Sei sulla pagina 1di 17

COUPLING OF AN UNSTEADY LIFTING LINE FREE VORTEX WAKE CODE

TO THE
AEROELASTIC HAWT SIMULATION SUITE FAST
Joseph Saverin
David Marten, George Pechlivanoglou,
Christian Navid Nayeri, Christian Oliver Paschereit
Chair of Fluid Dynamics
HFI Berlin
TU Berlin
ABSTRACT
A coupling of the Lifting Line Free Vortex Wake (LLFVW)
model of the open source wind turbine software QBlade and the
wind turbine structural analysis tool FAST has been achieved.
FAST has been modified and compiled as a dynamic library, tak-
ing rotor blade loading from the LLFVW model as input. Most
current wind turbine aeroelastic simulations make use of the
Blade Element Momentum (BEM) model, based upon a num-
ber of simplifying assumptions which are often violated in un-
steady situations. The purpose of the implemented model is to
improve accuracy under unsteady conditions. The coupling has
been thoroughly validated against the NREL 5MW reference tur-
bine. The turbine is compared under both steady conditions and
three unsteady operating conditions to the BEM code AeroDyn.
The turbine has been simulated operating at a constant RPM and
with a variable-speed, variable blade-pitch-to-feather controller.
Under steady conditions the agreement between the LLFVW and
AeroDyn is demonstrated to be very good. The LLFVW produces
different predictions for rotor power, blade deflection and blade
loading during transient conditions. A number of important ob-
servations have been made which illustrate the necessity of a
higher fidelity aerodynamic model. The validation and results
are considered as a step towards the implementation of an open-
source, high fidelity aeroelastic tool for wind turbines.
Address all correspondence to this author: j.saverin@tu-berlin.de
NOMENCLATURE
dTa
Time step of aerodynamic module
dTs
Time step of structural module
Fn
Blade normal force
Ft
Blade tangential force
P
Power
t
Simulation time

Out-of plane deflection of blade 1

Angle between freestream and rotor axis

Azimuthal position of blade 1


INTRODUCTION
The drive for transition to renewable energies and the re-
sulting growth of the wind energy sector is having a significant
impact on the wind turbine industry. The general drive to in-
crease turbine size motivates decreasing mass of turbine blades.
As blades become lighter however their elastic behaviour is af-
fected. This can often mean stronger aeroelastic coupling with
other structural components of the turbine. The occurrence of
off-design conditions such as highly turbulent inflow, turbine
control faults, grid loss or emergency shutdowns give rise to
unsteady aerodynamics which through coupling cause unsteady
structural dynamics or vice versa. The ability to simulate such
Proceedings of ASME Turbo Expo 2016: Turbomachinery Technical Conference and Exposition
GT2016
June 13 17, 2016, Seoul, South Korea

GT2016-56290
1
Copyright 2016 by ASME
Downloaded From: http://173.254.190.160/pdfaccess.ashx?url=/data/conferences/asmep/89529/ on 04/07/2017 Terms of Use:
http://www.asme.org/about-asme/terms-of-use

Page 2
events requires an aerodynamic model which can account for
such unsteady behaviour, and thus improve predictions of un-
steady loads. The industry standard aerodynamic model for wind
turbine certification is the Blade Element Momentum (BEM)
model, pioneered by Glauert [1]. The model relies upon a num-
ber of simplifying assumptions which, although in their summa-
tion allow for a relatively simple and robust model, are often
violated in even the most ideal operating conditions of a wind
turbine [2]. Although the BEM model is the least computa-
tionally expensive method available, it performs unsatisfactorily
for a range of unsteady modelling situations and a higher order
model is necessary [3]. A number of software packages have
been developed to model aeroelasticity of wind turbines includ-
ing NRELs FAST [4], DTUs HAWC2 [5] or GLs Bladed [6].
These and many others are based upon couplings of BEM mod-
els with structural models [7]. The recent implementation of
the LLFVW model into QBlade [8] marks a significant advance-
ment in the software as unsteady behaviour can be modelled and
the transient forcing and wake behaviour can be simulated. A
number of the restrictive assumptions of the BEM model are dis-
carded and wake behaviour and influence on inflow can be mod-
elled directly. Although the model is of lower fidelity than a full
finite-volume numerical model of the flow field, it can be shown
that the increase in accuracy more than justifies the increase in
computational expense in comparison to BEM models.
SOFTWARE MODULES
The work following shall be a validation of the coupling be-
tween the LLFVW module of QBlade and ElastoDyn, the struc-
tural module of FAST. The validation is carried out by comparing
the blade and turbine parameters predicted by the new aerody-
namics module with those which are predicted by the supplied
aerodynamics module of FAST, AeroDyn. This validation is car-
ried out with the knowledge that the traditional coupling of Elas-
toDyn and AeroDyn in FAST has been certified by Germanischer
Lloyd WindEnergie Gmbh [9], and can thus be assumed to be ac-
curate for a number of operating cases.
AeroDynamics:
LLFVW The formulation of the
LLFVW is based upon the lifting line model as developed by van
Garrel for ECNs AWSM code [10]. The blade is discretized into
aerodynamic panels. Relying on potential flow theory, the lifting
action of the blade is concentrated to the quarter chord point with
a filament of concentrated vorticity. The blade is discretized in
the spanwise direction to allow for variations in spanwise circula-
tion. By making use of the Kutta-Joukowski law [11], the lifting
force of each spanwise segment can be determined as a function
of the circulation of the segment:
Fl = Urelr ,
Fl = Cl()
1
2
U
2
rel
(1)
Where is the fluid density, r is the filament circulation and Urel
is the combined relative velocity of the as seen by the blade ele-
ment due to inflow, relative motion and wake elements. By com-
bining the two equations, an iterative procedure can be derived
which solves for the circulation at every spanwise station [12].
This makes use of the fact that the airfoil polars are known for
each airfoil section to give the airfoil lift and drag coefficients, C l
and Cd respectively at each spanwise station. This can be either
imported from experimental data into QBlade or calculated with
the built-in functionality of XFoil [13]. The advantage of this
FIGURE 1: The nodes can be seen to be deflected away from
their rigid position (shown in black). The wake of a single blade
is shown to demonstrate that the model predicts wake rollup and
dynamics.
form of modelling is that according to Kelvins circulation the-
orem, the wake can be modelled by accounting for variations in
spanwise circulation (trailing vortices) and shed circulation (shed
vortices). By making use of the Biot-Savart law for the velocity
induced by a vortex element the induced velocity of each wake
filament on the flow field can be determined [11]. By summing
the effect of all elements in the wake, the effect of flow retarda-
tion behind the rotor is modelled. By determining the velocity
at wake points, wake convection can also be modelled. This is
illustrated in Figure 1.
Aerodynamics: AeroDyn (FAST) FAST v.8.10.00 has
been used for the comparison here. The complimentary aerody-
namics package provided for this version is AeroDyn v.14.03.01
This BEM model uses an iterative procedure at every blade sta-
tion to solve for the streamwise induction. A number of extra
capabilities are included in the system, these include skew cor-
rection, wake swirl, dynamic stall and a dynamic wake model,
details of which can be found in [14
Structural: ElastoDyn (FAST) The current work
makes use of ElastoDyn v1.02.00. ElastoDyn makes use of an
Euler-Bernoulli beam model [15]. Prescribed modes in the form
of sixth order polynomials are used as input into a Rayleigh-
Ritz method to calculate blade deflections and motions. This
does not take into account torsional or axial degrees of freedom.
The model considers the first three dominant modes of a turbine
blade, the first two flapwise modes and the first edgewise mode.
A simulated blade deflection in the first flapwise mode is shown
in Figure 1.
IMPLEMENTATION
FAST has been written in the language Fortran and allows
only a set of pre-defined output parameters, which are output
only upon completion of the entire simulation. As this does not
allow for a true two-way coupling, certain sections of the FAST
source code have been rewritten to output the necessary instan-
taneous data at chosen stages of the simulation, which are then
passed to QBlade. The aerodynamic portion of the calculation
is then handled by QBlade, which passes the new loading val-
ues back to FAST and continues the calculation, see Figure 2.
This is referred to as a loose coupling [16] and is is sufficient
for the majority of simulations provided dynamic and aerody-
namic time steps are resolved sufficiently. As a result of the fact
that the structural natural frequencies of the turbine are usually
much higher than any aerodynamic excitations, the integration
time step of the structural dynamics module dTs must generally
be much smaller than the aerodynamic time step dTa. The struc-
tural loop hence usually requires many more iterations than the
aerodynamic loop which contains it.
Comparison
For the comparisons performed in this work, the NREL
5MW reference turbine has been chosen [17]. Significant pa-
rameters for this turbine are summarized in Table 1. Operating
conditions have been chosen in each case to account for realis-
tic effects while facilitating simple aeroelastic comparison. For
all simulations, a set of conditions have been prescribed. These
shall be individually detailed. For all simulations however, the
following conditions have been maintained:
- The aerodynamic effect of the tower on the blades and
tower wake shadow are included
- Blade first and second flapwise and first edgewise modes
are activated
- The tower is treated as being rigid
- The turbine does not physically yaw. Yawing effects are
simulated by changing wind direction.
- The hub-height wind velocity is maintained constant at the
rated velocity except in the case of gust simulation.
Simulation begin
t=0
LLFVW solve
Pass blade forces
to FAST
tn+1 = t +dTa
FAST solve
t = t + dTs
t tn+1
Pass structural
parameters to QBlade
True
False
t tsim
True
Simulation complete
False
FAST loop
FIGURE 2: Loose coupling routine of LLFVW (QBlade) and
FAST.
TABLE 1: NREL 5MW TURBINE PROPERTIES
Parameter
Rating
5 MW
Orientation, configuration
Upwind, 3 Blades
Control
Variable Speed, Collective Pitch
Drivetrain
Multiple-Stage gearbox
Rotor, Hub Diameter
126 m, 3 m
Hub height
90 m
Rated wind speed
11.4 ms1
Cut-in, Cut-out Wind Speed
3 ms1, 25 ms1
Cut-in, Rated rotor speed
6.9 RPM, 12.1 RPM
Overhang, shaft tilt, precone
5 m, 5, 2.5
For all unsteady cases, the simulations are allowed to run until
power, blade loading and tip deflection are converged (repeating
at every cycle) before any dynamic changes are allowed to occur
in the windfield. After the windfield returns back to the ideal
conditions, the solutions are allowed to run long enough that any
disturbance can pass through the resolved wake region.
Wake parameters
The wake model in the LLFVW implemented in QBlade
truncates the spanwise circulation by lumping together vortex
elements to reduce overhead [8], this is referred to as the fine
wake region. After extensive trialling, it was found that results
for essentially all observables differed negligibly when the wake
was fully resolved more than half of a full rotor rotation down
tream before being truncated. The wake parameters used for
all simulations are summarized in Table 2, where R represents
the azimuthal delay in rotor revolutions until vortex elements are
truncated.
TABLE 2: WAKE SETTINGS
Parameter
(R = Rotations)
Wake fully resolved
0.5R
Fine wake revolutions
7.5R
Maximum wake age
8R
Turbine controller
Use has been made of the controller built into the test cases
of the FAST software for the 5MW reference turbine [17]. This
is a conventional variable-speed, variable blade-pitch-to-feather
configuration. This makes use of two independent controllers;
a generator-torque controller and a rotor collective blade-pitch
controller.
Steady operation
The purpose of the steady case is to show that under steady
conditions, the LLFVW model produces very similar results. For
the the steady case, a constant, unsheared inflow of 11.4 ms1
is chosen with operating parameters as described for the rated
operating point of the turbine as given in Table 1.
Unsteady operation
Three unsteady operating conditions have been simulated.
These are chosen so as to represent cases where it is expected that
the unsteady formulation of the BEM will break down and the
aeroelastic behaviour of the rotor will be significantly different.
These have also been chosen loosely to reflect cases that will
arise in certification of the turbine according to the IEC-61400
guidelines [18].
Case 1: Extreme change of wind direction (EDC)
This represents the equivalent, however physically realistic
case that the turbine enters a yawing state. The angle between the
rotor axis and the wind direction (defined in Figure 4) is given
in this particular case as a linear step function, as is illustrated in
Figure 3.
Case 2: Extreme vertical wind shear (EWSV)
This case assumes that the wind is already operating within
a wind field with atmospheric shear. The transient vertical wind
28
30
32
34
36
38
0
10
20
30
Time [s]

[]
2
4
6
8
10
12
14
16
18
50
100
150
Wind speed [ms1]
Height
[m
]
Normal shear
Extreme shear
26 28 30 32 34 36 38 40 42 44
10
12
14
16
Time [s]
W
indspeed
[ms

1
]
FIGURE 3: Unsteady wind conditions: Extreme direction change
(top), extreme linear vertical shear (centre), extreme operating
gust (bottom).
profile is given (after start of transient condition) by:
V(z,t) = Vhub
[( z
zhub
)
+
(zzhub
Drot
)(
1cos
(2t
T
))]
Where z is the height of the point, the power law shear, Drot
the rotor diameter and is the linear vertical shear. In this case
= 0.2, = 0.991 and the transient occurs over a time period
T of 12 seconds.
Case 3: Extreme operating gust (EOG)
In this case the entire rotor is exposed to a mexican hat-type
pulse in the inflow velocity after which the velocity returns to
normal. Wind velocity is shown in Figure 3.
Results
The results for steady operation are presented first, this al-
lows for validation and observation of key differences in the
models. Following this, the unsteady results are shown and are
separated into each of the dependent variables observed:
1. Rotor power: low-speed shaft power
2. Blade deflection: Out-of-plane deflection (aligned with rotor
axis)
3. Normal blade loading (Thrust: normal to rotor plane)
4. Tangential blade loading (Torque: tangent to radius vector)
5. Controlled cases: Pitch angle and rotational speed
y
x
y
z
Rotor axis
Wind direction

FIGURE 4: Definition of the inflow angle and the azimuthal


position .
Steady operation
Rotor power Transient results are presented in Figure 5
for a rotor with both rigid and flexible blades. Converged values
for mean rotor power mean P, standard deviation std(P) and rela-
tive percentage difference as compared to AeroDyn are presented
in Table 3. For the rigid case, the LLFVW model is compared to
the idealized BEM model without a dynamic inflow correction
to model as closely as possible the ideal conditions of the BEM
assumptions. The results are seen to agree excellently and to be
within 1%. A number of interesting observations can be made
here. The reduction in power caused by tower passing is visible
in all simulations. It is also observed that in the rigid case, the
BEM model predicts no power decrease as the wake develops.
In the case that the blades are allowed to deflect, the dynamic
inflow model of AeroDyn has been activated and the transient
power can be observed as the model mimics the development of
a wake. In both simulations the LLFVW model is seen to predict
transient power as the wake behind the rotor develops and causes
the rotor induction to increase, decreasing power. It can be seen
that the oscillations in power predicted by the LLFVW model
are much smaller. In contrast to the BEM model, the LLFVW
model accounts for discrete additions to the wake, rather than
an instantaneously quasi-converged wake structure. This is seen
to account for much smaller deviations in power during conver-
gence to the steady value. In all configurations, convergence is
seen to be excellent and it is noted that as the wake is allowed
to develop, the value slowly approaches the idealized value as
predicted by BEM.
Effect of azimuthal discretization It was observed
that for azimuthal discretization greater than A 8deg the
0
2
4
6
8
10 12 14 16 18 20
4
6
8
10
Time [s]
Power
[M
W
]
AeroDyn, rigid
AeroDyn
LLFVW, rigid
LLFVW
FIGURE 5: Rotor power predicted by AeroDyn and LLFVW.
TABLE 3: CONVERGED ROTOR POWER VALUES
Simulation
P [MW] std(P) [kW] % dev.
Aerodyn-rigid
5.325
96.52

LLFVW-rigid
5.276
49.32
0.94
Aerodyn
5.497
161.69

LLFVW
5.497
115.78
0.008
Aerodyn-controller
5.296
34.88

LLFVW-controller
5.296
25.76
0.007
model diverged. A linear integration of velocity is used to de-
termine relative velocity in the LLFVW model, this can result in
the case of large structural deflections to unrealistic prediction
of blade relative velocity. This causes the aerodynamic load-
ing to be exaggerated and erratic. An analogous effect is seen
in AeroDyn. For all simulations here, the results are compared
to AeroDyn simulations with dTa = 0.0125, which corresponds
to an azimuthal discretization A 1. AeroDyn was found
to converge much quicker due to this lower time step. Smaller
timesteps give rise to larger instabilities in the far wake region
of the LLFVW model, which hinder convergence. It is also ob-
served that with the controller activated, convergence took much
longer for the LLFVW model due to the controller-wake interac-
tion. As the impact on blade loading pattern and deflection was
seen to vary negligibly for smaller timesteps, the simulations in
this work have been carried out with A = 3. The results are
summarised in Table 4, where it is noted the tconv refers not to
computing time, but rather simulation time t

TABLE 4: EFFECT OF AZIMUTHAL DISCRETIZATION


Simulation
P
std(P)
% dev tconv
[MW]
[kW]
-
[s]
AeroDyn
5.497
161.69

30
LLFVW-2
5.635
117.87
2.52
70
LLFVW-3
5.497
115.78
0.008
80
LLFVW-5
5.777
155.11
5.099
110
AeroDyn controller
5.296
34.88

30
LLFVW-3-controller 5.296
25.76
0.007
140
Blade deflection The blade deflection for the converged
model is given for two operating cases in Figure 6. The effect
of tower passing is again visible. The results are seen to agree
quite well and the mean blade deflection is well within 1% of the
AeroDyn value. The second case shown is the blade deflection
including atmospheric shear. The large increase in blade deflec-
tion as a result of the shear effect is quite visible. In this case,
as opposed to the steady case, the LLFVW predicts larger de-
flections, which shall be explained later with the blade loading
profile. Even in the sheared case, relative deflection error is be-
low 1%. The converged blade maximum and mean deflection
along with the difference between maximum and minimum de-
flection for both steady and controlled cases are given in Table 5.
0
50
100
150
200
250
300
350
5
5.5
6
Azimuthal position []
Deflection
[m
]
AeroDyn
AeroDyn+shear
LLFVW
LLFVW+shear
FIGURE 6: Blade out-of-plane deflection (no controller).
Blade normal loading Figure 7 illustrates the differ-
ence in loading between the two models for the steady, ideal
operating case. Although in the mid section of the blade the re-
sults agree very well, large deviations are seen at the tip and rot
regions. A principle consequence of not modelling the shed tip
TABLE 5: CONVERGED BLADE DEFLECTION
Simulation
max [m] [m] A [m]
Aerodyn
5.593
5.423
0.322
LLFVW
5.548
5.426
0.245
Aerodyn- controller
5.19
5.019
0.326
LLFVW- controller
5.10
4.969
0.254
or root vortices, the impact of the idealized modelling of BEM
is seen here. The LLFVW model predicts larger loading at the
tip, and reduced loading at the blade root as compared to Aero-
Dyn. The physical effect of releasing a vortex element at the tip
reduces the load at the tip as a result of redistributing pressure
between the suction and pressure sides of the blade. This is only
0
2
4
6
F
n
[kN
m

1
]
AeroDyn
LLFVW
0
10
20
30
40
50
60
0
0.5
1
Length [m]
F
t
[kN
m

1
]
FIGURE 7: Top: Blade normal loading. Bottom: Blade tangen-
tial loading ( = 0)
accounted for with idealized models in BEM such as the Prandtl
tip or root loss function [14]. It can be seen that the implemen-
tation of the tip loss models cause a discontinuity in the loading
distribution of the BEM method. This is visible at both tip and
root regions. Inspection of unsteady results indicates that this
occurs in all conditions. It is also seen that the iterative proce-
dure for the blade circulation in the LLFVW model produces in
essentially every case a continuous distribution of blade loading.
Blade tangential loading Similar results are seen here
for tangential loading. The LLFVW predicts higher loading at
the blade tip and lower loading at the blade root. The disconti
nuity in the BEM iterative procedure is again clear to see and the
AeroDyn curve is seen to generally be quite discontinuous. The
loading agrees again very well in the mid section of the blade. As
the loading in the LLFVW model is directly a function of the cir-
culation generated over the blade, this has consequences on the
spanwise shed vorticity. Modifications of the pitch angle cause
changes to the development of the wake as a result of the chang-
ing gradient of the spanwise circulation gradient seen here to be
of opposite sign to that predicted by the BEM model.
Unsteady operation
For all simulations the simulations were carried on long
enough until all transients had vanished. All plots here indicate
time after onset of the unsteady condition, which was initiated at
t = 140s to allow convergence in the controlled case.
Rotor power The case of extreme change of direction
is seen in Figure 8 to have a large effect on power. Figure 14
shows that in this case the controller reacts by reducing the ro-
tor speed, which consequently leads to the reduction in power.
This reduction is also seen to occur for the uncontrolled case,
where AeroDyn predicts a very large drop in power. The most
likely source for this is that the model of AeroDyn assumes that
the entire wake is immediately aligned with the flow direction.
This is seen to have quite a large effect on the power prediction
in comparison the the LLFVW model, which allows the previ-
ously shed shed to convect naturally to the next position. The
converged values of power for this case agree very well between
both models, as can be expected as the wake is fully aligned in
the skewed flow direction. For the case of extreme linear vertical
shear no drastic change in power is seen to occur. The controller
is seen to regulate the power quite effectively. The erratic load-
ing in the uncontrolled case is most likely due to the exaggerated
blade deflections as seen in Figure 9. The predictions made in
the controlled cases agree very well. As is to be expected in
the extreme operating gust, a drastic increase in power is seen
to occur. The predicted increase in power is less in the LLFVW
model than AeroDyn. This could be a consequence of decreased
loading on the blade as predicted by the LLFVW model, visible
in Figure 12. The loading is also seen to be drastically different
in the case that the blade reaches the apogee of its deflection,
which shall hereafter be referred to as the inflection point of the
blade motion. This shall be elaborated upon later.
Blade deflection Figure 9 shows the very large deflec-
tions predicted in the yawed state. Interesting here is that Aero-
Dyn predicts that the blade deflection remains relatively contin-
uous in reference to Figure 6, before entering into the large yaw
deflections whereas the LLFVW predicts an instant decrease in
the blade deflection. The rapid change is angle of attack expe-
4
4.5
5
5.5
P
[M
W
]
AeroDyn
AeroDyn, cont
LLFVW
LLFVW, cont
5
5.2
5.4
5.6
5.8
P
[M
W
]
0
2
4
6
8
10
12
14
16
4
6
8
10
Time [s]
P
[M
W
]
FIGURE 8: Rotor power during: Extreme direction change (top),
extreme linear vertical shear (centre), extreme operating gust
(bottom).
rienced by the blade may explain this as a dynamic stall effect.
A large deflection is seen here to occur for the case of extreme
linear vertical shear. It can be seen that the LLFVW model pre-
dicts slightly larger deflections than AeroDyn in both controlled
and uncontrolled cases. It appears that in the moments directly
following the onset of an unsteady excitation large deviations in
blade deflection and power occur. This too was observed to a
lesser degree in the case of the extreme vertical shear. Careful in-
spection of the blade position and relative deflection velocity in-
dicates that these large deviations occur near the inflection point
of the blade deflection. In this case the relative velocity changes
sign and the blade returns to the proximity of its own shed wake.
Such behaviour is not predicted at all in BEM models and can
only be accounted for with a resolution of the immediate wake
eroDyn
AeroDyn,cont
LLFVW
LLFVW,cont
2
4
6

[]
0
2
4
6
8
10 12 14 16 18 20
0
2
4
6
Time [s]

[]
FIGURE 9: Out of plane blade deflection under: Extreme direc-
tion change (top), extreme linear vertical shear (centre), extreme
operating gust (bottom).
Furthermore, oscillatory behaviour around this point produces a
deviation of results in comparison to AeroDyn. The controller
is seen to drastically feather the blade when the power increase
due to the gust is detected, visible in figure 9 and 14. It is ob-
served in almost all cases that the amplitude of blade deflections
predicted by LLFVW also does not agree with the BEM model,
often predicting higher deflection (this was also the case in ex-
treme vertical shear), this has an impact on fatigue damage of the
blade. The increased tip deflection is very likely the result of the
higher tip loading predicted by the LLFVW model.
Normal blade loading (thrust) Figure 10 shows the
deviation at different azimuthal positions for the yawed case.
Both models predict quite similar behaviour for converged re-
sults. Both normal and tangential loading are increased as the
rotor moves into the upstream region. Similar behaviour is seen
in Figure 11. Tip and blade deviations are again seen to occur
and both models predict higher loading in the sector of = 0
due to the windshear profile. The difference in loading for this
case is an illustration that the blade is constantly in an unsteady
environment during operation under normal environmental con-
ditions, and suggesting that assumptions of steady, converged op-
eration in modelling should be avoided if possible. In the case of
0
2
4
6
8
F
n
[kN
m

1
]
AeroDyn,90
AeroDyn,270
LLFVW,90
LLFVW,270
0
10
20
30
40
50
60
0
0.5
1
Length [m]
F
t
[kN
m

1
]
FIGURE 10: Converged blade loading for yawed operation.
0
2
4
6
8
F
n
[kN
m

1
]
AeroDyn,0
AeroDyn,180
LLFVW,0
LLFVW,180
0
10
20
30
40
50
60
0
0.5
1
Length [m]
F
t
[kN
m

1
]
FIGURE 11: Converged blade loading under atmospheric shear.
extreme linear vertical shear, the iterative procedure of the BEM
model is seen in Figure 12 to produce highly discontinuous re-
sults. The relative differences as predicting in the controlled state
are also indicative of very different integral rotor predictions such
as rotor power, demonstrated earlier in Figure 8.
0
1
2
3
4
F
n
[kN
m

1
]
AeroDyn
AeroDyn,cont
LLFVW
LLFVW,cont
0
10
20
30
40
50
60
0
2
4
6
Length [m]
F
n
[kN
m

1
]
FIGURE 12: Transient blade loading. Top: Extreme vertical lin-
ear shear ( = 180). Bottom: Extreme operating gust.
Tangential blade loading (torque) Similar trends are
seen here when compared to the normal blade loading. In the
case of extreme direction change (or yaw), as illustrated in Fig-
ure 10, the two models predict quite different distributions of
tangential loading. AeroDyn predicts very strong loading near
the blade root as compared to the LLFVW model. The predic-
tions here can again be explained by the shedding near the blade
root of spanwise circulation, which gives rise to strong crossflow
components which reduce loading near the blade root, quite sim-
ilar to the effect near the tip. An analogous reaction has been
observed in the transient loading while yawing. Figure 11 shows
again the difference in loading experienced by the blade under
atmospheric shear. Figure 13 shows that the LLFVW model pro-
duces quite reduced loading as compared to AeroDyn under tran-
sient conditions. The discontinuities predicted by AeroDyn are
again seen here in the case of the transient loading occurring af-
ter the onset of the gust. The data shown here is for the point
of the simulation directly after the blade deflection has reached
its maximum (named earlier inflection point) and oscillates more
freely as the unsteady condition is removed and the rotor returns
to steady loading. In this time period the blade is observed to
oscillate as seen in Figure 9.
0.2
0
0.2
0.4
0.6
F
t
[kN
m

1
]
AeroDyn
AeroDyn,cont
LLFVW
LLFVW,cont
0
10
20
30
40
50
60
0.2
0
0.2
0.4
0.6
Length [m]
F
t
[kN
m

1
]
FIGURE 13: Transient blade loading: Top: Extreme vertical lin-
ear shear ( = 180). Bottom: Extreme operating gust.
Controlled cases: Pitch angle and rotor speed
Many of the phenomena highlighted earlier are illuminated by
inspecting the controller reaction to the loading input of the
LLFVW model. It can be observed in Figure 14 that there is a no-
ticeable lag in the pitch response as compared to AeroDyn. In the
case of the yawed turbine, the reduction in average wind speed
is detected and the pitch automatically reverts to zero position
while the turbine speed is reduced. In the case of extreme verti-
cal shear the LLFVW model predicts a slightly larger increase in
power which causes the controller in this case to compensate by
reducing rotor speed and then returning back. In the case of the
gust the controller heavily feathers the blades out to reduce ro-
tor power, which upon the large reduction in power then pitches
back into position to regain power output. The lag in response
time is seen in all cases and is a consequence of the fact that
aerodynamic disturbances need a finite time to propagate into
the wake. This is not accounted for properly by BEM models
and the difference in predicted response illustrates the effect this
can have on turbine control. Attention is now drawn to the rotor
speed, which is modulated to optimise output from the genera-
tor. In the case of extreme direction change the controller reacts
equivalently by reducing the rotor speed, as can also be observed
in Figure 8. The shifting between controllers is observed in the
linear vertical shear case. The lag in response time is also seen
here in the case of the gust onset, as elaborated upon earlier.
Conclusion and Future Work
It has been shown that under unsteady conditions, the
LLFVW model predicts quite different results as compared to
the BEM model implemented in AeroDyn. It can be seen that the
LLFVW predicts increased tip loading and decreased blade root
loading as compared to the BEM model regardless of operating
AeroDyn+yaw
LLFVW+yaw
AeroDyn+lin shear
LLFVW+lin shear
AeroDyn+gust
LLFVW+gust
0
5
10
15
20
25
30
35
40
11
12
13
Time [s]
Rotor
speed
[RPM
]
FIGURE 14: Top: Blade pitch angle for unsteady cases. Bottom:
Rotor rotational speed.
conditions. The LLFVW model predicts significantly different
behaviour in the period where the rotor returns to the normal op-
eration after an unsteady excitation, where large oscillations are
induced. The difference in blade loading causes increased tip
deflection predictions and a general increase in deflection am-
plitude, both having significant consequences on turbine fatigue
damage. Under controlled conditions the LLFVW model pre-
dicts a delayed controller response as a result of a finite propa-
gation time of disturbances into the turbine wake. The predicted
loading distribution along the blade of the LLFVW has been seen
in every case to be more continuous than that predicted by Aero-
Dyn, having consequences on the distribution of bending mo-
ment predicted along the blade. It is planned to continue on with
the model used in this work to determine the influence such a
model has on prediction of fatigue damages suffered by a turbine
blade during unsteady conditions.
ACKNOWLEDGEMENT
Thanks go to the team at the Intel Developer Zone for their
help in compiling Fortran.
REFERENCES
[1] Glauert, H., 1948. The Elements of Airfoil and Airscrew
Theory. Cambridge University Press.
[2] Snel, H., 1998. Review of the present status of rotor aero-
dynamics. Wind Energy, Spring, pp. 4649.
[3] Leishman, J., 2002. Challenges in modeling the unsteady
aerodynamics of wind turbines.
[4] Jonkmann, J., and Jr., M. L. B. FAST Users Guide. NREL,
August.
[5] Larsen, T. J., and Hansen, A. M., 2007. How 2 HAWC2,
the users manual. Ris National Laboratory. 1597(ver.
3-1)(EN).
[6] GL, D.
Bladed.
https://www.dnvgl.com/
services/bladed-3775.
[7] Rasmussen, F., Hansen, M. H., Thomsen, K., Larsen, T. J.,
Bertagnolio, F., Johansen, J., Madsen, H. A., Bak, C., and
Hansen, A. M., 2003. Present status of aeroelasticity of
wind turbines. Wind Energy, 6, pp. 213228. Ris Na-
tional Laboratory.
[8] Marten, D., 2015. Implementation, optimization and val-
idation of a nonlinear lifting line free vortex wake module
withing the wind turbine simulation code qblade. Proceed-
ings of the ASME Turbo Expo 2015: Turbine Technical
Conference and Exposition.
[9] Jonkmann, J., and Jr., M. L. B., 2006. A comparison of
wind turbine aeroelastic codes used for certification.
[10] Grasso, F., van Garrel, A., and Schepers, J., 2011. De-
velopment and validation of generalized lifting line based
code for wind turbine aerodynamics. In 30th ASME Wind
Energy Symposium, ASME.
[11] J. Katz, A. P. Low Speed Aerodynamics: From Wing Theory
to Panel Methods. Aeronautical and Aerospace Engineer-
ing. McGraw-Hill, 1991.
[12] van Garrel, A., 2003. Development of a Wind Turbine Aero-
dynamics Simulation Module. ECN, August.
[13] Drela, M., and Youngren, H. Xfoil; subsonic airfoil de-
velopment system. http://web.mit.edu/drela/
Public/web/xfoil/.
[14] Moriarty, P., and Hansen, A., 2005. AeroDyn Theory Man-
ual. NREL.
[15] O.A. Bauchau, J. C., 2009. Structural Analysis with Appli-
cation to Aerospace Structures. Springer.
[16] Jonkman, J., 2013. The new modularization framework for
the fast wind turbine cae tool. January.
[17] Jonkmann, J., Butterfield, S., Musial, W., and Scott, G.,
2009. Definition of a 5-mw reference wind turbine for off-
shore system development. Tech. rep., NREL, February.
[18] IEC 614001, 2005.
International Standard: Wind
Turbines Part 1: Design Requirements, Augus

Potrebbero piacerti anche