European Wind Energy Conference (20-23 April 2010, Warsaw, Poland)

Output-only Modal Analysis on Operating Wind Turbines: Application to Simulated Data

#
Dmitri Tcherniak†
, Shashank Chauhan†, Michele Rossetti‡, Iciar Font‡, Jon Basurko , Oscar
#1
Salgado

Summary
Output-only (Operational) Modal Analysis (OMA) is a modern branch of experimental modal
analysis; the main advantage of OMA is its ability to extract modal model using only measured
responses. This makes OMA extremely attractive for modal analysis of big structures such wind
turbines.
However, there are issues preventing straightforward application of OMA to operational
turbines, e.g. structure invariance during the test. The effect of rotor rotation manifests itself in
the equation of motion with time-dependent coefficients. Formulating and solving eigenvalue
problem lead to time-dependent eigenvalues and eigenvectors which become meaningless as
modal parameters. Fortunately, so-called Coleman coordinate transformation (also known as
multi-blade coordinate transformation) allows one to eliminate time dependency of the system
matrices, thus converting the original time-varying eigenvalue problem to a time-invariant one.
This study extends this approach to experimental modal analysis. Forward Coleman
transformation is applied to the data measured on the wind turbine blades, which is then
combined with responses measured on the tower. The methods of Operational Modal Analysis
are then applied to the transformed data, resulting in modal frequencies, damping and mode
shapes. Backward Coleman transformation is finally employed for the mode shapes for their
visualization.
The study demonstrates the method using simulated vibrational responses of operational 3MW
wind turbine. The responses of the tower and blades were obtained from the simulation of
operational wind turbine dynamics under realistic wind load using commercial aeroelastic code.
Introduction
The design of modern wind turbines heavily relies on numerical models which are used for the
simulation of the dynamic behavior of wind turbines under different operating conditions. The
examples of such models are finite element, aeroelastic, control models, etc. The efficiency of
the final design strongly depends on the accuracy and validity of these models and simulation
codes. As a consequence, the design community (e.g. structural design, blade design, durability
and control) needs good experimental tools for their validation.
Generally, the dynamic behavior of structures is characterized in terms of their modal
parameters (modal frequencies, modal damping and mode shapes). Experimental Modal
Analysis (EMA) [1] is a technique for determining the modal parameters of a structure based on
experimental data. EMA (Figure 1a) involves exciting the structure by means of known forces
{F} (either using shakers or impact hammers) and measuring the response {X} to these forces
over the structure (usually by means of accelerometers). Based on calculated Frequency
Response Functions [H], the structure’s modal model (i.e. the set of modal frequencies and
damping (λk) and mode shapes {ψ}) is being extracted.

† ‡
Bruel and Kjaer Sound and Vibration Measurement, Denmark; ALSTOM Wind, Spain,
#
IKERLAN-IK4, Spain


Corresponding author, Bruel and Kjaer SVM, Skodsborgvej 307, DK-2850 Naerum, Denmark,
[email protected], phone: +45 77 41 22 44
Operational Modal Analysis (also known as output- a)
only modal analysis or OMA) techniques [2] are one
of the newer methods of performing modal analysis. {F(ω)} [H(ω)] {X(ω)}
(Measured) (Calculated) (Measured)
OMA techniques aim at obtaining modal
parameters characterizing the dynamics of the
structure/system based only on the knowledge of Modal parameters
response (i.e. output) of the structure to various λk {ψk}
ambient excitations, which are not measured. A
b)
good example, where by these techniques are
found to be readily applicable and very useful, is {F(ω)} {x(t)}
their application to river bridges or high
skyscrapers/towers. For such structures output (Assumed to be
uniformly distributed
responses are measured to ambient excitations in frequency and Modal parameters
space).
such as wind, rain, traffic etc. (which are not λk {ψk}
measured), and then system dynamic
characteristics are obtained from these
Figure 1 a) Experimental Modal Analysis
measurements. Figure 1b illustrates this process of
(EMA); b) Operational Modal Analysis (OMA)
identifying modal parameters using OMA
techniques.
The great advantage of OMA techniques is in providing the dynamic model of the structure
under actual operating conditions and real boundary conditions. The value of such model can
be demonstrated e.g. on the design of wind turbine control systems: Traditionally, the design of
control algorithms is performed based on linearized models of the wind turbine dynamics.
Control performance is strongly dependent on the accuracy of these models and for this reason
validation of the dynamics is essential for achieving optimal control. Having a reliable dynamic
model of a wind turbine for different wind loads would be a great advantage for designing
effective control algorithms (Figure 2).
Utilization of output-only data for system identification purposes started way back in 1970s, e.g.
[3], however it was not till early 1990s that researchers started taking note of these techniques.
During early 1990s, James et al. [4] proposed the NExT framework for utilizing output response
time histories for modal parameter estimation purposes, thus laying foundation of Operational
Modal Analysis. This research was a result of work performed at Wind Energy Research
Organization at Sandia labs for testing wind turbines. Surprisingly, though OMA subsequently
got popular in various civil engineering applications (bridges, buildings, stadiums, etc.) there
wasn’t much follow up with respect to wind turbines. One of the possible reasons for this could
be that application of OMA to wind turbines is not a straight forward task due to the presence of
considerable aeroelastic effects along with presence of rotational components.
The NExT framework involved a four stage process; data acquisition, calculation of correlation
functions, use of traditional parameter estimation algorithms for finding system parameters and
finally extraction of mode shapes. NExT was initially applied to a parked Vertical Axis Wind
Turbine (VAWT). However, wind turbines behave very differently in operation, in which case
aeroelastic effects are dominant and aeroelastic damping is significant in comparison to
structural damping. Thus, NExT was subsequently applied with limited success to rotating
VAWTs [4] and to Horizontal Axis Wind Turbine (HAWT) [5]. It was noted in [4] that mode shape

OMA

Dynamic model of WT for

different wind speeds

Aeroelastic Control System

Design Structural Design
Design

Aeroelastic stability Experimental dynamic

(interaction of aerodynamic Magnitude of structural model and verified FE
forces with elastic vibration response under operations, model for designing most
of the blades) durability, etc. effective WT behaviour

Figure 2. Application of dynamic model to wind turbine design related disciplines.

information is important to explain changes in damping with increase in turbine rotation rate and
that better techniques are required for estimating low amplitude modes and removal of harmonic
peaks.
In [6], the exciter mechanism method along with an Operational Modal Analysis algorithm,
Stochastic Subspace Identification (SSI) were used for estimating aeroelastic damping of
operational wind turbine modes. In this work, it was shown that blade pitch and generator torque
variations can be used as exciter mechanisms for exciting particular modes of interest.
However, this method suffers from limitations on account of the fact that excited turbine
vibrations are not pure modal vibrations and therefore the estimated damping is not actual
damping. Yet another limitation of this method was that it was not possible to achieve required
pitch amplitudes to excite sufficiently modes other than the tower modes. SSI method performed
better in comparison to the previous method, though longer time histories and averaging
techniques were required (only accelerations of the tower were measured). It was also able to
estimate the closely spaced modes including the rotor-related modes, which caused problems
for exciter method. Thus SSI showed promise in determining characteristics of an operational
wind turbine.
In case of a parked wind turbine, all OMA assumptions and also those pertaining to Modal
Analysis (system being linear, stationary and time invariant) are valid in general. Therefore,
application of OMA to standstill wind turbines is actually a straight forward task [7]. The case of
operational wind turbines is completely different: Periodic nature of aerodynamic forcing and the
fact that the system is now time-variant makes the application of OMA to operational wind
turbines questionable. However, the ability to estimate actual modal parameters under actual
boundary conditions and actual loading is the major incentive to be gained.
The latest became a motivation of the presented work: the goal was to identify which OMA
assumptions are violated; find the ways to circumvent this and verify the feasibility of suggested
approaches on simulated data.
The paper is built as follows: section 1 provides a brief introduction into OMA and the OMA
assumptions though not touching the algorithmic part of OMA. In Section 2 and 3 the two
violated OMA assumptions are identified and way(s) to circumvent them are suggested: section
2 deals with structure time-invariance, section 3 focuses on violation of OMA assumptions
concerning the nature of excitation forces. Section 4 describes the details of the application of
the OMA algorithms to the simulated response data and provides the results.
1. Theoretical background
The following expression relates the (linear) response of the structure to the excitation forces:
x(ω) = H(ω)f (ω) , (1)

where x(ω) is the vector of the response spectra, f(ω) is the vector of the excitation spectra and
H(ω) is the frequency response functions (FRF) matrix. From modal analysis theory, it is known
that FRF matrix contains all necessary information to extract modal parameters [1]. Multiplying
(1) by its Hermetian

x(ω ) H = f (ω ) H H (ω ) H , (2)
one obtains

x(ω ) x(ω ) H = H(ω ) f (ω )f (ω ) H H (ω ) H . (3)

The mathematical expectation of (3) is

G xx (ω ) = H(ω ) G ff (ω )H(ω ) H , (4)

H H
where Gxx(ω) = E(x(ω) x(ω) ) is the output power spectra matrix and Gff(ω) = E(f(ω) f(ω) ) is
the input power spectra matrix.
Assuming the forces are uncorrelated and distributed over the entire structure, their cross-
spectra matrix becomes diagonal:
 G ff (ω ) ∝ I , and (5)

G xx (ω ) ∝ H (ω )H H (ω ) . (6)
From modal analysis theory, it is known that FRF matrix contains all necessary information to
extract modal parameters. Expression (6) shows that, if the excitation assumptions fulfilled, the
response cross-spectrum matrix also contains the full information required to obtain (un-scaled)
modal model of the system.
2. Time-variance of the operational wind turbine and Multi-blade coordinate
transformation
Time-invariance of the structure during the test is a general and obvious requirement for any
kind of modal testing; it demands that structure under test remains the same during the test.
This is not a case for operational wind turbines. Indeed, many parts of wind turbine move with
respect to each other: the nacelle revolves about the tower following the wind; the rotor rotates
about its axis; the pitch of the blades changes depending on wind speed and rotor speed. The
first and the last mutual motions (nacelle yaw and pitch) are manageable: one can select a
period of time where the wind direction and speed do not significantly change so the structure
can be assumed time invariant for these substructures. However, the same approach cannot be
applied to the rotor: obviously the rotor can make hundreds of revolutions during the necessary
observation period.
Including rotor rotation into the equations of motion of entire wind turbine causes the mass,
stiffness and gyroscopic matrices to be dependent on time. Formulating and solving the
corresponding eigenvalue problem yields to time-dependent eigenvalues and eigenvectors,
which do not have a meaning as modal frequencies, damping and mode shapes in traditional
sense [1].
This issue can, however, be tackled by use of Multi-blade Coordinate (MBC) transformation [8,
9]. The idea behind MBC transformation is to replace individual blade deflections by some
special variables which include information about all three blades and include information about
instant azimuth angle ψ:
3 3 3
a 0 ,n = 1
3 ∑ qi,n ; a1,n = 23 ∑ qi,n cosψ i ; b1,n = 23 ∑ qi,n sinψ i . (7)
i =1 i =1 i =1

where qi,n is the n-th deflection of the i-th blade, a and b are multi-blade coordinates. The points
with same n are located on the same radius on the different blades, and the deflection in the
same direction (e.g. radial) is measured. Expressions (7) represent forward transformation (i.e.
blade coordinates to MB coordinates). The backward transformation is
qi ,n = a0,n + a1,n cosψ i + b1,n sin ψ i . (8)

MBC transformation converts the motion of individual blades described in rotating blade frame
into the ground-fixed frame which results in elimination of the periodic terms present in the
equations of motion, thus making application of modal analysis techniques, such as OMA,
possible.
The present study suggests using MBC transformation as a data pre-processing before applying
OMA algorithms. The schematic data flow is shown on Figure 3.
The process consists of the following steps:
1) Accelerations of points located on the blades and the tower (also nacelle, etc) and the
azimuth angle are acquired (as time histories). These can be the results of measurements
conducted on operational wind turbine or data simulated by aeroelastic code for selected
operating conditions (wind speed, direction, level of turbulence, etc).
2) The acceleration data from the rotating parts (blades, hub) are subjected to forward MBC-
transformation (7) using the azimuth data. Acceleration data from the tower and nacelle is
not transformed.
 Data from experiment / aeroelastic code
Backward MBC-trans.
Accelerations of the Rotor Accelerations of the
points on the blades azimuth points on the tower
Mode shapes in multi- Mode shapes in
Forward MBC-trans. blade coordinates blade coords.

Accelerations of multi- Modal frequency & Mode

OMA +
blade coordinates damping Animation

Results

Figure 3. Application of MBC-transformation. General data flow.

3) Obtained accelerations of multi-blade coordinates (a and b) together with accelerations of

not rotating parts (all as time histories) are input to OMA.
4) The output of OMA: modal frequencies and damping become the results
5) Resulting mode shapes are subjected to backward MBC transformation (8); the results can
be directly animated overlaid by the rotor rotation.
The suggested procedure is applied to the data obtained from the simulations; the results are
discussed in Section 4.
3. Nature of excitation forces
The second violated assumption concerns the nature of excitation. As it is mentioned in
Introduction, to make expression (5) valid the forces should act over entire structure, be
uncorrelated and have a flat spectrum. For standstill structures like towers, bridges, parked wind
turbines the aerodynamic forces acting on the structure very much satisfy these assumptions.
However, for operational wind turbines this is not the case. As it was shown in [10], a typical
spectrum of aerodynamic forces is characterized by peaks at the frequency of rotor rotation and
its harmonics. Besides this, the forces acting at different points of the blades are highly
correlated at fundamental frequency and its harmonics. Figure 4a shows a typical spectrum of
the aerodynamic force estimated at the different radii of a blade; Figure 4b presents the
coherence between forces acting at different points on the blades.
Analyzing the plots on Figure 4, one can note that the peaks have “thick tails” meaning that the
signals are not just a mixture of several pure tones (as it is in a case of for example unbalanced
rotor) but have rich frequency content. This means that expression (5) is only valid in quite
narrow frequency bands between the harmonics where the coherence drops to minimum and
spectral density has a reasonably flat valley.
As it was discussed in [10], the following means for circumventing the problem can be
suggested:
1) Application of tone removing methods (e.g. based on synchronous averaging, [11]) cannot
be considered as a proper solution, as these methods work well only for sharp peaks but
will not have any advantage in this case due to the “thick tails” phenomenon.

a) b)
2 1
10 2
P ower dens ity s pec tra, S ou (r, r, f )/ σ 2u

γAB (rA = 26.42m, rB = 54.40m, s ame blade)

2
0.8 γAB (rA = 26.42m, rB = 54.40m, diff. blades )
1
10 2
γAB (rA = 26.42m, rB = 26.42m, diff. blades )
Coherence γ2AB

2
0.6 γAB (rA = 54.40m, rB = 54.40m, diff. blades )
0
10
r=0 0.4
r = 26.42m
-1
10 r = 54.40m
0.2

-2
10 0
0.001 0.01 0.1 1p 2p 3p 4p 1 0 1p 2p 3p 4p 1 6p 7p 8p 9p 2 11p 2.5
Frequency f, Hz Frequency f, Hz

Figure 4. a) Power spectral density of aerodynamic forces acting at different radii of the blade; b)
Coherence between the aerodynamic forces acting at different points of the blades.
2) From a first glance, the use of run-up and run-down events looks attractive but, first of all,
these events are rather short compare to the acquisition time required for data collection for
proper OMA application (at least 10 minutes of data are required if the lowest frequency of
interest is 0.2-0.4 Hz). Secondly, a wind turbine engineer is typically interested in the
dependency of modal parameter to the rotor speed; in the case of run-up/run-down events,
only averaged modal characteristics can be obtained.
3) One can also consider a careful planning of the experiment, constructing the test matrix in a
way to avoid the modal frequencies (which are approximately known from finite element
analysis) to be in the vicinity of rotor speed and its lowest harmonics. This means that only
few modes can be estimated with higher degree of confidentiality for a given rotor speed,
while another rotor speed will be suitable for another set of modes. The example of such
test matrix is shown on Figure 5. In the white and orange cells of the matrix, the distance
between the rotor frequency and its harmonics and the expected mode frequency is too
small, therefore application of OMA is doubtful. For operating condition corresponding to
blue and green cells, OMA can be readily applied. This approach is used in the study.
4) Amongst recently suggested methods, operational modal analysis based on transmissibility
functions [12] appears very attractive. Their main advantage is insensitivity to colored
excitation spectra. However, so far these methods are still under development and not
ready for industrial applications.
It must be noted that MBC transformation removes periodicity from the system matrix but does
not help in removing periodicity from the excitation forces.
4. Results and discussion
In the current study, we applied the suggested approach (see Figure 3, 5) to synthesized data.
As it was mentioned in the Introduction, the goal of the study was to validate the feasibility of the
approach before conducting expensive data acquisition campaign on a real wind turbine.
According to the test matrix (Figure 5), seven representative operating conditions were selected
(Table 1).
For each chosen operating condition, a simulation was performed, and acceleration data were
generated. The simulations were performed by means of commercial aeroelastic code using
aerodynamics, mass, geometry, stiffness and control parameters of the wind turbine supposed
to be tested (new ALSTOM W IND ECO 100 wind turbine). For every chosen operating condition,
the time histories corresponding to 15 minutes of operations were generated for 6 elevations of
the tower (both X and Y directions) and 4 radial locations on each blade (both in-plane and out-
of-plane directions). In total, 36 acceleration time histories were used. The azimuth angle data
synchronized with the acceleration data were utilized for MBC transformation.
Wind speeds
List of the modes

cases selected for analysis

Figure 5. Example of the test matrix. The colors represent the expected difference from fundamental
frequency and its harmonics to the modes of interest: the difference is < 0.15Hz – white; between 0.15Hz
and 0.18Hz – orange; between 0.18Hz and 0.5Hz (blue); > 0.5Hz (green). Operating conditions selected
for the analysis are framed.
Time domain based Stochastic Subspace Iteration (SSI) Table 1
algorithm was employed for modal identification. As it is No. Test case
typical for modal analysis, the behavior of curve fitting 1 Standstill (wind speed 9 m/s)
algorithm is defined by a number of parameters; for
example, in case of SSI, the decimation factor, the 2 Production, wind speed 3 m/s
number of projection channels, maximum state space 3 Production, wind speed 5 m/s
dimension can be listed as such parameters. While doing 4 Production, wind speed 9 m/s
modal identification, we observed high sensitivity of the 5 Production, wind speed 15 m/s
results to these input parameters. This could be
6 Production, wind speed 19 m/s
explained by the violation of OMA assumptions for
aeroelastic excitation. 7 Production, wind speed 23 m/s

In order to elude the ambiguity of the obtained modal parameters, the modal identification was
performed for 6-8 sets of input parameters for each production case. The collected statistics
allowed the estimation of mean values for each mode of interest and, which is quite important,
the standard deviation and confidence interval. Obviously, a smaller standard deviation means a
higher confidence in identified modal parameters.
Figure 6a presents the modal parameters of the rotor-related modes as a function of the wind
speed (Campbell diagram). Figure 6b shows modal parameters of the tower related modes. The
mode nomenclature is given in Table 2.
Rotor-related modes of an operational wind turbine have
Table 2 quite complex nature; Hansen in [13, 14] contributed a lot
Mode Abbrevia- into the theoretical understanding of the phenomena.
name tion Pairs of asymmetric modes at standstill (e.g. out-of-plane
st st st st
Tower 1 tilt and 1 yaw or in-plane 1 horizontal and 1
st vertical) typically have very close resonance frequencies.
1 Tower Fore-Aft T1FA When the rotor rotates, these pairs transform into pairs of
st
1 Tower Side-to-Side T1SS whirling modes, backward and forward whirling, with the
nd
2 Tower Fore-Aft T2FA frequencies differ by 2Ω (where Ω is rotation frequency).
1
2
nd
Tower Side-to-Side T2SS In RPM-regulated regime , the increase of rotor rotational
speed causes the centrifugal stiffening which contributes
Drive Train
to increasing modal frequencies for all modes. When the
Drive Train Torsional DT turbine is pitch-regulated, the increasing pitch makes
Rotor (out-of-plane) blades stiffer in out-of-plane direction and more compliant
st
1 Backward Whirling O1W-bw in in-plane direction. All these phenomena can be
st
1 Forward Whirling O1W-fw
followed in the Campbell plot.
st
1 Collective O1C The confidence intervals denoted by the vertical line
nd segments on Campbell plot show that some of the modes
2 Backward Whirling O2W-bw
nd
are more easily identifiable then others. For some wind
2 Forward Whirling O2W-fw speeds few modes were not possible to identify at all,
nd
2 Collective O2C e.g. all first out-of-plane modes and O2W-fw for 3 m/s
Rotor (in-plane) wind. Generally speaking, out-of-plane modes are more
st
1 Backward Whirling I1W-bw difficult ones compare to in-plane modes; this can be
st explained by much higher damping (Figure 7b) inherent
1 Forward Whirling I1W-fw in out-of-plane modes.
st
1 Collective I1C
nd Among other interesting phenomena, one can note for
2 Backward Whirling I2W-bw example
nd
2 Forward Whirling I2W-fw
- change of mode order between O1W-fw and I1W-bw
nd
2 Collective I2C (seen as an intersection of the cyan and magenta
lines);

1
ALSTOM W IND ECO100 wind turbine is RPM-regulated (rotor RPM changes, pitch stays
constant) for low wind speed, and pitch-regulated (RPM is maintained constant by controlling
the blades’ pitch) for higher wind speed.
- the diverge of the backward and forward whirling pairs, e.g. I1W-fw/bw and O1W-fw/bw
which are separated by 2Ω interval.
The frequencies of the tower-related modes do not change significantly with the wind speed, as
can be seen on Figure 6. Actually, tower modes can be obtained by removing blade
accelerometers signals from the data sets. However, this requires some preliminary knowledge
about the tower mode frequencies since the mode shapes can be easily mixed up with the rotor-
related modes (which are observed as tower modes if the information about the rotor is
missing). Thus, it is more reliable to use the full datasets which include full information instead
of the reduced datasets.
Modal damping of the rotor-related modes is shown on Figure 7a and 7b. As it can be clearly
seen, the confidence of damping estimation is quite low, especially for heavily damped modes.
Out-of-plane modes are more heavily damped compared to their in-plane counterparts. Wide
confidence interval does not allow us making any conclusion about the development of damping
with increase of wind speed.
Mode shape animation plays an important role in modal analysis since it helps mode
identification and classification. Unfortunately, conventional modal analysis packages do not
allow animation of mode shapes of time-variant systems. A dedicated MATLAB-based
animation program was made to facilitate mode shapes visualization overlaid with rotor rotation.
The screen dump of the program is shown on Figure 8.

a)
Rotor modes vs wind speed

I2W-fw

I2W-bw
O3W

I2C
O2C
Modal frequency, Hz

O2W-fw
O2W-bw

I1W-fw
O1W-fw
I1W-bw
O1C
O1W-bw
3P
DT1
2P
0 1P

0 5 10 15 20
Wind speed, m/s

b)
T1SS
T2SS
T1FA
T2FA
1p
3p
6p
Tower modes frequency, Hz

0
0 5 10 15 20
Wind speed, m/s

Figure 6. Frequencies as a function of wind speed: a) Rotor-related modes. Dashed line – the mode was
not identified for the corresponding wind speed; b) Tower-related modes.
 Modal damping (rotor in-plane modes) vs wind speed Modal damping (rotor out-of-plane modes) vs wind speed
10 60

I1W bw
O1W yaw
I1W fw
I2C O1C
9
I2W bw O1W tilt
I2W fw
O2W yaw
I3C 50
O2W tilt
8
O2C
O3W

7
40

Modal damping, %
Modal damping, %

5 30

20
3

2
10

0 0
0 5 10 15 20 0 5 10 15 20
Wind speed, m/s Wind speed, m/s

Figure 7. Modal damping, rotor-related modes a) In-plane modes; b) out-of-plane modes

5. Drive train mode

The turbines from ALSTOM W IND have a special mechanical
configuration providing direct support of the rotor by means of
the frame, leaving the low speed shaft only the torque
transmission task. This concept has a clear influence on the
dynamic behavior of the turbine since the rotational moment is
decoupled from the bending moments. Figure 9a compares the
layout of the ALSTOM W IND turbine (left) with the classical layout
(right).
The commercial software used in this study describes the drive
train in terms of rotor inertia, low speed shaft (LSS) stiffness,
gearbox ratio, high speed shaft (HSS) and generator inertias.
The nacelle is described as a rigid support with the mass and
inertia attached to the tower. Depending on the wind turbine
state, either braked or in production, the torsional mode
changes. When braked, the torsional mode is mainly governed
by the rotor inertia and the LSS stiffness. However, in
production case, this mode is mostly influenced by the HSS and
the generator inertias and the LSS stiffness. Figure 8. Animation of the
modes of the operational wind
OMA carried out for the parked wind turbine identified the drive turbine
train mode at the frequency expected from the analytical
solution. The corresponding mode shape shows that all three blades rotate symmetrically, as it
is expected for a drive train mode (Figure 9b). The tower is coupled to this mode by its side to
side bending. Due to this, this mode can be detected using only the deflections of the tower.
Obviously, in this case the identification is more difficult since there is no information about the
blades deformation.
The identification of this mode for the production cases seems to be more complex, especially
when the first drive train mode appears relatively close to the 3P excitation at nominal speed. In
this frequency range the OMA assumptions are not fulfilled, which causes the mode not being
clearly identified by the OMA procedure.
6. Conclusion
The study presents the application of output-only modal analysis (OMA) to the operational wind
turbine. The work demonstrates that a straightforward application of OMA to operational wind
turbine is not possible since the main assumptions OMA is based upon are violated. The study
suggests the way to circumvent the abovementioned violations. In order to validate the
suggested approach, it is applied to synthesized data obtained by simulating ALSTOM W IND ECO
100 wind turbine behaviour for several selected production cases. The resulting modal
parameters (modal frequencies and damping) are presented as a function of wind speed (so-
called Campbell diagrams), discussed and compared with the results found analytically in other
studies.
a) b)

Figure 9. a) In a classical wind turbine layout (right), the rotor is hanged on the first rotor bearing, so the
torsional and bending moments are coupled. In ALSTOM W IND layout (left) these moments are decoupled;
b) Front view of the drive train mode.
References
1. Ewins DJ, Modal Testing: Theory, Practice and Application; Research Studies Press Ltd.:
Baldock, England, 2000.
2. Zhang L, Brincker R, Andersen P. An Overview of Operational Modal Analysis: Major
Development and Issues, Proceedings of 1st International Operational Modal Analysis
Conference, Copenhagen, Denmark, 2005
3. Gersch W, Fouth DA. Least Squares Estimates of Structural System Parameters Using
Covariance Function Data, IEEE Transactions on Automatic Control 1974; AC-19(6): 898-
903.
4. James GH, Carne TG, Lauffer JP. The Natural Excitation Technique (NExT) for Modal
Parameter Extraction from Operating Structures, Modal analysis, The International Journal
of Analytical and Experimental Modal Analysis 1995; 10:260-277.
5. James GH. Extraction of Modal Parameters from an Operating HAWT using the Natural
th
Excitation Technique (NExT), Proceedings of the 13 ASME Wind Energy Symposium,
New Orleans, LA, 1994.
6. Hansen MH, Thomsen K, Fuglsang P, Knudsen T. Two methods for estimating aeroelastic
damping of operational wind turbine modes from experiments, Wind Energy 2006; 9(1-
2): 179-191.
7. Chauhan S, Hansen MH, Tcherniak D. Application of Operational Modal Analysis and Blind
Source Separation / Independent Component Analysis Techniques to Wind Turbines,
Proceedings of XXVII International Modal Analysis Conference, Orlando (FL), USA, 2009.
8. Bir G. Multiblade Coordinate Transformation and Its Application to Wind Turbine Analysis,
Proceedings of ASME Wind Energy Symposium, Reno, Nevada, 2008
9. Skjoldan PF, Hansen MH. On the Similarity of the Coleman and Ljapunov-Floquet
Transformations for Modal Analysis of Bladed Rotor Structures, Journal of Sound and
Vibration, 2009; 327:424-439.
10. Tcherniak D, Chauhan S, Hansen MH, Applicability Limits of Operational Modal Analysis to
Operational Wind Turbines, Proceedings of International Modal Analysis Conference,
Jacksonville (FL), USA, Feb. 2010.
11. Peeters B, Cornelis B, Janssens K, Van der Auweraer H. Removing disturbing harmonics in
operational modal analysis, Proceedings of International Operational Modal Analysis
Conference, Copenhagen, Denmark, 2007.
12. Devriendt C, De Troyer T, De Sitter G, Guillaume P. Automated operational modal analysis
using transmissibility functions, Proceedings of International Seminar on Modal Analysis,
Leuven, Belgium, 2008
13. Hansen MH. Improved Modal Dynamics of Wind Turbines to Avoid Stall-induced Vibrations,
Wind Energy 2003; 6:179-195.
14. Hansen MH. Aeroelastic Instability Problems for Wind Turbines, Wind Energy 2007; 10:551-
577