ARTICLE IN PRESS

Progress in Aerospace Sciences 42 (2006) 285–330

www.elsevier.com/locate/paerosci

State of the art in wind turbine aerodynamics and aeroelasticity

M.O.L. Hansena,, J.N. Sørensena, S. Voutsinasb, N. Sørensenc,d, H.Aa. Madsenc
a
Department of Mechanical Engineering, Technical University of Denmark, Fluids Section, Nils Koppels Alle,
Building 403, DK-2800 Lyngby, Denmark
b
Department of Mechanical Engineering, National Technical University of Athens, Fluids Section, 15780 Zografou, Greece
c
Wind Energy Department, Risoe National Laboratory, Building VEA-762, P.O. Box 49, Frederiksborgvej 399, DK-4000 Roskilde, Denmark
d
Department of Civil Engineering, Alborg University, Sohngaardsholmsvej 57, DK 9000 Aalborg, Denmark
Available online 29 December 2006

Abstract

A comprehensive review of wind turbine aeroelasticity is given. The aerodynamic part starts with the simple
aerodynamic Blade Element Momentum Method and ends with giving a review of the work done applying CFD on wind
turbine rotors. In between is explained some methods of intermediate complexity such as vortex and panel methods. Also
the different approaches to structural modelling of wind turbines are addressed. Finally, the coupling between the
aerodynamic and structural modelling is shown in terms of possible instabilities and some examples.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Aeroelasticity; Wind turbines

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
2. Predicting aerodynamic loads on a wind turbine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
2.1. Blade Element Momentum Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
2.1.1. Dynamic wake/inflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
2.1.2. Yaw/tilt model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
2.1.3. Dynamic stall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
2.1.4. Airfoil data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
2.1.5. Wind simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
2.2. Lifting line, panel and vortex models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
2.2.1. Vortex methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
2.2.2. Panel methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
2.3. Generalized actuator disc models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
2.4. Navier–Stokes solvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
2.4.1. Introduction to computational rotor aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
2.4.2. Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
2.4.3. Turbulence and transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

Corresponding author. Tel.: +45 45254316.

E-mail address: molh@mek.dtu.dk (M.O.L. Hansen).

0376-0421/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.paerosci.2006.10.002
 ARTICLE IN PRESS
286 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

2.4.4. Geometry and grid generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

2.4.5. Numerical issues. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
2.4.6. Application of CFD to wind turbine aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
2.4.7. Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
3. Structural modelling of a wind turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
3.1. Principle of virtual work and use of modal shape functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
3.2. FEM modelling of wind turbine components applying non-linear beam theory. . . . . . . . . . . . . . . . . . 304
4. Problems and solutions in wind turbine aeroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
4.1. Aeroelastic stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
4.2. Aeroelastic coupling: linear vs. non-linear formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
4.3. Examples of time simulations and instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
4.3.1. Edgewise blade vibration instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
4.3.2. Instability problems of parked rotors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
4.3.3. Flutter instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
5. Present and future developments of aeroelastic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
5.1. Areas with influence on the development of aeroelastic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
5.1.1. Influence of up-scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
5.1.2. Siting of the turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
5.1.3. Future trends in turbine design and siting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
5.2. Areas of development in present and new codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
5.2.1. Non-linear structural dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
5.2.2. Calculation of induction and its dynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
5.2.3. Wake operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
5.2.4. Derivation of airfoil data for aeroelastic simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
5.2.5. Complex inflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
5.2.6. Aerodynamics of parked rotors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
5.2.7. Offshore turbines including floating turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
6. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325

1. Introduction larger and larger commercial wind turbines the need

The size of commercial wind turbines has
increased dramatically in the last 25 years from
approximately a rated power of 50 kW and a rotor
diameter of 10–15 m up to today’s commercially
available 5 MW machines with a rotor diameter of
more than 120 m. This development has forced the
design tools to change from simple static calcula-
tions assuming a constant wind to dynamic simula-
tion software that from the unsteady aerodynamic
loads models the aeroelastic response of the entire
wind turbine construction, including tower, drive
train, rotor and control system. The Danish
standard DS 472 [1] allows simplified load calcula-
tions if the rotor diameter is less than 25 m and
some other criteria are fulfilled. A rotor diameter of
25 m corresponds approximately to a rated power of
200–250 kW, which is less than almost any modern
commercial wind turbine today. Instead, modern
wind turbines are designed to fulfill the require-
ments of the more comprehensive IEC 61 400-1 [2]
standard. At some time during the development of
for aeroelastic tools thus became necessary. Aero-
elastic tools were mainly developed at the univer-
sities and research laboratories in parallel with the
evolution of commercial wind turbines. At the same
time governments and utility companies erected
large non-commercial prototypes for research pur-
poses, as the Nibe [3] and Tjaereborg machines [4].
Measurement campaigns were undertaken on these
machines and the results used to tune and validate
the aeroelastic programmes, in order to develop
advanced software for the rapidly growing industry.
Even today measurements from the Tjaereborg
machine is used as a benchmark when developing
new aeroelastic codes, see e.g. [5]. In [5] is also
compiled a long list of available software that at
different levels of complexity can model the aero-
elastic response of a wind turbine construction. All
the aeroelastic models need as input a time history
of the wind seen by the rotor, which as a minimum
must contain some physical correct properties such
as realistic power spectra and spatial coherence.
Apart from the wind input aeroelastic codes contain
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
an aerodynamic part to determine the wind loads
and a structural part to describe the dynamic
response of the wind turbine construction. For the
aerodynamic part most codes use the Blade Element
Momentum Method (BEM) as described by
Glauert [6], since this method is very fast and,
provided that reliable airfoil data exist, yields
accurate results. Therefore, this method, with all
the necessary engineering adds on, is thoroughly
described later in this article. However, more
advanced numerical models based on the Euler
and Navier–Stokes (NS) equations are becoming so
fast that they now begun to replace the BEM
method in some situations, e.g. when analysing yaw
or interaction between wind turbines in parks.
These models contain more physics and less
empirical input than the BEM method and are
extensively described in this paper. The discretiza-
tion of the wind turbine structure is presently where
the various available codes differ most. Roughly,
there exist three ways to model the structural
dynamics of a wind turbine. One is a full Finite
Element Method (FEM) discretization and another
is a multi-body formulation, where different rigid
parts are connected through springs and hinges.
Finally, the description of blade and tower deflec-
tions can be made as a linear combination of some
physical realistic modes; typically the lowest eigen-
modes. The last method greatly reduces the
computational time per time step, as compared
with a full FEM discretization. All the various ways
of discretizing the wind turbine structure will be
treated in details later in the paper. The very
detailed description of the aerodynamic and struc-
tural models is where this paper differs mostly from
other review articles concerning wind turbine
aeroelasticity such as e.g. [7–9].
2. Predicting aerodynamic loads on a wind turbine
Methods of various levels of complexity to
calculate the aerodynamic loads on a wind turbine
rotor are given, starting with the popular BEM, and
ending with the solution of the NS equations.
2.1. Blade Element Momentum Method
BEM is the most common tool for calculating the
aerodynamic loads on wind turbine rotors since it is
computationally cheap and thus very fast. Further,
it provides very satisfactory results provided that
good airfoil data are available for the lift and drag
287
coefficients as a function of the angle of attack, and
if possible, the Reynolds number. The method was
introduced by Glauert [6] as a combination of one-
dimensional (1D) momentum theory and blade
element considerations to determine the loads
locally along the blade span. The method assumes
that all sections along the rotor are independent and
can be treated separately; typically in the order of
10–20 radial sections are calculated. At a given
radial section a difference in the wind speed is
generated from far upstream to deep in the wake.
The resulting momentum loss is due to the axial
loads produced locally by the flow passing the
blades, creating a pressure drop over the blade
section. The local angle of attack at a given radial
section on a blade can be constructed, provided that
the induced velocity generated by the action of the
loads is known, see Fig. 1. V0 is the undisturbed
wind velocity, W, the induced velocity, Vrot ¼ o  r
the rotational speed of the blade section, Vblade the
velocity of the blade section apart from the blade
rotation and b is the local angle of the blade section
to the rotor plane.
Combining the global momentum loss with the
loads generated locally at the blade section yields
formulas for the induced velocity as
BL cos f
Wz ¼  , (2.1.1)
4r prF V 0 þ f g  nðn  WÞ

BL sin f
Wy ¼  , (2.1.2)
4r prF V 0 þ f g  nðn  WÞ
B is the number of blades, L the lift computed from
the lift coefficient, f is the flow angle, r the density
of air, r the radial position considered, V0 the wind
velocity, W the induced velocity and n the normal
vector to the rotor plane. F is Prandtl’s tip loss
correction that corrects the equations to be valid for
a finite number of blades, see [6, 10]. If there is no
yaw misalignment, that is, the normal vector to the
rotor plane, n, is parallel to the wind vector, then
W z
−Vblade y x
Vrot
β
φ α rotor plane
Vo
Vrel
Fig. 1. Construction of angle of attack, a.
 ARTICLE IN PRESS
288 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

Eq. (2.1.1) reduces to the well-known expression
C T ¼ 4aF ð1  f g  aÞ,
where by definition for an annual element of
infinitesimal thickness, dr, and area, dA ¼ 2pr dr,
dT
CT ¼ . (2.1.4)
1=2rV 20 dA
The axial interference factor is defined as
Wz
a¼
V0
and fg, usually referred to as the Glauert correction,
is an empirical relationship between CT and a, in the
turbulent wake state. It may assume the form
( equilibrium with the aerodynamic loads. If the loads
1 for ap0:3;
fg ¼ 1 (2.1.6)
4ð5  3aÞ for a40:3:
Eqs. (2.1.1) and (2.1.2) are also known to be valid
for an extreme yaw misalignment of 901, that is, the
incoming wind is parallel to the rotor plane as a
helicopter in forward flight. Without any proof
Glauert therefore assumed that Eqs. (2.1.1) and
(2.1.2) are valid for any yaw angle.
An aeroelastic code is running in the time domain
and for every time step the aerodynamic loads must
be calculated at all the chosen radial stations along
the blades as input to the structural model. For a
given time the local angle of attack is determined on
every point on the blades, as indicated in Fig. 1. The
lift and drag coefficients can now be found from
table look-up, and the lift can be determined.
The induced velocities can now be updated using
400
BEM
Measurement
350
Rotorshaft torque [kNm]
Eqs. (2.1.1) and (2.1.2) simply assuming old values
(2.1.3) for the induced velocities on the right-hand sides
(RHS). Updating the RHS of Eqs. (2.1.1) and
(2.1.2) could continue until the equations are solved
with all values at the same time step. However, this
is not necessary as this update takes place in the
next time step, i.e. time acts as iteration. More
important, the values of the induced velocities
change very slowly in time due to the phenomena
of dynamic inflow or dynamic wake.
(2.1.5)
2.1.1. Dynamic wake/inflow
The induced velocities calculated using Eqs. (2.1.1)
and (2.1.2) are quasi-steady, in the sense that they
give the correct values only when the wake is in
are changed in time there is a time delay proportional
to the rotor diameter divided by the wind speed
before a new equilibrium is achieved. To take into
account this time delay, a dynamic inflow model
must be applied. In two EU-sponsored projects
([11,12]) different engineering models were tested
against measurements. One of these models, pro-
posed by S. Øye, is a filter for the induced velocities
consisting of two first-order differential equations
dW int dW qs
W int þ t1 ¼ W qs þ k  t1 , (2.1.7)
dt dt
dW
W þ t2 ¼ W int , (2.1.8)
dt
Wqs is the quasi-static value found by Eqs. (2.1.1)
and (2.1.2), Wint an intermediate value and W the
final filtered value to be used as the induced velocity.
The two time constants are calibrated using a simple
vortex method as
1:1 R
t1 ¼  (2.1.9)
ð1  1:3aÞ V 0

300 and
  r 2 
250 t2 ¼ 0:39  0:26 t1 , (2.1.10)
R
200 where R is rotor radius.
In Fig. 2 is shown for the Tjaereborg machine the
150
computed and measured response on the rotorshaft
torque for a sudden change of the pitch angle. At
100
0 10 20 30 40 50 60 t ¼ 2 s the pitch is increased from 01 to 3.71,
time [s] decreasing the local angles of attack. First the
Fig. 2. Comparison between measured and computed time series rotorshaft torque drops from 260 to 150 kNm and
of the rotorshaft torque for the Tjaereborg machine during a step not until approximately 10 s later the induced
input of the pitch for a wind speed of 8.7 m/s. velocities and thus the rotorshaft torque have settled
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
at a new equilibrium. At t ¼ 32 s the pitch is
changed back to 01 and a similar overshoot in
rotorshaft torque is observed. The decay of the
spikes seen in Fig. 2 can only be computed with a
dynamic inflow model, and such a model is there-
fore of utmost importance for a pitch-regulated
wind turbine.
2.1.2. Yaw/tilt model
Another engineering model for the induced
velocities concerns yaw or tilt. When the rotor disc
is not perfectly aligned with the incoming wind there
is an angle different from zero between the rotor
normal vector and the incoming wind, see Fig. 3. A
yaw/tilt model redistributes the induced velocities so
that the induced velocities are higher when a blade is
positioned deep in the wake than when it is pointing
more upstream. An example of such a model, taken
from helicopter literature [13] is given below. Here,
the input is the induced velocity, W0, calculated
using Eqs. (2.1.1), (2.1.2), (2.1.7) and (2.1.8). The
output is a redistributed value, finally used when
estimating the local angle of attack, W,
 
r w
W ¼ W 0 1 þ tan cosðyb  y0 Þ , (2.1.11)
R 2
yb is the actual position of a blade, y0, is the position
where the blade is furthest downstream and w is the
wake skew angle, see Fig. 3. In some BEM
implementations, W0, is the average value of all
blades at the same radial position, r, and in other
codes it is the local value. This difference in
implementation may cause a small difference from
code to code. Further, there exist different mod-
ifications of Eq. (2.1.11) from different codes, see
ω dynamic stall model is not used one might compute
θyaw/tilt
n x
Vo Vo
, Wn
V
Rotor disc
Fig. 3. Wind turbine rotor not aligned with the incoming wind.
The angle between the velocity in the wake (the sum of the
incoming wind and the induced velocity normal to the rotor
plane) is denoted the wake skew angle, w.
289
[12]. A yaw/tilt model increases the induced
velocities on the downstream part of the rotor and
decreases similarly the induced velocity on the
upstream part of the rotor disc. This introduces a
yaw moment that tries to align the rotor with the
incoming wind, hence tending to reduce yaw
misalignment. For a free yawing turbine such a
model is, therefore, of utmost importance when
estimating the yaw stability of the machine.
2.1.3. Dynamic stall
The wind seen locally on a point on the blade
changes constantly due to wind shear, yaw/tilt
misalignment, tower passage and atmospheric
turbulence. This has a direct impact on the angle
of attack that changes dynamically during the
revolution. The effect of changing the blades angle
of attack will not appear instantaneously but will
take place with a time delay proportional to the
chord divided with the relative velocity seen at the
blade section. The response on the aerodynamic
load depends on whether the boundary layer is
attached or partly separated. In the case of attached
flow the time delay can be estimated using
Theodorsen theory for unsteady lift and aerody-
namic moment [14]. For trailing edge stall, i.e. when
separation starts at the trailing edge and gradually
increases upstream at increasing angles of attack,
so-called dynamic stall can be modelled through a
separation function, fs, as described in [15], see later.
The Beddoes–Leishman model [16] further takes
into account attached flow, leading edge separation
and compressibility effects, and also corrects the
drag and moment coefficients. For wind turbines,
trailing edge separation is assumed to represent the
most important phenomenon regarding dynamic
airfoil data, but also effects in the linear region may
be important, see [17]. It is shown in [15] that if a
flapwise vibrations, especially for stall regulated
wind turbines, which are non-existing on the real
machine. For stability reasons it is thus highly
recommended to at least include a dynamic stall
model for the lift. For trailing edge stall the degree
of stall is described through fs, as
C l ðaÞ ¼ f s C l;inv ðaÞ þ ð1  f s ÞC l;fs ðaÞ, (2.1.12)
where Cl,inv denotes the lift coefficient for inviscid
flow without any separation and Cl,fs is the lift
coefficient for fully separated flow, e.g. on a flat
plate with a sharp leading edge. Cl,inv is normally an
extrapolation of the static airfoil data in the linear
 ARTICLE IN PRESS
290 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
region, and in [17] a way of estimating Cl,fs and fsts is
shown. fst
s is the value of f s that reproduces the static
airfoil data when applied in Eq. (2.1.12). The
assumption is that, fs, always will try to get back
to the static value as
df s f st fs
¼ s , (2.1.13)
dt t faster BEM method. In this method, the induced
that can be integrated analytically to give
f s ðt þ DtÞ ¼ f st st
s þ ðf s ðtÞ  f s Þ expðDt=tÞ, (2.1.14)
t is a time constant approximately equal to Ac/Vrel,
where c denotes the local chord, and Vrel is the
relative velocity seen by the blade section. A is a
constant that typically takes a value about 4.
Applying a dynamic stall model the airfoil data is
always chasing the static value at a given angle of
attack that is also changing in time. If e.g. the angle
of attack is suddenly increased from below to above
stall the unsteady airfoil data contains for a short
time some of the inviscid/unstalled value, Cl,inv, and
an overshoot relative to the static data is seen. It can
thus been seen as a model of the time constant for
the viscous boundary layer to develop from one
state to another.
2.1.4. Airfoil data
The BEM as described above, including all
engineering corrections, is used in most aeroelastic
codes to compute the unsteady aerodynamic loads
on wind turbine rotors. The method is often quite
successful, but depends on reliable airfoil data for
the different blade sections. Three-dimensional (3D)
effects from the tip vortices are taken into account
when applying Prandtl’s tip loss correction and after
this correction the local flow around a given blade
section is assumed to be two-dimensional, i.e. 2D
airfoil data from wind tunnel measurements are
used. However, such measurements are often
limited to the maximum lift coefficient, Cl,max for
airplanes that usually are operated at unstalled flow
conditions. Further, at higher values it is difficult to
measure the forces because of the unsteady and 3D
nature of stall. In contrast to airplane wings, a wind
turbine blade often operates in deep stall, especially
for stall regulation. For the inner part of the blades
even data for low angles of attack might be difficult
to find in literature since for structural reasons the
airfoils used are much thicker than those used on
airplanes. Further, because of rotation the bound-
ary layer is subjected to Coriolis- and centrifugal
forces, which alter the 2D airfoil characteristics.
This is especially pronounced in stall. It is thus often
necessary to extrapolate existing airfoil data into
deep stall and to include the effect of rotation.
Methods have been developed that from a CFD
calculation of the flow past a full wind turbine rotor
can extract 3D airfoil data [18], which then later can
be applied in aeroelastic calculations using the much
velocity at the rotor plane is estimated from the
azimuthally averaged velocity in very thin annular
elements up- and downstream of the rotorplane. In
[19,20], two engineering methods to correct 2D
airfoil data for 3D rotational effects are given as
C x;3D ¼ C x;2D þ aðc=rÞh  cosn b  DC x ; x ¼ l; d; m;
(2.1.15)
DC l ¼ C l;inv  C l;2D ,
DC d ¼ C d;2D  C d;2Dmin
DC m ¼ C m;2D  C m;inv ,
c is the chord, r the radial distance to rotational axis
and b the twist.
In [19] only the lift is corrected, i.e. x ¼ l, and the
constants are a ¼ 3, n ¼ 0 and h ¼ 2, whereas in [20]
a ¼ 2.2, n ¼ 4 and h ¼ 1. In [21] another method
based on correcting the pressure distribution along
the airfoil is given. One must, however, be very
aware that the choice of airfoil data directly
influences the results from the BEM method. For
certain airfoils a lot of experience has been gathered
regarding appropriate corrections to be used in
order to obtain good results, and, because of this,
blade designers tend to be conservative in their
choice of airfoils. With maturing CFD algorithms
especially for the transition and turbulence models
and more wind tunnel tests, the trend is now to use
airfoils specially designed and dedicated to wind
turbine blades, see e.g. [22].
2.1.5. Wind simulation
Besides airfoil data, also realistic spatial-temporal
varying wind fields must be generated as input to an
aeroelastic calculation of a wind turbine. As a
minimum the simulated field must satisfy some
statistical requirements, such as a specified power
spectre and spatial coherence, see [23,24]. In this
method, each velocity component is generated
independently from the others, meaning that there
is no guarantee for obtaining correct cross-correla-
tions. In [25], a method ensuring this is developed
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 291

on the basis of the linearized NS equations. In the be derived

future, wind fields are expected to be generated
numerically from Large Eddy Simulations (LES) or
Direct Numerical Simulations (DNS) of the NS
equations for the flow on a landscape similar to the
actual siting of a specific wind turbine.
L ¼ rV rel G ¼ 1=2rV 2rel cC l ) G ¼ 1=2V rel cC l .
(2.2.1)
Any velocity field can be decomposed in a
solenoidal part and a rotational part as

2.2. Lifting line, panel and vortex models V ¼ r  C þ rF, (2.2.2)

where C is a vector potential and F a scalar
In the present section, 3D inviscid aerodynamic potential [29]. From Eq. (2.2.2) and the definition of
models are reviewed. They have been developed in vorticity a Poisson equation for the vector potential
an attempt to obtain a more detailed description of is derived
the 3D flow that develops around a wind turbine.
The fact that viscous effects are neglected is r2 C ¼ o. (2.2.3)
certainly restrictive as regards the usage of such In the absence of boundaries, C, can be expressed in
models on wind turbines. However, they should be convolution form as
given the credit of contributing to a better under- Z
1 o0
standing of dynamic inflow effects as well as the CðxÞ ¼ dVol, (2.2.4)
credit of providing a better insight into the overall 4p jx  x0 j
flow development [11,12]. There have been attempts where x denotes the point where the potential is
to introduce viscous effects using viscous–inviscid computed, a prime denotes evaluation at the point
interaction techniques [26,27] but they have not yet of integration x0 which is taken over the region
reached the required maturity so as to become where the vorticity is non-zero, designated by Vol.
engineering tools, although they are full 3D models From its definition the resulting induced velocity
that can be used in aeroelastic analyses. field is deduced from the induction law of Biot–
Savart
2.2.1. Vortex methods Z
1 ðx  x0 Þ  o0
In vortex models the rotor blades, trailing and wðxÞ ¼  dVol. (2.2.5)
shed vorticity in the wake are represented by lifting 4p jx  x0 j3
lines or surfaces [28]. On the blades the vortex In its simplest form the wake from one blade is
strength is determined from the bound circulation prescribed as a hub vortex plus a spiralling tip
that stems from the amount of lift created locally by vortex or as a series of ring vortices. In this case the
the flow past the blades. The trailing wake is vortex system is assumed to consist of a number of
generated by the spanwise variation of the bound line vortices with vorticity distribution
circulation while the shed wake is generated by a
oðxÞ ¼ Gdðx  x0 Þ, (2.2.6)
temporal variation and ensures that the total
circulation over each section along the blade where G is the circulation, d is the line Dirac delta
remains constant in time. Knowing the strength function and x0 is the curve defining the location of
and position of the vortices the induced velocity the vortex lines. Combining (2.2.5) and (2.2.6)
can be found in any point using the Biot–Savart results in the following line integral for the induced
law, see later. In some models (namely the lifting- velocity field:
line models) the bound circulation is found from Z
1 ðx  x0 Þ qx0 0
airfoil data table-look up just as in the BEM wðxÞ ¼  G  qS , (2.2.7)
4p S jx  x0 j3 qS0
method. The inflow is determined as the sum of
the induced velocity, the blade velocity and where S is the curve defining the vortex line and S0 is
the undisturbed wind velocity, see Fig. 1. The the parametric variable along the curve.
relationship between the bound circulation and the Utilizing (2.2.7) simple vortex models can be
lift is denoted as the Kutta–Joukowski theorem derived to compute quite general flow fields about
(first part of Eq. (2.2.1)), and using this together wind turbine rotors. The first example of a simple
with the definition of the lift coefficient (second vortex model is probably the one due to Joukowski
part of Eq. (2.2.1)) a simple relationship between [30], who proposed to represent the tip vortices by
the bound circulation and the lift coefficient can an array of semi-infinite helical vortices with
 ARTICLE IN PRESS
292 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
constant pitch, see also [31]. In [32], a system of
vortex rings was used to compute the flow past a
heavily loaded wind turbine. It is remarkable that in
spite of the simplicity of the model, it was possible
to simulate the vortex ring/turbulent wake state
with good accuracy, as compared to the empirical
correction suggested by Glauert, see [6]. Further, a
similar simple vortex model was used in [33] to
calculate the relation between thrust and induced
velocity at the rotor disc of a wind turbine, in order
to validate basic features of the streamtube-
momentum theory. The model includes effects of
wake expansion, and, as in [32] simulates a rotor
with an infinite number of blades, with the wake
being described by vortex rings. From the model it
was found that the axial induced velocities at the
rotor disc are smaller than those determined from
the ordinary stream tube-momentum theory. A
similar approach has been utilized by Koh and
Wood [34] and Wood [35] for studying rotors
operating at high tip speed ratios.
2.2.2. Panel methods
The inviscid and incompressible flow past the
blades themselves can be found by applying a
surface distribution of sources s and dipoles m
(Fig. 4). The background is Green’s theorem, which
allows obtaining an integral representation of any
potential flow field in terms of the singularity
distribution [36,37].
Vðx; tÞ ¼ V 0 þ rfðx; tÞ:
Z  
s0 ðtÞ m0 ðtÞ n0 :r
fðx; tÞ ¼   dS 0 ,
S 4pjx  x j
0
4pjx  x0 j3
ð2:2:8Þ
V0 denotes a given (external) potential flow field
possibly varying in time and space and f is the
perturbation scalar potential. S stands for the active
boundary of the flow and includes the solid
boundaries of the flow SB as well as the wake
SB :  , μ
SW : μW
r 1pUW
U ext r ¼ W þ V W;m  1VUW
n C
F qmW 
ΓC = − μ W ( P )
Fig. 4. Notations for the potential flow around a wing.
surfaces of all lifting components SW. In (2.2.8) m, s
are defined as jumps of f and its normal derivative
across S: m ¼ 1fU and s ¼ 1qn fU with n defined
as the unit normal vector pointing towards the flow.
Source distributions are responsible for displacing
the unperturbed flow so that the solid boundaries
are shaped as flow surfaces and therefore are defined
on SB. Dipoles are added so as to develop
circulation into the flow to simulate lift. They are
defined on SW and the part of SB referring to the
lifting components. In fact a surface distribution of
m is identified to minus the circulation around a
closed circuit which cuts the surface on which m is
defined at one point: G ¼ m.
An important result given by Hess [36] states that
the flow induced by a dipole distribution m defined on
Sm is given by a generalization of the Biot–Savart law:
Z
n0  ðx  x0 Þ
r m0 dS0
Sm 4pjx  x0 j3
Z
ðr0 m  n0 Þ  ðx  x0 Þ
¼ 0 j3
dS 0
Sm 4pj x  x
I 0 0
m s  ðx  x0 Þ
þ dS 0 ð2:2:9Þ
4pjx  x0 j3
@S m
where the line integral is taken along the boundary of
Sm and s is the unit tangent vector to qS m in the
anticlockwise sense. If S m is a closed surface the line
integral vanishes, whereas if m is piecewise constant as
in the vortex lattice method, the surface term will
vanish leaving only the line term which corresponds
to a closed-loop vortex filament present along all lines
of m discontinuity on S m . The two terms on the RHS
of (2.2.9) have the form as the Biot–Savart law
(2.2.5). From this analogy, c ¼ rm  n is called
surface vorticity and ms line vorticity, which justifies
the term vortex sheet for the wake of lifting bodies.
In potential theory a wake surface is the idealiza-
tion of a shear layer in the limit of vanishing
thickness. For an incompressible flow, the flow will
exhibit a velocity jump: 1VUW ¼ rmW while
1VUW  n ¼ 0, 1pUW ¼ 0. Using Bernoulli’s equa-
tion, it follows that
qm
r qt

¼ þ V W;m  r mW ¼ 0, ð2:2:10Þ
qt
 
where V W;m ¼ V þ 
W þ V W =2. Since, G ¼ mW ,
Kelvin’s theorem is obtained from (2.2.9) provided
that SW is a material surface moving with the mean
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
flow V W;m . Circulation will be materially conserved
and therefore mW is identified with its value at t ¼ 0.
For a lifting problem, this means that mW is known
from the history of the wing loading. Assuming that
SW starts at the trailing edge of the wing, the
generation of the wake can be viewed as a
continuous release of vorticity in the free flow.
The streak line of a point along the trailing edge will
reveal the history of the loading of the specific wing
section as indicated in Fig. 7.
The first model developed within the above
context is Prandtl’s lifting-line theory, see e.g. [38].
It concerns a lifting body of vanishing chord (or else
large aspect ratio), and thickness (Fig. 5). So s  0
whereas SB becomes a line carrying the loading GðyÞ
(bound vorticity), which is the only unknown since
the vorticity in the wake (trailing vorticity) is given
by @y GðyÞ. In Prandtl’s original model, GðyÞ is
determined from airfoil data as equation (2.2.1).
Then, as an introduction to lifting-surface theory,
bound vorticity was placed along the c/4 line while
along the 3c/4 the non-penetration condition was
applied in order to determine Gðy; tÞ. The next step
was to introduce the lifting body as a lifting surface.
The most widely used model in this respect is the
vortex-lattice model [39]. It consisted of dividing SB
and SW into panels and defining on them piecewise
constant m distributions (Fig. 6). Then according to
(2.2.9) the perturbation induced by the wing and its
wake, is generated by a set of closed-loop vortex
filaments each defined along the boundary of a
panel. The dipole intensities on the wing can be
determined by the non-penetration condition at the
panel centres whereas along the trailing edge, see
(2.2.10) m ¼ mW to ensure zero loading locally. In
the case of an unsteady flow, the loading GB will
Γ(y)
δy
δΓw = −∂y Γ.δy
Fig. 5. The lifting-line model.
293
⎝
⎛ ∂ΓB ⎜
ΓW = − ΓB = − ⎜ .t
Zero loading at ⎝ ∂t⎛
trailing edge
Se
Sew
emission line
Fig. 6. The lifting-surface model.
change so that the vorticity shed in the wake will
also have a cross component qGB =qt:dt, as
indicated in Fig. 6.
Having determined m it is possible to calculate the
induced velocity and thus the angle of attack and
finally the loads from an airfoil data table look-up.
Another option frequently used in propeller appli-
cations, is to determine lift by integrating the
pressure jump along the section. Then by consider-
ing that the lift force is perpendicular to the effective
inflow direction the angle of attack is determined. In
this case the pressure jump is obtained directly from
Bernoulli’s equation over the section except at the
leading edge where the geometrical singularity of a
blade with no thickness makes it necessary to
include the so-called suction force. In propeller
applications where this concept was first introduced,
the suction force is estimated by means of semi-
empirical modelling [40]. Whenever used in wind
energy applications, the suction force has been
determined as rV  Gdl, where V; G are the local
values at the leading edge and dl represents the
vector length along the leading edge line. In general,
the two schemes give comparable results. It is
difficult to clearly state which scheme is better to
use, since deviations appear as the angle of attack
increases so that a theoretical justification based on
matched asymptotic expansions is difficult. Another
point of concern, regarding both lifting theories, is
the detail in which the flow can be recorded. The
fact that the flow geometry is approximate suggests
that only at some distance from the solid boundary
the flow could be meaningful. Finally, for the same
reason, viscous corrections based on boundary layer
theory cannot be applied.
In order to overcome these difficulties, the exact
geometry of the flow had to be included. This was
done by Hess who first introduced the panel method
 ARTICLE IN PRESS
294 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

in its full form [41]. Now the panel grid is defined on prescribed upper limit. Both schemes, however,
the true solid boundary and a piecewise constant require substantial bookkeeping and quite intense
source distribution is introduced which can be procedures. With panel methods this problem
determined by the no-penetration boundary condi- becomes more complicated because the wake will
tion. In order to account for lift dipole distributions also contain a surface vorticity term. A completely
are added. Because there is no kinematic condition different approach is to discard wake connectivity.
to determine m, Hess defined m to vary linearly along Rehbach [42] was the first to note that for an
the contour of each section (Fig. 7): mðs; y; tÞ ¼ incompressible flow, vorticity concentrations dO of
s  Gðy; tÞ=LðyÞ and used (2.2.10) at the trailing edge the type o dD, g dS and Gdl behave similarly. Their
as an extra condition for determining Gðy; tÞ (Fig. kinematic analogy was already discussed with
7). The resulting problem is non-linear and an reference to (2.2.9). Dynamically they all satisfy
iterative procedure must be used which penalizes the the same evolution equation:
computational cost considerably, as compared to
D q
the lifting-surface model. d O  d O þ ðu:rÞd O ¼ ðd O:rÞu, (2.2.12)
An important aspect of potential flow models Dt qt
concerns wake dynamics. As discussed earlier, the where D=Dt denotes the total or material time
wake of a lifting body is a moving surface and SW derivative. So Rehbach integrated the wake vorti-
should be allowed to change in time. Regarded as a city into point vortices, which subsequently moved
vortex sheet, the evolution of SW in time will be freely as fluid particles carrying vorticity. This
subjected to convection and deformation. For procedure provides substantial flexibility in the
example in the case of the vortex lattice method, evolution of the wake. He also introduced the
each wake segment will conserve its intensity but its concept of modifying the kernel of the Biot–Savart
vector length dl will satisfy the following equation: law in order to cancel the r2 singularity. The
d theoretical justification of Rehbach’s method came
dl ¼ ðdl:rÞV . (2.2.11) later on leading to the development of the vortex
dt
blob method [43].
As time evolves, wake deformations will generate In vortex models, the wake structure can either be
singularities such as intense roll-up along the wake prescribed or computed as a part of the overall
extremities and crossings. In fact at finite time the solution procedure. In a prescribed vortex techni-
flow will blow-up. In order to avoid blow-up, que, the position of the vortical elements is specified
corrective actions must be taken. If the simulation from measurements or semi-empirical rules. This
retains the connectivity of the wake surface, the makes the technique fast to use on a computer, but
lines defining the wake must be smoothened limits its range of application to more or less well-
regularly during the run time. Alternatively one known steady flow situations. For unsteady flow
can apply remeshing which consists in dividing the situations and complicated wake structures, free
wake vortex segments so that they do not exceed a wake analysis become necessary. A free wake
method is more straightforward to understand and
ΓB ( y;t )= −μ TE use, as the vortex elements are allowed to convect
and deform freely under the action of the velocity
field as in Eq. (2.2.12). The advantage of the method
lies in its ability to calculate general flow cases, such
streak line
e
lin

as yawed wake structures and dynamic inflow. The

ion

disadvantage, on the other hand, is that the method

iss

is far more computing expensive than the prescribed

em

wake surface
wake method, since the Biot–Savart law has to be
evaluated for each time step taken. Furthermore,
μ =− sΓ/L μ TE
μTE t t −Δ t
μ TE t −2Δt μ
free wake vortex methods tend to suffer from
TE t − 3Δ t
stability problems owing to the intrinsic singularity
L
in induced velocities that appears when vortex
μ W = μ TE elements are approaching each other. To a certain
emission time
extent this problem can be remedied by introducing
Fig. 7. The exact potential model of a wing. a vortex core model in which a cut-off parameter
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
models the inner viscous part of the vortex filament.
In recent years, much effort in the development of
models for helicopter rotor flow fields have been
directed towards free-wake modelling using ad-
vanced pseudo-implicit relaxation schemes, in order
to improve numerical efficiency and accuracy, e.g.
[44,45].
To analyse wakes of horizontal axis wind
turbines, prescribed wake models have been em-
ployed by e.g. [46–48]. Free vortex modelling
techniques have been utilized by e.g. [49,50].
A special version of the free vortex wake methods
is the method described in [51], where the wake
modelling is taken care of by vortex particles or
vortex blobs.
Recently, the model of [52] was employed in the
NREL blind comparison exercise [137], and the
main conclusion from this was that the quality of
the input blade sectional aerodynamic data still
represents the most central issue to obtaining high-
quality predictions. Nevertheless, it is worth noti-
cing that by introducing relaxing techniques in the
wake evolution [53], it is nowadays possible to run a
large number of revolutions which is of importance
in aeroelasticity with reference to fatigue and
stability analysis; see later.
An alternative to panel methods is offered by the
Boundary Integral Equation Methods (BIEM). By
assuming stagnant flow inside the blade, m ¼ f
and s ¼ @n f, the resulting integral equation is only
weakly singular and so less expensive. Within the
field of wind turbine aerodynamics, BIEMs have
been applied by e.g. [54–56]; up to now, however,
only in simple flow situations.
Vortex methods have been applied on wind
turbine rotors particularly in order to better under-
stand wake dynamics. The next and quite challen-
ging step is to upgrade potential flow methods so as
to also include viscous effects. Examples of applying
viscous–inviscid coupling within the context of 3D
boundary layer theory can be found in [26,57]. Also
attempts to include separation were made in [58].
All these works, however, cannot be considered
conclusive. There are several unresolved issues such
as convergence at the inboard region where
significant radial flow develops as a result of
substantial separation, and the end conditions at
the tip. The fact that current trends in wind turbine
design indicate preference to pitch regulated ma-
chines could increase the interest in flow models
based on inviscid considerations. Finally, another
application of potential flow models is to use them
295
in order to obtain far field conditions for RANS
computations in view of reducing their computa-
tional cost [59].
2.3. Generalized actuator disc models
The actuator disc model is probably the oldest
analytical tool for analysing rotor performance. In
this model, the rotor is represented by a permeable
disc that allows the flow to pass through the rotor,
at the same time as it is subject to the influence of
the surface forces. The ‘classical’ actuator disc
model is based on conservation of mass, momentum
and energy, and constitutes the main ingredient in
the 1D momentum theory, as originally formulated
by Rankine [60] and Froude [61]. Combining it with
a blade-element analysis, we end up with the
celebrated Blade-Element Momentum Technique
[6]. In its general form, however, the actuator disc
might as well be combined with the Euler or NS
equations. Thus, as will be shown in the following,
no physical restrictions have to be imposed on the
kinematics of the flow.
A pioneering work in analysing heavily loaded
propellers using a non-linear actuator disc model is
found in [62]. Although no actual calculations were
carried out, this work demonstrated the opportu-
nities for employing the actuator disc on compli-
cated configurations, such as ducted propellers and
propellers with finite hubs. Later improvements,
especially on the numerical treatment of the
equations, are due to [63,64]. Recently, Conway
[65,66] has developed further the analytical treat-
ment of the method. Within wind turbine aero-
dynamics [67] developed a semi-analytical actuator
cylinder model to describe the flow field about a
vertical-axis wind turbine. A thorough review of
‘classical’ actuator disc models for rotors in general
and for wind turbines in particular can be found in
the dissertation [68]. Later developments of the
method have mainly been directed towards the use
of the NS or Euler equations.
In a numerical actuator disc model, the NS (or
Euler) equations are typically solved by a second-
order accurate finite difference/volume scheme, as in
a usual CFD computation. However, the geometry
of the blades and the viscous flow around the blades
are not resolved. Instead the swept surface of the
rotor is replaced by surface forces that act upon the
incoming flow. This can either be implemented at a
rate corresponding to the period-averaged mechan-
ical work that the rotor extracts from the flow or by
 ARTICLE IN PRESS
296 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
using local instantaneous values of tabulated airfoil
data.
In the simple case of an actuator disc with
constant prescribed loading, various fundamental
studies can easily be carried out. Comparisons with
experiments have demonstrated that the method
works well for axisymmetric flow conditions and
can provide useful information regarding basic
assumptions underlying the momentum approach
[69–72], turbulent wake states occurring for heavily
loaded rotors [73], and rotors subject to coning
[74,75].
In Fig. 8, an example of how various wake states
can be investigated by introducing a constantly
loaded actuator disc into the axisymmetric NS
equations is shown. By changing the thrust coeffi-
cient all types of flow states can be simulated,
ranging from the wind turbine state through the
chaotic wake state to the propeller state.
The generalized actuator disc method resembles
the BEM method in the sense that the aerodynamic
forces has to be determined from measured airfoil
characteristics, corrected for 3D effects, using a
blade-element approach. For airfoils subjected to
temporal variations of the angle of attack, the
dynamic response of the aerodynamic forces
changes the static aerofoil data and dynamic stall
models have to be included. However, corrections
for 3D and unsteady effects are the same for
generalized actuator disc models and the BEM
Fig. 8. Various wake states computed by actuator disc model with prescribed loading: (a) wind turbine state; (b) turbulent wake state; (c)
vortex ring state; (d) hover state. Reproduced from [70,73].
model, hence the description of how to derive
aerofoil data is the same as in Section 2.1.
In helicopter aerodynamics combined NS/actua-
tor disc models have been applied by e.g. [76] who
solved the flow about a helicopter employing a
chimera grid technique in which the rotor was
modelled as an actuator disk, and [77] who
modelled a helicopter rotor using time-averaged
momentum source terms in the momentum equa-
tions.
Computations of wind turbines employing nu-
merical actuator disc models in combination with a
blade-element approach have been carried out in
e.g. [69,70,78,79] in order to study unsteady
phenomena. Wakes from coned rotors have been
studied by Madsen and Rasmussen [74], Mikkelsen
et al. [75] and Masson et al. [78], rotors operating in
enclosures such as wind tunnels or solar chimneys
were computed by Hansen et al. [80], Phillips and
Schaffarczyk [81] and Mikkelsen and Sørensen [82],
and approximate models for yaw have been
implemented by Mikkelsen and Sørensen [83] and
Masson et al. [78]. Finally, techniques for employ-
ing the actuator disc model to study the wake
interaction in wind farms and the influence of
thermal stratification in the atmospheric boundary
layer have been devised by Masson [84] and
Ammara et al. [85].
The main limitation of the axisymmetric assump-
tion is that the forces are distributed evenly along
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 297

the actuator disc, hence the influence of the blades is

taken as an integrated quantity in the azimuthal
direction. To overcome this limitation, an ex-
tended 3D actuator disc model has recently been
developed [86]. The model combines a 3D NS
solver with a technique in which body forces
are distributed radially along each of the rotor
blades. Thus, the kinematics of the wake is
determined by a full 3D NS simulation whereas
the influence of the rotating blades on the flow
field is included using tabulated airfoil data to
represent the loading on each blade. As in the
axisymmetric model, airfoil data and subsequent
loading are determined iteratively by computing
local angles of attack from the movement of
the blades and the local flow field. The concept
enables one to study in detail the dynamics of
the wake and the tip vortices and their influence Fig. 9. Actuator line computation showing vorticity contours
on the induced velocities in the rotor plane. A model and part of computational mesh around a three-bladed rotor.
following the same idea has recently been sug- Reproduced from [88].
gested by Leclerc and Masson [87]. A main
motivation for developing such types of model is
to be able to analyse and verify the validity of
the basic assumptions that are employed in the
simpler more practical engineering models. Re-
views of the basic modelling of actuator disc
and actuator line models can be found in [88], that
also includes various examples of computations.
Recently, another Ph.D. dissertation [89] carried out
a simulation employing more than four million
mesh points in order to study the structure of tip
vortices. In the following we will give some
examples of how the actuator disc/line technique
may help in understanding basic features of wind
turbine flows.
Computed iso-contours of vorticity for a three-
bladed rotor with airfoil characteristics corres-
ponding to the Tjæreborg wind turbine is shown
in Fig. 9. In Fig. 10, a similar computation
shows the formation of the trailing tip vortices. It
is remarkable that the vortices are clearly visible Fig. 10. Iso-surface of constant vorticity showing the formation
more than 3 turns downstream. A new and of tip and root vortices. Reproduced from [89].
interesting application of the actuator line model
is to study the interaction between two or more
turbines, especially for simulating park effects. In 2.4. Navier– Stokes solvers
Fig. 11, the outcome of a computation in which the
interaction between two wind turbines is simulated 2.4.1. Introduction to computational rotor
by replacing the two rotors by actuator lines with aerodynamics
forces obtained from airfoil data is shown. Pre- The first applications of CFD to wings and rotor
sently, this technique is used to investigate the effect configurations were studied back in the late
of large wind farms including many up- and seventies and early eighties in connection with
downstream wind turbines. airplane wings and helicopter rotors [90–94] using
 ARTICLE IN PRESS
298 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
Fig. 11. Interaction of the wake between two partly aligned wind
turbines. Reproduced from [88].
potential flow solvers. To overcome some of the
limitations of potential flow solvers, a shift towards
unsteady Euler solvers were seen through the
eighties [95–98]. When computing power allowed
the solution of full Reynolds Averaged NS equa-
tions, the first helicopter rotor computations in-
cluding viscous effects were published in the late
eighties and early nineties [99–102].
In the late nineties, with the CFD solvers capable
of handling viscous flow around rotors, application
to wind turbine rotors became of practical interest.
The first full NS computations of rotor aero-
dynamics was reported in the literature in the late
nineties [103–107]. The European effort to apply NS
solvers to rotor aerodynamics had been made
possible through a series of National and European
project through the nineties. The European projects
dealing with development and application of the NS
method to wind turbine rotor flows was the Viscous
Effects on Wind turbine Blades (VISCWIND) from
1995 to 1997 [108], Viscous and Aeroelastic effects on
Wind Turbine Blades, (VISCEL), 1998 to 2000
[109,110], and Wind Turbine Blade Aerodynamics
and Aeroleasticity: Closing Knowledge Gaps, 2002 to
2004 [111–115].
2.4.2. Approaches
As a consequence of the origin of most CFD
rotor codes from the aerospace industry and related
research, many existing codes are solving the
compressible NS equations and are intended for
high-speed aerodynamics in the subsonic and
transonic regime [116–120]. For the helicopter
applications, where compressibility plays an impor-
tant role, this is the natural choice. For wind turbine
applications, however, the choice is not as obvious;
one reason being the very low Mach numbers near
the root of the rotor blades. As the flow here
approaches the incompressible limit, Mach0.01 it
is very difficult to solve the compressible flow
equations. One remedy to improve their capability
is the so-called preconditioning, that changes the
eigenvalues of the system of the compressible flow
equations by premultiplying the time derivatives by
a matrix. On the other hand, the compressible
solvers have many attractive features, among
these the ease of implementation of overlapping
and sliding meshes, application of high-order up-
wind schemes, and very well-developed solutions
methods.
Another very popular method, especially in the
US is the Artificial Compressibility Method
[121,122], where an artificial sound speed is intro-
duced to allow standard compressible solution
methods and schemes to be applied for incompres-
sible flows. In case of transient computations sub-
iterations are taken within each time step to enforce
incompressibility [122]. The method has several
attractive features: Among these a similar ease of
implementation of overlapping grids as the com-
pressible codes. Overlapping grids are a necessity, to
solve rotor/stator problems that are present when
the rotor, tower and nacelle are all included in the
computations. The main shortcoming of the method
may be problems to enforce incompressibility in
transient computations without the need for a huge
amount of sub-iterations, and the problem of
determining the optimum artificial compressibility
parameter.
Due to the low Mach number encountered in
wind turbine aerodynamics, an obvious choice is
thus the incompressible NS equations. These
methods are generally based on treating pressure
as a primary variable [123–125]. Extensions to
general curvilinear coordinates can be made along
the lines of [126]. The method is not as easily
extended to overlapping grids as the compressible
and the artificial compressibility method, due to the
elliptical pressure correction equation. But the
method is well suited for solving the nearly
incompressible problems often experienced in con-
nection with wind energy. In connection with
steady-state problems, the method can be acceler-
ated using local time stepping, while the method
using global time stepping still is well suited for
transient computations.
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
2.4.3. Turbulence and transition
It is well known, that the NS equations cannot be
directly solved for any of the cases of practical
interest to wind turbines, and that some kind of
turbulence modelling are needed. The standard
approach to derive turbulence models is by time
averaging the NS equation, resulting in the so-called
Reynolds Averaged NS equations (RANS). Several
different models have been used with good results
for wind turbine applications, the most successful
ones being the k-omega SST model of Menter [127],
the Spalart–Allmaras model [128], and the Bald-
win–Barth model [129]. The Baldwin–Lomax [130]
model, often used in connection with helicopter and
fix-wing applications, are not very well suited for
wind turbine applications, where relatively high
angles of attack are very common.
Several studies performed for stall controlled
wind turbines, have shown that all RANS models
lack the capability to model the stalled flow regime
at high wind speeds. One possible way around this
problem, the so-called Detached Eddy Simulation
(DES) technique [131,132], has shown some promis-
ing results but still needs further validation.
Additionally, the DES technique is much more
computationally expensive than the standard
RANS approach, as it needs much finer computa-
tional meshes and the computations needs to be
computed with time accurate algorithms.
From experiments it is known that laminar/
turbulent transition influences the flow over rotor
blades for some cases. It has been demonstrated
for 2D applications, that transition models can
greatly improve the accuracy for cases where
transition phenomena are important. Even though
nearly all rotor studies so far have been com-
puted assuming fully turbulent conditions, it is
generally accepted that it is important to in-
clude laminar/turbulent transition to model the
physics as close as possible [133,134]. Predicting
transition in 3D is a much more complex task than
dealing with 2D, and 3D transition is an active
research field.
2.4.4. Geometry and grid generation
To compute a rotor using CFD, the first step is to
obtain a digitized description of the blade geometry.
Often the blade descriptions are given as spanwise
sectional information, listing the airfoil section, the
twist, the thickness, and the position with respect to
the blade axis. Often the blades are highly twisted,
and with a large taper in the spanwise direction.
299
Depending on the flow solver, different ap-
proaches to the mesh generation process exist:
Cartesian cut cells, unstructured and structured
and combinations of these. So far the majority of
flow solvers applied to wind turbine research have
utilized structured grids with hexahedral cells. In
connection with structured grids there are several
issues that need to be decided upon. Generally, the
problem of making a high quality grid around a
modern rotor cannot be handled by a single block
configuration, but needs some kind of multi block
mesh. These can either be conforming at the block
boundaries, non-conforming or overlapping. The
overlapping grids gives the highest degrees of
freedom followed by the non-conforming and the
conforming grids. Firstly, the grid needs to accu-
rately resolve the blade shape, with good resolution
of the leading edge and tip region. Secondly, the
grids also need to resolve the regions around the
blade with sufficient resolutions to capture the flow
physics. As the Reynolds numbers are quite high,
1–6 million, the cells near the rotor blades become
very thin, as the non-dimensional distance y+ must
be approximately 1 to resolve the laminar sub-layer
and have accurate solutions. The mesh generation
process calls for some degree of experience and grid
refinement studies to verify that the grid is sufficient
to resolve the desired physics. Also the grids need to
extend far away from the rotor, in the order of
several rotor diameters, to avoid disturbing the
induced velocity field near the rotor blades. For
axial flow conditions, the flow solvers often take
advantage of the rotational periodicity of the rotor,
solving only for a single blade using periodic
conditions.
Using an unstructured flow solver with tetrahe-
dral cells, the grid generation process is less
cumbersome. But the problem of resolving very
thin boundary layers using tetrahedral cells is well
known, and it may be necessary to combine the
solver with some kind of prismatic grids near the
blade surface to avoid this problem. The use of
unstructured flow solvers is not wide spread in
connection with wind turbine aerodynamics, prob-
ably because of the limited geometrical complexity,
and the strength of unstructured solvers mainly
being their ability to cope with complex geometries.
2.4.5. Numerical issues
The codes typically used for wind turbines are of
at least second-order accuracy in both time and
space, often with an implicit time discretization
 ARTICLE IN PRESS
300 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

scheme to loosen the time step restriction inherent One of the first real proofs that CFD for wind
to explicit methods. Typically, the viscous terms are turbine rotor applications can be useful came in
discretized with central differences, while the con- connection with the blind comparison organized by
vective terms are discretized with second- or third- the National Renewable Energy Laboratory in
order upwind schemes. To solve routinely for 5–10 Boulder, Colorado in December 2000 [136–138].
million grid points, the solvers are often available in Some of these results were later published in
a parallelized version that allows for execution on [139,140]. Here several wind turbine research groups
several CPU’s in parallel. The rotating nature of the were asked to compute a series of different
problem requires the use of either a moving frame operational conditions for the NREL Phase-VI
including the non-inertial acceleration terms or a turbine, corresponding to actual cases measured in
moving mesh option where so-called mesh fluxes the NASA Ames 80  120 ft wind tunnel. When the
must be included in the code. For a good overview results were made publicly available, it proved that
of the numerical issues in connection with incom- one of the applied CFD codes were consistently
pressible flow, see [135]. reproducing the measured distribution of the aero-
dynamic forces along the blade span, even under
2.4.6. Application of CFD to wind turbine highly 3D and extreme stall conditions.
aerodynamics The output extracted from typical CFD rotor
The major part of wind turbine rotor computa- computations are the low-speed shaft torque, or
tions performed until now has been focused on zero power production, and root flap moments, see Fig.
yaw rotor only configuration, where the nacelle and 12. For modern full size commercial turbines, the
tower have been neglected, and the inflow to the power and root flap moment are typically the only
rotor has been assumed to be steady without shear. available properties. Besides the quantities normally
This is, of course, a great simplification, but in many measured, the CFD simulations provide a huge
cases still a sufficiently good approximation. The amount of detailed information that can be used to
effect of the tower on the rotor on an upwind provide more insight. The data typically extracted
turbine is comparable to other unsteady effects, are the spanwise distributions of force coefficients,
such as incoming turbulence, time variations of the Fig. 13, the limiting streamlines on the blade
rotor and of the incoming flow. A simulation, surfaces, Fig. 14, and the sectional pressure
working with a full turbine geometry has been tried distributions along the blade span, Fig. 15. For
[106]. This type of simulation is much more the NREL Phase-VI turbine, these detailed quan-
expensive, and needs some kind of sliding/over- tities can be compared to measurements, but this is
lapping mesh to accommodate the movement of the not generally the case for commercially available
rotor with respect to the turbine tower and nacelle. turbines.
Additionally, the simulation needs to be time Another application of rotor CFD is the study of
accurate, and good resolution of the flow around different aerodynamic details of the rotor, such as
the tower is needed to capture the tower wake far the blade tips, the design of the root section etc.
downstream of the turbine. Here, the CFD technique can be used to supply

1700 5000
Low Speed Shaft Torque [Nm]

1600 Measured
4500 Measured
Computed Computed
Root Flap Moment [Nm]

1500
4000
1400
1300 3500
1200 3000
1100 2500
1000
2000
900
800 1500
700 1000
6 8 10 12 14 16 18 20 22 24 26 6 8 10 12 14 16 18 20 22 24 26
Wind Speed [m /s] Wind Speed [m/s]

Fig. 12. Comparison of computed and measured low-speed-shaft torque and root flap moment for the NREL Phase-VI rotor during axial
flow conditions. The 7 one standard deviation of the measurements is indicated around the measured values.
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 301

3 0.04
Measured Measured
2.5 Computed 0.02 Computed

2 0

CT
CN

1.5 -0.02

1 -0.04

0.5 -0.06

0 -0.08
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r/ R r/ R

Fig. 13. Spanwise normal and tangential force coefficients for run S20000000, of the NREL/NASA Ames measurements, corresponding
to a wind speed of 20 m/s.

NREL PHASE-6, W=10 [m / s], Limiting Streamlines modelled in yaw simulations, thereby increasing the
RISOE, EllipSys 3D Computation number of mesh points typically by a factor of
three. Additionally, the azimuthally variation in-
herent in yaw simulations dictates a time accurate
simulation, as no steady state solution can exist,
thereby increasing the computing time severely.
Fig. 14. Limiting streamlines on the surface of the NREL Phase- Typically, a yaw computation will be 10–20 times
VI rotor for the 10 m/s axial case. The picture shows a sudden more expensive compared to steady state axial flow
leading edge separation around r/R ¼ 0.47. computations. Fig. 16 shows the normal and
tangential force at the r/R ¼ 30% radius during
one revolution at 601 yaw operation.
3
r/ R=0.63 The release of the measurements on the NREL
Phase-VI rotor has heavily influenced the CFD
2.5
activities dealing with wind turbine rotor aerody-
2
namics. This unique data set, with several well-
1.5
documented cases, has given a new possibility to test
1
-C p

details of state of the art CFD codes. In the years

0.5 since the release of the measurements nearly half of
0 all published CFD studies of wind turbine rotors
-0.5 deals with these measurements. Besides the refer-
-1 ences mentioned other places in this paper, the
-1.5 following studies use the NREL/NASA Ames
0 0.2 0.4 0.6 0.8 1
measurements [144–147].
X / Chord
Recently DES simulations of rotors at realistic
Fig. 15. Comparison of the measured and computed pressure operational conditions have been attempted [148].
distribution at the r/R ¼ 0.63 section of the NREL Phase-VI Again the NREL Phase-VI rotor was used to verify
blade, the measurements are shown as circles while the
computations are shown with lines.
the model. The reason for this choice is, besides the
availability of detailed measurements, the fact that
the rotor has a limited aspect ratio, hence making
the DES computations more affordable with respect
information, that the engineering methods are not to the number of grid cells. The mesh for this
capable of providing. simulation consists of 15 million cells, compared to
With the increase in computational power, it is around 2 million cells for a standard RANS rotor
today both possible and affordable to do yaw computation. The fact that DES computations must
computations using CFD [133,141–143]. In contrast be performed using time accurate computations,
to the axial flow cases, the total rotor must be makes these types of simulations 20–40 times more
 ARTICLE IN PRESS
302 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

S1500600, r / R=0.30 S1500600, r/ R=0.30

12 1
Measurements Measurements
10 0.8
Computed Computed
8 0.6
6 0.4

CT
CN

4 0.2
2 0
0 -0.2
-2 -0.4
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Azimuth Angle [deg.] Azimuth Angle [deg.]

Fig. 16. Azimuth variation of normal and tangential force coefficient, for the NREL Phase-VI turbine at the r/R ¼ 0.4 section during a
601 yaw error at 15 m/s wind speed.

expensive than standard steady-state rotor compu-
tations. For modern-type rotors with large aspect
ratios, the cell count would be even higher and the
computations even more expensive. For the NREL
Phase-VI rotor the improvement using the DES
technique is very limited, but for other rotors where
the RANS equations do not perform as well, DES
may provide much more improvement. The use of
pure Large Eddy Simulation, has been demon-
strated in [149], where a computational grid of 300
million points is used to compute the initial
transient of the development of a tip vortex.
the dynamic response of the entire construction
subject to the time-dependent load using an aero-
dynamic model, such as the BEM method. For
offshore wind turbines also wave loads and perhaps
ice loads on the bottom of the tower must be
estimated. Two different and frequently used
approaches to set up a dynamic structural model
for a wind turbine are described in the next
subsections.
3.1. Principle of virtual work and use of modal shape
functions

2.4.7. Future The principle of virtual work is a method to set

Today NS solvers are an important tool for
analysing different wind turbine rotor configura-
tions, and are used routinely along with measure-
ments and other computational tools for
development and investigation of wind turbines.
NS solvers are especially well suited for detailed
investigation of phenomena that cannot directly be
accessed by simpler and less computationally
expensive methods. Additionally NS methods can
be used as a supplement to measurements, where
they can be used both in the planning phase and to
have a better interpretation of the actual physics in
connection with the analysis of measurements.
Finally, one of the latest trends is to couple NS
solvers to structural codes to perform full elastic
computations of wind turbine rotors.
3. Structural modelling of a wind turbine
The main purpose of a structural model of a wind
turbine is to be able to determine the temporal
variation of the material loads in the various
components. This is accomplished by calculating
up the correct mass matrix, M, stiffness matrix, K,
and damping matrix, C, for a discretized mechanical
system as
Mx€ þ Cx_ þ Kx ¼ Fg , (3.1.1)
where Fg, denotes the generalized force vector
associated with the external loads, p. Eq. (3.1.1) is
of course nothing but Newton’s second law assum-
ing linear stiffness and damping, and the method of
virtual work is nothing but a method that helps
setting this up for a multibody system and that is
especially suited for a chain system. Knowing the
loads and appropriate conditions for the velocities
and the deformations, Eq. (3.1.1) can be solved for
the accelerations wherefrom the velocities and
deformations can be determined for the next time
step. The number of elements in, x, is called the
number of DOF, and the higher this number the
more computational time is needed in each time step
to solve the matrix system. Use of modal shape
functions is a tool to reduce the number of DOF
and thus reduce the size of the matrices to make the
computations faster per time step. A deflection
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
shape is here described as a linear combination of a
few but physical realistic basis functions, which are
often the deflection shapes corresponding to the
lowest eigenfrequencies (eigenmodes). For a wind
turbine such an approach is suited to describe the
deflection of the tower and the rotor blades and the
assumption is that the combination of the Power
Spectral Density of the loads and the damping of
the system do not excite the eigenmodes associated
with higher frequencies. In the commercially avail-
able and widely used aeroelastic simulation tool,
FLEX, see e.g. [150], only the first 3 or 4 (2 flapwise
and 1 or 2 edgewise) eigenmodes are used for the
blades. Results from this model are generally in
good agreement with measurements, indicating the
validity of the underlying assumption. First one has
to decide on the DOF necessary to describe a
realistic deformation of a wind turbine. For instance
in FLEX4 17–20 DOFs are used for a three bladed
wind turbine, with 3–4 DOFs per blade as described
above, 4 DOFs for the deformation of the shaft (1
for torsion, 2 for the hinges just before the first
bearing, with associated angular stiffness to describe
bending, and 1 for pure rotation), 1 DOF to
describe the tilt stiffness of the nacelle and, finally,
3 DOFs for the tower (1 for torsion, 1 for the first
eigenmode in the direction of the rotor normal and
1 in the lateral direction).
The method of virtual work will only be briefly
described. For a more rigorous explanation of the
method the reader is referred to textbooks on
dynamics of structures. The values in the vector
describing the deformation of the construction, xi,
are denoted the general coordinates. To each
generalized coordinate is associated a deflection
shape, ui, that describes the deformation of the
construction when only xi is different from zero and
typically has a unit value. The element i in the
generalized force corresponding to a small displace-
ment in DOF number i, dxi, is calculated such that
the work done by the generalized force equals the
work done on the construction by the external loads
on the associated deflection shape.
Z t ¼ (n+1)Dt from deformations, xni , velocities, x_ ni ,
F g;i dxi ¼ p  ui dS, (3.1.2)
S choice for the time integration scheme is the
where S denotes the entire system. Please note that
the generalized force can be a moment and that the
displacement can be angular. All loads must be
included, i.e. also gravity and inertial loads such as
Coriolis, centrifugal and gyroscopic loads. The non-
linear centrifugal stiffening can be modelled as
303
equivalent loads calculated from the local centrifu-
gal force and the actual deflection shape as shown in
[151]. The elements in the mass matrix, mi,j, can be
evaluated as the generalized force from the inertia
loads from an unit acceleration of DOF j for a unit
displacement of DOF i. The elements in the stiffness
matrix, ki,j, correspond to the generalized force from
an external force field which keeps the system in
equilibrium for a unit displacement in DOF j and
which then is displaced xi ¼ dij, where dij is
Kroneckers delta. The elements in the damping
matrix can be found similarly. For a chain system
the method of virtual work as described here
normally gives a full mass matrix and diagonal
matrices for the stiffness and damping. For one
blade rigidly clamped at the root (cantilever beam)
it is relatively easy to estimate the lowest eigen-
modes (first flapwise u1f(x), first edgewise u1e(x) and
second flapwise u2f(x)) e.g. using an iterative method
as described in [151]. The eigenmodes are normally
described in a coordinate system aligned with the tip
chord as e.g. shown in Fig. 1. It is practical to
normalize the deflection shapes so that the tip
deflection is unity. It is now assumed that any
deflection can be described as a linear combination
of these modes as
uðxÞ ¼ x1 u1f ðxÞ þ x2 u1e ðxÞ þ x3 u2f ðxÞ. (3.1.3)
The velocity and accelerations can be calculated,
respectively, as
u_ ðxÞ ¼ x_ 1 u1f ðxÞ þ x_ 2 u1e ðxÞ þ x_ 3 u2f ðxÞ (3.1.4)
and
u€ ðxÞ ¼ x€ 1 u1f ðxÞ þ x€ 2 u1e ðxÞ þ x€ 3 u2f ðxÞ. (3.1.5)
The advantage of the method using generalized
coordinates and modal shape functions is that the
number of DOF in the dynamic system can be
reduced to a relatively small number. Further, some
high eigenfrequencies are filtered away, which is
beneficial for the allowable time step, when comput-
ing deformations, xinþ1 , and velocities, x_ nþ1
i , at time
and accelerations, x€ ni , at time t ¼ nDt. A good
Runge–Kutta–Nyström method, which requires
for stability reasons that the time step should
resolve the highest eigenfrequency with 4 points;
but for accuracy reasons 10 points is preferred.
By reducing the highest eigenfrequency using a
modal description of e.g. the blades not only
 ARTICLE IN PRESS
304 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

reduces the number of DOFs but also larger time bending displacements u, w the torsion angle y and
steps can be taken. the tension v:

3.2. FEM modelling of wind turbine components r ¼ n0 þ S0  u þ S1  @y u;

2 3
applying non-linear beam theory 1 0 0 z
6 7
S0 ¼ 4 0 1 0 0 5;
Even though modal analysis offers a computa- 0 0 1 x (3.2.1)
tionally effective way to analyse wind turbine 2 3
dynamics, most of the recently developed aeroelastic 0 0 0 0
6 7
codes use a full FEM approach [152], which allows S1 ¼ 4 x 0 z 0 5:
a more complex deformation state of the wind 0 0 0 0
turbine. The main features of such modelling
procedures together with some aspects of non-linear Using Hooke’s law and assuming that shear stresses
beam theory are discussed in the present section. do not produce net torsion, the sectional elastic
The structural modelling of wind turbines is loads can be derived:
based on beam theory. For a 1D structure (beam)
subjected to bending in two directions, torsion and F y ¼ ðEAÞqy v  ðEAx Þq2yy u  ðEAz Þq2yy w,
axial tension, the formulation of the problem M y ¼ ðGI t Þqy y,
consists of two parts: (a) the elastic model which
reduces the 3D structure of each component into a M x ¼ ðEI xz Þq2yy u þ ðEI xx Þq2yy w  ðEAz Þqy v,
1D structure concentrated along the elastic axis of M z ¼ ðEI zz Þq2yy u  ðEI xz Þq2yy w þ ðEAx Þqy v,
the beam, and (b) the derivation of the dynamic
ð3:2:2Þ
equations. In classical beam theory (first order) the
deformed position r of any point initially at x0 ¼ where the terms in parentheses denote the averaged
ðx; y; zÞT (Fig. 17) can be described by the displace- sectional properties of the structure. By considering
ment vector u ¼ fu; v; w; ygT consisted of the two the balance of loads and moments on of a beam

a
z
y
Undeformed geometry
dy Deformed geometry

Deformed elastic axis

r w+dw
w A

u u+du

b M z + dM z
z
δP
Fz + dFz y
Mx M y + dM y
Fy + dFy
Fx
ra
Fy
(a) dr (b)
A Fx + dFx
My Fz
M x + dM x
Mz

Fig. 17. Kinematics and dynamics of a beam structure.

 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 305

2 3
element of length dy the beam equations are 0 0 0 0
derived: 6 7
6 EAx 0 EAz 07
Z  Z  6 7
K12 ¼ 6 7,
6 0 0 0 07
r€ dm dy ¼ dF þ g dm dy þ dp dy, 4 5
A A
0 0 0 0
(3.2.3) 2 3
0 EAx 0 0
Z  6 7
60 0 0 07
r0  r€ dm dy ¼ dM þ dr  ðF þ dFÞ 6 7
K21 ¼ 6 7.
A 60 EAz 0 07
Z  4 5
þ r0  g dm dy þ ra  dp dy; ð3:2:4Þ 0 0 0 0
A
In case of flexible blades if large deformations are
where m denotes the mass per unit length, dF, dM expected, as in the case of helicopter blades, second-
are the net elastic loads on dy, g is the acceleration order non-linear beam theory is to be used. The
of gravity and dp the sectional aerodynamic loads relevant developments originate from the work in
exerted at the aerodynamic centre ra ¼ ðxa ; 0; za Þ [154]. The beam axis again lies along the y-axis but
while in the undeformed state of the beam while x and z
denote the two bending directions of the beam. At
r0 ¼ fdu þ x þ zy; dv þ dy; dw þ z  xygT
its deformed state a local co-ordinate system O0 xZz
ffi fx þ zy; dy; z  xygT . is introduced which follows the pre-twist of the
beam so that x and z coincide with the local
In view of a more systematic and general framework
structural principle axes of the cross section
for formulating dynamic equations, Hamilton’s
(Fig. 18).
principle has been also used [153,154].
Then, (3.2.1) becomes
By introducing (3.2.2) into (3.2.3) and (3.2.4) the 8 9 8 9
beam dynamic equations are obtained: < u >
> = < x >
> =
Z Z r ¼ y þ v þ E  l@y y , (3.2.6)
ðS0 ÞT  r€ dm  qy ðS1 ÞT  r€ dm ¼ qy ½K11  qy u >
: >
; : z >
> ;
A A
w
þ q2yy ½K22  q2yy u þ qy ½K12  q2yy u where l denotes the warping function of the cross
Z
section and E is the transformation matrix between
þ q2yy ½K21  qy u þ ðS0 ÞT  g dm þ Sa  dP; Oxyz and O0 xZz. In [154], E is given in terms of the
A
elastic displacements up to second order. Next the
ð3:2:5Þ
strain tensor ij defined through (3.2.6) is introduced
2 3 in Hooke’s law in order to define the strain–displa-
Fy 0 0 0 cements relations, which are used in the derivation
6 7
6 0 EA 0 0 7 of the resulting sectional elastic loads as in (3.2.2).
6 7
K11 ¼6 7, Because they are derived in the O0 xZz system, before
6 0 0 Fy 0 7
4 5 introducing them in (3.2.3) and (3.2.4) they are
0 0 0 GI t transformed into the Oxyz system using the
2 3
EI zz 0 EI xz 0
6 7
6 0 0 0 07 z ζ
6 7
K22 ¼ 6 7,
6 EI xz 0 EI xx 0 7 θx
4 5 z
θ
0 0 0 0 O’ ξ
2 3 w θz w θˆ + θt
1 0 0 y ξ
u
6 7 O u x
60 1 0 7 v
6 7
Sa ¼ 6 7,
60 0 1 7
4 5 x
za 0 xa Fig. 18. Beam co-ordinate system definition.
 ARTICLE IN PRESS
306 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

transformation matrix E. The final expressions are y

quite lengthy and the reader is referred to [154]. It is
noted that compared to the classical beam model, z
second-order theory will produce fully complete
stiffness matrices containing many non-linear terms. x
The above non-linear beam model is a specific
example amongst several models of varying com-
plexity which have been gradually developed start-
O
ing from the Timoshenko beam theory, by x
eliminating the ad hoc simplifying assumptions of y y ,y
G
the original model [155–158]. O
Regarding wind turbines, almost all structural rGk
models are based on classical beam theory
[150,153,159–162]. This is partially due to the fact
that wind turbine components are far more rigid ρk
z
compared to helicopter blades for which non-linear
theory has been developed. Furthermore most of the
difficulties in analysing wind turbine systems are
generated by the aerodynamics and in particular the
onset of stall, which is certainly less pronounced on
helicopter rotors. Current designs are quite stiff so
that non-linear beam modelling is not expected to OG ,O
change drastically the quality of predictions besides xG , x
providing a sound basis for including the geometrical
non-linearities [163]. However, if wind turbines zG , z
become more flexible, it could become necessary to
adopt such kind of structural modelling. Fig. 19. Multi-body representation of a wind turbine.
Regardless the details the beam equations are
fourth order with respect to bending and second displacement field du we require that
order with respect to tension and torsion. So for a Z L
FE approximation of the equations, C1 shape
duT  Zðu; u_ ; u€ Þ dy
functions are used for u, w and C0 for v and y. At 0
the element level, u, w are approximated with third- X Z Le
order polynomials and the DOFs are the values and  duTe  Zðue ; u_ e ; u€ e Þ dy ¼ 0; 8du. ð3:2:8Þ
e 0
space derivatives at the end nodes while v and y are
approximated with first-order polynomials and the
Because due ðy; tÞ ¼ Ne ðyÞ  d^ue ðtÞ, it follows that:
DOFs again at the end nodes. In some models and
certainly when the second-order beam theory is XZ Le
applied, second- and third-order polynomials are NeT  ZðNe u^ e ; Ne u^_ e ; Ne u^€ e Þ dy ¼ 0, (3.2.9)
used, respectively, for v and y. In this case the mid- e 0

point as well the points at 1/3 and 2/3 interior points

are used as nodes. So within an element ‘‘e’’,
ue ðy; tÞ ¼ Ne ðyÞ  u^ e ðtÞ;
where Ne ðyÞ is the matrix containing the shape
functions and u^ e ðtÞ the vector of the DOFs at the
nodes of the elements [164]. As in Section 3.1
concerning the principle of virtual work, which in
the FEM terminology is the Galerkin formulation
of the problem is used to generate the discrete
equations. With reference to (3.2.5) taken symboli-
cally as Zðu; u_ ; u€ Þ ¼ 0, for any admissible virtual
which is a set of second-order ordinary equations in
time with respect to u^ e ðtÞ. Although Z can contain
(3.2.7) non-linear terms, it can always be written in the
usual form:
MðuÞ  u^€ þ CðuÞ  u^_ þ KðuÞ  u^ ¼ QðuÞ; (3.2.10)
   
where the mass, damping and stiffness matrices as
well as the generalized loads in the RHS, in general
will depend on the displacement field and its time
and space derivatives (noted by a tilde).
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
The main components of a wind turbine are the
blades, the drive-train and the tower (Fig. 19). They
are all modelled as beam structures and typically the
structural properties are assumed for each compo-
nent to continuously vary along the corresponding
elastic axis. However, localized properties can be
added in the form of concentrated masses, dampers
or springs. The gearbox (if present), the generator,
the hub are usually added in this way. Other
examples are the flexibility or damping character-
istics of the yaw bearing or the pitch mechanism.
The involvement of different body motions for each
component in combination with the connections
where loads and displacements are communicated
from one component to the other, calls for a global
formulation of the dynamic problem. To this end
most works adopt a multi-body approach [165],
which consists of considering each component
separately subjected to appropriate boundary con-
ditions, which fit the different components into the
complete configuration. In such an approach the
elastic DOF of each component are defined as local
to an extended vector of DOF
elastic displacements to which rigid body motions
are added through the kinematic boundary condi-
tions, see e.g. [161,166,167] for a more general
discussion. It is also possible to introduce the global
displacements and rotations at each node instead of
the local ones [153], allowing a unified description of
the complete configuration. In this case, however,
the non-linearities of the connections are introduced
implicitly and therefore linearization can only be
performed globally, which is not always desirable.
In multi-body modelling each component k is
assigned a local coordinate system Oxyz with its
undeformed elastic axis along the y-axis. Then the
position of a point with respect to the fixed system
rG;k can be expressed with respect to its local
position rk (defined as in (3.2.1)) as follows:
rG;k ¼ qk þ Ak  rk , (3.2.11)
where qk denotes the position of the origin of the
local system with respect to the fixed system and Ak
denotes the local to global transformation matrix.
The exact form of qk and Ak depends on the
kinematic conditions introduced when joining (con-
necting) the various components (i.e. blades-to-hub
or drive-train-to-tower). Depending on the type of
connection, common global displacements and
rotations are assigned for the restricted DOF, which
are identified to either an existing elastic DOF (shaft
bending at the hub) or a rigid body motion (blade
pitch). The component contributing the kinematics
307
will in response receive from the rest of the
connected components their internal (or reaction)
loads. This operation involves co-ordinate system
transformations between the components. The set
of kinematical DOF involved in the definition of qk
and Ak for all components is denoted collectively as
q. So qk ¼ qk(q; t) and Ak ¼ Ak(q; t). qk is defined as
a mixed series of translations and rotations, whereas
Ak is defined solely as a sequence of rotations. A
typical example is shown in Fig. 20, where q1–q6 are
the elastic DOF at the top of the tower, q8–q13 are
the elastic DOF of the drive train at the hub and q7
and q14 are the yaw angle and the pitch of the blade,
respectively. By introducing (3.2.11) into (3.2.5), the
centrifugal and Coriolis terms of the inertia loads
appear as a result of the time derivatives of Ak while
in the Hamilton’s principle approach they are
produced automatically.
By combining the equations of all components
the complete system of the dynamic equations for
the wind turbine is obtained in the form of (3.2.10)
with respect
u^ ext ¼ u^ ; q . By construction qk and Ak will depend
on unknown DOF while at the same time they
correlate the state of the components in contact so
the terms they generate are non-linear with respect
to the unknown DOF uk and q. Linearization in this
case is a tedious procedure which must be done
carefully and systematically. Fortunately software
using symbolic mathematics, such as Mathematica,
y
q14
z
x
q7
O y G, y
x
y q1, q2, q3, q4,
O q5 , q6
q8, q9, q10, q11,
q12, q13
z
OG ,O
xG, x
z G, z
Fig. 20. A typical definition of the q DOF on a HAWT.
 ARTICLE IN PRESS
308 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
are available which are strongly recommended. The
procedure is based on Taylor’s expansions of all
variables and retaining of first-order terms. Alter-
natively, the extended version of (3.2.10) can be
solved directly through an iterative procedure in
each time step, which can use the linearized solution
as a starting predictor.
4. Problems and solutions in wind turbine
aeroelasticity
In the following section, the use of aeroelastic
codes on wind turbine constructions including
stability studies will be illuminated mainly through
examples.
4.1. Aeroelastic stability
Wind turbines suffer from low structural damp-
ing which can become critical under certain opera-
tional conditions. Most problems appear on the
blades that receive almost 100% of the loading. In
particular the onset of stall plays a decisive role, not
only on stall-regulated machines, but also on pitch-
regulated ones around rated conditions. On the
other hand, the blades are made of composite
materials for which the knowledge on damping is
substantially inferior compared to steel or concrete
constructions. The damping of composite structures
depends on the ambient temperature. So depending
on the season, or the time during the day, the
structural damping of the blades can decrease
substantially. Furthermore, aging always degrades
the damping of composite structures. There is a
definite need to increase structural damping and
significant effort has been put recently on its
modelling in composite structures as well as in their
appropriate design. This is one direction of research
regarding the improvement of the stability char-
acteristics of modern wind turbines; see [168] for a
review of the current developments. Damping is
required to suppress the onset of vibrations
generated by the unsteady aerodynamic loads that
interact with the wind turbine structure. First, the
aeroelastic modelling of wind turbines is considered
and then on this basis the aeroelastic stability
problem is formulated and the relevant available
methods are reviewed.
Aerodynamic modelling involves the calculation
of the aerodynamic loads that also depend on the
dynamics of the system. The flap and lag vibrations
are interpreted by the incoming flow as a change in
the effective angle of attack indicating strong
aeroelastic coupling. All the currently available
aeroelastic tools for wind turbines use 2D dynamic
airfoil data models such as the Beddoes–Leishmann
or the ONERA model. They provide aerodynamic
coefficients as functions of the aerodynamic state
variables that satisfy appropriate dynamic equa-
tions [17,169]. In either model the equations are 2D
and so they are applied on a strip-by-strip basis,
meaning that there is no interaction in the radial
direction. This fits well with BEM modelling of the
overall aerodynamic analysis. In fact the following
presentation is referred to the BEM context, since
most existing stability models are based on it. If a
potential flow model is applied, then the attached
part of lift can be associated to potential lift while
the rest is simply added [163]. If the potential flow
model is enhanced with a separation model there
should be correlation on both aspects [58]. How-
ever, the latter has not been applied yet in
aeroelastic computations.
By considering a blade section at a radial position
as shown in Fig. 1 the local aerodynamic loads
acting on this section of the blade can be written as
follows, since the lift is perpendicular and the drag
parallel to the relative wind:
py ¼ L sin f  D cos f,
pz ¼ L cos f þ D sin f,
M y ¼ M, ð4:1:1Þ
where L, D are the lift and drag forces, M is the
pitching moment, f is the local flow angle with
respect to the rotor plane. Once the induced velocity
has been determined, the local effective incidence a
and relative flow velocity can be calculated as shown
in Section 2.1. They will depend on the elastic
deformations and the velocities, the rigid body
motions and, if appropriate, orientation changes, as
in pitch and yaw variations. In order to obtain a
unified description of the coupled aeroelastic
problem a so-called aeroelastic finite element is
defined [170] which, in addition to the elastic DOF,
the aerodynamic state variables added. For example
in the ONERA model there will be in total eight
additional DOF. Because the dynamic equations for
the aerodynamic state variables only can be first
order in time, in order to produce a consistent
coupled system, they can be differentiated as in
[171]. Therefore they can be assembled together
with the dynamic equations in the form of (3.2.10)
where the vector of the unknowns is further
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
extended to include the additional aerodynamic
DOF. Another approach used in [172] is to
introduce the modal expansion of the structural
DOF at stand still, which better conditions the
coupled system.
On the above basis it is possible to formulate the
aeroelastic stability problem for wind turbines.
Within the context of linear theory, stability
boundaries are obtained by examining the evo-
lution of small perturbations with respect to a
steady state or a periodic solution acting as
reference. To this end the original equations are
linearized with respect to the reference state and
the resulting system solved with respect to the
fluctuations:
Mðuref Þ  u^€ þ Cðuref Þ  u^_ þ Kðuref Þ  u^ ¼ Q;
where M, C and K depend only on the reference
solution. If the matrices do not depend on time,
(4.1.2) is reformulated into a first-order system:
n oT
y_ ¼ D  y þ b; y ¼ u^ ; u^_ . (4.1.3)
The eigenvalue analysis of D provides the eigen-
frequencies and damping of the system [173]. This is
possible when the reference state corresponds to a
steady solution as in the case of an isolated blade.
However, when analysing the complete wind
turbine configuration, due to the rotation of the
blades, the reference state should correspond to a
periodic equilibrium state with reference to the rotor
speed, which is assumed constant. In this case the
corresponding theoretical framework is Floquet’s
theory, which for a large system is computationally
heavy [174]. If the blades are identical and the
number of blades NX3, which is the most frequent
case, it is possible by means of multi-blade
transformation to eliminate the periodic coefficients
in the coefficients of (4.1.2) and therefore be able
to still use eigenvalue analysis [175]. In this context
the non-linear equations of the system are inte-
grated in time until such a periodic state is attained.
In the case of an unstable situation, because the
time domain response will contain significant
components in all of the basic eigenfrequencies of
the system, the reference state is obtained by
truncating the response so as to retain only its 1
and N/rev components. To this end all DOF in the
rotating system qm ; m ¼ 1; . . . NX3, are reformed:
qm ¼ q0 þ qc cos cm þ qs sin cm , where cm ¼ Ot þ
ð2p=NÞðm  1Þ is the corresponding azimuth loca-
tion. Next the equations for the blade DOFs are
309
PN
rearranged by PNapplying the operators ð1=NÞPN m¼1
ð  Þ, ð2=NÞ m¼1 ð  Þ cos cm and ð2=NÞ m¼1 ð  Þ
sin cm . This is performed after the periodic solution
has been obtained, over one period of rotation. The
resulting equations will be in the non-rotating frame
and contain only higher harmonics (3/rev and
higher). By averaging over one period, the final
constant coefficient system is obtained in the form
of (4.1.3). For the blades y will contain: the
corresponding q0 ; qc and qs DOF and their time
derivatives for both the purely structural and the
aerodynamic DOF [171]. In the above approach it is
worth noticing that the averaging procedure will
force the reference state to appear in the equations.
This is an important aspect when dealing with non-
linear systems.
(4.1.2) Stability of wind turbines has been considered
systematically in Europe in the late 1990s [176], while
today there is an on-going activity under the EU
funded project STABCON. Several linear stability
tools have been produced along the lines described
above and good correlation has been obtained
amongst the different codes, see [177] and the
references cited. Inevitably linear theory is approx-
imate so the results it produces are subject to cross
checking. Clearly flutter measurements are difficult
to obtain because of the involved risk, therefore it is
indispensable to rely on theoretical developments
which will be discussed in the next section.
4.2. Aeroelastic coupling: linear vs. non-linear
formulations
Linearization of the aeroelastic equations cer-
tainly offers computational efficiency. However, it is
not always appropriate. For the structure, linear
theory requires that the deformations and displace-
ments are small; an assumption not always valid.
For the aerodynamics and its coupling with the
structure, linearization will suppress some depen-
dencies involved in (4.1.1) and therefore influence
the estimation of the aerodynamic damping. In this
connection the use of semi-empirical unsteady
aerodynamic models introduces still unresolved
uncertainties. Furthermore there are cases in which
the multi-blade transformation is not applicable e.g.
when blades are non-identical, or when the rota-
tional speed varies.
Retaining fully the non-linearities of the aero-
elastic problem has the following consequences: the
structure is considered at its deformed state, the
coupling conditions among the components are
 ARTICLE IN PRESS
310 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
taken in their full form, and the aerodynamic
equations are not simplified. The last point is
important with respect to the best possible estima-
tion of the aerodynamic damping within the
margins offered by the semi-empirical models used
in the currently available aeroelastic tools. In this
respect multi-body analysis, which has been
adopted in most of the existing stability tools
[161,178–181], offers this option by construction,
provided that the structural model considers the
construction at its deformed state and the code
takes this into account in the definition of the local
aerodynamics.
Under such conditions, the information of the
stability characteristics of the wind turbine are
contained in the time signals of the loads calculated
through a non-linear time domain simulation. There
exist two types of methodologies for retrieving this
information: the one is based on the work-based
approach [176], while the other is using signal-
processing techniques. In the work-based method
any mode shape is excited and loads are recorded
until a steady periodic state has been reached. Then
the aerodynamic damping is estimated by the work
of the aerodynamic loads over the last period under
the assumption that there is no energy interchange
between the modes through the aerodynamic loads.
In the signal processing approach, the system is
harmonically excited for a finite duration. Then the
transient response following the abrupt termination
of the excitation is recorded wherefrom the aero-
elastic damping and frequency of the specific mode
are determined. There are two different methods,
both well documented in the literature [182–184]
that can be used: the moving block method and the
method of Hilbert transform. The moving block
method is a FFT based method, commonly used in
rotorcraft applications [182,184]. The transient
response amplitude is computed on a block of data
using a FFT calculation. The block is then moved
forward by a single point in time and the computa-
tion of the transient response amplitude is repeated.
The linear fit for the slope of the natural logarithm
of the sequence of response amplitudes in time
(peak plot) provides the damping (Fig. 21). In the
Hilbert method, the transient response y(t) and its
Hilbert transform yðtÞ ~ are used to ffi define the
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
envelope signal AðtÞ ¼ y2 ðtÞ þ y~ 2 ðtÞ [183,184].
For a transient response typical of a viscously
damped system a line can be fitted to the logarithm
of this envelope. The slope of this line, as in the case
of the moving block method, gives the damping
(Fig. 22). Comparing the two, it has been found that
the Hilbert method has certain advantages over the
moving block method. In particular the Hilbert
damping analysis provides an estimate of the
decaying envelope signal so it can be used in
assessing the non-linear damping characteristics of
a mode. Therefore it is not limited by the assump-
tion of the viscous damping as the moving block
method does. Moreover, experience has shown that
the Hilbert method is more efficient in calculating
the damping of spectrally close modes [163].
4.3. Examples of time simulations and instabilities
In this section the challenges in aeroelastic design of
wind turbine rotors will be addressed and in particular
the phenomenon of aeroelastic instability of wind
turbine rotors will be explored by showing examples
of simulation results on a wind turbine design.
Back from the beginning of development of
modern wind turbine rotors in the late seventies
there has been concern about the problems that
could arise due to aeroelastic instability. In parti-
cular the use of the stall regulation principle was
uncertain as it was foreseen that a flapwise
instability (so-called stall flutter) would occur when
operating in the stall region due to the negative
slope of the CL vs. a curve. However, it turned out
that the flapwise instability during operation in stall
was not the most critical problem, but instead the
edgewise instability, resulting in edgewise blade
vibrations. The first experimental evidence of this
instability was seen in the mid nineties and initiated
considerable research activities in order to explore
the phenomenon and provide practical solutions.
Below, is mainly focused on this instability as this
serves as a good example to illustrate the problem of
aeroelastic instability in more general terms.
In the more recent wind turbine designs the
regulation of the turbines has shifted from stall
regulation to pitch control where the operational
range for the flow over the blade moves to low angle
of attack at high wind and thus away from the stall
region. This has almost removed the instability
associated with stall during operation but it is still
so that the edgewise blade modes are aerodynami-
cally low damped on the pitch regulated turbines.
A major instability problem on the modern
turbines is seen when the rotor is parked or idling
at very low RPMs at wind speeds above stop wind
speed, which typically is around 25 m/s. Again it is
typically edgewise dominated instability problems
 Δt

-1E +06

-1.5E+06

-2E +06

-2.5E+06

-3E +06

tower top tilt bending moment (Nm)

-3.5E+06
35 40 45 50
time (sec)

data
ARTICLE IN PRESS

3E+08 19.4

19.35
2.5E+08
19.3 -ζ Ω n
2E+08
19.25
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

1.5E+08 19.2

19.15
1E+08
19.1
log of response amplitude

5E+08
19.05

0 19

FFT of the tower top tilt bending moment (Nm)

1 2 3 4 0 0.5 1 1.5 2 2.5 3
frequency (Hz) t (sec)
311

Fig. 21. Schematic description of the moving block damping analysis.

 ARTICLE IN PRESS
312 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

1.5e+06
transient signal
envelope signal
envelope signal exp. fit

tower top tilt bending moment (KNm)

1e+06

500000

-500000

-1e+06

-1.5e+06
2 4 6 8 10 12 14
time (s)

Fig. 22. Envelope signal calculation of the tower top tilting moment transient response using the Hilbert transform method for the same
signal as in Fig. 21.

EDGEWISE BLADE VIBRATION INSTABILITY EDGEWISE BLADE VIBRATION INSTABILITY

--MEASUREMENT
1000 1000
MOMENT [UNCALIBRATED]

MOMENT [UNCALIBRATED]

900
EDGEWISE BLADE ROOT

EDGEWISE BLADE ROOT

800 800
700
600 600
500
400
400
300
200
200
0 100
0
-200 -100
0 50 100 150 200 250 300 200 210 220 230 240 250 260
Time [sec.] Time [sec.]

Fig. 23. Measured edgewise blade root moment on a stall regulated rotor at high wind. Detail of the time track to the right.

related to stall of the blade. A simulation example nineties on stall-regulated rotors with a diameter of
will be shown later to illustrate this problem. 35–40 m. An example is presented in Fig. 23 and it is
Finally, the flutter instability will be addressed. It seen that the amplitude of the edgewise blade root
seems that the increased up scaling of the turbines moment (which at steady conditions vary sinusoidal
has led to rotor and blade designs where the flutter with 1p due to the gravity) increases 2–3 times due
speed is not so far away from normal operational to instability during operation in stall. The experi-
speed. Also this instability will shortly be illustrated. mental evidence of the edgewise instability led to
considerable research on this subject and a major
4.3.1. Edgewise blade vibration instability European project ‘‘Prediction of Dynamic Loads
As mentioned above the first experimental and Induced Vibrations in Stall’’ funded by the EU
evidence of this instability was seen in the mid was carried out in the period from 1995–1998 [176].
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 313

The origin of the instability is simple. If a rotating
airfoil section is harmonically translated along an
axis xB and the direction of this axis yRB relative to
the orientation xR of the rotor plane is varied, the
aerodynamic damping coefficient for the section
varies considerably, see Fig. 24 from [185]. For low
yRB which means in-plane or edgewise vibration
direction the damping coefficient is negative even at
low wind speed. For vibrational directions close to
901, which corresponds to out-of-plane or flapwise
direction the damping coefficient is strongly depen-
dent on the inflow wind speed. It is highly damped
at low wind but close to zero or negative damped at
high wind.
On a complete rotating blade the direction of
vibration depends on the structural design of the
blade as well as of the complete turbine structure.
The movement of the tip section of the blade will
now no longer necessarily be along a straight line.
However, typically the tip section of the blade in the
first flapwise mode will be on a path almost
perpendicular to the rotor plane and in the 1st
edgewise mode it will almost be in-plane. It is thus
expected that the basic damping characteristics of

Fig. 24. The aerodynamic damping coefficient c_xx_b for a rotating airfoil section as function of vibration direction yRB and at three
different inflow velocities, from [185].

Fig. 25. Damping characteristics computed for a rotating, isolated blade using the linear aeroelastic stability tool HAWCStab [186].
 ARTICLE IN PRESS
314 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
the whole blade will be comparable with the
characteristics of a single airfoil section. Using the
linear stability tool HAWCStab the damping
characteristics for a rotating blade can easily be
derived and the computed damping characteristics
for a rotating 19 m blade (500 kW rotor) corre-
sponding to the blades of the rotor where the
experimental instability was shown in Fig. 23, are
shown in Fig. 25 from [186]. It is seen that the
damping for the first flapwise mode varies from
highly positive values at low wind speed and to
values close to zero at 14 m/s and then increasing
slightly at higher wind speeds. It should be noted
that the influence of using unsteady blade section
aerodynamics is shown (modelled with the Beddoes
Leishman model in the present case) and that this
has considerable influence on the aeroelastic damp-
ing. Generally, the unsteady aerodynamic effects
increase the damping. The damping for the first
edgewise mode is seen to decrease gradually with
increasing wind from being slightly positive at low
wind to slightly negative at high wind. Finally, it
should also be noted that the damping shown is the
total damping including the structural damping
which in the present case is around 2% for the first
flapwise mode and 3% for the first edgewise mode.
The final step in model complexity is now
obtained by going to a full aeroelastic model of
the turbine comprising the dynamics of the shaft,
the nacelle and the tower. The aeroelastic stability
Fig. 26. Aeroelastic damping computed with the aeroelastic stability HAWCStab [186]. Three modes have negative damping at high wind
speed.
results using HAWCStab is for the complete wind
turbine, see e.g. Fig. 26 from [186], showing the
aeroelastic damping for the first 10 mode shapes
(numbered from lowest frequency). It is seen that
two mode shapes are negatively damped at high
wind speed and that one mode shape is close to zero
and slightly negatively damped at the highest wind
speed. For comparison, time simulations with the
aeroelastic code HAWC [187,188] is performed on
exactly the same turbine model at a wind speed of 8
and 16 m/s, respectively, in order to see if the
instability at high wind speed predicted by the linear
stability tool can be confirmed by the HAWC
model, with non-linear aerodynamics. The com-
puted edgewise blade root moment shown in Fig. 27
confirm the instability at high wind and compares
very well with the measured instability shown in
Fig. 23. It is thus demonstrated that the edgewise
instability problem can be predicted in both time
simulations and using linearized stability analysis.
However, as will be demonstrated below, the
edgewise vibration instability is much more complex
than just edgewise vibrations of the individual
blades but comprises the dynamic characteristics
of the whole turbine. This understanding is vital for
the development of design solutions preventing the
instabilities.
In order to analyse the instability in more details
the use of parallel aeroelastic time simulations on a
turbine with a stiff structure is made. In this case the
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 315

loads on the different components will mainly with considerable less difference between the re-
reflect the direct load input from the external sponse of the stiff and flexible turbine indicating low
aerodynamic forces and from gravity forces. It damping but not an instability.
should be mentioned that the simulations are Analysing further the results of the stability
performed at two average wind speeds of 8 and analysis with HAWCStab as shown in Fig. 26 it
16 m/s, respectively, and with a turbulence intensity can be derived from the results of the model that the
of 12%. three mode shapes that are negatively damped or
A power spectral density analysis of the simulated very low damped at 16 m/s are:
edgewise blade root moment shows a distinct peak
at 3.07 Hz which is close to the frequency of the first
edgewise mode for a single, rotating blade which is
mode no. 1 with frequency 0.77 Hz and damping
3.19 Hz, see Fig. 28. The response at 16 m/s of the 6.05%,
flexible turbine at this frequency is seen to be several
mode no. 3 with frequency 0.93 Hz and damping
orders of magnitude bigger than the response of the 4.33%,
stiff turbine indicating an instability situation.
mode no. 7 with frequency 2.73 Hz and damping
Somewhat the same tendency is seen at 8 m/s but 1.59%.

Fig. 27. Time simulation with the aeroelastic code HAWC on a 500 kW stall regulated rotor at two wind speeds, 8 and 16 m/s.

Fig. 28. Comparison of power spectra of edgewise blade root moment for a flexible and a stiff turbine, respectively, at 8 and 16 m/s.
 ARTICLE IN PRESS
316 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

The numbering in HAWCStab of the modes is
with increasing mode number with increasing
frequency starting with the lowest frequency. Mode
no. 1 and no. 3 are both modes where lateral tower
bending is the main motion together with edgewise
bending of the three blades and torsion of the main
axis. The edgewise movement of the blades are all in
phase and the difference between the two modes is
the direction of the edgewise bending relative to the
tower bending. Comparing again the results of the
linear stability analysis with the time simulation
results, the instability of the lateral tower bending
dominated modes is confirmed by the spectra of the
tower top and tower bottom lateral bending
moments, Figs. 29 and 30. A distinct peak at
0.79 Hz is seen in both tower top and tower bottom
spectra at 16 m/s and this peak cover probably also
a considerable content at the 0.93 Hz, which was the
frequency of mode no. 3.
It can be noticed that the HAWCStab results do
not contain an unstable mode with a frequency
around 3.07 Hz (the edgewise instability) as seen in
the time simulation results. However, mode no. 7 is
in fact the response on the non-rotating turbine
structure from the edgewise vibrations at 3.07 Hz.
The edgewise vibrations in the present case is a so-
called backward whirling, edgewise mode where
there is a phase shift of 1201, between the movement
of the blades. This whirling mode results in shaft
and tower bending but with a frequency shifted 1p

Fig. 29. Comparison of power spectra of tower top lateral moment for a flexible and a stiff turbine, respectively at 8 and 16 m/s.

Fig. 30. Comparison of power spectra of tower bottom lateral moment for a flexible and a stiff turbine, respectively, at 8 and 16 m/s.
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
(p is the rotational frequency) up and down,
respectively, compared with the frequency in the
rotating system. As 1p in the present case is 0.45 Hz
a peak in the spectra of tower bending at a
frequency of around 3.07–0.45 Hz equal to 2.62 Hz
is expected, which is also seen in Figs. 28 and 29.
However, in the HAWCStab results a slightly
higher frequency 2.73 Hz was seen for this mode.
The presented example of the edgewise blade
vibration instability has shown that the low aero-
dynamic damping of an airfoil section undergoing a
motion along a path close to the chordwise direction
can cause instability of quite different modes on the
turbine at the same time. The design challenges with
the objective to minimize the influence of low
aerodynamic damping or instability are thus big as
e.g. changes in tower design could lead to e.g.
edgewise vibrations on the rotor. A number of
different methods to reduce the risk for edgewise
vibrations on stall-regulated turbines have been
investigated [176,186] and comprises e.g. changes of
the stalling characteristics of the airfoils by so-called
‘‘stall strips’’ as well as structural design of the
blades to achieve optimal vibrational directions.
4.3.2. Instability problems of parked rotors
The turbines are normally designed to operate up
to a certain maximum wind speed and above this
wind speed the turbines are shut down. In the
stopped conditions the rotors can be completely
parked or they can idle with low rotational speed
depending on the actual design of the control
system. As part of the certification of the turbine
it must be shown that the turbine can withstand the
Fig. 31. Simulation of a parked rotor at 50 m/s and a yaw error of 601. Time track of the edgewise blade root moment to the left and a
power spectrum of the same signal to the right.
317
wind loads at extreme wind speed conditions with
parked rotor and the wind coming from any
direction in the case that the yawing system of the
turbine is not functioning.
As an example, the simulation of a rotor, parked
at a wind speed of 50 m/s and a yaw error of 601, is
shown in Fig. 31. The turbine is the same as above
and the time track of the edgewise blade root
moment indicates low damping as the amplitudes
vary much in time. To the right in Fig. 31 is shown a
power spectrum of the same signal and it shows that
the blade mainly vibrates in two modes; mode 3 at a
frequency of 0.93 Hz which was described above
and the first edgewise blade frequency at about
3.19 Hz. For the present inflow direction the
vibrations are not unstable but it has to be
documented for all inflow directions.
4.3.3. Flutter instability
The last example of an instability is flutter. This is
a well-known instability from the aircraft industry
but has not yet been a problem on wind turbines
and has probably not been seen on commercial
turbines. However, with the increasing size of the
blades it seems that the flutter speed decreases due
to increasing structural flexibility of the blades and
not least the torsional frequency decreases. There-
fore, it is a good idea to include a flutter speed
calculation in the design verification for e.g. 50 m
blades and above.
Flutter involves two DOF of the blade; torsion
and translation. The flutter speed decreases when
the frequency of these two DOF approach each
other. For 50 m blades the frequency of the first
 ARTICLE IN PRESS
318 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
flapwise mode is typically slightly below 1 Hz and
the frequency of the first torsional mode will
typically be in the range from 5 to 8 Hz. However,
it has been seen that the flutter instability can occur
by a coupling of the second flapwise mode and the
first torsional mode of the blade and of course this is
also a reason for a decreasing flutter speed on the
bigger blades. Another important design parameter
for the flutter instability is the centre of mass in the
blade sections relative to the centre of the elastic
axis. As the centre of mass moves away from the
elastic axis in the direction of the trailing edge the
flutter speed decreases.
An example of a simulated flutter instability is
shown in Fig. 32. The blade does not represent an
industrial blade but has selected structural para-
meters, which give a rather low flutter speed. The
second flapwise mode for the blade is slightly above
2 Hz and the first torsional frequency is around
6 Hz. Further the centre of mass is positioned 10%
of the chord length behind the elastic axis. The
example is mainly intended to show the character-
istics of the flutter instability, which is completely
different from the instabilities that have been seen
so far. In the simulation the rotor is free to speed up
as the generator is disconnected. When the flutter
speed is reached the instability develops within 1–2 s
and this is completely different from e.g. the
edgewise blade instability that can build up over a
minute or more. This characteristic means the blade
probably will be damaged immediately if the flutter
speed is reached. A further consequence of this fast
development is that the flutter can occur at a lower
rotational speed of the rotor if there is a yaw error.
This is demonstrated in the right part of Fig. 31
Fig. 32. Example of flutter instability on a 50 m test blade where the rotor is free to speed up. To the left normal inflow whereas to the
right there is a yaw error of 501.
where there is a yaw error of 501. Here, the
rotational speed at flutter instability is 13% lower
than if there is no yaw error and the wind speed is
just 8 m/s. If the for example the yaw error of 501
occurred at around the stop wind speed of 25 m/s
the rotational speed at the flutter instability would
decrease even further.
5. Present and future developments of aeroelastic
models
The improvement of existing aeroelastic codes
and the development of new aeroelastic models are
highly influenced by the design trends of new wind
turbines and trends in the siting of the turbines
because this determines the needs for new capabil-
ities of the models.
5.1. Areas with influence on the development of
aeroelastic models
5.1.1. Influence of up-scaling
So far the most important design trend has been
the up-scaling which within the last 10 years has
increased the maximum size of the mass produced
turbines with a factor of 10 from about 500 kW with
a rotor diameter of about 40 m to 5 MW turbines
with a rotor diameter of 120 m. The newest turbines
are all pitch controlled with variable speed and
typically with some type of cyclic pitch for load
alleviation. An accurate modelling of these new
flexible turbines has increased the requirements to
simulate complex coupled modes where the inte-
grated flexibility of the tower, of the drive train and
of the blades and in interaction with the control, is
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
important. Also non-linear effects from consider-
able deflections of the blades and the tower are
becoming important and the aeroelastic stability
must be predicted with a good accuracy. Some
typical frequencies for a 5 MW turbine could be:
First tower bending frequency 0.2–0.25 Hz
First flapwise frequency 0.8–1.0 Hz
First edgewise frequency 0.9–1.1 Hz
First blade torsional frequency 5.0–7.0 Hz
The decrease in the fundamental turbine frequen-
cies arising from the up-scaling has the effect that
excitation from the turbulent inflow has increased as
the power spectrum of the turbulence peaks at
around 0.05 Hz.
5.1.2. Siting of the turbines
The major part of all new turbines is placed in
small or bigger groups and in some cases in wind
farms with more than 100 turbines. This means that
wake operation is part of the inflow conditions,
which has to be simulated in the derivation of the
total design loads of the turbines. Load alleviation
non-linear structural dynamics,
by cyclic or individual pitch controls is important
calculation of induction and its dynamics,
for such inflow conditions and the aeroelastic
wake operation,
models should be suited for this.
derivation of airfoil data for aeroelastic simula-
Still the major part of turbines are set up on land
but the offshore part will increase, at least if a 5–10
complex inflow,
years interval is considered. On land, the best sites
aerodynamics of parked rotors and
have been used in many countries and therefore
off shore turbines including floating turbines.
more and more complex sites will be used. This
means that rather complex inflow situations must be
modelled as for example complex shear and non-
uniform turbulence over the rotor disc.
5.1.3. Future trends in turbine design and siting
It seems that the speed of up-scaling might be
slowed down for some years as the industry wants
to be more focused on turbine reliability. One way
to improve this is to have the individual turbine
models on the market for more years than seen in
the past.
However, the increasing offshore development
will support the up-scaling tendency as the tran-
sport and erection of big turbine components is no
longer a major problem. Further, the higher
foundation costs offshore will also support the up-
scaling trend.
Extending the possible offshore sites to deeper
water, new foundation types, typical with multiple
frames, will be developed and the aeroelastic codes
319
should be able to handle these structures. A further
step in the offshore development is the floating
turbines, either as single floating turbines or more
turbines on the same, floating frame.
Finally, the offshore market could have the effect
that the two bladed turbine with a teetered down-
wind rotor again will be considered as an alternative
to the three-bladed machines. The main barrier for
the development of two bladed turbines has been
the low frequency noise from the blade passage of
the tower shadow. This will be of less importance
offshore as the noise restrictions are much lower
here.
5.2. Areas of development in present and new codes
Influenced by the continuously increasing require-
ments to the capabilities of the aeroelastic models as
described above, existing models are being further
developed and improved and new codes are written.
Below some important areas of development will be
discussed.
tions,
5.2.1. Non-linear structural dynamics
So far almost all aeroelastic codes have contained
the model assumptions of small deformations and
rotations, but the increased flexibility of the turbines
have made uncertainty about the validity of this
assumption for the new designs. The influence of
non-linear effects on the dynamic response of a
turbine was treated in a recent paper [189] and
different effects were considered. For example
characteristics (frequency and damping) of the first
edgewise mode will typically change as function of
increased flapwise deflection because the edgewise
bending will couple more and more with the
torsional mode of the blade. Besides increased
influence to non-linear effects from the flexibility
of the turbine components the new foundations
used offshore and in particular floating foundations
can contribute significantly to big deflections and
related non-linear effects.
 ARTICLE IN PRESS
320 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
To overcome this uncertainty about the impor-
tance of non-linear effects a few new non-linear
aeroelastic codes have now been developed. Siemens
wind power has developed a non-linear code
BHawC based on co-rotaing elements [190] while
the new non-linear code HAWC2 [191] developed
by Risoe is based on multibody dynamics.
The trend of including non-linear effects will
continue in the future and codes based on modal
expansion techniques will be developed to include
more modes as e.g. the blade torsional mode.
5.2.2. Calculation of induction and its dynamics
The unsteady aerodynamic air loads are directly
dependent on the computed induction at the same
point on the blade through the angle of attack.
Therefore, an accurate prediction of induction is of
crucial importance.
As the rotor disc has become bigger and bigger
relative to the scales in the turbulent inflow,
increasing variation in inflow conditions over the
rotor disc are seen. Wind shear contributes to this
variation in inflow. The inflow variations results in
likewise considerable variation in the loading
expressed through the local thrust coefficient CTlocal
over the rotor disc and thus also in induction. The
induction is therefore even in normal operation
highly dynamic and the induction model should be
adapted to these conditions. As an example, Fig. 33
Fig. 33. Example of dynamic loading at a point close to the tip of a blade on an 80 m diameter turbine in turbulent wind, from [192].
(reproduced from [192]) shows the variation of the
local thrust coefficient at the outer part of the blade
on an 80 m diameter turbine in normal operation.
Another source contributing to the load varia-
tions and thus to the dynamic induction is the
eigenmotion of the blades. This part is also
increasing due to up-scaling and due to the more
flexible designs. As an example a blade with a
flapwise frequency of 1 Hz as mentioned above and
vibrating with an amplitude of 1 m will experience a
relative velocity component perpendicular to the
chord of around 76 m/s in maximum. This value
can directly be compared with the variations in
inflow and is thus considerable. Different develop-
ments to improve the computation of induction
have been seen. One type of modelling is the
coupling of a numerical actuator disc model to the
structural part of the aeroelastic model. So far such
models have mainly been used to verify the accuracy
of BEM modelling. As an example the aeroelastic
code HAWC at Risoe with a BEM-type induction
modelling in the standard version, has been devel-
oped in another version HAWC-3D where the
induction is computed with a 3D actuator disc
model and used e.g. to investigate the accuracy of
yaw modelling [142,193], see Fig. 34 from [142]. In
the coming years aeroelastic codes coupled to an
actuator disc or actuator line model for computa-
tion of induction will certainly be further developed
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
Fig. 34. Influence of yaw on the local inflow angle. The curves show the difference in local inflow angle at yawed operation at 451
compared with non-yawed operation. The HAWC model has a BEM induction model, HAWC-3D a 3D actuator disc model and
EllipSys3D is a full 3D Navier–Stokes simulation, from [142].
but probably mainly used for research and for
verification of simpler codes and not so much in the
aeroelastic codes used by the industry due to the
much longer simulation time.
However, besides a direct coupling of an actuator
disc or line method to an aeroelastic code in order to
compute the induction at each time step there are
other ways to utilize the capabilities of an actuator
disc model. One example is to compute the
induction characteristics of a rotor within its wind
speed operational interval with an actuator disc and
then afterwards use these quasi-steady induction
characteristics in aeroelastic simulations for the
rotor. The BEM method in the new HAWC2
aeroelastic code developed at Risoe has been
implemented in such a way that this method can
be used. The method is shortly first to compute the
local induction as function of local loading for the
actual rotor in a number of calculation points ðn 
mÞ over the rotor with an actuator disc:
aðrn ; ym Þ ¼ f ðCT ðrn ; ym ÞÞ.
Afterwards, these induction functions are then used
in the BEM method during time simulations on the
rotor instead of the unique, single relationship from
1D momentum theory, C T ¼ 4að1  aÞ.
321
In fact, one of the main deficiencies of the BEM
method is that this induction formula concerns the
whole rotor disc. Deviations are seen in regions with
considerable radial variation in the loading, which
means in the tip or root region. Another example
where the momentum induction formula does not
hold is if the rotor disc is not plane, e.g. coned
rotors or rotors where the blades bend considerably.
5.2.3. Wake operation
Most turbines are now set up in clusters or wind
farms and this means that such turbines in part of
their lifetime operate in wake from one or more
turbines. In the international standard for design of
turbines IEC 61400-1 [2] the increased loading from
operation in wake can be taken into account by
using an increased, effective turbulence which
depends on a number of parameters in the
considered wind farm such as e.g. wind turbine
spacing. However, it has turned out that in cases
where more detailed knowledge of the increased
loading on the different turbine components from
wake operation is needed, another more detailed
aeroelastic modelling is needed than just increasing
the turbulence. One such aeroelastic modelling has
been developed over the past few years [194,195].
The main components in the modelling is: (1)
 ARTICLE IN PRESS
322 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330

computation of the wake deficit from the up-stream
turbine with an actuator disc model; (2) meandering
of this deficit from the big scales in the turbulence
and (3) additional turbulence within the wake. One
of the advantages of this new modelling compared
to the method using an effective turbulence is that
both the mean yaw loads as well as the yaw
dynamics compare well with measurement. Fig. 35
from [194] shows the loads as function of wind
direction and full wake operation at a direction of
around 2101. The highest mean yaw loads occur in
half wake operation at around 1951 and 2201.
Recently, the wake resulting from the interaction of
several turbines in a row was computed in [196] by
combining large eddy simulation of the NS equa-
tions with the actuator line methodology.
It is expected that the more detailed aeroelastic
modelling of wake operation will be developed
considerably in the coming years because such
modelling is necessary for the development of
advanced control algorithms adapted for load
reduction in wakes. The development of the new
detailed models can be in the direction like the
model described above but also using actuator line
models or full 3D rotor models to compute the wake
characteristics, which can be fed into an aeroelastic
simulation.
5.2.4. Derivation of airfoil data for aeroelastic
simulations
The airfoil data used in aerodynamic and aero-
elastic simulations are of crucial importance for the

Fig. 35. Measured and simulated loads at 8 m/s (crosses and full lines) and at 10 m/s (squares and dashed lines) on a 80 m rotor during
wake operation (full wake at around 2101 on x-axis) using a new aeroelastic wake simulation method taking into account the wake deficit
meandering, from [194].
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330 323

accuracy of such simulations. This was clearly seen
in the result of the NREL blind test [137] where the
difference in the airfoil data used in the individual
simulation models probably were the most signifi-
cant factor for the big variation between the output
of the different simulations. However, it should be
noted that this result is not a representative
indicator of the uncertainty in the industry on
aerodynamic and aeroelastic results on new turbines
as considerable experience has been learned from
older rotor designs in order to adapt airfoil data sets
so that computed rotor power and loads compare
well with measurements. The drawback is that this
introduces conservatism in the design and for
example can hinder the use of new airfoils. There-
fore, methods and models to derive or correct airfoil
data to be used in aeroelastic simulations have been
a key issue for many years in the wind energy
research community and will also be it in the near
future. 2D wind tunnel airfoil data has so far been
the common starting point to set up an airfoil data
set for simulations. In order to correct for so-called
3D flow effects and rotational effects some kind of
correction is typically applied [19–21,197]. The
corrections reflect the increased lift and drag that
was measured on different rotors in field rotor
measurements performed 10–15 years ago [198,199].
Full 3D CFD rotor computations later confirmed
clearly the tendencies of the 3D flow effects and
rotational effetcts in the form of increased lift in the
blade root region but also increased drag [137–139].
With the possibility of running full 3D simula-
tions on a rotor a new source for derivation of 3D
airfoil data sets for input in aerodynamic and
aeroelastic codes has emerged. Methods to extract
the airfoil data from 3D rotor computations have
been developed [200,201]. One example of such a
derived airfoil data set is shown in Fig. 36 from
[202], where a considerable increased maximum lift
on the inboard stations are seen. Within the next
few years it is expected that a considerable effort
will be on further development of the methods to
extract airfoil data from 2D wind tunnel data as
well as 3D CFD rotor data. So far the CFD
computations have mainly been used to extract
quasi-steady data but they will in the future also be
used to tune the parameters in the dynamic stall
models.
5.2.5. Complex inflow
Wind turbines are often set up in so-called
complex terrain, which means some kind of
mountainous terrain because the wind potential is
good at such places. However, the consequence is
that the inflow conditions can vary considerably
from turbine to turbine due to local variations of
the terrain. As a result extreme wind shear has been
seen to occur over the rotor disc of turbines placed
in such terrain and likewise the turbulence intensity
can be extremely high.
The fast development of CFD codes has now
made it possible to simulate the flow in details
over the terrain spanned by a wind farm in such
complex terrain. The flow data in the form of wind
shear and turbulence can be used as input in
aeroelastic simulations. There is thus a basis for a
simulation tool to micro site turbines not only with
respect to energy production but also loads is

2 0.5
r=14%
0.45 r=28%
1.8 r=42%
0.4 r=56%
1.6 r=69%
0.35 r=89%
r=94%
1.4 r=97%
0.3
C1

Cd

1.2 0.25

1 r=14% 0.2
r=28% 0.15
0.8 r=42%
r=56%
r=69% 0.1
0.6 r=89%
r=94% 0.05
r=97%
0.4 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
α α

Fig. 36. Extracted 3D airfoil data in the form of CL, CD data as function of angle of attack at different radial stations, from a 3D CFD
rotor computation. Figure from [202].
 ARTICLE IN PRESS
324 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
expected. Thus type of modelling is certainly of
high interest for the wind turbine industry
because considerable costs to repair turbines due
to extreme inflow conditions have been experienced
in the past.
5.2.6. Aerodynamics of parked rotors
It is a common procedure to shut the turbines
down at high wind, typically at wind speeds in the
range from 20 to 25 m/s. Ultimate loads on parts of
the wind turbine can occur during standstill condi-
tions at very high wind speed whereas ultimate loads
on other components will occur during operation
below the stop wind speed.
The aerolelastic codes are therefore also used to
compute the turbine response during standstill
conditions and it is particularly the computation
of the aerodynamic loads on the blades that is
uncertain. This is due to the requirement that the
loads on the parked rotor shall be investigated with
the wind from all directions. The models must
therefore be able to compute aerodynamic loads for
angle of attacks in the complete range from –180o to
180o. Furthermore the inflow is turbulent so that it
is highly unsteady aerodynamic loads and a severe
complication is that some inflow conditions often
lead to instability with the blade vibrating in a mode
with negative aerodynamic damping. In a recent EU
funded project ‘‘KNOWBLADE’’ different CFD
codes were used to explore some basic character-
istics of the blade standstill aerodynamics [115].
Because of the complexity of the aerodynamics it is
also believed that the use of CFD will be the main
path to explore the standstill aerodynamics in the
coming years.
Fig. 37. Different support structure concepts from [203].
5.2.7. Offshore turbines including floating turbines
Presently, there are considerable research and
development efforts on adapting aeroelastic models
to simulate offshore wind turbines, mounted on a
sub-structure standing on the sea-bed, see Fig. 37
from [203], or on a floating foundation.
Different approaches are seen in this develop-
ment. One line is characterized by a complete
integration of the standard aeroelastic model of
the turbine with a hydroelastic model of the sub-
merged supporting structure including the hydro-
dynamic loads, the wave loading and the soil forces
on the part of the support structure in the sea bed
[190,191]. In both these models the complete
structure is described with finite elements and with
engineering sub-models for computation of the
hydrodynamic loads.
Another line of modelling is to couple a
‘‘standard’’ aeroelastic model to a separate module
simulating the supporting structure, including wave
loads and hydro loads. This approach has been
presented by Repower [204] where the Flex5
aeroelatic code is coupled with a standard package
ASAS for computations of wave loads on off shore
structures. In the third development line the main
idea is also to use a ‘‘standard’’ aeroelastic code and
then couple it to a super element of the foundation.
This approach has been presented by Vestas [203]
where the foundation is modelled with a finite
element model but then decomposed to a super
element using the Craig–Bampton sub-structuring
scheme. Considerable efforts on further develop-
ments of the aeroelastic models for off shore
applications will be seen within the next years and
with more emphasis on deeper water installations.
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
6. Discussion
On a wind turbine there is a strong coupling
between the aerodynamic loads and the time-
dependent structural behaviour of the construction.
Statically a blade might change its twist and thus the
angle of attack when deflected. But also the angles
of attack are changed when the blades have a
velocity relative to the fixed ground. For instance if
the tower is moving upstream and everything else is
stiff it will be felt by the blades as an increased wind
speed and thus higher angles of attack will be
present along the blades. The aerodynamic response
will depend on how the lift and drag vary with the
angle of attack. In stall the lift will decrease yielding
a possible instability, whereas for attached flow the
lift will increase and thus creating a higher load seen
by the blade opposite the movement indicating
stability. On a real wind turbine not only the tower
is vibrating but all other components as well, which
directly feeds back to the angle of attacks and thus
the loads that again alter the motion. To simulate
the aeroelastic response of a wind turbine, a non-
steady structural model including the inertia must
be made. Two methods, i.e. the method of virtual
work applied on modal shape functions and the
FEM, are addressed. To evaluate the aerodynamic
loads, the BEM is still the most widely used due to
its simplicity and computing efficiency. To obtain
realistic results some engineering adds-on are
necessary such as the Dynamic Wake, Dynamic
Stall and a yaw model, which are therefore
thoroughly described in this paper. A BEM method
relies, however, on airfoil data and the results are
therefore no better than the input. To avoid the
uncertainties of the engineering adds-on the Actua-
tor Line model can be used, since this method
resolves the physics of these models through the NS
equations. However, the actual blades are not
resolved and to estimate the aerodynamic lift and
drag airfoil data are still needed. To avoid airfoil
data one needs to solve the NS equations and
resolve the blades and the possible boundary layers.
This is extremely computationally costly and there-
fore this model is not likely to replace the BEM
method in the near future. However, the method
can be used to extract airfoil data that can be used
in the less advanced models. It is expected that the
Actuator Line model very soon will replace the
BEM method, since it contains less empirics. With
the AL model the wake is also a part of the solution
and therefore the effect from this wake on a wind
325
turbine placed further downstream can be calcu-
lated. These simulations are becoming very impor-
tant as wind turbines are grouped in large (offshore)
wind parks. The inviscid flow models are included in
this paper mainly for historical reasons, since they
have played an important role in determining basic
physical features. This type of model was e.g. used
to calibrate the time constants in the Dynamic
Wake model. Also a few Potential flow models exist
that model the induced velocities and the wake
behind a wind turbine by shedding vortex blobs
from the blade surface. This model is similar to the
AL model and may also in some aeroelastic codes
replace the BEM method.
References
[1] Last og sikkerhed for vindmollekonstruktioner, DANSK
STANDARD DS 472 [in Danish].
[2] IEC 61400-1 Ed.3 CD. 2. revision. Wind turbines, Part 1:
design requirements. Edited by IEC TC88-MT1, 25–26
May 2004.
[3] Hansen KS. Performance Measurements on two Danish
630 kW WECS. In: Proceedings of the WIND POWER’85,
SERI/CP-217-2902, 1985, p. 130–5.
[4] Hansen KS, et al. An evaluation of measured and predicted
fatigue loads for the Tjæreborg wind turbine. In: Proceed-
ings of ECWEC’93 Travemünde, Germany, 1993,
p. 579–82.
[5] Ahlström A. Aeroelastic simulation of wind turbine
dynamics. Doctoral thesis in Structural Mechanics, KTH,
Sweden, 2005.
[6] Glauert H. Airplane propellers. In: Durand WF, editor.
Aerodynamic theory. New York: Dover Publications; 1963.
[7] Rasmussen F, Hansen MH, Thomsen K, Larsen TJ,
Bertagnolio F, Johansen J, et al. Present status of
aeroelasticity of wind turbines. Wind Energy 2003;6:213–28.
[8] Quarton DC. The evolution of wind turbine design
analysis—a twenty year progress review. Wind Energy
1998;1:5–24.
[9] Hansen AC, Butterfield CP. Aerodynamics of horizontal-
axis wind turbines. Annu Rev Fluid Mech 1993;25:115–49.
[10] Shen WZ, et al. Tip loss corrections for wind turbine
computations. Wind Energy 2005;8(4):457–75.
[11] Snel H, Schepers JG. Joint Investigation of dynamic inflow
effects and implementation of an engineering method.
ECN-C–94-107, 1995.
[12] Schepers JG, Snel H. Dynamic inflow: yawed conditions
and partial span pitch control. ECN-C-95-056, 1995.
[13] Bramwell ARS. Helicopter dynamics. Paris: Edward
Arnold (Publishers) Ltd; 1976.
[14] Theodorsen T. General theory of aerodynamic instability
and the mechanism of flutter. NACA report 496, 1935,
p. 413–33.
[15] Øye S. Dynamic stall, simulated as a time lag of separation.
In: K.F. McAnulty, editor. Proceedings of the fourth IEA
symposium on the aerodynamics of wind turbines, ETSU-
N-118, 1991.
 ARTICLE IN PRESS
326 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
[16] Leishman JG, Beddoes TS. A semi-empirical model for
dynamic stall. J Am Helicopter Soc 1989;34(3):3–17.
[17] Hansen MH, Gaunaa M, Madsen HA. A Beddoes–Leish-
man type dynamic stall model in state-space and indicial
formulations. Risoe-R-1354(EN), 2004.
[18] Johansen J, Sørensen NN. Aerofoil characteristics from 3D
CFD rotor computations. Wind Energy 2004;7(4):283–94.
[19] Snel H, Houwink B, Bosschers J, Piers WJ, van Bussel
GJW, Bruining A. Sectional prediction of 3-D effects for
stalled flow on rotating blades and comparison with
measurements. In: Proceedings of the ECWEC 1993,
Travemunde, 1993, p. 395–9.
[20] Chaviaropoulos PK, Hansen MOL. Investigating three-
dimensional and rotational effects on wind turbine blades
by means of a quasi-3D Navier–Stokes solver. J Fluids Eng
2000;122:330–6.
[21] Bak C, Johansen J. Three-dimensional corrections of airfoil
characteristics for wind turbines based on pressure dis-
tributions. In: Proceedings of the EWEC conference,
Athens, 27 February–2 March 2006.
[22] Fuglsang P, Bak C. Status of the Risoe wind turbine
airfoils. In: Proceedings of the EWEC 2003 (CD-ROM
available from EWEA), Madrid, 2003.
[23] Shinozuka M, Jan C-M. Digital simulation of random
process and its applications. J Sound Vib 1972;25:111–28.
[24] Veers P. Three-dimensional wind simulation. SAND88-
0152, UC261, Sandia National Laboratories, NM, 1988.
[25] Mann J. Wind field simulation. Probl Eng Mech
1998;13:269–82.
[26] Sørensen JN. Three-level viscous-inviscid interaction tech-
nique for the prediction of separated flow past rotating
wing. PhD thesis, AFM-83-03, Technical University of
Denmark, 1986.
[27] Karagiannis F, Simandirakis G, Chaviaropoulos P, Papai-
liou KD. A 3D shear layer prediction method for steady
viscous flows around HAWTs. In: Proceedings of the
ECWEC 1993, Lubeck-Travemunde, Germany, p. 391–4.
[28] Milne-Thomson LM. Theoretical aerodynamics. New
York: Dover Publications; 1966.
[29] Richardson SM, Cornish ARH. Solution of three dimen-
sional incompressible flow problems. J Fluid Mech
1977;82:309–19.
[30] Joukowski NE. Vortex theory of a rowing screw. Trudy
Otdeleniya Fizicheskikh Nauk Obshchestva Lubitelei
Estestvoznaniya 1912;16:1.
[31] Margoulis W. Propeller theory of Professor Joukowski and
his pupils. NACA Technical Memorandum No. 79, 1922.
[32] Miller RH. The aerodynamic and dynamic analysis of
horizontal axis wind turbines. J Wind Eng Ind Aerodyn
1983;15:329–40.
[33] Øye S. A simple vortex model. In: Proceedings of the third
IEA symposium on the aerodynamics of wind turbines,
ETSU, Harwell, 1990, p. 4.1–5.15.
[34] Koh SG, Wood DH. Formulation of a vortex wake model
for horizontal-axis wind turbines. Wind Eng 1991;15(4):
196–210.
[35] Wood DH. On wake modelling at high tip speed ratios.
Wind Eng 1992;16(5):291–303.
[36] Hess JL. Review of integral equation techniques for solving
potential flow problems with emphasis on the surface
source method. Comput Methods Appl Mech Eng 1975;5:
145–96.
[37] Katz J, Plotkin A. Low-speed aerodynamics. New York:
McGraw-Hill; 1991.
[38] Prandtl L, Tietjens OG. Applied hydro-and aeromechanics.
New York: Dover; 1934.
[39] Landhal MT, Stark VJE. Numerical lifting surface
theory—problems and progress. AIAA J 1977;6(11):
2049–60.
[40] Kerwin JE, Lee CS. Prediction of steady and unsteady
marine propeller performance by numerical lifting surface
theory. Trans SNAME 1978; 86.
[41] Hess JL. Calculation of potential flow about arbitrary
three-dimensional lifting bodies. McDonnell Douglas
report, MDC J5679-01, 1972.
[42] Rehbach C. Calcul d’ecoulements autour d’ailes sans
epaisseur avec nappes tourbillonnaires evolutives. Re-
cherche Aerospatiale 1973;2:53–61.
[43] Cottet G-H, Koumoutsakos PD. Vortex methods: theory
and practice. Cambridge: Cambridge University Press;
2000.
[44] Bagai A, Leishman JG. Rotor free-wake modelling using a
pseudoimplicit relaxation algorithm. J Aircraft 1995;32(6):
1276–85.
[45] Bhagwat M, Leishman JG. Accuracy of straight-line
segmentation applied to curvilinear vortex filaments. J
Am Helicopter Soc 2001;46(2):166–9.
[46] Gould J, Fiddes SP. Computational methods for the
performance prediction of HAWTs. In: Hulle FV,
Smulders P, Dragt J, editors. Wind energy: technology
and implementation. Amsterdam: Elsevier Science Publish-
ers; 1991. p. 29–33 [EWEC0 91].
[47] Robison DJ, Coton FN, Galbraith RAM, Vezza M.
Application of a prescribed wake aerodynamic prediction
scheme to horizontal axis wind turbine in axial flow. Wind
Eng 1995;19(1):41–51.
[48] Coton FN, Wang T. The prediction of horizontal axis wind
turbine performance in yawed flow using an unsteady
prescribed wake model. Proc Inst Mech Eng, Part A—J
Power Energy 1999;213:33–43.
[49] Afjeh AA, Keith TG. A simplified free wake method for
horizontal axis wind turbine performance prediction. Trans
ASME J Fluid Eng 1986;108:303–9.
[50] Simoes FJ, Graham JMR. Prediction of loading on a
horizontal axis wind turbine using a free vortex wake
model. In: Proceedings of the BWEA conference, 1991.
[51] Voutsinas SG, Beleiss MA, Rados KG. Investigation of the
yawed operation of wind turbines by means of a vortex
particle method. In: AGARD conference proceedings, vol.
552, 1995, p. 11.1–11.
[52] Coton FN, Wang T, Galbraith RAM. An examination of
key aerodynamic modelling issues raised by the NREL
blind comparison. Wind Energy 2002;5:199–212.
[53] Voutsinas SG. Vortex methods in aeronautics: how
to make things work. Int J Comput Fluid Dynamics,
2006.
[54] Preuss RD, Suciu EO, Morino L. Unsteady potential
aerodynamics of rotors with applications to horizontal axis
windmills. AIAA J 1980;18(4):385–93.
[55] Arsuffi G. A general formulation for aerodynamic analysis
of wind turbine. In: Proceedings of the second IEA
symposium on the aerodynamics of wind turbines, Depart-
ment of Fluid Mechanics, Technical University of Den-
mark, 1988.
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
[56] Bareiss R, Wagner S. A hybrid wake model for hawt. In:
McAnulty K, editor. Proceedings of the sixth IEA
symposium on the aerodynamics of wind turbines, ETSU,
Harwell, 1993, p. 7.1–10.
[57] Mughal B, Drela M. A calculation method for the three-
dimensional boundary-layer equations in integral form.
AIAA Paper 93-0786, 1993, Reno NV, January.
[58] Riziotis VA, Voutsinas SG. Dynamic stall on wind turbine
rotors: comparative evaluation study of different models.
In: Proceedings of the EWEC’97, 1997, Dublin.
[59] Chaviaropoulos PK, Nikolaou IG, Aggelis KA, Sørensen
NN, Johansen J, Hansen MOL, et al. Viscous and
aeroelastic effects on wind turbine blades. The VISCEL
Project, Part I: 3D Navier–Stokes rotor simulations. Wind
Energy 2003;6(4):365–85.
[60] Rankine WJM. On the mechanical Principles of the Action
of Propellers. Transactions of the Institution of Naval
Architects, vol. 6, 1865.
[61] Froude RE. On the part played in propulsion by difference
of fluid pressure. Trans R Inst Naval Arch 1889;30:
390–405.
[62] Wu TY. Flow through a heavily loaded actuator disc.
Schiffstechnik 1962;9:134–8.
[63] Greenberg MD, Powers SR. Nonlinear actuator disc theory
and flow field calculations, including nonuniform loading.
NASA CR 1672, NASA, 1970.
[64] Greenberg MD. Nonlinear actuator disc theory. Z Flug-
wissensch 1972;20(3):90–8.
[65] Conway J. Analytical solutions for the actuator disc with
variable radial distribution of load. J Fluid Mech 1995;
297:327–55.
[66] Conway J. Exact actuator disc solution for non-uniform
heavy loading and slipstream contraction. J Fluid Mech
1998;365:235–67.
[67] Madsen HA. The actuator cylinder flow model for vertical
axis wind turbines. PhD dissertation, Aalborg University
Centre, 1982.
[68] van Kuik GAM. On the limitations of Froude’s actuator
disc concept. PhD thesis, Eindhoven University of Tech-
nology, Netherlands, 1991.
[69] Sørensen JN, Myken A. Unsteady actuator disc model for
horizontal axis wind turbines. J Wind Eng Ind Aerodyn
1992;39:139–49.
[70] Sørensen JN, Kock CW. A model for unsteady rotor
aerodynamics. J Wind Eng Ind Aerodyn 1995;58:
259–75.
[71] Sørensen JN, Mikkelsen R. On the validity of the blade
element momentum theory. In: Helm P, Zervos A, editors.
Proceedings of the 2001 European wind energy conference
and exhibition, WIP-renewable energies, Munchen, 2001,
p. 362–6.
[72] Madsen HA. A CFD analysis for the actuator disc flow
compared with momentum theory results. In: Pedersen B,
editor. Proceedings of the 10th IEA symposium on the
aerodynamics of wind turbines, Department of Fluid
Mechanics, The Technical University of Denmark, 1996,
p. 109–24.
[73] Sørensen JN, Shen WZ, Munduate X. Analysis of wake
states by a full-field actuator disc model. Wind Energy
1998;1:73–88.
[74] Madsen HA, Rasmussen F. The influence of energy
conversion and induction from large blade deflections. In:
327
Proceedings of the European wind energy conference,
James & James, 1999, p. 138–41.
[75] Mikkelsen R, Sørensen JN, Shen WZ. Modeling and
analysis of the flow field around a coned rotor. Wind
Energy 2001;4:121–35.
[76] Fejtek I, Roberts L. Navier–Stokes computation of wing/
rotor interaction for a tilt rotor in hover. AIAA J
1992;30(11):2595–603.
[77] Rajagopalan RG, Mathur ST. Three dimensional analysis
of a rotor in forward flight. J Am Helicopter Soc
1993;38(3):99.
[78] Masson C, Smaı̈li A, Leclerc C. Aerodynamic analysis of
HAWTs operating in unsteady conditions. Wind Energy
2001;4(1):1–22.
[79] Masson C, Ammara I, Paraschivoiu I. An aerodynamic
method for the analysis of isolated horizontal-axis wind
turbines. Int J Rotating Mach 1997;3:21–32.
[80] Hansen MOL, Sørensen NN, Flay RGJ. Effect of placing a
diffuser around a wind turbine. Wind Energy 2000;4(3):
207–13.
[81] Phillips D, Schaffarczyk AP. Blade-element and actuator
disc models for a shrouded wind-turbine. In: Thor S-E,
editor. Proceedings of the 15th IEA symposium on
aerodynamics of wind turbines, FOI, Swedish Defence
Research Agency, 2001. p. 95–105.
[82] Mikkelsen R, Sørensen JN. Modelling of wind tunnel
blockage. In: Thor S-E, editor. Proceedings of the 15th IEA
symposium on aerodynamics of wind turbines, FOI,
Swedish Defence Research Agency, 2001, p. 41–51.
[83] Mikkelsen R, Sørensen JN. Yaw analysis using a numerical
actuator disc model. In: Pedersen B, editor. Proceedings of
the 14th IEA symposium on the aerodynamics of wind
turbines, Department of Fluid Mechanics, Technical
University of Denmark, 2000, p. 53–9.
[84] Masson C. Viscous differential/actuator disc method and
its applications. In: Thor S-E, editor. Proceedings of the
15th IEA symposium on aerodynamics of wind turbines,
FOI, Swedish Defence Research Agency, 2001, p. 65–80.
[85] Ammara I, Leclerc C, Masson C. A viscous three-
dimensional differential/actuator disc method for aerody-
namic analysis of wind farms. J Solar Energy—Trans
ASME 2002;124(4):345–56.
[86] Sørensen JN, Shen WZ. Numerical modelling of wind
turbine wakes. J Fluids Eng 2002;124(2):393–9.
[87] Leclerc C, Masson C. Towards blade-tip vortex simulation
with an actuator-lifting surface model. AIAA-2004-0667,
2004.
[88] Mikkelsen R. Actuator disc methods applied to wind
turbines. PhD dissertation, Department of Mechanical
Engineering, DTU, Lyngby, 2003.
[89] Ivanell SSA. Numerical computations of wind turbine
wakes. PhD dissertation, KTH, Royal Institute of Tech-
nology, Stokholm, 2005.
[90] Arieli R, Tauber ME. Computation of subsonic and
transonic Flow about lifting rotor blades. AIAA Paper
79-1667, 1979.
[91] Borland C, Rizzettaq D, Yoshihara H. Numerical solution
of three-dimensional unsteady transonic flow over swept
wings. AIAA Paper 80-1369, 1980.
[92] Sankar NL, Malone JB, Tassa Y. An implicit conservative
algorithm for steady and unsteady three-dimensional
potential flows. AIAA Paper 81-1016, 1981.
 ARTICLE IN PRESS
328 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
[93] Steger JL, Caradonna FX. Conservative implicit finite
difference algorithm for the unsteady transonic potential
equations. AIAA Paper 80-1368, 1980.
[94] Caradonna FX, Tung C, Desopper A. Finite difference
modeling of rotor flows including wake effects. Journal of
the American Helicopter Society, April, 1984, p. 26–33.
[95] Sankar NL, Wake BE, Lekoudis SG. Solution of the
unsteady Euler equations for fixed and rotor wind
configurations. J Aircraft 1986;23(4).
[96] Jameson Baker TJ. Solution of the Euler equations for
complex configurations. AIAA Paper 83-1919, 1983.
[97] Pulliam TH. Euler and thin layer Navier–Stokes codes:
ARC2D, ARC3D, notes for computational fluid dynamics
users workshop, The University of Tennessee Space
Institute, March 12–16, 1984.
[98] Agarwal RK, Deese JE. Euler calculations for flowfield of a
helicopter rotor in hover. J Aircraft 1997;24(4).
[99] Srinivasan GR, McCroskey WJ. Navier–Stokes calcula-
tions of hovering rotor flowfields. J Aircraft 1988;25(10).
[100] Srinivasan GR, Raghavan V, Duque EPN. Flowfield
Analysis of modern helicopter rotors in hover by Navier–-
Stokes method. International Technical Specialists meeting,
Rotorcraft Acoustics and Rotor Fluid Dynamics, October
15–17, 1991, Philadelphia, Pennsylvania, USA.
[101] Srinivasan GR, Baeder JD, Obayashi S, McCroskey WJ.
Flowfield of a lifting rotor in hover: a Navier–Stokes
simulation. AIAA J 1992;30(10).
[102] Berkman ME, Sankar LN, Berezin CR, Torok MS.
Navier–Stokes/full potential/free-wake method for rotor
flows. J Aircraft 1997;34(5).
[103] Hansen MOL, Sørensen JN, Michelsen JA, Sørensen NN.
A Global Navier–Stokes Rotor prediction model. AIAA
97-097, 1997.
[104] Sørensen NN, Hansen MOL. Rotor performance predic-
tions using a Navier–Stokes method. AIAA 98-0025, 1998.
[105] Xu G, Sankar LN. Computational study of horizontal axis
wind turbines. AIAA 99-0042, Reno, NV, January 1999.
[106] Duque EPN, van Dam CP, Hughes S. Navier–Stokes
Simulations of the NREL combined experiment Phase II
rotor. proceedings 1999, ASME wind energy symposium,
37th AIAA Aerospace Science Meeting and Exhibit, AIAA
99-0037, Reno, NV, January 1999.
[107] Sørensen NN, Michelsen JA. Aerodynamic predictions for
the unsteady aerodynamics experiment phase-II rotor at
the National Renewable Energy Laboratory. AIAA-2000-
0037, 2000.
[108] Sørensen JN. VISCWIND, Viscous effects on wind turbine
blades. ET-AFM-9902, Department of Energy Engineer-
ing, Technical University of Denmark, June 1999,
ISBN:87-7475-218-9.
[109] Chaviaropoulos PK, Nikolaou IG, Aggelis K, Sørensen
NN, Montgomerie B, von Geyr H, et al. Viscous and
aeroelastic effects on wind turbine blades: the Viscel
Project. European wind energy conference, Copenhagen,
Denmark, 2–6 July, 2001.
[110] Kang S, Hirsch C. Features of the 3D viscous flow around
wind turbine blades based on numerical solutions. Eur-
opean Wind Energy Conference, Copenhagen, Denmark,
2–6 July, 2001.
[111] Johansen J, Sørensen NN, Reck M, Hansen MOL,
Stuermer A, Ramboer J, Hirsch C, Ekaterinaris J,
Voutsinas S, Perivolaris Y. KNOW-BLADE task-3.3
report: Rotor blade computations with 3D vortex gen-
erators. Risø-R-1486(EN), 2005, 65p.
[112] Sørensen NN, Johansen J, Conway S, Voutsinas S, Hansen
MOL, Stuermer A. KNOW-BLADE task-3.2 report: tip
shape study. Risø-R-1495(EN), 2005, 49p.
[113] Politis ES, Nikolaou IG, Chaviaropoulos PK, Bertagnolio
F, Sørensen NN, Johansen J. KNOW-BLADE Task-4
report: Navier–Stokes aeroelasticity. Risø-R-1492(EN),
2005, 39p.
[114] Johansen J, Sørensen NN, Zahle F, Kang S, Nikolaou I,
Politis ES, et al. KNOW-BLADE Task-2 report: Aero-
dynamic accessories. Risø-R-1482(EN), 2004, 33p.
[115] Sørensen NN, Johansen J, Conway S. CFD computations of
wind turbine blade loads during standstill operation KNOW-
BLADE, Task 3.1 report. Risø-R-1465(EN), 2004, 28p.
[116] Buning P, et.al. OVERFLOW user manual ver. 1.6ap.
[117] Meaking R. Moving grid overset grid methods for complete
aircraft tiltrotor simulations. AIAA Paper 93-3350, July
1993.
[118] Rizzi A, Eliasson P, Lindblad I, Hirsch C, Lacor C,
Haeuser J. The engineering of multiblock/multigrid soft-
ware for Navier–Stokes flows on structured meshes. J
Comput Fluids 1993;22:341–67.
[119] Kroll N, Radespiel R, Rossow CC. Structured grid solvers.
I: accurate and efficient flow solvers for 3D applications on
structured meshes. AGARD report 807, 1995.
[120] Eliasson P. Edge, a Navier–Stokes solver for unstructured
grids. FOI-R-0298-SE, 2001.
[121] Chorin AJ. A numerical method for solving incompressible
viscous flow problems. J Comput Phys 1967;2:12–26.
[122] Rogers SE, Kwak D. Steady and unsteady solutions of the
incompressible Navier–Stokes equations. AIAA J 1991;
29(4).
[123] Harlow FH, Welch JE. Numerical calculations of time-
dependent viscous incompressible flow of fluid with free
surface. Phys Fluids 1965;8:2182.
[124] Chorin AJ. Numerical solution of the Navier–Stokes
equations. Math Comput 1968;22(104):745–62.
[125] Patankar SV, Spalding DB. A calculation procedure for
heat, mass and momentum transfer in three-dimensional
parabolic flows. Int J Heat Mass Transfer 1972(15):1787.
[126] Rhie CM. A Numerical study of the flow past an isolated
airfoil with separation. PhD thesis, University of Illinois,
Urbane-Champaign, 1981.
[127] Menter FM. Zonal Two equation k–o Turbulence models
for aerodynamic flows. AIAA-paper-932906, 1993.
[128] Spalart PR, Allmaras SR. A one-equation turbulence
model for aerodynamic flows. AIAA-92-0439, 1992.
[129] Baldwin BS, Barth TJ. A one-equation turbulence trans-
port model for high Reynolds number wall-bounded flows.
AIAA 91-0610, 1991.
[130] Baldwin BS, Lomax H. Thin layer approximation and
algebraic model for separated turbulent flow. In: AIAA
16th aerospace science meeting, Huntsvill, Alabama, 1978.
[131] Travin A, Shur M, Strelets M. Detached-eddy simulation
past a circular cylinder flow. Turbulence Combust
1999;63:293–313.
[132] Strelets M. detached eddy simulations of massively
separated flows. AIAA-2001-0879, 2001.
[133] Xu G, Sankar LN. effects of transition, turbulence and yaw
on the performance of horizontal axis wind turbines.
AIAA-2000-0048, 2000.
 ARTICLE IN PRESS
M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
[134] Michelsen JA, Sørensen NN. Current developments in
Navier–Stokes modelling of wind turbine rotor flow. In:
European wind energy conference, Copenhagen, Denmark,
2–6 July 2001.
[135] Ferziger JH, Milovan P. Computational methods for fluid
dynamics. Berlin, Heidelberg, New York: Springer; 2002.
ISBN:3-540-59434-5.
[136] Fingersh LJ, Simms D, Hand M, Jager D, Contrell J,
Robinson M, et al. Wind tunnel testing of NREL’s
unsteady aerodynamics experiment. AIAA-2001-0035 Pa-
per, 39th aerospace sciences meeting & exhibit, 2001, Reno.
[137] Simms D, Schreck S, Hand M, Fingersh LJ. NREL
unsteady aerodynamics experiment in the NASA-Ames
wind tunnel: a comparison of predictions to measurements.
NREL/TP-500-29494, June 2001.
[138] Sørensen NN. Evaluation of 3D effects from 3D CFD
computations. In: IEA Joint action, aerodynamics of wind
turbines, 14th symposium, Boulder, December 2000.
[139] Sørensen NN, Michelsen JA, Schreck S. Navier–Stokes
predictions of the NREL phase VI rotor in the NASA
Ames 80-by-120 wind tunnel. AIAA-2002-0032, 2002.
[140] Xu G, Sankar LN. Application of a viscous flow
methodology to the NREL phase VI ROTOR. AIAA-
2002-0030, 2002.
[141] Sørensen NN, Michelsen JA, Schreck S. Application of
CFD to wind turbine aerodynamics. In: Fourth GRACM
congress on computational mechanics, GRACM 2002,
Patra, 27–29 June 2002.
[142] Madsen HA, Sørensen NN, Schreck S. Yaw aerodynamics
analyzed with three codes in comparison with experiment.
AIAA-2003-0519, 2003.
[143] Tongchitpakdee C, Benjanirat S, Sankar LN. Numerical
simulation of the aeordynamics of horizontal axis wind
turbines under yawed flow conditions. AIAA-2005-0773,
2005.
[144] Benjanirat S, Sankar LN. Recent improvements to a
combined Navier–Stokes full potential methodology for
modeling horizontal axis wind turbines. AIAA-2004-0830,
2004.
[145] Duque EPN, Burklund MD, Johnson W. Navier–Stokes
and comprehensive analysis performance predictions of the
NREL phase VI experiment. AIAA-2003-0355, 2003.
[146] Benjanirat S, Sankar LN. Evaluation of turbulence models
for the prediction of wind turbine aerodynamics. AIAA-
2003-0517, 2003.
[147] Le Pape A, Lecanu J. 3D Navier–Stokes computations of a
stall regulated wind turbine. In: Proceedings: the science of
making torque from wind, 19–21 April 2004, Delft
University of Technology, the Netherlands, 2004.
[148] Johansen J, Sørensen NN, Michelsen JA, Schreck S.
Detached-eddy simulation of flow around the NREL
phase-VI rotor. In: Proceedings CD-ROM. CD 2. Eur-
opean wind energy conference and exhibition 2003 (EWEC
2003), Madrid (ES), 16–19 June 2003.
[149] Fleig O, Arakawa C. Numerical simulation of wind turbine
tip noise. AIAA-2004-1190, 2004.
[150] Øye S. FLEX4 simulation of wind turbine dynamics. In
Proceedings of 28th IEA meeting of experts concerning
State of the Art of Aeroelastic Codes for wind turbine
calculations (Available through IEA), 1996.
[151] Hansen MOL. Aerodynamics of Wind Turbines. James &
James (Science Publishers) Ltd; 2000.
329
[152] Schepers JG. Verification of European wind turbine design
codes, VEWTDC: final report. Technical report ECN-C-
01-055, Netherlands Energy Research Foundation ECN,
Petten, 2002.
[153] Chaviaropoulos P. Development of a state-of-the-art
aeroelastic simulator for horizontal axis wind turbines,
Part 1: structural aspects. Wind Eng 1996;20(6):405–22.
[154] Hodges DH, Dowell EH. Nonlinear equations of motion
for elastic bending and torsion of twisted non-uniform
blades. NASA report, NASA TN D-7818, 1975.
[155] Hodges DH. A review of composite rotor blade modelling.
AIAA J 1990;28(3):561–5.
[156] Kunz DL. Survey and comparison of engineering beam
theories for helicopter rotor blades. J Aircraft 1994;31:
473–97.
[157] Jung SN, Nagaraj VT, Chopra I. Assessment of composite
rotor blade modelling techniques. J Am Helicopter Soc
1999;44(3):188–205.
[158] Cesnil CES, Hodges DH, Sutyrin VG. Cross sectional
analysis of composite beams including large initial twist
and curvature effects. AIAA J 1996;34(9):1913–20.
[159] Nim E. Coupling and reduction of the HAWC equations.
RISOE report, RISOE-R-1294 (EN), 2001.
[160] Snel H, Liendenburg C. Aeroelastic rotor system code
for horizontal axis wind turbines: PHATAS-II. In: Pro-
ceedings of the ECWEC 1990, Madrid, Spain, 1990,
p. 284–90.
[161] Riziotis VA, Voutsinas SG. GAST: A general aerodynamic
and structural prediction tool for wind turbines. In:
Proceedings of the EWEC’97, 1997, Dublin, Ireland.
[162] Hansen AC, Laino DJ. Validation study for AeroDyn and
YawDyn using phase iii combined experiment data. In:
AIAA-1997-943, aerospace sciences meeting and exhibit,
35th, Reno, NV, January 6–9, 1997.
[163] Riziotis VA, Voutsinas SG. Advanced aeroelastic modeling
of complete wind turbine configurations in view of
assessing stability characteristics. In: Proceedings of the
EWEC’06, 2006, Athens, Greece.
[164] Bazoune A, Khulief YA. Shape functions of three-
dimensional Timoshenko beam element. J Sound Vib 2003;
259(2):473–80.
[165] Pfeiffer F, Glocker C. Multi-body dynamics with unilateral
contacts. New York: Wiley; 1996.
[166] Bauchau OA. Computational schemes for flexible, non-
linear multi-body systems. Multibody System Dynamics
1998;2:169–225.
[167] Bauchau OA, Hodges DH. Analysis of non-linear multi-
body systems with elastic couplings. Multibody System
Dynamics 1999;3:166–88.
[168] Chaviaropoulos PK, Politis ES, Lekou DJ, Sorensen NN,
Hansen MH, Bulder BH, et al. Enhancing the damping of
wind turbine rotor blades, the Dampblade Project. Wind
Energy 2006;9:163–77.
[169] Petot D. Differential equation modelling of dynamic stall.
La Recherche Aerospatiale 1989;5:59–72.
[170] Chopra I. Aeroelastic stability of an elastic circulation
control rotor blade in hover. Vertice 1984;8(4):353–71.
[171] Chaviaropoulos PK. Flap/lead–lag aeroelastic stability of
wind turbine blades. J Wind Energy 2001;4:183–200.
[172] Hansen MH. Improved modal dynamics of wind turbines
to avoid stall-induced vibrations. J Wind Energy 2003;6(2):
179–95.
 ARTICLE IN PRESS
330 M.O.L. Hansen et al. / Progress in Aerospace Sciences 42 (2006) 285–330
[173] Friedman PP, Hammond CE, Woo TH. Efficient numerical
treatment of periodic systems with application to stability
problems. Int J Numer Methods Eng 1977;11:1117–36.
[174] Peters DA. Fast Floquet theory and trim for multi-bladed
rotorcraft. J Am Helicopter Soc 1994;39(4):82–9.
[175] Coleman RP, Feingold AM. Theory of self-excite mechan-
ical oscillations of helicopter rotors with hinged blades.
NASA report, NASA-TN-3844, 1957.
[176] Petersen JT, Madsen HA, Bjorck A, Enevoldsen P, Oye S,
Ganander H, et al. Prediction of dynamic loads and
induced vibrations in stall. RISO report, RISOE-R-
1045(EN), 1998.
[177] Politis ES, editor. Benchmark calculations on the NM80
wind turbine. CRES, Technical report STABCON project,
2005.
[178] Kirchgassner B. ARLIS—a program system for analysis of
rotating linear systems. In: Proceedings of the EWEC 1984,
Hamburg, Germany, 1984, p. 22–6.
[179] Hansen, MH. Stability Analysis of three-bladed turbines
using an eigenvalue approach. In: 42 AIAA aerospace
sciences meeting and exhibit. 2004 ASME wind energy
symposium, Reno, NV, January 2004.
[180] van Engelen, TG, Braam, H. TURBU offshore computer
programme for frequency domain analysis of horizontal
axis offshore wind turbines: implementation. ECN report,
ECN-C—04-0479, Petten, The Netherlands, 2004.
[181] van Holten Th, van Overbeek K. Automatic simulation of
complex non-linear dynamic systems. In: 26th European
Rotorcraft Forum, Paper 65, 2000.
[182] Tasker FA, Chopra I. Assessment of transient analysis
techniques for rotor stability. J Am Helicopter Soc
1990;88:39–50.
[183] Simon M, Tomlinson GR. Use of the Hilbert transform in
modal analysis of linear and non-linear structures. J Sound
Vib 1984;96:421–36.
[184] Smith CB, Wereley NM. Transient analysis for damping
identification in rotating composite beams with integral
damping layers. Smart Mater Struct 1996;5:540–50.
[185] Madsen HA, Petersen JT, Björck A, Ganander H,
Winkelaar D, Brand A, et al. Prediction of Dynamic Loads
and Induced Vibrations in Stall—STALLVIB. Final
publishable report on Contract JOR3-CT95-0047. Risø
National Laboratory, 1998.
[186] Hansen, MH. HAWCStab aeroelastic stability tool for
wind turbines—user’s guide. Risø-I-2232(EN), Risø Na-
tional Laboratory, August 2004.
[187] Petersen, JT. Kinematically nonlinear finite element model
of a horizontal axis wind turbine. PhD thesis, Part 1 and 2.
Risø National Laboratory, Roskilde, Denmark, July 1990.
[188] Petersen, JT. The Aeroelastic Code HawC—model and
comparison. In: Pedersen BM, editor. Proceedings of state
of the art of aeroelastic codes for wind turbine calculations.
28th meeting of experts, International Energy Agency,
Annex XI. Technical University of Denmark, Lyngby,
April 11–12, 1996, p. 129–35.
[189] Larsen TJ, Hansen AM, Buhl T. Aeroelastic effects of large
blade deflections for wind turbines. In: Proceedings of the
conference: the science of making torque from wind.
Special topic conference Delft 19–21 April 2004.
[190] Rubak R, Petersen JT. Monopile as part of aeroelastic
wind turbine simulation code. In: Proceedings of Copenha-
gen Offshore Wind 2005, conference & exhibition, 26–28
October 2005.
[191] Larsen JT, Madsen HA, Hansen AM, Thomsen K.
Investigation of stability effects of an offshore wind turbine
using the new aeroelastic code HAWC2. In: Proceedings of
Copenhagen Offshore Wind 2005, 2005.
[192] Madsen HA, Rasmussen F. A near wake model for trailing
vorticity compared with the blade element momentum
theory. Wind Energy 2004;7:325–41.
[193] Madsen HA. Yaw simulation using a 3D actuator disc
model coupled to the aeroelastic code HawC. In: Pedersen
BM, editor. Proceedings, 13th symposium IEA joint action
on aerodynamics of wind turbines, Stockholm, 1999.
[194] Thomsen K, Madsen HA. A new simulation method for
turbine in wakes—applied to extreme response during
operation. Wind Energy 2005;8:35–47.
[195] Madsen HA, Larsen GC, Thomsen K. Wake flow
characteristics in low ambient turbulence conditions.
In: Proceedings (CD-ROM), Copenhagen offshore wind
conference 2005, Copenhagen (DK), 25–28 September
2005.
[196] Troldborg N, Sørensen JN, Mikkelsen R. Numerical
simulations of wakes of wind turbines in wind farms. In:
Proceedings of the EWEC 2006, European wind energy
conference, Athens, 2006.
[197] Madsen HA, Rasmussen F. Derivation of three-dimen-
sional airfoil data on the basis of experiments and theory.
In: Proceedings Windpower 0 88, Honolulu, Hawaii, 1988.
[198] Schepers JG, et al. Final report of IEA Annex XIV: Field
rotor aerodynamics. Report ECN-C-97-027, ECN June
1997.
[199] Schepers JG, et al. Final report of IEA Annex XVIII:
enhanced field rotor aerodynamics database. Report ECN-
C–02-016. ECN February 2002.
[200] Hansen MOL, Sørensen NN, Sørensen JN, Michelsen JA.
Extraction of lift, drag and angle of attack from computed
3-D viscous flow around a rotating blade. In: Proceedings
of the European wind energy conference, EWEC-1997,
Dublin, Ireland, October 1997.
[201] Johansen J, Sørensen NN. Method for extracting
airfoil data using 3D CFD computations. In: Thor SE,
editor. IEA Joint Action Committee on Aerodynamics.
Annex IV Aero experts meeting Boulder, CO (US), 5–6
May 2003, FFA, The Aeronautical Institute of Sweden,
2003.
[202] Johansen J, Sørensen NN, Mikkelsen R. Rotoraerodyna-
mik. In the report Risø-R-1434(DA) Forskning i Aero-
elasticitet edited by Christian Bak, February 2004.
[203] Hald T, Høgedal M. Implementation of a finite element
foundation module in Flex5 using Craig–Bampton sub-
structuring. In: Proceedings of Copenhagen offshore wind
2005, conference & exhibition, 26–28 October 2005.
[204] Seidel M, von Mutius M, Ris D, Steudel D. Integrated
analysis of wind and wave loading for complex support
structures of offshore wind turbines. In: Proceedings of
Copenhagen Offshore Wind 2005, conference & exhibition,
26–28 October 2005.