Available online at www.sciencedirect.

com

Computers & Fluids 38 (2009) 80–100

www.elsevier.com/locate/compfluid

Numerical simulation of an oscillating cylinder in a cross-flow at

low Reynolds number: Forced and free oscillations
Antoine Placzek a, Jean-Francßois Sigrist b,*, Aziz Hamdouni c
a
ONERA, The French Aerospace Lab, Aeroelasticity and Structural Dynamics Department, 29 Avenue de la Division Leclerc,
BP72, 92322 Châtillon Cedex, France
b
Service Scientifique et Technique, DCNS Propulsion, 44620 La Montagne, France
c
Laboratoire d’Étude des Phénomènes de Transferts Appliqueś au Bâtiment, Université de La Rochelle, Avenue Michel Crépeau,
17042 La Rochelle Cedex 1, France

Received 7 November 2006; received in revised form 15 April 2007; accepted 16 January 2008
Available online 13 February 2008

Abstract

A numerical simulation of the flow past a circular cylinder which is able to oscillate transversely to the incident stream is presented in
this paper for a fixed Reynolds number equal to 100. The 2D Navier–Stokes equations are solved by a finite volume method with an
industrial CFD code in which a coupling procedure has been implemented in order to obtain the cylinder displacement. A preliminary
work is first conducted for a fixed cylinder to check the wake characteristics for Reynolds numbers smaller than 150 in the laminar
regime. The Strouhal frequency fS and the aerodynamic coefficients are thus controlled among other parameters. Simulations are then
performed with forced oscillations characterized by the frequency ratio F = f0/fS, where f0 is the forced oscillation frequency, and by the
adimensional amplitude A. The wake characteristics are analyzed using the ti me series of the fluctuating aerodynamic coefficients and
their power spectral densities (PSD). The frequency content is then linked to the shape of the phase portraits and to the vortex shedding
mode. By choosing interesting couples (A, F), different vortex shedding modes have been observed, which are similar to those of the Wil-
liamson–Roshko map. A second batch of simulations involving free vibrations (so-called vortex-induced vibrations or VIV) is finally
carried out. Oscillations of the cylinder are now directly induced by the vortex shedding process in the wake and therefore, the time inte-
gration of the motion is realized by an explicit staggered algorithm which provides the cylinder displacement according to the aerody-
namic charges exerted on the cylinder wall. Amplitude and frequency response of the cylinder are thus investigated over a wide range of
reduced velocities to observe the different phenomena at stake. In particular, the vortex shedding modes have also been related to the
frequency response observed and our results at Re = 100 show a very good agreement with other studies using different numerical
approaches.
Ó 2008 Elsevier Ltd. All rights reserved.

0. Introduction tions with an industrial CFD code which could be used

later on a tube bundle configuration, like those existing
Flow around a fixed or oscillating cylinder has received in nuclear steam generators.
continued attention in the past few decades. In addition to The wake of a fixed circular cylinder exhibits a large vari-
being a building block in the understanding of bluff body ety of complex phenomena stemming from the diverse insta-
dynamics, it has a large number of applications in many bilities growing in the near wake. The classification of these
engineering situations. This study is the first step to inves- phenomena was primarily based on experimental measure-
tigate the feasibility of coupled fluid–structure computa- ments and therefore the limits describing the transition
between the different regimes were sometimes not exactly
*
Corresponding author. Tel.: +33 2 40 84 87 84; fax: +33 2 40 84 87 18. established. However a rather clear classification relying
E-mail address: [email protected] (J.-F. Sigrist). either on the evolution of the Strouhal number [8] or on

0045-7930/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compfluid.2008.01.007
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 81

the base pressure coefficient curve [48] is nowadays available. ther used to study industrial problems using a general
According to these classifications, the following regimes can approach based on computational fluid dynamics and com-
be highlighted: for Re [ 49, two stationary recirculation putational solid dynamics code coupling. It is therefore of
zones attached to the cylinder wall can be observed; then paramount importance to validate the fluid code and the
for 49 [ Re [ 190, the wake is still laminar and consists coupling procedure for such applications prior to perform
of two periodic staggered rows of vortices forming the numerical studies on real configurations. Validation is
well-known Von Kármán streets, the vortices of each row achieved by comparing the numerical results of our VIV
being shed alternately from either side of the cylinder. For simulations with other numerical studies and discussing
greater Reynolds numbers, the wake becomes three-dimen- the observed simulated phenomena. The paper is organized
sional (for 190 [ Re [ 260) and progressively turbulent. as follows: the numerical model used for the computations
This regime is followed by the shear-layer transition is first presented; numerical results obtained for fluid flow
(Re J 1200) where the separating shear layers become around a fixed cylinder are then briefly exposed before pre-
unstable and finally by the boundary-layer transition (Re senting the structure response and fluid flow pattern for
of order 105) which is associated to the drag crisis, i.e. a dra- forced and free oscillations of the cylinder.
matic decrease of the drag coefficient. Over all these regimes,
the flow exhibits a certain periodicity which is known as the
Strouhal frequency, denoted here by fS. When a periodic 1. Numerical model
vortex street is well established, this frequency corresponds
to that of the vortex shedding frequency; in other cases where 1.1. Resolution of the Navier–Stokes equations
the Von Kármán streets are not clearly visible, the frequency
can be defined as the one of the fluctuations of the stream- The flow field is governed by the Navier–Stokes equa-
wise velocity component for example. tions, which read for a Newtonian incompressible fluid:
In many applications, the cylinder is not fixed but oscil- 8
lates at a given frequency that could interact with the vor- < div u ¼ 0
tex shedding process. For forced oscillations in a certain ou 1 ð1Þ
: þ divðu  uÞ ¼  rp þ mDu
range of amplitude and frequency, the cylinder motion is ot q
able to control the instability mechanism which leads to
vortex shedding. One of the most interesting characteristics where u = (u v)T is the velocity vector with u and v being
of this fluid–structure interaction is the synchronization, or respectively the streamwise and transverse velocity compo-
‘‘lock-in”, between the vortex shedding and vibration fre- nents, p is the pressure, q and m are the fluid density and kine-
quencies. The vortex shedding frequency diverges from matic viscosity. As the Reynolds number does not rise above
that corresponding to a fixed cylinder fS and becomes equal 190, the flow is assumed to be laminar and two-dimensional
to the forced oscillation frequency f0. Similar phenomena according to the description of the flow regimes given by
are observed for vortex-induced vibrations (VIV): in this Williamson [48] for this range of Reynolds numbers.
case, the flow causes the cylinder to oscillate at its natural The Navier–Stokes equations are discretized using the
frequency fN which depends on the mass, the rigidity and finite volume technique, i.e. integral form of the conserva-
possibly the damping of the cylinder. The cylinder oscilla- tion equations are solved on control volumes which form a
tion frequency is thus different from the Strouhal frequency partition of the computational domain (see [13] for more
fS that would be obtained if the cylinder was supposed to details). Surface and volume integral approximations
be fixed. This phenomenon occurs over a certain range of require the values of variables at locations other than the
reduced velocities, where the vortex shedding frequency computational nodes. Indeed, the integrand involves the
becomes equal to the natural frequency and a peak of product of several variables (convective fluxes) or variable
amplitude is reached. gradients at those locations (diffusive fluxes). Interpolations
This complicated fluid–structure interaction phenome- are therefore performed to evaluate these fluxes with the
non still draws the attention of researchers and has become high-order MARS (Monotone Advection and Reconstruc-
the typical test case for new numerical techniques. A lot of tion Scheme) algorithm which has been developed especially
studies, involving Reynolds Averaged Navier–Stokes for the CFD code used here [46]. The MARS algorithm is a
(RANS) methods [39,18], Large Eddy Simulations (LES) second order conditional upwinding approximation devel-
[6,32,2,15], Direct Numerical Simulations (DNS) [11], oped in order to decrease the numerical dispersion. For
using finite volume or finite element [3,29,27] approxima- unsteady computations, the time dependent term is approx-
tions to solve the Navier–Stokes equations, can be found imated by an Euler scheme (implicit scheme).
in the literature for a large range of Reynolds numbers. The pressure–velocity decoupling is achieved by the
It is also crucial to check that the numerical computation SIMPLE (Semi-Implicit Method for Pressure-Linked
leads to the same phenomena than those observed in exper- Equations) algorithm for steady computations or the PISO
imental works [7,21]. (pressure implicit with splitting of operators) procedure for
The present study aims at performing numerical simula- unsteady cases. The pressure field is first predicted, then
tions of VIV using a general numerical tool that can be fur- corrected by several iterations so that the Poisson equation
82 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

for pressure and the momentum conservation equations for regards the surestimation of the drag coefficient and Strou-
velocity are satisfied. hal number when the Reynolds number is small.
Once Eqs. (1) have been discretized and the pressure– The mesh presented in Fig. 1b is block-structured and a
velocity decoupling has been realized, the problem is repre- ring of diameter D = 12d (represented with a dashed line in
sented by a matrix system composed of the cell-centered Fig. 1a) has been introduced around the cylinder to facili-
unknowns which has to be inverted. The resolution is per- tate the use of the moving mesh procedure when the cylin-
formed with the pre-conditioned conjugate gradient der oscillates. Grid independence tests have already been
method and provides the velocity components u and v, performed by Gerouache [16] for the same configuration
and the pressure p. The same CFD code and numerical and CFD code. The configuration of the blocks and the
techniques have already been used by Sigrist and Abouri number of control cells have therefore been built according
[42]. A complementary description of the whole resolution to the observations made in this study. Thus the total num-
method can also be found in this reference. ber of cells in the whole model used here stands at 28,800.
It should also be mentioned that a ratio has been intro-
1.2. Geometry and boundary conditions duced along the radial direction in the ring surrounding
the cylinder, so that the mesh is finer near the cylinder wall
The computational domain is represented in Fig. 1a and becomes coarser away from it (see Fig. 1b).
with the cylinder of diameter d. The size of the domain The boundary conditions adopted are specified in
and the position of the cylinder have been chosen accord- Fig. 1a. The inlet velocity U1 is chosen according to the
ing to simulations performed by Gerouache [16] with the cylinder diameter d and the fluid characteristics (q, m) to
same code for different lengths and heights. The results obtain the desired Reynolds number: Re = U1 d/m. When
are very sensitive to the size of the computational domain, the cylinder is moving, particular attention should be paid
particularly when the Reynolds number is small. Indeed, as to respect the cinematic coupling condition at the cylinder
the Reynolds number decreases, the flow is mainly gov- wall. Continuity of the velocities imposes the following
erned by the viscous effects whose region of influence varies equality for an unidirectional vertical motion:
with Re1. If the computational domain is not chosen wide
v ¼ y_ on C ð2Þ
enough to contain this influence region, the error caused by
the artificial boundary conditions disturbs the solution, where v is the vertical component of the fluid velocity, y_ is
even near the cylinder. For Re < 1, Lange et al. [25] advise the cylinder velocity and C denotes the cylinder wall. In or-
choosing H/D > 320Re0.8 to maintain the error smaller der to reduce the computational time, the moving mesh
than 1%. For greater Reynolds numbers [33] showed that procedure has been developed as a user subroutine so that
the aspect ratio should not be smaller than H/d = 22 for it operates only in the ring surrounding the cylinder and
Re = 100 and that the influence of the outlet boundary leads to a deformed mesh whose structure is preserved dur-
condition becomes negligible for L2/d P 34. The size of ing the oscillations [41].
the computational domain has been chosen here to keep
the number of control cells to a reasonable amount,
although the length and height are not conform to the val- 1.3. Cylinder motion
ues advocated by [33]. However, the configuration retained
(H/d = 20, L2/d = 20) provides acceptable results and the Two different techniques will be used in this study to
authors are aware of the blockage effects, particularly as obtain the cylinder motion. On the one hand, when forced

Fig. 1. Configuration of the computational domain and overview of the mesh used for the simulations. d is the cylinder diameter, x is the horizontal
direction, y the vertical one.
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 83

oscillations are imposed, the motion is known and can at the fluid–structure interface is not well evaluated. As a
therefore be directly imposed. On the other hand, when consequence, the coupling procedure is likely to cause arti-
the motion results from the vortex shedding process, the ficial instabilities (due for example to a negative numerical
displacement has to be computed from the fluid forces damping) leading to the divergence of the system. This
before it could be applied to the cylinder. numerical problem can be by-passed with the use of implicit
The forced oscillations of the cylinder are characterized procedures introducing subiterations for the same time step
by the frequency f0 and the maximal adimensional ampli- and leading to the convergence of the displacement and
tude A = ymax/d, where ymax is the maximal vertical fluid force at a given time step (see examples in
imposed displacement. A sinusoidal motion governed by [1,17,23,44]). Implicit coupling has been used for instance
the following equation is explicitly imposed to the cylinder by Sigrist and Abouri [42] who have noticed its superiority
in a dedicated subroutine implemented within the CFD for non-linear coupled problems (shock responses). Com-
code at each time step before solving the flow field: parisons between different coupling procedures (implicit
yðtÞ ¼ y max sinð2pf0 tÞ ð3Þ and explicit) have also been performed by Placzek et al.
[36] on a confined cylinder configuration to chose the best
The cylinder oscillates independently from the flow but the compromise between CPU time and accuracy.
wake can be strongly affected by the cylinder motion. The The comparisons performed by Placzek et al. [36] reveal
different wake regimes are classified thanks to the adimen- that for our application, the explicit algorithm adapted
sional amplitude A defined above and the frequency ratio from [41] (referred to as the blended procedure in [36]) pro-
F = f0/fS, where fS refers to the Strouhal frequency for vides satisfactory results, as it combines a good accuracy
the fixed cylinder. and a relatively small CPU time. The method used for
When vortex-induced vibrations are studied, the motion the time integration of Eq. (4) is based on an explicit algo-
corresponds generally to the cylinder bending mode. In rithm, better than the one tested by Sigrist and Abouri [42].
two-dimensions, the cylinder flexibility can be easily mod- Indeed, the numerical damping is dramatically reduced by
eled by a mass-spring system excited by the fluid forces. combining a centered upwind and downwind discretization
The vertical cylinder motion y(t) is consequently governed scheme for the prediction of the displacement. The steps of
by the equation of the following undamped oscillator: the coupling algorithm are the following:
m€y þ ky ¼ F y ð4Þ
1. Initialization for the first iteration (x0, v0, a0 and F0 are
where m denotes the cylinder mass, k is the rigidity of the the initial displacement, velocity, acceleration and
fictitious spring and Fy is the resultant of the lift force (ver- force):
tical component of the aerodynamic force), this latter being
y n ¼ x0 ; y_ n ¼ v0 ; €y n ¼ a0 ; F ny ¼ F 0 ð6Þ
a priori unknown. In absence of external forces, Eq. (4)
provides the natural cylinder frequency fN which will be 2. Explicit prediction of the cylinder acceleration for the
used later to drive the oscillations: time step tn+1 using Eq. (4):
rffiffiffiffi
1 k F ny k n
fN ¼ ð5Þ €y nþ1 ¼  y ð7Þ
2p m m m
The couple (m, k) could thus be chosen to represent the fre- 3. Evaluation of the cylinder velocity and displacement
quency of the bending mode, but when fluid forces are not with linear approximations (dt is the fluid time step,
null, the actual oscillation frequency f depends on them and h is the blending factor):
and is consequently different from fN in general. The reso- y_ nþ1 ¼ y_ n þ dt€y nþ1
lution of Eq. (4) requires the knowledge of the lift force ð8Þ
y nþ1 ¼ y n þ dt½ð1  hÞ_y n þ hy_ nþ1 
Fy(t), meaning that the flow field has to be computed be-
fore. In fact, this is a typical fluid–structure interaction 4. Mesh update: computation of the new mesh configura-
problem because the lift force influences the cylinder dis- tion (according to the displacement yn+1 evaluated in
placement y(t) which in turn modifies the flow field and the preceding step) with the moving mesh procedure.
therefore Fy(t), and so on. 5. Resolution of the Navier–Stokes equations with the
The resolution of such problems can be conducted rela- CFD code on the new mesh configuration to obtain
tively easily by using staggered procedures, i.e. the fluid and F ynþ1 .
the structure are solved successively for a given time step. A 6. Return to step 2 for the next time step.
detailed description of such algorithms has been proposed
for example by Piperno [34] or Farhat et al. [12]. Unfortu- The previous procedure is implemented in a coupling
nately, these methods lead inevitably to a time shift between subroutine handled by the CFD code at each time step
the fluid and the structure, which are not computed exactly before solving the flow field. This provides a numerical
at the same time step. The main drawback is that the cou- basis for more general code coupling procedures for future
pling rely on a prediction of the displacement which has applications. Several numerical tests conducted by Sigrist
to be as accurate as possible, otherwise the energy transfer [41] and Placzek et al. [36] have shown that h = 0.5 pro-
84 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

duces the smallest numerical damping in the coupled fluid/ and four others in the 2D periodic regime
structure calculation. Hence application of such algorithms (49 [ Re [ 190). The mesh remains stationary and several
to numerical simulations of VIV is straightforward; from a characteristic parameters of the wake are checked to con-
numerical point of view, this algorithm is not prone to alter trol the validity and the accuracy of the numerical model.
the simulation of fluid–structure energy exchanges since the
numerical damping is reduced. 2.1. Permanent regime
Finally, it is worth mentioning that in both cases (forced
and free oscillations), the cylinder motion is taken into Simulations are carried out until the convergence resid-
account by the fluid model which is written in an arbitrary ual becomes smaller than 107. The Reynolds numbers
Lagrangian–Eulerian (ALE) formulation (see [10]). This investigated are 10, 20, 30 and 40, all below the Hopf bifur-
formulation slightly modifies the Navier–Stokes equations cation between the permanent and the periodic regime, and
by introducing an additional term related to the mesh the wake is thus characterized by two recirculation zones
motion: the convective term in Eq. (1) is changed in attached to the rear cylinder wall. They are recognizable
div[u  (u  w)], where w is the mesh velocity field. Thanks by the strong vorticity f = $ ^ u and by low pressure levels,
to the moving mesh algorithm adapted from the one devel- as it can be seen in Fig. 2.
oped by Sigrist [41], the mesh velocity w is computed for The first parameter controlled is the length Lr of the recir-
each cell and the mesh can therefore be updated from the culation zone which is defined by the downstream distance
knowledge of the wall cylinder motion y. Combined with on the central line of the wake where the velocity is null. Val-
the fact that the size of the ring where the mesh moves ues of Lr are easily obtained by plotting the velocity versus
can be easily chosen so that the mesh distortion remains the distance after the cylinder wall for each Reynolds num-
small, this procedure which has already been used success- ber (Fig. 3a). In the same way, values of the separation angle
fully by Sigrist [41] and Sigrist and Abouri [42] turns out to hs are obtained by plotting the vorticity at the cylinder wall
be also efficient in the case of an infinite domain. against the angular position: hs corresponds to the angle
where the vorticity becomes null (Fig. 3b). The evolution
2. Fixed cylinder wake of these two parameters is presented in Figs. 4a and b. The
recirculation length is a linear function of the Reynolds num-
The wake of a fixed cylinder is investigated here for four ber and a least squares approximation provides the follow-
Reynolds numbers in the permanent regime (5 [ Re [ 49) ing expression: Lr = 0.0671Re  0.4155. The intersection

Fig. 2. Details of the vorticity contours (left) and pressure field (right) in the wake at Re = 40.

Re = 10 20 Re = 10
Re = 20 Re = 20
0.4 Re = 30 Re = 30
Adimensional Vorticity (ζ d/U∞)

Re = 40 Re = 40
Adimensional Velocity (u/U∞)

u=0 15 ζa = 0
0.3

0.2
10

0.1

5
0

–0.1 0

0 2 4 6 8 10 0 20 40 60 80 100 120 140 160 180

Adimensional Distance (x/d) Angular Position (θ)

Fig. 3. Plots used for the determination of the recirculation length Lr and the separation angle hs.
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 85

3 60

2.5 50

2 40

Lr/d 30

θs
1.5

1 20

0.5 10

0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Re Re

1
0.9
3
0.8
0.7
2.5
0.6

Ca
0.5
CD

2 0.4
0.3
1.5 0.2
0.1
1 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Re Re

Fig. 4. Comparison of the geometric wake characteristics and aerodynamic force coefficients for the permanent regime (Re < 49). () present simulations,
(h) [16], (M) [5], () [45], ($) [47], () [9], (+) [20,19], (/) [25].

of this line with the axis of abscissa defines the value of the to other studies like those of Lange et al. [25], Henderson
critical Reynolds number Rec from which the recirculation [20], Dennis and Chang [9] where the aspect ratio H/D is
zones appear. The value found here Rec = 6.19 is slightly greater. However, the agreement becomes better when Re
surevaluated compared to the one proposed by Gerouache increases and the relative errors on the drag coefficient re-
[16] (Rec = 5.74), and Socolescu [45] (Rec = 5.84). Values main inferior to 5% for each Reynolds number. Finally, the
of the recirculation length are close to the common values errors on the suction coefficient Ca are slightly larger but
obtained in other studies but the discrepancy between the still under 10%. This can be explained by the fact that Ca
results increases with Re. For the separation angle hs, the val- is a local variable which is, from the computational point
ues are also in good agreement with other studies. For these of view, very sensitive to numerical error.
two parameters, the relative errors (computed as the abso-
lute value of the difference between the value obtained with
the present simulation and the average of the values found 2.2. Periodic vortex shedding regime
in other studies divided by this same value) remain inferior
to 5% for the four Reynolds numbers tested. Transient simulations are now investigated. The simula-
Values of the drag CD and suction coefficients Ca are tion time is chosen long enough to observe about 15 vorti-
also controlled. As expected, the lift coefficient CL remains ces shedding once the periodic regime is established. The
null because of the perfect symmetry of the flow field. The time step is set to about 1/150 of the Strouhal period dur-
expressions of the aerodynamic coefficients are reminded ing the transient phase at the beginning of the computa-
below: tions. Once the periodic regime is reached, the time step
is then divided by 4 to increase the accuracy of the results.
FD FL p1  p0 The CPU time stand at about 26 h. To obtain the Von Kár-
CD ¼ CL ¼ Ca ¼ ð9Þ
1=2qU 21 d 1=2qU 21 d 1=2qU 21 mán streets, the symmetry of the wake has to be numeri-
cally broken thanks to a numerical artefact: 1%
FD (resp. FL) is the drag (resp. the lift) force by length unit, amplitude arbitrary noise is added to the incident velocity
p1 is the reference pressure and p0 is the pressure at the for that purpose. This perturbation is only maintained dur-
rear stagnation point. Values of the drag coefficient (see ing a small time interval.
Fig. 4c) are close to those of Gerouache [16] and Tuann The Reynolds numbers investigated now are 60, 80, 100
and Olson [47] who also used a small aspect ratio, but as and 120. In this case, the wake is composed of two stag-
mentioned before, they are slightly surestimated compared gered rows of vortices being shed alternately from either
86 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

Fig. 5. Vorticity contours in the wake over one Strouhal period at Re = 100.

side of the cylinder. Fig. 5 shows the vorticity contours in frequencies: the main frequency is equal to the lift oscilla-
the wake of the cylinder over a complete Strouhal period tion frequency and the secondary is identical to the drag
TS at Re = 100. At t = t0, a vortex is forming in the lower oscillation frequency.
side of the wake and is then completely detached from the The Strouhal frequency fS can be defined as the lift coef-
cylinder wall at t = t0 + 1/3TS. On the following snapshot, ficient frequency or the fluctuation frequency of velocity
the vortex of the upper side is about to be inserted between for any point in the near wake. The two frequencies are
the lower vortex formed previously and a new vortex which equal and the Strouhal number is obtained by the following
is forming. At t = t0 + TS, the vortex in the upper side is relation:
completely detached and the wake topology is exactly the
d
same as the one observed at t = t0. St ¼ fS ð10Þ
The periodicity of the shedding leads naturally to the U1
fluctuation of the aerodynamic coefficients which will be
The values of the Strouhal numbers obtained here are com-
denoted now by C 0D , C 0L and C 0a for the fluctuating values
pared in Fig. 7a to the analytical expressions of Fey et al.
and by C 0D , C 0L and C 0a for the mean values. These latter
[14], Norberg [30], Roshko [38], Williamson and Roshko
are evaluated as the average value of the fluctuating coeffi-
[49] and to other values from [16,25,20]. Although the evo-
cients over several periods chosen after the transient. The
lution of the Strouhal is well respected, the values com-
convergence of the coefficients is shown in Fig. 6 at
puted are slightly surestimated; however the relative error
Re = 100. The first part of the time series (t* < 10) exhibits
does not rise above 3.5%. This should one more time be re-
the transient phase during which the perturbation initially
lated to the small aspect ratio used here which could also
introduced arrives on the cylinder and causes the shedding.
explain the great values of the mean drag coefficient C 0D
The periodic state reached is characterized by the oscilla-
compared to those of [20] (see Fig. 7b). The mean value
tion of the drag coefficient at twice the lift frequency. The
of the lift C 0L is null and we therefore use the maximal value
fluctuations of the suction coefficient are governed by two
reached C 0L; max or the root mean square value C 0L;rms as rep-

1.6 0.4

1.5 0.3 1.2

Suction coefficient Ca

0.2 1
Drag coefficient CD

Lift coefficient CL

1.4
0.1 0.8

1.3 0 0.6

–0.1 0.4
1.2
–0.2 0.2
1.1
–0.3 0

1 –0.4 –0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
t*=t/TS t*=t/TS t*=t/TS

Fig. 6. Convergence of the aerodynamic fluctuating coefficients at Re = 100. From left to right: drag coefficient C 0D , lift coefficient C 0L and suction
coefficient C 0a .
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 87

0.19 1.5
1.48
0.18
1.46
0.17 1.44
1.42
0.16

CD
St
1.4
0.15
1.38

0.14 1.36
1.34
0.13
1.32
0.12 1.3
50 60 70 80 90 100 110 120 130 140 150 50 60 70 80 90 100 110 120 130 140 150
Re Re

0.45 0.9

0.4 0.85
0.35
0.8
0.3
0.75
0.25

Ca
CL

0.7
0.2
0.65
0.15
0.6
0.1

0.05 0.55

0 0.5
50 60 70 80 90 100 110 120 130 140 150 50 60 70 80 90 100 110 120 130 140 150
Re Re

Fig. 7. Comparison of the geometric wake characteristics and aerodynamic force coefficients for the periodic regime (49 < Re < 190). (d) present
simulations, (h) [16], (+) [20,19], (/) [25], (–) and (.) [48], (- - - -) [38], (–) [14]. For the lift coefficient, the maximal values of the present simulations () are
compared to the results of [16] (h); the rms values are represented on the same graph with empty circles (s) and compared to the analytical expression of
[30] (  ).

resentative parameters. The graph Fig. 7c presents the evo- Fig. 8 according to the limits established by Koopmann
lutions of the maximal lift coefficient C 0L; max compared to [24]. The lock-in zone is comprised between two limits
the values of Gerouache [16] and of the rms lift coefficient almost symmetrical with respect to the axis F = 1.0,
C 0L;rms . This latter is compared to the empirical expression although slight differences confirmed by the numerical sim-
derived by Norberg [30] and show a very good agreement. ulations of Anagnostopoulos [3] are observed for low
Finally the mean suction coefficient C 0a is compared to the amplitudes. In order to highlight the different response
results of [16,48,20] in Fig. 7d: the C 0a is slightly underesti- regimes of the cylinder, the amplitude A is kept constant
mated but the trend is respected. The accuracy of the re-
sults is rather satisfactory, especially concerning the
values of the Strouhal number and of the lift coefficient
which should be precise as they will play a major role in 0.3

the following study of the VIV. 0.25

0.2
3. Forced oscillations
A

0.15

Simulations are now performed for a cylinder forced to 0.1

oscillate at the frequency f0 which is better described by the 0.05
frequency ratio F = f0/fS. Computations are run for a Rey-
nolds number constant and equal to 100 from the solution 0
0.4 0.6 0.8 1 1.2 1.4 1.6
for the fixed cylinder, i.e. when the Von Kármán streets are F = f0 / fS
already present. Only the forced frequency f0 and possibly
the amplitude A are changed. The lock-in zone is defined Fig. 8. Representation of the lock-in zone in the plane (A, F) for forced
transverse oscillations. The solid lines represent the frontiers of the lock-in
by the domain where the vortex shedding frequency
zone according to the data points (+) of Koopmann [24]. The simulations
diverges from the value expected at the Reynolds number performed here are represented with squares: filled squares (j) correspond
considered and locks on the frequency of the forced oscil- to locked configurations whereas empty squares (h) indicate unlocked
lations: this zone is represented in the plane (A, F) in ones.
88 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

beyond a certain level and F varies over a range wide indicates that the aerodynamic forces are now governed by
enough so that the lock-in zone should be crossed. the forced frequency instead of the Strouhal frequency fS
determined above for a fixed cylinder. In the same way,
3.1. Cylinder response and lock-in zone PSDs of the drag coefficient - not represented here - would
show a main peak at f*  2.0, meaning that the forced fre-
The cylinder response is studied for several frequency quency also controls the drag fluctuations. This periodicity
ratios F between 0.50 and 1.50 while the amplitude A is can also be observed in the wake: visualizations of the vor-
kept constant and equal to 0.25. We present in the follow- ticity contours in the wake (not shown here) would be
ing two types of responses, the first in the lock-in zone and exactly the same at two instants separated by one period
the second out of it. For A = 0.25, the upper and lower lim- T0 = 1/f0.
its are approximately located at F = 0.75 and F = 1.25 The phase portraits of the system are also a very practi-
according to the frontiers established by Koopmann [24]. cal tool to analyze the response. Indeed, they represent the
energy transfer (product of the fluctuating lift force charac-
terized by the fluctuating lift coefficient C 0L and the adimen-
3.2. Locked configurations
sional cylinder displacement y* = y(t)/d) between the
motion of the cylinder and the fluid and thus provide an
Two cases are presented here to illustrate the locked con-
interesting description of how the system behaves. Phase
figurations: F = 0.90 and F = 1.10. Time series of the aero-
portraits for the two cases under study are therefore given
dynamic coefficients are characterized by a pure sinusoidal
in Fig. 9, col. 3 as a complement to the PSDs. The existence
response (see Fig. 9, col. 1) and exhibit a strong increase of
of a unique limit cycle is the result of the perfect undamped
the drag coefficient: indeed, for the fixed cylinder the mean
sinusoidal response and the inclination of the cycle gives an
drag coefficient was C 0D ¼ 1:37 (see Fig. 7b), whereas the
estimation of the phase angle between the imposed dis-
values reach now C 0D ¼ 1:50 for F = 0.90 and C 0D ¼ 1:75
placement and the lift.
for F = 1.10. The maximal value of the lift coefficient
(C 0L; max ¼ 0:33 for the fixed cylinder) is smaller for
F = 0.90 (C 0L; max ¼ 0:28) but increases then when F = 1.10 3.3. Unlocked configurations
(C 0L; max ¼ 1:44). The spectra of the lift coefficient presented
in Fig. 9, col. 2 highlight this sinusoidal response and clearly The lock-in region is defined like Nobari and Naredan
show that the main frequency is f0 since f* = f/f0  1.0. This [29] as the domain where the evolution of C 0L is purely sinu-

1 0.2
90
0.8 0.15
80
Power Spectral Density

0.6
70 0.1
0.4
60 0.05
0.2
50
CL′

CL′

0
0 40
–0.05
–0.2 30
–0.4 –0.1
20
–0.6 10 –0.15

–0.8 0 –0.2
0 2 4 6 8 10 12 14 16 18 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 –0.25 –0.2 –0.15 –0.1–0.05 0 0.05 0.1 0.15 0.2 0.25
* *
t = t/T0 f =f/f0 y*

1.5 0.8

3000 0.6
1
Power Spectral Density

2500 0.4
0.5
0.2
2000
CL′

CL′

0 0
1500
–0.2
–0.5
1000
–0.4
–1 500 –0.6

–1.5 0 –0.8
0 5 10 15 20 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 –0.25 –0.2 –0.15 –0.1 –0.05 0 0.05 0.1 0.15 0.2 0.25
* *
t = t/T0 f =f/f0 y*

Fig. 9. Time series of the fluctuating lift coefficient C 0L (first column), PSD of the fluctuating lift coefficient (second column) and phase portraits (third
column) for the cylinder forced to oscillate at F = 0.90 and F = 1.10.
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 89

soidal and governed by the forced oscillation frequency. forced oscillation period T0 and by definition of F we have
This means that only one peak at f* = 1.0 should be present TS = FT0 = 0.5T0. The case F = 1.50 (Fig. 10b, col. 1) is
in the frequency spectrum. A locked or unlocked wake is more complex: this time, the cycle-to-cycle oscillation is
therefore easily identified not only thanks to the PSDs, associated to the forced oscillation period T0 and the
but also and preferably with the analysis of the phase Strouhal period is TS = 1.5T0. The beating period TB is
portraits. now equal to 8T0.
The frequency ratios F = 0.50 and F = 1.50 are chosen The PSDs presented in Fig. 10, col. 2 for the cases
to illustrate this as they lead to an unlocked wake. Time F = 0.50 and F = 1.50 exhibit now two peaks. The first
series of the lift coefficient are no longer purely sinusoidal peak at f* = 1.0 corresponds to the forced oscillation fre-
and a beating behavior is observed: the signal is not peri- quency f0 and is still present on the two spectra but is alter-
odic over two successive cycles of oscillation but over sev- nately the main peak (F = 1.50) or the secondary one
eral ones. The time series are therefore characterized by a (F = 0.50). The second peak comes from the Strouhal fre-
cycle-to-cycle period, which is different from the ‘‘real” per- quency fS evaluated for the fixed cylinder and is therefore
iod, this latter being defined by the time interval after located at f*  fS/f0 = 1/F = 2 for F = 0.50 and f*  2/3
which the signal is exactly the same. The beating behavior for F = 1.50. The unlocked configurations are thus charac-
has also been observed numerically by Anagnostopoulos terized by the presence of the two frequencies f0 and fS,
[3] who noted that when the frequency ratio F was greater each one playing now a role in the vortex shedding process.
than 1 or smaller than 1 and outside the lock-in zone, the The main peak is always responsible for the cycle-to-cycle
flow was not absolutely periodic at subsequent cycles but oscillation whereas the low-frequency peak affects the peri-
a quasi-periodic flow pattern occurred. The presence of odicity of the signal. For the case F = 0.50, the beating is
more than one peak in the frequency spectra has also been enough energetic and is therefore identified on the spectra
observed by Nobari and Naredan [29] or Mittal and as the low-frequency peak at f* = 1.0 (Fig. 10a, col. 2):
Kumar [27] for transverse or in-line oscillations. For the beating frequency fB giving the periodicity of the time
F = 0.50 (Fig. 10a, col. 1), the cycle-to-cycle oscillation is series of the C 0L is thus approximately equal to f0. For the
associated to the Strouhal period TS but it is obvious that case F = 1.50, the second peak at f* = 2/3 corresponds to
the signal is not TS periodic. The beating period TB describ- the Strouhal frequency fS. In this case, the beating is less
ing the periodicity of the signal is in this case equal to the energetic in the sense that the difference of amplitude

0.5 0.5
0.4 400 0.4
0.3 350 0.3
Power Spectral Density

0.2 300 0.2

0.1 0.1
250
CL′
CL′

0 0
200
–0.1 –0.1
150
–0.2 –0.2
–0.3 100 –0.3
–0.4 50 –0.4
–0.5 0 –0.5
0 1 2 3 4 5 6 7 8 9 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 –0.25 –0.2 –0.15 –0.1 –0.05 0 0.05 0.1 0.15 0.2 0.25
t* = t/T0 f*=f/f0 y*

2.5 2
2 16000
1.5
1.5 14000
Power Spectral Density

1
1 12000
0.5 0.5
10000
CL′

CL′

0 0
8000
–0.5 0.5
6000
–1
1
–1.5 4000

–2 1.5
2000
–2.5 00 2
0 5 10 15 20 25 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 –0.25 –0.2 –0.15 –0.1 –0.05 0 0.05 0.1 0.15 0.2 0.25
*
t = t/T0 f*=f/f0 y *

Fig. 10. Time series of the fluctuating lift coefficient C 0L (first column), PSD of the fluctuating lift coefficient (second column) and phase portraits (third
column) for the cylinder forced to oscillate at F = 0.50 and F = 1.50.
90 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

between successive cycles is smaller (for F = 0.50, the differ- tion which was also the period of lift coefficient fluctua-
ence of amplitude was about 50% whereas now it is less tions. On the contrary, the behavior is more complex
than 25%). Combined with the fact that the time simulation outside the lock-in zone: according to the relative impor-
is possibly not long enough, the beating frequency is not tance of the peaks in the PSD (Fig. 10), it can be argued
detected by the spectral analysis represented in Fig. 10b, that the cycle-to-cycle period of the lift is governed by
col. 2. the Strouhal frequency below the lower limit of the lock-
Finally, the phase portraits (Fig. 10, col. 3) are dramat- in zone (resp. the forced oscillation frequency upon the
ically different from those obtained earlier. Indeed, they upper limit). However, the periodicity of the C 0L is associ-
highlight the presence of more than one frequency in the ated to a certain beating period TB which is a multiple of
signal which causes a fluctuation of the C 0L value between the cycle-to-cycle period and which corresponds to the
two successive cycles. These fluctuations are characterized periodicity of the vortex shedding process.
by a different path between two cycles and as a result, there
are many ways in the interior of the limit cycle. 3.4. Aerodynamic coefficients
Although the beating frequency fB is sometimes hardly
visible on the spectra (cf. the case F = 1.50), it plays an In addition to a shift of the shedding frequency, the
important role for the vortex shedding process. Even if it lock-in region is also characterized by an increase of the
is not visible on the spectra, the beating period TB = 1/fB aerodynamic coefficients, when compared to the fixed cyl-
can be evaluated by plotting the vorticity contours at differ- inder case. Fig. 12 represents the evolution of the mean
ent instants: the time interval between two identical snap- drag and maximal lift coefficient when the frequency ratio
shots provides an estimation of TB. Fig. 11 represents the F is increased and the amplitude A = 0.25 is constant. The
vorticity contours at t = t0, t = t0 + T0, t = t0 + TS, and evolution of the mean suction coefficient is not given here
t = t0 + TB. In the present case, we find TB = 8T0. This as it is quite similar to the mean drag coefficient, the max-
value agrees with the time series of the C 0L (Fig. 10b, col. imal value being obtained in both cases at F = 1.10. The
1) where it can be seen that the amplitude at two instants aerodynamic coefficients are practically always greater than
separated by TB is the same. It is clear from the vorticity the fixed cylinder values which are represented with dashed
contours that the periodicity of the wake is now governed lines. The shape of the drag curve Fig. 12a is characterized
by the beating period TB (the first and last pictures are by a maximum inside the lock-in zone and the drag coeffi-
exactly the same) instead of the Strouhal period TS or the cient seems to relapse near the fixed cylinder value outside
forced oscillation period T0 as it could be expected by see- of the lock-in zone. On the contrary, after a small drop, the
ing the PSD. Inside the lock-in zone, we observed that the lift is increasing from the beginning of the lock-in zone.
wake was governed by the period T0 of the forced oscilla- Outside the lock-in zone, this amplification seems to

Fig. 11. Vorticity contours at different strategic moments for F = 1.50.

 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 91

2 4
1.9
3.5
1.8
3
1.7
1.6 2.5

C′L,max
Mean CD
1.5 2
1.4
1.5
1.3
1
1.2
1.1 0.5

1 0
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
F F

Fig. 12. Evolution of the mean drag and maximal lift coefficients with the frequency ratio F. The dashed lines give the values of the mean drag C 0D and of
the maximum lift coefficient C 0L; max for the fixed cylinder case. The results are compared to those of Anagnostopoulos [3] (h) and Nobari and Naredan [29]
(M).

become less pronounced when F is further increased. The F = 1.00 and / = 47.7 for F = 1.10. A plausible reason is
computed curves are compared to the results of Anagnos- the difference in the Reynolds number between our study
topoulos [3] and Nobari and Naredan [29]. The shape of (Re = 100) and the one of Carberry et al. [7] where
the drag curve is globally similar with a maximum in the Re = 2300. Although it is not yet clear, it has been noticed
lock-in zone but the discrepancies between the values are by Khalak and Williamson [22], Anagnostopoulos [3],
important. Although there are some small differences con- Nobari and Naredan [29] that the change of vortex shed-
cerning the parameters of the model (Re = 106 for Anag- ding mode is most of the time not observed at low Rey-
nostopoulos [3], A = 0.2 for Nobari and Naredan [29]), nolds numbers, whereas it can be seen at higher Reynolds
the discrepancies stem certainly from the values obtained numbers [48].
for the fixed cylinder case: Anagnostopoulos found
C 0D ¼ 1:28, Nobari C 0D ¼ 1:72 and the present study yields 3.5. Vortex shedding modes
C 0D ¼ 1:37. This explains that the drag curve is comprised
between the two others, and recalling that the mean drag Attention has also been paid here to the topology of the
is slightly surestimated, one could expect that the curve wake. The beating phenomenon described previously also
would be shifted towards lower values if a greater aspect appears at (A, F) = (0.25, 1.25). It has been analyzed and
ratio had been used. Concerning the lift coefficient, the associated to a vortex merging mechanism for this couple
fixed cylinder case provided C 0L; max ¼ 0:33 whereas Anag- of parameters by Placzek et al. [35]. Then greater values
nostopoulos found C 0L; max ¼ 0:17. There is a factor 2 for the amplitude A have been used to observe different
between the two values and therefore the discrepancy shedding modes, which are commonly called the 2P and
between the curves Fig. 12b is very important. Despite P + S modes according to the appellations employed by
the great differences, the two curves exhibit an increase of Williamson and Roshko [49]. The values used to investi-
the lift coefficient inside the lock-in zone and a stabilization gate other shedding modes are deduced from the vortex
of the amplification near the end of the lock-in zone. shedding map proposed by Williamson and Roshko [49].
To conclude this paragraph, it is worth mentioning that First, the couple (A, F) = (1.00, 0.90) is chosen in the
a jump in the phase angle could be observed at the begin- middle of the 2P shedding mode domain. The wake looks
ning of the lock-in zone. Carberry et al. [7] find experimen- like a 2S shedding mode but the vortices are stretched in
tally at about F = 0.80 a jump of the lift coefficient and of the vertical direction because of the increased amplitude
the phase angle / defined as the shift between the cylinder of oscillation (see Fig. 14a). When the wake structure is
displacement and the fluctuating lift force. Below this observed in detail, it is possible to discern a secondary vor-
value, / is approximately equal to 180° and falls to 0° at tex denoted V2, whose intensity is very small. The vortices
the critical value of F. The authors attribute this jump to are still alternately shed from the upper and lower side of
a modification of the vortex shedding mode which appears the cylinder as in a 2S mode, but as they move down-
in the wake. In the present study, it is also observed that stream, the weakly energetic vortex V2 slips over its follow-
the lift coefficient increases when F increases but Carberry ing neighbor V3 and forms the tail of its preceding neighbor
et al. [7] found out that the steepest slope is at F = 0.80, V1. The vortices are then assembled by asymmetric pairs
whereas in the present case this happens later, between (V2–V3). Over one cycle period, two pairs are shed, so
F = 1.00 and 1.25. The phase angle has only been evaluated the vortex shedding mode looks like a 2P mode in which
in the lock-in zone where the fluctuations are purely sinu- the vortices would not be identical in a same pair. As the
soidal. The present simulations show a smooth decrease small vortex is weakly energetic, it quickly disappears in
of / with F and no real jump has been observed: the values the far wake. The vortex shedding mode in the far wake
found here are / = 113.8 for F = 0.90, / = 80.5 for looks therefore like a 2S mode with deformed vortices. It
92 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

1.5 7000 1.5

1 6000 1

Power Spectral Density

5000
0.5 0.5

4000
CL′

C ′
L
0 0
3000
–0.5 –0.5
2000

–1 –1
1000

–1.5 0 –1.5
0 2 4 6 8 10 12 14 16 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 –1 –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1
* *
t = t/T0 f =f/f0 y*

10 x 105
2 10
8 1.8
Power Spectral Density

6 1.6
1.4
4 5
1.2
CL′

C ′
2

L
1

0 0.8
0
0.6
–2
0.4
–4
0.2
–6 0 –5
0 5 10 15 20 25 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 –1.5 –1 –0.5 0 0.5 1 1.5
* *
t = t/T0 f =f/f0 y*

Fig. 13. Time series of the fluctuating lift coefficient C 0L (first column), PSD of the fluctuating lift coefficient (second column) and phase portraits (third
column) for the cylinder forced to oscillate at (A, F) = (1.00, 0.90) and (A, F) = (1.25, 1.50).

Fig. 14. Different vortex shedding modes obtained for (A, F) = (1.00, 0.90) (a) and (A, F) = (1.25,1.50) (b).

is not surprising to obtain a deformed shedding mode leads to a wake composed of two distinct rows of vortices.
instead of the regular 2P mode. As already mentioned, In the upper row, the vortices are grouped by pairs (P)
the reason comes certainly from the low Reynolds number whereas a single vortex (S) is shed in the lower row. The
used here. Indeed, it has been noted by Khalak and Wil- wake pattern is shown in Fig. 14b. The time evolution of
liamson [22] that the 2P mode is not observed at low Rey- the fluctuating lift coefficient C 0L and the frequency content
nolds numbers but the reasons why the wake is different can be linked to the vortex shedding regimes. Times series
remain obscure. Experiments presented by Ramberg and for the two cases (A, F) = (1.00, 0.90) and (A, F) =
Griffin [37] exhibited only the P + S mode for Re < 190; (1.25, 1.50) (Fig. 13, col. 1) do not exhibit any beating
the 2P mode in their laminar-regime studies has nonethe- behavior but the signal is modulated by a second frequency
less never been observed. over one cycle of oscillation. These additional frequencies
The couple (A, F) = (1.25, 1.50) located in the middle of appear on the PSD spectra shown in Fig. 13, col. 2. The
the P + S shedding mode area on the shedding mode map, main difference between the preceding cases at A = 0.25
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 93

is that these frequencies are now high frequencies which do when the vibrations are induced by the flow, the frequency
not affect the cycle-to-cycle periodicity but only the shape and amplitude responses being in this case not known a
of each cycle. The phase portraits (Fig. 13, col. 3) therefore priori.
exhibit only one path but the shape is not ovoid like for The vortex shedding process in the wake leads to fluctu-
locked configurations because of the presence of additional ating drag and lift forces which cause the oscillations of the
frequencies. The question here is to determine whether the cylinder. The phenomenon is self-limited: the fluid flow
wake is locked or not, since the spectra contains more than adjusts so that the oscillation amplitude is restricted to a
one frequency but the phase portraits exhibit only one certain upper limit. It has been observed by Mittal and
path. It could be argued that although several frequencies Kumar [27] that the various mechanisms by which the
exist, the wake is locked since the main frequency is always oscillator is able to self-limit its vibration amplitude are a
at f* = 1.0 and the additional high frequencies do not affect reduction in the amplitude of the aerodynamic forces,
the cycle-to-cycle periodicity. The major difference between appearance of additional frequency components in the time
the present cases and the preceding (at F = 0.25) concerns histories of the fluid forces and de-tuning of the vortex-
the position of the second peak relative to the first. For a shedding frequency from the structural frequency.
small amplitude of oscillation A = 0.25, the second peak Although the phenomenon has been observed for a long
is a low-frequency peak. As a result, a low-frequency beat- time, the maximal amplitude of vibration is not yet clearly
ing behavior is observed: the fluctuations of the lift are not defined. Indeed, the cylinder response depends on various
periodic over one cycle of oscillation but over several ones. parameters and particularly the mass-damping parameter
On the contrary, for high amplitudes (A = 1.25 and m*f, where m* represents the adimensional cylinder mass
A = 1.00), the secondary peak in the spectrum is a high fre- and f is the structural damping. Khalak and Williamson
quency peak. The fluctuations of the lift are therefore mod- give in their article the schematic amplitude response of
ulated by a high frequency signal which influences the the cylinder for high or low mass-damping parameters
response during one cycle. The fluctuations of the lift (see Fig. 3 of [22]). For high m*f, only two branches (called
remain thus periodic between two successive cycles, but initial excitation branch and lower branch) are observed,
over one cycle, a small fluctuation is observed. We suppose whereas the behavior at low m*f is more complex and
that this fluctuation could be related to the emission of the involves three branches (namely the two preceding
pair in the upper side of the wake. branches and an additional one called the upper branch).
The influence of the amplitude on the vortex shedding Each branch of the response characterizes a different
mode is crucial. Indeed, according to the values, high- or regime, where the adimensional amplitude of oscillation
low-frequency phenomena are observed. For small ampli- A is more or less pronounced. The determination of the
tudes, the low-frequency beating behavior in the time histo- maximal amplitude that can be reached is still an open
ries of C 0L has been linked by Placzek et al. [35] to a vortex question: while the maximal amplitude on the lower branch
merging in the wake for (A, F) = (0.25, 1.25). On the con- seems to be about A = 0.6, the maximum reached on the
trary, for high amplitudes ((A, F) = (1.00, 0.90) and upper branch is highly dependent on the mass-damping
(A, F) = (1.25, 1.50)), the appearance of high frequencies parameter. Khalak and Williamson suggest A = 1.20 as
in the spectra leads to the emission of a pair of vortices the maximal value but they find no signs of amplitude sat-
in the upper side of the wake, which could be related to uration when the mass-damping parameter was reduced to
the modulation observed in the time evolution of the lift extremely small values.
coefficient. For hydrodynamic applications like ours, the mass-
This first work has shown that the phenomena com- damping parameter takes generally low values and the cyl-
monly observed in the case of a cylinder forced to oscillate inder response should exhibit three branches. However m*f
in a transverse flow can be reproduced with our industrial is not only responsible for the number of branches in the
code. In view of the preceding results, the simulation of amplitude response. There is clearly an influence of the
vortex-induced vibrations seems to be feasible and results Reynolds number, especially at low values like in the lam-
for Re = 100 are presented in the sequel. inar shedding regime (Re  100). Attention has been drawn
before for forced oscillations on the fact that the 2P shed-
4. Vortex-induced vibrations ding mode was not observed at low Reynolds number. The
differences in the vortex shedding mode also exist for VIV
The case of an elastically mounted cylinder vibrating as and they prohibit the jump to the upper branch. This is
a result of fluid forcing is one of the most basic and reveal- confirmed by the 2D direct numerical simulation data from
ing cases in the general subject of bluff-body fluid–structure Newman and Karniadakis [28] for Re = 100 and 200 and
interactions. The first part of the work has been conducted the low Re experiments of [4] which did not show the upper
with the aim of illustrating and understanding the phenom- branch. The absence of the upper branch in the simulations
ena involved when the cylinder was subjected to forced may be due to the fact that the vortex shedding mode
oscillations whose characteristics were known, as well as obtained at low Re does not give a net energy transfer from
demonstrating the ability of the CFD code to capture the the fluid to body motion over one cycle, unlike the 2P mode
physics of VIV. Similar phenomena are now observed at higher Re [22].
94 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

The numerical results exposed in the following will be Table 1

compared to those of Shiels et al. [40] who also carry out Adimensional sets of parameters commonly used to classify the VIV
response regimes
simulations at Re = 100. Indeed the main part of their
results is presented in the case of an undamped oscillator Adimensionnal Guilmineau and Khalak and Shiels et al.
parameter Queutey [18] Williamson [22] [40]
like in the present study. To begin with the VIV study,
the various non dimensional parameters commonly used Amplitude A* y/d y/d y/d
Frequency f* f/fN f/fH f d/U1
to classify the response regimes of the cylinder are briefly Velocity U* U1/(fNd) U1/(fHd) U1/(2pfNd)
reminded before turning to numerical results. Then the Mass m* m/m m/m m/(0.5qd2L)
pffiffiffiffiffiffi
d pdffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
results of our simulations for an undamped cylinder at Damping f c= km c= kðm þ md Þ c/
Re = 100 are presented and compared to those of Shiels (0.5qU1dL)
et al. y is the vertical cylinder displacement, d is the cylinder diameter and L its
length, U1 is the inlet velocity and q is the density of the fluid, m, k and c
are respectively the cylinder mass, rigidity and damping, f is the actual
4.1. Adimensional parameters oscillation frequency of the cylinder, fN is the natural cylinder frequency
Eq. (5), fH is the natural frequency in water Eq. (12) and finally md is the
The mass-damping parameter m*f is adapted to deter- displaced mass of water Eq. (12).
mine whether the cylinder response is composed of two
or three branches, but for a given response shape (2 or 3
branches), it remains to determine the best parameter k ¼ k=ð0:5qU 21 LÞ is also introduced by Shiels et al. [40]
which collapses the different response regimes for a large to define a new representative parameter k eff called the
range of structural parameters. The reduced velocity U* ‘‘effective rigidity”:
(typically a reference velocity like U1 divided by the cylin-
k eff ¼ k  4p2 f 2 m ð13Þ
der diameter and a frequency like fN for example) has often
been used to plot the results but the width of the amplitude This parameter is always defined, even if the mass or the
peak varies with the mass of the system. A ‘‘true” reduced rigidity (or both) are null. Moreover, it collapses very well
velocity U*/f* used by Khalak and Williamson [22] seems the data for different structural parameters.
to collapse ideally the results over a wide range of cylinder
mass. The adimensional parameters used in different stud-
ies are thus quite diversified because of the different tech- 4.2. Response of the cylinder undergoing VIV
niques employed to make the cylinder equation
adimensional. The general equation of the oscillator with A set of simulations is performed without structural
structural damping is: damping and thus the movement is governed by Eq. (4).
The time integration is realized with the blended procedure
m€y þ c_y þ ky ¼ F y ð11Þ (see Eqs. (6)–(8)), but further simulations are currently car-
where m is the cylinder mass, c is the structural damping, k ried out with the implicit procedure already employed by
is the rigidity and Fy is the resultant of the lift force. The Abouri [1] and Sigrist and Abouri [42]. The behavior of
natural frequency fN of the cylinder (see Eq. (5)) and the the cylinder is summarized on the curves plotted in
frequency in water fH (defined below Eq. (12)) are often Fig. 15. The response regimes for the amplitude of oscilla-
used in the adimensionalization process: tion, the actual frequency of the cylinder and the aerody-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi namic coefficients are plotted against k eff .
1 k As expected, the amplitude response is only composed
fH ¼ where ma  md ¼ qpR2 L ð12Þ of the lower branch. The results show a very good agree-
2p m þ ma
ment with those of Shiels et al. [40], for each parameter.
The frequency in water fH depends on the added mass ma A ‘‘resonant” zone is observed in the range k eff  ½0–5,
which can be approximated here by the displaced mass of where the maximal amplitude of oscillation reaches
water md as the cylinder is situated in an infinite domain A = 0.58 at k eff ¼ 2:32 (Fig. 15a). The peak of amplitude
where the viscosity is small. Since no universal set of adi- is naturally associated with an increase of the aerodynamic
mensional parameters exists, the comparison of the results coefficients. Fig. 15b presents the evolution of the actual
between them requires to juggle with those used in different reduced frequency of the cylinder f* = f d/U1. Inside the
studies. For that purpose, Table 1 summarizes the different ‘‘resonant” zone, the frequency shifts from the Strouhal
sets of adimensional parameters commonly used in the lit- frequency fS and increases until a value corresponding to
erature and to which we will refer in the following to com- the natural frequency of the cylinder: the ‘‘resonant” zone
pare our results. is thus similar to the lock-in zone described previously for
The set of parameters used by Shiels et al. [40] (see the forced vibrations and the same name is therefore adopted
last column of Table 1) have been especially developed to to describe this zone. Finally, the mean drag and maximal
by-pass the problem of definition for the structural fre- lift coefficients are characterized by a maximum inside the
quencies fN and fH in extreme cases where k and/or m lock-in zone which is associated to the maximal amplitude
are null. A complementary adimensional rigidity response (Fig. 15c). Outside this zone, the aerodynamic
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 95

0.7
0.2
0.6
0.18
0.5 0.16
0.14
0.4
0.12
A*

f*
0.3 0.1
0.08
0.2 0.06
0.04
0.1
0.02
0 0
–5 0 5 10 15 20 –5 0 5 10 15 20
k*eff k*eff

1.8 2.5

1.6
1.4 2

1.2
1 1.5

CD,moy
CL,max

0.8
0.6 1
0.4
0.2 0.5
0
–0.2 0
–5 0 5 10 15 20 –5 0 5 10 15 20
k*eff k*eff

Fig. 15. Computed response (d) of the cylinder (amplitude A (a), actual frequency f* = fD/U1 (b), maximal lift coefficient C 0L; max (c) and mean drag
coefficient C 0D (d)) compared to the results of [40] (s).

coefficients relapse near the values corresponding to the is governed by the natural frequency of the structure
fixed cylinder case. Values of the lift for k eff < 0 have been instead of the Strouhal frequency fS for Re = 100. The adi-
set according to Shiels et al. [40] to the opposite of the mensional frequencies f* = fS/fH corresponding to the
C 0L; max value in order to indicate the existence of a phase Strouhal frequency at Re = 100 in the three cases are:
shift equal to p between the cylinder displacement and f* = 1.35 for k eff ¼ 0:05, f* = 0.82 for k eff ¼ 2:32, and
the C 0L . Without this artefact, the curve would have had f* = 0.51 for k eff ¼ 17:5. The preceding values indicate that
the same shape as the one of the reduced frequency f*. In for k eff ¼ 0:05 and k eff ¼ 17:5, the oscillations of the cylin-
the following, a detailed analysis of three cases der are driven by the Strouhal frequency, as the peaks high-
(k eff ¼ 17:5, k eff ¼ 2:32 and k eff ¼ 0:05) is exposed : the fre- lighted in Fig. 16 are located approximately at f* = fS/fH.
quency content, vortex shedding modes and lock-in zone On the contrary, the spectrum for k eff ¼ 2:32 reveals that
are studied and the phenomena observed are linked the lift is governed by the structural frequency fH since
together. the peak is located at f*  f/fH = 1.0.
Over a certain range of k eff , the cylinder response exhib-
4.2.1. Frequency content and phase portraits its an amplification of the amplitude and of the aerody-
Time histories of the cylinder displacement are now very namic coefficients. This amplification can be related to
interesting data, because the motion is not known a priori the synchronization of the cylinder frequency on its natural
as it was the case for forced vibrations. The graphs in the frequency: this is typical for the lock-in zone, which is
first row Fig. 16 highlight the difference of amplitude defined as the domain where the cylinder frequency shifts
between the different regimes for the cylinder displacement. from the Strouhal frequency and locks on the structural
In the lock-in zone, the amplitude is maximal but remains one.
limited. This is due to the self-limitation phenomenon men- The phase portraits of the oscillator are represented on
tioned before. The periodic regime is reached quickly: the third row Fig. 16. Unlike the forced vibration case,
about 10 oscillation cycles only are necessary to reach the the shape of the phase portraits is similar for each k eff either
maximal amplitude response. The PSDs of the fluctuating inside or outside the lock-in zone. The limit cycle is no
lift coefficient C 0L are plotted versus the frequency ratio longer ovoid but always resembles a double Lissajou figure.
f* = f/fH in the second row Fig. 16. For the intermediate Mittal and Kumar [27] observed a simple Lissajou figure in
case k eff ¼ 2:32, the peak at f*  1.0 indicates that the lift the case of in-line and transverse free oscillations but the
96 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

0.5 0.5 0.5

0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
y* = y(t)/d

y* = y(t)/d

y* = y(t)/d
0.1 0.1 0.1
0 0 0
–0.1 –0.1 –0.1
–0.2 –0.2 –0.2
–0.3 –0.3 –0.3
–0.4 –0.4 –0.4
–0.5 –0.5 –0.5

0 5 10 15 20 25 30 35 0 5 10 15 20 0 2 4 6 8 10 12 14
t* = t/Th t* = t/Th t* = t/Th

8000 3.5
800
7000

Power Spectral Density

Power Spectral Density

3
700
Power Spectral Density

6000
600 2.5
5000
500 2
4000
400
1.5
300 3000
1
200 2000

100 1000 0.5

0 0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
* * *
f =f/fh f =f/fh f =f/fh

1.5 1.5 1.5

1 1 1

0.5 0.5 0.5

C′L

C′L

C′L

0 0 0

–0.5 –0.5 –0.5

–1 –1 –1

–1.5 –1.5 –1.5

–0.4 –0.2 0 0.2 0.4 0.6 –0.4 –0.2 0 0.2 0.4 0.6 –0.4 –0.2 0 0.2 0.4 0.6
Y* Y* Y*

Fig. 16. Time histories of the cylinder displacement (first row), PSDs of the lift coefficient (second row) and phase portraits (third row) for VIV at
Re = 100. From left to right: k eff =17.5, 2.32 and 0.05.

same shape of phase portrait exhibiting a double Lissajou nation in the second quadrant for the last case because a
has also been obtained by Guilmineau and Queutey [18] phase jump occurred. One reason for the absence of phase
for transverse oscillations like here. The three phase por- jump in our study can be the low Reynolds number used
traits of Fig. 16 can be compared to those plotted by Kha- [18,22], which prohibits on the one hand the modification
lak and Williamson [22], see Fig. 13 in this reference. of the vortex shedding modes as it will be seen later and
Firstly, the global inclination of the phase portraits on the other hand the jump to the upper branch.
becomes progressively horizontal like here as k eff decreases
(or equivalently as U* increases). Secondly, the intermedi- 4.2.2. Vortex shedding modes
ate case located in the lock-in zone is obviously more reg- The vortex shedding patterns for the three cases under
ular in this study as the upper branch is not reached. study are illustrated in Fig. 17 with the vorticity field.
Indeed, Khalak and Williamson noted that the irregulari- The vortices are shed alternately from the upper and lower
ties appearing in the phase portrait were the manifestation sides of the cylinder, according to the classical Von Kár-
of the less steady dynamics of the upper branch. Finally, mán streets or 2S mode: no clear modification of the wake
the extent of the first and last portraits is like in our case has been observed over the range of k eff studied. The vor-
smaller than for the intermediate case. The absence of the tices are more stretched in the vertical direction at
upper branch in our case also explains why the global incli- k eff ¼ 2:32 than in the other cases. This slight modification
nation of the portrait is always in the first quadrant for the is certainly due to the increased amplitude of oscillation in
three cases: Khalak and Williamson [22] obtained an incli- the intermediate case, which deforms the wake in the verti-
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 97

Fig. 17. Vortex shedding modes inside and outside the lock-in zone for the cylinder undergoing VIV at Re = 100.

cal direction. Similar results have been obtained by Shiels oscillations like those used in forced oscillations. When
et al. [40], as it can be seen in Fig. 17a. oscillations are not purely sinusoidal, as it is the case in
For higher Reynolds numbers, a change between a 2S VIV for certain reduced velocities, the 2S, 2P, etc. struc-
and 2P mode is often observed [22,18] and can be linked tures may not become fully established because the ampli-
to the transition between the upper and lower branch of tude of oscillation and phase angle are both time
excitation. This transition has also been connected to the dependent. The lack of constancy in amplitude and phase
phase jump in the case of forced oscillations or VIV. How- angle could likely lead to the lack of repeatability in vortex
ever, at low Reynolds number such a change of the mode formation which certainly could suppress the standard
shape is not observed. Moreover, the numerical simula- mode patterns.
tions fail sometimes to capture this mode change, even The oscillations of the cylinder are not always purely
for high Reynolds numbers: in their study Al-Jamal and sinusoidal and a beating behavior similar to the one
Dalton [2] did not obtain the 2S and 2P modes although observed for forced vibrations is observed for example at
they use large eddy simulations at Re = 8000. The authors k eff ¼ 7:79: the vertical displacement is not periodic
argue that these modes are only observed for sinusoidal between two successive cycles (see Fig. 18a), but over sev-

0.2 0.08

0.15 0.07
Power Spectral Density

0.1 0.06
y* = y(t)/d

0.05 0.05

0 0.04

–0.05 0.03

–0.1 0.02

–0.15 0.01

–0.2 0
0 5 10 15 20 25 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
t* = t/Th f*=f/fh

Fig. 18. Desynchronization of the cylinder oscillations at k eff ¼ 7:79: the peaks of frequency progressively shift from the values f* = fS/fH = 0.68 and
f* = 1.
98 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

eral cycles. The spectrum (Fig. 18b) exhibits therefore 2.5

two peaks: the main peak is located at fm ¼ 0:77 and the

3 4
Re ∈ [10 – 10 ]
secondary peak at fs ¼ 0:90. It should be noted that
these two peaks are between f* = fS/fH = 0.68 and 2

f* = fH/fH = 1, the first value corresponding to the Strou-

hal frequency whereas the second is the natural cylinder 2
Re = 10
frequency in water. The beating behavior can thus be 1.5

understood as a desynchronization phenomenon: the cylin-

f*
Initial
der oscillation frequency progressively shifts from the nat- Branch
ural frequency fH, which is predominant at k eff ¼ 2:32, to 1
Lower
the Strouhal frequency which drives the wake at Branch
k eff ¼ 17:5. When k eff is increased, there is a progressive Upper
Branch
change of the main peak in the spectrum associated with 0.5

the end of the lock-in zone and which is the sign of the
response modification.
This beating behavior is often observed [22,27,40,2] and 0

has been related to a mode change, or a mode competition. 10

eff
Khalak and Williamson [22] have shown that this beating is

k*
associated which the transition between the lower and 0

upper branch of excitation, and therefore with the transi- –10

0 2 4 6 8 10 12 14 16
tion between the 2S and 2P modes. We do not have a sig-
U*
nificant mode change, but the beating is however connected
to the little modification of the vortex patterns between Fig. 19. Reduced frequency response f* = f/fH and lock-in zone. (d)
k eff ¼ 2:32 and k eff ¼ 17:5. As mentioned above, at low Present simulation, (s) [22]. The black line indicates the oscillation at fH
whereas the gray lines correspond to oscillations at the Strouhal frequency
Reynolds number the classical mode change 2S to 2P is for the range of Reynolds number considered. The subplot give the
not observed; however, a little modification of the wake correspondence between the parameters k eff and U* used respectively by
appears and a similar beating phenomenon happens, which Shiels et al. [40] and Khalak and Williamson [22].
is related to this vortex pattern change and to the progres-
sive transition between a cylinder response driven by the obtained at higher Reynolds numbers for which the Strou-
natural frequency fN to a response characterized by the hal number is slightly greater (the results of Khalak and
Strouhal frequency fS. Williamson [22] spread over the interval Re  [103; 104]
where fortunately the Strouhal number is nearly constant
4.2.3. Lock-in zone and will be approximated here by the average value
The lock-in zone can be identified in Fig. 15a as the region St = 0.21).
where an amplification of the oscillation amplitude is If we observe first the results of Khalak and Williamson
observed. This first representation focuses on the amplitude [22], it can be seen that for small or high U*, the cylinder
resonance which characterizes the lock-in. A second inter- oscillations are driven by the Strouhal. In the range
pretation (Fig. 15b) consists in defining the lock-in as the U*  [4–6], the cylinder frequency f is exactly equal to the
range where the frequency shifts from the Strouhal value. natural frequency of the cylinder: this corresponds to the
Instead of plotting the reduced frequency f* = fD/U1, the upper branch. Over the range U*  [6–10], the frequency
lock-in can also be characterized by representing the adi- f is still not equal to the Strouhal frequency but its value
mensional frequency f* = f/fH against a representative remains constant over this interval: this corresponds to
parameter (the effective rigidity k eff like [40] or the reduced the lower branch. The interval U*  [4–10] can be defined
velocity U* like [22]). In this case, the lock-in zone can be as the lock-in zone because the cylinder frequency shifts
interpreted as the region where the actual cylinder oscilla- from the Strouhal frequency. The transition zone between
tion frequency f is approximately equal to the natural cylin- the upper and lower branch is characterized by the simul-
der frequency fH. This approach is represented in Fig. 19. As taneous existence of two frequencies in the response.
there is no evident relation between the effective rigidity k eff In our case, the response is less complex because only
used by Shiels et al. [40] and the reduced velocity U* adopted one branch is observed. However, the two first and last
by Khalak and Williamson [22], the subplot in Fig. 19 gives a points (U* = 3.0, 4.0 and U* = 11.3, 14.1) show that the
relation in the form of a curve to facilitate the link with the cylinder oscillations are in this case driven by the Strouhal
previous paragraphs. frequency. In the range U*  [4.4–11.3], the actual fre-
The horizontal line f* = 1 indicates that the cylinder quency of the cylinder shifts progressively from the Strou-
oscillation frequency is the natural frequency, while the hal frequency and follows an oblique line whose slope is
oblique lines give the Strouhal frequencies over the range very smooth and quasi-horizontal. Over this interval, the
of reduced velocities studied. Two lines are represented cylinder is locked near the natural frequency of the cylin-
because results from Khalak and Williamson have been der. In the present case, the lock-in zone is not strictly
 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100 99

horizontal but rather oblique. This is due to the mass of the [2] Al-Jamal H, Dalton C. Vortex induced vibrations using large eddy
system as it is shown by Shiels et al. [40]. The authors per- simulation at a moderate Reynolds number. J Fluids Struct
2004;19(1):73–92.
form several series of simulations for different mass ratios [3] Anagnostopoulos P. Numerical study of the flow past a cylinder
m* between 0 and 20. For small masses, the lock-in zone excited transversely to the incident stream. Part 1: Lock-in zone,
is not very pronounced. On the contrary, for high mass hydrodynamic forces and wake geometry. J Fluids Struct
ratios the lock-in zone is immediately visible and forms a 2000;14(6):819–51.
horizontal level located at f* = 1 (see Fig. 13 of [40]). In [4] Anagnostopoulos P, Bearman P-W. Response characteristics of a
vortex-excited cylinder at low Reynolds numbers. J Fluids Struct
most cases tested here, we choose a mass ratio m* = 3.3, 1992;6(1):39–50.
which explains the slight slope of the level. [5] Braza M, Chassaing P, Ha Minh H. Numerical study and physical
analysis of the pressure and velocity fields in the near wake of a
circular cylinder. J Fluid Mech 1986;165:79–130.
5. Conclusion
[6] Breuer M. A challenging test case for large eddy simulation: high
Reynolds number circular cylinder flow. Int J Heat Fluid Flow
The numerical simulation of the vortex-induced vibra- 2000;21(5):648–54.
tions phenomenon has been investigated here at [7] Carberry J, Sheridan J, Rockwell D. Forces and wake modes of an
Re = 100. The first task of the study has shown that a oscillating cylinder. J Fluids Struct 2001;15(3–4):523–32.
[8] Chen S-S. Flow-induced vibration of circular cylindrical struc-
rather accurate description of the wake for a fixed cylinder
tures. Springer Verlag; 1987.
could be obtained. Based on these results, the simulation of [9] Dennis S-C-R, Chang G-Z. Numerical solutions for steady flow past
a cylinder forced to oscillate in a transverse stream has a circular cylinder at low Reynolds numbers up to 100. J Fluid Mech
been performed to analyze the different phenomena 1970;42:471–89.
appearing. The frequency content and vortex shedding [10] Donea J, Guiliani S, Halleux J-P. An arbitrary Lagrangian–Eulerian
finite element method for transient dynamic fluid–structure interac-
modes have been studied and linked together by changing
tions. Comput Methods Appl Mech Eng 1982;33(1–3):689–723.
the frequency and amplitude of the imposed oscillations. [11] Dong S, Karniadakis G-E. DNS of flow past a stationary
The well-known lock-in zone has been highlighted and and oscillating cylinder at Re = 10000. J Fluids Struct 2005;20(4):
characterized with the analysis of PSDs and phase portraits 519–31.
of the cylinder displacement and lift coefficient. [12] Farhat C, Lesoinne M, Stern P. High performance solution of three-
dimensional nonlinear aeroelastic problems via parallel partitioned
This step has provided interesting elements to understand
algorithms: methodology and preliminary results. Adv Eng Software
the phenomena involved and to validate the CFD code: the 1997;28(1):43–61.
numerical results have indeed proved that our industrial [13] Ferziger J-H, Perić M. Computational methods for fluid dynam-
code was able to capture the vortex induced vibration phe- ics. Springer Verlag; 1996.
nomenon. The lock-in zone has been characterized and the [14] Fey U, König M, Eckelmann H. A new Strouhal–Reynolds-number
relationship for the circular cylinder in the range 47 < Re < 2  105.
response regimes (amplitude, actual oscillation frequency
Phys Fluids 1998;10(7):1547–9.
and aerodynamic coefficients) are in good agreement with [15] Fujisawa N, Asano Y, Arakawa C, Hashimoto T. Computational
similar studies. Restrictions must however be underlined as and experimental study on flow around a rotationally oscillating
the results are for instance in good agreement for low values circular cylinder in a uniform flow. J Wind Eng Ind Aerodyn
of the Reynolds number and without structural damping. 2005;93(2):137–53.
[16] Gerouache M-S. Étude numérique de l’instabilité de Bénard-Kármán
Ongoing numerical simulations are performed for higher
derrière un cylindre fixe ou en mouvement périodique. Dynamique de
Reynolds number (1000 6 Re 6 10.000) with turbulence l’écoulement et advection chaotique. Ph.D. thesis, École Polytech-
modeling using the k–x SST model of Menter [26]. The nique de l’Université de Nantes; 2000.
results are rather promising: the amplitude response is cor- [17] Glück M, Breuer M, Durst F, Halfmann A, Rank E. Computation of
rectly evaluated for high reduced velocities and the upper wind-induced vibrations of flexible shells and membranous struc-
tures. J Fluids Struct 2003;17(5):739–65.
branch is reached [43]; these results are therefore very
[18] Guilmineau E, Queutey P. Numerical simulation of vortex-induced
encouraging since previous attempts to simulate the upper vibration of a circular cylinder with low mass-damping in a turbulent
branch with numerical approaches somehow failed to pre- flow. J Fluids Struct 2004;19(4):449–66.
dict the upper branch response [18,31]. [19] Henderson R-D. Details of the drag curve near the onset of vortex
Detailed investigation and analysis of these complemen- shedding. Phys Fluids 1995;7:2102–4.
[20] Henderson R-D. Nonlinear dynamics and pattern formation in
tary simulations will be presented in a coming paper and
turbulent wake transition. J Fluid Mech 1997;352:65–112.
future application of the presented work will be devoted [21] Jeon D, Gharib M. On circular cylinders undergoing two-degree-of-
to numerical simulation of flow-induced vibration phenom- freedom forced motions. J Fluids Struct 2001;15(3–4):533–41.
enon in tube bundle of nuclear propulsion steam genera- [22] Khalak A, Williamson C-H-K. Motions, forces and mode transitions
tors or vortex-induced vibrations on mooring cables and in vortex-induced vibrations at low mass-damping. J Fluids Struct
1999;13(7–8):813–51.
periscopes.
[23] Koobus B, Farhat C. Second-order time-accurate and geometrically
conservative implicit schemes for flow computations on unstructured
References dynamic meshes. Comput Methods Appl Mech Eng 1999;170(1):
103–29.
[1] Abouri D. Un algorithme d’interaction Fluide–Solide Rigide: Appli- [24] Koopmann G-H. The vortex wakes of vibrating cylinders at low
cation à la Volumétrie. Ph.D. thesis, Université de Poitiers; 2003. Reynolds numbers. J Fluid Mech 1967;28(3):501–12.
100 A. Placzek et al. / Computers & Fluids 38 (2009) 80–100

[25] Lange C-F, Durst F, Breuer M. Momentum and heat transfer from [37] Ramberg S-E, Griffin O-M. Hydroelastic response of marine cables
cylinders in laminar crossflow at 104 6 Re 6 200. Int J Heat Mass and risers. Hydrodyn Ocean Eng 1981:1223–45.
Transf 1998;41(22):3409–30. [38] Roshko A. On the development of turbulent wakes from vortex
[26] Menter F-R. Two-equation eddy-viscosity turbulence models for streets. NACA Report 1191, California Institute of Technology;
engineering applications. AIAA J 1994;32:598–605. 1954.
[27] Mittal S, Kumar V. Flow-induced vibrations of a light circular [39] Saghafian M, Stansby P-K, Saidi M-S, Apsley D-D. Simulation of
cylinder at Reynolds numbers 103 to 104. J Sound Vibr 2001;5(245): turbulent flows around a circular cylinder using nonlinear eddy-
923–46. viscosity modelling: steady and oscillatory ambient flows. J Fluids
[28] Newman D, Karniadakis G-E. Simulation of flow over a flexible Struct 2003;17(8):1213–36.
cable: a comparison of forced and flow-induced vibration. J Fluids [40] Shiels D, Leonard A, Roshko A. Flow-induced vibration of a circular
Struct 1996;10(5):439–53. cylinder at limiting structural parameters. J Fluids Struct
[29] Nobari M-R-H, Naredan H. A numerical study of flow past a 2001;15(1):3–21.
cylinder with cross flow and inline oscillation. Comput Fluids [41] Sigrist J-F. Modélisation et Simulation Numérique d’un Problème
2006;35(4):393–415. Couplé Fluide/Structure Non Linéaire. Application au dimensionn-
[30] Norberg C. Fluctuating lift on a circular cylinder: review and new ement de structures nucléaires de propulsion navale. Ph.D. thesis,
measurements. J Fluids Struct 2003;17(1):57–96. École Centrale de Nantes - Université de Nantes; 2004.
[31] Pan Z-Y, Cui W-C, Miao Q-M. Numerical simulation of vortex- [42] Sigrist J-F, Abouri, D. Numerical simulation of a non-linear coupled
induced vibration of a circular cylinder at low mass-damping using fluid–structure problem with implicit and explicit coupling proce-
RANS code. J Fluids Struct 2007;23(4):23–37. dures. In: 2006 ASME pressure vessels & piping division conference.
[32] Pasquetti R. High-order methods for the numerical simulation of ASME, Vancouver (BC), Canada; July 2006.
vortical and turbulent flows – high-order LES modeling of turbulent [43] Sigrist J-F, Allery C, Beghein C. A numerical simulation of vortex-
incompressible flow. Comptes Rendus Mécanique 2005;333(1):39–49. induced vibration on a elastically supported rigid cylinder at
[33] Persillon H, Braza M. Physical analysis of the transition to turbulence moderate Reynolds numbers. In: 2008 ASME pressure vessels &
in the wake of a circular cylinder by three-dimensional Navier–Stokes piping division conference. ASME, Chicago (IL), USA; July 2008.
simulation. J Fluid Mech 1998;365:23–88. [44] Slone A-K, Pericleous K, Bailey C, Cross M. Dynamic fluid–
[34] Piperno S. Simulation Numérique de Phénomènes d’Interaction structure interaction using finite volume unstructured mesh proce-
Fluide-Structure. Ph.D. thesis, École Nationale des Ponts et Chau- dures. Comput Struct 2002;80(5–6):371–90.
ssées; 1995. [45] Socolescu L. Étude numérique de l’instabilité de Bénard-Von
[35] Placzek A, Sigrist J-F, Hamdouni A. Numerical simulation of vortex Karman derrière un cylindre chauffé. Ph.D. thesis, Université du
shedding past a circular cylinder in a cross-flow at low Reynolds Havre; 1996.
number with finite-volume technique. Part 1: forced oscillations. In: [46] Star-CD. Star-CD Methodology. CD Adapco Group; 2003.
2007 ASME pressure vessels & piping division conference. ASME, [47] Tuann S-Y, Olson M-D. Numerical studies of the flow around a
San Antonio (TX), USA; 2007a. circular cylinder by a finite element method. Comput Fluids
[36] Placzek A, Sigrist J-F, Hamdouni A. Numerical simulation of vortex 1978;6:219–40.
shedding past a circular cylinder in a cross-flow at low Reynolds [48] Williamson C-H-K. Advances in our understanding of vortex
number with finite-volume technique. Part 2: flow-induced oscilla- dynamics in bluff body wakes. J Wind Eng Ind Aerodyn 1997:3–32.
tions. In: 2007 ASME pressure vessels & piping division conference. [49] Williamson C-H-K, Roshko A. Vortex formation in the wake of an
ASME, San Antonio (TX), USA; 2007b. oscillating cylinder. J Fluids Struct 1988;2(4):355–81.