Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Flow Past a Stationary and Moving Cylinder: DNS at Re=10,000

Suchuan Dong, Didier Lucor, and George Em Karniadakis


Brown University, Division of Applied Mathematics, Providence, RI
{sdong, didi, gk}@dam.brown.edu

Abstract Due to the complexity of the hydroelastic


cylinder/wake problem, theoretical models remain
We conduct direct numerical simulations with 300 incomplete. Simplified models rely on the force input as
million degrees of freedom of turbulent flows past a well as the added mass coefficient and correlation lengths.
stationary and a forced oscillating rigid cylinder at the Until recently, prediction of VIV was primarily based on
Reynolds number Re=10,000. This one-order of semi-empirical methods. Direct numerical simulations
magnitude increase in Reynolds number (compared to have been applied to investigate the flow past a freely
previous DNS) is accomplished by employing a vibrating cable in both the laminar[19] and turbulent[6]
multilevel-type parallel algorithm within the spectral regimes. Two states are observed in the cylinder wake
element framework. Comparisons with the available from these simulations, corresponding to a traveling wave
experimental data show that the simulation has captured response and a standing wave response, respectively. The
the flow quantities, mean, and rms statistics of the former state produces a vorticity field consisting of
cylinder wake correctly. We also examine the effect of the oblique “rollers” shed off the cylinder, while the latter
randomness in the inflow on the vortex formation at a corresponds to a three-dimensional staggered pattern
lower Reynolds number. We demonstrate that noisy forming Ȝ-type vortices. The existence of both states has
inflows cause a vortex shedding-mode switching in the been demonstrated by Olinger[21], who used low-order
cylinder wake. modeling based on circle maps to represent shedding
patterns behind flexible cables.
Previous numerical studies of the flow past an
1. Introduction oscillating cylinder concentrated on relatively low
Reynolds numbers (up to Re=1,000 in[6]). In the current
Fluid flows passing over rigid or flexible cylinders work, we increase the Reynolds number by an order of
arise in numerous engineering situations, such as power magnitude to Re=10,000 based on the inflow velocity and
transmission lines, marine cables, drilling and production the cylinder diameter. This is enabled by a multilevel-
risers in petroleum production, mooring lines, and heat type parallel algorithm in the computations. In this paper,
exchangers[3,10]. From both design and operational we focus on turbulent flow past a stationary cylinder and
standpoints, it is important to be able to predict the a rigid cylinder undergoing forced sinusoidal oscillations,
hydrodynamic forces and motion of such structures and compare the simulation results with the available
caused by vortex-induced vibration (VIV). From the experimental data at this Reynolds number.
fundamental point of view, it is important to understand
how the near wake is modified by the motion of the 2. Parallel Computation Method
cylinder, which in turn depends on the flow conditions,
the structural characteristics, and the type of support.
Significant progress has been made in understanding Realistic simulations of flow past a cylinder subject
the wake of a stationary cylinder in recent years, to VIV at high Reynolds numbers require high resolution
especially in the low Reynolds number range[25]. Several in the streamwise and cross-flow directions and a large
researchers have measured the forces on finite-span number of Fourier modes along the cylinder span.
cylinders undergoing forced harmonic and multi- Parallel computations employing a single-level
frequency oscillations[2,23,24,7], as well as undergoing parallelism for this type of problem has clear performance
freely transverse vibrations[15,11]. However, considerably limitations that prevent effective scaling to the large
less is known for the flexible cylinder subject to VIV[1,28]. processor count on modern supercomputers. Careful
analysis indicates that VIV computations

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
demonstrate inherent hierarchical structures when the Table 1. Performance of MPI/MPI parallel algorithm for
problem is discretized with spectral/hp element cylinder flow at Re=10, 000 with a problem size of
methods[13]. Consider an incompressible flow past a 300,000, 000 degrees of freedom on TCS at PSC.
cylinder subject to VIV. We assume the flow and Parallel speed-up is computed based on the time on
256 processors.
cylinder variables are periodic in the homogeneous
direction. A combined spectral element-Fourier CPU Seconds/step Speed-up Efficiency
discretization can be employed to accommodate the 256 14.766 1.0 1.0
requirement of high order as well as efficient handling of 512 7.417 1.991 0.996
complex computational domain in the non-homogeneous 1024 3.882 3.804 0.951
planes. Spectral expansions in the homogeneous direction
1536 2.877 5.132 0.856
employ Fourier modes that are decoupled except in the
nonlinear terms. Each Fourier mode is a 2D field and
solved with the spectral element approach. Computations
on the Fourier modes, spectral element plane, spectral
3. Simulation Parameters
elements within the plane, and at the sub-element level
form the hierarchy of operations in the solution process of We consider the turbulent flow past a rigid circular
the VIV problem. Multilevel parallelism naturally suits cylinder. The cylinder is either stationary or undergoing a
such VIV computations with inherent hierarchical forced oscillation in the cross-flow direction. Following
structures[4]. Newman and Karniadakis[19], we solve the Navier-Stokes
To exploit the inherent hierarchical structures in the equations in a coordinate system attached to the cylinder,
computations, we employ an MPI/MPI two-level parallel which avoids the difficulty of a moving mesh. We
algorithm[5] for simulating the cylinder flow at employ a stiffly-stable pressure correction-type scheme[12]
Re=10,000. Specifically, the flow domain is first with a third-order accuracy in time.
decomposed along the cylinder. Each sub-domain We focus on the turbulent wake at Reynolds number
consists of one or more Fourier modes in the Fourier Re=10,000 based on the inflow velocity and the cylinder
space. At the first level are groups of MPI processes, diameter. For the moving cylinder case, the motion of the
with each group computing one sub-domain. The spectral cylinder in cross-flow y direction is expressed by
element mesh in the non-homogeneous planes are further y = Y0 cos(2ʌf0t), (1)
partitioned using a graph-partitioning algorithm[14] at the where y is the cylinder displacement at time t, Y0 is the
second level. Each sub-domain at the second level cylinder displacement amplitude, and f0 is the cylinder
comprised of structured or unstructured elements. oscillation frequency. All length parameters are
Accordingly, each MPI process within the group normalized with the cylinder diameter D.
computes a sub-domain at the second level. The key The computational domain extends from í20D at the
advantage of this parallel algorithm is that MPI processes inlet to í50D at the outlet, and from í20D to 20D in the
participate in communications at different levels and cross-flow direction. The spanwise length of the domain
communicate only with the other processes at the same Lz is chosen as Lz/D = ʌ. A “z-slice” of the computational
level. As a result, compared with a single-level domain (figure 1) consists of K = 6,272 triangular prisms
computation on the same number of processors, much (elements). The polynomial order per element is varied
fewer processes are involved in the communications at from P = 5 to 8. In Figure 1 we also plot the mean
each level. This reduces the communication latency streamwise velocity profile along a vertical line crossing
overhead and enables the applications to scale to a large the cylinder axis. It indicates that there is over one layer
number of processors more efficiently. This is of spectral elements (and more than 6 grid points when
demonstrated by the performance result in Table 1 the element order is taken into account) within the
obtained on the TCS cluster at Pittsburgh Supercomputing cylinder boundary layer with the current mesh.
Center (PSC) for the cylinder flow at Re=10,000 with a Resolution tests were performed by varying the number of
problem size of 300,000,000 degrees of freedom. grid points in the spanwise z direction. Table 2
summarizes the drag coefficient, base pressure coefficient
and the Strouhal number under several resolutions
together with their experimental values in the flow past a
stationary cylinder at Re=10,000. It is evident that the
computed values agree with the experimental data quite
well.

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
4.1. Turbulent Flow Past a Stationary Cylinder
at Re=10,000.

To obtain the DNS results presented here, the


simulations have advanced more than 1,000 convective
time units (D/U0, where U0 is the inflow velocity).
Figure 2 shows a window of the time history of the drag
and lift coefficients with Nz = 128 at Re=10,000. The
nondimensional time step typically varies between
4.0×10í4 and 5.0×10í4. The discussion will be focused on
the flow statistics in the cylinder wake.

Figure 1. (a) Two-dimensional “z-slice” of the


computation domain consisting of 6,272 triangular
spectral elements; (b) mean streamwise velocity
profile along a vertical line crossing the cylinder.

Table 2. Physical quantities in the flow past a


stationary cylinder at Re=10,000: drag coefficient Cd,
base pressure coefficient Cpb and Strouhal number Dt.
The order of spectral elements is denoted by P, and Nz
is the number of grid points in the spanwise z
direction.

Cd íCPb St
DNS (P = 5, Nz = 16) 1.155 1.129 0.195
Figure 2. (a) Time history of drag coefficient and (b) lift
coefficient, of the flow past a stationary cylinder at
DNS (P = 5, Nz = 64) 1.110 1.084 0.209
Re=10,000.
DNS (P = 5, Nz = 128) 1.128 1.171 0.205
Bishop & Hassan (1964) — — 0.201 The mean pressure distribution on the cylinder
Gopalkrishnan (1993) 1.186 — 0.193 surface is first compared to the experimental data from[20]
Williamson (1996) — 1.112 in Figure 3. An overall good agreement is observed
Norberg (2003) — — 0.202 between the DNS at Re=10,000 and the experiment by
Norberg[0] which was performed at Re=8,000. The
simulation results show a lower value of the minimum
4. Cylinder Flows at Re=10,000 pressure on the cylinder surface.
We next compare the flow statistics between the
simulation and the PIV experimental data[22] at
Simulations have been performed on two
Re=10,000. In Figure 4 we plot contours of the Reynolds
configurations of the cylinder flow at Re=10,000. In the
first configuration the cylinder is stationary. In the stress  u cvc ! from PIV experiment (left) and from
second configuration the cylinder undergoes a forced current simulations (right). The contours from PIV and
sinusoidal motion described by Eq. 1. In this section we DNS are plotted on the same levels,
will compare the flow statistics of the cylinder wake with u cvc! min 0.03 and '  u cvc ! 0.01 . The
between simulations and PIV experiment for the computed Reynolds stress agrees with the experimental
stationary case, and compare several physical quantities data quite well. The Reynolds stress distribution shows
between simulations and the MIT experiment for the four distinct “lobes” which are anti-symmetric with
oscillation case. respect to the centerline. Comparisons of the mean and

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
rms velocity distributions (not shown here) also increases dramatically as the oscillation frequency
demonstrate that the simulation has produced the same increases.
distributions as the PIV experiment.
Table 3. Simulated cases of the flow past an
oscillating cylinder. Y0 and f0 denote the cylinder
displacement amplitude and oscillation frequency,
respectively.

Cases Y0/D f0D/U0


M1 0.3 0.14

M2 0.3 0.25

Figure 3. Comparison of pressure coefficient on


cylinder surface between simulation (Re=10,000) and
experimental data (Re=8,000) in Norberg[20].

Figure 4. Comparison of the Reynolds stress < u cv c >


between PIV experiment[22] (left) and current
simulations (right). Contours of the experimental data
and the simulation results are plotted on the same
levels: <u cv c> min
= 0.03 and ǻ < ucv c > = 0.01 .
Figure 5. (a) Drag coefficient and (b) lift coefficient
4.2. Turbulent Flow Past an Oscillating Cylinder versus non-dimensional frequency; sinusoidal
at Re=10,000. oscillations; Y0/D = 0.3.

We next examine the turbulent flows past a circular The lift force phase angle, the angle by which the lift
cylinder that undergoes forced sinusoidal oscillations in force leads the imposed cylinder motion, is an important
the cross-flow direction (see Eq. 1) at Re=10,000. Our quantity as it determines the sign of the power transfer
emphasis is on the influence of the cylinder oscillation on between the cylinder and the fluid. The magnitude of the
the physical quantities, and the comparison of simulation power transfer depends on the lift coefficient in phase
results with the experiment data[7] at the same Reynolds with the cylinder velocity. In Figure 6 we plot the history
number. of the lift coefficient and the cylinder displacement in the
We have considered two oscillation frequencies with flow past an oscillating cylinder at Re=10,000 at an
a moderate displacement amplitude Y0/D = 0.3 (see oscillation frequency foD/U0 = 0.25. The lift force and the
Table 3 for the parameters). In Figure 5 we plot the drag cylinder motion are essentially in phase at this frequency.
coefficient (left) and lift coefficient (right) as a function The behavior of the lift force phase angle and the lift
of the nondimensional frequency foD/U0 from the coefficient in phase with velocity from[7] and from current
experimental data as well as from current simulations at simulations are illustrated in Figure 7. The DNS results
Re=10,000. The agreement between simulation results are in good agreement with the experimental data.
and the measurements of[7] is reasonably good. The
simulation under-predicts the drag coefficients and over-
predicts the lift coefficient at frequency foD/U0 = 0.14. At
the two frequencies simulated, there is not much
difference in the drag coefficient, while the lift coefficient

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
therein. The different synchronization regions for such
prescribed cylinder oscillations have been mapped in a
amplitude-wavelength plane by Williamson & Roshko[26].
The dominant vortex patterns near the fundamental
synchronization region are 2S, 2P, and (P+S). A 2S
shedding-mode implies that in each half-cycle of the
oscillation, a single vortex is shed into the wake. A 2P
shedding-mode means that in each half-cycle, a pair of
vortices of opposite signs are shed into the wake. Finally,
a (P+S) mode is an asymmetric version of the 2P mode
where the cylinder sheds a pair of vortices of opposite
signs and a single vortex each cycle. Adopting this
classification, the standard von Karman vortex street is
equivalent to a 2S mode. Some vortex formation
Figure 6. Time history of lift coefficient and cylinder processes, such as the ones leading to 2P and (P+S)
displacement in flow past an oscillating cylinder at shedding-modes, appear to be more unstable than the 2S
Re=10,000; oscillation frequency f0D/U0 = 0.25; Y0/D = and sensitive to different experimental conditions. More
0.3. generally, the boundaries between the different regimes
identified in the experiments of[26] seem not to be robust,
and indeed they depend on many parameters including
Reynolds number, thermal effects, turbulence level, start
up conditions, etc.
We consider the effect of noisy inflow on the vortex
formation behind an oscillating circular cylinder. More
specifically, we simulate stochastically laminar flow past
an oscillating circular cylinder with random inflow, i.e.,
consisting of a uniform mean inflow plus a random
component; the latter follows a uniform distribution. We
employ a new simulation method based on generalized
polynomial chaos (GPC) algorithms developed in[27,17].
GPC represents efficiently Gaussian and non-Gaussian
second-order random processes in terms of orthogonal
polynomial functional.
We examine the effect of different levels of noise on
the vortex formation behind the cylinder at a low
Reynolds number Re=140. The cylinder is forced to
oscillate in a purely harmonic deterministic motion with
amplitude A/d = 1.0 and reduced velocity based on the
excitation frequency Vrn Ud f e 7.5 . This choice of
parameters leads to a (P+S) shedding-mode in
deterministic simulations, i.e., in the absence of noise at
Figure 7. (a) Phase angle of lift force with respect to the inflow. Visualizations of deterministic vorticity at
cylinder motion; (b) Lift force coefficient in phase with different instants within one shedding cycle and spectral
velocity; sinusoidal oscillations; Y0/D = 0.3. analysis of the cylinder forces indicate that we do obtain a
stable (P+S) shedding mode. The top image in Figure 8
5. Vortex Shedding-Mode Switching Caused shows an example of the (P+S) shedding mode at y/d=0
with one pair (P1) of vortices (I & II) shed from the
by Noisy Inflow
cylinder upper side and one single (S1) vortex shed from
the lower side.
One of the main features of the flow past a cylinder We now present results from the stochastic
oscillating transversely in a free stream is lock-in, a simulations. The lower three images in Figure 8 show
frequency synchronization phenomenon. In experimental instantaneous stochastic mean vorticity contours (at the
work, the motion of the cylinder is often prescribed in same time as the deterministic simulation) for different
order to mimic the vortex-induced oscillations of a free, level of noise. The location of the letters identifying the
i.e., elastically-mounted cylinder, see[9,18] and references

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
vortices does not change in the four different plots. The In Figure 9 we plot the isocontours of the pressure
effect of the noise is striking; the vortex street in the wake field in the near wake for the flow past a cylinder at
is re-organized as the level of noise increases from 10% to Reynolds number Re=300. We have considered the
20% to 30%, see Figure 8. These results seem to indicate deterministic flow case and a case with a random inflow
that the noise affects primarily the formation of the (with 5% randomness, Vu = 0.05). Figure 9(a) shows the
second vortex of each pair in the (P+S) mode; this is deterministic pressure pdet = í0.15, and Figure 9(b) shows
particularly clear in the second plot from the top. For this the mean random pressure p0 = í0.15. The three-
case, corresponding to a moderate level of noise, we see dimensional nature of the near wake structures of the
that the second vortex (II) of the first (P1) and second deterministic flow is quite evident from Figure 9(a). On
(P2) pairs are really affected while the other vortical the other hand, the randomness in the inflow effectively
structures (I of P1, I of P2, S1, S2, S3) remain almost suppresses the three-dimensionality of near-wake
intact. We also note that the circulation of the second structures.
vortex of each pair of deterministic vortices is weaker
than the first vortex. This was also observed in the case
of the 2P shedding-mode in the wake of free rigid
cylinders in crossflow[8]. This suggests that the weakest
vortical structures of the wake are the first ones to suffer
the effect of noise[16].

Figure 9. Isocontours of the deterministic pressure


Pdet field (a) and the mean pressure p0field of random
inflow (randomness ıU = 0.05). p = í0.15, Re=300

6. Concluding Remarks

We have presented results of direct numerical


simulations of turbulent flow past a stationary cylinder
and a forced oscillating cylinder at the Reynolds number
Re=10,000, which is about one order of magnitude higher
than previously was achieved. This is accomplished by
employing a multilevel-type parallel algorithm within the
spectral/hp element framework. Comparisons with the
available experimental data for the stationary-cylinder
Figure 8. Comparison between deterministic (a) and case show that the simulation has captured the flow
stochastic mean instantaneous vorticity fields for quantities and the mean, rms and Reynolds stress statistics
different levels of noise au at identical time and correctly. Comparisons with the experimental data for the
y/d = 0; ı U = 0% U (b); ı U = 10% U (c); ı U = 20% U (c); forced oscillation case indicate that the simulation has
ı U = 30% U (d); Re = 140; p = 15
captured the physical quantities such as the drag and lift
coefficients, and the phase angles reasonably well. DNS

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
is now emerging as a promising and effective tool for 6. Evangelinos, C. and G.E. Karniadakis, “Dynamics and flow
studying the VIV phenomena at high Reynolds numbers. structures in the turbulent wake of rigid and flexible cylinders
subject to vortex-induced vibrations.” J. Fluid Mech., 400, 1999,
pp. 91–124.
7. Significance to DoD 7. Gopalkrishnan, R., “Vortex-induced forces on oscillating
bluff cylinders.” Ph.D.Thesis, Department of Ocean
Vortex-induced vibration research is important to Engineering, MIT, Cambridge, MA, USA, 1993.
DoD in general and is critical to naval and marine under- 8. Govardhan, R. and C.H.K. Williamson, “Modes of vortex
water applications. High levels of VIV fatigue damage formation and frequency response of a freely vibrating
can be accumulated in relatively short periods of time in cylinder.” J. Fluid Mech., 420, 2000, pp. 85–130.
the severe currents encountered in most deep-water 9. Griffin, O.M. and S.E. Ramberg, “The vortex street wakes of
operational environment. Our simulations are the first vibrating cylinders.” J. Fluid Mech., 66, 1974, pp. 553–576.
direct numerical simulations in this area about the 10. Hover, F., A. Techet, and M. Triantafyllou, “Calculation of
interaction between turbulent flow and a flexible dynamic motions and tensions in towed underwater cables.”
cylinder/cable. The detailed comparisons between DNS IEEE J. Ocean Engng., 19, 1994, p. 449.
and experimental data at Re=10,000 in this paper have 11. Hover, F., A. Techet, and M. Triantafyllou, “Forces on
demonstrated that direct numerical simulation has oscillating uniform and tapered cylinders in crossflow.” J. Fluid
emerged as an efficient tool with high accuracy for Mech., 363, 1998, pp. 97–114.
studying vortex-induced vibrations at high Reynolds 12. Karniadakis, G.E., M. Israeli, and S.A. Orszag, “High-order
numbers. splitting methods for the incompressible Navier-Stokes
equations.” J. Comput. Phys., 97, 1991, p. 414.
8. Systems Used 13. Karniadakis, G.E. and S.J. Sherwin, Spectral/hp element
methods for CFD, Oxford University Press, 1999.
14. Karypis, G. and V. Kumar, “A fast and high quality
Simulations were performed on the IBM SP4 at multilevel scheme for partitioning irregular graphs.” SIAM J.
NAVO and ARSC, and on the SGI Origin 3800 at ERDC. Sci. Comput., 20, 1998, p. 359.
15. Khalak, A. and C. Williamson, “Fluid forces and dynamics
9. CTA of a hydroelastic structure with very low mass and damping.” J.
Fluids and Structures, 11, 1997, pp. 973–982.
Computational Fluid Dynamics (CFD) 16. Lucor, D. and G.E. Karniadakis, “Noisy inflows cause a
shedding-mode switching in flow past an oscillating cylinder.”
Phys. Rev. Lett., 92, 2004, p. 154501.
Acknowledgements 17. Lucor, D., D. Xiu, C.-H. Su, and G.E. Karniadakis,
“Predictability and Uncertainty in CFD.” Int. J. Numer. Mech.
This work was supported by ONR. Computer time Fluids, 43, 2003, pp. 483–505.
was provided by HPCMP (NAVO, ERDC, and ARSC). 18. Ongoren, A. and D. Rockwell, “Flow structure from an
We would like to thank Professor M. Triantafyllou and oscillating cylinder – Part 1: Mechanisms of phase shift and
Professor D. Rockwell for providing the experimental recovery in the near wake.” J. Fluid Mech., 191, 1988, pp. 197–
data. 223.
19. Newman, D.J. and G.E. Karniadakis, “A direct numerical
simulation study of flow past a freely vibrating cable.” J. Fluid
References Mech., 344, 1997, pp. 95–136.
20. Norberg, C., “Pressure forces on a circular cylinder in cross
1. Alexander, C., “The complex vinrations and implied drag of a flow.” IUTAM Symposium Bluff-Body Wakes, Dynamics and
long oceanographic wire in crossflow.” Ocean Engineering, 8, Instabilities, Gottingen, Germany, 1992.
1981, pp. 379–406.
21. Olinger, D., “A low-order model for vortex shedding
2. Bishop, R.E. and A.Y. Hassan, “The lift and drag forces on a patterns behind vinrating flexible cables.” Phys. Fluids, 10,
circular cylinder oscillating in a flowing fluid.” Proc. R. Soc. 1996, pp. 1953–1961.
Lond. A, 277, 1964, pp. 51–75.
22. Saelim, N. and D. Rockwell, “Near-wake of a cylinder in the
3. Blevins, R., Flow Induced Vibration, Van Nostrand Reinhold, range of shear layer transition.” Phys. Fluids, submitted 2003.
1990.
23. Sarpkaya, T., “Fluid forces on oscillating cylinders.” ASCE
4. Dong, S., and G.E. Karniadakis, “Dual-level parallelism for Journal of Waterway, Port, Coastal, and Ocean Division, 104,
high-order CFD methods.” Parallel Computing, 30, 2004a, pp. 1978, pp. 275–290.
1–20.
24. Staubli, T., “Calculation of vibration of an elastically-
5. Dong, S. and G.E. Karniadakis, “Multilevel parallelization mounted cylinder using experimental data from forced
models for high-order CFD.” AIAA Paper 2004-1087, 2004b. oscillation.” ASME J. Fluids Engng., 105, 1983, pp. 225–229.

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE
25. Williamson, C., “Vortex dynamics in the wake.” Ann. Rev. Generalized Polynomial Chaos.” J. Fluids Eng., 124, 2002, pp.
Fluid Mech., 28, 1996, pp. 477–539. 51–59.
26. Williamson, C.H.K. and A. Roshko, “Vortex formation in 28. Yoerger, D., M. Grosenbaugh, M. Triantafyllou, and J.
the wake of an oscillating cylinder.” J. Fluids and Struct., 2, Burgess, “Drag force and flow-induced vibrations of a long
1988, pp. 355–381. vertical tow cable–Part 1: Steady-state towing conditions.”
27. Xiu, D., D. Lucor, C.-H. Su, and G.E. Karniadakis, Journal of Offshore Mechanics and Arctic Engineering, 113,
“Stochastic modeling of flow-structure interactions using 1991, pp. 117–127.

Proceedings of the Users Group Conference (DOD_UGC’04)


0-7695-2259-9/04 $ 20.00 IEEE

You might also like