Superelement approach in fully coupled offshore wind turbine simulation: Influence of the detailed
support structure modelling on simulation results for a 5-MW turbine on a tripod substructure
Fabian R. Vorpahl1 , Michael Strobel1 , Hans-Gerd Busmann1 , Stefan Kleinhansl2
Fraunhofer Institute for Wind Energy and Energy System Technology (IWES)1
Bremerhaven, Germany
A BSTRACT
Fatigue design and certification of offshore wind turbines (OWT) is
performed using aero-servo-hydro-elastic time domain simulations.
So far, the local flexibility of the joints in branched OWT support
structures has been neglected in those simulations. In this work, a
superelement approach, allowing for detailed modeling of local joint
flexibility in the OWT design tool ADCoS-Offshore, is presented.
The influence of this simulation approach, taking into account joint
stiffnesses with the same level of accuracy as detailed finite element
models, is described and interpreted using the example of a realistic
5 MW OWT on a tripod in 45 m of water. Further, recommendations
for modeling of turbines on this type of structure are given.
K EYWORDS
Offshore wind turbine; ADCoS-Offshore; fully coupled simula-
tion; offshore structure; steel support structures; joint flexibility;
substructuring; superelement
I NTRODUCTION
Growing turbine size and steps towards deeper waters and se-
rial production of support structures are key parameters describing
the current offshore wind energy development.
Branched bottom mounted support structures such as tripiles1 , tripods
or jackets2 3 are currently used as prototypes and accepted as promis-
ing solution for large scale future projects.
To allow for cost effective and reliable design, accurate and confidable
simulation of the system offshore wind turbine (OWT) is inevitable.
Especially for fatigue simulation and certification, aero-hydro-servo-
elastic tools are used. A selection of the tools that are currently avail-
1
http://www.bard-offshore.de/prod_fundament-en.php; November 19, 2009
2
http://www.alpha-ventus.de/index.php?id=80; November 19, 2009
3
http://www.beatricewind.co.uk/home/default.asp; November 19, 2009
Paper No. ISOPE-2010 TPC-286 Vorpahl
Aero Dynamik Consult Ingenieurgesellschaft (ADC) 2
Stuttgart, Germany
able is shown by Nichols et al. (2009).
For the purpose of this study, a 5-MW OWT on a tripod support struc-
ture was implemented in ADCoS-Offshore. This is the aero-servo-
hydro-elastic tool used at Fraunhofer IWES. ADCoS-Offshore is a
nonlinear finite element (FE) based tool applying a direct implicit time
domain integration scheme. Details on ADCoS-Offshore are provided
in Kleinhansl et al. (2004) and Vorpahl et al. (2007).
The model in ADCoS-Offshore incorporates deterministic or stochas-
tic wind fields and modified blade-element momentum (BEM) theory
for aerodynamic description of the turbine. Hydrodynamics are de-
scribed using linear or nonlinear regular waves as well as irregular
linear waves combined with Morison’s equation. More details about
ADCoS-Offshore are published in Vorpahl et al. (2009b).
So far, ADCoS-Offshore uses beam elements for support structure
modeling. This approach neglects local joint flexibilities, that were
found to be a critical issue for the following reasons: Joint models im-
plemented with beams and neglecting local flexibilities do not lead to
accurate results and the joints are often design driver for the support
structures due to fatigue loads (cf. De Vries (2008), Vorpahl (2009)
and Vemula (2009)).
Therefore, a superelement approach was implemented in ADCoS-
Offshore as described by Vorpahl et al. (2009a). With this approach,
joint models with the same stiffness properties as detailed FE models
but dramatically reduced number of Degrees of Freedom (DOF) are
included in the fully coupled OWT simulation. The superelement or
substructuring approaches using Guyan (cf. Guyan (1965)) or Craig-
Bampton (cf. Craig and Bampton (1968)) methods are widely used
and described for example in NASTRAN (2004) or ANSYS (2007).
M ODEL D EVELOPMENT
In this work, an OWT with a rated power of 5 MW on a tripod
support structure in 45 m of water is simulated. Three different tripod
models, including joint models with different levels of detail, are
used.
Total number of pages: 8
Turbine model MSL

The so called "NREL Offshore 5-MW baseline wind turbine"

as defined by Jonkman et al. (2009) as a numerical model for code
testing and comparison purposes is used in this study. This turbine
is mainly based on publicly available specifications of the REpower
5M machine4 and completed primarily using data from the Dutch
Offshore Wind Energy Converter (DOWEC) project5 , where real-life
data is not available for confidentiality reasons. Special emphasis was
put on the definition of a lifelike and "typical" machine of this size.
The 5-MW baseline turbine is widely used in the scientific commu-
nity. In the integrated project "UpWind - Integrated Wind Turbine
Design6 ", currently the largest wind energy research project funded
under the European Union’s sixth framework programme, the NREL
machine is used as reference turbine. Several modifications of the
baseline model - like a downscaled turbine (cf. Tarp-Johansen et al.
(2009)), a two-bladed version (cf. Larsen et al. (2007)) or a turbine
with modified controller (cf. Jonkman and Matha (2009)) - are used in Figure 1: Tripod structure without tower. Outer member diameters
various contexts. and mean sea level (MSL) displayed.
The described approach resulted in a conventional three-bladed 5 MW
upwind turbine with variable speed and collective pitch controller. The
rated rotor speed of 12.1 rpm and the rated wind speed of 11.4 m/s are 1. Basic beam model (beam model)
the same as for the REpower 5M and slightly lower than the values for
existing turbines respectively. The blade length is 61.5 m. With a hub 2. Model implemented with beam- and shell elements (shell model)
diameter of 3 m this leads to a rotor diameter of 126 m which is the
same as for the REpower machine. To ensure a sufficient tower clear- 3. Model including beam- and superelements (superelement model)
ance of the blades, the nacelle overhang is 5 m and a tilt and a cone The features of those models are described in more detail in the next
angle of 5 deg. and 2.5 deg. respectively are defined. The nacelle, in- sections, whereas a certain attention is given to the number of DOF
cluding the conventional drive train with its multi-stage gearbox with because the beam model and the superelement model must be suitable
a ratio of 97:1, weights 240 tons. for the computationally expensive fully coupled OWT simulation.
The NREL turbine - or the rotor nacelle assembly (RNA) of the NREL
turbine, as the tower is mentioned as part of the support structure in the Beam model of the tripod
following - is implemented without modifications in ADCoS-Offshore
and simulated in the framework of this study. The tripod for the use in ADCoS-Offshore is defined with stan-
dard FE Euler-Bernoulli beam elements with two nodes and six DOF
each. The conical parts of the structure are set up as stepped members
Support structure properties in the model.
The modeling is mainly based on the findings from the 3rd phase of
The different support structure models used in this study are the OC3 project as described by Nichols et al. (2009) and by Vorpahl
based on the tripod as defined by Garrad Hassan for the Offshore et al. (2009b) but some further modifications were realized. Firstly, the
Code Comparison Collaboration (OC3) within IEA Wind Annex overlapping parts of the tripod members at the joints are excluded from
XXIII7 and described by Nichols et al. (2009). The tripod is - as the the wave load and buoyancy calculation using small supplementary
turbine itself - fictitious but realistic (shown in figure 1). elements as shown in figure 2.
This structure is designed for a water depth of 45 m and consists firstly
of a tapered main column, that is extended over the mean sea level by
a tower featuring a decreasing diameter and wall thickness with in-
creasing height. The total height of the tripod with tower is 132.6 m,
leading to a realistic hub height for a turbine of this size. The support-
ing legs, braces, mud braces and pile sleeves have cylindrical shapes.
The piles are not defined as the tripod is simply cantilevered at each
pile sleeve at the mudline in all models. Figure 2: Simplified joint model with a supplementary element (dot-
The tripod structure includes five joints: The "central joint", having its ted line) excluded from the wave, buoyancy and dead
intersection point at the sea level or at 45 m vertical distance from the weight calculation (right) compared to a basic model (left).
mudline. The "lower central joint" connecting the braces and the main
column, and three identical "pile sleeve joints". Doubling of masses is avoided because the mass of the small sup-
In this study, the main column of the tripod is assumed to be free plementary members as shown in figure 2 (dotted line) was set close
flooded, whereas the legs and braces are not. The dynamic effects of to zero. This is a simplified approach as the real "doubled" volume
water added mass are included in a very simplified way specifying an (buoyancy), surface (wave load calculation) and steel mass (gravity
added mass coefficient of Ca = 1 for all submerged structural members and dynamics) is still roughly estimated, however, the model is signif-
when calculating the inertia loads with Morison’s equation. icantly improved by using the supplementary members. Furthermore,
In the course of this study the following types of tripod models were the tapered main column of the tripod is finely discretized because this
used: was found to have a significant influence on wave load calculation in
the OC3 project.
As it is intended to have a closer look to the local dynamic behavior of
4
http://www.repower.de/index.php?id=237; November 19, 2009 the structure during this study, an influence of the discretisation of the
5
http://www.ecn.nl/units/wind/rd-programme/aerodynamics/projects/dowec/; Novem- tripod tubes on this has to be avoided. The members with the highest
ber 19, 2009 length to diameter ratio for one single FE beam element, that would
6
http://www.upwind.eu/default.aspx; November 19, 2009 therefore be subject of a first step toward finer discretisation of the
7
http://www.ieawind.org/Annex_XXIII.html; November 19, 2009 model, are used to define the mud braces.
As the dynamic behavior of a structure is determined by its eigen-
states, the natural frequencies associated with the first bending mode
of the mud braces are calculated for different models. Figure 3 shows
the differences of those natural frequencies for models with a number
of single elements between nb, 1 = 2 and nb, max = 100 for the mud
brace.

0.7
0.6
0.5
Difference [%]

0.4
0.3
0.2
Figure 4: Tripod (left) with load output position (dotted circle) and
0.1 shell central joint (right).
0
0 10 20 30 40 50 The model described herein has a number of DOF of approximately
Number beam elements [-] nshell = 50, 000

Figure 3: Natural frequency of mud brace for different discretiza- Tripod modeled with beam- and superelements
tions.
The superelement model is based on the shell model where all
the joints -the shell substructures- are replaced by superelements
As expected, the differences decrease with increasing number of el- using a Guyan reduction procedure. The rest of the beam structure is
ements. However, even for the simplest model the difference to the unchanged compared to the other models.
most detailed model - which is assumed to be "correct" - is clearly un- The number of DOF of this structure is approximately nsuper = 400,
der 1 %. As the computational effort for the time domain simulations a value even significantly below its counterpart for the basic beam
should be kept in mind, such small errors are accepted and the model model. The reason for this is as follows: The DOF of each superele-
is not refined. The number of DOF of the resulting beam model is ment are only the DOF of its master nodes. As mentioned before,
approximately nbeam = 600. several supplementary members have to be included in the basic beam
To conclude, this model is as realistic as achievable with standard model to allow for more realistic load simulation ("overlapping mem-
beam elements and reasonable effort. Therefore, it is suitable to in- bers") at the joints. Furthermore, changing properties of the tubes
vestigate the differences between beam models and more sophisticated such as diameters Di or thicknesses ti are modeled with several beam
structural models. elements and at the intersection points of tubes a node has to be de-
fined which leads to more elements as well. The parts of the struc-
ture modeled with those supplementary beam elements become part
Tripod modeled with shell- and beam elements of the superelements and therefore those DOF are not in the superele-
ment model. Figure 5 shows the beam model of a pile sleeve joint
This model is based on the beam model, however the beam ele- with fourteen nodes (black dots) and a resulting number of DOF of
ments representing all joints are replaced by shell joints. A standard nbeam,1 = 84 (left) and the corresponding superelement with six mas-
four-noded shell element, well suited for the modeling of warped ter nodes (black dots) and therefore nsuper,1 = 36 DOF (right).
shell structures with moderate thickness and six DOF per node is
used.
Furthermore so-called "master nodes" are defined to connect the de-
tailed sub-models to the residual structure. Being located in the center-
lines of the tubular chord and brace members those nodes are rigidly
connected to the outgoing beam element and rigidly connected to the
detailed model via radially arranged rigid link or stiff beam connec-
tions. Figure 4 shows the tripod model with shell joints and the central
joint in detail.
In defining the shell structures, the appropriate size and shape of the el-
ements have to be guaranteed. This topic has not been investigated ex-
tensively, as very details of the joints are not subject of this study and
strong influences on the more global phenomena investigated herein
are not expected. However the static and dynamic plausibility of the
model is checked.
The stiff connections between the shell and beam members suppress
the ovalisation of the circular cross section of the joints in the proxim- Figure 5: Pile sleeve joint modeled with beams (left) and as superele-
ity of the connection. This mainly influences the joints if the distances ment (right). Nodes are shown as black dots in both models.
to the master nodes are too short and might be investigated using para-
metric studies. However, this is not expected to be of major impor-
tance for this study, as the influence of joint models on more global
phenomena is investigated and not the joints themselves at a very high
level of detail.

S IMULATION RESULTS
In the following exemplary results of simulations with the three
models are presented. All relative values in comparisons are related
to the model expected to be the most accurate. For comparisons
between shell models and other models, the values are related to
the results from the shell models and for comparisons between
superelement- and beam models all values are related to the results
from the superelement models respectively.
Preliminary study on support structure models
In an initial study, the dead weights and eigenstates of the beam-, the
shell- and the superelement support structure models are checked.
A model of the rotor nacelle assembly (RNA) is not used for the
following reasons: For the mass comparison, the RNA is just not
relevant, as the mass of the RNA would be the same in all three
models. Also, the shell model of the support structure is part of the
investigation and it is currently not possible to use shell elements in
the aero-elastic code ADCoS-Offshore. Therefore, a model including
a detailed RNA and a shell support structure is not available for
comparison of eigenstates.
Table 1 shows the total steel masses of the structures and the mass
difference between the beam and the shell model.
Table 1: Dead weight of all three support structure models and differ-
ence between shell- and beam model.
beam shell super
1081.96 t 1113.25 t 1113.25 t
As described by theory, the mass of the shell model - the static load
vector - is exactly the same as the mass of the superelement model.
The difference between those two types of models and the initial beam
model is only 3 %. In this case, differences were expected, as the beam
model does not represent the steel surfaces at the intersection of mem-
bers at the joints correctly as previously mentioned. The relatively
small difference shows the good quality of the beam model with its
massless members included at those intersections.
In Figure 6, the lowest 15 natural frequencies of the three models are
displayed.
4
beam
3.5
shell
3 super
Frequency [Hz]
2.5
2
1.5
1
0.5
0
1 2 3 4 5 6 7 8 9 10
Number eigenstate [-]
Figure 6: First 15 natural frequencies of beam-, shell- and superele-
ment support structure models.
Two results are recognized immediately: All natural frequencies are
exactly the same for the shell and the superelement model and there
are significant differences between some of the natural frequencies
determined with the beam- and the two other models. The maximum
difference of a pair of natural frequencies is approximately 50 % from
the 3rd to the 6th natural frequency. Even the lowest natural frequen-
cies, associated with the first global bending modes of the structure,
show a difference of approximately 6 %.
The preliminary study leads to the following results:
1. The superelement model leads exactly to the same results as the
most accurate shell model, therefore superelement models are
very suitable for modeling this type of structure.
2. There are significant differences between the results obtained
with the superelement model and with the beam model, which
means that a significant increase of simulation accuracy is ex-
pected using superelements compared to the state-of-the-art ap-
proach.
Concerning point 1 it should be noticed, that the accuracy of the shell
model - with a number of DOF around two orders of magnitude higher
than in the other models - is generally not questioned for the simula-
tion of the investigated type of steel structure as long as the modeling
is realized carefully. Furthermore, there are significant differences,
especially concerning mass matrix description (cf. Vorpahl (2009)),
but those differences have no influence on the relevant parameters for
a dynamic analysis of the degree of detail performed in aero-elastic
codes.
In this section, support structure models are analyzed. The influence
of the significant differences between beam- and superelement support
structure models (cf. point 2) on the OWT models - including the RNA
- is investigated below.
Full system eigenanalysis
difference
Two models are defined for full system eigenanalysis: The baseline
3% turbine RNA with beam element support structure and with superele-
ment support structure respectively. The first 15 of the eigenmodes
and natural frequencies calculated with the two models are analyzed
and compared.
Figure 7 shows a front view of the 10th eigenmode of the turbine as
an exemplary result. The eigenmode of the model with beam structure
(left) is related to a frequency of 1.23 Hz and the eigenmode of the
turbine with superelement structure (right) is related to a frequency of
1.08 Hz.
11 12 13 14 15
Figure 7: 10th eigenmode at 1.23 Hz (beam model, left) and at
1.08 Hz (superelement model, right).
In the case of the beam model, this is a coupled mode with two main
portions: The 2nd global bending of the structure combined with a
blade mode, which is the symmetric 1st edgewise bending. In the
superelement model, the blade mode - here an asymmetric 1st edge-
wise bending - is much more dominant. Other portions are not clearly
visible. Furthermore, the related frequency is much lower for the su-
perelement model.
A coupled eigenmode of the superelement model comparable to the
mode of the beam model shown here is not found within the first 15
modes that are analyzed.
The lower frequency calculated with the superelement model for the
natural frequency with the same number of order was expected due to
local joint flexibilities included in this model, leading to more compli-
ant joints.
Furthermore the coupled eigenmodes with the same number of order
are completely different. This was expected as well.
As an example, weaker joints lead to much lower natural frequencies
associated with local bending modes of single members like the mud
braces. Global modes with smaller portions of joint deflection are not
influenced as much by the modified joint modeling. As a result, some
of the frequencies connected to local modes that have been associ-
ated with a higher frequency than the considered global mode in the
beam model, get a frequency shift towards lower values and are found
at lower frequency values than the global mode in the superelement
model. This means that single modes "change places" if listed with
increasing frequencies.
As the model modification influences the various eigenstates differ-
ently, other eigenvalues become associated with frequencies lying
close together. This results in coupled modes in the superelement
model, that have not been described by the beam model and vice versa.
Additionally, the superelement model may lead to coupled modes that
combine the same, or nearly the same, single modes, but in other rel-
ative portions.
To conclude, there are significant differences between the results cal-
culated with the two models. However, the comprehensive investi-
gation of all the introduced effects is beyond the scope of this paper
and the practical relevance of some of the details may be questioned.
Therefore, the influence of the support structure modeling on the val-
ues of the natural frequencies, that are considered most important for
the highly dynamic system OWT with its various excitations in differ-
ent frequency ranges, is described in the following.
Figure 8 shows the differences between the first 15 natural frequen-
cies of the full system including the beam support structure and the
superelement structure respectively.
2.5
Frequency [Hz]
2 1
beam
1.5 super
1
0.5
0 0.4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number eigenstate [-]
25
Difference [%]
20
15
10
5 (1p, 3p, 6p) due to turbine operation (boxes). Full system
0 natural frequencies for the beam model (dotted lines) and
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number eigenstate [-]
Figure 8: First 15 full system natural frequencies for both models
(top) and differences between those frequencies (bottom).
Compared to the results obtained with the support structure models
without RNA in the previous section, some changes are obvious in
both figures. Mainly due to the influence of the RNA on global sup-
port structure frequencies and due to low blade bending frequencies,
the first 15 natural frequencies are generally lower for the full sys-
tem. Furthermore, the differences between some of the frequencies
obtained with the two models decrease as well which is due to the na-
celle mass and inertia and to the influence from the blades which are
the same in both models.
For the 1st and the 2nd natural frequency, the differences are about
5 %. Those frequencies are associated with the first global bending
mode of the structure, where a strong influence of the modified joint
modeling is not expected. From this point of view, those differences
are relatively high.
The difference between the 3rd natural frequency in both models is
about 7 %. This mode mainly couples a first flapwise blade bending
mode with a nacelle yaw motion. This nacelle yawing - in other words
the global torsion of the structure - is likely to be influenced by the
modeling of the central joint.
The 4th to the 6th natural frequencies show only minor offsets and the
difference for the 6th natural frequency is under 1.5 %. These frequen-
cies are mainly associated with first flapwise blade modes, that are
obviously the same in both models.
The 7th and the 8th frequencies show slightly higher differences. Here,
first edgewise bending modes of the blades couple with local support
structure modes and - for the superelement model - with small portions
of the second global bending mode of the structure.
The seven highest natural frequencies described here show frequency
offsets between 12 % and 23 % and the portions of the coupled modes
are hardly to distinguish. However, the differences are remarkable
and the large offsets might result firstly from large portions of local
support structure modes and secondly from global support structure
modes with strong influence of the joints.
The next step is to analyze the potential effect of the shifted natural
frequencies. To simplify this problem and to allow for a general view,
the most important excitations resulting from the operating turbine
are taken into account: The rotational frequency range (1p), causing
cyclic excitations due to e.g. rotor imbalances and the blade passing
frequency range (3p) causing cyclic excitations due to the disturbed
wind flow in front of the tower experienced by the passing blades.
Figure 9 shows the 1p range, the 3p range and the 1st harmonic of
the 3p range (6p) for the NREL turbine (boxes). Furthermore, the
lowest nine natural frequencies for the beam model (dotted lines) and
the superelement model (continous lines) are indicated.
3rd super 3rd beam 9th super 9th beam
0.8
1P 3P 6P
0.6
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Frequency [Hz]
Figure 9: Frequency ranges suspect to significant dynamic excitations
the superelement model (continous lines).
Besides many details that are not of primary interest for this study, the
following results are found: Firstly, the frequencies associated with
the first global bending modes of the support structure are shifted by
approximately 5 % towards lower frequencies as mentioned before.
Those frequencies are among the most important design parameters
for OWT support structures as it is common practice do design struc-
tures with a first natural frequency between the 1p and the 3p range
(soft-stiff design) of the turbine. The allowable frequency gap is rel-
atively small, especially as a distance of about 10 % between natural
frequency and excitation range has usually to be provided for safety Figure 10 shows the fore-aft tower top displacement (displacement
reasons (not included in figure 9). For the NREL turbine, the more along the wind and wave direction) over the first 50 s for the beam-
detailed modeling would lead to higher theoretical safety margins, as and for the superelement model.
Series.Run = LC_1_2_1
the frequencies are shifted towards the dead center between the 1p and Output Channel = TTDspFA 15
the 3p range. 0.55
Secondly, the 3rd full system natural frequency is shifted into the upper Company Name = beam beam super

Displacement [m]
0.5 Name =
Simulator ######## 201209_2 ########
3p frequency range due to the more realistic support structure model- 0.45
1 Time TTDspFA #BEZUG! #BEZUG! Time TTDspFA #BEZUG!
ing. The difference between the 3rd natural frequency calculated with 2
0.4
3
(sec)
0.0000
(m) #BEZUG! #BEZUG! (sec)
0.3854 #BEZUG! #BEZUG! 0.0000
(m) #BEZUG!
0.4256 #BEZUG!
the beam model and the superelement model is about 7 % as men- 0.35
4 0.0500 0.3854 #BEZUG! #BEZUG! 0.0500 0.4256 #BEZUG!
tioned above. The excitations due to disturbed wind flow around the 0.3
5 0.1000 beam model
0.3855 #BEZUG! #BEZUG! 0.1000 0.4257 #BEZUG!
tower in the 3p range are considered to bring more energy into the 6
0.25 0.1500 superelement
0.3856 #BEZUG! #BEZUG! 0.1500model0.4259 #BEZUG!
7 0.2000 0.3859 #BEZUG! #BEZUG! 0.2000 0.4263 #BEZUG!
system than its harmonics and even more than the excitations in the 0.2
8 0.2500 0.3863 #BEZUG! #BEZUG! 0.2500 0.4268 #BEZUG!
1p range (cf. Fischer and Kuehn (2009)). Furthermore, the natural 0 9 0.3000
10 0.3871 #BEZUG! #BEZUG!
20 30 0.3000 0.4277 #BEZUG!
40 0.4289 50
frequency is found near the upper bound of the 3p range which is as- 10 0.3500 0.3882 #BEZUG! #BEZUG!
Time [s] 0.3500 #BEZUG!
11 0.4000 0.3897 #BEZUG! #BEZUG! 0.4000 0.4305 #BEZUG!
sociated with rated rotor speed. It is obvious, that the turbine operates 12 0.4500 0.3916 #BEZUG! #BEZUG! 0.4500 0.4325 #BEZUG!
much more often at rated rotor speed than at lower speeds, as this is Figure 10: Tower top displacement over time for baseline turbine with
13 0.5000 0.3939 #BEZUG! #BEZUG! 0.5000 0.4350 #BEZUG!
the operational speed at all wind speeds above rated wind speed. The 14 0.5500 0.3967 #BEZUG! #BEZUG! 0.5500 0.4378 #BEZUG!
beam support structure (dashed line) and superelement
15 0.6000 0.3998 #BEZUG! #BEZUG! 0.6000 0.4411 #BEZUG!
maximum rotor thrust occurring at rated wind speed makes a strong support structure (continuous line) respectively.
16 0.6500 0.4033 #BEZUG! #BEZUG! 0.6500 0.4447 #BEZUG!
dynamic excitation at this speed even more probable. 17 0.7000 0.4081 #BEZUG! #BEZUG! 0.7000 0.4495 #BEZUG!
Thirdly, the 9th natural frequency is shifted to the upper 6p range. This 18 0.7500 0.4125 #BEZUG! #BEZUG!
The significant difference of the two results is evident; the average
19 0.8000 0.4174 #BEZUG! #BEZUG!
0.7500
0.8000
0.4541 #BEZUG!
0.4591 #BEZUG!
is not too critical, other natural frequencies are found in the 6p range 20 0.8500 0.4227 #BEZUG! #BEZUG!
displacement difference is approximately 10 %. As expected, the su- 0.8500 0.4645 #BEZUG!
as well. 21 0.9000 0.4283 #BEZUG! #BEZUG!
perelement support structure is more compliant and therefore leads to
0.9000 0.4702 #BEZUG!
22 0.9500 0.4339 #BEZUG! #BEZUG! 0.9500 0.4760 #BEZUG!
All in all, the support structure modeling with superelements leads to larger displacements. The paths of the curves are similar, but not ex-
23 1.0000 0.4395 #BEZUG! #BEZUG! 1.0000 0.4818 #BEZUG!
several changes in comparison to the basic beam model concerning the actly the same. As this structural answer is a result of the external
24 1.0500 0.4448 #BEZUG! #BEZUG! 1.0500 0.4873 #BEZUG!
full system eigenstates. The number of order of the natural frequen- 25 1.1000 0.4499 #BEZUG! #BEZUG!
loads (that are the same in both cases) and the characteristics of the 1.1000 0.4926 #BEZUG!
26 1.1500 0.4546 #BEZUG! #BEZUG! 1.1500 0.4976 #BEZUG!
cies varies, as local modes are influenced differently. This leads to structure (that are not the same) this was expected as well. As it was
27 1.2000 0.4590 #BEZUG! #BEZUG! 1.2000 0.5022 #BEZUG!
coupled modes that do not occur with the basic model. Furthermore, shown in this paper before, the natural frequencies associated with the
28 1.2500 0.4630 #BEZUG! #BEZUG! 1.2500 0.5064 #BEZUG!
the relative portions of single modes in one coupled mode may vary. 29 1.3000 0.4667 #BEZUG! #BEZUG!
first tower bending modes, that are among the most important struc- 1.3000 0.5105 #BEZUG!
30 1.3500 0.4701 #BEZUG! #BEZUG! 1.3500 0.5143 #BEZUG!
As expected, the RNA - which is the same in both models - leads tural properties, differ about 5 %.
31 1.4000 0.4732 #BEZUG! #BEZUG! 1.4000 0.5178 #BEZUG!
to lower frequencies associated with global modes due to its signifi- The following figure (cf. figure 11) shows the bending moment around
32 1.4500 0.4760 #BEZUG! #BEZUG! 1.4500 0.5210 #BEZUG!
cant mass at the tower top. It leads to smaller differences between 33 1.5000 0.4785 #BEZUG! #BEZUG!
the local z-axis, in the upwind leg near the pile sleeve joint over a 1.5000 0.5239 #BEZUG!
the two models, as the frequencies associated with global modes and 34 1.5500 0.4806 #BEZUG! #BEZUG! 1.5500 0.5265 #BEZUG!
fraction of 50 s of the load case for both models. The load output
35 1.6000 0.4824 #BEZUG! #BEZUG! 1.6000 0.5287 #BEZUG!
blade modes get closer together and are nearly the same respectively. position is visualized in figure 4 (dotted circle). The local z-axis lies
36 1.6500 0.4837 #BEZUG! #BEZUG! 1.6500 0.5305 #BEZUG!
However, there are differences of up to 23 % for the first 15 natural Series.Run = 37 1.7000
LC_1_2_1 0.4846 #BEZUG! #BEZUG!
perpendicular to the wind direction in the horizontal plane. 1.7000 0.5319 #BEZUG!
frequencies. The frequency differences of about 5 % associated with Output Channel = 38 1.7500
M152N148 0.4850
47 #BEZUG! #BEZUG! 1.7500 0.5329 #BEZUG!
39 1.8000 0.4850 #BEZUG! #BEZUG! 1.8000 0.5334 #BEZUG!
the first global bending modes are not negligible especially if the im-
2.5 Name = 40 1.8500 beam 0.4845 #BEZUG! beam
#BEZUG! 1.8500 super
0.5335 #BEZUG!
Bending moment [MNm]

Company
portance for support structure design is taken into account. The 3rd Simulator Name = 41 1.9000 ########
0.4836 #BEZUG! 201209_2
#BEZUG! 1.9000 ########
0.5331 #BEZUG!
42
1 1.9500 M152N148
Time 0.4822 #BEZUG!
#BEZUG! #BEZUG! 1.9500 M152N148
#BEZUG! Time 0.5323 #BEZUG!
#BEZUG!
natural frequency is shifted into the upper 3p range, which is assumed 2 43
2 2.0000
(sec) (kN)0.4804 #BEZUG! #BEZUG! 2.0000
#BEZUG! #BEZUG! (sec) (kN)0.5311 #BEZUG!
#BEZUG!
to be critical and the most important result. Furthermore, the 9th nat- 1.5
44
3 2.0500 0.4784 #BEZUG!
0.0000 ######## #BEZUG! #BEZUG!
#BEZUG! 2.0500 0.5297 #BEZUG!
0.0000 ######## #BEZUG!
45
4 2.1000 0.4762 #BEZUG!
0.0500 ######## #BEZUG! #BEZUG!
#BEZUG! 2.1000 0.5281 #BEZUG!
0.0500 ######## #BEZUG!
ural frequency is shifted into the upper 6p range, a fact that should be 46 2.1500 0.4744 #BEZUG!
5 0.1000 ######## #BEZUG! #BEZUG!
#BEZUG! 2.1500 0.5268 #BEZUG!
0.1000 ######## #BEZUG!
treated with caution as well. 1 47
6 2.2000 0.4721 #BEZUG! #BEZUG!
0.1500 ######## #BEZUG! #BEZUG! 2.2000 0.5250 #BEZUG!
0.1500 ######## #BEZUG!
48
7 2.2500 0.4696 #BEZUG!
0.2000 ######## #BEZUG! #BEZUG!
#BEZUG! 2.2500 0.5232 #BEZUG!
0.2000 ######## #BEZUG!
0.5 49
8 2.3000
0.2500 ########
beam
0.4671 model #BEZUG! 2.3000
#BEZUG!
#BEZUG! #BEZUG! 0.5212 #BEZUG!
0.2500 ######## #BEZUG!
50
9 2.3500 superelement
0.4645 #BEZUG! model
#BEZUG!
0.3000 ######## #BEZUG! #BEZUG! 2.3500 0.5191 #BEZUG!
0.3000 ######## #BEZUG!
Time domain simulation 0 10 0.3500 ######## #BEZUG! #BEZUG! 0.3500 ######## #BEZUG!
11 0.4000 ######## #BEZUG! #BEZUG! 0.4000 ######## #BEZUG!
50 60 70 80 90 100
This section presents the results from one exemplary time do- 12 0.4500 ######## #BEZUG! #BEZUG!
Time [s] 0.4500 ######## #BEZUG!
13 0.5000 ######## #BEZUG! #BEZUG! 0.5000 ######## #BEZUG!
main simulation load case that was run with the baseline turbine RNA 14 0.5500 ######## #BEZUG! #BEZUG! 0.5500 ######## #BEZUG!
including the beam support structure and the superelement support 15 0.6000 ######## #BEZUG! #BEZUG!
Figure 11: Bending moments in upwind leg over 50 s for beam 0.6000 ######## #BEZUG!
structure. 16 0.6500 ######## #BEZUG! #BEZUG!
(black) and superelement model (light grey).
0.6500 ######## #BEZUG!
17 0.7000 ######## #BEZUG! #BEZUG! 0.7000 ######## #BEZUG!
The selected load case is an exemplary power production design load 18 0.7500 ######## #BEZUG! #BEZUG! 0.7500 ######## #BEZUG!
case (DLC) as defined in IEC (2005), which is based on the descrip- 19 0.8000 ######## #BEZUG! #BEZUG! 0.8000 ######## #BEZUG!
tions in Fischer et al. (2009) for a Dutch North Sea Site. One could The average bending moment resulting from the simulation with the
20 0.8500 ######## #BEZUG! #BEZUG! 0.8500 ######## #BEZUG!
speak of a typical fatigue load case for North Sea conditions as the super element model is slightly higher than the moment calculated
21 0.9000 ######## #BEZUG! #BEZUG! 0.9000 ######## #BEZUG!
22 0.9500 ######## #BEZUG! #BEZUG! 0.9500 ######## #BEZUG!
occurrence of its metocean parameters, that are shown in table 2, is with the beam model. Furthermore, the amplitudes of the fluctuating
23 1.0000 ######## #BEZUG! #BEZUG! 1.0000 ######## #BEZUG!
highly significant. bending moments are much bigger for the beam model.
24 1.0500 ######## #BEZUG! #BEZUG! 1.0500 ######## #BEZUG!
In table 3 the damage equivalent bending moments (damage equiva-
25 1.1000 ######## #BEZUG! #BEZUG! 1.1000 ######## #BEZUG!
26 1.1500 ######## #BEZUG! #BEZUG! 1.1500 ######## #BEZUG! 8
Table 2: Wind (average wind speed, turbulence intensity, spectrum) lent load: DEL) derived from the ten-minute time series for N = 2·10
27 1.2000 ######## #BEZUG! #BEZUG! 1.2000 ######## #BEZUG!
and wave (significant wave height, peak period, spectrum) load cycles and different Wöhler material exponents (m) are shown to
28 1.2500 ######## #BEZUG! #BEZUG! 1.2500 ######## #BEZUG!
29 1.3000 ######## #BEZUG! #BEZUG!
point out the impact of the higher load amplitudes. 1.3000 ######## #BEZUG!
parameters for the exemplary load case. 30 1.3500 ######## #BEZUG! #BEZUG! 1.3500 ######## #BEZUG!
The differences in the damage equivalent loads for this load case and
31 1.4000 ######## #BEZUG! #BEZUG! 1.4000 ######## #BEZUG!
load output are very significant for all Wöhler material exponents. For
32 1.4500 ######## #BEZUG! #BEZUG! 1.4500 ######## #BEZUG!

Wind: Vave = 11 m/s Ti = 15 % (Kaimal) m = 3, the value is even nearly doubled.

33 1.5000 ######## #BEZUG! #BEZUG! 1.5000 ######## #BEZUG!
34 1.5500 ######## #BEZUG! #BEZUG! 1.5500 ######## #BEZUG!
Summing up, it is obvious that the stiffness differences in the two
35 1.6000 ######## #BEZUG! #BEZUG! 1.6000 ######## #BEZUG!
Waves: Hs = 1.6 m Tp = 5.8 s (JONSWAP) support structure models (beam model and superelement model) even
36 1.6500 ######## #BEZUG! #BEZUG! 1.6500 ######## #BEZUG!
lead to global effects as it is shown for the tower top displacement
37 1.7000 ######## #BEZUG! #BEZUG! 1.7000 ######## #BEZUG!
38 1.7500 ######## #BEZUG! #BEZUG! 1.7500 ######## #BEZUG!
of the baseline turbine. The dispacement is generally larger for the
39 1.8000 ######## #BEZUG! #BEZUG! 1.8000 ######## #BEZUG!
turbine on the more compliant superelement model. Furthermore, an
40 1.8500 ######## #BEZUG! #BEZUG! 1.8500 ######## #BEZUG!
The simulation time is 600 s. There is no wind and wave, or turbine exemplary load output shows the significance of the differing results.
41 1.9000 ######## #BEZUG! #BEZUG! 1.9000 ######## #BEZUG!
42 1.9500 ######## #BEZUG! #BEZUG! 1.9500 ######## #BEZUG!
yaw misalignment. Wind- and wave directions are parallel to one of The time series shows that the magnitude of the bending moments in
43 2.0000 ######## #BEZUG! #BEZUG! 2.0000 ######## #BEZUG!
the tripod legs, called the "upwind leg" in the following. the upwind leg near the pile sleeve joint is comparable, whereas the
44 2.0500 ######## #BEZUG! #BEZUG! 2.0500 ######## #BEZUG!
45 2.1000 ######## #BEZUG! #BEZUG! 2.1000 ######## #BEZUG!
46 2.1500 ######## #BEZUG! #BEZUG! 2.1500 ######## #BEZUG!
47 2.2000 ######## #BEZUG! #BEZUG! 2.2000 ######## #BEZUG!
48 2.2500 ######## #BEZUG! #BEZUG! 2.2500 ######## #BEZUG!
49 2.3000 ######## #BEZUG! #BEZUG! 2.3000 ######## #BEZUG!
50 2.3500 ######## #BEZUG! #BEZUG! 2.3500 ######## #BEZUG!
Table 3: Damage equivalent bending moments in upwind leg for
beam- and superelement model and several Wöhler material
exponents.
m [-] DEL super [KNm] DEL beam [KNm]
3 7.6 14.4
4 36.3 45.3
5 57.8 91.9
values for the superelement model are slightly higher. But the higher
load amplitudes for the beam model results are much more significant.
The damage equivalent bending moment, which is derived from the
exemplary load case, is significantly higher for the beam model.
C ONCLUSION AND OUTLOOK
In this paper, an aero-elastic model of a 5MW OWT on three
different types of tripod support structure models is investigated. A
state-of-the-art beam model, a detailed shell model and a superele-
ment model derived from the shell model.
The beam model used here is already a good representation of the
structure, its details are based on findings from the Offshore Code
Comparison Collaboration. The shell model leads to a much more
accurate representation of the critical joints combined with the draw-
back of several thousands of supplementary DOF that would lead to
unacceptable computational effort for OWT load calculations. Using
the superelement model, the stiffness representation (stiffness matrix)
of the joints is as accurate as for the shell model whereas the dynamic
representation (mass matrix) is by far accurate enough to meet the re-
quirements of fully coupled OWT simulation. The number of DOF -
and therefore the computational effort - is reduced, even compared to
the beam model. Preliminary model checks show the expected results
concerning masses and eigenstates of the support structure models.
A full system eigenanalysis shows that the superelement model leads
to several changes compared to the beam model. This concerns
the number of order of eigenmodes, natural frequencies and coupled
eigenstates. Even the 1st global support structure bending modes are
affected. Furthermore, the 3rd and the 9th natural frequencies are
shifted into the frequency ranges that should be avoided in OWT de-
sign due to excessive dynamic loads (3P, 6P). A design based on the
beam model would therefore probably not be conservative in terms of
global dynamic excitations.
An exemplary time domain fatigue load case shows, that there are sig-
nificant differences in the results due to the different support structure
modeling approaches. On the one hand, the structure as a whole be-
comes more compliant due to the superelements. This leads to higher
displacements of the turbine for the presented load case, a fact that
must be kept in mind as larger displacements lead to larger inertia
loads and therefore one might underestimate loads in some cases with
the beam model. On the other hand, a more compliant joint reduces
the load amplitudes and therefore the DEL in the member near the
joint significantly. This is an important outcome as design driving
joints may be dimensioned based on the loads gained from aero-elastic
modeling at those positions following the relevant standards. In such
cases, the more detailed superelement approach would allow for much
less conserative fatigue design of the joints.
Despite these results, showing the strong influence of the superelement
joint modeling compared to the state-of-the-art beam model, it must
be noted, that only exemplary load cases have been calculated so far.
The next steps of work are the definition and simulation of a set of load
cases based on the relevant guidelines and standards, the identification
of critical load outputs and the quantification of the effects described
in this paper on, for instance, lifetime DELs.
However, it is a main result of this study, that the superelement mod-
eling leads to significantly differing results in terms of structural de-
flections and member loads for an OWT mounted on the investigated
type of tripod support structure. The enhanced joint modeling has the
potential to lead to more reliable structures with even lower safety
margins in future OWT design processes.
ACKNOWLEDGEMENTS
This work was funded by the German Federal Ministry for the
Environment, Nature Conservation and Nuclear Safety in the
framework of the "Hook Up2 Wind" project at Fraunhofer IWES.
Additional thanks goes to the Commission of the European Commu-
nities for providing Fraunhofer IWES access to the work groups of
the project "UpWind - Integrated Wind Turbine Design" and to Olaf
Lieser from ADC for the outstanding collaboration.
References
ANSYS (2007). Release 10 Documentation for ANSYS. SAS IP Inc.
Craig, R. R. and Bampton, M. C. C. (1968). Coupling of substructures
for dynamic analysis. In AIAA Journal, volume 6.
De Vries, W. (2008). Soft-stiff bottom mounted support structures -
analysis of monopile and multi-member support structures for off-
shore wind turbines. Upwind deliverable D4.3.2 (WP4: Offshore
foundations and support structures), Delft University of Technol-
ogy. Restricted to project members.
Fischer, T., de Vries, W., and Schmidt, B. (2009). Design basis -
upwind deep water site. Upwind deliverable (WP4: Offshore foun-
dations and support structures), Endowed Chair of Wind Energy
(SWE) at the Institute of Aircraft Design Universität Stuttgart.
Fischer, T. and Kuehn, M. (2009). Site sensitive support structure and
machine design for offshore wind farms. In European Wind Energy
Conference (EWEC). Endowed Chair of Wind Energy (SWE) at the
Institute of Aircraft Design, Universität Stuttgart.
Guyan, R. J. (1965). Reduction of stiffness and mass matrices. In
AIAA Journal, volume 3.
IEC (2005). Wind turbines - Part 1: Design requirements. IEC 61400-
1, 3.0 edition.
Jonkman, J., Butterfield, S., Musial, W., and Scott, G. (2009). Defini-
tion of a 5-MW Reference Wind Turbine for Offshore System De-
velopment. Technical Report NREL/TP-500-38060, National Re-
newable Energy Laboratory (NREL).
Jonkman, J. and Matha, D. (2009). A quantitative comparison of
the responses of three floating platforms. In European Offshore
Wind Conference (EOW). National Renewable Energy Laboratory
(NREL).
Kleinhansl, S., Mayer, M., and Mangold, A. (2004). ADCoS - A Non-
linear Aeroelastic Code for the Complete Dynamic Simulation of
Offshore- Structures and Lattice-Towers. In Deutsche Wind Energie
Konferenz (DEWEK). Aero Dynamik Consult Ingenieurgesellschaft
m.b.H.
Larsen, T. J., Madsen, H. A., Thomsen, K., and Rasmussen, F. (2007).
Reduction of teeter angle excursions for a two-bladed downwind
rotor using cyclic pitch control. In European Wind Energy Con-
ference (EWEC). Risø National Laboratory Technical University of
Denmark.
NASTRAN (2004). MSC.NASTRAN. Advanced Dynamics Analysis
User’s Guide. MacNeal-Schwendler Corporation.
Nichols, J., Camp, T., Jonkman, J., Butterfield, S., Larsen, T., Hansen,
A., Azcona, J., Martinez, A., Munduate, X., Vorpahl, F., Kleinhansl,
S., Kohlmeier, M., Kossel, T., Böker, C., and Kaufer, D. (2009).
Offshore Code Comparison Collaboration within IEA Wind Annex
XXIII: Phase III Results Regarding Tripod Support Structure Mod-
eling. In 47th AIAA Aerospace Sciences Meeting. Garrad Hassan &
Partners Limited (GH).
Tarp-Johansen, N. J., Andersen, L., Christensen, E. D., Mørch, C.,
Kallesøe, B., and Frandsen, S. (2009). Comparing sources of damp-
ing of cross-wind motion. In European Offshore Wind Conference
(EOW). DONG Energy.
Vemula, N. K. (2009). Jacket foundation support structure. Upwind
deliverable D4.2.6 (WP4: Offshore foundations and support struc-
tures), Rambøll Wind Energy. Restricted to project members.
Vorpahl, F. R. (2009). Implementation of a superelement approach in
a design tool for offshore wind turbines with braced support struc-
tures. Upwind deliverable D4.3.4 (WP4: Offshore foundations and
support structures), Fraunhofer Institute for Wind Energy and En-
ergy System Technology (IWES). Restricted to project members.
Vorpahl, F. R., Blunk, M., Wingerde, A. V., Busmann, H.-G., and
Kleinhansl, S. (2009a). Implementation of a superelement approach
in a design tool for offshore wind turbines with arbitrary support
structures. In European Offshore Wind Conference (EOW). Fraun-
hofer Institute for Wind Energy and Energy System Technology
(IWES).
Vorpahl, F. R., Huhn, H., Busmann, H.-G., and Kleinhansl, S. (2007).
A Flexible Aero-elastic Simulation Approach for Offshore Wind
Turbines. In European Offshore Wind Conference (EOW). Fraun-
hofer Center for Wind Energy and Maritime Engineering.
Vorpahl, F. R., Wingerde, A. V., Blunk, M., Busmann, H.-G., Klein-
hansl, S., Kossel, T., Kohlmeier, M., Böker, C., Kaufer, D., Azcona,
J., Martinez, A., and Munduate, X. (2009b). Validation of a fi-
nite element based simulation approach for offshore wind turbines
within iea wind annex XXIII - simulation challenges and results
for a 5-MW turbine on a tripod substructure. In Proceedings of
the Nineteenth International Offshore and Polar Engineering Con-
ference. Fraunhofer Institute for Wind Energy and Energy Systems
Technology (IWES).