Pergamon

PII: SOO38-092X(96) 00019-9

Solar Energy Vol. 56, No. 6, pp. 621-633, 1996

Copyright 0 1996 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
003%092X/96 $15.00 + 0.00

A STUDY OF THE NEAR WAKE STRUCTURE OF A WIND TURBINE

COMPARING MEASUREMENTS FROM LABORATORY
AND FULL-SCALE EXPERIMENTS
J. WHALE,* K. H. PAPADOPOULOS,** C. G. ANDERSON,***
and D. J. SKYNER*

C. G. HELMIS**

*University of Edinburgh, Department

of Mechanical Engineering, Kings Buildings, Mayfield Rd.
Edinburgh EH9 352, U.K., **University of Athens, Department
of Applied Physics, Laboratory
of
Meteorology,
Ippokratous
33,10680 Athens, Greece and ***Aerpac Special Products B.V., P.O. Box 167.
7600AD Almelo, The Netherlands
(Communicated

by DAVID MILBORROW)

Abstract-Wake
flow measurements
have been performed using the technique of particle image velocimetry
(PIV) at stations downstream
from a model wind turbine rotor, and evaluated against experimental data
from two full-scale machines. Comparisons
include both mean velocity and turbulent intensity cross-wake
profiles at a range of tip speed ratios. The application
of PIV to the study of wind turbine wakes is
described in detail, including the steps required to ensure appropriate
and accurate simulation of the flow
field conditions. The results suggest that the PIV method is a potentially useful tool in the investigation
of detailed wake flow, though significant differences are observed between wake velocity deficits at fulland model scale. These are discussed with regard to scale effect, the influence of terrain, model similarity,
and the phenomenon
of wake meandering and effective cross-wake smoothing. Copyright 0 1996 Elsevier
Science Ltd.

1. INTRODUCTION

group at the University of Edinburgh, Scotland,

using the relatively recent technique of particle
image velocimetry (PIV).
Field studies of wakes behind single turbines
or of multiple wakes in wind farms (H6gstrBm
et al., 1988; Taylor et al., 1988; Larsen & Velk,
1989; Nierenberg 1989; Elliot and Barnard,
1990) usually concentrate on the decay rate of
the velocity deficits in the far-wake region. This
is then related to power production optimisation
of wind farms. While confirming the qualitative
trends revealed by wind tunnel simulations, field
studies have indicated the necessity of further
measurements, especially over complex terrain
(Van der Snack, 1989), in order to improve the
accuracy of theoretical wake models.
Wind tunnel studies have also demonstrated
that the simulation of the near wake region,
which is usually described by a uniform velocity
profile in the so-called potential core, does not
represent the real-flow situation accurately
(Ainslie, 1987). This suggests that the effect of
the turbulence produced by the turbine is
improperly parameterised, and again the need
for more detailed measurements is indicated.
The use of PIV in wind turbine studies is a
relatively recent development. Infield et al.
(1994) have applied the technique both in the
wind tunnel, and to a full-scale wind turbine in
the field. Their studies concentrated on the
immediate vicinity of the blade, and produced

The operation of a wind turbine produces a

downstream region of reduced wind speed, the
so-called wake. The wake constitutes an important factor in determining the siting of turbines
in windfarms, for two principal reasons:
(1) mean wake characteristics, and their relation
to the incident wind field and the local topography, influence the total energy resource at a
potential windfarm site, and (2) the turbulent
structure of the wake affects the fatigue loading
of downstream turbine rotors, thus dictating the
minimum spacing of wind turbines within the
windfarm. The design of windfarms, therefore,
can benefit significantly from a detailed knowledge of these fundamental wake parameters.
The present article describes a recent investigation into the properties of the near wake
region of three-bladed wind turbines. Measurements were obtained from two full-scale
machines, and from a replica model in the laboratory, at approximately l/100 scale. The fullscale data were derived from new experiments
carried out on the Greek island of Samos, by a
research group at the University of Athens, and
from data previously recorded at the National
Test Station for Wind Turbines, at Risa,
Denmark. At full scale, wake data were obtained
from anemometry measurements. The smallscale experiments were conducted by a research
621

622

J. Whale et al.

detailed profiles of bound circulation

and the
tip vortex. Visualisation
of the full wake of the
rotor up to 4 rotor diameters (D) downstream
was achieved
at Edinburgh
(Whale
and
Anderson, 1993) using a small-scale model, with
water, rather than air, as the flow medium.
One major objective of the present work was
to assess the validity of small-scale PIV measurements as a technique for investigating
fullscale wind turbine
phenomena.
If successful,
there would be significant attractions
in using
PIV, because of its ability to map the velocity
in the entire rotor wake at a given instant. PIV
vector maps may be processed to yield both
bulk wake measurements,
such as velocity deficits, or data relating to the detailed structure of
the wake, e.g. vorticity contours. In the present
campaign,
results were restricted to mean and
turbulent
wake characteristics
only, measured
at distances of l.lD and 1.5D downstream
of
the model turbine, i.e. in the near wake, and
subsequently
compared with full-scale data.
It is worth noting that, despite their importance for understanding
the basis of wind turbine
rotor performance, measurements
at distances as
close as 1D downstream are rare. The few examples include the comprehensive wind tunnel study
by Papaconstantinou
and Bergeles (1988), the
extended
Nibe project (Taylor,
1990) which
included some experimental
results at 1D and
the study of the near wake structure of a Darrieus
turbine by Strickland and Goldman (1981).
2. DESCRIPTION

OF EXPERIMENTS

2.1. Full-scale measurements

2.1.1. Experimental
method. At full scale,
windspeed
measurements
were made simultaneously upwind and downwind
of the wind
turbines
to determine
velocity profiles in the
wake. The method has been widely used in the
wind turbine
field. Windspeeds
are recorded
using mast-mounted
anemometers,
one ahead
of, and one behind, the rotor. Mean values of
upstream windspeed are assumed to represent
the freestream velocity. Wake velocity profiles
are obtained by recording data over a range of
incident wind directions,
such that the downstream anemometer
is immersed
in different
parts of the wake. The wake velocity ratio for
a given wind direction is normalised as the ratio
of downstream
to upstream windspeed, suitably
time-averaged.
Wake velocity ratios are then
plotted against upstream wind direction. On the
assumption
that the wind turbine yaw system

tracks the wind direction accurately

over the
averaging period, the data can be re-interpreted
as velocity ratio profiles obtained
by a horizontal traverse behind, and parallel to, the rotor.
Note that the wake profiles thus obtained are
inherently
averaged with respect to short-term
variations of upstream wind direction. Although
no information
on the yaw control of the experimental machines was available, it is assumed
that on average the rotors were correctly aligned
with the upstream
wind direction throughout
the measurements.
It is to be expected, however,
that rotor alignment
lags behind changes in
upstream wind direction, leading to a degree of
cross-wake smoothing of the velocity profiles.
It should also be noted that by normalising
the wake velocity with respect to the upstream
anemometer
readings,
it is implied that the
upstream
values of wind speed and ambient
turbulence
are considered representative
of the
actual flow which intersects
the rotor. This
assumption
has previously been justified on the
basis of experimental
analysis (Helmis et al.,
1995).
2.1.2. Samos Island 19 m wind turbine. Samos
Island lies in the eastern region of the Aegean
Sea, and hosts a wind farm located 390 m above
mean sea level, on a saddle confined by the
islands two major mountain
ranges. The wind
farm comprises nine three-bladed,
horizontalaxis, Vestas (formerly Windmatic) WM19S wind
turbines,
with rotor diameter
of 19 m, hubheight of 25 m, and output rating of 100 kW.
The WM19S is stall regulated, achieving rated
power at a windspeed of 13 ms- . Cut-in and
cut-out wind speeds are 3 ms- and 27 ms-
respectively.
The maximum
power coefficient
C pmax is 0.38, attained in the windspeed range
8-10 ms-,
at a constant
rotational
speed of
48 r.p.m.
Measurements
were made on a single wind
turbine using two met masts, one located 0.80
(where D is rotor diameter) upwind, the other
l.lD downwind, of the machine. The measurements used for comparison with the laboratoryscale PIV data were taken from cup anemometers mounted at 12 m and 29 m above ground
level, on the upwind
and downwind
masts,
respectively; the anemometer
sampling rate was
1Hz. The experimental
layout is described fully
by Helmis et al. (1995), and the measurements
were made over the period 16-24 August 1991.
At the given elevation (29 m), the downstream
anemometer
was above the centreline
of the
rotor, clear from the influence of tower shadow.

A study of the near wake structure

This also meant, however, that the wake profiles

measured by it were not on the rotor centreline,
but rather at an offset of 0.42R, where R is the
rotor radius. This was taken into account when
replica tests were carried out at small scale using
PIV (see Section 2.2.2.). Another factor to be
accounted for in the comparisons was the complexity of the terrain on Samos, and an attempt
was made to compensate the results for a nonuniform inflow profile, based on the use of nonwake data, measured with the turbine stopped
(see Section 2.1.3.).
During the experiments, the mean windspeeds
at the Samos site ranged from 9 to 27 ms-.
The influence of windspeed on the wake features
has been discussed previously (Helmis et al.,
1995; Papadopoulos et al., 1995). The mean
upstream turbulence intensity was 6%, while the
atmospheric stability was estimated by nearsurface air temperature profiles to be effectively neutral.
2.1.3. Rise 20 m wind turbine. The second
wind turbine for which wake profile data were
obtained was a Vestas V20/100, three-blade,
stall regulated machine, located at the Rise
National Laboratory in Denmark. The V20 has
a 20 m diameter rotor, rotor speed of 45.5 r.p.m.,
and rated output of 100 kW.
The measurements were made on the V20
prototype machine in 1988, and have been fully
documented previously (Paulsen, 1989). The
upwind and downwind met masts were situated
at 0.680 and 1.5D, respectively. The anemometers were located at hub height (24.25 m) in
both cases; as a result of this, wake velocity
profiles traversed the rotor centreline. Windspeed data were sampled at 2Hz, with run
statistics based on 10 min averaging. The experimental data used were measured in the mean
windspeed range 8-9 ms-; the turbulence
intensity was in the range 5-10%.
The use of the V20 for comparison with the
WM19S was based on the similarity of the two
machines in terms of their configuration, i.e.
both three-bladed and stall regulated, and size
(100 kW rating in both cases, and rotor diameters of 20 m and 19 m, respectively). It should
be noted, however, that the V20 has a somewhat
lower solidity than the WM19S, at 5.5% rather
than 9%. The possible influence of this on the
experimental results is discussed in Section 3.
2.1.4. Full-scale wake projiles
(a) WM19S, Samos Island.

As noted before an attempt was made to

correct for the influence of complex terrain on

623

of a wind turbine

the Samos Island wake data, and in particular

the existence of a non-uniform upstream profile.
To do this, measurements were taken with the
turbine in operation (the wake data set) and
stationary (the non-wake data set). A preliminary analysis of the non-wake data set was then
used to establish the background correction to
be applied to operational data. The importance
of this procedure is seen from previous results
described by Helmis et al. (1995) who highlight
the uncertainty introduced by estimating wake
velocity deficits by comparing upstream and
downstream measurements using data recorded
only with the turbine in operation.
The non-wake velocity ratio was found to
depend significantly on wind direction (Fig. 1 ),
and less strongly on windspeed. Data for low
and high wind speeds were therefore used to
yield two correction curves, which gave the nonwake velocity ratio as a function of wind direction for each windspeed range. This was then
used to provide correction factors for the data
obtained during operation of the turbine: a
given velocity ratio obtained with the turbine
running was divided by the non-wake ratio
corresponding to the same upstream wind direction. In this way, the effects of topography and
wind shear were compensated for.
The corrected wake data, i.e. with the turbine
operational, are shown for a range of wind
speeds in Fig. 2. Although full wake profiles are
not available for all wind speed ranges, the
results show a clear dependence on windspeed,
with the wake deficit (defined as one minus the
velocity ratio) increasing as a function of tip

9
al 0.7
E
0.6

j
310

Upstream

speed

??
Dooo

+++++
ooooo
I

320

330

Wind

340

direction

350

range

13-15
15-17
22-26
I

360

m/s
m/s
m/s
I

370

(deg.)

Fig. 1. Non-wake
velocity ratios at l.lD for the WM19S
wind turbine (rotor stationary),
as a function of upstream
wind direction and windspeed.

624

J. Whale et al.
1.1

1.1

?
tio.9

$1.0

.s
E 0.8
x
x
8 0.7

.;
E 0.9

-?

h
ti
5

0 1.0

3 0.6

is
c 0.5

0.6

E
:

c
g 0.7

0.3

0.4

..
I

245

265

310 ,

3io

3io

Wind

340

direction

350

360

3+0

(deg.)

Fig. 2. Wake velocity ratios at l.lD for the WM19S wind

turbine; downstream
velocity is measured at an offset position of 0.42R above the rotor shaft axis, and corrected for
non-uniform
inflow conditions
using factors derived from
Fig. 1.

speed ratio. It is assumed that the wake centreline corresponds

to the 350 wind direction, for
which the wind turbine is directly upwind of the
29 m measuring anemometer.
Centreline velocity ratios, at a radial distance of 0.42R above
the rotor shaft centreline,
may be derived
directly from the data in Fig. 2.
The given wake profiles are based on 1 min
averages. Longitudinal
and lateral coherence
considerations
suggest that the relatively short
averaging
time is appropriate
(Taylor, 1990).
Analysis of corresponding
15 min samples gives
almost identical results, though with a somewhat more spiky appearance:
this was attributed to changes in rotor orientation
during the
15 min period caused by operation
of the yaw
system. The graphs based on 1 min data are
nonetheless
fairly smooth. The statistical significance of the results may be assumed greatest
for the more extended
wake data sets. The
measured standard deviation of wind direction
was of the order 5-6, which translates into a
maximum cross-wake smoothing over f 5%D.
(b)

V20/100, Riss.

Unlike the Samos Island site, the Rise Test

Station in Denmark
is situated
on very flat
terrain. For this reason, no correction was made
to the measured velocity profile for the Vestas
V20/100, which is shown in Fig. 3. The minimum velocity ratio on the wake centreline
is
approximately
0.4. A point of note is that the
velocity ratio rises above unity at cross-wake

RiSB data (A-5.6)

285

305

325

Wind direction
0.6

..I

(deg.)

Fig. 3. Wake velocity profile measured for Vestas V20/100,

Riss Test Station. Downstream
distance is 1.5D, tip speed
ratio 1= 5.6.

distance y/R > 1, i.e. the boundary prescribed

the rotor radius. This indicates a speed-up
the Ilow at the edge of the wake.
2.2. Laboratory-scale

by
of

measurements

2.2.1. Experimental method. The experiments

at model scale were made using the technique
of particle image velocimetry
(PIV). PIV is a
non-intrusive
velocity measurement
technique
which allows two-dimensional
flow fields to be
captured at a single instant. The basis of PIV is
to illuminate
a two-dimensional
plane of flow
containing
small, neutrally
buoyant,
seeding
particles,
using a stroboscopically
repeating
light source. A double (or multiple) exposure
photograph
of this plane is taken, whereby the
spacing between the images of each particle on
the film gives the local flow velocity. The photograph is then analysed to determine
the flow
velocities across the entire field.
The film is interpreted
point by point over a
dense grid using a combination
of optical and
digital analysis: this involves scanning successive
small regions of the negative with a probe laser
to produce
an interference
pattern
from the
multiple particle images in that area. The interference fringes are measured
and recorded in
digital form and the data Fourier transformed
to yield the particle velocities at that point. The
whole negative is scanned in this way to build
up a flow velocity map, which forms the basis
of all subsequent
analysis.
The technique of PIV was introduced
to the
field of wind turbine aerodynamics
by Infield
et al. (1994) who conducted
tests on a 0.9 m
diameter wind turbine in a wind tunnel, using

A study of the near wake structure

TT

.A----- Tffles,

water

of a wind turbine

625

FI

surface
streamlined
tower

I
I

750 mm

measurement

zone

l-

Fig. 4. Schematic

diagram

of PIV set-up in the laboratory

pulsed lasers. The tests established the applicability and usefulness of PIV as a velocimetry
tool for wind turbines.
The same researchers
subsequently
made PIV measurements
on a fullscale wind turbine of 17 m diameter. These tests
were mainly concerned with visualising the flow
around a localised region of the blade, however.
Visualisation
of the entire wake at full scale
using pulsed lasers presents obvious difficulties.
The PIV equipment at Edinburgh
University,
as used in the present study, is applicable
to
scale model rotors only, but is capable of capturing images from the near wake up to 40 downstream. The experiments
were carried out in a
two-dimensional
water flume, 10 m long by
400 mm wide (Fig. 4). The use of water, rather
than air, as the flow medium greatly facilitates
seeding and illumination.
The flume has glass
walls and base, and is filled with water to a
depth of 750 mm. A steady current
can be
established in the tank, driven by a water pump,
and recirculated
via an external pipe system.
A continuous
wave (CW) laser was used in
conjunction
with a scanning
beam system of
illumination
to produce the laser sheet (Gray
et al., 1991). The laser sheet was directed
through the base of the flume, illuminating
a
two dimensional
cross-section
of the flow. The
model turbine rig was placed in the tank, with
the rotor aligned normal to the upstream flow.
The water was seeded with conifer pollen of
average diameter
70 pm; concentrations
were
maintained
at a level that ensured a high density
of non-overlapping
particles on the resulting
film record.
The model rotor (Fig. 5) was a l/lOOth scale
replica of the three-bladed
Vestas (Windmatic)
WM19S. The blades were manufactured
from
rigid plastic, using a numerically
controlled
cutter. Despite the small scale, the model blades
were accurately
profiled with a NACA-632XX

Fig. 5. Scale model of the WM19S rotor used in the laboratory test.

section, with twist, chord and thickness distributions based on the manufacturers original drawings. The turbine model was driven in the tank
by an electric motor, located on a frame above
the water level, and connected to the rotor shaft
by a toothed belt running inside a hollow tube,
effectively an inverted
tower. In order to
reduce the disturbance
to the rotor wake caused
by the tower, it was streamlined
with a foam
plastic
shroud
of symmetric
aerofoil
crosssection.
The image recording equipment
consisted of
and
a
a rotating-mirror
shifting
system,
Hasselblad
large format camera. The purpose
of the shifting system was to superpose a known

626

J. Whale et al.

velocity component
onto the recorded image in
order to (a) eliminate directional
ambiguity
in
the final vector maps, and (b) increase
the
dynamic
range of the system. The shifting
sequence was synchronised
with an index pulse
from a position encoder, connected to the turbine drive motor.
For reasons of geometry,
the image shiftvelocity was non-uniform
over the flow field. A
correction
therefore had to be applied to the
processed results, based on a calibration
of the
shifting system. In order to separate the effects
of image-shifting
distortion
from the flow
recording of the wake, a number of PIV photographs were taken of still water in the tank. The
still-water
records
were then averaged
and
subtracted
pointwise from the vector fields of
the turbine wakes in order to correct for shift
velocities.
2.2.2. Replicating
full-scale
conditions. The
experiments were aimed at reproducing
the fullscale measurements
taken on Samos with the
Vestas WM19S. Thus, the model was positioned
with the rotor centreline at a distance of 0.42R
from the laser sheet. Assuming an axially symmetric wake, data were thus recorded at an
equivalent offset position from the centre of the
wake to the readings taken at full scale (see
Section 2.1.2.).
In order to control the ambient turbulence
level, turbulence
manipulators
were placed
upstream
of the rotor. These consisted
of a
parallel system of baffles comprising of an aluminium honeycomb
section, a perforated plate
and a fine mesh (see Fig. 4). The honeycomb
acted as a flow straightener,
with the perforated
plate serving to impose a particular
upstream
profile. Final smoothing
was provided by the
fine mesh screen.
From the Samos experiments,
it was concluded that the wind speed and turbulence
intensity were fairly constant across the rotor
disk. The turbulence
manipulators
were therefore chosen to produce a uniform
upstream
profile with low turbulence.
A perforated plate
with 32 mm diameter holes and regular pitch of
38 mm was placed 500 mm downstream
of the
honeycomb
section. A fine mesh screen of 18
lines/inch was placed a further 800 mm downstream, and 1200 mm upstream of the rotor.
As noted above, an important
feature of this
work was to investigate whether tests at model
scale could yield valid data regarding the performance of full-scale wind turbines. It was therefore decided to replicate as accurately as possible

the conditions pertaining to the full-scale measurements on Samos, with the obvious exception
of scale. In this way, any discrepancies
between
the full-scale measurements
and those obtained
from the PIV tests could be attributed either to
scale effect, i.e. Reynolds
number,
or tunnel
blockage, rather than an improperly characterised experiment.
Geometric similarity between the model and
the full-scale machine was attained by the use of
an accurate replica. Kinematic
similarity was
achieved by running the model at an appropriate
range of tip speed ratios A, using the motor speed
controller. The water current velocity was maintained constant throughout,
with its value accurately determined from the PIV analysis.
A further consideration
was that, while the
full-scale results were based on time-averaged
recordings, the PIV vector maps corresponded
to instantaneous
wake images. It was therefore necessary to introduce
equivalent
timeaveraging in the latter case. This was done by
repeating each PIV test a number of times, with
the camera exposure synchronised
to a different
rotor position in each case, and taking a numerical average of the resulting vector maps. Ideally,
i.e. to achieve stationarity
of the averaged map,
this procedure would have been repeated a very
large number of times, with the rotor photographed at positions evenly distributed
around
the disk. In practice, the shifting sequence was
synchronised
to photograph
the blades in just
6 azimuthal
positions,
20 apart.
For a
three-bladed
rotor, this discretizes one whole
revolution.
Post-analysis
averaging of these six
exposures
yields
six vector
maps
which,
averaged together, provide the equivalent
of a
time-averaged
wake image.
2.2.3. Results and analysis. The PIV photographs thus obtained
were processed to yield
two-dimensional
velocity vector maps of the
type shown in Fig. 6, where each vector indicates
the velocity in the flow at that point. The figure
shows the wake behind the WM19S model
operating at a tip speed ratio of 4.8; the area of
reduced velocity behind the rotor is clearly seen
in this image. Velocity maps were obtained at
five tip speed ratios in the range 1.6-4.8.
The cross-wake profile at l.lD (offset position
0.42R) downstream
of the model rotor was
found by averaging
four columns
from the
vector map corresponding
to downstream
distances 1.0-1.20 from the rotor to account for
any uncertainty
in the downstream
position.
The results are shown in Fig. 7, as velocity ratio

A study of the near wake structure

of a wind turbine

621

Downstream distance, x(metres)

Fig. 6. PIV velocity vector map of wake behind WM19S model rotor at i. =4.8. The vertical cross-section
of the wake is evident as a downstream
region of reduced velocity, and lies 0.42R from the centreline to
correspond
to the full-scale experimental
case.

1.5

oooooooooooooooooo

1.4

000

~oo~noooooLm~~on~o

s1.3
\
3 1.2

00

+++++

00

0 OOOO
ooooooooooooooo

oooooo
oooo
oon~moon
++a
0

00 + +

?o0
?
0

++++++

+++++++
A~A~~AA~~AAAAAAA+++na
gn+++AAAA
+++++++++++

++
AA

6 1.1
.rl
$ 1.0
k
&_0.9

+
xx xxxxx~x~x

xXxX

A A A,A

AAAAAAAAAA
A

'A

000

A
+

xX XxXxX

Ax

xxx
7.1

xxxx

+++

'A

x
A

xA
x

.%
LS0.8
$

A
x
A

0.7

aA
A

0.6

'

x
x

f 0.5

ooooo),11.6

?~oo~h~2.7
?

x
x

2 0.4

xx

+++++

~~3.2

AAAAAh=4.2

0.3
0.2

x xxx x hz4.8
-4

I
-3

I
-2
crosslLe

I
OdistAce,

I
3

I
4

y2/R

Fig. 7. Wake velocity ratio profiles at l.lD from PIV experiments. The curves for L=4.2, 3.2, 2.7 and 1.6
have been shifted upwards by 0.1, 0.2, 0.3 and 0.4, respectively, for clarity.

628

J. Whale et al.

plotted against cross-wake distance. A single

averaged value of the upstream velocity was
assumed in calculating the velocity ratio for
each location: this was evaluated from the mean
flow statistics of the first column of vectors at
0.60 upstream of the rotor, with corrections for
the distortion introduced by image-shifting
(see above).
The influence of tip speed ratio can be seen
clearly in Fig. 7. The wake velocity ratio
decreases with increasing 2, with correspondingly greater centreline deficits. The asymmetry
seen in the profiles at their outer edges (where
conditions approach those in the freestream) is
attributed to the presence of the model support
structure, i.e. the tower of the wind turbine.
Despite its steamlined shroud, this evidently still
introduced some non-uniformity
behind the
rotor disc.

Cross-wake
-1.0
1.2
;

0.7

I (a)
320

-0.5

,
330

distance,

I
340

Wind

OeooO

PIV

Full

direction

-1.0
;

-0.5

1.0

data

(h=2.7)

scale

I
360

I
350

Cross-wake
1.2

y/R

0.5

0.0

(h=3.0)

I
370

(deg.)
distance,

y/R

0.5

0.0

3. COMPARISON OF MODEL AND FULLSCALE RESULTS

I
380

oe-

PIV

Full-scale

1.0

data

(h=3.2)
(h=3.3)

3.1. Mean wake properties

(a) WM19S (Samos Island) data. Figure 8
contains a comparison of the full-scale and
model wake data for the WM19S wind turbine,
in the form of wake profiles at two tip speed
ratios selected from Figs 2 and 7. The choice
was dictated by the availability of a reasonable
amount of cross-wake data from the full-scale
data set, and the similarity of the tip speed
ratios in the two cases. The wake profiles suggest
that the shapes of the model and full-scale wake
are somewhat different. The full-scale wake is
wide and has a homogeneous central portion
whereas, in general, the PIV results produced
narrower profiles with deeper troughs.
The most likely reasons for the difference are
( 1) scale effect, and (2) meandering of the fullscale wake. The latter is caused by variation of
wind speed and direction during the averaging
period, causing smoothing of the experimental
profiles (Helmis et al., 1995). The shape of the
full-scale wake profiles suggests significant
cross-wake mixing.
The effect of wake meandering is less likely
to be important in the near (as opposed to far)
wake and the relatively short averaging period
(1 min) should have ensured that gross changes
in wake direction were avoided. Nonetheless,
the minima of the full-scale curves very often
do not coincide with the machine alignment of
350 and the influence of wake meandering on
the experimental profiles cannot be discounted.

,
9
0

\
\
B.m,
0.7l0,
320

330

340

Wind

Lf

/
350

direction

360

370

380

(deg.)

Fig. 8. Comparison
of velocity ratio profiles at l.lD for the
WM19S rotor, comparing
PIV and full-scale (FS) data: (a);
R,s=3.0, &,,=2.7,
(b); I,s=3.3,
l,,v=3.2.

Comparison of Figs 2 and 7 shows that at I=

4.4 the full-scale data contains a minimum velocity ratio of about 0.68 at a cross-wake position
of 340; this is much closer to the model result.
The complex terrain may also be a factor in
explaining the shape of the profiles at full-scale.
Despite attempts at similarity, the scales of
turbulence in the atmosphere may have been
different from those in the water tank and have
varied according to stability. Large-scale
inhomogeneities of the terrain impose energetic
turbulent motions with characteristic scales of
the size of the wake (and even larger), leading
to smearing of velocity gradients of the kind

A study of the near wake structure

found in the centre of the wake (see model wake

profiles).
From the given wake profiles, centreline
velocity ratios at an offset position
of 0.42R
may be obtained.
As noted above, the wake
ratios measured at full scale (Fig. 2) incorporate
directional
smoothing,
because of the variation
in upstream wind direction during the averaging
period. To account for this at model scale, the
centreline
velocity at l.lD downstream
was
averaged over a cross-wake distance based on
the variance
of the wind direction
in the
full-scale tests. In practice, this involved averaging together the velocity vectors either side of
the centreline i.e. the two adjacent (cross-wake)
values at l.lD downstream.
The resulting centreline velocity ratios from
the model tests are shown as a function of tip
speed ratio in Fig. 9, together with the corresponding data from the Samos Island measurements. The error bars attached to the modelscale results are based on the standard error of
the cross-wake average centreline velocity ratio
over four downstream
positions in the vicinity
of 1.1 D, as noted above. For low 1, the comparison is promising. However, the two curves deviate significantly
at high /1.
Flow blockage in the tank could be a significant factor in explaining
the divergence of the
two curves towards high i. Another possibility
is that the experimental
profiles are displaced
because of wake meandering
(see above), and
that their velocity minima do not correspond

of a wind turbine

629

to the assumed upstream wind direction of 350;

accordingly,
the true centreline velocity ratio
could be better represented
by using the minimum velocity ratio found at each A. This is
done in Fig. 9, where it is seen that the corrected curve for full-scale data lies closer to the
laboratory
results than the original.
The error bars in Fig. 9 also reflect the level
of turbulence existing at the location of interest
in each case. Note that the turbulence
appears
to increase towards both high and low tip speed
ratios, with a minimum existing in between; this
feature is further discussed in Section 3.2.
(b) V20/100 (Rise) data. Figure 10 shows a
comparison
of a PIV wake velocity profile at
2=4.8 with full-scale measurements
from the
Vestas V20/100 at Rise, at a mean tip speed
ratio of A= 5.6. Again, model- and full-scale
data have been chosen on the basis of reasonably similar tip speed ratios. Note that the Ris0
profile shown here differs from the original data
in Fig. 3, which have been processed to obtain
a velocity profile at an offset of 0.42R from the
rotor centreline.
This is necessary because, as
noted in Section 2.1.2, the Riss anemometers
were set up at hubheight, to measure centreline
wake profiles.
In processing
the Riser data, it has been
assumed that the centreline wake profile is axisymmetric, and that a profile at arbitrary offset
according
to
distance d may be interpolated
the relationship
JJ~=y2-d2
where y is the
Cross-wakedistance, y/R

1.0

1.1

-j

-2.1
I

-1.05
I

0.0
I

1.05
I

2.1
I
lr-

\ .

b
. . *. . 95% confidence
-PlV
data
x_XXXY Full-scale
data
A&AM
MInimum
ratios

o~21~

intervals

for

full

scale

(h&3

data
0.3

r---.
3.0

Tip

speed

-I

3.5

ratio

4.0

4.5

5.0

Fig. 9. Comparison
of centreline velocity ratios at l.lD for
full-scale and PIV data. Best fit lines are drawn through the
corresponding
points. The crosses are median values for the
full-scale data; the triangles are subjectively chosen minimum values of the same data.

245

265

265

305

325

Wind direction (deg.)

Fig. 10. Comparison
of PIV wake velocity profile at 1=4.8
with full-scale measurements
from the V20/100 at Riser, at
1=5.6. The Rise, data have been processed to represent a
profile at centreline
offset 0.42R (axisymmetric
wake
assumption).

630

J. Whale et al.

translated value of cross-wake distance y. The

corresponding wind direction ((3)at 1.5D downstream was computed using the relation y=
3R tan 0. Whatever the merits of this assumption regarding the resulting profile shape, it
seems a legitimate method of extracting at least
the wake minimum velocity ratio at the required
offset position, This is found to have a value of
0.44, compared with 0.39 for the small-scale PIV
measurement.
These velocity ratios are relatively close,
though again the model-scale result corresponds to a greater wake deficit than that seen
at full scale. It is not possible to draw any firm
conclusions from this, i.e. with regard to the
influence of scale, as the PIV model was not a
replica of the V20 wind turbine, and the
solidity of the two rotors was somewhat
different (see Section 2.1.3). Nonetheless, considering the large discrepancies in results at
high I for the WM19S comparisons, it is a
promising result.
3.2. Turbulent properties of the wake
Figure 11(a) presents laboratory results for
the turbulence intensity ratios which involve
four column averages as before. The ratio is
defined as the ratio of the four-column average
I, to the turbulence intensity value 0.60
upstream of the rotor. A significant increase in
turbulence intensity is seen as tip speed ratio
increases, with maximum turbulence in the
region of the wake centreline. There is also some
evidence that wake turbulence does not increase
indefinitely with increasing ;1, but goes through
a minimum at some intermediate value.
An explanation for this may be that at low
tip speed ratio the rotor is heavily stalled, and
the turbulence is caused by the separated flow
behind the individual blades of the model rotor;
this may be referred to as local turbulence. At
high tip speed ratio the blades are likely to be
largely unstalled, with smooth (unseparated)
flow over their surfaces; however, by now the
wake itself is highly turbulent on a large scale,
because of the strong vorticity being transmitted
into it from the rotor. At some intermediate
value of tip speed ratio, the blades may be
operating out of stall, but with a relatively weak
vortex pattern in the wake. The turbulence
peaks seen at low A outside the rotor circumference (specifically y/R > 1) may be caused by the
wake of the supporting tower.
The full-scale data, shown in Fig. 11(b), also
reveal an increase of wake turbulence towards

high 2 values, with a minimum at an intermediate value of A= 3.3. This effect of somewhat
lower turbulence levels at intermediate 2, discussed above, is seen clearly in Fig. 12.
Moreover, the flat profiles of Fig. 11(b) probably reflect the wake meandering and cross-wake
smoothing of full-scale data. The local maxima
of wake turbulence at approximately Iv/RI= 0.6
(Fig. 11(b)) are related to tip-vorticity induced
turbulence.
The spectral analysis of the full-scale data has
already revealed fundamental variations of the
turbulent structure of the near wake as the tip
speed ratio varies (Papadopoulos et al., 1995);
not only is the wake intensity small at low il
values, but furthermore, the expected turbulent
characteristics of the wake are absent. The PIV
results demonstrate a large growth in turbulent
energy in the centre of the wake with increasing
2. This appears to be caused by the strong tip
vortex structure.
A comparison of individual cross-wake profiles of turbulent intensity ratios (Fig. 12) reveals
a proximity of laboratory and full-scale values
in the central part of the wake. However, the
shape of the full-scale profile again implies the
effect of wake meandering. Figure 13 presents
the comparison for the centreline wake turbulence as a function of the tip speed ratio. Using
spline interpolation, a smoothed curve is plotted
through the PIV data. The comparison is good
for A14 (Fig. 13(a)). The discrepancy for the
highest A value could be related to the limited
amount of full-scale data. Figure 13(a) was
reconstructed to include points of maximum
ratios for both full- and laboratory-scale data.
The full-scale data is seen to lie more closely to
the PIV curve (Fig. 13(b)), supporting the possibility that measurements from the wind park
are displaced because of wake meandering.
3.3. Region of acceleratedjow
A further observation, common to both the
full-scale and laboratory results, is that a region
of accelerated flow exists outside the wake
boundary. Referring again to Fig. 7, the velocity
ratio clearly rises above unity for Iv/RI> 1, with
the effect becoming more pronounced as L
increases. At an experimental scale this tendency
may be exaggerated by a blockage in the tank,
but it is clear from Fig. 2 that it also occurs at
full scale. Similar findings have also been
reported by Taylor (1990).
A simple explanation for the region of
accelerated flow is that the rotor partially

A study of the near wake structure

of a wind turbine

631

w
h=1.6
m
h=2.7
-ttcH x=3.2
A-A-+++h=4.2
M
h=4.8

8.0

"07.0

k-

d 6.0
3
21
5.0
h
c,
.I-#
; 4.0
aJ
2 3.0
%2.0
2
k 1.0
fz
0.0
-4

-3I

-2I

-1I

0I

cross-wake
3.0

dista:ce,

yZ/H

w
h=3.0
w
h=.3.;5
t-+--cf-+ h=4.0

o.03*F-T---1
Wind

3I

I
4

(b)

340

360

direction

3;;

380

(deg.)

Fig. 11. (a) Wake turbulent intensity ratio profiles at l.lD from PIV experiments for five i, values. The
curves for 1= 2.7, 3.2, 4.2 and 4.8 have been shifted upwards by 1, 2, 3 and 4 respectively to separate them
from each other. (b) Wake turbulent intensity ratio profiles at l.lD from full-scale data for three A values.

obstructs the airflow, as a solid object would.

In order to conserve mass flow (at constant
pressure), the air must speed up around the
obstacle. Alternatively, as the air in the wake
slows down and expands, so the air outside it
must speed up to flow through a more confined
space. The effect is consistent with the idea of a
helical vortex structure in the wake, which
retards the air inside it, but accelerates the air
outside, with respect to freestream.

4. CONCLUSIONS
A comparison has been presented of PIV
wake measurements from a three-bladed model
wind turbine with data captured in the wake of
two full-scale machines. The shape of the PIV
velocity profiles differed in significant respects
from the measurements obtained from the
WM19S machine on Samos Island. In general,
the PIV data yielded narrow, deep velocity

J. Whale et al.

632
Cross-wake

distance,

4.0

y/R

OoooO

P
lo****

PIV
Full

data

b3.5

(h=3.2)

scale

PIV data

x x x x x Full-scale

data

(h=3.3)

I 1.6

0)

2.0 -

2 1.5 4;
U 1.0 d
.*
z
s 0.5 G

(a)

o.03803800
Wind

direction

0.0

(deg.)

1.5

2.0

2.5

3.0

Tip speed

Fig. 12. Comparison

of wake velocity ratio profiles at l.lD
between full-scale (FS) and PIV data for 1,s = 3.3, A,,, = 3.2.

3.5

ratio,

I
5.0

I
4.5

I
5.0

4.0

4.5

4.0
Om
xxxxx

5
profiles
whereas
measurements
from
the
WM19S produced wider profiles with homogeneous central portions.
Large discrepancies
in
centreline velocity deficit between the PIV data
and the WM19S results occurred at high tip
speed ratio. Closer agreement in velocity deficit
and profile shape was found when the PIV
results were compared with measurements
made
on the flat terrain of the Vestas V20/100 site
at Ris0.
A number of factors have been discussed to
account for the discrepancies at high A, of which
the difference in scale is the most obvious candidate. The Reynolds number of the PIV tests is
lower than full-scale by a factor of 1000; as a
consequence,
the boundary
layer flow on the
model blades will differ from that at full-scale,
particularly
regarding
the stall angle and the
transition to turbulence (Galbraith
et al., 1987).
The influence of the Reynolds number on the
near wake properties
is not well understood,
however, and their sensitivity to scale may well
be less significant than blade flow. Certainly the
bulk properties of the wake further downstream,
which is fully turbulent,
are less sensitive to
Reynolds number.
The discrepancy between model and full-scale
results may also be attributable
to (a) blockage
in the water tank, or (b) uncertainties
pertaining
to the full-scale experiment.
In the latter case,
for instance, the complex terrain of Samos Island
may produce large-scale inhomogeneities
that
affect the wake properties. It is possible that a
highly turbulent
wake (such as the turbulent

1.0

_3.5
.$
,m3.0
h

PIV data
Full-scale

data

.;

; 2.5 z
.d
; 2.0 -

Pj 1.5 4
_d
;

/p

OX

g 1.0
6
9 0.5 r
0.0

@I
1.0

I
1.5

I
2.0

I
2.5

1
3.0

Tip speed

I
3.5

ratio,

I
4.0

Fig. 13. Centreline turbulence

intensity ratios at l.lD for
full-scale and PIV data, WM19S wind turbine; (a) using
spline interpolation,
a smoothed curve is plotted through
the PIV data; in (b) the maximum values of the ratios have
been plotted for both the PIV and full-scale data.

wake state with large areas of recirculating

flow)
may remain stable in the laboratory
environment but could not exist in the field. Full-scale
wake profiles have been shown to be displaced
from the PIV profiles and wake meandering has
been suggested as a reason for the offset.
Estimating
the degree to which each of the
factors contributes
to the observed discrepancy
is not trivial. In the case of wake meandering,
some insight is gained by re-plotting the profiles
using minimum
and maximum
ratios for the
full-scale
data. The influence
of Reynolds

A study of the near wake structure

number
and tank blockage
require
further
investigation.
Although
it would be unwise to draw too
many firm conclusions
at this stage, bearing in
mind the difference in scale of the two experiments and the limitations
of the method in each
case, the further use of PIV in this field seems
to be clearly indicated. It is inferred that a more
careful assessment of the effect of turbulence, in
terms of its spectral content
rather than its
integral
levels, on the wake properties
is
necessary.
A number
of investigations
suggest themselves. In particular,
the study of the change in
wake properties in the transition
from low to
high windspeed (i.e. high to low tip speed ratio)
is of interest. This work is planned, and will be
based on further analysis
of the turbulence
content
of the measured
wakes. Analysis
of
wake vorticity, readily available from the PIV
vector maps, is being undertaken.
In this case
it may be possible to investigate the properties
of the wake under conditions
where simple
analytic models for wind turbine rotors, e.g.
actuator-disc
theory, fail, for example where the
rotor is heavily stalled, or where the thrust
coefficient exceeds unity.
In these respects, it is hoped that further
comparisons
of PIV measurements
and fullscale data will be forthcoming
shortly.

NOMENCLATURE
D
R=D/2
YIR
d
u
UO
0
1

turbine rotor diameter, m

turbine rotor radius, m
non-dimensional
cross-wake distance
offset distance from wake centreline, m
air flow velocity in the wake, m s-
upstream wind speed, s s-i
wind direction
tip speed ratio
cv standard deviation of wind speed, m s-
1, turbulent intensity (uu/U)
turbulent intensity of the upstream flow
IO
C pmax maximum value of turbine power coefficient

Acknowledgements-The
authors would like to extend their
thanks to the following people: Jean-Baptiste
Richon and
Iain Morrison of Edinburgh University Physics Department,
for design of the image shifting system and assessment of
image shifting errors, respectively, and John Korsgaard
of
LM Glasfiber A/S, and Tom Pedersen of Vestas A/S, for
supplying
details of the WM19S wind turbine and rotor
blades.
The University of Athens research group would also like
to thank the Greek Ministry
of Industry,
Energy and

of a wind turbine

633

Technology
for financial support of the Samos experimental campaign.
Finally, both the teams at Athens and Edinburgh
would
like to express their thanks to the British Council, for
funding their ongoing collaboration.

REFERENCES
Ainslie J. F., (1987) Calculating
the flow field in the wake
of wind-turbines,
Proc. Int. Conf. on Wind Farms, Leeuwarden, The Netherlands.
Elliott D. L. and Barnard J. C. (1990) Observations
of wind
turbine wakes and surface roughness effects on wind flow
variability. Solar Energy, 45, 265283.
Galbraith
R.. A. McD, Coton F. N., Saliveros E. and Kokkodis G. (1987) Aerofoil scale effects and the relevance
to wind turbines. Proc. British Wind Energy Conf, Edinburgh, UK., pp. 1433153.
Gray C., Greated C. A., McCluskey D. R. and Easson W. J.
(1991) An analysis of the scanning beam PIV illumination system. J. Phys.: Meas. and Sci. Tech., 2, 717-724.
Helmis C. G., Papadopoulos
K. H., Asimakopoulos
D. N.,
Papageorgas
P. G. and Soilemes A. T. (1995) An experimental study of the near wake structure of a wind turbine
operating over complex terrain. Solar Energy (in press).
Hogstrom
U., Asimakopoulos
D. N., Kambezidis
H. D.,
Helmis C. G. and Smedman A. (1988) A field study of
the wake behind a 2MW wind turbine. Atmos. Environ.,
22, 803-820.
Infield D. G., Grant I., Smith G. and Wang X. (1994) Development of particle image velocimetry
for rotor flow
measurements.
Proc. British Wind Energy Co$, Stirling,
U.K., pp. 73-78.
Larsen A. and Velk P. (1989) Wind turbine wake interference-a
validation study. Proc. European Wind Energy
Conf, Glasgow, U.K., pp. 482-487.
Nierenberg
R. (1989) Terrain effects, wake deficits and
macro&ale
wake effects. Proc. European Wind Energy
Con& Glasgow, U.K., pp. 978-982.
Papaconstantinou
A. and Bergeles G. (1988) Hot-wire measurements of the tlowfield- in the vicinity of a HAWG
rotor. J. Wind Eng. Ind. Aerodyn, 31, 133-146.
Papadopoulos
K. H., Helmis C. G., Soilemes A. T., Papageorgas P. G. and Asimakopoulos
D. N. (1995) Study
of the turbulent characteristics
of the near-wake field of
a medium-sized
wind turbine operating
in high wind
conditions. Solar Energy (in press).
Pauslen U. S. (1989) The impact of the induced velocity in
the near flow held of a HAWT, Risa National Laboratory, Report M-2835.
Strickland
J. H. and Goldman
A. L. (1981) Preliminary
work concerning the near wake structure of the Darrieus
turbine. J. Energy, 5, 94-98.
Taylor G. J., Hojstrup J. and Luken E. (1988) Full-scale
wake
data
from
the NIBE
wind-turbines.
Proc.
Euroforum- New Energies Congress, Germany,
Vol. 3,
pp. 7555758.
Taylor G. J. (1990) Wake measurements
on the NIBE windturbines in Denmark. Part 2: data collection and analysis, CEC Contract EN3W.0039.UK(Hl).
Van Der Snoek L. (1989) Wake effects in wind farms and
behind obstacles. Proc. European Wind Energy Co&
Glasgow, U.K., pp. 4944498.
Whale J. and Anderson C. G. (1993) An experimental investigation of wind turbine wakes using particle image velocimetry. Proc. European Community Wind Energy Conf..
Lubeck-Travemunde,
Germany, pp. 457-460.