Sei sulla pagina 1di 26

Fluid Dynamics Research 26 (2000) 69–94

Unsteady viscous ow about a submerged circular cylinder with


a steady current
B. Yan
School of Mathematics, University of East Anglia, Norwich, NR4 7TJ, UK
Received 11 January 1999; received in revised form 22 March 1999

Abstract
Viscous ow about a submerged circular cylinder, which oscillates in an otherwise steady current, is considered. The
amplitude of the oscillation is assumed small compared with the dimension of the cylinder and the magnitude of the
oscillation generated velocity is comparable with that of the steady current. Attention is mainly paid to the in uence of
the steady current upon the induced steady streaming. It is demonstrated that when the magnitude of the steady current is
relatively large the induced steady streaming for rectilinear oscillations is largely independent of the type of oscillation.
However, when the magnitude of the steady current is small, the Reynolds stresses produced by the oscillation become
signi cant in the neighbourhood of the cylinder and hence the induced steady streaming is of the classic four-cell structure.
When the cylinder performs uniform orbital motion, the circulation around the cylinder due to the induced steady streaming
is almost independent of the magnitude of the steady current. c 2000 The Japan Society of Fluid Mechanics and Elsevier
Science B.V. All rights reserved.

Keywords: Steady current; Steady streaming; Free surface; Forces; Oscillating cylinder

1. Introduction

When a cylinder performs small-amplitude oscillations in a viscous uid that is otherwise at rest, it
is well-documented that, superimposed on an expected oscillatory ow is a time-independent steady
streaming ow whose origin is the Reynolds stresses generated in the uctuating ow. See, for
example, Stuart (1966), Riley (1967) and Amin (1988). Amin’s work is mainly concentrated on
the low-frequency oscillation of which the rst-order solution is of a di usive character. Similarly,
when free-surface waves propagate over a submerged cylinder an oscillatory environment is cre-
ated in which streaming motion about the cylinder is induced. In that case, even for an inviscid
uid, there is a time-independent ow about the cylinder induced by non-linear free-surface e ects.
The inviscid- ow problem for a circular cylinder has been studied by, for example, Ogilvie (1963),
Vada (1987) and Riley and Yan (1996). Ogilvie and Vada have concentrated mainly on the un-
steady ow whilst Yan and Riley have focused on the induced steady streaming, they note that
the steady-streaming induced at second order is not unique since an arbitrary circulation about the

0169-5983/00/$20.00 c 2000 The Japan Society of Fluid Mechanics and Elsevier Science B.V.
All rights reserved.
PII: S 0 1 6 9 - 5 9 8 3 ( 9 9 ) 0 0 0 1 5 - 5
70 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

cylinder can always be added associated with bound vortices within it. In a subsequent paper Yan
and Riley (1996) have shown, by considering a uid of small viscosity, how the non-uniqueness
of streaming at second order may be resolved. In a more recent paper Yan and Riley (1998) have
addressed the same problem at nite Reynolds number. The results for the circulation about the cylin-
der con rm their earlier asymptotic results for large Reynolds number. In addition, they demonstrate
good agreement between theory and experiment (Chaplin, 1984) for the second-order time-averaged
lift force on the cylinder.
In related studies Yan (1998a,b) has considered the ow, in uid of nite depth, that arises when a
submerged circular cylinder performs translational or orbital oscillations in the uid being otherwise
quiescent. The theory is developed for both the inviscid ow limit and at high Reynolds number.
Particular attention is focused on the second-order steady streaming due to the Reynolds stresses.
For the case of rectilinear oscillations parallel to the undisturbed free surface the solution is shown
to be non-unique, and critical Reynolds number at which the bifurcation appears is estimated.
In this paper we extend the work of Yan (1998a,b) by including a steady current which ows
parallel to the undisturbed free surface. The cylinder itself performs either rectilinear oscillations with
frequency ! parallel, or perpendicular, to the undisturbed free surface or a purely orbital motion in
which its centre moves periodically on a small circle. If a is the cylinder radius, and A the amplitude
of the oscillations, then we assume  = A=a1 and develop our theory accordingly. As in the earlier
cases the steady-streaming velocity that arises from the action of the Reynolds stresses is, O(2 !a),
the velocity in the primary oscillations is, of course, O(!a). We choose O() as the magnitude of
the steady current. Although this is a relatively slow ow, it is the current magnitude that brings
about, at second order, e ects that are comparable with the Reynolds stresses.
In Section 2 we address the inviscid problem, by expanding the velocity potential in powers of
. Each term of the expansion satis es Laplace’s equation, with appropriate free-surface conditions
derived from the non-linear kinematic and dynamic conditions at the free surface. To determine the
potentials we have used a boundary-element method as described by Riley and Yan (1996). For
viscous ow we have adopted, in Section 3, a stream function–vorticity approach. The problem is
formulated for arbitrary values of , with a Reynolds number Rb = a2 !=, where  is the kinematic
viscosity of the uid. Following the successful approach of Yan and Riley (1998), we adopt boundary
conditions at the free surface and at large distances upstream and downstream from the cylinder,
from the inviscid solution calculated in Section 2. Two methods have been used to solve the viscous
ow problem, mainly for Rb = 50. First, we have solved the full, unsteady Navier–Stokes equations,
integrating over a sucient number of periods until the solution is sensibly periodic. Although this
method is explicitly independent of , the fact that the smallness of  is implicit in the boundary
conditions we have adopted suggests that an expansion of the ow variables in powers of  might
again be appropriate. In doing this we are able to identify, at O(), the uctuating components with
the fundamental frequency as well as the time-averaged part of ow associated with the steady
current and, at O(2 ), the time-independent component and the uctuating parts of the solution
which include elements with the fundamental frequency and the higher harmonics with double that
frequency. This perturbation approach proves to be much more ecient, computationally, than that
based on the full Navier–Stokes equations. However, solutions from the latter do enable us to place
bounds on the parameter range for which the perturbation approach is valid.
From the results we have obtained, as are discussed in Section 5, we have looked, in particular,
at streamline patterns and the force acting on the cylinder. We note that the steady streaming at
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 71

second order, which is induced by the action of the Reynolds stresses, is overwhelmed by even
a modest steady current. In the absence of any movement of the cylinder the steady ow past it
exhibits fore and aft symmetry as appropriate for slow ow. The cylinder motion destroys this,
and in all cases of oscillatory motion of the cylinder a clearly de ned wake structure is observed
in the time-averaged ow. However, although eddies are apparent, we have observed no vortex
shedding from the cylinder at the Reynolds number employed. In calculating forces the leading order
time-averaged drag increases linearly with the current speed and at second order varies quadratically
with the current speed. The second-order e ect leads to a small drag reduction in all cases, except for
that of orbital motion of the cylinder when there is an enhancement of the drag when the cylinder is
suciently close to the free surface. In all cases the time-averaged lift force is directed away from
the free surface. For the time-varying force coecients we have concentrated on the inertia and
damping coecients. At leading order these exhibit no dependence on the steady current, but are
in uenced by the presence of the free surface. The characteristics exhibited are largely independent of
the type of oscillatory motion undertaken by the cylinder. In particular, we nd a slight reduction in
the inertia coecient as the cylinder approaches the free surface. By contrast, a signi cant increase
in the damping coecient is observed as the free-surface is approached. There are second-order
contributions to these coecients, which vary linearly with the current speed. In all cases a reduction
in the inertia coecient is predicted, which increases in magnitude as the cylinder approaches the
free surface. For the damping coecient we nd that in the case of rectilinear oscillations there is
a contribution which reduces the coecient. But for orbital motion of the cylinder an increase of
the damping coecient is predicted. In all cases the increment to the damping coecient increases
as the cylinder approaches the free surface.

2. Inviscid ow

In this section, as a preparation for the viscous ow calculations described in Section 3 below,
we consider the uniform, two-dimensional ow past a circular cylinder submerged beneath the free
surface of an inviscid uid that is of in nite depth. The cylinder itself performs a prescribed,
small-amplitude oscillatory motion perpendicular to its generators. It is convenient to work with the
velocity potential , in a frame of reference whose origin is xed at the mean position of the centre
of the cylinder, from which rectangular and polar coordinates, (x0 ; y0 ) and (r 0 ; ), respectively, are
measured as in Fig. 1.
If a is the radius of the cylinder, A the amplitude of its oscillation with frequency !, and g the
acceleration due to gravity, we de ne dimensionless quantities as follows:
x = x0 =a; y = y0 =a;  = A=a; h = H=a;
(2.1)
k = a!2 =g; t = !t 0 ;  = 0 =a;  = 0 =!a2 :
In Eq. (2.1) t 0 is time, 0 is the velocity potential of the assumed irrotational ow, H is the depth
of the mean position of the cylinder centre below the undisturbed free surface and 0 its elevation
above the undisturbed free surface.
The ow under consideration is governed by Laplaces equation
∇2  = 0; (2.2)
72 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 1. De nition sketch.

with the kinematic and dynamic boundary conditions at the free surface y = h + 
@ @ @ @
+ = ; (2.3)
@t @x @x @y
" ( 2  2 ) #
@ 1 @ @ 1
(x; t) = −k + + − 2 2 ; (2.4)
@t 2 @x @y 2
where a constant has been included in Eq. (2.4) for convenience, and @=@y → 0 as y → −∞.
With 1 we develop a solution by writing
 = x + (1) + 2 (2) + · · · ; (2.5)

 = (1) + 2 (2) + · · · ; (2.6)


where the order of magnitude of the steady current, Us , has been chosen to be comparable with the
velocities induced by the cylinder motion.
We now expand conditions (2.3) and (2.4) about the mean position of the free surface y = h and
collect the like terms of , after eliminating (1) and (2) from the resulting equations, we have the
following boundary conditions at y = h:
@2 (1) @(1)
k + = 0; (2.7)
@t 2 @y
!
@2 (2) @(2) @(1) @3 (1) @2 (1)
k + = k k +
@t 2 @y @t @y@t 2 @y2
!
@(1) @2 (1) @(1) @2 (1) @2 (1)
−2k + − 2k : (2.8)
@x @x@t @y @y@t @x@t
Condition (2.7) at the leading order is unchanged from the case when there is no steady current,
which now appears explicitly in the second-order condition (2.8). We assume that the cylinder
performs a harmonic oscillatory motion with its centre at the point (1 ; 2 ) where
1 (t) = (d11 cos t + d12 sin t); 2 (t) = (d21 cos t + d22 sin t); (2.9)
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 73

which indicates that (1) may be written as


(1) = 10 + 11 cos t + 12 sin t: (2.10)
Inspection of Eq. (2.8) then shows that we must write
(2) = 20 + 21 cos 2t + 22 sin 2t + 23 cos t + 24 sin t; (2.11)
where the oscillatory terms with the fundamental frequency in Eq. (2.11) are absent when there is
no steady current present. There are now eight potential functions ij = ij (x; y) to determine.
Substituting Eqs. (2.10), (2.11) into Eqs. (2.7) and (2.8) gives, as boundary conditions at y = h,
@10
= 0; (2.12)
@y

@1j
= k1j ; j = 1; 2; (2.13)
@y

@20
= F20 ; (2.14)
@y

@2j
= 4k2j + F2j ; j = 1; 2; (2.15)
@y

@2j
= k2j + F2j ; j = 3; 4; (2.16)
@y
where
( )
k @2 11 @2 12
F20 = − 12 − 11 ; (2.17)
2 @x2 @x2
( )
k @11 @12 @2 11 @2 12
F21 = − 4 + 6k 2 11 12 +  12 + 11 ; (2.18)
2 @x @x @x2 @x2
( " 2  2 # )
k @12 @11 2 @2 12 @2 11
F22 = − 2 − + 3k (212 − 211 ) +  12 − 11 ; (2.19)
2 @x @x @x2 @x2
(   )
@2 10 @10 @12
F23 = −k 12 + 2 + ; (2.20)
@x 2 @x @x
(   )
@2 10 @10 @11
F24 = k 11 + 2 + : (2.21)
@x 2 @x @x
When compared with the corresponding conditions in Yan (1998a) only those for 23 ; 24 involve
the steady current; the others remain unchanged.
74 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

We consider next the boundary condition at the cylinder surface. There we have
@ d1 d2
= n1 + n2 ; (2.22)
@n dt dt
where 1 and 2 are de ned in Eq. (2.9), n = (−cos ; −sin ) is the unit normal directed towards the
centre of the cylinder. Similar to Yan (1998a), we rst transfer this boundary condition to the mean
position of the surface of the cylinder and equate coecients of like power of , to give following
boundary conditions at r = 1:
@10
= −cos ; (2.23)
@r
@11
= d12 cos  + d22 sin ; (2.24)
@r
@12
= −(d11 cos  + d21 sin ); (2.25)
@r
 
@20 1 @ @11 @12
= (d11 cos  + d21 sin ) + (d12 cos  + d22 sin ) ; (2.26)
@r 2 @ @ @
@21
= (d11 d12 − d21 d22 ) cos 2 + (d11 d22 + d12 d21 ) sin 2
@r
 
1 @ @11 @12
+ (d11 cos  + d21 sin ) − (d22 cos  + d12 sin ) ; (2.27)
2 @ @ @
@22 1
= {(d221 − d222 ) − (d211 − d212 )} cos 2 − (d11 d21 − d12 d22 ) sin 2
@r 2
 
1 @ @11 @12
+ (d12 cos  + d22 sin ) + (d11 cos  + d21 sin ) ; (2.28)
2 @ @ @
  
@23 @ @10
= −(d11 cos 2 + d21 sin 2) + (d11 cos  + d21 sin ) ; (2.29)
@r @ @
  
@24 @ @10
= −(d12 cos 2 + d22 sin 2) + (d12 cos  + d22 sin ) : (2.30)
@r @ @
Finally, conditions at large distances from the cylinder are required. Radiation conditions are
required for 11 ; 12 ; 21 ; 22 ; 23 and 24 as x → ±∞, which are applied at x = xl ; xr (xl ¡ xr ),
whilst for 10 ; 20 we have @10 =@x; @20 =@x → 0. And as y → −∞ we require @ij =@y → 0 for all
i; j = 0; 1; 2: The unknowns ij all satisfy Laplace’s equation which is to be solved subject to the
conditions set out above.
With the velocity potential  determined up to, and including, terms O(2 ) we may calculate the
corresponding stream function using the Cauchy–Riemann equations. This we write as
(1)
= y +  + 2 (2)
+ ···
= {y + 10 +( 11 + d22 x − d12 y) cos t + ( 12 − d21 x + d11 y) sin t)}
2
+ ( 20 + 21 cos 2t + 22 sin 2t + 23 cos t + 24 sin t) + · · · ; (2.31)
where the form adopted in Eq. (2.31) is convenient for what follows.
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 75

3. Viscous ow

Following Yan (1998a), and Yan and Riley (1998), we use the inviscid solution, determined above,
to provide boundary conditions at the free surface and at the remote locations x=xl ; xr , for the viscous
ow calculations. For this it is convenient to adopt a stream function–vorticity formulation in a frame
of reference xed to the moving cylinder, with rectangular and polar coordinates (x;  y);
 (r;
 ) as in
Fig. 1.

3.1. Viscous formulation

If we de ne non-dimensional quantities as in Eq. (2.1), where U0 = a! is a scale for velocity,


then the two-dimensional governing equations for the viscous ow can be written in terms of the
stream function  and vorticity component  perpendicular to the (x;
 y)-plane,
 as
@ @ @ 1 
+ u + v = ∇2xy ; ∇2xy  = −;
 (3.1)
@t @x @y Rb
where
@2 @2 @ @
∇2xy = + ; u = ; v = − : (3.2)
@x2 @y 2 @y @x
Rb = U0 a= = a2 != is a Reynolds number, with  the kinetic viscosity of the uid.
If the plane-polar coordinate system (r;
 ), with its origin located at the centre of the cylinder, is
chosen then the governing equations in terms of  and  may be written as
 
@ 1 @  + @ (u )
 = 1 ∇2 ;
+ (rur) ∇2r  = −;
 (3.3)
@t r @r @ Rb r
where
@2 1 @ 1 @2 1@ @
∇2r = + + ; u
 r
 = ; u
  = − : (3.4)
@r2 r @r r2 @2 r @ @r
Again we assume that the depth of the uid is in nite, and the ow at the free surface and at
large distances from the cylinder is close to the inviscid potential ow, so that boundary conditions
to be satis ed by Eqs. (3.1)–(3.2), or Eqs. (3.3)–(3.4), are as follows:
@ @
= = 0 at r = 1;
@r @
 = y +  (1) + 2 (2) ;  = 0 at y = h − 2 (t); xl 6x6
 xr ; (3.5)
 = y +  (1)
+ 2 (2)
;  = 0 when x = xl ; xr ; −∞ ¡ y6h
 − 2 (t);
 = ;  = 0 as y → −∞; xl 6x6
 xr ;
where (1) , (2) and are inviscid solutions as obtained in Eq. (2.31), xl and xr are the left and
right boundaries of the solution domain of which the magnitudes are supposed to be large enough
such that the numerical solution has sucient accuracy, but usually smaller than those in Section 2.
The boundary conditions for  and  on the surface of the cylinder require care and we will return
76 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

to this point in the next section. Two methods have been applied to solve the above viscous ow
problem, namely, a full numerical technique as described in Section 3.2 and a perturbation method
which will be described in Section 3.3.

3.2. Numerical solution Rb = O(1)

The computational grid that we use for our numerical solutions is a hybrid one, in two parts.
In the neighbourhood of the cylinder we use a grid based on plane polar coordinates, elsewhere
a grid based on Cartesian coordinates. Central di erences are employed in our numerical scheme
and, in the overlap domain, ow variables are interpolated between the grids using techniques that
maintain second-order accuracy. Since the solution domain extends to y = −∞ it is convenient, for
computational purposes, to introduce a new variable de ned as

 y   if y¿y
 b;
 = (y)
 = yb (3.6)
 yb + 1 − if y ¡ yb :
y
In (3.6) yb and are constants to be chosen, with the requirement that yb ¡ − re where re is the
radial extent of the polar grid, so that 1 ¡ re ¡ h. The solution domain is now divided into three
sub-domains as
I : 16r6 re ; 06 ¡ 2,
II : xl 6x6
 xr ; −∞ ¡ y6y
 b and
III : {
\ {
I ∪
II }} ∪
{re − 5hr6r6
 re ; 06 ¡ 2}, here hr is the mesh size in the radial direction. Note that, as de ned,

I and
III share an overlap region in which interpolation of ow variables from one grid to the
other can be carried out easily.
Following transformation (3.6), the governing equations on the rectangular grid
II and
III may
be written as
@ @ @ 1 
+ u + 1 v = ∇2x ; ∇2x  = −;
 (3.7)
@t @x @ Rb
where
@2 2 @
2
@ @ @
∇2x = 2 + 1 + 2 ; u = 1 ; v = − ; (3.8)
@x @ 2 @ @ @x
1 = ( − − yb )2 =( yb ) in
II and 1 in
III , 2 = 1 (@ 1 =@).
In our numerical scheme the mesh sizes in the t−, x−,  −, r−,  and -directions are denoted by
ht ; hx; h ; hr and h , respectively. The value of a typical dependent variable, say  , at (xi ; j ; tl ) =
(ihx; jh ; lht ) in
II and
III or (ri ; j ; tl ) = (1 + ihr; jh ; lht ) in
I is denoted by  i; j; l . A Crank–
Nicolson scheme is employed for the vorticity transport equations (3.3) and (3.7), whilst in Poisson
equations (3.3) and (3.7) the nite-di erence scheme is based on central di erences. In the overlap
region
I ∩
III , as the solution is carried from one mesh to the other, interpolation techniques are
used that maintain second-order accuracy.
There are no restrictions on the parameter in Eq. (3.6). Here we relate it to the values of h
and yb . With h given we have yb = −Nb h , and we choose such that (yb − h ) = yb − h , i.e.
= yb − h . This choice ensures that in the original plane the mesh size in the y-direction  is the
same at points adjacent to y = yb , see Yan and Riley (1998).
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 77

The boundary condition at the free surface for  , determined from the inviscid solution, may be
written in the present frame of reference as
( " !
 |(x; y=h) (1) @ (1)
 = y +  + 2 (2)
+ +  (d21 cos t + d22 sin t)
@y
#)
(1)
@
+ (d11 cos t + d12 sin t) ;
@x (x;y=h)

(x; y=h)
 = 0; (3.9)

In our stream function–vorticity formulation we require a boundary condition for  at the cylinder
surface. Following Woods (1954) the vorticity on the boundary may be expressed in terms of its
values at r − 1 = hr; 2hr, and the values of  at the rst three grid points, to the accuracy of our
nite-di erence scheme, as

48  1; j; l − 3  2; j; l − 45  0; j; l + 2h2r (41; j; l − 2; j; l )


0; j; l = ; j = 1; 2; : : : ; l = 1; 2; : : : ; (3.10)
4h2r (hr − 3)

the no-slip condition at r = 1 has been used in deriving Eq. (3.10).


There remains, to complete the formulation of our problem, the constant value of  ,  0; j; l , at
the cylinder. Solutions may be obtained for any value of this constant; however only one has any
physical relevance, namely that which ensures the pressure is single valued. From the momentum
equation in the -direction this condition is seen to require
Z ! Z !
1 2
@3 @  @2  1 2
@
I2 ≡ − + d = d = 0; (3.11)
Rb 0 @r3 @r2 r=1

Rb 0 @r r=1


which, when discretised, yields an expression for  0; j; l in terms of  i; j; l and i; j; l , i = 1; 2.


At each time step an iteration procedure is adopted to obtain a converged solution. For the stream
function over-relaxation accelerates convergence with a relaxation factor, typically, 1.2 whilst for the
vorticity under-relaxation is necessary with a relaxation factor 0.5 for boundary points and, typically,
0.8 elsewhere. The iteration procedure is carried out until both

X  ki
1 − i; j; l ¡ ; |I2 | ¡ ; (3.12)
ki −1
i; j; l

where summation extends over all the mesh points, ki is an iteration count and  a pre-assigned
tolerance which is taken as 10−3 for all the results that we present.
From the converged solution the lift L, and drag D, on the cylinder may be determined from
which we de ne lift and drag coecients as

L L D D
CL = 2
= 3 2; CD = 2
= 3 2: (3.13)
U0 a a ! U0 a a !
78 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Each of these force coecients includes contributions from both shear and normal stresses, and can
be determined as
Z Z 2 Z ( !)
1 2  1 2 @ 
CL =   cos  d − p r=1 sin  d = − cos  d; (3.14)
Rb 0 r=1 0

Rb 0 @r r r=1


Z Z Z ( !)
1 2 2
1 2
@ 
CD = − r=1
 sin  d − p r=1
 cos  d = sin  d; (3.15)
Rb 0 0 Rb 0 @r r r=1


where p = p0 =U02 = p0 =a2 !2 .

3.3. Perturbation solution 1; Rb = O(1)

The computational method described in Section 3.2 is independent of ; the problem is charac-
terised by the single parameter Rb . However, for the purposes that we employ the numerical scheme,
 is indirectly involved in the construction of the boundary conditions at the free surface and x=
 xl ; xr .
This is based on the inviscid analysis of Section 2 in which it is assumed 1. It therefore seems
appropriate to exploit the smallness of  in a perturbation scheme. As we shall see, this leads to a
more ecient means of calculating the time-independent part of the solution, that is of a particular
concern to us.
With 1 we expand the viscous stream function and vorticity, respectively, as
(u) (s)
 (x;
 t) =   1 (x;
 t) + 2 {  2 (x;
 t) +  (x)}
 + O(3 ); (3.16)
(u) (s)
 x;
(  t) = 1 (x;
 t) + 2 {2 (x;
 t) +  (x)}
 + O(3 ): (3.17)
In these equations x is a position vector and the subscripts (u) and (s) emphasise the time-dependent
and time-independent components of the solution at O(2 ). As in Yan and Riley (1998), it proves
convenient to separate the variables in Eqs. (3.16)–(3.17) by writing
 =  +  cos t +  sin t = y + f10 + f11 cos t + f12 sin t;
1 10 11 12
(u)
 =  21 cos 2t +  22 sin 2t +  23 cos t +  24 sin t
2
=f21 cos 2t + f22 sin 2t + f23 cos t + f24 sin t; (3.18)
1 = g10 + g11 cos t + g12 sin t;
(u)
2 = g21 cos 2t + g22 sin 2t + g23 cos t + g24 sin t;
 (s) =  (s)
= f20 ;  = g20 ;
20

where fij and gij are unknown functions of either (r;  ) in


I or (x;  ) in
II ;
III . Substituting
Eqs. (3.16)–(3.18) in Eq. (3.7) and their counterparts in
I , collecting like powers of  and then
di erent harmonics in t yields
∇2 g10 = 0; ∇2 f10 = −g10 ; (3.19)

∇2 g11 = Rb g12 ; ∇2 f11 = −g11 ; ∇2 g12 = −Rb g11 ; ∇2 f12 = −g12 ; (3.20)

∇2 g20 = X0 ; ∇2 f20 = −g20 : (3.21)


B. Yan / Fluid Dynamics Research 26 (2000) 69–94 79

∇2 g21 = X1 + 2Rb g22 ; ∇2 g22 = X2 − 2Rb g21 ; ∇2 f21 = −g21 ; ∇2 f22 = −g22 ; (3.22)

∇2 g23 = X3 + Rb g24 ; ∇2 g24 = X4 − Rb g23 ; ∇2 f23 = −g23 ; ∇2 f24 = −g24 ; (3.23)


2
where ∇ represents ∇2r in
I or ∇2x in
II ,
III , other quantities are de ned as
 
Rb @(g12 ; f12 ) @(g11 ; f11 ) @(g10 ; f10 + r sin )
X0 = + +2 ;
2r @(r;
 ) @(r;
 ) @(r;
 )
 
Rb @(g11 ; f11 ) @(g12 ; f12 )
X1 = − ;
2r @(r;
 ) @(r;
 )
 
Rb @(g11 ; f12 ) @(g12 ; f11 )
X2 = + ; (3.24)
2r @(r;
 ) @(r;
 )
 
Rb @(g10 ; f11 ) @(g11 ; f10 + r sin )
X3 = + ;
2r @(r;
 ) @(r;
 )
 
Rb @(g10 ; f12 ) @(g12 ; f10 + r sin )
X4 = + ;
2r @(r;
 ) @(r;
 )
in
I and similar expressions for Xi (i = 0; 1; : : : ; 4) in
II and
III .
The boundary conditions to be satis ed by Eqs. (3.19)–(3.23) at the free surface, and the remote
boundaries, may be summarised as follows:
f10 |(A; B) = 10 |(A; B) ; g10 |(A; B) = 0;
f11 |(A; B) = ( 11 + d22 x − d12 y) |(A; B) ; g11 |(A; B) = 0; (3.25)
f12 |(A; B) = ( 12 − d21 x + d11 y)|(A; B) ; g12 |(A; B) = 0;
 
1 @ 11 @ 12 @ 11
f20 |(A; B) = 20 + d11 + d12 + d21
2 @x @x @y

@ 12
+ d22 + 2(d11 d22 − d12 d21 ) ;
@y (A; B)
  
1 @ 11 @ 12 @ 11 @ 12
f21 |(A; B) = 21 + d11 − d12 + d21 − d22 ;
2 @x @x @y @y (A; B)
  
1 @ 11 @ 12 @ 11 @ 12 (3.26)
f22 |(A; B) = 22 + d12 + d11 + d22 + d21 ;
2 @x @x @y @y (A; B)
  
@ 10 @ 10
f23 |(A; B) = 23 + d11 + d21 + d21  ;
@x @y (A; B)
  
@ 10 @ 10
f24 |(A; B) = 24 + d12 + d22 + d22  ;
@x @y (A; B)

g2j |(A; B) = 0; j = 0; 1; 2; 3; 4;


in which the quantities ij are as in Eqs. (2.31). In Eqs. (3.25)–(3.26) we have, at the free surface,
A = x; A = x;
 B = B = h, on the remote left-hand boundary A = xl ; A = xl ; B = y; B = y,
 and on the
remote right-hand boundary A = xr ; A = xr ; B = y; B = y.

80 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

In addition, we again require conditions for the stream function, and the vorticity, at the cylinder
surface r = 1. As in our earlier discussion, Section 3.2, the value of each term in the expansion for
 in Eq. (3.16), at r = 1, must be determined to ensure that the pressure remains single-valued. As
a consequence each term must satisfy the integral constraint (3.11) from which the value of each,
at the boundary, is determined. Similarly, our earlier representation (3.10) for the boundary vorticity
must be satis ed by each term in expression (3.17).
Eqs. (3.19)–(3.23), to be solved subject to Eqs. (3.25)–(3.26) and appropriate conditions at r = 1,
are discretised using a central-di erence scheme and the resulting nite-di erence equations are
solved numerically by a point relaxation technique similar to that outlined in Section 3.2.
With the solution so obtained we may write the force coecients CL ; CD in the form

CL = (CL10 + CL11 cos t + CL12 sin t)


+ 2 (CL20 + CL21 cos 2t + CL22 sin 2t + CL23 cos t + CL24 sin t) + · · · ; (3.27)

CD = (CD10 + CD11 cos t + CD12 sin t)


+ 2 (CD20 + CD21 cos 2t + CD22 sin 2t + CD23 cos t + CD24 sin t) + · · · : (3.28)
Each of the coecients of the time-harmonic terms in these expressions may then be expressed in
terms of gij , as in Eqs. (3.14) and (3.15). For example
Z 2   
1 @ g20
CL20 = − cos  d; (3.29)
Rb 0 @r r r=1


Z 2   
1 @ g20
CD20 = sin  d: (3.30)
Rb 0 @r r r=1


In Section 4 we describe, brie y, numerical techniques used in our computations.

4. Numerical techniques

The numerical techniques for the potentials ij of Section 2 that we have used are based on the
boundary-element method described by Riley and Yan (1996), the details are not repeated here.
The extent of the computational domain for these calculations is given by xr = −xl = 15. For the
viscous calculations a hybrid mesh structure, as described in Section 3, has been employed. For
the full unsteady Navier–Stokes calculations of Section 3.2 a fully implicit Crank–Nicolson scheme
is adopted, which has second-order accuracy. The solution is marched forwards in time until it
is sensibly periodic; this usually requires some 30 to 40 complete periods and represents a con-
siderable computational e ort. The time step adopted for these calculations was ht = =30: The
equations to be solved for the quantities  ij in the perturbation scheme of Section 3.3 are indepen-
dent of t, and although eight such quantities are to be calculated, this approach is computationally
much more ecient. Central di erences are used for their discretisation. In all cases solution of
the nite-di erence equations that result from discretisation is carried out using a point-relaxation
technique. For the viscous calculations we have taken xr = −xl = 8, and the spatial mesh sizes are
chosen as hx =0:05; h =0:05; hr =0:025; h ==80. Note for the solution domain
I a smaller mesh
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 81

size is used since ow variations will be larger close to the cylinder. The above values are chosen
such that any further increase in the size of the solution domain and decrease in mesh sizes result in
no signi cant change in the solutions. In Section 5 we compare results obtained by the perturbation
method with those of the full equations for small . Both approaches are, in principle, appropriate
for all nite values of Rb . However, there are numerical constraints associated with the development
of a boundary layer, thickness O(R−1=2
b ), at the cylinder surface r = 1. This is the familiar Stokes
shear-wave layer. For the mesh sizes that we have been able to employ this Stokes layer may be
satisfactorily resolved up to values of Rb = O(102 ). A detailed discussion of the Stokes layer, for
Rb  1, that is relevant to the present investigation, has been given by Yan and Riley (1996).

5. Results

For the most part we concentrate on results from the perturbation scheme of Section 3.3. We use
the results from the full Navier–Stokes equation to estimate the limits on the permitted size of the
parameters involved. As we shall see it is , rather than  itself, that has to be suciently small
for the perturbation approach to succeed. All our calculations have been carried out with k = 0:5,
Rb = 50 except two cases shown in Fig. 10 where Rb = 75. Also, for the cylinder displacement
(2.9) we have taken d12 = d21 = 0 with d11 ; d22 equal to unit or zero. Thus with d11 = 1; d22 = 0
the cylinder performs oscillations parallel to the undisturbed free stream and free surface, whilst
for d11 = 0; d22 = 1 these oscillations are perpendicular to the free stream and free surface. With
d11 = d22 = 1 the cylinder performs a purely orbital motion, with its centre moving in a small circle
counterclockwise.
We rst consider the streamline patterns associated with the constituent stream function elements
 , and initially in the absence of a steady current. The streamlines are shown rst in Fig. 2 when the
ij
cylinder is at a depth h = 6, well away from the free surface. In that case the ow closely resembles
the ow when the cylinder oscillates in an unbounded uid. Relevant to the present work is the
analysis of Amin (1988) who considers such a problem for Rb 1 with d11 = 1; d22 = 0 for which
we show in Fig. 2(a). There we see at leading order streamlines  12 = constant associated with the
velocity eld in phase with the cylinder velocity, whilst  11 = constant are the streamlines associated
with the out-of-phase velocity eld. At second order there is an induced steady streaming,  20 . This
displays the characteristic four-cell structure as shown by the streamline pattern. The streamlines
associated with the higher harmonics  21 ;  22 also display a characteristic four-cell structure. These
streamline patterns accurately mirror the analysis of Amin, except that in the present case they are
not as di use, since Rb is not small. Indeed if Rb were to increase, the cellular structures evident in
these streamline patterns would compress further, eventually forming the classical Stokes shear-wave
layer. In Fig. 2(b) the ow patterns are simply rotated through an angle − 12  except that it is now
the velocity eld associated with  11 that is in phase with the cylinder velocity. Fig. 2(c) shows the
streamline patterns for a purely orbital motion of the cylinder. At leading order each component of
the uid velocity is in phase with a component of the cylinder velocity. At second order the induced
steady streaming, represented by  20 , is quite di erent from the previous cases with counterclockwise
circulation about the cylinder; and we see that the cellular pattern of the higher harmonics, whilst
still present, has become distorted. These features of the ow are consistent with the analysis of
Riley (1971) carried out for Rb  1. We now consider these same cases when the cylinder is moved
82 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 2. Contours of stream functions  ij for h = 6 and  = 0. (a) for d11 = 1; d22 = 0, (b) for d11 = 0; d22 = 1, (c) for
d11 = 1; d22 = 1.
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 83

Fig. 2. Continued

closer to the free surface with h = 2, as shown in Fig. 3. At leading order the streamline pattern in
phase with the cylinder motion is still recognizable, but the others are grossly distorted except for the
induced streaming, represented by  20 , for orbital ow where, in Fig. 3(c) we see that the streamline
pattern is similar to the previous case in the sense that is composed of closed streamlines encircling
the cylinder. Perhaps the most striking changes are associated with the higher harmonics  21 and
 . In each of the cases shown in Fig. 3 these elements of the ow clearly display the higher
22
harmonic wave, frequency 2, that the oscillating cylinder induces at the free surface. In particular,
for the case of purely orbital ow, shown in Fig. 3(c), we see that the wave generated at the free
surface propagates only to the left; a feature that has been discussed earlier by Ogilvie (1963) and
Yan (1998a).
We now move on to consider the e ect of a current, uniform speed , on these ow patterns.
First, we note that from the boundary-value problems posed in Section 3.3 for the perturbation
stream functions  11 ;  12 ;  21 ;  22 there is no dependence upon , and these remain unchanged
as in Figs. 2 and 3. The steady current stream function  10 has  as a multiplicative factor, and
the elements  23 ;  24 which are now non-zero and vary linearly with  as we see from boundary
conditions (3.26). These latter two quantities are not of especial interest as far as the streamline
pattern is concerned, and we do not consider them further. The most dramatic impact that an imposed
current has is upon the second-order steady streaming  20 . The reason for this is to be found in the
source term X0 of Eq. (3.21), de ned in Eq. (3.24), which displays a quadratic dependence on . In
Fig. 4 we show the streamline patterns for the time-independent elements of the stream function for
84 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 3. Contours of stream functions  ij for h = 2 and  = 0. (a) for d11 = 1; d22 = 0, (b) for d11 = 0; d22 = 1, (c) for
d11 = 1; d22 = 1.
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 85

Fig. 4. Contours of steady current and streaming ow when h = 6; d11 = 1; d22 = 0;  = 0:1. (a)  10 for  = 1, (b)  20
for  = 0:5, (c)  10 +   20 for  = 0:5, (d)  10 +   20 for  = 1, (e)  10 +   20 for  = 2.

the case of in-line oscillations, d11 = 1; d22 = 0. Fig. 4(a) shows the steady current owing past the
cylinder, an almost symmetrical pattern at this depth h = 6. In Fig. 4(b) we present the modi ed
steady streaming,  20 , at second order for  = 0:5. Compared with Fig. 2(a) it appears that the
four-cell structure has simply expanded, although as we shall see this is somewhat misleading. As 
increases,  10 is simply scaled by a factor , and although the pattern associated with  20 changes
slightly it is essentially the same, modi ed by a factor 2 . Combining these, as in Eq. (3.16) gives
us the time-averaged ow. We show this time-averaged ow on Fig. 4(c) for  = 0:5 and  = 0:1 and
similarly for  = 1; 2 in Figs. 4(d) and (e). As  increases the e ect of the second-order streaming
becomes more pronounced. Next we consider the case, d11 = 0; d22 = 1, of perpendicular oscillations
with h= 6. We found that the second-order steady streaming  20 , for  =0:5, is almost identical with
that shown in Fig. 4(b) and hence not shown here. But when compared with Fig. 2(b), not only has
the cellular structure expanded but the ow direction has changed. The implication is that the steady
streaming associated with the oscillatory ow when there is no current plays virtually no role in the
second-order steady ow when a current of any reasonable strength is present. The time-averaged
ows are virtually identical with those shown in Figs. 4(c)–(e) and are not reproduced. Finally, in
Fig. 5 we consider the case of orbital ow with d11 = d22 = 1 and h = 6, the induced streaming for
 = 0, shown in Fig. 2(c), is stronger than in the other two cases and is not so easily swamped when
 6= 0. Fig. 5(a) shows the secondary streaming when  = 0:5 which is quite di erent from that of
Fig. 4(b). However, increasing  to  = 1, Fig. 5(b) shows a four-cell structure closely resembling
86 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 5. Contours of steady current and streaming ow when h = 6, d11 = d22 = 1,  = 0:1. (a)  20 for  = 0:5, (b)  20
for  = 1, (c)  10 +   20 for  = 0:5, (e)  10 +   20 for  = 1, (g)  10 +   20 for  = 2.

Fig. 6. Contours of  20 when  = 0:5, h = 2. (a) for d11 = 1, d22 = 0, (b) for d11 = 0, d22 = 1.

that of Fig. 4(b), and as  increases further the second-order streaming in all three cases converge.
In Figs. 5(c) and (d) we show the time-averaged ows for  = 0:5; 1 and  = 0:1 which di er a
little from those in Fig. 4(c). For  = 2:0 all three cases closely resemble the ow patterns shown in
Fig. 4(e). As before, we now consider the e ect of the free surface on these ow patterns from a
consideration of the case h = 2. As for the case h = 6 the e ect of the steady current on the induced
steady streaming represented by  20 is quite dramatic. We show this in Figs. 6(a) and (b) for the
cases d11 = 1; d22 = 0 and d11 = 0; d22 = 1, respectively, with  = 0:5. The resulting four-cell pattern
is a fairly obvious derivative of that when the cylinder is deeper. The similarities of these are again
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 87

Fig. 7. Time-averaged ow  10 +  20 when h = 2, d11 = 1, d22 = 0,  = 0:1. (a) for  = 0:5, (b) for  = 1, (c) for  = 2.

Fig. 8. Induced streaming  20 when d11 = d22 = 1 and h = 2. (a) for  = 0:5, (b) for  = 1, (c) for  = 2.

quite striking, given the di erences when =0 as in Figs. 3(a) and (b), for the same reasons as were
advanced for the greater depth h = 6. The time-averaged ow for this depth, with d11 = 1; d22 = 0
is shown in Fig. 7 for  = 0:1 and the three cases  = 0:5; 1; 2. These are almost indistinguishable
from the ow patterns for d11 = 0; d22 = 1. Finally, we consider orbital ow with d11 = d22 = 1. For
 = 0, as we have seen in Fig. 2(c) the induced streaming is stronger than in the other two cases
and this is also true when the cylinder is raised close to the free surface as in Fig. 3(c). This means,
again compared with two previous cases, that larger values of  are required before the induced
streaming,  20 is overcome by e ects of the steady current  10 . In Fig. 8 we show  20 for the
values  = 0:5; 1 and 2. Only at the largest value do we see the four-cell structure emerging that is
typical of the previous cases of translational oscillations. Fig. 9 shows the time-averaged ow for
this orbital motion with  = 0:1 and the three cases  = 0; 1 and 2. As anticipated above, it is only
at the highest value of , Fig. 9(c), that the e ects of the imposed current dominate at second order
(cf. Fig. 7(c)).
What we have seen above is that even a relatively slow current, and recall that in our dimensionless
framework this is , can signi cantly change the second-order streaming motion which, in the
absence of such a current, is generated by the Reynolds stresses. As we can observe from any of
the time-averaged streamline patterns presented above, the small-amplitude oscillatory motion of the
cylinder destroys the fore and aft symmetry when a steady current is present. A wake is created
whose structure depends not only on the amplitude of the oscillation but also on the magnitude of
the free-stream velocity.
We now comment on the validity of our approach of expanding the solution in powers of the
dimensionless amplitude of oscillation . We do this in two parts, rst by considering the nature
of the free-surface ow. In an earlier paper Riley and Yan (1996) analysed the propagation of a
88 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 9. Time-averaged ow  10 + 20 when h = 2, d11 = d22 = 1,  = 0:1. (a) for  = 0, (b) for  = 1, (c) for  = 2.

train of waves, dimensionless amplitude , over a xed, submerged cylinder. In an experimental


programme Chaplin et al. (1999) have been able to compare theory and experiment, in particular
free-surface pro les. For  suciently small excellent agreement has been recorded. However, for
each cylinder depth, when a critical value of  was exceeded this agreement broke down. This failure
coincided with the onset of wave-breaking, and for h = 2 the critical values of  was c ≈ 0:1. Of
course, in the present investigation waves are generated by a submerged, oscillating cylinder; but it
would seem not unreasonable to empirically adopt this value of  as one below which the present
approach can adequately describe the ow. However, in our series approach we have retained only
the leading terms, namely those O() and O(2 ), and this in turn must place a limit on the adequacy
of this approach. In addition, the steady current speed  appears as a second parameter. As we
have already seen the e ects of the steady current can readily overwhelm the steady streaming at
second order induced by the action of Reynolds stresses. For  ≈ 1 the two e ects are comparable
for purely orbital ow, see Fig. 8(b). However, for the translational oscillations  must be reduced
to the range 0.1–0.2 for the two e ects to be comparable. To demonstrate this we show in Fig.
10 the induced streaming when h = 6 and  = 0:1; 0:5 for d11 = 1; d22 = 0 and d11 = 0; d22 = 1.
Clearly, when  = 0:1, see Figs. 10(a) and (d), the ow near the cylinder is dominated by the
Reynolds stresses whilst that far away from the cylinder has been a ected more by the imposed
steady current. For  = 0:5, see Figs. 10(c) and (f), the ow structures are virtually the same, all
dominated by the e ect of the steady current. Also shown in Figs. 10(b) and (e) is the induced
streaming when h = 6;  = 0:1; Rb = 75 for d11 = 1; d22 = 0 and d11 = 0; d22 = 1. We see that a larger
value of Rb makes the ow less di use in both cases. The streaming layer (see the inner four-cell
structures) becomes much thinner when Rb increases, this agrees with the theoretical result which
says that the thickness of the Stokes streaming layer is proportional to R−1=2b . As we have argued
it is the quadratic dependence of X0 on , see Eqs. (3.21) and (3.24), that is crucial; and since
at leading order  10 varies linearly with  it seems reasonable to assume that  is the parameter
that places a limitation upon our perturbation approach in Section 3.3. To investigate this further we
have obtained solutions of the full Navier–Stokes equations, in the manner described in Section 3.2,
for the case  = 0:1; h = 2; d11 = 1; d22 = 0, and  = 0:5; 1:0; 2:0. These calculations are enormously
time consuming but suggest, when comparing the results with those obtained from the perturbation
approach, that agreement is only reached when 61, which in turn suggests a limitation 60:1 on
the perturbation approach.
For orbital motion of the cylinder the Reynolds stresses generate a circulation about the cylinder,
as is clear from Figs. 2(c) and 3(c). In Fig. 11 we show the radial distribution of circulation induced
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 89

Fig. 10. Induced streaming  20 when h = 6. (a) for d11 = 1, d22 = 0,  = 0:1, Rb = 50, (b) for d11 = 1, d22 = 0,  = 0:1,
Rb = 75, (c) for d11 = 1, d22 = 0,  = 0:5, Rb = 50, (d) for d11 = 0, d22 = 1,  = 0:1, Rb = 50, (e) for d11 = 0, d22 = 1,
 = 0:1, Rb = 75, (f) for d11 = 0, d22 = 1,  = 0:5, Rb = 50.

at second order for each of the cases shown in Figs. 2(c) and 3(c), that is with  = 0. In fact, the
induced circulation with  6= 0 is virtually indistinguishable from these. We also include in this
diagram the circulation calculated by Riley (1971) in the case of orbital motion of the cylinder in
unbounded uid, based on the limit Rb → ∞, for Rb = 50 and 100. The present calculations exhibit
many of the features of the asymptotic theory. In particular the region in which viscous e ects are
signi cant are comparable in extent, and there is an overshoot towards the edge of that region.
However, the values of the circulation, for the case h = 6 falls short of that predicted by the high-Rb
theory at the edge of the viscous layer, namely 6, by some 10%. This may be accounted for, in
part, by the presence of the free surface, but in large part by nite-Reynolds number e ects.
We consider next the forces acting on the cylinder, with reference to the force coecients de-
ned in Eqs. (3.27) and (3.28) In Fig. 12(a) we show the drag coecient CD10 associated with
the steady current. This increases linearly with  and is depth-dependent in the sense that there is
a signi cant increase in drag, as the cylinder depth beneath the free surface decreases. The cor-
responding second-order time-averaged drag coecient CD20 is shown in Figs. 12(b)–(d). For the
case of transverse oscillations, in Figs. 12(b) and (c) we see there is a quadratic variation with 
of CD20 which is as anticipated from our earlier discussion. There is little variation with depth in
either case and the coecient is quite small. Although we have described this as a drag coecient
it is negative, and therefore contributes a small thrust to the cylinder. For orbital ow, Fig. 12(d),
90 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 11. Radial distribution of the induced circulation for di erent values of h.

the behaviour is similar to the cases of transverse oscillation until the cylinder approaches the free
surface. In that case we see a signi cantly larger, now positive, drag which persists even when 
is zero due to the circulation induced around the cylinder. The corresponding lift coecients are
shown in Fig. 13. At leading order we have CL10 which again, of course, varies linearly with  and
is small in magnitude. It is now negative indicating a force directed away from the free surface
whose magnitude again increases as the depth decreases. The second-order force represented by CL20
is also directed away from the free surface in all three cases, and is now an order of magnitude
greater than the corresponding drag values.
For the unsteady force coecients, the form in which these are expressed, (3.27) and (3.28)
allows us, conveniently, to relate our results to Morison’s equation. In dimensional form Morison’s
equation for the ‘in-line’ oscillatory force may be written (see Sarpkaya, 1986) as
dU
F = aCd |U |U + a2 Cm ; (5.1)
dt
where, with axes xed in the cylinder, U is the oscillatory uid motion. In Eq. (5.1) Cd is the
damping coecient and Cm the inertia coecient. At leading order the unsteady force coecients
are una ected by the oncoming uniform current. Following accepted practice we approximate |U |U
by the rst term of its Fourier series expansion. In Table 1(a) we present calculated values of the
inertia and damping coecients for the case in which the cylinder performs oscillations parallel to
the undisturbed free surface, denoted by a subscript x, and in Table 1(b) the coecients when the
oscillations are perpendicular to the undisturbed free surface, denoted by a subscript y. Surprisingly,
perhaps, these are virtually indistinguishable. We note that the inertia coecient is not signi cantly
a ected by the presence of the free surface, whereas the damping coecient increases quite rapidly
as the free surface is approached. It is interesting to compare these results with those for a cylinder
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 91

Fig. 12. Force coecients CD10 and CD20 for various values of h. (a) CD10 , (b) CD20 for d11 = 1, d22 = 0, (c) CD20 for
d11 = 0, d22 = 1, (d) CD20 for d11 = d22 = 1.

oscillating in an unbounded uid. Asymptotic expressions for Rb  1 are set out by Sarpkaya (1986)
from which, with Rb = 50, we have Cm = 7:54; Cd = 1:63, values to which the results in Tables
1(a) and 2(b) appear to be close as the cylinder depth increases. For the case of an orbital motion
we have used Eq. (5.1) for each of the perpendicular directions of oscillation, and these are set out
in Table 1(c). The variation with depth is virtually identical in each case and the results are similar
to those set out in Table 1(a) and (b), with a slightly higher damping coecient.
At higher order there are contributions to the force coecients that uctuate with the fundamental
frequency, and so contribute to the inertia and damping coecients. Each of these varies linearly with
, vanishing when =0, and so contributes an amount km ; kd  to the inertia and damping coecients
Cm and Cd . Values of km and kd for the cases under consideration are set out in Table 2. In all cases
a small reduction in inertia coecient is predicted. For rectilinear oscillations either in line with the
undisturbed current, or perpendicular to it, there is a predicted very small reduction in the damping
coecient. This is at variance with the experiments of Chaplin (private communication) who nds
that with transverse vibrations the damping coecient increases with . However, in making such a
92 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Fig. 13. Force coecients CL10 and CL20 for various values of h. (a) CL10 , (b) CL20 for d11 = 1, d22 = 0, (c) CL20 for
d11 = 0, d22 = 1, (d) CL20 for d11 = d22 = 1.

Table 1
Inertia and damping coecients for di erent types of oscillation

(a) d11 = 1; d22 = 0 (b) d11 = 0; d22 = 1 (c) d11 = d22 = 1

h Cmx Cdx h Cmy Cdy h Cmx Cdx Cmy Cdy

2 6.6 2.9 2 6.7 3.0 2 6.5 3.2 6.6 3.2


4 7.2 1.6 4 7.2 1.6 4 7.1 1.9 7.2 1.9
6 7.4 1.5 6 7.4 1.5 6 7.4 1.7 7.4 1.7

comparison, care must be exercised since in the experiments Rb  1 and the values of  at which
measurements were made are such that vortex shedding from the cylinder takes place. For orbital
motion, Table 2(c), there is now a predicted increase in the damping coecient. That the increment
has a larger magnitude than for the case of rectilinear oscillations may perhaps be attributed to the
B. Yan / Fluid Dynamics Research 26 (2000) 69–94 93

Table 2
kd and km for di erent types of oscillations

(a) d11 = 1; d22 = 0 (b) d11 = 0; d22 = 1 (c) d11 = d22 = 1


h 100kmx 100kdx h 10kmy 100kdy h 10kmx kdx kmy kdy
2 −2.28 −1.92 2 −0.44 −1.07 2 −3.78 1.42 −0.62 0.67
4 −0.27 −0.82 4 −0.07 −1.07 4 −3.72 1.22 −0.23 0.06
6 −0.02 −0.31 6 −0.01 −0.41 6 −2.75 0.02 −0.11 0

more complex interaction between the oscillatory ow and the oncoming steady current that we have
already noted in the discussion of ow patterns above.

6. Conclusions

In this paper we have investigated the in uence of imposed steady current on the steady streaming
induced by an oscillating circular cylinder beneath a free surface. We have shown that when the
magnitude of the steady current is relatively large the induced steady streaming for both horizon-
tal and vertical oscillations is largely independent of the type of oscillation. However, when the
magnitude of the steady current is small, the Reynolds stresses produced by the oscillation become
signi cant in the neighbourhood of the cylinder and hence the induced steady streaming appears to
be of the classic four-cell structure. When the cylinder performs uniform orbital motion, the circula-
tion around the cylinder due to the induced steady streaming is almost independent of the magnitude
of the steady current and compares well with the analytic solution of Riley (1971).

Acknowledgements

This work is nancially supported by the Marine Technology Directorate. The author would like
to thank Professor N. Riley for his help during the preparation of this work and anonymous referees
for their constructive comments.

References

Amin, N., 1988. Low frequency oscillations of a cylinder in a viscous- uid. Quart. J. Mech. Appl. Math. 41, 195–201.
Chaplin, J.R., 1984. Nonlinear forces on a horizontal cylinder beneath waves. J. Fluid Mech. 147, 449– 464.
Chaplin, J.R., Riley, N., Yan, B., 1999. Note on a paper by Riley and Yan. J. Eng. Math., 35, 455–456.
Ogilvie, T.F., 1963. First- and second-order forces on a cylinder submerged under a free surface. J. Fluid Mech. 16,
451– 472.
Riley, N., 1967. Oscillatory viscous ows. J. Inst. Math. Appl. 3, 419– 434.
Riley, N., 1971. Stirring a viscous uid. ZAMP 22, 645– 653.
Riley, N., Yan, B., 1996. Inviscid uid ow around a submerged circular cylinder induced by free-surface travelling
waves. J. Eng. Math. 30, 587–601.
Sarpkaya, T., 1986. Force on a circular-cylinder in viscous oscillatory ow at low Keulegan-Carpenter numbers. J. Fluid
Mech. 165, 61–67.
Stuart, J.T., 1966. Double boundary layers in oscillating viscous ows. J. Fluid Mech. 24, 673– 687.
94 B. Yan / Fluid Dynamics Research 26 (2000) 69–94

Vada, T., 1987. A numerical solution of the second-order wave-di raction problem for a submerged cylinder of arbitrary
shape. J. Fluid Mech. 174, 23–37.
Woods, L.C., 1954. A note on the numerical solution of fourth order di erential equations. Aeronaut. Quart. 5, 176 –184.
Yan, B., Riley, N., 1996. Boundary-layer ow around a submerged circular cylinder induced by free-surface travelling
waves. J. Fluid Mech. 316, 241–257.
Yan, B., 1998a. Oscillatory ow beneath a free surface. Fluid Dyn. Res. 22, 1–23.
Yan, B., 1998b. Free surface ow generated by the orbital motion of a submerged circular cylinder in a uid of nite
depth. Proceedings of Third International Conference on Fluid Mechanics, Beijing, 1998.
Yan, B., Riley, N., 1998. Viscous ow about a submerged circular cylinder by free surface travelling waves. J. Fluid
Mech. 374, 173–194.

Potrebbero piacerti anche