Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Journal of Wind Engineering

and Industrial Aerodynamics 84 (2000) 1}21

An experimental and numerical study of the vortex structure


in the wake of a wind turbine
J. Whale!,*, C.G. Anderson", R. Bareiss#, S. Wagner#
! Department of Mechanical Engineering, The University of Edinburgh, The King+s Buildings,
Edinburgh, EH9 3JL, UK
" Aerpac UK Ltd, 13 Faraday Rd, Glenrothes, Fife KY6 2RU, UK
# Institut fu( r Aerodynamik und Gasdynamik, Universita( t Stuttgart Pfawenwaldring 21,
D-70550 Stuttgart, Germany
Received 11 August 1997

Abstract

An experimental investigation into the properties of the vortex wake behind a wind turbine
rotor has been carried out at model scale, using Particle Image Velocimetry (PIV). The
two-blade model was operated at tip speed ratios in the range j"3}8, and chord Reynolds
numbers Re"6400}16 000. The blades were untwisted, with #at-plate aerofoil pro"le.
Measurements of wake velocity and vorticity were obtained for a two-dimensional #ow "eld
representing an axial cross-section of the wake, extending 2.9 rotor diameters downstream of
the rotor. The vorticity maps were compared with calculations made using the Rotor Vortex
Lattice Method (ROVLM), an inviscid free-wake code recently developed at the University of
Stuttgart. The PIV and ROVLM data show qualitative agreement in terms of the shape of the
wake boundary, including downstream wake contraction, and quantitative agreement in terms
of the tip vortex pitch. It appears that the fundamental behaviour of the helical vortex wake
may be relatively insensitive to blade chord Reynolds number, so long as similarity of tip speed
ratio is observed. ( 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Wind turbine wakes; Particle Image Velocimetry; Vortex wake model; Tip vortex
behaviour

1. Introduction

Current industry-standard codes for calculating wind turbine rotor loads and
power output are based on blade-element momentum theory (BEMT). It is widely

* Corresponding author. Department of Aeronautical and Astronautical Engineering, University of Illinois


at Urbana, Champaign, 306 Talbot Lab., 104 S. Wright St. Urbana, IL 61801, USA. Tel.: #1 217 333
85804663; fax: #1 217 244 0720; e-mail: whale@uiuc.edu.

0167-6105/00/$ } see front matter ( 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 6 7 - 6 1 0 5 ( 9 8 ) 0 0 2 0 1 - 3
2 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Nomenclature

D diameter of the rotor (m)


R radius of the rotor, R"D/2 (m)
c blade chord (m)
X rotational speed of the rotor (rad/s)
x radial cross-wake distance, measured from the hub of the rotor (m)
z axial downstream distance, measured from the hub of the rotor (m)
d spacing of vector "eld grid
< freestream velocity (m/s)
=
< relative velocity at the blade (m/s)
r
< axial transportation velocity of the tip vortex (m/s)
z
1 vorticity (s~1)
l kinematic viscosity (m2/s)
j tip speed ratio, j"(XR)/<
=
Re chord Reynolds number, Re"(< c)/l
r
1 normalised vorticity, 1 "41d/<
/ / =
p tip vortex pitch, p"2p< /XR
z

acknowledged, however, that BEMT contains fundamentally invalid assumptions [1],


which are overcome in practice by empirical adjustments. There is a widespread desire
in the wind industry that a more theoretically correct approach should supplant BEMT
in the long term. To do so, it should embody a model of the rotor wake which is valid
in all operating conditions, as it is the structure of the wake, and in particular its
vorticity content, which ultimately dictates the #uid loading on the blades of the wind
turbine.
The present paper addresses the need for detailed measurements of the rotor wake, in
order to develop and validate new numerical prediction codes. Novel images have been
obtained at the University of Edinburgh, using particle image velocimetry (PIV) to
visualise the wake of a two-blade rotor at model scale, and thus obtain detailed records
of velocity and vorticity. The PIV results were compared with numerical simulations
from a sophisticated vortex-wake code, recently developed at the University of
Stuttgart.

2. Wind turbine wakes

2.1. Aerodynamics

The air #ow downwind of a wind turbine rotor su!ers a loss in momentum resulting
in a region of reduced mean velocity, i.e. the wake. The wake is highly structured,
owing its form to the interaction of the freestream with the vorticity sheets generated
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 3

at the blades and trailed downstream of the rotor. This trailing vorticity is synony-
mous with the `horseshoe vortexa system created by a conventional wing of "nite
span, due to the spanwise variation in bound circulation [2]. On a wind turbine rotor
the vortex sheets terminate at the root and tip of each blade, where bound circulation,
and hence lift, are forced to zero. The change in circulation is particularly great at the
blade tip, where a strong trailing vortex results.
The trailing vorticity creates a helical pattern, in which the vortex sheets form screw
surfaces. The diameter and pitch of the sheets do not remain constant downstream of
the rotor plane, however, due to spatial variation of the wake velocity. The pattern is
not trivially modelled, as the induced velocity (the vector component caused by the
action of the rotor) at any point in the wake is determined by the vorticity at all other
points according to the Biot}Savart law [3]. In addition, the vortex sheets tend to
`roll upa shortly downstream of the rotor, with the vorticity becoming concentrated
at the outer edge of the wake. The wake deforms into a system comprising an outer
region containing an intense tip vortex spiral, and an inner region containing a weak
di!used vortex sheet.
The in#ow velocities across the rotor disc, which dictate the #uid loading on the
rotor blades, are de"ned by the complete wake vortex system, and it is for this reason
that an accurate description of the latter is so eagerly sought.

2.2. Theoretical models

As noted above, BEMT forms the most common basis for wind turbine
performance prediction codes (e.g. [4,5]). BEMT models embody the simplifying
assumption that the wake may be represented as a set of radially independent
streamtubes. The theory neglects any interaction of the #ow in adjacent streamtubes,
or exchange of #ow between the bounded wake and the freestream. BEMT codes are
fast to run, enabling a wind turbine power curve to be evaluated in typically a few
minutes. As noted by several authors, however (e.g. [6,7]), the underlying theory is
fundamentally de"cient at both low and high tip speed ratios j, de"ned by

XR
j" , (1)
<
=
where ) is the rotor rotational speed, R the blade tip radius, and < the undisturbed
=
freestream windspeed.
Vortex wake models attempt a more realistic representation of the wake structure.
The most fundamentally correct are free wake models [3], in which steady-state
solutions for the wake geometry incorporate self-induced wake distortions. In a full
free wake analysis, every point in the #ow "eld is analysed by taking account of the
vorticity at all other points, without prior assumption about the "nal wake geometry.
As a consequence, such analyses are computationally expensive, and rarely used for
performance prediction.
A variety of techniques have been employed to formulate a less computationally
intensive free wake method. Most promising are the vortex lattice codes (e.g. [8])
4 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

which discretise the rotor blade and vortex sheet into a large number of surface
panels. Bareiss and Wagner [9] implement a lattice method which simulates the
e!ects of vortex shedding and roll-up in the wake, with the resulting code known as
the Rotor Vortex Lattice Method (ROVLM). This code, which has been developed at
the Institut fuK r Aerodynamik und Gasdynamik (IAG) at the UniversitaK t of Stuttgart,
predicts the wake geometry and strength of vorticity on the rotor blades and in the
wake.
The ROVLM code has been applied with some success to the calculation of rotor
loads and induced velocities in European Union R&D projects in recent years
[10,11]. For practical reasons, however, veri"cation of the code has been restricted to
the comparison of integrated quantities such as net rotor power output and blade root
loading. It has not yet been possible to determine whether the detailed wake geometry
modelled by ROVLM is representative of a full scale rotor. This problem is common
to all rotor prediction codes, due to the di$culties in measuring the wake of
a full-scale wind turbine } these include the unsteadiness of the #ow, the need to place
instrumentation over a wide area, and the problems of seeding or illuminating a cross
section of the wake. As a result, there have been only a few attempts at detailed
measurement of the coherent structure in the wake of a wind turbine. These are
described below.

2.3. Prior experimental studies

Qualitative descriptions of wind #ow patterns and tip vortex behaviour for full-
scale wind turbines have been gained from #ow-visualisation studies with the aid of
smoke grenades [12,13]. To obtain greater detail of the vortex structure, other
researchers have opted for small-scale wind-tunnel testing using the techniques of
Laser Doppler Anemometry (LDA) and hot-wire anemometry [14,15]. A compre-
hensive study of tip vortices and vortex sheets was performed by Green [16] using
LDA on a modi"ed aircraft propeller of 150 mm diameter.
The technique of PIV was "rst introduced to the "eld of wind turbine aerodynamics
by Smith et al. [17], who conducted wind-tunnel tests on a 0.9 m diameter battery-
charging wind turbine, using pulsed lasers to illuminate the #ow. These experiments,
although di!ering in detail from those described by the present authors, serve to
illustrate the essential PIV method, in which laser light `paintsa a two-dimensional
region of #ow containing small seeding particles. The #ow pattern is captured as
a photograph, in which each seeding particle appears as a multiple image. By
analysing the negative using optical and digital methods, the local velocities in the
entire #ow region may be determined. PIV thus allows a two-dimensional cross-
section of a #ow "eld to be captured at a single instant [18].
The PIV experiments reported in the present paper were conducted at the Univer-
sity of Edinburgh, using small-scale rotor models, and water, rather than air, as the
#ow medium. These measures enabled full-"eld velocity and vorticity maps to be
captured for an area extending from the rotor plane to 2.9 diameters (D) downstream
[19,20]. The experimental programme is described in detail below.
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 5

3. Description of the PIV experiments

3.1. Experimental facilities

The experiments were carried out in a glass-bottomed water channel, with a recir-
culating pump to maintain a steady #ow in the channel (Fig. 1). The use of water
enabled a high seeding concentration to be attained, with illumination of the entire
near wake. In addition, operating in water removed the problems of dispersion and
condensation of seeding reported in previous wind tunnel tests (e.g. [21]). The
two-blade model rotor had a diameter of 175 mm, and was driven by a DC electric
motor mounted on a platform above the water level, and connected to the rotor via
a toothed drive belt. An aluminium tube isolated the belt from the #ow, and supported
the rotor nacelle under the water (Fig. 2). The tube was streamlined to minimise
turbulence.
The rotor blades were of #at-plate pro"le with sharpened leading and trailing edges,
and thickness of 1.26 mm. The blade planform was untwisted, with linear taper from
15 mm chord at the hub to 10 mm at the tip. The channel dimensions provided
a rectangular `working sectiona 400 mm wide by 750 mm deep, with 112.5 mm clearance
between the blade tips and the side walls, and 300 mm from blade tip to free surface.
A study of the in#uence of the channel walls on the #ow past the model rotor, using the
method of images [22], estimated the velocity error due to blockage as less than 2.5%.
The speed and angular position of the rotor were measured by a tachogenerator
and position encoder connected to the drive motor. The output from the tachogenerator
was used to close an analogue control loop around the motor, enabling precise
control of rotor speed. A once-per-rev pulse output from the position encoder was
synchronised with the PIV recording equipment, so that #ow photographs could be

Fig. 1. The model wind turbine in the water channel at Edinburgh.


6 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Fig. 2. A PIV photographic #ow record of wake of two-blade #at-plate rotor at j"8.

captured at any predetermined azimuthal position of the blades. The model was
operated at tip speed ratios in the range j"3}8, with constant current velocity, and
chord Reynolds number in the range Re"6400}16 000. This is signi"cantly lower
than on a full-scale wind turbine, where Re may be of the order of 106 or more, and the
implications of this are discussed further in Section 6. An arrangement of honeycomb
section, perforated plate, and "ne mesh screen were used as #ow manipulators 7.5D
upstream of the rotor, producing an approximately uniform freestream with a mean
velocity of 0.25 m/s, and spatial turbulence intensity of 4%.

3.2. PIV imaging technique

The PIV apparatus employed the scanning beam illumination system [23]. A con-
tinuous-wave laser was scanned over a parabolic mirror below the tank to produce
a sheet of laser light which, directed vertically upwards through the glass base,
illuminated a vertical cross-section of the #ow "eld. The water was seeded with conifer
pollen (lycopodium) of average diameter 70 lm, in su$cient concentration to ensure
a high density of particle images on the photographic records. A Hasselblad 553 EL/X
camera and 80 mm lens were used to capture the images. The camera was used in
conjunction with a rotating-mirror image shifting system to increase image resolution
and remove directional ambiguity [24]. Fig. 2 shows a photograph of the rotor model
operating at j"8.
To yield two-dimensional velocity vector maps, the PIV negatives were inter-
rogated pointwise over a dense grid using optical and digital techniques [25]. The
negatives were "rst scanned with a probe laser to obtain an interference pattern from
the multiple particle images in each area. The fringes thus obtained were then
recorded in digital form and the data Fourier transformed to yield the particle
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 7

Fig. 3. A PIV velocity vector map of wake of two-blade #at-plate rotor at j"8.

velocities at each point on the negative. The resulting velocity maps were used as the
basis for all subsequent analyses, including calculation of the vorticity components in
the planar #ow "eld. Fig. 3 shows a PIV velocity vector map for the #at-plate blades
at j"8, obtained by processing the photographic image shown in Fig. 2.
Each photograph was synchronised with the rotor blades oriented vertically, the
trailing edge of the lower blade having just passed through the image plane. Six
images were captured per tip speed ratio, and the resulting vector maps numerically
averaged in order to suppress the in#uence of turbulence, and extract the coherent
structure in the #ow. This technique e!ectively produced a `frozen wakea image, from
which a cross-section of the vortex structure was obtained. Systematic and random
errors in the PIV technique contributed less than 3% and 1% error in velocity
measurement, respectively. Errors in velocity measurement were compounded during
vorticity calculations, but the relative error in vorticity was calculated to be less than
6% throughout the near wake, and was reduced by computing vorticity from an
averaged velocity vector map.

4. Free-wake numerical simulations

In parallel with the PIV experiments, theoretical predictions of wake characteristics


were performed at the University of Stuttgart using the ROVLM code (see above,
Section 2.2). The inputs to the code included the rotor blade geometry and #ow
conditions used in the PIV experiments. No prior assumptions concerning the shape
of the wake were required, as ROVLM is based on a free-wake numerical method.
8 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Fig. 4. Discretisation of the blade and wake in the ROVLM vortex-wake code.

The code incorporates a time-dependent simulation in which the rotor and wake
are treated as singularity solutions of Laplace's equation, with dipoles placed at the
rotor blades to simulate lift forces [26]. The wake #ow is assumed to be inviscid and
irrotational. The blades are discretised into panels covering the camber line of each
aerofoil section along the span of the blade (Fig. 4). Each panel contains a dipole of
constant strength equivalent to a quadrilateral vortex "lament de"ned by the corner
points of the panel. The e!ects of blade thickness are neglected. In the centre of each
panel, a control point is declared at which the singularity strength is calculated, by
applying the kinematic boundary condition in external Neumann form, so that the
velocity component normal to each panel vanishes.
Vorticity is shed into the wake as a spanwise series of panels evolving from the
trailing edge, the amount of shed vorticity being determined by the Kutta condition.
These panels in#uence the induced velocities at the blades, and the singularity
distribution (hence the blade loading) has to be re-calculated. In each time step of the
simulation an incremental portion of the wake is thus shed from the trailing edge of
the blades, resulting in a growing wake. The number of rotor revolutions required to
attain a steady state solution is dependent on the loading of the rotor and the
oncoming #ow conditions. In order to reduce the computation involved in the
free-wake calculations, a simpli"ed model is incorporated for far wake transport.
A post-processor is used to read in the geometry and the singularity strengths of the
blades and their trailing wakes. The Biot}Savart law is used to calculate velocities at
desired points in the #ow "eld, with a vortex core model used in conjunction with the
panel method when calculating the self-induction of the vortex "laments; a core radius
is assumed for each quadrilateral "lament in order to prevent the calculation of an
unrealistically high value of velocity for "eld points very close to the inducing
"lament. Once the velocities are known, the vorticity components normal to the "eld
plane are calculated from the central di!erences of the neighbouring points.
To simulate the PIV measurements, the geometry of the model rotor, the experi-
mental tip speed ratio, and the dimensions and location of the analysis area in the #ow
channel were used as input to the ROVLM code. The core radius was set to zero, i.e.
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 9

Fig. 5. A typical simulation of the two-blade #at-plate rotor PIV experiments by the ROVLM code.

vortex "laments whose position coincided with points in the wake where the induc-
tion was to be calculated were not included in the computation. From an impulsive
start, results were computed for several rotor revolutions, allowing convergence of
predicted blade loads, and transportation of the starting vortex far downstream of the
near wake. Fig. 5 shows a typical result of the ROVLM calculation for the #at-plate
blade model, including the discretised rotor and the resulting wake. A two-dimen-
sional region is depicted, corresponding to the laser sheet in the PIV system, from
which wake velocities and vorticity were subsequently extracted for the purpose of
comparison with the PIV data.

5. Results

The velocity information from the PIV experiments and the ROVLM simulations
was processed to produce the vorticity contour maps shown in Fig. 6a}e. These
correspond to tip speed ratios of j"3, 4, 5, 6 and 8, respectively. The vorticity data
f have been normalised with respect to the spacing of the velocity grid, d"7.5 mm,
and the freestream velocity, <"0.25 m/s, according to
4fd
f" . (2)
/ <
=
10 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Fig. 6. Comparison of PIV and ROVLM vorticity contour plots of full wake of two-blade #at-plate rotor,
for (a) j"3, (b) j"4, (c) j"5, (d) j"6 and (e) j"8.

The contours of the maps join points of equal vorticity with strength represented by
varying degrees of shade. The maps provide information about the geometry, strength
and instability of the vortex system in the wake, as described below.

5.1. Vortex wake geometry

The contour plots from both the PIV and ROVLM data reveal the expected pattern
from a cross-section of a helical vortex system; a series of tip vortices embedded in
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 11

Fig. 6. Continued.

a region of di!used vorticity behind the rotor (see e.g. Fig. 6a). Note that the higher
turbulence seen in the upper half of the rotor wake in the PIV plots is attributable to
interference from the rotor support structure, not entirely eliminated by the stream-
lined shroud. The PIV contour plots also reveal separate regions of concentrated
vorticity in the wake of the model rotor, located inboard of the tip at approximately
mid-span, moving under the in#uence of wake expansion, and merging with the tip
vortex system at approximately 1.0D downstream. No evidence of this phenomenon is
seen in the ROVLM simulations.
12 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Fig. 6. Continued.

5.1.1. Shape of the wake boundary


The shape of the wake boundary may be characterised by the position of the
discrete tip vortices. This is identi"ed in the contour plots as the location of maximum
vorticity within a region of #ow containing each vortex. Although drawing con-
clusions on the shape of the wake boundary from the results is made di$cult due to
wake meandering, and the errors is specifying vortex position caused by vortex
coalescence and dissipation, some general observations may be made.
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 13

Fig. 6. Continued.

At j"3 and j"4 the radius of the wake for the PIV results appears to be
approximately constant with downstream distance. At higher tip speed ratios, there
are signs of wake expansion immediately behind the rotor, becoming more pro-
nounced towards high j. In addition, at j"8 the PIV contour plot of Fig. 6e shows
contraction of the wake at around 1.5D downstream. This phenomenon is character-
istic of a wind turbine operating in the `turbulent wake statea, as de"ned by Eggleston
and Stoddard [27]. Under these conditions high blockage exists, and the near wake
14 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Fig. 6. Continued.

e!ectively acts as a solid body of revolution impeding the freestream, which must
initially diverge around it, but eventually converge again. The experimentally
observed wake contraction appears to correspond to this downstream convergence.
The phenomenon of wake contraction is not addressed in BEMT.
The ROVLM data display qualitative agreement with the trends mentioned above,
although there are some di!erences. The initial wake expansion is not as marked as in
the PIV results, while downstream wake contraction is observed at low, as well as
high, tip speed ratios in the ROVLM plots.
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 15

Fig. 7. Comparison of PIV and ROVLM values of tip vortex pitch in the wake of the two-blade #at-plate
rotor.

5.1.2. Pitch of the helical tip vortex


The axial distance a tip vortex is transported during one blade revolution may be
de"ned as the tip vortex pitch, p. The average axial coordinate of the position of the
tip vortices in the plots of Fig. 6 was used to calculate the pitch of the tip vortex spiral.
The calculation was restricted to tip vortices in the lower half of the wake since the
#ow in the upper half of the experimental wake was disturbed by the rotor support
structure. Fig. 7 shows the tip vortex pitch for both the PIV and ROVLM data as
a function of tip speed ratio. For reference, the tip vortex pitch in the ideal case of
vortices inducing no change in #ow at the rotor, i.e. travelling at freestream velocity
< , is also plotted and referred to as the `zero-inductiona curve.
=
There is good qualitative agreement between the PIV and ROVLM results and,
with the exception of the extreme cases of j"3 and j"8, the two datasets agree to
within 12%, and are bounded above by the zero-induction curve. Low induced
velocities at the blade equate to higher freestream speeds and therefore higher forces
on the rotor. Thus Fig. 7 indicates that the model rotor exerted a higher decelerating
in#uence on the #ow than that simulated by the ROVLM code.
The datasets show an overall trend of decreasing pitch of the tip vortex spiral with
increasing blade passing frequency, as expected. The exception is the case of j"8 for
the PIV results where there is a slight increase in pitch and the "gure suggests that the
tip vortices are travelling faster than the freestream. The tip vortices in the j"8 case
appear to be subject to vortex coalescence, however, which may yield errors in
specifying tip vortex position, as noted above. The PIV contour plot of Fig. 6e
appears to con"rm an increase in tip vortex spacing with downstream distance,
16 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

Fig. 8. Comparison of PIV and ROVLM values of tip vortex strength in the wake of the two-blade
#at-plate rotor.

consistent with vortex coalescence (it is possible, however, that at high j the wake may
be accelerating: Montgomerie[1] refers to wind tunnel tests at the Aeronautical
Institute of Sweden, where smoke studies on model wind turbine rotors revealed
strong acceleration of the wake spirals, following an initial deceleration; this questions
the assumption of BEMT that the wake maintains a densely packed spiral corre-
sponding to an asymptotic slow-down of the wake).

5.2. Aspects of wake vorticity

5.2.1. Tip vortex strength


The maximum vorticity in each of the tip vortices in the lower half of the wakes of
Fig. 6 was used to calculate the strength of the tip vortex spiral. Fig. 8 shows the
values of maximum vorticity averaged over the near wake and plotted against tip
speed ratio for both the PIV and ROVLM results. The datasets show an overall trend
of increasing `strengtha of the vortex spiral with higher tip speed ratio. This is
expected since, at high tip speeds, turbulence in the wake preserves the energy in the
tip vortices. Drawing conclusions on tip vortex strength is made di$cult due to vortex
coalescence in the plots of Fig. 6. In addition a number of factors may in#uence the
strength and lifetime of the tip vortices and care must be taken when isolating the
e!ect of tip speed ratio alone on vortex strength. This is discussed further in Section 6.

5.2.2. The inboard vorticity pattern


Perhaps the most unexpected result of the PIV analyses was the evidence of the
inboard trailing vorticity patterns referred to above, which are seen to be quite
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 17

Fig. 9. PIV vorticity contour plots of half wake of two-blade #at-plate rotor, for (a) j"6, (b) j"8.

separate, at least initially, from the tip vortex structure. The inboard vorticity is seen
clearly in Fig. 9a and b for j"6 and 8, respectively. These "gures highlight the lower
half of the wake (with the rotor blade at the top left hand corner of each plot). The
possibility that the inboard region is the blade root vortex may be discounted, as the
vorticity is of the same sign as the tip vortex system, with which it coalesces at
approximately 1.0D behind the rotor plane. It has been suggested that the phenom-
enon may be due to blockage by the model nacelle, but this appears unlikely, as the
diameter of the vorticity pattern greatly exceeds that of the nacelle, which was in any
case streamlined.
From examination of raw vector maps (Fig. 3) and the vorticity contour plots (e.g.
Fig. 9b), the inboard vorticity is seen to occupy the region between the slow-moving
wake core and the freestream, in the immediate downstream vicinity of the rotor
plane. One possibility may be that the vorticity is due to blockage caused by the high
18 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

solidity of the inboard blade, particularly when operating at high j. Although the
mechanism of its generation is unclear, the vorticity may represent a boundary layer
created entirely within the #ow, due to the stagnant wake core acting in the manner of
a solid body (of roughly elliptical shape). Such an `internala boundary layer might be
expected to generate vorticity in the same way as the boundary layer does at a solid
surface.
If this explanation is correct, the phenomenon would be highlighted at low Re,
when the boundary layer would be relatively thick. Conversely, it would be less
pronounced on a full-scale wind turbine than in the present PIV experiments. In tests
by Pedersen and Antoniou [13], however, the stretching of smoke traces appeared to
support the existence of this shear layer at full scale. It should also be noted that the
absence of inboard vorticity from the ROVLM results does not con#ict with the
above explanation, as the ROVLM code is inviscid, and therefore incapable of
generating vorticity within the #ow, other than that prescribed at the blade surface.

5.2.3. Vortex interaction


The PIV vorticity contour plots for j"6 and j"8 (Fig. 9a and b) both contain
indications of vortex interaction. At j"6, coalescence of adjacent vortices is observed
at around 1.0D. The preceding vortex of the adjacent pair appears to roll-up over the
top of its neighbour to produce a single vortex by 1.5D downstream. At j"8, the size
and shape of the regions of concentrated vorticity at the wake boundary suggest that
each region contains two vortices coalesced into one. This pairing process has been
commented upon previously in separate studies by Green [16] and Pedersen and
Antoniou [13].
Due to the inviscid assumptions embodied in the ROVLM code, the predicted
vortex wake system is not subject to dissipation. In addition, the blade root vortices
calculated by ROVLM are unrealistically strong, due to the invalid assumption that
high blade incidence near the hub produces large values of circulatory lift. In reality
the blade lift is limited by boundary layer separation and stall. The size and persist-
ence of the root vortices simulated by the ROVLM code result in vortex interaction
leading to instability in the helical vortex system (Fig. 6c}e). In contrast to the
ROVLM results, the PIV contour plots of Fig. 9a and b reveal that the experimental
root vortices are very small and dissipate within a blade revolution. These are areas in
which the present ROVLM code is de"cient, and further development is needed.

6. Discussion of scale e4ect

In drawing conclusions from the results of the present study, and assessing its
relevance to full-scale wind turbine aerodynamics, an obvious concern is that of scale
e!ect. The di!erence in scale between the PIV model and a full-scale wind turbine was
typically a factor of 100 [20], and the chord Reynolds number of the PIV tests was
lower by a factor of approximately 1000. Despite this, the results of the study tend to
suggest that the near wake structure of a model wind turbine shares some of the
fundamental characteristics of that of a full-scale machine.
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 19

This is inferred from the comparison of the wake vorticity maps from the PIV and
ROVLM results, whose qualitative similarity is evident in Fig. 6a}e, particularly with
regard to wake expansion immediately behind the rotor, and downstream contraction
at high j. While the chord Reynolds number of the PIV tests was in the range
6400}16000, that of the ROVLM results corresponded to Re"R, by virtue of the
assumption of inviscid and irrotational #ow. Thus the ROVLM predictions relate to
very large scale. The observed similarity in wake shapes may therefore seem surpris-
ing, as the boundary layer #ow over the blades, which dictates their lifting properties,
and hence the vorticity shed into the wake, is known to be highly sensitive to Re.
It may be the case, however, that a rotating wing generating circulatory lift, and
shedding vorticity into the downstream #ow, gives rise to an essentially inviscid wake
even at the small scale of the present PIV experiments. Model tests of this type may
therefore aid in the fundamental understanding of full-scale rotor #ows, in much the
same way that experiments on model aircraft give insight into the universal principles
of #ight. In a similar way, Gould [28] notes that low Reynolds number #ow
visualisation tests are often suitable for showing the #ow pattern changes associated
with changes in lift-coe$cients at higher Reynolds number.
A "nal note of caution should be made regarding the present PIV experiments,
namely that di!erent tip speed ratios were achieved by running the rotor at di!erent
speeds, in a constant freestream, and the chord Reynolds number therefore varied
signi"cantly with tip speed ratio. This was a limitation of the tests, and makes a strong
case for further tests in a channel in which the upstream #ow may be varied, enabling
the tip speed ratio to be altered without changing the Reynolds number. This
observation does not, however, a!ect the principle conclusions of the study.

7. Conclusions

An investigation into the properties of the vortex wake behind a wind tubine rotor
has been carried out, comparing results from laboratory measurements made with
PIV and simulations from ROVLM, a sophisticated vortex wake code. The study
focused on areas where current wake modelling requires attention; in particular, the
strength and geometry of the trailing tip vortex spiral, vital for the prediction of #uid
loading on the rotor blades.
Results from the present PIV experiments and ROVLM simulations enabled
a direct comparison to be made of the vortex wake of a two-blade rotor operating at
tip speed ratios in the range j"3}8, revealing the structure in the region from the
rotor plane to a distance of 2.9D downstream. The results show that (i) PIV can
provide a means to obtain detailed data from an instantaneous wide-"eld image of
a turbine wake at small scale, (ii) there is good qualitative agreement between the
wake structures obtained from the PIV data and the ROVLM simulations, parti-
cularly in regard to the shape of the wake boundary, including features such as wake
expansion and contraction, and despite the di!erence in Reynolds number of the
experimental and numerical data, (iii) at least at model scale there is a region of
`inboarda vorticity between the tip vortex system and the retarded inner wake
20 J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21

immediately behind the rotor, which expands to merge with the helical tip vortex
system at around 1.0D downstream, and may be due to blockage caused by the blade
root sections, and (iv) the ROVLM code requires to be further modi"ed to incorpor-
ate Reynolds number e!ects, and a realistic model for stalled #ow at the inboard blade
sections.

Acknowledgements

The authors wish to extend their thanks to the Engineering and Physical Sciences
Research Council (EPSRC) and Aerpac B.V. for "nancial support of this work.
Thanks are also due to Tom Bruce of the University of Edinburgh for his helpful
suggestions during the writing of this paper.

References
[1] B. Montgomerie, The need for more measurements, Proc. 4th Int. Energy Agency (IEA) Aerodynamics
Symp., Rome, Italy, 1990.
[2] J.D. Anderson, Incompressible #ow over "nite wings, Fundamentals of Aerodynamics, McGraw-Hill,
Singapore, 1985, pp. 229}275.
[3] A.A. Afjeh, T.G. Keith, A simpli"ed free-wake method for horizontal-axis wind turbine performance
prediction, ASME Trans. J. Fluids Eng. 108 (1986) 400}406.
[4] H. Glauert, Airplane propellers, in: W.F. Durand (Ed.), Aerodynamic Theory, vol IV, Division L,
Springer, Berlin, 1935.
[5] R.E. Wilson, S.N. Walker, State-of-the-art wind turbine aerodynamics, Proc. US Wind Energy Conf.,
San Francisco, USA, 1985, pp. 30}38.
[6] L.A. Viterna, R.D. Corrigan, Fixed pitch rotor performance of large HAWTs, DOE/NASA Work-
shop on Large HAWTs, Cleveland, Ohio, USA, 1980.
[7] H. Snel, R. Houwink, W.J. Piers, J. Bosschers, G.J.W. van Bussel, A. Bruining, Sectional prediction of
3d e!ects for stalled #ow on rotating blades and comparison with measurements, Proc. 1993
European Community Wind Energy Conf., TravemuK nde, Germany, 1993. pp. 395}399.
[8] F.J. Simoes, J.M.R. Graham, A free vortex model of the wake of a horizontal axis wind turbine, Proc.
12th British Wind Energy Conf., Norwich, England, 1990, pp. 161}166.
[9] R. Bareiss, S. Wagner, The free wake/hybrid code ROVLM } a tool for aerodynamic analysis of wind
turbines, Proc. 1993 European Community Wind Energy Conf., TravemuK nde, Germany, 1993,
pp. 424}427.
[10] R. Bareiss, G. Guidati, S. Wagner, An approach towards re"ned noise prediction of wind
turbines, Proc. 1994 European Wind Energy Association Conf., Thessaloniki, Greece, 1994,
pp. 785}790.
[11] J.G. Schepers, B. Montgomerie, H. Snel, Investigation and modelling of dynamic in#ow e!ects: yaw
modelling, Proc. 1994 European Wind Energy Association Conf., Thessaloniki, Greece, 1994,
pp. 699}707.
[12] J.M. Savino, T.W. Nyland, Wind turbine #ow visualisation studies, Proc. US Wind Energy Conf., San
Francisco, US, 1985, pp. 559}564.
[13] T.F. Pedersen, I. Antoniou, Visualisation of #ow through a stall-regulated wind turbine rotor, Proc.
1989 European Wind Energy Conf., Glasgow, Scotland, 1989, pp. 83}89.
[14] P.H. Alfredsson, J.A. Dahlberg, A preliminary wind tunnel study of windmill wake dispersion in
various #ow conditions, Technical Note of the Aeronautical Research Institute of Sweden (FFA)
UA-1499, Part 7, 1979.
J. Whale et al. / J. Wind Eng. Ind. Aerodyn. 84 (2000) 1}21 21

[15] L.J. Vermeer, J.J. Briaire, C.V. Doorne, How strong is a tip vortex?, Proc. 17th BWEA Conf., 1995,
pp. 59}64.
[16] D.R.R. Green, Modelling large wind turbines and wakes, Ph.D. Thesis, Department of Mechanical
Engineering, Loughborough University of Technology, Loughborough, England, 1985.
[17] G.H. Smith, I. Grant, A. Liu, D.G. In"eld, Diagnostics of wind turbine aerodynamics by particle
image velocimetry, Proc. 12th British Wind Energy Conf., Norwich, England, 1990, pp. 259}264.
[18] C. Gray, T. Bruce, The application of particle image velocimetry (PIV) to o!shore engineering, Proc.
5th Int. O!shore and Polar Engineering Conf. (`ISOPE '95a), vol. 3, 1995, pp. 701}708.
[19] J. Whale, C.G. Anderson, An experimental investigation of wind turbine wakes using particle image
velocimetry, Proc. 1993 European Community Wind Energy Conf., TravemuK nde, Germany, 1993,
pp. 457}460.
[20] J. Whale, C.G. Helmis, K.H. Papadopoulos, C.G. Anderson, D.J. Skyner, A study of the wake
structure of a wind turbine comparing measurements from laboratory and full-scale experiments, J.
Sol. Energy Eng. 56 (6) (1995) 621}633.
[21] M.B. Anderson, D.J. Milborrow, J.N. Ross, Performance and wake measurements on a 3 m diameter
HAWT: comparison of theory, wind tunnel and "eld test data, ETSU Contract
E/5A/CON/1090/177/020, Cavendish Laboratories, Cambridge, England, 1982.
[22] J. Whale, A study of the near wake of a model wind turbine using Particle Image Velocimetry, Ph.D.
Thesis, Department of Physics, University of Edinburgh, Edinburgh, Scotland, 1996.
[23] C. Gray, C.A. Greated, D.R. McCluskey, W.J. Easson, An analysis of the scanning beam PIV
illumination system, J. Phys.: Measurement Sci. Technol. 2 (1991) 717}724.
[24] M. Ra!el, J. Kompenhans, Theoretical and experimental aspects of image-shifting by means of
a rotating mirror system for particle image velocimetry, J. Phys.: Measurement Sci. Technol. 6 (1995)
795}808.
[25] J.M. Huntley, An image processing system for the analysis of speckle photographs, J. Phys. E 19
(1986) 43}49.
[26] R. Bareiss, S. Wagner, Load calculations on rotor blades of a wind turbine, Proc. 5th Int. Energy
Agency (IEA) Symp., University of Stuttgart, Germany, 1991.
[27] D.M. Eggleston, F.S. Stoddard, Wind Turbine Engineering Design, Ch. 1, Van Nostrand Reinhold,
1987, pp. 30}35.
[28] R.F.W. Gould, Design of a low-speed water channel. NPL AeroReport 1277, 1968, pp. 2}3.

You might also like