You are on page 1of 12

FORCES ON CYLINDERS IN OSCILLATORY FLOW:

A COMPARISON OF THE RESULTS OF NUMERICAL


AND PHYSICAL MODELS

G. van Oortmerssen
J.J.W. van der Vegt
F. van Walree

Maritime Research
Institute Netherlands
( MARIN )

The Netherlands

Summary

IT'

This paper describes a vortex modelling technique for the theoretical prediction of two-dimensional flow around circular cylinders.
Examples are given of predicted flow fields and drag and lift
forces in steady and oscillatory onflow and these are compared with
results of experiments carried out in a test tank. The correlation
between computer simulations and experiments is encouraging. The
possible implications for-the engineerjng practice are discussed.
1. Introduction
Many offshore structures are composed of tubular elements. The flow
around such slender structural elements is characterized by vortex
shedding, a phenomenon resulting from the viscous nature of the
fluid. The design practice of offshore structures is based on the
application of the empirical Morison equation for the determination
of wave and current forces. There is, however, much uncertainty
with regard to the value of the drag and inertia coefficients that
should be applied in this formula. Moreover, the Morison formula
does not give information on the fluid forces perpendicular to the
flow direction, while these lift forces might be very important for
generating flow induced vibrations.
In recent years several attempts have been made to develop theoretical methods for simulating the flow around circular cylinders.
Finite difference and finite element methods are not well suited
for the description of the typical flow pattern around cylinders at

Reynolds numbers, which consists of regions of concentrated !


vorticity embedded in irrotational flow, and therefore most approaches are based on inviscid discrete vortex models, which use a
Lagrangian description of the flow field. The flow is determined by )
tracking the path of 'individual vortices and calculating their in- i
I
duced velocity field. Such methods do not require a computation /
grid, contrary to the methods based on the Eulerian description, !
and discretization can be restricted to the areas with highly concentrated vorticity. The restriction to inviscid flow, however,
causes problems in the vicinity of the cylinder wall, because in :
this region viscous effects can no longer be neglected, and most
discrete vortex methods require therefore a priori knowledge of the
position of the separation point. Some researchers use empirical
data on the location of the separation point, ref. (l), others 1
apply more or less advanced boundary layer calculations in com- i
bination with an inviscid vortex model, ref. (2, 3 and 4).

l high

At the Maritime Research Institute Netherlands (MARIN) a more ambitious method has been developed, based on the approach of Chorin
(5), which properly accounts for the effects of viscosity. Chorin's
scheme uses the physical insight that viscous effects are primarily
restricted to the vicinity of the body surface, the flow outside
the boundary region being mainly inviscid in nature. The method
therefore approximates a viscous flow field by successively solving
the vorticity diffusion equation and the inviscid vorticity transport equation at small time intervals, This so-called operator
splitting algorithm is combined with a stochastic solution of the
diffusion process, representing the random behaviour of small scale
vortical structures. The behaviour of large vortical structures is
calculated in a deterministic way.

i
1

I,
.

.
,

;
!
!

:
I
!

2. The Numerical Model


The numerical model developed at MARIN simulates 2-dimensional :
steady or oscillatory flow around cylinders in the time domain, and
can be summarized as follows.

i
I
i

The inviscid flow around the cylinder is described by means of a


boundary integral method which satisfies the boundary condition of
zero normal velocity at the cylinder surface. The continuous vorticity field is represented by a finite number of vortices. Instead

L ! ]

Type page number hert:

-7

1 l

of point vortices, so-called vortex blobs are used. In the kernel


of point vortices the velocity becomes infinite, which causes problems when calculating the induced velocities of vortices in close
proximity. This problem can be overcome by applying blobs, in which ;
the vorticity is spread over a small finite area. By choosing a
special type of vorticity distribution (see Beale (6)) a very
accurate description of the continuous vorticity field can be I;
obtained.

1
1
f

The computational scheme starts with calculating the inviscid flow


around the cylinder. In order to satisfy the no-slip condition at
the cylinder wall, vortex blobs are created in the boundary layer.
Subsequently these blobs are diffused by means of the random walk
method, which means that all blobs are given a small random displacement, thus simulating the stochastic character of viscous
flow. The random displacement has a Gaussian probability density
function with zero mean and a variance which depends on the kinematic viscosity of the fluid and on the time step. Next, the vortices are convected by the local fluid velocity, which is the
ambient flow velocity plus the velocities induced by all the other
vortices. Finally the forces on the'cylinder are calculated from an
alternative formulation which expresses the forces in terms of
velocity and vorticity. The whole process of creating new vortices
at the cylinder wall, diffusion, convection and force calculation
is repeated each time step.

:
!
l

:
:
i

I
:

i
[

1I
j

I
l
,

lj
i

i
The convection step in the numerical procedure requires a large
computational effort. In an infinite fluid domain the relation
between velocity field and vorticity field is given by the BiotSavart law of interaction. The velocity at a certain vortex position is found from calculating the velocities induced by all other
vortices. The computing time involved is proportional to the square
of the number of vortices, and since large numbers of vortices are
required to obtain the desired level of accuracy, the computational
effort soon becomes prohibitive, even for today's powerful computers. A more efficient approach for obtaining the velocity field
is to use a fast Poisson solver. This results in a mixed EulerianLagrangian algorithm. The first results with such method were obtained using a finite difference scheme, known as the vortex-incell method, see Stansby and Dixon ( 7 ) . A problem with this method
is the limited accuracy. The present method uses a very accurate

i1
1i
1

'

Type page number here

I 1

tvariationally optimized spectral algorithm which is solved by means


'of Fast Fourier Transform. The computing time of this algorithm is
linearly proportional to the number of vortices. A more rigorous
description of the numerical model and theoretical background is ;
given in ref. (8).
3. Description of Experiments
The experiments comprised towing and forced oscillation tests with ;
i
submerged horizontal cylinders in still water. The tests were car- !
ried out in MARIN's High Speed Basin, which measures 220 m X 4 m X
3.8 m in length, width and depth respectively. The test cylinder
was mounted in a wishbone construction formed by two vertical
struts as shown in Figure 1. The struts were fixed to the pistons
of a hydraulic oscillator which was mounted in horizontal position
to the towing carriage. The cylinder axis was approximately 1.90 m
below the water surface. End plates with a diameter of 1.0 m were
mounted to the cylinder in order to assure 2-dimensional onflow.:

Two different cylinders were used. The towing tests were carried :
out with a 0.10 m diameter cylinder with a length of 1.0 m. The
cylinder used for the oscillation tests had a diameter and length
of 0.15 m and 1.1 m respectively.
The towing tests were carried out at speeds up to about 1.0 m/s, 1
which corresponds to subcritical flow conditions (Reynolds number j
R, G 1.3*105; a list of symbols is given in section 6). The oscil- i
I
lator device has a maximum amplitude xa of 0.5 m at a frequency of I

0.09 Hz, resulting in a maximum value of the Keulegan-Carpenter f


number K, of about 21. At higher frequencies the maximum achievable l
amplitude decreases (for instance at 0.45 Hz the maximum amplitude
is 0.1 m).
The test cylinders consisted of three sections. The drag and lift
. forces were measured by means of strain-gauges on the middle section, which had a length of 0.40 m. In addition, pressure gauges
!
,were mounted in the cylinder wall and electrolysis was applied for
flow visualization. All data were sampled at a rate of 25 Hz and

.
I

recorded by a data acquisition computer system. All experimental


;records covered a period of at least 300

S.

Type page number here

-b

! 4. Discussion of Results

comparison between measured and computed results is given for


three cases, one steady flow and two oscillatory flow conditions,
the main flow parameters of which are given in Table 1.

Table 1. Main Flow Parameters of Experiments

Case

Cylinder motion

1
2
3

steady
oscillatory
oscillatory

26,000
23,500
33,900

9.5
13.7
A

The computations were carried out on a CRAY-X-MP supercomputer. The


cylinder was represented by 64 surface panels and the time step
amounted to t = 0.05~/~,. The maximum number of vortices was between 13,000 and 30,000 while the number of Fourier components in
\
the spectral algorithm amounted to 256*256. For the steady flow
case a period of 300 S was simulated, requiring a computing time of
approximately 13 hours. For the oscillating cylinder the simulations covered 50 S, requiring 4000 S on the computer. Both computed
and measured force records were low-pass filtered with the same
numerical filter. The results are presented ,in the form of statistical data in Table 2 and representative parts of the force records
in Figures 2, 3 and 4 . The drag and lift forces are given in the
form of non-dimensional force coefficients, defined as follows:

I
I

I
/

(for symbols see section 6)


!

When comparing the experimental and computed force records it


should be borne in mind that the vortices in the wake of the cylI
inder behave in a stochastic way. Experimental data can therefore
not be reproduced exactly in a numerical simulation, and the correlation should therefore be focussed on the qualitative similitude
of force records and on a comparison of statistical data. In reality, the random behaviour of vortical structures also leads to 3dimensionality of the wake, although the flow visualization showed
that in the cases presented here the wake remained to a large degree 2-dimensional.
a

Type page number here

I I

For the steady flow case, the mean values of calculated CD and CL
lwere found to be close to the measured data, as can be seen from
Table 2. The variation of the predicted forces is somewhat larger
than found experimentally. As can be observed in Figure 2, the
force records in general show a good qualitative agreement, although the character of the computed lift force is somewhat more
irregular compared to the measured lift force.

'

Table 2 and Figures 3 and 4 show that the predicted drag force is
in good agreement with the measured force in the two oscillatory
flow cases. The discrepancies between measured and computed lift
forces are larger. In case 2 the calculated lift force is larger
than the measured one, while the reverse is true in case 3. The
character of the flow field as predicted by the numerical simulation was quite similar to the observed flaw field. An example of a
computed vortex pattern for oscillatory flow is given in Figure 5.
The results in this figure apply to a flow with K, = 4.7.
5. Concludins Remarks
The results presented in this paper are very encouraging. Much work
remains to be done, however, before the capabilities of the pre- Ii
sented numerical model can be fully appraised. More insight is for
instance needed in the effect of parameters such as the number of /
vortices, the time step and the simulation period on the accuracy 1
i
of the results. Further, the correlation with experimental results I
must be extended over a wider range of values of the Reynolds and
Keulegan-Carpenter numbers. In particular, it is of interest to see j
to what Reynolds numbers the two-dimensional model will satisfy,
because it is known, that the wake behind the cylinder becomes I
quite three-dimensional at the higher values of the Reynolds number
which correspond to cases of practical interest. Three-dimensional
effects become also of importance when cylinders in waves or in ,
oblique flow are considered. For these reasons a three-dimensional
version of the numerical model is under development. Notwithstanding the fact that much effort will be devoted to improving the computational efficiency, it is not to be expected that numerical models like the one presented here will be applied in the engineering
practice in the near future. As a research tool, however, these
models can help to improve our insight in the problem of hydrody-

Type page number here

I 1

'namic forces on offshore structures and may generate basic force


data that can be used in the design practice.
l

6. List of Symbols

CD = non-dimensional drag force


CL = non-dimensional lift force
D = cylinder diameter
FD = drag force
FL = lift force
K, = Keulegan-Carpenter number = 2nxa/D
= cylinder length
R, = Reynolds number = U.D/V

= U,.D/V

for steady flow


for oscillatory flow

U = flow velocity
U, = amplitude of oscillatory flow velocity
xa = amplitude of cylinder motion
v = kinematic viscosity
P = fluid density
References

(1) FALTINSEN, 0. and PETTERSEN, B., "Separated Flow around Marine j


Structures", SSPA Ocean Engineerinq Symposium, Gothenburg, !
!
1983.
I
(2) AARSNESS, J.V., Current Forces on Ships, Thesis, Norwegian '
il
Institute of Technology, Trondheim, 1984.
(3) SARPKAYA, T. and SHOAFF, A., A Discrete Vortex Analysis of
Flow about Stationary and Oscillating Circular Cylinders,
l
NPS-69 SL 79011, Naval Post Graduate School, Monterey, 1979.
(4) IKEDA, Y., "Calculation of Lift Force acting on Circular
Cylinder in Oscillating Flow", J. Kansai Soc. N.A. Japan, No.
159, December 1984.
1
(5) CHORIN, A.J., "Numerical Study of Slightly Viscous Flow", J.
Fluid Mech. 57 (1973), pp.785-796.
(6) BEALE, J.T. and MAJDA, A., "High Order Accurate Vortex Methods
with Explicit Velocity Kernels", J. of Comp. Physics, 1985,
pp.188-208.

(7) STANSBY, P.K. and DIXON, A.G.,

"Simulation of Flow around Cyl-

inders by a Lagrangian Vortex Scheme", Applied Ocean Research,


Vol. 5, No. 3, 1983.

Type page number here

l?

I 1

I
4

i
'

(8) VAN DER VEGT, J.J.W.

and DE BOOM, W.C.,

"Numerical Simulation

of Flow around Circular Cylinders at High Reynolds Numbers".


4th Int. BOSS Conference, Delft, 1985.

Table 2. Statistical Data of Measured and Computed Force


Coefficient Time Traces
Case 1

Case 2

Case 3

Stationary
cylinder

Oscillating
cylinder

Oscillating
cylinder

Exp.

Exp.

Camp.

Exp.

Comp

C~
mean

1.34

1.39

0.04

max.

1.66

1 .g0

min.

0.99

0.79

2.61
-2.50

st. dev.

0.11

0.20

1.30

0.93

0.99

C~
mean

0.00

0.09

-0.02

-0.04

-0.30

max.

1.36

1.85

0.80

1.75

2.07

1.65

min.

-1.35

-1.60

-1.25

-2.02

-2.70

0.53

0.68

0.69

1.00

1.55

-1.92
0.76

st. dev.

Camp.

0.07

0.10

2.50
2.37
-2.67. -2.49

2.30

0.03

-1.70
0.76

-0.12'

Fig. 1. Experimental Arrangement


Type page number here

I I

CL-EXP.

CL-COMP.

CD-EXP.

CD-COMP.

0.0

10.0

20.0

SECONDS

Fig. 2. Comparison of Experimental and Computed Force Coefficient


Time Traces

Case 1

Type page number here

I I

CL-EXP.

g
F.

CL-COMP.

CD-EXP

CD-COMP.

\I

10.0

0.0

20.0

SECONDS
l

: Fig.

Comparison of Experimental and Computed Force Coefficient

i
II

Time Traces

Case 2

!
!

1
l
I

T y ~ t page
:
nunlbe! here

I3

I I

CL-EXP.

CL-COMP.

CD-EXP.

t
CD-COMP.

20.0

10.0

0.0

SECONDS
l
!

I Fig. 4. Comparison of Experimental and Computed Force Coefficient


!
Time Traces - Case 3
!

I I

Type page number here

1I

I I

You might also like