OWID Offshore Wind Design Parameter

T. Neumann; DEWI Wilhelmshaven S. Emeis; IMKIFU des FZK K. Grigutsch; DEWIOCC V. Riedel; DEWI North America Inc. M. Trk; KEMA Consulting GmbH

T. Neumann

ENGLISH

Summary In the research project OWID a joint approach has been undertaken to check design standards for offshore wind tur bines with the analysis of measurements, computational fluid dynamics (CFD) modelling and load calculations using a multi body program (FLEX 5). More than four years of FINO1 mast data perform a sound basis to investigate long term and short term load cases of the design basis according to the IEC standards. Meteorologically the data offer an unprecedented view on the vertical struc ture of wind and turbulence in the marine boundary layer. Results on the connection between wave height and wind speed, the vertical structure of the marine boundary layer and the frequency distribution of the power law exponent, the frequency of inversions, and on different numbers charac terising turbulence and extreme wind conditions are shown. A major part deals with a check of the meteorological assumptions made in the load cases described in the IEC 614001 and IEC 614003. This check shows that not all assumptions are conservative. This is especially true for power law exponent and the 90th percentile of the turbu lence intensity.

While only a few measurements exist for offshore wind farms, the CFDapproach is able to model the additional tur bulence and mean wind speed effects that are caused by a surrounding wind farm. A transfer matrix gives the relative share for each wind turbine, wind speed bin and turbulence bin for the FINO1 site. This transfer matrix scheme has been used to generate input for load calculations in the multi body program. Vertical Structure of the Marine Boundary Layer (MBL) Main characteristic of the sea surface is its varying roughness that is enlarged with increasing wind speed. Additionally, the roughness elements (waves) are moving. Therefore the ratio between their speed and the wind speed (the socalled wave age) is an influencing factor, too. The wave height has a con siderable influence on the vertical structure of the marine boundary layer (Fig. 1, [1]). Mean Profiles The analysis of the mean data e.g. delivered wind speed pro files, Hellmann exponents (Fig.2) and estimations for 50year extreme wind speeds. Here we will only focus on the power

DEWI MAGAZIN NO. 35, AUGUST 2009

73

Fig. 1:

Vertical structure of the marine boundary layer as function of the wave height (from [1]).

Fig. 2:

Hellmann exponent as function of wind speed.

Fig. 3:

Observed variation of turbulence intensity with 10 min mean wind speed at 90 m at FINO1 in the period September 2003 to August 2007.

Fig. 4:

Measured 90th percentiles of the turbulence intensity for four heights (30, 50, 70, and 90 m, red lines) compared to the parameterisation used in the IEC 614003 (dotted black lines).

law exponents (Hellmann exponents) for the profiles which are clearly dependent on wind speed and the thermal strati fication of the MBL. For a complete description of the mean profiles please refer to [2]. For small wind speeds and unstable stratification the expo nent is small (below 0.1) but for large wind speeds around 25 to 30 m/s we find mean exponents of 0.13. For stable stratifi cation mean values of 0.19 are observed, which is considera bly beyond the value of 0.14 as assumed in the IEC 614003. Turbulence of the undisturbed MBL Large efforts have been made to analyse the turbulence in the MBL. Not only has the mean turbulence intensity (u/u) been analysed but also the size of turbulence elements and their inclination was investigated. Further, in addition to these mean parameters that characterise the turbulence, it was looked at the strength of temporal variations of isolated tur bulence elements (such as Mexican hats). Fig. 3 shows the variation of the turbulence intensity with the mean wind speed. For low wind speeds (below 10 to 12 m/s) considera ble parts of the turbulence are due to thermally induced tur bulence (convection in cold air over a warm sea surface). Above 12 m/s turbulence intensity is increasing again with wind speed due to the increasing roughness of the sea sur face (higher waves). For values of the estimated extreme

wind with a 50year return period (this is about 40 to 42 m/s from the present data) we expect from extrapolation of the data in Fig. 3 a turbulence intensity of about 0.10. Fig. 4 shows the measured 90th percentiles of the turbulence intensity in four different heights at FINO1. It turns out that there are three wind speed ranges within which the parame terisation used in the IEC 614003 is not satisfying. One range is below 7 m/s where the parameterised values are far below the observed values. Another range is above about 20 m/s where the parameterised values are slightly below the observed values. Here, obviously the increase of turbulence due to higher waves is somewhat underestimated. A third range is around the most frequent wind speeds around 12 m/s. Here, the parameterisation, which is based on Prandtl layer theory, delivers too high values because the measure ment heights (30 to 90 m above sea level) are already within the Ekman layer (the layer above the Prandtl layer) which usually exhibits lower turbulence than the Prandtl layer. A first suggestion to change this parameterisation has been made in [2]. Based on this, we propose here 2VTi ,min Vhub u ,s a (1,44m / s ) I 15 bVhub ln( z hub / z 0 ) Vhub This leads to a better approximation in the whole range of possible hub height wind speeds. The first term copies Prandtllayer theory, the second term deals with the increase

Fig. 5:

Measured 90th percentiles of the turbulence intensity for 90 m height (full line) compared to the modified parameterization proposed above (dotted line) with a=.63, b=.0012, I15=4.9%, Vti,min=12 m/s.

Fig. 6:

Example of a Mexican hatshaped gust at 80 m above sea level (upper curve). This gust is not present at 40 and 60 m height (lower two curves).

Fig. 7:

Amplitude statistics for Mexican hatlike turbulence gusts with 10.5 s duration at 80 m height (440 cases in the year 2005).

towards low wind speeds, and the third term raises the curve for high wind speeds. Scanning the 10 Hz data has yielded examples (Fig. 6) and statistics (Fig. 7) for Mexican hatshaped turbulence struc tures (gusts) in the wind speed time series. Fig. 6 shows an example of such a gust which in this case is only observ able at 80 m height. There are other cases where such gusts appear in all three heights. A statistical evaluation shows that Mexican hats with smaller duration (e.g. 8 s) are 1.6 times more frequent than Mexican hats with 10.5 s duration (the assumption which is made in the IEC 61400). Such gusts with even longer duration (e.g. 14 s) are even less frequent (63% compared to the 10.5 s gusts). Fig. 7 shows that Mexican hats do not need to be upright. Actually, the statistics reveal that about 57 % of all hats are negative hats (i.e. they looked like hats turned upside down). The turbulence length scale is slightly increasing with height at FINO1. Mean values of 249 m (40 m), 280 m (60 m), and 302 m (80 m) have been found (Fig. 8). The inclination of the large majority of the turbulence elements (Fig. 9) is forward. Fig. 10 shows the frequency distribution of the maximum wind direction change with time intervals of 6, 10, and 14 s from 10 Hz data at FINO1. The mean value for 6 s is 10.7 degrees, for 10 s it is 11.3 degrees and for 14 s it is 11.7 degrees. This behaviour is well described by the EDC model of the IEC 61400.

CFD-Model for the Main Wind Characteristics in Offshore Wind Farms In the previous chapters we described the results of the FINO1 data assessment of the mean and turbulence values of the undisturbed offshore wind flow. To incorporate the effects of surrounding wind turbines to the properties of the wind field a Computational Fluid Dynamics (CFD) model has been used. The OWID Wake model uses the commercial flow simulation software version Phoenics 3.4 as numerical core. The model is used in parabolic mode, which means that the calculation advances downstream only. The variables in the model used are the wind speed, the turbulent kinetic energy k and the dissipation of the turbulent kinetic energy . The wind farm area and a one kilometre buffer around it is covered with a cartesian grid with a resolution of 10m in all three spatial directions. In the vertical direction the grid spacing is increased above 300m (over sea level) up to the top of the model at 1000m altitude. A 10m resolution in the relevant area is sup posed to be sufficient to resolve the relevant effects on wind turbines within this model. In a CFD run the mean wind speeds and the corresponding turbulence intensities at each wind turbine are calculated. The effect of the wind turbines on the flow field is dependent on the hub height, the rotor diameter and the Ct curve. The

DEWI MAGAZIN NO. 35, AUGUST 2009

75

Fig. 8:

Frequency distribution of the turbulent length scale in three different heights (40, 60, and 80 m).

Fig. 9:

Frequency distribution of the inclination of the turbulence ele ments between 40 and 60 m height at FINO1. Positive values mean forward inclination (i.e. the upper edge arrives first), nega tive values indicate a backward inclination. (The gap at zero degree inclination (perfectly upright elements) is due to the limited temporal resolution (10 Hz) of the available turbulence data.)

Fig. 10:

Frequency distribution of the maximum wind direction change for three different time intervals (6, 10, and 14 s).

reduction of wind speeds at the rotor level is determined approximately from the ctcurve and the actuator disc model. Wake meandering has been incorporated by super imposing shifted versions of the flow field and according to a twodimensional Gaussian distribution. Validation of the Model The consistency of the model results were checked with measurements that were taken from a flat terrain onshore site and also with data that was provided by the Horns Rev wind farm operator. Onshore comparison shows that for the average wind speed, both the depth and the width of the wake are well predicted. The increase of the turbulence intensity as a function of the wind speed is in good accor dance in the relevant range. The comparison with Horns Rev data shows also good agreement. The steady decrease of the energy yield along a series of wind turbines is well repro duced [3]. Wind Farm Results The geometry of the modelled example wind farm is a simple rectangular grid with the rows heading eastwest and the columns heading northsouth. The spacing between the wind energy converters is seven rotor diameters in the directions of the rows and columns. The hub height is 90m. The position

of the wind farm is supposed to be near the FINO1 platform, therefore FINO1 [4] wind data is used directly as input for the model. For a whole wind farm calculation all relevant wind speeds and wind directions are calculated as described in the previous chapter. These results are combined according to the distributions evaluated from FINO1 measurements and the effective turbulence [5,6,7] is calculated with a Whler coefficient of 10. As a result, it can be seen in fig. 12 that the highest effective turbulence intensity values (>11%) within the wind farm appear slightly offcentre which is consistent with the dominant wind direction south west. Lowest values of the turbulence intensity (>8%) can be found at the south west corner, where free flow occurs most often for the FINO1 wind distribution. Discussion of the Sten Frandsen Model The Sten Frandsen model [5] has been applied for the same situation and park geometry as in fig. 12, the result is plotted in fig. 13. A direct comparison reveals substantial differences. The most striking difference is the overestimation of the gen eral turbulence intensity, especially in the northeast corner, opposite to the main wind direction. Lowest values can be found in the centre which is also in opposition to the flow model results. In general however the effective turbulence is found to be 34% higher than the one calculated with the flow model.

Fig. 11:

Example result for the increase of the turbulence intensity in an offshore wind farm.

8000

8000

11.2
7000 7000

11
6000

14 13.9

10.8 10.6

6000

13.8 13.7

5000

10.4 10.2

5000

13.6 13.5

4000

10 9.8 9.6

4000

13.4 13.3

3000

3000

13.2 13.1

2000

9.4 9.2

2000

13

1000 1000

2000

3000

4000

5000

6000

7000

8000

1000 1000

2000

3000

4000

5000

6000

7000

8000

Fig. 12:

Effective Turbulence calculated with the CFDmodel generated for the wind speed distribution of FINO1. 5MW class offshore wind turbines have been used.

Fig. 13:

Effective Turbulence calculated with the Frandsenmodel gen erated for the wind speed distribution of FINO1. 5MW class offshore wind turbines have been used.

According to the flow simulations in this project the Frandsen model can be considered to produce conservative results even in the case of large offshore wind farms. However, we regard the Frandsen model, as described in the IEC 614001 Ed.3 [6] to be not sufficiently defined. For a better reproduc ibility and applicability of the method we propose to use the turbines Ct curve instead of an approximation, consider wind speed losses in the wind farm, improve the definition of a neighbouring wind turbine for non rectangular wind farm layouts and to omit the Whler averaging in order to split up envi ronmental and machine related parameters. Wind Turbine Fatigue Loads For the constructive design of a wind turbine the load compo nents in between the blade system, the nacelle system and the tower system are most relevant. The fatigue load analysis for these systems is carried out with stochastic wind fields. In accordance with IEC these are generated for a representa tive number of wind speed classes with the software tool GenTurbWind. The generated wind fields are vector fields of time and space and are used to generate an external force on the moving rotor that is controlled according to the imple mented control scheme. A common simulation time is 600 seconds. The time dependent load series are classified and counted in a post processing procedure (RFC). It was one

main goal of the project to check how a more realistic approach compared to the simplified assumptions in the IEC [6,7] change the results for the fatigue loads. Some of the main calculation results are given below. Load Comparison for Height Exponent in the iEC a wind profile with a fixed height exponent = 0.14 is assumed for offshore conditions. We compare the results of this standard approach with calculations that are carried out using the real distribution for the height parameter as indi cated in fig. 14. In fig. 15 the result for the bending moment at the blade root is shown. It turns out that the loads for the real distribution are significantly higher than for the fixed alpha calculations, a 90 % quantile value for the alpha is also not conservative. It was derived by further calculations that a fixed of 0.18 is necessary to achieve comparable fatigue loads in the constant amplitude load spectrum. For all other load components the influence of the chosen fixed (in between 0.140.2) on the load spectra can be neglected. Also the use of the real alpha distribution of FINO1 does not lead to different results. Load Comparison for Turbulence Intensities A similar procedure as in the former chapter has been used to study the validity of the 90% quantile approach of the IEC

DEWI MAGAZIN NO. 35, AUGUST 2009

77

0.30 0.25 0.20

0.15 0.10 0.05 0.00

1000

number of a-values

Hhenexponent a

mean maximum minimum median 10%-quantile 25%-quantile 75%-quantile 90%-quantile number of a-values

100000

Mehrstufenkollektiv Blattwurzelschlagmoment

10000

100

Schwingweite

10 -0.05 -0.10 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
10

v40 (m/s) [Klassenobergrenzen]

alpha = 90% Quantil alpha = 0.14 alpha = 0.15 alpha = 0.16 alpha = 0.17 alpha = 0.18 alpha = 0.19 alpha = 0.2 alpha = "real" 100 110
3

110

110

110

110

110

110

Fig. 14:

Distribution of alpha as found from the FINO1 measurements with different quantile values.

Anzahl Schwingspiele

Fig. 15:

Fatigue load spectrum for different values of height exponent alpha, for 90% quantile and for the real distribution as found from FINO1 measurements.

Fig. 16:

Fatigue load spectrum for different distributions of the turbu lence intensities. The 75% and 90% quantile and the real distribution as found from FINO1 measurements.

versus the use of the real distribution of turbulence intensi ties in the FINO1 observation. Fig. 16 again shows the load spectrum for the blade root bending moment, but for differ ent turbulence intensities. It can be concluded that the 90 % quantile approach is conservative compared to the calcula tions with the real FINO1 distribution. A 75 % quantile assumption is not conservative in all cases but it produces nearly comparable results. This statement is consistent with all load components of the system. Fatigue Loads in the Model Wind Farm The Frandsen model has been compared to the CFDapproach that has been developed in this project. The wind and turbu lence spectra as measured at FINO1 are supplemented by the wind farm effects of the CFD model and than transformed as input for load calculations via a transfer matrix. This transfer matrix gives the relative share for each wind turbine, wind speed bin and turbulence bin for the FINO1 site. We considered several ways to calculate the fatigue loads in the wind farm: The CFD approach with(a1)/without(a2) wind speed reduction and Frandsen approach with approximated (a3) and real ctcurve (a4). A few examples are shown for the blade bending moment for a real 5MW wind turbine in fig. 17 and for the tower top tilt moment in fig. 18.

For the bending moment the load factors in the Frandsen scheme are about 1015% higher according to the CFD model approach. For the tower top tilt moment 1020% higher val ues are found. This holds to the general trend, where it could be found that in the Frandsen model the loads are overestimated by several 10%, especially when using the approximated ctcurve. Conclusion To summarize the following conclusions can be made: The assessment of FINO1data reveals substantial differ ences to the height exponent and the turbulence inten sities that are assumed in the IEC 61400 3 (1) The FINO1data are the basis for the suggestion of an enhanced parameterization for the turbulence intensity as function of wind speed, which leads to a better approxi mation in the whole range of possible hub height wind speeds. The analysis of gusts has shown that Mexican hatlike structures with a duration somewhat smaller than the 10.5 s duration assumed in the IEC 61400 3 (1) are more frequent. CFD Models can be applied successfully for large offshore wind farms. The Frandsen Model leads to systematically higher tur bulences and fatigue loads compared to the CFD approach (especially with approximated Ct).

78

DEWI MAGAZIN NO. 35, AUGUST 2009

Blattwurzel Schlagmoment

Nabe Nickmoment

Fig. 17:

Ratio of the bending moment fatigue loads in Frandsen model (a2)and CFD (a4), m=10

Fig. 18:

Ratio of the tower top tilt moment fatigue loads in Frandsen model (a2) and CFD model (a4), m=10

The 90% quantile approach of the turbulence intensity (IEC) is conservative. For the a bending moment: a height exponent = 0.18 had to be assumed to match the real case loads (IEC = 0.14). OWID has laid the basis for further projects. The findings on turbulence presented here in Section 3 will be further evalu ated in order to enhance the turbulence parameterisation in numerical wind field models in the project VERITAS (PTJ, FKZ 0325060) which has now become part of the joint project OWEA (PTJ, FKZ 0327696) in which the OWID results can be validated in a real case (alpha ventus). Acknowledgements OWID has been funded by the German Federal Ministry for the Environment, Nature Conservation and Nuclear Safety (BMU) within the Project OWID (Offshore Wind Design parameter) via PTJ (FKZ: 0329961) together with ENERCON GmbH, GEWind Energy, Areva Multibrid GmbH and REpower AG

References: [1] Emeis, S., M. Trk, 2009: Winddriven wave heights in the German Bight. Ocean Dynamics, 59, 463475. [2] Trk, M., 2008: Ermittlung designrelevanter Belastungsparameter fr OffshoreWindkraftanlagen. Dissertation Universitt zu Kln. Verfgbar unter: http://kups.ub.unikoeln.de/volltexte/2009/2799/pdf/ Dissertation_Tuerk_Publikation.pdf [3] V. Riedel, T. Neumann, M. Strack, Beyond the Ainslie Model:, 3D Navier Stokes Simulation of Wind Flow through Large Offshore Wind Farms, DEWEK 2006. [4] T. Neumann, V. Riedel. FINO 1 Platform: Update of the Offshore Wind Statistics. DEWI Magazin Nr. 28, February 2006 [5] Sten Trons Frandsen: Turbulence and turbulencegenerated structural loading in wind turbine clusters, Ris National Laboratory, Roskilde, Denmark, 2007. [6] IEC 614001 Wind turbine generator systems Part 1: Safety require ments, 2nd edition 199902 [7] IEC 614003 Wind turbines Part 3: Design requirements for offshore wind turbines, 1st edition CDV

DEWI MAGAZIN NO. 35, AUGUST 2009

79