Professional Documents
Culture Documents
Version 4.3
DISCLAIMER
Acceptance of this document by the client is on the basis that Garrad Hassan
& Partners Ltd is not in any way to be held responsible for the application or
use made of the findings of the results from the analysis and that such
responsibility remains with the client.
COPYRIGHT
All rights reserved. Duplications of this document in any form are not allowed
unless agreed in writing by Garrad Hassan & Partners Ltd.
www.gl-garradhassan.com
1. INTRODUCTION
1.1 Purpose
Bladed is an integrated software package for wind turbine performance and loading
calculations. It is intended for the following applications:
With its sophisticated graphical user interface, it allows the user to carry out the following
tasks in a straightforward way:
The Garrad Hassan approach to the calculation of wind turbine performance and loading has
been developed over many years. The main aim of this development has been to produce
reliable tools for use in the design and certification of wind turbines.
Many of the models and theoretical methods incorporated in Bladed have been extensively
validated for wind turbines against monitored data from a wide range of turbines of many
different sizes and configurations, including:
This document describes the theoretical background to the various models and numerical
methods incorporated in Bladed.
1.3 Support
Bladed is supplied with a one-year maintenance and support agreement, which can be
renewed for further periods. This support includes a ‘hot-line’ help service by telephone, fax
or e-mail:
1.4 Documentation
In addition to this Theory Manual, there is also a Bladed User Manual which explains how the
code can be used.
1.5 Acknowledgements
Bladed was developed with assistance from the Commission of the European Communities
under the JOULE II programme, project no. JOU2-CT92-0198. .
2. AERODYNAMICS
The modelling of rotor aerodynamics provided by Bladed is based on the well established
treatment of combined blade element and momentum theory [2.1]. Two major extensions of this
theory are provided as options in the code to deal with the unsteady nature of the aerodynamics.
The first of these extensions allows a treatment of the dynamics of the wake and the second
provides a representation of dynamic stall through the use of a stall hysteresis model.
The theoretical background to the various aspects of the treatment of rotor aerodynamics
provided by Bladed is given in the following sections.
At the core of the aerodynamic model provided by Bladed is combined blade element and
momentum theory. The features of this treatment of rotor aerodynamics are described below.
Wind turbines extract energy from the flow by producing a step change in static pressure across
the rotor-swept surface. As the air approaches the rotor it slows down gradually, resulting in an
increase in static pressure. The reduction in static pressure across the rotor disk results in the
fluid behind it experiencing a reduction in pressure compared to free stream conditions. As the
fluid proceeds downstream the pressure climbs back to the free stream value resulting in a
further slowing down of the flow. There is therefore a reduction in the kinetic energy in the flow,
some of which is converted into useful energy by the turbine.
In the actuator disk model of the process described above, the flow velocity at the rotor disk Ud is
related to the upstream wind velocity Uo as follows:
U d (1 a )U o
The reduced flow velocity at the rotor disk is clearly determined by the magnitude of a, the axial
flow induction factor or inflow factor.
By applying Bernoulli’s equation and assuming the flow to be uniform and incompressible, it can
be shown that the power P extracted by the rotor is given by :
P 2AU o3a(1 a) 2
where is the fluid density and A the area of the rotor disk.
The thrust T acting on the rotor disk can similarly be derived to give:
T 2AU o2 a(1 a)
and:
CT T /( 1 2 AU o2 ) 4a(1 a)
1 16
The maximum value of the power coefficient CP occurs when a is /3 and is equal to /27 which
is known as the Betz limit.
1
The thrust coefficient CT has a maximum value of 1 when a is /2.
The torque generated by the rotor is equal to the rate of change of angular momentum and can
be derived as:
Q R 4 (1 a)a,U o
The thrust dT developed by a blade element of length dr located at a radius r is given by:
where W is the magnitude of the apparent flow speed vector at the blade element, is
known as the inflow angle and defines the direction of the apparent flow speed vector relative
to the plane of rotation of the blade, c is the chord of the blade element and CL and CD are the
lift and drag coefficients respectively.
The lift and drag coefficients are defined for an aerofoil by:
CL L /( 1 V 2 S )
2
and
CD D /( 1 V 2 S )
2
where L and D are the lift and drag forces, S is the planform area of the aerofoil and V is the
flow velocity relative to the aerofoil.
The torque dQ developed by a blade element of length dr located at a radius r is given by:
a g1 /(1 g1 )
and
a , g 2 /(1 g 2 )
where
and
Here B is the number of blades and F is a factor to take account of tip and hub losses, refer
Section 2.1.4.
4a(1 a)
for a 0.3539,H
(0.6 0.61a 0.79a 2 )
In the situation where the axial induction factor a is greater than 0.5, the rotor is heavily
loaded and operating in what is referred to as the “turbulent wake state”. Under these
conditions the actuator disk theory presented in Section 2.1.1 is no longer valid and the
expression derived for the thrust coefficient:
CT 4a (1 a )
ak ak 1 tol
and
The circulation at the blade tips is reduced to zero by the wake vorticity in the same manner as
at the tips of an aircraft wing. At the tips, therefore the factor Ft becomes zero. Because of the
analogy with the aircraft wing , where losses are caused by the vortices trailing from the tips, Ft
is known as the tip loss factor.
Prandtl [2.2] put forward a method to deal with this effect in propeller theory. Reasoning that, in
the far wake, the helical vortex sheets could be replaced by solid disks, set at the same pitch as
the normal spacing between successive turns of the sheets, moving downstream with the speed
of the wake.
The flow velocity outside of the wake is the free stream value and so is faster than that of the
disks. At the edges of the disks the fast moving free stream flow weaves in and out between
them and in doing so causes the mean axial velocity between the disks to be higher than that of
the disks themselves, thus simulating the reduction in the change of momentum.
Ft 2 arccos[exp( s )]
d
where s is the distance of the radial station from the tip of the rotor blade and d is the distance
between successive helical sheets.
A similar loss takes place at the blade root where, as at the tip, the bound circulation must fall to
zero and therefore a vortex must be trailed into the wake, A separate hub loss factor Fh is
therefore calculated and the effective total loss factor at any station on the blade is then the
product of the two:
F Ft Fh
The combined tip and hub loss factor is incorporated in the equations of blade element theory
as indicated in Section 2.1.3 above.
Clearly in this interpretation of blade element theory the axial and tangential induced
velocities at a particular blade element vary with time and are not constant within the annulus
swept out by the element.
The equilibrium wake treatment of blade element theory is the most computationally
demanding of the three treatments described here.
It is important to note that it is the axial and tangential induced velocities aUo and a r and
’
’
not the induction factors a and a which are frozen in time.
The study of dynamic inflow was initiated nearly 40 years ago in the context of helicopter
aerodynamics. In brief, the theory provides a means of describing the dynamic dependence of
the induced flow field at the rotor upon the loading that it experiences. The dynamic inflow
model used within Bladed is based on the work of Pitt and Peters [2.3] which has received
substantial validation in the helicopter field, see for example Gaonkar et al [2.4].
The Pitt and Peters model was originally developed for an actuator disk with assumptions
made concerning the distribution of inflow across the disc. In Bladed the model is applied at
blade element or actuator annuli level since this avoids any assumptions about the
distribution of inflow across the disc.
For a blade element, bounded by radii R1 and R2 , and subject to uniform axial flow at a flow
speed Uo, the elemental thrust, dT, can be expressed as:
dT 2U o am U o m A a
where m is the mass flow through the annulus, mA is the apparent mass acted upon by the
annulus and a is the axial induction factor.
m U o (1 a )dA
For a disc of radius R the apparent mass upon which it acts is given approximately by
potential theory, Tuckerman, [2.5]:
mA 8 R 3
3
Therefore the thrust coefficient associated with the annulus can be derived to give:
16 ( R23 R13 )
CT 4a(1 a) a
3U o ( R22 R12 )
This differential equation can therefore be used to replace the blade element and momentum
theory equation for the calculation of axial inflow. The equation is integrated at each time step
to give time dependent values of inflow for each blade element on each blade. The tangential
inflow is obtained in the usual manner and so depends on the time dependent axial value. It is
evident that the equation introduces a time lag into the calculation of inflow which is
dependent on the radial station.
It is probable that the values of time lag for each blade element calculated in this manner will
under-estimate somewhat the effects of dynamic inflow, as each element is treated
independently with no consideration of the three dimensional nature of the wake or the
possibly dominant effect of the tip vortex. The treatment is, however, consistent with blade
element theory and provides a simple, computationally inexpensive and reasonably reliable
method of modelling the dynamics of the rotor wake and induced velocity flow field.
The representation and to some extent the general understanding of aerodynamic stall on a
rotating turbine blade remain rather poor. This is a rather extraordinary situation in view of the
importance of stall regulation to the industry.
Stall delay on the inboard sections of rotor blades, due to the three dimensionality of the
incident flow field, has been widely confirmed by measurements at both model and full scale.
A number of semi-empirical models [2.6, 2.7] have been developed for correcting two
dimensional aerofoil data to account for stall delay. Although such models are used for the
design analysis of stall regulated rotors, their general validity for use with a wide range of
aerofoil sections and rotor configurations remains, at present, rather poor. As a consequence
Bladed does not incorporate models for the modification of aerofoil data to deal with stall
delay, but the user is clearly able to apply whatever correction of the aerofoil data he believes
is appropriate prior to its input to the code.
Stall and its consequences are fundamentally important to the design and operation of most
aerodynamic devices. Most conventional aeronautical applications avoid stall by operating well
below the static stall angle of any aerofoils used. Helicopters and stall regulated wind turbines
do however operate in regimes where at least part of their rotor blades are in stall. Indeed stall
regulated turbines rely on the stalling behaviour of aerofoils to limit maximum power output from
the rotor in high winds.
A certain degree of unsteadiness always accompanies the turbulent flow over an foil at high
angles of attack. The stall of a lifting surface undergoing unsteady motion is more complex
than static stall.
On an oscillating aerofoil, where the incidence is increasing rapidly, the onset of the stall can be
delayed to an incidence considerably in excess of the static stall angle. When dynamic stall
does occur, however, it is usually more severe than static stall. The attendant aerodynamic
forces and moments exhibit large hysteresis with respect to the instantaneous angle of attack,
especially if the oscillation is about a mean angle close to the static stall angle. This represents
an important contrast to the quasi-steady case, for which the flow field adjusts immediately,
and uniquely, to each change in incidence.
Many methods of predicting the dynamic stall of aerofoil sections have been developed,
principally for use in the helicopter industry.
The model adopted for inclusion of unsteady behaviour of aerofoils is that due to
Beddoes [2.8]. The Beddoes model was developed for use in helicopter rotor performance
calculations and has been formulated over a number of years with particular reference to
dynamic wind tunnel testing of aerofoil sections used on helicopter rotors. It has been used
successfully by Harris [2.9] and Galbraith et al [2.10] in the prediction of the behaviour of vertical
axis wind turbines.
The model used within Bladed is a development of the Beddoes model which has been
validated against measurements from several stall regulated wind turbines. The model utilises
the following elements of the method described in [2.8] to calculate the unsteady lift coefficient
The use of the model of leading edge separation has been found to be inappropriate for use on
horizontal axis wind turbines where the aerofoil characteristics are dominated by progressive
trailing edge stall.
The time lag in the development of trailing edge separation is a user defined parameter within
the model implemented in Bladed. This time lag encompasses the delay in the response of the
pressure distribution and boundary layer to the time varying angle of attack. The magnitude of
the time lag is directly related to the level of hysteresis in the lift coefficient.
The drag and pitching moment coefficients are also calculated according to the Beddoes-
Leishman model.
Modifications to the Beddoes-Leishman model are as follows:
Modification of the chordwise force expression to give agreement with steady aerofoil
data.
The coefficients for the circulatory part of the indicial response are given by Jones’
approximation to the Wagner function that models a thin airfoil undergoing a step
change in angle of attack in incompressible flow: A1 = 0.165, A2 = 0.335, b1 = 0.0455
and b2 = 0.3.
Addition of vortex component in the chordwise direction according to the results of
[2.11].
Attached flow contributions applied only to the proportion of the aerofoil for which flow is
attached.
Expression for circulatory part of pitching moment coefficient based on assumption of
linear part and an additional part due to stall. These are weighted by the time-lagged
separation position so that in limiting steady conditions the quasi-steady pitching
moment is recovered.
An additional term modelling the contribution to the pitching moment from drag is
included. This is based on the variation of windspeed over the aerofoil chord length
when the aerofoil is pitching and is important for the torsional stability of heavily stalled
aerofoils.
3. STRUCTURAL DYNAMICS
In the early days of the industry, wind turbine design was undertaken on the basis of quasi-
static aerodynamic calculations, while the effects of structural dynamics were either ignored
completely or included through the use of estimated dynamic magnification factors. From the
late 1970’s research workers began to consider more reliable methods of dynamic analysis,
and two basic approaches were considered: finite element representations and modal
analysis.
The traditional use of standard, commercial finite element analysis software packages for
solving problems of structural dynamics is challenging in the case of wind turbines. This is
because of the presence of rigid body motion of one structural component, i.e. the rotor, with
respect to another, i.e. the tower or another support structure type. In principle, the standard
finite element method only considers structures in which the deflection occurs about an initial
reference position, and for this reason the finite element models that have been developed for
wind turbine in the past have been tailored to deal with this problem.
Dynamic wind turbine models commonly used as the basis of design calculations involve a
modal representation of the deformation state. This approach, borrowed from the helicopter
industry, has the major advantage that it offers a reliable representation of the dynamics of a
wind turbine with relatively few degrees of freedom. Another important advantage is that the
structural damping of flexible components can be described conveniently and realistically as
modal damping. Obviously the number and type of modal degrees of freedom used to
represent the dynamics of a particular wind turbine depends on the configuration and
structural properties of the machine.
The multi-body dynamics approach used for the Bladed structural model was originally
proposed by Dr. J.P. Meijaard, Nottingham University, (presently University of Twente, in the
Netherlands) under commission of Garrad Hassan and Partners Ltd [3.4], and it was
developed particularly for modelling wind turbine structural dynamics. The approach assumes
a tree-like structure, which means that structural loops cannot be described.
In general, the structural components may be assumed to be rigid, such as yaw and blade
bearings, or flexible such as support structures, towers and blades. While rigid components
are relatively easy to model, flexible components are more complicated as the motion of a
material point of this type of components is generally caused by rigid body motion combined
with relative motion due to the deformation. A description of the applied method for modelling
flexible components is given in Section 3.2.
The rigid body motion of a component is described in terms of the motion of a set of local
component nodes that are characteristic material points, where the motion of the component
is assumed to be known. Components can only be interconnected at the nodes, and a
connection is imposed by the usual finite element technique by linking the nodes of the
components involved in the connection to a global structural node. Due to the assumption of
the tree-like structure it is convenient to subdivide the nodes of a component into a proximal
node that is linked to a node of its predecessor, and distal nodes that may be linked to nodes
of successors. For example a yaw bearing has two nodes, i.e. one proximal node attached to
the tower and one distal node attached to the main frame.
For all components a local Cartesian coordinate system is attached to the proximal node such
that the position of origin and the orientation are defined by the position and orientation,
respectively, of this node. This local coordinate system is mainly used for flexible components
that are described in more details in Section 3.2.
The deformation state of a component is described by generalised strains that represent the
degrees of freedom associated with the component. For example a yaw bearing has one
generalised strain, which represent the yaw angle. The approach used also allows prescribed
strains, which are particularly important in the case of stick-slip friction in bearings, where the
bearing may stick if the absolute value of the angular velocity (i.e. the strain rate) approaches
zero. It is noted that prescribing a strain element will reduce the effective order of the system
of equilibrium equations by one.
In general, the relative motion of the distal nodes is constrained, for which reason the position
and orientation of a distal node are expressed as functions of the position and orientation of
the proximal node and the generalised strains. From this fundamental transformation it is
straightforward to derive the corresponding transformations for the velocity and the
acceleration. The constraints are conveniently expressed in terms of two constraint matrices
relating to the nodal velocities and the strain rates. In general, the constraint matrices are
time-dependent functions of the position and orientation of the proximal node as well as the
strains.
Components may have mass or may be considered massless. The generalised inertia forces
are derived from the principle of virtual work, where the inertia force density is expressed
according to D’Alambert’s principle. The material velocity and acceleration are derived from a
fundamental displacement hypothesis that defines the absolute displacement of all material
points of the component as a function of the relative position, the nodal position and
orientation and the strain. The result of this analysis shows that the inertia force can be
expressed in terms of a mass matrix multiplied by the acceleration vector plus a vector of non-
linear inertia forces and stresses, i.e. centrifugal and Coriolis forces.
In principle, external loads can only be applied as nodal loads due to the nodal velocities or
generalised stresses due to the generalised strain rates. For distributed loads like wind and
wave loads the corresponding applied nodal loads and generalised stresses are calculated
according to the principle of virtual work. For a yaw bearing the applied generalised stress is
simply the moment applied by the yaw actuator.
The resulting equilibrium equations of a component are derived by collecting all generalised
forces acting on the component. The effect of the distal node(s) constraints is described by
Lagrange’s methods in terms of internal forces that are expressed by the constraint matrices
and a set of yet unknown Lagrange multipliers [3.5]. A detailed analysis shows that the
resulting component equilibrium equations and the transformation for the acceleration may be
expressed in matrix form as
M crr M cr Dcr T v c 0 0 fac f0c fic
M c T M c D c T ε c C c εc K c εc σ c σ c σ c
rc
a c 0 i
Dr D 0 λc 0 0 a 2 0
c
where
v c is a vector of nodal velocities
εc is a vector of generalised and prescribed strains
λc is a vector of Lagrange multipliers corresponding to the constraints
fic and σ ic are vectors of non-linear inertia forces and stresses dual to nodal velocities
and strain rates
fac is a vector of applied nodal forces
σ ca is a vector of applied generalised stresses dual to generalised strain rates
f0c is a vector of joint reactions dual to nodal velocities
σ c0 is a vector of generalised constraint stresses corresponding to prescribed strains
a c2 collects the convective terms (quadratic in nodal velocities) in the transformation
for the acceleration
M crr , M cr and M c are the structural mass matrix partitions dual to nodal velocities and
strain rates
C c is the structural damping matrix dual to strain rates
K c is the structural stiffness matrix dual to strain rates
D cr and D c are the constraint matrix partitions relating to nodal velocities and strain
rates
Obviously it is not possible to solve this equation due to the unknown joint reaction forces f0c
(section forces), which originates from the connection to other components. In order to solve
the system it is necessary to collect all the component equilibrium equations into a global
system of the complete structure, which is done using the standard finite element assembly
process [3.5]. This global system of equations has almost the same form as the component
equations and is written in matrix form as
M rr M r DTr v 0 0 fa f i
M T M DT ε C ε K ε σ σ σ
r
a 0 i
r
D D 0
λ 0 0 a 2 0
The main difference between component equations and the global system of equations is that
the joint reaction forces do not appear in the latter as they have been cancelled out by the
assembly process. Consequently the above system can be solved directly with respect to the
nodal accelerations v , the strain accelerations ε , and the Lagrange multipliers λ .
In the standard wind turbine model in Bladed the blades and the support structure are
modelled as single linear flexible components (bodies) using the modal approach, where the
deformation is represented by a linear combination of some pre-calculated mode shape
functions or simply modes. The scalars of this linear combination are the modal amplitudes
[3.6] that represent the generalised strains and hence the degrees of freedom of the
component. It is important to note that the mode shape functions are constant in time and
varies only spatially.
The mode shape functions for a flexible component are calculated from a fundamental
structural model based on standard linear finite element technique [3.5], which means that the
deformation of a flexible component is generally assumed to be small. The fundamental finite
element model basically assumes that the flexible components are linear space beams or
more general space frames that comprise assemblies of multiple members modelled by
Timoshenko beam elements. It is also noted that the fundamental finite element model is also
used for calculating the internal forces (stress resultants) of the flexible component as
described in Section 3.5.
The applied beam element may be considered as an extension to the standard three-
dimensional engineering Timoshenko beam element [3.7] with two nodes or stations located
at the two ends. This element has twelve fundamental degrees of freedom, i.e. three
translational and three rotational degrees of freedom at both stations. At all intermediate
points the deflection is calculated interpolation functions that are derived from a set of
homogenous equilibrium equations valid for prismatic beam elements. It is important to note
that this beam element includes the effect of shear deformation that may be important for
some support structures, in particular complicated offshore foundations. The magnitude of the
shear deformations relative to bending deformations for an element may be evaluated by the
element shear parameter conveniently defined as
12 EI e
e 2
l e GAe
where
EI e is the bending stiffness
GAe is the corresponding shear stiffness
l e is the element length
The beam element supports an arbitrary spatial variation of the stiffness along the beam
element, but in the present implementation it is assumed that the bending stiffness and
2
torsional rigidity vary with H tw, where H is the cross-section structural height and tw is the
cross-section wall thickness that both are interpolated linearly over the element, whereas the
axial stiffness and the shear stiffness are assumed to vary linearly. The orientation of the
element is defined by the position of the two ends as well as the orientation of the principal
axis around the neutral axis (elastic centre). The effect of possible coupling between bending
and torsion is included in terms of the position of the shear centre (torsion centre), and a
transformation between displacements and forces relating to the shear centre and the neutral
axis is included. The resulting stiffness matrix is constant and calculated by numerical
integration. For prismatic elements, where the shear centre is located at the neutral axis, the
stiffness matrix is identical to the standard engineering Timoshenko beam element [3.7].
An important feature of the derived method is that some fundamental degrees of freedom
may be constrained, which is particularly useful in cases where the effect of elongation and/or
torsion can be neglected. In order to enable the description of rigid connection the deflection
of a beam element may also be constrained completely. The constraints are modelled in
terms of a constant constraint matrix together with Lagrange’s method [3.5].
Second-order effects of the axial forces are included in terms of a geometric stiffness matrix
(stress stiffening) that is calculated from the derivatives of the interpolation functions [3.6]. For
the blades the dominating part of the axial force is caused by centrifugal forces for which
reason the term centrifugal stiffness is traditionally used in this case.
The selection and calculation of mode shape functions follows the idea that was originally
suggested by Craig [3.8] as a modification of the celebrated Craig-Bampton method [3.9]. For
both methods the stations are subdivided into boundary stations that may couple to other
components and interior stations that do not couple. The boundary stations also represent the
component nodes that may link to nodes of other components. In particular the station
representing the proximal node is constrained completely in order to exclude rigid body
displacement modes. With the applied method the modes are generally selected as the union
between attachment modes that may couple to other components and normal modes that
may be considered as internal vibration modes. The attachment modes are calculated from
the component stiffness matrix by a static equilibrium, where the component is fixed at the
proximal node and point loads are applied in turn at the distal nodes. The normal modes are
calculated from the mass and stiffness matrices by a generalised eigenvalue problem, where
the component is fixed at all boundary stations. The frequencies of the attachment modes are
calculated by Rayleigh’s method [3.6], while the frequencies of the normal modes result from
the eigenvalue problem. These frequencies are solely used for describing damping as
explained below.
The modes are orthogonalised in a particular form that ensures that the stiffness matrix
relating to the generalised strains is diagonal.
Inertia forces acting on the element and the proximal node are derived as described in
Section 3.1 from the fundamental displacement hypothesis using the principle of virtual work.
An important feature of the derived method is that the inertia forces are expressed directly in
terms of the modal amplitudes, i.e. the strains and corresponding derivatives as originally
suggested by Shabana [3.2]. In principle this means that the time for calculating the
accelerations scales with the number of mode shapes rather than the number of beam
elements of the flexible component.
Modal frequency, i
Modal damping coefficient, i
Mode shape represented by a vector of the displacement at the stations
The modal frequencies and mode shapes of the blade are calculated based on the following
parameters that generally are defined at each radial station:
The frequencies and mode shapes of the rotor modes are computed from the eigenvalues
and eigenvectors of the fundamental finite element model of the blade structure. The finite
element model of the blade is based on the use of three-dimensional beam elements to
describe the mass and stiffness properties of the rotor blades as described in Section 3.2.
The outputs from the modal analysis of a blade are the modal frequencies and mode shapes
defined by a vector of the displacements of all stations, i.e. three translational components
and three rotational components. The modal damping coefficients are an input defined by the
user and may be used to represent structural damping.
The calculation of modal frequencies and mode shapes of the tower make use of the
following parameters:
The frequencies and mode shapes of the tower modes are computed from the eigenvalues
and eigenvectors of a finite element representation of the tower structure. The finite element
model of the tower is based on the use of two-dimensional beam elements to describe the
mass and stiffness properties of the tower.
The outputs from the modal analysis of the tower are the modal frequencies and mode
shapes representing fore-aft and side-side deflection and possibly torsion. The modal
damping coefficients are an input defined by the user and may be used to represent structural
damping.
Because of the complexity of the coupling of the modal degrees of freedom of the rotating and
non-rotating components, the algebraic manipulation involved in the derivation of the
equations of motion for a wind turbine is relatively complicated, and the following only gives a
brief description of the theory.
Blade deflection
Nacelle yaw
Tower fore-aft, side-side and torsional deflection (axisymmetric tower model)
General tower deflection (multi-member tower model): a large number of modes is
allowed, including the torsional and axial degrees of freedom.
Mε Cε Kε σ
In cases with no prescribed strains it is straightforward to show, that the three system
matrices appearing on the left-hand side of the above equation become
where D r Dr1D is the time-dependent part of the reduction matrix. The right-hand side
stress vector of the global system of equations becomes
σ σa σ0 σi MTr g 2 DT r fa fi M rr g 2
where g2 Dr1a2 .
In general, the system mass matrix M is full, due to the coupling of the degrees of freedom,
and it contains periodic coefficients because of the time-dependent interaction of the
dynamics of the rotor and tower. The system damping and stiffness matrices C and K are
generally diagonal and constant.
Because of their complexity, further details of the equations of motion are not presented here.
The structural loads at the nodes, e.g. the blade root, the power train and the tower base are
calculated as the joint reactions by the component equilibrium equation described in Section
3.1. The inertial loads are calculated by integration of the mass properties and the total
acceleration vector at each station according to the principle of virtual work. The total
acceleration vector includes modal, centrifugal, Coriolis and gravitational components.
Second-order effects of axial forces due to structural deflections are taken into account, i.e.
the contribution to bending moments caused by the applied axial forces acting on the
deflected structure. For example, the contribution to tower base bending moment caused by
the weight of the nacelle takes into account the true position of the nacelle centre of gravity
including the deflection of the tower top.
With the modal model, the deformed shape of the flexible components like the blades and
tower at any instant is a linear combination of the selected mode shape functions. With a
reduced number of modes, the resulting deformation may therefore not be accurately
predicted, which means that it is not possible to calculate the internal forces directly from the
deformations as done by standard finite element technique.
In order to calculate the internal forces of flexible components the deformation at all station is
therefore calculated from a static equilibrium equation, where the applied force is calculated
as the sum of all external forces including the inertia forces. In case that some fundamental
degrees of freedom are constrained the system is solved with respect to a reduced set of
independent degrees of freedom, and the Lagrange multiplies corresponding to the
constraints are calculated. Finally the internal forces of all beam elements at both ends are
calculated from the fundamental equilibrium equation of the beam element.
nd
In order account for the 2 order effects of the external loads on the calculated section
forces, the external loads are transformed into the non-deformed state, where the static
equilibrium equation is solved, after which the calculated section forces are transformed back
to the deformed state. However, the transformation of the external loads are only done for the
equivalent loads at the stations, so this transformation is not completely correct for distributed
loads like gravity and inertial forces.
q
1
q0 b
N b 1
N
q
2
q cn b cos(n b )
N b 1
N
q
2
q sn b sin( n b )
N b 1
N
q
1
qd b ( 1)
b
N b 1
The notations are as follows: qb is a particular degree of freedom for blade b (for example a
flapwise or edgewise degree of freedom). There are N blades, and b is the azimuth angle
for blade b, and n is an integer which goes from 1 to (N-1)/2 if N is odd, and 1 to (N-2)/2 if N is
even. The transformation produces a collective mode q0, a set of cosine and sine modes qcn
and qsn (just one of each for a three-bladed turbine, and none for two-bladed). There is also a
differential mode qd which exists only for even N.
Such a set of modes exists for each degree of freedom. Here are a few examples:
For each blade degree of freedom there are two rotating degrees of freedom, q 1 and q2, and
they transform into two non-rotating modes, q0 and qd. If qb represents the first flapwise mode
of blade b for example, qo will be a symmetric out of plane rotor mode and qd will be
antisymmetric (like a teeter mode). If qb is the first edgewise mode, q0 will be the collective in-
plane rotor mode and qd the antisymmetric in-plane rotor mode. There are similar sets for the
second and higher blade modes. Other blade modes (e.g. torsional) would be treated in the
same way.
For each blade degree of freedom there are three rotating degrees of freedom qb for b = 1 to
N, and they transform into three non-rotating modes, q0, qc1 and qs1. q0 is a collective mode
and the other two are like d and q axis modes.
q Aq Bu
y Cq Du
q NR A NR q NR B PNR u
y C PNR q NR D PNR u
with
) U 1
A NR ( UA U
B PNR UB
C PNR CU 1
D PNR D
The rotational transformations are applied only to blade modal states, not to any other
individual-blade states such as pitch positions, rates, actuator internal states, etc. The matrix
U just has unit diagonal elements for rows and columns corresponding the states and state
derivatives which are not transformed. For other rows and columns the elements of U
represent the basic transformation defined above for each group of modes. Note that the
elements connecting states and state derivatives also need to be defined by differentiating the
equations of the basic transformation, bearing in mind that the derivative of the azimuth angle
is equal to the rotor speed (which is assumed constant for this purpose). Model inputs and
outputs are not transformed.
4. POWER TRAIN DYNAMICS
The power train dynamics define the rotational degrees of freedom associated with the drive
train, including drive train mountings, and the dynamics of the electrical generator. The drive
train consists of a low speed shaft, gearbox and high speed shaft. Direct drive generators
can also be modelled.
During a parked simulation, or once the brake has brought the rotor to rest during a stopping
simulation, the actual brake torque balances the aerodynamic torque exactly (making
allowance for the gearbox ratio if the brake is on the high speed shaft) and there is no further
rotation. However, if the aerodynamic torque increases to overcome the maximum or applied
brake torque, the brake starts to slip and rotation recommences.
The rigid drive train model may be used in combination with flexible drive train mountings. In
this case the equations of motion are more complex - see Section 4.3.
The turbine rotor is accelerated by the torque imbalance between the aerodynamic torque
and the low speed shaft torque. The generator rotor is accelerated by the imbalance between
high speed shaft torque and generator reaction torque. The shaft torques are calculated from
the shaft twist, together with any applied brake torque contributions depending on the location
of the brake, which may be specified as being at either end of either the low or high speed
shaft.
During a parked simulation, or once the brake disk has come to rest during a stopping
simulation, the equations of motion change depending on the brake location. If both shafts are
flexible, then both rotor and generator will oscillate. However, if the torque at the brake disk
increases to overcome the maximum or applied brake torque, then the brake starts to slip
again.
The flexible drive train model may be used in combination with flexible drive train mountings.
In this case the equations of motion are more complex - see Section 4.3.
It should be pointed out that while the flexible shaft model provides greater accuracy in the
prediction of loads, there is potential for one of the drive drain vibrational modes to be of
relatively high frequency, depending on the generator inertia and shaft stiffnesses. The
presence of this high frequency mode could result in slower simulations.
The generator characteristics must be provided if either the rigid or flexible shaft drive train
model is specified. Three types of generator model are available:
In each case there is a choice between a mechanical model and an electrical model. The
electrical model allows interactions with the network to be modelled. This is useful for
calculations of electrical flicker, voltage variations, power and reactive power variations and
power factor, and response to network transients such as voltage and frequency transients,
(which wind turbines are increasingly being required to ride through without shutting down).
Q 1 [h( 0 ) Q]
where is the actual generator speed and 0 is the generator synchronous or no-load speed.
Pr
h
r (r 0 )
where r is the generator speed at rated power output Pr , given by r = 0 (1 + S/100) where
S is the rated slip in %, and is the full load efficiency of the generator.
The equivalent circuit parameters should be given for a star-connected generator. If the
generator is delta-connected, the resistances and reactances should be divided by 3 to
convert to the equivalent star-connected configuration.
The voltage should be given as rms line volts. To convert peak voltage to rms, divide by 2.
To convert phase volts to line volts, multiply by 3.
Since this model necessarily includes electrical losses in the generator and ancillary
equipment, it is not possible to specify any additional electrical losses, although mechanical
losses may be specified - see Section 4.4.
Four different models of the electrical dynamics of the system illustrated in Figure 4.1 are
provided:
Steady state
1st order
2nd order
4th order
The steady state model simply calculates the steady-state currents and voltages in Figure 4.1
at each instant. The 1st order model introduces a first order lag into the relationship between
the slip (s) and the effective rotor resistance (Rr/s), using the short-circuit transient time
constant given by [4.1]:
X s X r xm2
(4.2.2.1)
X s Rrs
where Xs = xs+xm = sLs, Xr = xr+xm = sLr, and s is the grid frequency in rad/s.
The 2nd order model represents the generator as a voltage source behind a transient
2
reactance X’ = Xs - xm /Xr, ignoring stator flux transients:
is (rs + jX’) = vs -
where is and vs are the stator current and terminal voltage respectively. The dynamics of the
rotor flux linkage r may be written as
1
r rr ir js r (4.2.2.2)
s (1 s)
where s is the fractional slip speed (positive for generating) and ir is the rotor current. This
can be re-written in terms of the induced voltage using
xm
r j (4.2.2.3)
Xr
to give
r jX s X X
T0 s jssT0 j s vs (4.2.2.4)
rs jX rs jX
where
Xr
T0 . (4.2.2.5)
s rr
The 4th order model is a full d-q (direct and quadrature) axis representation of the generator
which uses Park’s transformation [4.2] to model the 3-phase windings of the generator as an
equivalent set of two windings in quadrature [4.3]. Using complex notation to represent the
direct and quadrature components of currents and voltages as the real and imaginary parts of
a single complex quantity, we can obtain
Ls Lr L
d is Lr rs js Ls Lr sLm
2
Lm rr js Lr Lm (1 s) is Lr
2
dt ir Lm rs js Ls Lm (1 s)
m
vs
Ls rr js L2m sLs Lr )ir Lm
(4.2.2.6)
Where speed of simulation is more important than accuracy, one of the lower order models
should be used. The 4th order model should be used for the greatest accuracy, although in
many circumstances the lower order models give very similar results. The lower order models
do not give an accurate representation of start-up transients, however.
Qd
Qg
(1 e s)
where Qd is the demanded torque, Qg is the air-gap torque, and e is the time constant of the
first order lag. Note that the use of a small time constant may result in slower simulations. If
the time constant is very small, specifying a zero time constant will speed up the simulations,
without much effect on accuracy.
The phase angle between current and voltage, and hence the power factor, is specified, on
the assumption that, in effect, both active and reactive power flows into the network are being
controlled with the same time constant as the torque, and that the frequency converter
controller is programmed to maintain constant power factor.
An option for drive train damping feedback is provided. This represents additional
functionality which may be available in the frequency converter controller which adds a term
derived from measured generator speed onto the incoming torque demand. This term is
defined as a transfer function acting on the measured speed. The transfer function is
supplied as a ratio of polynomials in the Laplace operator, s. Thus the equation for the air-
gap torque Qg becomes
Qd Num( s)
Qg g
(1 e s) Den( s)
where Num(s) and Den(s) are polynomials. The transfer function would normally be some
kind of tuned bandpass filter designed to provide some damping for drive train torsional
vibrations, which in the case of variable speed operation may otherwise be very lightly
damped, sometimes causing severe gearbox loads.
Notation:
C DC link capacitance, [F]
d, q d, q rotor reference frame
ids , iqs Stator currents in d and q-axis, [A]
ids ref , iqs ref Reference stator currents in d and q-axis, [A]
idwt , iqwt Grid side converter d and q-axis currents, [A]
ikd , ikq Damper winding currents in d and q-axis, [A]
iconmax Grid side converter maximum current, [A]
Llkd , Llkq Damper winding leakage inductance in d and q-axis, [H]
Lmd , Lmq Mutual inductance in d and q-axis, [H]
Lls Stator leakage inductance, [H]
X lkd , X lkq Damper winding leakage reactance in d and q-axis, [ohm]
X md , X mq Mutual reactance in d and q-axis, [ohm]
X ls Stator leakage reactance, [ohm]
P Number of pole pairs, [-]
Rs , Rkd , Rkq Stator, d-axis damper, q-axis damper resistance, [ohm]
Rchop Chopper protection resistor, [ohm]
Tc Torque demand from controller, [Nm]
vas , vbs , vcs Stator phase voltages [V]
vds , vqs Stator voltages in d and q-axis, [V]
vdc Wind turbine DC link voltage, [V]
vdc max Wind turbine DC link maximum voltage, [V]
vdc th Wind turbine threshold voltage, [V]
vdctol DC link voltage tolerance for chopper protection, []
vTerm Wind turbine terminal voltage, [V]
vTermmin Wind turbine DC link maximum voltage, [V]
vTermth Wind turbine threshold voltage, [V]
d , q Generator controller d and q-axis time constants, [s]
An electrical model for synchronous generator with fully rated converter is available, as shown
in Figure 4.2. Two four quadrant IGBT converters are used as generator side and grid side
converters. Back-to-back converters connected through a DC link decouple the operating
frequency of the generator from the grid. Therefore the generator can be operated at a
different speed from the grid synchronous speed.
Synchronous DC link
generator
Figure 4.2: Schematic: synchronous generator with fully rated IGBT-IGBT converters
The generator side converter rectifies variable frequency AC power to DC and then the grid
side converter inverts back to grid frequency AC. The rating of the converters has to be higher
than the rating of the generator as the total generator power flows through the converters.
Even though a gearbox is shown in Figure 4.2, a multi-pole synchronous generator can be
used for a direct drive wind turbine.
The electrical system in Figure 4.2, has two main control loops, a generator controller which
manipulates the generator torque, and a grid side controller which maintains the DC link
voltage. Whenever necessary, reactive power can be supplied to the grid by the grid side
converter.
d
vds Rs ids r qs ds
dt
d
vqs Rs iqs r ds qs
dt
The equations of the short circuited damper windings are written as
d
0 Rkd ikd kd
dt
d
0 Rkqikq kq
dt
Flux linkages in each circuit are written as
where the permanent magnet produces a rotor flux, M . The salient effect of the rotor is
taken into account by considering different values of the mutual inductance Lmd and Lmq . A
synchronous generator can be represented as (i) a steady state, (ii) a reduced order or (iii) a
full order model [4.7]. The steady state model neglects both stator and rotor transients. The
reduced order model neglects the stator transients while including rotor transients. The full
order model includes both rotor and stator transients.
Phase-to-phase fault
Assume that phase-b and phase-c are short circuited due an internal generator fault. In order
to have zero phase-a current, the applied voltage to phase-a which will force ias to remain
zero is
1 d a
vas
b dt
For phase to phase fault,
vbs vcs
Phase-to-ground fault
During the fault, phase-c is short circuited to the ground and phase-b and phase-a are open.
In order to have zero phase-a and phase-b currents, the following equations need to be
satisfied
1 d a
vas
b dt
1 d b
vbs
b dt
Generator saturation
Above a certain flux density, the iron core of an electrical machine is subjected to magnetic
saturation. In Bladed, generator saturation is modelled by representing inductance as a
function of current in both d and q-axes. For more information see [4.7].
Demagnetization
Figure 4.3 shows typical demagnetization curves of a permanent magnet machine at different
temperatures. At high temperatures, a demagnetization curve tends to shrink towards the
origin. As this shrinking occurs, the flux available from the magnets reduces. The
demagnetization curve returns to its original shape as temperature drops.
During a generator short circuit, there is a risk of irreversibly demagnetizing the permanent
magnets of the generator. This is because the strong opposing magnetic field created by
short circuit currents forces the magnet to operate in the area of the critical knee. Irreversible
demagnetization permanently damages the torque producing capability of the generator. For
more information on demagnetization see [4.10].
B
Increasing
temperature
Irreversible
demagnetization
Critical knee
Recoil
0 H
3
Te P M iqs
2
where P is the number of pole pairs of the generator. Therefore for a particular torque
demand ( Tc ) the reference stator current in the q-axis is calculated as
2 1 1
iqs ref Tc
3 P M
In order to generate maximum torque for a given current, the reference current in the d-axis
( ids ref ) is set to zero for a permanent magnet synchronous generator. Alternatively, the d-
axis current can be used to maintain unity power factor at the generator terminal. When the
required stator voltage magnitude exceeds the converter ceiling voltage, a field weakening
process can be carried out by manipulating the reference current in the d-axis.
The generator side converter functions as a voltage source, therefore the required voltage
vectors are determined using reference and measured currents, PI controllers and decoupling
loops as shown in Figure 4.4 [4.5].
r M
iqs ref PI vqs
iqs r Lqs
ids r Lds
ids ref PI vds
Figure 4.4: Control loops of the generator vector controller
Instead of using measured currents for the decoupling loops, the reference current can also
be used for simulation purposes. For simulation purposes the controller can be further
simplified as shown in Figure 4.5. A steady state representation of the vector control scheme
is shown in Figure 4.5 (a). This model neglects all the dynamics associated with the converter
control. Two first order lags are added to the steady state model in Figure 4.5 (b). Time
constants of the first order lags represent any delays in the converter control. Time constant
q , in q-axis, represents any delay in the torque control loop. The time constant d is to
represent any delay in the flux control loop.
r M r M
1
iqs ref Rs vqs Rs vqs
iqs ref 1 qs
r Lqs r Lqs
r Lds r Lds
1
ids ref Rs vds ids ref Rs vds
1 d s
(a) Steady state
(b) First order lag
Figure 4.5: Generator control options
The terminal voltage of the wind turbine is controlled by manipulating the reactive power from
the grid side converter using the current in the d-axis. This controller provides additional
reactive power during a grid fault. In order to ensure that the grid side converter supplies
more reactive current while not exceeding the current limit of the converter, the limit of the
active power current is dynamically adjusted depending on the reactive power current as
shown in Figure 4.6. Instead of controlling the terminal voltage, power factor of the grid side
converter can be controlled. Reference current in d-axis is determined for a given power
factor and active power output.
idwt max
vTerm ref Edwt
1stPId 2ndPId
*
idwt
vTerm
X conv
idwt
iqwt
X conv
A chopper resistor on the DC link may be used to dissipate the excess power. The chopper
protection scheme switches a resistance ( Rchop ) ON and OFF at the DC link depending on the
DC link voltage. The controller switches the chopper resistance ON when the DC link voltage
exceeds vdc max 1 vdc tol . The chopper resistance is switched OFF when the DC link voltage
drops below vdc max 1 vdc tol .
An alternative approach to maintaining the DC link voltage below the upper limit is to reduce
the generator power. The generator power can be reduced rapidly by means of reducing the
generator torque. A droop controller acting either on the terminal or the DC link voltage is
used to reduce the generator torque demand. The droop gain in Figure 4.7 is used as a
multiplier for the torque demand.
Above the terminal threshold voltage or below the DC link threshold voltage, the droop gain is
1. If the droop controller is acting on the terminal voltage, the droop gain is reduced to zero
linearly as the terminal voltage reaches its minimum set point. Similarly, the droop gain drops
towards zero as the DC link voltage reaches its maximum limit (when the droop controller is
acting on the DC link voltage).
r Tc
1
vterm 1
1 prot s
vterm th
vterm min
vdc 1
vdc max
vdc th
Figure 4.7: Torque control droop for wind turbine fault ride through
When using the torque control protection method, it may be necessary to drop the generator
torque to zero rapidly as possible to keep the DC link voltage below its limit. This torque step
reduction could induce significant oscillations in the drive train. Therefore the generator torque
can be reduced at a defined ramp rate that is sufficiently slow so as not to excite the shaft-
line. However this method requires a braking chopper to protect the DC link voltage.
DFIG
Torque Voltage or PF
control control
The converter is controlled by two main control loops, as described in [4.5]: a torque control
loop which works by injecting a quadrature-axis voltage into the rotor circuit, and a voltage or
power factor control loop which works by injecting a direct-axis voltage. Both loops are
essentially PI controllers, with additional terms for decoupling the two loops.
Rotor voltage increases with the slip speed but only until it reaches the rotor side converter
maximum voltage. The voltage limit on maximum speed occurs at minimum grid frequency
and operating with a capacitive power factor. At the top of the operating speed range, the
converter voltage limit will force the generator to absorb a large amount of reactive power
from the grid.
Generator modelling
Generator modelling is similar to the modelling of the synchronous generator given in section
4.2.5. The reference frame used for an induction generator is generally fixed to the stator flux
linkage. The d-axis of the reference frame is aligned with the stator flux vector and the q-axis
0
is 90 ahead of the d-axis.
d
vds Rs ids s qs ds
dt
d
vqs Rs iqs s ds qs
dt
The equations of the short circuited damper windings are written as
d
vdr Rr idr s r qr dr
dt
d
vqr Rr iqr s r dr qr
dt
Flux linkages in each circuit are written as
ds Lds ids Lmd idr
qs Lqsiqs Lmqiqr
dr Ldr idr Lmd ids
qr Lqr iqr Lmqiqs
3 3 v L
Te P ds iqs P s . m iqr
2 2 s Ls
where P is the number of pole pairs of the generator. Therefore for a particular torque
demand ( Tc ) the reference stator current in the q-axis is calculated as
2 2 L
iqr ref . s s .Tc
3 P vs Lm
It is assumed that the magnetising current is provided the rotor side converter and the
reference magnetising current in the d-axis is calculated as
1
idr ref vs
s Lm
The generator side converter functions as a voltage source, therefore the required voltage
vectors are determined using reference and measured currents, PI controllers and decoupling
loops as shown in Figure 4.9. Instead of using measured currents for the decoupling loops,
the reference current can also be used for simulation purposes.
sLm
vs
Ls
vqr
iqr ref PI
L2
iqr ss Lr m iqr
Ls
L2
idr ss Lr m idr
Ls
idr ref PI vdr
Figure 4.9: Control loops of the DFIG vector controller
Stator
terminal
Rotor
Converter
Crowbar
Figure 4.10: Crowbar circuit
There are a number of possible design solutions for the crowbar, however for the purposes of
this analysis the crowbar is assumed to be made up of a three phase diode rectifier and a
thyristor as shown in Figure 4.10. During a grid fault, the crowbar thyristor is switched on
when the rotor current reaches its maximum limit.
When the rotor current drops below the maximum limit and there is enough grid voltage the
rotor converter is gradually re-engaged to the rotor circuit. This switchover from the crowbar to
the rotor converter upon clearing the fault turns off the crowbar thyristor and removes the
crowbar from the rotor terminals.
Above rated, the variable slip generator uses a fast-switching controller to regulate the rotor
current, and hence the air-gap torque, so the generator actually behaves just like a variable
speed system, albeit with a limited speed range. The same parameters as for a variable
speed system must therefore also be supplied (see Section 4.2.3), with the exception of the
phase angle since power factor control is not available in this case.
Alternatively, a full electrical model of the variable slip generator is available. The generator
is modelled as in Section 4.2.2, and the rotor current controller is modelled as a continuous-
time PI controller which adjusts the rotor resistance between the defined limits (with integrator
desaturation on the limits), in response to the difference between the actual and demanded
rotor current. The steady-state relationship between torque and rotor current is computed at
the start of the simulation, so that the torque demand can be converted to a rotor current
demand. The scheme is shown in
Figure 4.11.
Torque Current 1 PI with Rotor
demand demand |I| limits resistance
If desired, torsional flexibility may be specified either in the gearbox mounting or between the
pallet or bedplate and the tower top. This option is only allowed if either the stiff or flexible
drive train model is specified, and it adds an additional rotational degree of freedom.
In either case, the torsional stiffness and damping of the mounting is specified, with the axis
of rotation assumed to coincide with the rotor shaft. The moment of inertia of the moving
components about the low speed shaft axis must also be specified. In the case of a flexible
gearbox mounting, this is the moment of inertia of the gearbox casing. In the case of a
flexible pallet mounting, it is the moment of inertia of the gearbox casing, the generator stator,
the moving pallet and any other components rigidly fixed to it.
If either form of mounting is specified, the direction of rotation of the generator shaft will affect
some of the internal drive train loads. If the low speed and high speed shafts rotate in
opposite directions, specify a negative gearbox ratio in the drive train model. The effect of
any offset between the low speed shaft and high speed shaft axes is ignored.
Any shaft brake is assumed to be rigidly mounted on the pallet. Thus any motion once the
brake disk has stopped turning depends on the type of drive train mounting as well as on the
position of the brake on the low or high speed shaft. For example if there is a soft pallet
mounting, then there will still be some oscillation of the rotor after the brake disk has stopped
even if both shafts are stiff.
As in the case of the flexible shaft drive train model, it should be pointed out that while
modelling the effect of flexible mountings provides greater accuracy in the prediction of loads,
there is potential for one or two of the resulting drive train vibrational modes to be of relatively
high frequency, depending on the various moments of inertia and shaft and mounting
stiffnesses. The presence of high frequency modes could result in slower simulations.
Power train energy losses are modelled as a combination of mechanical losses and electrical
losses in the generator (including the frequency converter in the case of variable speed
turbines).
Mechanical losses in the gearbox and/or shaft bearings are modelled as either a loss torque
or a power loss, which may be constant, or interpolated linearly from a look-up table. This
may be a look-up table against rotor speed, gearbox torque or shaft power, or a two-
dimensional look-up table against rotor speed and either shaft torque or power. Mechanical
losses modelled in terms of power are inappropriate if calculations are to be carried out at low
or zero rotational speeds, e.g. for starts, stops, idling and parked calculations. In these
cases, the losses are better expressed in terms of torque.
The electrical losses may specified by one of two methods:
Linear model: This requires a no-load loss LN and an efficiency , where the electrical power
output Pe is related to the generator shaft input power Ps by:
Pe = (Ps - LN)
Look-up table: The power loss L(Ps) is specified as a function of generator shaft input power
Ps by means of a look-up table. The electrical power output Pe is given by:
Pe = Ps - L(Ps)
Note that if a full electrical model of the generator is used, additional electrical losses in this
form cannot be specified since the generator model implicitly includes all electrical losses.
4.5 The electrical network
Provided either the detailed electrical model of the induction generator or the variable speed
generator model is used, so that electrical currents and voltages are calculated, and reactive
power as well as active power, then the characteristics of the network to which the turbine is
connected may also be supplied. As well as allowing the voltage variations, and hence the
flicker, at various points on the network to be calculated, the presence of the network may
also, in the case of the directly connected induction generator, influence the dynamic
response of the generator itself particularly on a weak network.
The network is modelled as a connection, with defined impedance, to the point of common
coupling (PCC in
Figure 4.12) and a further connection, also with defined impedance, to an infinite busbar.
Further turbines may be connected at the point of common coupling. These additional
turbines are each assumed to be identical to the turbine being modelled, including the
impedance of the connection to the point of common coupling. However they are modelled
as static rather dynamic, with current and phase angle constant during the simulation. The
initial conditions are calculated with the assumption that all turbines are in an identical state,
and the ‘other’ turbines then remain in the same state throughout. Thus the steady state
voltage rise due to all the turbines at the point of common coupling will be taken into account
in calculating the performance of the turbine whose performance is being simulated.
Other turbines
(if required)
Wind
turbine R1 + jX1 R2 + jX2
PCC
Farm Network
interconnection connection
impedance impedance
Infinite
busbar
5.1 Introduction
Closed loop control may be used during normal running of the turbine to control the blade
pitch angle and, for variable speed turbines, the rotor speed. Four different controller types
are provided:
1. Fixed speed stall regulated. The generator is directly connected to a constant frequency
grid, and there is no active aerodynamic control during normal power production.
2. Fixed speed pitch regulated. The generator is directly connected to a constant
frequency grid, and pitch control is used to regulate power in high winds.
3. Variable speed stall regulated. A frequency converter decouples the generator from the
grid, allowing the rotor speed to be varied by controlling the generator reaction torque. In
high winds, this speed control capability is used to slow the rotor down until aerodynamic
stall limits the power to the desired level.
4. Variable speed pitch regulated. A frequency converter decouples the generator from the
grid, allowing the rotor speed to be varied by controlling the generator reaction torque. In
high winds, the torque is held at the rated level and pitch control is used to regulate the
rotor speed and hence also the power.
For a constant speed stall regulated turbine no parameters need be defined as there is no
control action. In the other cases the control action will determine the steady state operating
point of the turbine as well as its dynamic response. For steady state calculations it is only
necessary to specify those parameters which define the operating curve of the turbine. For
dynamic calculations, further parameters are used to define the dynamics of the closed loop
control. The parameters required are defined further in the following sections.
Note that all closed loop control data are defined relative to the high speed shaft.
This controller is applicable to a turbine with a directly-connected generator which uses blade
pitch control to regulate power in high winds. It is applicable to full or partial span pitch
control, as well as to other forms of aerodynamic control such as flaps or ailerons. In the
latter case, the pitch angle can be taken to refer to the deployment angle of the flap or aileron.
From the optimum position, the blades may pitch in either direction to reduce the
aerodynamic torque. If feathering pitch action is selected, the pitchable part of the blade
moves to reduce its angle of attack as the wind speed (and hence the power) increases. If
stalling pitch action is selected, it moves in the opposite direction to stall the blade as the wind
speed increases. In the feathering case, the minimum pitch angle defines the pitch setting
below rated, while in the stalling case the maximum pitch angle is used below rated,
Figure 5.1: The fixed speed pitch regulated control loop
and the pitch decreases towards the minimum value (usually a negative pitch angle) above
rated.
Figure 5.1 shows schematically the elements of the fixed speed pitch regulated control loop
which are modelled.
This controller model is appropriate to variable speed turbines which employ a frequency
converter to decouple the generator speed from the fixed frequency of the grid, and which do
not use pitch control to limit the power above rated wind speed. Instead, the generator
reaction torque is controlled so as to slow the rotor down into stall in high wind speeds. The
control loop is shown schematically in Figure 5.2.
follow this curve in very low winds, and the turbine is then operated at nominally constant
speed along the line AB shown in the figure. Similarly in high wind speeds, once the
maximum operating speed S4 is reached, then once again it is necessary to depart from the
optimum load line by operating at nominally constant speed along the line GH.
Once maximum power is reached at point H, it is necessary to slow the rotor speed down into
stall, along the constant power line HI. If high rotational speeds are allowed, it is of course
possible for the line GH to collapse so that the constant power line and the constant tip speed
ratio line meet at point J.
Clearly the parameters needed to specify the steady state operating curve are:
The minimum speed, S1
The maximum speed in constant tip speed ratio mode, S4
The maximum steady-state operating speed. This is usually S4, but could conceivably be
higher in the case of a turbine whose characteristics are such that as the wind speed
increases, the above rated operating point moves from H to I, then drops back to H, and
then carries on (towards J) in very high winds. This situation is somewhat unlikely
however, because if rotational speeds beyond S4 are permitted in very high winds, there is
little reason not to increase S4 and allow the same high rotor speeds in lower winds.)
The above rated power set-point, corresponding to the line HI. This is defined in terms of
shaft power. Electrical power will of course be lower if electrical losses are modelled.
The parameter K which defines the constant tip speed ratio line BG. This is given by:
K = R Cp() / 2 G
5 3 3
where
= air density
R = rotor radius
= desired tip speed ratio
Cp() = Power coefficient at tip speed ratio
G = gearbox ratio
Then when the generator torque demand is set to K where is the measured generator
2
speed, this ensures that in the steady state the turbine will maintain tip speed ratio and the
corresponding power coefficient Cp(). Note that power train losses may vary with rotational
speed, in which case the optimum rotor speed is not necessarily that which results in the
maximum aerodynamic power coefficient.
As an alternative to the parameter K, a look-up table may be specified giving generator
torque as a function of speed.
Two closed loop control loops are used for the generator torque control, as shown in Figure
5.4. An inner control loop calculates a generator torque demand as a function of generator
speed error, while an outer loop calculates a generator speed demand as a function of power
error. Both control loops use PI controllers, as described in Section 5.6.
Below rated, the speed set-point switches between S1 and S4. In low winds it is at S1, and
the torque demand output is limited to a maximum value given by the optimal tip speed ratio
curve BG. This causes the operating point to track the trajectory ABG. In higher winds, the
set-point changes to S4, and the torque demand output is limited to a
Figure 5.4: Stall regulated variable speed control loops
minimum value given by the optimal tip speed ratio curve, causing the operating point to track
the trajectory BGH. Once the torque reaches QR, the outer control loop causes the speed
set-point to reduce along HI, and the inner loop tracks this varying speed demand.
This controller model is appropriate to variable speed turbines, which employ a frequency
converter to decouple the generator speed from the fixed frequency of the grid, and which use
pitch control to limit the power above rated wind speed. The control loop is shown
schematically in Figure 5.5.
regulated variable speed case described in Section 5.3.1, Figure 5.3. Above rated however,
the blade pitch is adjusted to maintain the chosen operating point, designated L. Effectively,
changing the pitch alters the lines of constant wind speed, forcing them to pass through the
desired operating point.
Once rated torque is reached at point H, the torque demand is kept constant for all higher
wind speeds, and pitch control regulates the rotor speed. A small (optional) margin is allowed
between points H (where the torque reaches maximum) and L (where pitch control begins) to
prevent excessive mode switching between below and above rated control modes. However,
this margin may not be required, in which case points H and L coincide. As with the stall
regulated controller, the line GH may collapse to a point if desired.
Clearly the parameters needed to specify the steady state operating curve are:
The minimum speed, S1
The maximum speed in constant tip speed ratio mode, S4
The speed set-point above rated (S5). This may be the same as S4.
The maximum steady-state operating speed. This is normally the same as S5.
The above rated torque set-point, QR.
The parameter K which defines the constant tip speed ratio line BG, or a look-up table.
This is as defined in Section 5.3.1.
Below rated, the speed set-point switches between S1 and S4. In low winds it is at S1, and
the torque demand output is limited to a maximum value given by the optimal tip speed ratio
curve BG. This causes the operating point to track the trajectory ABG. In higher winds, the
set-point changes to S4, and the torque demand output is limited to a minimum value given
by the optimal tip speed ratio curve, causing the operating point to track the trajectory BGH,
and a maximum value of QR. When point H is reached the torque
remains constant, with the pitch control loop becoming active when the speed exceeds S5.
First order lag models are provided in Bladed to represent the dynamics of the power
transducer and the generator speed transducer. The first order lag model is represented by
1
y ( x y)
T
where x is the input and y is the output. The input is the actual power or speed and the output
is the measured power or speed, as input to the controller.
The pitch actuator may be modelled as either a pitch position or pitch rate actuator, and either
active or passive dynamics may be specified.
The simplest model is a passive actuator, with the relationship between the input and the
output represented by a transfer function. For the pitch position actuator, the input is the pitch
demand generated by the controller and the output is the actual pitch angle of the blades. For
the pitch rate actuator, the input is the pitch rate demand generated by the controller and the
output is the actual pitch rate at which the blades move. The transfer function may be a first
order lag, a second order response, or a general transfer function, up to 8th order.
y 2 y 2 ( x y )
where is the bandwidth and the damping factor. The general transfer function model is
represented by numerator and denominator polynomials in the Laplace operator.
For detailed calculations, especially to understand the loads on the pitch actuator itself and
the duty which will be required of it, it is possible to enter a more detailed model. This can
take into account any internal closed loop dynamics in the actuator, and also the pitch motion
resulting from the actuator torque acting on the pitching inertia, with or against the
aerodynamic pitch moment and the pitch bearing friction. The bearing friction itself depends
critically on the loading at the pitch bearing.
Figure 5.8 shows the various options for controlling the pitch angle, starting from either a pitch
position demand or a pitch rate demand. The pitch position demand may optionally be
processed through a ramp control, shown in Figure 5.9, which smoothes the step changes in
demand generated by a discrete controller by applying rate and/or acceleration limits. Then
the pitch position demand can act either through passive dynamics to generate a pitch
position, or through a PID controller on pitch error to generate a pitch rate demand. Rate
limits are applied to the output, with instantaneous integrator desaturation to prevent wind-up
in the PID case. Thus the pitch rate demand may come either from here or directly from the
controller. This rate demand can act either through passive dynamics to generate a pitch
rate, or through a PID controller on pitch rate error to generate an actuator torque demand. In
the latter case, the pitch actuator passive dynamics then generate an actual actuator torque,
which acts against bearing friction and any aerodynamic pitching moment to accelerate the
pitching inertia of the blades and the actuator itself. An optional first order filter on each PID
input allows step changes in demand from the controller to be smoothed, and instantaneous
integrator desaturation prevents wind-up when the torque limits are reached.
Both PID controllers include a filter on the differential term to prevent excessive high
frequency gain. Also there is a choice of derivative action, such that the derivative gain may
be applied either to the feedback (i.e. the measured position or rate), the error signal, or the
demand. The latter case represents a feed-forward term in the controller.
If passive pitch rate dynamics are selected, the response will be subject to acceleration limits
calculated from the aerodynamic pitching moment, bearing friction and the actuator toque
limits acting on the pitching inertia. If the total pitching inertia is zero, no limits will be applied.
The pitch bearing sliding friction torque is modelled as the sum of four terms: a constant, a
term proportional to the bending moment at the bearing, and a term proportional to the axial
and radial forces on the bearing. Sometimes the actuator cannot overcome the applied
torques and the pitch motion will stick. Before it can move again, the break-out or ‘stiction’
torque must be overcome. This is modelled as an additional contribution to the friction torque
while the pitch is not moving. This additional contribution is specified as a constant torque,
plus a term proportional to the sliding friction torque.
Pitch position Measured Pitch rate
Measured
demand from pitch demand from
pitch rate
controller position controller
Bearing
- loads
Ramp control +
Pitching Bearing
moment friction
PID controller
Actuator
torque
limits
Pitch rate
demand
-
+
Pitching
inertia
Acceleration
Passive
limits PID controller
dynamics
Actuator
Passive
torque
dynamics
demand
Passive
dynamics
Actuator
torque
Pitching
inertia
Pitch rate
Actual pitch
position
0.8
Demand
0.6
Bladed supports both rotary and linear pitch actuators. For linear actuators, for example
using a hydraulic ram, the linkage geometry must be supplied. The principal difference
between the linear and rotary actuator models is that a rotary actuator has a torque limit,
while a linear actuator has a force limit (which translates into a torque limit which varies with
pitch position).
All the closed loop control algorithms described above use PI controllers to calculate the
output y (pitch, torque or speed demand) from the input x (power or speed error). The basic
PI algorithm can be expressed as
y K p x Ki x
where Kp and Ki represent the proportional and integral gains. The ratio Kp/Ki is also known
as the integral time constant. Calculation of appropriate values for the gains is a specialist
task, which should take into account the dynamics of the turbine together with the
aerodynamic characteristics and principal forcing frequencies, and should aim to achieve
stable control at all operating points and a suitable trade-off between accuracy of tracking the
set-point and the degree of actuator activity.
The choice of variable V depends on the particular control loop. The following choices are
provided:
The variables shown in bold are normally recommended. Gain scheduling is unlikely to be
required for the variable speed below rated controllers. For the variable speed stall regulated
above-rated controller, no general rule can be given. Gain scheduling on wind speed is not
usually a practical proposition because of the difficulty of measuring a representative wind
speed, and this option is only provided for research purposes. The wind speed used is the
hub wind speed, which may differ from any wind speed measured by an anemometer
mounted on the nacelle, especially in the case of an upwind turbine.
Gain scheduling on pitch angle is recommended for the pitch regulation controllers, to
compensate for the large changes in the sensitivity of aerodynamic torque to pitch angle over
the operating range. The steady loads calculation may be used to calculate the partial
derivative of aerodynamic torque with respect to pitch angle, and F may be set proportional to
this. In many cases, simply setting F proportional to pitch angle is a good approximation, but
a lower limit for F must be set to prevent excessive gains at small pitch angles.
The equation governing the rotational dynamics for the rigid turbine is:
Q
J aero Q gen Q Q Q V V
' ' '
J Total inertia
Qaero Aerodynamic torque (steady value)
Qgen Generator torque (referred to low speed shaft)
Rotor speed (deviation from steady value)
Pitch angle (deviation from steady value)
V Wind speed (deviation from steady value)
Q ' Partial derivative of torque w.r.t. rotor speed
Q ' Partial derivative of torque w.r.t. pitch angle
Q 'V Partial derivative of torque w.r.t. wind speed
The plant P is the transfer function from generator torque to rotor speed:
1
P
Q ' Js
Ignoring all sensor and actuator dynamics, the PI controller is C 1s K i sK p where Ki and
Kp and the integral and proportional gains, and the open loop transfer function H = P*C is
given by
K sK p
H
i
s Q ' Js
The closed loop transfer function is H/(1+H), so the closed loop poles are the solutions s of
1+H = 0:
Js 2 K p Q ' s K i 0
K i J02
K p 2J0 Q '
A damping factor of = 1/2 is selected for critical damping, and 0 is calculated by setting
the cross-over frequency c to a defined value, e.g. c = 1 rad/s. The cross-over frequency is
the frequency at which |H| = 1, so 0 can then be found by solving
K i c K p
1
i
i c Q ' Ji c
The plant P is now the transfer function from pitch angle to rotor speed, i.e.
Q '
P
Q ' Js
or Q times the torque loop plant. The same analysis can then be carried out, except that
'
now
K i J 02 / Q '
2J 0 Q '
Kp
Q '
The pitch PI gains are calculated for each above-rated operating point and a least-squares
procedure is used to calculate a gain schedule to fit for both gains simultaneously, ignoring
points very close to fine pitch where a maximum gain limit is imposed.
Note: the PI gains in Bladed are defined in terms of measured generator speed, and the
torque PI gains are also defined in terms of generator torque referred to the high speed shaft.
Therefore the pitch PI gains derived above must be divided by the gearbox ratio, and the
torque PI gains must be divided by the square of the gearbox ratio.
The variable speed controllers, both stall regulated and pitch regulated, require the following
mode changes:
Change of speed set-point from S1 to S4 (refer to Figures 5.3 and 5.6). This occurs when
the measured speed crosses the threshold value (S1+S4)/2. This mode change is
completely benign as the control action along the optimum tip speed ratio line BG is the
same either side of the mode change point, so no hysteresis is required.
For the stall regulated case, the change from below rated to above rated is also benign.
Making the switch in the middle of the section GH of Figure 5.3 causes no immediate change
in control action. However, in the case of G and H coinciding, or being very close together, it
may be necessary to modify the mode change strategy, depending on the turbine
characteristics.
For the pitch regulated case, the change to above rated control occurs when the torque
demand is at maximum (QR) and the speed exceeds S5 (refer to Figure 5.6). The change to
below rated occurs when the pitch demand is at fine pitch (minimum pitch for the feathering
case, maximum pitch for pitch-assisted stall) and the speed falls below S4. While this
strategy is usually suitable, it may be desirable to modify it depending on the turbine
characteristics.
The mode changes occur on a discrete timestep set to a default value of 0.1 seconds.
The control algorithms described above have been developed to be suitable for a wide range
of cases. However, these are very basic algorithms, and it is recognised that in reality there is
great variation in the design of controllers for different turbines.
Bladed therefore offers the possibility of incorporating user-defined controllers in the dynamic
simulations. Through a defined interface, a user’s external control program, written in any
language and compiled either as an executable or (preferably) a DLL, can be used to control
the simulation. Like a real controller, the external controller runs on a discrete timestep.
Using this facility, Bladed simulations can use any real control algorithm, and provide a very
useful means for testing new control algorithms.
For example, the user-defined controller may do any of the following:
Collective or individual blade pitch angle or pitch rate control during any phase of operation
including power production, stops, starts, idling etc.
Generator torque control for variable speed or variable slip turbines
Switch between generators for two-speed turbines
Control the generator contactor, allowing the generator to be switched on or off for
simulating stops and starts
Control the shaft brake, to simulate transitions between parked, idling, starting, stopping,
and power production states.
Control of nacelle yaw to simulate closed loop yaw control algorithms and/or yawing
strategies for start-up, shutdown etc.
The User Manual describes how to write an external user-defined controller. In an ideal
situation, the principal algorithm code modules could be shared between the Bladed external
controller code and the code in the actual controller hardware used on the turbine.
Alternatively, it is possible to use a Bladed external controller which communicates directly
with the turbine controller hardware. Bladed also incorporates an option to run in real time,
which allows it to be used as a virtual turbine for testing the real controller.
5.10 Signal noise and discretisation
When a discrete external controller is used, Bladed offers the possibility of adding random
noise to the measured signals sent to the controller, and also to discretise the signals to a
specified resolution.
The random noise may be Gaussian, in which case the standard deviation of the noise must
be specified, or it may be from a rectangular distribution, in which case the half-width of the
distribution should be given. The noise is added to the signal before it is discretised.
6. SUPERVISORY CONTROL
This section of the manual covers the modelling of the following aspects of turbine operation:
Start-up
Shut-down (normal and emergency stops)
Non-operational situations (rotor parked or idling)
Operation of the shaft brake
Teeter restraints
Yaw control
The standard implementation of these features in the simulation model is described. As in the
case of Closed Loop Control, alternative supervisory control logic can be incorporated in a
user-defined controller - see Section 5.9.
6.1 Start-up
Simulation of a wind turbine start-up begins with the rotor at a specified speed (usually but not
necessarily zero) and the generator off-line. The brake is assumed to be released at the start
of the simulation (i.e. at time zero).
If blade pitch or aileron control is available, the initial pitch or aileron angle is specified, along
with a constant rate of change which continues until either a specified angle is reached or the
closed loop controller takes over.
When a specified rotational speed is reached, the generator comes on line, and the closed
loop controller begins to operate. The simulation continues until the specified simulation end
time.
In the case of a variable speed turbine, there may be a transition period after cut-in of the
closed loop controller before the turbine is fully in the normal running state. There are two
different cases:
Variable speed pitch regulation: in the case when the pitch angle has not yet reached the
normal operating value (‘fine pitch’) at the moment when the closed loop controller cuts in,
then the pitch change rate for start-up continues to apply until either fine pitch is reached, or
until the conditions of Section 5.8 for starting the closed loop pitch controller are satisfied.
Variable speed stall regulation: when the closed loop controller cuts in, the above-rated
control mode is assumed to apply initially. In practice this assumption does not affect the
start-up since in low winds the operating point would be constrained by the quadratic
optimum-Cp characteristic in any case.
A normal stop is initiated at a specified time after the start of the simulation. Normal operation
in power production mode is assumed prior to this point, with full structural and control
dynamics in effect if desired. The structural dynamics continue in effect during the entire
simulation.
The standard logic for a normal stop is to start pitching the blades (or moving the ailerons) at
a specified rate from the moment that the stop is initiated, continuing until a final pitch angle is
reached. The generator is taken off-line when the electrical power reaches zero. If the built-
in variable speed power production controller is used, the torque controller continues to
1
operate normally until this point .
Once the rotational speed drops below a specified value, the shaft brake is applied to bring
the rotor to rest.
The simulation continues until the rotor comes to rest, or for a certain time longer if so desired
in order that the transient loads can be simulated as the brake disk stops. However, the
simulation end time overrides this, so it must be set long enough for the stop event to be
completed.
If there is no pitch control, the brake trip speed may be set high so that the shaft brake is
applied immediately at the initiation of the stop.
Section 6.4 describes the dynamic characteristics of the shaft brake itself.
An emergency stop is initiated at a specified time after the start of the simulation. Normal
operation in power production mode is assumed prior to this point, with full structural and
control dynamics in effect if so desired. The structural dynamics continue in effect during the
entire simulation.
Several options are available for simulating emergency stops. In all cases it is assumed that
the generator load is lost at the initiation of the emergency stop, whether because of grid
failure or some electrical or mechanical failure of the turbine.
Pitch (or aileron) action is initiated either immediately or when the rotational speed exceeds a
specified value. A fixed pitch rate then applies until a final pitch angle is reached. Provision
is made for the pitch of one or more of the blades to ‘stick’ at a specified angle to simulate
failure of a pitch bearing or actuator.
The shaft brake can also be applied either at the initiation of the stop or when a specified
overspeed is reached. Section 6.4 describes the dynamic characteristics of the shaft brake
itself. There is also a rotational speed below which the shaft brake is applied for parking, in
the event that it has not already been applied because of load loss or overspeed.
The simulation continues until the rotor comes to rest, or for a certain time longer if so desired
in order that the transient loads can be simulated as the brake disk stops. However, the
simulation end time overrides this, so it must be set long enough for the stop event to be
completed.
When the shaft brake is applied, either during a normal or an emergency stop, the full braking
torque is not available instantly. Instead, the torque builds up to the full value over a short
1
Simulations using dtbladed.exe version 2.85 or earlier with the built-in variable speed controller used a
simpler strategy in which the generator was disconnected as soon as the minimum generator speed was
reached.
period of time. This torque build-up may be modelled as either a linear torque ramp, or by
specifying a look-up table giving achieved braking torque as a function of time.
For simulations in the idling and parked states, a fixed pitch angle is specified, the generator
is off line, and there is no pitch control action. In the case of a parked rotor the shaft brake is
applied, and the rotor azimuth must be specified. The azimuth is measured from zero with
blade 1 at top dead centre.
All specified structural dynamics will be in effect during these simulations. This also allows for
the possibility of the shaft brake slipping during a parked simulation if the shaft torque
exceeds the specified brake torque.
1. One fixed-rate yaw manoeuvre may be specified, starting at a given point in any
simulation. This represents a change in the nominal nacelle position through a given
angle at a specified angular speed.
2. A user-defined controller (Section 5.9) may be used to specify either the yaw rate or the
yaw actuator torque at any time.
If active yaw is used to control the yaw rate, the effect of this is to change the ‘demanded
nacelle angle’ in a specified way. The actual nacelle angle depends on the yaw dynamics -
see next section.
1. Rigid yaw: the actual nacelle angle exactly follows the ‘demanded nacelle angle’ 0.
2. Flexible yaw: a certain amount of flexibility is present, usually in the yaw actuation system,
such that the actual nacelle angle may not follow the ‘demanded nacelle angle’ 0
exactly. The extreme case is free yaw, when the demanded nacelle angle does not have
any effect.
3. Controlled yaw torque: this is available only with an external controller to define the yaw
actuator torque demand.
Demanded yaw rate
Applied torque*
Tow
Controlled er
torque Yaw spring
Friction Damper
*(Aerodynamic and inertial yaw torque)
In the case of flexible or free yaw, the yaw damping Dy may be specified. This specifies a
torque Qd which opposes the yaw motion, given by
Qd Dy (0 )
In the case of flexible yaw, a yaw spring may be specified either as a linear spring or as a
hydraulic accumulator system such as is often used to provide flexibility in hydraulic yaw
drives. The hydraulic system is assumed to be double-acting, with one accumulator (or set of
accumulators) on either side of the yaw motor. The torque opposing the motion is provided
by compression of the gas in the accumulators. If the nominal gas volume is V 0 and the
instantaneous gas volumes either side of the yaw motor are v1 and v2 then the opposing
torque Qk is given by
V0 V0
Qk KP0
v1 v2
Although not strictly a supervisory control function, the teeter restraint model available in
Bladed for teetered rotors is described here. The model allows a linear variation of restoring
torque with teeter angle, but also allows a free teeter range and an initial pre-load. Figure 6.1
defines the relevant parameters. Linear damping is also allowed, giving an additional torque
contribution proportional to teeter rate.
7. MODELLING THE WIND
The wind field incident on the turbine may be specified in a number of ways. For some simple
calculations, a uniform, constant wind speed is assumed, such that the same incident wind
speed is seen by every point on the rotor. For more detailed calculations however, it is
important to be able to define both the spatial and temporal variations in wind speed and
direction.
The steady-state spatial characteristics of the wind field may include any combination of the
following elements:
The wind direction must also be specified, both relative to the direction in which the nacelle is
pointing (to define the yaw error), and relative to the horizontal plane (to define the upflow
angle). The latter effect may be important for turbines operating in hilly terrain.
For simulations, it is also important to be able to define how the wind speed and direction vary
with time. The following alternative models are provided:
Wind shear is the variation of steady state mean wind speed with height. Three alternative
models are provided, to relate the wind speed V(h) at height h above the ground to the wind
speed V(h0) at some reference height h0..
h
V ( h) V ( h0 )
h0
Specifying the exponent as zero results in no wind speed variation with height.
Various forms of temporal variation of wind speed and direction may be superimposed on the
spatial variations described in Section 7.1 above and Chapter 9.
Fs0 is the wind shear factor from the reference height (for mean speed V0 ) to the hub height,
Fs is the wind shear factor from the hub height to the point (r,),
FT is the tower shadow factor for the point (r,), and
FW is the upwind turbine wake factor for the point (r,).
Wind speed
Wind direction
Horizontal shear (linear variation of wind speed from one side of the rotor to the other)
Vertical shear (linear variation of wind speed from bottom to top of the rotor)
Each may be either a half-wave transient or a full-wave transient. The transients are
sinusoidal, with a more complex shape defined in edition 2 of the standard [7.7]. The
parameters needed to define each transient are the starting value Y0, the start time t0, the
duration T, and the amplitude A. These parameters are illustrated in
Figure 7.1.
The actual wind speed at radius r, azimuth and time t is then given by:
where V0 is the starting wind speed at the reference height, Vtrans is the combined effect of the
wind speed and horizontal and vertical shear transients, and other parameters as defined in
Section 7.4.2.
12.5
Half wave
Y0 + 12
A
11.5
11
Full wave
10.5
Y0 10
9.5
IEC edition 2
9
-0.2 t00 0.2 0.4 0.6 0.8
t0 1+ T 1.2
Time
For fatigue load calculations, Bladed creates a series of irregular waves based on linear Airy
theory. The amplitude and frequency content of these waves are specified by the user in
terms of a power spectral density function. This may be either:
For extreme load calculations, a regular wave train may be defined. The kinematics of this
wave are calculated using stream function theory.
Offshore wind turbines are most likely to be installed in relatively sheltered inshore conditions,
where the sea depth is in the range 5m to 25m. Bladed assumes that the tower is fixed to the
sea bed as a simple monopile as shown in Figure 8.1 below. The tower may be defined over
the full depth (Figure 8.1a) or above a rigid base (Figure 8.1b). In both cases, the turbine
structure is regarded as being transparent to the waves, implying that both tower and base
are slender in comparison to the wavelength.
As for onshore cases, the tower is assumed to have a circular cross-section and may be
tapered. Foundation translational and rotational stiffnesses may also be specified.
f
5
f
4
S f 2 H T exp 1.25
2
f
s p
f
p p
where f is the wave frequency (in Hz), H s is the significant wave height, T p is the peak
spectral period, f p 1 Tp , is the JONSWAP peakedness parameter,
0.0624
2
0.185
0.230 0.0336
1.9
f
2
1
exp 0.5
fp
and
0.07 for f fp
0.09 for f fp
The Pierson-Moskowitz spectral density function may be regarded as a special case of the
JONSWAP spectrum with 1.0 :
f
5
f
4
If the JONSWAP / Pierson-Moskowitz option is selected, the user is required to enter values
for H s , T p and .
When the wavelength of waves impacting on an offshore structure becomes comparable with
the dimensions of the structure (e.g. the diameter of a monopile), diffraction effects begin to
occur. Waves which have wavelengths much smaller than the diameter of the structural
member do not contribute much to the net force because regions of positive and negative
velocity are experienced by the member at the same time. Bladed provides two different
approaches for including these effects: an approximation based on McCamy-Fuchs theory,
and a simple frequency cut-off.
In the case of a vertical cylindrical cylinder with no diffraction effects, Cm = 1. Therefore in the non-
diffraction case, CM = 2.
In the MacCamy Fuchs approximation, the hydrodynamic inertia coefficient CM has the following form:
2
CM = 4A(kro) / π(kro)
Where:
k = wave number
ro = radius of cylinder
2 2 -1/2
A(kr0) = (J1’ (kr0) + Y1’ (kr0))
This reduces CM as k increases. As k tends to zero, CM tends to a value of 2, equal to the non-
diffraction case.
Since it is not possible to modify CM in the time domain within Bladed, CM is held fixed and the
change in hydrodynamic forcing predicted by the MacCamy Fuchs approximation is
reproduced by applying an identical functional form to the entire Hydrodynamic force
equation, considering it to be a modification of ẍ rather than CM.
This modification to ẍ is produced via modifying the wave energy spectrum. The standard wave
2
energy spectrum is the Pierson Moskowitz distribution, in which S(k) α H . In sinusoidal wave theory,
ẍ α H (wave height) giving S(k) α ẍ
2
S f 0 for k
1
radius
For a monopile, the radius is taken as the minimum tower radius between the sea bed and a
height of 3 standard deviations of the wave elevation above the mean water level. At any
instant, the wave elevation has a probability of 99.85% of being within this range.
Alternatively the cut-off frequency may be specified by the user.
In the case of Irregular Waves and Regular Waves (Linear Airy option), water particle
kinematics are based on linear Airy theory. The following equations describe the wave particle
velocity vector uw uwx , uwy , uwz , the corresponding acceleration vector
uw uwx , uwy , uwz , the hydrodynamic component of the pressure p and the water surface
elevation for a regular wave of height H and period T at the point x, y, z :
H
u wx cos w coshk d z cos t
2 sinh kd
H
u wy sin w coshk d z cos t
2 sinh kd
H
u wz sinhk d z sin t
2 sinh kd
2H
u wx cos w coshk d z sin t
2 sinh kd
2H
u wy sin w coshk d z sin t
2 sinh kd
2H
u wz sinhk d z cos t
2 sinh kd
gH
p coshk d z cos t
2 coshkd
cos t
H
2
where 2f is the angular wave frequency, f is the wave frequency, t is time, d is the
water depth (assumed to be constant), is the water density, g is the acceleration due to
gravity and
kx cos w ky sin w
where w is the direction from which waves arrive at the tower. The wave number k is found
as the solution to the dispersion relation:
2 gk tanh kd
The co-ordinate system used for the wave and current calculations is a right-handed
Cartesian system in which the xy plane is horizontal with the x-axis pointing to the North, the
y-axis pointing to the West and the z-axis pointing vertically upwards. The origin of the co-
ordinate system lies where the tower centre line intersects the mean water level. Angles are
defined relative to the x-axis (North) and increase positively toward the East.
For the calculation of regular extreme waves, the above equations are used directly to
calculate the wave particle kinematics at each submerged tower station. For fatigue load
calculations, however, it is necessary to calculate an irregular (i.e. random, non-repeating)
series of waves. This is achieved using the filtered white noise ‘shift register’ procedure
described in section 8.6 below.
A limitation of Airy theory is that it only defines wave particle kinematics up to the mean water
level (z = 0). The theory can be extended above the mean water surface, up to the level of the
wave crest, by using the Airy formulae with positive values of z. However this approach
causes calculation difficulties and is known to over-estimate particle velocities and
accelerations in the crest region and to underestimate velocities and accelerations in the
troughs. To avoid these difficulties, Bladed uses Wheeler stretching [8.3] to take account of
the forces acting between mean water level and the instantaneous free surface. Experimental
results by Gudmestad [8.4] indicate that Wheeler stretching provides satisfactory estimates of
particle kinematics in the free surface zone in deep water.
Wheeler stretching assumes that particle motions calculated using Airy theory at the mean
water level should actually be applied at the instantaneous free surface. Airy particle motions
calculated at locations between the sea-bed and mean free surface are shifted vertically to
new locations in proportion to their height above the sea bed. Airy wave particle kinematics
calculated at a vertical location z are therefore applied to a new location z defined by:
d t
z z t
d
During a simulation in which waves are specified, the following records are synthesised:
For irregular waves, these records are created by the digital filtering of pseudo-random white
noise. A single white noise record is used, together with a different filter for each time history
to be generated. Because each filter introduces the correct amplitude variation and phase
shift, the resulting output time histories display the correct amplitude and phase relationships
to each other. Unlike the generation of turbulent flow fields, which are generated and written
to a file before running the simulation, wave data are generated as the simulation proceeds.
The relationship between the parameter of interest (i.e. the wave particle velocity at the first
tower station, the particle acceleration at the sea bed etc.) and the water surface elevation is
defined in terms of a complex function of the wave frequency known as a Response
Amplitude Operator (RAO). It is represented as a complex number of the form:
RAOr Rr e i r
The filters used to process the pseudo-random white noise are Finite Impulse Response (FIR)
filters and are defined in terms of their frequency transforms. The transformed filter for
response r is given by:
S f m f
z m ,r Rr f m expi r f m
4N
z m, r zm, r
where
f m m f
f f max N
and m is in the range 0m N .
The filter weights are then obtained as the transform of the expression:
N
imn
wn,r
m N 1
zm,r exp
N
Having generated the filter functions for each parameter at each required location, time
histories are generated using a shift-register technique. Firstly an N-element array of
normally-distributed random numbers is created. The random numbers are generated by
converting the output of a simple random number generator to a normally distributed deviate
with zero mean and unit variance using the Box-Muller method. For each filter function in turn,
the N filter weights are multiplied by the values of the equivalent elements in the random
number array and the N products are then summed to give the value of the property at one
particular instant in time. To calculate the value of the property at the next time step, the
elements of the random number array are ‘shifted’ one place higher in the array, a new
random number is introduced at element 1 and the multiplication and summation process is
repeated.
Bladed offers two methods for modelling regular waves. Linear Airy waves are as described
in Section 8.4; again Wheeler stretching is used. If the Stream Function option is selected, the
wave kinematics are calculated using stream function theory. This method is more accurate
than linear wave theory in cases where the wave height is a significant proportion of the mean
water depth. The method may even be used to model waves with amplitudes close to the
breaking wave limit. In cases when currents are specified in addition to regular waves (see
Section 8.8), the wave calculation takes proper account of the influence of the current profile
on the wave kinematics. The non-linear regular wave calculations within Bladed are based on
original coding by Chaplin [8.5].
Regardless of whether current components are specified, Bladed first solves the wave
equation using stream function theory for the case of no currents. Stream function theory was
first applied to wave modelling by Dean [8.6 & 8.7] who developed the following form of
stream function:
X1
( x, z ) z n2 X n sinh( nk ( z d )) cos(nkx)
N
T
where
X 1 wavelength
X n 1 n
The stream function as defined above satisfies the requirements that (i) the shape of the
free surface is compatible with the motion of the water just below it (the Kinematic Free
Surface Boundary Condition), (ii) the flow is periodic, and (iii) the flow is compatible with the
presence of a horizontal sea bed at the specified depth. The values of X n are determined by
a least-squares method to satisfy the additional requirements that (i) the pressure on the free
surface is uniform (the Dynamic Free Surface Boundary Condition), and (ii) the required wave
height is obtained. As implemented within Bladed, the order of the solution, N, is
automatically chosen based on the input values of wave height, period and mean water
depth.
Once the stream function solution has been obtained, the horizontal and vertical velocities (in
the absence of a current) are calculated using the relations:
u and v
y x
In cases where a current profile is specified the flow is in general rotational and the wave
solution must be modified. The method used follows the approach developed by Dalrymple
[8.8 & 8.9] and is based on coding by Chaplin [8.5]. It is assumed that the relationship
between the vorticity and stream function is the same for the combined flow as for the
undisturbed current when viewed from a reference frame moving at such a speed that total
flow rate is the same as that in the x,y frame. This requirement can be stated mathematically
as:
2 2
f ( )
x 2 y 2
In Bladed, the stream function is computed at discrete points in the x,y plane using a
finite-difference calculation scheme. The most difficult feature of this approach is that the
location of the free surface is not known in advance. A regular grid of points in the x,y plane
would therefore have awkward intersections with the free surface profile, which must itself be
calculated as part of the computation. To overcome this difficulty, Dubreil-Jacotin’s method is
used to transform the problem from the x,y plane to the x, plane, with y as the field
variable. The position of the free surface is now defined along the upper boundary of a
rectangular grid in the x, plane. Treating x and as the independent parameters, the
velocity components are now given by:
y
u
1
and v x
y y
The accuracy of the solution relies on a sufficiently fine mesh in the x, plane to resolve the
structure of the flow and to allow the evaluation of derivatives on the boundaries of the
computational domain, particularly at the free surface. For this purpose a regular grid in the
x, plane is rather inefficient and therefore a stretched grid is employed which is finer near
the free surface than the sea bed.
After solving the finite difference relations on this plane, the flow velocities are calculated
using the equations above and dynamic pressures are calculated using Bernoulli’s equation.
Reference [8.5] should be consulted for further details of this method.
8.8 Constrained Waves
Bladed allows the user to include one constrained wave in each irregular wave simulation.
The simulation is otherwise irregular, but the user can constrain it to have a wave of desired
height at a specified time. Bladed uses “NewWaves” to modify the irregular sea around the
specified time so that the constraint is met. A NewWave, by definition, describes the most
likely surface elevation around a peak (or trough). It can be computed from the spectrum:
( ) ( )
16
( ) 2 S ( f ) cos(2f )2df
Hs 0
Here α gives the elevation of the crest, and τ is the time, measured relative to the time of the
crest. The water kinematics below the surface follow the usual linear wave relationships. In
the context of an irregular sea, Bladed uses the NewWave as a tool to achieve a constraint.
A linear combination of a standard irregular sea and a NewWave (centred on the constraint
time) is used to achieve a crest of the desired height at the desired time. The time derivative
of the water surface at the constraint time needs to be set to zero (so that it forms a crest);
this is done by using a slope NewWave:
( ) ( )
16
H s 0
( ) 2
S ( f ) sin(2f )2df
The slope NewWave has zero amplitude but slope α (at the origin), so it can be combined
linearly with the irregular sea to set the time derivative of the surface elevation to zero at the
constraint time.
When the Linear NewWave option is used, Bladed carries out the above procedure and
iterates the height of the crest, so that the difference in elevation between the crest and the
lowest point up to one period either preceding or following the crest meets the user’s
specification.
When the non-linear constrained wave option is used, the stream function solution (see
Section 8.7) is used for one period around the constrained crest, beyond which the linear
irregular model is used. NewWaves are used to constrain the linear sea in such a way as to
allow the two separate models to blend smoothly together. This option combines the
irregularity of the linear model with the full non-linearity achieved by the stream function
model. A full description of this approach can be found in reference [8.11].
Bladed allows current velocities to be calculated based on three current profiles, either
separately or in combination:
where cw , cs and cn are the directions from which the three current components arrive at
the tower. Components of the calculated current velocities are then combined linearly:
u c u cw u cs u cn
z d
ucs z u s 0 z 0
d
for 0 z d , where d is the water depth and u s 0 z 0 is the velocity at the sea surface.
The standard power law exponent is normally taken as 1/7, but can be changed by the
user.
If desired, a suitable current velocity at the location of the breaking wave may be calculated
as:
ucn 2s gH B
where g is the acceleration due to gravity, s is the beach slope and H B is the breaking wave
height given by:
b
HB
1 a
d B gTB2
where:
a 441 exp 19s
b 1.6 1 exp 19s
d B is the water depth at the location of the breaking wave and TB is the period of this wave.
For very small beach slopes H B may be estimated using the formula H B 0.8d B .
8.10 Total Velocities and Accelerations
The wave particle velocity and acceleration vectors at a particular location x, y, z in the
wave field at time t, obtained from the white noise filtering procedure, are denoted by u w and
u w . The total current velocity vector at the same location is u c and the velocity and
acceleration of the tower structure itself are u s and u s .
The total velocity u t and acceleration u t of the fluid relative to the structure at this location
and time are therefore:
ut u w uc u s
u t u w u s
Having evaluated the total particle kinematics relative to the tower, the resulting forces are
calculated as the sum of two components:
Dand inertia forcescalculated using the relative motion form of Morison’s equation,
Longitudinal pressure forces.
These forces are then used to calculate the tower modal forces as described in Section 3.2.2.
D2 D2
F Cm 1
1
Lu t Lu w Cd DLut ut
4 4 2
where F is the normal force on a segment of cylinder of length L and diameter D, is the
water density, C m is the inertia coefficient and C d is the drag coefficient.
The added mass acting in the longitudinal direction is very small and so the longitudinal
forces are estimated using the hydrodynamic pressure in the ambient wave field acting over
the change in cross-sectional area of the tower between the top and bottom faces of each
sub-element. For a structure with diameter D a at the top of a sub-element and diameter Db
at the bottom, the longitudinal force acting on this portion of the structure is:
F
Da2 Db2
p
4
No pressure force is included where the end of the tubular member passes through the free
surface or terminates at the sea-bed.
9. TOWER SHADOW AND WAKE EFFECTS
Tower shadow defines the distortion of the steady-state mean flow field due to the presence
of the turbine tower or support structure. Three different models are available: a potential flow
model for upwind rotors, an empirical tower wake model for down wind rotors, and a
combined model which is useful if the rotor yaws in and out of the down wind shadow area.
V ( x, y ) V 0
y2 x2 2 xy D 2
2 V0
,
y2 x2 2 y2 x2 2
provided the point is far from the tower top. For the variation close to the tower top see
Section9.1.4.
V ( x, y ) AV0
where
y
A 1 cos2
WDT
for azimuth angles within +60 of bottom dead centre. For other azimuth angles, the same
correction is applied as for the potential flow model, Section 9.1.1. Here is the maximum
velocity deficit at the centre of the wake as a fraction of the local wind speed, and W is the
width of the tower shadow as a proportion of the local tower diameter DT. These quantities
are defined for a given downwind distance, also expressed as a proportion of DT . For other
distances, W increases, and decreases, with the square root of the distance.
The effect close to the end of the member varies smoothly for both potential flow and Powles
models. For the potential flow model, the member is assumed to be a line source with ends at
positions p1 and p2. Hence, in a coordinate system with the z axis pointing along the tower (or
member) the resulting velocity is given by:
z p2 y 2 x 2
z p2 x 2
y 2 x 2 2 x 2 y 2 ( z p )2 2
1 3
y 2 x 2 x 2 y 2 ( z p2 ) 2 2
2
z p1 y 2 x 2
z p1 x 2
,
1 3
y 2 x 2 2 x 2 y 2 ( z p1 ) 2 2 y 2 x 2 x 2 y 2 ( z p1 ) 2 2
2 xy z p2 xy z p2 wh
2
V ( x, y , z ) V 0 D V
2
2
1 3 0 xy
2
y x 2
x 2
y 2
( z p 2 ) 2 2
y 2
x 2
x 2
y 2
( z p 2 ) 2 2
2 xy z p1 xy z p1
,
y 2 x2 2 x2 y 2 ( z p )2 2
1 3
y 2 x 2 x 2 y 2 ( z p1 ) 2 2
1
0
ere V0xy is the magnitude of the velocity in the x-y plane.
For the Powles model, the end effects are taken account of by simplifying it to a diffusion
equation. For line source in the diffusion equation, the fall off at the ends of the source is
proportional to the error function:
If the turbine rotor being modelled is assumed to be wholly or partially immersed in the wake
of another turbine operating further up wind, a model is provided to define the modification to
the steady-state mean wind profile caused by that wake.
A Gaussian profile is used to describe the wake of the upstream turbine. The local velocity at
a distance r from the wake centreline (which may be offset from the hub position) is given by:
2
r
V V0 1 e 2W 2
where V0 is the undisturbed wind speed, is the fractional centre line velocity deficit, and W is
the width of the wake (the distance from the wake centre line at which the deficit is reduced to
exp(-0.5) times the centre line value).
Two options are provided for defining the velocity deficit and the wake width W . They can
be defined directly, or they can be calculated by Bladed by specifying the characteristics of
the upstream turbine. In the latter case, an eddy viscosity model of the wake is used,
developed by Ainslie [7.8,7.9] and described in the next section.
The eddy viscosity wake model is a calculation of the velocity deficit field using a finite-
difference solution of the thin shear layer equation of the Navier Stokes equations in axis-
symmetric co-ordinates. The eddy viscosity model automatically observes the conservation of
mass and momentum in the wake. An eddy viscosity, averaged across each downstream
wake section, is used to relate the shear stress term in the thin shear equation to gradients of
velocity deficit. The mean field can be obtained by a linear superposition of the wake deficit
field and the incident wind flow. An illustration of the wake profile used in the eddy viscosity
model is shown in
Figure 9.1.
Incident
flow profile Wake profile
The Navier Stokes equations with Reynolds stresses and the viscous terms dropped gives
[7.10]:
U U 1 (r uv)
U V
x r r r
The turbulent viscosity concept is used to describe the shear stresses with an eddy viscosity
defined by [7.11]:
(x) Lm (x).Um (x)
U
and uv
r
Lm and Um are suitable length and velocity scales of the turbulence as a function of the
downstream distance x but independent of r. The length scale is taken as proportional to the
wake width Bw and the velocity scale is proportional to the difference UI – Uc across the shear
layer.
Thus the shear stress uv is expressed in terms of the eddy viscosity. The governing
differential equation to be solved becomes:
U U ( rU / r )
U V
x r r r
Because of the effect of ambient turbulence, the eddy viscosity in the wake can not be wholly
described by the shear contribution alone. Hence an ambient turbulence term is included,
and the overall eddy viscosity is given by [7.12]:
FK1 Bw (U i U c ) amb
where the filter function F is a factor applied for near wake conditions. This filter can be
introduced to allow for the build up of turbulence on wake mixing. The dimensionless
constant K1 is a constant value over the whole flow field and a value of 0.015 is used.
The ambient eddy viscosity term is calculated by the following equation proposed by
Ainslie [7.12]:
Kk is the von Karman constant with a value of 0.4. Due to comparisons between the model
and measurements reported by Taylor in [7.13] the filter function F is fixed at unity.
The centre line velocity deficit Dmi can be calculated at the start of the wake model (two
diameters downstream) using the following empirical equation proposed by Ainslie [7.12]:
Assuming a Gaussian wind speed profile and momentum conservation an expression for the
relationship between the deficit Dm and the width parameter Bw is obtained as
3.56Ct
Bw
8Dm (1 0.5Dm )
Using the above equations, the average eddy viscosity at a distance 2D downstream of the
turbine can be calculated. The equations can then be solved for the centre-line deficit and
width parameter further downstream.
Assuming to the Gaussian profile, the velocity deficit a distance r from the wake centreline is
given by:
r
2
Dm,r
exp 3.56
Bw
Therefore the wake width W used by Bladed is given by:
0 .5
W Bw
3.56
If the eddy viscosity wake model is used, it is also possible to calculate the additional
turbulence caused by the wake. The added turbulence is calculated using an empirical
characterisation developed by Quarton and Ainslie [7.14]. This characterisation enables the
added turbulence in the wake to be defined as a function of ambient turbulence I amb, the
turbine thrust coefficient Ct, the distance x downstream from the rotor plane and the length of
the near wake, xn. The characterisation was subsequently amended slightly by Hassan [7.15]
to improve the prediction, resulting in the following expression:
in which all turbulence intensities are expressed as percentages. Using the value of added
turbulence and the incident ambient turbulence the turbulence intensity Itot at any turbine
position in the wake can be calculated as
n r0
xn
dr
dx
where
m 1
r0 R
2
1
m
1 Ct
n
0.214 0.144m 1 0.134 0.124m
1 0.214 0.144m 0.134 0.124m
2 2 2
dr dr dr dr
dx dx dx m dx
dr
2.5 I 0 0.005 is the growth rate contribution due to ambient turbulence,
dx
The wind simulation method adopted in Bladed is based on that described by Veers [7.3]. The
rotor plane is covered by a rectangular grid of points, and a separate time history of wind
speed is generated for each of these points in such a way that each time history has the
correct single-point turbulence spectral characteristics, and each pair of time histories has the
correct cross-spectral or coherence characteristics.
Another approach due to Mann [7.18, 7.19] is also included, being now referenced in the third
edition of the IEC standard [7.20].
Calculations using such a turbulent wind field will take into account the crucially important
'eddy slicing' transfer of rotor load from low frequencies to those associated with the rotational
speed and its harmonics. This 'eddy slicing', associated with the rotating blades slicing
through the turbulent structure of the wind, is a significant source of fatigue loading.
The wind speed time histories may, in principle, be generated from any user-specified auto-
spectral density and spatial cross-correlation characteristics. A choice of two different models
of atmospheric turbulence has been provided. These are the von Karman and the Kaimal
models. Both models are generally accepted as good representations of real atmospheric
turbulence, although they use slightly different forms for the autospectral and cross-spectral
density functions. The von Karman model can be used either to generate just the longitudinal
component of turbulence, or to generate all three components if required. Two versions of the
von Karman model are available: the basic model, given in [7.4] and described in Section
7.5.1, and the improved model, described in Section 7.5.2, which is based on more up-to-date
information [7.5, 7.6].
It should be remembered, of course, that all these models tend to be based largely on
observations for flat land sites.
where Suu is the auto-spectrum of wind speed variation, n is the frequency of variation, u is
~ is a non-dimensional frequency
the standard deviation of wind speed variation and n u
parameter given by:
n x Lu
n~u
U
x
Here Lu is the length scale of longitudinal turbulence and U is the mean wind speed.
If the three-component model is selected, the corresponding spectra for the lateral (v) and
vertical (w) components are:
where
n xLi
n~i
U
and i is either v or w.
Associated with the von Karman spectral equations is an analytical expression for the
cross-correlation of wind speed fluctuations at locations separated in both space and time,
derived assuming Taylor's frozen turbulence hypothesis. Accordingly for the longitudinal
component at points separated by a distance r perpendicular to the wind direction, the
coherence Cu (r,n), defined as the magnitude of the cross-spectrum divided by the auto-
spectrum, is:
j
Here Aj(x) = x Kj(x) where K is a fractional order modified Bessel function, and
nL (r , n)
2
r
u 0.747 1 70.8 u
Lu (r , n) U
where y and z are the lateral and vertical components of the separation r, and Lu and Lu
y z
are the lateral and vertical length scales for the longitudinal component of turbulence.
For the lateral and vertical components, the corresponding equations are:
Ci (r , n)
0.597
2.869 i2 1
4.781 i2 A5 / 6 (i ) A11/ 6 (i )
where
nL (r , n)
2
r
i 0.747 1 70.8 i
Li (r , n) U
and
i Li (r , n)
i
r
for i = v or w.
( yLv y / 2) 2 ( zLv z ) 2
Lv (r , n) 2MIN (1.0,0.05n 2 / 3 )
y 2 z 2
and
1/ 2 ( yLwy) 2 ( z Lw z / 2) 2
Lw (r , n) 2MIN (1.0,0.2n )
y 2 z 2
The three turbulence components are assumed to be independent of one another. This is a
reasonable assumption, although in practice Reynolds stresses may result in a small
correlation between the longitudinal and vertical components near to the ground.
where Suu is the auto-spectrum of wind speed variation, n is the frequency of variation, u is
~ is a non-dimensional frequency
the standard deviation of wind speed variation and n u
parameter given by:
n x Lu
n~u
U
x
Here Lu is the length scale of longitudinal turbulence and U is the mean wind speed.
If the three-component model is selected, the corresponding spectra for the lateral (v) and
vertical (w) components are:
where
n xLi
n~i
U
and i is either v or w.
The five additional parameters a, 1, 2, F1 and F2 are defined as follows:
F1 1 0.455exp 0.76n~u / a
0.8
F2 1 2.88 exp 0.218n~i / a
0.9
2 1 1
1 2.357a 0.761
a 0.535 2.76(0.138 A) 0.68
where
A 0.115[1 0.315(1 z / h) 6 ]2 / 3
Here z is the height above ground, and h is the boundary layer height obtained from:
h u * /(6 f )
f 2 sin (the Coriolis parameter: is the angular speed of rotation of
the earth, and is the latitude)
u 0.4U 34.5 f .z / ln( z / z0 )
*
The turbulence intensities of the three components of turbulence are also defined for the same
choice of z, z0, U and , as follows:
1 6 f .z / u *
p 16
7.5 (0.538 0.09 ln( z / z0 )) p u *
u
1 0.156 lnu * / f .z0
although these may be changed by the user for any particular simulation.
x
Lu
A1.5 u / u * z
3
2.5K 1z .5 (1 z / h) 2 (1 5.75z / h)
y
Lu 0.5x Lu 1 0.46 exp 35( z / h)1.7
z
Lu 0.5x Lu 1 0.68 exp 35( z / h)1.7
x
Lv 0.5x Lu ( v / u )3
x
Lw 0.5x Lu ( w / u )3
y
Lv 2 y Lu ( v / u )3
z
Lv zLu ( v / u )3
y
Lw yLu ( w / u )3
z
Lw 2 z Lu ( w / u )3
where
K z 0.19 (0.19 K 0 ) exp Bz / h
N
K0 0.39 / R 0.11
B 24R 0.155
N 1.24R0.008
u*
R
f .z0
Associated with the von Karman spectral equations is an analytical expression for the
cross-correlation of wind speed fluctuations at locations separated in both space and time,
derived assuming Taylor's frozen turbulence hypothesis [7.6]. Accordingly for the longitudinal
component at points separated by a distance r perpendicular to the wind direction, the
coherence Cu (r,n), defined as the magnitude of the cross-spectrum divided by the auto-
spectrum, is:
j
Here Aj(x) = x Kj(x) where K is a fractional order modified Bessel function, and
2
0.747r 2nr
2
i c for i = u
2 Li U
( yLu y) 2 ( z Lu z ) 2
Lu (r , n)
y 2 z 2
while
1.6(r / 2 Lu ) 0.13
c max(1.0, )
0b
with
b 0.35r / 2 Lu
0.2
and
2
0.747r 2nr
2
0
2 Lu U
y and z are the lateral and vertical components of the separation r, and yLu and zLu are the
lateral and vertical length scales for the longitudinal component of turbulence.
For the lateral and vertical components, the corresponding equations are:
Ci (r , n)
0.597
2.869 i 1
2
4.781 i2 A5 / 6 (i ) A11/ 6 (i ) for i = v,w
i 2 Li (r , n)
i
r
( yLv y / 2) 2 ( z Lv z ) 2
Lv (r , n)
y 2 z 2
and
( yLw y) 2 ( z Lw z / 2) 2
Lw (r , n)
y 2 z 2
The three turbulence components are assumed to be independent of one another. This is a
reasonable assumption, although in practice Reynolds stresses may result in a small
correlation between the longitudinal and vertical components near to the ground.
where Suu is the auto-spectrum of wind speed variation, n is the frequency of variation, u is
~ is a non-dimensional frequency
the standard deviation of wind speed variation and n u
parameter given by:
n L1
n~u
U
x x
Here L1 = 2.329 Lu where Lu is the length scale of longitudinal turbulence, and U is the mean
wind speed as before.
A simpler coherence model is used in conjunction with the Kaimal model. With the same
notation as in Section 7.5.1, the coherence is given by
2
n 0.12
2
C (r , n) exp 8.8r
U L(r , n)
A three-component Kaimal model has been introduced for compatibility with the IEC standard
1400-1 edition 2 [7.1]. The scale parameter 1 defines the characteristics of the turbulence,
through the following relationships:
x
n Li
n~i i = u, v, w
U
2
n 0.12
2
C( r,n) exp Hr
U Lc
where the coherence decay constant H = 8.8 and the coherence scale factor Lc = 3.51. The
standard does not define the coherence for the other two components, so the following
expression is used:
n
C( r,n) exp Hr
U
A more general formulation for the Kaimal model has also been introduced in which the
parameters Lu, Lv, Lw and Lc can be specified separately instead of specifying 1, and the
x x x
parameter H and can also be specified. This can be used for compatibility with the third
edition of the IEC standard [7.20], which gives H = 12 and Lc = 8.11.
This is in many ways quite an elegant approach, but there are some practical limitations to be
aware of. The summation requires a three-dimensional fast Fourier transform (FFT) to
achieve reasonable computation time. The number of points in the longitudinal, lateral and
vertical directions must be a power of two for efficient FFT computation. In the longitudinal
direction, the number of points is determined by the length of time history required and the
maximum frequency requested, and is therefore typically at least 1024. The maximum
wavelength used is the length of the file (mean wind speed multiplied by duration), and the
minimum wavelength is twice the longitudinal spacing of points (which is the mean wind speed
divided by the requested frequency). In the lateral and vertical directions, a much smaller
number of points must be used, perhaps as low as 32, depending on available computer
memory. The maximum wavelength must be significantly greater than the rotor diameter,
since the solution is spatially periodic, with period equal to the maximum wavelength in each
direction. The number of FFT points then determines the minimum wavelength in these
directions. With a realistic number of points, the resulting turbulence spectra are deficient at
the high frequency end. Mann [7.19] suggests that this is realistic, because it represents
averaging of the turbulence over finite volumes of space which is appropriate for practical
engineering applications. Bladed will report the loss of turbulence intensity due to this effect,
so if a certain turbulence intensity is requested for a simulation, the actual turbulence intensity
will be slightly lower due to a loss of high frequency variations.
Note also that the time histories at each grid point may have individual spectra and variances
which can differ from one point to another. This may well be realistic of course, but it means
that if the spectrum or variance is computed at any single point, for example at the hub
position, the result may again be a little different from the expected value.
The length of the wind field, Lfield, must be sufficient for the simulation to be carried out. For
a simulation of T seconds at a mean wind speed of U m/s, Lfield must be at least UT + D
metres where D is the turbine diameter (the extra diameter is needed in case the turbine is
yawed with respect to the mean wind direction). Alternatively there is an option to allow the
turbulent flowfield to wrap around at the end, starting again from the beginning. This allows
an arbitrarily long simulation.
The width and height of the wind field must evidently be sufficient to envelope the whole
rotor, i.e. at least equal to the rotor diameter. The flow field can also be made tall enough
to encompass the entire rotor and tower.
The number of grid points in the lateral and vertical directions should be chosen to achieve
adequate sampling of the wind speed variations in the rotor plane. A grid point spacing of
6 – 7 m is likely to be adequate for this. The time taken to generate the turbulence file
increases with the fourth power of the number of points, so it is important not to use more
points than necessary. In the along- wind direction, a resolution of 10 Hz is likely to be
sufficient. For certification calculations, definitive guidelines should be sought from the
certifying body.
If a simulation uses only a part of a turbulent time history, the mean wind speed and
turbulence intensity for that part of the time history may not be the same as for the whole
time history, and therefore may not match the mean wind speed and turbulence intensity
which was specified for the simulation since this assumes that the whole time history will be
used. Note also that the number of points along the wind direction must be a power of two
for efficient calculation, since Fast Fourier Transform techniques can then be used. If it is
not a power of two, then the along-wind spacing of points will automatically be decreased.
Different time histories with the same turbulence characteristics can be generated by
changing the random number seed.
A sinusoidal half- or full-wave wind direction transient as described in Section 7.2.3 may be
superimposed on the turbulent wind field. This is intended for use with turbulent wind fields
when only the longitudinal component has been generated, to ensure that some yaw error
occurs during the simulation. Using all three components of turbulence should give a more
realistic variation of yaw error.
11. MODELLING EARTHQUAKES
For turbines situated in seismically active areas, it is important to assess the loading due to
earthquakes. Bladed offers the user the option of performing calculations using a recorded
time history, or of generating a synthetic accelerogram that satisfies a given Response
Spectrum.
For stiff foundations, Bladed assumes the soil to be rigid and so not to interact with the
dynamic response of the structure.
The equation of motion for a single modal degree of freedom, assuming no coupling with
other degrees of freedom, is as follows:The equation of motion for a single modal degree of
freedom, assuming no coupling with other degrees of freedom, is as follows:
where:
and:
H
li m(h)i (h)dh is the earthquake participation factor.
0
With flexible foundations, the seismic acceleration defines the ground motion, which affects
the forces and moments developed in the foundation.
The equation of motion for a single degree of freedom system subject to the mug force is
as follows:
where is the circular frequency of the structure in rad/s, and is the damping ratio. This
equation is solved for many single degree of freedom structures having different periods,
each subjected to the same ground acceleration, and each with the same damping ratio. The
plot of maximum response against structural frequency is called the response spectrum [9.1].
Design response spectra are normally presented as a plot of the maximum acceleration
against the period (the inverse of the natural frequency of the structure).
11.3 Generation of response spectrum compatible accelerogram
The algorithm that generates the response spectrum compatible accelerogram is an iterative
procedure. The ability to specify a random number seed ensures repeatability of results. The
process is outlined as follows:
The first step is the generation of a set of random numbers, which are then scaled as a
function of their position in time to create the characteristic earthquake shape. The user is
able to choose between an overall shape typical of stiff or soft soil properties. If soft soil
properties are selected, the stationary part and the decay constant can also be specified. The
time history is then passed through a low and high pass band filter to remove unwanted
frequencies.
After this initial set up has taken place, the process is an iterative one [9.2], with the following
steps:
Compute response spectrum.
Compare against Target Response Spectrum at a number of frequency points.
Check to see if convergence criteria have been met.
Scale in the frequency domain.
At each iteration, the response spectrum is computed, and compared against the Target
Response Spectrum.
The user is able to specify upper and lower limits of the acceptable deviation of any of these
computed points from the Design Response Spectrum. If all the points are within this range,
the convergence criteria have been met, and the process is complete. If any points are
outside this range, the Fourier transform of the time history is scaled with the following
relationship:
S aD
Vg ( new ) Vg ( old )
Sa
D
where Vg is the Fourier transform of the acceleration time history, S a is the target spectrum
Bladed includes an integrated post-processing facility which allows the results of wind turbine
calculations to be processed further in various ways. The theory behind these post-
processing calculations is described in this section.
Minimum MIN(x)
Maximum MAX(x)
Mean x
Standard deviation ( x x )2
Skewness ( x x )3 / 3
Kurtosis ( x x )4 / 4
Wind turbine loads consist of both periodic and random or stochastic components. The
periodic components of loads result from effects which vary as a function of rotor azimuth,
such as gravitational loads, tower shadow, yaw misalignment, wind shear etc. The stochastic
components result from the random nature of turbulence. In understanding the loads on a
wind turbine it is often useful to separate out the periodic and stochastic parts of a load time
history, and future to analyse the periodic part in terms of the harmonics of the fundamental
rotational frequency.
The periodic part of a signal is obtained by binning the signal against rotor azimuth. The
number of azimuth bins may be specified by the user, otherwise it is calculated from the first
two azimuth values in the time history. These are used to define the azimuth bin width, which
is then adjusted to an exact sub-multiple of a revolution.
The number of azimuth bins must be compatible with the sampling interval of the time history.
If too many bins are used, it is possible for some of them to be empty, in which case the
calculation will not proceed.
Having obtained the periodic component of the signal, the Fourier harmonics are obtained by
means of a discrete Fourier transform, after first increasing the number of bins by two to four
times using linear interpolation.
The stochastic component of the signal is obtained for each time point by subtracting the
periodic component calculated from the azimuth at that time point. Linear interpolation is
used between azimuth bins.
An alternative approach, which avoids the problem of the rather arbitrary nature of these
discrete gusts, is based on probabilistic techniques, with the stochastic nature of the loads
due to wind turbulence represented by means of a probability distribution. Although this
approach has been used for many years for the evaluation of extreme loads on buildings and
similar structures, its application to wind turbine loads is relatively rare. The analysis involved
in applying it to an operational wind turbine is rather more complicated since the probability
distribution of the combined stochastic and deterministic load components must be
considered.
where z and x represent the periodic and stochastic parts of the load respectively (see
Section 12.2). It is generally a good approximation to assume that the stochastic part of the
load is Gaussian, so its probability distribution is:
1 / 2 x2
ex
2
p( x)
x 2
where x is the standard deviation of x. For such a signal, Rice [10.1] has derived the
probability distribution of signal peaks as:
1 2 / 2 (1 2 ) 2 / 2
2
pˆ ( x) e e 1 erf
x 2 2 x 22 2
where
x / x
0 / m
M2
0 (the zero up-crossing or apparent frequency)
M0
M4
m (the frequency of peaks)
M2
M i f i H ( f )df (the ith spectral moment)
0
f frequency (Hz)
H ( f ) power spectral density (see next section for calculation details), and
erf () error function.
Knowing the probability distribution of peaks for such a process, the probability distribution of
extremes can then be deduced. For the extreme of the signal in a given period to be x, one
peak must have this value and all other peaks in the period must have a lesser value. The
probability distribution can be written
where
Q( ) pˆ ( )d , and
pˆˆ ( ) e
where
0T e
2
/2
and
T time period.
The mean of this distribution is
ext
where
2 ln( 0T ) and
0.5772 (Euler’s constant).
As the term 0T increases, the distribution of extremes has a larger mean and becomes very
narrow.
For an operational turbine whose loads are a combination of stochastic and periodic
components, Madsen et al [10.3] proposed an approach based on Davenport’s model of the
stochastic signal, with the assumption that the extremes in the total signal occurred at minima
and maxima of the periodic component. This allows the periodic time history to be idealised
as a square waveform as follows:
z
t1
zmax
zmean t2
Time
t3
zmin
T0
The resulting expressions for the mean and standard deviation of the extreme distributions
are:
For extreme maxima:
ye max zmax x 1
1
e max x
6 1
where
1 2 ln(1 0T )
t1 z2
1
T0 ( zmax zmean )( zmax zmin )
ye min zmin x 3
3
e min x
63
where
3 2 ln( 3 0T )
t3 z2
3
T0 ( zmean zmin )( zmax zmin )
Here z is the standard deviation of the periodic component z. The time period T should be
taken as the total time for which the condition being modelled will be experienced during the
lifetime.
Bladed allows the calculation of auto-spectral density, cross-spectral magnitude and phase,
transfer functions and coherence functions.
All calculations involving spectral analysis use a Fast Fourier Transform technique with
ensemble averaging. To perform the spectral analysis, the signal is divided into a number of
segments of equal length, each of which contains a number of points which must be a power
of 2. The segments need not be distinct, but may overlap. Each segment is then shaped by
multiplying by a ‘window’ function which tapers the segment towards zero at each end. This
improves the spectrum particularly at high frequencies. A choice of windowing functions is
available. Optionally, each segment may have a linear trend removed before windowing,
which can improve the spectral estimation at low frequencies. The final spectrum is obtained
by averaging together the resulting spectra from each segment, and scaled to readjust the
variance to account for the effect of the window function.
Number of points: the number of datapoints per segment. This must be a power of 2: if it is
not, it is adjusted by the program. The maximum allowed is 4096. The larger the number of
points, the better will be the frequency resolution, which may be important especially at low
frequencies. However, choosing fewer points may result in a smoother spectrum because
there will be more segments to average together. If in doubt, 512 is a good starting point.
Percentage overlap: the overlap between the segments. This must be less than 100%.
50% is often satisfactory, although 0% may be more appropriate if a rectangular window is
used.
where f is the fractional position along the segment (0 at the start, 1 at the end). One of the
last three windows (which are all quite similar) is recommended.
Trend removal: If checked, a linear trend is calculated for each segment and removed from it
before windowing, this is usually desirable. If left unchecked, the mean is calculated instead
for each segment and removed from it before windowing.
These calculations work by binning values. The range and size of the bins to be used are
calculated by the program, unless they have been supplied by the user.
The probability density analysis simply bins the signal values. From the probability density
function it also calculates the cumulative probability distribution. Also a Gaussian distribution
is calculated for comparison, which has the same mean and standard deviation as the signal.
There is an option to remove the mean of the signal: this merely moves the mean of the
calculated distribution to zero.
The peak analysis bins only those signal values which are turning points of the signal. Peaks
and troughs are binned separately, so that the probability distribution of each can be output.
For the level crossing analysis, the number of up-crossings and down-crossings are counted
at each of the bin mid-points. The number of crossings per unit time in each direction is
output for each bin mid-point.
Bladed offers the possibility of rainflow cycle counting of a stress time history and of
subsequent fatigue analysis based on the cycle count data. A suitable stress time history can
be generated from one or more load time histories by use of the channel combination and
factoring facility provided by the code.
The cycle counting procedure is as defined in Recommended Practices for Wind Turbine
Testing and Evaluation, 3. Fatigue Loads, IEA, 2nd edition 1990. It involves the following
steps:
The stress history is searched to determine the successive peaks and troughs by
identification of turning points.
The successive peaks and troughs are re-ordered so that the sequence begins with the
highest peak value of the stress history.
The sequence of peaks and troughs is now scanned to determine the rainflow cycles. A
rainflow cycle is only recorded when the range exceeds a user specified minimum range.
The purpose of this user-specified minimum range is to filter out very small cycles where
this is desired.
The count of rainflow cycles is binned according to the cycle mean and range values. The
distribution of bins is defined by the user who is required to specify minimum and
maximum values of stress and the number of bins to be used.
The output from the rainflow cycle counting analysis consists of the two-dimensional
distribution of the number of cycles binned on the means and ranges of the cycles.
This calculation can also be extended to generate damage equivalent loads. The user
specifies one or more inverse S-N slopes m (see next section) and a frequency f (typically 1P
for fixed speed machines), and an equivalent load is calculated as the peak-to-peak
amplitude (i.e. the range) of a sinusoidal load of constant frequency f which would produce
the same fatigue damage as the original signal. The equivalent load is therefore given by:
1
ni S i m m
i
Tf
where ni is the number of cycles in load range Si and T is the duration of the original time
history.
The basis of the fatigue analysis provided in Bladed is that fatigue failure is predicted to occur
according to the Palmgren-Miner [10.4] linear cumulative damage law. Failure will occur when
the “accumulated fatigue damage number” is equal to 1.0 as follows::
ni
Ni
1.0
i
th
where ni is the number of rainflow cycles of the i stress range and Ni is the corresponding
number of cycles to failure. The summation is defined as the accumulated damage.
For rainflow cycles of stress range Si, the number of cycles to failure Ni is given by the S-N
curve for the material. The user of Bladed must supply the S-N curve in one of two ways. The
first possibility is that the S-N curve is provided as a log-log relationship of the form:
1 1
log S log k log N
m m
so that:
N kS m
The user must specify the value of m, the inverse slope of the log S against log N
relationship. The user must also specify the intercept of the log-log relationship, c. The
parameter k above is related to the intercept c by:
k cm
The second option is for the user to specify the S-N curve as an arbitrary function through the
use of a look-up table.
For a material where the mean stress has an influence on the fatigue damage accumulated,
Bladed offers the option of converting each cycle range to the equivalent range assuming a
zero mean stress value. (A cycle with a zero mean value has a R-ratio of -1, where R is the
ratio of minimum to maximum stress.) This conversion is performed by means of a Goodman
diagram and the user is required to provide the ultimate tensile strength (UTS) of the material.
Following the conversion, the fatigue analysis proceeds using the Palmgren-Miner law and
user specified S-N curve as described above.
The output from the fatigue analysis consists of the accumulated damage due to the stress
history as well as the two-dimensional distribution of the proportion of the accumulated
damage binned on the means and ranges of the rainflow cycles.
The annual energy yield is calculated by integrating the power curve for the turbine together
with a Weibull or user-defined distribution of hourly mean wind speeds. The power curve is
defined at a number of discrete wind speeds, and a linear variation between these points is
assumed.
where F is the cumulative distribution of wind speed V. Thus the probability density f(V) is
given by
k
V
V k 1 cV
f (V ) k e
(cV ) k
Here k is the Weibull shape factor, and c is the scale factor. For a true Weibull distribution,
these two parameters are related by the gamma function:
1
c 1 / 1
k
cutout
E Y P(V ) f (V )dV
cutin
where
Frequently a steady state power curve is used, combined with a distribution of hourly mean
wind speeds. For a more accurate calculation, it is desirable to use a dynamically calculated
power curve given as the average power from a series of simulations based on a model of the
turbulent wind field. It is common practice to use 10-minute simulations to capture the effects
of turbine dynamics and wind turbulence. Strictly speaking, the appropriate distribution to use
in this case would be one representing the distribution of 10-minute mean wind speeds in a
year. For a Weibull distribution, this will typically have a slightly smaller shape factor than that
for hourly means.
The ultimate loads calculation, which is often required for certification calculations, is simple in
concept: the results of a load case simulation are analysed to find the times at which each of
a number of specified loads reaches its maximum and minimum values. The simultaneous
values of all the loads at each of those instants is reported.
A further calculation named ‘ultimate load cases’ further analyses the results of a number of
ultimate loads calculations for different groups of load cases, and generates a histogram
showing the load cases in which the maximum and minimum values of each load occurred
within each group.
12.9 Flicker
The Flicker calculation generates short-term flicker severity values (Pst), either from a voltage
time history, or from time histories of active and reactive power. Such time histories are
available from simulations with the full electrical model of the fixed speed induction generator,
and also with the variable speed generator model.
The flicker severity is a measure of the annoyance created by voltage variations through
perception of the resulting flicker of incandescent lights. The calculation of flicker from a
voltage time history is defined in [10.5]. An algorithm conforming to this standard is
incorporated into the Bladed post-processor. It can also calculate flicker from a time history
of active and reactive power. In this case a voltage time history is calculated first, and this
can be calculated for any given network impedance to which the turbine might be connected.
In fact the flicker for several different network impedances can be calculated in a single
calculation. The network impedances are entered as a set of short circuit power levels and
network angles, the network angle being arctan(X/R), where X and R are the network
reactance and resistance respectively. The voltage is calculated as the solution of the
following equation:
4 2 2 2 2
U + U (2{QX - PR} - U0 ) + (QX - PR) + (PX + QR) = 0
where U0 is the voltage at the infinite busbar, and P and Q are the active and reactive power
respectively.
The Extreme load extrapolation calculation uses statistical methods to predict the extreme
load for a defined return period by extrapolating from the extreme load values obtained from a
large number of simulations performed with a range of external conditions.
Probability distributions are fitted to these extreme load results. These fitted distributions are
then weighted and summed to give an overall response function for each load. Extrapolated
extreme loads are then obtained for the defined return period from these overall response
functions.
Two probability density functions are used for fitting to the extreme data. These are the (2-
parameter) Gumbel distribution and the 3-parameter Weibull distribution. It has been found
[10.6] that these are likely to be the most applicable distributions to wind turbine loading. The
equations below are the cumulative distribution function form of these distributions:
Gumbel:
x u
F x e e s
3-parameter Weibull:
k
x u
F x 1 e s
where
u = location parameter
s = scale parameter
k = skewness parameter
The method of moments equates sample moments to parameter estimates. The method of
moments has the advantage of simplicity. The disadvantage of the method of moments is that
it tends not to be as accurate in its parameter estimates as the maximum likelihood and least
squares estimators. The moment estimates are used as starting values for the more precise
maximum likelihood and least squares estimates. They are also output as a check.
For the Gumbel distribution, the method of least squares fits a straight line in the usual way
but in Gumbel axes and for the three parameter Weibull distribution, a similar technique is
employed but also minimising the sum of the squares for the third parameter.
There is no way of deciding quantitatively which distribution or fitting method best suits the
data. This motivates the diagnostic plots which can help qualitatively. For more detailed
theory points including an example of when distribution choice is important, please refer to
the GH study into extrapolation carried out under the European Union funded project,
‘Recommendations for the Design of Offshore Wind Turbines’ (RECOFF) [10.7]. Generally the
3-parameter Weibull distribution is more flexible for fitting that the Gumbel distribution due to
the extra skewness parameter. The Gumbel distribution is always useful as a check.
Two updates have been made for version 4.2 and later, as follows:
Update 1 – A change to the short-term probability distribution when using the Peak Over
Threshold (POT) method.
Previously the probability assigned to each local extreme(maxima) in a given simulation was
simply assumed to be equal to the probability assigned to the global extreme(maxima) of the
simulation. Now the probability assigned to each local extreme(maxima) in a given simulation
is the probability distribution of the local extreme(response maxima) raised to the power of the
average number of local extremes(responses) taken from all simulations within the individual
wind bin. This results in a change to the extrapolated loads when using the ‘POT’ method.
The aim of the ‘Offshore Code Checking’ post processing tool in Bladed is to unite the main
competencies and advantages of Bladed with those of the offshore analysis packages ASAS
or SACS. On the one hand, Bladed is able to perform rigorous non-linear, time domain
simulation of the integrated WTG and support structure model. On the other hand, ASAS and
SACS are customised to perform robust and automatic fatigue and ultimate limit state (code-
checking) routines.
The Bladed-ASAS option also automatically creates a FATJACK or BEAMCHECK input file,
based on a combination of user-defined inputs (pre-defined by the user in a template file
within the Bladed installation directory) and automatic input commands from Bladed. The
latter include transfer of the following parameters from Bladed into the ASAS input file:
- Project name
- ASAS ‘structure’ name
- load case list
- associated probability of each load case
- design lifetime
A detailed explanation of the ASAS input commands is given in the FATJACK and
BEAMCHECK user manuals.
The Windfarmer Link is a software module dedicated to the calculation of loading on wind
turbines operating in specific environmental conditions. The tool unites the respective
capabilities of Windfarmer [11.1, 11.2] and Bladed software to provide an integrated method
of site-specific fatigue load assessment. This represents a major advance, placing the
calculation of site specific wind turbine loads on the same computational basis as the
evaluation of wind farm energy production which is routinely performed for project developers
and investors.
The motivation to determine site-specific fatigue loads follows from the inability of class
structures used in wind turbine certification standards to account for the complexity inherent in
an actual wind farm. Site-specific calculations are therefore required to determine whether the
loading on a turbine operating in a particular environment falls within the machine’s design
envelope. The unique specification of a given site (in terms of its turbulence levels, air
density variation, up-flow angles, wake-flow characteristics, wind probability distribution, and
other parameters) will determine the loading experienced by each turbine and should be
accounted for in any valid model.
Successful selection of turbine and turbine layout for a given site is therefore dependent on
an accurate understanding of the associated site-specific loads, emphasising the need for
effective calculation methods.
The principal aim of a site specific load calculation is to assess which turbines on site, if any,
exceed their certification/design levels. A wide range of load components are investigated.
This is achieved by running a number of time-domain simulations of turbine operation (and in
some cases turbine idling scenarios). It is standard practice to base these simulations upon
the associated fatigue load cases used for the turbine certification load calculations, but
significantly the generic external conditions of the certification standard are substituted with
the specific site environmental conditions. A salient consideration in this regard is the effect of
wake interaction. In summary, the site-specific conditions relevant to fatigue load calculations
include:
air density
wind speed and direction probability distribution (wind rose information)
ambient turbulent intensity distribution
wind shear
terrain up-flow angle
wake interaction
An initial evaluation of the site prior to the simulation stage can be used to select turbines
which are obviously the most severely loaded on site. Candidate turbines can be identified on
the basis of wake interaction (itself determined by the number of neighbouring machines, the
spacing between turbines and the predominant wind direction) and the variation in wind
shear, up-flow angle and annual mean wind speed across the site. In cases where candidate
turbines are not so obviously identified, a larger number of machines should be selected, the
worst-case scenario being an explicit analysis of the whole site.
Section 13.9 looks at the issue of turbine selection in more detail. An approximate pre-
simulation ranking method is described whereby turbines on site are ordered based on the
environmental severity of their location.
Once the study turbines have been selected, the WindFarmer Link module is used to
construct Bladed project files for creating the wind, performing time domain simulations, and
post-processing the results. A Bladed batch can then be constructed which incorporates all
these calculations.
The final stage is to compare the calculated site-specific fatigue loads (normally expressed as
a lifetime damage equivalent load) with the design constraints of the turbine’s components,
including safety factors. If site specific loads are found to exceed design loads for any given
turbines, repositioning of turbine locations and/or shut-down procedures may be considered
and the loads re-calculated for the revised specification.
Windfarmer is used to provide a detailed specification of a given site in terms of its ambient
and wake-affected turbulent intensity levels, air density, up-flow angles, wind speed and
direction distribution and wake-flow characteristics.
The Windfarmer Link module is used to manipulate the site information from Windfarmer
(under the guidance of user-defined constraints) to generate wind file specifications, multiple
time-domain simulations and post-processing calculations in Bladed that then represent the
particular operating environment of selected turbines.
Bladed is used to create the wind-files, perform the simulations and carry out post processing
in order to generate results in terms of lifetime equivalent loads and/or rainflow cycle counts.
It is well known that the fatigue loading on a wind turbine operating in a wind farm is critically
affected by the downstream wake flow of neighbouring machines. Two load-increasing
mechanisms are observed:
The nature of both mechanisms is highly dependent on the separation between incident and
upstream turbine.
Two widely used methods to account for wake effects on fatigue loading are the Frandsen
formulation [11.3] and the ‘Dynamic Loads in Wind Farms II’ (DLWF) method [11.4]. Both
introduce the concept of an artificial design turbulence to represent the complex mechanisms
of wake-affected fatigue loading.
Departing from the concept of design turbulence, the WindFarmer Link method draws on the
capability of a validated wake flow modelling tool conventionally used for wind farm energy
calculations and couples it with the customised load calculations of Bladed to provide a state-
of-the-art tool for site-specific load assessment. An important feature of this method is its
ability to model the mechanisms of increased turbulent intensity and mean velocity profile
explicitly.
1
I amb
lnh / z0
The turbulence intensity is defined here as the quotient of standard deviation and mean wind
speed at high wind speeds.
In the WindFarmer model set-up the ambient turbulence distribution should be the mean
level. However, for load calculations, it is standard practice to use the characteristic or
representative turbulence. The characteristic turbulence cannot be used directly in
WindFarmer as it would introduce non-physical turbulence values into the wake model. This
problem is negotiated by defining an array of characteristic turbulence factors in the
WindFarmer Link interface which scales on the mean ambient turbulence levels subsequent
to execution of the eddy viscosity wake model. Further details of this functionality are given in
the Bladed User Manual.
PARK model,
Modified PARK model,
Eddy viscosity model.
When using the WindFarmer Link module, the eddy viscosity model should be used.
Due to the complexity of the wake directly behind the rotor, the eddy viscosity model is
initiated from two diameters downstream. This is assumed to be the distance where pressure
gradients no longer dominate the flow. If a turbine is within this limit, the program resets the
axial distance offset to a value of two diameters. A list of the input parameters to the eddy
viscosity model in WindFarmer can be found in [11.1].
Using this wake model, WindFarmer calculates a Gaussian wake profile and outputs the
centreline velocity deficit, wake width (Bw in Section 9.2.1) and centreline horizontal offset.
The WindFarmer Link module uses this information to calculate the corresponding
parameters required by Bladed to define the Gaussian Wake model for the Upwind turbine
wake.
WindFarmer uses a wind flow model to ascertain the variation in wind speeds incident on
each turbine at their respective locations on the site. It is conventional to use the WAsP code
for the site wind flow analysis but any wind flow model that outputs the same standard file
formats as defined by WAsP may be used. The model uses Weibull A and k parameters to
represent the directional wind speed probability distribution.
1. A WAsP wind resource grid (WRG) file at the turbine hub height with extents covering
all intended turbine locations, or a WAsP RSF file at the turbine locations
2. Turbine locations as grid co-ordinates,
3. Turbine performance data, including power, thrust and rotor speed characteristics.
The flow and performance matrices output from WindFarmer are referenced to the
associated site mast wind speed. Variation in wind speed between the site mast and turbine
positions due to topographical effects are then described by the software in terms of speed-up
factors. The speed-up factors vary as a function of direction.
It is therefore a requirement of the WindFarmer Link module to re-scale the flow and
performance matrices so that the dependent variables are given in terms of the turbines’
localised wind speeds. These re-scaled matrices are subsequently used in the generation of
Bladed simulations.
The wind speed and direction distribution is important in load calculations for two reasons:
1. The wind direction information determines the amount of time spent operating in a
wake.
2. The wind speed distribution determines the operational hours per year (time between
cut-in and cut-out wind speeds) and the annual mean wind speed at the turbine.
13.5.3 Upflow angles
The localised ground slope at each turbine position on site is calculated in WindFarmer
through the following routine. A circle with radius 2D (where D is the turbine rotor diameter) is
centred on the turbine’s tower base coordinates. At each degree of the circle, the elevation of
the ground directly above or below the associated circumferential point is used in conjunction
with the elevation of the turbine base to calculate the gradient between the two points. The
ground slope is assumed to be constant between these two points. A terrain angle is thus
calculated for each of direction bin around each turbine on site. This is assumed to be equal
to an effective upflow angle.
For fatigue load assessment the WindFarmer Link module only selects simulations at certain
direction sectors. The code effectively combines information for multiple direction bins into a
bin for which a single simulation is defined. Section 13.6 describes this process in detail. With
particular regard to upflow angles, the user has two options:
1. Assign the maximum up-flow angle observed over the combined direction bins to
the simulation.
2. Calculate a weighted mean up-flow angle as follows:
iN iN
weighted mean Pi i / Pi
i 1 i 1
where Pi is the probability of the ith direction bin assigned to the simulation, i is the upflow
angle of the ith direction bin assigned to the simulation and N is the total number of direction
bins assigned to the simulation.
The WindFarmer Link module receives information from WindFarmer in the form of flow and
performance matrices, with mast wind speeds and wind directions introduced as the
independent variables. A number of these matrices, such as those defining the speed-up
factors or upflow angles on site, are one-dimensional as these site parameters do not vary
with wind speed, only wind direction. The majority of matrices output (those defining
turbulence levels, wake profile parameters, wind speed / direction probability distribution and
so on) are two-dimensional, varying as functions of both wind speed and direction. The
WindFarmer Link module can accept any resolution in the number of wind speed bins and
direction bins defined in the WindFarmer simulation. However, the user should be aware that
if a smaller number of bins is selected, a less accurate simulation will result. The wind speed
bins are recommended to go up to 70m/s to ensure that the model is valid over a conceivable
range of speed up factors. Simulations run with less than 72 direction steps should be
avoided.
Wake
offset
Bw/2
Wind direction
The user can opt to have either one or two simulations in each wake. If one simulation is
chosen, the wind direction will be chosen so that the wake offset is zero; if two simulations are
chosen, they will be done with wind directions giving wake offsets of +R (one turbine radius).
In real wind farms, the situation will be complicated by the occurrence of overlapping wakes
from different neighbouring turbines. This will be characterised by a range of wind directions
within which there are two or more distinct wakes affecting the turbine without any ambient
direction bins separating them. Then for any particular wind direction Windfarmer will select
the wake characteristics based on whichever wake has the greater velocity deficit at the
turbine position. If the user opts to have one simulation per wake, these simulations will be
defined for each wind direction such that turbine is on the wake centreline of each wake; with
two simulations per wake, wind directions giving wake offsets of +R for each wake are
chosen. Then if any of these simulations is for a wake which is not the strongest wake for
that wind direction, that simulation is discarded.
If in a given sector no time is spent in ambient conditions (that is, the sector is completely
dominated by wake-affected operation) then no ambient simulation is defined. Hence, for
example, a user may specify 5 ambient simulations per wind speed bin and yet only
observe 3 simulations allocated by the code; the other 2 sectors are completely
immersed in wake-affected flow.
Any combination may be implemented. The most simplistic will be assigning all sectors
to Simulation 1. The most detailed will have 12 separate simulations; one for each
sector.
After determining the position of wake-affected and ambient simulations, the code proceeds
to write the associated Bladed project files.
A template file for the simulations is provided by the user through the interface. This file
should contain all the normal information required to successfully run a turbulent wind, time-
domain calculation in Bladed. Furthermore, it is a requirement that the ‘up-wind turbine wake’
module and the ‘idling’ module are both specified in the template. However, the actual
parametric values assigned to these specific modules are, at this stage, unimportant.
The template file as provides the description of the actual turbine structure (rotor, nacelle and
tower assembly), the control system and the dynamic simulation controls, whereas the
processed information from Windfarmer specifies the environmental conditions. Clearly, it is
usually important to ensure that common information assumed in the two models is consistent
– for example, the turbine hub height and rotor diameter.
User-defined Bladed Environmental conditions
simulation template from Windfarmer
Multiple Bladed
site-specific simulation files
The Bladed simulation template may contain values for the environmental parameters; but
these will merely be overwritten during execution of the Windfarmer Link software.
In situations where the user wishes to overwrite particular outputs from Windfarmer, an
options module in the user interface allows particular values of air density, wind shear and up-
flow angle to be assigned to all simulations. See the User Manual for further details.
The wind speed bins specified by the user in the Windfarmer Link screen are used to define
the individual wind speeds and total number of the generated wind files. The turbulence data
from Windfarmer is used to define the turbulent characteristics of the wind files.
The Windfarmer Link module determines the required dimensions of the wind files from the
hub height and rotor diameter information contained in the turbine summary data output from
Windfarmer. It also assigns random number seeds to each wind file.
The turbulence data from the wind file is then combined with the appropriate wake profile,
wind shear and upflow angle during the execution of each dynamic simulation.
The algorithm computes an ‘indication of severity’ parameter X for each turbine T, for a range
of SN-slope values, such that:
1
i, j m
X T Pi , j ,T U i ,T TI i , j ,T
m
where U is the hub height rotor-averaged wind speed (including wake deficit effects), TI is the
turbulence intensity distribution, P is the wind speed / direction probability distribution, m is
the Wohler exponent or inverse S-N slope (Section 12.6.2), T is the turbine index, i is the wind
speed index and j is the wind direction index of the flow and performance matrices.
The calculation integrates the standard deviation of the wind speed at each turbine with the
probability distribution, with an adjustment for the material inverse S-N slope m. The
probability distribution is not normalised, so that it accounts for the variation in operational
hours across the turbine positions.
The ranking method does not take into account the wind shear and upflow angles, so in
addition to the turbines suggested by the ranking, the user should manually select any
turbines experiencing high wind shear and/or upflow angles for explicit simulation.
The Windfarmer Link method automatically generates rainflow cycle count simulations for
execution in Bladed. The user is required to define a rainflow simulation template which the
link then alters to correctly assign load cases and their associated probabilities based on the
site specification provided by WindFarmer. Final loads are then extracted from Bladed in the
normal way.
REFERENCES
2.1 Glauert H, “An aerodynamic theory of the airscrew”, Reports and memoranda, AE. 43,
No 786, January 1922
2.2 Prandtl L and Tietjens O G, “Applied Hydro and Aeromechanics”, Dover Publications,
1957
2.3 Pitt D M and Peters D A, “Theoretical prediction of dynamic inflow derivatives”, Vertica,
Vol. 5, No 1, 1981
2.4 Gaonkar G H, Sastry VV,, Reddy T S R, Nagabhushanam J and Peters D A, “The use
of actuator disc dynamic inflow for helicopter flap lag stability”, 8th European Rotorcraft
Forum, France, Sept. 1982
2.5 Tuckerman L B, “Inertia factors of ellipsoids for use in airship design”, NACA Report No
210, 1925
2.8 Leishman J G and Beddoes T S, “A semi-empirical model for dynamic stall”, Journal of
the American Helicopter Society, July 1989
2.9 Harris A, “The role of unsteady aerodynamics in vertical axis wind turbines”, Recent
developments in the aerodynamics of wind turbines, BWEA workshop, University of
Nottingham, February 1990
2.11 Kirk Gee Pierce “Wind Turbine Load Prediction using the Beddoes-Leishman Model for
Unsteady Aerodynamics and Dynamic Stall”, University of Utah Thesis, August 1996
3.2 Shabana, A.A., Dynamics of Multibody Systems, Second Edition. Cambridge University
Press, Cambridge, 1998
3.3 Géradin, M. and Cardona, A., Flexible Multibody Dynamics. John Wiley & Sons Ltd,
Chichester, 2001
3.4 Meijaard, J.P., A Proposal for the Application of Multibody Dynamics to the Simulation
of Wind Turbines. Technical report No. 2005-1-B prepared in commission of Garrad
Hassan and Partners Ltd, Bristol, December 2005 (Internal document, not publicly
available)
3.5 Cook, R.D., Malkus, D.S., and Plesha, M.E., Concepts and Application of Finite
Element Analysis, Third Edition. John Wiley & Sons, New York, 1989
3.6 Clough, R.W. and Penzien, J., Dynamics of Structures, Second Edition. McGraw-Hill,
Inc., New York, 1993
3.7 Przemieniecki, J.S., Theory of Matrix Structural Analysis. McGraw-Hill, Inc., New York,
1968
3.8 Craig Jr., R.R., ‘Coupling of Substructures for Dynamic Analysis: An Overview’,
AIAA Paper, No. 2000-1573, AIAA Dynamics Specialists Conference, Atlanta, Georgia,
April 5, 2000
3.9 Craig Jr., R.R. and Bampton, M.C.C., ‘Coupling of Substructures for Dynamic Analysis’,
AIAA Journal, Vol. 6, No. 7, pp. 1313-1319, 1968
th
3.10 Kreyszig, E., Advanced Engineering Mathematics, 9 edition, John Wiley and Sons,
inc, 2006
3.11 Bir, G. “Multiblade coordinate transformation and its application to wind turbine
analysis”, ASME wind energy symposium, Reno, Nevada, 7-10 January 2008.
3.12 Hansen M.H., “Improved Modal Dynamics of Wind Turbines to Avoid Stall-induced
Vibrations”, Wind Energ. 2003; 6:179–195.
4.1 Ahmed-Zaid S and Taleb M, Structural modelling of small and large induction
generators using integral manifolds, IEEE trans. Energy Conversion 6, 3, September
1991.
4.2 Park R H, Two-reaction theory of synchronous machines, Trans AIEE 48, 1929.
4.3 Bossanyi E A, Investigation of torque control using a variable slip induction generator,
ETSU WN 6018, ETSU, 1991.
4.6 Vas P: “Sensorless vector and direct torque control”, Oxford University Press, 1998,
ISBN 19 8564651.
4.10 Duane C H, “Brushless permanent magnet motor design”, 2nd Edition , Magna Physics
Publication , 2006, ISBN: 1881855155.
7.1 IEC 1400-1, Wind turbine generator systems - Part 1: Safety requirements, First
edition, 1994.
7.2 Powles S R J, “The effects of tower shadow on the dynamics of a HAWT”, Wind
Engineering, 7, 1, 1983.
7.3 Veers P S, “Three dimensional wind simulation”, SAND88 - 0152, Sandia National
Laboratories, March 1988.
7.4 Engineering Sciences Data Unit, “Characteristics of atmospheric turbulence near the
ground. Part II: Single point data for strong winds”, ESDU 74031, 1974.
7.5 Engineering Sciences Data Unit, “Characteristics of atmospheric turbulence near the
ground. Part II: Single point data for strong winds (neutral atmosphere)”, ESDU 85020,
1985 (amended 1993).
7.6 Engineering Sciences Data Unit, “Characteristics of atmospheric turbulence near the
ground. Part III: Variations in space and time for strong winds (neutral atmosphere)”,
ESDU 86010, 1986 (amended 1991).
7.7 IEC 1400-1, Wind turbine generator systems - Part 1: Safety requirements, Second
edition, 1999.
7.8 Ainslie J F, “Development of an eddy viscosity model for wind turbine wakes”,
th
Proceedings of 7 BWEA Wind Energy Conference, Oxford 1985.
7.9 Ainslie J F, “Development of an Eddy Viscosity model of a Wind Turbine Wake”, CERL
Memorandum TPRD/L/AP/0081/M83, 1983.
7.10 H Tennekes and J Lumley, “A first course in turbulence”, MIT Press, 1980.
7.11 L Prandtl, “Bemerkungen zur Theorie der freien Turbulenz”, ZAMM, 22(5), 1942.
7.12 Ainslie J F, “Calculating the flowfield in the wake of wind turbines”, Journal of Wind
Engineering and Industrial Aerodynamics, Vol 27, 1988.
7.13 Taylor G J, “Wake Measurements on the Nibe Wind Turbines in Denmark”, National
Power, ETSU WN 5020, 1990.
7.14 Quarton D C and Ainslie J F, “Turbulence in Wind Turbine Wakes”, Wind Engineering,
Vol. 14 No. 1, 1990.
7.15 U Hassan, “A Wind Tunnel Investigation of the Wake Structure within Small Wind
Turbine Farms”, Department of Energy, E/5A/CON/5113/1890, 1992.
7.17 Vermeulen P and Vijge J, “Mathematical Modelling of Wake Interaction in Wind Turbine
Arrays, Part2”, report TNO 81-02834, 1981.
7.19 J. Mann, “Wind field simulation,” Prob. Engng. Mech., v. 13, n. 4, 1998, pp. 269-282.
7.20 IEC 1400-1, Wind turbine generator systems - Part 1: Safety requirements, Third
edition, 2004.
8.1 Goda Y, “A Review on Statistical Interpolation of Wave Data”, Report of the Port and
Harbour Research Institute, Vol. 18, No. 1, March 1979.
8.2 Hogben N and Standing R, “Experience in Computing Wave Loads on Large Bodies”,
OTC 2189, Offshore Technology Conference, Houston, 1975.
8.3 Wheeler J D, “Method for Calculating Forces Produced by Irregular Waves”, J. Petr.
Techn., pp.359-367, March 1970.
8.4 Gudmestad O T, “Measured and Predicted Deep Water Wave Kinematics in Regular
and Irregular Seas”, Marine Structures, Vol. 6, pp.1-73, 1993.
8.7 Dean R G, “Stream Function Wave Theory: Validity and Application,” Proceedings of
the Santa Barbara Specialty Conference, Ch. 12, Oct. 1965.
8.10 MacCamy, R.C. and Fuchs, R.A., Wave Forces on Piles: a Diffraction Theory, Tech.
Memo No. 69, US Army Corps of Engineers, Beach Erosion Board, 1954.
8.11 Rainey P J and Camp T R, “Constrained non-linear waves for offshore wind turbine
design”, Journal of Physics: Conference Series 75 (2007) 012067, “The Science of
Making Torque from Wind”,
http://www.iop.org/EJ/article/1742-6596/75/1/012067/jpconf7_75_012067.pdf.
10.1 Rice S O, “Mathematical analysis of random noise”, Selected papers on noise and
stochastic processes, ed. N Wax, 1959.
10.2 Davenport A G, “Note on the distribution of the largest value of a random function with
application to gust loading”, Proc. Inst. Civil Eng. 28, pp187-196, 1964.
10.5 Flickermeter functional and design specification, BSEN60868, 1993, and evaluation of
flicker severity, BSEN60868-0, 1993, equivalent to IEC 868-0, 1991.
10.6 Cheng, P.W., “A Reliability Based Design Methodology for Extreme Responses of
Offshore Wind Turbines”, PhD thesis, Delft University of Technology, 2002.
10.7 Norton, E. J. “Investigation into IEC Offshore draft standard Design Load Case (DLC)
1.1”, GH report 2762/BR/20, or
http://www.risoe.dk/vea/recoff/Documents/Sec_3/RECOFFdoc059.pdf
11.1 “WindFarmer Theory Manual”, GH & Partners Ltd, 2002.
11.4 Adams, B.M., et al., “Dynamic loads in wind farms II,” Final report of CEC Joule
project JOU2-CT92-0094, GH report 286/R/1, 1996.