You are on page 1of 14

J. Wind Eng. Ind. Aerodyn.

112 (2013) 1124

Contents lists available at SciVerse ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage:

Large eddy simulation of the wind turbine wake characteristics

in the numerical wind tunnel model
Jang-Oh Mo a, Amanullah Choudhry a, Maziar Arjomandi a, Young-Ho Lee b,n

School of Mechanical Engineering, The University of Adelaide, South Australia 5005, Australia
Division of Mechanical and Energy-System Engineering, Korea Maritime, Busan 606-791, South Korea

a r t i c l e i n f o


Article history:
Received 25 January 2012
Received in revised form
30 August 2012
Accepted 26 September 2012
Available online 23 November 2012

Large Eddy Simulation of NREL Phase VI wind turbine was performed in a virtual wind tunnel
(24.4 m  36.6 m) in order to achieve a better understanding of the turbine wake characteristics. For
this purpose, ANSYS-Fluent package was used to run the simulation using the dynamic SmagorinskyLilly model. For the purpose of validation, the pressure distribution at different span-wise sections
along the turbine blade and the power produced by the wind turbine were compared with the
published experimental results for the NREL phase VI rotor tested in the NASA wind tunnel with the
same dimensions as in the model and a good agreement was found between the two. The airow
immediately behind the wind turbine was observed to be a system of intense and stable rotating helical
vortices, which determined the dynamics of the far-wake. The system of vortices in the near-wake
became unstable and broke down due to wake instability at a distance of ve rotor diameters
downstream of the wind turbine. This was dened as the boundary between the near- and far-wake
regions. The collapsed spiral wake was found to spread in all directions in the far-wake resulting in the
formation of the two pairs of counter-rotating vortices which caused the gradual increase of turbulence
in these regions. The turbulence intensity in the wake was observed to increase immediately behind the
turbine with a maximum of 12.12% at a distance of three rotor diameters downstream of the turbine,
after which a gradual decrease in the turbulence intensity was observed in the near-wake regions due
to wake instability. However, in the far-wake regions, due to counter-rotating vortices formed by
the wake instability, the turbulence intensity showed a tendency to increase intensity. Finally the timeaveraged wake velocities from the LES, with and without the blockage corrections, were compared
with WAsP and a comparatively good agreement for the axial velocity predictions was observed in
the far-wake.
& 2012 Elsevier Ltd. All rights reserved.

Large eddy simulation
Wind turbine
Wake instability
Counter-rotating vortex pairs
Vortex breakdown
Turbulence intensity
Blockage correction
Numerical wind tunnel

1. Introduction
Wind farms are locally clustered groups of wind turbines in
the same location used to produce electric power. There are many
advantages to this commercial structure; however protable
wind resources are limited to distinct geographical areas with
relatively higher wind speeds. The introduction of multiple
turbines into these areas increases the total wind energy produced and results in a reduction of the overall cost from an
economic point of view due to the concentration of maintenance
equipment and spare parts (Manwell et al., 2002). As of October
2010, 52 wind farms greater than 100 kW capacity operate in
Australia (WEC, 2010). The Waubra wind farm near Ballarat,
Victora, completed in 2009, is the largest wind farm in the

Corresponding author. Tel.: 82 51 410 4293.

E-mail address: (Y.-H. Lee).

0167-6105/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.

southern hemisphere, consisting of 128 turbines spread over

173 km and rated at 192 MW. However, in terms of generating
capacity, Lake Bonney wind farm near Millicent, South Australia is
the largest with 239.5 MW, despite having only 99 turbines (ABS,
2010; WEC, 2010).
However, despite the commercial benets of locally concentrated wind turbines, several drawbacks cannot be overlooked
because in a wind farm most wind turbines operate in the wakes
of other wind turbines. Wakes behind horizontal axis wind
turbines are complex turbulent ow structures with rotational
motion induced by the turbine blades, radial pressure slopes and
tip vortices originating from the tip vortex-trailing edge interaction by local cross ows along the trailing edge (Mo and Lee,
2011; Wagner et al., 1996). From the perspective of a wind farm
optimal layout, there are two major issues related to the wakes of
wind turbines. The rst one is the velocity decit, which results
in reduced power production for the downwind wind turbines.
The second is the increase in the turbulence levels, which leads


J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

directly to an increase in the dynamic loads and hence has

negative effects on the fatigue life of the turbine blades
(Barthelmie et al., 2009,2008; Chamorro and Porte-Agel, 2009).
The wake behind a wind turbine can be classied either as
near-wake or as far-wake; however no distinct classication of
the two regions exists. According to Vermeer et al. (2003), the
near-wake is dened as the area just behind the wind turbine
rotor and is considered to be one diameter downstream. It is the
region of the wake where the effects of the rotor aerodynamics
are apparent on the wake structure. On the other hand, the farwake is the region beyond the near-wake. Here the effects of the
wake on the downstream wind turbines in farm situations are
generally considered. Understanding the turbulent wake characteristics behind a wind turbine has been the subject of research,
both experimentally and numerically, over the last few decades
(Vermeer et al., 2003; Corten and Nederland, 2001; Crespo et al.,
In general the approaches reviewed by Vermeer et al. (2003)
and Crespo et al. (1985) have been adopted to study wind
turbine wakes. The work presented here is a continuation of the
UPMWAKE model proposed by Crespo et al. (1985) and Crespo
and Hernandez (1989) which was based on the k  e closure
methods and the explicit algebraic model for the predictions
of components of the turbulent stress tensor as proposed by
Gomez-Elvira et al. (2005). However, in all these methods, the
Reynolds-Averaged NavierStokes (RANS) was used over all the
turbulence scales and thus made it difcult to accurately predict
the turbulence characteristics of the wake.
Other semi-empirical methods like the Lissaman model
(Lissaman and Bate, 1977; Lissaman et al., 1982) and its derivatives (Vermeulen et al., 1981) and the Ris model (Jensen, 1983;
Katic et al., 1986) are based on using a near-Gaussian wake shape
and a top-hat shape for the velocity decits respectively. Furthermore, certain details in the ow eld around the turbine are
neglected and the wakes are assumed to expand linearly with
distance. A number of attempts have been made to establish more
accurate wake models. However, so far advanced and detailed
wake models, even when including an explicit representation of
turbulence and its impact on the wake expansion, have not been
able to produce convincingly better predictions (Barthelmie et al.,
2006). Therefore, to overcome the limitations of these models,
large eddy simulation (LES) has been used in this research as the
tool to investigate the details of the wind turbine wake. LES will
reproduce the unsteady oscillations of the ow characteristics over
all scales larger than the grid size; consequently, the much-needed
details of the turbulent characteristics of the wake in controlled
environments have been obtained and presented in this paper.
To the best of our knowledge, very little work has been
published in which LES is used to simulate a wind turbine

rotating in a wind tunnel. A number of researchers have used

CFD, based on the RANS equations to acquire comparatively fast
results (Menter et al., 2006; Potsdam and Mavriplis, 2009;
Srensen et al., 2002b). Others have used LES to simulate the
wake ows without the turbines, combined with the actuator line
and disc methodologies (Wu and Porte-Agel, 2001). However, the
approach of return to the basics as proposed by Vermeer et al.
(2003) is highly valuable, in that it provides the opportunity to
study the aerodynamics of the wind turbines in controlled
environments like wind tunnels. The objective of this investigation is to achieve a better understanding of the turbulent wake
characteristics behind the wind turbine (NREL Phase VI wind
turbine) that was tested in the NASA Ames 24.4 m  36.6 m wind
tunnel. For this, LES was carried out using the commercial CFD
code, ANSYS FLUENT 13. The results of the LES have been
compared with the aerodynamics of the wind turbine blade that
were obtained experimentally by the NREL (Hand and Simms,
2001; Simms, 2001). Particular emphasis has been placed on the
study of the distribution of the overall wake structure, timeaveraged axial velocity (corresponding to velocity decit) and the
increased turbulence intensity in cross-sectional planes perpendicular to the axis of the wake at uniform incoming velocity.
The study will be useful for the designers to plan and design
future wind farms for the purpose of improvement of the overall
wind farm efciency and the fatigue life of the wind turbines. It
also provides understanding of the turbulent wake characteristics
of a wind turbine and the much needed results required for
validation of turbine wake models. For this reason, the results of
LES were also compared with the simple equations provided by
WAsP to estimate the wake velocities in the far-wake.

2. Numerical simulation
2.1. Specication of the NREL phase VI wind turbine
In May 2000, NREL successfully completed the analysis of the
Phase VI wind turbine in the NASA Ames 24.4 m  36.6 m wind
tunnel. The details of the experiment and the results were then
released on the NREL website [1] in order to verify the performance of commercially available analytical codes developed
around the world, while keeping the dependence on upstream
parameters, such as the atmospheric boundary layer velocity and
turbulence intensity proles, to a minimum. The purpose was
therefore to gain an insight into the wind turbine aerodynamics
and loads with minimal interference from the upstream. Following in similar footsteps, the aim of the present article is to gain an
in-depth understanding of turbine wake characteristics by suing
the NREL Phase VI wind turbine for CFD simulations, as the

Fig. 1. (a) NREL Phase VI wind turbine in the NASA-Ames 24.4 m by 36.6 m wind tunnel and (b) CFD model of the turbine in a numerical wind tunnel.

J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

required geometrical data and the test results are publically

available and can be accessed easily.
The NREL Phase VI wind turbine is a stall-regulated wind
turbine that produces a rated output power of 19.8 kW. The CFD
model of the NREL Phase VI wind turbine is presented in Fig. 1 in
comparison with the experiment, based on the shape data
information (Hand and Simms, 2001; Simms, 2001). The details
of the turbine model are also shown in Table 1. For the case
considered in the present study, the rotor cone angle is set at 01,
and the blade pitch angle is set at 51. This rotates the blade tip
chord line by 51 towards feather, relative to the rotor plane, thus
pointing the leading edge to the oncoming wind. The twist angle
is a maximum of 20.041at the root, and reduces to a minimum of
2.151at the tip in order to maximize aerodynamic performance.
The details of the NREL wind turbine blade are shown in Fig. 2.
The S809 airfoil was used in the construction of the blades. The
S809 airfoil is a 21% thick airfoil, shown in Fig. 3, specically
designed by Somers [1] for sustained maximum lift, insensitivity
to surface roughness, and low prole drag.
2.2. Computational mesh
The computational domain has a height of 24.4 m and a width
of 36.6 m, corresponding to NASA-Ames 24.4 m  36.6 m wind
tunnel, with a length of 222.32 m in the stream-wise direction.
This length corresponds to twenty times the rotor diameter as
shown in Fig. 4. The wind turbine has a tower height of 11.5 m
and is placed approximately in the middle of the wind tunnel at a
distance of 2d from the upwind boundary. The wind turbine
blades are operating under uniform velocity conditions as in the
experiment and, therefore, we do not need to consider the
boundary layer prole at the inlet. However, in the real eld
conditions, such an effect, including the turbulence prole, should
be considered as the inlet boundary conditions. According to
Hand and Simms (2001) and Simms (2001), the key factor in the
Table 1
Specication of NREL Phase VI wind turbine.
Number of blades, Z
Rotor diameter, d
Rotor radius, R
Rotational speed, N
Cut-in wind speed, Vc
Rated power
Power regulation
Rotational direction
Global pitch angle
Tower height, h

10.058 m
5.029 m
71.9 rpm
6 m/s
19.8 kW
11.5 m


Fig. 3. Shape of S809 airfoil.

choice of wind tunnel was the test section size and minimal
blockage it offered for the subject experiment. It was observed by
Simms (2001) that the ratio of the test section speed to the test
section speed required to achieve the same turbine performance
in the absence of the test section boundaries, for different pitch
angles of the turbine blade, was consistently less than 1%. This
indicated that blockage was of minimal concern in the experiments. Similar observations were made by Srensen et al. (2002a)
during their simulation where they concluded that the tunnel
blockage was of negligible concern and was the key parameter for
the choice of the tunnel.
The computational domain for the wind turbine placed in the
wind tunnel is illustrated in Fig. 5. The domain consists of two
parts, namely the moving parts (cylinder and rectangle part) and
the stationary part (wind tunnel part). The sliding mesh technique is applied to the moving parts with two mesh interface zones
(interface zone 1: cylindrical part and rectangular part, interface
zone 2: cylindrical part and wind tunnel part) in the shared faces
where the meshes overlap. A sliding mesh technique is a recommended method for computation of the unsteady ow led when
a time-accurate solution (rather than a time-averaged solution) for
rotor-stator interaction is desired. The sliding mesh model is the most
accurate method for simulating ows in multiple moving reference
frames, but it is also the most computationally demanding.
The rectangle surrounding the two blades was composed of
hexahedral meshes with 20 ination layers on the blade surface, a
spacing ratio of 1.1 in the normal direction and a rst height of
0.01 mm in order to accurately capture the boundary layer region.
Y value was set to approximately one at the blade tip and
decreased towards the blade root. The grid distribution around
the rotor blade is shown in Fig. 6. The total number of cells in the
numerical grid was 3.6  106, consisting of hexahedral meshes
over the total domain and partially cooper meshes in the
cylindrical part. The blade was meshed in the chordwise direction
with 100 nodes on both the upper and lower surfaces with a
higher concentration near the leading and the trailing edge
regions. Furthermore, 150 nodes were used in the spanwise
direction as indicated in Fig. 6. The original NREL S809 airfoil
that contained a sharp trailing edge was clearly not the case for
the real blade. This feature, furthermore, unnecessarily complicated the construction of the hexahedral grids and therefore a
blunt trailing edge was used along the blade length by cutting at
0.99 position of each chord length.
2.3. Boundary conditions

Fig. 2. Geometry of the NREL Phase VI wind turbine blade.

A uniform velocity condition of 7 m/s with turbulence intensity of 0.2% was applied as the boundary condition at the inlet
where the ow enters the computational domain. It should be
noted that the uniform velocity prole and the small turbulence
intensity considered in this research corresponds to the experiments conducted by the NREL. For the outlet where the ow
leaves the domain, the ambient domain condition was selected.
Wall condition was applied at the walls of the tunnel, ground and


J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

Fig. 4. Schematic of NASA Ames 24.4 m  36.6 m wind tunnel with the NREL Phase VI wind turbine.

The steady state computation was rst carried out for approximately 1000 iterations to have a convergence below 10  3, and
then the unsteady computation used the steady state ow
solution as the starting condition. The simulation took approximately 120 h per case on a single CPU with eight cores with
processing power of 2.4 GHz, based on operating system Linux.
The blade completed one cycle of revolution in 0.834 s with a
time step size of 0.005795 s corresponding to a blade rotation of
2.51. LES calculation was run for a sufciently long ow-time to
obtain stable statistics of the ow and turbulence. The 1440
sampling data during 8.35 s, corresponding to exactly 10 rotations
of the wind turbine, were taken.

Fig. 5. Computational domain.

surfaces of the wind turbine whereas pressure condition with

ambient pressure was applied at the downstream. Spectral
synthesizing technique was employed for modeling the turbulent
intensity at the inlet. It is based on the random ow generation
technique originally proposed by Kraichnan (1970) and then
modied by Smirnov et al. (2001). In this method, the uctuating
velocity components are computed by synthesizing a divergencefree velocity-vector eld from the summation of Fourier harmonics.
The present work was carried out using an ANSYS FLUENT
13.0, a general-purpose commercial CFD code. FLUENT employs a
cell-centered nite-volume method based on a multi-dimensional
linear reconstruction scheme, which permits use of computational elements with arbitrary polyhedral topology, including
quadrilateral, hexahedral, triangular, tetrahedral, pyramidal, prismatic, and hybrid meshes. There are several choices of the solver
algorithms in FLUENT including coupled explicit, coupled implicit
and segregated solvers. For the computations presented in this
paper, we used the coupled implicit exclusively to speed up
the convergence. The convective terms were discretised using
Bounded Central Differencing scheme.

2.4. Large eddy simulation based on dynamic Smagorinsky-Lilly

Turbulent ows are characterized by eddies with a wide range of
length and time scales. The largest eddies are typically comparable
in size to the characteristic length of the mean ow (e.g., shear layer
thickness). The smallest scales are responsible for the dissipation of
turbulence kinetic energy. It is possible, in theory, to directly resolve
the whole spectrum of turbulent scales by using an approach known
as direct numerical simulation (DNS). No modeling is required in
DNS. However, DNS is not feasible for practical engineering problems involving high Reynolds number ows. The cost required for
DNS to resolve the entire range of scales is proportional to Re3t ,
where Ret is the turbulent Reynolds number. In large eddy simulation (LES), large eddies are resolved directly, while small eddies are
modeled. LES can be considered as an intermediate between DNS
and RANS (Reynolds-Averaged NavierStokes equations). Resolving
only the large eddies allows one to use much coarser mesh and
larger time-step sizes in LES than in DNS. However, LES still requires
substantially ner meshes than those typically used for RANS
calculations. In addition, LES has to be run for sufciently long
ow-times to obtain stable statistics of the ow being modeled. As a
result, the computational cost involved with LES in normally orders
of magnitudes higher than that for steady RANS calculation in terms

J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124


Fig. 6. Grid distribution around the rotor blade.

of memory and CPU time. Therefore, high-performance computing is

a necessity for LES, especially for industrial applications.
The ltered continuum and NavierStokes equations are given
as follows:


rui 0

@   @p @ti,j
rui uj
@xi @xj

where ui is the resolved velocity in i-direction (i1, 2, and

3 correspond to the x, y and z directions), si,j is the stress tensor
due to molecular viscosity and ti,j is the subgrid-scale stress and
are dened by the following equations:

@ui @uj
2 @u
si,j  m

 m l di,j
3 @xl

ti,j  rui uj rui uj

The subgrid-scale stresses resulting from the ltering operation are unknown, and require modelling. The subgrid-scale
turbulence models employ the Boussinesq hypothesis (Hinze,
1959) as in the RANS models and therefore the subgrid-scale
turbulent stresses can be computed from the following:


t d 2 mt Si,j
3 k,k i,j

where mt is the subgrid-scale turbulent viscosity. The isotropic

part of the subgrid-scale stresses tk,k is not modelled, but added
to the ltered static pressure term. The rate of strain tensor for
the resolved scale is represented by Si,j and is dened by the
following expression:

1 @ui @uj

2 @xj
Germano et al. (1991) and subsequently Lilly (1992) conceived
a procedure in which the Smagorinsky model constant, CS is
dynamically computed based on the information provided by the
resolved scales of motion. The dynamic procedure thus obviates
the need for users to specify the model constant CS in advance of
the simulation. The concept of the dynamic procedure is to apply
a second lter (called the test lter) to the equations of motion.

b ) is equal to twice the grid lter width (D).

The new lter width (D
Both lters produce a resolved ow eld. The difference between
the two resolved elds is the contribution of the small scales, the
size of which is in between the grid lter and the test lter. The
information related to these scales is used to compute the model
constant. In ANSYS FLUENT, variable density formulation of the
model is considered and explained as follows.
At the test ltered eld level, the SGS stress tensor can be
expressed as:

T ij rd
ui uj  r
ui r
uj =r
Both Tij and ti,j are modeled the same way with the
Smagorinsky-Lilly model, assuming scale similarity:

~ S~ ij  1 S~ dij
tij 2C rD2 9S9

b 2 9b
~ b
S~ ij  S~ kk dij
T ij 2C r

In Eqs. (8) and (9), the coefcient C is assumed

be the same

and independent of the ltering process C C 2s . The grid
ltered SGS and the test-ltered SGS are related by the Germano
identity such that:

b~ u~  1 r
b u~ i r
b u~ j
ij T ij tij ru
i j
where ij is computable from the resolved large eddy eld.
Substituting the grid-lter Smagorinsky-Lilly model and Eq. (9)
into Eq. (10), the following expressions can be derived for C with
the contraction obtained from the least square analysis of Lilly:

Lij  Lkk dij =3
M ij M ij
Mij 2

b 2r

S~ ij rD2
S~ ij


Further details of the model implementation in ANSYS FLUENT

and p
its validation can be found in the article by Kim (2004).
C s C was obtained using the dynamic Smagorinsky-Lilly
model in time and space over a fairly wide range. To avoid


J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

3.1. Validation of the simulation

blade length. Note that Time averaging was performed using the
total simulation time of 8.35 s, corresponding to ten revolutions
of the turbine blade. A small discrepancy can be observed near the
suction side at x/c 0 at r/R 0.95, the outermost region of the
blade, due to tip vortices. Since the wake characteristics of a wind
turbine are closely related to the change in axial and angular
momentum (Burton, 2001) that gives rise to the pressure distribution on the blade, accurate predictions of the pressure
coefcients along the blade provide the necessary validation for
the subsequent wake analysis.

3.1.1. Pressure coefcients comparisons with experiment

The measured and computed pressure distributions at different spanwise locations along the blade length for a wind speed of
7 m/s and yaw angle of 01 are shown in Fig. 7. The spanwise or
radial sections presented are at a distance of 0.3R, 0.47R, 0.63R,
0.8R and 0.95R. The computed LES results of time-averaged
pressure distributions show a very good quantitative agreement
with the pressure distribution measurements (Hand and Simms,
2001; Simms, 2001) for the different span-wise sections along the

3.1.2. Comparison of power output

The oscillating power output of the wind turbine is shown in
Fig. 8. The time-averaged power is 5.826 kW. The power curve has
some irregularities but shows a stable behaviour with an approximate time interval of 0.42 s and a frequency of 2.38 Hz as
observed in Fig. 9. In reality the uctuating power curve has
spectral attributes and it is often useful to analyze them for
interpretation of the time-sequence data from a transient

numerical instability, both the numerator and the denominator in

Eq. (11) are locally averaged (or ltered) using the test-lter. In
ANSYS FLUENT, Cs is also clipped at zero and 0.23 by default. In
this research, the dynamic Smagorinsky-Lilly model was applied
to yield the best results for a wide range of ows.

3. Results and discussions

Fig. 7. Comparison between the experimental (Hand and Simms, 2001; Simms, 2001) and the LES result of time-averaged chordwise pressure coefcient at selected
spanwise sections along the blade at an inlet velocity of 7 m/s. (a) r/R=0.3, (b) r/R=0.47, (c) r/R=0.63, (d) r/R=0.8 and (e) r/R=0.95.

J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

solution. In signal processing Fast Fourier Transform (FFT) enables

us to take any time dependent data and resolve it into an
equivalent summation of sine and cosine waves. In the current
research FFT analysis using 1440 sampling power data acquired at
each time step was performed and the result with different
frequencies are presented in Fig. 7. The results show that 1st
frequency of the oscillating curve is 2.42 Hz with amplitude of
0.04 kW, and the 2nd and 3rd frequencies are 4.83 Hz and 2.29 Hz
with amplitudes of 0.0297 kW and 0.0238 kW, respectively.
Therefore, the 1st frequency of 2.42 Hz calculated by the FFT
analysis determines the major uctuating power frequency and
closely matches the 1st frequency of 2.38 Hz observed in Fig. 9
with a difference of only 1.65%.

Fig. 8. Oscillating output power for 7 m/s wind speed.


Previous researchers (Menter et al., 2006; Potsdam and

Mavriplis, 2009; Srensen et al., 2002b) have shown that the
steady-state calculations with two-equation turbulence models
have the tendency to over predict the power output from a
minimum of 9.315% up to a maximum of 64.375% when compared with the experimental data (Hand and Simms, 2001;
Simms, 2001). As presented in Table 2, LES calculations in the
present study show a very good agreement with the measured
power output. Only a slight under-prediction of 3.093% is
observed. It is believed that this difference comes from the
convergence criteria rather than the computational mesh number
around the rotor blades. The convergence criteria sensitivity
analysis was performed for two values of 1  10  3 and 2  10  3
for one cycle of revolution of the turbine blade. The results
indicated a 2% difference between the two criteria. However,
the computational cost associated with the 1  10  3 criteria was
approximately twice as much when compared with 2  10  3 due
to the low slope of the residual curve. Since for stable statistics of
the ow, a larger number of revolutions was needed (in this case
10 revolutions were simulated), therefore, the convergence criterion for the residuals was set to a value to 2  10  3.
Similarly due to limited computer resources grid sensitivity
checks could not be performed. However, it is anticipated that the
grid sensitivity has a little effect on the power output due to the
ne mesh used for the boundary layer region of the blade surface
from root to tip. On the other hand, for the downstream wake
region an equal mesh size of 0.25 m interval was used in all
directions. This mesh size was the allowable minimum size due to
the restrictions put by our computer resources. Even though grid
sensitivity in the wake region affects the wake length and
instability, sudden changes in ow patterns are not expected
because the aerodynamic characteristics of the blades determine
the ow characteristics of the near-wake (Burton, 2001) and these
characteristics have shown a very good agreement with the
experimental results.
Since the turbulent wake measurements of wind turbines have
not been performed in wind tunnels, there are no experimental
results that can validate the LES results for the turbulent wake
characteristics. However, since aerodynamic characteristics like
pressure coefcients and power output have been predicted with
reasonable accuracy, and since the turbulent wake characteristics
depend largely on these aerodynamic characters like the leading
and trailing edge separation (Corten and Nederland, 2001), the
turbulent wake analysis that follows is considered to be sufciently accurate. Moreover, LES is a suitable tool for transition
prediction resolving a wide range of length and time scales
(Launder and Sandham, 2002) and hence it is believed that it will
lead to useful conclusions.
3.2. Velocity proles in the wake

Fig. 9. FFT analysis using 1440 sampling data of power for 7 m/s wind speed.

The horizontal proles of the time-averaged normalized

y-velocity and the y-velocity contours for one upstream and
several normalized downstream locations behind the turbine for
 1.75 oz/do1.75 and at the hub centre (x/d0) for the reference wind speed (Vref ) of 7 m/s are presented in Fig. 10.

Table 2
Comparisons of power output from different studies for a wind speed of 7 m/s.

Experiments (Hand and Simms, 2001; Simms, 2001)

Srensen et al.(2002b)(free condition)
Menter et al. (2006)
Potsdam and Mavriplis, 2009 (NSU3D triangular surface, ne)
LES (unsteady result)

Power (kW)

Power coefcient

Thrust coefcient

Percentage error in power







J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

Fig. 10. Comparison of horizontal proles of the time-averaged normalized y-velocity and time-averaged normalized y-velocity contours at corresponding downstream

The velocity proles for the case without the wind turbine are
also shown in the plots as the blue dotted lines. The axial velocity
(Vy) has been normalized with the reference wind speed. A near
symmetrical behaviour of velocity proles can be observed at
each location. At the upstream location there is no difference in
the normalized velocity except for the region of  0.5 oy/d o0.5,
where the velocity is 0.08% lower than the case when there is no
wind turbine. This is due to the momentum resistance of the wind
turbine and its rotational effect. This velocity decit is generally
catered by the velocity induction factor. On the other hand right
after the airow passes through the wind turbine, there is clear
evidence of the turbine extracting momentum from the incoming
uniform ow and producing a wake. This can be observed from

the regions of reduced velocity at y/d 2 where a W-shaped

velocity prole can be observed. The axial velocity falls down
almost to 0.73 of the reference wind speed with a wake width
approximately equal to the rotor diameter. This prole shape can
largely be attributed to the aerodynamic design of the blades and
the rotor hub. The W-shaped velocity prole becomes atter as
the distance increases downstream due to the mixing effect and
the momentum recovery of velocity decit.
It is difcult to dene the wake width from numerical results,
especially when the ow is being constrained by the wind tunnel.
In the current research, the wake width is dened using the wind
speed ratio as the region where the wind speed ratio is smaller
than 0.99. The inlet reference velocity could not be used in this

J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

case and the average velocity in the regions  1.5o z/do  1.0
and 1.0oz/do1.5 for each downstream location is used instead
to dene the wake width at that location. The average velocity
with the wind turbine at these locations increases downstream
and therefore the average velocity increases by 3.37%, 3.54%,
3.82%, 4.1%, 4.53%, 5.15% and 6.04% of the inlet reference velocity.
This is due to the boundary layer effect that results in continuous
supply of momentum into the regions by developing velocity
proles near the side walls of the wind tunnel model.
Using this denition of the wake width, it is interesting to note
that the wake does not expand linearly in the near-wake and
remains approximately constant downstream of the wind turbine
due to the entrainment caused by the wind tunnel. On the other
hand, in external ows, turbulence in the wake mixes the low
velocity uid in the wake with the high velocity uid outside it.
In this way, momentum is transferred into the wake; the wake
expands, and the velocity decit is reduced. In a wind tunnel
experiment, gradual increase in the velocity as compared to the
reference velocity, as explained above, can be observed due to the
boundary layer effect in the ducted ows. Therefore in the current
research, wake expansion is observed only slightly.
The normalized y-velocity contours can also be observed in
Fig. 10 where the effects of the tower shadow can be observed by
the relatively reduced velocity in the tower wake. Fig. 11


demonstrates the process of velocity decit and recovery behind

a wind turbine in the wind tunnel model. It can be observed that
the wake effects remain noticeable even in the far-wake at
distances as large as y/d 20.
3.3. Turbulence intensity in the wake
The horizontal proles of the time-averaged turbulence intensities (TI) behind the turbine for one upstream and several
normalized downstream locations behind the turbine for
 1.75 oz/do1.75 and at the hub centre (x/d0) for the reference wind speed of 7 m/s are presented in Fig. 12. The uniform
wind velocity at the inlet is 7 m/s with TI of 0.2%. The TI for the
case without the wind turbine is also shown as the dotted line in
the plots. It can be observed that there is no increase in the TI for
the case without the wind turbine.
According to Chamorro and Porte-Agel (2009), large increases
in turbulence intensities are observed in the wake of the wind
turbine compared to the relatively low turbulence intensities in
the incoming ow. Similar observations can be made from Fig. 12
where the large increase in TI is observed in the wake just
downstream of the wind turbine. As noted in the gure, the TI
just prior to the wind turbine is comparatively very low, close to
0.5%. However at y/d2, the maximum TI observed is approximately

Fig. 11. Overlapped time-averaged normalized y-velocity contour and y-velocity vector at x/d 0 plane.

Fig. 12. Comparison of horizontal proles of the time-averaged TI.


J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

Fig. 13. (a) Maximum TI in the wake at each section downstream location (b) Components of TI measured on a line between P1 (0, 1d, 0.5d) and P2 (0, 20d, 0.5d).

10.58%. The four peaks in the TI for the downstream locations

correspond to the regions of the blade tip and root. TI increases as
the distance in the near-wake is increased and the peak TI of 12.12%,
approximately sixty times larger than the TI at the inlet, is observed
at a distance of y/d3. Further downstream a gradual decrease in TI
is observed till y/d5, where the TI gradually drops down to 9.19%
due to wake instability (Okulov and Sorensen, 2004a). It is important
to point out that this is within the typical range of distances between
adjacent wind turbines in wind farms and, therefore, it should be
taken into account when calculating power losses and wind loads on
the turbines.
The maximum TI at each downstream location is plotted in
Fig. 13(a) where a large decrease in maximum TI can be observed
after y/d5. As explained in the next section, this occurs due to
wake instability and vortex breakdown in the near-wake. Moreover,
an interesting phenomenon is observed regarding the maximum TI
in the far-wake. The reduced TI of approximately 3% at y/d7 shows
a tendency to increase up to 4.5% at y/d20, even though it is
known that turbulence acts as an efcient mixer, leading to recovery
of the decit and a decrease in the overall TI. It is believed that this
phenomenon results from counter-rotating vortex pairs formed by
the wake instability as explained in the next section.
The components of TI measured on a line between P1 (0, 1d,
0.5d) and P2 (0, 20d, 0.5d) (as shown in Fig. 4) is shown in
Fig. 13(b). It can be observed that the largest TI component is in
the axial direction (TIy), the magnitude of which is almost double
in comparison to the other two components (TIx and TIx) at all
downstream locations. The TI in the axial direction is observed to
increase till y/d 1 and then decreases gradually till y/d 5.
On the other hand the other two components of TI are observed
to increase till y/d 5. After this all components reduce drastically
due to the vortex breakdown that occurs due to the wake
instability as explained in the next section.
3.4. Flow structures in the near and far-wake
Wake instability behind wind turbines is considered to be one of
the most important parameters that determine the operation and
performance of tandem wind turbines operating in the wake of
other turbines. It is dened as the degree of susceptibility of wind
turbine wake to upstream and surrounding turbulence and is
considered to be the primary cause of wake meandering (Medici
and Alfredsson, 2008; Sanderse, 2009). Furthermore, wake instability results in the breakdown of large-scale vortices into smaller
vortices/eddies. Therefore, if a tandem wind turbine is located in a
wake with stable vortices (near-wake), it will experience more

severe fatigue loads as compared to being placed in the region

where vortex breakdown into a continuous vortex sheet has
occurred (Okulov and Sorensen, 2004b). In order to understand
the basic ow structures in the wake of a wind turbine and to
thoroughly understand wake instability and its relation with the
turbulence intensity in the wake, instantaneous y-vorticity contours
are presented in Fig. 14. Note that the y-vorticity contours range
from  0.5 to 0.5 of y-vorticity magnitude in the wake of the wind
turbine at 8.34 s, corresponding to 10 rotations of the turbine blades.
It can be observed from Fig. 14 that the wake immediately
behind the wind turbine is a system of intense and stable rotating
helical vortices. Several important ow structures can be recognized in the wake immediately behind the turbine. For example, a
ring of high vorticity can be observed before y/d5 in the nearwake, corresponding to the blade tip regions. This vortex structure is termed as the blade tip vortex. Similarly, the vortex
structure in the centre is formed due to the interaction of the
root vortex structure with the hub vortices. It should be noted
that the root vortex structures stemming from the blade root are
not identiable in Fig. 14 due to their close proximity to each
other as compared to the tip vortices. The close proximity,
therefore, results in the spiral structure being destroyed very fast
due to the interaction with the hub vortex.
As the distance behind the turbine is increased, the blade tip
vortex structure in the wake becomes unstable yielding a sinuous
shape at y/d 4 and breaks down completely due to wake
instability at y/d 5. The breakdown results in a large reduction
of the maximum TI that was observed in Fig. 13. As the distance
increases, the collapsed spiral wake spreads and from y/d 10 two
counter-rotating vortex pairs are observed. The phenomenon of
vortex breakdown, as described, can also be observed in the eld
experiment test performed at the Ris test centre in Denmark [2],
as shown in Fig. 15. In the eld experiment, a single blade turbine
was studied to avoid disturbances from the other blades and the
blade tip vortex was identied by smoke exiting the blade at two
radial positions close to the tip. It can be observed in Fig. 15 that
the blade tip vortex becomes unstable at approximately 3 rotor
radii downstream of the turbine. When compared with Fig. 14, it
is possible to see that the breakdown of the blade tip vortex for a
wind turbine in a wind tunnel occurs at a much later downstream
position. The earlier breakdown during the eld experiment could
be explained by the presence of atmospheric turbulence and the
energy exchange between the surrounding air and the wake.
In order to thoroughly investigate the relation between the ow
structures caused by wake instability and the gradual increase of TI
observed in the far-wake, overlapped instantaneous distributions of

J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124


Fig. 14. Instantaneous y-vorticity at 8.34 s corresponding to 10 rotations of the wind turbine blade at selected axal locations.

Fig. 15. A eld experiment at Ris test centre.

y-vorticity and the vectors of velocity magnitudes in the x and z

direction at the normalized distance of y/d3, and y/d20 are
represented in Fig. 16. In the near-wake at y/d3, strong y-vorticity
is seen near the location of blade tip, in comparison with the farwake at y/d20. This indicates that y-vorticity is more intense than
x and z vortices. The ow structure of two counter-rotating vortex
pairs with unequal strengths is shown through the velocity vectors
at y/d20 in Fig. 16(b). The red vortex of the upper counter-rotating
vortex pair circulates anticlockwise and collides with the opposite
blue vortex resulting in ow acceleration in the tangential direction
at the point of contact. In the process of this collision between the
vortices, the normal Reynolds stress in the x direction (u02 ) at P2 and
in the z direction (w02 ) at P3 relatively increase. This can be observed
in the comparison of the normal Reynolds stresses at P1, P2 and P3
as given in Table 3 through the ratio u02 : v02 : w02 . It is believed that
the gradual increase of TI in the far-wake is attributed to this ow
structure caused by the wake instability. As mentioned by Vermeer
et al. (2003). It is difcult to distinguish both types of wakes that is
near and far-wake. We suggest that the clear boundary between the
near-wake and far-wake should be the starting location of the wake
breakdown and in our investigation, this location is at a distance of
5d downstream from the wind turbine.
3.5. Comparisons with WAsP
The time-averaged wake velocities from the LES are compared
with WAsP (Mortensen et al., 2003) of the European Wind Atlas

(Troen and Petersen, 1989), as shown in Fig. 17. The WAsP model
is one of the oldest and the simplest wake models, still in wide
use, and assumes a linearly expanding wake with a velocity
decit. It is important to note that the WAsP and the LES model
are based on completely different concepts since WAsP cannot
cover the complicated ow structures in the wake of wind
turbines. However, at present, since there is no experimental
information to compare the wake characteristics with; comparison of LES results was made with WAsP in order to gain a deeper
understanding and highlight the deciencies of both models.
The WAsP is a standard method for wind resource predictions
on land as well as offshore. In WAsP model (Jensen, 1983;
Mortensen et al., 2003; Troen and Petersen, 1989) the velocity
decit in the wake is calculated using Eq. (13) and is related to the
freestream velocity magnitude (Vfreestream), thrust coefcient (CT),
rotor diameter (d), wake decay factor (kwake) and downstream
distance from the turbine (y).
2 #
Vwake =Vfreestream 1 1 1C T
d 2 kwake y


The wake decay factor for land cases is generally taken to be

0.075, whereas for offshore cases a value of 0.05 is recommended.
The velocity decit in the wake calculated using these values and
the experimentally determined thrust coefcient of 0.487 is
compared with the LES in Fig. 17. It can be observed that a good
correlation of the velocity predictions exists for the far-wake
regions after y/d 14 where the LES results can be seen to
converge to the WAsP predictions. However, in the near-wake
region the LES results are more reliable due to accurate predictions of the aerodynamic characteristics of the blade and due to
the limitation that the WAsP wake prediction algorithms are valid
for distances in excess of 3d (Mortensen et al., 2003). The gap in
WAsP predictions and the LES results can be observed to increase
between y/d 5 and y/d 14 and therefore more work is required
for this region.
As stated earlier, Hand and Simms (2001); Simms, 2001
concluded that blockage was of minimal concern for the subject
experiment. However, since wake studies were not the primary
focus of their work, the study conducted to examine the effects of
the blockage only included the wind turbine itself and did not
deal with the wake. Furthermore, since WAsP is suitable for
external ows, a blockage correction calculation needed to be
performed to conrm the effects of the wind tunnel on the wake.
Blockage correction equation suggested by Glauert (1933) and


J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

Fig. 16. Overlapped instantaneous distirbution of y-vorticity and veocity magnitude vectors in x and z directions at 8.34 s or after 10 rotations. (a) y/d=3 and (b) y/d=20.

Table 3
Comparison of normal Reynolds stresses in the x, y and z directions at P1(y/d 3),
P2 and P3(y/d 20).




u02 : v02 : w02

TI (%)






Fig. 18. Comparison of wake width predictions using WAsP and LES.

Fig. 17. Comparison of LES results, with and without blockage correction, and the
WAsP wake model with two wake decay factors.

Gould (1969) was used for this purpose and the results are shown
in Fig. 17. Glauert (1933) presented the derivation of the blockage
correction especially applicable to propeller testing which was
subsequently applied by Fitzgerald (2007). Glauerts correction
can be summed-up as an expression for an equivalent free stream
velocity, V by the following relationship:
t1 a1
2  1 2  t1 2
In Eq. (14),




As seen in Fig. 17, the blockage correction results show 0.89%

lower values than the original data at all the distances, but the
curve shape still remains the same. The blockage effect on the
wake velocities, calculated using the Glauerts correction, is
reasonably close to the less than 1% measured by the NREL
(Simms, 2001) and therefore Glauerts correction can be reasonably applied for blockage estimations in this case. The purpose of
performing the blockage corrections is to illustrate that it is of
minimal concern in this simulation and the velocity decit
information provided is accurate. This was, as mentioned earlier,
one of the primary reasons for choosing the NASA Ames
24.4 m  36.6 m wind tunnel for the NREL experiments.
The wake, in WAsP model, is assumed to have a top hat form
and the wake width (dw) is assumed symmetric in the vertical
and lateral directions. The wake width can be calculated using the
following equation.
dw d 2kwake y


The linearly expanding wake width predictions using Eq. (15)

for the two different wake decay factors are compared with the
wake width obtained from the LES as shown in Fig. 18. From the
LES, it can be observed that the wake width remains approximately constant with the downstream distance compared to the
linearly increasing wake width calculated using the WAsP model.
It can be observed that the wake width from the LES increases in
the near-wake-region until y/d 5, corresponding to the location
of the blade tip vortex breakdown. After the initial increase, a

J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

decrease in the wake width is observed till y/d11. The wake

width then starts to increase again, reaching a maximum at
y/d 20. The maximum wake width observed from LES is approximately 1.44 times larger than the wake width at y/d 1, whereas
it is approximately 3 or 4 time larger for WAsP predictions
depending on the wake decay factor. The primary cause of the
different results from the two models is due to the constraints of
the numerical wind tunnel, which does not allow the wake size to
increase as rapidly in the LES as compared to the WAsP

4. Conclusions
Large eddy simulation on the turbulent wake characteristics
behind NREL Phase VI wind turbine was performed to achieve a
better understanding of the wind turbine wake formation and
propagation. The CFD simulation model included a numerical
NASA Ames 24.4 m  36.6 m wind tunnel in which the experiments were performed. The reliability and validity of the analysis
were veried using the published results of the experiment and
an excellent agreement was observed in the comparisons of timeaveraged pressure coefcients and power. It was observed that
LES showed much better results than the steady-state calculations of the two-equation turbulence models of previous
researches. These comparisons provide sufcient evidence that
the predictions of the wake characteristics behind the wind
turbine are accurate, since the aerodynamic characteristics of
the blade are closely related to the wake characteristics. Important observations made as a result of this research are as follows:
(1) A W-shaped velocity prole was observed downstream at
y/d 2 at the normalized distance of  0.5 oz/d o0.5. The
axial velocity was observed to reduce to 0.73 of the freestream velocity at the inlet with a wake width of approximately equal to the rotor diameter. The velocity prole then
became atter as the distance downstream was increased and
the velocity decit was reduced due to momentum recovery.
The wake effects were still quite noticeable even in the farwake at a distance of y/d 20.
(2) TI was observed to increase considerably in the wake of the
wind turbine. In the TI curves, peaks were observed in the
near-wake in the regions corresponding to wakes of blade tip
and root. The maximum TI was observed to increase till
y/d 3, where the peak TI was noted to be 12.12% which is
approximately sixty time larger than the TI at the inlet, after
which a gradual drop was observed till y/d 5. After y/d 5,
due to the collapsed spiral wake breakdown caused by wake
instability, a large decrease in the maximum TI was observed.
TI continued to decrease afterwards till y/d 8 after which it
started to increase slightly due to a system of counterrotating vortices in the far-wake.
(3) The ow structure behind the wind turbine was a system of
intense and stable rotating helical vortices that determined
the dynamics of the far-wake. The vortices in the wake
became unstable yielding a sinuousoid and ultimately broke
down by wake instability at y/d 5. The collapsed spiral wake
spread in all directions in the far-wake, making two counterrotating vortex pairs. Therefore, it was observed that the
gradual increase of TI in the far-wake is a consequence of
this ow structure caused by the wake instability.
(4) It was suggested by the authors that the boundary between
the near- and far-wake should be identied as the starting
location of the wake breakdown and in this work the wake
breakdown occurred at a distance of 5d downstream from the


wind turbine. This was followed by a large drop in the

maximum TI for the next downstream station.
(5) Compared with the WAsP predictions, a good agreement was
observed for the axial velocity magnitude in the far-wake
regions, after y/d 14. However, in the comparison of wake
width, a considerable difference due to the wall effect,
resulting from the numerical wind tunnel, was observed.
Therefore, further work is required using un-ducted ow. It
was also observed that the blockage was of minimum concern
when dealing with the velocity decits in the wake.
The information made possible due to this research will be
useful for wind farm designers to improve the overall wind farm
efciency and fatigue lives of the turbines by optimizing the wind
farm layout. In the future, investigations of the turbine wake
generation and propagation will be investigated in un-ducted
ow environment with atmospheric boundary layer velocity and
turbulence intensity proles. The results will be compared with
the results of the present research in order to further study wind
turbine wake aerodynamics.

The authors wish to gratefully acknowledge the nancial
support provided by the Centre for Energy Technology (CET)
and the School of Mechanical Engineering at the University of
Adelaide as well as the Korea Maritime University in Korea.
ABS, 2010. Australias Environment: Issues and Trends.
Barthelmie, R., Frandsen, S., Hansen, K., Schepers, J., Rados, K., Schlez, W., Neubert,
A., Jensen, L., Neckelmann, S., 2009. Modelling the Impact of Wakes on Power
Output at Nysted and Horns Rev.
Barthelmie, R., Larsen, G., Frandsen, S., Folkerts, L., Rados, K., Pryor, S., Lange, B.,
Schepers, G., 2006. Comparison of wake model simulations with offshore wind
turbine wake proles measured by sodar. Journal of Atmospheric and Oceanic
Technology 23, 888901.
Barthelmie, R.J., Frandsen, S.T., Rathmann, O., Hansen, K., Politis, E., Prospathopoulos, J., Cabezon, D., Rados, K., van der Pijl, S., Schepers, J., 2008. Flow and
Wakes in Large Wind Farms in Complex Terrain and Offshore.
Burton, T., 2001. Wind Energy: Handbook. John Wiley & Sons, Ltd.
Chamorro, L.P., Porte-Agel, F., 2009. A wind-tunnel investigation of wind-turbine
wakes: boundary-layer turbulence effects. Boundary-Layer Meteorology 132,
Corten, G.P., Nederland, E.C., 2001. Inviscid stall model. Netherlands Energy
Research Foundation.
Crespo, A., Hernandez, J., 1989. Numerical modelling of the ow eld in a wind
turbine wake. Forum on Turbulent Flows ASME FED 76, 121129.
Crespo, A., Manuel, F., Moreno, D., Fraga, E., Hernandez, J., 1985. Numerical
analysis of wind turbine wakes, in: Proceedings of the Delphi.
Fitzgerald, R.E., 2007. Wind Tunnel Blockage Corrections for Propellers. ProQuest.
Germano, M., Piomelli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgridscale
eddy viscosity model. Physics of Fluids A: Fluid Dynamics 3, 1760.
Glauert, H., 1933. Wind Tunnel Interference on Wings, Bodies and Airscrews.
Gomez-Elvira, R., Crespo, A., Migoya, E., Manuel, F., Hernandez, J., 2005. Anisotropy
of turbulence in wind turbine wakes. Journal of Wind Engineering and
Industrial Aerodynamics 93, 797814.
Gould, R., 1969. Wake blockage corrections in a closed wind tunnel for one or two
wall-mounted models subject to separated ow. HMSO.
Hand, M., Simms, D., 2001. Unsteady Aerodynamics Experiment.
Hinze, J., 1959. Turbulence, 1975. New York.
Jensen, N.O., 1983. A note on wind generator interaction. Ris National Laboratory.
Katic, I., Hjstrup, J., Jensen, N., 1986. A simple model for cluster efciency.
European Wind Energy Conference, Rome, 407409.
Kim, S.-E., 2004. Large eddy simulation using unstructured meshes and dynamic
subgrid-scale turbulence models. Technical Report AIAA-2004-2548. American
Institute of Aeronautics and Astronautics, 34th Fluid Dynamics Conference and
Kraichnan, R.H., 1970. Diffusion by a random velocity eld. Physics of Fluids 13,
Launder, B.E., Sandham, N.D., 2002. Closure Strategies for Turbulent and Transitional Flows. Cambridge University Press.


J.-O. Mo et al. / J. Wind Eng. Ind. Aerodyn. 112 (2013) 1124

Lilly, D., 1992. A proposed modication of the Germano subgridscale closure

method. Physics of Fluids A: Fluid Dynamics 4, 633.
Lissaman, P., Bate, E., 1977. Energy effectiveness of arrays of wind energy
conversion systems. AeroVironment Report AV FR, 7050.
Lissaman, P., W, G.G., Zalay, A.D., 1982. Numeric-modelling Sensitivity Analysis of
the Performance of Wind-turbine Arrays. AeroVironment Report AV FR.
Manwell, J.F., McGowan, J.G., Rogers, A.L., 2002. Wind Energy Explained. Wiley
Online Library.
Medici, D., Alfredsson, P.H., 2008. Measurements behind model wind turbines:
further evidence of wake meandering. Wind Energy 11, 211217.

Menter, F., Langtry, R., Volker,

S., 2006. Transition modelling for general purpose
CFD codes. Flow, Turbulence and Combustion 77, 277303.
Mo, J.O., Lee, Y.H., 2011. Numerical simulation for prediction of aerodynamic noise
characteristics on a HAWT of NREL phase VI. Journal of Mechanical Science
and Technology 25, 13411349.
Mortensen, N.G., Heatheld, D.N., Myllerup, L., Landberg, L., Rathmann, O., Troen,
I., Petersen, E., 2003. Getting Started With WAsP8. Ris National Laboratory.
Okulov, V.L., Sorensen, J., 2004a. Instability of a Vortex Wake Behind Wind
Turbines. Springer, pp. 772777.
Okulov, V.L., Sorensen, J.N., 2004b. Instability of the far wake behind a wind
turbine. Abs. 21th ICTAM-2004, Warsaw, Poland (http://ictam04. ippt. gov. pl/).
Potsdam, M.A., Mavriplis, D.J., 2009. Unstructured mesh CFD aerodynamic
analysis of the NREL Phase VI rotor. American Institute of Aeronautics and
Astronautics, 1801 Alexander Bell Drive, Suite 500, Reston, VA, 201914344,
Sanderse, B., 2009. Aerodynamics of wind turbine wakes. Energy Research Centre
of the Netherlands (ECN), ECN-E-09-016; Petten, The Netherlands, Technical
Simms, D.A., 2001. NREL unsteady aerodynamics experiment in the NASA-Ames
wind tunnel: A comparison of predictions to measurements. National Renewable Energy Laboratory.

Smirnov, A., Shi, S., Celik, I., 2001. Random ow generation technique for large
eddy simulations and particle-dynamics modeling. Journal of Fluids Engineering 123, 359.
Srensen, N., Michelsen, J., Schreck, S., 2002a. NavierStokes predictions of the
NREL phase VI rotor in the NASA Ames 80 ft  120 ft wind tunnel. Wind Energy
5, 151169.
Srensen, N.N., Michelsen, J.A., Schreck, S., 2002b. NavierStokes predictions of the
NREL phase VI rotor in the NASA Ames 80 ft  120 ft wind tunnel. Wind Energy
5, 151169.
Troen, I., Petersen, E.L., 1989. European wind atlas. Roskilde: Riso National
Laboratory, 1.
Vermeer, L., Srensen, J.N., Crespo, A., 2003. Wind turbine wake aerodynamics.
Progress in Aerospace Sciences 39, 467510.
Vermeulen, P., Builtjes, P., Vijge, J., 1981. Mathematical Modelling of Wake
Interaction in Wind Turbine Arrays, Part 1. Report TNO, 81-01473.
Wagner, S., Bareiss, R., Guidati, G., 1996. Wind Turbine Noise. Springer-Verlag,
Berlin/Heidelberg, Germany.
WEC, 2010. Clean Energy Australia 2010 Report, pp. 4953.
Wu, Y.T., Porte-Agel, F., 2001. Large-eddy simulation of wind-turbine wakes:
evaluation of turbine parametrisations. Boundary-Layer Meteorology 138(3),

Web references
/, (accessed 15.10.11).
/, (accessed 15.10.11).