DEVELOPMENT OF WIND TURBINE CONTROL ALGORITHMS FOR INDUSTRIAL USE

T.G. van Engelen, E.L. van der Hooft, P. Schaak

Energy research Centre of the Netherlands (ECN), Wind Energy
P.O. Box 1, NL-1755 ZG Petten, The Netherlands
Telephone: ++31 224 564141
Telefax: ++31 224 568214
email: vanengelen@ecn.nl

ABSTRACT: A tool has been developed for design of industry-ready control algorithms. These pertain to the prevailing
wind turbine type: variable speed, active pitch to vane. Main control objectives are rotor speed regulation, energy yield
optimisation and structural fatigue reduction. These objectives are satisfied through individually tunable control loops.
The split-up in loops for power control and damping of tower and drive-train resonance is allowed by the use of dedicated
filters. Time domain simulation results from the design tool show high-performance power regulation by feedforward of
the estimated wind speed and enhanced damping in sideward tower bending by generator torque control. The tool for
control design has been validated through extensive test runs with the authorised aerodynamic code PHATAS-IV.
Keywords: Control, Variable speed operation, Dynamic models.

1. INTRODUCTION
generator curve
Ω generator speed
The number of wind turbine manufacturers that apply pitch T
curve
Θ blade angle
full Tgen generator torque
control towards feathering position for power limitation is load x nod tower nodding
xnay tower naying
increasing considerably. Usually, pitch-to-vane control is Ω low pass u tilt control
combined with variable speed operation, which is facilitat-
ed by commercially available fast switching components in xnod
power electronics. Operation outside stall conditions and en- band
pass fore/aft tower ∆Θ
damp

hanced energy yield around and below nominal wind speed ω0t thrust control
+
are major drivers towards this concept. In comparison with Ω
nom low aero power Θ
full
Θ
set

power reduction by stall, the axial blade and tower loading

pass
ω3p control
pitch
actuator Θ
+ _ +
is smaller and the aerodynamic behaviour is much better curve set
very low electric power T T EM-torque
predictable. Especially at offshore siting the first feature is pass
control servo Tgen
+
being considered more important than ever because of the ω vw +
band
extreme high reliability requirements. pass shaft torsion
ω 0d torque control +
This situation raises the need for control algorithms for vari- ∆Tdamp
band lateral tower +
pass
able speed pitch-to-vane wind turbines. For this reason, a ω0t torque control
design tool for such control algorithms has been developed x nay
at ECN [4]. This paper addresses the following topics of the Ω
control tool: Figure 1: Feedback loops for control of rotor speed, power,
• problem identification and approach; tower bending and shaft distortion
• turbine modelling and design principles;
• time domain simulation results.
distortion damping are typical narrow band processes around
ω0t and ω0d , both (far) beyond ωVw (low- and band-pass
2. CONTROL PROBLEM AND APPROACH filters in fig. 1). Furthermore, the frequency ω3p applies
(∼ 0.6−1.2 Hz). This is the center frequency of the effects
The main control loops concern the power production and
of rotationally sampled turbulence and tower shadow. For a
rotor speed behaviour. Besides, control loops can be added
3-bladed rotor, this is 3 times the rotational frequency (3p);
for compensation of resonances (active damping). The latter
for a 2-bladed rotor this ω2p (2p), which does not differ much
loops are not allowed to significantly disturb the primary
from ω3p as the rotor speed is considerably higher for a 2-
control functions. The resonances may appear in the rotor
blader. A suitable filter should eliminate these Bp-effects in
blades, the drive-train and the tower. In a multivariable
the pitch-actuator activity (B=2,3). However, an exception
design approach [1] the difference between all loops will not
can be made for the control loop on shaft distortion as this
exist any more. In the developed design tool, the different
loop can also be used for reduction of inertia loads caused
control loops (fig. 1) are designed separately.
by ‘Bp rotor acceleration’.
Separate loop design is enabled because the frequency ranges
The control scheme in fig. 1 does not deal with resonance
of the phenomena to be controlled significantly differ, which
of rotor blades because it is limited to active damping on-
is illustrated in figure 1. The following typical frequencies
ly. Blade resonance is usually reduced by passive damper
exist:
devices (‘mass spring damper’ systems).
• ωVw : rotor uniform turbulence (∼0.07 Hz);
• ω0t : first tower bending mode (∼0.35 Hz); 3. MODELLING AND CONTROL SYNTHESIS
• ω0d : first shaft distortion mode (∼2.5 Hz).
The next subsections deal with the turbine model for control
Aerodynamic and electric power control (thick-line blocks)
design and the synthesis principles of the identified loops.
concerns frequencies around ωVw . Tower bending and shaft
3.1 Model for control design expressions in the pitch angle. The approach is based on the
The control tool includes models for wind and wave influ- dynamic inflow modelling principle in [6] .
ences and for the dynamic response of the wind turbine. The The stochastic wave simulation is based on water depth de-
model features are listed below and discussed afterwards. pendent power spectra of the wave velocity and the wave
External influences: acceleration. All these spectra are governed by the pow-
er spectrum of the surface elevation through the linear wave
• stochastic wind and wave generation; theory (Airy) [7]. Figure 3 shows for a water depth d of 20 m
• aerodynamics by BEM-theory; the (fully) correlated horizontal wave speed and acceleration
• dynamic inflow effect of blade pitching; signals on 55%, 65%, up to 95% of d above the sea bottom.
• hydrodynamics by Morison’s equation. A Pierson Moskowitz wave spectrum has been applied at an
Wind turbine system dynamics: average wind speed of 12 m/s.
• first bending mode of tower (2 directions); Afstand tot grond: 55% 65% 75% 85% 95% van waterdiepte van 20.0 m
0.6
• first distortion mode of drive-train; 0.4

[m/s] bij 12.0m/s

linear servo behaviour for generator torque; 0.2

• non-linear servo behaviour for pitch actuation; 0

•
−0.2
delayed and quantisized measurements. −0.4

golf
v
−0.6
The stochastic wind simulation is based on a rotor-wide −0.8
100 110 120 130 140 150 160 170 180 190 200
description of the effect of rotationally sampled wind turbu-
lence, tower shadow and wind shear. This approach is based 1

on the modelling principle in [3]. A wind signal is obtained

[m/s2] bij 12.0m/s
0.5
by inverse Fourier transform of ‘the rotor-wide’ power spec-
trum of the wind field as ‘sampled’ by the rotor blades. This 0

spectrum is derived from auto power spectra and coherence

golf

−0.5
C ⋅a
m

functions in accordance with IEC standards. Figure 2 shows

−1
a typical generated wind speed signal; the detailed lower 100 110 120 130 140
tijd [s]
150 160 170 180 190 200
file F:\tgengel\ctrltool\MODELS\PS\khv12d20.ps 08−Dec−2000
graph visualises the effect of rotational sampling and tower by F:\tgengel\ctrltool\MODELS\M\hydrload.m

shadow on this signal (Bp-effects). Figure 3: Realisation of wave speed and acceleration

wind signal for power and thrust coefficient data including turbulence and tower shadow
16 Note that the lower graph in fig. 3 shows the product of
14 mass coefficient Cm and wave acceleration a. This product
12 is the ‘force effective’ acceleration with wave diffraction
[m/s]

10 included as proposed by MacCamy and Fuchs [2]. If the

8 waves are perpendicular to the wind, Morison’s equation for
6
0 50 100 150 200 250 300 350 400
the (lateral) wave force per unit tower length fhy on z m
below the sea surface is given by
11.5

11 fhyz = ρwater · 14 πDz2 · (ẇzeff − ẍnyz ) + . . .

10.5

10 CdH · 12 ρwater · Dz · (wz − ẋnyz ) · |wz − ẋnyz | ,

[m/s]

9.5

9
with tower diameter Dz , wave speed wz and force effective
8.5

8
acceleration ẇzeff , naying speed ẋnyz and acceleration ẍnyz .
100 102 104 106 108 110 112 114 116 118 120
time [s] file F:\tgengel\verkoop\articles\ewec01\paper\fgvweff.ps 26−Jun−2001 Realistic values for the drag coefficient CdH lay between 0.6
and 1.2 [7].
Figure 2: Realisation of rotor-wide wind speed
Tower bending and drive-train dynamics are modelled by the
The rotor-wide wind speed Vw is fed through power and following set of mutually dependent differential equations
thrust coefficient data (Cp , Ct ) for conversion to the aero- (waves perpendicular to wind):
dynamic torque Ta and axial force Fa (with nodding speed mt ẍnd = −kt ẋnd − ct xnd + Fa cos θtilt ,
ẋnd ; rotor blade radius Rb , tip speed ratio λ):
Tnac
mt ẍny = −kt ẋny − ct xny + 32 + ...
Cp (θac , λ)/λ · 12 ρπRb3 · (Vw − ẋnd )2 , Lt
Ta =  d

Fa = Ct (θac , λ) · 12 ρπRb2 · (Vw − ẋnd )2 . F(fhyz )dz ,

0
Jr ·Jg
The pitch angle value θac as used in aerodynamic conversion Jr +Jg
· γ̈ = −cd · γ − kd · γ̇ + . . .
is obtained from the ‘physical’ pitch angle value θph by Jg
· (Ta − T ) + Jr
· Te ,
having θph led through a so called lead-lag filter. This filter Jr +Jg Jr +Jg

models the dynamic inflow effect of pitching through the Jr · Ω̇r = −cd · γ − kd · γ̇ + Ta − T .
following differential equation:

 The shaft distortion speed γ̇ is the difference between ro-
τiDI · d
dt
(θac ) + θac = τdDI · d
dt
θph + θph . tor speed Ωr and ‘slow shaft level’ generator speed Ωg /igb
(gearbox transmission ratio igb ). Jr and Jg are correspond-
The time constants τiDI and τdDI depend on the operating ing moments of inertia; cd and kd are the shaft stiffness and
conditions. The actual values are obtained from polynomial damping constant for the 1st distortion mode with natural
 meas
damping rate βd (∼ 0.005). Tower mass mt and damping Θ

and stiffness constant kt and ct are the tower top equivalent LPF partial load
pitch setting Ω3pfilt
parameters for the 1st bending mode with natural damping 3pfilt
r
Θ
rate βt (∼ 0.005). nom
Ωr setpoint scheduling
adaptation
The integral of function F in the hydrodynamic force dis- Ωr
ref part
Θ set

+
tribution fhyz yields the tower top equivalent hydrodynamic Θ
PD LL
load. Function F caters for the shape of the 1st bending
+ +
- + + full switch limitation
Ω3pfilt dynamic inflow
inactivity zone Θ
mode and the distance Lt − (d−z) between the tower top
r feedback
compensation force
Θ meas
Ωr
and fhyz . The loss torque T is modelled by a constant and D forced speed full / partial
limitation load selector
a rotor speed dependent term. The sideward tower bend- `feedforward'

ing torque Tnac approximately equals the (slow shaft level) Ta

est
3pfilt
Ωr Ω3pfilt
r
generator torque Te . P/Ω +
+
J d/dt LPF
low-pass

The servo behaviour of the generator torque is modelled by 3pfilt

Pe Pe
meas

2nd order dynamics with cut-off frequency ωsv Te

and damping LPF
Te
rate βsv . The pitch servo model includes both 2nd order Ωr
torq

θ̇ θ̇ VLPF
dynamics (ωsv , βsv ) and a delay τdθ̇ that depends on sign very low-pass set
Te
reverse in pitching speed setting and on the thrust force. full
load
torque/speed curve

Control tool modules. The models listed above have been

implemented in MATLAB program modules for numeric Figure 4: Feedback structure for power control
integration in time domain simulations. They also have been
included in linearised form (transfer functions) in program
r − Ωr
for Ωref 3pfilt
as much as allowed in more favourable
modules for frequency domain based controller synthesis.
operating conditions and by fitting the D-gain for Taest on
3.2 Power control and resonance damping the inverted power coefficient data (gain scheduling).
Next to models for wind, waves and wind turbine system The feedback gains for tower and drive-train damping are
dynamics, the control tool incorporates feedback structures. derived from isolated analysis of the governing equations for
These pertain to aerodynamic and electric power control and the 1st bending and distortion mode. The tower loops include
to damping of resonance in the tower and drive-train. The narrow band-pass filters with nearly zero phase shift around
features are listed below and discussed afterwards. the tower eigenfrequency ω0t . Additionaly, sharp band-
Aerodynamic and electric power control: stop filters reduce the peaks around the mBp-frequencies
(m = 1, 2) in the nodding signal. The drive-train loop
• rotational speed feedback with setpoint adaptation;
includes a high-pass filter and a Kalman filter for estimation
• non-linear feedforward of estimated wind speed;
of the shaft distortion from the generator speed.
• dynamic inflow compensation;
• inactivity zone and filtering of Bp-effects; The servo systems for the actuators behave ideal in the tower
• scheduling of control parameters; θ̇
loops (ωsv  ω0t , ωsv
Te
 ω0t ), whereas in the drive-train loop
• forced rotational speed limitation; Te
the actuator bandwidth is sufficiently large (ωsv ≥ 2ω0d ).
• partial load pitch setting;
The filtered tower signals and estimated distortion speed
• smooth transients between partial and full load; bp ˆ hp
• low-pass effectuation of torque/speed curve. (ẍbp
nd , ẋny , γ̇ ) are fed back to damping contributions in
pitch speed and torque setting:
Tower bending and drive-train distortion damping
r
∆θ̇nd = Knd · ẍbp
nd ,
• nodding acceleration feedback to pitch speed;
• naying speed feedback to generator torque;
r
∆Tny = −Kny · ẋbp
ny ,
• narrow band-pass filter in tower loops; ∆Ttrr = −Ktr · γ̇
ˆ hp
.
• shaft distortion speed feedback to generator torque;
• Kalman filter in drive-train loop; The relevant parts of the bending and distortion equations
• maximum level in control effort. are then approximated by ( ∂F
∂θ
a
< 0):

Figure 4 visualises the feedback structure for aerodynam- mt ẍnd ∼ −kt ẋnd − ct xnd − | ∂F
∂θ
a
| · Knd · ẋnd ,
ic and electric power control. The proportional differ- Kny
ential (PD) feedback of the filtered rotational speed error mt ẍny ∼ −kt ẋny − ct xny − 2
3 · ẋny ,
Lt
r − Ωr
Ωref 3pfilt
is the core of the aerodynamic control loop; Jr ·Jg ˆ hp .
it is the usual approach to control the inertia based rotor Jr +Jg
· γ̈ ∼ −cd · γ − kd · γ̇ − Jr · Ktr · γ̇ Jr +Jg

dynamics. The differential (D) ‘feedforward’ of the estimat-

This yields the following enhanced damping rates:
ed aerodynamic torque Taest effectuates pseudo wind speed
∂F
feedforward towards the pitch angle that belongs to the actual kt +| ∂θa |·Knd
βtnd ∼ √
2 mt ct
,
wind speed. The lead-lag filter for dynamic inflow compen-
sation (LL) implements the inverse of the dynamic inflow kt + 3 K /Lt
βtny ∼ √2 ny
,
model equation in §3.1. 2 mt ct

The control gains in the feedback and ‘feedforward’ loop βd ∼ √

kd +Jr /(Jr +Jg )·Ktr
.
2 Jr Jg /(Jr +Jg )·cd
are derived with Nyquist analysis in worst case operating
conditions (industry-adopted stability assessment). High- The feedback gains Knd , Kny and Ktr are tuned in non-
performance control is obtained by enlarging the PD-gains linear time domain simulations. The achievable damping
 Effect naying damping (all filters included); dotted: bF = 1, solid: bF = 15

25
rate in realistic wind conditions is constrained by the allowed

Ωr [rpm]
level of (harmonic) control effort and stability requirements. 20

rel
The nodding gain Knd is scheduled in a similar way as the 15

PD-gains for the rotational speed error. 0 50 100 150 200 250 300 350 400

Control tool modules. The feedback structures listed above 2000

have been implemented in MATLAB program modules for

Te [kNm]
1500
time domain simulation: the MATLAB edition of the con-
trol algorithms. The algorithms are also available in the 1000
0 50 100 150 200 250 300 350 400
programming languages C and Fortran for straightforward
incorporation in process computers and aerodynamic codes. 0.1
0.05

any [m/s]
0
Besides, interactive program modules have been developed −0.05
−0.1
for parametrising the filters and gains of the linear parts in
0 50 100 150 200 250 300 350 400
the control loops; these modules include the linearised wind time [s]
file E:\schaak\TowDamp\SCOPE\PS\scnyeff.ps 01−Feb−2001
by E:\schaak\TowDamp\SCOPE\M\slnywtrs.m

turbine system dynamics. For overall stability and robust-

ness assessment, program modules have been developed for Figure 6: Lateral tower resonance at waves perpendicular
Nyquist analyses of the open loop transfer function. This to the wind (lower graph: naying acceleration [m/s2 ]; dash:
transfer function is obtained by linearisation of the integrat- without damping)
ed model of the control loops and the wind turbine, with the
main feedback path cut through, that is to say the rotational
damping of tower bending and drive-train distortion. Spe-
speed measurement feedback path to the PD-action.
cial features are (i) dedicated filter design, (ii) wind speed
estimation in power control and (iii) shaft distortion estima-
4. SIMULATION RESULTS tion by Kalman filtering.
The results plotted below apply to a typcial multi-MW (off-
The algorithms with the implemented control loops are clear
shore) wind turbine. They have been obtained from the
in implementation and operation, and are on-site tunable
simulation stage in the design tool. Validation runs with the
by well-educated operators. The C- or Fortran-coded algo-
aerodynamic computer code PHATAS [5] (control algorithm
rithms can be incorporated in process computers and aero-
included) yielded equal behaviour.
dynamic codes with very minor effort.
rotor effective wind speed (gray); estimated windspeed; rated wind speed (dashed)
The approach as implemented in the tool has been exensively
20
validated by non-linear time domain simulations with the
authorised aerodynamic code PHATAS [5].
[m/s]

15

10

400 450 500 550 600 650

ACKNOWLEDGEMENT
aerodynamic power (gray); electric power
5 Koert Lindenburg (ECN) is acknowlegded for his contribu-
4
tion to dealing with the impact of dynamic inflow on power
[MW]

3
control and for the many validation runs with PHATAS. Jan
2
Pierik (ECN) is acknowlegded for his contributions to elec-
1
400 450 500 550 600 650
rotor speed (gray); rotor speed setpoint; rated rotor speed (dashed)
tric system modelling and desk top publishing.
18

16
REFERENCES
[rpm]

14

12
[1] P.M.M. Bongers; Modeling and Identification of flexi-
10
ble wind Turbines and a Factorizational Approach to Robust
400 450 500 550 600 650
pitch angle Control, PhD thesis, ISBN 90-370-0100-9, Delft Universi-
15
ty of Technology, fac. of Mech. Eng., the Netherlands, 1994.
10
[2] S.K. Chakrabarti; Hydrodynamics of Offshore Structures,
[dg]

5
Computational Mechanics Publications Southampton, 1987.
0
[3] J.B. Dragt; Atmospheric Turbulence Characteristics in
400 450 500 550 600 650

4
pitching speed the Rotating Frame of Reference of a WECS Rotor. Pp 274-
2
278 in proc. ECWEC Conf. Madrid, Spain, 1990.
[4] T.G. van Engelen, E.L. van der Hooft and P. Schaak;
[dg/s]

−2 Ontwerpgereedschappen voor de Regeling van Windturbines

−4
400 450 500 550 600 650
(in Dutch), Technical report, ECN Wind Energy, Petten, the
time [s]
Netherlands, Draft, June, 2001.
Figure 5: Aerodynamic and electric power control with wind [5] C. Lindenburg and J.G. Schepers; PHATAS-IV Aeroe-
speed estimator lastic Modelling, Program for Horizontal Axis Wind turbine
Analysis and Simulation, version IV, ECN Wind Energy,
Petten, the Netherlands.
5. CONCLUSIONS [6] H. Snel, J.G. Schepers; Joint Investigation of Dynam-
A design tool has been developed for control algorithms for ic Inflow Effects and Implementation of an Engineering
variable speed wind turbines. The by nature multivariable Method. Technical Report ECN-C-94-107, ECN Wind En-
control problem is split-up into physically interpretable con- ergy, Petten, the Netherlands, April, 1995.
trol loops that are individually parametrised. These loops [7] J.F. Wilson; Dynamics of Offshore Structures. John
pertain to aerodynamic and electric power control and to Wiley & Sons, 1984.