monitoring resonant frequencies and damping values of an offshore wind turbine in parked conditions

www.ietdl.org

IE

d

Published in IET Renewable Power GenerationReceived on 7th July 2013Revised on 22nd January 2014Accepted on 10th February 2014doi: 10.1049/iet-rpg.2013.0229

Special Issue: European Wind Energy Association 2013

T Renew. Power Gener., 2014, Vol. 8, Iss. 4, pp. 433–441oi: 10.1049/iet-rpg.2013.0229

ISSN 1752-1416

Monitoring resonant frequencies and damping valuesof an offshore wind turbine in parked conditionsChristof Devriendt1, Wout Weijtjens2, Mahmoud El-Kafafy2, Gert De Sitter1

1Offshore Wind Infrastructure Lab, Brussels, Belgium2Vrije Universiteit Brussel, Brussels, Belgium

E-mail: [email protected]

Abstract: This study shows the first results of a long-term monitoring campaign on an offshore wind turbine in the Belgian NorthSea. It focuses on the continuous monitoring of the resonant frequencies and damping values of the most dominant modes ofthe support structure. These parameters allow to better understand the dynamics of offshore wind turbines and are crucial in thefatigue assessment during the design phase. They can also help to minimise operation and maintenance (O&M) costs and toassess the lifetime of the offshore wind turbines structures during their operation. To do an accurate continuous monitoring ofthese parameters, a state-of-the-art operational modal analysis technique has been automated, so that no human-interaction isrequired and the system can track small changes in the dynamic behaviour of the offshore wind turbine. The study will analysethe resonant frequencies and damping values of the most dominant modes shapes while the wind turbine is in parked conditions.

1 Introduction

The size of commercial wind turbines has continuouslyincreased in the last decades, going from 50 kW ratedpower machines to the current 5 MW turbines. Thestructural growth and the trend of investing more and moreon offshore installations pose serious challenges for thefuture. To increase the power generation and limit theweight, offshore wind turbines are becoming structurallymore flexible, thus an accurate prediction of their dynamicbehaviour is mandatory. On the other hand, inspection andmaintenance for offshore installations are much morecumbersome and expensive than for onshore turbines. Thus,a remote monitoring application with the ability to trackstructural changes can help to reduce O&M costs andassess the lifetime of these structures.Many large scale offshore wind farm projects use monopile

foundations to obtain a cost effective design. Even for waterdepths beyond 30 m the monopile design is currently beingconsidered as an option. During the design of thesemonopile structures fatigue because of the combined windand wave loading is one of the most important problems totake into account. Coincidence of structural resonances withthose dynamic wind and wave forces can lead to largeamplitude stresses and subsequent accelerated fatigue. Thecurrent practice is to design the wind turbine supportstructure in such a way that the tower fundamentalresonance does not coincide with the fundamental rotational(1P) and blade passing (3P for three-bladed turbines)frequencies of the rotor [1]. However, the higher-order rotorharmonics might still coincide with higher modes of thesupport structure inducing greater vibrations andconsequently a reduced fatigue-life. Experiments performed

by the Maritime Research Institute Netherlands alsoconfirmed that breaking waves could induce significantoscillations and accelerations in the turbine [2]. This canhave a significant effect on the lifetime of the wind turbine.Damping ratios are crucial for the lifetime predictions, as

the amplitude of vibrations at resonance are inverselyproportional to these ratios. The overall damping ratios ofan offshore wind turbine consists of a combination ofaerodynamic damping and damping because of constructivedevices, such as a tuned mass damper (TMD), andadditional offshore damping, for example, structuraldamping [3–5]. Compared with the onshore supportstructures, the additional damping is further influenced byeffects such as soil damping and hydrodynamic damping.Real damping ratios are difficult to predict by numericaltools and therefore measurements on existing offshore windturbines are crucial to verify existing design assumptions[1]. Several measurement studies have previously beenperformed at different offshore wind farms. Damgaard et al.[5] performed ‘rotor-stop’ tests for five wind parks between2006 and 2011 and determined the natural frequency andmodal soil damping of a specific wind turbine at differentsoil conditions. Hackell and Rolfes [6] have extracted themodal parameters for an offshore wind turbine on a tripodstructure located at German Alpha ventus wind farm.Devriendt et al. [7, 8] performed an overspeed stop test toestimate the overall damping of the first for-aft (FA) modeof a V90 wind turbine in the Belgian Belwind wind farm.Offshore wind turbines are, however, complex structures

and their dynamics can vary significantly because of thechanges in operating conditions, for example, changingrotor-speed, changing pitch angle or changing ambientconditions, for example, change in wind speed, wave height

433& The Institution of Engineering and Technology 2014

www.ietdl.org
or wave period. Especially in parked conditions, with thereduced aerodynamic damping forces, the dynamic responsebecause of wind and wave actions can be high. Thereforemeasurements on the parked offshore wind turbines are inparticular interesting. A continuous monitoring of the modalparameters (i.e. resonance frequencies, damping ratios andmode shapes) of the fundamental vibration modes of thesupporting structures will allow us to verify the existingdesign assumptions and understand the dynamic behaviourof an offshore wind turbine in parked conditions.Identification of the modal parameters of a full-scale
operating wind turbine is particularly difficult and in theresearch community a lot of effort still goes into thedevelopment of suitable methods to tackle this problem [6,9, 10]. Classical experimental modal analysis methodscannot be applied because the input force owing to thewind and the waves cannot be measured. For this reason,operational modal analysis methods were developed toidentify the modal parameters from the response of amechanical structure in operation to unknown randomperturbations [11–13]. These methods work under theassumption that the system is linear time invariant duringthe analysed time interval and that the excitation is whitenoise within the frequency band of interest. Owing to thepresence of rotating components and their correspondingharmonic force contributions that do not comply with theoperational modal analysis assumptions the identificationprocess becomes difficult or impossible. The analysis in thispaper is focused on a wind turbine in idling or parkedconditions while the rotating speed was always lower then1.5 rpm. This means that the risk of having some harmoniccomponents in the frequency-band or interest resulting fromthe rotating equipment is very low. Therefore the whitenoise assumption in parked conditions is valid andoperational modal analysis (OMA) is applicable.State-of-the-art operational modal analysis techniques can

provide accurate estimates of natural frequencies anddamping ratios of structures in parked conditions. Howeverexisting techniques require a lot of human-interaction and assuch they cannot easily be applied for continuousmonitoring purposes. Therefore the development andvalidation of tools for the automatic and continuousidentification of these parameters is essential to process largeamounts of data and perform a long-term analysis.Furthermore, it is required that these routines are sufficientlyrobust to run on an online basis, in order to provide inalmost real-time parameters that characterise the condition ofthe wind turbine. In this paper an automatic monitoringapproach will be presented using a state-of-the-art modalparameter estimator, the polyreference least squares complexfrequency-domain estimator-commercially known asPolyMAX [14, 15]. The method will be applied to 2016datasets of 10 min collected during a period of 2 weekswhile the wind turbine was in parked conditions. It will beshown that the method allows us to identify and track themost dominant modes in the frequency band of interest.This will allow us to have a discussion on the dynamicbehaviour of an offshore wind turbine in parked conditionsin terms of its resonance frequencies, mode shapes anddamping values of the most dominant modes.

2 Offshore measurements

The measurement campaign is performed at the Belwindwind farm, which consists of 55 Vestas V90 3 MW wind


turbines. The wind farm is located in the North Sea on theBligh Bank, 46 km off the Belgian coast. The wind turbineis placed on a monopile foundation structure with adiameter of 5 m and a wall-thickness of 7 cm. Thehub-height of the wind turbine is on average 72 m abovesea-level. The transition piece is 25 m high. The interfacelevel between the transition piece and the wind turbinetower is at 17 m above sea level. The actual water depth atthe location of the tested wind turbine is 22.9 m and themonopile has a penetration depth of 20.6 m. The soil isconsidered stiff and mainly consists of sand.Within the OWI-project (www.owi-lab.be) two

measurement campaigns have been performed. The firstshort measurement campaign focused on performing anoverspeed test with the aim of obtaining a first estimate ofthe damping value of the fundamental FA vibration modeof the wind turbine. The results of this measurementcampaign have been presented at EWEA 2012 [7] andpublished in [8]. During the second long term measurementcampaign we are continuously monitoring the evolution ofthe frequencies, damping and mode shapes of the mostdominant modes of the tower and foundation.The structures instrumented in this campaign are the tower

and transition piece. Measurements are taken at four levels onnine locations using a total of ten sensors. The measurementlocations are indicated in Fig. 1. The locations are chosenbased on the convenience of sensor mounting, such as thevicinity of platforms. The chosen levels are 67, 37, 23 and15 m above sea level, respectively, level 1–4. There are twoaccelerometers mounted at the lower three levels and four atthe top level. The chosen configuration is primarily aimedat identification of tower bending modes. The two extrasensors on the top level are placed to capture the towertorsion. The installed accelerometers have a high sensitivity(1 V/g) and are able to measure very-low frequencies (0–250 Hz). Fig. 1 shows an example of the accelerationsmeasured during 10 min of the ambient excitation.The data-acquisition system is mounted in the transition

piece and was programmed to acquire data with a samplerate of 5 kHz. Considering the frequency band of interest(0–2 Hz) and to reduce the amount of data, the recordedtime series are filtered with a band-pass filter andre-sampled with a sampling frequency of 12.5 Hz. Afterthese steps a coordinate transformation is performed toobtain the accelerations in the coordinate system of thenacelle [10, 16].To classify the measurements, SCADA data (power, rotor

speed, pitch angle and nacelle direction) and ambient data(wind speed, wind direction, wave height, wave period,temperature and so on) is being collected at 10 minintervals. In Fig. 2, the corresponding SCADA data for thepresented monitoring period is given. Note that the x-axis islabelled as ‘index’ and each data-point represents a 10 minvalue. The monitoring results presented in this paper havebeen collected during a period of 2 weeks resulting in 2016values. The wind speed varied between 0 and 16 m/s. Thepitch angle was almost constant during the period ofanalysis, with a pitch angle of 78° or 88.3°. Most of thetime the wind-turbine was idling with a rotational speedlower than 1.3 rpm or was in parked conditions.

3 Continuous dynamic monitoring

During the long-term measurement campaign the aim is tocontinuously monitor the evolution of the frequencies and

IET Renew. Power Gener., 2014, Vol. 8, Iss. 4, pp. 433–441doi: 10.1049/iet-rpg.2013.0229

Fig. 2 SCADA data for monitoring period from top to down: rpm, pitch-angle, yaw angle and wind speed

Fig. 1 Measurement locations and data-acquisition system based on NI CompactRIO System (left).` Example of measured accelerationsduring ambient excitation on four levels, with level 1 the highest level, in the SS direction (right-top) movement seen from above (right-bottom)

www.ietdl.org

damping values of the most dominant modes of the tower andfoundation.In order to perform continuous dynamic monitoring the

following steps are followed:


Step 1: Pre-processing vibration data

1. Creation of a database with the original vibration datacollected at 10 min intervals and sampled at high frequency,


www.ietdl.org
together with the ambient data and the SCADA data withcorresponding time stamps.2. Pre-processing the vibration-data to eliminate the offset,reduce the sampling frequency, transform them in thenacelle coordinate system.3. Calculate the power spectra of the measured accelerationresponses using the correlogram approach [7, 8].
Step 2: Automated operational modal analysis

1. Applying a modal parameter estimator to the calculatedpower spectra to extract the modal parameters in anautomated way based on a clustering algorithm2. Calculate statistical parameters (e.g. mean values andstandard deviation) of the identified parameters

Step 3: Tracking frequencies, damping values and modeshapes

1. Creation of a database with processed results

To allow an accurate continuous monitoring of thedynamic properties a fast and reliable solution that isapplicable on industrial scale has been developed. The

Fig. 3 Example of a stabilisation diagram, red s indicates a well identifiethat only the frequency is well identified (top left). Clusters obtained frocolours have no meaning, the red square indicate the mean values of thclusters based on the different validation criteria and using a Fuzzy C-m


different steps of the proposed approach to automate theprocess are briefly explained in the next paragraphs, a moredetailed description can be found in [17].

3.1 Automated modal analysis

In a first step of the automated modal analysis the poles,containing the resonance frequencies and damping values,are estimated with the p-LSCF algorithm [14] for differentmodel orders. In a non-automated approach these resultscan be used to construct a classic stabilisation chart fromwhich the user can try to separate the physical poles(corresponding to a mode of the system) from themathematical ones [14, 15, 18]. Fig. 3 shows an example ofa stabilisation diagram using a 10 min time-segment with512 time lags taken from the correlation functions and amaximum model order of 32. The analysis focuses in thefrequency range 0–2 Hz, where the main vibration-modes ofinterest are expected. The stabilisation diagram seems tohave around five well identifiable stable poles. In anon-automated approach the poles are selected from thestabilisation diagram by clicking on a red s which indicatesa well identified pole. Such an approach is highly userdependent and labour intensive especially in the presence of

d pole (both frequency and damping well identified) a blue f indicatesm the poles for different model order by the cluster algorithm, thee pole within cluster (top right) classification result of the obtainedean cluster algorithm (bottom)


www.ietdl.org
noise. The next paragraph will present a fully automatedapproach that requires no user interaction.In [17], the authors developed a methodology for automatic
identification of modal parameters, using parametricidentification methods, based on a hierarchical clusteringalgorithm, to cluster poles that are related to the samephysical mode. Several basic procedures are available in theMATLAB Statistics Toolbox and several papers havesuccessfully applied these methods [17, 19]. In this work, arobust agglomerative hierarchical approach was used basedon the method presented in [20–22]. Fig. 3 shows theresults of the cluster algorithm on the previously identifiedpoles shown in the stabilisation diagram. When we look atthe results of the clustering-algorithm we can clearlyidentify several clusters. Based on the cluster results, astatistical analysis, yields the mean and the standarddeviation (std) for each of the estimated poles and hence forthe damped natural frequencies and damping values. Sincetypically, high model orders are chosen for reasons of noiseon the data as well as the use of a discrete time transferfunction model, not each cluster corresponds to a systempole. Therefore each of the clusters needs to be assessed forits physicality. Moreover, not all the identified clusters havefrequencies and damping values with small uncertaintiesbecause of the noise and modelling errors.Validation criteria can be defined that can distinguish

physical modes from non-physical ones and that allows usto discard estimates of low quality. In the proposedhierarchically clustering approach, the number of poles in acluster, or the identification success rate, as well as thestandard deviations on the frequencies and damping valuesof each cluster, may give a good indication for thephysicality and quality of the identified clusters. When asufficient number of sensors is available, mode shape

Fig. 4 Five dominant identified mode shapes: from left to right

a First FA bending mode, FA1b First SS bending mode, SS1c Mode with a second FA bending mode tower component FA2d Mode with a second SS bending mode tower and nacelle component, SS2Ne Mode with a second FA bending mode tower and nacelle component, FA2N


information can be used to further evaluate the clusters byconsidering an additional validation criteria, such as themodal assurance criteria, modal phase collinearity (MPC)and the modal phase deviation (MPD) [23]. For each polein a cluster the corresponding mode shape information canbe computed and the following ratios can, for example, becalculated: the fraction of mode shapes that have an MPClarger than 80% and an MPD smaller than 10°.Based on the above discussed validation criteria the

clusters obtained in the previous step can now be groupedinto two classes using an iterative Fuzzy C-means clusteringalgorithm, that is, physical clusters retained for continuousmonitoring and clusters discarded for continuousmonitoring. This algorithm has already been successfullyapplied in [17, 21, 22]. The algorithm is implemented inthe MATLAB Fuzzy Logic Toolbox. The output of thealgorithm gives a classification result for the clusters. If theclassification result is larger than 50% then it is decidedthat the cluster belongs to the class of clusters that can beused for continuous monitoring, more details can be foundin [17]. In Fig. 3, we can see the results of classificationalgorithm. In this example, five clusters have a classificationresult above 50% and are thus retained for continuousmonitoring. These five clusters represent the five dominantvibration modes that are excited during the consideredmonitoring period while the wind turbine was in parkedconditions.The modes shapes are shown in Fig. 4. The first mode is the

first FA bending mode (FA1). The second mode is the firstside-side (SS) bending mode (SS1). These modes arecharacterised by a lot of motion at the nacelle level. Nextwe identify a mode with a second FA bending modebehaviour in the tower and almost no motion at nacellelevel (FA2). This mode is in fact a coupled blade mode


www.ietdl.org
inducing some vibrations in tower and foundation in the FAdirection, The last two modes also show a second bendingmode behaviour in the tower but now with a small motionof the nacelle, in respectively, the SS (SS2N) and FA(FA2N) direction. These modes are in fact the actual secondbending modes of tower and foundation in, respectively,SS2N and FA2N direction. We can conclude that thealgorithm seems to be an efficient approach for theclassification of the identified clusters and thus can beconsidered as a valuable tool to be used in automaticidentification and continuous monitoring of the mostdominant modes.
3.2 Tracking and discussion of the first monitoringresults

In the last step we want to be able to track the different modesover time. Since the natural frequencies and damping ratios ofthe modes may change because of changes in the operating orambient condition, it is not always straightforward todetermine which identified modes from two subsequentdatasets are corresponding. Furthermore, it can happen thattwo modes cross each other in terms of natural frequency ordamping ratio or even that the estimation algorithm returnsa single mode at the moment of crossing, masking the othermode and seriously hampering the tracking process. In thecase of the wind turbine, this is especially the case with theclosely spaced first FA-mode and first SS-mode. Owing tothe small asymmetries in the structure or soil conditions,these modes can have crossing frequencies owing to thechanging yaw-angles. In the presented method, the modeshape information in the form of the modal assurancecriterion (MAC) [23] is applied for mode tracking. TheMAC evaluates the degree of correlation between two modeshapes over subsequent instants resulting in a value close toone for corresponding modes. This approach allows to linkestimates of the same physical mode and simultaneously,

Fig. 5 Estimated mode shapes of the five dominant modes during the mtop views (right figures), from top left to bottom right: FA1, SS1, FA2, S


possible frequency shifts lower than 5%, motivated by thedifferent ambient and operating conditions or possibledamages, are allowed. The reference modal parameters(natural frequencies, modal damping ratios and modeshapes) used in this case are the ones presented in Fig. 4.Fig. 5 illustrates the different mode shapes identified in the2016 successive datasets. It can be seen that the modeshapes are very coherent over the different datasets,indicating that we are clearly able to track the differentphysical modes. Small differences can be attributed to, forexample, small errors on the yaw angle used for thecoordinate transformation or small asymmetries in thefoundation structure and soil conditions.Fig. 6 presents the evolutions of the natural frequencies of

the five most dominant modes identified during themonitoring period. A zoom is also presented to show thatthe methodology has been able to successfully identify theclosely spaced FA and SS mode even when the frequenciescross each other. The methodology also successfullymanaged to capture small daily variation on the highestthree modes. These changes can be attributed to the tidaleffect. The lowest two modes seem to be less sensitive tothis effect. This can be understood because of the fact thatthe highest three modes, all showing a second bendingmode behaviour, have a higher relative motion at the waterlevel in comparison with the first FA and SS bending mode.The variation of the modal damping ratios of these modes

is represented in Fig. 7. It can be observed that the values forthe highest three modes are reasonably coherent (taking intoaccount that the estimates of the modal damping ratiosalways present some uncertainty) whereas, the onesassociated with the lowest two modes present a higherscatter. This can be explained by the fact that identifyingtwo closely spaced modes always increases the uncertainty,especially on the damping values. Also the fact that we useonly 10 min of data, mainly affects the quality of theestimates of the lowest modes. However, a part of this high

onitoring period: FA-direction (red lines) SS-direction (green lines),S2N and FA2N


Fig. 6 Evolution of frequencies of the five dominant modes during the monitoring period (top) Zoom on the closely spaced FA1 and SS1 modes(bottom left) Zoom on the highest three modes (bottom right)

www.ietdl.org

scatter can also be explained by the high dependence on theambient parameters, for example, the wind speed of thedamping values.We can make a statistical analysis on the results obtained

during the continuous monitoring, for different windspeeds. Fig. 8 illustrates one box and whisker plot perdamping value of each tracked mode. The obtained resultsfrom this short dataset allows us to get a betterunderstanding of the damping of a parked offshore windturbine.For example for the first mode in the SS(mode 2: SS1)

direction we find a slightly higher damping in comparisonwith the first FA mode (mode 1: FA1). This can beexplained because of the presence of some extra

Fig. 7 Evolution of damping values of the five dominant modes during


aerodynamic damping effects in the SS directionconsidering the pitch angle of 88.3° in the parkedconditions and 78° when idling. According to Hansen et al.[24], the aerodynamic forces are present even at standstillbecause of the larger blade surface that interacts withsurrounding air when the tower vibrates in the SS direction.The same conclusion can be found for the higher modeswith a nacelle and tower motion in, respectively, the SSdirection (mode 4: SS2N) and FA direction (mode 5:FA2N), the latter having a lower damping value. Mode 3(FA2) clearly has the lowest damping. This can beexplained both because of the pitch-angle of the blades andthe fact that for this mode the nacelle is almost completelyfixed (see Figure 11). Thus the contribution of the

the monitoring period


Fig. 8 Boxplots of damping values of the five dominant modes (mode 1: FA1, mode 2: SS1, mode 3: FA2, mode 4: SS2N and mode 5: FA2N)during monitoring period for low wind speeds of 0–5 m/s (left) and higher wind speeds of 10–15 m/s (right)On each box, the central mark is the median, the edges of the box are the 25th and 75th percentiles, the whiskers extend to the most extreme damping values thatwere not considered as outliers and the outliers are plotted individually

www.ietdl.org

aerodynamic damping to the overall damping can beneglected for this mode.Comparing the damping values between the low

wind-speed bin and the higher wind speed bin we can notean overall increase of the damping values for all modes, butmainly on the first FA and SS bending mode. This can beattributed to the higher drag forces in case of higher windspeeds resulting in higher aerodynamic damping values.The relative difference between the different modes, for thetwo wind bins, stays the same. This indicates again theconsistency of the obtained results and gives confidence inthe estimated values.We can conclude that aerodynamic damping is a dominant

damping mechanism for those tower modes where the nacellehas significant motion. This also explains the higher scatter onthe identified damping values of the first FA and SS modeduring the monitoring period, as they have a greaterdependency on the changing wind speeds. The aerodynamicdamping is a minor contributor to the overall damping ofthose tower modes where the nacelle is almost not moving.For these modes the additional damping is dominant.Note that during this monitoring period a TMD was

activated and tuned on the first FA and SS mode. In [7, 8]both an overspeed test and an ambient vibration test havebeen performed on the same wind turbine, while the TMDwas deactivated. The damping values at that time for thefirst FA-mode and first SS mode and a wind speed of 4.5m/s were found to be, respectively, 1.05 and 1.27%. On thesame day several datasets, while the wind turbine wasparked with the TMD not active and with low wind speedsaround 4 m/s, have been analysed using operational modalanalysis. Damping values for the FA mode varied between0.8 and 1.2%. This corresponds well with the suggested

Table 1 Results of the continuous monitoring

Mode Mean freq,Hz

Std freq,Hz

Mean damp,%

Std damp,%

FA1 0.361 0.004 1.86 0.85SS1 0.366 0.005 2.49 0.97FA2 1.201 0.006 0.72 0.22SS2N 1.449 0.018 1.38 0.33FA2N 1.560 0.016 1.14 0.49


values in [3]. Note that the measurements are only able togive values for the overall damping. It is impossible toidentify the different damping contributions independently.At best one can separate the aerodynamic damping form theadditional offshore damping by comparing the low windcases with the higher wind cases. Recently we proposed totune and select the different damping contributions in asimulation model in such a way that the overall dampingestimated using time-domain simulations would be inagreement with the overall measured damping. Thisapproach allowed us to have a rough estimate of thedifferent damping contributions. A detailed discussion onthis approach of choosing and evaluating the dampingcontributions by tuning a simulation model has beenprovided in [25].Table 1 synthesizes the results of the continuous

monitoring routines in the analysis of 14 days of data. Thehigher standard deviations on the frequencies of the highesttwo modes show that, in the case of this offshore windturbine, the influence of ambient conditions such as thetidal effect, is higher on these modes. For the dampingvalues we can see a higher standard deviation on the firsttwo modes, for reasons already discussed above.

4 Conclusions

This paper presents the first of the results of a long-termdynamic monitoring campaign on an offshore wind turbine.The processing algorithms included automaticpre-processing and online identification dynamic parametersof the wind turbine. The proposed method proved to bevery successful in the identification and tracking of theresonance frequencies and damping values of the five mostdominant modes during the monitoring period. The resultsachieved during 14 days, which involved the analysis of2016 datasets while the wind turbine was in parkedconditions, show its ability to identify estimates with highaccuracy, allowing to track both the closely spaced FA andSS modes and to detect small variations in the frequenciesand damping values because of the changing ambientconditions. It provided meaningful damping values andcoherent mode shapes, allowing us to get better insights inthe damping mechanisms. It is part of on-going work tobetter understand the different damping contributions


www.ietdl.org
present in offshore wind turbines. An approach based ontuning the different damping contributions in a simulationmodel has recently been proposed. Future studies will beconducted to understand the effect of the differentoperational and ambient conditions on the identifiedparameters. The on-going work is to evaluate the proposedmethods on the datasets obtained while the wind-turbine isoperating. This will allow us to get better insight in theoverall dynamics of an offshore wind turbine in all itsdifferent operating conditions. It is believed that thesefindings can help to achieve a more cost-effective design bya more correct dynamic modelling during the fatigueassessment in the design-phase of future offshore wind farms.
5 Acknowledgments

This research has been performed in the framework of theOffshore Wind Infrastructure Project (http://www.owi-lab.be). The authors acknowledge the Fund for ScientificResearch – Flanders (FWO). The authors also gratefullythank the people of Belwind NV for their support before,during and after the installation of the measurementequipment.

6 References

1 Van Der Tempel, J.: ‘Design of Support Structures for Offshore WindTurbines’, PhD T.U. Delft, 2006

2 http://www.marin.nl/web/Events/Events-2010/Presentations-Offshore-Wind-Seminar-2010.htm, 2010

3 Germanischer Lloyd Windenergie: Overall damping for piled offshoresupport structures’. Guideline for the Certification of Offshore WindTurbines, (Edn.), 2005

4 Tarp-Johansen, N.J.: ‘DONG Energy, 2009: Comparing sources ofdamping of cross-wind motion’. EOW 2009, Stockholm, Sweden,September, 2009

5 Damgaard, M., Andersen, J.K.F., Ibsen, L.B., Andersen, L.V.:‘Cross-wind modal properties of offshore wind turbines identified byfull scale testing’, J. Wind Eng. Ind. Aerodyn., 2013, 116, pp. 94–108

6 Hackell, M.W., Rolfes, R.: ‘Monitoring a 5 MW offshore wind energyconverter condition parameters and triangulation based extraction ofmodal parameters’, Mech. Syst. Signal Process., 2013, 40, (1),pp. 322–343

7 Devriendt, C., Jan Jordaens, P., De Sitter, G., Guillaume, P.: ‘Dampingestimation of an offshore wind turbine on a monopile foundation’.EWEA 2012, Copenhagen, Denmark, April 2012

8 Devriendt, C., Jan Jordaens, P., De Sitter, G., Guillaume, P.: ‘Dampingestimation of an offshore wind turbine on a monopile foundation’(Published in IET Renewable Power Generation, 2013)


9 Carne, T.G., James, G.H.III: ‘The inception of OMA in the developmentof modal testing for wind turbines’, Mech. Syst. Signal Process., 2010,24, (5), pp. 1213–1226

10 Chauhan, S., Tcherniak, D., Basurko, J., et al.: ‘Operational modalanalysis of operating wind turbines: application to measured data’,Proc. Society for Experimental Mechanics Series, 8, 2011, Vol. 5

11 Hermans, L., Van Der Auweraer, H.: ‘Modal testing and analysis ofstructures under operational conditions: industrial applications’, Mech.Syst. Signal Process., 1999, 13, (2), pp. 193–216

12 Brincker, R., Zhang, L., Andersen, P.: ‘Modal identification of outputonly systems using frequency domain decomposition’, Smart Mater.Struct., 2001, 10, pp. 441–445

13 Cauberghe, B.: ‘Applied frequency-domain system identification in thefield of experimental and operational modal analysis’. PhD thesis, VrijeUniversiteit Brussel, Belgium, 2004

14 Guillaume, P., Verboven, P., Vanlanduit, S., Van der Auweraer, H.,Peeters, B.: ‘A poly-reference implementation of the least-squarescomplex frequency domain-estimator’. Proc. IMAC XXI, Kissimmee,USA, February, 2003

15 Peeters, B., Van der Auweraer, H., Guillaume, P., Leuridan, J.: ‘ThePolyMAX frequency-domain method: a new standard for modalparameter estimation?’, Shock Vib., 2004, 11, (3–4), pp. 395–409

16 Tcherniak, D., Chauhan, S., Hansen, M.H.: ‘Applicability limits ofOperational Modal Analysis to operational wind turbines’. Proc.IMAC XXVIII, Jacksonville, USA, February, 2010

17 Devriendt, C., Elkafafy, M., De Sitter, G., Guillaume, P.: ‘Continuousdynamic monitoring of an offshore wind turbine on a monopilefoundation’. Int. Conf. on Noise and Vibration Engineering, ISMA2012, Leuven, Belgium, 2012

18 Cauberghe, B., Guillaume, P., Verboven, P., Vanlandult, S., Parloo, E.:‘The influence of the parameter constraint on the stability of the polesand the discrimination capabilities of the stabilisation diagrams’,Mech. Syst. Signal Process., 2005, 19, (5), pp. 989–1014

19 Magalhaes, F., Cunha, A., Caetano, E.: ‘On line automatic identificationof the modal parameters of a long span arch bridge’, Mechan. Syst.Signal Process., 2009, 23, (2), pp. 316–329

20 Verboven, P., Parloo, E., Guillaume, P., Van Overmeire, M.:‘Autonomous structural health monitoring. Part 1: modal parameterestimation and tracking’, Mech. Syst. Signal Process., 2002, 16, (4),pp. 637–657

21 Vanlanduit, S., Verboven, P., Guillaume, P., Schoukens, J.: ‘Anautomatic frequency domain modal parameter estimation algorithm’,J. Sound Vib., 2003, 265, (3), pp. 647–661

22 Verboven, P., Cauberghe, B., Parloo, E., Vanlanduit, S., Guillaume, P.:‘User-assisting tools for a fast frequency-domain modal parameterestimation method’, Mech. Syst. Signal Process., 2004, 18, (4),pp. 759–780

23 Maia, N.M.M., Silva, J.M.M. (eds.): ‘Theoretical and experimentalmodal analysis’ (Research Studies Press Ltd., England, 1997)

24 Hansen, M.H., Thomsen, K., Fuglsang, P.: ‘Two methods for estimatingaeroelastic damping of operational wind turbine modes, fromexperiments’, Wind Energy, 2006, 9, (1–2), pp. 1–191

25 Shirzadeh, R., Devriendt, C., Bidakhvidi, M., Guillaume, P.:‘Experimental and computational damping estimation of an offshorewind turbine on a monopile foundation’, J. Wind Eng. Ind. Aerodyn.,2013, 120, pp. 96–106


monitoring resonant frequencies and damping values of an offshore wind turbine in parked conditions

Documents