonic in CL is designated by /L3/Y1. Although there are other components in CL, these two

Fig. 12. The relative spectral properties of the free vibration, which are used to construct the forced vibrations.
1050 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

Fig. 13. The non-dimensional magnitude of the fundamental component in the displacement for the free vibration, which is used to construct the forced
vibrations.

Fig. 14. The magnitude of the fundamental component in the lift coefficient for free vibration and for corresponding forced vibrations with one and two
frequencies, over the range of applied damping (f from 0.001 to 1).

Fig. 15. The phase-shift in the fundamental component in the lift for free vibration and for corresponding forced vibrations with one and two frequencies,
over the range of applied damping (f from 0.001 to 1).

are the major ones as was shown in Figs. 9 and 11. The results of these two phase-shifts are plotted in Figs. 15 and 16, respec-
tively. /L1/Y1 is very similar for all cases of vibration and all f. On the other hand, there is considerable disparity in /L3/Y1 be-
tween the case of single-frequency forced vibration and the other vibration cases as the damping increases above f  0.4.
Although the case of two-frequency forced vibration results in a slightly shifted value of /L3/Y1 above the one obtained in
the case of free vibration, /L3/Y1 for the former case follows the same behavior with f as in the latter case. This behavior
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1051

Fig. 16. The phase-shift in the third-superharmonic component in the lift for free vibration and for corresponding forced vibrations with one and two
frequencies, over the range of applied damping (f from 0.001 to 1).

is described by a very slow increase with f, followed by a steep decrease above f  0.4. At f = 1, /L3/Y1 = 50.5° for the case of
free vibration, whereas it is 71.5° for the case of single-frequency forced vibration. Therefore, the inclusion of the third-
superharmonic in Y is important when constructing the forced vibration in order to capture the correct /L3/Y1. To complete
the analysis of the third-superharmonic component in CL, we examine the effect of f on the relative magnitude, L3/L1, in
Fig. 17. Given that the two types of forced vibration result in values of L1 that match well the ones obtained in the case
of free vibration, as indicated in Fig. 14, the results in Fig. 17 are effectively representing the ability of the two types of forced
vibration to reproduce L3. The two-frequency forced vibration successfully reproduces the L3 values obtained in the case of
free vibration, regardless of f. The single-frequency forced vibration fails to achieve this for all values of f, giving a larger L3/

Fig. 17. The magnitude ratio between the fundamental and third-superharmonic lift components for free vibration and for corresponding forced vibrations
with one and two frequencies, over the range of applied damping (f from 0.001 to 1).
1052 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053

L1 than those obtained for the other vibration cases. Whereas plotting L3/L1 can indicate that this overprediction of L3/L1 is
mitigated at high values of f, the consistent overprediction is in fact disguised by the increase L1 values. Therefore, it better
to switch to L1/L3 for high values of f, which is plotted also in Fig. 17. At f = 1, for example, L1/L3 for the single-frequency
forced vibration is 2.7-fold larger than its value for the other two vibration cases.

5. Concluding remarks

We simulated in the time domain a structural model of an elastically-mounted cylinder coupled to a computational fluid
dynamics model, and resolved the free vibration of the cylinder, the wake, and its load on the cylinder. The mass was kept
constant, corresponding to a unity mass parameter. The damping was varied by three orders of magnitude, changing the
phase-shift between the wake load and the free vibration, gradually, by 90° and altering the periodic flow motion in the
wake.
For each linear damped system that describes the free vibration, we extracted the spectral properties of the vibration,
which are the magnitudes and phases of the fundamental and third-superharmonic components. Although the magnitude
of the third-superharmonic component is about 400-fold less than the magnitude of the fundamental component, we found
that it is has important influence on the wake load at low damping. This influence, becomes less significant at high damping
but it still exists. A constructed forced vibration that does not contain this component failed to reproduce many details of the
wake load. On the other hand, a constructed forced vibration that accounts for both components of the free vibration yielded
a much better agreement with the wake load that is obtained under the wake-body interaction. This forced vibration, which
is considered to be identical to the free vibration, neglects the feedback from the wake on the body. This resulted in some
differences in the spectral properties of the wake load, mainly the existence of extra, small-magnitude superharmonic com-
ponents. As the system damping is increased, the magnitudes of these extra components fade away.

Acknowledgements

The author thanks the Advanced Supercomputing (NAS) Division at NASA Ames Research Center, and Professor Ali H.
Nayfeh at Virginia Polytechnic Institute and State University.

References

[1] G. Di Silvio, F. Angrilli, F.A. Zanardo, Fluidelastic vibrations: mathematical model and experimental result, Meccanica 10 (1975) 269–279.
[2] T. Sarpkaya, Fluid forces on oscillating cylinders, J. Waterw. Port Coast. Ocean Eng. 104 (1978) 275–290.
[3] W.F. Warren, An Experimental Investigation of Fluid Forces of an Oscillating Cylinder, Ph.D. Dissertation, The University of Maryland, Maryland, 1962.
[4] R.E.D. Bishop, A.Y. Hassan, The lift and drag forces on a circular cylinder oscillating in a flowing fluid, Philos. Trans. Royal Soc., Lond. Ser. A 277 (1964)
51–75.
[5] G.H. Koopmann, The vortex wakes of vibrating cylinders at low Reynolds numbers, J. Fluid Mech. 28 (1967) 501–512.
[6] H. Tanaka, S. Takahara, Study on unsteady aerodynamic forces acting on an oscillating cylinder, in: 19th Japan National Congress for Applied
Mechanics, Tokyo, Japan, 1969, pp. 162–166.
[7] R. Gopalkrishnan, Vortex-Induced Forces on Oscillating Bluff Cylinders, Ph.D. Dissertation, Department of Ocean Engineering, Massachusetts Institute
of Technology, Massachusetts, 1993.
[8] X. Lu, C. Dalton, Calculation of the timing of vortex formation from an oscillating circular cylinder, J. Fluid Struct. 10 (1996) 527–541.
[9] S. Krishnamoorthy, S.J. Price, M.P. Païdoussis, Cross-flow past an oscillating circular cylinder: synchronization phenomena in the near wake, J. Fluid
Struct. 15 (2001) 955–980.
[10] L. Kaiktsis, G.S. Triantafyllou, M. Özbas, Excitation, inertia, and drag forces on a cylinder vibrating transversely to a steady flow, J. Fluid Struct. 23 (2007)
1–21.
[11] C.C. Feng, The Measurement of Vortex-Induced Effects in Flow past Stationary and Oscillating Crcular and D-Section Cylinders, M.S. Thesis, Department
of Mechanical Engineering, University of British Columbia, Vancouver, Canada, 1968.
[12] F. Angrilli, G. Di Silvio, A. Zanardo, Hydroelasticity study of a circular cylinder in a water stream, in: IUTAM/IAHR Symposium on Flow-Induced
Structural Vibrations, Karlsruhe, Germany, August 14–16, 1972, pp. 505–512.
[13] I. Goswami, R.H. Scanlan, N.P. Jones, Vortex-induced vibration of circular cylinders. I: experimental data, J. Eng. Mech. 119 (1993) 2270–2287.
[14] A. Khalak, C.H.K. Williamson, Dynamics of a hydroelastic cylinder with very low mass and damping, J. Fluid Struct. 10 (1996) 455–472.
[15] J.B.V. Wanderley, G.H.B. Souza, S.H. Sphaier, S. Levi, Vortex-induced vibration of an elastically-mounted circular cylinder using an upwind TVD two-
dimensional numerical scheme, Ocean Eng. 35 (2008) 1533–1544.
[16] A. Placzek, J.-F. Sigrist, A. Hamdouni, Numerical simulation of an oscillating cylinder in a cross-flow at low Reynolds number: forced and free
oscillations, Comput. Fluids 38 (2009) 80–100.
[17] H.M. Blackburn, G.E. Karniadakis, Two- and three-dimensional simulations of vortex-induced vibration of a circular cylinder, in: Third International
Offshore and Polar Engineering Conference, Singapore, June 6–11, 1993, vol. 3, pp. 715–720.
[18] J.S. Leontini, M.C. Thompson, K. Hourigan, A numerical comparison of forced and free vibration of circular cylinders at low Reynolds number, in: 15th
Australasian Fluid Mechanics Conference, Sydney, Australia, December 13–17, 2004.
[19] H. Al Jamal, C. Dalton, The contrast in phase angles between forced and self-excited oscillations of a circular cylinder, J. Fluid Struct. 20 (2005) 467–482.
[20] A.J. Chorin, A numerical method for solving incompressible viscous flow problems, J. Comput. Phys. 2 (1967) 12–26.
[21] W.Y. Soh, J.W. Goodrich, Unsteady solution of incompressible Navier–Stokes equations, J. Comput. Phys. 79 (1988) 113–134.
[22] S.E. Rogers, D. Kwak, Upwind differencing scheme for the time-accurate incompressible Navier–Stokes equations, AIAA J. 28 (1990) 253–262.
[23] A.H. Nayfeh, F. Owis, M.R. Hajj, A model for the coupled lift and drag on a circular cylinder, in: ASME International Design Engineering Technical
Conferences & Computers and Information in Engineering Conference, Chicago, Illinois, September 2–6, 2003.
[24] S.V. Patankar, D. B Spalding, A calculation procedure for heat, mass and momentum transfer in three-dimensional parabolic flows, Int. J. Heat Mass
Transf. 15 (1972) 1787–1806.
[25] A.A. Amsden, F.H. Harlow, A simplified MAC technique for incompressible fluid flow calculations, J. Comput. Phys. 6 (1970) 322–325.
[26] J. Kim, P. Moin, Application of a fractional-step method to incompressible Navier–Stokes, J. Comput. Phys. 59 (1985) 308–323.
[27] R.I. Issa, Solution of the implicitly discretized fluid flow equations by operator-splitting, J. Comput. Phys. 62 (1986) 40–65.
 O.A. Marzouk / Applied Mathematical Modelling 35 (2011) 1036–1053 1053

[28] C.A.J. Fletcher, Computational Techniques for Fluid Dynamics, second ed., vol. 5, Springer-Verlag, Germany, 1991.
[29] H. Jasak, Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows, Ph.D. Dissertation, Imperial College, University
of London, 1996.
[30] F. Muldoon, S. Acharya, A modification of the artificial compressibility algorithm with improved convergence characteristics, Int. J. Numer. Methods
Fluids 55 (2007) 307–345.
[31] Y.P. Marx, Time integration schemes for the unsteady incompressible Navier–Stokes equations, J. Comput. Phys. 112 (1994) 182–209.
[32] B.N. Rajani, A. Kandasamy, S. Majumdar, Numerical simulation of laminar flow past a circular cylinder, Appl. Math. Model. 33 (2009) 1228–1247.
[33] C.H.K. Williamson, Vortex dynamics in the cylinder wake, Ann. Rev. Fluid Mech. 28 (1996) 477–539.
[34] H. Persillon, M. Braza, Physical analysis of the transition to turbulence in the wake of a circular cylinder by three-dimensional Navier–Stokes
simulation, J. Fluid Mech. 365 (1998) 23–88.
[35] H. Blackburn, R. Henderson, Lock-in behavior in simulated vortex-induced vibration, Exp. Therm. Fluid Sci. 12 (1996) 184–189.
[36] M. Braza, P. Chassaing, H. Ha Minh, Numerical study and physical analysis of the pressure and velocity fields in the near wake of a circular cylinder, J.
Fluid Mech. 165 (1986) 79–130.
[37] J.F. Ravoux, A. Nadim, H. Haj-Hariri, An embedding method for bluff body flows: interactions of two side-by-side cylinder wakes, Theor. Comput. Fluid
Dyn. 16 (2003) 433–466.
[38] H. Al-Jamal, C. Dalton, Vortex-induced vibrations using large eddy simulation at a moderate Reynolds number, J. Fluid Struct. 19 (2004) 73–92.
[39] J.S. Leontini, B.E. Stewart, M.C. Thompson, K. Hourigan, Wake state and energy transitions of an oscillating cylinder at low Reynolds number, Phys.
Fluids 18 (2006) 067101. 9 pp.
[40] O. Posdziech, R. Grundmann, A systematic approach to the numerical calculation of fundamental quantities of the two-dimensional flow over a circular
cylinder, J. Fluid Struct. 23 (2007) 479–499.
[41] O. Posdziech, R. Grundmann, Numerical simulation of the flow around an infinitely long circular cylinder in the transition regime, Theor. Comput. Fluid
Dyn. 15 (2001) 121–141.
[42] K. Herfjord, A Study of Two-Dimensional Separated Flow by a Combination of the Finite Element Method and Navier–Stokes Equations, Ph.D.
Dissertation, Department of Marine Hydrodynamics, Norwegian University of Science and Technology, Trondheim, Norway, 1996.
[43] R.D. Henderson, Nonlinear dynamics and pattern formation in turbulent wake transition, J. Fluid Mech. 352 (1997) 65–112.
[44] A. Roshko A., On the development of turbulent wakes, NACA Technical Note 2913, 1953.
[45] Z.C. Zheng, N. Zhang, Frequency effects on lift and drag for flow past an oscillating cylinder, J. Fluid Struct. 24 (2008) 382–399.