J. Wind Eng. Ind. Aerodyn.

121 (2013) 2938

Contents lists available at ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

A calibration method for downwind wake models accounting

for the unsteady behaviour of the wind turbine tower shadow
behind monopile and truss towers
Marit Reiso n, Torbjrn Ruud Hagen 1, Michael Muskulus
Hgskoleringen 7a, Department of Civil and Transport Engineering, Norwegian University of Science and Technology, NTNU, NO-7491 Trondheim, Norway

art ic l e i nf o

a b s t r a c t

Article history:
Received 27 February 2013
Received in revised form
7 July 2013
Accepted 26 July 2013
Available online 21 August 2013

Traditionally wind turbines are built in the upwind conguration, but the alternative of a downwind
rotor has distinct advantages. A main issue with such a conguration is the tower shadow effect. The
presence of the tower generates a complex wind eld, consisting of an averaged velocity decit, unsteady
uctuations from vortex shedding processes and turbulence. Since this tower shadow is commonly
implemented using parametric steady wake models, the dynamic behaviour of the wake is not directly
accounted for. The present paper introduces a general method for calibrating the parametric steady wake
models and an effective turbulence intensity (accounting both for the velocity prole and the unsteady
effects) from computational uid dynamic simulations of the unsteady structural shadow.
To demonstrate its potential the method is used in a blade fatigue comparison study. A 15 percent
more exible and lighter blade for the downwind mounted rotors showed a decrease in blade fatigue
loads of three, four, and two percent compared to the conventional upwind mounted rotor on a monopile
tower for the monopile tower and truss towers at 0 and 22.5 degrees, respectively. This example shows
that the truss tower at 0 degree angle seems to result in the lowest blade fatigue loads.
The proposed calibration method is completely general and can be used to obtain efcient, effective
engineering models for the aerodynamic ow behind complex, multi-membered structures.
& 2013 Elsevier Ltd. All rights reserved.

Keywords:
Tower shadow
CFD
Powles model
Downwind
Truss tower
Blade fatigue

1. Introduction
Wind turbines are complex dynamic structures, which need
detailed computational models for their simulation and the necessary
iterations between design, analysis and optimization. Most of the
simulation technology available today is based on approximations that
have intrinsic limitations; for new technological concepts such as
downwind rotors, truss towers, etc., these approximations are not fully
able to represent a physically accurate model of the wind turbine.
Of particular interest here is the complex and unsteady aerodynamic
ow around wind turbine towers, which is, as of today, implemented
in commercial software codes based on a combination of time
averaged wake models and turbulent uctuations (Bossanyi, 2009;
Moriarty and Hansen, 2005). Most commonly used wake models are
varieties of Powles model (Powles, 1983).

Corresponding author. Tel.: +47 73 59 46 70; fax: +47 73 59 70 21

E-mail addresses: marit.reiso@ntnu.no (M. Reiso),
torbjorn.hagen@owectower.no (T.R. Hagen),
michael.muskulus@ntnu.no (M. Muskulus).
1
Current afliation: OWEC Tower AS, Storetveitvegen 96, NO-5072,
Bergen, Norway.
0167-6105/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jweia.2013.07.016

In the present study a downwind mounted rotor is considered.

It offers the potential of reducing the mean blade root apwise
bending moment, due to a possible coning of the blades (Reiso and
Moe, 2010). The risk of the blades hitting the tower will be
reduced, hence softer (and thereby cheaper and lighter) blades
can be used (Ahlstrm, 2006; Lee and Flay, 1999). Such blades have
also been shown to decrease the blade fatigue loading (Reiso and
Muskulus, 2013). The main downside of using a downwind
mounted rotor is the additional impact-like loading on the blades
as they pass through the tower wake.
To mitigate these loads, a truss tower has been proposed. Such
a tower consists of a number of cylindrical members that are
arranged in a complex geometric pattern (Fig. 1). A model scale
experiment by Reiso et al. (2011) of the wake behind a full truss
tower and a monopile tower seems to indeed favour the truss
tower, who exhibits a more narrow velocity decit and a lower
turbulence intensity in the wake.
The resulting ow eld depends on the spacing between the
cylinders and how they are arranged, e.g., side-by-side, tandem or in
staggered conguration. Above some critical spacing (that depends
on the cylinder arrangement), the ow pattern around the cylinders
can be assumed to behave as for a single cylinder. This is therefore an
excellent approximation for complex geometries under potential

30

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

Nomenclature
DEL

Blade root apwise bending moment

RFM

Damage equivalent load

ow (Krause and Muskulus, 2012), but also possible when involving

viscous effects. A lot of work has been conducted and reported on
such ows around and behind cylindrical structures (e.g., Blevins
(1990), Ishigai et al. (1972), Meneghini et al. (2001), Moe et al. (1993),
Gao et al. (2010), Zdravkovich (1977, 1997), Zdravkovich and Namork
(1979)), but not many authors have considered the ow behind
complex truss structures with diameters and spacing relevant for
wind turbine truss towers, at the relevant distances of approximately
25 tower diameters downstream of the tower.
For reliable simulation results based on parameterized steady
wake models, the parameters need to be carefully chosen (Coton
et al., 2002; Munduate et al., 2004; Wang and Coton, 2001). This can
be done by tting the parameters with data from full scale experiments, as in Wilmshurst et al. (1985), tting the parameters of
Powles model for wake width and mean velocity decit for a
monopile and a fairing tower. Hagen et al. (2011a, 2011b) demonstrated tting the parameters of Powles model to the mean velocity
prole from two dimensional computational uid dynamic (CFD)
simulations, thereby obtaining good agreement for the velocity
proles, both for a monopile tower and a truss tower.
Wind turbine loads do not solely result from the steady velocity
decit, but also the dynamics of the unsteady ow around the tower
plays an important role, especially for blade loads on downwind
mounted rotors. These variations in wind speed could in principle be
represented by three dimensional wind elds obtained using CFD
simulations of the tower shadow, but this would signicantly
increase the simulation time beyond the feasible, as 10 min time
series are required by the different standards, e.g., IEC 61400-3
(2009), for many different wind speeds.
The method that is proposed here, instead accounts for the tower
shadow (including the unsteady effects), by tting the parameters of

Powles model for the mean velocity eld and additionally calibrating
an effective value of turbulence intensity with a number of short CFD
simulations of two-dimensional cross sections through the tower. In
other words, two different tower shadow models are used: rst an
unsteady two dimensional CFD simulation, and then the simple
engineering model of Powles. The parameters of the second are
calibrated using the results of the rst. It should be noted that this
second tower shadow model results in a signicant simplication of
the ow eld: not only is the turbulence intensity assumed to be
constant (instead of varying temporally and spatially, as it does in the
CFD simulations), also the unsteady uctuations (i.e., periodic velocity
variations caused by vortex shedding behind tower members) are
approximated by an effective value of turbulence intensity. Moreover,
the application of Powles model to a multi-member tower is based
on additional assumptions (linear superposition of changes to the
mean velocity eld and scaling relationships for the wake parameters) that cannot completely accurately reproduce the mean ow
eld behind such a complex structure. This means that the second
wake model does not represent all features (e.g., frequency content)
of the actual ow, and the goal of the calibration process is therefore
to determine parameters that result in an optimal match of certain
response variables of interest. In this manuscript the blade fatigue
loading has been chosen, but the method can easily be adapted to
other outcome variables or combinations thereof, depending on the
intended application.
The major advantage of this method is that the results of this
calibration (parameters for the steady tower shadow and effective
values of turbulence intensity) are directly applicable in commercially available software for carrying out full wind turbine simulations. It is demonstrated that, compared to using time series
obtained from CFD simulations as input, this method captures

Uin

Uin

y
x

x
2.8D

2.8D

c)
a)

b)

Uin

K-brace
position
y
x

4m

10.8m

X-brace
position

z
x

2.8D
Fig. 1. CFD mesh and geometric conguration of the towers; (a) monopile, (b) X-braced truss tower at 0 degrees, (c) K-braced truss tower at 22.5 degrees and (d) crosssections where (a)(c) were calculated. The output data were collected at 2.8D (D 4.0 m, equal to monopile tower diameter) behind the tower centres. Inow direction, Uin,
is from the left (similar plots for K-braced and X-braced truss towers at 0 and 22.5 degrees, respectively).

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

the blade fatigue loading with excellent agreement for the rotor
mounted on the monopile tower, while the truss towers at 0 and
22.5 degrees deviates by one and three percent only, respectively,
which seems an acceptable compromise. We note in passing that
any inuence from the rotating blade on the tower wake has not
been accounted for, as is current practice in wind turbine analysis
(e.g., also for upwind turbines). Instead, the actual velocity
experienced by the blade sections is calculated through BladeElement-Momentum theory, using further assumptions on the
structure of the wind turbine wake (Burton, 2001). Further
research on the detailed interaction between rotor and tower
wakes would be very interesting, but lies beyond the scope of the
present manuscript.

31

Table 1
Physical (three dimensional) and CFD (two dimensional) tower properties for the
truss and monopile towers.
Parameter

Three dimensional physical tower properties:

Diameter [m]
Number of sections
Tower height [m]
Top distance between main legs [m]
Bottom distance between main legs [m]
Angle against horizontal [o]

Truss tower
(main leg/brace)

Monopile
tower

0.90/0.36
10
120
4.00/
28.0/
84.3/50.0

3.877.00

120

Two dimensional tower properties for CFD simulations:

Leg spacing [m]
10.80
Diameter [m]
0.90/0.36

4.00

2. Methods
The study is divided in two partsone part introducing the
method for calibrating the parametric tower shadow model and
the effective turbulence intensity. And a second part, demonstrating the application of the calibrated model in a blade fatigue
comparison study.
2.1. CFD model
The CFD model is based on earlier work and is described in
detail in Hagen et al. (2011a,b); for completeness, here the most
important features are summarized.
The commercial software package Ansys Fluent (Version 12.1.4;
Ansys Inc., Canonsburg, USA) was used to simulate the ow elds
around two-dimensional cross sections of wind turbine towers.
The different members (legs and braces) were approximated by
circular cross-sections for simplicity and easy meshing. The
following ve two-dimensional (horizontal plane) geometrical
congurations were considered (Fig. 1):

 monopile tower
 truss tower at 0 degrees angle (with respect to the inow




direction); close to X-brace position (later referred to as

X-braced truss tower at 0 degrees)
truss tower at 0 degrees angle; close to K-brace position (later
referred to as K-braced truss tower at 0 degrees)
truss tower at 22.5 degrees; close to X-brace position (later
referred to as X-braced truss tower at 22.5 degrees)
truss tower at 22.5 degrees; close to K-brace position (later
referred to as K-braced truss tower at 22.5 degrees)

The braces were modelled at two different positions (X- and

K-brace) in order to assess the inuence of cross-sectional geometry (i.e., height) on the results. The inow was considered at
both 0 and 22.5 degrees for the truss towers, to assess the
inuence of inow angle. Results for some other inow angles
can be found in Hagen et al. (2011a).
The distance between the main legs in the truss tower for the
two-dimensional CFD simulations was 10.8 m (Fig. 1). This corresponds to the vertical position in the tower where the part of the
blade experiencing the highest power production (about 2/3 along
the blade length from the root) will pass. The dimensions of the
physical (three-dimensional) and of the CFD model (two-dimensional) for the monopile tower and the truss tower are given in
Table 1, the full truss tower geometry is based on work by Long
and Moe (2007).
The computational domain spanned 17.5D times 10D in x- and y
direction, respectively (x being in the ow direction, and y being
transversal to the ow direction, Fig. 1). Here D4.0 m represents the
monopile tower diameter. Periodic boundary conditions were used for

the side boundaries (at y 720 m), and the outlet boundary was
implemented with a velocity reference pressure of 1.0 atm.
The free stream velocity at the inlet (x  5D) was 12.0 m/s,
representing a typical operational wind speed for power production
under close to maximum operational loads (rotor thrust). A turbulence intensity (dened as the ratio of root-mean-square velocity
uctuations to mean velocity) of 10 percent was used, being typical
for the offshore environment, and a turbulence length scale of 1.0 m
was used, roughly corresponding to the diameter of the larger
members in the truss tower (for which the major vortex shedding
process and subsequent generation of turbulence is expected). The
air density was taken to be 1.225 kg/m3, with a constant dynamic
viscosity of 1.789  10  5 kg/(ms), resulting in a kinematic viscosity of 1.5  10  5 m2/s.
The boundary layers around the members were modelled with
30 layers (inner element size 0.0001 m, with a growth factor of
1.2) inside an unstructured mesh that was embedded in an outer
structured mesh. The mesh was adaptive with quadrilateral cells
and was implemented using Gambit (Version 2.4.6; Ansys Inc.,
Canonsburg, USA).
The Reynolds-Averaged NavierStokes (RANS) equations were
complemented with the k shear-stress transport (SST) viscosity
model (Menter, 1994) and solved numerically. The time step was
xed at dt 0.005 s, which is both small enough to allow for the
resolution of vortex shedding, and to resolve up to 100 Hz
spectrally. Outputs were sampled 2.8D downstream of the geometric centre of the towers, corresponding to the approximate
position where the wind turbine blades will pass through the
ow eld.
The simulations were validated for the monopile tower by
considering the pressure coefcient around its circumference at
Reynolds number Re3:3  106 (Hagen et al., 2011a). Results
agreed well with the ndings of Warschauer and Leene (1971).
The drag coefcient was found to be 0.37, which corresponds to
what has been reported by Ong et al. (2009). The non-dimensional
size of the boundary layer y+ was used to check whether the
transition near the cylinder wall was accurately resolved. For
super-critical Reynolds numbers one should generally implement
y+ r5 (Salim and Cheah, 2009), which is fullled in the present
work with a y+
max of 3.5.

2.1.1. Post processing

Numerical results were recorded for N6000 time steps, corresponding to a ow time of 30 s. To ensure fully developed vortex
shedding before extracting the output data, 10,000 time steps were
simulated prior to writing the output (Nakayama et al., 2010).

32

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

Time series of local turbulence intensity (including turbulence from

the vortex shedding and unsteady motions) and mean velocity
proles were computed.

2.2. Aeroelastic model

Full wind turbine simulations were run in the commercial
software Bladed (version 4.2, GL Garrad Hassan). This software
offers the capability to use a three-dimensional wind velocity eld
(discretized on a regular grid) as input, intended for adding
turbulent uctuations, but here used for passing the complete
wind eld obtained in CFD simulations to the software (for the
rst wake model). Bladed also has the capability to use Powles
model for the mean velocity decit, also for multi-member towers.
In this case (second wake model) the wind velocity le is only
used to include turbulent uctuations (but with an effective value
of the turbulent intensity; cf. Section 2.3).
The wind turbine is based on the NREL Offshore 5-MW Baseline
Wind Turbine, hereafter referred as the NREL reference (Jonkman
et al., 2009). The wind turbine was used in its original conguration, as well as with some applied changes. It was adapted from an
upwind to a downwind rotor conguration, and some of the
simulations were run with a truss tower instead of the conventional monopile tower. In addition, shaft tilt and cone angles were
changed (Table 2). The model is based on a combined multi-body
representation and modal approach using beam elements, timedomain integration, and uses a standard PID controller. The output
time step was 0.04 s.
All simulations were run with a mean free stream velocity of
12.0 m/s and with the same wind shear prole, an exponential model
with a vertical shear exponent of 0.14. Turbulent uctuations were
added to the wind velocity le using realizations obtained from the
Kaimal spectrum (Table 3) by standard methods (Shinozuka, 1972). For
simplicity, mostly longitudinal correlations were used (with negligible
correlations for the crossow), but more realistic turbulence models
with full three-dimensional correlation structure (e.g., the Mann
model) can be used as well. In total eight different realizations
(random number seeds) were used to obtain more reliable results.
A total of 150 and 630 s were simulated, respectively, for use in the
calibration of the steady wake model and for the blade fatigue
comparison study, respectively, with the rst 30 s removed due to
transients. For the blade fatigue comparison study, this resulted in
time series with a length of 10 min, as required by the IEC 61400-3
standard (IEC, 2009).
The tower shadow was implemented both with the unsteady
ow eld variations from the CFD simulations (Section 2.3.2), and
with the combined Powles model available in Bladed (Section 2.3.1).
Damage equivalent loads (DEL) for the blade root apwise
bending moments (RFM) were calculated. DEL was calculated from
Miners rule by way of a rainow count of the blade loading time
histories. An inverse SN slope m 10 and 128 bins for the cycle
range were used. The equivalent loads were calculated for a
reference frequency of 0.003169 Hz, corresponding to 2  106
cycles during the total lifetime of 20 years.

Table 3
Kaimal turbulence model parameters.
Width of wind eld [m]
Lateral number of points (calibration/implementation) [-]
Height of wind eld [m]
Vertical number of points [-]
Simulation time (calibration/implementation) [s]
Lateral spacing (calibration/implementation) [m]
Vertical spacing [m]
Longitudinal spacing [m]
Mean wind speed [m/s]
Turbulence length scale:
Longitudinal [m]
Transversal [m]
Vertical [m]

150
300/50
200
20
150/630
0.5/3.0
10.0
0.48
12.0
340.2
0.1
0.1

2.3. Tower shadow

The effect of the tower shadow is usually accounted for in full
wind turbine simulation software by a constant (but spatially
varying) velocity decit behind the tower, by way of a parameterized model. Here, both the steady model suggested by Powles
(1983) and an unsteady CFD simulation of the tower shadow were
used. Fitting of the models and construction of wind eld les has
been performed in MatLab (The MathWorks; Version R2012a).
2.3.1. Steady tower shadow model
In Powles model the wind velocity U behind a circular member is
described in terms of changes to the mean velocity U0. The model is
specied in terms of two non-dimensional parameters: (1) velocity
decit () at the centre of the wake, and (2) wake width (w). Both
parameters are here taken in terms of the member diameter. Recall
that y denotes the distance transversal to the free stream velocity and
D the tower member diameter, the velocity is given by:





y
Ux; y U 0 1x cos 2
1
wxD
The model is limited to at most one period of the cosine term,
and is usually extended laterally with the potential ow solution
for arguments of the cosine outside of 7 60 degrees (Bossanyi,
2009). The parameters and w depend on the longitudinal
distance x from the centre of the tower and usually need to be
estimated individually for each transversal.
Powles model is therefore often supplemented by the following square-root-law in terms of values r and wr for some
(relative) reference distance xr (Bossanyi, 2009):
p
x r = x=xr D
p
wx wr  x=xr D
2
This relationship can be theoretically motivated from simple
considerations of diffusion processes, and seems to hold well in
practice (Powles, 1983). It forms the basis of extending Powles model
from a single cylinder to a multi-member truss tower. As of today,
Bladed is the only commercial software that allows for using a tower
shadow model for such a multi-member truss type tower. This is
achieved by superposition of the solutions for each member:
n

Ux; y U i x; y

Table 2
NREL reference properties and applied changes.

i1

Parameter

NREL reference

Changes to NREL reference

Rotor orientation
Shaft tilt [degrees]
Cone angle [degrees]
Tower

Upwind
5.0
2.5
Monopile

Downwind
2.0a
2.0a
Truss and monopilea

Only for downwind rotor simulations.

1nU 0 U 0  xxi cos 2

in

yyi
wxxi D


3

where Ui (i1, 2, , n) represents the velocity eld calculated because

of the presence of the ith member, and xi and yi represent the
coordinates of the centre point of the ith member. This approach is
only a rough approximation of the actual ow eld, in particular, not
all individual wakes of the different members can be equally well

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

represented using just these two reference parameters. Parameter

tting has been done with the least-squares approach, minimizing the
root-mean-square error (RMSE) of the mean velocity prole across the
whole section. This estimation was performed for each tower prole
individually and also globally for the truss tower by minimizing the
maximum RMSE over all four congurations simultaneously (truss
tower at 0 and 22.5 degrees at both X- and K-brace position).

2.3.2. Unsteady tower shadow model

The time series of the tower shadow ow eld obtained from the
unsteady CFD simulations were collected in a three-dimensional
wind velocity le for use with Bladed software. Since the simulations
were only two-dimensional, vertical variations needed to be arti-

33

cially introduced in order to obtain a three dimensional wind eld.

To avoiding full correlation, the CFD results were shifted by a random
offset (in time) for each vertical layer, totally making up 20 layers.
Both instantaneous wind velocity (in fact, its ensemble average, since
RANS simulations were performed) and turbulent kinetic energy
(TKE) were used. The time series of the rst resulted in a threedimensional discrete wind velocity le. To this the variations from
parameterized (subgrid) turbulence were added separately, by calculating the instantaneous standard deviation of turbulent velocity
uctuations from the TKE (assuming equi-distribution over all
degrees of freedom), and using these values to scale a threedimensional zero mean, unit standard-deviation realization from
the Kaimal spectrum. Adding together these two signals resulted in
the nal three-dimensional discrete wind le that was used in full

Step 1
Lateral
mesh
refinement
Sub-grid
turbulence

Turbulence
spectra

CFD
K-brace
CFD
X-brace

2D-->3D
Mean velocity
deficit and vortices

CFD

Bladed simulations

Step 2
Step 1 + 2
Wind field and
tower shadow
Turbulence
spectra

Lateral
mesh
refinement

Turbulence
spectra
CFD

K-brace
Fitting Powles
model

Step 3

X-brace
Fitted Powles
model

Fitted Powles
model and TI

Fig. 2. Flow chart of the method for calibrating Powles model parameters and the turbulence intensity.

Step 3

34

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

wind turbine simulations. The method cannot currently represent

explicit vertical correlations in wind velocity (e.g., correlated vortex
shedding), but the added turbulent variations can partially account
for this, since these are based on a fully three-dimensional turbulence model with preserved correlation structure.
As the CFD simulations were limited to 30 s (again, due to
constraints on computational time), whereof the rst 5 s were
removed due to transients, the results of the CFD simulations were
recycled six times. In principle the cyclic nature of the tower
shadow wind eld could result in biased results for, e.g., blade
fatigue loading, but since the tower shadow is superposed with
turbulent uctuations and additionally averages out vertically
along the blades, this seems unlikely.
The simulations were carried out using a grid resolution of
300  20 points across the lateral and vertical domain of
150 m  200 m, at a longitudinal resolution of 25 Hz, which
has been found sufcient for such structures (Reiso and
Muskulus, 2012). The turbulent uctuations from the Kaimal
spectrum had a maximum resolution of 50 points across the
150 m lateral domain (a constraint of the software used to
generate these three-dimensional correlated elds), and therefore linear interpolation was used to obtain values for all 300
points used for representing the tower shadow. It should be
noted that this introduces certain articial changes in the
correlation structure of the turbulent uctuations, but these
are thought to be of minor importance. The nal wind eld les
were again normalized to unit standard deviation before being
imported to the Bladed software for full wind turbine simulations (where the normalization is again removed, through
proper declaration of the wind le standard deviation). With
the tower shadow present in the wind eld le, the tower
shadow module in Bladed is disabled.

2.4. Blade fatigue comparison study

The second part of the study is a blade fatigue comparison
study. Full wind turbine simulations were performed using both
the original NREL reference blade and an adjusted NREL blade.
The adjusted NREL blade is a 15 percent more exible and 15
percent lighter blade compared to the NREL reference blade. The
value of 15 percent was chosen as it has shown good performance in power production, as well as a decrease in blade
fatigue loading behind both a monopile and a fairing tower
(Reiso and Muskulus, 2013). This is a simplied adjustment,
where the blade geometry is kept the same as that of the NREL
reference. As both the stiffness and weight were decreased by
the same amount, the blade eigenfrequency stayed intact
(Ahlstrm, 2006). The NREL reference blade was used for both
the upwind and downwind rotor congurations, while the
adjusted NREL blade was implemented for the downwind
congurations only.
The tower shadow and tower dam (upwind rotor) effects were
implemented using the inbuilt capabilities of Bladed software,
with the physical tower geometries given in Table 1. For the
downwind congurations, Powles tower shadow model was used.
The conventional upwind mounted rotor on a monopile tower
was included for reference and the tower dam effect was derived
assuming potential ow (correction factor set to 1), which is the
current standard in wind turbine simulations (Bossanyi, 2009).
With the Bladed software a limitation of 50 lateral points across
the domain exist, this maximum resolution was used (lateral
domain size of 150 m), with a vertical and longitudinal resolution
of 20 points (across a vertical domain of 200 m) and 25 Hz
(Table 3).

3. Results
3.1. Calibration method for the steady Powles wake model
and the turbulence intensity
It should rst be mentioned that the velocities at the boundaries of the CFD domain (at approximately y/D 7 5, Fig. 3)
overestimated the free stream velocity of 12 m/s with a mean
offset of 2.5 percent for the monopile tower, an overestimation
that was present also for the truss tower congurations (with a
3.0 percent overestimation, Fig. 3). This overestimation originates
from the limited size of the computational domain, and is a
numerical artefact due to the periodic boundary conditions, but
does not affect the validity of this approach. If used in an industrial
application, a larger domain will lead to slightly more accurate
parameter estimates.
The calibration method consists of three distinct steps (Fig. 2).
Step 1 is the reference case. A three dimensional grid is constructed from the velocity elds obtained in the two dimensional
CFD simulations, to which random realizations of the turbulence
from the Kaimal spectrum are added to account for the additional
sub-grid turbulence (Section 2.3.2).
Step 2 is where the tting of the Powles model parameters and
turbulence intensity is done. The two parameters, the reference
velocity decit (r) and wake width (wr) parameter, were tted at
a reference length xr (Eq. (2)), at 2.8 monopile diameters
(D 4.0 m) downstream the tower.
The mean velocity prole represented by Powles model for the
monopile tower (Fig. 3a), reproduces the CFD prole excellently,
with reference velocity decit and wake width parameters of
r 0.218 and wr 1.804, respectively (Table 4). The only discrepancy was a slight underestimation at transversal position 72D.
For the truss tower arrangements some discrepancies appeared.
The central velocity decit behind the X-braced truss tower at
0 degrees underestimated the CFD results by roughly 25 percent for
the individually tted parameters (Fig. 3b) due to solidication (more
ow going around the sides of the tower than predicted by the
individual wake models for each member). For the globally tted
parameters the underestimation of the dip was even larger, around 31
percent. Powles model for the K-braced truss tower at 0 degrees
represented the results of the CFD simulations better than for the
X-brace, with discrepancies at the two dips (at transversal position
71.5D) of 2 and 18 percent for the individually and globally tted
values, respectively (Fig. 3c).
For the truss tower at 22.5 degrees, Powles model for the
K-brace showed the largest discrepancy from the CFD simulations,
with a maximum underprediction of the mean velocity prole (at
transversal position  1.5D) of 14 and 21 percent for the individually and globally tted parameters, respectively. The maximum
underprediction for the X-brace at transversal position -0.5D was
10 and 14 percent, respectively (Fig. 3c).
The reference wake width parameters for all truss tower congurations were larger than that of the monopile tower (w1.804,
Fig. 3a), with a broad scatter from slightly higher for the K-braced
truss tower at 22.5 degrees (w1.813, Table 4), to 27 percent higher
for the X-braced truss tower at 0 degrees (w2.295, Fig. 3b). The
globally tted value was six percent higher than for the monopile
tower, w1.914. For the reference velocity decit, only the two
K-brace congurations showed a larger velocity decit than the
monopile tower ( 0.218), approximately 1213 percent higher
( 0.241 and 0.249, respectively). Here the globally tted value
was 15 percent lower ( 0.185) than the reference velocity decit
parameter for the monopile tower.
Some of these differences can be accounted for by calibrating
the effective turbulence intensities (TI) used with the (globally)
tted Powles model. Full wind turbine simulations were carried

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

35

Fig. 3. Normalized (with respect to free stream velocity) mean wind velocity proles at reference position 2.8D (D 4.0 m) behind the towers (dips corresponding to the
approximate position of the tower legs and braces); (a) monopile, (b) X-braced truss tower at 0 degrees, (c) X-braced truss tower at 22.5 degrees (K-braced truss towers not
shown). For the truss towers, both individual (broken lines) and global ts (dotted lines) of the combined Powles model are shown (parameter values given in Table 4).
Table 4
Individual and global tted Powles model parameters for the monopile tower and
the X-braced and K-braced truss towers at 0 and 22.5 degree angle towards the
incoming wind direction.
Tower geometry

Monopile
X-braced truss tower
at 0 degrees
K-braced truss tower
at 0 degrees
X-braced truss tower
at 22.5 degrees
K-braced truss tower
at 22.5 degrees
a

Individually tted valuesa

Globally tted valuesa

Velocity
decit ()

Wake
width (w)

Velocity
decit ()

Wake
width (w)

0.218
0.197

1.804
2.295

0.185

1.914

0.241

1.995

0.185

1.914

0.217

2.027

0.185

1.914

0.249

1.813

0.185

1.914

Reference position 2.8D (D 4.0 m).

out using the tower shadow both from the globally tted Powles
model (second wake model; using four different trial values for the
TI) and the CFD simulations (rst wake model). For the calibration
of TI, Powles model was used for the same cross sectional
geometries as for the CFD simulations, i.e., for a monopile tower
with constant diameter, D 4.0 m, and for truss towers with
constant main leg spacing of 10.8 m. Two different setups,
X- and K-brace, were run with each of the two truss tower
arrangements, 0 and 22.5 degrees, respectively.

The effective TI to be used with Powles model was obtained

from plots of the response variable of interest (blade fatigue
loading) against TI, as the value where the curve for the CFD
tower shadow model intercepted the trend-line through the
results for the four different TI trial values used with the second
wake model. An interesting nding is the linear shape of the
trend-line (R2 E1) between the DEL RFM for the different TIs. This
relationship is well known theoretically and simplies the calibration procedure signicantly.
The lowest effective turbulence intensity was found for the
monopile tower, 8.11 percent, Fig. 4a. For the truss tower at 0 and
22.5 degrees, TI values of 8.23 and 8.43 percent were found,
respectively (Fig. 4b and c). The respective root mean square errors
for the X-brace and K-brace of the two tower angles were 0.10
and 0.37.
Simulations run with the calibrated turbulence intensities
showed a deviation of the DEL of 71.0 and 7 3.0 percent for
the truss tower at 0 and 22.5 degrees, respectively (Table 5).
Step 3 is using the tower shadow model in Bladed software
with the values for the wake width parameter, velocity decit
parameter and the TI from the second step.
3.2. Blade fatigue comparison study
Two sets of cases were studied, either using the full wind elds as
described above, or, additionally, only using the mean wind proles.
The damage equivalent loads (DEL) for the blade root apwise
moment (RFM) in the standard cases were signicantly higher

36

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

Powles
CFD

6.5
DEL RFM [MNm]

DEL RFM [MNm]

6.5

6.0

5.5

7.0

7.5

8.0

8.5

9.0

9.5

10.0

6.0

Powles Kbrace
CFD Kbrace
Powles Xbrace
CFD Xbrace

5.5

5.0

7.0

Turbulence Intensity, TI [%]

8.0

8.5

9.0

9.5

10.0

Turbulence Intensity, TI [%]

Powles Kbrace
CFD Kbrace
Powles Xbrace
CFD Xbrace

6.5
DEL RFM [MNm]

7.5

6.0

5.5

5.0

7.0

7.5

8.0

8.5

9.0

9.5

10.0

Turbulence Intensity, TI [%]

Fig. 4. Damage equivalent loads (DEL) for the blade root apwise bending moment (RFM) at different turbulence intensities (TI); (a) monopile, (b) truss tower at 0 degrees
and (c) truss tower at 22.5 degrees. Curves are shown for the CFD tower shadow at constant TI and Powles tower shadow model with four different values of the TI.

Table 5
Damage equivalent loads (DEL) for the blade root apwise bending moment (RFM)
obtained from the CFD simulations and the Powles model, latter using the
calibrated turbulence intensities.

Monopile
K-braced truss tower
at 0 degrees
X-braced truss tower
at 0 degrees
K-braced truss tower
at 22.5 degrees
X-braced truss tower
at 22.5 degrees

CFD [MN m]

Powles model with calibrated

TI [MN m]

5.46
5.81

5.45
5.85

5.80

5.76

5.78

5.60

5.42

5.58

(by approximately 20 percent) than for the steady cases with only the
mean tower shadow proles (Fig. 5). The latter feature fewer uctuations in the ow behind the towers, and hence resulted in reduced
DEL RFM. For comparison, the conventional upwind mounted rotor on
a monopile tower was included (using the potential ow model), and
showed a similar difference, but with a 40 percent increase in DEL
RFM from the steady to the standard wind case.
The steady wind simulations do not allow for realistic blade
fatigue estimates, but were performed in order to isolate and judge
the relative effect of the turbulent uctuations compared to the mean
tower shadow. The turbulent wind leads to additional smoothing of
the tower shadow effect from the mean velocity prole, and leads to
smaller relative differences in DEL RFM between the upwind and
downwind rotor congurations in the standard wind cases (deviating
by a maximum of 7 percent, Fig. 5), compared to the steady wind
cases (deviating by 1127 percent, Fig. 5).
In addition, with the smaller difference in DEL RFM between
the upwind and downwind rotor congurations in the standard
wind cases, the adjusted NREL blades resulted in a decrease in DEL

RFM compared to the upwind mounted rotor by 3, 4 and 2 percent

for the monopile tower and truss towers at 0 and 22.5 degrees,
respectively. This is a different nding than for the steady wind
cases, where an increase was found for the downwind mounted
rotors, of 11, 13 and 14 percent for the monopile and truss towers
at 0 and 22.5 degrees, respectively.
The lowest DEL RFM for the downwind rotors in the turbulent
wind cases was found for the exible blade on the truss tower at
0 degrees, while the steady wind cases showed the lowest DEL
RFM for the exible blade on the monopile tower.

4. Discussion
4.1. Calibrated steady wake model
Intuitively one could think that the monopile tower (having the
largest apparent solidity) would be the tower with the largest
velocity decit. But the largest velocity decit was found behind
the two dimensional K-braced truss tower at 0 degrees (well
reproduced by the individually tted Powles model). The large
velocity decit can be explained from the interaction between the
closely spaced main leg and brace (centre to centre distance of one
meter) in the K-brace conguration. This agrees with literature,
such as Zdravkovich (1977), Gao et al. (2010) and Blevins (1990),
where both interaction effects and increased velocity decits
(compared to two single cylinders that do not interact) were
reported for closely spaced cylinders. With the present tower
aligned at 0 degrees, two and two K-brace congurations were
arranged in tandem, and the ow eld behind the rst K-brace
would have insufciently recovered (Powles, 1983) before reaching the second K-brace, further increasing the total velocity decit
behind the K-braced truss tower at 0 degrees.
Powles model signicantly underestimated the central dip in
the X-braced truss tower at 0 degrees (25 and 31 percent for the

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

Turbulent wind

DEL RFM [MNm]

5.0

Steady wind

4.5

4.0

3.5

37

UW monopile
DW monopile (NREL blade)
DW truss 0 (NREL blade)
DW truss 22.5 (NREL blade)
DW monopile (adjusted blade)
DW truss 0 (adjusted blade)
DW truss 22.5 (adjusted blade)
UW monopile
DW monopile (NREL blade)
DW truss 0 (NREL blade)
DW truss 22.5 (NREL blade)
DW monopile (adjusted blade)
DW truss 0 (adjusted blade)
DW truss 22.5 (adjusted blade)

Fig. 5. Damage equivalent loads (DEL) for blade root apwise bending moment (RFM) for upwind (UW) and downwind (DW) mounted rotors on a monopile tower and a
truss tower; the latter both aligned at 0 and 22.5 degrees with respect to the inow direction. Downwind rotors were simulated with two different blade exibilities and
weights. Steady wind and turbulent wind refer to two sets of distinct cases, where either only the mean wind proles were used, or where the complete wind elds
were used.

individually and globally tted parameters). As the model uses

simple superposition of the ow elds for the individual truss
tower members (Eq. (3)), the interaction effects between the brace
members were not accounted for, which would have given a larger
velocity decit, as shown by, e.g., Gao et al. (2010).
The X- and K-braces at 22.5 degrees represent the conguration
more frequently encountered in practice, where the inow is not
aligned with the structure, and these results should therefore be
considered more relevant. Also the discrepancies between Powles
model and the CFD results were smaller at 22.5 degrees compared
to the 0 degree truss tower. The staggered arrangement of the
truss tower members at 22.5 degrees exhibited an increased
velocity recovery, as fewer tandem arrangements were present
compared to the truss tower at 0 degrees, this also agrees with,
e.g., Ishigai et al. (1972).
Using one set of wake parameters, i.e., the globally tted
parameters for Powles model (Fig. 3), raises the question of the
validity of the model, as its predictions deviate to some extent
from the CFD proles. Regarding the variable of interest in the
application example (see below), using the calibrated turbulence
intensity resulted in small deviations of one and three percent in
response for the truss tower at 0 and 22.5 degrees, respectively,
which seems an acceptable result. If an even more accurate result
is required, the parameters may be tted individually for each
truss tower angle with respect to the incoming wind direction.
Although a trivial modication of the procedure, this will only be
practically applicable if commercial software is updated accordingly. In general, the intention of our approach is not to realistically reproduce all features of the ow eld behind the structure,
but to obtain an efcient model.
It should also be noted that unsteady CFD simulations with
fully turbulent inow would result in additional turbulent mixing
behind the tower structures (Hagen et al., 2011a). And this, to a
certain extent, is expected to reduce the structured vortex shedding compared to the present study. This could additionally
benet the blades in a downwind mounted rotor.
4.2. Blade fatigue comparison study
A study by Reiso and Muskulus (2012) showed that the main
contribution to blade fatigue loads on downwind mounted rotors
resulted from the mean velocity decit and turbulence, with
smaller contributions from unsteady effects originating from the
vortex shedding. The dominant inuence of turbulence was also
seen in this study, with a signicant increase of 40 and 20 percent
in blade fatigue loads for the up- and downwind mounted rotors,

respectively, compared to a steady wind eld. However, for an

accurate representation of the tower shadow for a multi-member
structure and more reliable simulation results, the deviations from
Powles model (used with square-root-law scaling in the simple
superposition approach) due to interaction effects as well as the
additional uctuations from the vortex shedding need to be
accounted for. These effects can be approximately included in
the simulations by adjusting the turbulence intensity.
The largest blade fatigue loads for the downwind mounted
rotors were reported for the blade behind the truss tower at 22.5
degrees angle with the incoming wind direction (for both steady
and turbulent wind cases), despite the fact that the velocity decit
was approximately 62 percent larger and the wake width only 17
percent lower for the truss tower at 0 degrees (Fig. 3). A possible
explanation for this could again be the dominating inuence of
turbulent uctuations, effectively smoothing out the effect of the
deeper velocity decits in the more narrow velocity proles of the
truss tower at 0 degrees to a larger extent than for the wider truss
tower at 22.5 degrees. This would also explain why the lowest
blade fatigue loads were found behind the monopile tower in the
steady wind cases, while they were found for the truss tower at
0 degrees in the turbulent wind cases.
The method was used for three dimensional tapered towers,
showing a decrease in blade fatigue loads when using 15 percent
more exible and 15 percent lighter blades of three, four and two
percent for the monopile tower and the truss tower at 0 and 22.5
degrees, respectively, compared to the upwind mounted rotor.

5. Conclusion
A novel method for tting Powles model and calibrating the
value of the effective turbulence intensity with CFD simulations
was introduced. It improves the reliability of wind turbine simulations for complex, multi-membered towers, and the method is
directly applicable in commercial software. Time consuming CFD
simulations, which could be alternatively used to calculate the
wind eld inputs for the turbine simulations, are reduced to a few
two-dimensional studies for cross sections of the relevant tower
geometries. The mean velocity proles from the CFD simulations
are then used to t the velocity decit and wake width parameters
of Powles model, while the unsteady and turbulent behaviour is
accounted for through an effective value of turbulence intensity
(with respect to an outcome variable of interest, which here was
damage equivalent structural loading). Some discrepancies between
the mean velocity proles obtained from the CFD simulations and

38

M. Reiso et al. / J. Wind Eng. Ind. Aerodyn. 121 (2013) 2938

Powles model occur under the linear superposition of the different

ow elds, and this can be accounted for, to a large extent, through
adjusting the turbulence intensity.
The method is general and could also be used for other steady
wake models, e.g., the ones suggested by Blevins (1990),
Schlichting and Gersten (2000) and the JET wake model by
Madsen et al. (2007), as well as for other structures where it is
important to account for the unsteady and turbulent behaviour of
the wind eld behind a structure.
It is a very interesting question how far the calibrated models
can extrapolate also other ow regimes (e.g., how the model
parameters will change for different wind speeds, different controllers, or under structural modications). If such changes are
small, then the method will be extremely useful, avoiding recalibration, i.e., many additional CFD simulations. However, this is
outside of the present scope, and has been left for future studies.
Acknowledgements
The work of Marit Reiso was nanced by the Research Council
of Norway (NFR) through the project Offshore Wind Energy in
Norway: Setting the Basis, contract no. 186952/I30.
References
Ahlstrm, A., 2006. Inuence of wind turbine exibility on loads and power
production. Journal of Wind Energy 9, 237249.
Blevins, R.D., 1990. Flow-induced Vibration. Van Nostrand Reinhold, New York.
Bossanyi, E.A., 2009. GH Bladed Theory Manual. Technical Report 282/BR/009.
Garrad Hassan and Partners Limited, Bristol.
Burton, T., 2001. Wind Energy: Handbook. J. Wiley.
Coton, F.N., Wang, T., Galbraith, R.A.M., 2002. An examination of key aerodynamic
modelling issues raised by the NREL blind comparison. Wind Energy 5,
199212.
Gao, Y., Yu, D., Tan, S., Wang, X., Hao, Z., 2010. Experimental study on the near wake
behind two side-by-side cylinders of unequal diameters. Fluid Dynamics
Research 42, 113.
Hagen, T.R., Reiso, M., Muskulus, M., 2011a. Numerical analysis of turbulent ow past a
truss tower for offshore downwind turbines. In: Proceedings of the International
Offshore and Polar Engineering Conference, ISOPE, Maui, pp. 319326.
Hagen, T.R., Reiso, M., Muskulus, M., 2011b. Numerical tower shadow modeling for
a downwind wind turbine truss tower. In: Proceedings of the International
Conference on Offshore Mechanics and Arctic Engineering OMAE, Rotterdam,
pp. 859870.
IEC, 2009. Wind TurbinesPart 3: Design Requirements for Offshore Wind
Turbines. Technical Report IEC 61400-3. International Electrotechnical Commission, IEC, Geneva.
Ishigai, S., Nishikawa, E., Nishimura, K., Cho, K., 1972. Experimental study on
structure of gas ow in tube banks with tube axes normal to ow. I. Karman
vortex ow from two tubes at various spacings. Bulletin of the Japan Society of
Mechanical Engineers, JSME 15, 949956.
Jonkman, J., Buttereld, S., Musial, W., Scott, G., 2009. Denition of a 5-MW
Reference Wind Turbine for Offshore System Development. Technical Report
NREL/TP-500-38060, Golden.
Krause, L., Muskulus, M., 2012. Modeling of tower inuence for a full-height truss
tower wind turbine with the source panel method. In: Proceedings of the
International Offshore and Polar Engineering Conference ISOPE, Rhodes,
pp. 417422.

Lee, A.T., Flay, R.G.J., 1999. Compliant blades for wind turbines. IPENZ Transactions
26, 712.
Long, H., Moe, G., 2007. Truss type towers in offshore wind turbines. In: Proceedings of the European Offshore Wind Conference & Exhibition., EWEA, Berlin.
Madsen, H.A., Johansen, J., Srensen, N.N., Larsen, G.C., Hansen, M.H., 2007.
Simulation of low frequency noise from a downwind wind turbine rotor,
Collect, Tech. Pap. Aeros. Sci. Meet. AIAA, Reno, pp. 75497560.
Meneghini, J.R., Saltara, F., Siqueira, C.L.R., Ferrari, J.A., 2001. Simulation of ow
interference between two circular cylinders in tandem and side-by-side
arrangements. Journal of Fluids and Structures 15, 327350.
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineering
applications. American Institute of Aeronautics and Astronautics 32, 15981605.
Moe, G., Domben, T., Steen, P.E., 1993. Vibrations of a circular cylinder in the wake of
a larger cylinder. In: Proceedings of the Second European Conference on
Structural Dynamics: EURODYN. Publ by A.A. Balkema, Trondheim, pp. 1083
1083.
Moriarty, P.J., Hansen, A.C., 2005. AeroDyn Theory Manual. Technical Report NREL/
EL-500-36881, Golden.
Munduate, X., Coton, F.N., Galbraith, R.A.M., 2004. An investigation of the aerodynamic response of a wind turbine blade to tower shadow. Journal of Solar
Energy Engineering 126, 10341040.
Nakayama, A., Okamoto, D., Takeda, H., 2010. Large-eddy simulation of ows past
complex truss structures. Journal of Wind Engineering and Industrial Aerodynamics 98, 133144.
Ong, M.C., Utnes, T., Holmedal, L.E., Myrhaug, D., Pettersen, B., 2009. Numerical
simulation of ow around a smooth circular cylinder at very high Reynolds
numbers. Marine Structures 22, 142153.
Powles, S.R.J., 1983. The effects of tower shadow on the dynamics of a horizontalaxis wind turbine. Journal of Wind Engineering 7, 2642.
Reiso, M., Moe, G., 2010. Blade response on offshore bottom xed wind turbines
with down-wind rotors. In: Proceedings of the International Conference on
Offshore Mechanics and Arctic Engineering OMAE, Shanghai, pp. 499504.
Reiso, M., Muskulus, M., 2012. Resolution of Tower Shadow Models for Downwind
Mounted Rotors and its Effects on the Blade Fatigue. Unpublished results.
Reiso, M., Muskulus, M., 2013. The simultaneous effect of a fairing tower and
increased blade exibility on a downwind mounted rotor, Journal of Renewable
and Sustainable Energy 5, 033106.
Reiso, M., Muskulus, M., Moe, G., 2011. Tower shadowexperiment comparing
wake behind tubular and truss towers. In: Proceedings of the International
Offshore and Polar Engineering Conference ISOPE, Maui, pp. 335341.
Salim, S.M., Cheah, S.C., 2009. Wall y+ strategy for dealing with wall-bounded
turbulent ows. In: International MultiConference of Engineers and Computer
Scientists 2009. IMECS 2009. Newswood Limited, Hong Kong, pp. 2165-2170.
Schlichting, H., Gersten, K., 2000. Boundary-Layer Theory. Springer.
Shinozuka, M., 1972. Monte Carlo solution of structural dynamics. United States,
51p.
Wang, T., Coton, F.N., 2001. A high resolution tower shadow model for downwind
wind turbines. Journal of Wind Engineering and Industrial Aerodynamics 89,
873892.
Warschauer, K.A., Leene, J.A., 1971. Experiments on mean and uctuating pressures
of circular cylinders at cross ow at very high Reynolds numbers. In: Proceedings of the Third International Conference on Wind Effects Building and
Structures, pp. 305315.
Wilmshurst, S.M.B., Powles, S.J.R., Wilson, D.M.A., 1985. The problem of tower
shadow. In: Proceedings of the Wind Energy Conversion. BWEA, Oxford,
pp. 95102.
Zdravkovich, M.M., 1977. Review of ow interference between two circular
cylinders in various arrangements. Journal of Fluids Engineering-Transactions
of the ASME 99, 618633.
Zdravkovich, M.M., 1997. Flow Around Circular Cylinders. vol. 1. Oxford University
Press, New York. (Fundamentals).
Zdravkovich, M.M., Namork, J.E., 1979. Structure of interstitial ow between closely
spaced tubes in staggered array. In: Flow-Induced Vibration, Symptoms National
Congress on Pressure Vessel and Piping Technology, San Francisco, pp. 4146.