Você está na página 1de 17

Journal of Turbulence

ISSN: (Print) 1468-5248 (Online) Journal homepage: http://www.tandfonline.com/loi/tjot20

The growth of a cylinder wake in turbulent flow

I. Eames , C. Jonsson & P. B. Johnson

To cite this article: I. Eames , C. Jonsson & P. B. Johnson (2011) The growth of a cylinder wake in
turbulent flow, Journal of Turbulence, 12, N39, DOI: 10.1080/14685248.2011.619985

To link to this article: https://doi.org/10.1080/14685248.2011.619985

Published online: 24 Oct 2011.

Submit your article to this journal

Article views: 1173

View related articles

Citing articles: 8 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=tjot20
Journal of Turbulence
Vol. 12, No. 39, 2011, 1–16

The growth of a cylinder wake in turbulent flow


I. Eames, C. Jonsson and P.B. Johnson∗

Department of Mechanical Engineering, University College, London, London, UK


(Received 8 November 2010; final version received 23 August 2011)

In the absence of ambient turbulence, the width of the downstream wake of a cylin-
der grows diffusively either through viscous effects or entrainment at the outer wake
edge, depending on the characteristic Reynolds number, and in both the cases it in-
creases as x 1/2 , where x is the distance downstream of the cylinder. We examine the
effect of ambient turbulence on the downstream wake of a cylinder, where this turbu-
lence is treated as homogeneous and isotropic. Wake spreading becomes affected when
the velocity deficit has decayed sufficiently that it is comparable with the root mean
square velocity of the ambient turbulence. A new model is developed to explain how
and when the ambient turbulence is sufficiently strong, and the wake is passively ad-
vected by the turbulence in the freestream. In this case, when the wake width is smaller
than the integral length scale, the wake grows linearly with distance downstream (i.e.
proportional to x), whereas when the wake width is larger than the integral length
scale, it grows diffusively (i.e. proportional to x 1/2 ). A laboratory study of the decay
of the velocity deficit behind a cylinder confirms the salient aspects of the new model
predictions.
Keywords: external aerodynamic turbulence; environmental turbulent flows; turbulent
mixing

1. Introduction
Many problems encountered in engineering involve a rigid structure moving relative to
a fluid, for instance, an off-shore platform in a current or ship moving in an ocean. The
flow around these structures is usually described in terms of a near wake (within one or
two body diameters downstream) and a far wake. The vorticity shed from the surface of
these structures creates a velocity deficit downstream, whose maximum amplitude decays
downstream as the wake spreads. It is important to understand how quickly the velocity
deficit behind these structures decays with distance or time, as this largely controls how
close energy extracting devices can be placed in tidal streams, and also the persistence of
wake signatures from ships, submarines and aircraft.
Most research has tended to focus on cases where the level of ambient turbulence is
low or on the influence of turbulence on the near wake structure. However, there are many
environmental flows where turbulence levels are quite high, for example tidal streams
exhibit turbulence intensities exceeding 20% [1, 2]. Few studies have examined the effect
of external turbulence on the flow far downstream of a rigid structure. The recent numerical
calculations of Legendre et al. [3] showed that behind a rigid sphere the velocity deficit


Corresponding author. Email: peter.johnson@ucl.ac.uk

ISSN: 1468-5248 online only


C 2011 Taylor & Francis
http://dx.doi.org/10.1080/14685248.2011.619985
http://www.tandfonline.com
2 I. Eames et al.

changes fundamentally from uw ∼ x −1 to uw ∼ x −2 when ambient turbulence is present,


where x is the distance downstream. This observed decay is consistent with the recent
experiments of Amoura et al. [4]. The corresponding case of a planar body, such as a
cylindrical strut, was studied by Symes and Fink [5], who conducted experiments in a wind
tunnel using two different levels of ambient turbulence. However, the ambient turbulence
in their experiments decayed faster than the velocity deficit and so had a rather weak effect
on the wake spreading. Only for their highest case of turbulence intensity (still only around
5%) did the freestream turbulence have a significant effect; in this case it seems to have
increased the rate of turbulent diffusion.
The purpose of this paper is to develop a new conceptual framework to understand
how intense ambient turbulence affects the wake of a rigid cylinder spreading downstream.
These processes depend on the level of turbulence and the integral length scale. In this
investigation, we limit the analysis to the case where the ambient turbulence is homogeneous
and isotropic (i.e. with a constant intensity and integral length scale throughout the flow
field), and where it is sufficiently strong to dominate the spreading. The paper is structured
as follows. In Section 2, we introduce the conceptual framework and mathematical model
to understand how the wake growth and velocity deficit are modified by ambient turbulence
and how various scalings emerge depending on the level of turbulence and the integral scale.
In Section 3 we describe an experimental study to test the new analysis and in Section 4 we
present and discuss its results, before drawing general conclusions in Section 5.

2. Mathematical model
Consider a rigid cylinder of diameter d, fixed in a homogenous and isotropic turbulent
flow of time averaged speed U∞ = u (see Figure 1(b and c)) and root mean square (RMS)
velocity u∗ = (u 2 )1/2 . The autocorrelation coefficient is defined as

ui (t) ui (t − τ )


R (τ ) = , (1)
u2∗

where the overbar is a time average. This can be used to define an integral length scale [6]:
 ∞
L = U∞ TL = U∞ R (τ ) dτ, (2)
0

where TL is the integral time scale. The flow is characterised by the Reynolds number
Re = U∞ d/ν, where ν is the kinematic viscosity, and dimensionless groups  = L/d and
It = u∗ /U∞ . It satisfies the Navier–Stokes equation

Du
ρ = −∇p + ∇ · τ , (3)
Dt

where ρ is the density, u = (u, v, w) is the velocity field, p is the pressure and τ =
µ(∇u + (∇u)T ) is the shear–stress tensor, µ being the dynamic viscosity. The force per
unit length on the cylinder is defined by

F (t) = (p I − τ ) · n̂ dS, (4)
SB
Journal of Turbulence 3

where I is an identity matrix, n̂ is the unit normal directed into the cylinder and SB is the
surface of the cylinder, of unit length directed into the z-direction. The time-averaged drag
force can be expressed in terms of a drag coefficient, i.e.

1
FD  = F · x̂ = CD ρ d U∞
2
, (5)
2

where CD  is the mean drag coefficient, which includes the effect of ambient turbulence
wherever relevant; for a cylinder the time-averaged lift force is zero. The use of the mo-
mentum flux in the mean flow downstream of the body to estimate the force requires the
contribution of the pressure gradient in the wake to be negligible. This gives
 ∞
FD  = F · x̂ = ρ u (U∞ − u) dy. (6)
−∞

When the velocity deficit is sufficiently weak, Equation (6) reduces to the well-known Betz
formula [7]:
 ∞
FD  = ρ U∞ (U∞ − u) dy. (7)
−∞

We proceed to develop a series of models to predict and understand how the wake spreads
in a non-turbulent and turbulent environment. The mathematical analysis for spreading in
a non-turbulent environment is well known; the model of wake spreading in a turbulent
environment forms the new element to this analysis.

2.1. Non-turbulent ambient flow


The characteristics of the wake flow depend on whether it is laminar, unsteady or turbulent.
The wake width, yw , is defined as the distance from the centreline where the velocity deficit
has decayed to e− 2 of its maximum value. Associated with the wake is a Reynolds num-
1

ber based on the velocity deficit, defined as Rew = |U∞ − u|yw /ν. For approximately
Re < 30, the wake flow is steady and laminar, and the momentum equation reduces
to [8]

∂u ∂ 2u FD 
U∞ = ν 2 − δ (x) , (8)
∂x ∂y ρ

where the right-hand term is the point drag force induced by the cylinder [9]. The streamwise
pressure gradient is negligible far downstream. In this laminar viscous regime, the wake
width, yw , increases due to diffusion, where

1 dy 2
U∞ w = ν, (9)
2 dx

which on integration yields

2νx
yw2 = yw2 (0) + . (10)
U∞
4 I. Eames et al.

Figure 1. General schematic of a cylinder in an incident turbulent flow showing the nota-
tion for this analysis, illustrated for large Reynolds number. (a) ambient turbulence is negligi-
ble, (b) ambient turbulence is significant with an integral scale similar to the cylinder diame-
ter and (c) ambient turbulence is significant with an integral scale much larger than the cylinder
diameter.
Journal of Turbulence 5

The relationship between the velocity deficit and wake width is established by determining
the similarity solution satisfying u(y) → U∞ as y → ±∞ and the integral constraint, i.e
Equation (7), to give
 
uw CD  d y2
= √ exp − 2 , (11)
U∞ 2 2π yw 2yw

where uw = U∞ − ū(x, y). From Equations (10) and (11), the maximum (centreline) ve-
locity deficit decays with distance downstream as

  12
um CD  d Re
∼ √ , (12)
U∞ 4 π x

where um = uw (x, 0).


Above Re  60, the flow is unsteady and a characteristic von Karman vortex street is
generated. Since lumps of positive and negative vorticity are shed alternately, the velocity
deficit only begins to decay when sufficient time has elapsed that their cores diffuse into
one another through a process called vorticity annihilation [10]. This occurs at a distance
∼ d.Re downstream, and explains why the wake Reynolds number, Rew , can usually be
assumed to be constant. These conclusions are supported by the recent detailed numerical
calculations in [11].
In the case of turbulent (high Re) wakes behind cylinders, the first detailed studies were
reported by Townsend [12–14] who proposed the idea of self-similarity. For Re > 200, the
wake is turbulent and spreads through a process that can be understood by appealing to the
steady Reynolds-averaged Navier–Stokes equation. This argument has also been given by
Tennekes and Lumley [6]. Far downstream of the body, the streamwise pressure gradient
in the wake is negligible and the momentum equation reduces to

∂u ∂   
U∞ =− uv . (13)
∂x ∂y

To solve this, a similarity solution is sought for the velocity deficit and Reynolds stress
such that

uw = um (x) ũ (ζ ) , (14)
u v  = u2w τ̃ (ζ ) , (15)

where ũ and τ̃ are non-dimensional functions and ζ = y/yw . Regardless of the shape of ũ,
from the integral constraint (7),

uw CD  d
= √ ũ (ζ ) , (16)
U∞ 2 2π yw
∞ √
where ũ is normalised such that ũ(0) = 1 and −∞ ũ(ζ ) dζ = 2π . Further, balancing
the integral constraint (7) requires um yw ∼ CD dU∞ , while the momentum equation (13)
requires U∞ um /x ∼ um /yw . In combination, these constraints yield yw2 ∼ CD dx. Hence,
6 I. Eames et al.

the wake grows as

2νx
yw2 = yw2 (0) + γ CD Re, (17)
U∞

where γ is a constant that depends on the characteristics of the turbulence in the wake. The
expression has been manipulated to allow for an easy comparison with Equation (10), and
it is clear that the growth of a turbulent wake is much more rapid than that of a laminar
wake. From Equations (16) and (17), the centreline velocity deficit decays downstream as

  12
um 1 CD  d
∼ √ . (18)
U∞ 4 π γ x

To determine the distribution of the velocity deficit, i.e. ũ, it is necessary to establish a
relationship between ũ and τ̃ . In taking a simplified approach, it is common either to apply
Prandtl’s mixing length hypothesis [15] or to assume that the eddy viscosity is constant
across the wake. As noted by Townsend [14], the latter provides a good description of the
interior of the wake, but a poor description of the outer regions where the turbulence is
intermittent. The former provides a poorer description of the interior of the wake with a
pointed profile over the centreline, but better predicts the edges. In the same paper [14],
Townsend also proposes a model that takes into account the distribution of the intermittency
factor, but that and other more sophisticated models shall not be discussed here.
Prandtl’s mixing length hypothesis [15] estimates the Reynolds stress as
 
 ∂u 
2 ∂u
u v  = −λl  , (19)
∂y  ∂y 

where λ > 0 is a dimensionless constant and l is a mixing length. Using this expression,
the solution to Equation (13) is
⎧  3/2 2   13   13

⎨ 9 9 9
ũ = 1− √ |ζ | , − <ζ < ; (20)
⎪ 10 2π α α

0, otherwise.

If the eddy viscosity is assumed to be constant, solution to Equation (13) is


 
y2
ũ = exp − 2 . (21)
2yw

Further details on the turbulent plane wake are given in, for example [16, Ch. 5] and
[6, Ch. 4].

2.2. Turbulent ambient flow


The influence of the ambient turbulence on the wake spreading depends on the nature of
the ambient turbulence. Consider the general case where ambient turbulence is generated
by a grid of mesh size M, located at x = xg , and where the RMS velocity and the integral
Journal of Turbulence 7

scale vary with downstream distance according to


   D
u∗ x − xg −B L x − xg
=A , =C . (22)
U∞ M M M

The exponents vary with the Reynolds number (Reg = U∞ M/ν) and geometry of the grid,
but typical values are D ∼ 0.4 and B ∼ 0.6 (see, for example [17]). If the downstream
variation in L and u∗ is very small over the region of interest these may be treated as
constants; the analysis that follows focuses on this case.
The analysis of Section 2.1 breaks down beyond a distance xT , where the maximum
velocity deficit, i.e. Equation (12), becomes comparable to the RMS of the ambient turbu-
lence (um ∼ u∗ ). By combining Equation (12) with Equations (10) and (17), this transition
distance is estimated to be

xT CD 2 Re xT CD 
lim ∼ , lim ∼ , (23)
Re<30 d 16π It2 Re>200 d 16π γ It2

respectively. Further downstream (x xT ), the ambient turbulence is stronger than the


velocity deficit (um u∗ ) and passively advects the velocity deficit. During this phase the
velocity deficit spreads according to

∂uw   FD 
+ U∞ x̂ + u · ∇uw = δ (x) , (24)
∂t ρ

and can be understood by analysing how passive material is advected in the turbulent flow.
The application of the Lagrangian model to analyse the effect of gustiness on wakes was
recently suggested by Larsen et al. [18] to determine the wake meandering caused by
unsteady wind.
Passive elements are advected to
 t
 
X = (X, Y ) = U∞ x̂ + u dt, (25)
0

in time t. An approximate solution can be developed in the limit where It 1, and the
displacement can be approximated by
 t
X = U∞ t, Y = v dt. (26)
0

Taylor [19] examined the dispersion of fluid elements in a homogeneous turbulent flow and
showed that the mean squared displacement increased at a rate
 t  t
1 dY 2
= v (t) v (t + τ ) dτ = u2∗ R (τ ) dτ, (27)
2 dt 0 0

where R(τ ) is the autocorrelation coefficient as defined in Equation (1). Strictly, this
equation is, of course, only valid if the turbulence is stationary in the Lagrangian setting,
which is the case only when the external turbulence completely dominates the dynamics.
The spreading described by Equation (27) is modelled as a stochastic process and the mean
8 I. Eames et al.

wake width is defined as yw = Y 2 . By the central limit theorem, the velocity deficit
then takes the form uw = E exp(−y 2 /2yw2 ). The constant E is determined using either the
integral constraint obtained from Equation (7) or Equation (24) to give
 
uw CD d y2
= √ exp − 2 . (28)
U∞ 2 2π yw 2yw

Using Taylor’s hypothesis [20], where um U∞ , the spread of passive elements in


time can be recast into the spread with distance downstream so that these results can be
related to the experiments. For yw < L, the cross-stream velocity is temporarily correlated:
R(τ ) ∼ 1 and
 x
1 dy 2 dx
U∞ w = u2∗ . (29)
2 dx 0 U∞

For large time when yw > L (since R(τ ) approaches zero),

1 dy 2
U∞ w = u2∗ TL , (30)
2 dx

where TL = L/U∞ is the integral time scale (from the definition in Equation (2)). In
this case the spreading is Fickian and characterised by an eddy diffusivity u2∗ TL = It u∗ L.
We derive analytical expressions for wake spreading for the case where u∗ and L are
constants. For x < L/It , or x/d < /It , from Equation (29), the wake width increases
as

yw2 = yw2 (0) + It2 x 2 , (31)

and ultimately the velocity decays according to

um CD  d
∼ √ . (32)
U∞ 2 2π It x

Thus, the initial effect of ambient turbulence leads to the wake width increasing linearly
with distance (a process sometimes referred to as ballistic spreading) and the velocity deficit
decaying rapidly.
The transition from ballistic to diffusive spreading occurs beyond a distance
x/d > /It , where

yw2 (x) = yw2 (0) + 2 It2 L x. (33)

Here the wake spreads diffusively at a rate set by the ambient turbulence. The corresponding
maximum velocity deficit is

  12
um CD  d2
∼ √ . (34)
U∞ 4 π It Lx
Journal of Turbulence 9

Table 1. Summary of the results in this paper, where x̃ = x+x0


d
, x0 being a virtual origin.

Wake width Equation Centreline velocity Equation


Regime (yw /d) number deficit (um /U∞ ) number

CD 
x < xT Re < 30 x /Re)1/2
(2 (10) √ (12)
4 π (x̃/Re)1/2
 
1 CD  1/2
Re > 200 (2γ CD x̃)1/2 (17) √ (18)
4 π γ x̃
 CD 
x > xT x̃ < It x̃ (31) √ (32)
It 2 2π It x̃
   12 CD 
x̃ > 2It2 x̃ (33) √ (34)
It 4 π It ( x̃)1/2

The analytical expressions, summarised in Table 1, provide an assessment of what


occurs when the ambient turbulence intensity is very weak (um /u∗ 1) or very strong
(um /u∗ 1). In the intermediate range, when none of the processes is dominant, the
spreading of the wake can only be understood by considering how the internal and external
fields of turbulence interact. Such an analysis is more complicated and beyond the scope
of this paper.

3. Experimental study
3.1. Experimental setup and methodology
The experimental study was undertaken in a recirculating water channel, 1.2-m wide,
0.7-m deep and 14-m long. The cylinder, having a diameter of 28 mm, was held by vertical
streamlined struts and was centred at 0.3 m below the water surface. The struts have a
minimum effect on the flow (negligible at the centre of the channel where measurements
were taken). Tests were run at Re = 1300 and 2100, the kinematic viscosity of the water
being taken as 1.0 × 10−6 m2 s−1 at 20◦ C.
The flow velocity was measured using a Nortek Vectrino+ acoustic Doppler velocimeter
(ADV), which produces a three-component measurement of velocity. Data were sampled
at 25 Hz for a period of 400 s, which was found to be long enough for the statistics to
be repeatable within a satisfactory margin of error. The sampling volume was cylindrical,
6 mm in diameter (in the y–z plane) and 3.1-mm long (in the x-direction). The sampling
volume is located 150 mm upstream of the probe so that the presence of the instrument
has a minimal effect on the flow in that volume. The flow was also seeded with ‘110P8’
particles designed to improve the acoustic signal; the particles of spherical borosilicate
glass had density of 1.1 g cm−3 and mean diameter of 10 µm, and 80% of the particles
have diameter between 5 and 19 µm.

3.2. Diagnostic tools


A rotation matrix was applied to each velocity data point to correct for the attitude of the
ADV probe. A rotation through two angles was performed where the same pair of angles
was used on all points within a given sample. The angles were determined by the continuity
condition that the mean of both cross-stream components of the velocity should be zero for
10 I. Eames et al.

each sample set. The magnitude of angles of rotation was about 1◦ and the effect of rotation
on the streamwise component and RMS values was negligible.
The time average and RMS of a streamwise velocity series were calculated from

1 
N
u= uk , (35)
N k=1

and


1  N
u∗ =  (uk − u)2 , (36)
N k=1

where N is the number of valid data points labelled as uk . The cross-stream mean and RMS
velocity components (v, w and v∗ , w∗ ) are defined in a similar fashion.
The ADV can produce some spuriously large velocity values, which tend to distort the
RMS value of a given sample. In order to ameliorate this problem, an algorithm was used to
remove anomalous points. The phenomenon of spikes in ADV time series is addressed by
Goring and Nikora [21], where they propose a method for detecting spikes based on plotting
the data and detecting outliers in phase space. As a criterion for distinguishing between valid
data and spikes, they use the expected absolute maximum of a normal random variable,
λσ σ , where σ is the standard deviation and

λσ = 2 ln N , (37)

and N is the number of samples.


Any data points with velocity values far away from the mean, using the criterion
in Equation (37), were removed. σ was estimated by the absolute median deviation as
suggested by Wahl [22]. To best reproduce the energy spectrum, all spikes were replaced
by the last valid data point as suggested by Parsheh [23].
Velocity measurements by ADV also have the associated problem of Doppler noise,
which must be accounted for. As explained by Garcı́a et al. [24], in many cases the level of
noise can be identified as a white noise plateau in the high frequency range of the energy
spectrum. In the cross-stream velocity components of our measurements, such a plateau
was identified, whereas it was absent for the streamwise velocity component. However, in
all cases the energy associated with the noise was determined to constitute no more than
7% of the measured turbulent kinetic energy, which is also consistent with the fact that the
flow is of relatively high energy level. The error introduced into the RMS velocity values
is never more than 5%, which is significant but acceptable for the purpose of the present
analysis (keeping in mind that they may be slightly biased towards a high value).
Further, it should be noted that the sampling strategy employed did not permit complete
spatial and temporal resolution. The Kolmogorov length scale and Kolmogorov time scale
of the turbulence were estimated (based on the isotropic dissipation rate) to be 0.2–0.5 mm
and 0.05–0.3 s, respectively, depending on position in the flow field. Since the size of the
sampling volume is more than an order of magnitude larger than this length, there is clearly
a filtering effect present. The same applies, but to a smaller extent, to the time scale, as the
reciprocal of the Nyquist frequency is comparable to the smallest scale. However, only a
very small proportion of the total turbulent kinetic energy is contained at these scales, and
Journal of Turbulence 11

0.09
0.09

0.08 0.08

u, v and w (ms−1 )
u, v and w (ms−1 )

0.07 0.07

0.06 0.06

0.05 0.05

0.04 0.04

0.03 0.03

0.02 0.02

0.01 0.01

0 0

−0.01 −0.01
0 2 4 6 8 10 12 −3 −2 −1 0 1 2 3
(a) x/d (c) y/d
0.2
0.2
0.18
0.18
0.16
u∗ /u, v∗ /u and w∗ /u
0.16
u∗ /u, v∗ /u and w∗ /u

0.14 0.14

0.12 0.12

0.1 0.1

0.08 0.08

0.06 0.06

0.04 0.04

0.02 0.02

0 0
0 2 4 6 8 10 12 −3 −2 −1 0 1 2 3
(b) x/d (d) y/d

Figure 2. Variation of the mean velocity (ū, v̄, w̄) (a) downstream and (c) cross-stream vertical from
the origin; and the RMS velocity (u∗ /u, v∗ /u, w∗ /u) (b) downstream and (d) cross-stream vertical in
the absence of the cylinder. The symbols correspond to streamwise (◦), cross-stream vertical (∗) and
cross-stream horizontal (+).

so for our purposes (i.e. measuring mean and RMS velocities) the filtering effect can be
neglected.

4. Experimental results
Figure 2 shows the variation of the mean and RMS velocity components along the center-
line (Figures 2(a and b)) and with water depth (Figures 2(c and d)). These experimental
measurements are taken in the absence of the cylinder and are indicative of the ambient
turbulence. The measurements show that the mean flow is uniform within the region of
investigation and the turbulent intensity is It = 0.18. The integral scale, as defined by
Equation (2), was estimated to be 0.19 m in the streamwise direction, which is an order of
magnitude larger than the cylinder diameter. The standard deviation of these measurements
was 0.08 m.
12 I. Eames et al.

0.7 0.7

u∗ /U∞ , v∗ /U∞ and w∗ /U∞

u∗ /U∞ , v∗ /U∞ and w∗ /U∞


0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
2 4 6 8 10 12 2 4 6 8 10 12

(a) x/d (b) x/d

Figure 3. Variation of the RMS velocity (u∗ /U∞ , v∗ /U∞ , w∗ /U∞ ) downstream along the centreline
in presence of the cylinder for (a) Re = 2100 and (b) Re = 1300. The symbols correspond to
streamwise (◦), cross-stream vertical (∗) and cross-stream horizontal (+).

Shear layers are generated at the channel walls and free surface, but over the domain
considered, the mean flow varied spatially by less than 5%. The non-uniformity of the RMS
values is probably due to Doppler noise, and the limited sampling period. The differences
between the RMS values in streamwise and cross-stream horizontal directions were small.
The cross-stream vertical RMS velocity was smaller than the other two components; this
may be due to the tank being wider than its depth.
The effect of the cylinder upon the turbulence is illustrated in Figure 3, which shows
the variation of the centreline RMS velocity with streamwise distance; it is the highest in
the near wake and decreases downstream, being higher than the ambient level throughout
the range of measurement. Figure 4 shows the Reynolds stresses in the wake. These
calculations are based on gradients that are steep compared to the size of the sampling
volume, which explains the low quality of the results. The Reynolds stress is self-similar
and proportional to u2m ∼ (x/d)−2 , which is consistent with the linear decay of the velocity
deficit. Also note that this self-preservation is approached much more rapidly than is the
case when the ambient flow is non-turbulent ([6, Ch. 4] quotes a value of around 200d),
which adds confidence to the assumption that the dynamics is dominated by the external
turbulence.
Measurements of the streamwise velocity are used to evaluate the momentum thickness,
defined as
 ∞  
u u
= 1− dy. (38)
−∞ U∞ U∞

The momentum thickness can be estimated from the measured cross-stream velocity pro-
files, giving 2 /d = 1.15. This is typical of the drag on the cylinder in turbulent flow at
Re ≈ 1000 – it is 15% higher than the drag coefficient of the cylinder in a uniform flow
for equivalent Reynolds numbers [1]. Far downstream of a rigid body, the measure 2 /d
ultimately tends to CD  (from Equation (6)).
Figure 5 shows three cross-stream velocity profiles taken at different distances down-
stream of the cylinder. The velocity is normalised by the maximum centreline velocity and
Journal of Turbulence 13

2.5

1.5

0.5
y/d

−0.5

−1

−1.5

−2

−2.5
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
2
τxy/ρU∞ (x/d)2

Figure 4. The component of Reynolds stress, τxy , cross-stream vertically in presence of the cylinder
for Re = 2400. The symbols correspond as follows: ◦ for x/d = 2, ∗ for x/d = 3 and + for
x/d = 5.5.

a Gaussian profile is plotted for reference. The measurements confirm an approximately


Gaussian form, as anticipated by our model. The most likely reason for the difficulty in
obtaining better quality cross-stream profiles is that the sampling volume was too large in
relation to the velocity gradient in the regions of maximum shear.

0.8
uw/um

0.6

0.4

0.2

0
−3 −2 −1 0 1 2 3
y/yw

Figure 5. The normalised cross-stream velocity deficit profiles at three different downstream po-
sitions, with y normalised by the wake width and the velocity deficit normalised by the velocity
deficit at the centreline. The symbols correspond as follows: ◦ for x/d = 2, ∗ for x/d = 3 and + for
x/d = 5.5. The full line is the Gaussian profile exp(−y 2 /2yw2 ).
14 I. Eames et al.

10

6
U∞/um

0
2 4 6 8 10 12
x/d

Figure 6. Inverse of the velocity deficit (U∞ /um ) is plotted as a function of distance downstream of
a rigid cylinder for two Reynolds numbers. The symbols correspond as follows: + for Re = 2100
and ∗ for Re = 1300. The full line represents the theoretical prediction, i.e. Equation (32), using
CD  = 1.15 and It = 0.18.

The large integral scale of the ambient turbulence ensures that the decay of the RMS
velocity over the considered region is negligible. To test Equation (32), the inverse of the
centreline velocity deficit is shown in Figure 6 as a function of downstream distance, and
is compared to the theoretical prediction. The agreement is good. The ballistic spreading
regime (as described by Equations (31) and (32)) is predicted to occur when uw /u∗ 1
and yw /L 1. However, the experimental results seem to suggest that this may occur
even before uw /u∗ reaches unity, given that the second condition is met. For the results in
Figure 6, uw /u∗ is initially around 5. The theoretical prediction, i.e. Equation (32), relies
on the drag coefficient, which is determined using Equation (38); this technique is only
accurate to within about 10%, and could explain the slight overprediction of um /U∞ as
compared with the experimental results. Further, the prediction is dependent on the value
of It , which is also subject to experimental error of up to 5%.

5. Conclusion
Ambient turbulence has a profound effect on how rapidly the wake signature of a body
spreads and the maximum velocity deficit decreases with distance downstream. We have
developed a new model that describes how the wake spreads in a highly turbulent flow,
where it ultimately behaves passively. The model incorporates passive transport of the
wake with an integral constraint that links um and yw . The major results are summarised
in Table 1. For two-dimensional bodies, the wake grows linearly with distance during the
ballistic regime until the wake width is comparable to the integral scale, beyond which the
growth is diffusive.
A laboratory study of the flow downstream of a cylinder in the presence of intense
ambient turbulence was presented. The integral length scale of the freestream turbulence
was large compared to the cylinder diameter. The experimental results are compared with
Journal of Turbulence 15

the theoretical predictions and show good agreement; this confirms the ballistic regime and
x −1 decay of velocity deficit.
The analysis in this paper does not explain what occurs in a situation where both
the internal and external fields of turbulence are dynamically significant to the spreading
process. Further research on this matter is required to establish a complete picture of how
a plane wake spreads under the effect of various levels and characteristics of ambient
turbulence. In addition, we consider a freestream turbulence that is homogeneous and
isotropic; the effects of departure from such homogeneity and isotropy is an interesting
topic that should be pursued further.

Acknowledgements
IE is supported by a Senior Leverhulme Fellowship. PBJ acknowledges support from PhD supervisors
Kevin Drake and Adam Wojcik, and an EPSRC doctoral training award.

References
[1] W.H. Bell, The influence of turbulence on drag, Ocean Eng. 6 (1979), pp. 329–340.
[2] G. McCann, M. Thomson, and S. Hitchcock, Implications of site specific conditions on the
prediction of loading and power performance of a tidal stream device, in Proceedings of the
2nd International Conference on Ocean Energy (ICOE), 15–17 October, EMEC, Brest, France,
2008.
[3] D. Legendre, A. Merle, and J. Magnaudet, Wake of a spherical bubble or a solid sphere set fixed
in a turbulent environment, Phys. Fluids 18 (2006), pp. 048102:1–4.
[4] Z. Amoura, V. Roig, F. Risso, and A. Billet, Attenuation of the wake of a sphere in an intense
incident turbulence with large length scales, Phys. Fluids 22 (2010), pp. 055105:1–9.
[5] C.R. Symes and L.E. Fink, Effects of external turbulence upon the flow past cylinders, in Lecture
Notes in Physics: Structure and Mechanisms of Turbulence I, Vol. 75, Springer Verlag, Berlin,
1978, pp. 86–102.
[6] H. Tennekes and J.L. Lumley, A First Course in Turbulence, MIT Press, Cambridge, MA, 1972.
[7] A. Betz, A method for the direct determination of profile drag, Z. f. Flugtechn. Motorluftschif-
fahrt 16(42) in German.
[8] G.K. Batchelor, An Introduction to Fluid Mechanics, Cambridge University Press, Cambridge,
UK, 1967.
[9] I. Eames, V. Roig, J.C.R. Hunt, and S.E. Belcher, Vorticity annihilation and inviscid blocking in
multibody flows, in Y.A. Gayer and J.C.R Hunt, editors, NATO Science Series II: Mathematics,
Physics and Chemistry, Vol 236: Flow and Transport Processes with Complex Obstructions,
Springer, Netherlands, 2007, pp. 251–270.
[10] J.C.R. Hunt and I. Eames, The disappearance of laminar and turbulent wakes in complex flows,
J. Fluid Mech. 457 (2002), pp. 111–132.
[11] A. Nicolle and I. Eames, Numerical study of flow through and around a circular array of
cylinders, J. Fluid Mech. 679 (2011), pp. 1–31.
[12] A.A. Townsend, Measurements in the turbulent wake of a cylinder, Proc. R. Soc. Lond. 190
(1947), pp. 551–561.
[13] A.A. Townsend, Momentum and energy diffusion in the turbulent wake of a cylinder, Proc. R.
Soc. Lond. 197 (1949), pp. 124–140.
[14] A.A. Townsend, The fully developed wake of a circular cylinder, Austral. J. Sci. 2 (1949),
pp. 451–468.
[15] L. Prandtl, Recent results of turbulence research, NACA Tech. Memo No. 720, National Advi-
sory Committee for Aeronautics, Washington, DC 1933.
[16] S.B. Pope, Turbulent Flows, Cambridge University Press, Cambridge, UK, 2000.
[17] G. Comte-Bellot and S. Corrsin, Simple Eulerian time correlation of full- and narrow-
band velocity signals in grid-generated, ‘isotropic’ turbulence, J. Fluid Mech. 48 (1971),
pp. 273–337.
16 I. Eames et al.

[18] G.C. Larsen, H.A. Madsen, K. Thomsen, and T.J. Larsen, Wake meandering: A pragmatic
approach, Wind Energy 11 (2008), pp. 377–395.
[19] G.I. Taylor, Diffusion by continuous movements, Proc. L. Math. Soc. 20 (1921), pp. 196–212.
[20] G.I. Taylor, The spectrum of turbulence, Proc. R. Soc. Lond. 164 (1938), pp. 476–490.
[21] D.G. Goring and V.I. Nikora, Despiking acoustic doppler velocimeter data, J. Hydraulic Eng.
128 (2002), pp. 117–126.
[22] T.L. Wahl, Discussion of ‘Despiking Acoustic Doppler Velocimeter Data’ by Derek G. Goring
and Vladimir I. Nikora, J. Hydraulic Eng. 129 (2003), pp. 484–487.
[23] M. Parsheh, F. Sotiropoulos, and F. Porté-Agel, Estimation of power spectra of acoustic-
doppler velocimetry data contaminated with intermittent spikes, J. Hydraulic Eng. 136 (2010),
pp. 368–378.
[24] C.M. Garcı́a, M.I. Cantero, Y. Niño, and M.H. Garcı́a, Turbulence measurements with acoustic
doppler velocimeters, J. Hydraul. Eng. 131 (2005), pp. 1062–1073.

Você também pode gostar