Ewts Ii

European Wind Turbine Standards II
The book is based on work done in an R&D project in the framework of an ongoing programme of the
European Commission. This programme, called JOULE, supports research and development into
renewable energy technologies. It is managed by the Directorate General XII for Science, Research and
Development. The work done, as published in this book, has been supported by the Commission under EC
contract No JOR3-CT95-0064.
Published by
ECN Solar & Wind Energy

PO Box 1
1755 ZG Petten
The Netherlands
Reproduction is authorised, except for commercial purposes, provided the source is acknowledged.
Realisation: BetaText, Bergen, The Netherlands
II European Wind Turbine Standards II

EUROPEAN WIND TURBINE STANDARDS II
J.W.M. Dekker
J.T.G. Pierik
(Editors)
European Wind Turbine Standards II III

This book contains results of the "European Wind Turbine Standards - II" project carried out within the
JOULE-III RD programme of the European Commission, contract nr. JOR3-CT95-0064.
PARTICIPANTS OF THE EWTS-II PROJECT
CRES, Centre for Renewable Energy Sources (Greece)

CIEMAT, Centro de Investigaciones Energeticas, Medioambientales y Tecnologicas (Spain)
DEWI, Deutsches Windenergie-Institut (Germany)
ECN, Energieonderzoek Centrum Nederland (Netherlands)
Germanische Lloyd (Germany)
NEL, National Engineering Laboratory (United Kingdom)
RNL, Risø National Laboratories (Denmark)
Teknikgruppen AB (Sweden)
Windtest (Germany)
Project administration:
EUREC-Agency, European Renewable Energy Centres Agency
Kapeldreef 75
B3001 Leuven
Belgium
IV European Wind Turbine Standards II

PREFACE
The number of installed wind turbines, world wide as well as in Europe, is increasing rapidly. At the end
of 1997, the total amount of wind power had reached 7600 MW, almost 4800 MW was found in Europe 1.
Wind energy has realized a substantial improvement in system reliability as well as a considerable
reduction in costs over the recent years. The electricity production in regions with high average wind
speeds is now competitive with conventional production methods 2.
The increase in industrial activity in wind energy in turn generates a demand for standardized methods of
design, testing and certification of wind turbines. This demand is partly covered by the activities of
standardization bodies (CEN/CEGELEC, IEC), however a number of bottle-necks in knowledge and
technical harmonization still exists. It was the objective of the European Wind Turbine Standards project
(EWTS), the predecessor of the current project, to remove some of the constraints and bottle-necks and
contribute to the harmonization. EWTS was completed in the beginning of 1996, but during its execution a
number of technical items evolved that still required further investigation. Therefore, a continuation of the
project was defined: EWTS-II, to address items that could not be completed. These items were:
1. load spectra and extreme wind conditions;
2. quantification of failure probabilities;
3. integration of blade tests in design;
4. power performance in complex terrain;
5. site evaluation..
This report presents the results of these investigations.
Furthermore, the previous EWTS project started to establish an organisation of qualified measurement
institutes in the field of wind energy, the MEASNET organisation. MEASNET unifies measurement
procedures of the participating institutes and guarantees qualified measurements and mutual acceptance
among its members. MEASNET was formally established during EWTS-II. A summary on its results is
included in the executive summary. More details on the structure of MEASNET, its in house measurement
procedures and the rules for acceptance are presented in a separate report 3.
We want to express our gratitude to the European Commission and to the national governments for
sponsoring this project and we hope that it has brought the work on wind turbine standardisation a step
further.
Jos Dekker and Jan Pierik,

June 1998,
Petten, The Netherlands
1
International Wind Energy Development: World Market Update 1997. BTM Consult
2
Inaugural speech of Prof. dr. ir. G.A.M. van Kuik, June 1998, Delft
3
Implementation of the Network of European Measurement Institutes, MEASNET
European Wind Turbine Standards II V

VI European Wind Turbine Standards II
LIST OF CONTENTS
PREFACE ………………………………………………………………………. V
EXECUTIVE SUMMARY ……………………………………………………… IX
1. WIND FARMS-WIND FIELD AND TURBINE LOADING …………………… Part 1 Sub A.1
2. COMPLEX TERRAIN AND FATIGUE LOADING …………………………… Part 1 Sub B.1
3. EXTREME WIND CLIMATE EVENTS ……………………………………… Part 1 Sub C.1
4. QUANTIFICATION OF FAILURE PROBABILITIES ………………………… Part 2.1
5. INTEGRATION OF BLADE TEST IN DESIGN ……………………………… Part 3.1
6. POWER PERFORMANCE IN COMPLEX TERRAIN ………………………… Part 4.1
7. SITE EVALUATION …………………………………………………………… Part 5.1
European Wind Turbine Standards II VII

Executive Summary
VIII European Wind Turbine Standards II

Executive Summary
EXECUTIVE SUMMARY
1. LOAD SPECTRA AND EXTREME WIND CONDITIONS

1.1 Introduction
The subproject Load Spectra and Extreme Wind Conditions will generate technical background
information to facilitate the development and improvement of the European Wind Turbine Standards in
the framework of CENELEC.
The objective of this subproject is to provide guidelines for:
• description of the wind condition in wind farms,
• description of the wind condition in complex terrain
• investigation and evaluation of extreme wind conditions.
All the investigations were performed with reference to the IEC-1400-1 standard.
The turbine classifications in IEC-1400-1 are defined in terms of mean wind speeds, turbulence
intensities, and a procedure to determine the extremes. The IEC-1400-1 defines four standard classes I, II,
III, IV and an S class. In the S-class, all the wind field parameters have to be specified by the
manufacturer. The mean wind speeds in class I to IV are 10m/s, 8.5m/s, 7.5m/s and 6m/s, respectively.
The wind speed is assumed to be Raleigh distributed. In addition the standard classes are divided into a
high and low turbulence class A and B.
The turbulence classes are defined to cover an extended range of wind turbine operation, including
operation in wind farms and complex terrain. Even so, the standard states, that the modified turbulence
structure in complex terrain and wake effects must be taken into account in the fatigue analysis of the
wind turbines. However no commonly accepted procedures or guidelines exist for the estimation of the
relevant turbulence parameter for the two above mentioned types of wind turbine operation. Based on the
knowledge of the ambient wind and turbulence distribution, and assumptions of the wind farm
configuration, the presented subproject “Load Spectra and Extreme Wind Conditions” provides guidelines
for estimation of the wake turbulence for wind farm operation. In the same way, the fatigue relevant
turbulence characteristics for complex terrain are investigated. The extreme wind conditions given in the
IEC 1400-1 are evaluated from a theoretical and practical point of view. Guidelines are given how to
estimate the extreme wind conditions from the actual wind distribution and the turbulence intensity level.
1.2 Wind farms-Wind field and turbine loading

When operating under wake conditions, an increase in fatigue consumption of wind turbines has been
observed. The changes in the load patterns originate both from modifications in the mean wind field and
from modifications in the turbulence field. By means of a parameter study, the significant wind field
parameters, in relation to the increased wind turbine fatigue consumption in wakes, are identified. The
analysis is based on a large number of aeroelastic simulations for five significantly different wind turbine
concepts. The effect on selected equivalent loads, originating from realistic perturbations in the selected
parameters, is determined by means of a two level factorial method.
European Wind Turbine Standards II IX

Executive Summary
It has been demonstrated that the deterministic wake deficit as well as the modified turbulence
characteristics (turbulence length scale, turbulence intensity, and coherence decay) all are of primary
importance in relation to fatigue loading in wake conditions. Furthermore the effects of each of the above
parameters appeared to be approximately additive in a fatigue life consumption sense.
Finally a set of simple models has been investigated and subsequently proposed for the quantification of
the identified key parameters. The result is that the wind farm effects are easily included in the IEC 1400-
1 by including wake load cases simply defined by suitable modifications in the parameters presently used
in definition of the conventional load cases.
1.2.1 Conclusions
A procedure for wake load predictions, and its possible inclusion in the IEC-code framework is
established. The overall philosophy is to preserve the existing division in four wind turbine classes, and
simply supplement the existing load specifications with quantitative wake load specifications.
This procedure is limited to single wake situations, but it is supposed to apply also for multiple wake
situations. A detailed investigation has been performed to quantify the wake load situation. The
investigation falls into three phases: a crude identification of key wake wind field parameters, a
quantification of the identified key parameters and finally a detailed key parameter identification based on
predicted wake characteristics.
Applying a two level-factorial method it has been demonstrated that, in the present analysis, 10-minutes
simulation time series are a sufficient representation of the individual load situations to yield significant
results.
The main findings from the investigation are:
• All four investigated parameters representing the wake wind field, wake deficit, wake turbulence,
length scale, and wake coherence—were demonstrated to be significant in relation to increased fatigue
life consumption in wakes compared to ambient conditions.
• It is possible to quantify the relevant wake parameters by use of simple models.
• The fatigue contributions, caused by the investigated four parameters, are shown to behave
additively. This implies, that a simple approximate method, involving only a very limited number of
aeroelastic calculations, can be applied for fatigue estimates taking into account the detailed wind
farm topology and the particular wind turbine concept.
• The effect due to increased turbulence intensity is crucial and usually dominates the fatigue effects
caused by the other investigated parameters with a factor of 2 to 3.
• The different turbine concepts and different load types act differently with respect to parameter
sensitivity. This fact demonstrates the appropriateness of a detailed wake load estimation based on the
philosophy applied here, compared to the “all factors in one parameter” approach previously used.
1.3 Complex terrain and fatigue loading

The aims of this subtask are:
• to investigate the effect of the terrain complexity on the wind field properties that drive the fatigue
loading of wind turbines (like wind shear, Reynolds stresses, etc.).
• to investigate and quantify the impact of the above properties on fatigue loading for machines of
different size and design philosophy.
The first goal is met by analysing the available experimental complex terrain wind databases, as well as by
performing systematic parametric runs for simple, 2-D, configurations including single and tandem hill
X European Wind Turbine Standards II

Executive Summary
layouts. In the latter case, results are presented in terms of speed-up and turbulent kinetic energy
distributions. The second goal is met through parameter identification (PI) procedures applied on both
measurements and computational -aeroelastic- results. “Fatigue influence” matrices are, thus, established
for different machine sizes and concepts. Recommendations are made on how to include the present
findings into the IEC Standards.
When a turbine is erected at a site which differs significantly from the design conditions of the IEC
standard, the fatigue and the extreme loads have to be recalculated. The estimation of the turbulence
parameters at a specific site must be part of the site assessment procedure. If a turbine, designed according
to the standard classes, is intended to operate at a site with a Weibull shape factor k lower than 1.8, the
fatigue load has to be recalculated using the actual Weibull parameters. The actual extreme wind speed
events at the site need to be re-evaluated also.
It is felt that the 16% to 18% turbulence intensity values used in the IEC 1400-1 standard can cover
complex terrain operation with turbulent intensities from 13% up to 15% respectively, without additional
design calculations. It was estimated that this 3% extra turbulent intensity value can compensate a 20 %
increase in fatigue loading due to overall complex terrain effects. When a machine is intended to operate
at a complex terrain site with a turbulent intensity larger than 15%, the fatigue loads have to be re-
evaluated based on the actual values of the Weibull parameters and turbulence conditions at the site.
The wind shear law used in the standards can be maintained, having in mind that a conservative estimation
of complex terrain loading is thus achieved.
Sites with wind inclination more than 20 degrees should be considered within the (S)pecial class context.
A site assessment of a complex terrain site must, as a minimum include the following parameters:
• Weibull scale and shape factors C and k;
• turbulence intensity;
• flow tilt (inclination) angle;
Characteristic values of the remaining wind field parameters can be selected on basis of these values.
1.4 Extreme wind climate events

Critical loads result in responses that exceed the limits beyond which structural damage or irreversible
change may be expected. Thus the designer is interested in the number of times, or the probability, that
large loads and resulting large responses may occur during the life of the system.
In the IEC 1400-1 standard divides the wind regime for load and safety considerations into normal wind
conditions which will occur frequently during normal operation of the wind turbine, and extreme wind
conditions which are defined as having a 1 year or 50 year recurrence period. The extreme wind
conditions are used to determine critical design loads which the turbine must withstand during its lifetime.
These extreme conditions include peak wind speeds due to storms and rapid changes in wind speed and
direction.
The objective of this part of the project is to review the models for extremes of the IEC 1400-1 standard,
in particular to investigate whether the specified values for the extremes correspond to the assigned
probability .
To this end a theoretical approach is adopted by using Extreme Value Theory rather than to perform yet
another analysis of wind speed time series. Although measurements should ultimately be decisive, it is
obviously dangerous to infer statistics of extremes from a necessarily limited set of data.
Using Extreme Value Theory it is shown that, given a certain knowledge of the climatology, it is possible
European Wind Turbine Standards II XI

Executive Summary
to derive statistical characteristics of extreme wind conditions which are believed to be important for wind
turbine design, e.g. the survival wind speed, extreme gusts and extreme wind direction changes. In the
present report the limit distributions of these three extreme wind conditions have been investigated,
assuming that the climatology (the Weibull distribution, the annual average turbulence intensity, the
turbulence model etc.) is defined by the IEC wind turbine classes I through IV. In addition the sensitivity
of the extremes for the shape parameter k investigated.
With the assumptions given above the derivation of the probability density functions (pdf) and the limit
distributions (cdf) for the 10-minute average and 3-second average extreme wind speed is straightforward.
It is shown that, given a certain confidence level, the magnitude of the extreme wind speed strongly
depends on the shape, k, and scale parameter, C, of the parent Weibull distribution. Smaller shape
parameters lead to higher extreme values and vice versa. The influence of the shape parameter becomes
pronounced when k ≤ 1.8.
Comparison with the IEC values shows that the reference wind speed Vref is acceptable for sites with k ≥
1.77, i.e. for most flat terrain sites in Europe. This is also the case for the extreme 3-second average wind
speeds Ve1 and Ve50 (the so-called ‘survival wind speeds’) given in the IEC extreme wind speed model
(EWM). But it should be noted that the annual extreme 3-second average wind speed with a recurrence
period of one year, Ve1, has been redefined to the “annual characteristic largest 3-second average wind
speed.”
If a wind turbine has to be designed for complex terrain (1.4 ≤ k ≤ 1.5, say), then it is recommended to
increase the ratio of the reference wind speed to the annual wind speed to Vref/Vave = 6.6. In that case, the
expressions for Ve50 and Ve1, given by the IEC, still hold.
Assuming longitudinal wind speed fluctuations uT1 and uT2 at the same location but at different instants t1
and t2 to be jointly normal, the conditional pdf f(uT2 - uT1 uT1) is derived and the limit distribution
function is computed through numerical integration. For extreme gusts not only a strong dependence on
the parameters of the parent Weibull distribution is found, but also on the assumed turbulence model, viz.
Von Kármán or Kaimal, which determines the temporal correlation as function of the time lag τ = t2 - t1
and the integral time scale (L/U).
The magnitude of the extreme gust is inversely proportional to the shape parameter k of the parent
Weibull distribution. The gust speed increases for increasing time lag τ until the time lag is greater than
the integral time scale (L/U), then the gust speed hardly changes anymore. Given the same confidence
level, gust values are greater for a starting wind speed at Vrated than for a starting wind speed at Vout.
The values found for the extreme 3-sec average gusts with a confidence level of 98% are much larger than
the values specified by the IEC 1400-1 standard. This was to be expected because experience using the
IEC standard has shown that the Normal Turbulence Model (NTM) leads in most cases to higher loads
than the extreme operating gust, indicating that the extreme operational gust model is too benign with
respect to actual external conditions.
New analysis of the extreme operating gust, performed by a working group of the IEC (TC88/WG7),
based on measurements from the German lowland and from Californian sites, shows a ratio of gust
magnitude, Vgust1, to standard deviation, σ1, of about 4.8, indicating that ‘real’ extremes will yield much
higher ratio’s, close to values found in the present project, which are estimated to vary between 4.4 for
wind turbine class IV and 7 for wind turbine class I. Therefore it must be concluded that acceptance of the
new model of the extreme operating gust for the second edition of the IEC standard, proposed by the
above mentioned working group, would definitely be an improvement.
The derivation of a probability density function fθ(θ) for the wind direction can be found in the literature.
It can be shown that this angular pdf only depends on the reciprocal value of the turbulence intensities and
XII European Wind Turbine Standards II

Executive Summary
the non-dimensional ratio of the standard deviation of the u and v component of the wind (a measure of
the anisotropy of the fluctuations). Because of the non-Gaussian tail of fθ(θ) it cannot be assumed that the
density function f θ1 ,θ2 (θ1 ,θ2 ) is jointly normal and it turns out to be impossible to derive, from first
principles, this density function in closed form. Numerical integration proved to be too costly.
To circumvent this problem an exponential type density function is fitted to fθ(θ) and the extreme value
analysis is performed in much the same way as for the extreme gusts. Given the uncertainties in the
present analysis the comparison with the IEC values is very limited. It is clear, however, that the IEC
values are too benign. Again the newly proposed formulation for the next edition of the IEC standard will
be an improvement.
Finally some attention is given to some rare meteorological phenomena— tornadoes and downbursts—
which quite likely will induce extreme wind loads, but which are not included in the extreme climate
events of the IEC standard.
To conclude:
• Probability density functions and distribution functions have been derived for extreme wind speeds,
extreme gusts and extreme wind direction changes. With these density functions it is not only possible
to assign a probability to certain extreme wind conditions, but it is also possible to construct a
standard where the model for extreme winds and the model for ‘normal’ turbulence and the site
classification are logically related, both mathematically and physically.
• The derived distribution functions have been used to compute the magnitude of extreme wind speeds,
extreme gusts and extreme wind direction changes with a confidence level of 98%. Comparison with
the values given by the IEC 1400-1 standard confirms the observation that the IEC values for extreme
gusts are generally too benign. It is recommended however to re-evaluate the distribution function for
the extreme wind direction change.
• Finally a rare meteorological phenomenon, called a downburst, is described, which should be included
in the extreme climate events of the IEC standard.
2. QUANTIFICATION OF FAILURE PROBABILITIES

Presently, wind turbines are being designed in accordance with deterministic design rules embedded in
standards like the IEC 1400-1 and various national standards and certification criteria. These rules concern
the design of the load carrying components and the design of safety and control systems.
2.1 Target values for structural reliability

For a safe design of load carrying components a margin is introduced between the design value of the
strength and the characteristic value of the load. Over the years various methods to define safety margins
have been used, but nowadays the concept of the partial safety factors is commonly embodied in the
structural design codes. The determination of the magnitude of the partial safety factors can be done
empirically. However, a probabilistic approach is preferred, because in this way uncertainties in the load
and the strength can be considered in a more rational manner. This probabilistic method has been applied
already in other branches of industry like offshore and civil engineering, but is not introduced in the wind
energy branch yet. As the concept of probabilistic calibration of partial safety factors is fairly new in the
European Wind Turbine Standards II XIII

Executive Summary
wind turbine community the principles behind the method were examined as part of the EWTS-I
subproject “Calibration of Safety Factors”. To calibrate partial safety factors based on a probabilistic
method target values for the structural reliability of load carrying components have to be drawn up. This
aspect has been considered in the current subproject by means of a literature study. From this literature
study it is concluded that:
• The Scandinavian countries and The Netherlands seem to be leading in the application of structural
reliability methods.
• The code which is not local to a country, but shall be applied Europe-wide, is the Eurocode 1.
Therefore, it is recommended to apply the safety level of the Eurocode to wind turbines, which would
mean a yearly safety index of 4.7, corresponding to a failure probability of 10-6/year.
2.2 Guidelines for data collection and parameter estimation

Generally, the control systems have to keep the wind turbine within its design limits during normal
operation and have to ensure that electrical energy is generated and supplied to the grid in an efficient
manner. In case of a failure or in case of abnormal external conditions, the control systems in combination
with the safety systems also have to keep the turbine within its design envelope to prevent damage or
unsafe operation. As a consequence of a failure of a safety system, the wind turbine can run into
overspeed and is operating beyond design conditions. This can lead to catastrophic failures like blade
rupture. Scenarios, how a wind turbine can run into overspeed or failed operation, are derived by means of
a qualitative Probabilistic Safety Assessment (PSA). With a quantitative PSA it is also possible to
calculate the frequency of occurrence of for example overspeed. PSA’s for wind turbines have been
carried out successfully in various countries. Because it is expected that PSA’s will be used more and
more, guidelines or recommended practices were formulated as part of the EWTS-I subproject
“Assessment of Wind Turbine Safety; Recommended Practices”.
Because the treatment of data for the quantification was beyond the scope of the EWTS-I project, this
aspect has been considered in the current subproject. Guidelines to collect data for quantitative safety and
reliability analyses have been formulated, based on a limited benchmark study.
As guidelines for data collection and parameter estimation, it is recommended to use as starting basis:
• the guidelines as outlined in the LW 15/75 and NEWECS-45 studies reports; and
• the recommended practices from the EWTS-I project.
In addition to these reports, the following is recommended:

1. The recommendations set out in the LW15/75 and NEWECS-45 studies can be used for design of
specific data collection and parameter estimation. Also, the recommendations of the LW 15/75 study
on the use of generic data are still valid.
2. Preferably, use should be made of standard, proven methods as laid down in reliability handbooks.
Preferably, computerised versions of (generic) databases and existing software for reliability data
analysis should be used.
3. Generic wind turbine data sources can be used if no specific data are available or if there are no time
or resources available to collect design specific data.
4. Generic wind turbine data sources can be used to provide data on a high level, e.g., an estimate of
blade failure frequency to be used in risk studies, or for frequencies of certain initiating events, e.g.,
grid loss.
5. If generic data are used to estimate a particular parameter (e.g., a failure rate), the following should be
documented clearly for each parameter estimated:
• the reference(s) to the source(s) used,
• a discussion why the source is considered valid or why the source is used,
XIV European Wind Turbine Standards II

Executive Summary
• the calculation procedure to arrive at the final estimate if multiple sources are used.
6. If engineering judgement is used to provide an estimate, the reason for this should be indicated.
Engineering judgement should be used to provide an estimate if design specific or generic sources do
not provide an adequate answer.
2.3 Risks to the public

As mentioned above failure of the safety system might lead to catastrophic failures like blade rupture,
which can cause risks to the public. In that way it is of great interest for e.g. authorities, especially in
densely populated areas. To judge whether a risk is unacceptable, limits have to be indicated above which
the risks become unacceptable (maximum permissible levels).
From a literature study it is concluded that risk criteria related to the risk to the public, applicable for
wind turbines are:
• the individual risk and
• the group risk (as defined in the official Dutch premises for risk management).
For these two criteria, the following target values are recommended:
Individual risk
For the use of the individual risk a risk target level of 10-5/year is recommended. The individual risk is
independent of the site of the wind turbine (park) and can be calculated without knowing the exact
location of the installation. Note that the term ‘risk limit’ is not used but instead the term ‘risk target’,
because risk limits are not commonly used in all EU countries.
Group risk
The target value for the group risk is that the likelihood of an accident with 10 deaths occurring should not
exceed one in every hundred thousand years (10-5/year). The aim of setting target values for group risk is
to prevent social disruption (the death of a group of people all at once). Accidents with even more serious
consequences lead to correspondingly greater degrees of disruption. It is therefore assumed that an n-times
larger impact than 10 deaths should correspond with an n-squared smaller probability of such an accident
occurring. The actual group risk is expressed as a so-called Complementary Cumulative Distribution
Function (CCDF).
2.4 Final recommendations

The main recommendations resulting from this EWTS-II subproject are summarised as follows:
• The guidelines formulated as part of this EWTS-II subproject can be an aid to collect data and to use it
wisely for the purpose of safety and reliability studies of wind turbine designs.
• Qualitative safety and reliability studies are recommended very strongly as a minimum, as these
studies result often in the identification of design errors (e.g., no or too less redundancy, no defence
for certain events). Quantification, even with only the use of generic data sources, often results in a
good ranking of the important event sequences, component failure modes, and systems. A
combination of generic data and engineering judgement can always be used, although the reliability
analyst should be aware of the limitations. In this way, a design can be made (more) balanced.
• Quantitative safety and reliability studies should be used primarily for the ‘relative’ (qualitative)
purposes outlined above. The use of absolute figures and absolute criteria should be avoided.
European Wind Turbine Standards II XV

Executive Summary
3. INTEGRATION OF BLADE TEST IN DESIGN

3.1 Introduction
The objective of this part of the project is to develop a recommended methodology to include full-scale
blade tests in the wind turbine design process. Actual test and measurement procedures do not need to be
addressed here, especially because they are in development under the co-ordination of a working group of
IEC TC-88WG8.
During the performance of the work a continuing discussion has been held whether the general objective,
the deliverables and the activities to be performed are in compliance with each other. Full scale blade
tests, as performed nowadays, are executed to demonstrate adequate safety margins and to verify design
calculations. This means that the performance of mechanical tests on rotor blades is already a fact today,
although the tests are usually performed at the end of the design process. Consequently the test results are
not used to upgrade the design, unless the experiments showed that the design criteria are not met. So the
tests are performed to demonstrate, to a reasonable level of certainty, that specified limit states of the rotor
blades are not reached, provided that they are produced according to the specifications. In other words that
the rotor blade will survive the design load spectrum. The usefulness of these tests for the designer to
improve the design formed the main discussion point within this EWTS II subproject.
Use has been made of the reports and guidelines of the SFAT and IEC-TC88 WG8 documents.
The deliverables of the work as reported, contain information concerning recommended methodologies to
include blade tests in the wind turbine design process in order to, demonstrate adequate safety margins and
to verify design calculations.
3.2 Technical description

General
The full scale tests as discussed here, are split up in (blade) property tests, i.e. the eigenfrequencies, mass
etc. , strength tests, and fatigue tests.
Property tests
These tests are intended to check whether properties, like mass, centre of gravity, eigenfrequencies and
static strength are (almost) equal to the values assumed in the load spectrum calculations. When they
differ to much, the loads used to verify the structural design are not valid any more. The actual allowable
margin, between measured values and the values assumed in the aeroelastic calculations can easily be
determined using calculations in which the most critical parameters are varied some percents up or down.
When the most important statistics, such as maximum, minimum and standard deviation, or for example a
fatigue equivalent load range, differ a lot, the sensitivity of that parameter is large and consequently the
allowable margin is small.
Mass and centre of gravity are the most easiest measurements to perform. The values give a very good
indication whether the blade has been produced according to the specifications.
Stiffness is very important, it indicates whether tower clearance is sufficient and together with the mass
distribution it provides the input for the calculation of the natural frequencies. The stiffness distribution is
not that easy to determine, especially when the lay-up of the laminates across the cross-section is not
symmetric and coupling between deformation modes has to be taken into account.
Natural vibration tests. Results are very important to verify whether the aeroelastic calculations have been
XVI European Wind Turbine Standards II

Executive Summary
performed on an accurate model of the wind turbine (rotor blades). The eigenfrequencies are not that
difficult to determine, however the actual mode shape is more difficult to measure. Differences compared
with the assumed values should be consistent with the differences between assumed mass and stiffness
distribution.
Static tests
The static test results are used to verify whether the structural design is sufficient. This test can be
performed in a number of selected directions.
When it is the intention to go to ultimate load or failure load one has to be very careful which direction is
tested first. The most likely direction to fail first should be tested last.
The results, usually strain gauge readings on a number of selected spots give a fair indication whether the
structural design is accurate enough to predict the strains/stresses. Strain/stress concentrations are difficult
to determine due to the fact that only a limited number of strain gauges will or can be monitored.
When failure occurs at an early stage in the test, a careful examination of the failure and the (test) loading
compared to the design loading, has to be made in order to make sure that differences between design and
test loading are not the cause of the early failure.
Fatigue tests
The fatigue test has limited value for the designer, especially when the designer wants to improve the
design from the test results.
The main reason is that the test loading differs much from the design loading, due to the fact that a test is
made up of only 3 .10 6 to 10 7 cycles and the design loading comprises up to 5 .10 8 cycles. For coupon
fatigue tests the scatter in allowable number of cycles is almost a factor of 10. When only 1/3 of that
scatter is present at a full scale test, the results of two or even more tests are needed to obtain clear
conclusions, except when a failure occurs early in the fatigue test.
3.3 Conclusions
For designers the blade property and static test are the most important tests. The design load spectrum
and structural integrity can be checked to a large extent. The fatigue test which is of less use for the
designer, is however a very valuable test to check all details of design and manufacturing. Errors or
design flaws which remain unnoticed easily during the other tests, will show up clearly in a fatigue test.
Spots with high strains e.g. in corners where strain gauges can not be applied, will crack and this will
indicate that the design requirements are not met.
4. POWER PERFORMANCE IN COMPLEX TERRAIN

4.1 Introduction
It is widely accepted that in spite of the fact that during the last years the wind energy technology and
industry attained an outstanding progress, further research is needed on specific technical and non-
technical issues. The prominent issues are related to system integration, cost-effectiveness improvement as
well as standardisation and certification. All these issues are strongly dependent on the power performance
verification and assessment practices. Recently, several EU funded research projects as well as
standardisation bodies were concentrated on the different aspects of this issue.
European Wind Turbine Standards II XVII

Executive Summary
The research , presented within this EWTS II subproject, is intended to clarify the status of power
performance verification and assessment in complex terrain, putting emphasis on the following items:
• power performance verification for wind turbines operating in complex terrain,
• assessment of developed, applied and verified tools for WECS power performance in complex
terrain.,
• assessment of the available international and national standards.
The major achievements of the project are:

• presentation and assessment of various power performance measurements regarding wind turbines of
different size and control strategy,
• presentation and assessment of alternative power performance verification practices
• identification and quantification of the terrain induced effects on the power performance of wind
turbines,
• assessment of the existing recommendation and standardisation documents and identification of all
inefficiencies that influence the power performance assessment and verification in complex terrain,
• construction of the technical basis for the development of a reliable procedure for power performance
measurements in complex terrain.
4.2 Technical description

The project scientific and technical description and progress is presented for each task separately, as
follows:
Inventory of existing information on the issue of power performance measurements in complex
terrain.
The literature research resulted in many topic-related articles and documents, yet it provided a limited
readily exploitable information for complex terrain measurements. The survey and the resulting
conclusions regarded topography effects, effect of obstacles, site calibration practices, alternative power
curve measurement procedures and site dependent wind characteristics effect on power curve.
Assessment of power performance measurement campaigns and site calibration techniques. Focus
on comparison of different power curve measurements at different sites through annual energy
production estimation and uncertainty estimation for power curve and annual energy production.
Several power performance campaigns have been analysed in order to assess the recommended procedures
regarding site calibration, power curve determination and annual energy production uncertainty
estimation. The analysed data regarded wind turbines of different size and control strategy operating at
different sites, as shown in table 1.1 where an overview of the experimental and analysis work performed
within the project is presented.
In the field of site calibration different techniques were assessed, namely a nacelle cup anemometer at a
downwind parked turbine and the use of a meteorological mast prior wind turbine erection. Moreover the
application of numerical site calibration was judged also.
XVIII European Wind Turbine Standards II

Executive Summary
Table 1.1. Overview of experimental and analytical research within EWTS-II.IV.

Wind turbine Terrain Institute Site calibration AEP Uncertainty Parameter
technique estimation estimation identification
WINCON complex CRES nacelle cup on running WT Yes Yes Yes
110XT nacelle cup on parked WT
(110kW, stall) rotor disk measurements
VESTAS V27 complex CRES nacelle cup on parked WT Yes Yes Yes
(225kW, pitch)
flat RISO reference mast Yes Yes No
flat DEWI reference mast Yes Yes No
flat WINDTEST reference mast Yes Yes No
MADE AE-30 flat CIEMAT reference mast Yes Yes Yes
complex CIEMAT reference mast Yes Yes Yes
NORDTANK complex CRES calibration prior WT Yes Yes Yes
500/37 erection
(500kW, stall) nacelle cup on parked WT
flat RISO reference mast Yes Yes Yes
complex NEL numerical simulation Yes Yes Yes
flat DEWI reference mast Yes Yes Yes
flat WINDTEST reference mast Yes Yes Yes
NEDWIND complex DEWI nacelle cup on running WT Yes Yes No
500kW
ELKRAFT flat RISO nacelle cup on running WT Yes Yes No
1MW
Assessment of power curve evaluation using nacelle anemometers.

The assessment of the nacelle anemometer on running machines was performed on a wide range of wind
turbine sizes ranging from 110kW up to 1MW machines (see table 1.1).
The following conclusions were drawn from this research:
• the correct application of the methodology and the transfer of the calibration formula to other wind
turbines of the same make and type, presupposes that the wind turbine rotor settings, yaw behaviour,
the position on the nacelle anemometer are unchanged as well as that the terrain remains flat.
• the dependency of the calibration formula to wind and wind turbine parameters was found to be
limited. On the other hand these effects are expected to be magnified in cases where wind structure
and wind turbine response at the calibration site compared with the testing site are quite different.
• lower scatter in the measured power curve as a function of the nacelle wind speed may be obtained.
• the introduction of the nacelle cup calibration formula introduces a statistical error that should be
taken into account in the uncertainty estimation.
• The use of nacelle wind speed measurements may decrease he error induced by site calibration
significantly in cases where large discrepancies are encountered between the measured and reference
wind speed.
Parametric analysis of the available power curve measurements for the identification of the site
related effects influencing the power performance behaviour.
European Wind Turbine Standards II XIX

Executive Summary
The parameter identification of wind turbine power performance regarded the quantification of the
dependency of power characteristics on wind deterministic and stochastic characteristics, especially mean
value and standard deviation.
The following issues were clarified by means of this parameter identification procedure:
• identification and quantification of the wind parameters that affect power performance,
• assessment of the sensitivity of power characteristics for wind turbines of different size and control
system,
• assessment of the sensitivity of power characteristics for the same wind turbine type when operated in
different sites,
• assessment of causal parameters in relation to complex terrain characteristics
• assessment of a power curve normalising procedure based on parameter identification results .
An analytic method is introduced covering parameter identification tasks in power performance

measurements. The developed tool is based on multivariate regression analysis with a backward parameter
elimination technique in order to account for the statistically insignificant parameters. The method was
applied for the parameter identification of the power performance of wind turbines operating at different
sites, both flat and complex. The parameters chosen to comprise the model set describe the deterministic
part of the wind, the main turbulent characteristics, the turbulence length scale as well as the wind speed
distribution. The examined dependent variables were the mean power output, the standard deviation of the
power output and in the case of pitch controlled machine the response of the control system. The effect of
the above parameters was identified and quantified. It appeared that except from the mean wind speed
other parameters, related to the deterministic as well to the stochastic wind characteristics are affecting the
power performance of the turbine and should be considered within power performance assessment
procedures. The parameter identification procedure was applied to the following cases: 225kW pitch
regulated, 110kW stall regulated, 300kW stall regulated and 500kW stall regulated. Moreover, a separate
investigation on the sensitivity of power performance on turbulence, wind shear and atmospheric stability
has been performed on a 3MW wind turbine.
Assessment of existing guidelines and their validity for complex sites. Identification of drawbacks
connected to the application of existing guidelines and recommendations for power performance
measurements in complex sites. The assessment was focused on the recommendations of IEC 1400-1
ed.2. Certain inefficiencies, related to site calibration and uncertainty estimation procedures, were
identified.
The assessment of the existing standards and recommendations regarded the following issues:
• status of wind turbine power performance assessment and verification (existing standards, EU project
results, contractual matters, market needs etc.),
• review of characteristics of complex terrain and their effect to power performance ,
• points that are not adequately described by the present IEC 1400-1 ed.2 document, particularly in
relation to complex terrain (site calibration, uncertainty estimation etc.),
• recommendations for performing power performance assessment and verification in complex terrain.
The discussed critical points regarded the following issues:

• test site requirements,
• characterisation of wind speed sensors,
• influence of wind mean and turbulent structure,
• site calibration,
• response system ,
• measurement analysis,
• contractual aspects.
XX European Wind Turbine Standards II

Executive Summary
The derived practical recommendations for power performance assessment in complex terrain followed
from the identified inefficiencies of the IEC 1400-1 ed.2. and regarded the following issues:
• definition of site complexity,
• selection and characterisation of instrument configuration,
• wind turbine control system response,
• site calibration,
• normalisation procedure,
• uncertainty estimation procedure,
• power curve measurements of wind farms.
Drafting of a status report providing a comprehensive overview of the up to date available

information on power performance in complex terrain and of the proposed solutions and results of
the parameter identification.
The final technical report includes the following items:
• introductory review on project background, objective, working methodology and project scientific and
technical performance,
• review on power performance measurements focusing on the effects of terrain orography on wind
field, site calibration, alternative methodologies for power measurements and effects of turbulence and
shear on power curves,
• presentation of the experimental research and analysis performed by the participants within the
project,
• review of the experimental research on alternative power performance methodologies, namely using a
nacelle cup anemometer on a running wind turbine of by using the upwind rotor averaged wind speed
as reference,
• presentation of the parameter identification procedure background and results. Analysis was
performed on all available data bases regarding wind turbines of different size and control strategy,
• review of the assessment of existing guidelines for power performance,
• presentation of practical recommendations and conclusions.
5. SITE EVALUATION
Over the ten to fifteen years that commercial wind farm development has been a reality throughout
Europe, the term “site assessment” or “site evaluation” has generally tended to be associated with the
prediction of the energy yield potential of a site prior to the installation of wind turbines. Hence,
historically, the evaluation has usually been restricted to the assessment of the site annual average wind
speed and wind speed distribution. However, as the economic and design margins on wind turbines and
wind farms become tighter, the need to quantify and minimise all aspects of risk has become more
important. The site evaluation is at the core of this risk evaluation process and consequently needs to
consider more than just energy yield potential if other aspects such as turbine integrity and safety are to be
assessed.
This is addressed in the latest draft edition of the IEC wind turbine safety standard (IEC-1400-1 “Safety of
Wind Turbine Systems”, Edition 2) which necessitates a site evaluation. Although IEC 1400-1 (ed2)
explicitly states what the scope of the site evaluation must be, it does not give guidance on how to do it.
Therefore, the intention of this sub-project has been to provide information and methods which will be of
practical use to those involved in the application of the IEC standard.
European Wind Turbine Standards II XXI

Executive Summary
The approach taken may be summarised as follows:

• a survey of consultants and test organisations offering site assessment/evaluation services in each of
the EC countries taking part in the ongoing project was made in order to determine the various
methodologies used .
• the survey results were collated in a database and interpreted to identify national or other trends.
• with regard to mean wind speed prediction, which is but one aspect of site evaluation, a detailed
review of the results of numerous well documented modelling and measurement based evaluation
techniques was carried out and hence guidelines for the application of the techniques were proposed.
• the more promising new or revised modelling based evaluation techniques for mean wind speed
determination were implemented using various test cases to assess their repeatability and usefulness.
• a detailed investigation of the background principles and theory relating to “other” site wind
parameters was carried out and summarised. Hence a list of instructions for the practical determination
of these other parameters was written, including a summary of instrumentation requirements.
5.1 Survey of European Consultants

Consultants and test organisations in Denmark, Germany, Greece, Netherlands, Spain and United
Kingdom were individually contacted to determine the scope and methods used for site evaluation, with
the emphasis on mean wind speed prediction techniques as it was anticipated that this was the most
commonly assessed parameter. Consequently 71 responses to the survey were received out of a total of
about 100 consultants approached.
5.2 Database
The survey results reveal strong national and topographical trends governing the extent and method of site
assessment. Countries with generally flat or simple terrain rely mainly on modelling or calculation based
techniques whereas the countries with complex terrain primarily use measurement based techniques,
sometimes in combination with modelling for short distance spatial extrapolation of wind speed. In certain
countries the mean wind speed evaluation is carried out as a legal requirement but primarily with regard to
the aspect of energy yield rather than structural integrity. Of the measurement based techniques used,
variations on the Measure-Correlate-Predict (MCP) technique are most common whereas the most popular
model in use is WAsP. Of the other parameters assessed by measurement, turbulence intensity and the
mean shear layer characteristic were typically determined.
As a separate exercise a review of the sources and types of input data available commercially within each
country for use with modelling or MCP type techniques was carried out. The main specifications of terrain
contour maps and long term wind speed data available in each country are summarised.
5.3 Mean Wind Speed Prediction

A comprehensive review of the technical status of modelling and measurement based techniques was
carried out. As dictated by the results of the survey, specific emphasis was placed on WAsP and MCP
with a view to proposing guidelines for use.
In the case of WAsP the main recommendations for successful implementation are:
• both the reference and predicted sites are subject to the same weather regime
• neutral atmospheric conditions prevail
• the surrounding terrain is sufficiently gentle and smooth to ensure mostly attached flows
• the reference data are reliable
• the description of the background roughness length must be as accurate as possible The MCP
technique can be successfully applied where the following conditions can be satisfied:
XXII European Wind Turbine Standards II

Executive Summary
• validated, long-term and short-term concurrent measured reference wind speed data are available at a
reference site within the same climatological zone as the prediction site.
• validated prediction site wind speed measurements are available for a period of at least 8 months and
more typically 12 months or more.
5.4 Mean Wind Speed Modelling Technique Enhancements

Perhaps the major shortcoming in the commercial use of wind flow models (and MCP) is the difficulty in
the quantification of uncertainties. In this report a philosophy for the assessment of uncertainty is
proposed which, although not necessarily easy to apply to wind models, does highlight the components of
uncertainty of relevance. Derivation of a specific parameter (Ruggedness Index) for categorising terrain
has been summarised from recent literature and is proposed as a useful technique for identifying when the
limits of applicability of a model may have been exceeded.
5.5 Other Parameters

Methods for assessing parameters other than mean wind speed are proposed. In each case, the theoretical
background is also presented.
The following parameters have been considered:
• reference wind speed
• characteristic turbulence intensity at 15 m/s
• annual average wind speed distribution
• normal wind profile
• turbulence length scales
• coherence
• standard deviation ratios
• negative gust
Recommendations for instrumentation and measurement requirements are given in each case. As certain
parameters require the use of an ultra-sonic anemometer and as this is still a rather specialised instrument a
review of important points on the use and interpretation of data arising from this device is presented.
It is believed that the information presented in this report will provide useful guidance to those involved in
site evaluations and, especially with regard to the other parameters, collates relevant practical information
and knowledge in one convenient document for the first time. However, more practical experience of
applying the recommended techniques should be gained to determine if all of the proposals are indeed
sufficient.
6. IMPLEMENTATION OF EUREC-AGENCY MEASNET

6.1 Introduction
More than three years ago the six the European institutes CIEMAT, CRES, DEWI, ECN, NEL, RISØ later
joined by WINDTEST, decided to improve their measurement quality jointly in order to avoid any
problems of future mutual recognition. Measurements performed by the institutes, even applying the
existing IEA, IEC and other standards and recommendations, showed remarkable differences in their
results, a situation which is unacceptable in an open international market. To improve this unsatisfactory
situation the above mentioned test centres worked together in the projects “European Wind Turbine
Standards I and II” co-financed by the European Commission and agreed to form a grouping called:
MEASNET, Network of European Measuring Institutes.
European Wind Turbine Standards II XXIII

Executive Summary
During the two projects special attention was given to the anemometer calibration procedure. Due to the
third power dependence of power from wind speed, the main emphasis had to be concentrated on a
accurate anemometer calibration. Specially the problem of the use of different wind tunnels for
anemometer calibration had to be solved. Basis for all performed harmonisation and quality evaluation
work were round robin tests and agreed quality evaluation procedures.
Within the two above mentioned projects, the main task of the project team consisted of creating an organ-
isational structure and of establishing rules and requirements which will guarantee that high quality meas-
urements are carried out by the participants. In effect, the objective of this project was to arrive at the
situation where the measuring institutes are able to perform measurements of equal quality which are
sufficient for the mutual comparison and acceptance.
MEASNET is not restricted to the actual founding members but also open for other institutions as long as
they are independent of industry and fulfil the membership requirements set up by MEASNET. All
member institutes ensure compliance with the agreed measurement procedures by obtaining and
maintaining EN 45001 accreditation. Measurements will not be done by MEASNET but by each
participating institute. Customers therefore have the advantage to contract that MEASNET member for a
measurement, which offers the best commercial conditions.
6.2 Internal Structure and Requirements of MEASNET

6.2.1 Structure of MEASNET
To ensure high quality measurements, uniform interpretation of standards and recommendations as well as
interchangeability of results, the members established an organisational structure for MEASNET
(Fig. 6.1). Within this structure mutual periodical quality procedures for measurements and evaluations
will be performed. The highest tier is the Council of Members. An Executive Board, composed of three
representatives from different member institutes executes the tasks delegated by the Council of Members.
One or more Expert Groups, specialising in certain measurement tasks, advise and support the Executive
Board and the Council of Members on the definition of the measurement procedures. Assessment Teams
are established to perform assessments for the admission of new members and for quality confirmation of
MEASNET members. MEASNET members must be accredited to EN 45001 for the MEASNET approved
measurements.
XXIV European Wind Turbine Standards II

Executive Summary
Council of Members
Executive Board
Expert G roup A Expert G roup B Expert G roup N
Assessment Team A Assessment Team B Assessment Team N
Figure 6.1 Organisational structure of MEASNET
6.2.2 Requirements of MEASNET

In order to ensure generally acceptable, high quality measurements the member institutes shall fulfil the
following requirements:
• legally independent from industry.
• adequate experience in the field of wind energy and wind energy related measurements.
• qualified and experienced measurement staff.
• carry out measurements according to the MEASNET rules and procedures, at least power performance
measurements.
• EN 45001 accreditation of the agreed measurement procedures. If the EN 45000 accreditation system
is not implemented in the home country of the organisation, the agreed alternative acceptance
requirements will be applied*).
• presentation of measurement results according to the MEASNET format.
• accept the policy of co-operation and exchange of information on measurement and evaluation
procedures and on problems arising in measurement campaigns.
• MEASNET members will accept each other results as far as they are carried out according to the
MEASNET procedures.
• MEASNET members will subject themselves to an internal MEASNET quality evaluation
programme.
*) If the EN 45000 accreditation system is not implemented in a country, the MEASNET Council of
Members will decide on alternative acceptance procedures for the candidate organisation.
MEASNET membership is not restricted to the member institutes but is open for other organisations as
long as they are independent of industry and fulfil the membership requirements set up by MEASNET.
The seat of MEASNET is the address of the institute of the current Executive-Chairman of MEASNET.
Actual address for 1997 and 1998 is DEWI in Germany. As this seat will change with the Executive-
Chairman the formal contact address of MEASNET for initial contacts is the seat of the EUREC-Agency
in Leuven.
European Wind Turbine Standards II XXV

Executive Summary
MEASNET c/o EUREC-Agency

Kapeldreef 75
3001 Leuven - Heverlee, Belgium
Tel.:++32-16-281-522
Fax.: ++32-16-281-510
All commercial contacts can be made of course directly with the desired member institutes.
6.3 MEASNET Recognised Measurements

During the course of the two projects, the members of MEASNET agreed on the following measurement
and quality evaluation procedures to be performed under the MEASNET quality criteria:
• anemometer calibration
• power performance
• noise
• power quality.
Special attention was given to the calibration of anemometers as the most crucial part of a power curve
measurement. As an intermediate result of the round robin tests was agreed that not all MEASNET
members are approved for absolute anemometer calibrations. Absolute anemometer calibration is
considered to be a general commercial service offered to customers. The not approved institutes may
perform relative calibrations which allow for own power curve measurements only. In the beginning of
the project it was not foreseen to distinguish between the two calibration methods, but it turned out to be
necessary. Until the end of the project the participants did not obtain an agreement about relative
calibration method due to missing experience.
All agreed measurement procedures except power quality, take into account the final and draft documents
of international organisations, e.g. IEC, IEA and in addition requirements derived from results of related
projects and measurement experiences. The agreed noise measurement procedure is an example for the
necessity of additional MEASNET measurement requirements. The performed round robin evaluation of a
measured data set led, after some necessary correction measures of the different institutes, to comparable
results because the IEC recommendation concerning the evaluation of tonality was improved.
For the power quality measurement procedure a IEC standard is not yet existing. Therefore the members
of MEASNET decided to use the respective German guidelines until IEC recommendations will be
worked out. At the moment Germany is the only country, where such measurements are required by the
utilities. In Germany this guidelines for power quality measurements had been worked out during the last
two years by a group of experts coming from the involved measurement institutes, industry, utilities and
state governments. They include two years of experience gained by the three German measurement
institutes and therefore are already a good starting point for a future international harmonisation within
MEASNET.
A MEASNET member must at least be approved for power performance measurements. This requirement
was established to guarantee that MEASNET members have enough experience and understanding in
wind energy related questions. In the actual status of the project not all founding members have finished
yet the quality evaluation procedure for the different agreed measurements. A member must not be
approved for all measurement types and can be approved for additional measurements at any time he
performed successfully the quality evaluation procedure. The MEASNET measurement approval is valid
for five years as far as the institute passes successfully all internal quality evaluation programmes during
this time and maintains the accreditation to EN 45001. An official MEASNET acceptance document
indicates the measurement types for which the MEASNET member is approved for. Customers should ask
the institute of their choice to present this document.
XXVI European Wind Turbine Standards II

Executive Summary
Measurement reports performed under the quality criteria of MEASNET will be stamped by the measuring
institute with a stamp which consists of the MEASNET logo and, written around the logo, the name of the
institute. As an example the general logo of MEASNET is shown in Fig. 6.2.
Figure 6.2 Logo of MEASNET
MEASNET is now operative and the institutes are allowed to offer their MEASNET approved, high
quality measurements from October 1997 on.
6.4 Performed Quality Evaluations

Several quality evaluations have been made during the two projects. The most intensive one was for
anemometer calibration, because a procedure had to be found, how to judge the quality of the wind tunnel
which should be used for the anemometer calibration. The standard anemometers in use by each institute
were sent around for calibration to each of the MEASNET founding members. One of the evaluation
difficulties was how to define the correct wind speed reference to which the calibration results of each
institute had to be compared. After an additional evaluation of the influence of anemometers under skew
wind flow conditions and after several correction measures a mutually agreed comparison base was found.
As a result of this round robin the institutes decided to have the above mentioned two types of approved
anemometer calibrations, the absolute and the relative anemometer calibration.
Other round robin evaluations were performed for power performance and noise measurements. In both
cases a set of measured data were sent around for evaluation by the individual institutes. Differences in the
evaluation results were discussed and led to correction measures concerning the established measurement
and quality evaluation procedures. All round robin tests showed, that the idea of MEASNET to harmonise
the interpretation of standards and the applied measurement methods is of very high importance.
Guidelines for evaluation, for example stated in the existing IEC standards, even when correctly applied
led to different interpretations of requirements and consequently to differences in the results.
6.5 Conclusion
MEASNET members are interested commercially in performing measurements in the field of wind
energy. In spite of all difficulties encountered during the project the institutes always found a solution
which could be accepted finally by all members. It was of great advantage for the progress of the work
that the experts of the founding members are used to work together in many other joint European research
projects since more than a decade. As a result of their intensive and detailed work, MEASNET members
now mutually accept their measurement results and guarantee high quality by regularly performed quality
evaluation programmes.
European Wind Turbine Standards II XXVII

Executive Summary
The advantages for the industry to contract a MEASNET institute are:

• the measurements are accepted in other countries,
• are of high and comparable quality,
• and can be ordered on the basis of competitive offers from the member institutes.
For the first time measurement institutes work together and were able to find agreed procedures with the
goal to harmonise the interpretation of measurement procedures established in international standards and
recommendations. The different performed quality evaluation tests during the course of the two projects
showed the necessity of such a grouping, if measurements performed by different institutes shall be
comparable to each other. The fact that all MEASNET members are also engaged in the different existing
IEC, IEA and CENELEC working groups will help to integrate the gained MEASNET experience in the
respective international measurement standards and recommendations.
At the moment MEASNET is only a wind energy activity. But the participants are convinced that
measuring institutes engaged in other renewable energies will encounter the same harmonisation problems
when they have to enter a real commercial market situation. Perhaps MEASNET serves as an example for
them of how to work together and solve the harmonisation and quality evaluation problems.
XXVIII European Wind Turbine Standards II

PART 1
LOAD SPECTRA AND EXTREME WIND CONDITIONS
SUB A
WIND FARMS – WIND FIELD AND TURBINE LOADING
G.C. Larsen (ed.) Risø National Laboratory

I. Carlén Teknikgruppen AB
G.J. Schepers Netherlands Energy Research Foundation ECN
Abstract
When operating under wake conditions, an increase in fatigue life consumption on wind turbines has been
observed. The changes in the load patterns originate both from modifications in the mean wind field,
and from modifications in the turbulence field. By performing a parameter study, the significant wind
field parameters, in relation to the increased wind turbine fatigue consumption in wakes, are identified.
The analysis is based on a large number of aeroelastic simulations and it has been performed for five
significantly different wind turbine concepts. For each of these the effect on selected equivalent loads,
originating from realistic perturbations in the selected parameters, is determined by means of a two level
factorial method.
It has been demonstrated that the deterministic wake deficit as well as the modified turbulence character-
istics – turbulence length scale, turbulence intensity, and coherence decay – all are of primary importance
in relation to fatigue loading under wake conditions. The effects of each of the above parameters was
furthermore shown to be approximately additive in a fatigue life consumption sense.
Finally a set of simple models has been investigated which subsequently resulted in a recommendation
for the quantification of the identified key parameters. The recommendation can serve as a resource for
certifying bodies and is further, if it is found appropriate, easily included in the IEC–code framework.
The wake load cases is thus simply defined by suitable modifications in the parameters presently used in
definition of the conventional load cases.
Wind Farms – Wind Field and Turbine Loading
CONTENTS OF PART 1Sub A
1 INTRODUCTION 7
2 WAKE WIND FIELD STRUCTURE 9

2.1 Delimitations : :::::::: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9
2.2 Preliminary parameter study : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10
2.2.1 Reference sets : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 11
Wake deficit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 12
Turbulence intensity : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 12
Turbulence length scale : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 12
Coherence decay factor : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 12
2.2.2 Results : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13
2.3 Wake measurements : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13
2.4 Wake models : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 14
2.4.1 GCL–model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 14
Deficit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 14
Turbulence Intensity : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 15
Turbulence Length Scale : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 17
Coherence decay : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 19
Transversal turbulence : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 21
Turbulence intensity : : : : : : : : : : : : : : : : : : : : : : : : : : 22
Turbulence length scales : : : : : : : : : : : : : : : : : : : : : : : 23
2.4.2 SF–model :::::: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
2.4.3 DR–model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
2.4.4 TNO–model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 28
Turbulence Intensity : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 28
Deficit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 29
2.4.5 MM–model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 29
Velocity deficit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 32
Length scale : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 34
European Wind Turbine Standards II 1Sub A.3

2.4.6 UPMWAKE–model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 34
Velocity deficit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 34
2.4.7 Concluding remarks : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 38
3 WAKE LOADING 41
3.1 Method : :::::::: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 41
3.1.1 Reference sets : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 42
Flat terrain : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 42
Wake : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 42
3.1.2 Factorial method and calculation scheme : : : : : : : : : : : : : : : : : : : : 43
3.1.3 Fatigue determination : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 44
Replicated simulations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 44
3.2 Results and Analysis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 44
3.2.1 Benchmark : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 45
3.2.2 Results : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 46
3.2.3 Discussion : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 54
4 CONCLUSIONS 55
5 RECOMMENDATIONS 57
5.1 Mean Wind Deficit :::::: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 57
5.2 Turbulence Intensity : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 57
5.3 Turbulence Length Scales : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 58
5.4 Coherence Decay Factors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 59
ACKNOWLEDGEMENTS 60
REFERENCES 61
A VESTAS V27 TURBINE 63
B VARIABLE SPEED 500 KW TURBINE 65

B.1 Background ::::::::::::::::::::::::::::::::::::::: 65
1Sub A.4 European Wind Turbine Standards II

B.2 Rated Conditions and Losses : :::::::::::::::::::::::::::::: 67

B.3 Generator curve : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 67
B.4 Pitch control : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 67
C DANWIN 180 KW TURBINE 69
D NÄSUDDEN 3 MW TURBINE 73
E NORDTANK NTK500 TURBINE 77
F EXPERIMENTAL WAKE CHARACTERIZATION 81
G GRAPHIC COLLOCATION OF RESULTS 83


1 INTRODUCTION
An increasing part of the future wind turbines will be erected in wind farms due to the diverted reduction
in grid costs, the limited number of sites with optimal wind potential, and environmental requirements.
Therefore, prediction of the wind induced loads in wind farms is of vital importance.
Measurements have revealed an increased loading of wind turbines sited in wind farm configurations
compared to stand alone turbines, other things being equal. The fatigue loads on a turbine is a combination
of periodic deterministic loads, associated with gravity and mean wind field, and turbulence induced loads.
The increase in fatigue consumption in wind farms originates both from modified mean wind field and
from modified turbulence field. One possible strategy to account for increased wake loads is to collect the
total impact from wake operation in only one parameter – often chosen as the turbulence intensity which
is a parameter experienced to be of significant importance [1]. However, this method has the drawback
that its generality is affected by the fact that it is based on the choice of a specific fatigue evaluation model
and a limited number of predefined Wöhler exponents and wind turbine concepts. Moreover, it excludes
the detailed wind farm topology from being taken into account and fails to reflect, that the induced loads
on different component categories respond individually on the different elements in the total wind load
modification.
Therefore a more detailed description is required, not only for design approval of a specific site, but
especially in relation to optimization of wind turbines suited for wake operation. Consequently it is
of importance to investigate in details the impact arising from wake modifications of the individual
parameters constituting the conventional wind field description.
One of the important conclusions from the "Dynamic Loads in Wind Farms II" project [1]1 is that the
stochastic component of the fatigue loading, caused by wake turbulence, is dominant compared to the
periodic deterministic contribution, and that partial wake operation does not represent a significantly more
severe fatigue loading regime than a full wake operation. Based on this knowledge it has been decided to
(conservatively) limit the present investigation to an artificial situation defined by a partial wake deficit
combined with a full wake turbulence formulation. For decreasing spacing, the degree of conservatism
associated with the partial wake formulation of the deficit is believed to increase.
In order to simplify the task further, only single wake situations are considered and in turn assumed
representative for wake loading in general. Bearing in mind that the turbulence part of the fatigue loading
in a wind farm is dominant compared to the periodic deterministic contribution, the single wake approach
is justified by measurements from the Sexbierum wind farm. Here it has been observed that double
wake situations mainly differ from single wake situations by reducing further the mean wake wind speed,
whereas the turbulence field is only marginally modified2 .
Based on the above described simplifications, a detailed analyses has been performed in order to quantify
the wake load situation. The analysis falls into three phases.
The first phase embraces a parameter study aiming at identifying the key wake wind field parameters
responsible for increased wind turbine fatigue life consumption in wakes compared to ambient conditions.
In this part of the analysis, crude estimated wake characteristics are obtained from relevant measurements
and these are only meant to be of the correct order of magnitude.
1 Final report p. 105–106.
2 Dynamic Loads in Wind Farms II – Final report p. 104.

The second phase deals with a possible quantification of the identified key parameters from the first
phase. This phase embraces collection of significant wake measuring data and a subsequent verification
of simple models predicting the key parameters.
The third and final step is to repeat the parameter study performed in the first phase, with the simple
experimentally based wake characteristics replaced by more detailed predictions from selected models,
and to draw the final conclusions.

2 WAKE WIND FIELD STRUCTURE
When operating in a wake situation, the mean wind profile as well as the turbulence characteristics
experienced by a turbine, are modified compared to the ambient situation.
For relevant turbine spacings, the modified mean wind profile is characterized by adding a bell–shaped
velocity deficit to the a priori wind shear profile associated with the undisturbed flow. The modified
turbulence field is primarily the consequence of creation of turbulence in the free shear layer between
the retarded wake flow and the undisturbed free stream. Hence the modified wake turbulence field is in
general characterized by an increased turbulence intensity, a decreased turbulence length scale and an
increase in the coherence decay. As no significant qualitative modifications in the spectral shape has been
observed, the turbulence description in the present study follows the traditional description of turbulence
in flat and homogeneous terrain3 . Thus, the wind farm effects are condensed only in the modification of
the mean wind shear profile and three parameters defining the turbulence field.
2.1 Delimitations
Basically, the present investigation is associated with wake conditions in flat and homogeneous terrain,
as only turbulence structures typical for those terrain conditions have been considered.
In order to simplify the situation only single wake situations are considered, and these are in addition
assumed to be representative for wake loading in general. This delimitation is closely linked to an
important conclusion from the "Dynamic Loads in Wind Farms II" project [1]4 where it is stated that the
stochastic component of the fatigue failure loading, caused by wake turbulence, is dominant compared to
the periodic deterministic contribution. Bearing that in mind, the approach is justified by measurements
from the Sexbierum wind farm, where it has been observed that double wake situations mainly differ
from single wake situations by reducing further the mean wake wind speed, whereas the turbulence field
is only marginally modified5 .
An other important result, originating from the same report, is that partial wake operation does not
represent a significantly more severe fatigue loading regime than a full wake operation. Based on this
knowledge it has been decided to (conservatively) limit the present investigation to an artificial situation
defined by a partial wake deficit combined with a full wake turbulence formulation. It is assumed that
this artificial wake situation, to a reasonable degree, is representative for fatigue wake loading in general.
For decreasing spacing, the degree of conservatism associated with the partial wake formulation of the
deficit is believed to increase.
The situation is further simplified by assuming the mean velocity field modifications in the mean wind
direction to be of "first order importance", whereas mean wind field modifications related to the two
remaining directions are considered to be of "second order importance". Only modifications of "first
order importance" are considered.
At down stream distances, corresponding to traditional spacings in wind farms, it is assumed that only
the surface– and wake shear mechanisms contribute significantly to the turbulence production. When
the two shear turbulence contributions have scales of comparable orders of magnitude, it is reasonable to
3 The Kaimal turbulence spectrum and the Davenport coherence model are applied in the analysis.
4 Final report p. 105–106.
5 Dynamic Loads in Wind Farms II – Final report p. 104.

expect the resulting spectral shape to be qualitatively invariant, which is also confirmed by measurements.
However, the parameters characterizing the particular spectra will be modified. Modifications related
to turbulence intensity and turbulence length scale for all three turbulence components are considered,
whereas only possible changes in the uu–autocorrelation is investigated. In other words wake aspects
related to auto– and cross correlations, involving any of the v – and w– turbulence components, are
neglected.
2.2 Preliminary parameter study
The present section deals with a preliminary parameter study aiming at identifying the key wake wind field
parameters, responsible for increased wind turbine fatigue consumption in wakes compared to ambient
conditions. In the previous section, the array of parameters, with potential significant impact on wake
fatigue life consumption, has been limited to mean wake velocity deficit, turbulence intensity, turbulence
length, and coherence decay.
In this initial parameter study the situation is further simplified, as the turbulence intensity– and turbulence
length scale wake modifications, associated with turbulence components perpendicular to the mean wind
direction, have been neglected due to an assumed dominating influence from the u–component. To
summarize, the parameters included in the present parameter investigation are :
Longitudinal mean wind field wake deficit (the only parameter with a deterministic character).
Standard deviation of the u–turbulence component expressed in terms of turbulence intensity.
Turbulence length scale related to the u–turbulence component.
Spatial uu–coherence expressed in terms of coherence decay.
Crude estimates of the above wake characteristics are obtained from relevant measurements6, supple-
mented with a numerical determination of the wake deficit form, and at this stage the wake parameters
are only meant to be of the correct order of magnitude.
Basically, the analysis is defined by selecting a number of turbine concepts and turbine spacings. In
order to represent a broad spectrum of different wind turbine concepts and sizes, four fundamentally
different horizontal axis wind turbine types have been considered – a stall regulated three–bladed turbine,
a pitch regulated three–bladed turbine, a three bladed variable speed turbine, and a large two–bladed
pitch regulated turbine. The main data, describing the investigated turbines, are summarized in the
Appendices A–D. For each turbine type, two turbine spacing situations (5 rotor diameter spacing (5D)
and 8 rotor diameter spacing (8D)) have been investigated based on criteria as relevance, representative
situations, and finally that the scenarios are covered by the available measurements.
The basic philosophy in the parameter investigation is then, for each wind turbine concept, to choose
two sets of reference situations – one related to a wind farm configuration with 5D spacing, and an other
related to a wind farm configuration with 8D spacing.
Each set of reference situations contains a (common) reference situation, related to ambient conditions in
flat and homogeneous terrain, and a reference situation related to the wake situation (5D or 8D spacing).
6 The order of magnitude of the selected parameter variations are based on analysis of measurements performed at the Danish
Noerrekaer Enge and Vindeby wind farms.

For each reference set, the effect on selected fatigue loads, originating from first order variations in each
of the investigated parameters (for a number of different mean wind speeds), is subsequently determined
from aeroelastic simulations. For each mean wind speed regime and for each turbine concept, a two level
factorial design method [16] is subsequently used to identify the effect of each parameter. The procedure
is described in more detail in chapter 3.
2.2.1 Reference sets
The defining parameters are given in Table 2.1 and Table 2.2, where I u denotes the turbulence intensity
associated with the turbulence u–component, Lu is the length scale related to the turbulence u–component,
and au is the coherence decay constant in a Davenport formulation of the spatial uu–coherence. The
maximum value of the deterministic wake deficit is denoted Vd . The reference situation corresponding
to ambient conditions has been selected with no mean wind shear. Usually the shear is relatively small
across the rotor plane, and as the present investigation moreover focus only on relative differences, the
simplification is justified for the present purpose.
The wake deficit, the wake turbulence intensity, the wake length scale, and the wake coherence decay
factor depend on the ambient mean wind velocity and on the turbine concept. However, as the quantities
in the tables below are only meant to be of the correct order of magnitude, constant values have been
specified for reasons of simplicity.
The interpretation of the above described parameters, in terms of load case definitions, is specified below.
Table 2.1 Reference set no. 1 (5D spacing).
Ambient conditions Wake conditions (5D spacing)

yaw–error 0 deg. 0 deg.
slope of terrain 0 deg. 0 deg.
V 0 m/s 1.5 m/s
Iu 15 % 18 %
Lu 600 m 300 m
au 8.8 25
Table 2.2 Reference set no. 2 (8D spacing).
Ambient conditions Wake conditions (8D spacing)

yaw–error 0 deg. 0 deg.
slope of terrain 0 deg. 0 deg.
V 0 m/s 1 m/s
Iu 15 % 17 %
Lu 600 m 360 m
au 8.8 25

Wake deficit
The focus is on a (partial) single wake situation, which is assumed to be the most severe case of operation,
although no dramatic deviation compared to the full wake situation is expected. The partial wake situation
to be considered is (rather arbitrarily) defined as the situation where the blade tip passes through the center
of the wake (and thus including the wake deficit part characterized by a large horizontal gradient). It
is presumed, that the deficit is positioned at the side of the rotor giving rise to maximum impact on the
fatigue loads.
The explicit shape of the wake deficit has been based on UPMWAKE7 calculations through an iterative
procedure, where the particular wind turbine thrust coefficient was adjusted until the specified deficit
magnitude was obtained.
Turbulence intensity
The turbulence intensity is always defined with the ambient mean wind speed at hub height, Va , as the
normalizing factor.
Turbulence length scale
The longitudinal turbulence length scale is given with reference to a standard Kaimal formulation of the
turbulence spectrum, S (f ), as formulated in the Danish Code of Practice [18] :
fS (f ) = fLu =Va ;
u2 5=3 (2.2.1.1)
1 + 1:5 fL V
u
a
where u2 denotes the variance of the longitudinal turbulence component, and f is the frequency measured
in hertz.
If the Kaimal formulation given in the IEA 1400–1 document is adopted, the length scales specified in
Table 2.1 and Table 2.2 should be divided by a factor of 4.
Coherence decay factor
The coherence function is, in the present context, defined as the magnitude of the complex cross–
spectral density of the longitudinal wind velocity components, corresponding to two spatial separated
points, divided by the squareroot of the product of the respective autospectrum functions. The standard
Davenport formulation is adopted :

Coh(s; f ) = exp ;au fs
Va ; (2.2.1.2)
where s denotes the magnitude of the projection of the separation vector between the two spatial points
on a plane perpendicular to the average wind direction.
7 A brief description of this code is presented in section 2.4.6.

2.2.2 Results
The main conclusions from the investigation are :
The mean wind deficit, increased turbulence intensity, decreased turbulence length scale as well
as increased coherence decay should all be considered as significant in relation to increased wind
turbine fatigue life consumption in wakes compared to ambient conditions.
The effects on fatigue loading, originating from the investigated parameters, act approximately
additively.
2.3 Wake measurements
In order to form a general view of the available experimentally based wake characteristics, and furthermore
to establish a basis for verifying available wake models against experimental data, it was decided to
establish a database of wake measurements. Each of the project participants have selected at most 15 of
their most suitable measured campains and these are collected in the table presented in Appendix F. The
table contains parameters describing the wake situation (turbine separation measured in rotor diameters,
turbine type, atmospheric stability expressed in terms of the Richardson number, ambient mean wind
speed at hub height etc.) as well as parameters describing the wake wind field. The wake wind field
parameters are defined as follows :
The wake deficit is described in terms of the maximum wake deficit V within the wake.
The ambient turbulence intensity, Iau , is defined as the standard deviation of the horizontal turbu-
lence component (as it is usually obtained from cup anemometer measurements) divided by the
undisturbed mean wind speed Va . Both values refer to turbine hub height.
The wake turbulence intensity, Iwt

u , is defined as the maximum standard deviation of the horizontal
turbulence component within the wake divided by the undisturbed mean wind speed at turbine hub
height.
The ambient turbulence length scale, Lua , is defined as the (Kaimal) length scale of the u–turbulence
component, corresponding that particular frequency, fL , where half of the turbulence energy relates
to frequencies heigher than fL . Lua refers to the turbine hub height.
The wake turbulence length scale, Luw , is defined in analogy with the ambient turbulence length.
Luw refers to the position within the wake where the reduction in the length scale takes its maximum.
The ambient transversal turbulence component, av , refers to the turbine hub height.
The wake transversal turbulence component, wv , refers to the position within the wake where the
maximum value is obtained.
The ambient coherence decay factor, aa , relates to a Davenport coherence description. Both
horizontal and vertical separation are accepted, and to facilitate the use of cup anemometers, the
decay factor relates to the horizontal turbulence component. Vertical and horizontal separation are
indicated by adding a "V" or a "H" to the table value.

The wake coherence decay factor, aw , relates to a Davenport coherence description. Both hori-
zontal and vertical separation is accepted, and the decay factor relates to the horizontal turbulence
component. Vertical and horizontal separation are again indicated by adding a "V" or a "H" to the
table value.
2.4 Wake models
The overall conclusion from the initial parameter study is that the induced mean wind deficit, the
increased turbulence intensity, the decreased turbulence length scale, as well as the increased coherence
decay should all be considered as significant modifications in relation to increased wind turbine fatigue
life consumption in wakes compared to ambient conditions.
The present section is concerned with a possible quantification of these parameters, in a wake situation,
as function of the involved turbine characteristics and the ambient meteorological conditions. The
performance of a number of available models is investigated. The study is based on the wake measurements
supplied by Teknikgruppen, ECN and Risoe, and described in the previous section 2.3. The measurements
include a wide range of different wind turbines and ambient meteorological conditions.
2.4.1 GCL–model
The model [2] encompass a semi–analytical description of the wake deficit and a set of simple empirical
relations providing the relevant characteristics (turbulence intensity as well as the turbulence length scale)
for the turbulence field in a wake. The empirical expressions are based on full scale measurements and
are applicable for both stall–, pitch– and variable speed regulated turbines.
In order to enable the treatment of data from model experiments performed in wind tunnels, the expression
for the turbulence intensity has been made non–dimensional and is thus slightly modified compared to
the expression given in the above reference. Moreover, one of the boundary conditions for the deficit
model was modified in order to reflect the blocking effect from the ground (in a simple manner).
Deficit
Neglecting the blocking effect originating from the ground, the wake behind a wind turbine is considered
a free turbulence region. Problems involving free turbulence are often of the same nature as boundary
layer problems in the sence that the size of the free turbulence region in the direction perpendicular to
the mean flow is considerably smaller than the size in the mean flow direction. The model is thus based
on the presumption that the wake region can be adequately described by Prandtl’s (rotational symmetric)
turbulent boundary layer equations (large Reynolds numbers assumed).
Assuming similarity between deficits at different distances from the upstream turbine and moderate
amplitudes of the deficit, the following assymptotic expressions are obtained for the mean wind speed
deficit, V , and for the wake radius, Rw :
35 1
(3c21 ) 15 (CT Ax) 31 ;

5
Rw = 2 (2.4.1.1)

Va ; 1 n 3
35 103 o2

V = ; 9 CT Ax ;2 3
r (3c1 CT Ax) ; 2 (3c21 ); 51 ;
2 2 ; 1
2 (2.4.1.2)
where x and r denote axial and radial directions, A is the rotor area, CT denotes the rotor drag coefficient,
Va the ambient mean wind velocity at hub height, and c1 is the non–dimensional mixing length defined
by
c1 = l(CT Ax); 31 ; (2.4.1.3)
with l being Prandtl’s mixing length.

This expression separates to some degree the rotor specific drag dependance, and l is consequently
expected to be relatively insensitive to both the size and the design of the rotor.
With CT determined either from measurements or from a simple aerodynamic calculations, two parameters
have to be evaluated from boundary conditions – one is the position of the rotor relative to the applied
coordinate system, and the second is the non–dimensional mixing length c1 .
The first boundary condition to be satisfied is that the wake radius at the rotor position equals the rotor
radius.
The second boundary condition is an empirical relationship stating that the wake radius at 9.5 rotor
diameters distance from the turbine can be determined by Rw;9:5D = max(1:08D; 1:08D + 21:7D (Iah ;
0:05)), where D and Iah denotes the rotor diameter and the (horizontal) ambient turbulence intensity,
respectively. This boundary condition essentially states that the expansion of the wake is dominated by
ambient cross flow turbulence – however for sufficiently small ambient turbulence intensity the turbulence
contributions from wake shear, blade tip vortices, and hub vortices ensures a turbulence intensity level
corresponding to minimum 5%. If the wake radius, determined according to the above prescribtion,
exceeds the hub height, the blocking effect is taken into account by applying a wake radius equal to the
mean of Rw;9:5D and the hub height as a "design wake radius" at 9.5 rotor diameters distance from the
turbine.
The performance of the model appears from the Table 2.3, where S denotes the turbine spacing expressed
in rotor diameters, and V is the mean wind deficit.
Some deviation between measurements and model predictions are encountered, especially for the results
associated with the Sexbierum wind farm. However, no systematic trend was observed in the deviations.
As the model is based on an asymptotic theory, it might be somewhat conservative for very close
spacings [19].
Turbulence Intensity
At down stream distances corresponding to traditional spacings in wind farms, it is assumed that only
the surface– and wake shear mechanisms contribute significantly to the turbulence production. It is
moreover assumed that the turbulence fluctuations, originating from these two sources, are statistically
independent such that the turbulent energies are additive in the energy sense. Normalizing with respect to
the undisturbed mean wind velocity, the total turbulence intensity of the along wind turbulence component
in the wake, Iwt
u , is expressed as

Table 2.3 Measured wake deficit values compared to GCL–model predictions.
S D Hub CT Stability Amount Va Ia Mea. Est. Rel.

[D] [m] Height Ri of Data [m/s] [%] V V dev.
[m] [hours] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 4.4 2.1 -52
5.3 28 31 0.76 neutral 9.0 7.3 11.9 1.2 1.2 0
5.3 28 31 0.70 neutral 4.5 9.1 12.9 1.4 1.2 -14
5.3 28 31 0.59 neutral 7.5 11.5 13.7 1.6 1.2 -25
7.5 28 31 0.76 neutral 4.5 7.5 10.9 0.6 1.1 83
7.5 28 31 0.70 neutral 1.5 9.1 14.2 0.7 0.8 14
14.5 28 31 0.75 neutral 16.0 7.9 11.7 0.5 0.6 20
14.5 28 31 0.69 neutral 11.0 9.3 12.7 0.5 0.6 20
14.5 28 31 0.59 neutral 4.0 11.7 13.8 0.3 0.6 100
4.2 23 30 0.44 0.04 0.33 14.2 5.6 4.5 3.4 -24
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 2.4 2.3 -4
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 1.6 1.8 13
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 1.8 2.7 50
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 1.9 2.3 21
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 3.0 3.9 30
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 1.8 2.3 28
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 3.0 2.9 -3
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 3.4 2.5 -27
4.2 23 30 0.83 0.07 0.5 7.9 3.8 4.3 5.2 21
4.2 23 30 0.3 0.01 0.166 17.4 7.0 4.6 2.2 -52
4.2 23 30 0.52 0.03 0.166 13.4 4.3 5.3 4.6 -13
4.2 23 30 0.5 0.03 0.166 13.7 5.9 5.3 3.4 -36
4 30 35 0.75 0.285 7.3 10.5 3.6 1.6 -56
4 30 35 0.75 0.356 7.7 8.9 3.6 2.0 -44
4 30 35 0.75 0.356 8.1 10.9 3.5 1.7 -51
7 30 35 0.57 0.711 13.6 7.2 2.8 1.9 -32
7 30 35 0.75 0.569 10.1 9.5 2.7 1.8 -33
7 30 35 0.74 0.996 10.0 9.1 2.3 1.8 -22
7 30 35 0.74 0.356 8.9 9.6 2.3 1.5 -35
5 0.27 0.31 0.91 0.006 4.13 9.0 1.05 1.2 14
7.5 0.27 0.3 0.87 0.008 4.1 9.0 1.03 0.9 -13

q
u = I2 + I2 ;
Iwt a w (2.4.1.4)
where subscripts a and w refer to ambient and wake, respectively. The specific wake contribution, Iw ,
depends on both the down wind distance and the undisturbed mean wind velocity, and is (for spacings
larger than two rotor diameters) determined as :
q p
Iw = 0:29S ;1=3 1 ; 1 ; CT ; (2.4.1.5)
where S denotes the spacing expressed in rotor diameters and CT is the rotor thrust coefficient.
The proposed model has the correct qualitative– and asymptotic features as Iw increases for increasing
CT , decreases for increasing S , and tends to zero for S approaching infinity.
The performance of the model appears from the Table 2.4.
The model is based on the maximal change in a standard deviation profile (which is usually associated to
the upper part of the rotorplane) normalized with the undisturbed mean wind velocity at hub height. The
reason is, that the wind field modelling in most aeroelastic models operate with only one value for the
turbulence intensity – we have then selected a slightly conservative value for that parameter.
In some of the experiments the spatial resolution was limited, and as a consequence the estimated values
were expected to be moderately conservative. This is also reflected in the table above, except for a few
very large deviations related to measurements with practically no increase in the turbulence intensity,
originating from the wake situation. In general, a satisfactory agreement is encountered.
Turbulence Length Scale

When the two shear turbulence contributions have scales of comparable orders of magnitude, it is
reasonable to expect the resulting spectral shape to be qualitatively invariant compared to the ambient
situation. However, the resulting length scale is modified. The modified length scale depends on the rotor
thrust coefficient, the spacing distance, the rotor diameter, and the undisturbed turbulence length scale.
Assuming an approximate linear relationship between thrust coefficient and mean wind speed (for a stall
controlled turbine), and refeering to a Kaimal formulation of the turbulence spectrum, the length scale
for the u turbulence component inside a wake, Luw , is determined from :
0 0 D 11
12:2 1 ;
Luw = Lua @1 ; min @ 0:6 L ; 0:9AA ;
Va Sr
a
(2.4.1.6)
where Lua denotes the u–turbulence length scale in the undisturbed flow, Va is the undisturbed mean wind
velocity at hub height, and the reduced spacing, Sr , between the turbines is defines as :
Sr = 20:8 + (S ; 2)0:8 for S 2: (2.4.1.7)
The undisturbed length scale and the resulting wake length scale refer to conditions at hub height. Luw
is seen to be monotonous in Va , Lua , Sr , and D . However, the dependence on D is not pronounced for

Table 2.4 Wake turbulence intensities compared to GCL–model predictions.

[D] [m] Height Ri of Data [m/s] [%] Iwt Iwt dev.
[m] [hours] [%] [%] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 16.5 20.5 24
5.3 28 31 0.76 neutral 9.0 7.3 11.9 16.2 16.8 4
5.3 28 31 0.70 neutral 4.5 9.1 12.9 16.2 17.1 6
5.3 28 31 0.59 neutral 7.5 11.5 13.7 17.5 16.9 -3
8.5 35 37 0.52 neutral 4.0 11.1 6.8 11.0 10.4 -5
9.5 35 37 0.57 neutral 10.0 7.2 11.3 10.8 -4
7.5 28 31 0.76 neutral 4.5 7.5 10.9 14.0 15.2 9
7.5 28 31 0.70 neutral 1.5 9.1 14.2 14.3 17.3 21
14.5 28 31 0.75 neutral 16.0 7.9 11.7 12.3 14.4 17
14.5 28 31 0.69 neutral 11.0 9.3 12.7 12.6 15.0 19
14.5 28 31 0.59 neutral 4.0 11.7 13.8 14.7 15.5 5
4.2 23 30 0.44 0.04 0.33 14.2 5.6 6.6 10.6 61
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 11.4 12.2 7
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 11.3 14.4 27
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 12.5 13.8 10
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 12.9 13.5 5
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 15.6 15.0 -4
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 13.5 13.3 -1
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 16.0 15.0 -6
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 13.0 14.4 11
4.2 23 30 0.83 0.07 0.5 7.9 3.8 10.7 14.3 34
4.2 23 30 0.3 0.01 0.166 17.4 7.0 9.3 10.1 9
4.2 23 30 0.52 0.03 0.166 13.4 4.3 6.6 10.9 65
4.2 23 30 0.5 0.03 0.166 13.7 5.9 6.6 11.4 73
4 30 35 0.75 0.285 7.3 10.5 15.1 16.6 10
4 30 35 0.75 0.356 7.7 8.9 13.7 15.7 15
4 30 35 0.75 0.356 8.1 10.9 15.4 16.9 10
7 30 35 0.57 0.711 13.6 7.2 9.3 11.4 23
7 30 35 0.75 0.569 10.1 9.5 12.2 14.3 17
7 30 35 0.74 0.996 10.0 9.1 11.9 14.0 18
7 30 35 0.74 0.356 8.9 9.6 11.1 14.3 29
5 0.27 0.31 0.91 0.006 4.13 9.0 14.1 16.8 19
7.5 0.27 0.3 0.87 0.008 4.1 9.0 12.3 14.9 21

Table 2.5 Wake length scales for u–turbulence components compared to GCL–model predictions.
S D Hub CT Stability Amount Va Lua Mea. Est. Rel.

[D] [m] Height Ri of Data [m/s] [%] Luw Luw dev.
[m] [hours] [m] [m] [%]
2.0 28 31 0.67 neutral 0.5 9.9 1070 140 149 7
5.3 28 31 0.76 neutral 9.0 7.3 1052 359 343 -5
5.3 28 31 0.70 neutral 4.5 9.1 1035 421 475 13
5.3 28 31 0.59 neutral 7.5 11.5 1363 788 776 -2
8.5 35 37 0.66 neutral 14.0 9.0 940 505 530 5
4.2 23 30 0.44 0.04 0.33 14.2 1306 598 797 33
6.1 23 30 0.67 -0.3 0.33 9.7 1380 846 717 -15
6.1 23 30 0.87 -0.15 0.33 6.3 1776 672 448 -33
6.1 23 30 0.87 -1.25 0.33 6.7 1849 975 559 -43
6.1 23 30 0.81 -0.5 0.5 7.8 1730 837 687 -18
4.2 23 30 0.86 -1.0 0.33 6.9 1592 635 317 -50
6.1 23 30 0.79 -2.0 0.5 8.0 1711 828 705 -15
4.2 23 30 0.83 -6.5 0.33 7.5 1868 616 479 -22
4.2 23 30 0.71 -1.7 0.33 9.6 1196 690 504 -27
4.2 23 30 0.83 0.07 0.5 7.9 672 598 210 -65
4.2 23 30 0.3 0.01 0.166 17.4 1840 1270 1251 -1
4.2 23 30 0.52 0.03 0.166 13.4 902 543 531 -2
4.2 23 30 0.5 0.03 0.166 13.7 1693 708 1008 42
4 30 35 0.67 0.285 7.3 644 138 159 15
4 30 35 0.67 0.356 7.7 920 184 253 37
4 30 35 0.67 0.356 8.1 736 230 233 1
the usual size of wind turbines. For large turbine spacings or for large mean wind speeds, Luw tends to
approach Lua , and the expression thus exhibit the correct asymptotic behaviour.
In the Table 2.5, the performance of the model has been compared to measured values. The model is
calibrated to a Kaimal formulation of the length scale, and where data were given with reference to a Von
Karman formulation, they have been converted into a Kaimal formulation.
Considering the large uncertainties related to the determination of the length scale in general, the agree-
ment between measurements and predictions is considered satisfactory – especially when bearing in
mind that fatigue estimates, based on aeroelastic calculations, is rather unsensible to relatively moderate
variations in the length scale.
In analogy with the turbulence intensity parameter, and for the same reasons, the present model is based
on the maximal changes in a length scale profile. This is also reflected in the estimates, as most of these
underpredict the measured values.
Coherence decay
Initiated by the indications from the initial parameter study, a simple attempt has been performed to extend
the original model–complex also to take into account modification of the cross–correlation between
turbulence components in a wake situation. In accordance with the assumptions introduced in beginning

of this chapter, the considerations to follow are limited to only uu–coherences. The investigation is based
on a relatively slender data material8 .
The tendencies displayed, in the Swedish and the Danish measured uu–coherence decay factors, seems
at first to be somewhat contradicting, as the Danish results indicate no wake effect on vertical coherence
for a 7.5 diameter spacing, whereas the Swedish results show a marked effect for a 6.1 diameter spacing.
However, forcing the wake effect to be neglectable for spacings larger than 7.5 diameters and maximal
for a 2 diameter spacing, introduces some harmony in the interpretation.
The two sets of experiments operate with different values of the vertical spacing, as the Danish results
relates to a 13 m spacing, whereas the Swedish measurements operate with 6.4 m and 4.7 m spacing.
Thus the coherences evaluated based on the Danish measurements will decay somewhat faster with the
frequency than the Swedish coherences. However, the Danish results are based on very long runs which
(more than) compensate for the loss in statistical significance due to the faster decay9.
The coherence is given a Davenport–formulation and is thus expressed as
s
asf
Coh(s; f ) = exp ; V ; (2.4.1.8)
a
where f is the frequency in [Hz], s is the separation in [m], Va is the (ambient) mean wind speed in [m/s].
The ambient (undisturbed) coherence decay factor, aa , is determined by [8]
(
12 + 11z jz2 ;z1 j for vertical separation
aa = 12 + z 11y for lateral separation ;
average
(2.4.1.9)
average
with the measuring heights denoted by z1 , z2 , the average of these equal to zaverage , and y being the
lateral separation.
Intuitively, it is felt that the relative change of the coherence decay factors for lateral and vertical
separations should be of the same order of magnitude. Both the lateral and the vertical coherences is,
as expected, strongly effected by the wake for small wind turbine separations, whereas for moderate
separations they are unaffected. As stated above, the wake influence is assumed to be negligible for
spacings larger than 7.5 rotor diameters, and maximal for a spacing equal to 2 rotor diameters.
Basing the lateral wake coherence decay modification on the Danish measurements and the vertical wake
coherence decay modification on both Swedish and Danish data, the following expression for the wake
coherence decay factor, aw , is proposed
(
aw = (1 + Kw )aa for 2 S < 7:5 ;
aa for 7:5 S (2.4.1.10)
where S denote the spacing expressed in rotor diameters and the wake correction factor, Kw , is given by
8 The data material reproduced in the Appendix F.
9 The coherence decay is mainly determined based on values in the frequency band corresponding to coherence values between
0.3 and 1.0

Table 2.6 Measured wake coherence decay compared to model predictions.
S Stability Amount Va Mea. Est. Mea. Est.

[D] Ri of Data [m/s] 1 + KwH 1 + KwH 1 + KwV 1 + KwV
[hours]
2 neutral 3.0 11.3 2.6 2.8
2 neutral 1.5 11.7 3.2 1.9
2 neutral 1.5 14.7 3.8 2.9
4.2 -1.0 0.33 6.9 1.8 1.8
4.2 -6.5 0.33 7.5 1.9 1.9
4.2 -1.7 0.33 9.6 2.4 2.3
4.2 0.03 0.166 13.4 3.0 3.1
4.2 0.03 0.166 13.7 3.4 3.1
4.2 0.04 0.33 14.2 3.3 3.2
4.2 0.01 0.166 17.4 2.1 3.4
6.1 -0.15 0.33 6.3 1.2 1.3
6.1 -1.25 0.33 6.7 1.5 1.3
6.1 -0.5 0.5 7.8 1.7 1.4
6.1 -2.0 0.5 8.0 1.9 1.4
6.1 -0.3 0.33 9.7 1.9 1.6
8
>
< 7:5;S for Va < 6
Kw = > 7:55:;5 S ( 1 Va ; Cs ) for 6 Va 15 :
4(75::55;S3)
(2.4.1.11)
: 5:5 for 15 < Va
For vertical separation Cs 1 and for horizontal separation Cs 2. For spacings less than 2 rotor
diameters the value of the wake coherence decay factor is assumed to equal the value related to a 2
diameter spacing.
The model performance is illustrated below, where KwV refers to vertical separation, and KwH refers to
horizontal separation.
The agreement is satisfactory. However, the data material is somewhat limited which might affect the
generality of the proposed estimation procedure.
Most turbulence generators do not have the ability to take into account coherence decay factors depending
on spacing direction, nor on the magnitude of the spacing. In general, the turbulence generators presume
constant coherence decay factors, and the derived coherences (refeering to a particular site) thus depend
only on frequency and on the absolute value of the spacing. In this situation the mean of cohence decay
factors estimated for horizontal– and vertical separation, respectively, is recommended, and in case the
ambient coherence factor is determined from the above expression (2.4.1.9) , a vertical separation of 2/3
of the rotor diameter should be specified in order to assure satisfactory agreement for eddies with the
maximal potential for aerodynamic impact.
Transversal turbulence
So far the investigations have concentrated on the along wind turbulence component u only, as this
component is the main responsible for the fatigue load generation, and as most available measured
data are based on cup anemometer recordings (giving the horizontal velocity component and in a first
order approximation the along wind turbulence intensity). However, the fatigue effect originating from
the transversal components is not negligible. Therefore, a relation between the v – and w–turbulence
components and the u–component is of interest. The derivation to follow is also an extension compared
to the original model concept.
In analogy with the considerations related to the u–component, it is assumed that the transversal turbulence
can be properly described by adjusting the spectral parameters related to a traditional description of the
ambient turbulence in flat homogeneous terrain. Thus, the qualitative characteristics of the turbulence
are presumed to be preserved, and here the wind farm effects are condensed in a suitable modification of
only a few parameters.
At this stage, there is limited information available on the wake behaviour of the transversal components,
and the available material is dealing with only two of the above mentioned parameters – the turbulence
intensity and the turbulence length scales.
For flat and homogeneous terrain, the following relations express the standard deviations of the turbulence
components, u , v , and w as
av 34 au ; (2.4.1.12)
aw 21 au ; (2.4.1.13)
where subscript "a" indicate ambient values.

In a wake, the turbulence tends to be more isotropic, and the above relations are modified to [9]10
wv 0:8wu ; (2.4.1.14)
ww 0:6wu ; (2.4.1.15)
where subscript "w" indicate wake–values.

10 In the analysis, the relation between turbulence components turn out to depend on the measuring height, and the presented
values is the most conservative ones in the sence that they produce the largest standard deviations.

Turbulence length scales

Let the u–, v –, and w– turbulence components inside as well as outside the wake be described by power
density spectra of the Kaimal form,
k
Sk (Lk ; f ) = (k )2 (1 + 3LfL=kV=V )5=3 ; (2.4.1.16)
2
in the neutral regime, where k substitutes either u, v or w. Assuming local isotropy in the inertial subrange
for sufficiently small eddies, the following relations hold for the spectral components:
lim Su(Lu ; f ) = lim Su (Lu ; f ) = 3 :

f !1 S (Lv ; f ) f !1 S (Lw ; f ) 4
(2.4.1.17)
v w
The proportion 3:4 in relation (2.4.1.17) does not express lack of isotropy, but is a consequence of the
turbulence component u being directed along the mean wind direction.
For flat and homogeneous terrain, combining expressions (2.4.1.12), (2.4.1.16), and (2.4.1.17) leads to
the following relation
S (L u; f ) L u 1 + 3 Lv =Va !5=3 3
u a a f 2 a
lim = lim
f !1 Sv (La ; f ) f !1 ; 2 1 3 u = 4;
4 La f + 2 La =Va
v 3 v
from where the proportion between the v – and u length scales can be derived as11
Lva = 3 9 3=2 = 0:27 :

Lua 4 16 (2.4.1.18)
In analogy, the value for the proportion between the w– and u length scales in homogeneous terrain can
be determined to
Lwa = 0:08 :
Lua (2.4.1.19)
Similar considerations related to the wake situation, and thus based on relations (2.4.1.14), (2.4.1.15),
(2.4.1.16), and (2.4.1.17), lead to
Lvw = 0:33
Luw (2.4.1.20)
Lww = 0:14 : (2.4.1.21)
Luw
11 Using the value 0.8 as the proportion between the v and u standard deviations, we arrive at the value 0.33 for proportion
between the v and u length scales, which corresponds to the value given in the code IEC–1400–1.

2.4.2 SF–model
Also the present empirical method [3] is based on full scale measurements and works by specifying
an effective turbulence intensity (or design turbulence intensity) in the mean wind direction, aiming
at substituting a detailed wake calculation, when determining fatigue life consumption by means of
aeroelastic models.
The wake effects, associated with the mean wind deficit, the decreased turbulence length scale, and the
increased coherence decay, concerning the fatigue loading of a wind turbine is thus included in this
effective turbulence intensity. As a consequence the effective turbulence intensity is usually conservative
compared with measured physical values.
u 12 , is determined from :
The effective turbulence intensity, Iwe
s
u = 1:2CT + I 2 :
Iwe S2 a (2.4.2.1)
The performance of the model appears from the Table 2.7. Apparently the model is not capable of
handling near wake situations, as the 2D situation is leading to a 185% deviation. It is often found in
wind tunnel measurements that up to 2D spacing, the shear stress profile develops very slowly in single
wakes, by which transfer of momentum from the outher flow to the wake is limited [10],[11].
As expected, the model is somewhat conservative when compared to the measured (physical) wake values.
However, as the model contain all wake effects in one number, a quantitative comparison is difficult.
2.4.3 DR–model
This model describes the increased fatigue loading in wakes by modifying the ambient turbulence intensity
and is the procedure applied in the present "Danish Recommendation".
When wind turbines are erected in a wind farm cluster with a distance between the turbines of at least 5
rotor diameters, or in a wind farm array with a mutual distance of at least 3 rotor diameters, the increase
in fatigue loading, originating from the change in mean wind field and turbulence structure, can be taken
into account by specifying a design turbulence intensity13 , Id , as :
q
Id = Ia2 + Iw2 ; (2.4.3.1)
where Iw is an additional wake contribution expressed by
Iw = v l 0:15 : (2.4.3.2)
In the above expression, v denotes a correction factor related to the mean wind hub velocity, whereas
l is a correction factor related to the distance between the wind turbines. These correction factors are
obtained from Figures 2.1, 2.2 and 2.3, respectively.
12 Normalized with respect to the ambient mean wind speed at hub height.
13 Normalized with respect to the ambient mean wind speed at huh height.

Table 2.7 Wake turbulence intensities compared to SF–model predictions.

[D] [m] Height Ri of Data [m/s] [%] Iwe u
Iwe dev.
[m] [hours] [%] [%] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 16.5 47.0 185
5.3 28 31 0.76 neutral 9.0 7.3 11.9 16.2 21.6 33
5.3 28 31 0.70 neutral 4.5 9.1 12.9 16.2 21.6 33
5.3 28 31 0.59 neutral 7.5 11.5 13.7 17.5 21.0 20
8.5 35 37 0.52 neutral 4.0 11.1 6.8 11.0 11.5 5
9.5 35 37 0.57 neutral 10.0 7.2 11.3 11.3 0
7.5 28 31 0.76 neutral 4.5 7.5 10.9 14.0 16.8 20
7.5 28 31 0.70 neutral 1.5 9.1 14.2 14.3 18.7 31
14.5 28 31 0.75 neutral 16.0 7.9 11.7 12.3 13.4 9
14.5 28 31 0.69 neutral 11.0 9.3 12.7 12.6 14.2 12
14.5 28 31 0.59 neutral 4.0 11.7 13.8 14.7 15.0 2
4.2 23 30 0.44 0.04 0.33 14.2 5.6 6.6 18.2 176
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 11.4 16.0 41
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 11.3 18.1 60
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 12.5 17.6 41
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 12.9 17.3 34
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 15.6 24.7 58
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 13.5 17.2 27
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 16.0 24.5 53
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 13.0 23.3 79
4.2 23 30 0.83 0.07 0.5 7.9 3.8 10.7 24.1 125
4.2 23 30 0.3 0.01 0.166 17.4 7.0 9.3 15.9 71
4.2 23 30 0.52 0.03 0.166 13.4 4.3 6.6 19.3 192
4.2 23 30 0.5 0.03 0.166 13.7 5.9 6.6 19.4 193
4 30 35 0.75 0.285 7.3 10.5 15.1 25.9 72
4 30 35 0.75 0.356 7.7 8.9 13.7 25.3 85
4 30 35 0.75 0.356 8.1 10.9 15.4 26.1 69
7 30 35 0.57 0.711 13.6 7.2 9.3 13.8 49
7 30 35 0.75 0.569 10.1 9.5 12.2 16.6 36
7 30 35 0.74 0.996 10.0 9.1 11.9 16.2 37
7 30 35 0.74 0.356 8.9 9.6 11.1 16.5 48
5 0.27 0.31 0.91 0.006 4.13 9.0 14.1 22.8 61
7.5 0.27 0.3 0.87 0.008 4.1 9.0 12.3 16.3 33

Figure 2.1 Wind speed correction factor.
Figure 2.2 Distance correction factor for wind turbines in a wind farm array.
Figure 2.3 Distance correction factor for wind turbines in a wind farm cluster.

The performance of the model appears from the Table 2.8.
Table 2.8 Wake turbulence intensities compared to DR–model predictions.

[m] [hours] [%] [%] [%]
5.3 28 31 0.76 neutral 9.0 7.3 11.9 16.2 16.2 0
5.3 28 31 0.70 neutral 4.5 9.1 12.9 16.2 16.9 5
5.3 28 31 0.59 neutral 7.5 11.5 13.7 17.5 17.6 0
8.5 35 37 0.52 neutral 4.0 11.1 6.8 11.0 10.1 -8
9.5 35 37 0.57 neutral 10.0 7.2 11.3 10.4 -8
7.5 28 31 0.76 neutral 4.5 7.5 10.9 14.0 13.2 -6
7.5 28 31 0.70 neutral 1.5 9.1 14.2 14.3 16.1 12
14.5 28 31 0.75 neutral 16.0 7.9 11.7 12.3 13.9 13
14.5 28 31 0.69 neutral 11.0 9.3 12.7 12.6 14.7 17
14.5 28 31 0.59 neutral 4.0 11.7 13.8 14.7 15.7 7
4.2 23 30 0.44 0.04 0.33 14.2 5.6 6.6 12.5 89
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 11.4 9.9 -14
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 11.3 10.2 -10
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 12.5 9.2 -26
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 12.9 9.8 -24
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 15.6 13.8 -12
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 13.5 9.8 -27
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 16.0 14.2 -11
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 13.0 15.0 16
4.2 23 30 0.83 0.07 0.5 7.9 3.8 10.7 13.4 26
4.2 23 30 0.3 0.01 0.166 17.4 7.0 9.3 11.1 19
4.2 23 30 0.52 0.03 0.166 13.4 4.3 6.6 12.5 90
4.2 23 30 0.5 0.03 0.166 13.7 5.9 6.6 13.0 96
4 30 35 0.67 0.285 7.3 10.5 15.1 16.9 12
4 30 35 0.67 0.356 7.7 8.9 13.7 16.0 17
4 30 35 0.67 0.356 8.1 10.9 15.4 17.2 11
7 30 35 0.42 0.711 13.6 7.2 9.3 9.9 6
7 30 35 0.66 0.569 10.1 9.5 12.2 12.1 -1
7 30 35 0.66 0.996 10.0 9.1 11.9 11.8 -1
7 30 35 0.66 0.356 8.9 9.6 11.1 12.2 10
5 0.27 0.31 0.91 0.006 4.13 9.0 14.1 14.6 4
7.5 0.27 0.3 0.87 0.008 4.1 9.0 12.3 11.7 -5
Despite the simplicity of the model, the agreement between model predictions and measurements is
satisfactory except for a few very large deviations. This is somewhat surpricing, as the philosophy behind
the model is the same as for the the SF–model – namely to consentrate all wake effects in only one
parameter (turbulence intensity). Therefore more conservative results were expected.

2.4.4 TNO–model
This empirical model [4] originates from work carried out in the "Dynamic Loads in Wind Farms (I)"
Joule project and has been further assessed by the Dutch TNO laboratory. It is based on extensive
measurements of wind turbine arrays in an atmospheric boundary layer wind tunnel. The model specifies
u 14 , compared to the ambient conditions as well as the
the increased turbulence intensity in the wake, Iwt
maximal wake deficit, V .
The wake turbulence intensity is determined from the expression

q
u = I2 + I2 ;
Iwt a w (2.4.4.1)
with the additional wake contribution determined from
;0:96
Iw = 1:31CT0:7 Ia0:68 XX ; (2.4.4.2)
F
where CT denotes the rotor thrust coefficient, X is the downwind distance, and XF denotes the beginning
of the far wake section given by
XF = d(1e(1;;e)dX) N ; (2.4.4.3)
with d and e conveniently introduced as
p
d = 0:214 + 0:144m ; (2.4.4.4)
p
e = 0:134 + 0:124m ; (2.4.4.5)
and m determined by
m = p1 ;1 C : (2.4.4.6)
T
The extension of the wake near field, XN , is expressed by
XN = q; dR 2R0 ; dR 2 ; (2.4.4.7)
dx m + dx
where the involved quantities are determined by
14 Normalized with respect to the undisturbed ambient mean wind speed.

dR p
(1 ; m) 1:49 + m
dx m = 9:76(1 + m) ; (2.4.4.8)
dR q
dx = 0:075Ia0:21 + 0:457 10;3 (0:71I +2:62) ;
a (2.4.4.9)

and
r
R0 = D2 m 2+ 1 ; (2.4.4.10)
where denotes the tip speed ratio, and D is the rotor diameter.
The performance of the TNO–model, regarding turbulence intensities, appears from the Table 2.9.
Some deviations are encountered between measurements and model predictions. However, there is no
systematic trend, except that the model tend to yield conservative results in situations with low ambient
turbulence intensities.
Deficit
With the nomenclature introduced above, the maximal wake deficit is expressed as
;1:25
V = VN XX ; (2.4.4.11)
N
where X denotes the down wind distance from the turbine, and
VN = 0:56CT + 0:255 : (2.4.4.12)
The performance of the deficit prediction is illustrated in the Table 2.10.

Some deviations between measurements and model predictions are encountered, and there is a clear
tendency for the model to underpredict the deficit.
2.4.5 MM–model
The MM–model predicts velocity deficit as well as turbulence intensity and turbulence length scale in
wake situations. Contrary to most other models, the present approach operate with the consept "transport
time" instead of downstream distance. The transport time is calculated using Taylors hypothesis. The
model is described in detail in model [5].

Table 2.9 Wake turbulence intensities compared to TNO–model predictions.

[m] [hours] [%] [%] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 16.5 36.7 122
5.3 28 31 0.76 neutral 9.0 7.3 11.9 16.2 16.2 0
5.3 28 31 0.70 neutral 4.5 9.1 12.9 16.2 18.0 11
5.3 28 31 0.59 neutral 7.5 11.5 13.7 17.5 18.7 7
8.5 35 37 0.52 neutral 4.0 11.1 6.8 11.0 8.1 -27
9.5 35 37 0.57 neutral 10.0 7.2 11.3 8.3 -27
7.5 28 31 0.76 neutral 4.5 7.5 10.9 14.0 13.3 -5
7.5 28 31 0.70 neutral 1.5 9.1 14.2 14.3 17.1 19
14.5 28 31 0.75 neutral 16.0 7.9 11.7 12.3 12.5 2
14.5 28 31 0.69 neutral 11.0 9.3 12.7 12.6 13.5 8
14.5 28 31 0.59 neutral 4.0 11.7 13.8 14.7 14.6 0
4.2 23 30 0.44 0.04 0.33 14.2 5.6 6.6 9.8 48
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 11.4 9.7 -15
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 11.3 10.4 -7
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 12.5 9.0 -28
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 12.9 10.0 -23
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 15.6 10.7 -32
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 13.5 9.9 -27
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 16.0 12.4 -22
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 13.0 14.5 11
4.2 23 30 0.83 0.07 0.5 7.9 3.8 10.7 9.4 -12
4.2 23 30 0.3 0.01 0.166 17.4 7.0 9.3 10.1 9
4.2 23 30 0.52 0.03 0.166 13.4 4.3 6.6 8.8 33
4.2 23 30 0.5 0.03 0.166 13.7 5.9 6.6 10.8 64
4 30 35 0.75 0.285 7.3 10.5 15.1 18.0 19
4 30 35 0.75 0.356 7.7 8.9 13.7 15.9 16
4 30 35 0.75 0.356 8.1 10.9 15.4 18.5 20
7 30 35 0.57 0.711 13.6 7.2 9.3 8.8 -5
7 30 35 0.75 0.569 10.1 9.5 12.2 12.5 2
7 30 35 0.74 0.996 10.0 9.1 11.9 11.9 0
7 30 35 0.74 0.356 8.9 9.6 11.1 12.5 12
5 0.27 0.31 0.91 0.006 4.13 9.0 14.1 21.9 56
7.5 0.27 0.3 0.87 0.008 4.1 9.0 12.3 15.7 28

Table 2.10 Measured wake deficit values compared to TNO–model predictions.

[m] [hours] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 4.4 5.4 23
5.3 28 31 0.76 neutral 9.0 7.3 11.9 1.2 1.0 -17
5.3 28 31 0.70 neutral 4.5 9.1 12.9 1.4 1.4 0
5.3 28 31 0.59 neutral 7.5 11.5 13.7 1.6 1.9 19
7.5 28 31 0.76 neutral 4.5 7.5 10.9 0.6 0.7 17
7.5 28 31 0.70 neutral 1.5 9.1 14.2 0.7 0.9 29
14.5 28 31 0.75 neutral 16.0 7.9 11.7 0.5 0.3 -40
14.5 28 31 0.69 neutral 11.0 9.3 12.7 0.5 0.4 -20
14.5 28 31 0.59 neutral 4.0 11.7 13.8 0.3 0.5 67
4.2 23 30 0.44 0.04 0.33 14.2 5.6 4.5 3.2 -29
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 2.4 1.4 -42
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 1.6 0.8 -50
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 1.8 0.9 -50
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 1.9 1.1 -42
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 3.0 1.6 -47
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 1.8 1.1 -39
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 3.0 1.7 -43
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 3.4 2.3 -32
4.2 23 30 0.83 0.07 0.5 7.9 3.8 4.3 2.0 -54
4.2 23 30 0.3 0.01 0.166 17.4 7.0 4.6 3.4 -26
4.2 23 30 0.52 0.03 0.166 13.4 4.3 5.3 3.3 -38
4.2 23 30 0.5 0.03 0.166 13.7 5.9 5.3 3.3 -38
4 30 35 0.75 0.285 7.3 10.5 3.6 1.6 -56
4 30 35 0.75 0.356 7.7 8.9 3.6 1.7 -53
4 30 35 0.75 0.356 8.1 10.9 3.5 1.8 -49
7 30 35 0.57 0.711 13.6 7.2 2.8 1.2 -57
7 30 35 0.75 0.569 10.1 9.5 2.7 1.1 -59
7 30 35 0.74 0.996 10.0 9.1 2.3 1.1 -52
7 30 35 0.74 0.356 8.9 9.6 2.3 1.0 -57
5 0.27 0.31 0.91 0.006 4.13 9.0 1.05 1.24 18
7.5 0.27 0.3 0.87 0.008 4.1 9.0 1.03 0.78 -24

Velocity deficit
The decay of the relative maximal velocity deficit downstream a wind turbine is expressed as the sum of
the thrust coeficient, CT , and a function of the transport time downstream the wind turbine, together with
a time scale, t0 :
V = 0:4ln t0 + C :
T
Va t (2.4.5.1)
The transport time t = Vxa is used since it was found that the development of the flow is depending on
how long time the processes have been active.
Immediately downstream the turbine, the relative velocity deficit has two peaks, situated at the midle
section of the blades, but with increasing transport time behind the turbine, momentum transport towards
the centre gradually wipes out the two peaks and one maximum will occure at the centre line. The
time needed to obtain only one maximum is defined as t0 , and depends on atmospheric stability, turbine
characteristic, and the roughness length at the site. For neutral stratification t0 can be written :
1 H R
t0 = ! ln z H ; (2.4.5.2)
0
where ! is the rotational frequency in [Hz], R rotor radius, H hub height, and z0 the roughness length.
For sufficently long transport time the calculated relative velocity deficit becomes smaller than zero; hence
the model is not valid for very large transport times. When this situation occured in the calculations, the
deficit was set equal to zero. Also, the model is not valid for transport times shorter than the time t0 ,
although a value can be calculated.
For transport times where the model is applicable, the spatial distribution v (y; z ) in the vertical (z) and
lateral (y) directions is approximated by a Gaussian distribution as
; ! ; z+H 2 ! ; z;H 2 !!
v(y; z ) = V exp ; Ry 2 exp ; 22 R + exp ; 2R2 ;
Va Va 2y2 z z
(2.4.5.3)
where y and z are the normalized lateral and vertical standard deviations.
The performance of the model is illustrated in the Table 2.11.
The results reveal some deviations, especially for large turbine spacings. The agreement with the Swedish
measurements, which has been used for the calibration of the model, is exelent.
It is a prerequisite for application of the model that a representative surface roughness can be defined,
and consequently situations with roughness changes might cause problems.
The centreline turbulence intensity found in the wake, I~wt

u , can be expressed as

Table 2.11 Measured wake deficit values compared with the MM-model predictions.

[m] [hours] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 4.4 6.86 56
5.3 28 31 0.76 neutral 9.0 7.3 11.9 1.2 1.3 9
5.3 28 31 0.7 neutral 4.5 9.1 12.9 1.4 1.89 35
5.3 28 31 0.59 neutral 7.5 11.5 13.7 1.6 2.2 37
7.5 28 31 0.76 neutral 4.5 7.5 10.9 0.6 0.39 -35
7.5 28 31 0.7 neutral 1.5 9.1 14.2 0.7 0.63 -10
14.5 28 31 0.75 neutral 16.0 7.9 11.7 0.5 0 -100
14.5 28 31 0.69 neutral 11.0 9.3 12.7 0.5 0 -100
14.5 28 31 0.59 neutral 4.0 11.7 13.8 0.3 0 -100
4.2 23 30 0.44 0.04 0.33 14.2 5.6 4.5 4.7 6
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 2.4 2.6 10
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 1.6 1.8 15
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 1.8 1.7 -4
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 1.9 2.0 6
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 3.0 2.8 -6
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 1.8 1.9 5
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 3.0 3.1 1
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 3.4 3.7 7
4.2 23 30 0.83 0.07 0.5 7.9 3.8 4.3 4.3 0
4.2 23 30 0.3 0.01 0166 17.4 7.0 4.6 4.9 5
4.2 23 30 0.52 0.03 0.166 13.4 4.3 5.3 5.2 -1
4.2 23 30 0.5 0.03 0.166 13.7 5.9 5.3 5.2 -2
4 30 35 0.75 0.285 7.3 10.5 3.6 2.1 -42
4 30 35 0.75 0.356 7.7 8.9 3.6 2.2 -38
4 30 35 0.75 0.356 8.1 10.9 3.5 2.4 -33
7 30 35 0.57 0.711 13.6 7.2 2.8 0.04 -98
7 30 35 0.75 0.569 10.1 9.5 2.7 0.65 -76
7 30 35 0.74 0.996 10.0 9.1 2.3 0.54 -77
7 30 35 0.74 0.356 8.9 9.6 2.3 0.48 -79
5 0.27 0.31 0.91 0.006 4.13 9.0 1.05 1.09 -4
7.5 0.27 0.3 0.87 0.008 4.1 9.0 1.03 0.9 8


u = I + 0:6 1 ; t0 V :
I~wt a t Va (2.4.5.4)
The turbulence intensity determined according to the above procedure, corresponds to the measured value
normalised with the local mean wake velocity, which is indicated by equipping the turbulence intensity
parameter with a "tilde".
The performance of the model is illustrated in the Table 2.12, where the results presented have been
recalculated and normalised with the ambient wind speed, using the ratio between measured wake
velocity and ambient velocity.
Some deviations are encountered but there is no systematic trend.
Length scale
In the MM–model, the characteristic length scale calculated is Lmax – the wavelength with most energy.
!
Lmax = DV (2.4.5.5)
V
a
This model can only be compared with data from Alsvik, since there has not been any calculations of this
variable for the other sites.
The agreement between predictions and measurements is satisfactory.
2.4.6 UPMWAKE–model
In contradiction to the other presented models, the UPMWAKE model requires more comprehensive
calculations. The model [6] takes into account the atmospheric stability, described by the Monin–
Obukhov length, and the surface roughness in the description of the undisturbed flow. The perturbations,
introduced in the basic flow by a wind turbine, are described in terms of fluid dynamic equations for
conservation of mass, momentum, energy, turbulent kinetic energy, and dissipation rate of the turbulent
kinetic energy. The modeling of the turbulent transport terms is based on the k– model for closure of
the turbulent flow equations.
The set of equations is discretized by means of the finite difference method and solved numerically using
the SIMPLE algorithm [7]. The parabolic approximation has been made, and the numerical solution of
the parabolic equations has been carried out using an alternating direction implicit (ADI) method. The
resulting wake model is three dimensional, and the output is velocity deficit and turbulence intensities.
Velocity deficit
The performance of the model is illustrated by the example calculations given in the Table 2.14.

Table 2.12 Measured turbulence intensity values compared with the MM-model predictions.

[m] [hours] [%] [%] [%]
2.0 28 31 0.67 neutral 0.5 9.9 14.0 16.5 6.4 -61
5.3 28 31 0.76 neutral 9.0 7.3 11.9 16.2 16.8 4
5.3 28 31 0.7 neutral 4.5 9.1 12.9 16.2 18.4 14
5.3 28 31 0.59 neutral 7.5 11.5 13.7 17.5 18 3
7.5 28 31 0.76 neutral 4.5 7.5 10.9 14.0 12.4 -11
7.5 28 31 0.7 neutral 1.5 9.1 14.2 14.3 16.1 -13
8.5 35 37 0.52 neutral 4 11.1 6.8 11 17.3 -58
9.5 35 37 0.57 neutral 10 7.2 11.3 10.6 -6
14.5 28 31 0.75 neutral 16.0 7.9 11.7 12.3 11.7 -5
14.5 28 31 0.69 neutral 11.0 9.3 12.7 12.6 12.7 1
14.5 28 31 0.59 neutral 4.0 11.7 13.8 14.7 13.8 -6
4.2 23 30 0.44 0.04 0.33 14.2 5.6 6.6 7 6
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 2.4 2.6 8
6.1 23 30 0.87 -0.15 0.33 6.3 6.9 11.3 15.1 34
6.1 23 30 0.87 -1.25 0.33 6.7 5.4 12.5 12.8 2
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 12.9 13.5 5
4.2 23 30 0.86 -1.0 0.33 6.9 4.8 15.6 12.1 -22
6.1 23 30 0.79 -2.0 0.5 8.0 6.3 13.5 13.2 -2
4.2 23 30 0.83 -6.5 0.33 7.5 6.0 16.0 13.7 -14
4.2 23 30 0.71 -1.7 0.33 9.6 7.7 13.0 13.2 2
4.2 23 30 0.83 0.07 0.5 7.9 3.8 10.8 9.3 -14
4.2 23 30 0.3 0.01 0166 17.4 7.0 9.3 5.8 -38
4.2 23 30 0.52 0.03 0.166 13.4 4.3 6.6 6.5 -2
4.2 23 30 0.5 0.03 0.166 13.7 5.9 6.6 7.2 9
4 30 35 0.75 0.285 7.3 10.5 15.1 22 46
4 30 35 0.75 0.356 7.7 8.9 13.7 20 46
4 30 35 0.75 0.356 8.1 10.9 15.4 22 43
7 30 35 0.57 0.711 13.6 7.2 9.3 7.2 -23
7 30 35 0.75 0.569 10.1 9.5 12.2 12.4 2
7 30 35 0.74 0.996 10.0 9.1 11.9 11.5 -3
7 30 35 0.74 0.356 8.9 9.6 11.1 12.0 8
5 0.27 0.31 0.91 0.006 4.13 9.0 14.1 16.2 15
7.5 0.27 0.3 0.87 0.008 4.1 9.0 12.3 15 22

Table 2.13 Measured Lmaxw values compared with the MM-model predictions.
S D Hub CT Stability Amount Va La Mea. Est. Rel.

[D] [m] Height Ri of Data [m/s] [%] Lw Lw dev.
[m] [hours] [m] [m] [%]
4.2 23 30 0.44 0.04 0.33 14.2 234 59 73 24
6.1 23 30 0.67 -0.3 0.33 9.7 168 82 93 13
6.1 23 30 0.87 -0.15 0.33 6.3 114 69 92 33
6.1 23 30 0.87 -1.25 0.33 6.7 290 66 87 32
6.1 23 30 0.81 -0.5 0.5 7.8 605 70 96 37
4.2 23 30 0.86 -1.0 0.33 6.9 852 44 53 21
6.1 23 30 0.79 -2.0 0.5 8.0 490 75 99 32
4.2 23 30 0.83 -6.5 0.33 7.5 440 57 57 0
4.2 23 30 0.71 -1.7 0.33 9.6 317 60 64 7
4.2 23 30 0.83 0.07 0.5 7.9 106 36 42 17
4.2 23 30 0.3 0.01 0166 17.4 685 132 87 -34
4.2 23 30 0.52 0.03 0.166 13.4 166 71 58 -18
4.2 23 30 0.5 0.03 0.166 13.7 137 53 59 11
Table 2.14 Measured wake deficit values compared to UPMWAKE predictions.

[m] [hours] [%]
5.3 28 31 0.76 neutral 9.0 7.3 11.9 1.2 1.67 39
5.3 28 31 0.70 neutral 4.5 9.1 12.9 1.4 1.69 21
5.3 28 31 0.59 neutral 7.5 11.5 13.7 1.6 1.79 12
7.5 28 31 0.76 neutral 4.5 7.5 10.9 0.6 1.26 110
7.5 28 31 0.70 neutral 1.5 9.1 14.2 0.7 1.12 60
14.5 28 31 0.75 neutral 16.0 7.9 11.7 0.5 0.75 50
14.5 28 31 0.69 neutral 11.0 9.3 12.7 0.5 0.66 32
4.2 23 30 0.44 0.04 0.33 14.2 5.6 4.5 3.15 -30
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 2.4 2.11 -12
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 1.9 2.20 16
4.2 23 30 0.3 0.01 0.166 17.4 7.0 4.6 2.46 -47
4 30 35 0.75 0.285 7.3 10.5 3.6 3.58 0
4 30 35 0.75 0.356 7.7 8.9 3.6 3.83 6
4 30 35 0.75 0.356 8.1 10.9 3.5 3.93 12
7 30 35 0.57 0.711 13.6 7.2 2.8 3.19 14
7 30 35 0.75 0.569 10.1 9.5 2.7 3.85 43
7 30 35 0.74 0.996 10.0 9.1 2.3 4.02 75
7 30 35 0.74 0.356 8.9 9.6 2.3 3.35 46
5 0.27 0.31 0.91 0.006 4.13 9.0 1.05 1.63 55
7.5 0.27 0.3 0.87 0.008 4.1 9.0 1.03 1.08 5

Table 2.15 Wake turbulence intensities compared to UPMWAKE predictions.

[m] [hours] [%] [%] [%]
5.3 28 31 0.76 neutral 9.0 7.3 11.9 16.2 21.7 34
5.3 28 31 0.70 neutral 4.5 9.1 12.9 16.2 20.7 28
5.3 28 31 0.59 neutral 7.5 11.5 13.7 17.5 19.1 9
8.5 35 37 0.52 neutral 4.0 11.1 6.8 11.0 12.1 10
9.5 35 37 0.57 neutral 10.0 7.2 11.3 12.8 13
7.5 28 31 0.76 neutral 4.5 7.5 10.9 14.0 19.3 38
7.5 28 31 0.70 neutral 1.5 9.1 14.2 14.3 19.8 38
14.5 28 31 0.75 neutral 16.0 7.9 11.7 12.3 16.4 33
14.5 28 31 0.69 neutral 11.0 9.3 12.7 12.6 16.2 29
6.1 23 30 0.67 -0.3 0.33 9.7 6.4 11.4 15.9 39
6.1 23 30 0.81 -0.5 0.5 7.8 6.3 12.9 19.4 50
4.2 23 30 0.3 0.01 0.166 17.4 7.0 9.3 9.3 0
4 30 35 0.75 0.285 7.3 10.5 15.1 18.7 24
4 30 35 0.75 0.356 7.7 8.9 13.7 16.8 23
7 30 35 0.57 0.711 13.6 7.2 9.3 10.9 17
7 30 35 0.75 0.569 10.1 9.5 12.2 18.7 53
7 30 35 0.74 0.996 10.0 9.1 11.9 18.3 54
5 0.27 0.31 0.91 0.006 4.13 9.0 14.1 19.2 36
7.5 0.27 0.3 0.87 0.008 4.1 9.0 12.3 18.1 47
Roughly speaking the deviations between measurements and model predictions are of the same order of
magnitude as for the simple models.
The performance of the model is illustrated by the example calculations given in table 2.15.
The deviations between measurements and model predictions reflect model estimates that are somewhat
more conservative than observed for the semiempirical models.
Most likely, the explanation is that wake meandering (tending to reduce deficit as well as turbulence
intensity in measurements) has not been taken into account in the calculations, whereas it to some
extend is included in the calibration of the simple models. This is confirmed by the fact that, generally
speaking, the degree of conservatism increases with down stream distance and decreases with increasing
ambient wind speed 15 . Furthermore, the UPMWAKE results have larger spatial resolution than the
measurements. Since maximum values are compared, the UPMWAKE turbulence intensities are expected
to exceed the measured values moderately.
15 The latter tendency reflects that for low wind speeds, the wind direction fluctuations are stronger.

2.4.7 Concluding remarks
The performance of 7 models has been investigated with respect to prediction of one or more of the wake
characteristics : mean wind deficit, turbulence intensity, turbulence length scale, and coherence decay.
Each of the wake characteristics have been represented by only one value, and the spatial distribution
across the rotor plane has thus not been investigated. However, most of the models offer a spatial
description of the wake deficit, and the UPMWAKE–model furthermore predict a spatial distribution of
the turbulence intensity. To facilitate a comparison, Appendix G contain a graphical representation of the
results embracing wake deficits and wake turbulence intensities.
Some of the models fail to represent the turbulent wake situation with CT > 1 creating backflow in
the wake, when momentum theory does not give a solution anymore (GCL–model and TNO–model).
However, these turbulent wake situations usually occur at very low wind speeds, which make them of
less practical importance as they hardly contribute to the fatigue damage. Therefore, and because none
of the available measurements reflects this situation, this complex of problems has not been investigated
further.
The effect of wake meandering caused by wind direction changes is obviously integrated in the (full scale)
measurements and has the effect of smoothing out the wake in the mean wind direction, leading to less
pronounced time averaged values of the wake characteristics. The effect is increasing with increasing
distance from the upstream turbine. Most of the semiempical models are calibrated under such conditions
which makes a direct comparison with measurements meaningfull. However, for models which are not
adjusted to these situations conservative results is expected, especially for large rotor spacings.
Some uncertainty is related to a comparison between model prediction and measurements. This is caused
both by uncertainties in the mesurements16 and by uncertainties in the model input (f. ex. the CT values
in the stall region of a stall controlled turbine).
Based on the performed investigation the following general conclusions has been drawn :
Three of the models provide information on the wake deficit – the GCL–model, the TNO–model,
and the MM–model. The three models display deviations of the same order of magnitude.
The turbulence intensity inside a wake is predicted by all the investigated models. The SF–
model gives somewhat conservative results which originate from the fact that the model aim at
concentrating the wake fatigue effect, from changes in all atmospheric parameters, in the turbulence
intensity. Also the UPMWAKE–model seems to be somewhat conservative, which might be related
to the fact that wake meandering has not been modelled in the analysis. The remaining models
display acceptable deviations of the same order of magnitude.
The maximal turbulence intensity is often seen in the upper part of the rotor plane (where maximal
vertical shear appears). Due to limited spatial resolution of measurements not all of these refer to
maximum turbulence intensity. Usually, only one representative value of turbulence intensity can
be specified in turbulence models for aeroelastic calculations. Therefore it might be suggested to
recommend a model which gives slightly conservative results in the comparison.
16 The 5D wind tunnel experiment performed by TNO was repeated in the Marchwood wind tunnel under seemingly identical
experimental conditions. However, the measured deficits were 1.44 m/s and 1.05 m/s, respectively. The discrepancy is to some
extend caused by the fact that the flow in the wind tunnel has not fully developed at the measurement positions – however, it still
illustrates the uncertainty related to the measurements. The observed deviation is of the same order of magnitude as has been
identified between model predictions and measurements.

Turbulence wake length scales are only provided by the GCL–, and the MM–models. These models
operate with two different definitions of the length scale, but the deviations between measurements
and predictions are of the same order of magnitude.
A simple model for estimating the coherence decay factor inside a wake is proposed. However, the
model is founded on a limited data material which might affect its generality.
The GCL–model has been extended also to handle wake–modifications of the turbulence compo-
nents perpendicular to the mean wind direction, reflecting the more isotropic character of wake
turbulence compared to ambient turbulence. The model extension is based on analysis of mea-
surements and simple assumptions concerning the spectral behaviour, and it links the wake v – and
w–components to the previously predicted wake u–component. However, no systematic experi-
mental verification was performed.
In conclusion, two models (GCL–model and MM–model) initially provided information on all the relevant
wind wake parameters except the coherence. The GCL–model has subsequently been extended with a
simple coherence prediction module which, however, is founded on a very limited data material possibly
limiting its generality. Further, the GCL–model has been extended also to handle wake–modifications of
the turbulence components perpendicular to the mean wind direction.
No significant differences between the performance of the MM– and the GCL–model have been revealed
in the present verification, and both of these models rest on simple engineering formulas which can be
programmed very easy. Neither of the models is designed for near wake situations, and the MM–model is
furthermore not valid for very large spacings. Concerning the wake length scale modification, the GCL–
model reflects the observed tendency towards decreasing wake modification with increasing ambient
mean wind speed.


3 WAKE LOADING
In the initial phase of the parameter investigation, calculations were performed using the same basic wake
parameters for all the four investigated wind turbines. Having identified the key wake parameters on this
basis, these were quantified in the second phase. Based on the findings in the model verification part, it
was decided to use the GCL–model for prediction of wake parameters. The MM–model for wake deficit
and turbulence intensity was also considered, but the performance was not found to be robust enough
for the quite large variations of fundamental data (CT ; D; rotational speed etc.) corresponding to the
analysed turbines.
Basically, the purpose of the present section is to verify the result obtained in the initial parameter study,
but under more realistic loading conditions. Thus, the rather restrictive assumptions, put forward in
that analysis, are now modified to approach the load situation defined in section 2.1. More specific, the
following adjustments have been performed :
An ambient wind gradient, corresponding to the specified turbine hub height and the ambient
turbulence intensity (at hub height), is now included. A conventional logarithmic shear is presumed.
The wake load parameters is now adjusted to depend explicitly on the individual wind turbines and
operational situations (turbine spacing, ambient mean wind speed). This is accomplished by use
of the GCL–wake–model.
The influence, caused by the transversal turbulence components, is included in the description of
the ambient loading as well as in the wake loading.
3.1 Method
The method applied is basically analogues to the procedure briefly described and subsequently used in
the initial parameter study. However, in the present investigation a few new aspects are included, and a
more thorough description is therefore appropriate.
Like the previous investigation, the present investigation is also based on the four wind turbines described
in Appendices A–D. In addition to these, the medium sized stall controlled turbine, described in Ap-
pendix E, has been included in order to serve as a common reference object for a small benchmark test
performed with the involved program packages.
Three different load categories have been considered for each of the investigated turbines – corresponding
to (undisturbed) mean wind speeds 10 m/s, 14 m/s, and 18 m/s, which in turn represent the regimes "below
stall", "stall", and "deep stall" for the stall regulated turbines.
As in the previous investigation reference sets, corresponding to wind farms with 5D spacing and 8D
spacing, respectively, have been selected. However, in the present investigation the reference sets depend
on the ambient mean wind speed as well as the particular turbine. The choice of wind farm spacings is
believed to be representative for most existing wind farms.
For each load situation the fatigue loads, expressed in terms of equivalent moments, have been determined
based on 10 minutes time series obtained from aeroelastic calculations using "state of the art" wind field
generators and the aeroelastic codes PHATAS, HAWC, and VIDYN. In the simulations, the focus was

put on flapwise–, edgewise–, yaw–, and tilt moments, as well as on the thrust force, and these were
subsequently analysed by means of a two–level factorial method [16].
3.1.1 Reference sets
Each set of reference situations contains one reference situation related to flat and homogeneous terrain
(defined by the particular mean wind speed), and one reference situation related to the wind farm situation
(5D or 8D spacing) depending on the particular turbine as well as on the particular (undisturbed) mean
wind speed.
Flat terrain
The ambient situation has been defined with a turbulence intensity of the longitudinal turbulence com-
ponent equal to 12 % at hub height, and a logarithmic mean wind shear consistent with this choice. The
mean wind speed at altitude z , Va (z ), is thus derived from

Va (z ) = Ia Va (H ) ln z exp(1=0:12) ;
u
2:5 H (3.1.1.1)
where the previously introduced notation is applied, and denotes the von Karman constant.
The magnitude of the energy in the transversal turbulence components are determined from expres-
sions (2.4.1.12) and (2.4.1.13).
The u–turbulence length scale, corresponding to the Kaimal formulation given in [18], is specified to
1000 m, and assuming local isotropy in the inertial subrange for sufficiently small eddies, the ambient
turbulence length scales in the v – and w–directions are determined from expressions (2.4.1.18) and
(2.4.1.19).
Finally, the cross correllations between different Cartesian turbulence components have been neglected,
and the coherence decay constants, corresponding to the remaining three cross correlations, were specified
in accordance with expression (2.4.1.9) as an avaraged compromise between values corresponding to a
vertical– and a horizontal separation equal to 2/3 of the particular rotor diameter (corresponding to eddies
with a size giving rise to maximal aerodynamic rotor impact).
Wake
The relevant wake situations have been determined from the GCL–model expressions outlined in section
2, and a conservative choice of the wake position has been adopted for the wake deficit, as the wake center
was situated at the tip of a horizontal positioned blade (corresponding to maximal induced horizontal
shear) on the side of the rotor giving the most unfavourable blade flap loads. As the applied wind
field generators are not able to distinguish between vertical– and horizontal separation, the average
value Cs = 1:5 is used in the estimation of the modified coherence decay factor describing the u–cross
correlation.

3.1.2 Factorial method and calculation scheme
The basic idea in the Mounturb project [20] was to perform first order variations in each of the selected
parameters. Two different expansion points, defined by each of the reference situations, were used.
For each turbine and for each mean wind situation, this "one–factor–at–a–time" approach requires 5
aeroelastic calculations in order to determine the requested partial derivatives related to each reference
situation. For each reference set this equals 10 aeroelastic calculations. As seen, it is assumed that, over
the parameter range of interest, the variables act additively on the equivalent moment. If furthermore the
requested partial derivatives differ for the two reference situations contained within the same reference
set, an interpretation is difficult.
To circumvent these shortcommings, a more general approach – a two level factorial design procedure [16]
– is applied here. The factorial design procedure is here based on aeroelastic simulations with all possible
combinations of parameter variations contained within the same reference set. In the present case this
requires a total of 16 aeroelastic simulations within each reference set (describing one particular turbine
operating at one particular mean wind speed situation). In return for the 6 additional simulations,
compared to the "one–factor–at–a–time" approach, the following is gained :
Each main parameter effect – an effect caused by the isolated modification of only one parameter
– is determined with superiour significance, as it is based on 8 realisations of differens load series
instead of 2. Effectively, it means that the a main parameter effect is computed as the average of 8
gradient estimates, corresponding to 8 different expansion points.
In case the involved parameters do not act additively, the method can detect and estimate interactions
that quantify the particular non–additivity.
For each turbine, each reference set, and each mean wind speed the simulation matrix to be performed
appears from Table 3.1 below.
Table 3.1 Specification of aeroelastic calculations. Within a given reference set, "A" denotes ambient
conditions and "W" denotes wake conditions of a particular parameter.
Simulation V I u;v;w Lu;v;w a Simulation V I u;v;w Lu;v;w a
1 A A A A 9 W A A W
2 A A A W 10 W A W A
3 A A W A 11 W W A A
4 A W A A 12 A W W W
5 W A A A 13 W A W W
6 A A W W 14 W W A W
7 A W A W 15 W W W A
8 A W W A 16 W W W W

3.1.3 Fatigue determination
Based on the results produced by the performed aeroelastic time simulations, the Rainflow Counting
Procedure is used to determine the involved load cycles for each of the selected load signals. The
Rainflow Counting Procedure syntesize local maximas and minimas to load range cycles, interpretated
as closed hysteresis curves in the load history of the particular material. In the present situation 50 levels
were used to resolve the load range interval.
Having evaluated the Rainflow Spectrum, the fatigue effects, of the associated spectral components, are
subsequently decoupled applying the Palmgren–Miner approach to concentrate the spectral information
in only one number, expressing the corresponding the fatigue damage. In the present analysis, the fatigue
damage estimation has been based on the Wöhler exponents 4, 8, and 12. The results are presented in
terms of equivalent moments giving identical fatigue damage with reference to a 1 Hz cycle frequency.
Usually, more than one (typically between 3 and 6) 10 minutes load time series are required in order to
make fatigue estimates converge to a acceptable level. This is primarely due to the demand for sufficient
statistical significance of the low–frequency turbulence contribution to the fatigue life consumption.
However, as the postpossessing involving the two–level–factorial method average 8 simulations to predict
main effects, it is a priori expected that only one simulation of each individual load situation is sufficient.
Performing a number of replicated simulations, the robustness of the results from the factorial analysis
are further investigated.
Replicated simulations
In order to investigate the statistical significance of the results, based on one single aeroelastic calculation
for each combination of parameters, some replicated sets of calculations were performed. By extending
the set of 16 combinations, specified in Table 3.1, to a new set of 32 combinations, a fifth factor was
hereby introduced – the random seed used as input for the turbulence field generator. A significant effect
of this fifth factor would then indicate a need for averaging equivalent loads from several 10 min load
series, using different random seeds in the involved turbulence simulations. The conclusion from this
excercise was that, for this specific study, the need for replicated runs was not found to be essential.
The analysis was performed using the VIDYN code, with the Nordtank 500 kW turbine as the example
turbine, and covered the three mean wind speed load cases involved in the parameter analysis.
3.2 Results and Analysis
The knowledge extracted from the analysis of the replicated runs legalize the use of 10 minutes time series
representations of the involved load cases as basis for the present parameter analysis. To avoid influence
from irrelevant parameter changes, all the involved turbulence fields, associated with the investigation of
a particular wind turbine consept, have been generated using the same random seed.
As previously mentioned, the aeroelastic simulations involve different program packages. In order to
secure a common reference for the analysis, a small benchmark test has been performed based on one of
the turbines exposed to identical load cases.

Table 3.2 Benchmark results for a 5 diameter spacing. The relative changes in equivalent moments/forces
are presented for the HAWC simulations (Risoe), and for the VIDYN simulations (TG).
Load Wake 10 m/s 14 m/s 18 m/s

Type Par. Risoe TG Risoe TG Risoe TG
Flap V ; m=12 0.06 0.08 0.02 0.10 0.01 0.04
I u;v;w ; m=12 0.32 0.30 0.22 0.19 0.18 0.12
Lu;v;w ; m=12 0.09 0.07 0.03 0.05 0.03 0.04
a; m=12 0.02 -0.06 0.02 -0.03 -0.02 0.02
Edge V ; m=12 0.04 0.00 0.02 0.03 -0.01 0.01
I u;v;w ; m=12 0.11 0.06 0.10 0.04 0.09 0.04
Lu;v;w ; m=12 0.03 0.03 0.02 0.02 0.03 0.02
a; m=12 0.06 0.03 0.17 0.05 0.18 0.09
Tilt V ; m=4 0.01 -0.03 0.01 0.01 0.00 0.01
I u;v;w ; m=4 0.37 0.33 0.24 0.20 0.19 0.11
Lu;v;w ; m=4 0.14 0.13 0.08 0.08 0.05 0.05
a; m=4 0.12 0.04 0.15 0.02 0.07 0.02
Yaw V ; m=4 -0.01 0.00 -0.02 0.00 -0.01 0.00
I u;v;w ; m=4 0.36 0.32 0.23 0.19 0.19 0.10
Lu;v;w ; m=4 0.16 0.13 0.08 0.08 0.05 0.05
a; m=4 0.09 0.05 0.13 0.09 0.05 0.06
Thrust V ; m=4 -0.02 -0.03 -0.01 0.02 -0.01 0.02
I u;v;w ; m=4 0.35 0.33 0.24 0.17 0.18 0.11
Lu;v;w ; m=4 0.15 0.12 0.08 0.08 0.05 0.06
a; m=4 0.00 -0.01 0.06 0.08 0.03 0.08
3.2.1 Benchmark
The Nordtank 500 kW turbine was used as example turbine for the validation of the simulation pakages,
and the investigated load cases correspond to the load cases involved in the parameter analysis. Only two
of the codes, HAWC (Risoe) and VIDYN (Teknik Gruppen), were involved in the benchmark test, as the
time did not permit ECN to finetune FATAS to the Nordtank 500 kW turbine simulations.
The results obtained from the HAWC and VIDYN simulations are presented in Table 3.2 and in Ta-
ble 3.3 below, where the relative changes in equivalent moments/forces, caused by the specified wake
parameter variations, are presented. For the blade material, the Wöhler exponent m = 12 was considered
representative, whereas for the tower loading, a Wöhler exponent equal to 4 was selected.
In generel there is an acceptable degree of agreement between results produced by the two codes for
the 5D spacing. However, there is a tendency of increasing discrepancies with increasing mean wind
speed. The phenomenon is especially pronounced for the "deep stall" load situation, and it might thus
be associated with different stall modeling. For the "below stall" and "stall" regimes the agreement is
good, except for the relative increase in flap– and tilt moments caused by the coherence decay variation
and for the relative increase in the edge moment originating from the turbulence intensity variation. The
latter might be caused by different structural damping in the edge wise direction where the aerodynamic

Table 3.3 Benchmark results for a 8 diameter spacing. The relative changes in equivalent moments/forces
are presented for the HAWC simulations (Risoe), and for the VIDYN simulations (TG).
Load Wake 10 m/s 14 m/s 18 m/s

Type Par. Risoe TG Risoe TG Risoe TG
Flap V ; m=12 0.04 0.02 0.01 0.04 0.01 0.00
I u;v;w ; m=12 0.26 0.24 0.16 0.14 0.15 0.08
Lu;v;w ; m=12 0.06 0.05 0.03 0.05 0.02 0.02
a; m=12 0.00 0.00 0.00 0.00 0.00 0.00
Edge V ; m=12 0.03 0.00 0.01 0.01 0.00 0.01
I u;v;w ; m=12 0.06 0.04 0.06 0.03 0.03 0.02
Lu;v;w ; m=12 0.02 0.02 0.02 0.02 0.01 0.01
a; m=12 0.00 0.00 0.00 0.00 0.00 0.00
Tilt V ; m=4 0.00 -0.02 0.00 -0.01 0.00 0.01
I u;v;w ; m=4 0.29 0.25 0.18 0.14 0.15 0.08
Lu;v;w ; m=4 0.12 0.10 0.06 0.06 0.03 0.04
a; m=4 0.00 0.00 0.00 0.00 0.00 0.00
Yaw V ; m=4 -0.01 0.00 -0.01 -0.01 0.00 0.00
I u;v;w ; m=4 0.28 0.25 0.18 0.14 0.15 0.07
Lu;v;w ; m=4 0.13 0.10 0.06 0.06 0.03 0.03
a; m=4 0.00 0.00 0.00 0.00 0.00 0.00
Thrust V ; m=4 -0.01 0.00 0.00 0.01 -0.01 0.00
I u;v;w ; m=4 0.28 0.26 0.19 0.13 0.15 0.08
Lu;v;w ; m=4 0.12 0.08 0.05 0.06 0.03 0.03
a; m=4 0.00 0.00 0.00 0.00 0.00 0.00
damping is modest.
In the 8D spacing situation the agreement is very convincing, resulting in an identical mutual ranking of
the involved parameters regarding fatigue life consumption in wake situations. However, also in the 8D
situation, the turbulence intensity variation in the "deep stall" situation display some differences.
3.2.2 Results
The described parameter analysis has been performed for the four different wind turbines introduced
previously. From the analysis it appears, that the effects, originating from the investigated parameter
variations, act approximately additively. In the tables 3.4 through 3.23 the relative changes in equivalent
moments/forces, originating from the specified single parameter variations, are presented for the involved
load cases and turbines, based on selected representative Wöhler exponents.
The notation in the tables is to be interpreted as follows : "Pitch; 5D" refers to results associated with
the Vestas V27 turbine and a 5D turbine spacing, "Var.; 5D" refers to results associated with the artificial
500 kW variable speed turbine and a 5D turbine spacing, "Stall; 5D" refers to results associated with the
Danwin 180 kW turbine and a 5D turbine spacing, and "2–Bl.; 5D" refers to results associated with the

Näsudden turbine and a 5D turbine spacing. The 8D spacing examples are indicated analogues with "5D"
replaced by "8D".
Table 3.4 Relative increase in equivalent flapwise moment based on Wöhler exponent 12 and caused by
the wake deficit.
Load Case 10 m/s 14 m/s 18 m/s
Pitch; 5D 2% 0% 1%
Var.; 5D -8% 3% 0%
Stall; 5D 7% 9% 1%
2–Bl.; 5D 5% 0% 4%
Pitch; 8D 1% -1% 1%
Var.; 8D -9% 1% 0%
Stall; 8D 0% 8% 2%
2–Bl.; 8D 5% -6% 2%
Table 3.5 Relative increase in equivalent edgewise moment based on Wöhler exponent 12 and caused
by the wake deficit.
Pitch; 5D 3% 1% 1%
Var.; 5D -7% -3% -3%
Stall; 5D 2% 1% -1%
2–Bl.; 5D 2% 3% 1%
Pitch; 8D 3% 0% 1%
Var.; 8D -5% -2% -2%
Stall; 8D 1% 1% -1%
2–Bl.; 8D 1% 2% 1%

Table 3.6 Relative increase in equivalent tilt moment based on Wöhler exponent 4 and caused by the
wake deficit.
Pitch; 5D -2% 0% 0%
Var.; 5D -11% -1% -1%
Stall; 5D -2% 3% 2%
2–Bl.; 5D 12% 9% 3%
Pitch; 8D -1% 0% 0%
Var.; 8D -10% 1% 0%
Stall; 8D 4% 1% 0%
2–Bl.; 8D 7% 7% 2%
Table 3.7 Relative increase in equivalent yaw moment based on Wöhler exponent 4 and caused by the
wake deficit.
Pitch; 5D -3% -1% 0%
Var.; 5D -12% 1% 0%
Stall; 5D -5% -3% 1%
2–Bl.; 5D 5% 7% 2%
Pitch; 8D -2% 0% 0%
Var.; 8D -9% 1% 0%
Stall; 8D 3% -2% 1%
2–Bl.; 8D 4% 3% 2%
Table 3.8 Relative increase in equivalent thrust force based on Wöhler exponent 4 and caused by the
wake deficit.
Pitch; 5D -1% -1% 0%
Var.; 5D -31% 2% -2%
Stall; 5D -1% 3% -7%
2–Bl.; 5D -25% -14% -5%
Pitch; 8D 1% 0% 0%
Var.; 8D -24% 0% -2%
Stall; 8D -10% -7% 3%
2–Bl.; 8D -17% -12% -3%

Table 3.9 Relative increase in equivalent flapwise moment based on Wöhler exponent 12 and caused by
modified turbulence intensity.

Pitch; 5D 31% 17% 12%
Var.; 5D 33% 13% 7%
Stall; 5D 24% 22% 10%
2–Bl.; 5D 26% 16% 9%
Pitch; 8D 24% 11% 11%
Var.; 8D 26% 10% 5%
Stall; 8D 2% 9% 9%
2–Bl.; 8D 22% 12% 7%
by modified turbulence intensity.

Pitch; 5D 3% 2% 3%
Var.; 5D 2% 2% 1%
Stall; 5D 2% 2% 1%
2–Bl.; 5D 4% 5% 3%
Pitch; 8D 0% 1% 2%
Var.; 8D 1% 1% 1%
Stall; 8D 1% 2% 1%
2–Bl.; 8D 3% 3% 2%
Table 3.11 Relative increase in equivalent tilt moment based on Wöhler exponent 4 and caused by

Pitch; 5D 31% 17% 12%
Var.; 5D 40% 17% 8%
Stall; 5D 24% 20% 13%
2–Bl.; 5D 20% 16% 8%
Pitch; 8D 22% 13% 10%
Var.; 8D 32% 14% 6%
Stall; 8D 4% 15% 10%
2–Bl.; 8D 17% 12% 6%

Table 3.12 Relative increase in equivalent yaw moment based on Wöhler exponent 4 and caused by

Pitch; 5D 35% 21% 15%
Var.; 5D 41% 18% 8%
Stall; 5D 19% 14% 11%
2–Bl.; 5D 21% 13% 5%
Pitch; 8D 25% 16% 14%
Var.; 8D 33% 14% 7%
Stall; 8D 7% 10% 8%
2–Bl.; 8D 18% 10% 4%
Table 3.13 Relative increase in equivalent thrust force based on Wöhler exponent 4 and caused by

Pitch; 5D 40% 22% 15%
Var.; 5D 37% 16% 8%
Stall; 5D 27% 16% 6%
2–Bl.; 5D 19% 19% 7%
Pitch; 8D 29% 17% 13%
Var.; 8D 28% 12% 6%
Stall; 8D 3% 12% 4%
2–Bl.; 8D 13% 13% 6%
Table 3.14 Relative increase in equivalent flapwise moment based on Wöhler exponent 12 and caused
by modified turbulence length scale.

Pitch; 5D 6% 0% 1%
Var.; 5D 0% 1% 0%
Stall; 5D 7% 4% 5%
2–Bl.; 5D 7% 4% 3%
Pitch; 8D 5% -1% 0%
Var.; 8D 2% 1% 1%
Stall; 8D 3% 5% 3%
2–Bl.; 8D 5% 4% 1%

by modified turbulence length scale.

Pitch; 5D 1% 1% 0%
Var.; 5D 0% 0% 0%
Stall; 5D 1% 1% 1%
2–Bl.; 5D 2% 2% 2%
Pitch; 8D -2% 0% -1%
Var.; 8D 0% 0% 0%
Stall; 8D 1% 1% 0%
2–Bl.; 8D 1% 1% 1%
modified turbulence length scale.

Pitch; 5D 13% 5% 1%
Var.; 5D 2% 0% 1%
Stall; 5D 13% 8% 6%
2–Bl.; 5D 8% 5% 4%
Pitch; 8D 9% 3% 0%
Var.; 8D 3% 2% 2%
Stall; 8D 4% 6% 4%
2–Bl.; 8D 7% 5% 3%

Pitch; 5D 16% 6% 2%
Var.; 5D 3% 1% 0%
Stall; 5D 10% 6% 5%
2–Bl.; 5D 8% 4% 3%
Pitch; 8D 10% 4% 0%
Var.; 8D 4% 2% 2%
Stall; 8D 5% 4% 4%
2–Bl.; 8D 7% 3% 2%


Pitch; 5D 15% 7% 3%
Var.; 5D 2% 1% 1%
Stall; 5D 11% 7% 3%
2–Bl.; 5D 6% 2% 2%
Pitch; 8D 11% 4% 1%
Var.; 8D 3% 3% 2%
Stall; 8D 2% 2% 2%
2–Bl.; 8D 4% 3% 1%
Table 3.19 Relative increase in equivalent flapwise moment based on Wöhler exponent 12 and caused
by modified coherence decay.

Pitch; 5D -1% -6% -2%
Var.; 5D -4% -5% 3%
Stall; 5D -2% -1% -2%
2–Bl.; 5D -4% -8% -1%
Pitch; 8D 0% 0% 0%
Var.; 8D 0% 0% 0%
Stall; 8D 0% 0% 0%
2–Bl.; 8D 0% 0% 0%
by modified coherence decay.

Pitch; 5D 1% 4% 7%
Var.; 5D 0% 0% 0%
Stall; 5D 1% 1% 1%
2–Bl.; 5D 0% 0% 0%
Pitch; 8D 0% 0% 0%
Var.; 8D 0% 0% 0%
Stall; 8D 0% 0% 0%
2–Bl.; 8D 0% 0% 0%

modified coherence decay.

Pitch; 5D 8% 9% 7%
Var.; 5D 8% 8% 7%
Stall; 5D 3% 4% 0%
2–Bl.; 5D -3% -5% -4%
Pitch; 8D 0% 0% 0%
Var.; 8D 0% 0% 0%
Stall; 8D 0% 0% 0%
2–Bl.; 8D 0% 0% 0%

Pitch; 5D 8% 3% -6%
Var.; 5D 9% 10% 11%
Stall; 5D 5% 3% 2%
2–Bl.; 5D -12% -14% -10%
Pitch; 8D 0% 0% 0%
Var.; 8D 0% 0% 0%
Stall; 8D 0% 0% 0%
2–Bl.; 8D 0% 0% 0%
Pitch; 5D 7% 9% 6%
Var.; 5D -1% 12% 17%
Stall; 5D -2% -4% -2%
2–Bl.; 5D 0% 0% 5%
Pitch; 8D 0% 0% 0%
Var.; 8D 0% 0% 0%
Stall; 8D 0% 0% 0%
2–Bl.; 8D 0% 0% 0%

3.2.3 Discussion
A number of qualitative findings can be extracted from the analysis. The wind field modifications caused
by the wake, and quantified in section 2, all tend to diminish with increasing wind turbine spacing. This
behaviour is also reflected in the life time consumption, as a general tendency of decreasing wake effects,
from the 5D situation to the 8D situation, is observed. The turbulence intensity– and the length scale
modifications are seen to decrease for increasing wind speed, and this behaviour is also reflected in the
investigated equivalent moments/forces.
The fatigue load dependance of the mean wind deficit with the mean wind speed is more complicated, as
three factors contribute simultaneously – decrease in mean wind speed, introduction of horizontal shear,
and modification of the vertical shear. In contradiction to the fatigue loading, caused by wake turbulence
intensity, the fatigue loading associated with the wake deficit indicates no general tendency with the mean
wind speed for the investigated turbine concepts and load types.
Tilt– and yaw moments exhibit the same behaviour except for the two–bladed turbine, where the increased
wake coherence decay results in significant differences. The two–bladers are, however, born with distinct
periodic properties of the rotor inertia often introducing phase–shifts in the turbine loads, even for very
modest changes in the mechanical properties. These phase–shifts often affect the loading in an unexpected
way, and the discrepancy between the tilt– and yaw loading, arising from an increased coherence in the
turbulence field, is believed to originate from such a phenomenon.
All the four investigated parameters were demonstrated to be significant in relation to fatigue life con-
sumption. However, in most of the investigated situations the effect originating from increase in the
turbulence intensity dominates the other parameter effects with a factor of 2–3.
It is moreover evident that different turbine concepts act differently with respect to a particular parameter
sensitivity, and that this also applies for different load types. More specific it is seen that the induced
wake deficit generally is favourable for the thrust force, whereas it, for the majority of the investigated
turbine concepts, introduces increased blade– and rotor fatigue loading.
Finally, it is observed that the increased wake coherence decay is favourable for the flap root moment,
whereas increased wake turbulence intensity is in general introducing higher fatigue loading.

4 CONCLUSIONS
The present report deals with the establisment of a procedure for wake load predictions, and its possible
inclusion in the IEC–code framework. The overall philosophy is to preserve the existing division in four
wind turbine classes, and simply supplement the existing load specifications with quantitative wake load
specifications.
The study is confined to single wake situations, but is anticipated to apply also for multiple wake situations.
A detailed investigation has been performed to quantify the wake load situation. The investigation falls
into three phases – a crude identification of key wake wind field parameters, a quantification of the
identified key parameters, and finally a detailed key parameter identification based on predicted wake
characteristics.
Applying a two–level–factorial method it has been demonstrated that, in the present analysis, 10–minutes
simulation time series are a sufficient representation of the individual load situations to yield significant
results.
The main findings from the investigation are :
That all four investigated parameters – representing wake deficit, wake turbulence intensity, wake
turbulence length scale, and wake coherence – were demonstrated to be significant in relation to
increased fatigue life consumption in wakes compared to ambient conditions.
That it is possible to quantify the relevant wake parameters by use of simple models.
That the fatigue impact, caused by the investigated four parameters, are shown to act additively,
which imply that a simple approximative method, involving only a very limited number of aeroe-
lastic calculations, can be applied for fatigue estimates taking into account the detailed wind farm
topology and the particular wind turbine concept [17].
That the effect due to increased turbulence intensity is crucial and usually dominates the fatigue
effects caused by the other investigated parameters with a factor of 2–3.
That different turbine concepts and different load types act differently with respect to parameter
sensitivity – this fact demonstrates the appropriateness in a detailed wake load estimation based on
the philosophy applied here, compared to the "all factors in one parameter" approach previously
used.


5 RECOMMENDATIONS
The present section deals with inclusion of wake load description in the IEC–code framework and
contains, in a condensed version, the recommendations resulted from the present investigations.
When operating in a wake situation, the mean wind profile as well as the turbulence characteristics
experienced by a turbine are modified compared to the ambient situation. The significant modifications
are quantified in the following four subsections.
5.1 Mean Wind Deficit
The wake deficit in the mean wind direction, V , and the wake extension, expressed in terms of the wake
radius, Rw , are estimated according to the following expressions :
35 1
Rw = 2 (3c21 ) 15 (CT A(x + x0 )) 31 ;

5
Va ; 1 n 3
35 103 o2

V = ; 9 CT A(x + x0 ) ; 2 3
r (3c1 CT A(x + x0 )) ; 2 (3c21 ); 51 ;
2 2 ; 1
2
where x and r denote axial and radial directions, A is the rotor area, CT denotes the rotor drag coefficient,
Va the ambient mean wind velocity at hub height, and c1 and x0 are parameters approximated by :
x0 = ; 2R99 :55D3 ;
D
:
;1
D ; 1
(cT Ax0 ); 6 ;
2
c1 = 2
5
where D is the diameter of the upstream rotor. Denoting the ambient turbulence intensity by Ia and the
hub height by H , R9:5 is determined from :
R9:5 = 21 (Rnb + min(H; Rnb )) ;

Rnb = max(1:08D; 1:08D + 21:7D(Ia ; 0:05)) :
The wake efffect on the mean flow in directions perpendicular to the mean wind direction shall not be
considered.
5.2 Turbulence Intensity
At down stream distances corresponding to traditional spacing in wind farms it is assumed that only the
surface– and wake shear mechanisms contribute significantly to turbulence production. It is moreover

assumed that the turbulence fluctuations originating from these two sources are statistically independent
such that the turbulent energies are additive in the energy sense. Normalizing with respect to the
undisturbed mean wind velocity, Va , the total turbulence intensity of the along wind turbulence component
in the wake, Iwt
u , is expressed as :
q
u = I2 + I2 ;
Iwt a w
where subscripts a and w refer to ambient and wake, respectively. The specific wake contribution, Iw ,
depends on both the down wind distance and the undisturbed mean wind velocity and is (for spacings
larger than two rotor diameters) determined as :
q p
Iw = 0:29S ;1=3 1 ; 1 ; CT ;
where S denotes the spacing expressed in rotor diameters.

Compared to ambient situations the wake turbulence tends to be more isotropic, and the energy content
in the transversal turbulence components, v and w, are determined from :
u V ;
wv ' 0:8Iwt a
u V ;
ww ' 0:6Iwt a
where subscript "w" indicate wake–values.
5.3 Turbulence Length Scales
When the two shear turbulence contributions have scales of comparable orders of magnitude, it is
reasonable to expect the resulting spectral shape to be qualitatively invariant. However, the resulting
length scale is modified. The modified length scale depends on the rotor thrust coefficient, the spacing
distance, the rotor diameter, and the undisturbed turbulence length scale.
Refeering to the Kaimal formulation of the turbulence spectrum given in [18], the length scale for the u
turbulence component inside the wake, Luw , is expressed as :
0 0 D 11
12:2 1 ;
Luw = Lua @1 ; min @ 0:6 L ; 0:9AA ;
u
a
Va Sr
where Lua denotes the u–turbulence length scale in the undisturbed flow, and Sr is defines as :
Sr = 20:8 + (S ; 2)0:8 for S 2:

Assuming local isotropy in the inertial subrange for sufficiently small eddies, the following relations can
be derived for the wake turbulence length scales in the v – and w–directions :
Lvw = 0:33Luw
Lww = 0:14Luw :
5.4 Coherence Decay Factors
Assuming a Davenport–formulation of the coherence, the uu–coherence decay factor related to wake
turbulence, aw , shall determined according to :
(
aw = (1 + Kw )aa for 2 S < 7:5 ;
aa for 7:5 S
where aa is the coherence decay factor related to ambient conditions, and the wake correction factor, Kw ,
is given by
8
>
< 7:5;S for Va < 6
Kw = > 7:55:;5 S ( 1 Va ; Cs ) for 6 Va 15 :
: 4(75::55;S3) for 15 < Va
5:5
For vertical separation Cs 1 and for horizontal separation Cs 2. For spacings less than 2 rotor
diameters the value of the wake coherence decay factor is assumed to equal the value related to a 2
diameter spacing.

ACKNOWLEDGEMENTS
The work has been supported by the European Commission under contract JOR–CT95–00642. Stimu-
lating discussions and assistance from our colleague Per Vølund, Wind Energy and Atmos. Phys. Dept.,
Risø National Laboratory, is furthermore acknowledged. Last but not least the contribution of Mikael
Magnusson, Department of Meteorology, Uppsala University, is appreciated.

References
[1] Adams, B.M. (1996). Dynamic loads in Wind Farms 2. Final report of Joule project JOU2–CT92–
0094. GH-report 286/R/1, Garrad Hassan & Partners.
[2] Larsen, G.C., Højstrup, J., and Madsen, H.Aa. (1996). Wind Fields in Wakes. EUWEC’96, Göteborg
1996.
[3] Frandsen, S. et. al. (1996). Measurements on and Modelling of Offshore Wind Farms. Risoe–R–
903(EN).
[4] Verheij, F.J. (1993) Wind Data for Design of Wind Turbines Operating in Wind Farms (in Dutch).
TNO – Environmental and Energy Research, 93-189.
[5] Magnusson, M. (1996). Wind turbine "Wakes". Ph.d. dissertation, Acta Universitatis Upsaliensis,
Uppsala.
[6] Crespo, A. et al. (1985). Numerical Analysis of Wind Turbine Wakes. Proc. of the Delphi Workshop
on Wind Energy Applications.
[7] Patankar, S.V. and Spalding, D.B. (1972). A Calculation Procedure for Heat, Mass and Momentum
Transfer in Three Dimensional Parabolic Flows. Int. Journal of Heat and Mass Transfer, Vol. 15.
[8] Panofsky, H.A. and Dutton, J.A. (1983). Atmospheric Turbulence. John Wiley & Sons.
[9] Hoejstrup, J. and Noergaard, P. (1990). Taendpibe Wind Farm Measurements 1988. Risoe–M–2894.
[10] Papaconstantinou, A. and Bergeless, G. (1988). Hot–wire measurements of the flow field in the
vicinity of a HAWG rotor. Journal of Wind Engineering and Industrial Aerodynamics, 31, p. 133–
146.
[11] Smith, D. (1990). Multiple wake measurements and analysis. Proceedings of 12’th BWEA Confer-
ence.
[12] Tindal, A.J. (1993). Dynamic loads in wind farms I. GH-report 205/R/12, Garrad Hassan & Partners.
[13] Seignette P.F.A.B (1996). On-line PVOPT user’s manual, version 1.2. Netherlands Energy Research
Foundation, ECN.
[14] Schepers, J.G. (1996). PVOPT, Theory and test cases. ECN-C–96-057,Netherlands Energy Research
Foundation, ECN.
[15] Schepers, J.G. (1996). Design of variable speed turbine to be used in the WF-PNL program.
Netherlands Energy Research Foundation, ECN.
[16] Box, G.E.P. et al.(1978). Statistics for Experimenters – An Introduction to Design, Data Analysis,
Data Analysis, and Model Building. John Wiley & Sons.
[17] Petersen, S.M. et al. (1997). EWTS II – Load Spectra and Extreme Wind Conditions. EWEC’97,
Dublin 1997.
[18] Danish Code of Practice DS472 (1992). Loads and Safety of Wind Turbine Structures (in Danish).
Teknisk Forlag.

[19] Alberts, A. and Söker, H. (1997). Wind Speed and Turbulence Evaluations from Power data and the
Nacelle Anemometer.
[20] Fragoulis, A. et al. (1996). MOUNTURB FINAL REPORT – Load and Power Measurement Program
on Wind Turbines Operating in Complex Mountaneous Regions. Final report of Joule project JOU2–
CT93–0378. CRES.

A VESTAS V27 TURBINE
The present appendix contains a brief description of the V27 wind turbine, which is representing the
pitch regulated turbine concept in the performed wake parameter analysis. Basically, the V27 turbine is
a three bladed upwind horizontal axis wind turbine with active yaw and a high speed rotor. The main
specifications of the turbine is given in Table A.1 below.
Table A.1 Main specifications for the Vestas V27 turbine.
ROTOR
Power regulation Pitch
Number of blades 3
Rotor diameter 27.00 m
Swept area 573 m2
Hub height 28.65 m
Rotor speed 44 rpm/33 rpm
Blade tip angle 0 – 90 during operation
BLADES
Manufacturer Vestas
Spar and shell material GRP
Total blade length 13.00 m
Profiled blade length 11.50 m
Root chord 1.30 m
Tip chord 0.50 m
Blade twist 13.0
Blade profiles NACA 63–600
First flapwise natural frequency 2.38 Hz
First edgewise natural frequency 3.41 Hz
GEAR BOX
Gear ratio 1 : 27.6
GENERATOR
Manufacturer Siemens; double winding
Nominal power 225 kW
Voltage 480 V AC
Frequency 60 Hz
Rotational speed 1210 rpm (225 kW) / 910 rpm (50 kW)
TOWER
Type Lattice tower
Height 27.43 m
First bending natural frequency 1.31 Hz
CONTROL SYSTEM
Manufacturer Vestas
Cut–in wind speed 3.7 m/s
Cut–out wind speed 25.0 m/s

Table A.1 Main specifications for the Vestas V27 turbine (cont).
WEIGHTS
Blade 600 kg
Rotor 2900 kg
Nacelle excl. rotor 7900 kg
Tower 600 kg
Total 19700 kg
A Theoretical thrust curve, which is vital for the quantification of wake effects, has been supplied from
the manufacturer and is presented in Table A.2.
Table A.2 Calculated thrust curve for the Vestas V27 turbine.
Mean wind speed Thrust Thrust coefficient

[m=s] [kN ] CT
3.5 3.8 0.88
4.0 4.9 0.88
5.0 7.7 0.88
6.0 10.0 0.79
7.0 12.5 0.73
8.0 18.0 0.80
9.0 21.0 0.74
10.0 24.6 0.70
11.0 26.7 0.63
12.0 27.8 0.55
13.0 26.7 0.45
14.0 22.7 0.33
15.0 20.5 0.26
16.0 18.9 0.21
17.0 17.2 0.17
18.0 17.0 0.15
19.0 15.2 0.12
20.0 15.4 0.11
21.0 13.9 0.09
22.0 13.6 0.08
23.0 13.0 0.07
24.0 12.1 0.06
25.0 13.2 0.06
Note, that the jump in CT between the mean wind speeds 7 m/s and 8 m/s is caused by shift between
generators.

B VARIABLE SPEED 500 KW TURBINE
The present appendix contains a brief description of an artificial wind turbine which represents the variable
speed regulated turbine concept in the wake parameter analysis. The defined turbine is a three bladed,
500 kW, upwind horizontal axis wind turbine.
B.1 Background
The turbine has been designed in the ECN Basic Subsidy project "FyndFarm". The design is derived
from the fixed speed, stall and pitch regulated, turbines which have been used in the development of wind
farm design guidelines in the Dynamic Loads in Wind Farms II (DLWF-II) project [1]17 .
Obviously, the design of these DLWF-II turbines needed some adjustment in order to account for variable
speed operation :
An optimal tip speed ratio (design = 7.5) operation is assumed up to rated wind speed.
Above Vrated the rotor speed and power are regulated by means of a pitch control algorithm. This
algorithm is based on the one which is used for the constant speed, pitch regulated DLWF-II turbine.
The aerodynamic design (chord and twist distribution) is found from an aerodynamic optimization,
using the program PVOPT, [13] and [14].
The mass and stiffness distributions are almost similar to the distributions of the DLWF-II turbines.
However, some slight modifications were required in order to avoid operation a first flapwise natural
frequency of 3P at rated conditions.
It should be noted that for the present study, only blade flexibility in the flapwise direction was taken into
account. Hence the blade flexibility in the edgewise direction, the tower flexibility, and the drive train
flexibility have been neglected.
The main specifications of the turbine are given in Table B.1. For a more detailed design report, reference
is made to [15]. The calculated (stationary) thrust, power, pitch angle, and rotor speed are presented in
Table B.2 as function of wind speed.
17 See section 6.

Table B.1 Main specifications for the 500 kW variable speed turbine.
ROTOR
Power regulation Pitch / Variable speed
Number of blades 3
Swept area 1075 m2
Hub height 34.95 m
Rotor speed range 20.2 rpm – 46.0 rpm; see also Table B.2
Tilt 2 deg
Coning 0 deg
Direction of rotation Clockwise
BLADES
Root chord 2.150 m
Tip chord 0.765 m
Blade twist 12.625 deg
Blade profiles NACA 632xxx (inner) / NACA 634xxx (outer)
First flapwise natural frequency 2.239 Hz (standstill)
GEAR BOX
Gear ratio 1 : 39.1
GENERATOR; see also section B.3
TOWER
Type Tapered tubular tower
Height 34.95 m
Top diameter 1.68 m
Base diameter 2.41 m
CONTROL SYSTEM, see also section B.4
Cut–in wind speed 5 m/s
Rated wind speed 11.9 m/s
Cut–out wind speed 23 m/s
WEIGHTS
Blade 1853 kg
Rotor incl. hub 8780 kg

Table B.2 Stationary rotor characteristics.
V CP;aer CT Paer Thrust
[m/s] [kW] [kN] [deg] [rpm]

5.000 0.494 0.855 40.690 14.080 0.000 20.211
6.000 0.494 0.850 70.340 20.140 0.000 23.967
7.000 0.496 0.835 112.000 26.940 0.000 26.982
8.000 0.496 0.837 167.100 35.260 0.000 30.954
9.000 0.496 0.829 238.100 44.220 0.000 34.241
10.000 0.496 0.837 326.400 55.110 0.000 38.719
11.000 0.496 0.836 434.500 66.650 0.000 42.542
12.000 0.495 0.787 563.800 74.660 0.810 46.017
13.000 0.381 0.498 550.700 55.430 4.962 46.101
14.000 0.306 0.380 552.500 49.100 6.977 46.108
15.000 0.250 0.303 555.500 44.850 8.638 46.122
16.000 0.207 0.247 558.600 41.660 10.099 46.138
17.000 0.174 0.206 562.200 39.140 11.441 46.155
18.000 0.147 0.174 566.000 37.100 12.695 46.173
19.000 0.126 0.149 570.400 35.420 13.883 46.195
20.000 0.109 0.129 574.900 33.990 15.019 46.221
21.000 0.093 0.111 568.300 32.180 16.184 46.015
22.000 0.079 0.095 552.500 30.140 17.352 45.797
23.000 0.068 0.082 542.700 28.600 18.447 45.668
B.2 Rated Conditions and Losses
The losses at rated conditions remained unchanged from the losses of the DLWF-II turbines ( 7%).
As a consequence, the rated aerodynamic power was 537.5 kW, which resulted in a rated wind speed of
11.9 m/s. The rated rotorspeed was 46.03 rpm, see also Table B.2.
B.3 Generator curve
The generator curve (i.e. the generator power vs. the rotational speed) is basically a conventional third
order curve. Above the rated power of 500 kW, the generator power is kept constant. Some modifications
have been applied, in order to avoid operation at a rotational speed which is close to a multiple of the first
flapwise natural frequency.
B.4 Pitch control
Above rated conditions it is attempted to keep the rotor speed (and consequently the power) at its rated
values by means of a pitch control. The pitch control algorithm is almost similar to the one which was
used to control the power of the fixed speed DLWF-II turbine; see [1] :

PI proportional gain = 8.448 10;7 rad/W.
PI integral gain = 7.04 10;6 rad/(Ws).
1
Gain scheduling via pitch angle with scheduling factor equal to 0:6375+20 :77 :
Limits on schedule factor : 0:08677 < fact < 1.
The controller has been designed with a first order pitch actuator response with a time constant
equal to 0.3 s.
However, for the present purpose, the controller should act on the rotor speed instead of the power. To
this end, a dependency is introduced between the generator power set point and the rotor speed: if the
rotor speed exceeds the rated value of 46.03 rpm, the set point is reduced with P :
Pset = Prated ; P ;
ZT
P = kn (
;
rated) dt (
in rad=s) ;
0
kn = 1:0 kW
rad :

C DANWIN 180 KW TURBINE
The present appendix contains a brief description of the Danwin 180 kW wind turbine which, together
with the Nordtank 500 kW turbine, is representing the stall regulated turbine concept in the performed
wake parameter analysis.
The Danwin 180 kW turbine is a three bladed upwind horizontal axis wind turbine with active yaw. The
main specifications of the turbine is given in Table C.1 below.
Table C.1 Main specifications for the Danwin 180 kW turbine.
ROTOR
Power regulation Stall
Number of blades 3
Swept area 422.7 m2
Hub height 30.00 m
Rotor speed 42.8 rpm
Tilt 5
Coning 0
Blade tip angle 0.8
BLADES
Manufacturer LM Glasfiber A/S
Type LM 11H
Root chord 1.30 m
Tip chord 0.59 m
Blade twist 16.0
Blade profiles NACA 632xx
Air brakes Pivotable blade tips
GEAR BOX
Manufacturer Jahnel/Kestermann
Gear ratio 1 : 41.7765
GENERATOR
Manufacturer Hawker Siddeley
Type C315 M Sp
Voltage 480 V
Frequency 60 Hz
Rotational speed 1000/750???? rpm

Table C.1 Main specifications for the Danwin 180 kW turbine (cont.).
MECHANICAL BRAKE
Type Disc
Function Fail Safe
TOWER
Height 29.8 m
Number of sections 2
Top diameter 1.24 m
CONTROL SYSTEM
Manufacturer Orbital
Cut–in system Thyristors
Logic system Microprocessor
WEIGHTS
Blade 680 kg
Tower 10000 kg
A measured thrust curve was not available for the present turbine, and as a consequence the wake analysis
is based on a calculated thrust curve which, for completeness, is presented in Table C.2. The thrust curve
was determined with the aeroelastic code VIDYN.

Table C.2 Calculated thrust curve for the Danwin 180 kW turbine.

[m=s] [kN ] CT
4.0 3.76 0.92
5.0 5.76 0.90
6.0 8.02 0.87
7.0 10.3 0.82
8.0 12.7 0.78
9.0 15.1 0.73
10.0 17.3 0.68
11.0 19.1 0.62
12.0 20.5 0.56
13.0 21.3 0.50
14.0 22.0 0.44
15.0 22.3 0.39
16.0 22.5 0.34
17.0 22.7 0.31
18.0 23.0 0.28
19.0 23.4 0.25
20.0 24.0 0.23
21.0 25.0 0.22
22.0 26.5 0.21
23.0 28.2 0.21
24.0 29.4 0.20
25.0 30.3 0.19


D NÄSUDDEN 3 MW TURBINE
The present appendix contains a brief description of the Näsudden 3 MW wind turbine which represents
a large two bladed pitch regulated turbine concept in the performed wake parameter analysis.
The Näsudden II demonstration turbine is a two bladed upwind horizontal axis wind turbine with active
yaw. The tower is very stiff on this prototype machine, and in order to approach a more realistic future
design, the original tower has been replaced by the more flexible tower belonging to the sister turbine,
Aeolus II, in the model calculations. The main specifications of the turbine, equipped with the soft tower,
is given in Table D.1 below.
Table D.1 Main specifications for the Näsudden 3 MW turbine.
ROTOR
Power regulation Pitch
Number of blades 2
Swept area 5026.5 m2
Hub height 92.00 m
Tilt 8
Coning 0
BLADES
Manufacturer Messerschmitt–Bölkow–Blohm Gmbh.
Root chord 4.40 m
Tip chord 0.65 m
Blade twist 7.2
Blade profiles FX–W–xxx
GEAR BOX
Manufacturer Lohmann&Stolterfoht
Type Planetary/bevel
GENERATOR
Manufacturer ABB
Type Induction/2 speed
Nominal power 3MW
Voltage 6 kV
Frequency 50 Hz
Rotational speed 1000/1500 rpm

Table D.1 Main specifications for the Näsudden 3 MW turbine (cont.).
MECHANICAL BRAKE
Manufacturer Kvaener Turbin AB
Type Disc
Function Fail Safe
TOWER
Type Tapered tubular tower18
Height 87.7 m
Top diameter 4.50 m
CONTROL SYSTEM
Manufacturer Kvaener Turbin AB
Type PLC
WEIGHTS
Blade 9250 kg
Tower 1410000 kg
is based on a calculated thrust curve which, for completeness, is presented in Table D.2. The thrust curve
18 Reinforced concrete.

Table D.2 Calculated thrust curve for the Näsudden 3 MW turbine.

[m=s] [kN ] CT
3.5
4.0 60 1.21
5.0 86 1.11
6.0 112 1.00
7.0 138 0.91
8.0 163 0.82
9.0 191 0.76
10.0 217 0.70
11.0 243 0.64
12.0 270 0.61
13.0 295 0.56
14.0 309 0.51
15.0 290 0.42
16.0 258 0.33
17.0 229 0.26
18.0 211 0.21
19.0 197 0.18
20.0 186 0.15


E NORDTANK NTK500 TURBINE
The present appendix contains a brief description of the Nordtank 500 kW wind turbine which, together
with the Danwin 180 kW turbine, is representing the stall regulated turbine concept in the performed wake
parameter analysis. The NTK500 turbine is moreover used as a "reference" turbine in the benchmark
exercise performed with the three involved aeroelastic codes – HAWC, VIDYN, and PHATAS.
The Nordtank 500 kW turbine is a three bladed upwind horizontal axis medium sized wind turbine with
active yaw. The main specifications of the turbine is given in Table E.1 below.
Table E.1 Main specifications for the Nordtank 500 kW turbine.
ROTOR
Power regulation Stall
Number of blades 3
Swept area 1075 m2
Hub height 35.00 m
Tilt 2
Coning 0
Blade tip angle -2
Direction of rotation Clockwise
BLADES
Manufacturer LM Glasfiber A/S
Type LM 17.0 m
Root chord 1.85 m
Tip chord 0.40 m
Blade twist 19.8
Blade profiles NACA 63; modified
Air brakes Pivotable blade tips
GEAR BOX
Manufacturer FLENDER
Type PEAK 4280
MECHANICAL BRAKE
Manufacturer Svendborg Brakes
Type BSFH 317
Function Fail Safe

Table E.1 Main specifications for the Nordtank 500 kW turbine (cont.).
GENERATOR
Manufacturer ABB
Type HXR 355–LC 4B 3E
Voltage 690 V
Nominal current 460 A
Cos phi 0.91
Frequency 50 Hz
Rotational speed 1500 rpm
TOWER
Height 33.8 m
Number of sections 2
Top diameter 1.69 m
CONTROL SYSTEM
Manufacturer DANCONTROL
Type NTK DC 04
WEIGHTS
Blade 1620 kg
Tower 22500 kg
is based on a calculated thrust curve which, for completeness, is presented in Table E.2. The thrust curve

Table E.2 Calculated thrust curve for the Nordtank 500 kW turbine.

[m=s] [kN ] CT
4.0 17.1 1.61
5.0 22.3 1.34
6.0 27.5 1.16
7.0 33.0 1.02
8.0 38.1 0.90
9.0 44.0 0.82
10.0 49.3 0.75
11.0 53.3 0.67
12.0 56.1 0.59
13.0 58.1 0.52
14.0 59.0 0.46
15.0 59.6 0.40
16.0 60.2 0.36
17.0 60.8 0.32
18.0 61.8 0.29
19.0 62.5 0.26
20.0 63.4 0.24
21.0 64.4 0.22
22.0 65.4 0.20
23.0 66.5 0.19
24.0 67.5 0.18
25.0 68.6 0.17


F EXPERIMENTAL WAKE CHARACTERIZATION
For the Swedish results, the von Karman based length scales has been applied and then subsequently
transformed to the required Kaimal formulation. The conversion is based on the specifications in the
IEC–1400 document. The same transformation has been applied to the Dutch results.
Most of the Danish results are based on very long time series in order to improve the statistical significance.
The specified parameter values are obtained as averages of a number of half hour statistics, for which the
particular wake situation has occurred.

1Sub A.82
Table F.1 Wake characterization.
Spacing Rotor Hub Thrust Stability Amount Wind Ambient Max. Ambient Wake Ambient Wake Amb. transv. Wake transv. Amb. coh. Wake coh.
S [D] Diam. Height Coef. Ri of Data Turbine Mean Wind Deficit Turb. int. Turb. int. Lgd. scale Lgd. scale turb. comp. turb. comp. Decay Decay
u v
D [m] H [m] CT [hours] Type
Va [m/s] V [m/s] Iau Iwt Lua [m] Luw [m] av [m/s] w [m/s] aa aw
2.0 28 31 0.67 neutral 0.5 NTK500 9.9 4.4 14.0 16.5 1070 140
5.3 28 31 0.76 neutral 9.0 NTK500 7.3 1.2 11.9 16.2 1052 359
5.3 28 31 0.70 neutral 4.5 NTK500 9.1 1.4 12.9 16.2 1035 421
5.3 28 31 0.59 neutral 7.5 NTK500 11.5 1.6 13.7 17.5 1363 788
8.5 35 37 0.66 neutral 14.0 Bonus450 9.0 940 505
8.5 35 37 0.52 neutral 4.0 Bonus450 11.1 6.8 11.0
9.5 35 37 0.57 neutral Bonus450 10.0 7.2 11.3
2.0 28 31 neutral 3.0 NTK500 6.9(wake) 12(H) 31(H)
7.5 28 31 0.76 neutral 4.5 NTK500 7.5 0.6 10.9 14.0 12(H)/16(V) 12(H)/16(V)
7.5 28 31 0.70 neutral 1.5 NTK500 9.1 0.7 14.2 14.3 12(H)/16(V) 12(H)/16(V)
14.5 28 31 0.75 neutral 16.0 NTK500 7.9 0.5 11.7 12.3 12(H)/16(V) 12(H)/16(V)
14.5 28 31 0.69 neutral 11.0 NTK500 9.3 0.5 12.7 12.6 12(H)/16(V) 12(H)/16(V)
14.5 28 31 0.59 neutral 4.0 NTK500 11.7 0.3 13.8 14.7 12(H)/16(V) 12(H)/16(V)
4.2 23 30 0.44 0.04 0.33 Danw180 14.2 4.5 5.6 6.6 1306 598 0.57 0.98 16(V) 53(V)
6.1 23 30 0.67 -0.3 0.33 Danw180 9.7 2.4 6.4 11.4 1380 846 0.50 1.03 14(V) 27(V)
6.1 23 30 0.87 -0.15 0.33 Danw180 6.26 1.57 6.9 11.3 1776 672 0.38 0.76 16(V) 19(V)
6.1 23 30 0.87 -1.25 0.33 Danw180 6.71 1.78 5.4 12.5 1849 975 0.30 0.84 17(V) 25(V)
6.1 23 30 0.81 -0.5 0.5 Danw180 7.76 1.85 6.3 12.9 1730 837 0.35 0.96 13(V) 22(V)
4.2 23 30 0.86 -1.0 0.33 Danw180 6.94 3.00 4.8 15.6 1592 635 0.30 0.90 11(V) 20(V)
6.1 23 30 0.79 -2.0 0.5 Danw180 7.95 1.84 6.3 13.5 1711 828 0.43 0.94 13(V) 25(V)
4.2 23 30 0.83 -6.5 0.33 Danw180 7.49 3.03 6.0 16.0 1868 616 0.48 1.13 13(V) 25(V)
4.2 23 30 0.71 -1.7 0.33 Danw180 9.55 3.43 7.7 13.0 1196 690 0.57 1.20 12(V) 29(V)
4.2 23 30 0.83 0.07 0.5 Danw180 7.93 4.33 3.8 10.7 672 598 0.31 0.56 33(V) 33(V)
4.2 23 30 0.30 0.01 0.166 Danw180 17.4 4.60 7.0 9.3 1840 1270 0.97 1.14 13(V) 27(V)
4.2 23 30 0.52 0.03 0.166 Danw180 13.37 5.28 4.3 6.6 902 543 0.46 0.88 18(V) 54(V)
4.2 23 30 0.50 0.03 0.166 Danw180 13.74 5.34 5.9 6.6 1693 708 0.61 1.04 17(V) 57(V)
4 30 35 0.75 0.285 WPS-30 7.3 3.6 10.5 15.1 644 138 0.80 1.08
4 30 35 0.75 0.356 WPS-30 7.7 3.6 8.9 13.7 920 184 0.85 1.03
4 30 35 0.75 0.356 WPS-30 8.1 3.5 10.9 15.4 736 230 0.89 1.16
7 30 35 0.57 0.711 WPS-30 13.6 2.8 7.2 9.3
7 30 35 0.75 0.569 WPS-30 10.1 2.7 9.5 12.2
7 30 35 0.74 0.996 WPS-30 10.0 2.3 9.1 11.9
7 30 35 0.74 0.356 WPS-30 8.9 2.3 9.6 11.1
5 0.27 0.31 0.91 0.006 MODEL 4.13 1.05 9.0 14.1 0.355 0.535
7.5 0.27 0.30 0.87 0.008 MODEL 4.1 1.03 9.0 12.3
European Wind Turbine Standards II

G GRAPHIC COLLOCATION OF RESULTS
The following contains a graphical representation of results associated with wake deficits and wake
turbulence intensities.
The predictions by the investigated models are presented together with measurements (and relevant
ambient characteristics).
Cases indicated by "R", "A", "S", and "T" refer to Risø measurements, Alsvik measurements, Sexbierum
measurements, and wind tunnel measurements, respectively.

Figure G.1 Predictions of turbulence intensity compared to the Risø measurements.

Figure G.2 Predictions of turbulence intensity compared to the Alsvik measurements.

Figure G.3 Predictions of turbulence intensity compared to the TNO measurements.

Figure G.4 Predictions of wake deficit compared to the Risø measurements.

Figure G.5 Predictions of wake deficit compared to the Alsvik measurements.

Figure G.6 Predictions of wake deficit compared to the TNO measurements.

PART 1
SUB B
COMPLEX TERRAIN AND FATIGUE LOADING
P. Chaviaropoulos, (Editor) CRES

G. Glinou, (Editor) CRES
F.Mouzakis (Editor) CRES
D. Winkelaar Netherlands Energy Research Foundation ECN
B. Hendriks Netherlands Energy Research Foundation ECN
B. Heijdra Netherlands Energy Research Foundation ECN
S. Petersen RISØ National Laboratory
P. Voelund RISØ National Laboratory
G. Larsen RISØ National Laboratory
I. Carlen Technikgruppen AB
H. Ganander Technikgruppen AB
E. Morfiadakis CRES
K. Papadopoulos CRES
D. Douvikas CRES
P. Vionis CRES
A. Fragoulis CRES
Complex Terrain and Fatigue Loading
1 Sub B.2 European Wind Turbine Standards II

CONTENTS OF PART 1 SUB B
1. SUMMARY............................................................................................................................................... 5
2. NOMENCLATURE .................................................................................................................................. 7
3. INTRODUCTION ..................................................................................................................................... 9
4. MEAN WIND FIELD AND TURBULENCE STRUCTURE IN COMPLEX TERRAIN..................... 11
4.1 Approach followed ............................................................................................................................ 11
4.2 Measured wind field characteristics .................................................................................................. 11
4.2.1 Mean wind field.......................................................................................................................... 11
4.2.2 Turbulence structure ................................................................................................................... 12
4.3 Theoretical investigation ................................................................................................................... 13
4.3.1 Parametric set-up ........................................................................................................................ 14
4.3.2 Flow model ................................................................................................................................. 14
4.3.3 Obtained results .......................................................................................................................... 15
4.3.4 Complex terrain and wind shear ................................................................................................. 17
4.3.5 Complex terrain and turbulent structure ..................................................................................... 17
4.4 Synthesis of the results ...................................................................................................................... 17
5. MACHINE LOADING IN COMPLEX TERRAIN ................................................................................ 21
5.1 Approach followed ............................................................................................................................ 21
5.2 Modified wind field and machine loading; a measurement-based investigation............................... 21
5.2.1 Machines and sites considered.................................................................................................... 22
5.2.2 Investigated parameters .............................................................................................................. 22
5.3 Modified wind field and machine loading; a modelling-based investigation.................................... 24
5.3.1 Machines considered .................................................................................................................. 25
5.3.2 Parametric set-up ........................................................................................................................ 26
5.3.3 Results and conclusions.............................................................................................................. 28
5.4 Synthesis of the results ...................................................................................................................... 31
6. RECOMMENDATIONS......................................................................................................................... 35
7. REFERENCES ........................................................................................................................................ 37
Appendix A Speed-up and Turbulence Kinetic Energy computations for single and tandem hills ........... 39
Appendix B A simplified wind shear law for hill-tops................................................................................ 63
Appendix C Theoretical investigation of the Reynolds Stress Tensor in complex terrain [C1]................ 65
Appendix D Tables from measurement-based PI ........................................................................................ 71
Appendix E Tables from modelling-based PI.............................................................................................. 77
European Wind Turbine Standards II 1 Sub B.3


1. SUMMARY
The aim of this report is twofold. First to investigate the effect of the terrain complexity on the wind field
properties that drive the fatigue loading of wind turbines (like wind shear, Reynolds stresses, etc.).
Second, to investigate and quantify the impact of the above properties on fatigue loading for machines of
different size and design philosophy. The first goal is met by analysing the available experimental
complex terrain wind data bases, as well as by performing systematic parametric runs for simple, 2-D,
configurations including single and tandem hill layouts. In the latter case, results are presented in terms of
speed-up and turbulent kinetic energy distributions. The second goal is met through parameter
identification (PI) procedures applied on both measurements and computational -aeroelastic- results.
“Fatigue influence” matrices are, thus, established for different machine sizes and concepts.
Recommendations are made in the last part of the report on how one could possibly include the present
findings into the IEC Standards.


2. NOMENCLATURE
ϕ Wind direction (degrees)
x,y,z Cartesian, streamlined, coordinates (m)
z0 Roughness length (m)
U,V,W Cartesian, streamlined, mean wind speed vector components (m/s)
u Normalized longitudinal wind component : u(z) = U(z)/Uref
a
aws Vertical wind shear exponent estimated from the power law model, U(z)/U(H) = (z/H) ws
σu ,σv ,σw Standard deviation of the u, v and w components of the wind speed (m/s)
σϕ Standard deviation of (horizontal) mean wind direction (degrees)
I or T.I Turbulence intensity σu / U
u’w’,u’v’,v’w’ Reynolds stresses (non-diagonal entries to the Reynolds correlation tensor)
Lu ,Lv ,Lw Length scales of the u,v and w wind velocity components estimated with the use of the
Von Karman formulation (m)
Auu,Avv Aww Coherence decay factors of the wind components at different heights estimated with the
use of the Davenport model, x(D,n) = e-AnD/U , where x expresses the square root of the
coherence, D the separation distance between the points (m), n the frequency in Hz and U
the mean wind speed in m/s.
K Kinetic energy of turbulence defined as ½( σu2 + σv2 + σw2) (m2/sec2)
k Normalized kinetic energy of turbulence k(z) = K(z)/ Uref 2
Lsh Linear shear coefficient U(ztip)/U(zhub), z is the altitude
Yaw Yaw error in degrees. Positive when the rotation vector turns clock-wise seen from
the ground
S Speed Up, S= u (z) / uin (z) = U (z) / Uin (z)
T Turbulence increase factor, T= k (z) / kin (z) = K (z) / Kin (z)
3
1 N
 ui − U 
Uskew Skewness of the wind speed distribution estimated as
N
∑ 
i =1  σ u


4
1 N
 ui − U 
Ukurt Kurtosis of the wind speed distribution estimated as
N
∑ 
i =1  σ u
 −3

ref subscript denoting far-field conditions (at the edge of the atmospheric layer)
in subscript denoting properties at the inlet plane (resembling those of the flat-terrain)
This document is intended to produce the background information to help the introduction of complex
terrain effects into the existing or future standards.


3. INTRODUCTION
Wind turbine (WT) classification is defined in the IEC context [1] in terms of wind speeds and turbulence
intensities. The intention of these classifications is to cover most applications enabling the selection of
representative values of the above two parameters so as to cover as many different sites as possible. The
goal is to achieve WT classification with clearly varying robustness governed by wind speed and
turbulence intensity. Four WT classes, I to IV, are thus introduced. To cover cases where special design
conditions are needed, a further class “S” is foreseen where site-specific parameters are defined by the
designer. Relying on the class-defining parameter values, a series of design load cases is, then, introduced
for verifying the structural integrity of the load-carrying components of a machine within an appropriate
safety level. Although not explicitly stated in the IEC standard, the semi-empirical frame suggested for the
verification process is heavily relying on “flat terrain” experience / evidence. This is especially true for the
turbulence models proposed and, consequently, for the evaluation of the fatigue loading of the WT’s. Any
attempt, therefore, to include complex terrain operation within the current standardisation practice should
first answer the following question: To what extent do the “complex terrain” wind conditions differ from
the “flat terrain” ones and how these differences affect machine loading?
Recent work on complex terrain wind field [2],[3] has shown that both the deterministic and the
stochastic wind properties do deviate significantly from the so called “flat terrain” conditions. More
specifically:
• The vertical wind shear in complex terrain does not respect the log-law. In addition, the mean wind
field strain tensor (including spatial derivatives of all three wind components) may be full and its
entries may strongly vary with height.
• The wind vector can be strongly inclined, following the local terrain slope.
• The Reynolds stress tensor can be full and its entries may also vary with height. The kinetic energy of
turbulence is split into its three components (longitudinal, transversal, vertical) at different portions
than in flat terrain. More energy is now contained in the transversal and the lateral components.
• Turbulent length scales and spatial coherence are different in complex terrain. Smaller length scales
and higher coherence decay factors are encountered at complex terrain turbulence measurements.
• Non-Gaussian behaviour of turbulence is often met in complex terrain.
These deviations become stronger as the terrain complexity (a non-quantifiable term, in general) increases.
In the absence of a concise “complex terrain theory” and considering that the number of possible complex
terrain layouts is infinite, the quantification of “complex terrain effects” can only be obtained indirectly.
To do that the following two-step procedure is proposed and applied below:
1. Produce a complex terrain experimental database including as many sites as possible, to estimate the
envelope of variation of the individual wind field parameters. Try to interpret these variations by
introducing a suitable theory. Apply this theory in simple cases where the terrain complexity (single
hill configurations, for instance) is well defined to identify trends and possible correlation between the
parameters sought.
2. Interrelate wind-field properties and machine loading applying sensitivity analysis and parameter
identification techniques for a variety of machine sizes and concepts. Define a procedure for translating
specific site conditions to fatigue loading.
Most of the experimental work regarding step 1 has been accomplished in the framework of the concluded
MOUNTURB [2] and the ongoing COMTERID [3] projects. Representative data are presented and
discussed in chapter 2.

The theoretical work addressed in step 1 has been performed in the context of this project. This work
includes Navier-Stokes computations of the mean flow field and turbulence around single- and double-hill
configurations of different hill-slope(s) and separations (a measure of the terrain complexity for these
model-problems). Results are presented in terms of velocity speed-up and turbulence increase factors. The
set-up and the conclusions are discussed in chapter 2, while tabulated results are shown in Appendix A
The theoretical work is also supported by Appendix B, where a simplified wind shear law for hill-tops is
presented, and Appendix C, where a discussion on the “complex terrain” Reynolds stress tensor is held.
Chapter 3 is devoted to the sensitivity analysis procedures mentioned above. Both load measurements and
calculations are used for this purpose. Results obtained through a multi-regression analysis of the
experimental data are presented in Appendix D. Fatigue influence matrices obtained for several machines
through aeroelastic computations are tabulated in Appendix E.
Conclusions and recommendations are presented in chapters 4 and 5 respectively.

4. MEAN WIND FIELD AND TURBULENCE STRUCTURE

IN COMPLEX TERRAIN
4.1 Approach followed
The investigation of mean wind field and turbulence structure is based on the analysis of the experimental
“wind data bases” collated for a number of complex terrain sites (in terms of 10 minute blocks) and, also,
on theoretical considerations. A hill catalogue, including two dimensional single and tandem hill
configurations has been created through parametric mean flow field calculations. This catalogue can serve
as a guide for providing a rough estimation of the wind shear and the turbulent intensity at a specific site,
through an idealisation of its upwind surface section. The experimental approach is focusing on the
identification of differences in the basic inflow parameters between flat and complex terrain.
4.2 Measured wind field characteristics

Experimental wind data bases have been analysed for the following complex terrain sites:
• Agia Marina , CRES test station at Lavrio (GR)

• Andros island, (GR)
• Marmari, (GR)
• Toplou, Crete island (GR)
• Sky River, RISØ measurements, (USA)
The findings regarding the mean wind field and turbulence properties are summarised in the next
paragraphs.
4.2.1 Mean wind field
Typical wind profile shapes from Lavrio, Andros and Toplou, referring to a wide range of wind velocities,
are shown in figure 4.1. The power law (1/7 exponent, recommended for flat terrain) is also plotted on the
figure. It is seen that the actual wind shear is over-predicted by the fit in all cases. In reality, flat and even
inverse profiles are observed.
Up to 20 degrees distortion of the horizontal mean wind component measured at 20 and 40 meters above
ground level have been observed in Andros and Lavrio.
Up to 12 degrees flow tilt angle (wind vector inclination) have been observed at Marmari, at 40 meters
a.g.l. In general, the flow tilt angle follows the local terrain slope, decreasing with increasing height.

45
Wind Profiles at Andros, Lavrio & Toplou (2m/s Bin)
1/7 Power Law, h=30m
40
Andros
Height a.g.l., m
35
Lavrio
Toplou
30
25
20
15
6 8 10 12 14 16 18
Wind Speed, m/s
Figure 4.1 Vertical wind profiles of complex terrain sites
The horizontal spatial variation of mean wind field properties over distances of the order of a rotor
diameter (20 m), measured at Lavrio were shown to be rather insignificant. This conclusion can be
generalised provided that no significant terrain variations occur within the scale of interest.
4.2.2 Turbulence structure
0.25
Turb. Intens. at Andros, Lavrio & Toplou (2 deg Bin)
0.20 Lavrio
Andros
0.15
max
0.10
0.05 min Toplou
Wind Speed Bin : 12m/s-14m/s

0.00
-60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80
Incoming Wind Direction, deg
Figure 4.2 Turbulence intensity (TI) vs wind direction at the three sites

Turbulence intensity in complex terrain shows strong evidence of wind direction dependency.
Representative plots are shown in figure 4.2, including mean and min_, max_ values. This is also true for
all turbulence structure properties.
Mean, representative, values of the data bases over a wide sector centred at the prevailing wind direction
for each site, are tabulated in table 1.
Table 1 Mean, representative, values of the data bases
SITE ID Lavrio Andros Marmari Toplou Sky River

Height a.g.l (m) 22 31 40 40 43
Mean Wind Speed (m/s) 7.4 8.7 9.3 7.2 7.2
Weibull C(m/s),k 8.3 1.53 9.8 1.85 10.0 1.32 8.1 1.64
TI=σσu/U (%) 13 9.5 10 10 8
σv/σ
σu 0.85 0.90 0.97 1.0
σu
σw/σ 0.72 0.70 0.68 0.8
σϕ/TI 47.3 56.0 38.5
Lu(m) 20-120 20-120 20-200 20-120 10-250
Lv(m) 10-50 10-50 5-150 6-150
Lw(m) 5-20 5-20 4-40 6-54
2-7 3-12 2-20 2-15
Noticeable are the relatively low values of the Weibull shape parameter k which are definitely below the
Rayleigh equivalent value 2.0 and most of the times even below 1.8. This finding is in accordance with
the experience gained from the Wind Atlas [4] where the wind regime of a number of sites has been
analysed. Low values of k are in favour of extreme wind events.
The turbulence intensities at all sites appear to be smaller compared to these used by the IEC1400-1
standard, namely 16 and 18%. The kinetic energy of turbulence is split to its three components at ratios
different than the 1:0.8:0.5 characterising flat terrain. More turbulence energy is now distributed to the
transversal and lateral wind components.
The rather crude procedure followed for estimating the length scales and coherence decay factors, as
explained below, results in considerably scattered values. Nevertheless, a trend is evident for rather lower
length scale values than the equivalent flat terrain values (71, 18, 4.5) presented in table 4 in the sequel.
Non Gaussian behaviour is indicated by the non-zero values of the higher order statistics (skewness and
kurtosis coefficients). Slightly negative mean skewness values of the order (-0.3) have been observed at
certain sites. However, the obtained values are neither that extreme, nor that persistent, to support a strong
statement on non-Gaussian turbulence at complex terrain.
4.3 Theoretical investigation

The theoretical investigation is limited to flow calculations around two-dimensional hills of Gaussian
shape. Single and double hill configurations are analysed using as free parameters the hill height over
length (h/L), a maximum slope indicator, the separation (Dht) between the hills and the roughness length
(z0). It is the intention of this work to interrelate wind shear and turbulence increase with terrain
complexity, measured in this case by the two geometric parameters h/L and Dht. The obtained results,

which are presented in Appendix A along with the nomenclature adopted, include velocity and kinetic
energy of turbulence distributions along the 10 m and 40 m heights above ground level, as well as the
corresponding hill-top profiles.
4.3.1 Parametric set-up

All computations have been performed for the “hill half length” value L = 250 m. Disregarding Reynolds
number effects the results can apply for different L values as soon as (h/L) and (z0/L) are maintained.
Tables I-1 to I-7 present the single hill results. Parametric computations have been performed for four hill
heights (h = 50, 100, 150 and 200 m) and three roughness length values, called (S)mall, (M)edium and
(L)arge corresponding to z0 = 0.003 m, 0.03 m and 0.3 m respectively. A case defined as {50 S}, for
instance, corresponds to h=50 m and z0=(S)=0.003 m. Table I-1 presents the considered single hill
geometries. Table I-2 shows speed-up factors at the, so called, anemometer height, e.g. 10 meters above
ground level. Speed-up factors at 40 m a.g.l. (so called hub-height) are shown in table I-3. Tables I-4 and
I-5 present the evolution of the turbulence increase factor along the axial distance (x) at the two
characteristic heights. Table I-6 presents normalised velocity profiles at the hill-top versus the height a.g.l.
(linear scale). The inlet profile is also plotted. Table I-7 shows profiles of the kinetic energy of turbulence
versus the height a.g.l. (logarithmic scale) at the inlet and the hill-top planes. Similar results are presented
for the tandem-hill cases.
Tables I-8 to I-14 present results for tandem hills of equal height. The standard roughness (z0=0.03 m) has
been considered in all cases. Four different h/L ratios have been considered (0.2, 0.4, 0.6 and 0.8) along
with five hill-tops separations (D=4L, 6L, 8L, 10L, and 12L). Each case is marked with the corresponding
D(ht) and h/L values (4L h/L=0.2 for the up-left case, for instance). The wind shear profiles at the two
hill-tops (u2 for the first hill and u4 for the second) are compared against the corresponding single-hill
result (u single) in table I-13. Turbulence profiles at the hill-top sections, called k2 (first hill) and k4
(second hill), are compared against the corresponding inlet-plane profile (k inlet).
Tables I-15 to I-21 present results of tandem-hills of different height. Four combinations of hill heights
have been tested, (50 m - 100 m), (100 m - 50 m), (100 m -200 m) and (200 m - 100 m), the first height
corresponding to the first hill. Computations are performed for five separations per combination of
heights. The standard roughness (z0=0.03 m) has been considered in all cases.
4.3.2 Flow model

The flow model used for the parametric hill calculations is based on the Reynolds averaged
incompressible Navier-Stokes equations. The k-omega model [5], supported with wall-function type
boundary conditions (which take the roughness length into account), is used for turbulence closure. The
constants of the model are suitably calibrated for atmospheric flows. The numerical integration of the
governing equations is performed through a pressure-correction algorithm. Results are obtained on body-
fitted structural grids.
It must be emphasised that the employed turbulence model is an isotropic one. In that respect the flow
solver can only account for the global kinetic energy of turbulence but no information on how this energy
is split into its three components σu, σv and σw is provided. The missing information is crucial since it is
the longitudinal components of turbulence σu that mainly affects the fatigue loading of wind turbines
while the two other have a rather secondary effect. To fill the gap some additional theoretical work on the
Reynolds stresses is presented in Appendix C.

4.3.3 Obtained results

Analysing the obtained results the following qualitative conclusion can be drawn, regarding the influence
of the hill height and the surface roughness on the flow properties over a hill (always referring to hill-top
conditions):
• For a single hill, the speed up (S) and turbulence factor (T) increase as the h/L or the height h for a
constant L, increases.
• For a single hill, the speed up increases with the surface roughness z0, while the (T) factor decreases as
z0 increases.
• The wind shear profiles become fuller and more energetic as h/L increases and z0 decreases.
• The turbulence energy K increases with both h and z0.
The following conclusions can be made regarding the influence of a second hill on the turbulence and
flow fields:
• The presence of an equal second hill reduces the speed up on both hills
• The presence of a second hill reduces the turbulence increase ratio on the first hill, but increases it for
the second hill.
• The wind shear profiles become fuller and more energetic as the height h and distance between hills
Dht increase.
• The turbulence energy profiles grow as the hill height h increases, but decreases as the separation
between hills increases.
• When the height of the first hill is half the height of the second hill, the speed up at the top of the first
hill depends on Dht, while on the contrary, the speed up at the top of the second hill has a small
dependence on Dht.
• When the second hill is smaller, the dependence of the speed up results on Dht is reduced for the first
hill, but is strongly increased for the second hill.
• When the first hill is smaller, the dependence of (T) in the distance between hills is small for the first
hill. Nevertheless, it increases for the second hill.
• When the first hill is bigger than the second, (T) becomes independent of Dht for the first hill but not
for the second.
• When the hills are different, turbulence is always bigger on the second hill, except for the instances
where separation takes place.
• As the distance between hills Dht increases, the results converge to the single hill solution.
Regarding the turbulence variation with height it can always be seen from the corresponding hill-top
profiles that there exist the, so called, inner layer where turbulence is getting very high, depending on the
h/L ratio. Fortunately for the larger wind turbines, the depth of this inner layer is rather short (less than 10
meters in most of the cases examined here with L=250m). In the outer layer, turbulence is rather relaxed
compared to its surface value.
Let us now quantify the results in terms of the turbulent intensity. By definition of the kinetic energy of
turbulence (K) the following formula holds:
1   σ v   σ w  
2 2
K = σu 1 +   +    (1)
2  σu   σu  

That is, to translate K into σu the ratios (σv /σu)and (σw /σu) must be known. Consider, now, two sets of
such ratios, a flat terrain one with σu: σv: σw=1:0.8:0.5 and a complex terrain one with the, rather extreme,
values σu: σv: σw=1:1:0.8. Then, by substitution

σ u = K / 0.97 flat
(2)
σ u = K / 115
. complex
indicating that less turbulent energy goes to the longitudinal direction in the complex terrain case. To
avoid the dependence on the actual values of the turbulent ratios we now on adopt the unique,
conservative, relation σu=(K)1/2 .
Let us now demonstrate the use of the results of Appendix A, which are presented in non-dimensional
form.
For hill-tops where vertical profiles of k and u are given,
TI(z) = k(z)1/2 / u(z) (3)
For other than hill-top locations (here z = 10 or 40 m)
σu (z) / σu,in (z) = (T)1/2 (4)
In a similar way one may get the following formula for the turbulent intensity (TI):
TI (z) / TIin (z) = (T)1/2 / S (5)
where subscript (in) refers to the undisturbed far-stream conditions. Note that TIin (z) can be evaluated
from the single-hill tables I-6 and I-7 applying equation (3).
Some examples of application follow below.
Example 1. Evaluate the turbulent intensity at 40 m above the top of a single hill with h/L=0.4 and
z0=0.03m.
From tables I-6 and I-7 we focus on the {100 M} plots since 100 = 0.4 X L (250 m) and z0=0.03m (M
case). Then, at 40 meters a.g.l u=1.1, uin=0.75, k=0.012 and kin=0.0095. Then, applying formula (3):
TI(40m) = k(40m) 1/2 / u(40m) = (0.012) 1/2 / (1.1) = 10 %
Applying the formula for the “inflow” conditions (flat terrain equivalent)
TI(40m) in= (0.0095) 1/2 / (0.75) = 13 %
In other words the turbulence intensity at the hill-top is decreased not because turbulence is less (k=0.012
in the first case while k=0.0095 in the second) but because of the speed-up effect. It is reminded that it is
σu (or k in this case) that drives the fatigue loading and not the turbulence intensity itself.
Example 2. Using the setting of the previous case evaluate the turbulence intensity at x = -100m (100 m
up-stream the hill-top).
From tables I-3 and I-5 (40 m a.g.l) we get for x=-100m the values: S = 1.35 and T = 1.2. Then, applying
formula (5):
TI (40m) = / TIin (40m) (T)1/2 / S = (13%) (1.2) 1/2 / (1.35) = 10.5 %.
Example 3. Consider two hills of equal height h=100m (and L=250m). Let z0=0.03m. Evaluate the
turbulent intensity at 40m above the two hill tops for Dht = 6L and 10L.

From tables I-13 and I-14 and for the {6L h/L=0.4} case we get:
1st hill : u = 1.06 , k= 0.012 and, from relation (3) TI = 10 %
2nd hill : u = 1.04 , k= 0.016 and, from relation (3) TI = 12 %
From the same tables and for the {10L h/L=0.4} case we get:
1st hill : u = 1.08 , k= 0.011 and, from relation (3) TI = 10 %
2nd hill : u = 1.08 , k= 0.014 and, from relation (3) TI = 11 %
Comparing with the equivalent single hill case of example 1 (TI=10%) it is seen that the TI at the first hill
top is rather unaltered while TI at the second hill top increases as the distance between the hills decreases.
4.3.4 Complex terrain and wind shear

From the examined cases it appears that the complex terrain vertical shear, at least at hill-top conditions,
acts favourable on machine loading compared to the flat terrain wind shear. In general, the velocity slopes
(positive or negative when h/L is large) are smaller in absolute values than the log-law equivalent slope
du/dz = uT/κz , κ being the Von Karman constant and uT the friction velocity. In that respect, the use of the
log-law for all, flat and complex terrain, cases within the standards is a conservative approach for the latter
case.
If, however, a generalised wind shear law should be included in the standards, the analytical formulation
presented in Appendix B is a possible candidate. The velocity variation with height is expressed in terms
of the hill height h, its half-length L and the roughness length z0.
4.3.5 Complex terrain and turbulent structure

A theoretical investigation of the complex terrain turbulent structure, in terms of Reynolds stresses, is
attempted in Appendix C. For the considered 2-D and quasi-3D cases it has been shown that
2 2
 σu   σw 
  +  ≤ 2 (6)
 σv   σv 
where the equality holds for all 2-D flows. Deviations from the equality denote three-dimensional effects.
In the accelerating part of a 2-D flow the kinetic energy of turbulence increases (see Appendix A), while
turbulent energy is transferred from the longitudinal u_ to the vertical w_ wind component. This implies
that the turbulence in that area becomes more isotropic. In general, in the accelerating part of any complex
terrain flow it is expected that σv and σw take higher values than the “flat terrain” equivalent while no such
clear statement can be made for σu.
For the spectral properties of complex terrain turbulence (length scales, coherence) there is no concise
theory available, to the authors knowledge. By intuition it is expected that the terrain effect should lead to
some redistribution of the turbulent energy on the frequency domain, especially at its low frequency part.
For a given topography the power spectra of turbulence should enjoy some self-preserving form which
cannot be properly fitted either by the Kaimal or by the Von-Karman single-parametric representations. If
one of the two representations is used for length-scale evaluation, the result will be heavily dependent on
the selected area where the best-fit procedure applies. This is also true for the spatial coherence
description in terms of the decay factor.
4.4 Synthesis of the results

Referring to complex terrain sites we hereafter imply locations, within a greater area with intense surface
relief exhibiting a wind potential favourable for wind turbine installation e.g. to sites where the flow is

locally over-speeding, like hill-tops. Unattractive sites (where separated flow occurs, for instance) are
excluded from this discussion.
Mean flow field properties

• Low values of the Weibull shape factor are often met at complex terrain sites.
• Mean wind field and turbulence structure are clearly affected by the upwind terrain features.
• The wind profile is clearly affected by the upwind fetch inducing local speed-up regions which vary
with wind direction. Consequently, the logarithmic and power law are proved inadequate to describe
the vertical wind shear. In most of the cases complex terrain wind shear seems to be less intense than
its flat terrain equivalent.
• Depending on the actual site topography the mean flow field may be significantly distorted in the
transversal direction, as well. The effect is stronger at the side-boarders (in respect to the wind
direction) of local hills. Vertical deviations of wind direction of as much as 20 degrees have been
recorded in some measurements.
• The flow inclination follows the local terrain slope and acts on the machine as a systematic vertical
“yaw error”.
Turbulence structure
• Theoretical analysis and measurements indicate that the turbulence kinetic energy levels are higher in
complex terrain and more importantly that the distribution mechanism of turbulent kinetic energy to its
components is different from this observed in flat terrain cases. Numerical experimentation shows that
the kinetic turbulence energy increase is more pronounced in the, so called, inner layer of the
atmospheric boundary layer which, in most of the cases, is confined within the 10 meters above ground
level. In the outer layer, where the large wind turbines operate, the turbulence increase is less
significant. On the other hand, turbulence intensity seems to take lower values at complex than in flat
terrain. This is due to the over-speeding of the mean wind vector.
• Differences from flat to complex terrain are also depicted in the stochastic wind characteristics as
demonstrated by the values of the turbulence ratios σv /σu , σw /σu, σu2/K, σv2/K and σw2/K and
Reynolds stresses ratios.
• The non-zero values of the non-diagonal entries of the Reynolds stress (u’v’ and v’w’ correlation’s)
tensor found in a number of cases gives evidence of the 3D character of the wind flow in complex
terrain. Finally, a high correlation between the longitudinal and the vertical velocity component is
noticed at complex terrain.
• The analysis of the data in the frequency domain revealed that in complex terrain a spectral gap is
observed in the lower frequency part of the spectrum which can not be adequately modelled by the
formulas existing in the literature. This is due to the fact that high frequencies corresponding to
wavelengths smaller than the fetch over the terrain changes respond rapidly to changes imposed on the
flow by streamline distortion and surface geometry whereas the low frequency of the spectra of the
horizontal components are affected by the upstream terrain characteristics. Thus, the high frequency
part is always in a state of quasi-equilibrium and the spectral shape conforms to this over uniform
terrain while low frequency longitudinal spectral densities are reduced introducing the so called
“spectral gap or lag” representing the different response times of inner and outer layer (small and large
scales respectively).
• The significant scatter, associated with the uncertainty of the fitting process, observed in the length
scale values in all the investigated sites did not allow for a well established comparison of the length
scales between flat and complex terrain sites. Nevertheless, the results of this comparison indicate that
the length scales characterising the turbulence kinetic energy distribution in its frequencies, is smaller
in complex terrain (more energy in the high frequency part of the spectrum).

• The measured coherence exhibits the typical exponential decay with increasing turbulence frequency.
The decay rates obtained through the application of the Davenport exponential decay model exhibit
significant scatter, increasing with increasing longitudinal wind speed.
• Some deviations are observed in the distribution of wind speed with respect to the Gaussian in complex
terrain sites. Noticeable is a trend for slightly negative values of skewness found in most of the
analysed ten minutes time series, indicating an asymmetry in distribution, namely a shift (tail) towards
the lower wind speeds. The minimum wind speed occurring in the time series is lower than the
expected value from a Gaussian distribution (U-3σu). From the analysis performed at four complex
terrain sites it is difficult to conclude whether the “tail” in the wind speed distribution is connected to
the terrain, however this is the most probable assumption since this trend is observed in almost all the
measurements obtained at these sites.


5. MACHINE LOADING IN COMPLEX TERRAIN

5.1 Approach followed
Fatigue loading of wind turbines is investigated through the analysis of existing “load” data bases and
calculations. In both cases the emphasis is put in identifying the wind structure parameters that mainly
affect fatigue loading and, then, in quantifying their influence by calculating the corresponding “partial
derivatives” of the fatigue equivalent loads. In this way, systematic deviations of the wind structure
parameters from one type of terrain to another can be directly translated to load deviations.
It must be emphasised that the approach followed is confined to the 10 minute time scale and does not
“integrate” the findings to conclude on “machine life times”. The reason is that the integration procedure
should rely on some mean-wind-speed probability distribution (of Weibull type, for instance) which, in
principle, is terrain-type independent. Nevertheless, one should keep in mind that complex terrain
measurements have shown that the shape factor of the Weibull distribution is taking significantly lower
values (more energy at high wind speeds) than in flat terrain.
5.2 Modified wind field and machine loading; a measurement-based

investigation
The procedure used in the measurement-based investigation comprises the following phases:
I. Identification of causative processes
The deterministic wind characteristics, wind turbulence characteristics as well as the wind speed
distribution characteristics are regarded as the main causative processes when wind turbine fatigue loads
are considered.
II. Determination of candidate independent variables for each process
Upon definition of the causative processes, a set of independent variables are selected in order to describe
each process. The term independent is used according to the regression analysis nomenclature and does
not imply that the predictive variables should be uncorrelated. The deterministic wind characteristics are
described by the mean wind speed and wind shear. The wind turbulence is described by the standard
deviation of the wind speed components and the turbulence length scale components. The wind speed
distribution is described by the higher order descriptive statistics, namely skewness and kurtosis. In cases,
where sonic measurements are available and the effect of the complex terrain characteristics is to be
captured, the ratios of the standard deviations of the lateral and the vertical wind components against the
longitudinal one are used. When cup and vane measurements are available the ratio of the standard
deviation of wind direction against the turbulence intensity may be used instead.
III. Application of the full model
The regression analysis is performed for the selected full model.
IV. Iterative application of evaluation criteria and model reforming

The evaluation criterion is based on the p-values of the t-ratios of each independent variable. Large values
for the p-value indicate that the rejection of the hypothesis "H0: the regression coefficient a k 1 is zero" is
questionable, and consequently the parameter is rejected from the model and the regression analysis is
performed again. According to statistics literature and numerical experimentation the rejection threshold
values were set to be 5%.
V. Final selection and assessment
After the completion of the iterative process, the assessment of the regression findings is attained through
the selected magnitudes, i.e. regression coefficients, dependence coefficients or t-statistics.
The above procedure has been applied for three machines operating at complex terrain sites. The
theoretical background of the adopted methodology and the obtained results in tabulated form are
presented in Appendix D.
5.2.1 Machines and sites considered

The above procedure has been applied to the experimental “load” data bases of three machines operating
at complex terrain sites:
• The WXT110 (110 kW, three-bladed, stall ) operating at the CRES test station at Lavrio (GR)
• The V27 (225 kW, three-bladed, pitch) operating at Andros island (GR)
• The NTK500 (500 kW, three-bladed, stall ) operating at Toplou (GR)
5.2.2 Investigated parameters

The parameters considered for the fatigue analysis of the three machines are listed in the following table.
In each case the analysis is performed for three individual velocity bins, representing linear, stall and post-
stall operation. Mean values and standard deviations of the parameters at each velocity bin are, also,
provided. As main fatigue driving parameters, the Reynolds stresses (normalised where appropriate) and
the longitudinal length scale (Von Karman fit) have been selected. In one of the cases the wind shear
exponent αws is included in the parameters list. When cup-anemometer wind measurements are only
available, the normalised ratio of the wind direction standard deviation and the turbulence intensity
(σϕ/TI) is used instead of σv/σu to account for the transversal variations of the wind vector. The influence
of higher order statistics of turbulence, measured by the skewness (Uskew) and the kurtosis (Ukurt)
coefficients of the recorded 10 min time series, is also taken into account.

Table 2.a Parameter set for WINCON W110XT data base
U=5-8m/s U=8-11m/s U=13-17m/s

mean sdv mean sdv mean sdv
U 7.2 0.63 9.4 0.84 14.8 1.15
αws 0.028 0.034 0.024 0.028 0.028 0.024
σu 0.98 0.29 1.36 0.33 2.14 0.45
σv/σu 0.94 0.20 0.87 0.16 0.81 0.13
σw/σu 0.77 0.15 0.72 0.12 0.67 0.09
Lu 67.0 29.4 85.5 34.7 106.1 39.4
Table 2.b Parameter set for VESTAS V27 data base
U=5-7m/s U=8-11m/s U=13-16m/s

U 6.52 0.33 9.7 0.83 14.1 0.76
αws 0.088 0.044 0.091 0.046 0.082 0.034
σu 0.60 0.18 0.95 0.18 1.35 0.27
σϕ/TI 1.07 0.23 1.11 0.22 1.01 0.26
Lu 70,9 37,9 60.7 28,6 57.7 24.1
Uskew -0.017 0.36 -0.048 0.27 -0.126 0.29
Ukurt -0.16 0.71 -0.14 0.40 -0.20 0.37
Table 2.c Parameter set for NORDTANK NTK500/37 data base
U=5-8m/s U=8-11m/s U=13-16m/s

U 7.0 0.73 9.5 0.87 14.5 0.85
αws -0.48e-3 0.055 0.091 0.51 0.11 0.42
σu 0.67 0.24 0.85 0.29 1.45 0.35
σϕ/TI 0.94 0.25 1.03 0.22 0.91 0.12
Lu 28.3 13.8 32.5 15.6 46.3 17.2
Uskew 0.133 0.35 -0.18 0.35 -0.29 0.30
Ukurt 0.617 1.73 0.14 0.72 -0.014 0.61
3.2.3 Results and conclusions
Results are presented for the three machines in tables (D 1 to 3) of Appendix D. The column (Leq), where
bold characters are used, provides the mean value of the corresponding equivalent load in kNm. The next
columns provide the estimated values of the partial derivatives ∂Lj / ∂xi . A measure of the significance of
each parameter is the product of the corresponding partial derivative and the standard deviation of the
parameter itself. This product indicates the increase of the equivalent load when the considered parameter
increases one standard deviation with respect to its mean value. To demonstrate this, consider the
following example:
Compare the significance of σu and Lu on the flap-wise bending load of a NTK500 blade in
the wind-speed bin 8-11 m/s.

From table 2 we get the values SDV_ σu = 0.29 m/s, SDV_ Lu=15.6 m. From table D.3 we get the
values ∂L / ∂σu = 45.3 (kNm/m/s), ∂ L / ∂ Lu = -0.1 (kNm/m). The mean value of L is 46.88 kNm.
For 1-sdv increase of σu we get an increase in L
∆L = (45.3) * (0.29) = 13.14 kNm, or (13.14)/(46.88)=28% increase of the load
while for 1-sdv increase of Lu we get
∆L =(-0.1)*(15.6) = -1.56 kNm, or (1.56)/(46.88)=3.3% decrease of the load
In that respect Lu is almost 10 times less significant than σu.
Estimate the effect of a combined increase of σv/σ

σu and σw/σ
σu by 0.1 and 0.2 respectively on
the tower bottom fatigue load in the wind speed bin of 8-11m/s for the WINCON 110XT
turbine.
From table D.3 we get the values ∂ L / ∂ σv/σu = 17,31 (kNm/m/s), ∂ L / ∂ σw/σu = 42,62 (kNm/m).
The mean value of L is 86,68 kNm.
For the selected parameter changes the resulting increase in L is:
∆L = (17,31) * (0.1) + (42,62)*(0,2) = 10,26 kNm, or (10,26)/(86,68)=12% increase of the load
From the tables of Appendix D it is seen that the partial derivatives related to the Reynolds stresses are
always positive. When either σu σv σw or σϕ/TI increases the fatigue load is also increasing. The
longitudinal length scale has the opposite effect. The effect of σu is primary compared to the rest. The
effect of σv σw and σϕ/TI becomes stronger at stalled operation. Skewness and kurtosis have also
secondary effects.
5.3 Modified wind field and machine loading; a modelling-based

investigation
In the framework of the MOUNTURB project, aeroelastic modellers were called upon to investigate the
influence of some identified, terrain dependent, wind inflow parameters on the fatigue loading of the
machines involved (WXT110, V27, MADE 300, NTK500). To do that, a sequence of representative test
cases has been proposed, putting the emphasis on the wind inflow parameters that are hard to be
investigated through the experiment. The expected result is a ‘fatigue influence matrix’ which is
composed by the ‘partial derivatives’ of the fatigue loading in terms of the selected wind inflow
parameters. For each machine the ‘fatigue influence matrix’ is formed for three different operating points
(reference states) , which are distinguished in terms of the mean wind speed at hub height. These reference
states are selected in the linear, stall and post-stall zone, respectively. The input parameters which were
identified as critical are: the mean wind speed at hub height, the longitudinal turbulence intensity, the yaw
error, a wind shear parameter, the vertical mean wind component and the 3D turbulent characteristics of
the incoming flow. To limit the requested computational effort in reasonable bounds the turbulent
characteristics were grouped together to form three characteristic sets, one corresponding to flat terrain
operation and two to complex terrain operation. The flat terrain set was identical to that proposed in the
IEC International Standards draft report. The first complex terrain set relied on the mean values of
turbulence measured from the experimental campaign at Andros island. The second complex terrain

turbulence set introduces some low probability ‘extremities’ on the length scales and coherence factor, in
order to investigate their effect on fatigue. Fatigue damage was evaluated in terms of Rainflow Count
(RFC) plots and 1 Hz equivalent loads of the blade root, tower bottom and low-speed shaft moments.
Within the present project the MOUNTURB data sets were applied to three additional (DANWIN 180,
NASSUDEN II and MAGLARP/WTS3) machines by T.G. Evidently, the MOUNTURB approach does
not provide detailed information on the fatigue influence of the individual parameters (Reynolds stress
tensor components, length scales, coherence decay factors) since all of them are blended within the three
turbulent data sets. For this reason we shall call the MOUNTURB approach Coarse Grain Parameter
Identification (PI).
A detailed investigation of the influence of the individual turbulence structure parameters on fatigue is
attempted within COMTERID project. Coupled-3D turbulence generators are used to investigate the
possible influence of the wind components cross-correlation’s (off-diagonal components of the Reynolds
stress tensor). Once again the analysis is performed for three wind speeds, representing linear, stall and
post stall operation. The COMTERID approach will be called Fine Grain PI. Fatigue influence matrices
are once again produced for selected machines following the MOUNTURB notation - guidelines.
5.3.1 Machines considered

The following machines have been modelled for the coarse grain PI analysis
• WXT110 : Three bladed, rotor diameter D=20m, stall controlled, fixed (rotational)
speed and yaw , 110 kW.
• V27 : Three bladed, D=27m, pitch controlled, fixed speed and yaw, 225 kW.
• MADE AE/30 : Three bladed, D=30m, stall controlled, fixed speed and yaw, 300 kW.
• NTK500 : Three bladed, D=30m, stall controlled, fixed speed and yaw, 500 kW.
• V39 : Three bladed, D=29m, pitch controlled, fixed speed and yaw, 500 kW.
• DANWIN 180 : Three bladed, D=23m, stall controlled, fixed speed and yaw, 180 kW.
• NASUDDEN II : Two bladed, D=80m, pitch controlled, fixed speed yaw and teeter, 3MW.
• MAGLARP/WTS3 : Two bladed, downwind, D=84m, pitch,, teeter hinge, free yaw, 3MW.
The rotational speed and their first characteristic modes / eigen-frequencies are shown in the following
table
Table 3 Rotational speed and first characteristic modes / eigen-frequencies
WXT100 V-27 Made AE/30 NTK 500 V-39

Description of mode meas.(Hz) meas.(Hz) simu.(Hz) meas.(Hz) meas.(Hz)
Rotational Speed 0.73 0.72 0.60 0.50 0.50
1st Tower bending 1.47 1.05 1.06 0.94 0.89
1st Shaft torsional 2.13 1.70 2.10 1.16 1.19
1st asym. Rotor / yaw 2.41 2.15 2.29 1.70 1.26
1st asym. Rotor / tilt 2.56 2.20 - 1.80 1.55
1st sym. Rotor / flap 2.94 2.60 2.40 2.08 1.78
1st asym. Rotor / edge 4.78 3.50 4.40 3.50 2.83
DAN/180 NAS/ II MAG/WTS3
Description of mode meas.(Hz) meas.(Hz) meas.(Hz)
Rotational Speed 0.71 0.35 0.42

1st Tower bending 1.10 0.42 0.30

st
1 blade flap 2.40 1.66 1.00
1st blade edge 5.50 2.28 2.10
Presently, the fine grain PI has been only applied for the NTK500 and the V39 machines. Work is ongoing
in the framework of the COMTERID project.
5.3.2 Parametric set-up

Data sets for Coarse Grain PI
To suppress the numerical effort the investigation is done for three turbulence data-sets, only, one typical
for flat terrain and two for complex terrain operation. In all three cases the Von Karman non isotropic
turbulence model is used for generating the turbulent time series. A common value of the coherence
decay factor, called A, has been used for all three wind speed components in all rotor disk directions.
The FLAT terrain turbulence length scales have been derived from the Kaimal length scales suggested in
the IEC International Standards Draft document. The equivalent Von Karman length scales are obtained
from the Kaimal values after a suitable transformation which resulted in the following coefficients after a
best fit procedure:
(Lu )Kaimal= 2.3 (Lu )Von Karman , (Lv )Kaimal= 3.0 (Lv )Von Karman , (Lw)Kaimal= 3.0 (Lw)Von Karman
For complex terrain turbulence simulations two data sets are proposed, a ‘normal’ set called CLX1 and an
‘extreme’ one called CLX2.
Table 4 Data sets for complex terrain turbulence simulations
CASE σv / σu σw / σu Lu Lv Lw A
FLAT 0.8 0.5 71 18 4.5 8.8
CLX1 0.95 0.70 50 30 10 6
CLX2 0.95 0.70 20 15 8 10
All turbulence properties are considered to be uniformly distributed on the rotor disk.
In order to form the ‘fatigue influence matrix’ it is proposed to perform the following 39 runs for each
machine for the data sets presented below. To suppress the number of runs one, only, turbulent time series
realisation per run is suggested with a common random seed initialisation for all turbulence simulations.
Yaw effect is to be investigated for positive yaw error values, only, for the sake of economy. Since this
work looking for trends and not for ‘exact’ solutions the wind shear and the vertical wind component
effect, they are treated in an abstract way supposing linear shear for the longitudinal wind component,
only, and a uniform vertical velocity component along the rotor disk.
The proposed 39 runs are divided into three similar groups representing deviations around three ‘reference
states’ (rows in bold black). The reference states differ in terms of the hub wind velocity.
Table 5 Proposed runs for the formation of the ‘Fatigue influence matrix’
RUN No Uhub σu /Uhub Yaw Lsh W/Uhub Turb. CASE

(m/s) (°°) (%)

L1F-L1C 10 0.08 0 0 0.0 FLAT+CLX1

L2F-L2C-L2C+ >> 0.15 0 0 0.0 FLAT+CLX1+CLX2
L3F-L3C >> 0.15 30 0 0.0 FLAT+CLX1
L4F-L4C >> 0.15 0 15 0.0 FLAT+CLX1
L5F-L5C >> 0.15 0 0 0.1 FLAT+CLX1
L6F-L6C >> 0.25 0 0 0.0 FLAT+CLX1
S1F-S1C 14 0.08 0 0 0.0 FLAT+CLX1
S2F-S2C-S2C+ >> 0.15 0 0 0.0 FLAT+CLX1+CLX2
S3F-S3C >> 0.15 30 0 0.0 FLAT+CLX1
S4F-S4C >> 0.15 0 15 0.0 FLAT+CLX1
S5F-S5C >> 0.15 0 0 0.1 FLAT+CLX1
S6F-S6C >> 0.25 0 0 0.0 FLAT+CLX1
P1F-P1C 18 0.08 0 0 0.0 FLAT+CLX1
P2F-P2C-P2C+ >> 0.15 0 0 0.0 FLAT+CLX1+CLX2
P3F-P3C >> 0.15 30 0 0.0 FLAT+CLX1
P4F-P4C >> 0.15 0 15 0.0 FLAT+CLX1
P5F-P5C >> 0.15 0 0 0.1 FLAT+CLX1
P6F-P6C >> 0.25 0 0 0.0 FLAT+CLX1
Data sets for Fine Grain PI
The following specifications have been set for the fine grain PI which is restricted to the effect of the
turbulence structure on fatigue loading:
1. The influence of the mean wind field is skipped (examined in MOUNTURB).

2. One, only, turbulent intensity value is considered T.I. (σu/U=15%).
3. Three mean longitudinal wind speeds are considered ( U = 10, 14, 18 m/s). No wind shear is assumed.
4. Emphasis is put on the 3-D turbulence structure. The “mean” values (reference state) and the 1 S.D.V.
“worse” values are selected from the Ag. Marina experimental data base.
5. The following 3 (wind speeds) X 6 (states) are considered per machine. All dimensional quantities are
given in S.I. units. Case No 1 is the reference state. The 6th case considers the u-v-w components in a
de-coupled way (as done in MOUNTURB).
Table 6 Specifications for fine grain PI
No σu
σv/σ σw/σ
σu Lu Lv Lw Au Avv Aww u’w’ cor. u’v’ cor.
u
1 0.85 0.66 60 30 15 5 5 5 active active
2 1.00 0.66 60 30 15 5 5 5 active active
3 0.85 0.76 60 30 15 5 5 5 active active
4 0.85 0.66 30 30 15 5 5 5 active active
5 0.85 0.66 60 30 15 7 5 5 active active
6 0.85 0.66 60 30 15 5 5 5 non-active non-active
Analysing wind data from the CRES test-station at Ag. Marina it appeared that the correlation between the
v-w components is negligible compared to the u-w and u-v ones. In a first order approach one may also
neglect the imaginary part of the u-w correlation and the real part of the u-v one. The procedure will be

therefore limited to the following (local) components of the spectral matrix: Suu, Svv, Sww, Re(Suw),
Im(Suv). To calculate the (complex) coherence between the components:
Re(Coh(uw))=Re(Suw)/sqrt(Suu.Sww)
Im(Coh(uv))=Im(Suv)/sqrt(Suu.Svv)
the following expressions are considered, f denoting the frequency:
Re(Coh(uw))=Buw exp(-Auw.f)
Im(Coh(uv))=Buv exp(-Auv.f)
A data fitting process provided the following representative values for the A and B coefficients Auw =
11.85 and Buw =-0.55, Auv = 16.6 and Buv =0.2. These values are used in the present context when the u-w
and u-v correlation’s are active.
Output formats
Cumulative Rainflow Counting tables with Equivalent loads m=4, 8, 12 are presented in all cases for the
following loads:
RFM : blade root flapping moment

REM : blade root edge moment
TBMx : tower bottom bending moment (normal to the mean flow)
TBMy : tower bottom bending moment (parallel to the mean flow)
TTM : tower torsion
DTM : drive train - low speed - torque
5.3.3 Results and conclusions

The full sets of obtained results are presented in table form in Appendix E. Dimensional moments (kNm)
are only presented for the corresponding reference case. These are clearly designated using bold
characters. All the other columns represent the ratio of the calculated moments versus the respective
reference value. Tables D.1 to D.7 summarise the coarse grain PI results. Tables D.8 to D.9 summarise the
fine grain PI ones.
To identify the influence of one single parameter (xi)on a specific equivalent load Lj one has to calculate
the corresponding partial derivative (∂ Lj / ∂ xi ). Taking the linear part, only, of the Taylor expansion:
∂ Lj 
L j ( x1,ref , x2 ,ref ,..., xi ,ref + ∆xi ,...) = L j ( x1,ref , x2 ,ref ,..., xi ,ref ,...) +   ∆xi or
 ∂ x j  ref
∂ Lj 
L j ( xi ) = L j ,ref +   ( xi − xi ,ref ) and (7)
 ∂ x j  ref
∂ Lj  L j ,ref  L j ( xi ) 
  =  − 1
 ∂ x j  ref ( xi − xi ,ref )  L j ,ref 

To estimate the partial derivatives one has to calculate (xi-xi,ref) from table 5 or table 6, depending on the
sought parameter, and get the values Lj,ref and Lj(xi)/Lj,ref from the corresponding table of Appendix E. To
demonstrate the procedure consider the following example:
Evaluate the partial derivative of the flap equivalent load (m=12) in respect to the
turbulence intensity for the NTK500 machine operating at flat terrain conditions at mean
wind speed 10 m/s.
From table 5 we consider the reference T.I. value (0.15 case 2F) and the deviate T.I. (0.08 case
1F). Note that one could equally well use as the deviate the 6F case with T.I.=0.25. Then,
(xi-xi,ref) = (0.08)-(0.15) = -0.07
From Table D.4, (NTK500), m=12, V=10 m/s we get the following values:
Lj,ref = 94.6 kNm (column F2)

Lj(xi)/Lj,ref = 0.57 (column F1)
and the partial derivative is evaluated as:
(∂ Lj / ∂ xi ) = (94.6kNm)/(-0.07)*(0.57-1) = (581 kNm)
This means, for instance, that if T.I. is increased by 3% from 0.15 to 0.18 under the assumed
conditions, the equivalent load will increase
(581)*(0.18-0.15) kNm = 17.43 kNm
corresponding to a 18% increase of the equivalent load
Once the values of the partial derivatives have been evaluated, the effect of any combination of parameters
deviation around the reference state can be estimated through a multi-variate Taylor expansion where the
linear terms are only retained.
It must be emphasised that the selected range of variation of each individual parameter within the PI
process represents some “realistic range of variation” (for instance T.I. from 0.08 to 0.25, Lu from 30 to 60
m, etc) which covers a variety of flat and complex terrain conditions. In that respect, the load ratios
presented in Appendix E are also indicating the weight of each parameter on fatigue loading. Parameters
leading to higher equivalent load ratios are more significant than the others (this is not true for the partial
derivatives themselves).
Based on the last remark a review of the modelling PI results will be attempted herein.
Coarse Grain PI
Starting with the smaller machine, W110XT, it is seen from table D.1 that:
The main parameter influencing the equivalent loads is the turbulence intensity, or better, the standard
deviation of the longitudinal velocity component σu. It is important to note that the increment of the
equivalent loads with σu is of the same order of magnitude for all wind speeds and loads (with the
exception of the blade-edge moment).

The second important parameter is the yaw misalignment, case 3 (30 degrees, a rather large value)
which mainly affects the blade root flapping moment (RFM) equivalent load. The effect is stronger at
higher wind speeds, most probably due to dynamic stall effects.
The third important parameter is related to the turbulence structure characteristics, cases 2C and 2C+.
Concerning the turbulence structure properties it is important to note that:
• Both 2C and 2C+ cases correspond to increased turbulence kinetic energy in the rotor compared to the
2F case, since both the σv / σu and σw / σu ratios are increased. It is expected that part of this energy
will be translated to fatigue load increase.
• The length scales of 2C are decreased compared to 2F but the coherence decay factor is also decreased
leading to a contradictory effect on fatigue which can not be isolated through those limited runs.
• Case 2C+ has decreased length scales and increased decay factor in respect to 2C. Since both settings
result in an increase of the equivalent loads it is expected that the 2C+ loads will be higher than those
corresponding to 2C. The obtained results confirm the statement.
• Most of the employed turbulent wind generation models omit the cross-diagonal terms of the Reynolds
stress tensor providing uncorrelated wind components. It was envisaged that the wind components
coupling would further increase the fatigue loads.
• The additive effect of the turbulence structure properties is obvious in all C-(complex terrain) cases
compared to the F-(flat terrain cases).
• For all investigated turbulence structure characteristics all load exhibit similar relative variations,
except the blade load driven equivalent edge moment, which is less sensitive.
The linear shear, case 4 (15% corresponding to 8.5 degrees offset from vertical) has a quite dominant
effect, especially on the blade-flap and the tower bending loads. The vertical wind component, case 5
(corresponding to 6 degrees slope) effect is not that pronounced. It is noted that the vertical wind
component combined with the machine tilt angle introduces an effect similar to that of the yaw
misalignment which sometimes is not included in the aerodynamic models and, thus, the obtained results
for this case may be misleading.
The above stated general trends are confirmed for the MADE AE/30 and NTK500 machines, as well.
Similar conclusions can be drawn for the V27 machine. Some differences however exist. First, the yaw
misalignment effect is milder, especially at the higher wind speeds, and this is probably due to the fact that
the rotor is always running in its linear regime on account of the pitch control strategy. The effect of the
turbulence structure properties (see cases 2C and, especially, 2C+) is extremely pronounced, at all wind
speeds (as much as 70% increase of certain loads for 2C+).
Fine Grain PI
Although two, only, machines with similar dimensions have been considered, some preliminary
conclusions can be drawn from tables D.8 and D.9.
• Surprisingly enough, it appears that a significant increase of the turbulent energy in the transversal
(case 2) or the vertical (case 3) direction does not increase the loads more than 5%, in the worst case.
The influence is negligible at low wind speeds.
• The only important effect is that of the longitudinal length scale (case 4). Tower loads seem to increase
20% and blade loads 10% when the length scale changes from 60 to 30 m.

• The effect of the coherence decay factor (case 5) and the cross-correlations (case 6) seem to be
negligible.
In view of the above, it appears that the high loading encountered for the CLX1 and CLX2 turbulent sets
considered for the coarse grain PI analysis are mainly attributed to the length scale decrement. Evidently,
these conclusions may not be valid for machines of different size or control strategy.
5.4 Synthesis of the results

Both experimental and modelling-based parameter identification procedures have shown that the primary
fatigue loading parameter is the turbulent intensity, or better the standard deviation of the longitudinal
wind component. The analysis of the fatigue equivalent loads suggests that Leq is a quadratic function of
σu. Considering that at higher wind speeds a constant value of turbulent intensity can be assumed for a
given site, the above statement suggests that Leq is also a quadratic function of the mean wind speed. On
that basis, the influence of the site Weibull parameters on fatigue loading will be investigated.. Let
Leq = aU 2 + bU = b( eU 2 + U ) , e = a / b (8)
where b is the linear part parameter and e stands for the quadratic parameter.
The damage caused by a load spectrum of n cycles with ranges Li or by its equivalent load Leq is given by
the formulae:
n
1 N eq
D=∑ = (9)
i =1 q( pLi ) − m q( pLeq ) − m
where q is a component constant and p is a constant for a particular location and load. The magnitude that
decides the damage level is the m-moment of the equivalent load distribution over a large time period.
From the simplified expression for the short term equivalent load and after exponentiation we get:
  m  m 
Leq m = b m  e mU 2 m +   e m−1U 2 m−1 +...+  e m−iU 2 m−i +...+U m  (10)
  1  i 
Given the fact that in general the quadratic parameter has small values (e<<1) the rejection of all terms
that contain greater than second order factors of e is justified. Thus, the mean value of Leq m , with the
variable U following a Weibull distribution, is given by the following formula, <.> denoting the expected
value operator:
Lm = b m U m + meb m U m+1 (11)
After introducing the analytic expressions for the higher moments for the U:
 m
Γ1 + 
 m m  k
U m = c m Γ1 +  = U (12)
 k  1
m
Γ1 + 
 k
the mean value of Leq m is finally given by the formula:

  m  m + 1 
 Γ 1 +  Γ 1 + 
m  k  k 
Leq m m
=b U + me U (13)
  1 m  1 
m+1
 Γ 1 +  Γ 1 +  
  k  k 
In order to assess the effect of the Weibull shape parameter (k), in conjunction with the effect of the
quadratic term, the equivalent load ratio defined below is introduced,
1/ m
 Leq(k , e, m, U )m 
EquivalentLoadRatio =  
 (14)
 Leq(2,0, m, U )m 
indicating the load increase relative to a reference state corresponding to the Raleigh (k=2) shape factor
and linear fatigue response to σu. Figures 5.1 and 5.2 show the equivalent load ratios for representative
Weibull shape parameter values (k=1.4-2.2), quadratic term values (e=0,0.01,0.02) and S-N curve slopes
(m=4,12). For all applications a mean wind speed of 7m/s was assumed.
Mean wind speed 7.0m/s, S-N curve slope m=4 Mean wind speed 7.0m/s, S-N curve slope m=12
1.60 1.90
Quadratic parameter
1.80 Quadratic parameter
1.50
e=0.0 1.70 e=0.0
e=0.01 e=0.01
1.40 1.60
Equivalent load ratio
Equivalent load ratio
e=0.02 e=0.02
1.50
1.30
1.40
1.20
1.30
1.10 1.20
1.10
1.00
1.00
0.90 0.90
1.20 1.40 1.60 1.80 2.00 2.20 2.40 1.20 1.40 1.60 1.80 2.00 2.20 2.40
Weibull k parameter Weibull k parameter
Figure 5.1 Equivalent load ratio vs k for m=4 Figure 5.2 Equivalent load ratio vs k for m=12
It is seen that for two sites with the same mean wind speed <U> and turbulence intensity, the dependence
of Leq on U suggests higher fatigue loading at the site with the smaller k. The effect is present for both m-
values considered and becomes stronger as the machine response deviates from linearity. Evidently, k is
not only important for extreme load considerations but also for fatigue loading. There is a strong evidence
that complex terrain k values can be quite small.
A second important finding is that the increment of the equivalent loads with σu is of the same order of
magnitude for all the wind speeds and loads (with the exception of the blade-edge moment). This implies
that complex terrain standardisation could follow the current IEC approach if an “equivalent” -increased
value- of turbulence intensity is used to account for the extra terrain-induced loading.
Yaw error was found to be the second parameter of significance. There is no reason to distinguish between
flat and complex terrain operation regarding the yaw error (mostly depending on the control system of the

machine). Nevertheless flow inclination associated with complex terrain acts on the machine as a
systematic “vertical yaw misalignment” which should be taken into account in fatigue analysis.
Wind shear has positive impact on fatigue loading, especially for two bladed teetered rotors. Complex
terrain wind shear is, in general, milder that the flat terrain equivalent expressed through the log-law.
Both theoretical and experimental investigations have shown that the reinforcement of the turbulent
kinetic energy in the lateral and vertical wind directions, due to the terrain, have a secondary effect (less
than 5% increase) on fatigue loading. The effect is negligible at low wind speeds and related to the non-
linear machine operation at post-stall conditions.
Low turbulence length scale values, often met at complex terrain, have a positive impact on fatigue. This
is seen from both experimental and modelling PI’s. The aeroelastic calculations overestimate this effect,
showing that a longitudinal length scale drop from 60 to 30 meters would increase fatigue loading up to
20%. On the contrary the experimental investigation provides a 5% equivalent. This is rather due to the
length scale estimation in-ambiguities discussed above. Similar conclusions can be drawn regarding
coherence decay.
It is not so clear that complex terrain is responsible for non-Gaussian turbulence behaviour. Experimental
investigations, at least for the data bases analysed, have not identified some strong effect on fatigue
loading.


6. RECOMMENDATIONS
First, the Rayleigh distribution, wherever used within the present IEC standard should be replaced by the
Weibull distribution with a k value less or equal than 1.8 when addressing at complex terrain sites. This is
important for both the extreme and fatigue loading of a machine. When a machine is intended to operate at
a site with k significantly lower than 1.8, it should be S(pecial) class certified.
It is felt that the 16% up to 18% turbulent intensity values used in the standards can cover complex terrain
operation with TI’s 13% up to 15% respectively, without additional care. It was estimated that this 3%
extra TI value can afford a 20 % increase in fatigue loading due to overall complex terrain effects. When a
machine is intended to operate at a complex terrain site with TI more than 15%, it should be S(pecial)
class certified.
The wind shear law used in the standards can be maintained, having in mind that a conservative estimation
of complex terrain loading is thus achieved.
Sites with wind inclination more than 20 degrees should be considered within the (S)pecial class context.


7. REFERENCES
[1] Wind turbine generator systems Part 1: Safety requirements, IEC 88/1400-1/Ed.2, May 96.
[2] A.N. Fragoulis et al., Load and Power Measurement Program on Wind Turbines operating in
Complex Mountainous Regions, (MOUNTURB JOU2-CT93-0378), final report Vol: I to III,
Nov. 1996.
[3] A.N. Fragoulis, Investigation of Design Aspects and Design Options for Wind Turbines
Operating in Complex Terrain Environments, (COMTERID JOR3-CT95-0033), Third
Periodic report, July 1997.
[4] E.L. Petersen et al., European Wind Atlas, Vol. II: Measurements and Modelling in Complex
Terrain, Published for the European Commission.
[5] D.C. Wilcox, Turbulence Modelling for CFD, @ 1993 by DCW Industries Inc., IBSN 0-9636051-
0-0.


Appendix A Speed-up and Turbulence Kinetic Energy

computations for single and tandem hills
Results of two-dimensional single and tandem Gaussian hill computations are presented in this Appendix.
The flow model used is relying on the incompressible Reynolds averaged Navier-Stokes equations with k-
omega turbulence closure. The constants of the model have been suitably calibrated for atmospheric flows.
The work is fully reported in reference [A1].
A.1 Nomenclature
x : axial distance
z : height above ground level (a.g.l.)
z0 : roughness length
h : hill(s) height
L : hill half-length ; for the Gaussian hills considered here L is defined as the axial distance
from the hill-top where z = h/2
Dht : distance among hill-tops
U : longitudinal wind component
u : normalised longitudinal wind component : u(z) = U(z)/Uref
K : kinetic energy of turbulence
k : normalised kinetic energy of turbulence k(z) = K(z)/ Uref 2
ref : subscript denoting far-field conditions (at the edge of the atmospheric boundary layer)
in : subscript denoting properties at the inlet plane (resembling those of the flat-terrain)
Speed Up (S) : u (z) / uin (z) = U (z) / Uin (z)

Turbulence increase factor (T) : k (z) / kin (z) = K (z) / Kin (z)
Anemometer height : 10 m a.g.l.
Hub height : 40 m a.g.l
A.2 References
[A1] J. Cardells, “Numerical Modelling of Wind Shear in Complex Terrain”, CRES report, Sept. 1997.
A.3 Results of flow calculations
The results of the flow calculations are given as graphs, but denoted as table I-1 through I-21.
























Appendix B A simplified wind shear law for hill-tops

A semi-empirical wind shear law for hill-tops is presented here. The proposed formulas have been
obtained by analysing parametric results obtained for Gaussian hills using the viscous-inviscid interaction
method developed at CRES for mean wind field simulations in complex terrain. Further details on the
method can be found in references [B1] and [B2].
B.1 Nomenclature
z : height above ground level
z0 : roughness length
H : hill height
H
L : hill length at height
2
u : normalised longitudinal wind component
ut , fl : normalised “flat terrain equivalent” friction velocity (a function of the roughness length and
the atmospheric boundary layer thickness)
k : Von Karman Constant = 0.41
B.2 Single hill-top profile law

The resulted wind-shear law is the following
ut , fl
( )( )
u( H , L, z 0 , z ) = 1 + f1 ( H , L, z ) + 1 + f 2 ( H , L, z 0 ) 1 + f 3 ( H , L, z 0 ) +
k
(1 + f ( H , L, z )) ln zz
2 0
0
where f1 represents the potential flow part
 H 1
f1 ( H , L, z ) = a1  
 L  z
2
1 + 
 L
and the functions f2 , f3, corresponding to the viscous part, read
2
 z   z  H   z   z  H 
f 2 ( H , L, z 0 ) =  b1 + b2 ln  0  + b3 ln 2  0     −  b4 + b5 ln  0  + b6 ln 2  0    
  L  L  L    L  L  L 
 z   z  H 
f 3 ( H , L, z 0 ) = − c1 + c2 ln  0  + c3 ln 2  0    
  L  L  L 
The best-fitted values of the model constants are:
a1 = 105
. b1 = 0.49 b2 = 0167
. b3 = 0.0066 b4 = 0105
.
b5 = 0.04 b6 = 0.0014 c1 = 0165
. c2 = 0.018 c3 = 0.00065

B.3 Example
The application of the above formulas for a Greek complex terrain site (Andros Island) with H/L = 0.1 and
z0 = 0.15 m is presented in figure B.1. Calculations (continuous curves) are compared against experimental
data obtained at two neighbour masts, called M1 and M2. The dashed lines represent the corresponding
estimation of the single-slope (flat terrain) log-law.
60
complex M1
50 complex M2
flat M1
flat M2
exper. M1
Height A.G.L (m)
40 exper. M2
30
20
10
10 12 14 16
U (m/s)
Figure B.1 Application of the Hill Top velocity profile model at Andros
B.4 References
[B1] D.I. Douvikas, “3D Flow Computations over Complex Terrain - An Integral based Formulation”,
CRES report, Oct. 1997.
[B2] D.I. Douvikas and P.K. Chaviaropoulos, “3D Viscous Computations over Complex Terrain using
Integral Boundary Layer Method”, Proc. EWEC 97, Oct. 1997, Dublin, U.K.

Appendix B Theoretical investigation of the Reynolds Stress

Tensor in complex terrain [C1]
The terrain influence on turbulence structure is investigated in this Appendix. The presented theoretical
approach is based on the non-isotropic Reynolds-Stress-Model proposed by Saffman [C2]. The
investigation is limited to 2-D and quasi-3D flows. The emphasis is put on establishing the interrelation
between the standard deviation ratios of the wind components with the mean flow gradients.
C.1 Definitions
( x i ; i = 1,2,3) ≡ (x, y, z) Cartesian coordinates
(U i ; i = 1,2,3) ≡ (U,V,W) Mean flow velocity components
( ui ; i = 1,2,3) ≡ (u, v, w) Velocity fluctuation components
R : R ij = ui u j ; i, j = 1,2,3 Correlation tensor
1 ∂U ∂U j
S : S ij = ( i + ) Mean strain rate
2 ∂x j ∂x i
1 ∂U ∂U j
Ω : Ω ij = ( i − ) Rotation
2 ∂x j ∂x i
1
k = ( R 11 + R 22 + R 33 ) Kinetic energy of turbulence
2
ω =ε/k Specific dissipation rate
C.2 A Second order expression for Reynolds Stresses

Following ref. [C2], the following non-isotropic turbulence model is introduced.
1 2 2 1
R = I − S + D 2 [S Ω − Ω S ] ; D is a constant to be defined
k 3 ω ω
C.3 Application in 2-D flows

The following assumptions / definitions are introduced for 2-D flows
∂U ∂W
∂ () ∂U ∂x ∂x
V=0, ∂y = 0 ; s = ∂z , λ = ∂U , µ = ∂U
∂z ∂z
Then, it can be shown that the only non-zero correlation tensor components are:

2
σ u2 u 2 2  s D s
≡ = − 2  λ −   (1 − µ 2 )
k k 3 ω 2 ω
σ v2 v 2 2
≡ =
k k 3
2
σ w2 w 2 2  s D s
≡ = + 2  λ +   (1 − µ 2 )
k k 3 ω 2 ω
2
− (τ / ρ ) uw  s  s
≡ = −  (1 + µ ) + D  λ (1 − µ )
k k ω ω
Consequently, in all 2-D cases, flat or complex
2 2
σ  σ 
σ + σ = 2σ
2
u
2
w
2
v or  u  +  w  = 2
σv   σv 
Deviations from the above 2-D SDV’s formula indicate 3-dimensionality effects
Application in 2-D Flat Terrain. Estimating the D constant
For 2-D flat terrain it can be assumed that
σ u :σ v :σ w ≡ 1 : 0.8 : 0.5
and, in addition,
λ, µ ≅ 0
uT  z  u
U= ln   ⇒ s = T in the log - layer
κ  z0  κz
uT2 u  s
k= , ω = T ⇒   = A = const. ≅ 0183
.
A Aκz ω
where uT is the friction velocity. Then, the non-zero components of the correlation tensor yield

 σ u2  2 1
  = − DA 2
 k  Flat 3 2
 σ v2  2
  =
 k  Flat 3
 σ w2  2 1
  = + DA 2
 k  Flat 3 2
 uw  − (τ / ρ ) u
  ≡ ≡ − T = −A
 k  Flat k k
Applying the 1 : 0.8 : 0.5 rule to the above expression D is estimated to be -23.4.
Application in 2-D Complex Terrain
Assuming that the ratio (s/ω) is constant and retains its flat terrain value A, the non-zero entries of the
Reynolds stress tensor yield:
 σ u2  σ 2  1
  = u  − 2 A λ + DA 2 µ 2
 k  Complex  k  Flat 2
 σ v2   σ v2  2
  =  =
 σ w2   σ w2  1
  =  + 2 A λ − DA 2 µ 2
 uw   uw 
  =  − Aµ + DA 2 λ (1 − µ )
k k
  Complex   Flat
Following Wilcox [C3] it can be shown that for high Reynolds numbers (s/ω)2=const.*P, where P
represents the net production per unit dissipation of the turbulent kinetic energy.
Considering a 2-D hill case and the accelerating part of the flow (upwind from the hill-top) the turbulence
kinetic energy is expected to be increased compared to its free-stream (flat terrain) value, while λ takes
positive values. As a consequence (and since D is negative), the normalized (by k) u-component variance
is decreased in respect to its free-stream value, the w-component normalized variance is accordingly
increased while the v-component variance retains its free-stream value. This means that turbulent energy
is transferred from the u- to the w- component. In absolute terms both variances of the v- and w-
components are increased because of k while no direct conclusion may be stated for the u-component
(since k is increased but the normalized variance is decreased).

C.4 Application in 3-D flows

The treatment of the fully 3-D case is beyond the scope of the present work. A simplified 3-D case will be
considered instead, which is based on the assumption that all the mean flow partial derivatives in respect
to one of the Cartesian directions are negligible. This can simulate the case of a 2-D infinite hill which
faces an oblique flow in respect to its axis of symmetry, for instance. In this case one may model the
transversal wind component in respect to the longitudinal one using, for example, the Johnston’s
triangular profile which is applied below. Under these assumptions:
∂U i
≡ 0 ; ∀i = 1,2,3 and V = a(U e − U )
∂y
where the subscript e denotes the external (inviscid part) flow and α is the transversal profile parameter
which measures the local three-dimensionality of the flow field. Under the above assumptions the non-
zero entries of the normalized Reynolds stress tensor yield:
 σ u2   σ u2  1 σ 2 
  =   + DA 2 a 2 λ2 <  u
 k  3D − Complex  k  2 D − Complex 2  k  2 D −Complex
 σ v2   σ v2  1  σ v2 
  =  − DA a ( λ + 1) >  
2 2 2
 k  3D − Complex  k  2 D − Complex 2  k  2 D − Complex

 σ w2  σ 2  1 σ 2 
  = w + DA 2 a 2 <  w
 k  3D − Complex  k  2 D − Complex 2  k  2 D − Complex
 uv  1
  = aAλ + DA 2 a(1 + λ2 )
 k  3 D −Complex 2
 uw  1
  = − A(1 + µ ) + DA 2 λ (1 − µ + a 2 )
 k  3 D −Complex 2
implying that the uv correlation does not vanish any more. It is also seen that the diagonal terms satisfy the
following non-equality:
 σ u2   σ w2   σ v2  3
  +  − 2  = DA 2 a 2 (1 + λ2 ) < 0 or
 k  3D − Complex  k  3 D −Complex  k  3D − Complex 2
 σ u2  σ 2 
 2 +  w2  < 0
 σ v  3D − Complex  σ v  3 D −Complex
C.5 Conclusions
For the considered cases it has been shown that
2 2
 σu   σw 
  +  ≤ 2
 σv   σv 

where the equality holds for all 2-D flows. In the accelerating part of a 2-D flow the kinetic energy of
turbulence increases, while turbulent energy is transferred from the longitudinal u_ to the vertical w_ wind
component. This implies that the turbulence in that area becomes more isotropic. This result is in
agreement with the available complex terrain experimental data.
C.6 References
[C1] P.K. Chaviaropoulos, Unpublished work at CRES.
[C2] P.G. Saffman, “Development of a Complete Model for the Calculation of Turbulent Shear Flows”
April 1976 Symposium on Turbulence and Dynamic Systems, Duke Univ., Durham, NC.
[C3] D.C. Wilcox, Turbulence Modelling for CFD, @ 1993 by DCW Industries Inc., IBSN 0-9636051-
0-0


Appendix C Tables from measurement-based PI

The application of the multivariate regression analysis to the experimental load data bases of three
machines (WXT110, V27 and NTK500) erected at complex terrain sites is presented in this appendix. The
theoretical background of the approach is briefly discussed first.
D.1 Theoretical background

To establish the relative importance of the inflow parameters to the equivalent load range, the multiple
regression analysis is adopted. The topics that are discussed herein are related to the regression analysis
application and assessment [D1] as well as to the parameter identification procedure ([D2] and [D3]).
The multivariate regression analysis is based on least square fitting process as a way to identify the
maximum likelihood estimator of the fitted parameters. It is assumed that the measurement errors are
independent and normally distributed with constant standard deviation. The scope of the regression is the
estimation of the coefficients αk,k=1,…,M2, for the expression of the dependent variable y3 as follows:
M
y( x i ) = ∑a
k =1
k X k ( x i ) + E( x i ) , i = 1,..., N
where Xk(xi) 5is the value of the kth independent variable at point xi 6and E(xi) 7is the associated error at
the same point.
Chi-square fitting is applied as the weighted least square fitting process, in which the magnitude χ28
defined as:
2
N
 y i - y i  M
χ 2
= ∑ 
 σi 
 , y i = ∑a k X k(xi)
i=1 k =1
is minimised. The magnitude σi9 presents the measurement standard error at point (y(xi),xi)10. If the
measurement errors are not known (as in the present practice) they are all set equal to unity. The number
ν=Ν-Μ 11represents the fitting degrees of freedom.
The minimum of χ212 occurs where its derivatives with respect to parameters αk,k=1,…,M 13vanish. The
resulting linear equations are solved by the Gauss-Jordan elimination technique, yielding the coefficients
αj,j=1,…,M14:
-1
M
 N yi X k ( xi )   N X j ( xi ) X k ( xi ) 
aj = ∑ C jk ∑ σ i2 
, C jk = ∑
σ i2  , j = 1,..., M
k =1  i=1  i=1  jk
where the standard deviation related to the estimate of aj is given by the equation:
 ∑ σ  =
2
N
 ∂a j
σ(a j ) = 2
  C jj
 
i
i=1  ∂ yi 
D.2 Assessment of the accuracy of the multivariate regression

For the assessment of the accuracy of the regression analysis, the following magnitudes were considered:

a) The total sum of squares (SSY), sum of squares due to error (SSE), sample squared correlation
coefficient (R2) and finally the fitting standard deviation σ y d16efined as:
N N
SSY - SSE SSE
SSY = ∑ ( yi - y mean )2 , SSE = ∑ ( yi - y i )2 , R2 = , σ y =
i=1 i=1 SSY N-M
b) F-test statistic, testing the hypothesis, H0 : there is no significant overall regression using the M
independent variables, is defined as follows:
SSY - SSE  N - M 
F =  
SSE  M -1 
Consequently the p-value for the above defined F-test is calculated from Fisher's F sampling distribution
with M-1 and N-M degrees of freedom ( p F | M -1,N - M 17).
c) For the quantitative assessment of the relation of the dependent variable on each of the independent
parameters Xk, various magnitudes were considered, namely the regression coefficients αk themselves, the
relative per sigma dependence coefficient Sk or the t-test statistic for each parameter. The relative per
sigma dependence coefficient Sk, is defined as:
N
∑ (X )
2
k ( xi ) - X k
σ Xk i=1
S k = ak , σ Xk =
y mean N -1
The magnitude Sk represents the relative change of the dependent variable induced by the increase of the
value of parameter Xk by its standard deviation σΧκ within the regression domain20. Assuming that the
probability of that change is expected to be comparable for all independent variables, the relative
assessment of the effect of each variable Xk on y can be attained. Although the above assumption is not
valid in cases where the distribution of Xk, within the regression domain, differs significantly from
normal, still the magnitude Sk offers a valuable means for weighting the dependence of each Xk on the
dependent magnitude.
d) The t-test statistic, testing the hypothesis, H0 : the regression coefficient a k 21 is zero, is defined as
follows:
ak
t =
σ ( ak )
Consequently the p-value for the above defined t-test is calculated from Student's t sampling distribution
with N-M degrees of freedom ( pt | N - M 22).

D.4 Tables
Table D.1 Sensitivity parameter table for WINCON W110XT operating at CRES test station
Sensitivity parameter (∂Leq/∂xi)
Fatigue load (1Hz equiv.) U Leq U αws σu σv/σu σw/σu
Flapwise b.m. (m=12) 5-8m/s 13,51 0,99 8,31 5,41 1,65 3,41
8-11m/s 16,44 0,26 12,30 6,67 2,85 4,84
13-17m/s 27,36 1,60 s.i. 8,77 4,90 14,82
Edgewise b.m. (m=12) 5-8m/s 31,6 0,30 1,40 1,31 0,32 1.13
8-11m/s 32,96 0,45 2,83 1,90 0,79 1,81
13-17m/s 39,81 1,30 s.i. 3,12 4,17 3,31
Shaft yaw (m=8) 5-8m/s 12,25 0,67 10,31 6,40 2,46 2,85
8-11m/s 15,47 0,64 15,09 6,05 3,33 4,84
13-17m/s 26,35 1,37 16,78 7,97 6,45 13,71
Shaft tilt (m=8) 5-8m/s 16,86 0,77 8,93 8,33 2,61 4,51
8-11m/s 21,01 0,69 15,56 8,05 4,23 7,38
13-17m/s 32,88 1,45 11,72 9,29 4,42 18,24
Tower torsion (m=4) 5-8m/s 12,72 0,76 6,52 4,74 1,89 3,75
8-11m/s 15,91 0,76 18,65 5,56 1,91 6,77
13-17m/s 29,31 2,03 21,20 8,53 7,63 13,93
Tower bottom (m=4) 5-8m/s 60,41 5,98 s.i. 35,82 9,54 21,24
8-11m/s 86,68 6,54 66,36 41,56 17,31 42,62
13-17m/s 163,0 8,98 113,3 36,38 29,27 50,62
s.i. indicates that the specific parameter was rejected by the backward elimination process

Table D.2 Sensitivity parameter table for VESTAS V27 operating at Andros site
Fatigue load (1Hz equiv.) U Leq U αws σu σϕ/TI Lu Uskew
Flapwise b.m. (m=12) 5-7m/s 16,17 1,32 s.i. 22,0 4,71 -0,0170 0,315
8-11m/s 25,34 s.i. s.i. 16,61 8,32 -0,0183 1,53
13-16m/s 37,21 2,78 -22,9 22,41 11,36 -0,058 s.i.
Edgewise b.m. (m=12) 5-7m/s 51,71 0,25 s.i. 3,06 0,80 -0,004 -0,116
8-11m/s 54,82 1,03 s.i. 3,08 2,00 -0,0056 0,836
13-16m/s 64,83 2,40 18,96 2,49 s.i. -0,014 2,25
Shaft parallel b.m. (m=8) 5-7m/s 31,74 s.i. -30,2 17,08 4,37 -0,021 s.i.
8-11m/s 39,45 0,42 -29,7 15,0 8,56 -0,0277 2,12
13-16m/s 46,42 s.i. -25,1 11,40 12.64 -0,0247 2,27
Shaft yaw (m=8) 5-7m/s 19,94 1,09 5,92 15,47 4,99 -0,018 s.i.
8-11m/s 27,83 0,42 4,10 13,23 7,54 -0,0253 1,31
13-16m/s 35,62 1,50 s.i. 15,56 12,89 -0,0165 1,28
Shaft tilt (m=8) 5-7m/s 19,92 0,90 s.i. 15,35 6,93 -0,0128 -0,342
8-11m/s 29,21 s.i. s.i. 15,60 8,21 -0,0275 1,31
13-16m/s 37,99 2,18 -9,67 16,62 12,54 -0,039 4,038
Tower top (m=4) 5-7m/s 16,15 s.i. 12,21 4,62 3,99 s.i. 0,939
8-11m/s 25,81 0,68 12,02 11,73 6,46 -0,0165 1,29
13-16m/s 32,98 1,33 s.i. 14,67 10,86 -0,0288 2,18
Tower bottom (m=4) 5-7m/s 61,28 2,55 s.i. 55,92 17,05 s.i. 1,47
8-11m/s 99,75 4,14 16,08 67,69 32,44 -0,101 3,01
13-16m/s 180,5 21,0 s.i. 83,72 61,99 -0,254 33,05
4

Table D.3 Sensitivity parameter table for NORDTANK NTK500/37 operating at Toplou site
Fatigue load (1Hz U Leq U αws σu σϕ/TI Lu Uskew Ukurt
equiv.)
Flapwise b.m. (m=12) 5-7m/s 45,61 -1,29 s.i. 47,99 8,09 -0,13 s.i. s.i.
8-11m/s 46,88 -2,26 s.i. 45,3 13,3 -0,10 -2.62 s.i.
13-16m/s 68,56 4,07 s.i. 28,72 27,43 -0,069 3,49 s.i.
Edgewise b.m. (m=12) 5-7m/s 151,9 s.i. -7,46 7,83 -1,25 -0,058 s.i s.i.
8-11m/s 157,8 4.25 0,63 9,16 2,8 -2,20 s.i. s.i.
13-16m/s 190,3 4,83 2,24 7,51 s.i. -0,074 s.i. 0,42
Shaft normal b.m. (m=8) 5-7m/s 85,49 2,03 -39,0 30,54 25,25 s.i. -1,50 0,33
8-11m/s 98,79 2,51 -1,17 35,44 28,08 -0,14 -5,23 s.i.
13-16m/s 141,0 7,07 s.i. 37,9 36,9 -0,13 1,49 s.i.
Shaft yaw (m=8) 5-7m/s 44,20 1,43 s.i. 34,47 7,53 -0,10 -1,87 s.i.
8-11m/s 52,66 0,73 s.i. 32,43 11,87 -0,076 -1,91 1,16
13-16m/s 77,09 4,65 1,23 24,55 26,0 -0,036 4,58 1,67
Shaft tilt (m=8) 5-7m/s 51,74 1,47 s.i. 35,27 14,48 -0,083 -1,24 s.i.
8-11m/s 61,65 1,21 -0,71 37,9 18,56 -0,11 -2,83 0,10
13-16m/s 95,28 5,46 1,07 32,12 32,51 -0,69 1,37 s.i.
Tower torsion (m=4) 5-7m/s 39,62 s.i. s.i. 24,84 4,84 -0,060 -1,21 0,55
8-11m/s 46,9 2,08 s.i. 22,62 6,51 -0,048 -1,0 0,71
13-16m/s 74,3 5,46 1,57 19,66 15,13 -0,036 4,69 1,87
Tower top (m=4) 5-7m/s 56,95 7,59 s.i. 51,78 10,53 -0,16 -1,15 0,36
8-11m/s 78,62 5,23 -1,73 61,17 22,18 -0,20 -1,64 1,21
13-16m/s 164,0 18,32 s.i. 77,72 57,01 -0,17 9,75 2,91
Tower bottom (m=4) 5-7m/s 203,3 14,97 s.i. 246,8 55,64 -0,73 -13,1 2,67
8-11m/s 274,0 17,05 -8,37 267,7 101,8 -0,81 -10,2 6,01
13-16m/s 594,3 60,15 5,48 277,6 235,8 -0,50 35,83 13,26

D.5 References
[D1] Kleinbaum K., Kupper L.: “Applied regression analysis and other multivariable methods”, PWS-
KENT Publishing Company, Boston, 1985.
[D2] Mouzakis F., Morfiadakis E., Fragoulis A.: MOUNTURB (JOU2-CT93-0378) Final Report on
Complex terrain wind-WT parameter identification and quantification (Wind and WT
measurements), Volume II, November 1996.
[D3] Mouzakis F., Morfiadakis E., Dellaportas P.: “Parameter identification on power performance of
wind turbines operating at complex terrain”, 2nd EACWE, Genova, June 1997.

Appendix D Tables from modelling-based PI

The results obtained from the coarse and fine grain parameter identification procedure are presented in
this Appendix. Cumulative Rainflow Counting tables of Equivalent loads m=4, 8, 12 are included for the
following loads:
RFM : blade root flapping moment

REM : blade root edge moment
TBMx : tower bottom bending moment (normal to the mean flow)
TBMy : tower bottom bending moment (parallel to the mean flow)
TTM : tower torsion
DTM : drive train - low speed - torque
Dimensional moments (kNm) are only presented for the reference cases (2F for coarse grain, for
instance). These are clearly designated using bold characters. All the other columns represent the ratio of
the calculated moments versus the respective reference value.
Although each machine has been modelled by several participants, one set of results per machine is
presented here. The name of the participant whose results are presented is marked on the tables.
Tables E.1 to E.7 include the results of the coarse grain PI. The fine grain PI results are shown in tables
E.8 and E.9

Table E.1 COARSE GRAIN PI RESULTS WXT110 modelled by NTUA

Uhub = 10 m/s
Leq-4 1F 2F 3F 4F 5F 6F 1C 2C 3C 4C 5C 6C 2C+
RFM 0.68 16.8 1.13 1.14 1.04 1.48 0.70 1.03 1.14 1.17 1.07 1.53 1.23
REM 0.97 42.6 0.96 1.00 1.00 1.04 0.97 1.00 0.96 1.00 1.01 1.05 1.03
TBMx 0.62 46.5 0.98 0.99 0.99 1.56 0.71 1.14 1.11 1.13 1.14 1.77 1.34
TBMy 0.59 88.1 1.00 0.98 0.99 1.60 0.67 1.14 1.15 1.12 1.13 1.81 1.26
TTM 0.64 22.0 1.03 1.00 1.00 1.51 0.63 1.00 1.01 0.99 1.00 1.51 1.29
DTM 0.64 9.7 1.01 0.97 1.00 1.45 0.68 1.07 1.09 1.04 1.07 1.56 1.29
RFM 0.66 17.5 1.14 1.14 1.03 1.49 0.67 1.04 1.16 1.18 1.07 1.55 1.19
REM 0.97 39.4 0.96 1.00 1.01 1.06 0.97 1.00 0.96 1.00 1.01 1.06 1.04
TBMx 0.62 47.5 1.00 0.99 0.99 1.58 0.71 1.13 1.13 1.12 1.12 1.79 1.32
TBMy 0.58 107. 1.07 0.97 1.00 1.63 0.63 1.12 1.23 1.09 1.12 1.82 1.19
TTM 0.62 21.4 1.03 0.99 1.00 1.53 0.63 1.01 1.03 1.00 1.02 1.56 1.29
DTM 0.64 11.9 1.08 0.97 1.01 1.44 0.67 1.07 1.17 1.05 1.08 1.53 1.17
RFM 0.65 19.5 1.12 1.14 1.02 1.49 0.67 1.05 1.15 1.18 1.06 1.55 1.16
REM 0.96 38.6 0.96 1.00 1.01 1.07 0.96 1.01 0.96 1.00 1.01 1.08 1.04
TBMx 0.63 51.1 1.00 0.99 0.99 1.57 0.71 1.11 1.13 1.11 1.11 1.78 1.31
TBMy 0.57 125. 1.09 0.96 1.00 1.62 0.62 1.12 1.25 1.08 1.11 1.80 1.16
TTM 0.60 23.0 1.03 0.99 1.01 1.54 0.62 1.02 1.03 1.00 1.03 1.59 1.28
DTM 0.64 13.7 1.10 0.99 1.01 1.43 0.67 1.06 1.18 1.04 1.07 1.50 1.12
Uhub = 14 m/s
RFM 0.67 20.9 1.55 1.02 1.12 1.56 0.69 1.06 1.56 1.08 1.18 1.67 1.22
REM 0.94 45.7 1.01 0.99 1.03 1.09 0.94 1.00 1.01 1.00 1.03 1.09 1.06
TBMx 0.53 77.7 1.11 1.03 1.02 1.69 0.60 1.11 1.21 1.16 1.13 1.88 1.33
TBMy 0.58 129. 1.01 1.06 0.99 1.65 0.66 1.13 1.13 1.20 1.11 1.89 1.27
TTM 0.61 27.0 1.20 1.02 1.01 1.61 0.62 1.03 1.21 1.04 1.03 1.66 1.26
DTM 0.51 12.3 1.24 1.04 1.02 1.76 0.54 1.06 1.28 1.11 1.07 1.83 1.36
RFM 0.66 20.8 1.56 1.03 1.13 1.53 0.70 1.08 1.58 1.10 1.19 1.68 1.22
REM 0.93 42.6 1.01 0.99 1.03 1.10 0.93 1.00 1.01 1.00 1.03 1.11 1.07
TBMx 0.54 78.4 1.13 1.02 1.01 1.72 0.60 1.13 1.25 1.16 1.15 1.90 1.33
TBMy 0.57 133. 1.03 1.05 0.99 1.66 0.65 1.14 1.17 1.19 1.12 1.94 1.25
TTM 0.60 26.0 1.20 1.02 1.01 1.63 0.63 1.05 1.24 1.07 1.07 1.75 1.25
DTM 0.52 12.4 1.33 1.08 1.02 1.81 0.55 1.06 1.41 1.17 1.07 1.90 1.36
RFM 0.66 22.2 1.54 1.03 1.11 1.51 0.70 1.08 1.56 1.09 1.18 1.66 1.23
REM 0.92 42.0 1.02 0.99 1.03 1.12 0.92 1.00 1.02 1.00 1.03 1.12 1.08
TBMx 0.54 83.0 1.14 1.01 1.01 1.73 0.60 1.14 1.30 1.17 1.17 1.91 1.33
TBMy 0.57 144. 1.05 1.04 0.99 1.65 0.64 1.14 1.20 1.18 1.11 1.94 1.24
TTM 0.60 27.7 1.18 1.02 1.01 1.64 0.63 1.07 1.25 1.08 1.08 1.78 1.24
DTM 0.53 13.2 1.39 1.10 1.02 1.85 0.56 1.07 1.48 1.20 1.08 1.95 1.37
Uhub = 18 m/s
RFM 0.64 25.2 1.85 0.99 1.13 1.64 0.67 1.08 1.89 1.08 1.21 1.79 1.30
REM 0.93 47.2 1.10 0.99 1.04 1.14 0.93 1.01 1.10 1.00 1.04 1.15 1.07
TBMx 0.52 108. 1.17 0.99 1.02 1.67 0.57 1.12 1.32 1.12 1.15 1.88 1.38
TBMy 0.55 182. 1.00 0.97 1.00 1.58 0.62 1.12 1.13 1.11 1.13 1.79 1.30
TTM 0.60 32.6 1.26 0.96 1.03 1.71 0.62 1.06 1.31 1.02 1.09 1.83 1.34
DTM 0.50 16.3 1.28 0.98 1.01 1.74 0.53 1.07 1.36 1.06 1.09 1.85 1.39
RFM 0.63 24.9 1.88 1.00 1.13 1.65 0.67 1.10 1.95 1.12 1.22 1.84 1.30
REM 0.92 44.0 1.11 1.00 1.03 1.16 0.92 1.01 1.12 1.00 1.04 1.18 1.08
TBMx 0.50 113. 1.13 0.95 1.01 1.65 0.56 1.11 1.29 1.09 1.14 1.86 1.35
TBMy 0.54 188. 0.99 0.98 1.01 1.58 0.62 1.11 1.13 1.11 1.13 1.79 1.29
TTM 0.60 30.6 1.33 0.97 1.04 1.74 0.65 1.09 1.42 1.07 1.13 1.92 1.36
DTM 0.49 17.2 1.25 0.95 0.99 1.72 0.53 1.09 1.33 1.02 1.09 1.84 1.32
RFM 0.62 26.7 1.84 1.00 1.11 1.65 0.66 1.10 1.93 1.13 1.20 1.83 1.31
REM 0.91 43.5 1.13 1.00 1.03 1.18 0.91 1.01 1.14 1.01 1.05 1.20 1.09
TBMx 0.48 124. 1.09 0.91 1.01 1.60 0.54 1.09 1.25 1.06 1.12 1.82 1.32
TBMy 0.54 204. 0.99 0.97 1.02 1.57 0.61 1.10 1.11 1.10 1.12 1.77 1.27
TTM 0.62 31.5 1.38 0.99 1.03 1.76 0.67 1.12 1.50 1.11 1.16 1.99 1.38
DTM 0.48 18.8 1.22 0.94 0.99 1.72 0.53 1.11 1.32 0.99 1.10 1.84 1.27

Table E.2 COARSE GRAIN PI RESULTS V27 modelled by RISØ

Uhub = 10 m/s
RFM 0.60 23.5 1.13 0.99 1.03 1.59 0.77 1.33 1.45 1.32 1.35 2.14 1.56
REM 0.90 58.1 1.00 1.01 0.99 1.05 0.94 1.05 1.05 1.05 1.05 1.11 1.06
TBMx 0.52 64.2 1.14 1.01 1.02 1.58 0.68 1.30 1.49 1.30 1.35 2.04 1.68
TBMy 0.55 148. 1.29 1.01 1.03 1.63 0.69 1.31 1.67 1.29 1.31 2.12 1.50
TTM 0.59 27.6 1.00 1.00 1.01 1.58 0.74 1.31 1.28 1.32 1.31 2.09 1.66
DTM 0.58 11.9 1.01 1.00 0.99 1.54 0.73 1.27 1.26 1.27 1.28 1.97 1.46
RFM 0.60 29.8 1.12 0.99 1.04 1.56 0.70 1.21 1.36 1.19 1.23 1.95 1.46
REM 0.89 61.7 1.00 1.02 0.99 1.05 0.91 1.03 1.03 1.03 1.03 1.10 1.08
TBMx 0.52 74.5 1.15 1.01 1.05 1.58 0.66 1.26 1.45 1.26 1.31 1.96 1.67
TBMy 0.55 227. 1.31 1.04 1.02 1.68 0.58 1.11 1.39 1.07 1.10 1.77 1.27
TTM 0.56 33.6 1.00 0.99 1.01 1.55 0.71 1.26 1.22 1.27 1.25 1.99 1.67
DTM 0.55 19.6 1.04 1.01 0.98 1.52 0.59 1.06 1.06 1.05 1.06 1.60 1.19
RFM 0.61 35.7 1.10 1.00 1.05 1.55 0.68 1.17 1.36 1.13 1.18 1.87 1.38
REM 0.88 63.9 1.01 1.02 0.99 1.06 0.89 1.04 1.04 1.03 1.03 1.13 1.09
TBMx 0.51 84.2 1.15 1.01 1.09 1.58 0.66 1.25 1.48 1.26 1.31 1.95 1.67
TBMy 0.55 286. 1.32 1.05 1.03 1.71 0.57 1.09 1.36 1.04 1.07 1.73 1.21
TTM 0.55 38.8 0.99 0.98 1.03 1.54 0.71 1.27 1.23 1.29 1.26 2.01 1.68
DTM 0.55 24.4 1.06 1.01 0.98 1.52 0.60 1.06 1.07 1.05 1.06 1.60 1.15
Uhub = 14 m/s
RFM 0.62 32.9 1.11 0.99 1.01 1.57 0.72 1.20 1.29 1.17 1.20 1.92 1.53
REM 0.91 61.9 0.99 1.00 0.98 1.11 0.95 1.03 1.03 1.02 1.03 1.18 1.10
TBMx 0.56 113. 1.24 1.00 1.03 1.66 0.68 1.24 1.46 1.23 1.23 2.06 1.68
TBMy 0.58 230. 1.24 1.01 1.03 1.65 0.71 1.23 1.44 1.24 1.26 2.00 1.54
TTM 0.57 40.2 0.99 1.00 1.00 1.62 0.68 1.20 1.18 1.20 1.20 1.99 1.63
DTM 0.58 16.9 1.13 1.01 1.02 1.67 0.68 1.21 1.24 1.19 1.20 2.04 1.63
RFM 0.61 42.4 1.08 0.98 0.99 1.55 0.70 1.15 1.21 1.13 1.15 1.84 1.45
REM 0.89 66.4 0.99 1.00 0.98 1.13 0.95 1.04 1.03 1.03 1.04 1.20 1.12
TBMx 0.57 130. 1.28 1.01 1.02 1.70 0.67 1.29 1.53 1.28 1.25 2.17 1.74
TBMy 0.61 304. 1.23 0.99 1.00 1.61 0.74 1.22 1.37 1.22 1.23 1.90 1.43
TTM 0.56 49.0 0.96 0.99 0.99 1.60 0.67 1.18 1.15 1.20 1.21 1.99 1.65
DTM 0.54 22.4 1.36 1.02 1.02 1.69 0.64 1.17 1.38 1.15 1.15 1.98 1.51
RFM 0.61 50.0 1.07 0.97 0.99 1.52 0.69 1.13 1.16 1.10 1.12 1.78 1.44
REM 0.88 69.3 0.99 1.00 0.98 1.14 0.95 1.05 1.04 1.04 1.05 1.23 1.14
TBMx 0.58 145. 1.31 1.00 1.01 1.72 0.68 1.32 1.58 1.31 1.28 2.23 1.76
TBMy 0.62 357. 1.24 0.99 0.99 1.60 0.76 1.24 1.34 1.23 1.23 1.88 1.40
TTM 0.56 56.4 0.95 0.99 0.98 1.59 0.67 1.19 1.14 1.20 1.22 2.01 1.68
DTM 0.53 26.7 1.42 1.03 1.02 1.68 0.63 1.17 1.40 1.15 1.13 1.97 1.45
Uhub = 18 m/s
RFM 0.63 44.1 1.19 0.99 0.98 1.61 0.74 1.25 1.36 1.25 1.24 2.03 1.61
REM 0.88 64.1 1.02 1.00 0.98 1.25 0.91 1.10 1.09 1.09 1.07 1.42 1.28
TBMx 0.56 223. 1.00 1.00 1.00 1.66 0.72 1.28 1.21 1.28 1.27 2.13 1.69
TBMy 0.54 388. 1.33 1.00 1.03 1.67 0.65 1.22 1.58 1.22 1.26 2.06 1.57
TTM 0.59 54.9 1.06 1.01 1.01 1.63 0.73 1.27 1.27 1.26 1.27 2.09 1.80
DTM 0.59 31.8 0.93 1.00 1.01 1.62 0.71 1.25 1.09 1.25 1.24 1.94 1.65
RFM 0.61 55. 1.26 0.99 1.00 1.63 0.71 1.23 1.41 1.22 1.22 2.02 1.56
REM 0.86 70. 1.01 1.00 0.98 1.26 0.90 1.11 1.09 1.10 1.07 1.44 1.29
TBMx 0.55 262. 1.00 1.00 1.00 1.69 0.72 1.28 1.21 1.28 1.27 2.14 1.68
TBMy 0.55 489. 1.36 1.00 1.02 1.65 0.64 1.18 1.60 1.20 1.23 1.98 1.49
TTM 0.59 64.4 1.05 1.01 1.03 1.63 0.74 1.33 1.30 1.31 1.33 2.23 1.87
DTM 0.58 41.5 0.95 1.01 1.01 1.65 0.70 1.23 1.11 1.23 1.22 1.94 1.54
RFM 0.61 63.3 1.31 0.98 1.01 1.62 0.69 1.21 1.45 1.20 1.22 2.02 1.54
REM 0.84 74.3 1.01 0.99 0.98 1.27 0.89 1.11 1.09 1.11 1.08 1.48 1.31
TBMx 0.55 294. 0.99 1.01 1.01 1.74 0.72 1.28 1.21 1.28 1.28 2.15 1.67
TBMy 0.55 566. 1.39 1.01 1.02 1.66 0.64 1.19 1.62 1.20 1.22 1.94 1.45
TTM 0.60 73. 1.04 1.02 1.04 1.62 0.75 1.35 1.33 1.32 1.36 2.30 1.92
DTM 0.58 48.7 0.96 1.01 1.00 1.65 0.70 1.22 1.13 1.22 1.21 1.92 1.51

Table E.3 COARSE GRAIN PI RESULTS MADE AE/30 modelled by NTUA

Uhub = 10 m/s
RFM 0.62 52.8 1.24 1.13 1.01 1.53 0.64 1.04 1.27 1.16 1.06 1.61 1.20
REM 0.97 137. 0.95 1.01 1.01 1.04 0.97 1.00 0.95 1.01 1.01 1.05 1.02
TBMx 0.59 54.1 0.89 1.01 0.99 1.55 0.66 1.15 1.01 1.16 1.14 1.81 1.19
TBMy 0.56 220. 0.99 0.99 1.00 1.56 0.64 1.14 1.15 1.13 1.14 1.77 1.07
TTM 0.61 62.1 1.01 0.98 1.00 1.53 0.61 1.00 1.01 0.98 1.00 1.54 1.25
DTM 0.58 27.3 0.94 1.00 1.00 1.46 0.64 1.09 1.06 1.09 1.09 1.58 0.96
RFM 0.59 60.9 1.19 1.10 1.00 1.51 0.61 1.02 1.25 1.13 1.04 1.57 1.10
REM 0.97 127. 0.95 1.01 1.01 1.06 0.97 1.00 0.95 1.01 1.01 1.06 1.03
TBMx 0.57 73.0 0.85 1.02 0.99 1.54 0.62 1.11 0.93 1.12 1.10 1.74 1.07
TBMy 0.56 325. 1.00 0.98 1.00 1.53 0.62 1.10 1.14 1.09 1.10 1.68 0.98
TTM 0.60 60.8 1.02 0.98 1.00 1.55 0.61 1.02 1.03 1.00 1.02 1.59 1.25
DTM 0.59 42.1 0.96 1.00 1.00 1.43 0.62 1.05 1.05 1.05 1.05 1.48 0.90
RFM 0.58 70.6 1.16 1.09 0.99 1.48 0.59 1.00 1.22 1.10 1.02 1.53 1.05
REM 0.96 125. 0.95 1.01 1.01 1.07 0.96 1.00 0.95 1.02 1.01 1.08 1.03
TBMx 0.56 85.4 0.84 1.02 0.99 1.54 0.61 1.10 0.91 1.11 1.09 1.74 1.04
TBMy 0.56 384. 1.00 0.98 1.00 1.51 0.61 1.08 1.14 1.07 1.08 1.64 0.95
TTM 0.60 64.9 1.03 0.98 1.00 1.56 0.61 1.03 1.04 1.00 1.03 1.62 1.25
DTM 0.59 50.2 0.97 1.00 1.00 1.41 0.62 1.04 1.06 1.04 1.04 1.45 0.87
Uhub = 14 m/s
RFM 0.61 66.2 1.61 1.05 1.06 1.62 0.67 1.12 1.67 1.17 1.17 1.84 1.23
REM 0.96 146. 0.98 1.00 1.03 1.08 0.96 1.01 0.98 1.01 1.04 1.09 1.05
TBMx 0.53 89.3 0.93 0.96 1.01 1.59 0.62 1.17 1.11 1.14 1.18 1.99 1.24
TBMy 0.55 257. 1.02 1.00 0.99 1.68 0.64 1.17 1.19 1.16 1.16 1.99 1.13
TTM 0.61 78.6 1.08 1.01 1.01 1.59 0.63 1.06 1.12 1.07 1.07 1.71 1.25
DTM 0.56 28.7 1.15 0.98 1.01 1.60 0.62 1.09 1.29 1.07 1.10 1.77 1.01
RFM 0.60 67.3 1.61 1.05 1.06 1.63 0.67 1.15 1.70 1.20 1.19 1.92 1.20
REM 0.95 136. 0.99 1.00 1.03 1.09 0.96 1.01 0.99 1.01 1.04 1.11 1.06
TBMx 0.50 106. 0.86 0.92 1.02 1.61 0.60 1.15 1.03 1.08 1.17 2.23 1.14
TBMy 0.53 335. 1.03 0.99 1.00 1.71 0.60 1.14 1.19 1.12 1.14 1.97 1.03
TTM 0.60 77.3 1.08 1.01 1.01 1.62 0.65 1.10 1.18 1.11 1.12 1.82 1.26
DTM 0.56 41.6 1.16 0.97 1.00 1.60 0.60 1.07 1.28 1.05 1.07 1.72 0.93
RFM 0.60 72.5 1.59 1.04 1.05 1.62 0.68 1.16 1.69 1.19 1.18 1.96 1.20
REM 0.94 134. 0.99 1.00 1.03 1.11 0.95 1.01 1.00 1.01 1.04 1.13 1.07
TBMx 0.49 122. 0.83 0.90 1.03 1.61 0.58 1.14 1.00 1.05 1.16 2.34 1.10
TBMy 0.53 389. 1.03 1.00 1.00 1.71 0.60 1.13 1.19 1.13 1.13 1.95 0.99
TTM 0.60 83.6 1.06 1.01 1.00 1.63 0.66 1.12 1.22 1.13 1.14 1.85 1.26
DTM 0.56 48.5 1.16 0.97 1.00 1.60 0.59 1.06 1.28 1.04 1.06 1.70 0.91
Uhub = 18 m/s
RFM 0.59 82.5 1.01 1.09 1.60 0.67 1.17 1.17 1.25 1.88 1.29
REM 0.93 146. 0.99 1.06 1.10 0.94 1.01 1.00 1.07 1.12 1.05
TBMx 0.54 127. 1.01 1.02 1.70 0.62 1.16 1.17 1.16 2.44 1.31
TBMy 0.55 369. 1.00 1.00 1.76 0.64 1.19 1.20 1.18 2.11 1.22
TTM 0.59 96.2 0.99 1.01 1.63 0.64 1.12 1.11 1.14 1.85 1.32
DTM 0.50 26.0 1.02 1.01 2.22 0.53 1.09 1.12 1.10 2.48 1.19
RFM 0.59 82.2 1.02 1.08 1.62 0.68 1.21 1.20 1.28 1.97 1.29
REM 0.92 136. 0.99 1.06 1.12 0.93 1.01 1.01 1.07 1.15 1.06
TBMx 0.53 140. 1.01 1.03 1.78 0.62 1.17 1.18 1.19 3.02 1.30
TBMy 0.52 439. 1.00 1.00 1.84 0.61 1.15 1.16 1.15 2.17 1.14
TTM 0.58 94.3 1.00 1.01 1.61 0.67 1.19 1.18 1.19 1.93 1.32
DTM 0.45 34.7 1.02 1.02 2.38 0.47 1.09 1.11 1.11 2.61 1.08
RFM 0.59 86.8 1.03 1.07 1.63 0.68 1.23 1.23 1.29 2.01 1.29
REM 0.91 135. 0.99 1.06 1.13 0.92 1.02 1.01 1.08 1.18 1.06
TBMx 0.53 154. 1.01 1.04 1.82 0.62 1.18 1.18 1.20 3.25 1.31
TBMy 0.51 499. 1.00 1.00 1.83 0.59 1.14 1.15 1.14 2.16 1.11
TTM 0.59 100. 1.00 1.01 1.59 0.68 1.21 1.21 1.22 1.97 1.32
DTM 0.43 40.4 1.02 1.03 2.40 0.46 1.10 1.11 1.13 2.63 1.04

Table E.4 COARSE GRAIN PI RESULTS NTK500 modelled by NTUA

Uhub = 10 m/s
RFM 0.62 72.6 1.28 1.14 1.02 1.54 0.64 1.06 1.32 1.20 1.08 1.65 1.19
REM 0.97 230. 0.95 1.01 1.00 1.05 0.97 1.00 0.95 1.01 1.00 1.05 1.03
TBMx 0.60 122. 0.91 1.00 0.99 1.49 0.68 1.16 1.05 1.16 1.15 1.76 1.23
TBMy 0.57 382. 0.97 0.99 1.00 1.56 0.65 1.15 1.15 1.14 1.15 1.79 1.05
TTM 0.63 105. 1.04 0.99 1.00 1.52 0.64 1.03 1.06 1.01 1.03 1.57 1.24
DTM 0.59 47.9 0.86 0.99 1.00 1.45 0.65 1.10 1.02 1.10 1.10 1.58 1.00
RFM 0.58 82.5 1.25 1.08 1.02 1.56 0.62 1.07 1.29 1.16 1.09 1.68 1.08
REM 0.96 213. 0.95 1.01 1.00 1.06 0.97 1.00 0.95 1.01 1.01 1.07 1.03
TBMx 0.58 154. 0.88 1.00 0.98 1.46 0.63 1.11 1.00 1.11 1.09 1.66 1.12
TBMy 0.57 554. 0.97 0.99 1.00 1.54 0.63 1.11 1.14 1.10 1.11 1.71 0.96
TTM 0.62 103. 1.04 0.99 1.01 1.53 0.64 1.04 1.08 1.03 1.05 1.63 1.23
DTM 0.59 70.9 0.88 0.99 1.00 1.42 0.63 1.06 1.04 1.06 1.06 1.49 0.92
RFM 0.57 94.6 1.24 1.05 1.02 1.57 0.60 1.07 1.27 1.13 1.08 1.67 1.04
REM 0.96 209. 0.95 1.01 1.00 1.07 0.96 1.00 0.95 1.01 1.01 1.08 1.03
TBMx 0.57 179. 0.88 0.99 0.97 1.44 0.62 1.10 0.99 1.09 1.07 1.61 1.08
TBMy 0.57 650. 0.98 0.99 1.00 1.53 0.63 1.10 1.14 1.09 1.09 1.69 0.93
TTM 0.61 110. 1.05 0.98 1.01 1.54 0.63 1.05 1.09 1.04 1.07 1.65 1.22
DTM 0.59 83.6 0.90 0.99 1.00 1.41 0.62 1.05 1.05 1.05 1.05 1.46 0.90
Uhub = 14 m/s
RFM 0.61 92.1 1.81 1.05 1.10 1.62 0.67 1.15 1.85 1.19 1.23 1.88 1.21
REM 0.96 248. 0.98 1.00 1.03 1.08 0.96 1.00 0.99 1.00 1.03 1.09 1.06
TBMx 0.56 195. 0.98 0.97 1.01 1.56 0.66 1.21 1.15 1.17 1.22 1.88 1.24
TBMy 0.55 437. 1.11 0.98 0.99 1.68 0.66 1.19 1.33 1.18 1.19 2.02 1.14
TTM 0.60 134. 1.14 1.01 1.01 1.56 0.65 1.11 1.20 1.11 1.12 1.75 1.24
DTM 0.56 52.8 1.24 0.99 1.02 1.56 0.63 1.11 1.41 1.10 1.12 1.75 1.05
RFM 0.60 93.0 1.81 1.08 1.08 1.69 0.68 1.19 1.87 1.26 1.25 1.99 1.19
REM 0.95 231. 0.99 1.00 1.03 1.09 0.95 1.00 1.00 1.00 1.03 1.11 1.06
TBMx 0.54 232. 0.85 0.93 1.01 1.45 0.66 1.19 1.01 1.11 1.20 1.75 1.15
TBMy 0.53 569. 1.15 0.94 1.00 1.68 0.61 1.16 1.35 1.12 1.17 1.98 1.04
TTM 0.60 133. 1.13 0.99 1.01 1.57 0.68 1.16 1.23 1.15 1.18 1.82 1.23
DTM 0.56 71.8 1.28 0.98 1.03 1.56 0.61 1.08 1.43 1.08 1.10 1.71 0.96
RFM 0.59 99.4 1.79 1.10 1.07 1.73 0.67 1.22 1.87 1.32 1.25 2.07 1.18
REM 0.94 228. 1.00 1.00 1.03 1.11 0.94 1.00 1.01 1.01 1.03 1.13 1.07
TBMx 0.53 272. 0.80 0.92 1.00 1.38 0.65 1.18 0.94 1.09 1.18 1.65 1.12
TBMy 0.53 665. 1.15 0.93 1.00 1.67 0.60 1.16 1.35 1.11 1.17 1.95 1.02
TTM 0.61 143. 1.12 0.99 1.01 1.57 0.70 1.19 1.25 1.16 1.22 1.85 1.22
DTM 0.56 82.8 1.30 0.98 1.03 1.56 0.60 1.07 1.45 1.07 1.09 1.70 0.94
Uhub = 18 m/s
RFM 0.62 121. 2.11 0.99 1.11 1.61 0.71 1.19 2.18 1.17 1.29 1.91 1.29
REM 0.93 257. 1.07 0.99 1.05 1.11 0.93 1.01 1.07 1.00 1.06 1.14 1.07
TBMx 0.61 280. 1.00 0.96 1.03 1.69 0.70 1.19 1.19 1.14 1.21 1.93 1.28
TBMy 0.57 611. 0.99 0.99 0.99 1.77 0.68 1.19 1.21 1.20 1.19 2.10 1.25
TTM 0.62 177. 1.04 0.96 1.01 1.62 0.70 1.16 1.16 1.11 1.17 1.89 1.32
DTM 0.62 53.0 1.34 1.00 1.01 1.83 0.68 1.11 1.51 1.12 1.12 2.02 1.24
RFM 0.62 121. 2.09 1.00 1.10 1.64 0.72 1.19 2.19 1.19 1.31 1.96 1.29
REM 0.92 241. 1.08 0.98 1.05 1.12 0.92 1.01 1.08 1.00 1.06 1.15 1.08
TBMx 0.61 299. 1.00 0.98 1.04 1.73 0.70 1.21 1.19 1.17 1.24 2.02 1.29
TBMy 0.56 682. 1.03 1.00 0.99 1.88 0.66 1.17 1.23 1.18 1.17 2.18 1.22
TTM 0.61 175. 1.06 0.96 1.02 1.66 0.71 1.20 1.24 1.15 1.22 1.99 1.37
DTM 0.59 58.5 1.49 1.04 1.03 2.04 0.65 1.12 1.71 1.15 1.15 2.26 1.23
RFM 0.62 129. 2.06 1.00 1.09 1.66 0.72 1.19 2.17 1.19 1.30 1.99 1.29
REM 0.90 240. 1.08 0.98 1.05 1.14 0.91 1.01 1.09 1.00 1.06 1.17 1.09
TBMx 0.60 327. 0.99 1.00 1.05 1.76 0.70 1.22 1.18 1.19 1.26 2.09 1.30
TBMy 0.57 754. 1.05 1.01 0.99 1.93 0.65 1.16 1.25 1.17 1.16 2.20 1.21
TTM 0.61 186. 1.09 0.96 1.02 1.70 0.72 1.22 1.32 1.19 1.25 2.05 1.42
DTM 0.58 64.9 1.53 1.07 1.05 2.12 0.63 1.13 1.76 1.17 1.18 2.35 1.22

Table E.5 COARSE GRAIN PI RESULTS DANWIN 180 modelled by TG

Uhub = 10 m/s
RFM 0.60 21 1.34 1.25 1.02 1.45 0.63 1.07 1.38 1.31 1.08 1.56 1.27
REM 0.95 54 1.02 1.00 1.00 1.07 0.95 1.00 1.02 1.00 1.00 1.06 1.06
TBMx 0.53 210 0.72 0.99 1.00 1.50 0.64 1.21 0.90 1.20 1.19 1.80 1.10
TBMy 0.59 59 1.51 0.99 1.00 1.40 0.72 1.24 1.85 1.25 1.22 1.73 1.29
TTM 0.66 25 1.07 1.00 1.00 1.43 0.67 1.05 1.10 1.05 1.05 1.52 1.33
DTM 0.65 10 0.94 0.98 1.01 1.27 0.69 1.10 1.09 1.09 1.09 1.40 1.03
RFM 0.59 29 1.18 1.15 1.03 1.41 0.62 1.08 1.24 1.23 1.07 1.52 1.14
REM 0.95 58 1.02 1.00 1.00 1.08 0.95 1.00 1.02 1.00 1.00 1.08 1.07
TBMx 0.53 369 0.66 0.99 0.99 1.46 0.60 1.14 0.84 1.13 1.13 1.67 0.95
TBMy 0.59 94 1.31 0.99 1.00 1.30 0.67 1.15 1.72 1.16 1.14 1.54 1.11
TTM 0.63 30 1.05 1.00 0.98 1.44 0.65 1.03 1.10 1.03 1.03 1.51 1.28
DTM 0.64 18 0.91 0.98 1.00 1.25 0.67 1.07 1.06 1.06 1.06 1.34 0.96
RFM 0.60 35 1.11 1.11 1.03 1.38 0.62 1.07 1.17 1.18 1.07 1.48 1.09
REM 0.94 59 1.03 1.00 1.00 1.09 0.94 1.00 1.02 1.00 1.00 1.09 1.07
TBMx 0.53 464 0.65 0.98 0.98 1.46 0.59 1.13 0.84 1.12 1.11 1.63 0.92
TBMy 0.59 117 1.27 0.98 1.01 1.25 0.65 1.13 1.70 1.13 1.12 1.48 1.05
TTM 0.61 36 1.01 1.00 0.97 1.44 0.62 1.00 1.07 1.00 1.00 1.46 1.22
DTM 0.64 23 0.90 0.97 1.00 1.24 0.67 1.06 1.05 1.05 1.04 1.31 0.95
Uhub = 14 m/s
RFM 0.59 28 1.83 1.10 1.09 1.49 0.62 1.10 1.84 1.18 1.18 1.65 1.24
REM 0.93 59 1.01 0.98 1.02 1.09 0.93 1.01 1.02 0.99 1.03 1.10 1.11
TBMx 0.57 251 0.86 1.06 0.97 1.67 0.72 1.35 1.00 1.42 1.37 2.23 1.24
TBMy 0.60 88 1.63 1.19 1.00 1.48 0.69 1.14 1.97 1.37 1.15 1.93 1.35
TTM 0.61 33 1.04 1.06 1.02 1.52 0.65 1.09 1.16 1.16 1.10 1.65 1.33
DTM 0.60 10 0.87 0.98 1.01 1.40 0.63 1.08 1.03 1.04 1.09 1.55 1.01
RFM 0.57 34 1.76 1.10 1.08 1.51 0.61 1.12 1.78 1.19 1.19 1.65 1.20
REM 0.92 63 1.01 0.98 1.02 1.11 0.93 1.01 1.03 0.99 1.03 1.11 1.12
TBMx 0.53 364 0.74 1.09 1.00 1.68 0.65 1.27 0.91 1.31 1.30 2.14 1.14
TBMy 0.55 121 1.49 1.10 0.98 1.42 0.63 1.07 1.80 1.28 1.09 1.85 1.25
TTM 0.59 38 0.99 1.03 1.03 1.54 0.64 1.12 1.16 1.15 1.13 1.68 1.35
DTM 0.53 17 0.74 0.96 1.00 1.36 0.53 1.02 0.98 0.97 1.04 1.44 0.91
RFM 0.56 39 1.68 1.09 1.06 1.52 0.59 1.14 1.69 1.20 1.19 1.63 1.17
REM 0.91 65 1.01 0.98 1.03 1.12 0.92 1.01 1.04 0.99 1.03 1.12 1.14
TBMx 0.51 452 0.69 1.10 1.01 1.66 0.61 1.24 0.87 1.26 1.27 2.10 1.12
TBMy 0.52 147 1.44 1.06 0.98 1.38 0.59 1.03 1.71 1.21 1.06 1.80 1.19
TTM 0.59 44 0.96 1.01 1.04 1.56 0.64 1.14 1.17 1.13 1.16 1.70 1.36
DTM 0.53 22 0.71 0.95 1.00 1.33 0.51 0.99 0.94 0.94 1.01 1.36 0.89
Uhub = 18 m/s
RFM 0.66 36 1.94 1.00 1.11 1.40 0.71 1.10 2.01 1.13 1.23 1.55 1.29
REM 0.92 63 1.04 0.97 1.04 1.11 0.93 1.00 1.05 0.97 1.06 1.12 1.09
TBMx 0.63 387 0.98 1.01 1.01 1.50 0.83 1.27 1.25 1.33 1.26 1.42 1.34
TBMy 0.60 179 1.46 0.88 0.95 1.34 0.69 1.12 1.83 1.08 1.13 1.31 1.21
TTM 0.61 44 1.14 0.99 1.00 1.43 0.66 1.09 1.24 1.08 1.11 1.59 1.33
DTM 0.79 12 0.98 0.96 1.01 1.22 0.85 1.10 1.05 1.03 1.08 1.20 1.13
RFM 0.65 42 1.94 1.01 1.10 1.44 0.71 1.11 2.03 1.15 1.24 1.56 1.27
REM 0.91 68 1.04 0.97 1.04 1.13 0.91 1.00 1.05 0.97 1.05 1.12 1.09
TBMx 0.62 475 0.94 1.06 1.01 1.59 0.82 1.23 1.18 1.32 1.28 1.42 1.32
TBMy 0.57 237 1.40 0.83 1.00 1.32 0.65 1.13 1.73 1.01 1.17 1.14 1.23
TTM 0.60 50 1.13 0.98 1.03 1.46 0.66 1.13 1.26 1.09 1.14 1.56 1.33
DTM 0.83 18 0.94 0.95 1.01 1.14 0.86 1.05 0.97 0.99 1.04 1.09 1.05
RFM 0.66 46 1.94 1.03 1.10 1.48 0.71 1.11 2.04 1.18 1.25 1.57 1.27
REM 0.89 71 1.04 0.97 1.03 1.16 0.90 1.00 1.05 0.97 1.05 1.12 1.09
TBMx 0.61 538 0.92 1.08 1.01 1.66 0.81 1.20 1.14 1.32 1.29 1.42 1.31
TBMy 0.57 275 1.38 0.81 1.02 1.31 0.64 1.13 1.69 0.99 1.19 1.07 1.24
TTM 0.59 56 1.12 0.97 1.04 1.48 0.65 1.16 1.27 1.10 1.15 1.50 1.33
DTM 0.85 22 0.95 0.96 1.02 1.12 0.86 1.04 0.97 0.99 1.03 1.05 1.03

Table E.6 COARSE GRAIN PI RESULTS Nassudden II modelled by TG

Uhub = 10 m/s
RFM 0.59 1141 0.98 1.22 1.00 1.49 0.61 1.04 1.00 1.26 1.05 1.57 1.15
REM 0.95 1930 1.00 1.01 1.00 1.07 0.95 1.00 1.00 1.00 0.99 1.07 1.03
TBMx 0.70 33837 0.87 1.17 0.99 1.41 0.68 0.95 0.86 1.15 0.94 1.33 0.95
TBMy 0.57 38741 1.19 1.91 1.03 1.41 0.60 1.03 1.15 1.94 1.04 1.48 1.00
TTM 0.64 1091 0.99 1.06 1.00 1.26 0.70 1.08 1.10 1.13 1.08 1.32 1.10
DTM 0.55 287 0.88 1.02 0.98 1.49 0.59 1.06 0.85 1.05 1.04 1.55 0.93
RFM 0.59 1593 1.01 1.20 1.00 1.49 0.60 1.01 0.97 1.22 1.04 1.54 1.04
REM 0.94 2229 1.00 1.01 1.00 1.08 0.94 1.00 1.01 1.00 0.99 1.08 1.03
TBMx 0.66 43535 0.88 1.13 0.99 1.49 0.63 0.95 0.85 1.11 0.94 1.37 0.95
TBMy 0.56 52633 1.10 1.68 1.01 1.38 0.59 1.03 1.02 1.74 1.00 1.48 0.97
TTM 0.68 1359 0.99 1.05 1.00 1.19 0.74 1.05 1.07 1.10 1.06 1.23 1.07
DTM 0.55 514 0.88 1.02 0.98 1.49 0.57 1.03 0.78 1.02 1.00 1.49 0.86
RFM 0.59 1936 1.02 1.19 1.00 1.50 0.60 1.00 0.95 1.18 1.03 1.52 1.03
REM 0.93 2357 1.00 1.01 1.00 1.10 0.94 1.00 1.01 1.01 1.00 1.10 1.03
TBMx 0.64 49235 0.90 1.12 0.99 1.53 0.61 0.95 0.84 1.09 0.94 1.40 0.95
TBMy 0.55 62702 1.05 1.58 1.00 1.38 0.58 1.03 0.94 1.65 0.98 1.49 0.95
TTM 0.72 1510 1.00 1.04 1.00 1.16 0.78 1.04 1.05 1.08 1.04 1.20 1.06
DTM 0.55 644 0.89 1.02 0.99 1.49 0.56 1.02 0.76 1.01 0.99 1.48 0.86
Uhub = 14 m/s
RFM 0.62 1580 1.12 1.18 1.02 1.47 0.65 1.05 1.16 1.23 1.07 1.53 1.15
REM 0.92 2058 1.01 1.02 1.00 1.11 0.93 1.02 1.02 1.04 1.01 1.13 1.05
TBMx 0.67 40486 0.96 1.20 0.96 1.40 0.65 0.96 0.85 1.18 0.93 1.34 0.91
TBMy 0.66 52020 1.70 1.68 1.11 1.39 0.66 0.97 1.61 1.67 1.10 1.33 0.94
TTM 0.70 1368 1.04 1.08 1.03 1.19 0.74 1.05 1.07 1.11 1.07 1.19 1.05
DTM 0.55 390 1.08 1.01 1.02 1.50 0.62 1.13 1.22 1.12 1.14 1.69 1.02
RFM 0.65 2154 1.14 1.15 1.00 1.40 0.67 1.03 1.15 1.19 1.05 1.45 1.06
REM 0.91 2406 1.01 1.03 1.00 1.13 0.92 1.02 1.02 1.04 1.02 1.15 1.05
TBMx 0.66 50574 1.08 1.17 0.97 1.43 0.63 0.95 0.93 1.14 0.94 1.38 0.93
TBMy 0.66 65028 1.62 1.58 1.14 1.39 0.65 0.96 1.54 1.55 1.11 1.34 0.94
TTM 0.75 1585 1.05 1.06 1.02 1.16 0.79 1.04 1.07 1.09 1.05 1.16 1.04
DTM 0.56 636 1.11 1.02 1.04 1.49 0.61 1.10 1.27 1.10 1.13 1.65 0.93
RFM 0.66 2570 1.15 1.13 0.99 1.38 0.67 1.02 1.14 1.16 1.03 1.41 1.03
REM 0.89 2582 1.01 1.03 1.00 1.15 0.90 1.02 1.02 1.04 1.02 1.18 1.04
TBMx 0.65 57320 1.14 1.16 0.98 1.44 0.63 0.94 0.98 1.11 0.94 1.40 0.94
TBMy 0.66 73405 1.58 1.53 1.14 1.38 0.64 0.95 1.51 1.49 1.11 1.35 0.94
TTM 0.78 1702 1.06 1.06 1.02 1.15 0.81 1.03 1.07 1.08 1.05 1.15 1.04
DTM 0.56 792 1.09 1.02 1.04 1.48 0.59 1.09 1.27 1.09 1.11 1.63 0.91
Uhub = 18 m/s
RFM 0.54 2118 1.20 0.99 1.47 0.55 1.03 1.22 1.04 1.54 1.05
REM 0.88 2196 1.02 0.99 1.15 0.89 1.02 1.04 1.01 1.17 1.05
TBMx 0.72 328026 1.09 0.95 1.29 0.70 0.96 1.05 0.91 1.24 0.96
TBMy 0.69 356428 1.79 1.05 1.33 0.64 0.96 1.73 0.99 1.28 0.93
TTM 0.74 1499 1.09 1.00 1.13 0.77 1.01 1.10 1.02 1.14 1.02
DTM 0.56 497 1.01 1.00 1.48 0.63 1.14 1.18 1.15 1.76 1.11
RFM 0.51 2944 1.18 0.99 1.38 0.53 0.98 1.19 1.04 1.46 0.94
REM 0.87 2583 1.02 0.99 1.17 0.88 1.03 1.05 1.02 1.19 1.05
TBMx 0.71 403381 1.10 0.95 1.30 0.68 0.98 1.05 0.94 1.28 0.96
TBMy 0.67 463429 1.64 1.04 1.34 0.62 0.94 1.58 0.99 1.26 0.88
TTM 0.78 1723 1.07 1.00 1.12 0.81 1.01 1.08 1.02 1.14 1.02
DTM 0.56 681 0.99 1.01 1.52 0.63 1.16 1.21 1.18 1.86 1.08
RFM 0.50 3601 1.18 0.99 1.34 0.52 0.96 1.17 1.06 1.40 0.90
REM 0.85 2778 1.02 0.99 1.19 0.86 1.04 1.06 1.03 1.22 1.06
TBMx 0.70 456212 1.10 0.95 1.31 0.67 0.99 1.05 0.95 1.30 0.97
TBMy 0.65 548896 1.55 1.03 1.33 0.59 0.92 1.50 0.97 1.21 0.83
TTM 0.81 1843 1.06 1.00 1.11 0.83 1.00 1.07 1.01 1.14 1.02
DTM 0.56 800 0.99 1.01 1.55 0.63 1.15 1.20 1.18 1.90 1.07

Table E.7 COARSE GRAIN PI RESULTS MAGLARP/WTS3 modelled by TG

Uhub = 10 m/s
RFM 0.56 693 1.00 1.00 1.00 1.51 0.60 1.07 1.07 1.08 1.61 1.02
REM 0.98 2135 1.00 1.00 0.99 1.03 0.98 0.99 1.00 0.99 1.02 1.01
TBMx 0.51 9353 1.00 1.00 1.00 1.47 0.62 1.17 1.17 1.17 1.76 0.92
TBMy 0.75 1758 1.00 1.03 0.99 1.35 0.84 1.21 1.23 1.20 1.68 1.10
TTM 0.76 105 1.00 1.03 1.01 1.34 0.77 1.03 1.07 1.04 1.41 1.20
DTM 0.54 217 1.00 1.00 1.00 1.50 0.59 1.10 1.09 1.10 1.62 0.85
RFM 0.55 1183 1.00 1.00 1.00 1.50 0.59 1.07 1.06 1.07 1.58 0.93
REM 0.98 2394 1.00 1.00 0.99 1.03 0.97 1.00 1.00 0.99 1.03 1.01
TBMx 0.51 16985 1.00 1.00 1.00 1.44 0.60 1.14 1.13 1.13 1.68 0.87
TBMy 0.66 2822 1.00 1.01 0.99 1.40 0.77 1.21 1.22 1.19 1.71 1.06
TTM 0.72 120 1.00 1.02 1.01 1.37 0.75 1.07 1.10 1.07 1.50 1.20
DTM 0.54 403 1.00 1.00 1.00 1.50 0.59 1.09 1.09 1.09 1.59 0.83
RFM 0.55 1471 1.00 1.00 1.00 1.49 0.58 1.06 1.06 1.06 1.57 0.93
REM 0.97 2499 1.00 1.00 0.99 1.04 0.97 1.00 1.00 0.99 1.04 1.01
TBMx 0.51 21271 1.00 1.00 1.00 1.42 0.60 1.13 1.11 1.12 1.65 0.86
TBMy 0.65 3461 1.00 1.01 0.99 1.40 0.76 1.22 1.22 1.20 1.71 1.06
TTM 0.69 134 1.00 1.02 1.01 1.38 0.73 1.08 1.10 1.08 1.53 1.20
DTM 0.54 510 1.00 1.00 1.00 1.50 0.59 1.09 1.09 1.09 1.58 0.82
Uhub = 14 m/s
RFM 0.81 1150 0.97 0.99 1.27 0.87 1.08 1.06 1.08 1.42 1.01
REM 0.97 2354 1.00 0.99 1.05 0.97 0.99 1.00 0.98 1.04 1.01
TBMx 0.63 13122 1.00 1.00 1.46 0.79 1.32 1.31 1.32 1.98 1.15
TBMy 0.65 2791 1.04 0.98 1.34 0.78 1.27 1.28 1.25 1.70 1.13
TTM 1.02 320 0.90 0.99 1.02 1.01 0.99 0.91 0.99 1.03 0.95
DTM 0.55 293 1.02 1.00 1.44 0.61 1.14 1.16 1.14 1.67 0.91
RFM 0.79 1775 0.98 0.99 1.24 0.84 1.07 1.08 1.07 1.39 0.97
REM 0.96 2666 1.00 0.99 1.07 0.96 0.99 0.99 0.98 1.05 1.01
TBMx 0.66 21272 1.00 1.00 1.44 0.77 1.27 1.26 1.27 1.85 1.07
TBMy 0.62 4234 1.04 0.98 1.39 0.74 1.22 1.23 1.20 1.65 1.07
TTM 0.94 378 0.89 0.99 1.07 0.95 1.00 0.93 1.02 1.13 0.96
DTM 0.55 535 1.01 1.00 1.42 0.60 1.13 1.15 1.13 1.63 0.89
RFM 0.79 2164 0.98 0.99 1.23 0.82 1.06 1.07 1.07 1.37 0.97
REM 0.95 2806 1.00 0.99 1.08 0.95 0.99 0.99 0.98 1.07 1.01
TBMx 0.67 25892 1.00 1.00 1.44 0.77 1.26 1.25 1.26 1.84 1.06
TBMy 0.62 5156 1.04 0.98 1.42 0.73 1.21 1.22 1.18 1.63 1.03
TTM 0.90 424 0.89 0.99 1.11 0.91 1.00 0.93 1.03 1.19 0.97
DTM 0.55 667 1.01 1.00 1.41 0.61 1.14 1.16 1.14 1.63 0.90
Uhub = 18 m/s
RFM 0.76 1536 0.99 1.00 1.27 0.80 1.08 1.07 1.08 1.38 0.98
REM 0.96 2550 1.00 0.99 1.07 0.96 0.99 0.99 0.98 1.05 1.01
TBMx 0.52 19474 1.01 1.00 1.44 0.71 1.36 1.36 1.36 1.98 1.14
TBMy 0.61 3441 1.03 0.98 1.52 0.75 1.30 1.31 1.28 2.10 1.21
TTM 1.03 407 0.92 1.00 1.07 1.03 1.00 0.93 1.00 1.06 0.98
DTM 0.48 289 1.00 0.99 1.61 0.64 1.28 1.28 1.30 2.11 1.12
RFM 0.66 2427 1.00 1.00 1.22 0.72 1.07 1.07 1.07 1.30 0.89
REM 0.95 2898 1.00 0.99 1.09 0.95 0.99 0.99 0.98 1.07 1.01
TBMx 0.52 31493 1.01 1.00 1.33 0.66 1.27 1.27 1.26 1.76 1.06
TBMy 0.59 5001 1.02 0.98 1.51 0.77 1.34 1.34 1.32 2.15 1.21
TTM 0.96 485 0.93 1.00 1.13 0.96 1.02 0.95 1.01 1.15 1.01
DTM 0.47 456 1.00 1.00 1.68 0.65 1.24 1.24 1.27 2.15 1.12
RFM 0.65 2949 1.01 1.00 1.22 0.71 1.07 1.07 1.07 1.28 0.89
REM 0.94 3061 1.00 0.99 1.10 0.93 0.99 0.99 0.98 1.09 1.01
TBMx 0.53 38366 1.01 1.00 1.31 0.65 1.24 1.24 1.24 1.70 1.05
TBMy 0.59 5969 1.01 0.97 1.49 0.78 1.37 1.35 1.33 2.19 1.22
TTM 0.91 545 0.94 1.00 1.14 0.92 1.02 0.96 1.01 1.19 1.02
DTM 0.45 558 0.99 1.00 1.70 0.66 1.22 1.22 1.24 2.17 1.13

Table E.8 FINE GRAIN PI RESULTS NTK500 modelled by RISO

Uhub = 10 m/s
Leq-4 No 1 2 3 4 5 6
RFM 23.7 1.01 1.00 1.14 1.02 1.00
REM 170.1 1.00 1.00 1.00 1.00 1.00
TBMx 189.4 1.00 1.00 1.20 0.95 0.99
TBMy 63.0 1.00 1.00 1.20 0.97 1.00
TTM 31.8 1.00 1.00 1.13 1.01 0.99
DTM 85.6 1.00 1.00 1.03 1.01 1.00
Leq-8 No 1 2 3 4 5 6
RFM 30.6 1.01 1.00 1.11 1.02 1.00
REM 185.5 1.00 1.00 1.00 1.00 1.00
TBMx 260.3 1.01 1.00 1.15 0.95 1.00
TBMy 81.6 1.00 1.00 1.16 1.00 0.98
TTM 39.5 1.01 1.00 1.13 1.00 1.01
DTM 96.8 1.00 1.00 1.04 1.01 1.00
Leq-12 No 1 2 3 4 5 6
RFM 36.1 1.00 1.00 1.09 1.01 0.99
REM 191.2 1.00 1.00 1.00 1.00 1.00
TBMx 310.5 1.01 1.00 1.13 0.95 1.00
TBMy 96.3 1.00 1.00 1.13 1.01 0.98
TTM 45.8 1.00 1.00 1.13 1.00 1.02
DTM 103.2 1.00 1.00 1.04 1.00 1.00
Uhub = 14 m/s
Leq-4 No 1 2 3 4 5 6
RFM 39.9 1.01 1.00 1.09 0.99 0.98
REM 179.0 1.00 1.00 1.00 1.00 1.00
TBMx 357.3 1.00 1.00 1.18 0.93 0.99
TBMy 135.8 1.00 1.00 1.20 0.95 0.99
TTM 54.4 0.98 1.00 1.09 0.98 0.96
DTM 104.7 1.00 1.00 1.03 0.99 0.99
Leq-8 No 1 2 3 4 5 6
RFM 49.8 1.02 1.00 1.07 0.99 0.98
REM 195.8 1.00 1.00 1.00 1.00 1.00
TBMx 452.4 1.00 1.00 1.19 0.93 0.99
TBMy 173.0 1.00 1.00 1.20 0.95 0.99
TTM 66.3 0.97 1.00 1.07 0.97 0.95
DTM 122.1 1.01 1.00 1.03 0.99 1.00
Leq-12 No 1 2 3 4 5 6
RFM 57.2 1.02 1.00 1.06 0.98 0.98
REM 202.5 1.00 1.00 1.00 1.00 1.00
TBMx 515.7 1.00 1.00 1.20 0.93 1.00
TBMy 196.3 1.00 1.00 1.20 0.95 0.99
TTM 76.1 0.97 1.00 1.06 0.96 0.94
DTM 132.8 1.01 1.00 1.03 0.99 1.00
Uhub = 18 m/s
Leq-4 No 1 2 3 4 5 6
RFM 67.4 1.02 1.00 1.07 0.97 0.97
REM 178.7 1.00 1.00 1.00 1.00 1.00
TBMx 788.6 1.04 1.00 1.11 0.92 0.98
TBMy 202.7 1.04 1.00 1.11 0.95 0.98
TTM 92.6 0.99 1.00 1.08 0.96 0.96
DTM 138.4 1.02 1.00 1.04 0.98 0.99
Leq-8 No 1 2 3 4 5 6
RFM 78.5 1.02 1.00 1.07 0.97 0.96
REM 195.2 1.00 1.00 1.00 1.00 1.00
TBMx 939.9 1.05 1.00 1.12 0.92 0.98
TBMy 252.1 1.05 1.00 1.12 0.97 0.99
TTM 114.1 0.97 1.00 1.04 0.95 0.95
DTM 167.4 1.02 1.00 1.03 0.98 0.98
Leq-12 No 1 2 3 4 5 6
RFM 86.4 1.02 1.00 1.06 0.97 0.96
REM 201.6 1.00 1.00 1.00 1.00 1.00
TBMx 1041. 1.05 1.00 1.13 0.92 0.98
TBMy 283.6 1.05 1.00 1.13 0.97 0.99
TTM 132.8 0.97 1.00 1.02 0.94 0.94
DTM 186.9 1.03 1.00 1.03 0.98 0.98

Table E.9 FINE GRAIN PI RESULTS V39 modelled by RISO

Uhub = 10 m/s
Leq-4 1 2 3 4 5 6
RFM 72.5 1.02 1.01 1.10 1.02 1.00
REM 162.1 1.00 1.00 1.03 1.00 1.01
TBMx 540.6 1.01 1.01 1.11 0.96 1.02
TBMy 319.0 1.01 1.01 1.18 0.96 0.95
TTM 74.9 1.01 1.01 1.14 1.03 0.98
DTM 48.9 1.00 1.00 1.03 0.97 1.03
Leq-8 1 2 3 4 5 6
RFM 83.4 1.02 1.01 1.07 1.00 1.02
REM 151.0 1.00 1.00 1.03 1.00 1.01
TBMx 711.8 1.01 1.01 1.04 0.96 1.05
TBMy 328.0 1.01 1.01 1.16 0.96 0.93
TTM 78.5 1.01 1.02 1.13 1.03 1.00
DTM 71.9 1.00 1.00 0.98 0.97 1.05
Leq-12 1 2 3 4 5 6
RFM 94.0 1.02 1.01 1.06 1.00 1.03
REM 149.1 1.00 1.00 1.04 1.00 1.02
TBMx 833.0 1.01 1.01 1.02 0.96 1.05
TBMy 359.0 1.01 1.01 1.14 0.97 0.92
TTM 86.2 1.01 1.02 1.12 1.03 1.03
DTM 84.8 1.00 1.00 0.96 0.98 1.06
Uhub = 14 m/s
Leq-4 1 2 3 4 5 6
RFM 105.3 1.02 1.01 1.09 1.03 1.02
REM 187.0 1.00 1.00 1.06 1.01 1.01
TBMx 785.7 1.01 1.01 1.14 0.96 1.04
TBMy 538.8 1.00 1.00 1.22 0.96 1.05
TTM 114.8 1.01 1.02 1.12 1.02 0.95
DTM 78.5 1.00 1.00 1.08 0.97 1.07
Leq-8 1 2 3 4 5 6
RFM 114.1 1.02 1.01 1.06 1.01 1.07
REM 177.8 1.00 1.00 1.07 1.01 1.02
TBMx 906.8 1.01 1.01 1.08 0.95 1.04
TBMy 542.2 1.01 1.00 1.21 0.96 1.11
TTM 121.5 1.01 1.02 1.11 1.01 0.94
DTM 100.0 1.00 1.00 1.07 0.95 1.13
Leq-12 1 2 3 4 5 6
RFM 126.3 1.02 1.01 1.04 1.01 1.10
REM 179.1 1.00 1.00 1.07 1.01 1.03
TBMx 1030. 1.01 1.01 1.06 0.95 1.04
TBMy 583.7 1.01 1.00 1.19 0.97 1.15
TTM 132.5 1.01 1.02 1.10 1.00 0.93
DTM 113.9 1.00 1.00 1.07 0.95 1.16
Uhub = 18 m/s
Leq-4 1 2 3 4 5 6
RFM 158.4 1.01 1.00 1.10 1.02 0.97
REM 209.3 1.00 1.00 1.08 1.02 1.00
TBMx 1278. 1.01 1.00 1.17 0.94 1.01
TBMy 930.6 1.00 1.00 1.20 0.96 1.01
TTM 180.4 1.00 1.00 1.15 1.04 0.95
DTM 144.3 1.00 1.00 1.14 0.98 1.03
Leq-8 1 2 3 4 5 6
RFM 175.4 1.01 1.00 1.07 1.01 0.98
REM 202.5 1.00 1.00 1.08 1.02 0.99
TBMx 1459. 1.01 1.00 1.12 0.95 1.02
TBMy 941.0 1.00 1.00 1.20 0.96 1.03
TTM 193.6 1.00 1.00 1.14 1.03 0.93
DTM 165.9 1.00 1.00 1.12 0.98 1.06
Leq-12 1 2 3 4 5 6
RFM 191.7 1.01 1.00 1.06 1.01 1.00
REM 209.2 1.00 1.00 1.10 1.02 0.98
TBMx 1636. 1.01 1.00 1.10 0.95 1.02
TBMy 996.4 1.00 1.00 1.20 0.96 1.06
TTM 214.7 1.00 1.00 1.13 1.03 0.91
DTM 183.0 1.00 1.00 1.11 0.98 1.07

PART 1
SUB C
EXTREME WIND CLIMATE EVENTS
D. Winkelaar (ed.) Netherlands Energy Research Foundation ECN

J.W.M. Dekker Netherlands Energy Research Foundation ECN
P. Chaviaropoulos CRES
I. Carlén Teknikgruppen AB
G.C. Larsen Risø National Laboratory
Extreme Wind Climate Events
1Sub C.2 European Wind Turbine Standards II

CONTENTS OF PART 1Sub C
LIST OF SYMBOLS 5
1 INTRODUCTION 9
2 THE DISTRIBUTION OF THE EXTREME WIND SPEED 11

2.1 The cdf of the extreme 10-minute average wind speed :::::: : : : : : : : : : : : : 11
2.1.1 Exact extreme distribution based on Weibull parent : : : : : : : : : : : : : : : 11
2.1.2 The FT type-I distribution based on a Weibull parent : : : : : : : : : : : : : : 11
2.1.3 The Davenport correction : : : : : : : : : : : : : : : : : : : : : : : : : : : : 12
2.2 The number of independent events per year : : : : : : : : : : : : : : : : : : : : : : : 12
2.3 Evaluation of Vref in the IEC 1400-1 standard : : : : : : : : : : : : : : : : : : : : : : 13
2.4 The pdf of the 3-second average wind speed : : : : : : : : : : : : : : : : : : : : : : : 15
2.5 The cdf of the extreme 3-sec average wind speed : : : : : : : : : : : : : : : : : : : : 17
2.6 Evaluation of Ve1 and Ve50 in the IEC 1400-1 standard : : : : : : : : : : : : : : : : : : 18
2.7 Conclusions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 19
3 THE DISTRIBUTION OF EXTREME GUSTS 21

3.1 :::::::::
The pdf of single point gusts : : : : : : : : : : : : : : : : : : : : : : 21
3.1.1 The correlation function : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
3.1.2 The number of independent events : : : : : : : : : : : : : : : : : : : : : : : : 23
3.2 Results and comparison with IEC : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
4 THE DISTRIBUTION OF EXTREME WIND DIRECTION CHANGE 29

4.1 The pdf of wind direction :::::::::::::::::::::::::::::::: 29
4.2 The pdf of wind direction change : : : : : : : : : : : : : : : : : : : : : : : : : : : : 33
4.3 Results and comparison with the IEC 1400-1 standard : : : : : : : : : : : : : : : : : : 33
5 OTHER EXTREME ATMOSPHERIC EVENTS 39

5.1 Downbursts ::::::::::::::::::::::::::::::::::::::: 39
6 SUMMARY, CONCLUSIONS AND RECOMMENDATIONS 43
A NOTATION AND BACKGROUND 45
European Wind Turbine Standards II 1Sub C.3

A.1 Symbols and Definitions ::::::::::::::::::::::::::::::: : : 45

A.1.1 Distribution and density functions : : : : : : : : : : : : : : : : : : : : : : : : 46
A.1.2 Joint statistics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 46
A.1.3 Order Statistics from Independent and Identically Distributed Samples : : : : : 47
A.1.4 Return period : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 48
A.1.5 Characteristic largest value : : : : : : : : : : : : : : : : : : : : : : : : : : : 48
A.2 The Normal Distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 49
A.3 The Weibull Distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 49
A.4 The Gumbel Distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 50
B DERIVATION OF GUMBEL DISTRIBUTION 51

B.1 Derivation Based on Order Statistics ::::::::::: : : : : : : : : : : : : : : : : 51
B.1.1 Order statistics and asymptotic distributions : : : : : : : : : : : : : : : : : : : 51
B.1.2 The Gumbel distribution for a Weibull parent : : : : : : : : : : : : : : : : : : 52
B.2 Derivation Based on Exceedance Statistics : : : : : : : : : : : : : : : : : : : : : : : : 53
B.2.1 Exceedance statistics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 53
B.2.2 Frequency of maxima : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 55
REFERENCES 58

LIST OF SYMBOLS
c1 Constant defined in Eq. (2.13).
c2 Constant defined in Eq. (2.14).
E[ ] Denotes the ensemble averaged value of the quantity in square brackets.
f Frequency (Hz).
f0 Lower cut-off frequency (Hz).
f (x) (First-order) probability density function (pdf).
fx(x) Marginal probability density function.
f (x; y) Second-order (joint) probability density function.
f (x j y ) Conditional probability density function for the distribution of x when y is specified.
iu Longitudinal signal to noise ratio, = Uu , or the reciprocal of the longitudinal turbulence
intensity.
iv Transversal signal to noise ratio = Vv .
Jn Jacobean.
k Shape parameter of the Weibull distribution.
n Number of indpendent events.
n+a (T ) Number of crossings of the level y = a with positive slope in time T for a single member
function of the random process y (t).
qT Ratio of the standard deviation T to .
r Polar co-ordinate ; probability of exceedance.
t Time.
u Longitudinal (x) component of the wind speed.
uT Longitudinal component of the wind speed averaged over time interval T .
v Transversal (y ) component of the wind speed.

w Vertical (z ) component of the wind speed.
y(t) Random process composed of an ensemble of member functions each of which is a different
function of t.
z Vertical co-ordinate.
z0 Surface roughness length (m).
C Scale parameter of the Weibull distribution.
F (x) (First-order) probability distribution function (cdf).
Fx (x) Marginal probability distribution function.
F (x; y) Joint probability distribution function (cdf) for the random variable x.
I Identity matrix.
Iave Annual average turbulence intensity.
L Integral length scale.
Na+ (T ) Average number of crossings of the level y = a with positive slope in time T for a random
process y (t).
Pr Abbreviation for “probability”.
Suu Auto-spectral density function of u.
T Time period ; Averaging time ; Record length.
Tp Sample period (generally one year, but 20 and 50 years are also used).
Tr Return period.
U Ten-minute mean value of the longitudinal component of the wind speed.
V Ten-minute mean value of the transversal component of the wind speed.
Vave Annual average wind speed at hub height.
Vcg Maximum coherent gust over the whole rotor swept area.
Ve1 Extreme 3-sec. average wind speed at hub height with a 2% probability of being exceeded
within a year.

Ve50 Extreme 3-sec. average wind speed at hub height with a 2% probability of being exceeded
within 50 years.
Vgust1 Extreme 3-sec. average gust with a specified starting value and a 2% probability of being
exceeded within a year.
Vgust50 Extreme 3-sec. average gust with a specified starting value and a 2% probability of being
exceeded in 50 years.
Vr Annual extreme 10-min. average wind speed at hub height with a return period of Tr years.
Vref Annual extreme 10-min. average wind speed at hub height with a return period of 50 years.
Reciprocal value of the dispersion factor of the Gumbel distribution.
Modal value of the Gumbel distribution.
Euler’s constant = 0.5772156649... ;
Scale parameter of the Gumbel distribution.
Polar co-ordinate.
Location parameter of Gumbel distribution.
Mean of the normal distribution.
a+ Average frequency of positive slope crossings of the level y = a for a random process y(t).
0+ Average frequency of positive slope crossings of the zero level.
T Average frequency of statistically independent events.
xy Correlation coefficient (normalized covariance) for the random variables x and y .
u Standard deviation of longitudinal wind speed at hub height.
v Standard deviation of transversal wind speed at hub height.
T Part of the standard deviation u which remains after having applied a low pass filter with
averaging time T .
Time lag (time difference).
; Gamma function.


1 INTRODUCTION
Critical loads result in responses that exceed the limits beyond which structural damage or irreversible
change may be expected. Thus the designer is interested in the number of times, or the probability, that
large loads and resulting large responses may occur over the life time of the system.
The IEC 1400-1 standard divides the wind regime for load and safety considerations into normal wind
conditions which will occur frequently during normal operation of the wind turbine, and extreme wind
conditions which are defined as having a 1 year or 50 year recurrence period. The extreme wind conditions
are used to determine critical design loads which the turbine must withstand during its lifetime. These
extreme conditions include peak wind speeds due to storms and rapid changes in wind speed and direction.
The purpose of this investigation is to review the models for extremes of the IEC 1400-1 standard. To
that end it was decided not to analyse wind speed measurements as for example is done by a revision
working group (TC88/WG7) [2]. Although measurements should ultimately be decisive, it is obviously
dangerous to infer statistics of extremes from a rather limited set of data. Besides if we would perform
exactly the same sort of investigation as the working group TC88/WG7 did, then it would be difficult to
validate or invalidate the results of that group.
Inspired by a report by Bergström [1] and papers by Cook [6] and Harris [14] it was decided to adopt a
completely different and more theoretical approach, by using Extreme Value Theory. Assuming that the
statistics of the wind climate is given (i.e. the Weibull distribution and the turbulence model) it is possible
to derive the probability of occurrence of extreme wind climate events. An additional advantage of this
mathematical/statistical approach is that the models for extremes and the models for normal turbulence
are logically connected, making the results to some extent more “physical.”
Section 2 describes the derivation of the probability density function (pdf) and cumulative distribution
function (cdf) of the 10-minute and extreme 3-second average wind speed. These functions are used to
compute the survival wind speeds as function of the Weibull shape and scale parameter and the values
obtained are compared with the values given by the IEC 1400-1 standard in the so-called Extreme Wind
speed Model (EWM).
Section 3 deals with the problem of deriving a pdf and cdf of the extreme 3-second average gust.
The resulting distribution function is used to compute extreme gust magnitudes with a probability of
occurrence as specified by the IEC standard and the values found are compared with the values given
by the IEC in the so-called Extreme Operational Gust model (EOG) and Extreme Coherent Gust model
(ECG).
Section 4 is concerned with the derivation of a pdf and cdf of extreme wind direction changes. This
problem is not quite solved but an approximate distribution function is given which seems to yield
reasonable results.
Section 5 deals with the question whether there are extreme wind events which are not described by the
IEC standard but should be included. One rare event is described, a so-called downburst, which may
induce extreme loads on a wind turbine.
Section 6 gives a summary, some conclusions and recommendations.
Finally, the Appendices A and B give some background on statistics and extreme value theory, mostly
taken from references [3, 8, 14, 17–19, 21].


2 THE DISTRIBUTION OF THE EXTREME WIND SPEED
Two types of statistics are of interest; those relating to the total population of wind speeds and second the
properties of the extremes. In the following it is assumed that the statistics of the wind climate at the site
of the structure is established and what will be discussed is an attempt to find a relation between these
statistics and the properties of extremes. In statistical jargon: we assume that the parent distribution is
somehow known and we want to find the extreme value distributions.
Some of the difficulties in finding these limit distributions are addressed. Using these limit distributions
we determine extreme wind speeds defined in the Extreme Wind Speed Model (EWM) of the IEC 1400-1
standard and compare them with the values actually given by that standard. Some conclusions are drawn
and recommendations given.
2.1 The cdf of the extreme 10-minute average wind speed
2.1.1 Exact extreme distribution based on Weibull parent
Taking the Weibull cdf (A.23) as the parent, the distribution of the yearly extreme 10-minute averages is
given by the distribution of the n’th order statistic:
" ( k )#n
Exact : n V 1
F (V ) = 1 ; exp ; V ; 1 + k : (2.1)
ave
If Vr is the yearly extreme 10-minute average wind speed with a return period of Tr years then the
following relation holds:
1 !

ln (1 ; 1 =T r )
1 ; F n (Vr ) = Tr F (Vr ) = exp n : (2.2)
Combining (2.1) and (2.2) we find after some simple algebra:

1
Vr = 1
ln (1 ; 1 =Tr)
k
Vave ; (1 + 1=k) ; ln 1 ; exp n : (2.3)
2.1.2 The FT type-I distribution based on a Weibull parent
The Fisher-Tippett type-I or Gumbel distribution of the yearly extreme 10-minute average wind speed
with the Weibull distribution as a parent, i.e. the asymptotic version of (2.1), is given by :
Gumbel : Fn (V ) = exp [; exp f; (V ; )g] ; (2.4)
where
= Ck (ln n)1; k ;
1
(2.5)
1
= C (ln n) k ; (2.6)
C = ; (1V+ave1=k) : (2.7)

For a return period of Tr years we find:

1 ! Fn (Vr) = 1 ; 1=Tr :
1 ; Fn (Vr ) = Tr (2.8)
Combining (2.4) and (2.8) yields:

1
Vr = (ln n) k ;1 [k ln n ; ln f; ln (1 ; 1=T )g] :
Vave k; (1 + 1=k) r (2.9)
2.1.3 The Davenport correction
Using exceedance statistics, as proposed by Davenport [8], slightly different expressions are found for
and , see section B.2 :
;1 ;
Dav = 1 ; kk ln

n (2.10)
and 8 h k ;1 i
9
;1
>
< ln k; k + 1 (ln n) k > =
Dav = >1 + k ln n ; (k ; 1) >
: (2.11)
: ;
We then find for the ratio of Vr to Vave :

1
Vr = (ln n) k ;1 [c c k ln n ; ln f; ln (1 ; 1=T )g] ;
Vave c1 k; (1 + 1=k) 1 2 r (2.12)
where
k ; 1
c1 = 1 ; k ln n ; (2.13)
and 8 h 9
;1 k;1 i
>
< ln k; k + 1 (ln n) k > =
c2 = >1 + k ln n ; (k ; 1) >
: (2.14)
: ;
2.2 The number of independent events per year
In all three expressions (2.3), (2.9) and (2.12) the ratio of Vr to Vave is a function of the Weibull shape
parameter k , the number of independent events n and the return period Tr .
Extreme value theory only deals with extremes of samples of n independent and identically distributed
observations which originate from a parent population with cumulative distribution function (cdf) F (x).
Then the largest value has a probability distribution given by F n (x). Thus in the case of 10-minute
average wind speeds, if the values of each ten minute average were statistically independent of those
of neighbouring ten minutes, then it would be expected that the annual largest value of the ten-minute
average wind speed would have a probability distribution given by [F (x)]52596 , where F (x) is the Weibull
distribution and 52596 the number of ten minutes in a year. This is not observed in practice, because
values of adjacent ten minutes are not independent. More recent studies (see Castillo [3]) have shown
that provided the data are statistically stationary, and hence have a finite correlation time, , then the

distribution of the largest value is given by F n (x), where n T= if T is the record length (see Harris
[14]).
Following Bergström [1], a relatively simple and convenient way to determine is to use the autocorrela-
tion function ( ), where is the time lag over which the correlation is calculated. If we put = 0:5 and
solve for , the solution may be interpreted as a measure of the average time between two 50% correlated
values in the time series. The solution of this particular case becomes, using spectral representation (see
Rice [20]) :
"Z #1=2
0:5=T Z 0:5=T
T = f 2 Suu(f ) df= Suu(f ) df ; (2.15)
f0 f0
where Suu (f ) is the spectral density function, f the frequency and T = 1= is the sought effective
frequency valid for averaging time T . The lower integration limit, f0 , may be chosen to be about
1=(2 3600) Hz to eliminate the synoptical part of the spectrum. The number of independent observations,
averaged over a period of length T , within a time period Tp , may then be estimated from the relation
n = T Tp (2.16)
Bergström [1] states that the effective frequency is rather insensitive to the specific choice of spectral
density function. This is not confirmed by our own calculations, see Table 2.1. The table shows that the
estimated effective frequencies, using the IEC Von Kármán spectrum with L=U = 10 (see Eq. (3.11)),
are roughly four times greater than the corresponding values given by Bergström. It is possible that
Bergström actually used a lower correlation. We will use Bergströms values because he finds good
agreement between measurements and computation for 1-hour and 10-minute averages, but for shorter
averaging times his computed estimates are slightly too high.
Table 2.1 The effective frequency T and the ratio qT = T =u for various averaging times T; according
to Bergström [1] and according to present calculations based on the IEC Von Kármán and Kaimal spectrum
with L=U = 10.
T T ;B T ;VK T ;K qT ;B qT ;VK qT ;K
[s] ;1
[s ] ;
[s ]1 [s;1 ]
600 0.00073 0.000526 0.000522 0.18 0.167 0.163
60 0.00100 0.004637 0.004456 0.48 0.541 0.482
10 0.00460 0.019240 0.020772 0.78 0.853 0.775
5 0.00790 0.030721 0.035490 0.85 0.909 0.852
3 0.01000 0.042996 0.051734 0.89 0.936 0.893
1 0.02400 0.088163 0.112265 0.95 0.970 0.948
2.3 Evaluation of Vref in the IEC 1400-1 standard
In the IEC 1400-1 standard [16] the peak values of the wind speed are given in the so-called Extreme
Wind Speed Model (EWM). This model must be applied as an ultimate load case at parked conditions.
The extreme wind speeds are given as a function of the reference wind speed Vref which is defined as :
‘the extreme 10-minute average wind speed at turbine hub height with a recurrence period of 50 years.’

This definition is here interpreted as : ‘the annual extreme 10-minute average wind speed at hub height
with a confidence limit of 98%’.
In the IEC 1400-1 standard the ratio of Vref to Vave is constant for wind turbine class I through IV:
Vref = 5:
Vave
The choice of this ratio is based on typical values found in the British building code. This ratio is probably
characteristic of mid latitudes because in de Dutch building code the same ratio appears. This value is, to
a large extent, based on a Gumbel type analysis of wind speed measurements.
Using the above derived expressions (2.3), (2.9) and (2.12), the ratio of Vr to Vave is computed as a
function of the Weibull shape parameter k and presented in Figure 2.1. In this figure the curves based on
the ‘exact’ extremal distribution and the Gumbel distribution almost coincide. To determine the number
n of independent events in a year, the effective frequency T given by Bergström [1], see Table 2.1, is
used : )
T = 7:3 10;4 n = t Tp = 23037
Tp = 3:15576 10
7 (2.17)
It should be noted that the computed curves are not very sensitive for the precise choice of effective
frequency in Table 2.1: n is always so large that the extreme value curve has practically converged to the
limit value curve. For flat terrain sites at mid latitudes the value of the Weibull shape parameter k (for
‘Exact’
Gumbel
15 Davenport
10
Vref/Vave
0
1.0 1.5 2.0 2.5
Weibull shape parameter k
Figure 2.1 The ratio of Vref to Vave as function of the Weibull shape parameter k .
10-minute averages) varies typically between 1.65 inland, 1.9 in coastal areas and 2.1 at sea. Figure 2.1
shows that Vref =Vave 5 for k > 1:77 if we use the ‘exact’ or Gumbel distribution. Hence at mid latitudes
the ratio Vref =Vave = 5 is acceptable for most inland flat terrain sites, and even slightly conservative for

flat terrain coastal areas.
If we take Davenport’s correction into account then the range of acceptable flat terrain sites would be
smaller unless we increase the ratio to 5.5. Note that Bergström [1] finds a good agreement between long
term measurements and extreme value theory, using the Davenport correction.
Figure 2.1 also makes clear that a ratio of 5 is too low for sites in complex terrain, where k -values 1:5
are found. Taking k = 1:4 as a characteristic value for complex terrain then according to Figure 2.1 the
ratio Vref =Vave should be approximately 7.5. Hence with respect to Vref , an IEC Class I wind turbine is
only suitable for complex terrain sites with an annual average wind speed Vave which is lower than 6.7
m/s (at hub height!).
2.4 The pdf of the 3-second average wind speed
So far the extreme value distribution for ten-minute averages has been studied. We would like to have
also extreme value distributions for wind speeds averaged over much shorter intervals, e.g. 3 seconds. To
that end we first have to derive the parent distribution of the 3-sec average wind speeds.
These short term averages will be denoted by uT , where T is the averaging time. It will be assumed that
uT is normally distributed around the 10-minute mean value, U , with standard deviation T . Then the
conditional pdf of uT given U is:
" #
1 1
f (uT j U ) = p exp ; 2 u T ; U 2
: (2.18)
T 2 T
We assume that the distribution of U follows a Weibull function with scale parameter C and shape
parameter k : " #
k ;1 k
f (U ) = Ck CU exp ; UC ; (2.19)
According to probability theory the conditional density function of Y given X is:
f (y j x) ff(x;(xy)) ; (2.20)
x
R1
where f (x; y ) is the joint probability density function of X and Y and fx (x) = ;1 f (x; y ) dy is the
marginal density function of X . Using (2.20) we find
f (uT ; U ) = f (uT j U )f (U ); (2.21)
from which it follows that

Z 1 Z 1
f (uT ) = f (uT ; U ) dU = N [uT ; U; T (U )] W [U ; C; k] dU; (2.22)
0 0
where N [uT ; U; T (U )] and W [U ; C; k ] are shorthand notations for the Normal distribution and Weibull
distribution respectively, and T is the part of the total standard deviation u which remains after having
applied a low pass filter with averaging time T .

The ratio qT = T =u can be computed (see Bergström [1]) from :

"Z #
0:5=T Z 1
qT Suu(f ) df= Suu (f ) df : (2.23)
f0 f0
The lower integration limit f0 can be chosen to be about 1=(2 3600) Hz, to eliminate the synoptical part
of the spectrum, see Table 2.1. The table shows that the values given by Bergström are quite close to the
values found with the Kaimal spectrum given by the IEC 1400-1 standard.
The standard deviation, u , is a random variable. Hence we would prefer to use a conditional pdf
f (u j U ) for the standard deviation, but if such a function is not available then we can either use :
U
u = ln z=z (2.24)
0
or the expression given by the IEC 1400-1 standard :
u = 1:2Iave (0:75U + 0:16Vave ) (2.25)
Using (2.22) and (2.25), the pdf f (uT ) of the 3-sec average wind speed is computed for IEC wind turbine
class I through IV, and presented in Figure 2.2. The figure shows that the distribution of the 3-sec average
0.14
IV
0.12
III
0.10
II
pdf (uT)
0.08 I
0.06
0.04
0.02
0.00
−5 0 5 10 15 20 25 30 35
uT (m/s)
Figure 2.2 The probability density function of the 3-second average wind speed for IEC 1400-1 wind
turbine class I through IV.
wind speed resembles a Weibull distribution but is somewhat flatter and has a more pronounced right
tail than the underlying Weibull distribution of 10-minute averages. Note, however, that contrary to the
Weibull distribution, the left tail in theory extends to ;1.

2.5 The cdf of the extreme 3-sec average wind speed
The cumulative distribution function (cdf) of uT is given by:

Z uT Z 1
F (uT ) = N [uT ; U; T (U )] W [U ; C; k] dU duT : (2.26)
;1 0
Assuming n independent observations, the extreme value distribution is then given by the distribution of
the n’th order statistic (cf. Appendix A.1.3) :
FXn:n (uT ) = F n(uT ): (2.27)
Some examples of F n (uT ) are presented in Figure 2.3. The value of n is based on the effective frequency
given by Bergström in Table 2.1. Unfortunately, the computation of F (uT ) is very time consuming. An
alternative, much faster, procedure takes advantage of the fact that two cdf’s with the same right tail, the
same values for the (0.9,1) range say, and very different values in the (0,0.9) interval have exactly the
same (maximum) limit distribution, see Castillo [3][Chapter4.5].
Since f (uT ) resembles a Weibull distribution function, we can fit a Weibull distribution to the right tail
of f (uT ). The parameters C and k of this Weibull fit can then be used to obtain the extreme value
distribution, by either solving x from Eq. (B.7) or Eq. (B.9). The use of (B.7) is slightly more accurate
than (B.9) which introduces some additional error by neglecting the higher order terms in the Taylor
expansion (see Appendix B). In practice, good agreement with the ‘exact’ computation (2.27) is obtained
if the right tail in the (0.9,1) range is taken, see Figure 2.4.
1.0
0.9
0.8
0.7
0.6
F (uT)
0.5
n
IV III II I
0.4
0.3
n=315576
T=3 sec
0.2
k=2
0.1
0.0
20 25 30 35 40 45 50 55 60 65
uT (m/s)
Figure 2.3 The ‘exact’ cumulative extreme value distributions of the 3-sec average wind speed for the
four IEC wind turbine classes.

0
10
pdf(uT) (Vave=8.5, k=2.0)

−1
10 Weibull fit (C=8.878, k=1.671)
−2
10
−3
10
>90%
pdf (uT)
−4
10
−5
10
−6
10
−7
10
−8
10
−5 5 15 25 35 45 55
uT (m/s)
Figure 2.4 The calculated pdf of uT and a fitted Weibull distribution
2.6 Evaluation of Ve1 and Ve50 in the IEC 1400-1 standard
In the IEC 1400-1 standard [16] two extreme 3-second average wind speeds are defined:
1. The 50-year extreme wind speed Ve50 , which is defined as the extreme 3-seconds average wind
speed with a recurrence period of 50 years.
2. The yearly extreme wind speed Ve1 , which is defined as the extreme 3-seconds average wind speed
with a recurrence period of one year.
Ve1 and Ve50 are both based on the reference wind speed Vref and have to be computed as a function of
height z using the following equations :
Ve50(z) = 1:4Vref (z=zhub )0:11 ; (2.28)
Ve1 (z) = 0:75Ve50 (z); (2.29)
For each of these two extreme wind speeds a recurrence period is specified, so that, in principle, the
probability of occurrence is known. Unfortunately the definitions given above are difficult to interprete,
especially the definition of Ve1 seems a bit odd. None of the definitions contain information on the length
of the samples, which is essential in order statistics. Assuming that the extremes distribution is based
on samples of one year length, a yearly extreme wind speed with a recurrence period of one year has a
100% probability of being exceeded (cf. Section A.1.4). This cannot be intended of course. Therefore
we assume that the following definitions should be employed for Ve50 and Ve1 :
Ve50 is the annual extreme 3-second average wind speed with a 2% probability of being exceeded
in a year. In other words Ve50 has a return period of 50 years.

Ve1 is the yearly characteristic largest 3-second average wind speed. This value gives an idea
of the central location of the possible largest values, and corresponds to a confidence level of
approximately 63.21%, cf. Section A.1.5.
To verify whether the values given by the IEC standard correspond to the definitions as we assume they
should be, Ve1 and Ve50 have been computed for various values of Vave and k . The results are shown in
Figures 2.5 and 2.6. For the purpose of comparison these figures also show the IEC values for the four
wind turbine classes. These figures show that the IEC values for Ve1 and Ve50 are valid for sites where
k > 1:9 and k > 1:7, respectively. Hence the assumed confidence levels for Ve50 and Ve1 are probably
correct. That Ve1 is less conservative than Ve50 , in terms of k -values, seems appropriate because Ve1
should also be used in fatigue calculations (DLC 6.2).
It can be concluded that the IEC values of Ve1 and Ve50 are acceptable for most flat terrain sites at mid
latitudes and certainly valid if the parent distribution is Rayleigh. But, with respect to these values, a
Class I wind turbine is only suitable for complex terrain sites (k 1:4) where Vave 7:5 m/s at hub
height.
2.7 Conclusions
In the previous sections the theoretical values of Vref , Ve1 and Ve50 have been derived as a function of
the annual average wind speed Vave and the Weibull shape parameter k . These theoretical values are
compared to the values specified by the IEC 14001-1 standard in the so-called Extreme Wind Speed
Model (EWM).
It is shown that according to these theoretical computations the IEC values are acceptable for flat terrain
sites at mid latitudes. For sites with a Rayleigh parent distribution the IEC values are conservative. As
was to be expected, the IEC values are definitively not valid for complex terrain sites. With respect to the
survival wind speeds, Ve1 and Ve50 , a IEC Class I wind turbine is only suitable for complex terrain sites
(k 1:4) with an annual average wind speed which is lower than 7.5 m/s at hub height.

100
9 10 11 12
8
90
7
80
Characteristic largest wind speed (m/s)
6
70
5
60
I
50 4
II
III
40
IV
30
20
Confidence level 63.21%

10
0
1.0 1.5 2.0 2.5
Figure 2.5 The annual characteristic largest 3-sec average wind speed as function of the Weibull shape
parameter k and annual average wind speeds Vave varying between 4 m/s and 12.5 m/s. Also shown are
the values of Ve1 for the four IEC wind turbine classes.
100
7 8 9 10 11 12
90
Annual extreme 3−sec average wind speed (m/s)
80
5
I
70
II 4
60
III
50
IV
40
30
20
Confidence level 98%

10
0
1.0 1.5 2.0 2.5
Figure 2.6 The annual extreme 3-sec average wind speed with a confidence level of 98%, as function
of the Weibull shape parameter k and annual average wind speeds Vave varying between 4 m/s and 12.5
m/s. Also shown are the values of Ve50 for the four IEC wind turbine classes.

3 THE DISTRIBUTION OF EXTREME GUSTS
3.1 The pdf of single point gusts
To find the probability density function of gusts we consider the short-term average wind speeds uT1 and
uT2 at the same location but at two instants t1 and t2. Let = t2 ; t1 be the time difference and let us
assume that uT is normally distributed around the 10-minute mean value U . Then the conditional pdf
f (x1 ; x2 j U ) is jointly normal :

1 1
f (x) = (2)n=2 jK j1=2 exp ; 2 (x ; mx) Kxx (x ; mx ) ;
T ; 1 (3.1)
xx
with mean vector and covariance matrix
!
mx = fU; U gT Kxx = 2 1 ( ) j( )j < 1;
, T ( ) 1 , (3.2)
and where ( ) = x1 x2 =x1 x2 is the so-called correlation coefficient which depends on the time
difference . To evaluate gust amplitudes we need to determine the conditional pdf f (x2 ; x1 j x1 ),
expressing the difference x2 ; x1 for a given ‘starting’ wind speed x1 . This pdf is derived by following
the steps :
(i) Introduce a linear transformation of random variables :
Y1 = X1 ; Y2 = X2 ; X1 ;
(ii) Evaluate the conditional f ((y 1; y 2) j U ) according to the transformation rule :
f (y) f ((y1 ; y2 ) j U ) = f (g(x))jJn j (3.3)
where Jn is the Jacobean of the inverse transformation :

@ ( x 1 ; x2 ; : : : ; xn )
Jn = det @ (y ; y ; : : : ; y ) (3.4)
1 2 n
Using (3.3) and (3.4) we find:
1 ; 1 ) 1 ; 1 )(y2 ; 2 ) (y2 ; 2 )
2
1 1
f (y) = 22 1 ; 2 exp ; 2(1 ; 2 )
p ( y 2 ( y
2 ; 2 2 + 2 ;
(3.5)
where
1 = U and 2 = U ; y1 = U ; x1 : (3.6)
(iii) Using
f (x1; x2 ; : : : ; xm j xm+1 ; : : : ; xn ) = gh((xx1 ; x2;; :: :: :: ;; xxn)) ; (3.7)
m+1 n
it can be shown that
f ((y2 j y1 ); U ) = f ((y2 ; y1f) j(yU )) fU (U ) ; (3.8)
1 1
where f (y1 ) f1 (y1 j U ) is the marginal density function of y1 and which is equal to N [y1; U; T (U )],
see section 2.4 for the notation.

(iv) Evaluation of (3.8) results in :

f ((y2 j y1); U ) = N [y2 ; (U ; y1)(1 ; ); T (1 ; (2 )1=2] W [U ; C; k]:
(v) Hence the pdf of y2 given y1 is :
Z 1 Z 1
f (y2 j y1) = f ((y2 j y1 ); U ) dU = N [y2 ; (U ; y1)(1 ; ); T (1 ; 2)1=2 ] W [U ; C; k] dU:
0 0
(3.9)
In other words, f (y2 j y1 ) is the product of a normal distribution with mean (U ; y1 )(1 ; ) and standard
deviation T (1 ; 2 )1=2 and a Weibull distribution with parameters C and k . The resulting pdf provides
the probability of a wind increase y2 = x2 ; x1 in a given time interval and a starting wind speed
y1 = x1 . It should be noted that the pdf depends on the correlation function which is not only a function
of but also of L=U (see section 3.1.1). As a consequence cannot be entered as a constant in (3.9), as
was incorrectly done in [4]. The discussion on T held in the previous section applies here as well. A
typical plot of f (y2 j y1 ) is presented in Figure 3.1. This figure shows that for decreasing annual average
IV
0.5
0.4
III
II
pdf (Vgust | uT )
1
0.3
I
uT = 14 m/s
1
0.2
τ=6s
Von Karman
0.1
0.0
−10 0 10
Vgust = uT − uT (m/s)
2 1
Figure 3.1 Pdf of (uT2 ; uT1 j uT1 ) with T = 3 sec, uT1 = 14 m/s, and external conditions according to
IEC wind turbine class I – IV.
wind speed the negative gust portion increases and vice versa. It can also be shown that when decreases
(more correlation) the width of f (y2 j y1 ) is reduced and the peak becomes higher, sharper and more
symmetrically centered around zero. When the starting value uT1 increases, the peak of the distribution
shifts to the left, indicating that the probability of finding large (positive) gusts becomes smaller.
Using (3.9) we can for example find that the probability of Y2 being between c and d given that y1 <
Y1 < y1 + dy1 is
Z d
Pr(c < Y2 < d j y1 < Y1 < y1 + dy1) = f (y2 j y1 ) dy2 ; (3.10)
c

hence Pr(Y2 y2) = 1 ; Pr(;1 < Y2 < y2 j y1 < Y1 < y1 + dy1).
3.1.1 The correlation function
Under stationary conditions, one may obtain the dependence of the correlation coefficient on the time
interval from the power spectrum of the longitudinal component. Assuming ergodicity it can be shown
that the correlation function is the inverse Fourier transform of the non-dimensional spectrum Suu =u2 ,
where u is the wind speed component in longitudinal direction. For a Von Kármán-type spectrum :
Suu = 4 L 1
u2 U (1 + 70:78(fL=U )2 )5=6 (3.11)
it is evident that the correlation function will not only be a function of but also of the ratio L=U ,
which is equal to the integral time scale when Taylor’s hypothesis of frozen turbulence applies. The
positive-time part of ( ) is shown in Figure 3.2. Note that for a given time lag the correlation will be
1.0
0.9
Von Karman
Kaimal
0.8
Correlation coef. ρu(τ)
0.7
0.6
0.5
0.4
1/e
0.3
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Reduced time Uτ/L
Figure 3.2 The correlation coefficient as function of dimensionless time =(L=U ).
higher according to Kaimal than according to Von Kármán, due to the much larger integral length scale
L in the Kaimal spectrum.
3.1.2 The number of independent events
To find the number of independent events n for the extreme gusts, we introduce four jointly normal
random variables X1 ; X2 ; X3 ; X4 , corresponding to longitudinal wind component ensembles at t = 0,
t = , t = a and t = + a. Let the process be stationary and ergodic and let U be the ten-minute mean and

the standard deviation of each individual random variable. Then, the jointly normal f (x1 ; x2 ; x3 ; x4 )
will be characterized by the mean vector fU; U; U; U gT and the correlation matrix :
2 3
1 a +a
6
Kxx = 2 664 1 ;a a 777
1 5 (3.12)
symm 1
Performing the necessary calculus to express f (x3 ; x1 ; x4 ; x2 j x1 ; x2 ) it is shown that this is a jointly
normal pdf with correlation matrix
" #
K = 2 (2 + 2 ; ) 1 ;
1 ; 2 1 (3.13)
where
= 1 ; a and = ( +a ; )( ;a ; ): (3.14)
This denotes that the correlation coefficient of the gust pdf is equal to the correlation coefficient of the
two jointly normal wind speed variables (X1 ; X2 ) placed at time distance . In other words, wind speed
variables and gust variables have the same correlation coefficient and, consequently, one can apply the
earlier derived effective frequency values for extreme wind speeds to extreme gust computations as well.
3.2 Results and comparison with IEC
1.0
0.8
0.6
F (Vgust1 | uT )
1
IV III II I
uT =14 m/s
0.4 1
n
T=3 s
τ=6 s
0.2
n=315576
0.0
0 10 20 30 40 50
Vgust1 = uT −uT (m/s)
2 1
Figure 3.3 The limit distributions of the extreme operational gust (EOG) as prescribed by the IEC
Using Eq. (3.9), the magnitude of the extreme operating gusts is computed, assuming the same confidence
levels as for the extreme 3-second average wind speeds, i.e. 63.2% for Vgust1 and 98% for Vgust50 . For

the long term wind conditions we take the conditions according to IEC Class I to IV. According to the list
of load cases prescribed by the IEC standard, the starting wind speeds should be either Vrmrated or Vout .
We assume that Vrated = 14 m/s and Vout = 25 m/s. The results are presented in Table 3.1.
The table also contains the magnitude of the extreme coherent gust Vcg . The IEC does not specify a
recurrence period for this gust and therefore it is assumed that the confidence level of this gust magnitude
corresponds to that of the characteristic largest value. Hence Vgust1 and Vcg have the same assumed
probability of occurrence, only the time lag is different ( = 6 s and = 10 s, respectively).
In all computations the correlation function is the inverse FFT of Von Kármán’s auto power spectrum of
the longitudinal wind speed component, see Eq. (3.11). It should be noted that the correlation function
based on Kaimal’s expression for the longitudinal turbulence spectrum results in lower gust values.
Figure 3.4 shows that, at least in the range considered, there is a linear correspondence between the gust
magnitude and the annual average wind speed. For a given probability, the gust values are lower at a
starting wind speed of 25 m/s than at 14 m/s. The difference between Vgust1 and Vcg is small considering
the relatively large difference in time lag (6 s and 10 s, respectively). This is due to the correlation
function which does not change very much when the time lag is greater than the integral time scale L=U ,
see Figure 3.2. The results presented in Table 3.1 clearly show that the computed values are significantly
Table 3.1 The extreme operational gust and extreme coherent gust for a starting wind speed uT1 = 14
m/s and 25 m/s and external conditions according to the IEC wind turbine class I – IV.
IEC (EOG, ECG) uT1 = 14 m/s uT1 = 25 m/s IEC rev. 2

IEC Vgust1 Vgust50 Vcg Vgust1 Vgust50 Vcg Vgust1 Vgust50 Vgust1 Vgust50 Vcg
class (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s) (m/s)
I 15.6 20.8 15.0 29.9 37.1 31.0 19.7 26.7 18.7 25.0 15.0
II 15.4 20.5 15.0 23.3 29.3 24.5 13.4 19.3 18.7 25.0 15.0
III 15.3 20.3 15.0 18.9 24.2 20.1 9.4 14.4 18.7 25.0 15.0
IV 15.1 20.1 15.0 12.6 16.6 13.4 3.6 7.4 18.7 25.0 15.0
different from the values specified by the IEC standard, both in trend and magnitude. At a starting wind
speed of 14 m/s most computed gust amplitudes are much higher than those specified by the IEC. At 25
m/s the values are comparable for Class I and II but for Class III and IV the IEC values are much higher.
All computed gust amplitudes have a clear dependence on the annual average wind speed (see Figure
3.4), while no such dependence is evident in the IEC specification.
There are, however, a few remarks to be made here :
1. Experience using the IEC standard has shown that the Normal Turbulence Model (NTM) leads in
most cases to higher loads than the extreme operating gust, indicating that the EOG model is too
benign with respect to actual external conditions.
2. New analysis of the extreme operating gust, performed by a working group of the IEC (TC88/WG7)
[2], based on measurements from the German lowland and Californian sites, shows a ratio of gust
magnitude (Vgust1 ) to standard deviation (1 ) of about 4.8. For Vgust50 the ratio is estimated at 6.4.
They also observe that this ratio is more or less independent of the turbulence intensity of the site.
Based on this analysis the working group has proposed a new model for the extreme operating

gusts. The corresponding gust amplitudes are presented in Table 3.1 under the heading IEC rev. 2.
The new values are closer to our predicted values for class I and II, in good agreement for class III
and higher for class IV.
3. The reason that our computed gust magnitudes for Class III and IV are so low has to do with the
fact that we assumed the same rated and cut-out wind speed for all IEC classes, viz. 14 m/s and
25 m/s, respectively. In practice an IEC Class III or IV wind turbine will have a lower rated and
cut-out wind speed. For example, considering a Class IV wind turbine with Vrated = 8 m/s and
Vout = 20 m/s, we find that Vgust1 = 17:5 m/s and Vgust50 = 21:7 m/s at a starting wind speed of 8
m/s, and Vgust1 = 7:7 m/s and Vgust50 = 11:6 m/s at a starting wind speed of 20 m/s.
4. The IEC should make a distinction between the gust amplitudes at different starting wind speeds.
A starting wind speed equal to Vrated is generally much more likely than a starting wind speed at
Vout . Consequently a gust with an amplitude of 25 m/s has a much lower probability to start at Vout
than to start at Vrated , everything else being equal.
5. The IEC standard (and the proposed revision) is unclear about how to compute the standard
deviation in the EOG model. It must be made clear that in all cases 1 should be computed for
Vhub = Vout, but this is nowhere stated.
6. The extreme values computed here may seem very high, but they are based on all possible mean
values of the 10-minute average U . Obviously the largest gust values will be found during a period
of high mean wind speed. For example when U = 25 m/s then a drop to 14 m/s and a consequent
increase to 36 m/s is not exceptional. What makes it exceptional is that this wind speed change
takes place within 6 seconds and the high mean wind speed itself.
7. The gust magnitude is very sensitive for the Weibull shape parameter k . For example, for a Class
III wind turbine, with k = 1:7 and Vrated = 10:5 m/s, we find that Vgust1 = 29 m/s at a starting
wind speed of 10.5 m/s.
8. The computed values should be considered with some caution. Because the gusts should be applied
with the turbine in operation it would be better to restrict the range of U and to compute the extreme
gusts using the density function f ((y2 j y1 ) j U Vout ). This will reduce the computed gust values.
9. The gusts considered here are single point gusts, i.e. wind speed changes as they would be measured
by a single anemometer. It is obvious that coherent gusts will have a smaller magnitude if we assume
the same confidence limit. This so-called rotor averaging effect can be taken into account by means
of a spatial correlation function. The gust magnitude will then also be a function of the rotor
diameter.
10. The above derived gust distribution function (3.9) may be used to construct gust time histories,
because the pdf and cdf depend on the time lag . One could think of some kind of Markov chain
simulation (random walk).
11. We are not so much interested in extreme gusts but in gusts which cause extreme loads in the
various components of the wind turbine. It is quite possible that the extreme gust values found here
do not induce the highest loads, simply because the wind turbine does not have time to respond
to such large and rapid wind speed changes. The extreme values given in Table 3.1 should serve
as an upper limit, we do not need to evaluate larger gusts. When the gust is found that induces

the highest loads (this may vary with the design and the cross section that is considered) the above
presented theory can be used to compute the number of occurrences.
12. Due to their physically unrealistic nature, sinusoidal gust are not the most reliable means to compute
extreme loads, see Winkelaar [23]. It is recommended to use several realisations of a wind field
which somehow contains the gust to compute the (hopefully) extreme loads.
40 40
Von Karman Von Karman

30 30
k=2 k=2
n = 315576 n = 315576
u = 14 m/s u = 25 m/s
Τ Τ
1 1
1
1
20 20
2
10 Vgust50 10 Vgust50
Vgust1 Vgust1
Vcg
0 0
6.0 7.0 8.0 9.0 10.0 6.0 7.0 8.0 9.0 10.0
Annual average wind speed Vave (m/s) Annual average wind speed Vave (m/s)
Figure 3.4 The extreme 3-sec average gust as a function of the annual average wind speed for two
starting wind speeds (14 m/s and 25 m/s), two confidence levels (63.2% and 98%) and two time lags (6 s
and 10 s).


4 THE DISTRIBUTION OF EXTREME WIND DIRECTION CHANGE
4.1 The pdf of wind direction
There are several ways to approach the problem of deriving the pdf of the horizontal wind direction. The
easiest way is to use the streamwise coordinate system x; y; z; and the corresponding velocity components
u; v; w. These velocities can be separated into a mean component, here denoted by capital letters and an
eddy component denoted by a prime :
uT = U + u0T ; (4.1)
vT = V + vT0 ; (4.2)
wT = W + wT0 : (4.3)
By definition, we have u0T = vT0 = wT0 = 0. If the x axis is defined in the direction of the mean flow,
V = W = 0. Over flat, level, homogeneous terrain, we take x and y to be horizontal and z to be vertical
and positive upwards.
To construct the joint pdf of uT and vT we assume that both wind components are normally distributed
" #
fu(uT j U ) = 1p exp ; 1 uT ; U 2 ;
uT 2 2 uT (4.4)
" 2 #
fv (vT j U ) = 1 1
p exp ; 2 v T ; (4.5)
vT 2 vT
where it is assumed that U has a Weibull distribution and uT and vT are functions of U .
If we assume that uT and vT are jointly normal and uncorrelated then the joint distribution function
fuvU (uT ; vT ; U ) is given by
f (uT ; vT ; U ) = fu (uT j U ) fv (vT j U ) fU (U ): (4.6)
To construct the pdf of wind direction we transform to polar coordinates and consider the random
variables : q
r = u2T + vT2 = arctan vT =uT (4.7)
where we assume that r > 0 and ; < . With this assumption the system has a single solution
uT = r cos vT = r sin for r > 0: (4.8)
Since ;1
J (uT ; vT ) = ;cos

sin

= 1r

r sin r cos

(4.9)
we conclude from (3.3) that

" ( )#
f (r; ; U ) = rf (uT ; vT ; U ) = 2r exp ; 12 r cos ; U 2 + r sin 2 W [U ; C; k]
u v u v
(4.10)
for r > 0 and zero for r 0.
The pdf of wind direction is then given by

" ( )#
f () =
Z 1Z 1 r exp ; 1 r cos ; U 2 + r sin 2 W [U ; C; k] dr dU:
0 0 2u v 2 u v (4.11)
Now f ( ) is not computed for a given U and but computed for all possible mean wind speeds between
zero and infinity. In reality only mean wind speeds up to Vout are of importance, so we are actually
interested in f ( j U Vout ). Clearly the numerical evaluation of these multidimensional integrals is
computationally expensive. Therefore we seek a simpler approach, assuming that U , uT and vT are
given (chosen) values.
If we again assume that uT and vT are normally distributed
" #
1 1
fu(uT ) = p exp ; 2

u T ; U 2
; (4.12)
u 2 u
" 2 #
1 1
fv (vT ) = p exp ; 2 v T ; (4.13)
v 2 v
and if we also assume that these wind speed components are statistically independent, then the joint
density function is given by :
" ( )#
f (uT ; vT ) = fu(uT ) fv (vT ) = 2 exp ; 21 1 uT ; U 2 + vT 2 :
u v u v (4.14)
To find the pdf of wind direction we transform to polar coordinates :

" ( )#
f (r; ) = rf (uT ; vT ) = 2r exp ; 12 r cos ; U 2 + r sin 2 r>0 (4.15)
u v u v
and zero for r 0.
We then find for the pdf of :
" ( )#
f () =
Z 1
f (r; ) dr =
Z 1 r exp ; 1 r cos ; U 2 + r sin 2 : (4.16)
0 0 2u v 2 u v
The same pdf can be derived by starting with an arbitrary Cartesian coordinate system x and y and
assuming that the velocity components ux and uy are jointly normal distributed with mean fx ; y gT and
covariance matrix !
2 2
= 2x xy2 : K xy y
Following Ibarra [15] one can always define a transformation (ux ; uy ) ! (u; v ) that diagonalizes the
covariance matrix. In tensorial terms, the entire process is reduced to finding the eigenvalues and principal
K
directions of . The principal directions are those of an ellipse, the eigenvalues u2 ; v2 are given by
" #1=2
2
2 ; 2
u v =I
2
I
2 ; jK j ; (4.17)

where I = x2 + y2 , and the principal directions are determined by :

tan 2 = 2xy ; x2 ; y2 (4.18)
where is the angle formed by the Cartesian co-ordinate system and the principal directions. In flat terrain
this procedure is identical to finding the ‘streamlined’ co-ordinate system (longitudinal, transversal) where
U = x cos + y sin ; (4.19)

V = 0; (4.20)
and u : v = 1 : 0:8 and uv = 0. In general, however, the new jointly normal random variables will
have a mean vector fU; V gT with V 6= 0.
To obtain the pdf of the horizontal wind direction we proceed in the same way as before by a transformation
to polar coordinates :
( " ! #)
r 1 cos 2 sin2 cos sin
f (r; ) = 2 exp ; 2 r 2 + 2 ; 2r iu + iv + (i2u + i2v ) ;
2
u v u v u v
(4.21)
where
iu = U and iv = V : (4.22)
u v
This expression is clearly the same as (4.15) for iv = 0.
The pdf of wind direction is again found by integrating f (r; ) for r in [0; 1]. The integral can be cast
into closed form, see Gradshteyn and Ryzhik [12] :
; ) 1
(
p p " 2 # )
f() = exp(
2u v 2a ; 2a a exp
pa erfc pa ; (4.23)
where !
2 sin2
a = 2 2 + 2 ; = ; 2 iu + iv ; = 12 (i2u + i2v ):
1 cos 1 cos sin (4.24)
u v u v
It can be shown that f ( ) does not depend on the actual values of u and v (or U and V ) but only on
the non-dimensional ratios = v : u , iu and iv , the latter two are the reciprocal of the ‘directional’
turbulent intensities. Evidently this is also true for the standard deviation of f ( ), which is denoted by
Z
2
= 2 f() d (4.25)
;
A similar analysis is presented by Weber [22]. For = 1 and iv = 0 the relation of the standard deviation
of and the turbulence intensity (TI) of the longitudinal component is given in table 4.1. It can be
shown that, under extremely calm conditions (iu = iv = 0), f ( ) is uniformly distributed in the interval
[;; ]. For larger values of i a symmetric bell-type distribution with zero mean is obtained.
For values of the turbulence intensity we are interested in, f ( ) does not resemble a Gaussian pdf. To
demonstrate his, a logarithmic plot of f ( ) along with the corresponding Gaussian pdf (which has the
same mean and variance as f ( )) is presented in Figure 4.1 for = 1, iv = 0 and TI=15%. It is seen
that the tails of the distribution – which are very important for extreme value considerations – are not

Table 4.1 versus turbulence intensity TI

TI 5% 10% 15% 20% 25% 30%
(deg) 2.88 5.76 8.70 11.71 14.88 18.30
Figure 4.1 The pdf of wind direction and two fits.
well predicted by the Gaussian fit. This creates some difficulties in applying the theory of section 3 to
wind direction changes; we cannot assume that f1 2 (1 ; 2 ) is jointly normal. A better tail-fit can be
obtained by introducing the following symmetric pdf :
h i
1 exp ; (x ; =C )k
f (x) = 2 C ;(1 + 1=k) ; (4.26)
with
= E[x];
and
;(3=k) :
2 = E[(x ; )2 ] = C 2 ;(1 =k) (4.27)
It is seen that for given standard deviation the introduced pdf has only one free parameter, the shape
parameter k (for k = 2 the pdf (4.26) becomes the normal pdf). For k = 1:73 a curve is obtained, called
the exponential fit in Figure 4.1, which represents the tail distribution of f ( ) in a much better way than
the Gaussian fit.

4.2 The pdf of wind direction change
Due to the non-Gaussian tail of f ( ) the gusts theory presented in section 3 cannot be applied di-
rectly to wind direction changes. To circumvent this, it was first tried to introduce four jointly normal
(U1 ; V1 ; U2 ; V2 ) variables representing the Cartesian components of the horizontal wind speed vector at
two locations (1) and (2) with given correlation coefficient (or given time lag). A transformation of
variables from Cartesian to polar co-ordinates would then provide f (r1 ; r2 ; 1 ; 2 ). Integrating twice with
respect to r1 and r2 would then give f (1 ; 2 ) and the procedure of section 3 could then be followed.
Unfortunately, the double integral cannot be cast into closed form while numerical integration is too
costly. To face this problem an alternative, approximate, technique is proposed, based on the following
assumptions :
1. The sought f (2 ; 1 j 1 ), for a given longitudinal turbulence intensity, respects the mean and
variance of the ‘equivalent Gaussian problem’ discussed in section 3. In other words it is assumed
that :
m[(2 ; 1 ) j 1] = ;1(1 ; );
and
[(2 ; 1 ) j 1 ] = (1 ; 2 )1=2 ;
where is the corresponding correlation coefficient.
2. The new pdf retains the shape parameter k of the corresponding wind direction distribution.
Under these assumptions f ((2 ; 1 ) j 1 ) will have the form of Eq. (4.26) with = ;1 (1 ; ) and
a scale parameter C which corresponds to the ‘modified’ standard deviation (1 ; 2 )1=2 through Eq.
(4.27). The relation of the time distance between station (1) and (2) and the correlation coefficient can
be obtained by applying an inverse FFT to a typical wind direction spectrum. It is recalled that f ( )
and consequently f ((2 ; 1 ) j 1 ), does not depend on the mean wind speed or the standard deviation
of any wind component, but depends on the non-dimensional values = v : u , iu and iv which can be
considered to be site dependent constants.
4.3 Results and comparison with the IEC 1400-1 standard
Figures 4.2 to 4.4 present contour plots for various Weibull parameters and turbulence intensities. The
driving parameter here is the turbulence intensity and the three plots correspond to 10%, 15% and
20% turbulence intensity. The calculated extremes correspond to = 1 (complex terrain oriented) and
iv = 0. It is clearly seen that the extreme values increase rapidly as the turbulence intensity increases.
Although not exact in this case, the correlation ! time transformation is again performed by means of the
longitudinal wind component spectral properties. As already discussed a typical wind direction spectrum
should be used in this case to provide the ‘exact’ ( ) curve.
Note that if we had used the expression (4.10) as starting point, then the correlation function would be
a function of and of L=U and the results would not only be different, it would also be impossible to
present them the way they are presented now.
Projecting all the extreme values of a 1 = const: line on the time axis the time history of an extreme wind
direction change is synthesized. Time histories of the extreme wind direction change (IEC code name

EDC) are presented in Figure 4.5. Comparing amplitudes and rise times it is seen that the IEC values are
in good agreement with the present values obtained with 1 = ;90 deg. (the most severe case). It is also
worth noting that both the IEC model and the present are based on a single site parameter, the turbulence
intensity. The IEC curve is computed for TI = 18 %, while the present curves correspond to the contour
plot of Figure 4.3 (TI = 15%). It is seen that the present model predicts large direction changes at early
times.
Figure 4.2 Extreme 50-year wind direction change. 10% turbulence intensity and 95% confidence level



Figure 4.5 Extreme 50-year wind direction change time history. Present (Figure 4.3, L/U=10) versus
IEC standard.


5 OTHER EXTREME ATMOSPHERIC EVENTS
At most sites, wind speed arises from a variety of physically distinct meteorological mechanisms, such as
depressions, thunderstorms, tropical cyclones, etc. In the previous sections it is assumed, however, that
all large wind speeds are produced by only one of these mechanisms. For sites at temperate latitudes,
such as most sites in Europe, this is a reasonable assumption, the dominant mechanism being depressions.
In meteorological terms it is assumed that when the highest gust speeds and the highest average (over
10 min or 1 h) wind speed occur near the ground (say at 10 m) the lower boundary layer is neutrally
stratified and the gusts are caused solely by the vertical shear of the wind and the resistance of the rough
ground. This condition of the boundary layer is usually defined (in meteorology) by the criterion that
jh=LMO j 0:1, where h is the height above the boundary layer depth (typically 500–1000 m) and LMO
is the Monin-Obukhov length. Typically in temperate climates this criterion is satisfied if the wind speed
at 10 m above the ground U10 is greater than 10 m/s.
However, even for mid-latitudes, where tropical cyclones do not occur, it has been recognized that there
are exceptional meteorological situations when mechanisms quite different to shear turbulence cause
very high wind speeds. In particular, tornadoes are associated with high swirl velocities ( 100 m/s) and
vertical upward motion, while downdrafts, produced by cold air formed in deep clouds well above the
boundary layer, descend and spread out rapidly on hitting the ground giving rise to typical gust speeds
30 m/s [11]. Although both these phenomena are most frequent in continental climates and the tropics,
they are occasionally observed even in temperate climates, such of those of North Western Europe, and
need to be included in the statistics of extreme winds, see Collier et al. [5].
In this section we will not concern ourselves with tornadoes. To date, most tornado study is a post-disaster
analysis of structures which were victims of the storm, with the damage assessed by calculating the upper
limit of the failure load due to a straight line wind approximated as 140 m/s. Judging from the survival
wind speeds specified in the IEC standard (70 m/s at the most) a wind turbine does not have to survive a
tornado.
Although it is sometimes difficult to distinguish a downburst from a tornado the associated wind speeds
are generally not that high. Two large measurement campaigns (NIMROD and JAWS) in the USA (Fujita
[11]) recorded gust speeds up to 30 m/s and showed that intense downbursts can leave a trail of damage,
like uprooted trees etc. For these reasons and the assumption that downbursts probably happen much
more often than tornadoes it seems a good idea to study the phenomenon more closely, to see whether it
should be included as a load case in the IEC standard.
5.1 Downbursts
The following is taken from Fujita [11]. A downburst is a localized, intense downdraft that may be
dry or may be driven by cooling when the water vapour in falling moist air evaporates. The vertical
velocity is defined to be more than 3.6 m/s at 91 m above ground level. Damaging winds, either straight
or curved, are highly divergent. The sizes of downbursts vary from less than one kilometer to tens of
kilometers. Downbursts are subdivided into macrobursts and microbursts according to their horizontal
scale of damaging winds.

MACROBURST : A large downburst with its outburst winds extending in excess of 4 km in horizontal
dimension. An intense macroburst often causes widespread, tornado-like damage. Damaging
winds, lasting 5 to 30 minutes, could be as high as 60 m/s.
MICROBURST : A small downburst with its outburst, damaging winds extending only 4 km or less.
In spite of its small horizontal scale an intense microburst could induce damaging winds as high as
75 m/s.
Not all microbursts are alike; some are accompanied by heavy rain, while others form beneath small
virga, see Figure 5.1. The parents clouds which induce microbursts are not always thunderstorms.
Figure 5.1 Three types of microburst clouds confirmed by radar observation (Courtesy Ref. [11] ).
Often, isolated rainshowers spawn relatively strong microbursts which endanger aircraft operations at
low altitude.
By virtue of its small size and its short life of less than 10 minutes, a microburst often escapes detection
by non-Doppler radars or ground-based anemometers. Consequently a wind turbine could encounter a
microburst unexpectedly. A small but very strong microburst often causes a starburst-shaped tree damage.
If we assume a 25 m/s outburst wind along a circle of 100 m radius from the microburst center, the mean
divergence inside the circle is as large as 0.5 per second. In order to induce such a large divergence at the
near-ground level, the downflow speed will have to be 5m/s at 10 m AGL and 10 m/s at 20 m AGL. In
other words the downward current at the starburst center is very strong even at the tree-top level.
The maximum wind speed in a microburst measured in the NIMROD experiment was 31.3 m/s and in
JAWS was 32.6 m/s. However, most wind speeds were in the range of 12 to 14 m/s. A Gumbel type of
analysis reveals that extreme wind speeds with a very low probability of occurrence could be as high as
75 m/s.
In contrast to the very strong outburst winds inside a microburst, its boundary propagates outward rather
slowly. Photographic evidence of the microburst winds made visible by flying dust often shows a vortex
with a horizontal axis. When the horizontal vortex encircles the downflow center, forming a vortex
ring, the outbursting winds beneath the vortex ring are accelerated continuously as the ring expands
and stretches, see Figure 5.2. The ring vortex keeps stretching as a surface microburst gets older, until
reaching a stretching limit. Thereafter the vortex is cut into several pieces of roll vortices, each with a
horizontal axis. Some of these vortex rolls run away from their source region, inducing bands of high
winds, lasting two or three minutes.

Figure 5.2 Outflow microburst are the most commonly observed type. Some rotorbursts develop inside
macrobursts behind their gust fronts (Courtesy Ref. [11]).


6 SUMMARY, CONCLUSIONS AND RECOMMENDATIONS
Using extreme value theory it is shown that, given a certain knowledge of the climatology, it is possible to
derive statistical characteristics of extreme wind conditions which are believed to be important for wind
turbine design, e.g. the survival wind speed, extreme gusts and extreme wind direction changes. In the
present report the limit distributions of these extreme wind conditions have been investigated, assuming
that the design external conditions (i.e. the climatology) is defined by the IEC 1400-1 standard wind
turbine classes I–IV.
With the assumptions given above the derivation of the probability density function (pdf) and cumulative
distribution function (cdf) for the 10-minute average and 3-second average extreme wind speeds is
straightforward. It is shown that, given a certain confidence level or return period, the magnitude of the
extreme wind speed strongly depends on the shape (k ) and scale (C ) parameters of the parent Weibull
distribution. Smaller shape parameters lead to higher extreme values and vice versa. The influence of k
becomes pronounced when k < 1:8.
Comparison with the IEC values shows that the reference wind speed Vref is acceptable for sites with
k 1:77, i.e. for most flat terrain sites in Western Europe. This is also the case for the extreme 3-second
average wind speeds Ve1 and Ve50 (the so-called ‘survival wind speeds’) given in the IEC Extreme Wind
speed Model (EWM). But it should be noted that the annual extreme 3-second average wind speed with
a recurrence period of one year, Ve1 , has been redefined to the “annual characteristic largest 3-second
average wind speed.”
If a wind turbine has to be designed for complex terrain (1:4 k 1:5), it is recommended to increase
the ratio of the reference wind to the annual average wind speed to
Vref
Vave = 6:6:
In that case, the equations for Ve1 and Ve50 , given by the IEC, still hold.
Assuming longitudinal wind speed fluctuations uT1 and uT2 at the same location but at different instants
t1 and t2 to be jointly normal, the conditional pdf f (uT2 ; uT1 j uT1 ) is derived and the limit distribution
function computed through numerical integration. For extreme gusts not only a strong dependence on
the parameters of the parent Weibull distribution is found but also on the assumed turbulence model, viz.
Von Kármán or Kaimal, which determines the temporal correlation as function of the time lag = t2 ; t1
and the integral time scale, which is equal to L=U when Taylor’s hypothesis of frozen turbulence applies.
The magnitude of the extreme gust shows the same dependence on the Weibull parameters as the extreme
wind speed. The gust speed increases for increasing time lag until the time lag is greater than the
integral time scale, from that point on the the gusts speed does not change very much. Given the same
confidence level, gust values are greater for a starting wind speed at Vrated than at Vout .
The values found for the extreme 3-sec average gusts with a confidence level of 98% are much larger than
the values specified by the IEC 1400-1 standard. This was to be expected because experience using the
IEC standard has shown that the Normal Turbulence Model (NTM) leads in most cases to higher loads
than the extreme operating gust, indicating that the extreme operational gust model is too benign with
respect to actual external conditions1 .
New analysis of the extreme operating gust, performed by a working group of the IEC (TC88/WG7),
based on measurements from the German lowland and from Californian sites, shows a ratio of gust
1
Assuming that extreme sinusoidal gusts induce extreme loads.

magnitude (Vgust1 ) to standard deviation (1 ) of about 4.8, indicating that ‘real’ extremes (which are
generally not measured) will yield higher ratio’s, closer to values found in the present project, which are
estimated to vary between 4.4 for wind turbine class IV and 7 for wind turbine class I.
The analysis presented here could be refined in two ways. Firstly we should replace the single point gust
by a so-called coherent gust. To compute the distribution function of coherent gust we have to take into
account the spatial correlation function. Secondly the 10-minute average wind speed should be restricted
to U Vout . In the present computations U varies in theory between zero and infinity, but in practice
machine accuracy limits that range to much lower values. Both refinements will lead to lower gust values.
The gust values presented here should be regarded as upper limits; in design calculations we do not need
to evaluate larger gusts. When the gust is found that induces the highest loads, the distribution function
could be used to compute the number of occurrences.
It is shown that there are several approaches to derive the pdf of wind direction change which will not
yield the same results.2 The derivation of f ( ) that is given here can be found in the literature. It can
be shown that this angular pdf only depends on the reciprocal value of the turbulence intensities and the
non-dimensional ratio of the standard deviation of the u and v component of the wind (a measure of the
anisotropy of the fluctuations). This pdf allows for the calculation of the standard deviation of the wind
direction fluctuations as function of the turbulence intensity.
Because of the non-Gaussian tail of f ( ) it cannot be assumed that the density function f (1 ; 2 ) is
jointly normal and it turns out to be impossible to derive, from first principles, this density function in
closed form. Numerical integration proved to be too costly.
To circumvent this problem an exponential type density function is fitted to f ( ) and the extreme value
analysis is performed in much the same way as for the extreme gusts. This approach, however, requires the
presently unknown PSD of wind direction fluctuations in order to compute the correlation function. Given
the uncertainties in the analysis the comparison with the IEC values is very limited. Computations are
performed assuming that with good approximation the correlation function of the longitudinal component
can be used instead.
The present approach allows for a very simple construction of time histories of wind direction change.
It turns out that the computations for a turbulence intensity of 15%, a confidence limit of 95% and an
integral time scale L=U = 10 results in amplitudes and rise times which are in good agreement with the
IEC values if a starting value 1 = 90 deg. is taken.
The expression derived here for the distribution function of the wind direction change is not quite
satisfactory however. It is recommended to look into this problem once again.
Finally some attention is given to some rare meteorological phenomena – tornadoes and downbursts –
which quite likely will induce extreme wind loads, but which are not included in the extreme climate
events described by the IEC standard. Given the high price it takes, it is questionable whether wind
turbines should be designed to survive a tornado, but it seems likely that they should be designed to
withstand an intense microburst.
One of the main reasons for this assumption is that, although microbursts are sometimes difficult to
distinguish from tornadoes, they are pobably less rare and the associated wind speeds compare quite well
with the extreme gust speeds established here. In addition it seems quite likely that this phenomenon
induces extreme loads due to the combination of a strong vertical wind speed, high gust speeds in all
directions and a vortex with a horizontal axis. It is therefore recommended to study this phenomenon in
more detail, both physically and statistically, and to determine what its impact on a wind turbine could
be.
2
Actually the same remark applies to the pdf of gusts

A NOTATION AND BACKGROUND
A.1 Symbols and Definitions
For the definitions and notation we follow Castillo [3], Papoulis [18], and Spiegel [21].
f (x) (First-order) probability density function (pdf).
f (x; y) Second-order (joint) probability density function.
fx(x); fy (y) Marginal probability density functions.
k Shape parameter of the Weibull distribution.
n Number of independent observations in a sample.
r Probability of exceedance (risk).
x Possible value of the random variable X .
C Scale parameter of the Weibull distribution.
F (x) (First-order) probability distribution function (cdf).
F (x; y) Joint probability distribution function.
Fx (x); Fy (y) Marginal probability distribution function.
FXn (x) Extreme value distribution for the largest values of X .
Pr Abbreviation for “probability”.
Tp Sample period (generally one year, but 20 and 50 years are also u sed).
Tr Return period.
X Random variable.
Euler’s constant = 0.5772156649...
Scale parameter of the Gumbel distribution.
Location parameter of Gumbel distribution.
Mean of the normal distribution.

Standard deviation of normal distribution.
; Gamma function.
A.1.1 Distribution and density functions
A random variable (abbreviation : RV) is a number X (x) assigned to every outcome x of an experiment.
All random variables will be written in capital letters. The elements of the set S that are contained in
the event fX xg change as the number x takes various values. The probability PrfX xg of the
event fX xg is therefor a number that depends on x. This number is denoted F (x) and is called the
(cumulative) distribution function (abbreviation : cdf) of the RV X .
Definition A.1 (Distribution function) The distribution function of the RV X is the function
F (x) = PrfX xg (A.1)
defined for every x from ;1 to 1.
The derivative
f (x) = dFdx(x) (A.2)
of F (x) is called the (probability) density function (abbreviation : pdf) of the RV X . From the mono-
tonicity of F (x) it follows that
f (x) 0 (A.3)
Integrating (A.2) from ;1 to x and using the fact that F (;1) = 0, we obtain
Z x
F (x) = f ( ) d: (A.4)
;1
From (A.4) it follows that Z x2
F (x2 ) ; F (x1 ) = f (x) dx: (A.5)
x1
Hence Z x2
Prfx1 < X x2 g = f (x) dx: (A.6)
x1
At points of continuity of f (x), the in (A.6) can be replaced by < if desired.
A.1.2 Joint statistics
The joint distribution F (x; y ) of two RVs X and Y is the probability of the event
fX x; Y yg = f(X; Y )Dg
where x and y are two arbitrary real numbers and D a specified region :
F (x; y) = PrfX x; Y yg (A.7)

The function F (x; y ) is such that
F (;1; y) = 0 F (x; ;1) = 0 F (1; 1) = 1 (A.8)
The joint density of X and Y is by definition
f (x; y) = @ @x@y
2 F (x; y )
(A.9)
From this property it follows that

Z x Z y
F (x; y) = f (; ) d d (A.10)
;1 ;1
and
Z x Z 1
PrfX xg = Fx (x) = f (; ) d d (A.11)
;1 ;1
Z 1 Z y
PrfY yg = Fy (y) = f (; ) d d (A.12)
;1 ;1
We call Fx (x) and Fy (y ) the marginal distribution functions (or simply the distribution functions) of X
and Y respectively. The derivatives of (A.11) and (A.12) with respect to x and y are then called the
marginal density functions (or simply density functions) of X and Y . We maintain that
Fx (x) = FR 1(x; 1) Fy (y) = FR 1(1; y)

fx (x) = ;1 f (x; y) dy fy (y) = ;1 f (x; y) dx (A.13)
Cf. Papoulis [18] and Spiegel [21].
A.1.3 Order Statistics from Independent and Identically Distributed Samples
Consider a series of independent observations belonging to the same population and divided into samples
each containing n observations.
Definition A.2 (Order statistic) Let (X1 ; X2 ; : : : ; Xn ) be a sample from a given population. If the values
of the sequence X1 ; X2 ; : : : ; Xn are rearranged in an increasing order X1:n X2:n : : : Xn:n of
magnitude, then the r -th sample of this new sequence is called the r -th order statistic of the sample.
Note that the sample size n is included in the notation Xr:n and that any order statistics must have an
associated sample size.
The two important members of the sample
X1:n = min (X1 ; X2 ; : : : ; Xn )

and
Xn:n = max (X1 ; X2 ; : : : ; Xn )
are called extremes of the random variable X and play an essential role in applications.

Now suppose that the members X1 ; X2 ; : : : ; Xn of the sample are independent and identically distributed
and that they come from a parent population of samples with cumulative distribution function (cdf) F (x).
If we fit separate distributions for the smallest values of the samples only and for the largest values of the
samples only the resulting distributions, FX1 (x) and FXn (x) are called the extreme value distributions for
the smallest and largest values of X , respectively. It is possible to express the extreme value distributions
FX1 (x) and FXn (x) in terms of the parent distribution FX (x).
To find the distribution of the largest value Xn , we use the relation
FXn (x) = PrfXn xg = Prfall Xi xg

= PrfX1 xg PrfX2 xg PrfXn xg; (A.14)
if all Xi are independent. This means that the probability of realizing Xn less than equal to x is the same
as the probability of realizing all Xi ; i = 1; 2; : : : ; n to be less than or equal to x. If the Xi is identically
distributed with the distribution function FX (x),
FXn (x) = [FX (x)]n : (A.15)
Cf. Castillo [3] and Rao [19].
A.1.4 Return period
The return period is defined as the average elapsed time between occurrences of an event with a specified
magnitude or greater. If an exceedance occurs every 50 years, on the average, then the probability that the
event occurs in any given year is 1/50 = 0.02 or 2 percent. Thus the return period Tr and the probability
of occurrence are related as
Tr = 1 1 ; F (x) (A.16)
The use of this term can lead to dangerous misunderstanding when one is not aware of the fact that Tr
is only the average value of the length of the time interval between two consecutive crossings. After
the occurrence of an exceedance there is a big chance, that before Tr has past, there is at least one more
exceedance. Rather than choosing a return period it is better to state the risk r one considers acceptable
that a certain value, xlimit say, is exceeded within Tp years. The corresponding return period Tr then
follows from the relation Tp
1 ; r = 1 ; T1 : (A.17)
r
Note that the risk r that xlimit is exceeded within 20 years is 33.2%, when a return period of 50 years is
specified Note also that a return period of one year (as the IEC 1400-1 standard in some cases specifies)
is rather difficult to understand unless we assume that the parent distribution and the limit distribution are
based on a sample length which is shorter than one year. Cf. Castillo [3], Rao [19], Wieringa [9].
A.1.5 Characteristic largest value
The particular value of the random variable X , denoted xn , is called the characteristic largest value for a
period of n units if the mean value of the number of exceedances of xn is unity. The characteristic largest

value gives an idea of the central location of the possible largest value. By definition the characteristic
largest value is given by the relation
n [1 ; F (xn )] = 1 (A.18)
If we consider n time periods, the distribution of the largest value of xn is given by
n
FXn (xn ) = [F (xn )]n = 1 ; n1 (A.19)
Thus the probability of exceeding the characteristic largest value in n time periods will be
n
1
1 ; FXn (xn ) = 1 ; 1 ; n (A.20)
As n tends to infinity, Eq. (A.20) converges to the value 1 ; e;1 0:6321. Cf. Castillo [3] and Rao [19].
A.2 The Normal Distribution
The normal cumulative distribution function (cdf) is given by

" #
p1 exp ; 21 t ; dt ; ;1 < x < 1;
Z x 2
PrfX xg F (x) = (A.21)
;1 2
where is the standard deviation and the arithmetical mean. The associated probability density function
(pdf) is " 2 #
1
f (x) = p exp ; 1
x ; ; ;1 < x < 1: (A.22)
2 2
The main parameters of the normal distribution are summarised in table A.1.
A.3 The Weibull Distribution
The Weibull cdf is given by

"
k #
x
F (x) = 1 ; exp ; C ; 0 < x < 1 ; C > 0 ; k > 0; (A.23)
where C and k are constants known as the scale and shape parameters, respectively. The associated pdf
is k;1 " #
k x k
f (x) = C C exp ; x C ; 0 < x < 1 ; C > 0 ; k > 0: (A.24)
The main parameters of the Weibull distribution are summarised in table A.1. For k = 2 the well known
Rayleigh distribution is found.

A.4 The Gumbel Distribution
The Gumbel or Fisher-Tippett type-I cdf of maxima is given by
F (x) = exp ; exp ;(x; ) ; ;1 < x < 1 ; > 0;

(A.25)
where and are constants known as the location and scale parameters, respectively. The associated pdf
is h i
exp ;(;x) exp ; exp ;(x;)
f (x) = ; ;1 < x < 1: (A.26)
The main parameters of the Gumbel distribution are summarised in table A.1.
Table A.1 Some parameters of the Weibull, Gumbel and Normal distribution
Weibull Gumbel Normal

;
Mean C ; 1 + k1 +
Median C (ln 2)1=k ; ln(ln(2))
Mode
; ; 2 2
Variance C 2 ; 1 + k2 ; ;2 1 + k1 6 2
Kurtosis 1.1396
Euler’s constant = 0:5772156649::

B DERIVATION OF GUMBEL DISTRIBUTION
The importance of the Gumbel distribution is due to its extreme value behaviour. It has been applied as
the parent distribution or as an asymptotic approximation, to describe extreme wind speeds, sea wave
heights, floods, rainfall, air pollution, geologic problems, etc. , see Castillo [3][Ch. 5.2].
For a derivation of the Gumbel distribution, based on order statistics, we follow Harris [14] who in turn
based his text on a derivation by Cramer [7]. This derivation is given in section B.1.
An alternative derivation is given by Davenport [8]. He uses exceedance statistics to derive an asymptotic
distribution and casts this Poisson distribution into a Gumbel-like form for a Rayleigh parent distribution.
In section B.2 we will derive the same asymptotic distribution based on exceedance statistics, showing
the underlying assumptions, and then cast this distribution into a Gumbel-like form for the more general
Weibull parent distribution. Davenport claims that his version of the Gumbel distribution (for a Rayleigh
parent) is more exact than the one based on order statistics. We doubt that. In both derivations assumptions
are made which are not entirely met in reality.
B.1 Derivation Based on Order Statistics
B.1.1 Order statistics and asymptotic distributions
The distribution function of the extreme value X1 and Xn becomes increasingly insensitive to the exact
shape of the common distribution function FX (x) as n tends to infinity. The limiting forms as n ! 1
are known as asymptotic distributions. The asymptotic distributions often describe the behaviour of the
random variable X1 or Xn reasonable well even when the exact shape of the parent distribution is not
known precisely. The asymptotic distributions are classified into three types based on the general features
of the tail part of the distribution of the random variable X see e.g. Fisher & Tippett [10] and Gumbel
[13]. Different types of asymptotic distributions have been used for different engineering applications.
The Gumbel or Fisher-Tippitt type-I asymptotic distribution for the maximum values is useful whenever
the right tail of the parent distribution FX (x) is unbounded (x ! 1) and is of an exponential type. In
such a case we can express FX (x) as
FX (x) = 1 ; e;h(x) (B.1)
where h(x) increases with x monotonically. The distributions such as normal, lognormal and Weibull
distributions belong to this category. If FX (x) is of the form given by Eq. (B.1) the asymptotic distribution
function of the largest value can be found as follows. If we assume that the observations in a sample are
independent and identically distributed then the exact distribution of the largest value is
FXn:n (x) = [F (x)]n = f1 ; exp[;h(x)]gn : (B.2)
To investigate what happens when n ! 1 we have to introduce the so-called characteristic largest value,
denoted by xn , cf. A.1.5. The characteristic largest value, xn , is defined by the relation
1 ; F (xn ) = 1=n = exp[;h(xn )]: (B.3)
Thus n

FXn:n (x) = 1 ; exp[ h (x n ) ; h ( x )] : (B.4)
n
Now since h(x) is a positive and increasing function of x, as x increases beyond xn , exp[h(xn ) ; h(x)]
1. A well-known result for the exponential function originally due to Cauchy is
n

lim 1 ; n = exp (;) (B.5)
n!1
and the convergence of this expression to the limit is such that if is small and n is large, the exponential
limit is a good approximation to the left hand side, the leading term being 2 =2n. Thus applying this
result to (B.4) gives
FXn:n (x) ' exp f; exp[h(xn ) ; h(x)]g : (B.6)
Note that apart from the Cauchy formula, (B.6) is exact—it contains no other approximations. Taking
logarithms twice gives
; lnf; ln [FXn:n (x)]g ' h(x) ; h(xn ) = h(x) ; ln(n): (B.7)
The right-hand side of (B.7) is now expanded in a Taylor series about x = xn and gives
0 00
; lnf; ln [FXn:n (x)]g = (x ; xn) h (1!xn ) + (x ; xn)2 h (2!xn ) + : : : (B.8)
The familiar Fisher-Tippett type-I asymptotic formula is obtained by truncating the Taylor series at the
first term, and thus
; ln f; ln[Fn (x)]g ' (x ; xn); (B.9)
where Fn (x) is used to denote the asymptotic form of FXn:n (x) and
= h0 (xn) = n(dF=dx)x=xn : (B.10)
Comparing (B.9) with (A.25) shows that

= 1 (B.11)
B.1.2 The Gumbel distribution for a Weibull parent
To find the shape;and

k
scale parameter of the Gumbel distribution for a Weibull parent we make use of the
fact that h(x) = Cx . Hence
k ;1
0
= h (xn) = k x n :
C C (B.12)
Because of (B.3) we find

n = exp [h(xn )] (B.13)
and 1
= xn = C (ln n) k : (B.14)
Substituting (B.14) in (B.12) gives:
= Ck (ln n)1; k :
1
(B.15)

B.2 Derivation Based on Exceedance Statistics
A series of independent and identically distributed observations will only yield a set of maxima that
conform to an asymptotic extreme value distribution if the maxima have been drawn from infinitely large
samples. In practice this is never achieved. If we take hourly averages for example, then there are
8766 observations per year3 and moreover these observations are not independent. It can be shown that
for hourly averages the number of independent observations is approximately 800. With such a small
number, the question arises whether extreme value theory is satisfied within acceptable limits. It depends
of course on how fast the extreme value distributions approach their limiting form.
To avoid these problems Davenport [8] uses exceedance statistics to derive a limit distribution for
extremes, which he casts into a Gumbel distribution for a Rayleigh parent. By doing so he finds
alternative expressions for the shape factor and the scale parameter of the Gumbel distribution.
B.2.1 Exceedance statistics
Critical wind speeds result in responses that exceed the limits beyond which structural damage may be
expected. We would like to know the frequency with which the wind speed exceeds the critical value V .
The statistical function V+ is defined to be the number of times per second that the process passes with
positive slope through the value V (see Figure B.1), and the associated concepts are known as the theory
of exceedance statistics.
Crossing analysis
Suppose that we could consider a sample of length T of the one-hour or 10-min. average wind speed U
as a sine wave of varying amplitude and phase, i.e. a narrow-band random process, see Fig. B.1. Suppose
Figure B.1 Typical sample of a narrow band process (Courtesy Ref. [17]).
that we enquire how many ‘cycles’ of U (t) have amplitudes greater than the level U = V during the time
period T . Let n+V (T ) denote the number of positive slope crossings of U = V in time T for a typical
sample and let the mean value for all samples be NV+ (T ) where
NV+(T ) = E [n+V (T )]: (B.16)

3
We assume that a year has on average 365.25 days.

Since the process is stationary, if we take a second interval of duration T immediately following the first
we shall therefore obtain
N + (2T ) = 2N + (T )
V V
from which it follows that the average number of crossings is proportional to the time interval T . Hence
NV+ (T ) / T
or
NV+ (T ) = V+T (B.17)
where V+ is the average frequency of positive slope crossings of the level U = V . We now consider how
the frequency parameter V+ can be deduced from the underlying probability distribution for U (t).
Consider a small length of duration t of a typical sample function. Suppose that U < V at the beginning
of the interval, time t. There is a minimum slope at time t if the level U = V is to be crossed in time t
depending on the value of U at time t, see Fig. B.2. This is
Figure B.2 Conditions for a positive slope crossing of U =V in time interval t (Courtesy Ref. [17]).
V ;U
t
and so there will be a (high probability of a) positive slope crossing of U =V in the next time interval
t if, at time t,
U <V and dU > V ; U : (B.18)
dt t
The probability that both conditions are satisfied can be calculated from the joint probability density
function p(U; U_ ), hence
( ) Z 1 Z V
Pr Positive slope crossing
of U = V in time t
= Pt = dU_ p(U; U_ )dU (B.19)
0 V ;U_ t
and thus it follows that 1
P (V ) = lim Pt = Z
_ (V; U_ ) dU;
Up _
t!0 t
(B.20)
0
when the term p(V; U_ ) is understood to mean the joint probability density function p(U; U_ ) evaluated at
U = V , see Newland [17] [Ch. 8].
Now let T be the time within a total period T which U spends in the domain (V ; U_ t
U V; U_ > 0) in (U; U_ ) space. Then T=t is the average number of crossings of the interval
V ; U_ t U V , because we assume that U_ t is small and the process U (t) is smooth so occupation
of the interval is equivalent to crossing it in t.
By definition, Pt = T=T , and so Pt =t = (T=t)=T is the average number of crossings
of V in unit time. Thus Z 1
+_ (V; U_ )dU_
Up
V = (B.21)
0
This is a general result which applies for any probability distribution.
In order to calculate the exceedance curve + we must know the joint probability density function of U
V
and U_ . It is a property of any stationary random process x(t) that x and its derivative x_ are uncorrelated
E [xx_ ] = 0:
Hence we can write
p(U; U_ ) = p(U )p(U_ ): (B.22)
Substituting (B.22) in (B.21) yields
Z 1
V+ = p(V ) _ (U_ )dU_
Up (B.23)
0
Cf. Newland [17][Ch. 8].
B.2.2 Frequency of maxima
Our analysis of peaks is based on the assumption that the narrow band process resembles a sine wave of
varying amplitude and phase. If V+ is the average number of exceedances per unit time, then V+ T is the
average number of exceedances per time interval T .
Now divide such a time interval into n subintervals with length T = T=n. In such an interval there can
be one or more exceedances or no exceedance at all. We assume however
1. t is so small, hence n so large, that exceedance can only happen once4

2. the probability p of exceedance is equal for all subintervals.
3. the exceedances are independent.
In this case the stochastic variable NV+ (T ), which is the number of exceedances in time T , has a binomial
distribution. If we now let n ! 1, p ! 0 while pn = V+ T , we find the well known Poisson limit for
the binomial distribution.
Pr NV+ (T ) = N = (V T ) e;V T for N = 0; 1; 2; : : :;

+ N +
(B.24)
N!
where NV+ (T ) is the number of exceedances in time interval T. The probability that there will be no
crossings of the level U = V in the time interval T is
+
Pr NV+ (T ) = 0 = e;V T for T 0 (B.25)
4
In the sense that the probability of exceedance twice or more is extremely small.

Hence, the probability that there will be one or more crossings in time T is then simply
+
Pr NV+ (T ) > 0 = Pr fU > V g = 1 ; e;V T (B.26)
Because Pr fU > V g = 1 ; F (V ) we find that the cdf F (V ) is

F (V ) = e;V+ T (B.27)
Limit distribution for a Weibull parent

Suppose that the one-hour or ten-minute average wind speed U is Weibull distributed with shape factor
k and scale parameter C and suppose that the derivative of U is also Weibull distributed with the same
shape factor k and a different scale parameter C 0 . Hence
k;1 k" #
k
p(U ) = C CU exp ; UC (B.28)
and !k;1 2 !k 3
_ _
p(U_ ) = Ck0 CU0 exp 4; CU0 5 (B.29)
Substituting (B.28) and (B.29) in (B.23) yields

" # !k " #
k;1 k Z 1 _ k
dU_
+
V = Ck VC exp ; VC C 0 k CU0 exp ; CU0 C0
0
k;1 " #
0 k
= ; k1 + 1 k CC VC exp ; VC (B.30)
Put
C 0 = U_ = (B.31)
C U
and note that is a frequency. We will call the effective frequency and if U_ and U are not known we
can find alternatively from:
2 S (f ) df 1=2
"R 1 #
0 f
= R 1 S (f ) dfUU ; (B.32)
0 UU
because the spectral density of U_ is just f 2 times the spectral density of the original process U . Substitution
of (B.30) in (B.27) and noting that T = n gives
" (k k;1 !)#
F (> V ) = exp ; exp ; C + ln Tk; k + 1 VC
V 1 (B.33)
To cast this expression into a Gumbel-like form we apply the same procedure as in the previous section.
; k
Put h(x) = VC . Then taking logarithms twice gives:

; ln f; ln [F (> V )]g = h(x) ; ln nk; k1 + 1 ; ln h(x);1 (B.34)

Expanding the right hand side in a Taylor series about x = xn and truncating at the first term and some
rearranging gives
; ln f; ln [F (> V )]g ' 2 8 9

(B.35)
h ;1 k;1 i 3
ln k; k + 1 (ln n)
;h (x ) 1 ; k ; 1 1
>
< k > =
0 n x ; xn >1 +
6 7
k ln n 4
: k ln n ; (k ; 1) >
;
5
Comparing (B.36) with (B.9) shows that

k ;

1 1
Dav = 1 ; k ln n

(B.36)
and 8 h k;1 i
9
;
>
< ln k; k1 + 1 (ln n) k > =
Dav = >1 + k ln n ; (k ; 1) >
(B.37)
: ;
Typically, for k = 2, Dav is about 5% greater than and Dav is about 10% greater than . The increase
becomes smaller for decreasing k .

References
[1] Bergström, H. “Distribution of Extreme Wind Speed”. Wind Energy Report WE 92:2, Uppsala
University, Department of Meteorology, Uppsala, Sweden, 1992.
[2] Butterfield, S., B. Holley, P. Hauge Madsen, and C. Stork. “Report on 88/69/CD–Wind Turbine
Generator Systems Part 1: Safety Requirements, 2nd edition”. unpublished report, 1996.
[3] Castillo, E. Extreme Value Theory in Engineering. Academic Press, Inc., 1988.
[4] Chaviaropoulos, P.K. “Probabilistic Analysis of Extreme WInd Events”. Wind Engineering,
20(3):139–159, 1997.
[5] Collier, C.G., J. Dixon, M.S.J. Harrison, J.C.R. Hunt, J.F.B. Mitchell, and D.S. Richardson. “Extreme
surface winds in mid-latitude storms: forcasting and changes in climatology”. J. Wind Eng. Ind.
Aerodyn., 52(Complete):1–27, 1994.
[6] Cook, N.J. “Towards better estimation of extreme winds”. J. Wind Eng. Ind. Aerodyn., 9:295–323,
1982.
[7] Cramer, H. In Gumbel, E.J., editor, “Statistics of Extremes”, page 170. Columbia Univ. Press, New
York, 1958.
[8] Davenport, A.G. “The Dependence of Wind Loads on Meteorological Parameters”. In Proc. Int.
Res. Seminar, pages 19–82, Ottawa, 1967. University of Toronto Press. Wind effects on buildings
and structures.
[9] Wieringa, J en P.J. Rijkoort. “Windklimaat van Nederland”. Staatsuitgeverij ’s-Gravenhage, 1983.
[10] Fisher, R.A. and L.H.C. Tippett. “Limiting Forms of the Frequency Distributions of the Largest or
Smallest Member of a Sample”. Proc. Cambridge Philos. Soc., 24:180–190, 1928.
[11] Fujita, T.T. “The downburst”. SMRP Research Paper Number 210, Department of the Geophysical
Sciences, The University of Chicago, Chicago, Illinois 60637, 1985. Published by Satellite and
Mesometeorology Research Project (SMRP).
[12] Gradshteyn, I.S. and I.M. Ryzhik. Table of Integrals, Series and Products. Academic Press, Inc.,
1980.
[13] Gumbel, E.J. Statistics of Extremes. Columbia Univ. Press, New York, 1958.
[14] Harris, R.I. “Gumbel re-visited – a new look at extreme value statistics applied to wind speeds”. J.
Wind Eng. Ind. Aerodyn., 59(1):1–22, 1996.
[15] Ibarra, J.I. “A New Approach to the Determination of Horizontal Wind Direction Fluctuations”. J.
of Applied Meteorology, 34:1942–1949, 1995.
[16] International Electrotechnical Commission. “International Standard IEC 1400-1; Wind turbine
generator systems – Part 1: Safety requirements”, first edition, 1994.
[17] Newland, D.E. Random Vibrations and Spectral Analysis. Longman, Inc., New York, second
edition, 1984.

[18] Papoulis, A. Probability, Random Variables, and Stochastic Processes. Electrical Engineering
Series. McGraw-Hill, Inc., second edition, 1984.
[19] Rao, S.S. Reliability-based design. McGraw-Hill, Inc., 1992.
[20] Rice, S.O. “Mathematical Analysis of Random Noise”. Bell System Tech. J., 23:282–332, 1944,
and 24:46–156, 1945. both of which are reprinted in N. Wax (ed.), Selected Papers on Noise and
Stochastic Processes, Dover, New York, 1954.
[21] Spiegel, M.R. Probability and Statistics. Schaum’s Outline Series in Mathematics. McGraw-Hill,
Inc., 1975.
[22] Weber, R. “Estimator of the Standard Deviation of Wind Direction Based Moments of the Cartesian
Components”. J. of Applied Meteorology, 30:1341–1353, 1995.
[23] Winkelaar, D. “Some Ideas on the Modelling of Atmospheric Turbulence”. In IEA Joint Action on
Wind Conditions for Wind Turbine Design; 1st Symposium, Hamburg, Germany, June 1994.

PART 2
QUANTIFICATION OF FAILURE PROBABILITIES
H. Braam (editor) Netherlands Energy Research Foundation ECN

A.J. Seebregts Netherlands Energy Research Foundation ECN
M.W. Jensen RISØ National Laboratory
P. Christensen RISØ National Laboratory
C.J.C. Christensen RISØ National Laboratory
E. Hinrichsen Germanischer Lloyd
Quantification of failure probabilities
2.2 European Wind turbine Standards II

CONTENTS OF PART 2
1. INTRODUCTION......................................................................................................................................................5
1.1 Background ..........................................................................................................................................................5
1.2 Objectives.............................................................................................................................................................5
1.3 Methodology ........................................................................................................................................................6
2. SCOPE AND FIELD OF APPLICATION ................................................................................................................7
3. DEFINITIONS ...........................................................................................................................................................9
4. SYMBOLS AND ABBREVIATIONS ....................................................................................................................11
5. RELIABILITY OF LOAD CARRYING COMPONENTS .....................................................................................13
5.1 Target values for structural reliability................................................................................................................13
5.1.1 Introduction.................................................................................................................................................13
5.1.2 Reliability Methods.....................................................................................................................................13
5.1.3 Target Reliability’s in Structural Design Codes .........................................................................................14
5.2 Partial safety coefficients ...................................................................................................................................14
5.2.1 The Limit State - Probabilistic Design Requirement ..................................................................................14
5.2.2 Evaluating the failure probability ...............................................................................................................15
5.2.3 Partial coefficient method ...........................................................................................................................17
5.2.4 Calibration of partial coefficients ...............................................................................................................17
5.2.5 Target reliability and probabilistic safety calculations ...............................................................................18
6. RISK CRITERIA AND RISK TARGET VALUES ................................................................................................19
6.1 Recommendations for risk criteria and risk target values to be used for wind turbines.....................................19
6.2 Example mortality rate calculation for wind turbines ........................................................................................20
7. DATA COLLECTION AND PARAMETER ESTIMATION.................................................................................23
7.1 Introduction........................................................................................................................................................23
7.2 Design specific failure data................................................................................................................................24
7.3 Generic wind turbine failure data.......................................................................................................................24
7.4 Generic Failure Data from Other Industries.......................................................................................................26
7.5 Summary of Guidelines......................................................................................................................................27
8. QUANTIFICATION OF MODELS (BENCHMARK STUDY) .............................................................................29
8.1 Outline of Benchmark ........................................................................................................................................29
8.2 Benchmark results..............................................................................................................................................29
8.3 Conclusions and recommendations....................................................................................................................34
9. CONCLUSIONS AND RECOMMENDATIONS...................................................................................................35
10. REFERENCES .......................................................................................................................................................39
Appendix A. REVIEW OF SAFETY AND RISK CRITERIA ...................................................................................41

Appendix A.1 Target reliability’s in various countries and in various branches of industries ................................43
Appendix A.2 Risk criteria in the various countries.................................................................................................49
Appendix A.3 Mortality rates observed...................................................................................................................52
Appendix A.4 Wind turbine failure data..................................................................................................................52
Appendix B. MORTALITY RATE CALCULATION ................................................................................................55
Appendix B.1 Assumptions .....................................................................................................................................55
Appendix B.2 Calculational procedure....................................................................................................................55
Appendix B.3 Reference..........................................................................................................................................56
European Wind turbine Standards II 2.3

Appendix C. BENCHMARK ON DATA COLLECTION AND PARAMETER ESTIMATION .............................57

Appendix C.1 Guidance for benchmark exercise ....................................................................................................57
Appendix C.1.1 List of Basic Events (BEs) from EWTS-I case study................................................................57
Appendix C.1.2 Information needed for each BE ...............................................................................................58
Appendix C.1.3 Sources of information ..............................................................................................................58
Appendix C.1.4 Failure type model.....................................................................................................................58
Appendix C.1.5 Example of data to be provided for each BE: ...........................................................................59
Appendix C.2 Overall approach ECN......................................................................................................................59
Appendix C.3 Overall approach RISØ ....................................................................................................................61
Appendix C.4 Data and results ................................................................................................................................65
Appendix D. DERIVATION OF FAILURE RATES FROM REPORTED FAILURES ............................................69

1. INTRODUCTION
1.1 Background
Presently, wind turbines are being designed in accordance with deterministic design rules embedded in standards
like the IEC 1400-1 [1.1] and various national standards and certification criteria. These rules concern the design of
the load carrying components and the design of safety and control systems.
Design of load carrying components and partial safety factors
For a safe design of load carrying components a margin is introduced between the design value of the strength and
the characteristic value of the load. Over the years various methods to define safety margins have been used, but
nowadays the concept of the partial safety factors is commonly embodied in the structural design codes. These
partial safety factors ensure that there is a safety margin both on the load side and on the strength side. The
magnitude of the partial safety factors has to reflect the required safety level and likewise the uncertainty in the load
and the strength should be taken into account. This means that the partial safety factor for wind loads has a larger
value than for well defined static loads, like the self-weight. The determination of the magnitude of the partial safety
factors can be done empirically. However, a probabilistic approach is preferred, because in this way uncertainties in
the load and the strength can be considered in a more rational manner. This probabilistic method has been applied
already in other branches of industry like offshore and civil engineering, but is not introduced in the wind energy
branch yet. As the concept of probabilistic calibration of partial safety factors is fairly new in the wind turbine
community the principles behind the method were examined as part of the EWTS-I subproject “Calibration of Safety
Factors” [1.2]. To calibrate partial safety factors based on a probabilistic method target values for the structural
reliability of load carrying components have to be drawn up. This aspect was not considered in the EWTS-I project.
Control and safety systems, beyond design situations
Generally, the control systems have to keep the wind turbine within its design limits during normal operation and to
ensure that electrical energy is generated and supplied to the grid in an efficient manner. In case of a failure or in
case of abnormal external conditions, the control systems in combination with the safety systems have also to keep
the turbine within its design envelope to prevent damage or unsafe operation. As a consequence of a failure of a
safety system the wind turbine can run into overspeed and is operating beyond design conditions. This can lead to
catastrophic failures like blade rupture. Scenarios how a wind turbine can run into overspeed or failed operation can
be derived by means of a qualitative Probabilistic Safety Assessment (PSA). With a quantitative PSA it is also
possible to calculate the frequency of occurrence of for example overspeed. PSA’s for wind turbines have been
carried out successfully in various countries (e.g., see [1.3]). Because it is expected that PSA’s will be used more
and more the guidelines developed in the various countries should be harmonised to obtain unified assumptions and
comparable results. For this reason guidelines or recommended practices were formulated as part of the EWTS-I
subproject “Assessment of Wind Turbine Safety; Recommended Practices” [1.2]. To illustrate the method a limited
case study was performed for a ‘postulated’ design, but with similar characteristics to the AOC 15/50 [1.4]. The
results of that case study are given in part 4 of [1.2]. However the development of guidelines for the derivation of
data for the quantification was beyond the scope of the EWTS-I project.
Risks to the public
As mentioned above failure of the safety system might lead to catastrophic failures like blade rupture, which can
cause risks to the public. In that way it is of great interest for e.g. authorities, especially in densely populated areas.
To judge whether a risk is unacceptable limits have to be indicated above which the risks become unacceptable
(maximum permissible levels).
1.2 Objectives
The objectives of the 2nd EWTS-II subproject “Quantification of Failure Probabilities” are two-fold, viz.:
• derivation of target values for structural reliability and for risk criteria on which the designer should aim and
which the certifying bodies can use as a criterion;
• preparation of guidelines to collect data and to estimate parameters to be used for quantitative safety and

reliability analyses (PSA’s).
1.3 Methodology
The approach taken was a literature study to obtain target values, followed by a limited case study to obtain
guidelines for data collection.
Target values
The literature study included references from various countries and branches of industries and resulted in an
overview of:
• target reliability’s (screening of available codes for building and construction to establish reliability levels for
the design of load carrying components);
• risk criteria and risk figures used and observed (review of national and international regulations on risk limits
and references for observed risk figures, e.g., mortality rates);
• wind turbine failure data (review of national and international generic data on wind turbine failures, e.g., actual
frequency of occurrence of catastrophic blade failures, failure of main components or safety systems).
In appendix A of this report the references used are listed together with the corresponding summaries. From these
reviews conclusions and recommendations for target values are derived.
Chapter 5 presents the recommendations for target values for structural reliability. In addition, it presents an outline
how these target values can be used to calibrate partial safety factors.
Chapter 6 presents the conclusions and recommendations for risk criteria and mortality rates. Fatalities of the
‘public’ as a result of wind turbine accidents are not known. Only personnel was involved in such cases, which is a
labour safety item and outside the scope of this project. The ‘public’ mortality rate for wind turbines can therefore
only be assessed by means of an analysis. To illustrate the possibilities and the limitations of such a mortality rate
analysis, a simple calculation example is presented in appendix B.
Guidelines for data collection and parameter estimation
Guidelines to collect data for quantitative safety and reliability analyses were formulated based on a limited
benchmark study (essentially with the same design and with the same scope as in the EWTS-I subproject
‘Assessment of Wind Turbine Safety: Recommended Practices’). ECN and Risø participated in the benchmark
study.
Chapter 7 first presents an overview of the various data sources that are available and finally the resulting set of
practical guidelines for collecting data and estimating parameters for quantitative safety analyses.
Chapter 8 summarises the results of the benchmark study. Based on this study, appendix C provides more detailed
guidelines for data collection and parameter estimation, and it describes the results in more detail. The guidelines
formulated are based on previous experience present at ECN (seven quantitative wind turbine safety and reliability
studies) and the experience obtained within this project (jointly by ECN, GL, and Risø).

2. SCOPE AND FIELD OF APPLICATION

The target values for structural reliability are primarily meant for the calibration of partial safety factors using a
probabilistic method. As part of the EWTS-I project a procedure for this calibration was already developed, however
the target values itself were not considered at that time.
The risk of a wind turbine can be assessed by means of a quantitative Probabilistic Safety Assessment (PSA). As
part of the EWTS-I project guidelines and recommendations were defined to perform such a PSA. However, these
recommendations were applied to a limited extend only. In the current report the applicability of these
recommendations was considered by performing a case study for a postulated although realistic design of a wind
turbine. With the results of this case study the recommended practices from the EWTS-I project were adjusted and
extended.
Risk is defined in this project as: Risk to the public, and to be more precise as the probability of dying as a result of a
wind turbine incident or accident. Risk to personnel (labour safety) or economic risk or the risk of loosing the
structural integrity of the wind turbine is not considered here.
The target values for the reliability and safety of wind turbines are strongly related to the requirements in national
and international standards and European directives. The recommended target values can serve as a starting point for
standardisation in this field.
The recommendations for quantitative safety and reliability analyses together with the guidelines for data collection
and parameter estimation can be used by manufactures and designers to assess the reliability of advanced protection
systems and to identify critical components or weak spots in the design.


3. DEFINITIONS
Environmental
Impact Statement : Study and report in which the environmental impacts of an (industrial) activity are
outlined and shown below acceptable levels.
Risk : Risk is defined as the unwanted consequence of a particular activity in relation to

the likelihood that this may occur.
Risk to the public : Risk posed to the group of people outside the premises of the installation, the
consequence here referring to death.
Group risk : The likelihood per year that a group of at least a certain size will all be victim of a single
accident at one and the same time. The group risk should indicate the probability that a certain
group of people outside a facility will die due to an accident caused by the facility. As such,
this risk takes account of the surrounding area. Group risk does not involve the calculation of
the individual risk but rather the risk of a number of deaths. In addition, the group has to be a
fixed and identifiable group of persons, viz. the group of persons actually living at/near the
corresponding site.
Individual risk : The likelihood that a person will suffer a given detrimental effect (here: death) as a result of
exposure to an agent (expressed in probability units per year or related to an average
concentration per year).
The IR is calculated for the person who runs the greatest risk at a given location. In practice,
this normally refers to an individual who is present at a particular place in the vicinity of an
industrial site 24 hours per day.
Mortality rate : Number of deaths per year for a given population.
Risk criteria : A measure of risk, e.g., group risk or individual risk.
Risk targets : A level for the risk value which is aimed at not to exceed.
Risk limits : A strict level for the risk value which is not to be exceeded, e.g., in the Netherlands the limit
for individual risk is 1e-6/yr.
Target reliability : Maximum failure probability prescribed in a code.


4. SYMBOLS AND ABBREVIATIONS

ALARP: As Low As Reasonable Practicable
BE : Basic Event
ET : Event Tree
FT : Fault Tree
FORM : First order reliability method
HSE : UK Health and Safety Executive
PSA : Probabilistic Safety Assessment
QRA : Quantitative Risk Assessment
WTB : Wind Turbine
β : Reliability index


5. RELIABILITY OF LOAD CARRYING COMPONENTS

5.1 Target values for structural reliability
5.1.1 Introduction
Structures are generally designed with design codes or building codes, which are prescribed by the relevant national
legislation. To obtain a certain level of safety against uncertainties and tolerances in the material properties and the
loading, safety factors are included in the design formulas of these codes.
Traditionally this has been a single, global safety factor to account for all uncertainties. Modern design codes apply
the so called partial safety factor method, which uses separate safety factors for the material properties and the
loading. The advantage is that uncertainties, especially in environmental loads, can be better accounted for. The
disadvantage is, that it requires more work. A further problem is that these partial safety factors need to be
calibrated, to ensure that the desired safety level is met. This calibration is done through a probabilistic method
based on a certain target reliability.
Following a brief outline of structural reliability methods is given, in order to have a better understanding of the
concepts and terms used, see also [5.1] for this purpose. Then an overview is given of target reliability’s as used in
design codes in different countries and different industries. This is based on the summaries of references contained
in appendix A. This may then be used by regulatory bodies to define target reliability values for wind turbines, since
this is outside the scope of this sub-project.
5.1.2 Reliability Methods

System Reliability
In ISO 2394 [5.3] this is defined as: ‘The reliability of a structural system which has more than one relevant failure
mode or the reliability of a system of more than one relevant structural element’. Although this is relevant to wind
turbines, especially if one considers the safety and control systems, there are to our knowledge, no national codes
available that specify target values for system reliability. We found two general references that would give target
values for control systems, one for safety systems of nuclear power plants and one for the safety system of a wind
turbine.
Structural Reliability
The term structural reliability may be defined as follows: ‘The probability that the loading of a structure does not
exceed its strength for a specified period of time’ [5.2]. In order to be able to determine this probability, a set of so
called ‘limit states’ for the loading (L) and the resistance ® of the structure in question must be formulated. Then the
probability that (L-R) is positive can be calculated and compared to a predefined target value. It should be noted that
the limit state of a structural element may be exceeded by ultimate loads or through fatigue damage, both must be
considered. This is also the key principle of the proposed revision of ISO 2394 [5.3], which defines structural
performance as follows: ‘The structural performance of a whole structure or part of it should be described with
reference to a specified set of limit states beyond which the structure no longer satisfies the design requirements’. A
distinction is made between ‘ultimate’ and ‘serviceability’ limit states. Ultimate limit states correspond to the
maximum load carrying capacity of the structure, whereas serviceability limit states concern the normal use of the
structure, see [5.3] for examples. For complex structures it is normally not feasible to solve the limit state function
exactly, due to the excessive computation times required for the numerical integration. Typically these functions are
solved by applying FORM (First Order Reliability Methods), which means approximation and simplification of the
complex integrals involved. A commonly used presentation of the failure probability is the reliability index, β. The
failure probability, PF, and the reliability index are related through the standardised normal distribution, Φ, by
β = Φ −1 ( PF )

5.1.3 Target Reliability’s in Structural Design Codes

An extensive literature review has been performed with the aim to find target reliability’s from different industries
and different countries, see appendix A. It became obvious that, although most of the current design codes reviewed
were using partial safety factors to a varying degree, and some also mentioned that these have been calibrated
through probabilistic methods, but they are very reluctant to mention an explicit value for target reliability.
Following is a summary of what was found
Target Values for System Reliability
In Ref. (A1.2) in appendix A it is stated that the unavailability of the safety system of a nuclear power plant shall be
less than 1E-3 to 1E-4 per demand. In Ref. (A1.12) and (A1.13) in appendix A it is mentioned that the unavailability
of the safety system of a wind turbine shall be less than 2E-4 per demand.
Target Values for Structural Reliability
In order to be able to compare the different figures, we shall give all target values in the form of the safety index β.
Furthermore we shall consider only the safety indices that correspond to the ultimate limit states and not to the
serviceability limit states, since those will not result in a risk to the public.
Ref. (A1.1), Netherlands Building Codes:
Here the average safety level of the former Dutch building code was determined to arrive at a target reliability with
which the partial safety factors of a new code were then calibrated. The result was an average β of 3.8, based on a
reference period of 50 years.
Ref. (A1.3), NEN 6700:
This reference actually defines a target safety level. Thereby three Safety Classes, depending on the consequences in
terms of personnel injury, are defined. The safety indices are given as 2.3, 2.4 and 2.6 resp. for the three safety
classes for wind load and as 3.2, 3.4 and 3.6 for other loads. The indices are based on the service life of the
structures.
Ref. (A1.5), NKB Nordic Committee for Building Regulations:
Here also three safety classes and a target reliability are defined, similar to the previous reference. The safety indices
are given as 3.71, 4.26 and 4.75 for the three safety classes. The reference period is one year.
Ref. (A1.10), Eurocode 1:
This gives a yearly target reliability of 4.7, corresponding to the NKB safety class ‘high’.
5.2 Partial safety coefficients

The target reliability level is a central concept for the partial coefficient method normally used for evaluating the
safety and reliability of a structure. In this chapter we shall give a short review of the use of target reliability levels
for calibrating partial coefficients in the partial coefficient method. A more comprehensive discussion was given in
European wind turbine standards, EUR 16898, part 2 (1996). Here a short overview is given with a mind to
bringing out the rationale in treating structural reliability in the context of probabilistic safety calculations. It should
be remarked that in this section the mathematical symbols are chosen as normally used in structural reliability. This
means that f denotes the resistance rather than a probability function.
5.2.1 The Limit State - Probabilistic Design Requirement

A key element of the reliability or safety evaluation for load carrying structures is the limit state. According to ISO
2394 [5.3], structural performance is defined by
The structural performance of a whole structure or part of it should be described with reference to a specified set of
limit states beyond which the structure no longer satisfies the design requirements.

The specification of one particular limit state out of the set calls for the following definitions:
• Failure type for limit state
The failure type can be subdivided into different types of rupture, plastic deformations, loss of stability etc.
• Cause of failure
Failure may happen for different reasons, each of which requires a separate evaluation of safety, e.g.
• ultimate (extreme) load failure caused by extreme loads that only appear few times in the life time of a wind
turbine (e.g. caused by a 50 yr. extreme wind speed)
• fatigue induced failure caused by many repeated changes of loads, often mainly during normal operating
conditions (e.g. every day power production, routine stops and starts)
• Design Requirement
The purpose of design calculations or prototype testing is to keep the probability of a limit state being reached
below a certain value prescribed for the type of structure in question (ISO 2394).
The probability of a limit state being reached is often called the failure probability, PF, whereas the maximum
allowed value, PL, is typically defined as the complement of the target reliability, PT. Thus the design requirement
can be written as
PF ≤ PL
or (5.1)
PF ≤ 1 − PT
5.2.2 Evaluating the failure probability

For the evaluation of the design requirement, a probabilistic formulation of the limit state condition is set up as
follows. To describe the structure, a set of stochastic basic variables, Xn, is selected, including, e.g. its strength,
stiffness, geometry, and (possibly several types of) loading. The n basic variables selected are grouped into one
vector X.
For the limit state under consideration, a so-called limit state function g(X) (also called performance function or
reliability function) is defined as a relation between the basic variables Xn with the following properties
 R > 0 for X in the safe set


g ( X) = S = 0 for X on the limit state surface (5.2)
T < 0 for X in the failure set

g(X)=0 can be seen as an equation determining the limit state surface in an n-dimensional space spanned by vectors,
each vector corresponding to one of the basic variables.
The limit state surface in (5.2) separates the possible realisations of X (i.e. the different value sets of X1, X2,.., Xn)
into two sets; the set for which the structure will be safe, and the set for which the structure will fail.
The limit state surface can be seen as a sharply defined limit of safety in the sense that we can assume that, if the
values of the basic variables were accurately known, an infinitesimal change of X from the limit state surface into
the failure region would mean assured failure. This is, however, an overly simplified mathematical abstraction. The
basic variables are only known with a certain uncertainty, i.e. they are stochastic variables. Therefore we can not talk
about assured failure or survival, but must evaluate the probability of failure.
Using the limit state function, (5.2), the probability of failure can be regarded as the probability that the limit state
function takes on negative values and should be calculated as the probability of the system being in the failure set
(the failure probability):
PF = P[g(X) ≤ 0] = ∫f X ( x )dx (5.3)

g( X )≤ 0
where fX(x) is the joint probability density function for X. The integral extends over the part of the n-dimensional

space defined by the n basic variables, where g(X) < 0.

The calculation of the probability of failure is in principle simple, in practice often very difficult. The difficulties are
concerned with, e.g., picking the relevant basic variables, deciding on the distribution functions for these variables
(the description of the low probability tails is important and complicated), doing the probabilistic integral in(5.3).
As the very simplest example with one load and one resistance variable, take (X1, X2) = (F, f):
g ( X) ≡ f − F (5.4)
F is the load variable, and f is the resistance variable. The variables are both stochastic variables which are
characterised by their distribution functions (shown full drawn) in Fig. 5.1.
The distribution function shown at high loads is the strength distribution, at lower loads the load distribution. For
both distributions the mean values are shown dashed. The 3rd dashed line between the two mean values the design
value for both load and resistance, which are identical in the limit state. The two dotted lines are the so-called
characteristic values, see next section, equation (5.5). The condition for failure according to (5.4) inserted in (5.2) is
then g(X) = f-F < 0 or f < F, i.e., resistance smaller than load. g(X) can also be seen as a derived stochastic variable
Z = f-F, often called the safety margin. The failure probability can then be found from the distribution function for
this Z function, which is also shown in Fig. (5.1) as the dot-dash curve. In this simple case, where we assume that f
and F are both normal distributions, it is simple to find the distribution of the difference, which will also be normal.
The mean value will be the difference between the mean values of the two, and the variance will be the sum of the
variances of the two elements.
1
F k f k
0.8
0.6
0.4
0.2
0
0 5 10 15 20 25 30
Figure 5.1 Distribution functions for load and resistance, probability per unit load vs. load.
0.1
0.01
0.001
4
1 10
5
1 10
6
1 10
0 5 10 15 20
Figure 5.2 Safety margin distribution, cumulative and logarithmic.
Fig. 5.2 shows the cumulative distribution corresponding to the differential safety margin distribution
in fig. 5.1.
The cumulative distribution gives the probability, that Z is less that the abscissa value, thus it is seen that Z < 0 with
a probability of 0.00005 in the case illustrated.

5.2.3 Partial coefficient method

For practical design calculations, simple approximation methods have been developed, e.g. the partial coefficient
method.
According to this method the design requirement for the limit state can in the very simplest cases be expressed by
the design inequality
f d = f k / γ m ≥ Fk γ f = Fd (5.5)
f and F designate values of resistance and load, respectively, index k indicates characteristic values and index d
design values. The γ’s are partial safety factors for load (γf) and for resistance (γm). Very simply stated the design
inequality, (5.5), expresses that the design value of the resistance function of a component should be stronger than
that of the load. The inequality (5.5) is turned into the design equation for the limit state by replacing the ≥-sign with
=.
This design equation (5.5) is an approximate simplification of the basic limit state equation (5.2) in the simplest case
with g(X) defined as in (5.4).
The design equation in the partial coefficient method is constructed such that when used in a (broad) region of the
variables, the target reliability is approximately obtained without doing the probabilistic calculations. The partial
coefficients will typically be available from existing standards for structural safety.
In the partial coefficient method, the stochastic parameters are described by means of characteristic values. These
values should be chosen as quantiles for the stochastic variables, resistance and load. Thus a typical characteristic
value for resistance is a 5% quantile. The meaning of this characteristic value is that out of 100 samples produced,
approximately 5 will not be as strong as indicated by the characteristic value.
The partial safety factors used in the design equation together with the chosen quantiles regulates the failure
probability obtained through the use of the design equation. In 5.2.1 the characteristic values are shown. Also the
design values derived by application of partial safety factors the characteristic values are shown. As the distributions
are chosen such that the design equation is valid, the two design values merges into one as shown.
The simple equation (5.5) is often not sufficient. In building codes, e.g., one could wish to treat several loads which
can interact
Fd = Fk / γ m ≥ f p ,k γ p, f + f v ,k γ v , k = f d (5.6)
Here it is assumed, that the two forces are a permanent force (e.g. gravity) and a variable force with their natural
uncertainties. The statistical properties of these two forces are very different, but if they are well defined, the partial
coefficients also serves to allow to add the characteristic values with suitable weights so that the combined action
represents the statistical combination of the two different forces.
5.2.4 Calibration of partial coefficients

The calibration of partial coefficients basically is the process of deciding on the partial coefficients such that the
target reliability wished is obtained. There are several ways to do that. For a probabilistic treatment, the process
contains a calculation of the failure probability, and adjusting the partial coefficients so that the wanted target
reliability is found. In Fig. 5.3 such a calibration is illustrated using the equation
∞
( )
Pf = ∫ 1 − Φ f ( x ) ⋅ ϕ F ( x )dx
0
(5.7)
Here the probability of failure is calculated as follows. A failure could happen at any value of the resistance x, if the
load happens to get larger than x. The probability of the resistance being around x if ϕF(x)dx. The probability that the
load surpasses can be found from the cumulative strength distribution (1-Φf(x)). Multiplying these two probabilities
the probability of a certain value of the resistance combined with the load surpassing the resistance is obtained.
Integrating this product over x gives the wanted failure probability. In each figure in Fig. 5.3 the resistance

distribution and the cumulative load distribution are shown full-drawn. Their characteristic values are shown dotted
as is the integrand of (5.7).
0.1 0.1 0.1
Fk fk Fk fk Fk fk
0.075 0.075 0.075
0.05 0.05 0.05
0.025 0.025 0.025
0 0 0
10 11 12 13 14 15 10 11 12 13 14 15 10 11 12 13 14 15
γ m .γ f = 1.32 γ m . γ f = 1.27 γ m.γ f = 1.21
6 5 5
Z 0 = 2 10 Z 0 = 1 10 Z 0 = 5 10
Figure 5.3 Illustration of the effect of decreasing the partial coefficients.

The product of the partial coefficients is varied (decreased) by 5% for each figure. The failure probability
(P[Z<0]=zo) was calculated in each case. If a target reliability corresponding to a failure probability (PL) of 1⋅10-5 is
prescribed, the centre situation in Fig. 5.3 is the result.
5.2.5 Target reliability and probabilistic safety calculations

It should be stressed that it is often said in codes that the failure probability we have talked about here is only a
formal mathematical abstraction. This is mainly because of the many difficulties and approximations involved in the
calibration process, but also because the limit state concept is specifically developed for use in structural reliability.
Often a structural failure can happen in many ways, that all have to be considered. This should, however, not prevent
us from using the concept also in our probabilistic safety analyses.
This has two reasons. First, a structural failure can very well be imagined as initiating event before a major accident.
Secondly, the process that leads us to using the probabilistic concepts in structural reliability analysis is so clearly
geared towards calculating the probability of failure, that there seams to be no other way. Therefore the
amalgamation of structural failure probability and other probabilistic safety calculations is absolutely natural, as long
as the necessary caution in using the structural failure concept is shown. It is, however, deemed to be outside reach
in this project to go into further examination of the peculiar problems when amalgamating these fields.

6. RISK CRITERIA AND RISK TARGET VALUES

The safety of a construction operating within the design limits can be checked by means of design codes. However,
situations might occur such that the construction is operating beyond the design limits. This can be caused by the
failure of the safety system or by loadings not foreseen in the design. Ultimately, this may lead to catastrophic
failures like blade rupture which can cause risks to the public. Such risks are of great interest for, e.g., authorities,
especially in densely populated areas. To judge whether a risk is acceptable limits have to be indicated above which
the risks become unacceptable (maximum permissible levels).
Section 6.1 provides recommendations for risk criteria and risk targets to be used for wind turbines. Such risks are
calculated for, e.g., an Environmental Impact Statement (in Dutch: MER) of a wind turbine park. These risk figures
can also be used in the design phase of a wind turbine, e.g., to assess whether the conditional probability of a
catastrophic blade failure in an accidental situation is acceptable or not.
In addition, Section 6.2 summarises a simple mortality rate calculation made on request of the EWTS Project
Steering Committee (PSC). This calculated mortality rate can be used for comparison purposes with mortality rates
observed in other branches of industry and from other activities.
6.1 Recommendations for risk criteria and risk target values to be used
for wind turbines
Various sources of literature on risks and risk criteria have been reviewed. Appendix A presents a list of references
followed by a brief summary of each of them. This material formed the basis for the EWTS-II sub 2 project group to
draw conclusions on risk criteria and risk target values to be used for wind turbines.
Risk Criteria
Risk criteria applicable for wind turbines are the individual risk and the group risk (as defined in the official Dutch
premises for risk management [6.1]). The risk limits for these two criteria as used in the Netherlands are rather
stringent. To our knowledge, there is no other country that adheres to such strict limits. In some countries, similar
criteria are used, but the levels are set higher and they are also mere target levels. An example is the United
Kingdom [6.2].
Risk Target Values
We recommend the use of the individual risk with a risk target level of 10-5/year (which is 10 times higher than the
Dutch risk limit of 10-6/year, see Fig. 6.1). The individual risk is independent of the site of the wind turbine (park)
and can be calculated without knowing the exact location of the installation. Note that we do not use the term ‘risk
limit’; we prefer to use the term ‘risk target’, because risk limits are not commonly used in all EU countries.
The individual risk can be used for design purposes. Suppose we have a generic value of 10-3-10-2/year for blade
failure frequency. Then the conditional probability of being lethally hit by such a failed blade (part) thrown away
should be less than 10-2 (10-3 failure frequency) or than 10-3 (10-2 failure frequency) in order to get below the risk
target for individual risk. This is based on the assumption that the combined effect of more than one wind turbine at
the same site is negligible. In practice, this may not always be the case.
Studies1 performed by ECN as support of an Environmental Impact Statement (in Dutch: MER) indicate that the
conditional probability mentioned above is not higher than about 10-4 based on a modern 500 kWe wind turbine
design.
Although the limits for group risk in The Netherlands are rather stringent, we do not foresee problems of any wind
turbine not complying with the limit for group risk as defined in the Netherlands. Hence, the target values
recommended for the group risk is equal to the Dutch risk limits. The target value is that the likelihood of an
1
Unfortunately, these risk studies are confidential and contain proprietary information. Therefore, no
references are given.

accident with 10 deaths occurring should not exceed one in every hundred thousand years (10-5/year). The aim of
setting target values for group risk is to prevent social disruption (the death of a group of people all at once).
Accidents with even more serious consequences lead to correspondingly greater degrees of disruption. It is therefore
assumed that an n-times larger impact than 10 deaths should correspond with an n-squared smaller probability of
such an accident occurring (see Fig. 6.2, so-called f-n curve). The actual group risk is expressed as a so-called
Complementary Cumulative Distribution Function (CCDF).
Although accident scenarios can be thought of in which a group of people can be killed (e.g., a bus with 30 people is
hit by a blade and subsequently drives into the Noordhollands Canal and 20 people are drowned; however, such a
group is not within the definition of the Dutch group risk), we do not anticipate any problems with compliance with
the group risk. The probability of such scenarios seem negligible (but needs additional investigation).
The project group could not reach a conclusion on whether the risk criteria and target values should be used for an
individual turbine, or, for a park as well. This remains an open item.
6.2 Example mortality rate calculation for wind turbines

In the Netherlands, a number of risk studies for wind turbine parks have been performed (e.g., by ECN). These risk
studies indicated that the acceptable limit (of 10-6/year) for individual risk (as defined in the Netherlands) may be
exceeded. It should be noted that the results of such studies depend strongly on the specific wind turbine design, the
site of the park and the population in the vicinity of the parks into account.
When we relate this risk to a given population (e.g., all wind turbines in operation in the Netherlands), the risk of
each individual wind turbine can be combined to calculate the mortality rate. This rate is defined as the number of
deaths per year (due to a specified activity) for a given population. The Dutch premises for risk management [6.1]
present a table of such observed mortality rates in the Netherlands, see Table 6.1.
The PSC of the EWTS-II project asked for a generic type of mortality rate calculation for the total number of wind
turbines in the Netherlands and to take the type of population in the vicinity of the parks into account. This mortality
rate could then also be compared with mortality rates of other activities (e.g., as displayed in Table 6.1).
Given the available resources in this project (and because this calculation was not part of the original work
programme), such calculation could only be generic and guided by simplifying assumptions. Appendix B presents
the details of these assumptions and of the calculational procedure.

Table 6.1 Annual mortality rate associated with certain occurrences and activities in the Netherlands
([6.1])
Activity/Occurrence Annual mortality rate
Drowning as a result of a dike collapse 10-7 1 in 10 million
-7
Bee sting 2⋅10 1 in 5.5 million
Being struck by lightning 5⋅10-7 1 in 2 million
Flying 1.2⋅10-6 1 in 814,000
Walking 1.85⋅10-5 1 in 54,000
Cycling 3.85⋅10-5 1 in 26,000
Driving a car 1.75⋅10-4 1 in 5,700
Riding a moped 2⋅10-4 1 in 5,000
Riding a motorcycle 1⋅10-3 1 in 1,000
Smoking cigarettes (1 packet a day) 5⋅10-3 1 in 200
Table 6.2 presents the results of the calculations (spreadsheet). The most conservative calculation results in 2E-4/yr,
which is comparable to riding a moped (Sens. 1). A more realistic value seems 1.2E-6/yr, which is comparable to
flying (Sens. 6). The last value seems still larger than the actual fatality statistics. To our knowledge, not a single
person (from the public, not personnel) has ever been killed by a electricity-generating wind turbine. Many people
have been killed as a result of flying. We expect that a realistically calculated value would be of the order of the
mortality rate due to being struck by lightning (1E-7/yr).
Conclusion
Although based on rather simplified assumptions (mainly conservative ones) and a simple calculational procedure, a
‘best-estimate’ of the mortality rate in the Netherlands due to an overspeed accident and consequential blade failure
is of the order of the mortality rate of being struck by lightning.
1.0E-05
Acceptable
Individual Risk (per year)
1.0E-06
Negligible
1.0E-07
1.0E-08 old installation
1.0E-09 improved
installation
1.0E-10
0.1 1 10 100
Distance (km )
Figure 6.1 Example of individual risk limit and result for an installation

1.0E-05
Probability of exceedance
1.0E-06
1.0E-07 Acceptable
(per year)
1.0E-08 Negliglible
1.0E-09 old installation
1.0E-10
1.0E-11
10 100 1000 10000
Num ber of fatalities (N)
Figure 6.2 Example of group risk limits and a CCDF for an installation
Table 6.2 Results of example mortality rate calculations for wind turbines in the Netherlands
Example Mortality Rate Calculation EWTS-IIsub2 74228.01.12
(Assumptions given in accompanying text)
Base Case Sens. 1 Sens. 2 Sens. 3 Sens. 4 Sens. 5 Sens. 6
High BFF Medium BFF,

(Most High BFF, High BFF, Low EF (most
Parameter description/Value Calculated conservative) Medium EF Low EF Medium EF Medium EF realistic)
Blade failure frequency (BFF) 1.0E-03 5.0E-03 1.0E-03 1.0E-03 5.0E-03 5.0E-03 3.0E-03
Max. conditional probability of being hit assuming 24 hour exposure (MCP - 1m) 1.0E-04 1.0E-04 1.0E-04 1.0E-04 1.0E-04 1.0E-04 1.0E-04
Exposure fraction (EF) 1.00 1.00 0.33 0.01 0.33 0.33 0.01
Number of wind turbines (NWTB) 400 400 400 400 400 400 400
Dutch IR per WTB 1.0E-07 5.0E-07 3.3E-08 1.0E-09 1.7E-07 1.7E-07 3.0E-09
MR for all sites 4.0E-05 2.0E-04 1.3E-05 4.0E-07 6.6E-05 6.6E-05 1.2E-06
Note: exposure fraction can also be used as measure for 'rare' situations, e.g., a
group visit to the WTB

7. DATA COLLECTION AND PARAMETER ESTIMATION

7.1 Introduction2
The practices (guidelines) from the previous EWTS subproject were not applied in the case study (excluded from the
scope). These guidelines were:
Recommended Practices from EWTS subproject ‘Assessment of Wind Turbine Safety’
1. For initial quantification purposes, generic data/engineering judgement should be used (order of magnitude
values). Specific wind turbine failure data should be collected and used in the next phases if the scope of the
study or the preliminary results ask for more detailed and more accurate data.
2. The analyst should provide a discussion why certain data is valid for the specific design.
3. The analyst should provide the procedure how key failure data is derived from the data used. This includes both
the methodology and the actual calculations.
It is the objective of the current EWTS subproject to apply these guidelines to the same case, learn from the
experience, and update or extend the guidelines.
Data collection and parameter estimation process
Data are necessary to feed the quantification process. The steps of the data collection and parameter estimation are
outlined in Fig. 7.1 (based on [7.2]).
For safety and reliability purposes, the following types of data have to be collected:
• maintenance and failure data, which includes not only periodical maintenance and inspection but also failure
and incident registration (unplanned maintenance). Examples are the instances of the events, possible causes,
conditions under which events take place, detection and restoration times, affected components, performed
actions, etc.;
• wind turbine modes data, which includes the times and duration the wind turbine is in a certain operational
mode, and the frequencies of alarm signals and transitional modes; and
• external conditions under which events take place, mainly data concerning the wind but also other
climatological data and grid behaviour.
2
The material in this chapter is largely based on the the previous EWTS subproject 'Assessment of
Wind Turbine Safety' [7.1] and the ECN Study [7.2] (Chapter 5 and appendix A that deals with the
systematic collection and analysis of operational and failure data of wind turbines). For completeness
and to be self-contained, some of the material is repeated here. More details can be found in, e.g.,
[7.2, 7.3].

Establish sources of information for data collection
Compile generic or specific data in database
Analyse database
Estimate parameters from key failure data
Interpretation and discussion of the results
Figure 7.1 Steps in the Data Collection and Parameter Estimation Process
The above mentioned types of data originate from various sources, e.g., design specific sources like failure
registration forms of the manufacturer, maintenance sheets, monitoring systems, logbooks, and the owner’s
operating experiences. Since collecting wind turbine data for quantification purposes is at present not common
practice, many data have to be derived from generic sources, which may be from the wind turbine industry, or from
other industries.
The next section discusses briefly the collection of design specific data and parameter estimation.
Section 7.3 presents an overview of generic wind turbine databases and of generic data on external conditions (e.g.,
grid loss and lightning).
Section 7.4. gives an overview of generic databases from other industries or contained in reliability handbooks.
Finally, Section 7.5 summarises the guidelines.
7.2 Design specific failure data

Details of the collection of design specific failure data and parameter estimation can be found in [7.2] and [7.3]. In
particular, the recommendations set out in [7.2.] can be used as guidelines for design specific failure data. These will
not be repeated here. In addition, there is sufficient literature (in the form a reliability handbooks) on this topic. The
recommendations set out in [7.2] and [7.3] can be used for design specific data collection and parameter estimation.
7.3 Generic wind turbine failure data

Some attempts have been made to collect wind turbine data on a national or even European level, e.g., WindStats in
Denmark, the EUROWIN database, the database on accidents and incidents, and the EPRI database, see the Table
7.1. Unfortunately, the information that can be derived from these generic wind turbine databases often appear to be
of very limited use for quantification. Details like failure causes, effects and external conditions are missing. In
addition, only main components are included.

Table 7.1 Examples of generic wind turbine failure databases
Database Year Industry Reference

EUROWIN 1991 Wind Schmid, J., and Klein, H.P: ‘Performance of European Wind Turbines - A
Statistical Evaluation from the European Wind Turbine Database
EUROWIN’, Fraunhofer Institute for Solar Energy Systems, Freiburg,
July 1991.
Accidents 1990 Wind W.J. Stam, et al.: ‘Collection of Data on Accidents and Incidents of Wind
and Incidents Turbines for Updating of Safety Criteria’, ECN-C--90-017, May 1990.
EPRI 1989 Wind Electric Power Research Institute: ‘Assessment of Wind Power Station
Performance and Reliability’, EPRI GS-6256, Project 1590-10, Final
Report, March 1989.
Windstats - Wind WindStats Newsletter, Spring 1992, Vol. 5, No. 2, ISSN-0903-5648
WindStats Newsletter: Vrinners Hoved, Dk-8420 Knebel, Denmark, or:
PO Box 496007, Ste 2170, Redding, California 96099-6007, USA.
External Conditions
Some information was received from ISET who report results of the German ‘250 MW Wind’ Programme [7.4] .
a. General Downtime
The following data is from an earlier report on the 250 MW Project [7.5]. All participants in the program have to
report downtimes due to maintenance and due to failures. The report must contain the reason for the downtime and
the length and cost of the necessary repair work.
More than 800 wind turbines were included in this study, of which 43% were located in the coastal region, 36%
were located in the northern lowlands and 21% were located in the mountain areas of central Germany. The
following table is the result of this study:
Region Grid loss Lightning Ice Storm

Coast (43%) 69% 22% 16% 24%
N. Germ (36%) 8% 21% 19% 29%
Mount. (21%) 23% 57% 65% 47%
This shows that in the mountain areas failures due to lightning and ice are 2 to 4 times more often than in the other
areas. The high proportion of grid losses in the coastal areas is due to the high density of wind turbines, so that many
turbines are affected if part of the grid is lost.

b. Failure Cause Lightning Strikes

Ref. [7.4] and is based on 393 reports from 1992 to 1995, 269 indirect strikes and 124 direct strikes were reported.
The wind turbine locations were the same as before.
Coast Lowlands Highlands Total

No. of WECS 605 531 335 1471
Oper. time (a) 1827 1445 786 4058
No. Events 116 110 167 393
Direct Strikes 43 33 48 124
Events per WECS Year 6% 8% 21% 10%
Direct Strikes per WECS Year 2.4% 2.3% 6.1% 3.1%
This shows that lightning strikes are about three times more likely for turbines in mountain areas due to the higher
frequency of thunder storms there. A map published by the German weather service indicates less than 20
thunderstorms per year for coastal regions and 20 to 35 for mountainous regions. Most thunderstorms occur in the
period from May to September.
Mean repair costs for all direct lightning strikes were DM 50000 to DM 60000. If all three rotor blades were
damaged, they did amount to more than DM 100000.
7.4 Generic Failure Data from Other Industries

This section presents an overview of some generic databases outside the wind turbine industry. Such generic
databases may be the only available source of information. In general, these sources provide more details than the
generic wind turbine databases.
The overview as displayed in the next table lists publicly available generic databases, based on a review of [7.6],
[7.7], [7.8]. Most of these sources are in the form of a report. Some of them may also be available as a computerised
database.
Case studies by ECN indicate that such generic sources can be used, certainly to obtain order of magnitude numbers
for failure rates for generic categories of component and failure modes. Also, some sources provide error factors for
these numbers. However, the use of engineering judgement is often involved in combining data from different
sources and in declaring generic data valid to be used for a specific design.
Table 7.2 : Overview of generic reliability databases from other industries
Database Year Industry Reference

Engineering 1959 Aerospace Procedure and Data for Estimating Reliability and Maintainability, Engineering
Reliability Reliability, Martin, M-M-P-59-21, 1959
Reliability 1972 All Green, A.E. and Bourne, A.J.: Reliability Technology, Wiley, London, 1972
Technology
WASH-1400 1974 Nuclear
IEEE Std 500-1984 1984 Nuclear IEEE, IEEE Guide to the Collection and Presentation of Electrical, Electronic,
Sensing Component, and Mechanical Equipment Reliability Data for Nuclear-
Power Generating Stations, IEEE Std 500-1984, IEEE, Inc., New York, 1983.
IAEA-TECDOC- 1988 Nuclear IAEA: Component Reliability Data for Use in Probabilistic Safety Assessment,
478 IAEA-TECDOC-478, Vienna, 1998
Davidson 1988 J. Davidson (ed.): The Reliability of Mechanical Systems, The Institution of
Mechanical Engineers, 1988.
American Institute 1989 Chemical Guidelines for Process Equipment Reliability Data, Center for Chemical
of Process Safety of the American Institute of Chemical Engineers, New York,
Chemical 1989
Engineers

Table 7.2 : Overview of generic reliability databases from other industries
MIL-HDBK-217F 1991 Military Military Handbook: reliability prediction of electronic equipment, MIL-
HDBK-217F, Griffiss Air Force Base, Rome, New York, 1991
EIRADA 1991 All European Industry Reliability Data Handbook, CEC-JRC/ISEI 21020, Ispra
(Varese) Italy, EDF - DER/SPT 93206 Saint Denis (Paris), France, 1991
RAC 1991 Military Reliability Analysis Center: ‘Nonelectronic Parts Reliability Data 1991’, RAC,
P.O.Box 4700, Rome, NY 13440-8200, 1991.
T-Book 1995 Nuclear T-book, reliability of components in Nordic Nuclear Power Plants, 4rd edition,
1995
OREDA 1992 Offshore OREDA Participation Offshore Reliability Data, OREDA-92, P.O.Box DNV
Technica, N-1322, Hovik, Norway, 1992
OREDA, Guideline for Data Collection, 6th edition, prepared by SINTEF on
behalf of OREDA, 1995
H.A. Sandtorv: Practical experiences with a data collection project: the
OREDA project. Reliability Engineering and System Safety 51 (1996) 159-167.
7.5 Summary of Guidelines

The recommended practices from the EWTS-I project can be extended by the following, more detailed guidelines:
1 The recommendations set out in [7.2] and [7.3] can be used for design of specific data collection and parameter
estimation. Also, recommendations in [7.2] on the use of generic data are still valid.
2 Preferably, use should be made of standard, proven methods as laid down in reliability handbooks. Preferably,
computerised versions of (generic) databases and existing software for reliability data analysis should be used.
3 Generic wind turbine data sources can be used if no specific data is available or if there are no time or resources
available to collect design specific data.
4 Generic wind turbine data sources can be used to provide data on a high level, e.g., an estimate of blade failure
frequency to be used in risk studies, or for frequencies of certain initiating events, e.g., grid loss.
5 If generic data is used to estimate a particular parameter (e.g., a failure rate), the following should be clearly
documented for each parameter estimated:
• reference(s) to the source(s) used;
• discussion why the source is considered valid or why the source is used;
6 If engineering judgement is used to provide an estimate, the reason for this should be indicated. Engineering
judgement should be used to provide an estimate if design specific or generic sources do not provide an
adequate answer.


8. QUANTIFICATION OF MODELS (BENCHMARK STUDY)

This chapter summarises the results of the benchmark study. It was used to support the development of guidelines
for the collection of data and for the estimation of parameters, to be used for quantitative safety and reliability
analyses of wind turbines.
8.1 Outline of Benchmark

The preliminary guidelines set out in Chapter 7 were finalised and extended on the basis of the benchmark results.
The benchmark case consisted of a model of a hypothetical design, as defined and used in a previous EWTS-I
subproject ‘Assessment of Wind Turbine Safety: Recommended Practices’ [8.1]. The model includes only one
initiating event potentially leading to (catastrophic) overspeed of the wind turbine. The accident sequences were
modelled with 3 event trees, 4 fault trees, and about 30 component failure modes (so-called basic events).
Two of the three partners in the subproject, ECN and Risø, participated in the benchmark. The benchmark exercise
was performed following the next steps:
• formulate preliminary guidelines;
• define the benchmark PSA model and establish the types of data to be collected;
• first iteration: collect data independently and quantify model;
• compare results and explain differences in quantification results;
• second iteration: adjust data after having removed misunderstandings and unrealistic data, and re-quantify
model;
• calculate importance measures and rank the component failure modes;
• learn from the benchmark results and finalise guidelines.
The necessity of the second iteration clearly demonstrated the difficulties in the quantification even with the same
model available to the participants (e.g., also observed in [8.7]).
8.2 Benchmark results

Preliminary guidelines
Based on previous ECN case studies and experience [8.2-8.6], preliminary guidelines were formulated prior to
collecting data and estimating parameters. The guidelines covered:
• the list of basic events for which to collect data;
• a list of failure mode types based on event tree/fault tree analysis software used (NUPRA [8.8]);
• depending on the failure mode selected, a list of parameters to be collected (e.g., failure rates, time factors);
• reporting formats to submit the data.
Appendix C presents the final guidelines. The relationship of the data to be collected, their sources, and the
parameters to be estimated is displayed in Figure 8.1.
The guidelines were restricted to the use of generic sources of data (e.g., from literature).

Component failure Component failure model and

rate/hour time factors
(derived from generic (derived from WTB system
databases (e.g., from description and operating
literature) or from procedures)
design specific
operational
experience
model: monitored/periodically
tested or maintained?
time: failure detection time,
test/maintenance interval, demand
frequency
basic event = component plus failure mode
failure probability/unavailability
Figure 8.1 Relationship between data to be collected, data sources and parameters to be estimated
Model and data to be collected

The case concerns a hypothetical design but with similar characteristics to the AOC 15/50 turbine [8.2]. It was
selected because it was previously used in the EWTS-I project [8.1]. As a result, the qualitative part of the case was
documented extensively. The case model consisted of only one initiating event potentially leading to (catastrophic)
overspeed of the wind turbine. The accident sequences were modelled with 3 event trees, 4 fault trees, and about 30
component failure modes (so-called basic events).
The first event tree described the event sequences emerging from the initiating event (Too high an inverter
temperature). Some of the sequences transferred into two other event trees which were developed for a fast
shutdown and an overspeed situation. The last one is presented in Figure 8.2. Note that the frequencies of the
sequences are not simply the product of the nonavailabilities/failure probabilities of the individual events displayed
in the heading of the event tree, e.g., see sequence S05. The product equals 1.07e-8 (2.02e-5 * 2.83e 2 * 2.49e-2 *
7.55e-1), while the frequency of SEQ S05 equals 2.01e-5. This shows how the fault tree linking approach takes
account of the intersystem dependencies and of events already failed.
Fault trees were made for the Detection System, Tip Brake System, Mechanical Brake System, and the Generator
Disconnection. The actuation/operation of these systems follow more or less the chronology of the various events.
From the various end states of these sequences, only the end state ‘Catastrophic overspeed’ was selected for this
benchmark exercise. Other end states (e.g., failed operation without any safety implications or successful shutdowns)
were not investigated further.

IE-S5, overspeed Detection of S5 At least one tip Mechanical brake Emergency shut- S
Nrotor > 67 rpm brake deploys via closes via down. All 3 tip
contactor OT contactor OT brakes deploy
E SEQUENCE E
pass., magn. ener Q DESCRIPTOR S FREQUENCY
gized, Nrotor>75 # #
S5 D T MB T-ES
S01 S5 OK1 0.00E+00
MB-OT S02 S5MB ES1 0.00E+00

2.49E-02
S03 S5T OK2 0.00E+00
T-OT
2.83E-02
S04 S5TMB ES3 6.81E-08
MB-OT
IE-S5 2.49E-02
2.02E-05
T-ES S05 S5TMBT-ES OS 2.01E-05

7.55E-01
S06 S5D ES3 3.31E-10
D-S5
6.11E-03
T-ES S07 S5DT-ES OS 1.22E-07

7.55E-01
OK1 successfully parked

OK2 parked
ES1 idling, rotor speeding up, slowing down
ES3 idling, running into overspeed, tip deployment, slowing down
OS overspeed
C:\WIND\EWTS-II2\CASE\EVT\S5.EVT 12:18:06pm 3-03-97 NUPRA 2.33 ECN
Figure 8.2 Event tree depicting sequences leading to overspeed

First data collection, parameter estimation, and quantification

The sources were restricted to generic data(generic wind turbine data as well as average data from other types of
industry). It should be noted that design specific data are normally the preferred source of data. However, specific
data are often not available or only at the Expense or too much time and resources Moreover, the case model
concerns a hypothetical turbine for which of course no specific data is available.
ECN and Risø started a first iteration with the same set of documentation on the system description (as laid down in
[8.1]), and with the same preliminary guidelines. This resulted in two data sets comprising for each basic event:
component failure model (e.g., periodically tested, monitored i.e. self-annunciating failures), failure rate, time factors
(e.g., test intervals, detection time) and some rationale on how the data was derived.
Comparison of first results
ECN used a combination of generic wind turbine data (from previous ECN safety and reliability studies on wind
turbines, e.g., LW 15/75, AOC 15/50, NEWECS45), industry average data (as collected and used in a recent ECN
reliability study of the NEDWIND40), and engineering judgement. Risø used one commercial reliability database
(FARADIP.THREE), data of one 150 kW wind turbine of Nordic design, and engineering judgement.
Direct comparison of the two data sets revealed large differences, which showed up more strikingly after
quantification of the model. The resulting frequency of occurrence of overspeed, OS Freq, and the conditional
probability of overspeed (given the occurrence of the initiating event), Cond. Prob., differed by more than four
orders of magnitude (see Table 8.1), while the failure rates differed no more than one to two orders of magnitude.
The results of Risø seemed to be unrealistically low, in particularly when looking at the resulting conditional
probability of overspeed and the common cause failure involved.
Table 8.1 Overall results, first and second iteration: frequencies of the initiating event (IEF) and of overspeed
(OS) and conditional probability of overspeed given the initiating event
First iteration Second iteration

IEF [/yr] OS Freq Cond. Prob. IEF [/yr] OS Freq Cond. Prob. [-
[/yr] [-] [/yr] ]
ECN 3.8 e-2 2.0 e-5 5.3 e-4 3.8 e-2 2.0 e-5 5.3 e-4
Risø 8.4 e-6 7.1 e-15 8.5 e-10 3.1 e-3 8.3 e-8 2.7 e-5
Factor ECN/Risø 4.5 e+3 2.8 e+9 6.3 e+5 12 242 20
The differences in data were explained by the following factors:

1. Initiating event frequency.
The initiating event frequency (IEF) was about 4500 times higher in the ECN case (ECN: 3.8e-2; Risø 8.4e-6),
which directly affects the frequency of a catastrophic overspeed (OS Freq).
2. Failure rates.
In general, ECN used higher failure rates resulting in higher nonavailabilities (one to two orders of magnitude
higher).
3. Failure model.
ECN used mostly a periodically tested component model, except for the CPU and the initiating event (IE-S7),
with intervals of 730, 1750 or 8760 hours. Risø used mostly a monitored component model, with a
detection/repair time of 24 hours. This had a great impact on the resulting nonavailabilities (ECN higher than
Risø). There was probably a difference in the understanding of the wind turbine system operation, although the
same information was used.
4. Time factors.
ECN took the demand frequency into account resulting in smaller intervals than one year for the periodically
tested components (and hence smaller nonavailabilities). However, the effects of using higher failure rates (see
2.) and other failure models were larger.
5. Common Cause Failure (CCF).
The CCF unavailability of the tip brakes was very small in the Risø case (three orders of magnitude smaller than

the corresponding independent failure modes), which might be too different from reality. Risø had modelled a
functional dependency, viz. loss of grid, as common cause. If sufficient details were available, such dependent
failures should however be modelled in the logic of the fault trees, not at the basic event level. ECN used a so-
called ‘beta factor’ of 0.1, based on engineering judgement and because such value is considered relatively
conservative, according to the reliability literature. This means that 10 % of the failures involve a (residual)
common cause.
Adjustment of Risø data and re-quantification

After having explained the differences, the Risø data were adjusted. Basically, only the failure models and time
factors were changed. The full details of the fault tree analysis software was not available to Risø. Because of that,
some difficulties with understanding the different meanings of failure models were encountered. In addition, the
CCF failure rate was adjusted.
Comparison of second results and calculation of importance measures
With the new Risø data, the overall ECN and Risø results appeared to be better in agreement, see Table 8.1. The
differences in results were less significant, e.g., the conditional probability differed by a factor 20 (in stead of 6e+5
in the first iteration). The still persisting differences were within the line of expectation, since ECN and Risø used
different (generic) data sources. If (the same) design specific operational experience was used as data source, the
results would certainly be very close.
The new results were considered suitable to compare also the relative, qualitative results, as expressed in the
importance measures, e.g., of the component failure modes contributing to the overspeed frequency.
The importance’s are displayed in Table 8.2 for the component failure modes contributing 1% or more to the end
state ‘Catastrophic overspeed’. The columns ‘Relative Importance’ (The relative importance of the component
failure mode as contributing to the frequency of catastrophic overspeed) and ‘Rank’ are the ones which are
interesting. In both cases, the Tip Brake (TB) and Mechanical Brake (MB) systems appeared to be very important.
The Electrical System (EL) is relatively unimportant. The CCF of the Tip Brakes was important in both cases
(ranked 1 and 2, respectively). Broken down to the individual components or even failure modes, there were still
some differences between the ECN and Risø results.
These were explained by two factors:
1. in general, ECN used higher failure rates, which lead to higher component nonavailabilities;
2. in general, ECN used smaller test/maintenance intervals by taking into account the average demand frequency
of certain standby components, in case this frequency is higher than the usual test/maintenance frequency: this
lead to lower component nonavailabilities.
The effect of 1. appeared to be more important than the effect of 2.
Lessons learned from the exercise and final guidelines
Additional guidelines were formulated in order to prevent the misunderstandings observed in the first iteration of the
exercise. The lessons learned were used to adjust the preliminary guidelines set out before. In particular, additional
guidance was given on the selection of failure model and associated time factors.

Table 8.2 Ranking of the most component failure modes (based on second iteration)
ECN Risø
System, Unavaila- Relative Rank Unavaila- Relative Rank
component failure mode bility Importance bility Importance
MB, cylinder stuck 1.9e-2 79% 1 6.1e-4 20% 2
TB, common cause failure 1.1e-2 41% 2 8.7e-3 96% 1
TB, hinge 1/2/3 8.3e-2 19% 3 1.6e-2 1% 6
TB, damper 1/2/3 8.3e-2 19% 3 4.4e-3 0% 9
TB, catch plate 1/2/3 8.3e-2 19% 3 1.5e-2 1% 7
MB, spring loss of function 4.5e-3 19% 4 6.1e-4 20% 2
MB, valve Y2 stuck 1.3e-2 1% 5 3.1e-3 1% 4
EL, contactor KM fails to open 3.6e-4 1% 6 3.1e-3 4% 3
MB, throttle valve blocked 1.3e-2 1% 7 1.2e-2 1% 5
MB, valve Y1 stuck 1.3e-2 1% 7 3.1e-3 0% 8
MB, filter unit blocked 7.3e-6 0% 8 6.1e-4 20% 2
MB, disk fracture 7.3e-6 0% 8 6.1e-4 20% 2
MB, brake pad wear out 7.3e-6 0% 8 6.1e-4 20% 2
8.3 Conclusions and recommendations

The following is concluded from the benchmark study:
• The main differences in the results after the first iteration of the data collection and parameter estimation were
caused by different understanding of the system operation. E.g., this resulted in the assignment of different
failure models and associated time factors for the basic events. Therefore, the guidelines were extended such
that the analyst will make less mistakes in assigning either a monitored or periodically tested failure model, and
a better detection time and test interval, respectively.
• After the second iteration in the benchmark exercise, the qualitative results were more or less similar, e.g., with
respect to the two most important systems and to the most important individual component failure modes.
• The results of the benchmark show the strength of such quantitative reliability analysis, even in case of a lack of
data. E.g., lack of design specific data should not be an excuse for not performing quantitative reliability or
safety studies.
The following is recommended based on the findings of the study:

• The guidelines formulated as part of this EWTS-II subproject can be an aid to collect data and to use it wisely
for the purpose of safety and reliability studies of wind turbine designs.
• Qualitative safety and reliability are recommended very strongly as a minimum, as these studies result often in
the identification of design errors (e.g., no or too less redundancy, no defence for certain events).
Quantification, even with only the use of generic data sources, often results in a good ranking of the important
event sequences, component failure modes, and systems. A combination of generic data and engineering
judgement can always be used, although the reliability analyst should be aware of the limitations. In this way, a
design can be made (more) balanced.
• Quantitative safety and reliability studies should be used primarily for the ‘relative’ (qualitative) purposes
outlined above. The use of absolute figures and absolute criteria should be avoided.
• It would be interesting to conduct a similar benchmark on an existing design rather than a hypothetical one. For
this existing design, specific data and operational experience should preferably be available in maintenance
records, logbooks, etc. If not already available, collection of the data should be possible with limited effort. Our
experience learns that this is feasible if such an effort gets the support and involvement of the owners and
developer/manufacturer of the design.

9. CONCLUSIONS AND RECOMMENDATIONS

The two objectives of the EWTS-II subproject ‘Quantification of Failure Probabilities’ are:
• derivation of target values for structural reliability and for risk criteria at which the designer should aim and
which the certifying bodies can use as a criterion; risk is here meant as risk to the public;
• preparation of guidelines to collect data and to estimate parameters to be used for quantitative safety and
reliability analyses (or PSA’s).
To meet these objectives, the subproject was divided into two parts:
• a literature study to obtain target values for structural reliability and risk criteria;
• a limited benchmark exercise to obtain guidelines for data collection and parameter estimation.
Target values for structural reliability

From the literature study (as reported in Chapter 5), it is concluded that:
• The Scandinavian countries and The Netherlands seem to be leading in the application of structural reliability
methods.
• The code which is not local to a country, but shall be applied Europe-wide, is the Eurocode 1.
Therefore, it is recommended to apply the safety level of the Eurocode to wind turbines, which would mean a yearly
safety index of 4.7, corresponding to a failure probability of 10-6/year.
Risk criteria and target values
From the literature study (as reported in Chapter 6), it is concluded that:
• Risk criteria related to the risk to the public that are applicable for wind turbines are the individual risk and the
group risk (as defined in the official Dutch premises for risk management.
For these two criteria, the following target values are recommended:
Individual risk
For the use of the individual risk a risk target level of 10-5/year is recommended. The individual risk is independent
of the site of the wind turbine (park) and can be calculated without knowing the exact location of the installation.
Note that the term ‘risk limit’ is not used but instead the term ‘risk target’, because risk limits are not commonly
used in all EU countries.
Group risk
The target value for the group risk is that the likelihood of an accident with 10 deaths occurring should not exceed
one in every hundred thousand years (10-5/year). The aim of setting target values for group risk is to prevent social
disruption (the death of a group of people all at once). Accidents with even more serious consequences lead to
correspondingly greater degrees of disruption. It is therefore assumed that an n-times larger impact than 10 deaths
should correspond with an n-squared smaller probability of such an accident occurring. The actual group risk is
expressed as a so-called Complementary Cumulative Distribution Function (CCDF).
Mortality rate calculation

On a special request by the Project Steering Committee, a generic type of mortality rate calculation for the total
number of wind turbines in the Netherlands was performed. This mortality rate was compared with mortality rates of
other activities. Although based on rather simplified assumptions (mainly conservative ones) and a simple
calculational procedure, a ‘best-estimate’ of the mortality rate in The Netherlands due to an overspeed accident and
consequential blade failure is of the order of the mortality rate of being struck by lightning.
Guidelines for data collection and parameter estimation
As guidelines for data collection and parameter estimation, it is recommended to use as starting basis:
1. the guidelines as outlined in the LW 15/75 and NEWECS-45 studies reports; and
2. the recommended practices from the EWTS-I project.
In addition to these reports, the following is recommended:

1. The recommendations set out in the LW15/75 and NEWECS-45 studies can be used for design of specific data
collection and parameter estimation. Also, the recommendations of the LW 15/75 study on the use of generic
data are still valid.
2. Preferably, use should be made of standard, proven methods as laid down in reliability handbooks. Preferably,
computerised versions of (generic) databases and existing software for reliability data analysis should be used.
3. Generic wind turbine data sources can be used if no specific data is available or if there are no time or resources
available to collect design specific data.
4. Generic wind turbine data sources can be used to provide data on a high level, e.g., an estimate of blade failure
frequency to be used in risk studies, or for frequencies of certain initiating events, e.g., grid loss.
5. If generic data is used to estimate a particular parameter (e.g., a failure rate), the following should be clearly
documented for each parameter estimated:
• reference(s) to the source(s) used;
• discussion why the source is considered valid or why the source is used;
6. If engineering judgement is used to provide an estimate, the reason for this should be indicated. Engineering
judgement should be used to provide an estimate if design specific or generic sources do not provide an
adequate answer.
Benchmark study
A benchmark study by ECN and Risø was used to test and extend the initial guidelines. The study was performed in
two. In addition, the following was be concluded from the benchmark study:
1. The main differences in the results after the first iteration of the data collection and parameter estimation were
caused by different understanding of the system operation. E.g., this resulted in the assignment of different
failure models and associated time factors for the basic events. Therefore, the guidelines were extended such
that the analyst will make less mistakes in assigning either a monitored or periodically tested failure model, and
a better detection time and test interval, respectively.
2. After the second iteration in the benchmark exercise, the qualitative results were more or less similar, e.g., with
respect to the two most important systems and to the most important individual component failure modes.
3. The results of the benchmark show the strength of such quantitative reliability analysis, even in case of a lack of
data. E.g., lack of design specific data should not be an excuse for not performing quantitative reliability or
safety studies.
It would be interesting to conduct a similar benchmark on an existing design rather than a hypothetical one. For this
existing design, specific data and operational experience should preferably be available in maintenance records,
logbooks, etc. If not already available, collection of the data should be possible with limited effort. Our experience
learns that this is feasible if such an effort gets the support and involvement of the owners and
developer/manufacturer of the design.

Final recommendations
To following summarises the main recommendations that result from this EWTS-II subproject:
1. The guidelines formulated as part of this EWTS-II subproject can be an aid to collect data and to use it wisely
for the purpose of safety and reliability studies of wind turbine designs.
2. Qualitative safety and reliability studies are recommended very strongly as a minimum, as these studies result
often in the identification of design errors (e.g., no or too less redundancy, no defence for certain events).
Quantification, even with only the use of generic data sources, often results in a good ranking of the important
event sequences, component failure modes, and systems. A combination of generic data and engineering
judgement can always be used, although the reliability analyst should be aware of the limitations. In this way, a
design can be made (more) balanced.
3. Quantitative safety and reliability studies should be used primarily for the ‘relative’ (qualitative) purposes
outlined above. The use of absolute figures and absolute criteria should be avoided.


10. REFERENCES
[1.1] IEC 1400-1, Wind Turbine Generator Systems -Part 1: Safety Requirements, ed. 2, 1997
[1.2] European Commission, ‘European Wind Turbine Standards - Project Results’, Part 3: Assessment of Wind
Turbine Safety: Recommended Practices and Part 4: Assessment of Wind Turbine Safety: Case Study, Contract
JOU2-CT93-0387, Final Report, EUR 16898 EN, 1996.
[1.3] A.J. Seebregts, L.W.M.M. Rademakers, B.A. van den Horn, ‘Reliability Analysis in Wind Turbine
Engineering’, Microelectron. Reliab., Vol. 35, Nos 9-10, pp. 1285-1307, 1995.
[1.4] Rademakers, L.W.M.M. and Carter, J., ‘Reliability Analysis of the AOC 15/50 Wind Turbine Generator’,
ECN-C--94-057, Petten, The Netherlands, November 1994.
[5.1] Calibration of Safety Factors, Joule II, Volume 4, JOU2-CT93-0387, ECN, Petten
[5.2] Structural Reliability Analysis for Horizontal Axis Wind Turbines Components, B. A. van den Horn, L.
W. M. M. Rademakers, T. W. Verbrüggen, ECN-RX_94-097, 1994.
[5.3] General principles on reliability for structures, ISO-2394, 2nd edition, Okt. 1985
[6.1] Ministry of Housing, Physical Planning and Environment (VROM): Premises for Risk Management - Risk
Limits in the Context of Environmental Policy, Annex to the Dutch National Environmental Policy Plan
“Kiezen of Verliezen” (to Choose or to Loose) Second Chamber of the States General, session 1988-1989, 21
137, no 5, The Netherlands.
[6.2] F.R. Allen et al.: The Management of Risk to Society From Potential Accidents, Elsevier Science Publishers
LTD, ISBN 1-85166-892-6, 1992.
[7.1] L. Rademakers, et al.: European Wind Turbine Standards. Vol. 4 Assessment of Wind Turbine Safety, EUREC-
Agency, JOULE-II Programme, Contr.: JOU2-CT93-0387, 1996.
[7.2] L.W.M.M. Rademakers, A.J. Seebregts, B.A. van den Horn, J.N.T. Jehee, B.M. Blok: Methodology for PSA
for Wind Turbines - Demonstrated by a Case Study of the LAGERWEY 15/75 Design. ECN-C--93-010, Petten,
The Netherlands, March 1993.
[7.3] L.W.M.M. Rademakers, A.J. Seebregts, T.W. Verbruggen, R. Disseldorp: Reliability Analysis and Design
Review of the NEWECS-45, Part 1: Feedback of Operational Experience,
ECN-C--94-066, September 1994.
[7.4] M. Durstewitz, C. Ensslin, M. Hoppe-Kilpper, K. Rohrig: External Conditions for wind turbine
operation, Results from the German `250 MW Wind’ Programme, EU Wind Energy Conf., Goteborg, May
20-24, 1996.
[7.5] German 250 MW Project, presented at 3rd Austrian Wind Energy Symposium, St. Pölten,
26.-28.4.95.
[7.6] Cooke, R.M. et al: Design of Reliability Data Bases for Aerospace Applications, Report 93-110,
Delft University of Technology, ISSN 0922-5641, 1993.
[7.7] J.R. Fragola: Reliability and risk analysis data base development: an historical perspective.
Reliability Engineering and System Safety 51 (1996) 125-136.
[7.8] R.M. Cooke: The design of reliability data bases, part I: review of standard design concepts.
[7.9] H.A. Sandtorv: Practical experiences with a data collection project: the OREDA project.
[8.1] European Commission, ‘European Wind Turbine Standards - Project Results’, Part 3: Assessment of Wind
Turbine Safety: Recommended Practices and Part 4: Assessment of Wind Turbine Safety: Case Study,
Contract JOU2-CT93-0387, Final Report, EUR 16898 EN, 1996.
[8.2] Rademakers, L.W.M.M. and Carter, J., ‘Reliability Analysis of the AOC 15/50 Wind Turbine Generator’,
ECN-C--94-057, Petten, The Netherlands, November 1994.
[8.3] L.W.M.M. Rademakers, A.J. Seebregts, B.A. van den Horn, J.N.T. Jehee, B.M. Blok, ‘Methodology for PSA
for Wind Turbines - Demonstrated by a Case Study of the Lagerweij 15/75 Design’, ECN-C--93-010, March
1993.
[8.4] L.W.M.M. Rademakers, A.J. Seebregts, T.W. Verbruggen, R. Disseldorp, ‘Reliability Analysis and Design
Review of the NEWECS-45, Part 1: Feedback of Operational Experience’, ECN-C--94-066, September 1994.
[8.5] L.W.M.M. Rademakers, A.J. Seebregts, and T.W. Verbruggen, ‘Applications of Reliability Analysis Methods
for Wind Turbines’, Proceedings EWEC’94 Conference, Thessaloniki, Greece, 10-14 October 1994, Vol. 1,
pp. 640-646.

[8.6] A.J. Seebregts, L.W.M.M. Rademakers, B.A. van den Horn, ‘Reliability Analysis in Wind Turbine
Engineering’, Microelectron. Reliab., Vol. 35, Nos 9-10, pp. 1285-1307, 1995.
[8.7] Contini, S., Amendola, A. and Ziomas, I. ‘Benchmark exercise on major hazard analysis’, Volume I:
Description of the project, discussion of the results and conclusions, CEC/JRC-Ispra EUR 13386 EN, 1991.
[8.8] Halliburton NUS, ‘NUPRA User’s Manual’, Version 2.31 Supplement, Gaithersburg, Maryland 208878,
August 1995.

APPENDIX A. REVIEW OF SAFETY AND RISK CRITERIA

For the derivation of target values for structural reliability and for risk criteria on which the designer
should aim and which the certifying bodies can use as a criterion, a number of references are reviewed and
summarized. This appendix contains a list with the references considered (Table A.1) and the
corresponding summaries. The conclusions are part of the main report. In chapter 5 of the report
recommendation for target values for structural reliability are presented. The conclusions and
recommendations for risk criteria and mortality rates are given in chapter 6 of the report. The wind turbine
failure data are incorporated in chapter 7 of the report.
The references and summaries are grouped into the following categories:
• Target reliability’s in various countries and in various branches of industries,
• Risk criteria in the various countries,
• Mortality rates observed, and
• Wind turbine failure data.
Table A.1 List of references
Nr. Reference
A.1 Target reliability’s in various countries and in various branches of industries
[A1.1] Vrouwenvelder, A.C.W.M. and Siemes, A.J.M.: Probabilistic calibration procedure for the
derivation of partial safety factors for the Netherlands building codes.
[A1.2] IAEA: General Design Safety Principles for Nuclear Power Plants - A Safety Guide, IAEA
Safety Series Report No. 50-SG-D11, Vienna, 1986.
[A1.3] Technische Grondslagen voor de berekening van Bouwconstructies (in Dutch, Technical basis
for the design of building constructions).
[A1.4] ASME/USNRC: Risk-Based Inspection - Development of Guidelines, General Document,
NUREG/CR-0005, CRTD-Vol.20-1, R5, RG, February 1992.
[A1.5] Nordisk Komite for bygningstemmelser: Guidelines for loading and safety regulations,
NKB-rapport No 55 (Danish) +55E (English translation), juni 1987, available from Danish
Standards.
[A1.6] DS409, The Safety of Structures and DS-410, Loads for the Design of Structures, English
translation, NP-157-T of ‘Normstyrelsens publikationer’, N-157-N, June 1982, Teknisk Forlag,
Copenhagen.
[A1.7] DS 412: Code of Practice for the Structural Use of Steel (NP-163-N) in Danish, NP-163-T:
English translation, Teknisk Forlag, Copenhagen, April 1983, translation March 1984.
[A1.8] DS 456: Constructions made of Glass-fibre reinforced unsaturated polyester, 1st edition,
Teknisk Forlag NP-178-N, Copenhagen, 1985.
[A1.9] DS 472 (DIF-norm): Last og sikkerhed for vindmøllekonstruktioner, 1. udgave 1992 (In Danish,
Loads and Safety of Wind Turbine Construction, first edition), Teknisk Forlag.
[A1.10] EUROCODE no. 1: Basis of Design and Actions; part 1: Basis of Design, European
Commission, CEN/TC 250/SC 1, in preparation.
[A1.11] ISO 2394: General Principles on Reliability for Structures, 1986.

[A1.12] Teknisk Grundlag for typegodkendelse og certificering af vindmøller (In Danish, Technical
Criteria for type approval and certification of wind turbines), November 1994; Energistyrelsen,
Risø-I-943.
[A1.13] Rekommendation til opfyldelse af krav i Teknisk Grundlag (In Danish, Recommendation
Compliance with Requirement Technical Basis for the Type Approval and Certification of
Wind Turbines In Denmark), 1. July 1992, Energistyrelsen.
[A1.14] BS 8100, part 2: 1986, Lattice Towers and Masts.
[A1.15] DIN 18800.
[A1.16] API RP 2A - LRFD, Planning, Designing and Constructing Fixed Offshore Platforms - Load
Resistance Factor Design, 1993.
A.2 Risk criteria in the various countries

[A2.1] Ministry of Housing, Physical Planning and Environment (VROM): Premises for Risk
Management - Risk Limits in the Context of Environmental Policy, Annex to the Dutch National
Environmental Policy Plan “Kiezen of Verliezen” (to Choose or to Loose) Second Chamber of
the States General, session 1988-1989, 21 137, no 5, The Netherlands.
[A2.2] Vrijling, J.K., Hengel, W. van, Houben, R.J.: A Framework for Risk Evaluation, Journal of
Hazardous Materials 43 (1995) 245-261.
[A2.3] Vrijling, J.K., Hengel, W. van, Maanen, S.E. van: The application of the concept of societal risk
to various activities in the Netherlands, Proc. ESREL’96/PSAM-III Int. Conf. on Probabilistic
Safety Assessment and Management, June 24-30, 1996, Crete, Greece, pp. 960-966.
[A2.4] Sprent, P.: Taking Risks - The Science of Uncertainty, London, Penguin, 1988 (Chapter 11)
(Dutch translation: Risico’s - Zekerheden over onzekerheid, Bloemendaal, Aramith, 1990).
[A2.5] IPO (Interprovinciaal Overleg): Handleiding voor het opstellen en beoordelen van een extern
veiligheidsrapport (EVR) (In Dutch) (Manual to draft and to review an external safety report),
IPO-publikatienummer 54, The Hague, The Netherlands, April 1994.
[A2.6] SI 1992/2885: Offshore Installations (Safety Case) Regulations 1992 and Relevant Guidelines.
[A2.7] F.R. Allen et al.: The Management of Risk to Society From Potential Accidents, Elsevier
Science Publishers LTD, ISBN 1-85166-892-6, 1992.
A.3 Mortality rates observed

[A3.1] Sprent, P.: Taking Risks - The Science of Uncertainty, London, Penguin, 1988 (Chapter 11)
(Dutch translation: Risico’s - Zekerheden over onzekerheid, Bloemendaal, Aramith, 1990).
[A3.2] Danmarks Statistik: Statistical Year Book 1995.

A.4 Wind turbine failure data

[A4.1] Schmid, J., and Klein, H.P: Performance of European Wind Turbines - A Statistical Evaluation
from the European Wind Turbine Database EUROWIN, Fraunhofer Institute for Solar Energy
Systems, Freiburg, July 1991.
[A4.2] Goezinne, F. en Veldkamp, H.F.: ‘Veiligheidsanalyse Windparken Noord-Groningen’, (In
Dutch, Safety and Risk Analysis Wind Parks in the North of Groningen) Goezinne &
Veldkamp, Enschede, september 1993.
[A4.3] WindStats Newsletter: Vrinners Hoved, DK-8420 Knebel, Denmark, or: PO Box 496007, Ste
2170, Redding, California 96099-6007, USA.
[A4.4] Curvers, A. and Schuurman, J.J.: ‘Bedrijfsgegevens van IPW-windturbineprojecten: eerste fase’
(Operating experience of IPW wind turbine projects: first phase)’, ECN-C--91-041, June 1991.
[A4.5] W.J. Stam, et al.: ‘Collection of Data on Accidents and Incidents of Wind Turbines for Updating
of Safety Criteria’, ECN-C--90-017, May 1990.
Appendix A.1 Target reliability’s in various countries and in various

branches of industries
[A1.1]
The paper describes the procedure how partial safety factors in the Netherlands building code were
calibrated with a probabilistic calibration procedure. First many types of components of various materials
were designed in accordance with the Netherlands building code (beams, columns, T-joints, etc.). Next,
these elements were subjected to a FORM reliability analysis in order to determine the safety level. In this
way an idea was obtained of the average safety level and of the scatter with respect to it. The types of
distributions and their parameters to describe the stochastic variables (loads and resistance) were obtained
from an extensive literature survey. The purpose of the project was so to derive partial safety factors that
this scatter would, in the future code, be reduced while the average remained approximately unchanged.
The following was concluded from the reliability analysis.
• The value of the safety index β for the limit state “ultimate load” ranges between 2.2 and 6.1 with an
average of 3.8 (failure probability ≈ 7⋅10-5) and a standard deviation of 1.4.
• For the serviceability limit states, an average value for β was found to be 1.7 (failure probability ≈
5⋅10-2) and a standard deviation of 1.2.
It should be noted here that the reference time was 50 years so the failure probability per year is
approximately 1.4⋅10-6/yr for the ultimate limit state and 10-3 for the serviceability limit state. The general
trend was that a high reliability index was obtained for loading cases comprising a relatively high
proportion of self-weight. The low values were found for loading cases for which the variable loads
dominate. The variable loads have a higher coefficient of variation and therefore need a higher safety
factor.
[A1.2]
With respect to reliability design, this reference discusses (qualitatively) the following echelons of
defense: redundancy, diversity, independence, fail-safe design, and auxiliary services. However, no
quantitative target reliability values are provided for each line of defense.

A Dutch amendment to this Safety Guide does include some quantitative reliability requirement:
• a non-diverse safety system should have an unavailability of less than 10-4 - 10-3 per initiating event
(per demand).
[A1.3]
NEN 6700 “TGB 1990 - General Principles” gives general requirements for reliability (safety and
serviceability) on a fundamental level. These requirements have to be fulfilled for all building structures
independent of the materials used in these structures. In NEN 6700, which has become operational in 1991
the deterministic approach used in the preceding norm (NEN 3850: 1972) is replaced by a probabilistic
approach. The target values for the safety index β as specified in this norm are presented in Table A.2. These
target values are determined such that the reliability level of a building structure designed according to NEN
6700 is on average equal to the reliability level for the design based on the preceding norm (NEN 3850:
1972). For this reason distinction had to be made between wind loadings and other loadings. Three safety
classes are distinguished depending on the consequences of failure. In [A1.1] the procedure for determination
the safety index and the corresponding partial safety factors is outlined.
Table A.2 Safety index β
Safety index
Safety class
ultimate load serviceability
wind load other loads
1: negligible personnel injury, 2.3 3.2 1.8

small public loss
2: some personnel injury, 2.4 1) 3.4 2) 1.8
significant public loss
3: extensive personnel injury, 2.6 3.6 1.8
significant public loss
1
) reference period t ≥ 15 years; for t < 15 years, β = 3.1 - 0.60 log t
2
) reference period t ≥ 15 years; for t < 15 years, β = 3.6 - 0.85 log t
From an economical point of view the safety index is kept constant, independent of the design lifetime of
the building structure. This implies that the failure probability per year will be higher for building
structures with a shorter design lifetime.
The general requirements of NEN 6700 are elaborated in more detail with respect to the determination of
the characteristic loading and with respect to the specification of the partial safety factors in NEN 6702
(TGB 1990 - Loadings and Deformations, December 1990). For a number of building structures the safety
class and reference period is specified. Wind turbines are not mentioned, but building structures for
electric power stations are classified as class 3 with a reference period of 50 years. Special applications of
NEN 6702 are given in NEN 6706 for bridges and in NEN 6707 for roofs.
The determination of the resistance of a building construction against the applied load is elaborated in the
following material specific norms:

• NEN 6710 for aluminium construction;

• NEN 6720 for concrete construction;
• NEN 6740 for geotechnics;
• NEN 6750 for lumber construction;
• NEN 6770 for steel construction;
• NEN 6790 for bricks.
These material specific norms, which always have to be used in connection with NEN 6702, provide
characteristic material properties and partial safety factors. Also rules are specified which can be used to
check whether the requirements of NEN 6700 are fulfilled. The framework of this series of TGB norms is
such that in case the requirements of the material specific norms in connection with NEN 6702 are
fulfilled, the general requirements of NEN 6700 are satisfied automatically.
Besides, in NEN 6701 (TGB 1990 - Names and symbols and quantities) names and symbols are defined
for a number of quantities.
[A1.4]
Although the title focuses on inspection, this report presents a variety of target reliability values and
approaches in several industries: nuclear, fossil, aircraft, civil engineering and marine structures, industrial
insurance. However no relevant information for this project is presented.
[A1.5]
NKB, or Nordic Committee for building regulations, is a co-operative formed by the Nordic building
authorities. The NKB is responsible for harmonizing the Nordic building regulations. All Nordic countries
are members (Denmark, Finland, Iceland, Norway, and Sweden).
The NKB report details the partial coefficient code format to be used in Nordic codes. For establishing the
code format, the NKB states, that close contact with the international ‘Joint Committee on Structural
Safety’ has been maintained. Furthermore it says that the concept of ‘safety’ is equivalent to the concept
reliability in ISO 2394 [A1.11]. The NKB report also outlines the Nordic probabilistic code calibration
project that forms the basis for the Nordic safety evaluation.
In this report, the safety requirement is expressed in terms of the safety index and is also given as an
equivalent ‘formal yearly probability of failure’, which is the target reliability. Nordic codes distinguish
between different ‘safety classes’ which are defined to depend on the ‘consequences-of-failure’. Thus the
‘consequences-of-failure’ partial coefficient as defined in the ISO-2394 is closely related to Nordic safety
class concept. The NKB report specifies the target reliability as shown in Table A.3 for the three safety
classes in the form of both safety index and formal probability of failure.
The connection between safety class and consequences-of-failure is given in Table A.4 from NKB-55E.
Table A.3 Stipulated values of the safety index and the formal failure probability for the ultimate limit
state for ductile failure without safety reserve capacity. Corresponding consequences-of-failure materials
partial coefficient γn.
Safety Class Safety Index Target reliability (failure probability per year) γn
-4
Low 3.71 10 0.9
-5
Normal 4.26 10 1.0
-6
High 4.75 10 1.1

Table A.4 Safety classes vs. consequences of failure
Consequences of failure Safety class

Slight personnel injury and insignificant public loss Low
Some personal injury or significant public loss Normal
Extensive personal injury or very significant public loss High
Table A.3 also shows the consequence-of-failure partial coefficient γn. The implication of γn = 1 for the
normal class is, that the basic target failure probability for normal class constructions and for ductile
failure is put at 10-5 per year. In another safety class one modifies the design by changing the safety-class
dependent materials partial coefficients up or down by 10% for high or low safety class, respectively. This
in turn will decrease or increase, respectively, the failure probability by a factor of 10 as also seen in the
table in the particular safety check under study. For completeness, it should be mentioned that γn depends
on both safety class and type of failure in the materials part of DS 409 and the (old) NKB 35. In the new
NKB 55, these two variables are split into two separate (materials) safety factors.
[A1.6]-[A1.9]
The Danish code system consists of the following:
• The basic Danish code DS-409 [A1.6],
• A load detailing code DS-410 [A1.6],
• A number of material codes for separate materials (e.g. DS-412 for steel [A1.7], DS-456 for GRP
materials [A1.8]),
• A number of codes for specialised structures (e.g. off shore constructions, or wind turbines (DS-472
[a1.9]). These codes may specify both loads and resistance either by reference to the relevant
specialised loads or materials codes, or in some cases by defining deviations from these codes when
such deviations are needed.
[A1.6]
The basic safety code DS-409 does not explicitly mention a target reliability as the partial coefficients
method for practical use is formulated largely without reference to probabilistic methods.
DS-409 prescribes the use of the partial coefficient method and gives definitions for:
• The loading system to be assumed with building structures ;1 permanent load + 1 variable load
• Within this loading system, the characteristic values to be used and partial coefficients and possible
reduction factors are given
• An annex (DS 409, annex 1) specifies rules for further material codes in the Danish code system. Here
it suggests partial coefficients to account for uncertainties of different variables. They are given as a
function of percentile used for characteristic values and of the coefficient of variation for the materials
strength.
Hereby, the target reliability is fixed without actually referencing the target reliability value.
For details and for the general philosophy of the partial coefficient method to be used, DS-409 does refer
to and is generally in accordance with NKB-report 35: Guidelines for loading- and safety regulations for
structural design. Now superseded by report 55 + 55E, same title [A1.5].

[A1.10]
The basic Eurocode 1 is ENV 1991-1, part 1 1) is the European equivalent of ISO 2394. Compared to
Danish codes it covers partly NKB 55 plus DIS 409. In Eurocode 1, informative Annex 1 on partial factor
design, “Table A.2” specifies an indicative value for the target reliability index for the ultimate limit state
of β = 4.7 corresponding to a failure probability of 10-6 per year corresponding to Danish safety class
‘high’.
The table is shown here as Table A.5.
Table A.5 Target reliability as defined in Eurocode 1 (informative)

Limit state Target reliability index Target reliability index
(design working live) (1 year)
Ultimate 3.8 4.7
Fatigue 1.5 to 3.8 -
Serviceability 1.5 3.0
(irreversible)
Depends on degree of inspectability, reparability and damage tolerance 2)
1
) ENV 1991-1- part 1: Basis of design and actions on structures-Basis of design, CEN, Bruxelles, 1994
2
) IEC 2394-1986: General principles on reliability for structures. Switzerland 1986.
The Eurocodes do not contain safety classes. They do, however, in several places indicate that in some
cases one may wish to choose a different reliability level for other types of failure or for other
consequences of failure. This can in turn be done by adding or subtracting 0.5 - 1 to the target index. This
seems to indicate, that the basic target reliability is 10-6. The Eurocodes, however, have in most cases not
been probabilistically calibrated. Therefore the connection between the partial coefficient requirements
and target reliability stated is maybe not quite good. We have compared Eurocodes and the Danish codes
and found that they agree reasonably well when considering the use of partial coefficients.
[A1.11]
The ISO 2394 code does not state a target reliability or any partial coefficient values and consequently it
has no target reliability (or safety level) but is mainly a systematising the partial coefficient method. It
should be remarked that the ISO code does suggest the use of a safety factor for consequences-of-failure
and for failure type which is not present in the Eurocodes.
[A1.12],[A1.13]
Section 4.2.4 Page 18: “The target for the reliability of the safety system must be so great that the
probability of a failure of the safety system combined with the probability of a critical fault requiring the
engagement of the safety system is less than 2⋅10-4 per year.”
[A1.14]
This is the British Standard for the design of lattice towers and masts. Part 2 in particular is a guide for
Part1, which is “ Code of Practice for Loading”. It talks about reliability theory and the probability of
failure of towers and masts. The actual strength calculations are performed through application of partial
safety factor methods. Target reliability’s are not given.
[A1.15]
The main code for the design of steel structures in Germany is DIN 18800. This applies a partial safety
factor method for the loads and for the material/resistance to account for uncertainties and imperfections.
In the introduction to this code it is mentioned that characteristic values for load and resistance shall be

derived from the distribution of these values by calculating p-quantiles. The characteristic values together
with the desired target reliability can then be used to determine partial safety factors. But due to practical
reasons the partial safety factors used in the code have been determined beforehand, based on a global
calibration and experience. Furthermore we received the information that there are currently no plans to
define a target reliability for civil structures and re-calibrate the safety factors.
Other relevant civil codes are DIN 4228, Concrete Masts and DIN 4133, Chimneys. Both use partial
safety factors for the wind loads.
The requirements of the German authorities for the approval of wind turbines is described in Appendix 1
of the GL Rules for Certification of Wind Turbines.
[A1.16]
This recommended practice shall replace API RP 2A - WSD, Working Stress Design, which was using
one global safety factor to account for uncertainties in loading and resistance of offshore platforms. It has
been developed for the Gulf of Mexico and forms the basis of ISO/DIS 13819 Part 2, which is currently
developed for world-wide application.
API sponsored research revealed that the WSD method does not provide structures with consistent
component reliability. Therefore the LRFD code was developed and the load and resistance factors have
been matched to their respective uncertainties in order to reduce the reliability variations. The factors have
been calibrated using a reliability model and design experience. LRFD principles have been applied to
several other US structural codes as well, like ACI (reinforced concrete), AASHTO (bridges) and AISC
(steel structures). The reliability model used for establishing the load and resistance factors is based on the
safety index method.
In order to maintain the safety level from the old code, β values of components of existing platforms were
determined with structural reliability methods. From there target β values were determined. LRFD will be
more conservative than WSD for environmental loads and less conservative for gravity loads. This reflects
the greater uncertainty that is present in environmental loads.

Appendix A.2 Risk criteria in the various countries

[A2.1]
This first reference will be treated in more detail, because it also provides useful definitions and general
notions of risks.
The Dutch concept of risk assessment was first introduced into environmental policy in the 1986-1990
Multi-year Programme for Environmental Management. Limits are indicated above which the risks
become unacceptable (maximum permissible levels) and for establishing levels below which they are
negligible.
Risks from the following categories are discussed:
• major accidents;
• substances;
• genetically modified organisms;
• radiation;
• noise; and
• unpleasant odours.
The risk category relevant for wind turbine safety are (major) accidents. The risk associated with such
accidents are known as external safety risks and are established by means of quantitative risk analysis
techniques. Detrimental effects can therefore be expressed in mortality terms, taking account of the fact
that lethal accidents will also result in a number of less seriously injured casualties. The parameters in
such risk analyses refer specifically to individual risk (IR) and group risks (GR). The definitions are:
Individual Risk (IR), the likelihood that a person will suffer a given detrimental effect (here: death) as a
result of exposure to an agent (expressed in probability units per year or related to an average
concentration per year)
The IR is calculated for the person who runs the greatest risk at a given location. In practice, this normally
refers to an individual who is present at a particular place in the vicinity of an industrial site 24 hours per
day. The IR is generally displayed in the form of iso-risk contours. The maximum acceptable level is
defined as 10-6/year. The negligible level is defined as 10-8/year.
In the Netherlands, the following societal risk is defined:

Group Risk (GR), the likelihood per year that a group of at least a certain size will all be victim of a single
accident at one and the same time
The GR should indicate the probability that a certain group of people outside a facility will die due to an
accident caused by the facility. As such, this risk takes account of the surrounding area. GR does not
involve the calculation of the individual risk but rather the risk of a number of deaths. In addition, the
group has to be a fixed and identifiable group of persons, viz. the group of persons actually living at/near
the corresponding site.
The limits chosen for the GR specify that the likelihood of an accident with 10 deaths occurring shall not
exceed one in every hundred thousand years and aim to prevent social disruption (the death of a group of
people all at once). Accidents with even more serious consequences lead to correspondingly greater
degrees of disruption. It is therefore assumed that an n-times larger impact than 10 deaths should
correspond with an n-squared smaller probability of such an accident occurring .
E.g., the maximum permissible levels are defined as 10-5/year for n = 10 or more deaths and 10-7/year for
n=100 or more deaths. The corresponding negligible levels are defined as 10-7/year for n=10 or more
deaths and 10-9/year for n=100 or more deaths.

In estimation of GR and IR, acute deleterious risks are determined on the basis of death up to two or three
weeks after exposure. Exposure is often of short duration, up to a few hours at most.
In contrast to the GR, the IR refers to the degree of danger in the vicinity of the facility irrespective of the
surroundings. In assessing the individual risk at a particular place, it does not matter whether for instance,
a house or a stadium is situated there. However, this information would be relevant when estimating the
group risk.
The levels below which the risks to man as well as to ecosystems can be regarded as negligible have been
defined in principle as 1% of the maximum permissible values wherever possible. This approach has been
adopted to take account of factors such as multiple exposures (additivity of risks and synergism),
uncertainties in the estimates (limited testing and specific sensitivity) and to leave a sufficient margin to
distinguish between maximum permissible levels and the values below which the risks involved become
negligible.
When comparing perceived risks with the limits, a distinction is generally made between existing and new
activities. In the case of existing activities that give rise to situations in which the specified limits are
exceeded, social considerations are often of importance when determining the period over which these
hazards should be curbed. New activities, however, must comply with the specified limits immediately.
[A2.2], [A2.3]
These two references discuss the Dutch group risk in a more general framework that includes a policy
factor β in the societal risk criterion. This enables relative voluntariness and economic benefits to be
included and gives greater flexibility than the original VROM criterion and level (in [A2.1]). The
framework was tested on: (1) the extension of the national airport Schiphol (Amsterdam), (2) transport of
dangerous chemicals over water, (3) flooding of polders, (4) car traffic, (5) chemical plants, and (6)
operation of moderns wind turbines in housing areas. Examples for (1) -(4) are given, but unfortunately
not for (6).
[A2.4] (Ref. [A2.4] also occurs as [A3.1] (Mortality rates observed)
This reference is interesting because it also discusses aspects like risk perception, voluntary versus
involuntary risks, and the different bases on which to calculate and to compare risks. Generally, external
safety risks are expressed in probabilities of death per year (see [A2.1]), but also other bases are possible.
E.g., for energy producing installations, one could also calculate the probability of death per energy output
generated (/MWe*year).This reference also presents a variety of observed risk figures (mortality rates,
number of accidents, etc.) of different industries and activities.
[A2.5]
Most important part of this reference for the EWTS-II sub II project is the way GR can be calculated. E.g.,
ref. [A2.4] allows the use of factors correcting for the probability of people living in the vicinity of the
installation. However, both ref. [A2.1] and [A2.4] explicitly state that for IR a 24 hour per day exposure to
the risk should be assumed.
[A2.6]
Following the Piper Alpha Disaster in 1989, in which 167 lives were lost on a UK Offshore Installation,
the UK Government commissioned an enquiry about offshore safety, which was lead by Lord Cullen.
Based on the report of Lord Cullen a whole new set of offshore legislation for the UK Continental Shelf
was originated. The aim was to move from prescriptive legislation to a goal setting legislation, through
application of the ALARP principle, meaning that the duty holder has to demonstrate that risk associated
with his installation is “As Low As Reasonable Practicable”. One of the means to demonstrate ALARP
shall be through the application of QRA.
The Safety Case Regulations are part of this legislation and in the guidance to those the UK Health and
Safety Executive (HSE) requires in Para. 117 a demonstration that “the frequency with which accidental

events will result in loss of integrity of the Temporary Refuge, within a minimum endurance time, which
needs to be determined for the actual installation, does not exceed the order of 1 in 1000 a year”. The
demonstration shall be by QRA techniques, where the HSE will make proper allowance for the
uncertainties in QRA.
[A2.7]
This report describes the nature and definitions of risks and discusses risk targets and goals in various
countries. It appears that only the Netherlands and the USA have limits for individual risk and societal
risk, with the Netherlands as being the most stringent (1E-6/yr for individual risk). The UK uses the
concept of tolerable risk, with 1E-4/yr as just tolerable level for the public and 1E-6/yr as broadly
acceptable. Denmark and France use as acceptable levels 1E-6/yr, but they are more a guide than
mandatory.

Appendix A.3 Mortality rates observed

[A2.1]
Ref. [A2.1] also contains relevant material for part 3 “Mortality rates observed”. E.g., the next table shows
observed annual mortality rates in the Netherlands.
Table A.6 Annual mortality rate associated with certain occurrences and activities in the Netherlands
Activity/Occurrence Annual mortality rate
Drowning as a result of a dike collapse 10-7 1 in 10 million

Bee sting 2⋅10-7 1 in 5.5 million
Being struck by lightning 5⋅10-7 1 in 2 million
Flying 1.2⋅10-6 1 in 814,000
Walking 1.85⋅10-5 1 in 54,000
Cycling 3.85⋅10-5 1 in 26,000
Driving a car 1.75⋅10-4 1 in 5,700
Riding a moped 2⋅10-4 1 in 5,000
Riding a motorcycle 1⋅10-3 1 in 1,000
Smoking cigarettes (1 packet a day) 5⋅10-3 1 in 200
[A3.1]
Ref. [A3.1] also occurs as Ref. [A2.4]
[A3.2]
Railway casualties in Denmark (population 5 million):
1992 1993
Passengers 26 30
of whom killed 6 1
Employees 3 32
of whom killed 1 1
Fatal casualties in road traffic accidents (1993): 596 (≈ 3⋅10-4)
Appendix A.4 Wind turbine failure data

[A4.1] No relevant information.
[A4.2]
Based on figures and statistics in WindPower Monthly, the following values were derived for blade

failures (loss of blade).
Table A.7 Probability of loss of blade per year per turbine
Year Probability of blade loss Reference

1987 3⋅10-2 (500 years of operation) Dutch data*
1987 - 1989 2⋅10-3 - 5⋅10-3 Danish data*
1987 - 1989 3⋅10-3 (22000 years of operation) American data*
1991 - 1992 5⋅10-3 (4000 years of operation) Wind Power Monthly
∗ The data was based on an earlier investigation “De Vries et al Vooronderzoek omgevingsveiligheid
wind turbines (In Dutch, Pre investigation on safety of wind turbine surroundings), 1989”.)
[A4.3]
The results of an analysis of reported blade failures are given in Table A.8.
Table A.8: Number of (reported) turbines with a blade failure
1990 1990 1990 1991 1991 1991 1991 1992 1992 1992 1992 1993 To-
apr jul oct jan apr jul oct jan apr jul oct jan tal
may aug nov feb may aug nov feb may aug nov feb
jun sep dec mar jun sep dec mar jun sep dec mar
BLADES
1. Overspeed
1 1 1 1 4
2. Cracks,
Storm 1 3 3 2 8
3. Lightning,
Others 1 1 1 2 1 1 3 10
Total no. of
Turbines 1829 1831 1873 1964 1988 1915 2146 2034 2092 2113 2187 2241
Total number of years of operation: ≈ 6000 yr

Poverspeed = 4/6000 = 6.7⋅10-4 /yr
Pnormal operation = 8/6000 = 1.3⋅10-3 /yr
Plightning = 10/6000 = 1.7⋅10-3 /yr


APPENDIX B. MORTALITY RATE CALCULATION

This appendix presents the details of the assumptions for the mortality rate calculation and the details of
the calculational procedure. Section 6.2 of the main report provides the background, results and the
conclusion of this calculation.
Appendix B.1 Assumptions

• Only one accident scenario, viz. overspeed with given number of revolutions/min with one blade
failure frequency (BFF);
• All wind turbines are of equal design, viz. the 3-bladed 500 kW, pitch controlled design, with a
diameter of 40 m from the ECN/TNO risk study from September 1993 [B1]);
• No park effects, i.e. the risk of n WTB’s in a park is n * the risk of the individual WTB;
• The individual risk (IR) will be used to calculate the mortality rate (MR);
• The base case calculation will take the figures from the aforementioned ECN/TNO study:
BFF = 1E-3/yr (generic value: 1E-3 to 5E-3/yr);
Max. ‘conditional’ IR from one WTB = 1E-4 at 1 m; 1E-5 at 20 m and 1E-6 at 80 m; as a rule of
thumb, a conditional IR of 1E-6 at 2 * diameter is suggested;
The ECN/TNO study revealed that the combined effect of more than 1 WTB on the IR is negligible;
• The number of WTBs is equal to 400;
• No more than one fatality per accident.
The actual risk is dependent on:

• design (height tower, blades)
• location (site) of WTB, incl. wind regime at the site, population and activities around WTB
(determines exposure fraction), (site also determines BFF!)
• blade failure frequency (BFF)
• number of WTB’s
• exposure fraction
The boldfaced parameters are included in the calculation. The other ones are neglected or are implicitly
included in other parameters.
Sensitivity analyses may be performed to change some of the other figures.
Appendix B.2 Calculational procedure

The calculation will be performed as a spreadsheet calculation. This facilitates sensitivity analyses.
Parameters:
• Blade failure frequency (normally derived from a design and site-specific analysis);
• Number of wind turbines;
• Conditional probability of being hit assuming 24 hour exposure (normally derived from a design and
site-specific analysis);
• Exposure fraction (to differentiate between different locations; normally derived from a design and
site-specific analysis).

Exposure fractions:
• corresponding to and in accordance with the definition of Dutch IR (very conservative);
• corresponding for a farmer working 8 hours a day without shielding (less conservative);
• more realistic value, but still estimated to be conservative.
Appendix B.3 Reference

[B1] ‘A. Vrouwenvelder, L. Rademakers: Risico-analyse voor het windplan nabij de Roompotsluizen,
93-CON-R1055, TNO, ECN, September 1993 (In Dutch, Confidential report)

APPENDIX C. BENCHMARK ON DATA COLLECTION

AND PARAMETER ESTIMATION
This appendix presents the details of the benchmark study. In addition, it provides also more practical and
detailed guidance for the data collection and parameter estimation.
Appendix C.1 Guidance for benchmark exercise

Appendix C.1.1 List of Basic Events (BEs) from EWTS-I case study
This list is given in Appendix A of Volume 4B of the EWTS-I project:
NUPRA 2.2 Property of ECN File: EWT.BED Page 1

Created: Mon Sep 25 16:41:18 1995 Printed: Mon Sep 25 16:41:33 1995
Basic Event Id Point Est Description
DE-CPU---------- ERR 1.00E-003 CPU-failure
-
DE-SEN----------- LF 1.00E-003 Failure of overspeed relay
DE-SEN-HISPS- LF 1.00E-003 SENsor on HIgh SP. Shaft loses func. (LF)
-
DE-TC------------ LF 1.00E-003 Thermo couple fails
-
EL-CON-H------- FTO 1.00E-003 Contactor H fails to open.
-
EL-CON-K1------ FTO 1.00E-003 Contactor K1 fails to open.
EL-CON-K4------ FTO 1.00E-003 Contactor K4 fails to open
EL-CON-KM---- FTO 1.00E-003 Contactor KM fails to open
-
EL-CON-OT----- FTO 1.00E-003 Contactor OT fails to open
-
IE-S7 2.00E+000 Generator temp. over 150 C AND wind speed over 5 m
MB-BP------------ WEA 1.00E-002 Too much wear of brake pads.
-
MB-CYL---------- STU 1.00E-002 Cylinder stuck.
MB-DIS----------- FRA 1.00E-002 Fracture of disk or shaft.
MB-FU------------ BLK 1.00E-002 Filter unit blocked
MB-SPR---------- LF 1.00E-002 Insufficient spring force.
-
MB-TVA--------- BLK 1.00E-002 Throttle valve blocked.
-
MB-VAL-Y1----- STU 1.00E-002 Valve Y1 stucked.
MB-VAL-Y2----- STU 1.00E-002 Valve Y2 stucked.

TB-CP--1--------- STU 1.00E-002 Catchplate stucked to magnet.

-
TB-CP--2--------- STU 1.00E-002 Catchplate stucked to magnet.
-
TB-CP--3----- STU 1.00E-002 Catchplate stucked to magnet.
TB-DAM-1--- STU 1.00E-002 Damper stucked.
TB-DAM-2--- STU 1.00E-002 Damper stucked.
TB-DAM-3---- STU 1.00E-002 Damper stucked.
TB-HIN-1----- BST 1.00E-002 Hinge broken or stucked.
TB-HIN-2----- BST 1.00E-002 Hinge broken or stucked.
TB-HIN-3----- BST 1.00E-002 Hinge plate broken or stucked.
TBXTIP-123--- FTD 1.00E-003 Common cause failure of three tip brakes.
Appendix C.1.2 Information needed for each BE

For each of the basic events, one has to specify:
1. The component failure mode type, i.e.:
i. repairable, monitored component (failure ‘directly’ detected, after which repair follows; the
component is unavailable during the detection and repair time).
ii. periodically tested component;
iii. constant unavailability/frequency type; or
2. Depending on type, the following parameters have to be provided:
i. a failure rate plus ‘repair time’, with the repair time here detection time plus logistic plus actual
repair time;
ii. a failure rate plus a test interval;
iii. an unavailability or frequency;
Moreover, for failure rates and unavailability’s, error factors should be provided.
3. For each of the selected types, the reasons and source(s) of information;
4. For each of the parameters, the reasons, source(s) of information.
Appendix C.1.3 Sources of information

In general, for a quantitative safety and reliability analysis, the following sources of information for the
failure data may be used:
• turbine specific data (not available here!);
• generic wind turbine failure data, e.g. from generic wind turbine databases, from existing case studies
or from tests of involved components;
• generic industry-averaged failure data; and
• engineering judgment.
If more sources are used for one particular parameter, the way these sources of data are combined to
obtain one estimate, should be described.
Appendix C.1.4 Failure type model

(referred to as ‘component types’ in EWTS-I report, see Table 10.1, Vol. 4A)
After having identified the sources of information, the following has to be done for each basic event:
1. define component type (constant unavailability/frequency, periodically tested, repairable
(monitored));
2. dependent of type, data has to be collected.

NUPRA supports 5 mathematical models for deriving point estimates of BEs, viz.
Repairable, Type 1:
represents monitored event whose failure is revealed and repaired during operation. A linear
approximation is used to give the average unavailability U = λ ⋅ θ (failure rate ⋅ mean time to repair). If the
component cannot be repaired during operation (which is the case for wind turbines, the repair time can be
viewed as detection time).
Periodically tested, Type 2:
used to model standby components whose failure is detected on test or on demand.
U = (λ ⋅ T - 1 + exp(- λ ⋅ T))/(λ ⋅ T), where T = mean time between tests. This expression approximately
equals ½ λ ⋅ T for λ ⋅ T « 0.1.
Failure on demand, type 3:
models average probability of failure on demand of a component. It is also usable in cases where the
probability of failure is known but the event type is not, and for initiating event frequencies. Other types
of events can be temporarily changed to type 3 to allow editing the unavailability if the need arises.
Running failure, type 4:
models a component with a constant failure rate during its mission. It is usable for components known to
be available at the start of the mission and which have to stay unavailable during the mission time (e.g., up
to a moment the installation has been brought to a safe condition).
U = 1 - exp(- λ ⋅ Tm), where Tm equals the mission time.
Module, type 5:
not used in this study.
Appendix C.1.5 Example of data to be provided for each BE:

As an example of the way the data should be provided to ECN, the BE DE-SEN-HISPS-LF is given:
DE-SEN-HISPS-LF SENsor on HIgh SP. Shaft loses func. (LF)
7e-6/hr based on NEDWIND40 case which used for this component generic data
contained in [Denson, IEEE]. It is not yet clear if the value represents an average or if
the two references provide the same value.
Basic Event Id Failure Model Un’av’ty Fail Rate Time Factor
DE-SEN-HISPS-LF Type 2-Periodical 3.00E-002 7.00E-000 8.76E+003
So, a failure rate of 7e-6/hr. The component is periodically tested once a year (T=8760 hours), according
to the description in the EWTS-I report. Failure of the component is not directly detected (may be
dormant). It is not clear if this sensor is demanded more frequently than once per year. If such information
is known, the average demand period could be taken less than one year.
Appendix C.2 Overall approach ECN

For this benchmark, use was made of the most recent (confidential) ECN wind turbine case study
(November 1996, referred to as further as ‘LAST’). For this LAST study, 6 generic databases (from other
than the wind turbine industry) were consulted. For the EWTS-II benchmark study, these individual
generic databases (reports) were not consulted again. Rather, the information was directly taken from the

LAST study. The information taken were failure rates (per hour), for a particular component and failure
mode.
One problem was that in the LAST report, no details were given on how data from more than one generic
source were combined. The impression is that the average failure rates were presented.
This information was supplemented with experience from other wind turbine case studies (mainly
LW15/75 and NEWECS-45) and engineering judgment.
The design information was taken from the EWTS-I report. This applied to the type of components
(whether it was a monitored one with self-annunciating features of its failure, whether it was a standby
periodically tested component) and test interval parameters (a test interval of one year was mentioned).
Additional assumptions had to be made on parameters like detection, repair times, test intervals, the
frequency of normal shutdowns, and the frequency of visits to the wind turbine.
Test intervals of standby components
The turbine is tested/maintained once a year, so this gives an upperbound of the test interval. It is assumed
that test/maintenance will reveal any dormant failure contained in the basic event list. For certain
components, actual demands occur more frequently than this yearly interval. In that case, the test interval
could be set to the average interval between two demands on that component. This applies in particular to
the components involved in the normal shutdowns.
However, for this demonstration case, such information is not available, but an assumption could be made
on the number of normal shutdowns per year.
The components demanded in a normal shutdown are:
K1, K3, K4, and CPU.
Investigation of the FTs shows that the following components/BEs are also ‘demanded’ in these cases, i.e.,
failure of them would be identified if the normal shutdown should fail:
Via K1: all MB BEs; Via K3: KM and the CCF of the TB; Via K4: K6.
In addition, a potential overspeed situation is also assumed to occur relatively frequently. (E.g, 5 times per
year). The following components are demanded in such cases:
Overspeed relay; OT, H, KM and all MB BEs; All independent TB BEs.
Frequency of visits and detection times
(Engineering Judgment, for a real case this can easily be derived from design specific information sources)
It is assumed that the wind turbine is at least visited once per day. So, for the monitored components the
detection time is no more than 24 hours and failure of such a component in an operating wind turbine will
not last longer than this 24 hours.
Frequency of a normal shutdown
The frequency of a normal shutdown is estimated as once per month, so T is set to 730 hours.
Frequency of overspeed
The frequency of a potential overspeed situation is estimated as five times per year, so T is set 1752 hours.
Summary of data and rationale
The data for the failure rates and rationale are listed below.
BASIC EVENT ID DESCRIPTION/RATIONALE FOR FAILURE RATE/FREQUENCY/ETC.

DE-CPU-------ERR CPU-failure
NEWECS-45 study showed no CPU failures in contrast to
electrical/hydraulic/mechanical failures: therefore, a factor of 10-100 compared
to, e.g., electrical failures see also page 47, 2-nd bullet, ECN-C--94-066 So,
take the least failure rate, divide by 10
DE-SEN-------LF Failure of overspeed relay
7e-6/hr based on NEDWIND40 (Denson, IEEE). Not clear if it represents an average
or if all references provide the same value. Note T taken as average recurrence
time of an overspeed alarm
DE-SEN-HISPS-LF SENsor on HIgh SP. Shaft loses func. (LF)
7e-6/hr based on NEDWIND40 (Denson, IEEE). Not clear if it represents an average
or if all references provide the same value. Note T taken as average recurrence
time of an overspeed alarm
DE-TC--------LF Thermo couple fails
35E-6/hr based on NEDWIND40 (Cremer, Green). Not clear if it represents an
average or if all references provide the same value.
EL-CON-H-----FTO Contactor H fails to open
EL-CON-K1----FTO Contactor K1 fails to open
EL-CON-KM----FTO Contactor KM fails to open
EL-CON-OT----FTO Contactor OT fails to open
For all contactors, a generic failure rate of 1E-6/hr is used. Such value is
referenced in more than one generic source, according to the NEDWIND40 study
(Denson, Davidson, Dummer, IEEE)
IE-S7 Generator temp. over 150 C AND wind speed over 5 m
LW15/75 study: 0.038/year (unconditional of wind speed)
MB-BP--------WEA Too much wear of brake pads.
Least of all other failure rates has been used (Engineering Judgment)
MB-CYL-------STU Cylinder stuck.
53E-6/hr based on NEDWIND40 (Green, Denson, Davidson, IEEE). Not clear if it
represents an average or if all references provide the same value.
MB-DIS-------FRA Fracture of disk or shaft.
0.02E-6/hr based on NEDWIND40 (Green)
MB-FU--------BLK Filter unit blocked
Least of all other failure rates has been used (Engineering Judgement)
MB-SPR-------LF Insufficient spring force.
12.5E-6/hr based on NEDWIND40 (Green, Denson). Not clear if it represents an
average or if both references provide the same value.
MB-TVA-------BLK Throttle valve blocked.
Used value 37E-6/hr, based on NEDWIND40 (Cremer, Green, Denson, Davidson, IEEE).
It is not clear if this value was the average of these 5 references For the
throttle valve, the same value is used (Engineering Judgmemt)
MB-VAL-Y1----STU Valve Y1 stucked.
throttle valve, the same value is used (Engineering Judgememt)
MB-VAL-Y2----STU Valve Y2 stucked.
throttle valve, the same value is used (Engineering Judgmemt)
TB-CP--1-----STU Catchplate stucked to magnet.
Use similar value as for hinge

TB-DAM-1-----STU Damper stucked.

Use similar value as for hinge
TB-HIN-1-----BST Hinge broken or stucked.
TB-HIN-2-----BST Hinge broken or stucked.
TB-HIN-3-----BST Hinge plate broken or stucked.
For hinge, 1E-4/hr based on NEDWIND40 (Denson1988).
TBXTIP-123---FTD Common cause failure of three tip brakes.
CCF unavailability takes into account independent failure rate, the number of
tip brakes (3), and CCF parameters for CCF involving 2 and 3 tip brakes; A
generic beta factor of 0.1 is taken, hence CCF is 0.1 of independent failure,
hence 3*0.1*1E-4
Appendix C.3 Overall approach RISØ

Risø used three types of data sources:
• “Nordic generic”: a pseudonym for a Scandinavian manufacturer. Risø got the permission to use this
“real” failure data for a 150 kW turbine on the condition that the real name would remain confidential;
• FARADIP.THREE database: an English database for component reliabilities. It can be bought from
TECHNIS, 26 Orchard Drive, Tonbridge, Kent TN10 4LG, UK. Risø has had it here for some years,
and has used it a good deal.
• Engineering judgment: Rise considers the design method as follows: Such components are designed
with a 20 years life expectancy and it is allowed that 5 percent of the items designed that way may fail
before 20 years have passed. If you assume (which the design rules do) that the life expectancy of
such a mechanical component has a normal distribution, and at the same time you take (as the design
rules state, too) , 0.2 as parameter for the normal distribution, it is possible to calculate the mean life
expectancy of the items, whereby you end up with 29 years (if only 5 percent are allowed to fail
before 20 years). The mean time between failures at 29 years correspond to a failure rate of 3.8e-6/hr.
The data and rationale for the first iteration is presented below. The data for the second iteration is
presented in Appendix C.4.
DE-CPU-ERR CPU-failure
Data are “Nordic generic”, 0.7e-6/hr
Basic event Id Failure model Un’av’ty Fail rate Time factor
DE-CPU-ERR Type 1-repairable 0.7e-6 *)
*) = 24 hours
DE-SEN-LF
DE- SEN-LF Type 1-repairable 0.7e-6 *)
DE-SEN-HISPS-LF
DE- SEN-HISPS-LF Type 1-repairable 0.7e-6 *)
DE-TC-LF
Data from FARADIP data base, average value
DE-TC-LF Type 1-repairable 10e-6 *)

EL-CON-H-FTO
EL-CON-H-FTO Type 1-repairable 0.7e-6 *)
EL-CON-K1-FTO
EL-CON-K1-FTO Type 1-repairable 0.7e-6 *)
EL-CON-K2-FTO
EL-CON-K3-FTO
EL-CON-K4-FTO
EL-CON-K5-FTO
EL-CON-K6-FTO
EL-CON-KM-FTO
EL-CON-KM-FTO Type 1-repairable 0.7e-6 *)
EL-CON-OT-FTO
EL-CON-OT-FTO Type 1-repairable 0.7e-6 *)
IE-S7
Data for generator 0.7e-6 are “Nordic generic”, while it is assumed that the probability of a wind speed >
5m/s is 0.5. The resulting rate is calculated as the product of these 2 values, i.e. 0.35e-6/hr.
IE-S7 0.35e-6

MB-BP-WEA
“Nordic generic data” has a total value for MB-BP-WEA, MB-CYL-STU, MB-DIS-FRA, MB-FU-BLK
and MB-SPR-LF as a system, so 1/5 of the value is taken as the value for each.It is not known, whether
this division in equal values is correct.
MB-BP-WEA Type 1-repairable 0.14e-6 *)
MB-CYL-STU
and MB-SPR-LF as a system, so 1/5 of the value is taken as the value for each.It is not known, whether
MB- CYL-STU Type 1-repairable 0.14e-6 *)
MB-DIS-FRA
and MB-SPR-LF as a system, so 1/5 of the value is taken as the value for each. It is not known, whether
MB- DIS-FRA Type 1-repairable 0.14e-6 *)
MB-FU-BLK
MB-FU-BLK Type 1-repairable 0.14e-6 *)
MB-SPR-LF
MB- SPR-LF Type 1-repairable 0.14e-6 *)
MB-TVA-BLK
MB-TVA-BLK Type 1-repairable 2.8e-6 *)
MB-VAL-Y1-STU
MB-VAL-Y1-STU Type 1-repairable 0.7e-6 *)
MB-VAL-Y2-STU
MB-VAL-Y2-STU Type 1-repairable 0.7e-6 *)
TB-CP-1-STU

Data are average for heavy duty magnets taken from the FARADIP database.
TB-CP-1-STU Type2-periodical 3.5e-6 8760
TB-CP-2-STU
Data are average for heavy duty magnets taken from the FARADIP database.
TB-CP-3-STU
Data used are average values for heavy duty magnets, taken from the FARADIP database.
TB-DAM-1-STU
Data used are for hydraulic pistons, taken from the FARADIP database
TB- DAM-1-STU Type2-periodical 1e-6 8760
TB-DAM-2-STU
TB-DAM-3-STU
TB-HIN-1-STU
Data are engineering judgement based on a design to last for 20 years with a 5 percent fractile of a normal
distribution (= 0.2).
TB- DAM-3-STU Type2-periodical 3.8e-6 8760
TB-HIN-2-STU
distribution ( = 0.2).
TB- HIN-2-STU Type2-periodical 3.8e-6 8760
TB-HIN-3-STU
distribution ( = 0.2).
TB- HIN-3-STU Type2-periodical 3.8e-6 8760
TBXTIP-123-FTD
Failure assumed to be 24 V AC failure, i.e. transformer failure, data from FARADIP database
Common cause event Failure model Un’av’ty Fail rate Time factor
TBXTIP-123-FTD Type 1-repairable 0.4e-6

Appendix C.4 Data and results

The data used and the results are summarised in the following spreadsheets. RISØ2 stands for the second
iteration data of Risø.
The total operational time should be assessed according to the formula;
Nf Ns
T total = (t Now - t Past ) N WTB - ∑ (t Now -t ) - ∑ (t is -t Past )
i
f
i =1 i =1
where tPast is defined as the date of reporting the first failure in the database, NWTB is the number of wind
turbines under observation, Nf is the number of wind turbines removed from operation within the interval
[tPast , tNow ], Ns is the number of wind turbines put into operation within the interval [ tPast , tNow]. The
remaining variables are shown in Fig. D1

Data for case study EWTS-2
Basic Event Id Failure Model Un'av'ty Fail Rate

ECN RISO RISO-2 ECN RISO RISO-2 ECN RISO RISO-2
DE-CPU-------ERR Type 1 Type 1 Type 1 4.8e-8 1.7e-5 1.7e-5 2.0e-9 7.0e-7

DE-SEN-------LF Type 2 Type 1 Type 2 6.1e-3 1.7e-5 3.1e-3 7.0e-6 7.0e-7
DE-SEN-HISPS-LF Type 2 Type 1 Type 2 3.0e-2 1.7e-5 3.1e-3 7.0e-6 7.0e-7
DE-TC--------LF Type 2 Type 1 Type 2 1.4e-1 2.4e-4 4.3e-2 3.5e-5 1.0e-5
EL-CON-H-----FTO Type 2 Type 1 Type 2 8.8e-4 1.7e-5 3.1e-3 1.0e-6 7.0e-7
EL-CON-K1----FTO Type 2 Type 1 Type 2 3.7e-4 1.7e-5 3.1e-3 1.0e-6 7.0e-7
EL-CON-KM----FTO Type 2 Type 1 Type 2 3.7e-4 1.7e-5 3.1e-3 1.0e-6 7.0e-7
EL-CON-OT----FTO Type 2 Type 1 Type 2 8.8e-4 1.7e-5 3.1e-3 1.0e-6 7.0e-7
IE-S7 Type 3 Type 1 Type 1 3.8e-2 8.4e-6 3.1e-3 3.5e-7
MB-BP--------WEA Type 2 Type 1 Type 2 7.3e-6 3.4e-6 6.1e-4 2.0e-8 1.4e-7
MB-CYL-------STU Type 2 Type 1 Type 2 1.9e-2 3.4e-6 6.1e-4 5.3e-5 1.4e-7
MB-DIS-------FRA Type 2 Type 1 Type 2 7.3e-6 3.4e-6 6.1e-4 2.0e-8 1.4e-7
MB-FU--------BLK Type 2 Type 1 Type 2 7.3e-6 3.4e-6 6.1e-4 2.0e-8 1.4e-7
MB-SPR-------LF Type 2 Type 1 Type 2 4.6e-3 3.4e-6 6.1e-4 1.3e-5 1.4e-7
MB-TVA-------BLK Type 2 Type 1 Type 2 1.3e-2 6.7e-5 1.2e-2 3.7e-5 2.8e-6
MB-VAL-Y1----STU Type 2 Type 1 Type 2 1.3e-2 1.7e-5 3.1e-3 3.7e-5 7.0e-7
MB-VAL-Y2----STU Type 2 Type 1 Type 2 1.3e-2 1.7e-5 3.1e-3 3.7e-5 7.0e-7
TB-CP--1-----STU Type 2 Type 2 Type 2 8.3e-2 1.5e-2 1.5e-2 1.0e-4 3.5e-6
TB-DAM-1-----STU Type 2 Type 2 Type 2 8.3e-2 4.4e-3 4.4e-3 1.0e-4 1.0e-6
TB-HIN-1-----BST Type 2 Type 2 Type 2 8.3e-2 1.7e-2 1.7e-2 1.0e-4 3.8e-6
TBXTIP-123---FTD Type 2 Type 1 Type 2 1.1e-2 2.0e-6 8.7e-3 3.0e-5 1.0e-6 2.0e-6
Time Factor Ratio Unáv'tRatio Fail RaRatio Time FNewvalue for sensitivity
ECN RISO RISO-2 ECN/RISO ECN/RISO ECN/RISO Sens. 1 Sens. 2
DE-CPU-------ERR 24 24 24 0.003 0.003 1 4.8e-8

DE-SEN-------LF 1750 24 8760 363.7 10.0 72.9 1.7e-4
DE-SEN-HISPS-LF 8760 24 8760 1785.7 10.0 365.0 1.7e-4
DE-TC--------LF 8760 24 8760 579.2 3.5 365.0 8.4e-4
EL-CON-H-----FTO 1750 24 8760 52.1 1.4 72.9 2.4e-5
EL-CON-K1----FTO 730 24 8760 21.7 1.4 30.4 2.4e-5
EL-CON-K2----FTO 8760 24 8760 260.1 1.4 365.0 2.4e-5
EL-CON-K3----FTO 730 24 8760 21.7 1.4 30.4 2.4e-5
EL-CON-K4----FTO 730 24 8760 21.7 1.4 30.4 2.4e-5
EL-CON-K6----FTO 730 24 8760 21.7 1.4 30.4 2.4e-5
EL-CON-KM----FTO 730 24 8760 21.7 1.4 30.4 2.4e-5
EL-CON-OT----FTO 1750 24 8760 52.1 1.4 72.9 2.4e-5
IE-S7 24 8760 4523.8 na na 3.8e-2
MB-BP--------WEA 730 24 8760 2.2 0.1 30.4 4.8e-7
MB-CYL-------STU 730 24 8760 5684.5 378.6 30.4 1.3e-3
MB-DIS-------FRA 730 24 8760 2.2 0.1 30.4 4.8e-7
MB-FU--------BLK 730 24 8760 2.2 0.1 30.4 4.8e-7
MB-SPR-------LF 730 24 8760 1354.2 89.3 30.4 3.0e-4
MB-TVA-------BLK 730 24 8760 199.4 13.2 30.4 8.9e-4
MB-VAL-Y1----STU 730 24 8760 797.6 52.9 30.4 8.9e-4
MB-VAL-Y2----STU 730 24 8760 797.6 52.9 30.4 8.9e-4
TB-CP--1-----STU 1750 8760 8760 5.4 28.6 0.2 4.3e-1
TB-CP--2-----STU 1750 8760 8760 5.4 28.6 0.2 4.3e-1
TB-CP--3-----STU 1750 8760 8760 5.4 28.6 0.2 4.3e-1
TB-DAM-1-----STU 1750 8760 8760 18.9 100.0 0.2 4.4e-1
TB-DAM-2-----STU 1750 8760 8760 18.9 100.0 0.2 4.4e-1
TB-DAM-3-----STU 1750 8760 8760 18.9 100.0 0.2 4.4e-1
TB-HIN-1-----BST 1750 8760 8760 5.0 26.3 0.2 4.3e-1
TB-HIN-2-----BST 1750 8760 8760 5.0 26.3 0.2 4.3e-1
TB-HIN-3-----BST 1750 8760 8760 5.0 26.3 0.2 4.3e-1
TBXTIP-123---FTD 730 2 8760 5450.0 30.0 365.0 6.0e-5

Results case study EWTS-2
Comparison ECN and RIso results; based on importances

and on totals 2-nd (final) iteration
Importances ECN
Rank EVEN POINT EST. F-V IMPORT RSK ACMT RISK RED Corr. rank
4 MB-CYL----- 1.9e-2 79% 41.8 4.9 1

5 TBXTIP-123 1.1e-2 41% 38.5 1.7 2
6 TB-HIN-1/2/ 8.3e-2 19% 3.2 1.2 3
9 TB-DAM-1/2/3-----STU
12 TB-CP--1/2/3-----STU
15 MB-SPR--- 4.5e-3 19% 42.4 1.2 4
16 MB-VAL-Y 1.3e-2 1% 2.1 1.0 5
17 EL-CON-K 3.6e-4 1% 29.1 1.0 6
18 MB-TVA--- 1.3e-2 1% 1.6 1.0 7
19 MB-VAL-Y 1.3e-2 1% 1.6 1.0 7
20 MB-FU----- 7.3e-6 0% 42.2 1.0 8
21 MB-DIS---- 7.3e-6 0% 42.2 1.0 8
22 MB-BP----- 7.3e-6 0% 42.2 1.0 8
Note: IEF and XCOM have been deleted from the lists (no component failure modes)
Importances RISO
Rank EVEN POINT EST. F-V IMPORT RSK ACMT RISK RED Corr. rank
4 MB-CYL----- 6.1e-4 20% 322.2 1.2 2

3 TBXTIP-123 8.7e-3 96% 110.1 24.1 1
12 TB-HIN-1/2 1.6e-2 1% 1.5 1.0 6
19 TB-DAM-1 4.4e-3 0% 1.5 1.0 9
15 TB-CP--1/2 1.5e-2 1% 1.5 1.0 7
5 MB-SPR---- 6.1e-4 20% 322.2 1.2 2
10 MB-VAL-Y 3.1e-3 1% 5.9 1.0 4
9 EL-CON-KM 3.1e-3 4% 12.9 1.0 3
11 MB-TVA--- 1.2e-2 1% 2.0 1.0 5
18 MB-VAL-Y 3.1e-3 0% 2.0 1.0 8
6 MB-FU------- 6.1e-4 20% 322.2 1.2 2
7 MB-DIS------ 6.1e-4 20% 322.2 1.2 2
8 MB-BP------- 6.1e-4 20% 322.2 1.2 2
Totals
IEF S7 OS Freq Cond. prob
ECN 3.8e-2 2.0e-5 5.3e-4
RISO 8.4e-6 7.1e-15 8.5e-10
RISO2 3.1e-3 8.3e-8 2.7e-5
Difference
ECN/RISO 4523.8 2.8e+9 6.3e+5 first iteration
ECN/RISO2 12.4 242.2 19.6 second iteration

APPENDIX D. DERIVATION OF FAILURE RATES FROM

REPORTED FAILURES
The general formula for assessing the failure rate of the ith basic event, λBEi , (h-1) is as follows:
k
∑n
j =1
ij
λ BEi =
mi T total
where nij is the number of occurrences of the ith basic event for the jth WTB; mi is the number of similar
components on one WTB; k - the number of wind turbines of a given type; Ttotal (h) -total operational
repair time for all k WTBs. To understand the link between a basic event and a component, it is enough to
say that a basic event is completely defined by the system, component, and the failure mode, i.e. the same
component can belong to different basic events.
The parameter Ttotal can be calculated in different ways depending on particularities of the data collecting.
So, if failure data are collected from all wind turbines in operation, then Ttotal = Σk tj where tj is the total
time of operation of the jth wind turbine of a given type. In case the failure data are collected from a fixed
time in the past tPast we have to take into account five different wind turbines life lines (see Fig. D.1)
5
4
3
2
t1s t2s t3s t1f tPast t4s t2f t5s t4f tNow time
Figure D.1 Possible life lines of wind turbines; tis is the time of the start of the ith wind turbine, tif is the
stop of the operation, tPast is the time of the beginning of the data collection, and tNow is the present time.

PART 3
INTEGRATION OF BLADE TEST IN DESIGN
B.H. Bulder (edt.) Netherlands Energy Research Foundation ECN

P. Vionis Centre of Renewable Energy Research CRES
F. Arias Vega Centro de Energeticas Investigationes
Medioambientales y Technologicas
J.C. Sanz-Martı́n idem
M.W. Jensen Risø National Laboratory
Chr. Brokopf Germanischer Lloyd
Integration of Blade Test in Design
3.2 European Wind Turbine Standards II

CONTENTS OF PART 3
1 INTRODUCTION 7
1.1 General : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7
1.2 Project work : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 8
2 DISCUSSION ON THE VALUE OF TESTING OF ROTOR BLADES FOR DESIGNERS 9

2.1 General : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9
2.2 Property tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9
2.2.1 Static tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9
2.2.2 Natural vibration tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10
2.3 Fatigue tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 11
2.3.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 11
2.3.2 Damage equivalent testing : : : : : : : : : : : : : : : : : : : : : : : : : : : : 11
3 CONCLUSION 15
3.1 General : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 15
3.2 Rotor blade property tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 15
3.3 Stiffness/static strength tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 15
3.4 Fatigue strength tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 15
REFERENCES 17
A CURRENT TEST PRACTICES 19

A.1 General : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 19
A.2 Background : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 19
A.3 Test Principles or Approaches : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 20
A.3.1 Tests giving statistic information : : : : : : : : : : : : : : : : : : : : : : : : : 20
A.3.2 “Empiric” tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 20
A.3.3 Tests based on a probabilistic approach : : : : : : : : : : : : : : : : : : : : : 20
A.4 Material and component tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 21
A.5 Full scale strength tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 21
A.5.1 Ultimate strength tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 22
A.5.2 Fatigue strength test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 22
A.6 Full scale blade property tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
A.6.1 General : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
A.6.2 Deflection : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
A.6.3 Natural frequencies : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
A.6.4 Mass distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
A.6.5 Stiffness distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
European Wind Turbine Standards II 3.3

A.6.6 Mode shapes : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24

A.6.7 Strain distribution measurements : : : : : : : : : : : : : : : : : : : : : : : : 24
A.6.8 Damping : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
A.6.9 Creep : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
A.6.10 Other non destructive testing : : : : : : : : : : : : : : : : : : : : : : : : : : : 25
A.6.11 Blade sectioning : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 25
B CURRENT STANDARDS AND INFORMATION FROM PREVIOUS PROJECTS 27

B.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 27
B.2 Standards and Codes : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 27
B.2.1 Wind turbine engineering related standards : : : : : : : : : : : : : : : : : : : 27
B.2.2 Non wind turbine engineering standards : : : : : : : : : : : : : : : : : : : : : 27
B.3 Research information for verification/validation test : : : : : : : : : : : : : : : : : : : 29
B.3.1 Strength and Fatigue of Large Size Wind Turbine Rotors S&FAT : : : : : : : : 29
B.3.2 European Wind Turbine Testing Procedure Development : : : : : : : : : : : : 29
B.3.3 PRObability FAtigue Reliability : : : : : : : : : : : : : : : : : : : : : : : : : 29
C GUIDELINES FOR TESTS TO VERIFY ENGINEERING MODELS 31

C.1 Overview : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 31
C.2 Full scale tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 31
C.2.1 Purpose of tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 31
C.2.2 Design requirements (elements of design principles) (design methodology) : : : 32
C.2.3 Types of testing : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 32
C.2.4 Aeroelastic model verification : : : : : : : : : : : : : : : : : : : : : : : : : : 34
C.2.5 Structural model verification : : : : : : : : : : : : : : : : : : : : : : : : : : 34
C.2.6 Sensitivity analysis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 35
C.2.7 Required accuracy : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 35
D GUIDELINES TO INTERPRET RESULTS OF MECHANICAL TEST 37

D.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 37
D.2 Need of tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 37
D.3 Causes of the differences between the blade theoretical calculations and the tests results 37
D.4 Classification of static tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 39
D.5 Blade identification measurements : : : : : : : : : : : : : : : : : : : : : : : : : : : : 40
D.6 Natural Vibration Test (eigenfrequencies) : : : : : : : : : : : : : : : : : : : : : : : : 41
D.6.1 Natural Blade Frequencies : : : : : : : : : : : : : : : : : : : : : : : : : : : : 41
D.6.2 Damping : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 41
D.7 Elastic test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 41
D.7.1 Equivalent stiffness test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 42

D.7.2 Influence coefficient test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 43

D.8 Static loads strength test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 44
D.8.1 Limit load strength test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 44
D.8.2 Ultimate load test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 46
E CRUSHING 1 49


1 INTRODUCTION
1.1 General
The general aim of the project European Wind Turbine Standards II is to remove vital constraints and
bottle necks in the technical harmonisation of the European wind turbine market.
Concerning sub project 3: Integration of Blade Test in Design, this objective can be met when a
recommended methodology is developed to include full-scale blade tests in the wind turbine design
process. The project will be performed along the following activities:
1. an evaluation of the known test practices,
2. evaluation of existing standards,
3. specification of guidelines which items have to be taken into account,
4. specification of guidelines how to interpret the test results with respect to the design criteria,
5. investigation of potential valuation of a static and fatigue strength test.
Full scale blade tests, as they are performed nowadays, are performed to demonstrate adequate safety
margin and verify design calculations. Actual test and measurement procedures do not need to be
addressed here, especially because they are in development under the coordination of a working group of
IEC TC-88WG8.
During the performance of the work a continuing discussion has been held whether the general objective,
the deliverables and the to be performed activities are in compliances with each other.
The performance of mechanical tests on rotor blades is already a fact today, although the tests are usually
performed at the end of the design process. Consequently the test results are not used to upgrade the
design, unless the experiments showed that the design criteria are not met. These test are performed to
demonstrate, to a reasonable level of certainty, that the rotor blade’s, when produced according to certain
specifications, specified limit states are not reached. In other words that the rotor blade will survive the
design load spectrum. The usefulness of these test for the designer to improve the design is discussed in
section 2.
Due to the fact that the conditions during the test, specifically for the strength and fatigue test, are different
than in real service, and that the test has to be performed in a limited time period, implies that the test load
spectrum is not equal to the design load spectrum. This makes full-scale testing, definition according to
[3]:
both the test specimen and the loading spectrum, including sequence should be representative
of actual service conditions
in fact not possible.

The full-scale tests which are performed nowadays usually have the objective to show that the construction
and manufacturing is sound and has no flaws in them which would result is a catastrophic failure during its
design life. However it is not possible to determine the actual margins of the individual failure mechanism
let alone the combination of all failure mechanism. The results can also be used to verify whether the
proper quantities have been used in the design analysis, e.g. whether the eigen frequencies of the blade
assumed in the aero-elastic calculations is close to the measured ones.

To make a procedure where the tests should be incorporated in the design process at an earlier stage is not
that simple because it is not possible to test a blade before the prototype is build. In fact only procedures
what to do with the tests results can be developed while it is the decision of the designer whether he
wants to redesign his rotor blade using the results of the tests. However the trend in most technologies is
to reduce the time to market, which implies for rotor blade manufacturers that rotor blades are produced
already in series before the testing is complete. Thus design alterations are only performed when there is
a necessity.
In view of the above the authors assigned to this sub project, found it not possible to come up with a
recommended methodology to include full-scale blade test in the wind turbine design process.
This part of the project describes the value of full-scale and component testing for designers. Components
are included because they are usually tested at a much earlier stage in the design process. Especially
when a new kind of construction is applied there is a need to have an engineering model, to dimension
this part. These models can be based on a physical relation or empirical relations. The engineering model
is usually verified with a number of test. These test are designed and analysed by the designers.
1.2 Project work

In this sub project we will come up with recommendations how to perform model verification tests of
structural details or the full scale rotor blade and how to interpret the measured results.
Material characterisation and damage models are left out of this project. Material characterisation test
are well defined in a number of ISO/DIN/ASTM test standards. Fatigue damage models are probably of
such a fundamental level, and the cost to update the used models will probably be so high that it can not
be expected that research funds available for wind energy industry will be directed to this area.
Test which are meant here are test to verify engineering models of e.g. blade root connections or shear
web – shell connections. One could also think of buckling load test of parts of the rotor blade, with a
realistic ratio between the bending moments and the shearing forces.
The engineering models verified by mechanical tests are models to determine the static strength, stiffness,
fatigue life and stability of structural details which are common to modern large rotor blades.
The deliverables of the work as reported should contain information concerning recommended method-
ologies to include blade tests in the wind turbine design process in order to:
1. demonstrate adequate safety margin;
2. verify design calculations
All tests are split up in (blade) property tests, strength tests, and fatigue tests. For the activities 1 and 2,
see previous page, use can be made of the reports and guidelines of the S&FAT [1] and IEC-TC88 WG8
[2] documents or any other document or standard which is directed to material and or structural testing.

2 DISCUSSION ON THE VALUE OF TESTING OF ROTOR BLADES FOR DESIGN-

ERS
2.1 General
The objective to perform full-scale tests on rotor blades is usually to show that the blade has the properties
which are assumed in the design calculations and that the strength, stability and fatigue design are
sufficient. These tests convince customers and the certification authority that the blade design and
manufacturing meets the design requirements.
These test do not prove that the design will not fail during service because the tests are designed with
Design load spectrum as input and not with the Service load spectrum, see figure 2-1.
Other reasons, more aimed at improving the design process, to perform full-scale tests are, validation of
computer programs or the model behind the program and determination of the weakest spot or section.
These full scale tests are normally not used to modify or optimize the design when all tests results are
satisfactorily. When one or more tests show that the blade does not meet the design targets, and resulting
in a blade not fit for purpose, it will be modified.
Wind & grid Static Strength

Wind Turbine Service Loads Stability In Service
conditions Fatigue Strength
Material & Static Strength

Wind & grid Design Loads Design
Wind Turbine Structural Stability
Model Modelling Modelling Fatigue Strength
Material &
Structural
Modelling
Wind Turbine Static Strength

Test Loads Material Stability Test
Rotor Blade
Specimen Properties Fatigue Strength
Figure 2.1 The scope of mechanical test for rotor blades
2.2 Property tests

2.2.1 Static tests
For strength/stiffness tests one can quite easily show that when the stress/strain distribution that is
measured differs from the calculated or assumed distributions one has to go back to the design calculations.
One has to check whether the measured blade properties are within the design margins for all failure
phenomena, like strength, fatigue or stability. Also the reason behind the difference between the measured

and predicted stress/strain distributions should be determined. This is however not a simple problem since
there are many factors that could influence the results. The differences can occur due to a bad engineering
model or due to differences in the test w.r.t. the design load or due to differences between the design
specification and the manufactured rotor blade. The differences due to bad engineering model are of
most interest for this project. Differences due to the fact that the test loading is different from the design
loading, e.g. shearing forces during the test are usually higher than the design loads due to concentrated
load load application, are not very interesting for the design – test evaluation. However analysis of a
failure should include the possibility of such effects. When the origin of the difference is found one can
update the design analysis procedure and could even optimize the blade when the error is larger than
expected or required and it is attractive for cost saving to change the design.
Determination of the actual strength margins is more difficult, especially when the blade specimen also
has to be used for the fatigue test. However even when it is allowed to destroy the blade one has to choose
the loading situation for which the margin is determined. See for this problem figure 2-2 from the S&FAT
report [1].
Peak&Trough Load combinations for R = 0 [m]
N load
4.0e+05 E load
S load
M lead-lag [Nm]
0.0e+00
-4.0e+05
-8.0e+05
-8.0e+05 -4.0e+05 0.0e+00 4.0e+05 8.0e+05 1.2e+06
Mflatwise [Nm]
Figure 2.2 Binned peak&trough bending moments at rotor blade "Fl" cross section 0.00 [m], excluding
lim . PLF
2.2.2 Natural vibration tests
The same reasoning holds for natural vibration tests, when large discrepancies are found w.r.t. the used
or calculated eigenfrequencies in the design analysis, the designer has to repeat (some of) the aero-elastic
calculations to determine the sensitivity of the response of the wind turbine towards the eigenfrequencies
for the given external conditions and determine the loads on the rotor blades. If the sensitivity is large the
design analysis has to be repeated with the measured values. It is also interesting to find the reason why

the eigen frequency is so much different than predicted. This could lead to better prediction methods for
future designs.
2.3 Fatigue tests

2.3.1 Introduction
The reasoning for the fatigue test differs from the blade property tests. This is due to a number of reasons:
Experimental fatigue results have a much larger scatter in life (or number of cycles): it is assumed
that for coupons the test life is between 0.3 – 3.0 times the design life [3] than the scatter in static
strength.
The used allowable material properties in the fatigue analysis are very conservative. The character-
istic values which are used are not less than 95 % survival probabilities with 95 % confidence limits,
assuming a normal distribution of test results of these parameters thus normally fatigue failure will
not occur.
These allowable material parameters values are reduced by uncertainty factors which are assumed
in the design analysis, e.g. for environmental effects (moisture, UV etc.), which are not present in
the tests.
Another major difference of fatigue tests w.r.t. strength tests is that one has to modify the design
load spectrum to a test load spectrum. A test has to be performed in a relatively short time, say
between 3 and 6 month while the design life is in the order of 20 – 30 years. This means that
one has to modify the design load spectrum by relying, or on the basis of, the fatigue damage
formulation used in the design analysis. This is in fact very awkward, because one of the objectives
of performing full-scale fatigue test is to check whether the fatigue damage formulation is also
correct for a complete rotor blade.
The conversion method between a design load spectrum to a test load spectrum is based on the
following general procedure. One omits a large number of small load variations of which the
fatigue damage is small or negligible, increase the load application rate and increases the range of
a number of the remaining load cycles with a TLF (test load factor) to compensate for the cycles
that are omitted.
One other factor, of unknown magnitude, which could be included in the test load, a factor to
compensate for testing only one blade. Due to scatter in (fatigue) strength and the possibility that
the blade which is tested is of better quality than an average production blade one has to increase
the loading to make sure that when a blade resists the test one can assume that a large majority of
blades of the same type can resist the loading.
2.3.2 Damage equivalent testing
To do a meaningful (realistic) test, where the test fatigue damage is determined using the appropriate
factors, see the design criteria of IEC 1400-1, for the test conditions and the test fatigue damage is at least
equal to the design fatigue damage. The total load factor, the ratio between test loads and design loads,
can be quite high resulting in fatigue loads higher than the maximum loads for a static strength test.

In formula form this comes down to the following simple equation:
Dtest Ddesign (2.1)
In which the D stands for theoretical determined fatigue damage.

For the result of the fatigue analysis it doesn’t matter whether one assumes partial load/material factors
on the loads or on the allowable material values. According to ISO 2394 [4], the design values of the
loads are given by the equation
=
Fd f Fr (2.2)
and the design (allowable) values for the material is given by the equation
fd = fk m (2.3)
The f and m are functions which represent a number of effects. Especially for the material factor some
effects are sometimes present in service but not in the more benign laboratory conditions. Which in the
end means that the factor m reduces and consequently the loads have to be increased to make sure that
the theoretical fatigue damage in the test equals the theoretical fatigue damage in the design analysis.
Then there are also the factors
C which describes acceptable deformations, accelerations, crack widths with respect to the service-
ability limit state;
ad characteristic values of geometrical parameters;

n which is a coefficient by which the importance of the structure and the consequence of failure are
(is) taken into account;
d which is a coefficient related to model uncertainties or other circumstances which is not yet taken
into account by other values.
For practical design application the above results in
(
n S Fd ; ad ; Sd ) R (fd; ad; C; Rd ) (2.4)
) 1 R (fd; ad; C )
or in
(
n Sd S Fd ; ad (2.5)
Rd
In which d has been split into a part for the load (S) and an part for the resistance (R) side of the analysis.
On paper it doesn’t matter much where, on the loads or on the resistance side of the equation, an effect
is taken into account. For testing however it does matter. All factors which one has applied in the design
analysis on the resistance side but do not need to be applied in the test analysis e.g. the factor for humidity
- in a laboratory environment the humidity is usually low - does not need to be applied in the fatigue
test analysis. As a result the loads will need to be multiplied with the factor to reach the design fatigue
damage. This can result in a test load spectrum which can be close to or even above the maximum design
loads expected in the life of a wind turbine. This is not acceptable because other (non-linear) failure
mechanisms can or will than become dominant and destroy the rotor blade which in reality would not
occur.

In the IEC-TC88 WG8 guidelines test are divided into strength based and load based test. The load based
test are not very useful to determine margins in the structure and the strength based test could be useful
when one assume that failure of a cross section or spot in a construction is a one or two dimensional
effect.
A simple example of non-linear effects on stresses in rotor blades is the crushing load. These loads work
on the shear beam and are quadratic with the bending moment, see appendix E. A more serious non-linear
effect is buckling which is an actual design driver for large size rotor blades. In the worst case buckling
appears as a bifurcation phenomena, leading to a drastic reduction in stiffness and severe damage due to
secondary bending.
One has also to be careful with applying loads higher than 1.35 times the normal design loads. The load
factor which has to be applied in the design for the extreme loads is 1.35. For a well designed blade
the extreme loads will have almost the same magnitude as the maximum loads in the so called fatigue
load cases. This is especially true in the flapwise direction. The fatigue phenomena which are active can
change due to the much higher stress levels and result into fatigue failures which are not realistic.
Taking the above consideration into account one cannot say anything about the actual fatigue design
margins of a blade which doesn’t show major fatigue damage after the test. Damage occurring in the start
of a test usually means that the blade has a serious fatigue problem.
These considerations on fatigue testing do however not mean that one can say that full-scale fatigue testing
is not necessary (useful). Most tests which were performed resulted in modifications of the design. A
favourable effect of the highly non-linear fatigue process is that (small) errors in the design analysis
show already early in the test that something is not designed or manufactured in accordance with the
specifications. These results protect the wind turbine and rotor blade manufacturers from early failures
and large cost associated with re-fitting the wind turbines with new blades or repair of the existing blades.


3 CONCLUSION
3.1 General
The value of full scale blade test, that is rotor blade property, strength and fatigue tests, is high for
designers however not all test are of equal weight. This is due to the used method and the position in time
of the design process that they are performed. Usually these kind of tests are performed for certification to
show that the construction “contains” sufficient margin and not to determine the actual static strength and
fatigue life margins. This implies that only a failure during the test gives information about the margin
to the required static/fatigue strength. However due to common practice nowadays, to use the same rotor
blade for the static and fatigue test implies that it is normal not to go to failure in the static test. Thus the
static strength margin is not determined. The fatigue test is less valuable for designers due to the fact that
the test load spectrum is determined from the design load spectrum using the same models as used in the
design analysis. So it is impossible to make solid conclusions about the true margin of the fatigue life for
the structure under the design load spectrum.
Model verification test are valuable for designers. In the end all engineering models have to be validated
and or verified by these kind of test. The result of these test can be used to improve or “tune” the engineering
model and after a while, when already many products have been produced which are dimensioned with
the (engineering) model a designer relies on these models. Testing of that part is then not required any
more.
However, usually the need for cheaper and or lighter construction lead the designer to other, more
advanced, structural details which need more and other test to derive and validate/verify engineering
models.
3.2 Rotor blade property tests

The rotor blade property test is most valuable for the designer. The properties determined here are mass
(distribution, centre of gravity, eigenfrequencies possibly including the mode shapes, damping etc.
With the results of this test the designer can get confidence that the assumption made in the design
analysis, mainly the aero-elastic load calculations, are good or good enough. This implies that the design
load spectrum is well determined. This is the basis for continuation of the tests. When the difference
between test results and assumption, is too large the assumed design load spectrum is probably not correct
and continuation of the test, no matter what the result is, can not be used to prove that the rotor blade is
fit for purpose.
3.3 Stiffness/static strength tests

This test is also valuable for designers, the test results can be used to improve the (engineering) model to
determine the strains and subsequent determine the static and fatigue strength margins. It is also useful to
check the actual stiffness distribution, especially when the mode shapes are not measured in the property
test.
3.4 Fatigue strength tests

The fatigue strength test is less valuable for the designer, due to the fact that the test load spectrum differs
quite a lot with the design load spectrum.

However, fatigue test are still the only way to get sufficient confidence that the structure, design and
manufacturing, is sound. It still happens that rotor blades fail during the fatigue test while the static
strength or property test did not show large differences with the design analysis. Detailed analysis of the
failure usually leads to the conclusion of a design error or something which is not taken into account in
the design analysis.

References
[1] Grol, H.J. Van. “STRENGTH AND FATIGUE TESTING OF LARGE SIZE WIND TURBINE
ROTORS; Volume I: Rationale and directions.”. ECN 95-C–052, ECN, December 1995.
[2] Ann. “Draft Standard of the International Electrical Commity Wind Turbine Generator Systems IEC
1400-1, Part; Full scale structural testing of rotor blades for WTGS’s”. Draft, IEC-TC88-WG8,
September 1997.
[3] Collins, J.A. Failure of materials in mechanical design. Analyses - Prediction - Prevention. John
Wiley & Sons, 1981.
[4] Ann. “General principles on reliability for structures”. Technical Report ISO 2394, International
Standardisation Organisation, 1986.
[5] Smet B.J. de and P.W. Bach. “Lifetime predictions of glass fibre reinforced polyester with the
database FACT.”. -C- 94–044, ECN, April 1994.
[6] Smet B.J. de and P.W. Bach. “DATABASE FACT

FAtigue of Composites for wind Turbines.”. -C- 94–045, ECN, April 1994.
[7] Joosse P.A. and B.J. de Smet P.W. Bach, B.H. Bulder. “Emperical Fatigue Formulation for GRP,
discussion paper in Workshop on Fatigue, 10 April 1994”. Technical report, SPE, April 1994.
[8] Delft van D.R.V., G.D. de Winkel, and P.A. Joosse. “Fatigue behaviour fo fibreglass Wind Turbine
Blade Material under variable amplitude loading.”. In Proceedings of the 4th IEA expert meeting
on wind turbine fatigue, pages 55–62, February 1996.
[9] Reifsnider, K.L., editor. Fatigue of Composite Materials. Elsevier Amsterdam, 1991.
[10] Hartog J.P. “Strength of Materials”. Dover Publications Inc., New York, 1977.
[11] Delft D.R.V. van and J.L. van Leeuwen. “Full scale fatigue testing wind turbine rotor blades”. In
Proceedings of the EWEC Conf. Thessaloniki 1994, pages –, October 1994.
[12] Delft van D.R.V., G.D. de Winkel, and P.A. Joosse. “IEC TC88 WG 8 Testing of Rotor Blades.”. In
Proceedings of the 4th IEA expert meeting on wind turbine fatigue, pages 55–62, February 1996.
[13] Musial W.D., M.E. Clark, N. Egging, and M.D. Zuteck. “Comparison of Strength and Load Based
Methods for Testing Wind Turbine Blades”. In Walt Musial and Dale Berg, editors, A collectin of
the 1997 ASME Wind Energy Symposium Technical Papers – at the – 35th AIAA Aerospace Sciences
Meeting and Exhibition, pages 97–0957, January 1997.
[14] Philipidis Th. and P. Vionis. “Rotor blade design verification by means of full scale testing.”. In
Proceedings of the ECWEC conference held at Dublin 1997, pages ?? – ??, October 1994.
[15] Mandell J.F., D.D. Samborsky, and D.S..Cairns. “Avanced wind turbine blade structures development
program at Montana State University”. paper 97-0952, AIAA, 1997.
[16] Musial W.D. and C. Butterfield. “Using Partial Safety Factors in Wind Turbine Design and Testing”.
In Proceedings of the Windpower ’97, Austin, Texas, page ??, June 1997.

[17] Madsen P.H. et al. “Recommended practices for wind turbine testing and evaluation”; 3. FATIGUE
LOADS. International Energy Agency, 2 edition, 1990.
[18] Vega F. A. and E.S. Lascorz. “STRENGTH AND FATIGUE TESTING OF LARGE SIZE WIND
TURBINE ROTORS; Volume II: Full size natural vibration and static strength test; a reference
case.”. REFCIE - 43D21/02, CIEMAT, July 1996.
[19] Gere J.M. and S.P. Thimoshenko. “Mechanics of Materials”. Van Nostrand Reinhold (U.K.) Co.
Ltd, 2 edition, 1987.
[20] Bruhn E.F. and et al. “Analyses and Design of Flight Vehicle Structures”. Tri-State Offset Company,
USA, 1973.

A CURRENT TEST PRACTICES
A.1 General
This appendix contains a short description of full scale tests as they have been performed until now
including static and fatigue strength tests as well as blade property tests (strain/stress distributions, mode
shapes, damping etc.). Further tests performed to determine material and component characteristics will
be mentioned briefly. A more detailed description of especially test methods and procedures is given
in the draft technical IEC report CDV IEC-61400-23 "Full Scale Structural Testing of Rotor Blades for
WTGS’s".
A.2 Background
Normally blades are designed according to some national or international standards or codes such as IEC
61400-1 "Wind Turbine Generator Systems - Part 1: Safety Requirements" which uses the principles
of ISO 2394 defining the limit states and partial coefficients, which have to be applied to obtain the
corresponding design values.
The reason to perform full scale tests is basically to verify or demonstrate to a reasonable level of
certainty that the tested item - here the wind turbine blade - when manufactured according to a certain set
of specifications has the prescribed strength and/or safety margins with reference to specific limit states
and has the assumed blade properties as e.g. stiffness or natural frequencies. I.e. that the blade therefore
possesses the strength, service life and other properties provided for in the design. The partial coefficients
given in most material standards reflect uncertainties and are chosen in order to keep the probability of a
limit state being reached below a certain value prescribed for the structure.
The need for tests will depend on the level of uncertainty in the design assessment due to the use of new
design concepts, new materials, new production processes etc. and the possible impact on the structural
integrity. Normally full scale tests are only performed on a very limited number of samples. Because of
the costs only one or two blades of a given design are tested, so no statistical distribution of production
blade strength or other properties can be obtained.
A variety of different tests and test methods for wind turbine blades is (and has been) in use world wide.
Full scale tests but also component tests and material tests are and have been performed in relation to
the development, documentation and verification of wind turbine blades. Roughly the used tests can be
divided into three main groups defined by the methods/approaches they are based on:
Tests giving statistic information
"Empiric" tests
Tests based on a probabilistic approach.
These approaches are described in the following section.

In Europe most full scale blade tests are performed in connection with certification, but also tests
supporting the design process are performed.

A.3 Test Principles or Approaches

A.3.1 Tests giving statistic information
Full scale tests on series of blades giving statistic information as mean values and standard deviations are
more or less unknown. The costs for such a kind of investigation have until now been judged as too high
compared with the profits. A few full scale statistical investigations on strain distribution, stiffness and
deflections and geometrical variations have been performed, but only to a very limited level.
But for material characterisation the statistical approach is the current practice when performing tests
on coupons. Series of tests on a lot of different materials have been performed in many years to obtain
specifications of static strength, fatigue strength (normally defined as S-N curves) including investigations
on the influence of R-ratios, stiffness etc. (see paragraph A.4).
Also on some wind turbine details as glued in bolts and studs the statistical approach has been used
determining these values (see section 2.4).
A.3.2 “Empiric” tests
Most full scale blade tests performed until now are performed in accordance with procedures and methods
defined either by the blade manufacturer himself or by national standards, recommendations or guidelines.
Only one or a few blades are tested and the methods and procedures are more based on "thumb rules"
defining reasonable empiric test factors than on straight statistic or probabilistic approaches to the testing.
That means that the test factors are more based on values origin from a "best guess" based on reasonable
reasons and experience than on scientific investigations.
A.3.3 Tests based on a probabilistic approach
As described above it remains to investigate and formulate how tests on one or a few blades can be used
to draw conclusions about the strength and other properties of a series of blades of the same kind.
The determination of the actual margins to the limit states is desirable because such margins can provide
a measure of the actual safety obtained for the resistance of the test blade. Interpretation of such values is
not straightforward and presumable probabilistic methods have to be applied. It can be done combining
the ultimate limit states and probability density functions for e.g. load and resistance.
The principles can be explained with reference to figure A-3 The probability density functions for load
and resistance f(L) and f(R) are shown and the characteristic values for the same two functions are denoted
Lc and Rc. These two values usually correspond to 5
The probability density function of the load is nowadays more or less well known as defined in e.g. the
IEC 61400-1 standard. The problem is that the full resistance density function (including e.g. the pure
material values, scaling and shape factors, environmental influences, blade to blade variations, influences
of the chosen test method etc.) is not known yet. For some materials as glass fibre reinforced polyester
the pure material characteristics are known, but not the full resistance density function for the whole
blade.
More running projects as the EU subsided SMT-project "Blade Test Methods and Techniques" and the
JOULE-project PROFAR have the goal to investigate some of the unknown aspects.

Figure A-1 Load- and resistance-functions and characteristic and design values for load and resistance.
A.4 Material and component tests

Material tests are normally performed to derive material properties and strength formulations from series
of tests using statistic methods. This type of tests is well known and described and forms the basis of
most national and international material standards.
As mentioned material tests have been performed for some of the normally used blade materials. A
European database for values for fibre reinforced plastics is established in the so called FACT database
[5–7] by ECN in the Netherlands.
Component tests are and have been performed e.g. on blade roots, root bolts/studs or special details as
e.g. bonded connections. Some of these tests are or have been performed on a larger number of items
giving statistical information on the detail (average, standard deviation etc.). Others are only performed
on one or a few items just to determine the behaviour of the detail.
Some blade root tests in 2D and 3D have been subsidised by EU.
A.5 Full scale strength tests

A lot of different test methods and procedures has been used until now in different countries (see e.g.
CDV IEC 61400-23). They have different advantages and disadvantages, but some problems are general
and independent of the chosen method. In full scale testing it is not possible to load the blade with all
load components and/or to establish the same conditions as in the design. As no single test can load the
whole blade optimally, representative loading should be applied to the areas defined as critical for the

construction. Beyond that fatigue testing must be accelerated by increasing the test load to expose the
blade to sufficient fatigue damage within a reasonable period. Other differences from the ideal loading
such as increased shear loading and other problems caused by concentrated load application are also
general.
The difference between just proving the blade to a prescribed load and finding the actual reserve factor or
safety margin leads to two principally different attitudes to full scale strength testing namely load based
tests and strength based tests. These two types of tests have been used for different reasons, the load
based mainly in connection with certification of blades and mainly in Europe, the strength based mainly
in connection with design valuation and mainly in USA.
Using load based testing each design load case has to be defined to allow the test load to be accurately
assessed at the critical areas to be tested. If strength based testing is used, the appropriate blade strength
at each relevant cross section has to be given.
For both types of tests the assumptions for the design operating conditions must be clearly stated as well
as the test environmental conditions. In many cases it may be necessary to modify the test load to account
for differences between the test conditions and the environment assumed in the load or strength data sets.
The basis for the load based testing (or design load-envelope testing) is the total collection of design
loads. Its purpose is to show that the blade can sustain the intended loads without failure. The test load
is generally a reduction of the distributed aerodynamic and inertial blade moments into discrete forces
positioned along the span. Testing to destruction is not required.
Strength based testing uses as-manufactured blade strength data as its basis. It is often intended to load
the blade to destruction. Strength based testing allows a direct verification of the design methodology
and an assessment of ways in which the design computations and the design itself might be improved.
This method can be used to find the lowest strength location, relative to expected strength, within a broad
region.
A.5.1 Ultimate strength tests
Two main methods are and have been used until now loading the blade either by a surface load (e.g. sand
bags) or in one or more concentrated point loads. Using point loading a variety of equipment is and has
been in use and both single axis testing methods as well as combined tests are well known.
As mentioned care must be taken that the loading method will not lead to local buckling or excessive
shears which were not intended or on the other hand lead to local strengthening of the blade.
A.5.2 Fatigue strength test
Many methods to load a blade in fatigue testing are or have been in use. Also here loads can be applied
in a single or more points. The loads can be applied to a single axis or as combined loading. The load
can be of constant amplitude and frequency or variable. Each type has advantages and disadvantages. In
the CDV IEC 61400-23 more of the normal used methods and procedures are described.
A general problem is to accelerate the fatigue test reaching a reasonable test period. Too high loads can
lead to unrealistic failure modes. Besides, the fatigue effects of stress concentrations caused by altered
material composition and/or other structural details, which may have little effect in static loading, must
be accounted for in the fatigue test.

A.6 Full scale blade property tests

A.6.1 General
In the previous sections only strength related tests are dealt with. However, other tests giving additional
information on (other) structural or dynamic properties are important and are also commonly carried out
in many countries. These tests can be performed independently of the strength tests, but normally most
of them will, due to practical reasons, be performed in connection with the strength tests - especially with
the static strength test.
Most of the mentioned blade property tests are used both for verification and as a method to tune the
design tools.
A.6.2 Deflection
For the verification the flatwise deflection of the blade has most significance because of the (normal)
limited clearance to the wind turbine tower but both flatwise and edgewise deflections are often used also
by the designer to check the design tools. During the test both the deflections of the blade and test rig
have to be recorded. The number of displacement measurement locations has to be adequate to determine
the displacement curve and stiffness for the whole blade.
A.6.3 Natural frequencies
The most important frequencies (normally defined as the first and the second flatwise, the first edgewise
and in some cases the first torsion frequency) are normally measured to determine the behaviour of the
blade. For most blades these frequencies are well separated and more or less uncoupled. Consequently,
they can be measured directly one by one by putting the blade into the desired vibration mode while
monitoring signals from e.g. strain gauges or accelerometers representing the vibration mode.
If it is not possible to excite the modes separately, a frequency analysis can be made on signals gained
from exiting the blade in free vibration outside its harmonics.
A.6.4 Mass distribution
A rough mass distribution is given by the total mass of the blade and the centre of gravity normally
already by the blade manufacturer. More refined methods based on sets of strain gauges and the blade
eigenweight have also been used. If necessary the mass distribution can be measured by cutting the blade
in small sections and weighing each section.
A.6.5 Stiffness distribution
The blade bending stiffness in given load directions can be derived from the load/strain measurements or
from deflection measurements (see above).
The load/strain method is very suitable if a detailed stiffness distribution is required - especially in the
root and inner part sections. The strain measurement points have to be chosen carefully and sets of
strain gauges on each side of the blade are to be distributed along the blade. Loading the blade tip and
measuring the strains on each side of the blade the local curvature can be found. From the curvature and
the bending moment the stiffness can be derived.

The torsional stiffness of the blade can be expressed in terms of angular rotation as a function of a rising
torque.
A.6.6 Mode shapes
The mode shapes can be derived by putting the blade into the wanted vibration mode with one accelerom-
eter mounted at the tip and a number of other accelerometers well distributed along the blade.
Further, the normal mode shape values can for to lightly damped linear structures with well separated
natural frequencies be approximated by the imaginary part of the transfer function (at resonance) relating
the force input to the acceleration response at the points where the mode shape values are to be determined.
Flatwise and edgewise measurements can be and have been performed by applying an excitation force
(measured by a force transducer at the frequency of concern) at an appropriate point (mostly the tip) of
the blade while mounted on a rigid test stand. The resulting acceleration responses have to be monitored
in the points of interest. The measurements can then be fed into an analyser which offers the possibility of
extracting the modulus as well as the phase of the complex transfer function at the resonance frequency.
A.6.7 Strain distribution measurements
Strain distribution tests for blades have been performed for many years using strain gauges placed in
areas of interest. The number of measurement points will normally depend on the blade (size, complexity
etc.) and blade strains are then measured on the blade skin, typically at blade locations where geometry
transitions and critical design details are present or where the strain level is expected to be high.
Each measurement point can include up to three strain measurements. If the direction of the principal
strains is not known, a strain gauge rosette can be used to determine the magnitude and direction of the
principal strain values at the measurement points.
In some cases e.g. if only stress concentrations or a certain strain level are of interest, stress coat can be
used. Stress coat is a strain sensitive paint with a well defined cracking limit. This method are normally
only used as a supplement to strain gauge measurements e.g. in areas where it is difficult to use gauges.
Stresses can also be derived using photo-elastic measurements, but until now this method is only used in
a few projects.
A.6.8 Damping
The structural damping are normally recorded for the flatwise and edgewise directions by measuring the
logarithmic decrement of an undisturbed oscillation. The amplitude of the oscillation has to be small to
avoid influence from aerodynamic damping. The damping can be very dependent of the temperature.
A.6.9 Creep
Some materials used for the wind turbine blade are sensitive to creep (e.g. wood). If that is the case tests
can be performed tests to define the creep and recovery characteristics. These test are normally performed
for a longer duration of time under static loading. During the test the deflection is measured frequently
and the deflection versus the time is recorded. After a period the load is removed and the recovery versus
time is again recorded as the blade relaxes.

A.6.10 Other non destructive testing
Non Destructive Testing (NDT) techniques are in some cases used to prove if the blade is built in
accordance to the design assumptions and to find manufacturing defects. NDT can be performed in
connection with other tests or as a part of the quality assurance during the production of the blade .
Some of the methods used are measurements checking the blade geometry and tolerances, coin-tap and
ultra-sound to check primarily for delaminations and bondings etc.
A.6.11 Blade sectioning
Blade sectioning can be used to prove if the blade is built in accordance to the design assumptions and
to find manufacturing defects. E.g. the mass distribution, the (airfoil) geometry and the build-up of
laminates, beams, glued connections etc. can be checked.
Blade sectioning is also used to investigate the failure modes in case of failures during strength tests.


B CURRENT STANDARDS AND INFORMATION FROM PREVIOUS PROJECTS
B.1 Introduction
The current (draft) “Wind Turbine Standards” which are looked at are IEC 1400-1 ed.2 c IEC:1997 - 1 -
88/CJC/XX Safety of wind turbine generator systems and IEC-1400-23 c IEC:1997 Full - scale structural
testing of rotor blades for wind turbine generator systems.
Next to these also some non wind turbine standards, specifically made for material characterisation, are
reviewed whether they contain useful information for full scale tests or component/model verification test.
These other standards are, ISO 2602 1973 (first edition) Statistical interpretation of test results – estimation
of the mean – confidence interval, ISO 1922 1982 (second edition) Cellular Plastics – Determination of
shear strength of rigid materials, ASTM C 297 – 61 (Re-approved 1988) Standard test method for Tensile
Strength of Flat Sandwhich Constructions in Flatwise plane, ASTM D 3410 – 75 (Re-approved 1982)
Standard test method for Compressive Properties of Unidirectional or Cross-ply Fibre-Resin Composites,
ISO 3268 1978 (first edition) Plastics – Glass-reinforced materials – Determination of Tensile Properties,
ASTM D 3039 – 76 (Re-approved 1982) Standard test method for Tensile Properties of Fibre-Resin
Composites, DIN 50 100 1978 Dauerschwingversuch Begriffe Zeichen Durchfüring Auswertung, and
ISO 2394 General Principles on Reliability for Structures.
Finally also volume I - rationale and directions - of the S&FAT project, [1], has been reviewed in view of
this work.
B.2 Standards and Codes

B.2.1 Wind turbine engineering related standards
The IEC-1400-1 mentions in the part structural design that an acceptable safety level shall be ascertained
and verified by calculations and or tests to demonstrate that the design loading will not exceed the relevant
design resistance.
So testing of rotor blades is not required to pass a certification approval using the IEC-1400-1 criteria.
The draft guideline produced by Working Group 8 of TC-88 of IEC states that the purpose of the full scale
tests which are dealt with in their document is “to demonstrate, to a reasonable level of certainty, that a
blade type, when manufactured according to a certain set of specifications, has the prescribed reliability
with reference to specific limit states”. This means that the test is only used to show that it can with
stand the design load envelope and not to evaluate the actual margins to the limit states. It is even stated
that the actual margin can not straightforwardly be obtained. It is mentioned that by using probabilistic
methods it might be possible to obtain the actual margins to the limit states. The main reason that this is
not a practical answer is that for probabilistic methods you need more than one test to produce statistical
data and in reality one needs many tests to come up with reliable data.
B.2.2 Non wind turbine engineering standards
Some remarks found in non wind turbine engineering standards are listed here:

Hardware
Testing machines must have a constant rate of traverse, this is also of importance for testing of larger
specimen like full scale rotor blades or smaller components. The load application rate should be in the
same order of magnitude as is foreseen in actual service.
The grips shall be made so to allow alignment of the specimen axis with the testing machine axis or
direction of applied force. This could be read in such a way that one has to take load angle changes due to
deformation into account during the test. This item is also mentioned in the draft guideline IEC-1400-23.
The test load must be applied with an accuracy of 1% of the design value. This is also true for full scale
or component tests. Especially the fatigue damage formulation is very sensitive to the applied load. 1%
change in load can result already a substantial difference in the number of cycles to failure.
Together with the loads also the measurement of strains should be with an accuracy of at least 1%. The
measurement devices should not alter the properties of the specimen.
Documentation of tests
Next to what is requested in the TC88WG8 document [2] The test report shall include the following
particulars:
complete identification of specimen(s);

method of preparation of specimen;
test conditions like temperature, relative humidity, test speed;
Test results like maximum strength or maximum force or if applicable force at break, modulus of
elasticity;
statistical information like number of specimen, number of test results. Determine standard
deviations and possibly the 95% confidence interval of mean value;
if there is a number of rejected test motivate why they are rejected.
Other items to take into account
Size effects can reasonably well be taken into account for steel structures, they are already incorporated in
the design codes for composite structures. The general effect for metals is that due to larger size/volume
the fatigue properties reduce. An explanation for this is that larger specimen have a higher probability of
containing a more serious stress concentration or fatigue nucleus. For composite structures it is not sure
whether the same explanation holds.
An other effect to take into account is variable – constant amplitude loading. For metals this effect can
be quit dramatically, for composite materials this size of this effect is not yet known, although in a recent
publication [8] it is indicated that the effect could be large.
Statistical Interpretation of Test Results – Estimation of the Mean – Confidence Interval; ISO 2602
This standard deals with estimation of mean and standard deviation for test results which are normal
distributed.

The use of this standard for model verification test is limited.

One can not assume that the distribution is normal at fore-hand.
There is also a strong indication that the strength distribution is more like a Weibull distribution, see [9].
B.3 Research information for verification/validation test

B.3.1 Strength and Fatigue of Large Size Wind Turbine Rotors S&FAT
The S&FAT project, [1] dealt a.o. with the same problem as discussed here, how to make use of test
results in such a way that this or futures design can be improved. One separate item in this project was
the Cost and Benefit question. With a questionnaire the European rotor blade manufactures and wind
turbine industries in general were asked to reveal which test the find necessary and how much (money)
they are willing to spent on testing for a rotor blade of a 700 kW wind turbine rotor. The result showed
that a property and static strength test should always be performed, a fatigue strength test only when
commercial series production is foreseen.
An other subject which got quite some attention in the S&FAT project which specimen, i.e. rotor blade,
of a series production has to be used for testing. This in view of the fact that the first produced rotor blade
will be different to the third and e.g. the tenth rotor blade, due to the fact that manufacturing blades is
still to a large extend performed by humans. It is mentioned that it should not be the first but the third or
fourth blade. That means the first rotor blades (sh)(c)ould be used for the prototype wind turbine.
B.3.2 European Wind Turbine Testing Procedure Development
Within the European program Standard Measurement and Testing a project is ongoing about test proce-
dures and the influence of different test procedures on the results.
Within this project 5 different institutes perform a property, static and fatigue test on identical rotor blades.
At the time this report is written no results are available of this project.
B.3.3 PRObability FAtigue Reliability
The PROFAR project, which will start in the near future, has the folowing objectives:
to determine static and fatigue strength distributions of wind‘ turbine rotor blades;
To determine the so called blade to blade variation factor which should be applied in a test to
compensate for possible scatter;
to generate data which can be used for a probabilistic design tool.
It is not the intention to use real blades in the project but small blades which are specifically designed to
fail in fatigue at load levels which result in fatigue failures below 50 million cycles.


C GUIDELINES FOR TESTS TO VERIFY ENGINEERING MODELS
C.1 Overview
After the loads on an engineering structure have been determined or properly estimated and when from
these loads and the chosen dimensions of the structure the stresses have been calculated, the question
presents itself of whether the construction is safe or not. Since a structure will always start to fail at one
spot (the weakest) first, the question reduces to that of the safety of an element of material subjected to a
given stress pattern. During the last 2 centuries a number of assumptions have been made concerning the
criterion of failure of such an element, and these assumptions have been honoured with the designation
"theory of strength" [10] pp72-73. Full scale testing is used in many countries [1, 11–14] for answering
questions such as whether the construction is safe or not (certification) or as a means for verification of the
engineering models or assumptions. In particular, for blade constructions made of composite materials
using various techniques ranging from hand lay-up in open mould to resin transfer moulding (RTM) the
iterative approach design/ materials/manufacturing/laboratory testing/service testing is considered of high
importance [15] . Due to the complexity of the structure, the method of fabrication and the material itself,
coupon testing or component testing is not considered sufficient for either proof of the blade structural
integrity, or verification of the design assumptions. Coupon testing may provide data which deviate
significantly from the characteristics of the actual lay-up used in the fabricated blades. In addition, the
materials used, including the adhesively bonded joints, still have same uncertainties with respect to their
fatigue or strength properties. Moreover structural details such as adhesively bonded joints cannot always
be analysed in a manageable way with the available engineering tools. The use of testing, in particular
for design verification, is dealt with in the following paragraphs.
C.2 Full scale tests

A guideline of recommended practices for "full scale structural testing of rotor blades for WTGS’s", is
in preparation by the WG8 of the IEC Technical Committee 88,[2]. This part of IEC 1400 is intended to
be used as a guideline for the full-scale structural testing of wind turbine blades and gives guidance for
the interpretation or evaluation of results, as a possible part of a design verification of the integrity of the
blade. The primary emphasis of this WG effort was to identify commonly accepted practices among the
various laboratories and to give guidance in establishing blade test criteria. In the present analysis, the
integration of blade testing in the design procedures is in particular addressed.
C.2.1 Purpose of tests
As specified in[2], the fundamental purpose of a wind turbine blade test is to demonstrate to a reasonable
level of certainty that a blade type, when manufactured according to a certain set of specifications, has
the prescribed safety margin with reference to specific limit states, or, more precisely, to verify that
the specified limit states are not reached and the blade therefore possesses the strength and service life
provided for in the design. Furthermore, blade testing as considered in the present document can play an
important role in helping the designer with the verification of the engineering models used and the design
assumptions.

C.2.2 Design requirements (elements of design principles) (design methodology)
Following the requirements of the new edition of IEC 1400-1, the design of the blades of a wind turbine is
based on the principles of limit state design as described in the standard ISO 2394 [4]. According to this
standard the structural performance of a whole structure or part of it should be described with reference
to a specific set of limit states beyond which the structure no longer satisfies the design requirements. All
limit states should be considered in design. A calculation model should be established for each specific
limit state. This model should incorporate all appropriate variables and also allow for the uncertainties
with respect to actions. According to the standard the purpose of design calculations or prototype testing
is to keep the probability of a limit state being reached below a certain value prescribed for the type of
structure in question. To verify the reliability of the structural design, the format of partial coefficients,
is recommended.
In many cases the design procedure consists of
structural analysis which gives the forces and moments in the cross-sections, and
analysis of cross-sections, joints etc. which gives their resistance and more generally their be-
haviour.
According to the standards provisions the analysis of a structure can be made with the aid of calculation,
model testing or prototype testing. In some cases a combination of these methods is considered as useful.
Calculation models and basic assumptions for the calculation should express the structural response
according to the limit state under consideration.
A structure or part of it may be designed on the basis of results from appropriate model testing coupled
with the use of model analysis to predict the behaviour of the actual structure. In most cases, model
testing is used for structural analysis to verify calculations or as a substitute for them. Conversion of
model test results to be compatible for use in the actual structures should be ensured.
Eventually, a structure or part of it may also be designed on the basis of results from testing prototype
units relevant to the particular design under consideration. This type of testing is sometimes used as a
substitute for calculation to verify the resistance of small units or details. It is important, therefore, that
the prototype testing is conducted as nearly as possible under the same conditions and assumptions as for
the actual structure regarding strength, dimensions, loading and environmental conditions. If only a few
prototype test results are available, statistical uncertainties should be indicated.
As seen in the previous sections, it is of significant importance for the designer both the verification of
the aeroelastic codes used for the load determination and the verification of the structural model used for
stress and strain calculations (FEM or analytical methods).
C.2.3 Types of testing
There are two characteristic methods of blade test loading that can be applied in an actual blade test,
i.e. load based testing and strength based testing. The two methodologies are described in detail
in[2]. The former type of testing is performed to demonstrate that the tested blade, within a certain
level of confidence, has met the structural design requirements concerning its operating or extreme load
conditions. Testing to destruction is neither sought nor required. The basis for the test loads is the entire
envelope of blade design loads, derived according to generally accepted standards such as IEC1400-1.
A load-based test verifies the blade’s ability to withstand a particular load combination, but even after a
successful test many areas of the blade may still be under-loaded relative to the blade’s ability to carry

the load. For a strength based test, the test load is derived from a span wise distribution of the predicted
design strength of the test blade. If the blade is tested to its design strength, the design methodology
is verified for all of the load cases encompassed by the strength-based loads, providing that the design
strength calculations were correct [13].
The differences between the 2 approaches and the principles of the testing procedures can be further
explained in figure C-4, which shows the relationship between the design and test-load factors for the
two cases [16]. In the figure the probability density functions of loads and resistance are shown along
with the characteristic values for both. These values usually correspond to 2.5-5% fractile values of the
involved variables. By applying the appropriate partial coefficients on the variables the corresponding
design values are obtained. In the case of the IEC 1400-1, the value of the general material factor
accounting for the inherent variability of the strength parameter is prescribed to be not less than 1.1 when
applied to characteristic material properties of 95% survival probability, with 95% confidence limit and
a coefficient of variation of 10%. For different survival probabilities and/or coefficients of variation the
relevant general factor is also given. However to derive the global material factor from this general factor
it is necessary to account for scale effects, tolerances degradation due to external actions, i.e. ultraviolet
radiation, humidity and defects that would normally be detected. One should note that in the case of
load-based test the design partial safety factors and appropriate test load factors (TLF) are applied to the
characteristic loads to arrive at the load-based test load. The values of the TLF vary a lot depending on
the method of testing etc. [8], [2, 13], therefore, it is presented as a region in the relevant figure.
On the other hand, a strength test is concerned only with the right side of figure C-4. It uses the 95%
survival value on the "test blade" distribution to set a proof test load with a high probability of not failing
the blade. The remainder between the load based test load and the strength based test load is the reserve
strength.
region IEC design value of

strength
IEC Design Value
of Load • Characteristic
Value of Strength
Characteristic Value • Strength-based
of Load Test Load
Representative TLF
Adjusted Coupon
Loads
Strength
γn⋅γf γm
Mr
Figure C-1 Application of Load Factors for WT blade testing according to IEC standards.
f : Load uncertainty (1.1 for fatigue and 1.35 for ultimate strength)
n : Consequences of failure (1.15 for fatigue and 1.0 for ultimate strength)
m : Materials Strength Uncertainty (1.1 with 95% survival, 95% confidence, 0.1 COV)
Mr : Reserve Margin
TLF : Test Load Factor (according to[2])
Apparently, load-based test loading is not of direct use for the verification of engineering models, though
important for blade certification. As already discussed, the probability of occurrence of any failure when

the blade is subjected to this type of loading is extremely low. Therefore, potential failures indicate
insufficient design procedures, manufacturing quality etc. However firm conclusions on the design
assumptions and the engineering methods followed by the designer can be drawn only if additional tests
have been carried out as discussed in paragraphs C.2.4 and C.2.5. In addition, it should be noted that
further information about failure modes, manufacturing quality and reserve strength can be obtained if the
testing is continued to blade failure. On the other hand strength based testing allows a direct verification
of the design methodology and an assessment of ways in which the design computations, and the resulting
design itself, might be improved. This method can be used to find the lowest strength location, relative to
expected strength, within a broad region. Strength based testing is particularly useful if the test loading
type is desired to be different from that which determined the design, such as if a blade whose design
was determined by fatigue operating loads is tested for extreme loading, or vice versa. For such cases,
the blade strength distribution will often be quite different from design load cases that did not determine
its design. Another purpose might be to assess the strength reduction between bulk material as it exists
within the blade and small coupons upon which material strength testing has been performed. Eventually,
under a strength based test, a re-evaluation of the load cases can be made without conducting a new test.
C.2.4 Aeroelastic model verification
The role of full-scale testing in the verification of the aeroelastic models used for blade design is detailed
in this paragraph. In the design of current WT’s according to existing standards, the use of aeroelastic
modelling in order to effectively simulate the prescribed design load cases for both ultimate and fatigue
loading is more and more common. Aeroelastic modelling has become the most important tool for
developing new designs as well as to predict wind turbine loading with increased reliability, therefore
leading to design optimisation. Full scale blade testing can provide the designer with the appropriate
blade properties for the correct modelling of the blades and the verification of the design assumptions
used. In the draft document of WG8 of TC88[2] these tests are described as Other Tests Determining
Blade Properties. The blade properties that are important for the verification of the aeroelastic models
and should be measured during full scale testing are:
natural frequencies, at least first two in flap and lead-lag directions;

mass distribution;
stiffness distribution;
damping, as a function of temperature;
mode shapes;
profile deformations due to load or manufacturing.
Details on the performance of these tests are included in [2] .
C.2.5 Structural model verification
Detailed FEM techniques or approximate analytical tools are used for the determination of the resistance
of the blade structure, which characterises its behaviour. The following properties derived from full scale
blade testing are necessary for the verification of the aforementioned models. Details on these tests are
also included in chapter Other Tests Determining Blade Properties of [2] :

strain/stress distribution;
stress concentrations at certain areas
mass distribution
stiffness distribution
creep
tip deflection due to tower clearance
profile deformations due to load or manufacturing.
C.2.6 Sensitivity analysis
The required accuracy for the measurement of the blade properties required for the verification of
aeroelastic modelling depends on the specific wind turbine design parameters, the characteristics of the
aeroelastic code used by the designer, as well as the safety margin of a certain design. It is recommended
that a sensitivity analysis is performed by the designer, prior to setting up the accuracy requirements of
the full scale test measurements. The purpose of the sensitivity analysis is to investigate the influence
of the identified in paragraph C.2.4 parameters, on the load calculations. The sensitivity analysis should
result in creating a "loading influence matrix", depicting the influence on the loading of a specific wt
component, by a deviation of the identified parameters by a certain percentage of their nominal values.
The influence on the fatigue loading of the wt component can be expressed in terms of the increase of
the "equivalent fatigue load" Leq , while for the ultimate loading in terms of the increase of the calculated
loads. Once the rainflow spectrum is calculated [17] based on the output of the aeroelastic code, it can
be converted to an equivalent load range Leq and an equivalent number of load cycles, by applying the
Pålmgren-Miner damage rule on various Wöhler curves with different slopes (1/m). The equivalent load
range is calculated as:
P n Lm m1
Leq = i i
Neq
(C-1)
where ni is the number of cycles in the ith load range, Li is the maximum value of each load level bin,
Neq is the equivalent number of constant amplitude cycles and m the slope of the material S-N curve,
which can be assumed to be of the type:
=
N k S ;m (C-2)
Table C-1 shows a typical "Loading influence matrix" for blade flapwise bending moment calculation
results. In this table the "Base case" columns represent the calculations with all parameters at their
nominal values, while the "Increase" columns depict the percentage increase of the loading for the
specified percentage increase of the parameter. Similar "influence matrices" could be established by the
designer using different techniques than the presented methodology of equivalent loads based on the
design assumptions and the methodology followed.
C.2.7 Required accuracy
It is not suggested to specify particular accuracy levels for the measured blade properties. Based on the
results of the sensitivity analysis, and taking into account the relevant design safety margin, the designer

Table C-1 Influence matrix for blade loading

Flapwise blade loading (m= X for fatigue)
Parameter variation Fatigue Increase in Ult.LC 1 Increase in .........
base case Leq Base case Loading
Leq (%) (kN) (%)
(kN)
1st flapwise natural frequency a a1 b b1 ..........
etc
can determine the required accuracy levels for the measurement of the blade properties of interest during
full scale testing. In the case of measurement of primary quantities used to determine blade properties,
e.g. acceleration measurement used for the calculation of the blade natural frequencies, the required
accuracy of the relevant measurement is dependent on the required accuracy of the product quantity of
interest, e.g. the blade natural frequencies.

D GUIDELINES TO INTERPRET RESULTS OF MECHANICAL TEST
D.1 Introduction
The objective of this section of the report is:
to disseminate the experience of CIEMAT towards the interpretation of full scale rotor blade test
results and the use of engineering models for rotor blade design.
a description of by CIEMAT accepted deviations of principal parameters between theoretical design
and actual determined values in the tests. The deviation is expressed in a non dimensional form,
namely:
Mm ; Mt
Error = Mt
100
where Mm = measuring magnitude and Mt = theoretical magnitude.
comments, by CIEMAT, on the IEC-1400-1 partial factors to check whether the test result is
acceptable to reach the required safety level.
When the CIEMAT criteria are met, indicated by the test blade error values which should/must
be less than the values reflected in this document, the design/blade is adequate for certification.
The error margin, accepted for certification, is adequately adjusted so that the product has a
guaranteed quality/reliability, but at the same time the margin is not that low that this makes
the product excessively expensive.
D.2 Need of tests

It is well known (see e.g. the S&FAT project, [1, 18]) that wind turbine rotor blades test are performed
to compare the measured values with the theoretically determined values. Another objective is to
determine the safety level of an individual blade including the effects of (series) fabrication. The answer
on the question how representative is the result of one test for the series produced blades depends on
the manufacturing process characteristics. This implies that the test is not representative when the
manufacturing process is changed throughout the series production.
The reasoning and rules given in this section are based on the fact that the results obtained
from a full scale rotor blade test are only valid for a series produced group of blades of which
the manufacturing procedure, neither the quality of the materials nor the structural design
has changed.
Thus, the validity of the test is based on the validity of the test self, but also on the quality of the
manufacturing process of the series production.
VERIFICATION OF BLADE IS ESSENTIAL
D.3 Causes of the differences between the blade theoretical calculations and the tests
results
One has to accept that there will always be a difference between measured properties of actual structures
and theoretically determined values of those properties.

In this section is shown how to interpret the engineering model verification tests results.
By a test, we can verify several structural parameters of a given structure. However the determination of
such parameters is not exact, obviously in every test and measurements some errors are included.
There are many causes of these errors which appear in four fundamental phases of the design (verification)
process:
A) Modelling, in which a engineering model is elaborated.
B) Manufacturing, in which the blade is fabricated.
C) Test, in which we carry out a test for the design verification.
D) Interpretation of test results.
Itemising the causes that contribute to the error(s) we come to the following:
A) MODELLING PHASE
A-1) Codes to calculate the aerodynamic loads and the wind turbines performances.
A-2) Structural codes, finite elements programs for composite structures
A-3) Finite elements type selected to model the rotor blade.
A-4) To model the blade by finite elements means check codes capacity criteria and CPU time.
Total number of elements and distribution and fineness of the mesh for these elements along
the blade. Boundary conditions.
A-5) Number load application points (or localised straps) and its distribution along the wingspan,
that represent the load envelop, as test loads in order to determine some structural parameters,
according to specification test. Localised pulling point in every load section.
A-6) Knowledge of the used materials and of the aerodynamic parameters and coefficients.
B) FABRICATION PHASE
B-1) Intake and quality control of materials.

B-2) Factory conditions, Cleanliness, hygiene, air-conditioning in order to fulfil the fabrication
processes requirements.
B-3) Laminate quality, criteria of weight content both in fibre and resin. Possible air inclusion,
resin excess.
B-4) Mold tolerance with regard to the calculations and the blade plans.
B-5) Laminated adaptation to the mold. Spring bag phenomenon, wrinkled, three-dimensional
effects by internal stress in the laminate. Internal dimensions of pieces.
B-6) Adhesive, in case that exist blade stuck parts.
B-7) Process repetitiousness. Possibility that in a series production rotor blades are not identical
e.g. due to deformations during post-curing.
C) TEST PHASE
C-1) Test bed stiffness.

C-2) Precision of the loads systems and in associated equipment. Adequate loads aligning.
C-3) Accuracy of the strain measure systems and in associated equipment.
C-4) Accuracy of the accelerometers and in the vibrations measure equipment.
C-5) Loading time. Load profiles versus time.
C-6) Temperature measures.
C-7) Selected zones of measure. Stress concentration. Saint Venant principle.
C-8) Selection and precision of the strain gauges employees and in associated equipment. Strain
gauge assembly selection type. Thermal and hygrometric compensation.
C-9) Precision registering of all data types.
D) INTERPRETATION PHASE
D-1) Manufacturer to test analysis correct communication about the blade mass, geometric, and
elastic characteristics.
D-2) Mathematical data treatment precision. Numerical analysis.
D-3) Test analyst experience. Though this characteristic is valid for the others factors also.
As to be expected there are a lot of possible sources of errors. In spite of these possible errors the total
effect will remain restricted when the test is performed correctly and the test can be of great use for the
manufacturer.
D.4 Classification of static tests

A static test must be complete, i.e. must cover all areas of the structural knowledge. This is based on
the experience acquired in projects such as AWEC-60 (WEGA I), S&FAT (JOULE), etc. We consider it
convenient to divide (or to classify) the static general test activities into the following subtasks (see Table
D-2)
Table D-1
STATIC TEST Natural vibration test
Elastic test Equivalent stiffness
Influence coefficients
Static load Limit load
strength test Ultimate load

Not shown in table D-2, but before executing the static test, we must perform some geometric and mass
measurements on the test specimen. After these measurements are carried out, we will proceed to perform
the static test, i.e. all the activities as shown in table D-2.
The first test is the vibrations dynamic test, since the blade has not been loaded and is not deteriorated by
any destructive test yet.
The next test is the elasticity test to obtain the elastic and structural blade properties.
The last static tests is the limit load and ultimate load test. During these test the resistance of the blade
against the load is measured. During the ultimate load test the blade (local) strains might be above the
linear elastic allowable strain, even break the blade.
D.5 Blade identification measurements

As already mentioned, before testing a blade so called specimen identification measurements have to be
carried out. The objective of these measurements is to check the conformity of the blade with the design
specifications.
In this paragraph the minimum measurements are indicated that must performed before the blade is tested.
In table D-3 the measurements are listed including the maximum admissible deviations with the design
specifications, according to the criteria of CIEMAT.
The aim of this work is that the given results are valid for any type of blade.
Table D-2 Determined Previous Measures

MAXIMUM
MEASURE ADMISSIBLE AFFECTED CAUSES
DEVIATION
Geometric and mass properties
– Mold tolerance
Blade wingspan err. < 0.1 % – Blade compound integration
-Mold tolerance
Chords distribution along the blade err. < 0.2 % -End work, finish edge and blade
span polish
abs. err. < 2o
-Mold tolerance Post-cure twist
Total twist (absolute) value between
phenomena
the first aerodynamic section and tip
bolt hole diameter tolerance of the abs. err. < 0.5 mm
-Commercial drill bits
-Cutting speed and erosion
blade root connection
abs. err. < 1 mm
-Placement drilling machine
Bolt hole centre location tolerance of
-Machine tolerance
blade root connection
error < 5%
-Blade fabrication processes
Blade weight
-Reproduction process not exact
error < 5%
-Blade fabrication processes
Blade centre of gravity position
-Non exact reproduction process
These are the minimum requirements, on errors of geometric and mass characteristics, to be
fulfilled to proceed the test.

Obviously other characteristics like e.g. the blade longitudinal moment of inertia in relation to the rotor
shaft etc. could also be specified, but these are of less importance. Due to the fabrication process
-lamination of perfectly established layers - the blade weight as well as its centre of gravity position are
(usually) very similar to the theoretical values. In the practice the similarity between test and theoretically
determined values of the blades longitudinal inertia moment is guaranteed.
D.6 Natural Vibration Test (eigenfrequencies)

Is well known that the objective for the vibration test is to determine the blade’s natural frequencies and
overall damping.
D.6.1 Natural Blade Frequencies
The objective of the natural frequency test is to determine possible blade resonance problems, e.g. using
a Campbell diagram. On the other hand, the stiffness distributions will be determined by subsequent test
(elasticity test). In a blade test there is necessary to determine the first two flap and lag frequencies.
Table D-3 Admissible differences between the actual and theoretically determined values of blade
frequencies (CIEMAT criteria)
FREQUENCY PERCENT OF ERROR
1st of flap < 12 %
2nd of flap < 13 %
1st of lag < 18 %
2nd of lag < 20 %
These tolerances are justified by the stiffness and mass distribution variations, and also by considerations
based on the Campbell diagram.
Due to fact that the natural torsion frequencies are (much) higher than the natural bending frequencies, it
is not required to obtain this value.
D.6.2 Damping
To determine both damping components, aerodynamic (viscous) and structural, is not an easy task. Also,
the results vary quite a lot due to e.g. the test methodology and/or due to the temperature of the blade.
According to the CIEMAT criteria the damping error must be less than 50%
Error < 50 %
(See S&FAT document, VOL II, [18])
D.7 Elastic test

The elastic test is composed of two subtasks:
7-1) Stiffness equivalent test.
7-2) Influence coefficients test.
The objective of these tests is to know the principal elastic and structural blade characteristics. These
tests can be execute one after the other or simultaneously.

D.7.1 Equivalent stiffness test
The objective of this test is to obtain the equivalent stiffness, i.e. EI, in several blade sections. The
equivalent stiffness can be determined by assuming one or more prismatic cross section(s) ending up with
the same behaviour as the real stiffness distribution.
It is difficult to establish criteria in this field because it is related with the blade sections which can vary
a lot and due variations of its construction system.
And so, the admissible error in this case is taken conservatively and can therefor be applied to any type
of existing blade in the market, because the given value is based in elastic considerations.
When the error, between the design and the test values, exceed this value the difference is considered to
be unacceptable.
The maximum error value can be put in the form
EI R ; EI T
EI T
100% < 10%
Where EIR = actual test stiffness, and EIT = theoretical stiffness value.
Namely the maximum error value must be less to the 10 %.
Due to the fact that the information recorded by strain gauges is based on a small area, the theoretically
calculated stiffnesses, EIT0 s, compared to a measured bending moment based on a strain gauge couple
may deviate.
Consequently we propose to place several strain gauge couples in each of the interesting sections.
The minimum number of strain gauge couples to determine the equivalent stiffness in a section must be
two. And the minimum number of blade sections to check the equivalent stiffness must be four.
The strain gauge couples must be placed between 25 % and 50 % of the profiles chord, counting from the
leading edge.
The span wise location to check the stiffness are shown in table D-5.
Table D-4 Span wise locations to measure the equivalent stiffness (in blade wing span percentage)
SECTION 1% 3% x 5% 35% 70%
The location x = 1% takes into account the characteristic phenomena of stress concentration in the
blade-hub zone.
The sections with (3% x 5%) takes into account the blades handle (neck) behaviour.
The locations 35% and 70% of the span are distributed along the blade wingspan.
As we can see there are a number of error sources for this test, the errors are associated with:
the strain gauge itself and its local placement on the blade;
the determination of the materials characteristics;
the engineering model.

D.7.2 Influence coefficient test
When it is valid to use the linear elastic theory then the reciprocal-work theorem of Betti/Rayleigh is
valid. The reciprocal-work theorem is stated as:
The work done by the forces in the first state of loading when they move through their
corresponding displacements in the second state of loading is equal to the work done by
the forces in the second state of loading when they move through their corresponding
displacements in the first state of loading.
See e.g. a text book by Timoshenko & Gere, [19].

In one dimensional formula form this comes down to
Pa uB = PB uA
.
Or also by definition of a coefficient indicating the displacement in direction k at a location j due to a
certain load applied at point i in an other direction l. We will symbolise the coefficient by Cjlik .
Generally, the displacement ik is due to an arbitrary force F jl , applied in point j in the direction l, in the
point i and in the direction k . Mathematically we have:
ik = Cjlik F jl (D-1)
For a determined test we could simplify the prior expression for a single load, F , that acts in a given point
in a certain prefixed direction, for example in flap direction. The expression D-8 is now:
ik = C ik F (D-2)
The criteria of CIEMAT, to get a good indication of the blade production quality, requires only one test.
This test is to measure the displacements ik resulting of a single load, F, in one point and one direction.
This test is assumed to be sufficient and is much more simple than the more general test.
The flap direction is elected, due to the fact that this direction has the lowest stiffness and
thus the displacements are larger and, in consequence, the influence coefficient could be
measured with a high precision.
For certification the maximum error in the influence coefficients must be less than the 8%.
error <8%
A maximum error of less than 8% is correct because it implicates a good similarity between the theoretical
engineering model and the test results.
Other error sources must also be considered, such as:
the lack of precision or resolution in the measurement systems. These systems could be as simple
as, e.g. threads, rulers, etc..
the variability of basic material properties, resulting in different laminate properties, or
the errors in the engineering model.

However, such errors are less than associated with the equivalent stiffness measurement, because in the
last case strain gauges are used which detect locally.
Furthermore the closer we get to the blade root, where the deflections are very small, the larger this last
type of error becomes. The measurement is not valid when the measurement systems error is of the same
magnitude as the error in the deflection to be measured.
Consequently, when the measure systems are simple like e.g. threads, rulers, etc., they should not be used
to measure the influence coefficients in those blade zones, whose X coordinate is less than 35 % of the
wingspan. In such cases, to guarantee correct measure of the influence coefficients, we recommend the
use of laser measurement devices or other with equivalent accuracy, whose inherent errors are less than
the mentioned 8 %.
D.8 Static loads strength test

The objective of this test is to put the blade to the limit and ultimate loads to verify its strength and to
obtain, furthermore, the safety factor and the design reserve factor.
D.8.1 Limit load strength test
This test is to put the blade to the design limit load, LLDES , which definition is the following:
LLDES = P SFLIM LLT (D-3)
Where P SFLIM is the loads partial safety factor, and LLT is the theoretical limit load.
Testing the blade at this load, our experience indicated that measuring 3 sections is enough. Certainly 3
sections is not a great number, but it reduces the costs of the test. To test more than 3 sections would
make the test unnecessarily expensive or excessively complicated.
These three sections corresponding to 0%, 35%, and 70% of the wing span.
Analogous to this reasoning loading 3 sections is enough to reproduce the bending moments and the
shearing forces envelope adequately.
The loads must be introduce in the effective shear centre for each blade section. The maximum admissible
error in shear centre determination must be less than 7% of the local chord.
error < 7 %
For severity of static load test we think that the relative static strength factor:
RSSF = SSF design

SF S test
should never exceeds 1.2, so this factor will have to be limited to:
RSSF < : 12
Stress tolerances
In the opinion of CIEMAT, the admissible error values, test results compared to the pertinent (engineering)
models, for a number of phenomena are indicated below:
Error tensile-compressive stress < 5%

Error bending stress < 5%
Error shear stress < 8%
Error buckling stress < 10%
Error post-buckling stress < 16%
Error micro cracking stress < 8%
Safety factors
To palliate certain uncertainties of both, project and test, we utilise the so-called partial safety factors.
The partial safety factor, which is a parameter that corrects theoretical and laboratory values, takes into
account the uncertainty or variations, both in the loads and in the materials, in order to make the design
reliable and to avoid failure.
A review of the principals safety factors, and the values applied by CIEMAT in comparison with IEC-
1401-1, table 3 is given below.
Partial safety factor for loads, F
This factor is defined by following equation
Fd = F Ft
where Fd is the design aerodynamic load, and Ft is the theoretical aerodynamic load.
Load type F factor

Aerodynamic 1.30
Operational 1.80
Gravity 1.10
Other inertias 1.30
The main discrepancy between the CIEMAT value and the IEC-1400-1 is in the operational load value.
The value of 1.45 given by IEC, is assumed to be very low.
(
For loads related with transport, considering the very complex terrains as in Spain, a F transport is )
introduced with and estimated value of
(
F transport ) = 1:7
for all load types.
Partial material factor, m
This factor is mathematically defined as follows
fd = fk
m

where fd is the design values for material, and fk is the characteristic values of material property.
Material type m
Steel 1.05
Aluminium 1.12
Hand lay-up 1.30
"Pre-pegs" 1.08
Un-tempered laminate 1.15
GFRP Tempered without thermal slope 1.05
laminate with thermal slope 1.02
Creep strength 1.6
GFRP By vacuum 1.05
(compaction By pressure with male and nut 1.05 -1.2
process) Pressure cooker 1.02
Glue 1.20
Partial factor for failure consequences, n
( ) R(fd)
n S Fd
where S is a stress function and R is a material resistance. For static test and all failure consequence types
n = 1:00
in agreement with the IEC-1400-1 given value.
Test load scatter factor, su
su = 1:05
When the manufacturing process is correct the value of 1.1, given by IEC-TC88-WG8 (p. 32), could be
excessively high.
D.8.2 Ultimate load test
This test is to put the blade to ultimate load, U L, which definition is the following:
UL = SFDES LLDES (D-4)
Where SFDES is the design safety factor, and LLDES is the factor defined in equation D-10.
In our opinion should be:
SFDES : =15
The aim of this test is to obtain
TEST
SFDES = U LTEST =LLDES (D-5)

The load applied to the structure through this test is increase step by step, until eventually the ultimate
load, U LTEST , that according to our criterion, is the load under the which the structure endured for at
least 10 seconds.
From equation D-12 yields SFDESTEST . And this test concludes the blade adequacy if:
TEST SF
SFDES DES
Evidently that implies

U LTEST UL
We are believing that the SFDES minimum value must be great than 1.5 in order to certify a blade
SFDES 1:5
According to our criterion the elected sections for both tests, ultimate load and limit load strength one,
are equals. Namely, corresponding to the 0%, 35% and 70% of the wingspan.


E CRUSHING 1
As an example of a non linear phenomena which makes test results difficult to interpret when load levels
are higher than used in the design analysis.
A slender beam like structures will bend due to a moment M with curvature radius R according to the
following formulation:
EI
M
R
= (E-1)
The beam will be deformed in a circular shape by the bending moment. The bottom and top skin will
now have the tendency to move towards each other. This phenomena is called“crushing”.
Crushing loads will are expressed in a force C per unit of length and are in plane of the shear web. These
loads are in fact the loads which bend the upper and lower skins and increase the bending stresses in the
skin.
This crushing load is quadratic with the bending moment
M 2
C = EIh (E-2)
Thus the skin is not only loaded with the so called membrane stresses which are constant over the wall
thickness for thin walled structures but also by local bending stresses which are quadratic with the bending
moment due to the curvature.
M = E I y’’
M h M N = M/h
2
C = N/R = M/h . M/(EI) = M/(EIh)
P
Ν Ν
α
N R N
C C P = N. α
C
Rα
N C C
N C = P/ ( R α) = N/R
Figure E-1 The loading on a bend slender wing beam

1
Copied from lecture notes of Prof. Ir. J. Ten Asbroek

The load on the shear web is even more complex, it carries the largest share of the shear force, a small
part of the bending moment and the crushing loads. The first to are linear with the increase in (test) load
but the crushing load is quadratic with the increase in (test) load. The buckling load of the shear web is
again determined with the stress ratio or interaction curves first presented by Shanley see e.g. [20].
For combined loadings the general conditions for failure are expressed by Shanley as follows:
R1x + R2y + R3z + : : : 1:0 (E-3)
In which R is an expression of the stress ratio Ff where f is the applied stress and F the allowable stress.
In eq. E-15 the Ri could stand for bending, shear and crushing. The exponents x, y, and z give the
relationship for combined stresses. These exponents can be determined for some simpler combined load
systems. However, in many cases of combined loading and for particular types of structures the exponents
must be determined by making actual failure tests of combined load systems.

PART 4
POWER PERFORMANCE IN COMPLEX TERRAIN
E. Morfiadakis CRES
I. Antoniou RISØ National Laboratory
A. Cuerva CIEMAT
R. Hunter NEL
F. Mouzakis CRES
N. van der Borg Netherlands Energy Research Foundation ECN
D. Westermann DEWI
Power Performance in Complex Terrain

CONTENTS OF PART 4
1. INTRODUCTION......................................................................................................................................................5
1.1 Background..........................................................................................................................................................5
1.2 Objective ..............................................................................................................................................................5
1.3 Working Method..................................................................................................................................................6
1.3.1 Project technical description.........................................................................................................................6
1.3.2 Project scientific and technical performance ................................................................................................6
2. SCOPE AND FIELD OF APPLICATION ................................................................................................................9
3. DEFINITIONS, SYMBOLS AND ABBREVIATIONS .........................................................................................11
4. POWER PERFORMANCE MEASUREMENTS ....................................................................................................13
4.1 Introduction........................................................................................................................................................13
4.2 Effect Of Terrain Orography On Wind Field.....................................................................................................13
4.3 Effect Of Obstacles On Wind Field ...................................................................................................................14
4.3.1 Wind turbines .............................................................................................................................................14
4.3.2 Other obstacles ...........................................................................................................................................14
4.4 Site Calibration ..................................................................................................................................................15
4.4.1 Wind tunnel simulation ..............................................................................................................................15
4.4.2 Numerical modelling ..................................................................................................................................15
4.4.3 Field measurements ....................................................................................................................................17
4.5 Alternative Methodology For Power Curve Measurements ..............................................................................18
4.6 Effects Of Turbulence And Wind Shear On Power Curve ................................................................................19
4.7 Conclusions........................................................................................................................................................19
5. EXPERIMENTAL RESEARCH ON WIND TURBINE POWER PERFORMANCE ...........................................21
5.1 Introduction........................................................................................................................................................21
5.2 Review Of Power Performance Campaigns.......................................................................................................21
5.2.1 WINCON 100XT operating at CRES test station ......................................................................................21
5.2.2 VESTAS V27 operating at Andros site ......................................................................................................23
5.2.3 VESTAS V27 measured by DEWI ............................................................................................................24
5.2.4 VESTAS V27 measured by WINDTEST...................................................................................................25
5.2.5 NORDTANK 500/37 operating at Toplou site...........................................................................................25
5.2.6 NORDTANK 500/37 operating at Taff-Ely (turbine A) ............................................................................27
5.2.7 NORDTANK 500/37 operating at Taff-Ely (turbine B) ............................................................................28
5.2.8 NORDTANK 500/37 operating at Pilsum site ...........................................................................................29
5.2.9 NORDTANK 500/37 measured by WINDTEST .......................................................................................30
5.2.10 MADE AE-30 operating at Fuerteventura and Tarifa ..............................................................................30
5.3 Assessment Of Site Calibration Methodologies.................................................................................................32
5.3.1 Site calibration for WINCON 110XT ........................................................................................................32
5.3.2 Site calibration for VESTAS V27 ..............................................................................................................32
5.3.3 Site calibration for NORDTANK 500/37...................................................................................................33
5.3.4 Conclusions ................................................................................................................................................33
5.4 Conclusions........................................................................................................................................................33
5.5 Tables.................................................................................................................................................................35
5.6 Figures ...............................................................................................................................................................41
6. ALTERNATIVE POWER PERFORMANCE METHODOLOGIES
6.1 Introduction........................................................................................................................................................59
6.2 Nacelle Cup Methodology .................................................................................................................................59
6.2.1 Introduction ................................................................................................................................................59
6.2.2 Nacelle cup anemometry on 1MW wind turbine........................................................................................59
6.2.3 Nacelle cup anemometry on 500kW wind turbine .....................................................................................73
6.2.4 Nacelle cup anemometry on 110kW wind turbine .....................................................................................79
6.3 Upwind Rotor Measurements ............................................................................................................................85
6.3.1 Introduction ................................................................................................................................................85
6.3.2 Rotor disk measurements on a 110kW wind turbine..................................................................................85
6.3.3 Conclusions ................................................................................................................................................86

6.4 Conclusions........................................................................................................................................................87
7. PARAMETER IDENTIFICATION OF WIND TURBINE POWER PERFORMANCE .......................................88
7.1 Introduction........................................................................................................................................................88
7.2 Parameter Identification Procedure On Experimental Data Bases.....................................................................88
7.2.1 Multivariate regression analysis .................................................................................................................88
7.2.2 Assessment of the accuracy of the multivariate regression ........................................................................90
7.2.3 Parameter identification procedure.............................................................................................................90
7.3 Identification And Quantification Of Site Related Effects On Power Performance ..........................................91
7.3.1 Parameter identification on a 225kW pitch regulated WT .........................................................................92
7.3.2 Parameter identification on a 110kW stall regulated WT...........................................................................92
7.3.5 Parameter identification on a 3MW pitch controlled WT ..........................................................................94
7.4 Conclusions........................................................................................................................................................95
8. ASSESSMENT OF POWER PERFORMANCE PRACTICES ............................................................................109
8.1 Introduction......................................................................................................................................................109
8.2 Review On The Characterization Of Complex Terrain ...................................................................................109
8.3 Critical Points On Existing Recommendations................................................................................................110
8.3.1 Test site requirements ...............................................................................................................................110
8.3.2 Characterization of wind speed sensors....................................................................................................110
8.3.3 Influence of wind mean and turbulent structure .......................................................................................112
8.3.4 Site calibration..........................................................................................................................................112
8.3.5 Response system.......................................................................................................................................112
8.3.6 Measurement analysis...............................................................................................................................113
8.3.7 Contractual aspects ...................................................................................................................................113
8.4 Conclusions......................................................................................................................................................113
9. PRACTICAL RECOMMENDATIONS ................................................................................................................115
9.1 Definition Of Site Complexity .........................................................................................................................115
9.2 Selection And Characterisation Of Instrument Configuration.........................................................................115
9.2.1 Definition of reference wind speed ..........................................................................................................115
9.2.2 Mean wind speed and turbulence structure measurements.......................................................................115
9.2.3 Wind turbine control system response......................................................................................................115
9.3 Site Calibration ................................................................................................................................................116
9.4 Normalization Procedure .................................................................................................................................116
9.5 Uncertainty Estimation Procedure ...................................................................................................................116
9.6 Power Curve Measurements Of Wind Farms ..................................................................................................116
10. REFERENCES.....................................................................................................................................................123
11. ACKNOWLEDGEMENTS .................................................................................................................................129

1. INTRODUCTION
1.1 Background
A large and promising portion of the European wind energy market is located in the southern European countries,
namely Greece, Spain, South Italy and Portugal. The most suitable sites for wind energy exploitation in these
countries, as well as in other northern European countries such as Great Britain, are in the majority of cases located
in very complex mountainous terrain. Nevertheless, the influence of such complex terrain on the operation of wind
turbines is a major research topic while the research results will significantly influence the market.
EWTS.II-IV contributed significantly to the identification and quantification of the complex terrain effects on the
power performance of wind turbines. The program, included extensive review of current power performance
measurement practices, wind turbine measurements from such mountainous regions (Greece, Great Britain and
Spain) along with evaluation of previous or ongoing similar measurements in flat and complex terrain for stall and
pitch regulated wind turbines operating in wind parks of different sizes and ratings.
The present final report of EWTS.II-IV “Power Performance In Complex Terrain”, comprises a review of the project
objectives and analysis methodologies related to wind turbine measurements, presentation of literature surveys,
detailed presentation of all information regarding the wind turbine experiments and measurement analysis carried
out or included within the framework of the project as well as a complete assessment of the results. Finally, the
assessment of the existing power performance verification guidelines is presented along with the resulting
recommendations for the feasible inclusion of the research results in a power performance practice that will address
the wind turbine operation in complex terrain sites.
The assessment of the results is based on the comparison of the performance and the wind characteristics effect of
the wind turbines operating at flat and complex terrain. Moreover, the effect of complex terrain on wind turbine
operation is quantified by utilising parameter identification analysis.
In the present chapter a review of the project, regarding objectives, technical description and achievements, is
presented. In chapter 2 a brief review of the scope and field of application of the status report is given whereas in
chapter 3 the used nomenclature is defined. In chapter 4 the literature survey results on the issue of power
performance measurements is given. In chapters 5 and 6 the experimental research regarding power performance
measurements and alternative methodologies respectively, that were carried out within the framework of the project
is presented. In chapter 7 the description of the statistical tool, utilised for the parameter identification purposes, is
presented along with the research results. In chapter 8 the assessment of the existing power performance
recommendations and standardisation documents is presented. Finally, the technical basis for the development of
new recommendations on the issue of power performance in complex terrain resulted from the project results and the
assessment of the existing methodologies is summarised in chapter 9.
1.2 Objective
The main objective of the sub-project is the preparation of a status report which will be the technical basis for further
standardisation work on power performance determination and verification in complex terrain.
The major achievements of the project are:
• presentation and assessment of various power performance measurements regarding wind turbines of different
size and control strategy
• identification and quantification of the terrain induced effects on the power performance of wind turbines
• development of a new procedure of compiling the power performance measurements accounting for terrain
effects

• assessment of the existing recommendation and standardisation documents and identification of all
inefficiencies that influence the power performance assessment and verification in complex terrain
• construction of the technical basis for the development of a reliable procedure for power performance
measurements in complex terrain
1.3 Working Method

1.3.1 Project technical description
The sub-project comprised six tasks with the following objectives:
Task description/objective
Task 1. Inventory of existing information on the issue of power performance measurements in
complex terrain.
Task 2. Inventory of power performance measurement campaigns and selection of well documented
campaigns suitable for further analysis. Emphasis will be given to cases which claim to use
methods yielding better accuracy by applying site calibration.
Task 3.1. Comparison of different power curve measurements at different sites through annual energy
production estimation. Uncertainty estimation for power curve and annual energy
production.
Task 3.2. Power curve evaluation using nacelle anemometers.
Task 3.3. Parametric analysis of the available power curve measurements for the identification of the
site related effects influencing the power performance behaviour.
Task 4. Assessment of existing guidelines and their validity for complex sites. Identification of
drawbacks connected to the application of existing guidelines and recommendations for
power performance measurements in complex sites.
Task 5. Drafting of a status report providing a comprehensive overview of the up to date available
information on power performance in complex terrain and of the proposed solutions and
results of the parameter identification.
Task 6. Co-ordination of the sub-project work.
1.3.2 Project scientific and technical performance
The project scientific and technical progress of the sub-project is presented for each task separately.
Task 1
A literature research has been performed, by all participants, in order to gather the available information on power
performance measurements in complex terrain. The literature research resulted in many topic-related articles and
documents, yet it provided with a limited readily exploitable information for complex terrain measurements. The
survey and the resulting conclusions regarded topography effects, effect of obstacles, site calibration practices,
alternative power curve measurement procedures and site dependent wind characteristics effect on power curve.
Task 2
The power performance campaigns that have been selected and analysed are listed in table 1.1. The available
campaigns cover a wide range of wind turbine size, control techniques, terrain complexity, site calibration methods
and nacelle cup practices.
Task 3.1
Summary reports on power performance measurements have been produced for the selected campaigns. The
summary reports contained information on the reference measurement procedure, annual energy production and
estimated uncertainty, power as well as power coefficient curves. The curves were accompanied by the curve
tabulated data list and their basic, within each bin, statistics. CRES contribution included the analysis of the power
performance measurements for the V27, WINCON 110XT and NTK500/37 wind turbines. CIEMAT contribution

regarded the measurement analysis for the MADE AE/30 operating at complex terrain (Tarifa site). Attention was
paid to the double stall phenomena that were encountered during the operation of the machine. DEWI contribution
regarded the analysis of two power performance measurements, one for a NT500/37, operating at Pilsum, and
another for V27. NEL contribution regarded the power performance analysis of two NT500/37 wind turbines
operating at Taff-Ely wind farm, a complex terrain site located at south Wales. The machines chosen for the analysis
were those presenting the least upwind terrain slopes in the direction of the prevailing winds. RIS contribution
regarded the power performance measurements for ELKRAFT 1MW wind turbine. Moreover, comparisons of the
annual energy production estimation for NT500 and V27, from power curve measurements by WINDTEST, were
included. Except from the summary reports for the analysed wind turbines, documentation containing information
regarding the wind field structure (i.e. mean wind speed and direction, air density, turbulence and wind shear), the
turbine operational characteristics (i.e. power mean and standard deviation, yaw error etc.) as well as the analysis of
the uncertainty components estimation, in accordance to IEC (1997) guidelines, was also produced.
In the fields of site calibration different techniques were assessed. The nacelle cup on standstill wind turbine practice
was applied in all power performance campaigns by CRES (V27, W110XT and NTK500/37) using a cup
anemometer that was installed on a mobile mast on the wind turbine nacelle, while the turbine was at standstill in
the downwind direction. The measurements from the nacelle cup were correlated to the measurements from the
reference mast either directly in cases where the measuring points were at the same elevation or by correcting for
wind shear. The rotor disk averaged wind speed practice was applied by CRES for the W110XT campaign by the
aid of 5 cup measurements spread all over the upwind rotor disk plane. Conventional site calibration with
meteorological mast prior wind turbine erection was applied in the case of NTK500/37 and comparisons of this
approach with the one using the nacelle cup with the parked machine were made. Application of numerical
simulation codes (WA SP) was performed by NEL for the NTK500/37 campaigns.
Table 1.1. Analysed power performance campaigns within EWTS-II, part 4
Wind turbine Terrain Institute Site calibration AEP Uncertainty Parameter

technique estimation estimation identification
WINCON 110XT complex CRES nacelle cup on running WT Yes Yes Yes
(110kW, stall) nacelle cup on parked WT
rotor disk measurements
VESTAS V27 complex CRES nacelle cup on parked WT Yes Yes Yes
(225kW, pitch)
flat RISØ reference mast Yes Yes No
flat DEWI reference mast Yes Yes No
flat WINDTEST reference mast Yes Yes No
MADE AE-30 flat CIEMAT reference mast Yes Yes Yes
complex CIEMAT reference mast Yes Yes Yes
NORDTANK complex CRES calibration prior WT Yes Yes Yes
500/37 erection, nacelle cup on
(500kW, stall) parked WT
flat RISØ reference mast Yes Yes Yes
flat DEWI reference mast Yes Yes Yes
flat WINDTEST reference mast Yes Yes Yes
NEDWIND 500kW complex DEWI nacelle cup on running WT Yes Yes No
ELKRAFT flat RISØ nacelle cup on running WT Yes Yes No
1MW
Task 3.2
The assessment of the nacelle anemometer on running machines was performed by RISØ, DEWI, RISØ and CRES.
The performed campaigns regarded a wide range of wind turbine sizes ranging from 110kW up to 1MW machines.
RISØ contributed with the results from the application of the nacelle cup technique for the power performance
estimation of the ELKRAFT 1MW wind turbine operating at Avedore Holmen. The objectives of the research were:

• to examine to what extend the nacelle anemometer can be used for the derivation of the power of a turbine after
its relation to the mast anemometer has been established via measurements in another turbine situated possibly
in another kind of terrain
• ii) to investigate the influence of wake situations and different rotor settings on the nacelle cup calibration.
DEWI contributed with the results from the application of the method on a 500kW stall regulated wind turbine,
whereas CRES applied the method on a 110kW machine.
Task 3.3
Parametric analyses have been performed for the mean and standard deviation of the power output of the analysed
turbines. The analysis was based on a multivariate regression method equipped with a backward parameter
elimination technique. The effect of the primary deterministic and stochastic parameters was captured and
quantified. The analysed campaigns regarded the VESTAS V27, NORDTANK 500/37 (as measured by different
institutes at different sites), WINCON 110XT and MADE AE-30 wind turbines. The selected machines present a
variety of size and control strategy. Results from the investigation on the influence of different meteorological
conditions, namely turbulence, vertical wind shear and atmospheric stratification on the mean power output of a
large wind turbine, namely the AEOLUS II were also presented.
Task 4
The existing guidelines were assessed under the perspective of their applicability and validity for complex sites. The
identified drawbacks are discussed and proposals for guidelines and recommendations are made.

2. SCOPE AND FIELD OF APPLICATION

The scope of the present status report is to provide with a comprehensive overview of the up to date available
information on power performance in complex terrain and proposed solutions and recommendations for best
practice. In more detail, the present document regards the following:
• inventory of existing information on the issue of power performance measurements in complex terrain.
• inventory of power performance measurement campaigns with emphasis on the application of site calibration
techniques
• comparison of different power curve measurements at different sites through annual energy production
estimation and uncertainty estimation for power curve and annual energy production
• power curve evaluation using nacelle anemometers
• parametric analysis of the available power curve measurements for the identification of the site related effects
influencing the power performance behaviour
• assessment of existing guidelines and their validity for complex sites; identification of drawbacks connected to
the application of existing guidelines and recommendations for power performance measurements in complex
sites


3. DEFINITIONS, SYMBOLS AND ABBREVIATIONS

The symbols that are referenced herein are listed and explained in the following table:
ak regression equation coefficient of parameter k

D spatial distance, rotor diameter (m)
E( xi ) regression error at point xi
f frequency (Hz)
 Utop − Uhub Uhub − Ulow 
G asymm asymmetric shear factor, i.e. Gasymm = 0.5 + 
 Htop − Hhub Hhub − Hlow 
umax
Gmax maximum gust factor, i.e. Gmax =
U
u
Gmin minimum gust factor, i.e. Gmin = min
U
 Utop − Uhub Uhub − Ulow 
G symm symmetric shear factor, i.e. Gsymm = 0.5 − 
 Htop − Hhub Hhub − Hlow 
σu2 + σv 2 + σw2
k turbulence kinetic energy, i.e. k =
2
Lu turbulence length scale of the longitudinal wind speed component (m)
Lv turbulence length scale of the lateral wind speed component (m)
Lw turbulence length scale of the vertical wind speed component (m)
N number of samples within 10 minutes sampling period
σ Xk
Sk dependence coefficient of parameter Xk. i.e. S k = a k
y mean
t time (sec)
σu
T . I. turbulence intensity, i.e. T . I .=
U
U 10min period mean wind speed (m/s)
3
1 N
 u −U 
U skew skewness of the wind speed distribution, i.e. U skew =
N
∑=  iσ 
i 1 u
4
1 N
 u −U 
U kurt kurtosis of the wind speed distribution, i.e. U kurt =
N
∑=  iσ  − 3
i 1 u
ui instantaneous wind speed (m/s)

X k ( xi ) the value of the kth independent variable at point xi1
y.e. yaw error (°), i.e. nacelle direction - wind direction
y.s. mean yaw speed of the nacelle (°sec-1)
z elevation from ground (m)
M
y i regression estimation at point xi, i.e. y i = ∑a k X k ( xi )

k=1

Greek symbols
α
α , w.s. U ( z1 )  z1 
wind shear exponent based on the power law model, i.e. = 
U ( z2 )  z2 
θ , w.i. wind inclination, i.e. the angle between the wind speed vector and the horizontal
direction (°)
σp standard deviation of pitch angle (°)
σu standard deviation of longitudinal wind speed component (m/s)
σv standard deviation of lateral wind speed component (m/s)
σw standard deviation of vertical wind speed component (m/s)
σϕ standard deviation of wind speed direction (°)
σθ standard deviation of wind inclination (°)
N
∑ (X ( xi ) - X k )
2
k
σ Xk standard deviation of parameter Xk, i.e. σ Xk = i=1
N -1
2
N
 y i - y i 
χ2 chi squared estimator, i.e. χ2 = ∑ 
 σi 

i=1
ϕ wind direction (°)
Subscripts
H at hub height
L at lower position of the meteorological mast
T at top position of the meteorological mast

4. POWER PERFORMANCE MEASUREMENTS
4.1 Introduction
The methodology for power curve measurements on wind turbines is elaborated in various studies (Christensen C.,
1986, Petersen T., 1993b). The internationally agreed procedure for the measurement and data treatment has been
laid down in the draft IEC standard on power performance measurements of wind turbines (IEC, 1997). The
methodology is based on the assumption that the statistical wind descriptors (average horizontal speed, average
turbulence, average direction) measured at a certain distance from the wind turbine are equal to the descriptors of the
wind experienced by the turbine. In complex terrain this assumption will be violated. The draft IEC standard gives a
description of the terrain for which the above assumption is considered to be valid. For terrain that do not fit to this
description, possible procedures are suggested in the draft IEC standard in order to find the relation of the average
wind speed at the anemometer location and the average wind speed experienced by the turbine. This site calibration
procedure may be based on field measurements, on physical modelling (wind tunnel measurements) or on numerical
modelling. However, very little systematic experience is available on the item of site calibration. At the same time
power curve measurements have to be performed at sites in mountainous areas or at sites with obstacles to the wind.
The literature study focused on the following items.
• the influence of orography on local wind speeds over distances up to 100 m

• the influence of obstacles on local wind speeds over the same distances
• methods to quantify the differences in the measured wind speed and the wind speed experienced by the turbine
• alternative methodologies for power curve measurements
• the influence of the site wind characteristics on the power curve of the turbine.
4.2 Effect Of Terrain Orography On Wind Field

Variations in terrain height can cause differences in the average measured wind speed at the anemometer location
and the average wind speed experienced by the measured wind turbine. The draft IEC standard describes which
slopes and height variations are expected to cause too large differences for a normal power curve measurement.
Various publications are available that give an impression of the possible differences in the average wind speed in
mountainous areas between potential positions of a wind turbine and corresponding meteorological mast. Of course
the differences are very dependent on the degree of complexity of the site and of the measuring height above ground
level. Some examples are:
• At 42m a.g.l. at the AWEC-60 site (Cabo-Villano, Spain) at 250m distance differences in average wind speed
occur in the order of 4% (Klug H., 1996).
• At 41 m above sea level (about 30 m a.g.l) in the dune landscape of ECN on the North Sea shore differences in
average wind speed occur in the order of 3% over distances of 40 to 120m (Vermeulen P., 1982).
• At 40 m a.g.l. at a site on Andros (Greece) differences in average wind speed occur of 3 to 4 % over a distance
of 100m (Morfiadakis E., 1994).
• At the height of the Vestas27 hub and over a distance of 1.5 rotor diameters differences in average wind speed
occur of -1 to 5% on Samos (Greece) (Nielsen M., 1994).
• On a rather smooth mountainous site (Altamont Pass, USA) at the hub height of a 33 m turbine and a distance of
63 m differences in average wind speed occur of 30 to 40 % with winds over slopes of approximately 25%
(Jamieson P., 1988).
• Around a 30 degrees ramp (height 16 m) differences in average wind speed occur of a factor 2 to 4 over a
distance of 100 m (observed at 10 m height) (Courtney M., 1990).
• Smooth terrain variations with a height of 10% of the rotor diameter can cause 4% speed on values at heights
equal to the rotor diameter (Pedersen T.F., 1993).

The conclusion is that the orography of wind turbine sites can cause significant to extreme differences between the
wind speed at the anemometer and at the turbine, even if the position of the anemometer and the selection of the
wind direction are chosen carefully. None of the references contradicts the criteria of the draft IEC standard.
4.3 Effect Of Obstacles On Wind Field

Neighbouring wind turbines and obstacles as buildings and vegetation can cause differences in the average measured
wind speed at the anemometer location and the average wind speed experienced by the measured wind turbine. The
draft IEC standard describes which wind direction sectors must be excluded to prevent disturbance by running
turbines and other obstacles of the wind speed at the anemometer and at the measured turbine. The draft standard
regards parked turbine as obstacles. Available information on wind speed effects by running turbines is given in
paragraph 4.3.1. The effects of other obstacles (buildings and trees) are discussed in paragraph 4.3.2.
4.3.1 Wind turbines

Wind speed deficits measured up to 8 rotor diameters down wind of turbines have been presented in reference 20.
The deficits are for the largest distance (8 diameters) still significant (order of 10 %).
The ratio between the undisturbed wind speed and the wind speed measured inside a wind farm is given, as a
function of direction, in the work of Larsen A., 1989. The deficits for distances of about 6 rotor diameters downwind
vary from -5% up to 18%.
Wind speed deficits behind a turbine is given in the work of Magnusson M., 1994, for stable and unstable
stratification. The wind deficit in percentage of the ambient wind speed at hub height behind a turbine is given by
the formula ∆U/U=100(X/D)-0.8 for unstable stratification and ∆U/U=150(X/ D)-0.8 for stable stratification. The
formulae have been obtained by fitting measured data obtained from three wind farms for down wind distances
between about 2 and 10 rotor diameters. Although extrapolation outside this range from 2 to 10 diameters is not
really appropriate, an extrapolation to 20 diameters has been made for comparison with the draft IEC standard. This
standard considers upwind turbines at larger distances than 20 diameters as non-existing. Extrapolation of the above
given formula to 20 diameters and an unstable stratification results in a deficit of 9%.
Some measured and calculated wind speed deficits in wakes from turbines are presented in the work of van der
Snoek L., 1989. In the work of Emeis S., 1993, and Frandsen S., 1992, simple models to calculate the wind speed
inside a large wind farm or between a large number of obstacles are presented.
The conclusion is that the available data suggest that wind deficits at 20 diameters down wind may be still
significant.
4.3.2 Other obstacles

According to Frandsen S., 1994, the distance between an obstacle (of approximate size of the WT) and the WT
should minimal be 15-20D or 30-40D, depending of the wind direction, to secure a wind speed difference of less
than 1%.
European Wind Atlas (Troen I., 1989), gives some guidance to calculate the effect of obstacles. Using this guidance
the wind deficit has been calculated for a building (zero porosity) with height and width equal to D at a distance of
20D and at a height of 1D. The calculated deficit in this example is 2.5%.
In the work of Burger T., 1988, wake measurements made at a distance of 7m behind a house are presented.
Measured wind speed deficits behind a row of trees at rather small distances along with the influence of the
atmospheric stability on the deficit are included. This influence is rather small for distances of more than twice the
obstacle height.
In the work of Nierenberg R. (1993), the effects of trees with a height of 12m on the wind speeds between the trees
up to a height of 32m are presented.
The conclusion is that the available data suggest that wind deficits at 20 diameters down wind may be still

significant.
4.4 Site Calibration

If the 10 minutes averaged ratio between the wind speed at the anemometer position and the wind speed as
experienced by the turbine is known per wind direction sector, the wind speed measured at the anemometer position
can be translated to the wind speed experienced by the turbine during the power curve measurements. In order to
determine the 10 minutes averaged values ratio a site calibration has to be performed. It is uncertain how stable the
estimated ratio is in time. In the work of Smedman A. (1996) and Magnusson M. (1994) it is shown that terrain
effects can depend upon the atmospheric stability.
Site calibration can be done by modeling (physical or numerical) or by field measurements. The draft IEC standard
gives the possibility to perform a site calibration by numerical modelling and by field measurements using a
temporary meteorological mast. The associated uncertainties (minimum values prescribed by the draft IEC standard)
are respectively 50% and 33% of the maximum correction found in the measurement sector. More information found
in the literature is presented in the following paragraphs.
4.4.1 Wind tunnel simulation

The site can be calibrated by using a physical model of the site in a wind tunnel. Numerous examples of this
approach are found in experimental literature.
In the work of Chien H. (1980) the wind flow field over Kahuku Point has been studied using a wind tunnel and
using field measurements. The results have been compared and proved to be satisfactory. In the work of Pearse J.
(1982) wind tunnel studies of 3 smooth conical hills with slopes of 0.15, 0.25 and 0.50 have been performed. The
results have been compared with similar work on ridges. The effects of the ridges are stronger than the effects of the
hills because the flow passes around the hills rather than over it. In the work of Hassan U. (1982) the wind flow field
at Burgar Hill has been investigated by means of physical modelling (wind tunnel), numerical modelling, short term
field measurement campaigns and long term field measurement campaigns.
In the work of Vermeulen H. (1982) a wind tunnel experiment of a low rise hill (height l00m, slope 10- 20%) has
been carried out and the results have been compared with field measurements. The comparison shows good
agreements between the wind tunnel results and the field measurement results. Only heights a.g.l. of 3-10m have
been considered. In the work of Veenhuizen S. (1987) different methodologies for micro-siting have been discussed:
field measurements (using kites), physical models and numerical models. The uncertainties in wind speed resulting
from these three different approaches are 10% or higher. In the work of Vermeulen P. (1989) a wind tunnel study
has been performed on a scale model of the ECN test station for three heights, six positions and with a wind
direction resolution of 15 degrees.
In the work of McCarthy E. (1993) a wind tunnel study has been undertaken to examine the wind flow over hills and
ridges, covered with trees. The study involved two dimensional ridges with various heights, slopes and forest
configurations. Combination of the various parameter settings resulted in totally 96 different tests. In the work of
Sierputowski (1995) wind tunnel experiments have been described for a hill-valley configuration.
The conclusion is that site calibration by physical modelling is considered reliable if the modelling is done properly
in terms of orography discretisation, tunnel characteristics and instrumentation.
4.4.2 Numerical modelling

Numerical models have been developed and extensively used for site assessment purposes. A state of the art review
is found in the work of Glinou G. (1996). According to the review the available models are categorised as follows:
Kinematic or mass consistent models
The kinematic models are based on the solution of the continuity equation simulating inviscid flow fields. This
concept was introduced by Sherman C. (1978) and further developed by Traci R. (1987). Numerous models are in

use today, among which NOABL, AIOLOS and WINDS are well known. Research on the kinematic models has led
to improvements related to inclusion of atmospheric stability effects, simulation of the 3D flow field by solving a
momentum equation in the vertical direction as well as inclusion of a turbulence model, for instance the two
equation k-ε model (Croba D., 1996).
Dynamic models
Dynamic models are based on the solution of boundary layer equations. The concept of the linear perturbation
theory of Jackson P. (1975) in two dimensions and Mason P. (1979) in three dimensions introduced a two-layer
approximation using different perturbation schemes in the inner and outer region of the boundary layer. This concept
gave rise to the development of several analytical models (i.e. WASP, MS3DJH), which have given satisfactory
results in geometry’s where no flow reversal exists (Walmsley J., 1986). In the work of Mason P. (1985) a review of
different models and compared their results against their finite difference model which is effectively identical to the
model of Taylor P. (1977). At present analytical models based either on the potential flow theory or on boundary
layer parametrisations are used for predicting the wind flow over complex terrain. The WASP programme capable of
generalising a long-term meteorological data series at a (reference) site which may then be used to estimate
conditions at other (predicted) sites is nowadays the most extensively used practical tool applied for this purpose.
The MS3DJN family of models conserve mass but also attempt to solve momentum equations in an approximate
manner by using a constant-density atmosphere. It is assumed that the physics of the flow can be simplified by using
a mixing length closure for the turbulent stresses which is only valid for statically stable conditions and that the
momentum equations subjected to the thin shear layer approximation can be linearized and then solved using
perturbation techniques. In general the models belonging to this category have inherent limitations concerning the
terrain slopes and so their validity in complex terrain is not guaranteed.
Time averaged Navier-Stokes equation solutions
The solution of time averaged Navier-Stokes equations have been extensively used in all fluid dynamics applications
and specifically to atmospheric flows (Tryfonopoulos D., 1989, Alm L., 1993). In the literature several models have
been proposed for closure of the equations such as zero order models, one equation models, two equation models,
Reynolds stress models etc. To suppress the computational cost, most Navier-Stokes runs are performed in relatively
coarse grids applying law of the wall type boundary conditions. Simulations with more elaborate turbulence models,
like the low Reynolds k-ε or the algebraic Reynolds stress models, which could better estimate the wind profiles and
the Reynolds stress tensor, require large memory and computer time and this renders them cost ineffective.
Validation studies have confirmed that the k-ε is a good compromise between the simple models and the more
complicated ones (Tombrou M., 1993, Bergeles G., 1985).
Meso scale models
All the previous models are unable to account for the influence that local wind systems have on the annual wind
conditions especially at a complex terrain site. One important aspect of siting wind converters is a description of the
local wind climate with high spatial resolution which is often not available from observations in the atmosphere. The
density of operational observation networks is too poor for that purpose so that additional information is required.
Therefore methods have to be applied to achieve this information from the large scale state of the atmosphere and
topographic parameters by applying mesoscale models. The mesoscale simulation models are able to simulate local
wind systems produced by the orography and/or solar radiation, because they incorporate the appropriate physics but
also because of their working domain (characteristic dimensions of time and length of the meso scale) which is
restricted in the previously mentioned models. Only at this scale where the typical dimension of time lies between a
few minutes and a 24hour day and the typical length lies between 2,5km-mountain and valley winds to 2500km-
tropical cyclone, local wind systems phenomena occur. All meso scale models calculate the wind field by numerical
solving the Navier-Stokes conservation equation for mass, energy, momentum etc. The specific formulation of all
equations and the used approximations depend on the specific meso scale model that is selected. Summarising, meso
scale models are able to describe the wind field with a high spatial resolution during short episodes in complex
terrain. For gaining a wind climatology from simulations with these models available statistical information about
the large scale state of the atmosphere have to be scaled down to the required resolution by applying meso scale
models, which describe principally the modification of the large scale state by meso scale processes induced by the
topography.
A number of meso scale models that are used for site assessment applications are found in the literature, i.e. the
HIRLAM model by Danish Meteorological Institute, the KAMM model by University of Karlsruhe, the GESIMA

model by GKSS, the MEMO model by IEH-REG, the UK Meteorological Office model and the meso scale model
run by the University of Uppsala (Petersen E., 1996, Landberg L., 1996, Mengelkamp H., 1996 Hojstrup J., 1996).
The tremendous computer resource needed for applying these models should be taken into account. Moreover the
grid step used in the analysis which can not always be as small as required and the fact that the models require
several basic assumptions (i.e. definition and number of episodes, assumed frequency distribution of wind velocity
etc.), lead us to believe that at the moment their results are considered to allow for a qualitative description of wind
statistics in regions where observations are not available but the accuracy of the absolute values resulting from these
procedures is not always ensured.
Hybrid models
Recently, research addressed towards the combination or nesting of models with different spatial resolutions
(i.e.WASP /local scale with KAMM or HIRLAM / meso scale (Petersen E., 1996, Landberg L., 1996) or WASP with
Navier Stokes solvers) and hybrid micrositing models (Lalas D., 1996) combining the mass consistent methodology
with principles of the Navier Stokes approach. During the JOULE I project, contract no.JOUROO67 entitled “Wind
measurements and modelling in complex terrain” it was attempted to estimate the possible zones of separating flows
based on the topography input and to verify by means of the measurements obtained in the same project. In the
European Wind Atlas Vol.II (to be published shortly), the KAMM model able to calculate the resource over
hundreds of kilometres was used as a supplement to the WASP model which in contrast can handle very detailed
information on the topography in a radius of approximately five to ten kilometres from the sites. Preliminary results
for assessing the wind potential of Ireland are quite promising (Frank H., 1996).
Evidently, depending on its background theory each model has certain possibilities and limitations. Simple models
are still in almost exclusive use. Due to their approximate character, calibration procedures (wind tunnel studies, in
situ) are often needed. In some cases, additional calculations with more advanced (Navier-Stokes), generalised
(meso scale) or even statistical tools (i.e. MOS -Model Output Statistics) seem necessary for improving the
prediction accuracy.
Representative site calibration applications with the models BZ- WASP ,NOABL*, MS-MICRO, (NL)MSFD,
MASON-KING MODEL D, AVENU and MS3DJH/3R (predecessor of MS-MICRO) are found in the works of
Hassan U. (1982), Veenhuizen S. (1987), Troen I. (1989), Walmsley J. (1989), Barnard J. (1990), Botta G. (1992),
Kline J. (1993), Weng W. (1994), Watson R. (1994), Sandstrom S. (1994) and Smedman A. (1996).
4.4.3 Field measurements

Prior to or after the actual power curve measurements the relation can be determined between the average wind
speed measured by the anemometer on the meteorological mast and the average wind speed experienced by the wind
turbine using an anemometer at the location of the turbine.
This anemometer can be supported by a temporary meteorological mast or by the parked turbine. These two
possibilities are treated in the following paragraphs.
4.4.3.1 Anemometer on a temporary meteorological mast

This procedure is recommended by the draft IEC standard. The ratio between the average wind speed at the
anemometer at the meteorological mast and at an anemometer at the turbine rotor center (placed on a temporary
mast) are measured during a certain period. The temporary meteorological mast is positioned at the wind turbine
location before the erection or after the dismantling of the turbine. In Klug H. (1996) it is supported that the ratio is
rather independent of the wind speed of interest and that the ratio should be measured using wind speeds between 5
and 10 m/s. The ratio has to be determined in wind direction bins of 30 degrees or finer. The recommended
minimum duration of the used measurements is 10 to 48 hours per wind direction sector (Klug H., 1996). It is
uncertain what the effect of the atmospheric stability on the wind speed distribution is. For this reason it is felt that
the duration of 48 hours per sector should be considered as a minimum.
The disadvantage of this procedure is the need to erect a temporary meteorological mast and to delay the erection of
the wind turbine (or to remove it).
The conclusion is that the procedure is expected to give reliable results if the measurement duration per wind

direction sector is 48 hours or more.
4.4.3.2 Anemometer on parked wind turbine

To avoid the disadvantage of the use of a temporary meteorological mast the anemometer at turbine location can also
be supported by the turbine itself while the turbine is parked. To prevent disturbance of the measured wind speed by
the turbine the anemometer cannot be positioned at the exact location of the turbine rotor center. In the work of
Nielsen M. (1994) and Jamieson P. (1988) the anemometer is positioned on a vertical boom (length 3 or 4m)
attached to the rear end of the nacelle while the turbine rotor automatically remained down wind. In the work of
Morfiadakis E. (1994) a procedure has been described with the anemometer mounted on a vertical 10 m mast
supported by the nacelle. The effect of wind shear has been accounted for by measuring the wind shear at the
permanent meteorological mast. In both cases (with the anemometer displaced horizontally or vertically from the
rotor center) an uncertainty is introduced because the site effect is not measured at exactly the position of interest.
Furthermore it is not exactly known which distance between the turbine and the nacelle mounted anemometer
guarantees an undisturbed wind speed measurement. Information on the disturbance of the measured wind speed by
booms and masts in the vicinity of the anemometer can be found in the works of Dahlberg J. (1996) and Pedersen
B.M. (1991). The available information shows that the effects in front of or above a nacelle at rather small distances
(a few times the nacelle dimension) can certainly not be neglected.
The conclusion is that no simple instruction has been found to position the anemometer in the vicinity of the nacelle
with an acceptable 1% wind speed disturbance. On the other hand current experimental and theoretical research is
focusing on the issue and the determination of proper position, depending on nacelle geometry and acceptable wind
speed disturbance, is feasible.
4.5 Alternative Methodology For Power Curve Measurements

The effects of complex terrain as described in paragraphs 4.2 and 4.3 may cause differences in the measured wind
speed and the wind speed as experienced by the turbine. Preventing of or accounting for these effects is difficult and
costly in terms of money and time. For this reason an alternative procedure for power curve measurements is highly
desirable. In this procedure the undisturbed wind speed at the center of the turbine rotor is obtained from the wind
speed reading of an anemometer positioned on the nacelle of the (running) turbine. The anemometer is positioned in
such a way that disturbance of the measured wind speed is minimised and that small uncertainties in the positioning
do not have significant effects on this disturbance. A suggestion for the position is given in the work of Pedersen
T.F. (1994). The relation between the average undisturbed wind speed at a meteorological mast and the average
wind speed reading of the nacelle mounted anemometer is measured on a non-complex site (or on a site-calibrated
complex site) and is assumed to be applicable for the power curve measurements of another turbine of the same
make and model at another site. It is uncertain how good this assumption is.
In the work of Dekker J. (1986) it is shown that the wind speed ratio for the examined wind turbine, with variable
rotational speed, and anemometer mounting was wind speed and rpm dependent.
In the work of Nielsen M. (1994) the power curve obtained with a 1.5D meteorological tower and a site calibration
as described in 4.4.3.2. compared with the alternatively obtained power curve. The two power curves are identical
for wind speeds up to about 11m/s. At higher wind speeds systematic differences are observed. With the alternative
procedure the power curves of four nominally identical turbines have been determined. The resulting four potential
annual energy production values are identical within 2% for the annual wind speed of Samos island. In the work of
Riddle A. (1996) the use of a nacelle mounted anemometer for power curve measurements has been mentioned but
limited data are given in order to allow the assessment of the reliability of the method. In the work of Slater J. (1996)
two power curves of one turbine of a wind farm on complex site were presented. One power curve has been
measured using a meteorological mast and a site calibration (including the measurement of the relation of the nacelle
mounted anemometer and the meteorological mast anemometer) and another with the nacelle mounted anemometer.
The observed significant differences between the power curves are obscured in the presented AEP because the
deviations in the power curves at the low wind speed compensate for the deviations at the high wind speed.
The conclusion is that none of the papers gives a proof that the relation between undisturbed wind speed and the
reading of the nacelle mounted anemometer is independent of the power curve itself. As a consequence of this a

power curve measurement cannot be performed using the nacelle mounted anemometer unless it is sure that the
power curve of the turbine is equal to the power curve of the turbine with which the relation between undisturbed
wind speed and reading of the nacelle mounted anemometer has been determined. In that case there is obviously no
need to measure the power curve again. Nevertheless it can be argued that the procedure can be of use for checking
whether the power curve of a turbine differs from the power curve of the reference turbine (for which the power
curve and the relation between the undisturbed wind speed and nacelle mounted anemometer reading has been
measured). If the measured “power curve” of the tested machine is equal to the power curve of the reference turbine,
one can conclude that the power curves are indeed equal. The reason of this is that a positive deviation of the power
curve from the reference curve will change the wind speed relation to lower ratio’s which will make the measured
“power curve” even higher. In the same way a negative deviation will be enhanced in the results. If the measured
“power curve” of the turbine deviates from the reference curve one cannot conclude that the power curve of the
turbine is equal to the reference power curve and the measured “power curve” has no further meaning.
4.6 Effects Of Turbulence And Wind Shear On Power Curve

The previous paragraphs focused on the effects of complex terrain on the measurement of the power curve. These
effects cause uncertainties in the determination of the undisturbed wind speed at the center of the turbine rotor.
However complex terrain can also have effects on the power curve itself since the power curve of a turbine is (by
definition) basically the relation of the undisturbed wind speed at the center of the turbine rotor and the power output
of the turbine. The effects of the turbulence levels and the wind shear on the produced power are not accounted for.
This makes the measured power curve to a certain extent dependent on the measurement site.
The possible effects of turbulence on the power are the increase of wind power with increasing turbulence for equal
10-minutes average wind speeds and hysteresis effects around cut-in wind speed, rated wind speed and cut-out wind
speed. Since the power of a turbine reflects the rotor disc averaged wind speed rather than the wind speed at the
rotor center, effects of wind shear on the measured power curve can also be expected.
In the work of Morfiadakis E. (1994) it is shown the significant importance of the turbulence intensity for the power
curve. In the work of Sheimman Y. (1992) an analytical method for taking these effects into account is reported. It is
shown that not accounting for the effect of turbulence can lead to errors in predicted annual energy productions at
other sites of 10%. Elliot D. (1994) reports significant turbulence and shear effects on the power curves of three 2.5
MW turbines. The effects are correlated as the turbulence conditions are characterised by a strong shear in the lower
half of the rotor disk and weak or negative shear in the upper half.
In the work of Jamieson (1988) the reduction of all curvatures in the power curve due to turbulence is reported.
Especially in hourly averaged data this effect can be strong. The paper says that many commercial agreements are
based on hourly averaged data.
In the work of Riddle A. (1996) it is shown that a significant dependence of the power curve with turbulence but a
strong dependence of the power curve with the wind direction, as measured in a complex site. In the work of
Pedersen T.F. (1996) it is shown that the response of a cup anemometer to turbulence and 3D winds differs from the
response of the turbine, in terms of power output, for these winds. This causes a dependence of the power curve with
the wind field. This effect however seems to be rather small. Albers A. (1996) indicates that the power curve of a
turbine is influenced by the combination of turbulence, wind shear and atmospheric stability.
The conclusion is that, even if all measurement uncertainties would be eliminated, the power curve of a wind turbine
is not site-independent. Expressed in potential annual energy production values the differences can be in the order of
10 %.
4.7 Conclusions
The literature research resulted in many topic-related articles and documents providing however only limited ready-
to-use information. The conclusions are summarised below.
Effect of orography
The orography of wind turbine sites can cause significant to extreme differences between the wind speed at the

anemometer and at the turbine, even if the position of the anemometer and the selection of the wind direction sector
are chosen carefully. None of the references contradicts the criteria of the draft IEC standard.
Effect of obstacles
The available data suggest that wind deficits at 20 diameters downwind may be still significant. Extension of the
20D range of the draft IEC standard should be considered.
Site calibration
• Site calibration by physical modelling is considered be reliable if the modelling is done properly, in terms of
orography discretisation, wind tunnel and instrumentation characteristics.
• Site calibration by numerical modelling may result to increased uncertainty. On the other hand, systematic
research is being performed on the issue indicating that numerical modelling should be considered as a
candidate methodology .
• Site calibration by field measurements:
When the anemometer at the turbine location is supported by a temporary meteo-mast the site calibration is
expected to give reliable results if the measurement period per wind direction sector is 48 hours or more.
When the anemometer at the turbine location is supported by the parked turbine a procedure is needed to find an
“undisturbed” location for the anemometer. The experimental research performed within the project as well
as on-going experimental and theoretical research indicate that procedures for location selection may be
defined.
Alternative power curve measurement procedure.

The alternative power curve measurement procedure is based on the use of an anemometer mounted on the nacelle
of the running turbine. The relation between the reading of this anemometer and the undisturbed wind speed is
measured on another (nominal identical) reference turbine at another site. Although, it is possible that the wind
speed disturbance is power curve dependent and the alternative procedure is expected to introduce uncertainty, the
methodology may be used for verifying whether the power curve of a turbine is equal to the power curve of the
reference wind turbine.
Site dependency of power curves.
Even if all measurement uncertainties are eliminated the power curve of a wind turbine is site dependent through the
wind field characteristics. Expressed in potential annual energy production values the site-induced differences can be
in de order of 10% in complex sites.

5. EXPERIMENTAL RESEARCH ON WIND TURBINE

POWER PERFORMANCE
5.1 Introduction
Power performance campaigns were selected and analysed focusing on the issues of the procedure that are related to
complex terrain as well as to the identification of the terrain related wind characteristics that affect the power
performance of wind turbines.
The analysed campaigns regard wind turbines ranging from 100kW up to 1MW operated at different sites at flat or
complex terrain. A list of the analysed campaigns is found in table 1.1. In the following paragraphs a review of the
performed experimental research is presented.
5.2 Review Of Power Performance Campaigns

5.2.1 WINCON 100XT operating at CRES test station
5.2.1.1 Description of the experimental research

C.R.E.S Test Station is situated about 50 km SE of Athens. The hill on top of which the WT is located and where the
measuring campaign was performed, is 120m high above the sea level. The surrounding terrain is fairly complex and
includes hills and coastal regions. The ground cover within a 500m radius is isolated trees, bushes and small summer
cottages. Winds are mainly from the North, with occasional strong southern winds. To the west of the site at
approximately 2-3km there is a cluster of 450m high hills, while on the other three sides lies the sea at distances
varying from 2 km east to 5 km towards the north. The roughness length value has been estimated in terms of visual
inspection as 0.13 m. Three 40m towers were erected at a distance of 1.5D from the WT and were equipped with
five Gill sonic anemometers in a way to form a cross. At the cross intersection point a sonic anemometer is mounted
at the hub height (22.4m) of the central tower. Moreover, four couples of cup anemometers and vane resolvers were
mounted at the two side towers above and below the hub height, at a radius distance of 9.4m from the hub
anemometer of the central tower. This distance corresponds to a 85% of the WT’s blade length. Two more cup
anemometers were used: one mounted at the cross intersection point and the other mounted at the hub height on the
permanent meteorological mast of the Test Station, namely the north mast, at a distance of 2D from the WT. All
masts were of lattice type, whereas the central mast was NRG type mast. The distance of the reference
meteorological mast, namely the north mast, is 2D from the wind turbine at NNE direction.
The measurements used in site calibration and machine performance analysis are within the [330o-30o] wind
direction sector. The reference mast is within the selected sector which also includes the prevailing wind directions.
5.2.1.2 Site calibration

The anemometer used for the site calibration procedure was installed at 32.4m height at the location of the wind
turbine, using a mobile meteorological mast erected on the nacelle of the wind turbine. Data was collected for a
period of approximately 2 months. During this period the wind turbine was at stand-still, in the downwind position
and the wind speed varied from 2m/s to 14m/s. The measurements from the nacelle cup were compared with the
measurements from the main reference mast at hub height.
In figure 5.1 the distribution of the wind direction against wind speed, measured at the reference mast during the
calibration session, is presented. In figures 5.2 and 5.3 the distributions of the turbulence intensity and the wind
shear exponent (estimated from least square fitting) are shown. Wind shear is estimated from measurements
collected from the west, central and east mast, that are located between the reference mast and the wind turbine. The
measurements from the nacelle cup, at height 32.4m, are presented versus the relevant measurements from the

reference cup, at hub height, in figures 5.4 and 5.5 for the mean wind speed and the turbulence intensity
respectively. The wind speed at the turbine location and at hub height is estimated from the nacelle cup measurement
after correction for wind shear. In figure 5.6 the ratio of the estimated wind speed at wind turbine location at hub
over the reference value is presented for the 5-10m/s wind speed range.
The correction factors to be applied to the measured reference wind speed were estimated by applying linear
regression, with no intercept, of the estimated wind speed values at wind turbine location at hub height against the
measured reference wind speed. The results are summarised below:
Wind direction bin Correction factor Uwt/Uref Max. correction %
330o-340o 1.018 9%
340o-350o 1.017 6%
350o-360o 1.019 6%
0o-10o 1.014 4%
10o-20o 1.014 4%
20o-30o 1.018 4%
5.2.1.3 Power performance measurements and analysis

Detailed presentation of the wind and turbine measurements are found in the work of Morfiadakis E. (1996d). Data
from the 60o wind direction sector from 330o to 30o were analysed. The reference wind speed and direction were
measured by the north mast at hub height. In figure 5.7 the wind direction versus wind speed is presented. For the
selected wind direction sector the turbulence intensity versus wind speed and wind direction are presented in figures
5.8-9. In figure 5.10 the estimated air density as a function of wind speed is presented.
The power output, mean and extreme values, of the WT110XT as a function of the wind speed is presented in figure
5.11. The standard deviation of the power output and the power coefficient, estimated for the 346m2 rotor swept
area, as a function of wind speed are given in figures 5.12 and 5.13, respectively. The bin averaged power curve and
the power coefficient of the tested wind turbine are given in figure 5.14.
The uncertainty estimation of the measured magnitudes follows IEC (1997) recommendations. In table 5.1 the
summary of the uncertainty components for the W110XT is given.
In order to assess the effect of reference wind speed correction and wind direction bin selection on power
performance, AEP calculations were performed (based on Rayleigh wind speed distributions) for various power
curve estimation cases.

AEP estimations for general applications (Rayleigh distribution)

Mean annual AEP Measured Uncertainty AEP Extrapolated
wind speed Power curve Power Curve
[m/s] [MWh] [MWh] [%] [MWh]
5 112,63 4,11 3,65 112,63
6 188,03 5,87 3,12 188,10
7 264,27 7,13 2,70 265,09
8 333,55 7,94 2,38 337,54
9 390,49 8,38 2,15 401,86
10 432,40 8,54 1,97 455,72
11 459,21 8,50 1,85 497,73
5.2.2 VESTAS V27 operating at Andros site

Andros is a typical mountainous Cycladic island, situated in the southern Aegean sea. The total area of the island is
approximately 400km2, measuring 40km in length and 16km maximum width. The wind farm, with the tested
machine, is situated near the Kalyvari village at the northernmost part of the island on top of a small ridge with NE
to SW orientation. The altitude of the ridge varies between 330 and 320m ASL. The ridge is about 2.5 Km from the
closest coast and overlooks the Strait of Euboea to the NNW, i.e. the 10km wide channel separating Andros from the
island of Euboea. To the N & NNE the sea extends uninterrupted for over 50 Km. To the south of the wind farm
and at a distance of 10km lies the Agioi Saranda ridge (714m ASL) with a NE to SW orientation. The landscape in
the vicinity of the wind farm is bare with sparse and low lying vegetation and small rocks. Some low loose-stone
built huts serving the local shepherds are found near the wind farm but they can hardly be considered as obstacles
because of their low height (less than 3m) and the fact that they are integrated in the overall topography. The typical
roughness length has been estimated at a value of z0=0.10 that corresponds with farmland with close appearance.
The distance of the reference meteo mast (mast M2) is approximately 1.6D from WT at a east-northeast direction.
The measurement sector, used in site calibration and machine measurement campaign as well, is confined in the
range [340o-15o], so that no shadowing effects either on WT operation or wind measurements are present. This
sector includes the prevailing wind directions.

The anemometer used for the site calibration procedure was installed at 40m height at the location of the wind
period of approximately 30 hours. During this period the wind turbine was at stand-still, in the downwind position
and the wind speed varied from 5m/s to 14m/s. The measurements from the nacelle cup were compared with the
measurements from the main reference mast (i.e. mast M2), at the same height. Full account of the experiment can
be found in the work of Morfiadakis E. (1994a).
The ratio of the mean velocity measured at the WT location over the mean wind speed at M2, for the same height of
40m, was found to have an average of 1.015. Taking into account that mast M2 has been installed on the top of the
ridge close to the wind turbine, and the fact that the wind sector used for the analysis corresponds to directions
perpendicular to the main axis of the ridge, the hypothesis of equal velocity gradient at both mast and wind turbine
locations is quite satisfactory. With this hypothesis, the correlation coefficient calculated for the anemometers at
40m is applied also to the anemometers mounted at 31.5m height. Using the above mentioned correlation coefficient
the power curve referred to the undisturbed wind speed at the WT location is calculated.
In this application a 3% standard uncertainty for wind speed measurements due to the terrain effects, is applied when
no corrections for the reference wind speed are considered. When the reference wind speed is corrected in order to
take into account any terrain induced effects a 1% standard uncertainty is utilised as an indicative approximation,
justified by the complex terrain.


Detailed presentation of the wind and turbine measurements are found in the work of Morfiadakis E. (1996c). Data
from the 35o wind direction sector from 340o to 15 o were analysed. In figure 5.15 the wind direction versus wind
speed is presented. For the selected wind direction sector the turbulence intensity versus wind speed is presented in
figure 5.16. The mean turbulence intensity level is about 10% varying from 5% to 15%. In figure 5.17 the wind
shear exponent as a function of wind speed is presented. The wind shear exponent calculated, via least square fitting,
from wind speed measured at three different heights varies from -0.05 to 0.15. Finally in figure 5.18 the air density
as calculated from the temperature and pressure measurements is presented. In figures 5.19 and 5.20 the power
output and its standard deviation are given as functions of mean wind speed. The power coefficient is presented in
figure 5.21 whereas in figures 5.22 and 5.23 the pitch angle and its standard deviation are given.
summary of the uncertainty components for the V27 is given.
The estimation of annual energy production is based on representative Weibull distributions (corresponding to the
test site conditions) and on Rayleigh type distributions with mean wind speed ranging from 5 to 11m/s. The former
estimations are suitable for Greek applications whereas the latter for general applications. The results are given in
the following table.
5 272,81 10,95 4,01 272,83
6 438,06 15,01 3,43 438,88
7 603,55 17,98 2,98 609,93
8 747,74 19,76 2,64 771,50
9 857,19 20,54 2,40 914,36
10 928,31 20,57 2,22 1032,74
11 964,59 20,09 2,08 1123,73
The following observations are made:

• the differences found between the “uncorrected” and the “nacelle cup corrected” power curves are within the
uncertainty limit bounds of the relative AEP estimation; this is due to the limited magnitude of the correction
factor
• the differences encountered when using different wind direction sectors for power curve estimation, present
significant differences (up to 12%) that can not be attributed only to measurement uncertainty;
5.2.3 VESTAS V27 measured by DEWI

The power performance experiment was taken place during the period of 12/10-15/4/94. The experiment followed
the recommendations of IEA (1990). No site calibration was performed. The reported data refer to a valid wind
direction sector of 135°-225°.
The accuracy of the used instruments was as follows:
• cup anemometer (after calibration): 0.1m/s
• power transformer: 1.5kW
• temperature sensor: 1.0°C
• pressure sensor: 2.0hPa
In figure 5.24 the bin averaged power curve and power coefficient are presented.

5.2.4 VESTAS V27 measured by WINDTEST

The measurement campaign has been performed during the period 10/3-16/3/94 according to the IEA (1990).
The uncertainty of the instrumentation was as follows:
• wind speed: ±0.2m/s
• power transformer: ±0.5% at the nominal power of 300kW
• air density estimation ±0.25%
The bin averaged power curve is presented in figure 5.25. AEP estimations for general applications are presented in
the following table:
5 271,40 12,04 4,44 271,65
6 432,99 20,00 3,96 936,81
7 588,81 16,66 2,83 608,95
8 713,98 17,23 2,91 772,15
9 797,86 17,03 2,13 916,44
10 892,20 16,34 1,94 1035,80
11 854,51 15,39 1,80 1127,36
5.2.5 NORDTANK 500/37 operating at Toplou site

Toplou is located at the eastern part of the Crete island. Crete is the largest island in Greece situated at the southern
Aegean sea. The total area of the island is approximately 8500 km2, measuring 250 km in length and 60 km
maximum width. The eastern part of Crete is characterised by complex topography with many hills and ridges of
various slopes from 3 to 5o (smooth) to 20o+ (very steep). Toplou site is a very complex terrain site. The landscape
in the vicinity of the wind turbine is bare with sparse, low lying vegetation and small rocks. The typical roughness
length has been estimated by means of visual inspection at about 0.10m. The distance of the reference meteo
mast is approximately 2D from WT at a SSW direction. The measurements, used in site calibration and machine
performance analysis as well, are within the [275o-350o] wind direction sector, so that no shadowing effects either on
WT operation or wind measurements are present. This sector includes the prevailing wind direction.

Prior to the installation of the wind turbine a meteorological mast was erected at the location of the WT. The mast
was equipped with cups and vane resolvers at three heights, i.e. 16.5m, 35m and 39m. Data was collected for a
period of 13 days (15 to 28/12/1994). The measured wind speed at WT location at hub height was compared with
the corresponding measurement from the reference mast.
In figures 5.26 and 5.27 the distribution of the wind direction and turbulence intensity versus wind speed at hub
height for the reference mast are shown. In figure 5.28 the wind speed measured at the reference mast is shown
against the one measured at WT location, for all wind directions. In figure 5.29 the bin averaged data for the wind
speed ratio (WT over ref. measurement is presented against wind direction measured at the reference mast. The wind
speed range was set to 5-15m/s in order to increase the available data and minimise the uncertainty of estimating the
average wind speed ratio.
By using 10o wind direction bins, for the wind speed range of 5-15m/s the estimated correction factors (average and
maximum correction) are summarised in the following table:

Wind direction bin Correction factor Uwt/Uref Max. correction %

-280o 0.976 9%
280o-290o 1.000 9%
290o-300o 0.970 8%
300o-310o 1.000 11%
310o-320o 0.977 10%
320o-330o 0.969 9%
330o+ 0.990 9%
In this application a 3.67% standard uncertainty corresponding to one third of the maximum correction encountered
(i.e. 11%), for wind speed measurements due to the terrain effects, is applied when no corrections for the reference
wind speed are considered. When the reference wind speed is corrected in order to take into account any terrain
induced effects a 1% standard uncertainty is utilised as an indicative approximation, justified by the complex terrain.

Detailed presentation of the wind and turbine measurements are found in the work of Morfiadakis E. (1996b). Data
from the 75o wind direction sector from 275o to 350 o were analysed. In figure 5.30 the wind direction versus wind
speed is presented. For the selected wind direction sector the turbulence intensity versus wind speed is presented in
figure 5.31. The mean turbulence intensity level is about 10% varying from 5% to 15% . Extreme values of about
20% are also present for wind speeds of 8 m/sec. In figure 5.32 the wind shear exponent as a function of wind speed
is presented. The wind shear exponent calculated from the three wind speed at different heights varies from -0.5 to
+0.40. Finally in figure 5.33 the air density as calculated from the temperature and pressure measurements is
presented. In figures 5.34 to 5.36 the power output, its standard deviation and the power coefficient are presented as
functions of mean wind speed (measuring period from 7/12/95 to 1/6/1996).
summary of the uncertainty components for the NT500 is given.
The estimation of annual energy production is based either on representative Weibull distributions (corresponding to
the test site conditions) or on Rayleigh distributions with mean wind speed ranging from 5 to 11m/s. The former
estimations are suitable for Greek applications whereas the latter for general applications.
The following tables for AEP estimation present the calculations based on the corrected data (i.e. corrected reference
wind speed according to site calibration findings) for the whole wind direction sector.

5 539,44 21,00 3,89 539,43
6 887,33 29,88 3,37 887,37
7 1254,71 36,54 2,91 1255,39
8 1604,85 40,78 2,54 1609,10
9 1910,07 42,99 2,25 1924,35
10 2153,13 43,71 2,03 2185,80
11 2327,72 43,38 1,86 2386,00
The following observations are made:

• the differences found between the “uncorrected” and the “nacelle cup corrected” power curves are within the
uncertainty limit bounds of the relative AEP estimation; this is due to the small magnitude of the correction
factors (up to 3%)
• the differences encountered when using different wind direction sectors for power curve estimation, present
significant differences that can not be attributed only to measurement uncertainty; the available data in each
wind direction sector may be not adequate to describe consistently the power curve

5.2.6 NORDTANK 500/37 operating at Taff-Ely (turbine A)
5.2.6.1 Description of experimental research

The evaluation was carried out in the wind farm of 20 NTK500/37 turbines located in a complex terrain at an
elevation of 300 meters above sea level. The evaluation was conducted by NEL as an independent assessor acting
within a purchase and sale contract between the turbine manufacturer and the wind farm developer. IEA (1990)
recommendations were prescribed as being the reference procedural standard. The machines chosen for the analysis,
herein noted as turbine A and B, were those which had the smoothest incident wind flows (i.e. least steep slopes in
the direction of the prevailing south westerly winds). One met mast was used for both turbines being assessed.
Turbine A measured tip angles were found to be -2.1°, -2.1° and -2.0°.

To avoid interference from other turbines on both the mast and the turbine under test, only data measured in the
sector 164° - 259° N were used in the power curve determination. The site does not meet the requirements of the
IEC procedure with regard to the slopes of the surrounding terrain and therefore a site calibration exercise had to be
carried out to determine the wind speed-up between the meteorological mast and the turbine under test. The site
calibration was carried out using WASP Consequently it was found that site calibration using WASP was not
straightforward in a contractual context because of the sensitivity of results to grid resolution. Agreement was finally
reached that all parties would accept the results of an analysis using the Big WASP code based on a contour
resolution of 2m within 200m and a contour resolution of 10m within a 5km domain. In the event it was only
possible to access 5m resolution data for the inner domain, but this was extended to a 1600m radius. Although
internal calculation in WASP is believed to be carried out at 5 or 6 degree direction intervals, the data is only
available for output in 30 degree sectors. However the analysis can be run with any initial directional offset. To
determine directional dependent speed up factors (i.e. site calibrations), the model was run many times, with initial
directional offsets being increased by 5 degrees far each run. Linear interpolation was then used to determine the
appropriate wind speed correction for each valid data record. The maximum correction applied was 4.7 per cent. The
results for turbine A are presented in table 5.4. Each entry in the `relative to mast’ column shows the 30 degree
averaged speed up centered on the stated direction. Changing the direction by 5 degrees is seen in some cases to
change the averaged correction factor by up to 1.5 percentage implying that the real directional dependent
fluctuations in correction factor will probably be very significant. According to the site calibration results there are
implications that very small directional sectors should be used in experimental calibration.

The measurements and analyses were carried out according to IEC TC88/1400-12/Ed.1 (CDV), dated May 1996.
Modifications adopted at the TC88 Meeting, Tokyo in October 1996 have also been applied. The measurement
period lasted 7 months. The power curve was corrected to a reference air density of 1.225 kg/m3. Data recorded
during dry and wet (i.e. rain) conditions were used in this analysis. Topographic corrections were applied to the
wind speed data used in this analysis. Mean power output and power coefficient data are presented in figures 5.37
and 5.38 respectively. The bin average power curve is presented in figure 5.39.
At the time of analysis, the IEC guidance on directional rejection criteria was not fully evolved, particularly with
regard to wakes from adjacent turbines, therefore an approach was adopted whereby data were split into narrow
direction sectors and the consistency in the resulting adjacent power curves was used as the selection/rejection
criteria.
The annual energy production was derived from the measured power curve and from the power curve extrapolated to
the cut-out wind speed of 25m/s. The measured power curve includes wind speeds up to 23.5 m/s in 0.5 m/s intervals
with at least 30 minutes worth of data in each bin. The extrapolated curve assumes that the mean power in the last
bin with at least 30 minutes of data (24.5 m/s in this case) is sustained to the cut-out wind speed. A Rayleigh
distribution has been assumed, with annual mean wind speeds of 4 to 11 m/s in 1m/s intervals. The standard
uncertainty of the energy yields has also been calculated.


4 203 66 33 203
5 458 99 22 458
6 777 129 17 777
7 1125 152 14 1125
8 1469 inc. 168 11 1470
9 1785 inc. 176 10 1787
10 2053 inc. 180 9 2058
11 2262 inc. 181 8 2273
There were problems in agreeing whether or not the warranted power curves had been attained because of lack of a
contractual prescription defining the interpretation of the derived uncertainty figures.
5.2.7 NORDTANK 500/37 operating at Taff-Ely (turbine B)

The evaluation was carried out in the wind farm of 20 NTK500/37 turbines located in a complex terrain at an
elevation of 300m above sea level. The evaluation was conducted by NEL as an independent assessor acting within a
purchase and sale contract between the turbine manufacturer and the wind farm developer. IEA recommendations
were prescribed as being the reference procedural standard. The machines chosen for the analysis, herein noted as
turbine A and B, were those which had the smoothest incident wind flows (i.e. least steep slopes in the direction of
the prevailing south westerly winds). One met mast was used for both turbines being assessed. Turbine A measured
tip angles were found to be -2.2°, -2.2° and -2.1°.

To avoid interference from other turbines on both the mast and the turbine under test, only data measured in the
sector 190° - 262° were used in the power curve determination. The site does not meet the requirements of the IEC
procedure with regard to the slopes of the surrounding terrain and therefore a site calibration exercise had to be
carried out to determine the wind speed-up between the meteorological mast and the turbine under test. The site
calibration was carried out using WASP. Speed-up factors were determined for each 5° sector within the bulk of the
valid measurement sector, although on the periphery of the valid sector, 30° increments were used. The corrections
are tabulated in table 5.5. Linear interpolation was then used to determine the appropriate wind speed correction for
each valid data record. The maximum correction applied was 1.5 per cent.

The measurement period lasted 7 months. The power curve was corrected to a reference air density of 1.225 kg/m3.
Data recorded during dry and wet (i.e. rain) conditions were used in this analysis. Topographic corrections derived
using the WASP wind flow modelling package were applied to the wind speed data used in this analysis. Mean
power output and power coefficient data are presented in figures 5.40 and 5.41 respectively. The bin average power
curve is presented in figure 5.42.
The annual energy production was derived from the measured power curve and from the power curve extrapolated to
the cut-out wind speed of 25 m/s. In this case the measured and extrapolated curves are the same. The measured
power curve includes wind speeds up to 25 m/s in 0.5 m/s intervals with at least 30 minutes worth of data in each
bin. The extrapolated curve assumes that the mean power in the last bin with at least 30 minutes of data (24 m/s in
this case) is sustained to the cut-out wind speed. A Rayleigh distribution has been assumed, with annual mean wind
speeds of 4 to 11 m/s in 1m/s intervals. The standard uncertainty of the energy yields has also been calculated.


4 230 52 23 230
5 495 72 15 495
6 822 88 11 822
7 1171 99 8 1171
8 1510 inc. 106 7 1511
9 1814 inc. 110 6 1818
10 2067 inc. 112 5 2077
11 2261 inc. 112 5 2279
5.2.8 NORDTANK 500/37 operating at Pilsum site

The test was performed at Pilsum, Krummhörn, Lower Saxony, Germany. The site is located nearly 1.5 km east of
the coast of the North Sea. The ground has only low differences in height. The average turbulence intensity of the
measuring sector is nearly 8 % at hub height. The measurement was done following IEA (1990) recommendations.
The measurement sector was 120°-225°. The anemometers were calibrated in the wind tunnel of the university of
Oldenburg. Measurement period was three months. The measured power curve was adjusted to a reference air
density of 1.225 kg/m³. The blade pitch angle was -1.7°.
In figures 5.43 the power and power coefficient curves are given. In table 5.44 the bin averaged curves are also
presented. Annual energy production estimations were made under the assumption of 100% availability. The
estimations were performed Rayleigh distributions with annual average wind speeds from 5 to 11 m/s at hub height
(35.0 m) for general applications.
The absolute uncertainties of the measured wind speed, electrical power and air density are tabulated below:
• wind speed: 0.1m/s
• power: 5kW
• air density: 0.01kg/m3

The estimated annual energy production, based on a Rayleigh distribution is given in the following table:
5 495 41 8 -
6 831 50 6 -
7 1195 56 5 -
8 1550 61 4 -
9 1869 65 4 -
10 2134 68 3 -
11 2337 69 3 -
5.2.9 NORDTANK 500/37 measured by WINDTEST

The measurement campaign has been performed during the period 22/12/93-2/2/94 according to the IEA (1990)
recommendations.
The uncertainty of the instrumentation was taken as follows:
• wind speed: ±0.2m/s
• power transformer: ±0.5% at the nominal power of 750kW
• air density estimation ±0.25%
The measured power curve is presented in figure 5.45., whereas the annual energy estimations are presented in the
following table:
5 990,59 23,31 4,75 990,78
6 821,96 30,31 3,69 826,01
7 1163,77 39,89 2,99 1189,30
8 1462,93 37,20 2,54 1595,09
9 1685,55 37,73 2,24 1866,82
10 1826,02 37,12 2,03 2136,88
11 1893,30 35,74 1,89 2346,09
5.2.10 MADE AE-30 operating at Fuerteventura and Tarifa

The scope of the research on MADE AE-30 turbines was the comparison of the power performance of two identical
machines operating at Monteahumada on the Tarifa complex terrain area and at Cañada de la Barca in Fuerteventura
island flat terrain site. The characteristics of the MADE AE-30, the monitoring system and the two sites are found in
the work of Cuerva A. (1996a-c). MADE AE-30 is a three bladed, horizontal axis, stall controlled wind turbine with
330kW nominal power output. In both cases, the reference wind speed was taken from properly situated
meteorological masts without performing any corrections for terrain induced effects.


Four cases for complex terrain as well as a reference one in flat terrain have been considered. These four cases are
split in two different wind directions and two different turbulence intensity ranges in order to identify their influence
on the power output.
A sector within the East wind direction ranged form 56.25º to 123.75º was considered. The general data base for this
sector presented the double stall effect which has been included in the study. By stratifying the total measurement set
by the turbulence intensity level it was revealed that turbulence plays a significant role on the appearance of this
phenomenon. In figures 5.46-47 the power curve measurements, normalised for standard atmospheric conditions, for
the low (less than 10%) and high (higher than 10%) turbulence intensities are presented. A sector within the West
wind direction ranging from 236.25º to 258.75º was also considered. On the contrary that in the case of the
measurements from East direction, double stall effects were not identifiable. In figures 5.48-49 the power curve
measurements are presented for the two turbulence ranges.
The comparison of the measured power curves for all the examined cases is presented in figures 5.50-51. Three of
the four curves in complex terrain present similar behaviour. Only the one corresponding to the East wind direction
and a range of turbulence intensity values up to 10% features lesser values than the other. The reason for this is, as
can be checked in the scatter plots, the machine operating with wind from this specific and turbulence intensity
range was affected by double stall effect. In relation to the Power curve for flat terrain it can be realised the same
way to go into the aerodynamic stall state that in the complex terrain cases which are not featuring double stall
effect. Slight deviations of these values occur for wind speeds greater than 15 m/s, these ones reach a maximum
difference of 11.4 % at 16.5 m/s (West T.I. >10%-Flat reference). These two curves have been chosen to make the
previous statement ,since they are the only ones which present real values (No extrapolated) of power up to this
wind speed value.
Annual energy production estimations are presented in the following table for all the examined cases, whereas the
comparison of the AEP estimations with respect to the reference case of flat terrain are presented in figure 5.52.
Mean wind AEP estimation [MWh/year]
speed Flat terrain Eastern dir. (56.25º-123.72º) Western dir. (236.25º-258.75º)
[m/s] T.I. < 10% T.I. > 10% T.I. < 10% T.I. > 10%
4 160.7 167.9 172.4 145.9 157.8
5 356.5 362.4 371.3 335.6 352.2
6 592.9 589.8 607.5 573.3 590.7
7 827.1 808.6 841.3 815.8 830.1
8 1023.6 987.9 1038.9 1024.5 1033.3
9 1164.5 1113.5 1181.9 1178.1 1180.9
10 1248.5 1185.8 1268.2 1273.2 1270.5
11 1283.8 1213.7 1305.9 1317.2 1310.1
It can be seen that the production for the case of East wind direction, turbulence intensity values lesser than 10% is
always lesser than the others, including the flat terrain reference, when the site mean wind speed is greater than 7
m/s must be remembered that the machine presented double stall behaviour in this situation. The differences in this
case can reach values up to 6 % (diminution) in relation to the reference or even 8 % if it is compared to the value at
11 m/s in the case of West direction for turbulence intensity ranges up to 10 %. Slight differences in the linear zone
of the power curves (bin method) lead to considerable differences in all the AEP cases, between them and in relation
to the reference for site mean wind speeds up to 8 m/s since this part of the curve is highly weighed up during the
integration of the AEP. therefore, if we were in situation to take advantage of sites with low mean wind speed this
could mean big differences in the production estimation depending on the direction and the turbulence intensity
range. When the mean wind speed goes over 7m/s at 30m high (more less limit of profitability for constant speed
technology) curves always present differences lesser than 3 % between them and in relation to the reference, but in
the case of East sector with turbulence intensity up to 10% (double stall effect) already commented. Hence for this
range of mean wind speed, direction and turbulence intensity do not seem to be a differential factor for the AEP
estimation, whenever double stall effect does not appear.
In order to illustrate the way the AEP estimation can be affected by disturbed measurements of the wind speed due

to the wake of other machines for instance (this is the case) data from a sector within which it occurs is enclosed. In
the previous analysis a West wind direction ranging from 236.25º to 258.75º is considered; this is a clean sector but
if we take into account measurements for a wind direction greater than 258º and up to 348º the result for the AEP
estimation is as follows. The comparison is done presenting the same results for the East sectors and results for a
West direction between 236.25º to 348.75º. In figure 5.53 the AEP estimations are compared to the flat terrain case.
It can be realised that the differences due to the disturbance effects (West T.I >10%) reach more than 25% in relation
to the reference values (flat terrain) surpassing strongly the differences due to double stall effect. The curve in the
case of West for T.I<10% was only slightly affected by the disturbance.
The estimated uncertainties for the AEP calculations are presented in figure 5.54.
5.3 Assessment Of Site Calibration Methodologies

5.3.1 Site calibration for WINCON 110XT
The anemometer used for the site calibration procedure was installed at 32.4m height at the location of the wind
period of approximately 2 months. During this period the wind turbine was at stand-still, in the downwind position
and the wind speed varied from 2m/s to 14m/s. Wind shear was estimated from measurements collected from the
upwind erected masts (Morfiadakis E. 1996d). The wind speed at the turbine location and at hub height is estimated
from the nacelle cup measurement after correction for wind shear. The correction factors to be applied to the
measured reference wind speed were estimated by applying linear regression, with no intercept, of the estimated
wind speed values at wind turbine location at hub height against the measured reference wind speed. AEP
estimations for two cases, namely with and without correction at upwind reference are presented in table 5.7
The main conclusions are the following:
• due to the small correction factors the differences between AEP estimations based either on corrected or on
uncorrected data are less than 3%
• the use of the rotor averaged wind speed leads to estimations close to those without any correction; this is
attributed to the fact that wind shear on the turbine location is flat and therefore the rotor centre wind speed
value is close to the averaged one
• insignificant differences, less that 2% were found to exist for AEP estimations for different wind direction
sectors
5.3.2 Site calibration for VESTAS V27

The procedure was reported in the work of Morfiadakis E. (1994a). The anemometer used for the site calibration
procedure was installed at 40m height at the location of the wind turbine, using a mobile meteorological mast erected
on the nacelle of the wind turbine. The correlation coefficient calculated for the anemometers at 40m, on the
reference mast and the nacelle, was applied also to the anemometers mounted at 31.5m height, under the assumption
that wind shear deformations are not sensitive to wind direction. Using a 5-15m/s wind speed bin for the whole wind
direction sector the average and the maximum correction factor was found equal to 1.015 for the used wind direction
sector. The reference mast was erected side to the turbine at 1.6D distance. Furthermore, the usage of an upwind
located reference mast, at a distance of 4D was also assessed. AEP estimations for two cases, namely with and
without correction at upwind reference are presented in table 5.7
The main observations are the following:
• the differences between AEP estimations based either on corrected or on uncorrected data are less than 5%
• the reference mast selection (erected side to the turbine) led to small correction factor; if the upwind mast was
used the resulting AEP differences would exceed 30%
• the differences encountered in AEP estimations for different wind direction sectors indicate the need for using
correction factors for narrower wind direction sectors; this was apparent for the side reference mast whereas the
differences when using the upwind mast were lesser

5.3.3 Site calibration for NORDTANK 500/37

Prior to the installation of the wind turbine a meteorological mast was erected at the location of the WT. The mast
was equipped with cups and vanes at three heights, i.e. 16.5m, 35m and 39m. Data was collected for a period of 13
days (15 to 28/12/1994). The measured wind speed at the wt location at hub height was compared with the
corresponding measurement from the reference mast. The correction factors were estimated using 10o wind
direction bins, for the wind speed range of 5-15m/s.
In addition, a brief site calibration with a mobile mast placed on the wind turbine nacelle while the machine was at
stand still has been performed. Wind speed measurements were collected for a short period of two days from the
nacelle cup as well as from the reference mast. With the applied technique the correlation between the wind speed at
the wind turbine location at a height of 43.5m agl and the one measured at the hub height (35m agl) at the reference
mast was measured. The calculation of the correlation at the same height, was feasible through the wind shear
estimation. AEP estimations for two cases, namely with and without correction at upwind reference are presented in
table 5.8. In figure 5.55 the comparison of the results from the two site calibration techniques is presented. Although
applied for a short period the alternative site calibration technique yields results which are in fairly good agreement
with these of the conventional site calibration.
5.3.4 Conclusions
Different site calibration techniques have been experimentally assessed. When AEP estimation was used as a tool for
the assessment all the differences encountered were well within the uncertainty limits of the measurements. The site
calibration technique using a mobile meteorological mast placed on the nacelle while the machine is at stand still
when applied on a NTK500kW at Toplou yielded results which were in fairly good agreement with these of the
conventional site calibration.
The main conclusions are:
• the position of the reference masts are quite suitable as the mean corrections applied to reference wind speeds
were limited (up to 3%)
• the maximum observed speed-up within a wind direction bin reached 9% to 11%
• the only case where significant corrections would have been applied is in the case of V27 operating at Andros
site, if the upwind mast is chosen as the reference one (the mean bin corrections well exceed 5%)
• the site calibration technique using a mobile meteorological mast placed on the wt nacelle while the machine is
at stand still was validated and fairly good agreement found with these of the conventional site calibration
• although the wind speed corrections are of limited magnitude the turbulent characteristics may present
significant variation between the reference point and the turbine location.
5.4 Conclusions
Several power performance campaigns have been analysed in order to assess the recommended procedures regarding
site calibration, power curve determination and annual energy production uncertainty estimation. The analysed data
regarded wind turbines of different size and control strategy operating at different sites.
In the fields of site calibration different techniques were assessed:
Nacelle cup at downwind parked turbine
This practice was applied in all power performance campaigns by CRES. In the case of V27 the site calibration was
performed by the aid of a cup anemometer that was installed at 8.5m above the wind turbine nacelle, using a mobile
meteorological mast, while the turbine was at standstill in the downwind direction. The measurements from the
nacelle cup were correlated to the measurements from the reference mast at the same height. Under the assumption
that the wind shear as measured at the reference mast holds at the turbine location too, the correlation factor found
was used for the correction of the reference wind speed at hub height. In the case of WT110XT and NTK500/37 the
nacelle cup measurements were corrected to hub height by using the wind shear as it was estimated from the upwind
reference mast measurements. The correlation between the hub height undisturbed wind speed and the reference one
is found using a 10° wind direction bin.

Meteorological mast prior wind turbine erection

In the case of NTK500/37, a meteorological mast was erected at the location of the wind turbine prior to the turbine
installation. The measured wind speed at WT hub height was correlated to the corresponding measurements from the
reference mast. Wind direction effects were counted for, as the wind direction bin was set to 10°. A comparison of
this approach and the one using the nacelle cup yielded results which were in fairly good agreement.
Usage of numerical simulation code (WASP)
This practice was applied by NEL for the NTK500/37 campaigns. The estimated correlation showed significant
sensitivity to small wind direction changes.

5.5 Tables
Table 5.1 Uncertainty components for W110XT power performance testing.
Measured parameter Uncertainty component Uncertainty limit Notes

Electric power current transformers ±0.35% at 20% load 2 of class 0.2
voltage transformers
power transducer ±0.5% x 170kW 1 of class 0.5, nominal power 170kW
signal transmission (DAS) ±0.1% of the power
digitisation (DAS) ±575/4096 slope 57.5kW/volt
Wind speed anemometer calibration ±1% of the velocity
operational characteristics ±1% of the velocity
mounting effects
flow distortion due to 3% of the velocity no site calibration
terrain 1% of the velocity with site calibration corrections
signal transmission (DAS) ±1% of the velocity
digitisation (DAS) ±94/4096 slope 9.4m/sec/Volt
Air temperature temper. Sensor calibration ±0.50 oC
radiation shielding ±0.80 oC
mounting effects 0.09°C the uncertainty is (1/3)/10 x (Hub-sensor
height)
signal transmission (DAS) ±0.07 oC
digitisation (DAS) ±250/4096 slope 25°C/Volt
Air pressure pressure sensor calibration ±2hPa
mounting effects 0.25hPa 10% of the applied correction
signal transmission (DAS)
digitisation (DAS) ±520/4096 slope 52hPa/Volt

Table 5.2 Uncertainty components for V27 power performance testing.
Measured parameter Uncertainty component Uncertainty limit Notes

Electric power current transformers ±0.35% at 20% load 2 of class 0.2
voltage transformers
signal transmission (DAS) ±0.1% of the power
digitisation (DAS) 10/4096 slope 150kW/volt
operational characteristics ±1% of the velocity
mounting effects
flow distortion due to 1% of the velocity ref. wind speed corrected
terrain 3% of the velocity no corrections applied
signal transmission (DAS) ±1% of the velocity
digitisation (DAS) 10/4096 slope 9.4m/sec/Volt
Air temperature temper. sensor calibration ±0.50 oC
mounting effects the uncertainty is (1/3)/10 x (Hub-sensor
height) + 1
signal transmission (DAS) ±0.07 oC
digitisation (DAS) 10/4096 slope 25°C/Volt
Air pressure pressure sensor calibration ±2hPa
mounting effects ±0.4hPa for a height difference of 33.5m the
correction is 4.03hPa
signal transmission (DAS)
digitisation (DAS) 10/4096 slope 52hPa/Volt

Table 5.3 Uncertainty components for NTK500/37 power performance testing.
Measured Uncertainty component Uncertainty limit Notes

parameter
Electric power current transformers ±0.35% of current at 20% 2 of class 0.2
load
voltage transformers ±0.20% of voltage at all 2 of class 0.2
loads
signal transmission ±0.1% of the power
(DAS)
digitisation (DAS) ±1500/4096 slope 150kW/volt
operational ±1% of the velocity
characteristics
mounting effects
flow distortion due to 1% of the velocity ref. wind speed corrected
terrain 3.67% of the velocity no corrections applied
signal transmission ±1% of the velocity
(DAS)
digitisation (DAS) ±61/4096 slope 6.1m/sec/Volt
Air temperature temper. sensor ±0.50 oC
calibration
mounting effects the uncertainty is (1/3)/10 x (Hub-sensor
height)
signal transmission ±0.07 oC
(DAS)
digitisation (DAS) ±250/4096 slope 25°C/Volt
Air pressure pressure sensor ±1hPa
calibration
mounting effects ±0.4hPa for a height difference of 33.5m the
correction is 4.03hPa
signal transmission
(DAS)
digitisation (DAS) ±520/4096 slope 52hPa/Volt

Table 5.4 WASP orography corrections for NTK500/37(A) at Taff Elly.
WASP Orography corrections

Absolute (%) Speed up factor
Direction (°N) Mast Turbine A Mast Turbine A Running
average
165 22,36 18,17 1,224 1,182 -3,42%
170 25,25 20,56 1,253 1,206 -3,74%
175 28,00 23,10 1,280 1,231 -3,83%
180 30,60 25,50 1,306 1,255 -3,91%
185 32,97 27,72 1,330 1,277 -3,95%
190 34,90 28,58 1,349 1,286 -4,68%
195 36,29 31,75 1,363 1,318 -3,33%
200 36,39 33,75 1,364 1,338 -1,94%
205 35,09 34,83 1,351 1,348 -0,19%
210 33,46 33,73 1,335 1,337 0,20%
215 31,50 32,25 1,315 1,323 0,57%
220 29,28 30,45 1,293 1,305 0,91%
225 27,07 27,21 1,271 1,272 0,11%
230 24,85 25,94 1,249 1,259 0,87%
235 22,57 23,57 1,226 1,236 0,82%
Table 5.5 WASP orography corrections for NTK500/37(B) at Taff Elly.
WASP Orography corrections

Absolute (%) Speed up factor
Direction (°N) Mast Turbine B Mast Turbine B Running
average
249 14,07 14,29 1,141 1,143 0,19%
254 11,56 11,82 1,116 1,118 0,23%
259 9,99 9,72 1,100 1,097 -0,25%
264 8,79 7,97 1,088 1,080 -0,75%
269 7,88 6,65 1,079 1,067 -1,14%
274 7,43 5,61 1,074 1,056 -1,69%
279 7,12 4,92 1,071 1,049 -2,05%
284 7,21 4,52 1,072 1,045 -2,51%

Table 5.6 Site calibration assessment via AEP estimations for W110XT (CRES test station).
AEP estimation with corrected reference wind speed

Mean annual Annual Energy Production (Uncertainty) [MWh]
wind speed Wind direction sector
[m/s] 330°-30° 350°-10° 10°-30°
5 112,6 (4,11) 111,93 (4,12) 113.89 (4,12)
7 264,3 (7,13) 263.30 (7,15) 265,97 (7,23)
9 390,5 (8,38) 389.55 (8,40) 392,10 (8,66)
11 459,2 (8,50) 458,50 (8,49) 460,48 (8,92
AEP estimation without correction for reference wind speed

[m/s] 330°-30° 350°-10° 10°-30°
5 116,22 (8,35) 115,32 (8,56) 117,21 (7,85)
7 270,00 (14,49) 268.86 (14,68) 271,26 (14,37)
9 396,36 (16,59) 395.07 (16,69) 397,40 (17,24)
11 464,47 (16,32) 462,98 (16,34) 465,20 (17,56)
Table 5.7 Site calibration assessment via AEP estimations for V27 (Andros site).

[m/s] 345°-15° 345°-360° 0°-15°
5 272,81 (10,95) 270,98 (10,95) 295,69 (11,06)
7 603,55 (17,98) 603,58 (18,12) 619,78 (17,98)
9 857,19 (20,54) 859,54 (20,76) 867,49 (20,48)
11 964,59 (20,09) 968,17 (20,32) 971,06 (20,02)

[m/s] 345°-15° 345°-360° 0°-15°
5 303,63 (15,88) 281,69 (15,95) 307,18 (15,96)
7 634,49 (25,86) 620,24 (26,19) 637,14 (25,76)
9 884,78 (28,83) 876,68 (29,35) 885,86 (28,65)
11 987,89 (27,56) 983,45 (28,14) 987,92 (27,38)

Table 5.8 Site calibration assessment via AEP estimations for NTK500/37 (Toplou site).

[m/s] 275°-350° 300°-310° 310°-320° 320°-330°
5 539,44 (21,0) 509,76 (21,0) 564,89 (21,7) 539,94 (20,7)
7 1254,71 (36,5) 1212,29 (36,6) 1301,56 (37,5) 1236,72 (35,6)
9 1910,07 (43,0) 1866,20 (43,9) 1962,62 (44,4) 1878,40 (42,0)
11 2327,72 (43,4) 2290,40 (45,0) 2375,71 (45,5) 2291,42 (42,7)

[m/s] 275°-350° 300°-310° 310°-320° 320°-330°
5 515,7 (51,7) 509,8 (52,0) 530,0 (50,7) 503,5 (48,8)
7 1217,6 (92,1) 1212,3 (91,9) 1251,5 (92,1) 1171,4 (88,2)
9 1870,2 (108,3) 1866,7 (109,6) 1914,1 (105,5) 1802,1 (105,9)
11 2292,2 (108,4) 2291,9 (111,3) 2336,9 (112,4) 2218,1 (107,8)

5.6 Figures
WT110XT AT C.R.E.S SITE WT110XT AT C.R.E.S SITE
Site calibration session 10/4-5/6/96 Site calibration session 10/4-5/6/96
180 0.4
10min average
135
Turbulence intensity at hub (m/s)

90 0.3
Wind direction (deg)
45
0 0.2
-45
-90 0.1
-135
-180 0.0
0 3 6 9 12 15 0 3 6 9 12 15
Wind speed at hub - reference mast (m/s) Wind speed at hub - reference mast (m/s)
Figure 5.1 Wind direction distribution. Figure 5.2 Turbulence intensity vs wind speed at hub.

1.5 16.0
Wind speed - nacelle cup at 32.4m (m/s)
1.0
12.0
Wind shear exponent
0.5
8.0
0.0
4.0
-0.5
-1.0 0.0
0 3 6 9 12 15 0 3 6 9 12 15
Wind speed at hub - reference mast (m/s) Wind speed at hub - reference mast (m/s)
Figure 5.3 Wind shear exponent. Figure 5.4 Wind speed at nacelle cup vs reference.


0.4 1.10
Wind speed ratio at hub height: WT vs reference mast

Turbulence intensity - nacelle cup at 32.4m
0.3
1.05
0.2
1.00
0.1
0.0 0.95
0.0 0.1 0.2 0.3 0.4 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0
Turbulence intensity at hub - reference mast Wind direction at hub (deg)
Figure 5.5 Turbulence intensity at nacelle cup vs Figure 5.6 Wind speed ratio vs reference.
reference.

30 0.30
20 0.25
10 0.20
0 0.15
-10 0.10
-20 0.05
-30 0.00
0 3 6 9 12 15 18 21 24 0 3 6 9 12 15 18 21 24
Wind speed at hub (m/s) Wind speed at hub (m/s)
Figure 5.7 Wind direction vs wind speed. Figure 5.8 Turbulence intensity vs wind speed.


0.30 1.24
1.22
0.20
Air density (kgr/m 3 )

1.20
1.18
0.10
1.16
0.00 1.14
-30 -20 -10 0 10 20 30 40 0 3 6 9 12 15 18 21 24
Wind direction (deg) Wind speed at hub (m/s)
Figure 5.9 Turbulence intensity vs wind direction. Figure 5.10 Air density vs mean wind speed.
WT110XT AT C.R.E.S SITE

150
mean
max
min
100
Power output (kW )
50
-50
0 3 6 9 12 15 18 21 24
Wind speed at hub (m/s)
Figure 5.11 Active power vs wind speed at hub.


30 0.5
25
0.4
Power output SDV (kW )
Power coefficient (Cp )

20
0.3
15
0.2
10
0.1
5
0 0.0
0 3 6 9 12 15 18 21 24 3 6 9 12 15 18 21 24
Figure 5.12 SDV power vs wind speed. Figure 5.13 Power coefficient vs wind speed.
WT110XT AT CRES SITE

120 0.6
Power output
100 Power coefficient 0.5
80 0.4
Power coefficient (Cp)
Power output (kW)
60 0.3
40 0.2
20 0.1
0 Air density 1.225kg/m

3 0.0
o o
Wind direction 330 - 30
-20 -0.1
0 3 6 9 12 15 18 21
Figure 5.14 Power curve and power coefficient of W110XT.

V27 AT ANDROS SITE V27 AT ANDROS SITE

30 0.20
10min average 10min average
0.16
15
Turbulence intensity at hub

Wind direction at hub (deg)
0.12
0.08
-15
0.04
-30 0.00
0.0 4.0 8.0 12.0 16.0 20.0 0.0 4.0 8.0 12.0 16.0 20.0
Figure 5.15 Wind direction distribution. Figure 5.16 Turbulence intensity distribution.

0.30 1.17
1.16
0.20
WInd shear exponent
)
3
Air density (kgr/m
1.15
0.10
1.14
0.00
1.13
-0.10 1.12
0.0 4.0 8.0 12.0 16.0 20.0 0.0 4.0 8.0 12.0 16.0 20.0
Figure 5.17 Wind shear exponent distribution. Figure 5.18 Air density distribution.


300.00 50.00
mean
max
40.00
min
200.00
Power output (kW)
Power sdv (kW)

30.00
20.00
100.00
10.00
0.00 0.00
0.0 4.0 8.0 12.0 16.0 20.0 0.0 4.0 8.0 12.0 16.0 20.0
Figure 5.19 Power output of V27. Figure 5.20 Standard deviation of power output.

0.60 25.00
mean
20.00
max
min
0.40 15.00
Pitch angle (deg)t
10.00
0.20 5.00
0.00
0.00 -5.00
0.0 4.0 8.0 12.0 16.0 20.0 0.0 4.0 8.0 12.0 16.0 20.0
Figure 5.21 Power output of V27. Figure 5.22 Pitch angle vs mean wind speed.

V27 AT ANDROS SITE

6.00
10min average
4.00
Pitch angle sdv (deg)
2.00
0.00
0.0 4.0 8.0 12.0 16.0 20.0
Figure 5.23 Standard deviation of pitch angle.
300 0.6
250 0.5
elektr. Leistung P [kW]
200 0.4 Cp gesamt [-]
150 0.3
100 0.2
50 0.1
0 0.0
0 5 10 15 20 25
Windgeschwindigkeit in Nabenhöhe [m/s]
Figure 5.24 Power output and power coefficient for V27 as measured by DEWI.

VESTAS V27 POWER CURVE BY WINDTEST

250
bin averaged data
200
Power output (kW)
150
100
50
0.0 5.0 10.0 15.0 20.0 25.0

Wind speed at hub height (m/s)
Figure 5.25 Bin averaged power curve for V27 by WINDTEST.
TOPLOU SITE CALIBRATION TOPLOU SITE CALIBRATION

360 0.4
Turbulence intensity at hub - WT mast
Wind direction at hub (deg) - WT mast
270 0.3
180 0.2
90 0.1
0 0.0
0 5 10 15 20 25 0 5 10 15 20 25
Wind speed at hub (m/s) - WT mast Wind speed at hub (m/s) - WT mast
Figure 5.26 Wind direction vs wind speed at hub. Figure 5.27 Turbulence intensity vs wind speed.

TOPLOU SITE CALIBRATION

25.0
10min average
TOPLOU SITE CALIBRATION

20.0
Wind speed at hub (m/s) - Ref. mast
Wind speed bin 5-15m/s

1.20
Wind speed ratio at hub (Uwt/Uref)

15.0
1.10
10.0
1.00
5.0 0.90
0.0 0.80
0.0 5.0 10.0 15.0 20.0 25.0 275 290 305 320 335
Wind speed at hub (m/s) - WT mast Wind direction at hub (Ref. mast)
Figure 5.28 Wind speed at reference vs wind speed at Figure 5.29 Bin averaged wind speed ratio vs wind
WT location. direction at reference.
NT500 AT TOPLOU SITE NT500 AT TOPLOU SITE

350 0.30
335
Wind direction at hub (deg)
Turbulence intensity at hub
0.20
320
305
0.10
290
275 0.00
0.0 5.0 10.0 15.0 20.0 25.0 0.0 5.0 10.0 15.0 20.0 25.0
Figure 5.30 Wind direction vs wind speed at hub. Figure 5.31 Turbulence intensity vs wind speed.


0.08 1.26
0.04 1.24
Wind shear exponent
Air density (kg/m3)

0.00 1.22
-0.04 1.20
-0.08 1.18
0.0 5.0 10.0 15.0 20.0 25.0 0.0 5.0 10.0 15.0 20.0 25.0
Figure 5.32 Wind shear vs wind speed. Figure 5.33 Air density vs wind speed at hub.

800.0 140.0
10min mean
10min average
max
min 120.0
600.0
100.0
Power output (kW)
SDV power (kW)
400.0
80.0
60.0
200.0
40.0
0.0
20.0
-200.0 0.0
0.0 5.0 10.0 15.0 20.0 25.0 0.0 5.0 10.0 15.0 20.0 25.0
Figure 5.34 Power output of NTK500/37. Figure 5.35 Standard deviation of power output.

NT500 AT TOPLOU SITE

0.7
10min average
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0 5.0 10.0 15.0 20.0 25.0
Figure 5.36 Power coefficient for NTK500/37.
164° - 259°, includes rain

600
500
400
300
200
100
0
0 5 10 15 20 25 30
10-Minute Mean Wind Speed (m/s)
Figure 5.37 Power output of NTK500/37 at Taff-Ely (turbine A).

164° - 259°, includes rain
0.5
0.4
0.3
0.2
0.1
0
0 5 10 15 20 25 30
Figure 5.38 Power coefficient of NTK500/37 at Taff-Ely (turbine A).
164° - 259°, including rain
600.00
500.00
400.00
Mean Power (kW)
300.00
200.00
100.00
0.00
0.00 5.00 10.00 15.00 20.00 25.00
10-M inute M ean Wind Speed (m/s)
Figure 5.39 Power curve of NTK500/37 at Taff-Ely (turbine A)

190° 262°, including rain
600
500
400
300
200
100
0
0 5 10 15 20 25 30
Figure 5.40 Power output of NTK500/37 at Taff-Ely (turbine B).
190° 262°, including rain
0.5
0.4
0.3
0.2
0.1
0
0 5 10 15 20 25 30
Figure 5.41 Power coefficient of NTK500/37 at Taff-Ely (turbine B).

190° - 262°, including rain
600.00
500.00
400.00
Mean Power (kW)
300.00
200.00
100.00
0.00
0.00 5.00 10.00 15.00 20.00 25.00
Figure 5.42 Power curve of NTK500/37 at Taff-Ely (turbine B).
600 0 .6
500 0 .5
electr. power P [kW]
400 0 .4
Cp total [-]
300 0 .3
200 0 .2
100 0 .1
0 0
0 5 10 15 20 25
w ind speed v at hub height [m/s]
Figure 5.43 Mean power and power coefficient for NTK500/37 measured by DEWI

600 0 .6
500 0 .5
400 0 .4
Cp total [-]
300 0 .3
200 0 .2
100 0 .1
0 0
0 5 10 15 20 25
w ind speed v at hub height [m/s]
Figure 5.44 Bin averaged power and power coefficient curves for NTK500/37.
NORDTANK NTK500/37 POWER CURVE BY WINDTEST

600
bin average data
500
400
Power output (kW)
300
200
100
0
4.0 8.0 12.0 16.0 20.0
Wind speed at hub height (m/s)
Figure 5.45 Bin averaged power curve for NTK500/37 by

WINDTEST.

Figure 5.46 Power output for MADE AE-30 (wind Figure 5.47 Power output for MADE AE-30 (wind
direction: 56.25º-123.75º, turbulence intensity <10%) direction: 56.25º-123.75º, turbulence intensity >10%).
Figure 5.48 Power output for MADE AE-30 (wind Figure 5.49 Power output for MADE AE-30 (wind
direction: 236.25º-258.75º, turbulence intensity direction: 236.25º-258.75º, turbulence intensity
<10%). >10%).

Figure 5.50 Power curve comparison at linear part. Figure 5.51 Power curve comparison at stall region.
AEP
MADE AE-30
AEP
Different directions and turbulence intensity
Wind speed 3 m/s -18 m/s MADE AE-30
Influence of a disturbance of the measurements
%
8 on the AEP estimations
Wind speed 3 m/s -18 m/s
6
%
4 30
25
2
20
0
15
-2
10
-4
5
-6
0
-8
-5
-10
-10
0 2 4 6 8 10 12 14 16
0 2 4 6 8 10 12 14 16
Mean wind speed at hub m/s
Mean wind speed at hub m/s
EAST (56.25º-123.72º) T.I<10%
FLAT TERRAIN REFERENCE
EAST (56.25º-123.72º) T.I>10%
EAST (56.25º-123.72º) T.I<10%
WEST (236.25º-258.75º) (NO DISTURBED) T.I. <10% EAST (56.25º-123.72º) T.I>10%
WEST (236.25º-258.75º) (NO DISTURBED) T.I. >10% WEST (236.25º-348.75º) T.I.<10%
FLAT TERRAIN REFERENCE WEST (236.25º-348.75º) T.I.>10%
Figure 5.52 AEP comparison (flat terrain as Figure 5.53 AEP comparisons including wake
reference). measurements.

14
13
12
11
10
9
8
7
Total uncertainty in AEP %
6
5
4 5 6 7 8 9 10 11 12 4 5 6 7 8 9 10 11 12 4 5 6 7 8 9 10 11 12
WIND SECTOR WIND SECTOR WIND SECTOR
E1 E2 W1
14
13
12
11
10
9
8
7
6
5
4 5 6 7 8 9 10 11 12 4 5 6 7 8 9 10 11 12
WIND SECTOR WIND SECTOR
W2 FLAT
Wind Speed (m/s)
Figure 5.54 AEP estimation uncertainty for MADE AE-30.
TOPLOU SITE CALIBRATION ASSESSMENT

1.30
Calibr. prior WT installation
Calibr. with nacelle cup
Wind speed ratio (WT/Ref)
Calibr. prior WT installation (bin)

1.20
1.10
1.00
0.90
0.80
300.00 305.00 310.00 315.00 320.00

Wind direction at hub (deg) - Ref. mast
Figure 5.55 Comparison of calibration practices for NTK500/37.

6. ALTERNATIVE POWER PERFORMANCE

METHODOLOGIES
6.1 Introduction
In the present chapter the alternative power performance assessment and verification procedures that have
been recently developed and experimentally tested are examined. The considered methods regard the use
of a nacelle cup behind the running wind turbine rotor and the use of the rotor mean wind speed, yielded
from detailed rotor upwind measurements, as the reference wind speed.
6.2 Nacelle Cup Methodology

6.2.1 Introduction
The basic prerequisite for measuring the power curve of a wind turbine is the use of a cup anemometer
mounted on a meteorological mast, preferably at a distance of 2.5 rotor diameters upwind of the turbine.
However, in the occasions where the installation of the meteorological mast is unsuitable the use of the
nacelle anemometer has been suggested for verification of individual turbine power curve. The relation
between the response of the nacelle and the mast anemometer should be established during the power
curve measurements in terrain where site effects are known. Then the question to be answered is whether
this relation can be used to verify another turbine power curve which is installed elsewhere, alone or
within a wind park. It is understood that the turbine should be of the same make and type. For the turbine
manufacturer or the wind park developer this method has an obvious economic interest as it reduces the
costs involved with the assessment of the turbine power curve and the related guarantied annual energy
production.
The wind speed registered by the nacelle anemometer and thus its relation to the mast anemometer is
mainly expected to be influenced by two factors, the placement of the anemometer on the nacelle and the
retardation of the flow due to the power extracted from the wind. Mounting instructions for the placement
of the anemometer on the nacelle are found in the work of Pedersen T. (1994). The presence of the nacelle
influences the flow and can easily result in either an acceleration or retardation of the wind speed at the
position where the cup anemometer is mounted. The retardation on the other hand of the flow due to the
energy extraction is a function of the wind speed (provided that the rotor settings and operational
characteristics remain unchanged). Since this relation is not linear, it is expected that neither is the relation
between the mast and nacelle anemometer. In the work of Petersen S. (1994b) a linear relation was found
whereas in the work of Slater J. (1996) the relation is partly below and partly above unity. The above
experimental research show clearly the degree of complexity of the flow around the nacelle
Three power performance campaigns are analyzed for the assessment of the application of nacelle cup
anemometer regarding the ELKRAFT 1MW, a 500kW stall and the WINCON XT110XT wind turbines.
6.2.2 Nacelle cup anemometry on 1MW wind turbine

The research is based on the results of the power curve measurements that have been carried out on the
ELKRAFT 1MW stall-regulated wind turbine. The main objectives of the research were the following:

• to investigate whether for the same rotor settings the relation between the nacelle and the mast
anemometer (established with data from the 90° power curve measurement sector) is also valid for
other sectors. This is equivalent to examining whether the power curve (as a function of the corrected
nacelle anemometer) can be reproduced for other sectors and different data sets. It gives also evidence
whether the nacelle-mast relation defined within a measurement sector in flat terrain is still valid and
thus transferable to another terrain type
• to investigate whether and how the relation between the mast and the nacelle anemometer is
influenced by the specific position of the nacelle anemometer.
• to investigate whether different rotor settings affect the relation between the nacelle and the mast
anemometer and if so how the use of a wrong relation would affect the correct presentation of the
power curve and the evaluation of the annual energy production.
Data collected from three phases of operation and two nacelle anemometer height positions. The varying
parameters are tabulated in the following table for the characterization of each phase.
Measurement phase 1 2 3a 3b
Pitch settings +0.5° +0.5° -1.0° -1.0°
Nacelle anemometer height 1.2m 1.2m 1.2m 3.m
Yaw error at 10m/s +12 +7 +7 +7
Vortex generators on the no no yes yes
blades
The technical specifications of the wind turbine are found in Antoniou I. (1997).
The sector used for the power curve measurements is from 181.5° to 271.5° for the first and the second
phase data and from 175° to 265° for the third phase data. The direction defined by the met mast and the
turbine is 270°→90° with the mast located west for the turbine. The distance between the mast and the
turbine is 2.56D. The measurement sector selection has been performed according to Pedersen T. (1993b).
The terrain around the turbine is flat and in the measurement sector there are no roughness elements of
importance except for a wave brake of stones into the sea with an estimated height of one to two meters.
In the north of the measurement sector and up to 330° some big obstacles are found, namely a ship dock, a
crane and ships of varying size, the Avedøre power plant, high banks of coal, a big fuel tanks and deposits
of cinder. From 330° and up to 15°, the terrain becomes more uniform with smaller irregularities. From
15° to 180° the turbine neighbors to the sea and at a distance of some hundred meters there is a recreation
rural area without any big terrain irregularities. The area is only a few meters higher than sea level. It
should also be mentioned that the ground where the power plant and hence the turbine is located is
reclaimed at a later moment from the neighboring ground to the north and therefore this ground is at a
higher level relative to the neighboring one.
6.2.2.2 Comparison of data

Data form the sectors between 181.5° and 271.5° (first and second phase) and between 175° and 265°
(third phase). These sectors are used for the original power curve measurements and are open water
sectors with land present at a longer distance from the turbine. The slight translation of the measurement
sector in the third phase is due to the temporary formation of a cinder bank between the turbine and the
met. mast and is not consider to have any influence on the results. The cinder is now removed.
The power curve as function of the corrected response of the nacelle anemometer is, except for the above
sectors, shown also for the sectors
• 271.5° to 331.5° (turbine in the wake of the power plant)
• 330° to 360° (sector with a different roughness length)
• 110° to 180° (open water and uniform low elevation land).
4.60 European Wind Turbine StandardsII

For all selected runs the two generators of the turbine were connected to the grid. Moreover, all turbulence
intensities were included and no rain weather data filtering was performed.
6.2.2.3 Results of the measurement program

Measured operational and performance characteristics
The power curves of the turbine for all the three measurement phases are given as a function of the mast
anemometer at 55m height (hub height). The curves are presented as the mean bin data values of ten -
minutes periods, corrected for temperature and pressure (T=15°C and P=1013.3 hPa). In figure 6.1 the
turbine mean yaw error is shown for the three phases.
In order to correct the initial yaw error, the turbine’s wind vane was turned 12° in the anti-clockwise
direction. This resulted in a reduction in the yaw error by approximately 7°. Referring still to figure 6.1,
the data from phase2 and phase3a follow well each other up to 12m/s after which deviations are observed.
It is not clear whether these are due to scatter because of the limited data available or to the modifications
of the flow field behind the rotor hub caused by the presence of the vortex generators.
In figure 6.2 the turbine power curves, for the three phases, are shown together with the design power
curve calculated for a pitch angle equal to -1.0° (Petersen S., 1994b). The power curves from phase1 and
phase2 show good agreement between them up to 7m/s and are in fact slightly better than the design
power curve. Between 7m/s and 12m/s the phase2 power curve is better than the phase1 one, which can be
attributed to the lower yaw error. At higher wind speeds the two power curves agree well although the
phase2 curve shows a higher scatter due to the limited amount of data. The two curves are well below the
design curve (result of theoretical calculations) for wind speeds between 8m/s and 17m/s. An explanation
of the reduced performance of the rotor relative to the expected design curve under this configuration is
that early separation of the flow occurs in the thick inner sections of the blades. Therefore it was decided
to install vortex generators on these parts of the blades. The vortex generators were designed by LM
which also decided their positioning on the blades.
Regarding the phase3 curve and the design curve, good agreement exists between them up to 8m/s with
the phase3 curve being slightly higher than the design one. At wind speeds above 8m/s the two curves
follow well each other with the design curve being higher than phase3. The lack of high westerly winds
has hindered the completion of the power curve of this phase.
The power coefficients Cp for the three phases are shown in figure 6.3. Below 7m/s the phase3 curve lies
lower, while at higher wind speeds it reaches higher values than the two others. In line with the above
results from the power curves, it is seen that the phase2 curve is above the phase1 curve between 7m/s and
12m/s.
Relation between the mast- and the non-corrected nacelle anemometer power curves (nacelle anemometer
at 1.2m. or 3m. Height)
In figure 6.4, the power curves for the phase1 data are shown both for the mast and the nacelle
anemometer (not corrected, placed 1.2m above the nacelle and behind the non-profiled part of the blade).
The difference (distance) between the two power curves, in other words the deficit of the wind speed
recorded by the nacelle anemometer relative to the mast anemometer, is minimized for very low and very
high wind speeds, where extraction of energy from the wind is minimum according to figure 6.3, while it
reaches a maximum at values around the maximum Cp. Therefore it may be argued that there is a direct
relation between the curves in figures 6.2 and 6.3, as increased power extraction from the incoming wind
results in higher retardation of the air coming through the turbine rotor. This is illustrated in figure 6.4
where, for wind speeds around maximum Cp, the nacelle anemometer records a much lower wind speed as
compared to the mast anemometer. This is also an indication that the relation between the two cup
anemometers is not as a rule linear.

In figure 6.3 the power curve from phase3a is presented as a function of the nacelle and the mast
anemometer with the nacelle anemometer at 1.2m height from the roof and behind the non-profiled part of
the rotor. When comparing figure 6.4 and figure 6.5, the gap between the corresponding power curves is
seen to be more narrow for the phase3a data. The two power curves approach again each other at higher
wind speeds but the rate in the case of phase3a data is lower and the two curves have a tendency to run
almost parallel with each other. Thus the relation between the nacelle and the mast anemometer is now
different when compared to figure 6.4. The presence of the vortex generators, the modification of the yaw,
and the modification of the pitch angle has influenced the flow field around the rotor and the relation
between the mast and the nacelle anemometer.
In figure 6.6, the relation between the nacelle and the mast anemometer is presented for the phase3b data
when the nacelle anemometer is at 3m. height above the roof and behind the profiled part of the blade.
Before proceeding with the analysis of the figure it should be mentioned that the data points which deviate
from the main body of the power curves respond to an icing situation which occurred during the
measurement period. The two power curves are again close to each other at low wind speeds while at
higher speeds, contrary to the previous results, the deficit in wind speed as recorded by the nacelle
anemometer remain almost constant, also beyond stall. By comparing figures 6.5 and 6.6 it is seen that the
wind speed recorded by the nacelle anemometer in phase3b at 3m height is lower compared to the phase3a
data recorded at 1.2m height. The higher retardation is probably a result of the larger amount of energy
extracted by the profiled part of the blades in front of the anemometer. This conclusion is however
different than the one reached after analyzing figures 6.3 and 6.4 as here the behavior of the flow recorded
by the nacelle anemometer seems to be influenced more by the flow condition of the blade segment in
front. The extraction of any further conclusions is nevertheless hindered by the lack of knowledge of the
behavior of the surface flow in the presence of vortex generators.
This is in accordance with the results presented in figure 6.7 where a direct comparison of the bin power
curves as a function of the non-corrected nacelle anemometer at 1.2m height and for two rotor conditions
is presented. It is evident that the non-corrected nacelle anemometer cannot be used for a comparison of
the power curves of a turbine after rotor modifications have taken place. In fact in the figure, the phase3a
power curve lies well below the phase1 curve up to 10m/s which should lead to the wrong conclusion that
the modifications had a negative effect.
The main conclusions from the above results is that the relation between the mast and the nacelle
anemometer depends both on the position of the nacelle anemometer and the characteristics of the rotor.
The relation changes as soon as one or both of these factors are modified. Therefore it seems to be of less
importance where the cup anemometer is placed on the nacelle as long as the relation between the mast
and the nacelle anemometer is determined. The non corrected nacelle anemometer cannot be used to draw
any conclusions concerning the turbine’s power curve once modifications have taken place.
Relation between the mast and the nacelle anemometer (turbine in operation)
In figure 6.8, the relation between the mast and the nacelle anemometer is shown with the help of raw data
from phase1, their bin values and a polynomial curve fitted on the bin values. The relation is not linear
and the points have in this case been approximated by a 5th order polynomial curve. When trying to
establish the relation between the nacelle and the mast anemometer it is a common practice that the sector
used around the mast-turbine direction should be as narrow as possible (e.g. ±20°). In the present case,
since the terrain in the sector used is flat and homogeneous (water), a wider sector has been used due to
the lack of high wind speeds within a more narrow sector. This however has not taken place at the cost of
the accuracy of the correlation at low speeds as can be verified from the following figure 6.9, where the
bin data are presented for two different sectors (10° and 90° wide). The agreement is good at lower wind
speeds but becomes poorer at higher wind speeds due to the lack of an adequate number of data points in
the narrow sector.

In figure 6.10 the reproduction of the power curve as a function of the corrected nacelle anemometer (via
the 5th degree polynomial) for the case of the phase1 data is shown together with the mast power curve
(bin values). The data are from the sector 181.5° to 271.5°. The two power curves are as expected
practically identical. This means that, once the relation between the mast and the nacelle anemometer is
established, it is possible to use the nacelle anemometer instead of the mast anemometer and correctly
describe the turbine’s power curve. Similar relations as the one shown in figure 6.10 can be established for
all the data sets for the different measurement phases.
Relation between the mast and the nacelle anemometer (turbine stopped)
Next the relation between the mast and the nacelle anemometer is presented when the turbine is stopped.
WASP calculations, not presented here, have confirmed the absence of any site effects within the sector
181.5° to 271.5° and therefore any influence is due to the presence of the nacelle. The way the nacelle
presence influences the cup anemometer is really of not great importance as it is by “default” included in
the relation derived when the turbine is in operation.
The relation is presented in figure 6.11 and 6.12. In the case of figure 6.11, the data have been filtered
with respect to the position of the blades and the yaw error of the turbine. Thus the stopped turbine is
oriented towards the mast and one of the blades is always oriented around the vertical and downwards
With the exception of a few data points, the relation between the two anemometers is linear and the flow
above the nacelle at the anemometer position is accelerated. In figure 6.12 the raw data for the two heights
of the nacelle anemometer are shown. It is interesting to observe that in this case the acceleration of the
flow at 3m height is higher relative to the 1.2m. The acceleration of the flow explains why the relation of
the nacelle to the mast anemometer is sometimes, as e.g. in the work of Petersen S. (1994b), larger than
unity.
As the presence of the nacelle is seen to influence the response of the anemometer, it can be argued that
the relation mast-nacelle anemometer is influenced also by the shape of the nacelle and therefore the
relation derived for specific rotor settings and a specific nacelle is not necessarily valid for the same rotor
settings and a different nacelle shape.
The power curve as a function of the corrected nacelle anemometer (same rotor settings - different
sectors)
The conclusion from the above chapters is that once the relation between the nacelle and the mast
anemometer is established, then the nacelle anemometer can be used for the verification of the turbine’s
power curve within the same sector. When using the same sector then by default the same data are used
for the power curve as a function of the mast or the corrected nacelle anemometer and therefore the result
may be considered as biased. An extension of the above work is to show whether and under which
conditions the relation between the nacelle and the mast anemometer is valid for different data sets and for
different sectors. For that purpose the turbine’s bin power curves from the sector 181.5° to 271.5° (phase1
data) or 165° to 275°(phase3a data) will be compared to the bin power curves derived as a function of the
“corrected” nacelle anemometer from other sectors. The sectors used are shown in the following table
together with a sort description of the terrain roughness within the sector.

Table 6.1 Description of the terrain within the sectors used.

Sector Terrain description
181.5°→271.5° Fetch above land and sea
271.5°→330° Fetch above land, from 4D to 13D dominated by the power plant, coal banks, oil
tanks and cinder deposits
330°→360° Fetch above land, urban conditions
0°→90° Fetch above land and sea
100°→180° Fetch above land and sea
In figure 6.13, the power curve from the sector 181.5° to 271.5° is given as a function of the mast
anemometer and it is compared to the power curve from the sector 100° to 180°which is presented as a
function of the corrected nacelle anemometer. The surface roughness is the same for the two sectors and
the power curves are identical, the minor differences at higher wind speeds being attributed to the limited
number of data points in the corresponding bins.
The same good agreement is observed in the case the phase3a data shown in figure 6.14 where the power
curve from the sector 175° to 265° is given as a function of the mast anemometer and it is compared to the
power curve from three other sectors which are presented as a function of the corrected nacelle
anemometer. It is only the sector 330° to 360° which has a bigger surface roughness compared to the rest.
For this sector theoretical calculations, not presented here, show that minor differences to the design
power curve should be observed at around the nominal power.
Finally in figure 6.15, the power curve as a function of the mast anemometer from the sector 181.5° to
271.5° is compared to the power curve as a function of the corrected nacelle anemometer from the sector
271.5° to 330°. The last one is the sector where the turbine operates in the wake of the power plant and the
coal bunks. Unlike the previous cases, the two power curves do not coincide with the one from the sector
271.5° to 330° laying systematically below.
The reason for the lack of coincidence is seen in figure 6.16 where the 5th order polynomial line fitted to
the results from the sector 181.5° to 271.5° is presented together with raw data from the sector 271.5° to
330°. The correlation mast-nacelle anemometer is poor compared to the data of the sector 181.5° to
271.5°. This is expected since the mast anemometer is not within the 271.5° to 330° sector and it therefore
“sees” another part of the complex terrain in front, compare to the nacelle anemometer. That the relation
depends strongly on the wind direction is shown in figure 6.16 as considerable level differences in the
measured wind speeds. As a result it is concluded that the presence of the obstacles which generates
another terrain type modifies the wind speed profile relative to the flat terrain measurement sector and
thereby modifies the relation between the mast and the nacelle anemometer.
The wind shear through the rotor plane and for the two sectors, here defined as the difference in the speed
registered by the cup anemometers at 75m and 35m, is presented as a function of the anemometer at hub
height (55m) in figure 6.17. The slope is seen to be bigger in the case of the 271.5° to 330° sector. This
means that, within the rotor area, the speed deficit of the wind profile, with descending height, is higher in
the case of the sector 271.5° to 330° compared to the sector 181.5° to 271.5°. The flow within this sector
resembles the flow behind a fence where the wind speed suffers a deficit when compared to the inlet
profile. A result of this deficit is the reason for the lower energy produced by the turbine. We examine
now how much this difference in the power curves means to the annual energy production (AEP) of the
turbine.
The AEP and the related uncertainties for the two power curves are shown in table 6.3 (power curve as a

function of the mast anemometer) and table 6.4 (power curve as a function of the corrected nacelle
anemometer). The same sources of uncertainty have been considered for both power curves except for the
case a form factor of two table 6.4. There the second uncertainty column is calculated with an additional
1% calibration uncertainty. The mean AEP values, shown in table 6.2, agree with the mean AEP values
shown in table 6.3 within the stated uncertainties however the deviations are large.
Table 6.2. The annual energy production, power curve vs. the mast anemometer, phase1 data
AEP-measured and related uncertainty[MWh], sector 181.5° to 271.5°

Wind speed Form factor
#REF! 1.50 1.75 2.00 2.25 2.50
4 607.8 82.8 497.6 80.1 414.0 77.0 350.8 74.0 302.6 71.4
5 1120.4 106.0 997.8 107.1 893.5 106.8 808.0 105.6 738.6 104.1
6 1690.8 124.4 1594.5 129.9 1497.6 133.5 1410.8 135.7 1335.6 136.9
7 2254.9 137.3 2222.4 146.5 2160.8 153.6 2094.9 159.2 2032.7 163.6
8 2766.2 145.4 2827.4 157.3 2827.0 166.9 2800.9 174.9 2767.2 181.7
9 3197.9 149.5 3368.4 163.0 3451.5 174.3 3483.1 183.8 3489.9 191.9
10 3540.3 150.7 3819.0 164.9 3998.4 177.0 4104.7 187.2 4165.0 196.0
11 3795.7 149.7 4168.4 163.8 4443.5 176.0 4634.7 186.4 4762.1 195.3
Table 6.3. The annual energy production, power curve vs. the corrected nacelle anemometer, phase1 data
AEP-measured and related uncertainty [MWh], sector 271.5° to 330°
Wind speed Form factor
1.5 1.75 2 2.25 2.5
4 567.8 80.7 457.3 77.6 374.2 74.1 76.3 312.0 70.8 265.0 68.0
5 1065.9 103.7 940.3 104.7 833.4 104.0 108.2 746.1 102.4 675.6 100.4
6 1624.4 121.4 1524.8 127.5 1423.7 131.2 137.5 1332.6 133.2 1253.8 134.0
7 2177.4 133.0 2142.9 143.4 2078.0 151.4 159.3 2007.7 157.5 1940.6 161.9
8 2677.9 139.3 2738.1 152.4 2736.7 163.5 172.5 2708.5 172.9 2671.2 180.6
9 3099.3 141.7 3268.4 155.9 3352.2 168.6 178.1 3384.6 179.8 3391.3 189.6
10 3432.7 141.1 3708.2 155.5 3888.2 168.4 178.1 3997.0 180.1 4060.2 190.6
11 3680.5 138.6 4047.6 152.3 4321.5 164.6 174.1 4514.8 175.7 4646.3 185.8
These discrepancies in the power curves and the AEP, shown in figure 6.15 and the above tables would
certainly be smoothed out if a wider sector was chosen and it would thus be wrongly argued that the
nacelle-mast relation is still well applicable. Therefore if the verification of a power curve is wanted and
the relation of the mast-nacelle anemometer is known care should be taken to apply this relation in
mutually alike sectors.
The presence of the obstacles in the sector 271.5° to 330° simulates in many ways complex terrain
conditions for the turbine and the conclusion is that the relation between the mast and the nacelle
anemometer derived at one location and under the influence of specific site conditions is not transferable
at another location where different conditions prevail. The task of transferring the relation between
different terrain types is however not easy as procedures on this subject are not yet agreed. The second
important conclusion from figure 6.17 is that since the turbine power curve depends on the shape of the
wind profile, then in the presence of site effects the correct relation between the two anemometers at a hub

height is necessary but may be not sufficient for the description of the power curve under complex terrain
conditions as it does not give any information on the wind profile shape.
Power curve as a function of the corrected nacelle anemometer: (different rotor settings - same sector)
We now proceed to examine the effect of the changes of the rotor settings on the relation between the
nacelle and the mast anemometer for the three phases of the measurement program, for the same position
of the nacelle anemometer and for the same sector. Following this we will apply two different nacelle-
mast anemometer relations corresponding to two different rotor settings on the same data set.
The results of the relation nacelle-mast anemometer are shown in figure 6.18. By comparing the curves
from the first and the second measurement phase differences are observed. This means that the relation
between the mast and the nacelle anemometer depends not only on the rotor settings but also on the yaw
settings of the turbine. Once the settings have changed a new relation should be established. The
differences become even more pronounced between the phase1 and the phase3a data. We proceed now to
investigate to what extent these differences can affect the power curve presentation and the AEP since this
cannot be estimated from the figure.
In figure 6.19 the power curve from the phase3a data is presented as a function of the corrected nacelle
anemometer using both the relation derived during the phase1 (wrong) and the phase3a (correct)
measurements.
The use of the wrong relation produces a wrong power curve which in this case is overestimated but the
opposite might as well be true. The most important observation is that the differences between the two
power curves are in this case systematic, which means that a bias error will be introduced in the AEP
calculations which cannot be in any way detected. Therefore if a derived relation, between the mast and
the nacelle anemometer in one turbine, should be used in another turbine of the same type and make , care
should be taken that the rotor settings, the yaw error and the placement of the nacelle anemometer remain
unchanged.
6.2.2.4 Conclusions
The power curves measured as a function of the nacelle anemometer were compared with the power
curves measured as a function of the mast anemometer for the same wind turbine settings. It was found
generally, that with the nacelle anemometer in the blade cylinder wake region there was very high wind
speed deficit in the max. Cp region, whereas this deficit was reduced at low and high wind speeds. With
the nacelle anemometer in the profiled blade wake region the wind speed deficit seemed much more
constant for medium to high wind speeds. Thus the response of the nacelle anemometer is a function of its
location on top of the nacelle. From the comparison it can also be concluded that the modifications on the
wind turbine rotor settings, which results in changes of the power curves seen as a function of the mast
anemometer, does also change the power curves seen as a function of the nacelle anemometer, but not in a
consistent way. This is due to the three dimensional nature of the flow through the rotor and the fact that
the wind speed registered by the cup anemometer is of local nature and only representative of that
measurement point. In other words, it was found that the nacelle anemometer power curves cannot be used
to draw any conclusions concerning the changes of the power curve of the wind turbine due to the
modifications made. The main conclusions is thus, that the power curves, using the nacelle anemometer,
are dependent on both the position of the nacelle anemometer on the nacelle and the characteristics of the
rotor.
The relations between the mast anemometer and the nacelle anemometer were not linear in this study, but
they could be fitted well with a 5th order polynomial curve in the case where the cup anemometer is behind
the non profiled part of the blade. For the same wind turbine settings a very good agreement was found
between the measured power curve (phase 1 data) measured from an open sea sector 181.5° to 271.5° and
a power curve, derived from combining the relation between the mast and nacelle anemometers and the

measured power curve using the nacelle anemometer with data from same sector. The same very good
agreement was found when combing the relation between the mast and nacelle anemometers and
measured power curves from another open sea sector (100° to 180°). The same very good agreement was
also found with power curve data from phase 3a (175° to 265°) when combining the relation between the
mast and nacelle anemometers and measured power curves from other sectors: a mix of sea and flat land
0° to 90° (low roughness) and an industrial area 330° to 360° (high roughness). On this basis it can be
concluded that the nacelle anemometer can be used to verify the same power curve with winds from
different terrain roughness’. It was not possible, though, to find good agreement when using wind data
(phase 1) from the sector in the wake of a huge power plant (271.5° to 330°). The cause of this is found to
be due to the disturbed wind profile which results from the presence of site effects.
For the same measurement sector (about 180° to about 270°) and different wind turbine settings (phase 1,
2 and 3) the relation between the mast anemometer and the nacelle anemometer was quite varying.
Therefore, if the wind turbine settings are changed not only the power output changes, but also the nacelle
anemometer readings are changed.
With the wind turbine stopped it was found that the nacelle anemometer measured 20-30% higher wind
speeds than the mast anemometer, which means that the nacelle presence influences the flow around the
nacelle. The wind speed 3m above the roof was higher than at 1.2m. This shows that the induced flow
around the nacelle extends quite far from the nacelle, which indicates that the nacelle design has a
substantial influence on the reading of the nacelle anemometer.
In conclusion, it can be said that provided the following factors remain unchanged:
• the wind turbine rotor settings
• the wind turbine’s yaw error
• the anemometer’s position on the nacelle
• the terrain is flat
then the relation mast-nacelle anemometer can be can be transferred between wind turbines of the same
make and type. Under these conditions the “corrected” nacelle anemometer can be used for the
verification of the turbine’s power curve and the AEP.
Whether and under which conditions this relation is transferable from one kind of terrain to another (e.g.
from flat to complex) remains to be investigated as the lack of data on this subject prevents us from
drawing any conclusions.

30 1400
Phase1, pitch angle at +0.5, no VG Phase1, pitch angle at +0.5, no VG
Phase2, pitch angle at +0.5, new yaw condt., no VG Phase2, pitch angle at +0.5, new yaw condt., no VG
Phase3a, pitch angle at -1.0, new yaw condt., with VG 1200 Phase3a, pitch angle at -1.0, new yaw condt., with VG
Design power curve
20
Electrical power (1.225 kg/m^3) (kW)

1000
800
Yaw error (deg.)
10
600
0
400
200
-10
-20 -200
0 5 10 15 20 25 0 5 10 15 20 25
Wind speed at 55m. height (m/s) Wind speed at 55m. height (m/s)
Figure 6.1 The yaw error for the three power Figure 6.2 Power curves for the three phases
curves. against design curve.
1200 phase1, pitch angle at +0.5°, no VG

data sector: 181.5°- 271.5°
0.6 mast anemometer at 55m. height

Electrical power (1.225kg/m^3)(kw)
Phase1, pitch angle at +0.5, no VG Nacelle anemometer at 1.2m height

Phase2, pitch angle at +0.5, new yaw condt., no VG
Phase3a, pitch angle at -1.0, new yaw condt., with VG
0.5
800
0.4
Cp (1.225kg/m^3)
0.3
400
0.2
0.1
0
0
0 5 10 15 20 25
Wind speed at 55m. height (m/s) file : d:el1000/powper/graph1.grf
0 5 10 15 20 25
Wind speed (m/s)
Figure 6.3 The power coefficients for the three Figure 6.4 Phase1, the power curve as a function
phases. of the mast and the non-corrected nacelle
anemometer

1200 1200
phase3a, pitch angle at -1°, VG present file : d:el1000:powper:graph33.grf
data sector: 175°- 265°
mast anemometer at 55m. height
Electrical power (1.225kg/m^3)(kW)

nacelle anemometer at 1.2m height
800 800
400 400
phase3b, pitch: -1.0°, VG present

sector: 175°-265°
0 0
mast anemometer at 55m. height
nacelle anemometer at 3m. height

file : d:el1000/powper/graph34.grf
0 5 10 15 20 25 0 5 10 15 20 25
Wind speed (m/s) Wind speed (m/s)
Figure 6.5 Phase3a, the power curve as a function Figure 6.6 Phase3b, the power curve as a function
of the mast and the non-corrected nacelle of the mast and the non-corrected nacelle
anemometer. anemometer.
25
Wind speed at 55m. height (mast anem.) (m/s)
1200 phase3a: VG present, pitch angle -1°, new yaw condition

phase1, data sector: 181.5°- 271.5°
phase1: no VG present,pitch angle +0.5
Bin values
Fit 2: 5th order polynomial

20
800
15
400 10
5
0
file : d:/el1000/powper/graph3.grf
file : d:el1000/powper/graph36.grf 0
0 5 10 15 20 25
0 5 10 15 20 25
Wind speed (m/s) Wind speed (nacelle anem.) (m/s)
Figure 6.7 The power curve as a function of the Figure 6.8 Relation between the nacelle and the
nacelle anemometer at 1.2m height above the roof. mast anemometers for the phase1 data.

25
1200 phase1, data sector :181.5°- 271.5
phase1 data power curve using the corrected nacelle anemometer
power curve using the mast anemometer

10° sector, 261.5° to 271.5°
20
90° sector, 181.5° to 271.5°
800
Electrical power (kW)

15
10 400
file : d:/el1000/powper/graph21.grf
0 file : d:/el1000/powper/graph4.grf
0 5 10 15 20 25
0 5 10 15 20 25
Wind speed (nacelle anem.) (m/s)
Wind speed (m/s)
Figure 6.9 Relation between the mast and the Figure 6.10 Bin averaged power curve as a
nacelle anemometer for the phase1 data (two function of the mast and nacelle corrected
different sectors). anemometer (phase 1).
25 25
phase1 data, turbine stopped turbine stopped
Wind speed (mast anemometer) (m/s)
non filtered data phase1, nacelle anemometer at 1.2m height

filtered data
phase3a, nacelle anemometer at 3m. height
20 20
15 15
10 10
5 5
file : d:el1000:powper:graph27.grf file : d:el1000:powper:graph32.grf

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Wind speed (nacelle anemometer) (m/s) Wind speed (nacelle anemometer) (m/s)
Figure 6.11 Relation between the mast and the Figure 6.12 Relation between the mast and the
nacelle anemometer at 1.2m height with the nacelle anemometer at 1.2m and 3m height,
turbine stopped turbine stopped

1000 1200
phase3a data, pitch angle at -1.0°, new yaw cond.
phase1 :pitch : +0.5°, v.g. NOT on the blades VG present, nacelle anem. at 1.2m height
Electrical power (1.225kg/m^3) (kW)

data sector 181.5°-271.5° mast anem., sector 175°- 265°
800 data sector 100°-180° corrected nacelle anem., sector 0°- 90°
corrected nacelle anem., sector 330°- 360°

corrected nacelle anem. sector 100°-180°

800
600
400
400
200
file:d:/el1000/powper/graph35.grf
file : d:el1000:powper:graph20.grf
0 0
0 5 10 15 5 10 15
Figure 6.13 Power curve as a function of the mast Figure 6.14 Power curve as a function of the mast
and the corrected nacelle anemometer (two and the corrected nacelle anemometer (four
sectors). sectors).
1200 25
phase1 :pitch : +0.5°, v.g. NOT on the blades phase1 data,
data values , sector 271.5° to 330°

data sector 181.5°-271.5°
Fit 2: 5th order polynomial, sector 181.5° to 271.5°
data sector 271.5°-330° 20
800
15
10
400
file : d:el1000:powper:graph19.grf file : d:/el1000/powper/graph37.grf

0 0
5.00 10.00 15.00 0 5 10 15 20 25
Wind speed (m/s) Wind speed (nacelle anem.) (m/s)
Figure 6.15 Power curve as a function of the mast Figure 6.16 Comparison between the nacelle and
and the corrected nacelle anemometer (two the mast anemometer (phase 1, two sectors).
sectors).

5 20
phase1 data : phase1, pitch: +0.5°, no VG, sector: 181.5°- 271.5°
phase2, pitch: +0.5°, no VG,new yaw, sector: 181.5°- 271.5°

sector: 181.5°- 271.5°
phase3a, pitch: -1.0°, with VG,new yaw, sector: 175°- 265°
Fit 1: Linear, Y=0.121*X-0.0705
Wind shear (cup75-cup35) (m/s)
4 sector: 271.5° - 331.5°

15
Fit 2: Linear, Y=0.191*X+0.512
10
5
1
d:el1000:powper:graph11.grf file . d:el1000:powper:graph16.grf

0 0
0 5 10 15 20 0 5 10 15 20
Wind speed (mast anem.) (m/s) Wind speed (nacelle) (m/s)
Figure 6.17 The wind shear for the two sectors Figure 6.18 Relation between the nacelle and the
(phase1). mast anemometer for three different rotor settings
1200
Electrical power data from phase3a (175°- 265°)
(phase5: pitch -1.0°, v.g. on the blades)
nacelle derived from phase1 data
polynomial derived from phase3a
800
400
file : d:el1000:powper:graph17.grf
5 10 15 20
Wind speed (nacelle anem.) (m/s)
Figure 6.19 Power curve of the turbine when

using different corrections for the nacelle
anemometer (the stall points are not confirmed by
later results).

6.2.3 Nacelle cup anemometry on 500kW wind turbine
Experimental research was performed on a 500 kW active stall controlled wind turbine for the assessment
of power performance evaluation through the utilization of nacelle cup anemometry. Details on the
experimental setup and analysis procedures are found in Hinsch C. (1996b).
6.2.3.2 Calibration of the nacelle anemometer
The correlation of the wind speed measured on the nacelle and meteorological mast was approximated by
linear formula (figure 6.20). The effects of the nacelle induced boundary layer and flow structure behind
the running rotor are approximately canceled out, for the selected location of the nacelle cup. In figure
6.21 the regression of the wind speed ratio, that is the wind speed measured at reference mast over the
nacelle value is presented. Differences that exceed 10% of the reference values are encountered. The
correlation of the reference and nacelle measured wind speeds in the case where the wind turbine is parked
is presented in figure 6.22. It is seen that the nacelle cup measurements are influenced by the nacelle
induced boundary layer and lower wind speed values are recorded.
6.2.3.3 Comparison of power curves

In figures 6.23 and 6.24 the power curve data are presented as function of the reference wind speed and
the nacelle cup corrected values, respectively. The comparison of the power curves, by means of bin
averaged values, is presented in figure 6.25. In figure 6.26 the relationship between power curves
measured with corrected nacelle wind speed and with the reference wind speed, for the 240°-340° wind
direction sector, is presented. The differences between the power values are less than 7% with the higher
values at low wind speeds. The site calibration results, used for determining the reference wind speed, are
presented in figure 6.27.
The assessment of the feasibility of the method is attained by comparing the different power curves
measured with corrected nacelle wind speed and with the reference wind speed for two different wind
direction sectors, namely 190°-210° and 240° - 340°. The above sectors present significantly different
terrain effects, as these were captured by the site calibration process (figure 6.27), while the former sector
was taken as the reference sector. The results of the comparison are presented in figure 6.28. The use of
nacelle cup measurements decreases the encountered differences from 20% down to 7% of the reference
power curve values.
6.2.3.4 Conclusions
The experimental research on nacelle anemometry as it was applied on a 500kW wind turbine, resulted in
the following:
• lower scatter of the measured power curve is attained by the use of the nacelle cup measurements; this
results to clear capture of the switching point between generators
• the use of nacelle cup measurements may decrease significantly the error induced by site calibration,
in cases that large discrepancies are encountered between the measured and reference wind speed.

20
18
16
y = 1.000004x + 0.000498
R2 = 0.979718
wind speed nacelle [m/s]
14
12
10
0
0 2 4 6 8 10 12 14 16 18 20
wind speed metmast [m/s]
Figure 6.20 Correlation between the anemometer on the metmast and

the nacelle.
20
18
y = 0.9663x - 0.061
16
R2 = 0.9808
wind speed nacelle [m/s]
14
12
10
0
0 2 4 6 8 10 12 14 16 18 20
Figure 6.21 Correlation between the anemometer on the met. mast and
the nacelle (WEC stopped).

1.5
1.4
1.3
1.2
v-mast/v-nacelle
1.1
0.9
0.8
y = -0.0012x + 1.0131
0.7
R2 = 0.0039
0.6
0.5
0 2 4 6 8 10 12 14 16 18 20
Figure 6.22 Relationship between wind speed on the nacelle and the
wind speed on the met. mast.
1.2
0.8
P/Prated
0.6
0.4
0.2
0
0 2 4 6 8 10 12 14 16 18 20
Figure 6.23 Power curve with anemometer on measuring mast

1.2
0.8
P/Prated
0.6
0.4
0.2
0
0 2 4 6 8 10 12 14 16 18 20
wind speed (nacelle) [m/s]
Figure 6.24 Power curve with anemometer on nacelle
1.2
0.8
P/Prated
0.6
0.4
0.2
0
0 2 4 6 8 10 12 14 16 18 20
wind speed [m/s]
Figure 6.25 Bin-averaged power curves

1.2
1.15
240°-340° wind nacelle
1.1
1.05
0.95
Pmet/Pnacelle
0.9
0.85
0.8
0 2 4 6 8 10 12 14 16
Wind speed [m/s]
Figure 6.26 Relationship between different power curves measured with

wind speed on the nacelle and the wind speed on the metmast
1.2
1.15
1.1
1.05
vmet/vref
0.95
0.9
0.85
0.8
50 100 150 200 250 300 350
direction [deg]
Figure 6.27 Result of the site calibration (wind speed 5-10m/s).

1.2
1.15
1.1
P(240°-340°)/Po(190°-210°)
1.05
0.95
0.9
metmast
nacelle
0.85
0.8
0 2 4 6 8 10 12 14 16
wind speed [m/s]
Figure 6.28 Relationship between different power curves measured

with wind speed on the nacelle and the wind speed on the met. mast for
sector 190°-210° compared to the measured power curve from 240° -
340°.

6.2.4 Nacelle cup anemometry on 110kW wind turbine

The experimental research performed on WINCON 110XT wind turbine aimed to the assessment of the
nacelle anemometer method for power performance purposes.
A cup anemometer was placed on the nacelle according to the practice found in Pedersen T. (1994). The
horizontal distance of the cup from the blade root was 2.5 times the blade root diameter while the vertical
distance from the nacelle was set to 1m. The experiment lasted one month (from 1/10 to 30/10). Details
about the experimental setup and analysis procedures are found in Morfiadakis E. (1996a). In parallel with
the nacelle cup measurements the rotor disk experiment at the test station was running, thus giving
detailed information about the wind structure. The cup and vane resolver measurements from the central
mast at low, hub and top position were used within the present analysis.
In figure 6.29 the measured data from the reference mast are presented by means of wind speed statistics,
including the third and forth order statistics, the length scale as well as the wind shear parameters (referred
as gust symmetric and asymmetric factors, Bergstrom H., 1995). In figure 6.30 the nacelle cup
measurements are presented against the reference mast measurements. Finally iIn figure 6.31 the power
measurements are given along with the operational characteristics of the machine.
6.2.4.2 Estimation of nacelle cup correlation and assessment

In order to assess the performance of nacelle anemometer technique the whole data set was divided to two
wind direction sectors, i.e. 315°-360° and 0°-45°. Moreover, the correlation of nacelle cup measurement
against the reference cup were estimated for both sectors. Finally, the power curves for each sector using
the corrected nacelle cup measurements from the correlation provided from the other sector measurements
are drawn. The results of the procedure are shown in figure 6.32.
The assessment of the site calibration technique was performed via annual energy production estimations.
The AEP estimations are made for the power curves calculated from the following data sets:
• total data set (i.e. sector 315°-45°)
• western directions (i.e. sector 315°-360°), reference cup measurements
• eastern directions (i.e. sector 0°-45°), reference cup measurements
• western directions (i.e. sector 315°-360°), nacelle cup (correlation from eastern directions)
• eastern directions (i.e. sector 0°-45°), nacelle cup (correlation from western directions)
The results are given in table 6.4
Table 6.4 AEP estimations for different wind direction sectors using nacelle cup calibration.
AEP estimation [MWh]
(Rayleigh distribution, power curve 4-14m/s)
Method Sector Umean=6m/s Umean=8m/s Umean=10m/s
Ref. wind speed 315°-360° 184.1 260.5 271.1
(reference case)
Ref. wind speed 0°-45° 184.8 260.8 271.3
Ref. wind speed 315°-45° 184.3 260.4 270.9
Nacelle cup
(correlation from data 315°-360° 187.7 264.3 273.9

within 0°-45° sector)

Nacelle cup
(correlation from data 0°-45° 180.9 254.9 264.9
within 315°-360°
sector)
It is seen that although the AEP estimations are almost identical for the three cases based on the reference
mast measurements in the case of nacelle cup based estimations differences are depicted. The differences,
that are within the uncertainty of measurement procedure, are attributed to the statistical error induced to
the power curve calculations from the nacelle cup correlation as well as to the sensitivity of the correlation
to parameters that are not seen by the wind turbine as far as the mean power output is regarded.
In order to depict the effect of the influencing parameters a parameter identification was performed for the
nacelle cup wind speed against the reference wind speed and some of the primary wind and wind turbine
operational characteristics (details of the method are found in paragraph 7.2). The identification regarded
the mean wind speed as well as its standard deviation and the results by means of the dependence
coefficients are given in figure 6.33. The dependence coefficients represent the percent change of the
dependent parameter (in this case the nacelle cup readings) when each one of the causative wind and wind
turbine parameters is increased by one standard deviation of the parameter variability within the
regression domain. The effect of the wind turbulent magnitudes (for the lower wind speed ranges) and the
wind shear factors and the yaw error (for all wind speed ranges) is captured, though limited. On the other
hand, if a nacelle cup calibration formula is to be applied in a site that compared to the site where the
calibration took place presents different wind structure and consequent wind turbine response, then the
dependencies on the secondary parameters may act in an additive way resulting in higher uncertainty in
estimating the power curve.
6.2.4.3 Conclusions
The experimental research, performed on a 110kW stall regulated turbine, resulted to the following
conclusions:
• the introduction of the nacelle cup calibration formula introduces a statistical error that should be
taken into account in the uncertainty estimation
• the dependency of the calibration formula to the wind and wind turbine parameters was found to be
limited; on the other hand these effects are expected to be magnified in cases where the wind structure
and wind turbine response is quite different in the calibration and the testing site.

WINCON 110XT AT CRES TEST STATION

Nacelle cup measurements
0.25
120
Turbulence intensity at hub - ref. mast
Length scale at hub - ref. mast (m)
100
0.20
80
0.15
60
0.10
40
0.05
20
0.00
0
3 6 9 12 15 3 6 9 12 15
Wind speed at hub - ref. mast (m/s) Wind speed at hub - ref. mast (m/s)
1.0
4.0
Wind speed skewness at hub - ref. mast
Wind speed kurtosis at hub - ref. mast

0.5
3.0
0.0
2.0
-0.5
1.0
-1.0
0.0
-1.0
-1.5
-2.0
-2.0
3 6 9 12 15 3 6 9 12 15
0.10
0.00
Gust antisymmetric factor
-0.05
Gust symmetric factor
0.05
-0.10
0.00
-0.15
-0.05
-0.20
3.0 6.0 9.0 12.0 15.0 3 6 9 12 15

Figure 6.29 Wind statistics from reference mast (wind direction sector 315°-45°).


0.25
24
Turbulence intensity at nacelle

Wind speed at nacelle (m/s)
20
0.20
16
0.15
12
0.10
8
0.05
4
0.00
0
3 6 9 12 15 3 6 9 12 15 18
Wind speed at hub - ref. mast (m/s) Wind speed at nacelle (m/s)
3.0
2.0
SDV wind direction/T.I. at hub - ref. mast
SDV wind speed at nacelle (m/s)
2.5
2.0
1.5
1.5
1.0
1.0
0.5
0.0
0.5
0.0 0.5 1.0 1.5 2.0 2.5 3 6 9 12 15

SDV wind speed at hub - ref. mast (m/s) Wind speed at hub - ref. mast (m/s)
1.0
2.0
Wind speed skewness at nacelle
Wind speed kurtosis at nacelle
1.5
0.5
1.0
0.0
0.5
0.0
-0.5
-1.5 -1.0 -0.5

-1.0
-1.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Wind speed skewness at hub - ref. mast Wind speed kurtosis at hub - ref. mast
Figure 6.30 Nacelle cup measurements (wind direction sector 315°-45°).


80 100 120 140
30
SDV power output (kW)
Power output (kW)
20
60
40
10
20
0
-20
0
3 6 9 12 15 3 6 9 12 15
1030
10
1020
0
Yawerror (deg)
1010
RPM
-10
1000
-20
990
980
-30
3.0 6.0 9.0 12.0 15.0 3 6 9 12 15

100
Normalised power output (kW)
80
60
40
20
0
-20
3.0 6.0 9.0 12.0 15.0

Wind speed at hub - ref. mast (m/s)
Figure 6.31 Performance measurements for WINCON 110XT WT (wind direction sector 315°-45°).


Wind direction sector 315-360o Wind direction sector 0-45o
100
100

80
80
60
60
40
40
20
20
0
0
-20
-20
3 6 9 12 15 3 6 9 12 15

15
15
Uref=-0.3248+0.8996*Unac Uref=-0.5501+0.9324*Unac
12
12
9
9
6
6
3
3 6 9 12 15 18 3 6 9 12 15 18
Wind speed at nacelle (m/s) Wind speed at nacelle (m/s)

100
100

80
80
60
60
40
40
20
20
0
0
-20
-20
3.0 6.0 9.0 12.0 15.0 3 6 9 12 15

Corrected wind speed at hub (m/s) Corrected wind speed at hub (m/s)
Figure 6.32 Assessment of nacelle cup anemometry for W110XT power curves.

Nacelle cup reading SDV nacelle cup reading

0.9 2.0
U=5-9m/s U=5-9m/s
1.6 U=9-15m/s
U=9-15m/s
0.6
1.2
Dependence coefficient
0.8
0.3
0.4
0.0
0.0
-0.4
-0.3 -0.8
-1.2
-0.6 -1.6
U sk
Vu
Lu
ye
Vye
U ku
m
sy m
rp m
U sk
V ye
U ku
Lu
ye
m
sy m
rp m
m
Vw
Vw
V rp
G sy
V rp
G sy
SD
SD
Ga
SD
Ga
SD
SD
SD
SD
Figure 6.33 Dependency of nacelle cup wind speed and standard deviation on secondary wind and
wind turbine operational parameters.
6.3 Upwind Rotor Measurements

6.3.1 Introduction
A large scale experiment was set at CRES test station allowing for assessing the use of the upwind rotor
mean wind speed as the reference wind speed. Three 40m towers, forming a straight line, were erected
upwind to the WINCON 110XT wind turbine and equipped as follows:
• four couples of cup anemometers and vane resolvers mounted at the two side towers above and below
the hub height, at a radius distance of 9.4m from the hub anemometer of the central tower; the
distance corresponds to a 85% of the blade length
• one cup mounted at hub height, on the central mast
6.3.2 Rotor disk measurements on a 110kW wind turbine

The rotor mean wind speed is estimated as the average of the five cup readings. The ratio of the rotor
mean value over the hub height value is presented in figure 6.34. Due to the fact that on top of the hill,
where the meteorological masts and the turbine are erected, the wind shear is rather flat, the resulting
ratios are close to unity. The assessment of the method is attained by comparing the estimated annual
energy production for the two cases, namely using the rotor mean and the hub height wind speed values;
the latter comprises the reference case. The results are given in the following table:

AEP estimation with rotor mean reference wind speed

Mean AEP (Uncertainty) [Difference from reference] [MWh]
annual
[m/s] 330°-30° 350°-10° 10°-30°
5 117,3 (8,57) [1,1] 116,3 (8,54) [1,0] 118,6 (8,57) [1,4]
7 271,8 (14,70) [1,8] 270,6 (14,67) [1,1] 273,6 (14,83) [2,3]
9 398,4 (16,79) [2,0] 397,2 (16,75) [2,1] 400,4 (17,20) [3,0]
11 466,5 (16,54) [2,0] 465,4 (16,47) [2,4] 468,3 (17,18) [3,1]
6.3.3 Conclusions
The experimental research on the utilization of the rotor mean wind speed as the reference wind speed for
defining the power curve resulted to the following conclusions:
• the procedure is feasible for experimental research purposes; the complexity of the experimental setup
makes the procedure unsuitable for large scale power performance evaluations
• in cases where flat wind shear shapes are found the utilization of the rotor mean wind speed does not
necessarily results to higher accuracy compared to the estimations based on the hub height
measurements; the differences encountered in the present application are significantly smaller than the
estimated uncertainty
• the size of the tested wind turbine is one of the parameters that affect the feasibility of the method;
larger rotors may present higher sensitivity to wind shear
WT110XT AT C.R.E.S SITE

1.03
10min average
1.02
Wind speed ratio (Rotor mean)/(Hub)
1.01
1.00
0.99
0.98
-30.0 -20.0 -10.0 0.0 10.0 20.0 30.0
Figure 6.34 Wind speed ratio (rotor mean over reference at hub
height) vs wind direction.

6.4 Conclusions
Extensive experimental research carried out on three wind turbines of different size, ranging from 110kW
up to 1MW, in order to clarify the feasibility and applicability of the utilization of the nacelle cup
anemometer for power performance evaluation purposes. Although the research on the topic is continuing,
at this stage several critical conclusions regarding the application field of the method are drawn as
follows:
• the cup anemometer is preferably installed behind the non profiled part of the rotor blade
• in several cases the calibration formula was found to be linear; for the large machine a 5th order
polynomial curve was used in order to account for the pronounced wind speed deficit in wind speed
range corresponding to higher values of power coefficient
• the introduction of the calibration formula should be accompanied by a proper uncertainty estimation,
based on the fitting statistics
• the use of nacelle cup measurements may decrease significantly the error induced by site calibration,
in cases where large discrepancies are encountered between the measured and reference wind speed
• lower scatter of the measured power curve is attained by the use of the nacelle cup measurements; in
cases where two generators are used the switching point is clearly recognizable
• the nacelle cup calibration formula is strongly dependent on the rotor settings and the position of the
nacelle cup anemometer; these parameters should be unchanged when transferring the calibration
formula
• the calibration formula can be used for sites where the topography is the same with the calibration site,
allowing only for surface roughness deviations; on the other hand wind shear changes may introduce
significant effect on the calibration formula
• the dependency of the calibration formula to the wind (wind shear and turbulence) and wind turbine
parameters (rotational speed, yaw error) was identified for a 110kW turbine. The quantified effects
were found to be of limited magnitude, yet these effects are expected to be magnified in cases where
the wind structure and wind turbine response is quite different in the calibration and the testing site
• the nacelle anemometer power curves cannot be used to draw any conclusions concerning the changes
of the power curve of the wind turbine due to any modifications
• the transfer of the calibration formula for verification purposes presupposes that wind turbine rotor
settings, control operation, anemometer position on the nacelle as well as terrain flatness are kept
strictly unchanged
• the application of the nacelle cup anemometry in complex terrain proved successful. On the other
hand the transfer of the calibration formula to complex terrain, either from flat or from complex
terrain, may introduce significant uncertainty on the power curve determination. The same holds in
case where the method is to be applied on a wind turbine operating within wakes (wind farm case).
The utilization of rotor mean wind speed as the reference wind speed is feasible only for experimental
research purposes due to the complexity of the experimental setup. In cases where flat wind shear shapes
are found, and depending on rotor size, the method does not necessarily result to higher accuracy
compared to the estimations based on the hub height measurements.

7. PARAMETER IDENTIFICATION OF WIND TURBINE

POWER PERFORMANCE
7.1 Introduction
Parameter identification of wind turbine power performance regards the quantification of the dependency
of power characteristics, namely mean value and standard deviation, on the wind deterministic and
stochastic characteristics.
The scope of performing a parameter identification procedure on different wind turbines operating at
different sites is the clarification of the following issues:
• identification of the wind parameters that affect power performance and quantification of their effect
• assessment of the sensitivity of power characteristics of wind turbines of different size and control
system
• assessment of the sensitivity of power characteristics of the same wind turbine type when operated in
different sites
• assessment of the causative parameters in relation to complex terrain characteristics
• assessment of power curve normalizing procedure based on parameter identification results
7.2 Parameter Identification Procedure On Experimental Data Bases

The scope of a parameter identification procedure is to identify the statistically significant parameters and
to quantify their composed effect. Multivariate regression analysis methods present significant advantages,
under the restriction of the availability of adequate amount of data, when the goal is to capture the effect
of different parameters that are not supposed and most importantly do not need to be uncorrelated. The
topics that are discussed herein are related to the regression analysis application and assessment as well as
to the parameter identification procedure. Detailed description of the method and sample applications are
found in Kleinbaum K. (1985), Morfiadakis E. (1996e), Mouzakis F. (1996).
7.2.1 Multivariate regression analysis

The used multivariate regression analysis is based on least square fitting process as a maximum likehood
estimator of the fitted parameters. It is assumed that the measurement errors are independent and normally
distributed with constant standard deviation.
The scope of the regression is the estimation of the coefficients (ak,k=1,M)1, for the expression of the
dependent variable (y)2 as follows:
M
y( x i ) = ∑a k X k ( x i ) + E( x i ) , i = 1, N
k =1 (1)
where Xk(xi) 4is the value of the kth independent variable at point xi5 and E(xi) 6is the associated error at
the same point.
Chi-square fitting is applied as the weighted least square fitting process, in which the magnitude χ27
defined as:

2
N
 y i - y i  M
χ 2
= ∑ 
 σi 
 , y i = ∑a k X k(xi)
i=1 k =1
(2)
is minimized. The magnitude σi8 presents the measurement standard error at point (y(xi),xi)9. If the
measurement errors are not known (as in the present practice) they are all set equal to unity. The number
(n=N-M)10 represents the fitting degrees of freedom.
The minimum of χ211 12occurs where its derivatives with respect to parameters (ak,k=1,M) 13vanish. The
resulting linear equations are solved by the Gauss-Jordan elimination technique, yielding the coefficients
(aj, j=1,M)14:
-1
M
 N yi X k ( x i )   N X j ( xi ) X k ( xi ) 
aj = ∑ C jk  ∑ σ i2 
 , C jk = ∑
σ i2
 , j = 1, M
k=1  i=1  i=1  jk (3)
where the standard deviation related to the estimate of aj is given by the equation:
 ∑ σ  =
2
N
 ∂ a j
σ( a j ) = 2
  C jj
 i=1
i
 ∂ yi   (4)
7.2.2 Assessment of the accuracy of the multivariate regression

For the assessment of the accuracy of the regression analysis, the following magnitudes were considered:
a) total sum of squares SSY, sum of squares due to error SSE, sample squared correlation coefficient
R2 as well as fitting standard deviation σy16, defined as follows:
N
N N ∑ ( y - y
i i
)2
SSY = ∑ ( yi - y mean )2 SSE = ∑ ( yi - y i )2 R2 = SSY - SSE σ y = i=1
i=1 , i=1 , SSY , N-M

(5)
b) F-test statistic, describing the hypothesis “H0 : there is no significant overall regression using the
M independent variables”, is defined as follows:
SSY - SSE  N - M 
F =  
SSE  M -1  (6)
Consequently the p-value for the above defined F-test is calculated from Fisher’s F sampling
distribution with M-1 and N-M degrees of freedom (pF/M-1,N-M17).
c) The t-test statistic, describing the hypothesis “H0 : the regression coefficient ak18 is zero”, is
defined as t=ak/σ(ak) and consequently the p-value for the above defined t-test is calculated from
Student’s t sampling distribution with N-M degrees of freedom (pt/N-M19).
d) For the quantitative assessment of the relation of the dependent variable on each of the
independent parameters Xk, various magnitudes were considered, namely the regression
coefficients ak themselves, the relative per sigma dependence coefficient Sk or the t-test statistic
for each parameter. The relative per sigma dependence or dependence coefficient Sk, is defined as:

∑ (X )
2
k ( xi ) - X k
σ Xk i=1
S k = ak , σ Xk =
y mean N -1 (7)
The magnitude Sk represents the relative change of the dependent variable induced by the increase of the
value of parameter Xk by its standard deviation σXt, within the regression domain22. Assuming that the
probability of that change is expected to be comparable for all independent variables, the relativistic
assessment of the effect of each variable Xk on y can be attained. Although the above assumption is not
valid in cases where the distribution of Xk, within the regression domain, differs significantly from
normal, still the magnitude Sk offers a valuable means for weighting the dependence of each Xk on the
dependent magnitude.
Another option for assessing the regression results is by means of the beta coefficients, defined as:
βk = a k σ Xk
Beta coefficients are the regression coefficients for the transformed variables in normal distribution (zero
mean and unity standard deviation), and are linearly dependent to dependence coefficients.
7.2.3 Parameter identification procedure

The introduced procedure comprises the following phases:
I.Identification of potential causative processes: The deterministic wind characteristics, wind turbulence
and the wind speed distribution are regarded as the main causative processes as far as wind turbine power
performance is considered.
II. Determination of candidate independent variables for each process: Upon recognition of the potential
causative processes, a set of independent variables are selected in order to describe each process. The term
independent is used according to the regression analysis nomenclature and does not imply that the
predictive variables should be uncorrelated. For the present application, for instance, the deterministic
wind characteristics are described by the mean wind speed, the air density, the shape of the wind shear and
the wind mean inclination.
III. Application of the full model: The regression analysis is performed for the selected full model as
described in the above paragraphs.
IV. Iterative application of evaluation criteria and model reforming: The evaluation criterion is based on
the p-values of the t-ratios of each independent variable. Large values for the p-value indicate that the
rejection of the hypothesis “H0: the regression coefficient ak23 is zero” is questionable, and consequently
the parameter is rejected from the model and the regression analysis is performed again. The rejection
threshold, set to 5%, is defined after numerical experimentation.
V. Final selection and assessment: After the completion of the iterative process, the assessment of the
regression findings is attained through descriptive magnitudes namely the dependence coefficients.
7.3 Identification And Quantification Of Site Related Effects On Power

Performance
The parameter identification procedure, as described in paragraph 7.2 has been applied to the following
cases:
• 225kW pitch regulated

• 110kW stall regulated

In the case of the NTK500/37 the procedure has been applied for all the campaigns that have been
performed by the participants. The results of the procedure are assessed either by the dependence
coefficients or directly by the sensitivity factors of the regression formula.
Moreover, investigation on the sensitivity of power performance on turbulence, wind shear and
atmospheric stability has been performed on a 3MW wind turbine.
7.3.1 Parameter identification on a 225kW pitch regulated WT

The parameter identification for the V27 wind turbine regarded the power mean and standard deviation as
well as the mean pitch angle and standard deviation. The independent variables characterize the wind
deterministic (mean wind speed, air density and wind shear), stochastic (standard deviation of wind speed
and wind direction, turbulence length scale) as well as the distribution characteristics (higher moments of
wind speed). The tabulated statistics, namely mean values and standard deviation, of the independent
parameters, for the different wind speed ranges where the parameter identification was performed are
presented in table 7.1. The regression results, by means of the dependence coefficients are presented in
figure 7.1.
The following are observed:
For the mean power output:
• the predominant parameter for all wind speed ranges is the mean wind speed; the results follow the
power curve slope
• the standard deviation of wind speed induces a positive effect under 13m/s and an adverse effect for
higher wind speed ranges
• the wind direction standard deviation has an adverse effect more pronounced under 13m/s
• wind speed distribution skewness has a limited positive effect for the upper ranges
• air density effect is insignificant due to the small variation of the parameter
For the power standard deviation:

• the predominant parameters for all wind speed ranges are the turbulent components namely the
standard deviations of wind speed and wind direction
• wind speed effect is positive in lower wind speeds and adverse in higher ranges
• wind speed distribution skewness has a significant adverse effect in high wind speed ranges
• the limited air density positive effect is present within the middle wind speed ranges
For the mean pitch angle (control system response):

• the predominant parameter for all wind speed ranges is the mean wind speed
• the standard deviation of wind speed and wind skewness induce a positive effect; the opposite holds
for the standard deviation of wind direction
• the turbulence length scale induces an adverse effect in the 11-13m/s range
For the standard deviation of pitch angle (control system response):

• the predominant parameters for all wind speed ranges are the mean and the standard deviation of wind
speed
• the positive effects of air density, standard deviation of wind direction and wind speed distribution
skewness are detected in nearly all ranges
• the turbulence length scale induces an adverse effect in middle wind speed ranges

7.3.2 Parameter identification on a 110kW stall regulated WT

The parameter identification for the WINCON 110XT wind turbine regarded the power mean and standard
deviation. The independent variables characterize the wind deterministic (mean wind speed, air density,
wind shear and wind inclination angle), stochastic (standard deviation of wind speed components and
turbulence length scale) as well as the distribution characteristics (higher moments of wind speed). The
tabulated statistics, namely mean values and standard deviation, of the independent parameters, for the
different wind speed ranges where the parameter identification was performed are presented in table 7.2.
The regression results, by means of the dependence coefficients are presented in figure 7.2. The following
are observed:
power curve slope
• air density positive effect is captured for all ranges; this is due to the fact that the parameter values
covered a significant domain within the data base
• the standard deviations of the wind speed components are inducing an adverse effect for all wind
speed ranges
• wind speed inclination induces an adverse effect that is present for all wind speed ranges

• the predominant parameter for all wind speed ranges is the standard deviation of longitudinal wind
speed component
• wind speed effect is positive in lower wind speeds and adverse in the higher ranges
• standard deviation of the vertical wind speed component is significant and positive for the upper
ranges whereas the lateral component effect is present in the higher wind speed range
• wind inclination effect is positive and pronounced in higher wind speed ranges whereas wind shear
effect is insignificant
• air density effect is positive in all ranges
• wind speed distribution estimators have adverse effect; skewness effect is pronounced in higher wind
speed ranges

The parameter identification for the NTK500/37 wind turbine regarded the power mean and standard
deviation as well as the mean pitch angle and standard deviation. The independent variables characterize
the wind deterministic (mean wind speed, air density and wind shear), stochastic (standard deviation of
wind speed and wind direction, turbulence length scale) as well as the distribution characteristics (higher
moments of wind speed). The tabulated statistics, namely mean values and standard deviation, of the
independent parameters, for the different wind speed ranges where the parameter identification was
performed are presented in table 7.3. The regression results, by means of the dependence coefficients are
presented in figure 7.3. The following are observed:
power curve slope
• the longitudinal component of turbulence induces a significant positive effect under 11m/s
• the wind direction standard deviation has an adverse effect more pronounced under 11m/s
• wind speed distribution skewness has a positive effect pronounced in the lower wind speed ranges
• the adverse effect of wind shear is significant in the lower wind speed range
• air density effect is insignificant
• turbulent length scale effect is limited and adverse


• the predominant parameters for all wind speed ranges are the turbulent components namely the
standard deviations of wind speed and wind direction
• wind speed effect is positive in lower wind speeds and adverse in higher ranges
• wind speed distribution skewness effect is positive for the lower ranges whereas adverse for the high
range
• the air density effect is positive in 8-11m/s range and adverse in the high wind range
7.3.3.1 Analysis on combined data bases

In order to assess the sensitivity of the power characteristics of one specific wind turbine type in different
sites, flat and complex, several power performance campaigns, performed by the participants, were
collected and analyzed by means of the parameter identification procedure described in 7.2. The power
curve measurements for the NTK500/37 wind turbine were performed by the following institutes:
Institute Site Type of site
RISO RISO flat
DEWI Pilsum flat
WINDTEST - flat
NEL (data for 2 WT) Taff Ely complex
CRES Toplou complex
A brief overview of the data base structure is presented in table 7.4. In figure 7.4 the power data scatter
plots are presented for all campaigns. Moreover, in figures 7.5 to 7.7 the basic wind characteristics,
namely turbulence intensity, standard deviation of wind direction and wind shear ratio, that were observed
during the power performance campaigns are also presented. The bin averaged power curve of each
measuring campaign along with AEP estimations for several annual wind speed cases, for Rayleigh
distribution, are given in figures 7.8 and 7.9 respectively. The AEP estimations are based for power curves
limited in the range 4-14m/s in order to maintain a common basis for AEP comparisons.
The parameter identification regarded the air density corrected power mean and standard deviation. The
independent variables characterize the wind deterministic (mean wind speed and wind shear ratio) and
stochastic characteristics (standard deviation of wind speed and wind direction). The regression results, by
means of the sensitivity factors are presented in tables 7.5 and 7.6 for mean power and standard deviation,
respectively.
The following are observed:
• the captured effects are representative for the specific site with the specific instrumentation used; an
uncertainty of limited importance is introduced in the comparison as each instrumentation may
respond differently in a turbulent wind field
• the dependency of mean power output on mean wind speed presented differences between the
measurements at the different sites. The differences are in accordance with the figure 7.8.
• the effect of wind shear, as described by the ratio of wind speed at two different heights (U35m/U18m),
presented different behavior in flat and complex sites. In the former case it affected power output in a
positive way whereas in the latter presented an adverse effect
• turbulence effects, as captured by the wind speed standard deviation, presented significant variation
for the different sites. In general, positive effects were identified in lower wind speed ranges and
negative in higher ranges.
• standard deviation of wind direction introduce in general an adverse effect; at higher wind speed
ranges the effect is positive

• standard deviation of power output is strongly dependent on wind speed standard deviation; secondary
effects are captured by mean wind speed, wind shear ratio and wind direction standard deviation. In
all campaigns the captured effects are quantitatively comparable.

The parameter identification for the MADE 300kW wind turbine regarded the power mean and standard
deviation. Three groups of data have been utilized regarding operation in flat and complex terrain as
follows:
Group Terrain type Wind direction
EAST Complex 56.25°-123.75°
WEST Complex 236.25°-258.75°
FLAT Flat -
The independent variables characterize the wind deterministic (mean wind speed and wind shear factor)
and stochastic characteristics (standard deviation of wind speed and wind direction). The data base has
been split into wind speed intervals, namely 6-8m/s, 8-11m/s, 11-13m/s and 13-16m/s. The regression
analysis tool, in accordance to paragraph 7.2, followed the forward stepwise processing and the regression
results are presented by means of the beta coefficients for the variables that have been considered as
statistically significant by the method in figures 7.10 and 7.11 for mean power and standard deviation,
respectively.
In the case of mean power, wind speed is the principal factor. The non-linear behavior of the power curve
at low wind speeds and at wind speeds close to the stall region is also detected in the beta coefficient for
this parameter, which is lesser for 6-8m/s wind speed interval for all cases and 13-16m/s for complex
EAST case, and even negative for 13-16m/s for complex WEST and FLAT cases. In the three cases, beta
coefficient for wind speed presents higher values for the linear region of power curve. When examining
standard deviation of wind speed, power performance seems more influenced by this parameter in
complex terrain that in flat one. It should be remembered that all variables were transformed into normal
distribution and therefore beta coefficient means sensibility, it does not matter the range of variation of the
independent variable. Taking this into account, one can interpret this as a different behavior of the wind
turbine power performance before the same level of turbulence, depending on the complexity of the
terrain. This last fact can be provoked only by a different pattern of the turbulence spatial distribution. For
standard deviation of wind direction it is seen that in complex terrain there is more influence of this
parameter than in flat terrain. Standard deviation of wind direction is a stochastic parameter which has to
do with turbulence pattern of the flow. In complex terrain, specially for WEST direction it is an important
negative parameter. This may be attributed to nacelle orientation characteristics. Shear factor does not
present a common pattern of influence for the three cases. Only for complex WEST and FLAT this
dependence is similar. Moreover the influence appears only in few wind speed intervals. It should be
considered that in complex terrain vertical profiles are quite flat, and therefore, shear estimation by means
of logarithmic fitting (this was the kind of fitting used here) in these sites is not the most appropriated.
Finally, standard deviation of power is primarily influenced by standard deviation of wind speed for all
three cases.
7.3.5 Parameter identification on a 3MW pitch controlled WT
The tendency towards larger wind turbines leads to an increasing interest in the relation between
meteorological conditions and the power performance of these systems. The rotor of a MW-scale wind
turbine covers a height range of about 50-100 m in which strong vertical wind speed gradients can occur.

In that sense the experimental investigation on the relations between the power performance of the
Aeolus II and vertical wind shear, turbulence intensity and atmospheric stability are representative and
presented herein.
Aeolus II is a variable speed, pitch controlled wind turbine with a nominal power output of 3MW. The
rotor diameter is 80.5m whereas the hub height reaches 92m.
The recorded data at the Aeolus II were classified according to the turbulence intensity. For each
turbulence class a power coefficient curve was evaluated, presented in figure 7.12. The resulting annual
energy estimations are presented in figure 7.13. When evaluating a power curve the measured power and
wind speed is averaged over a period of usually 10 minutes. Because of the wind speed fluctuations and
the non linearity of the power curve this time averaging process leads to an increase of the evaluated
power curve where the power curve is concave and to an decrease of the evaluated power curve where the
power curve is convex. This effect is evident from figures 7.13 and 7.14 and will grow with increasing
turbulence intensity.
From the wind speed measurements at the heights 32m, 62m, 92m and 126m a wind gradient weighted
with its contribution to the rotor disk area was calculated and served as classification criteria for the
recorded data at the Aeolus II. For each class of wind gradient a power curves were calculated and the
resulting AEP estimations are presented in figure 7.15. Obviously for the applied wind speed distribution
an increase of wind gradient leads to an increase of the AEP. This effect can be generally understood from
the observations concerning the influence of turbulence intensity on the power performance discussed
before: high wind gradients correspond to high spatial variations of wind speed over the rotor height. The
spatial averaging effect of the rotor together with the nearly cubic increase of the power with wind speed
result in a higher annual energy production similar to the effect of high turbulence.
At the Aeolus II information about the actual atmospheric stability is derived from measurements of the
vertical temperature gradient, i. e. temperature measured at 90 m and 2.5 m. A positive difference of the so
called potential temperature dθ=θ(90.0m)-θ(2.5m) corresponds to a stable layer. A potential temperature
difference around 0K indicates a neutral layer and a negative dθ correspond to an unstable layer. For each
class of dθ a power curve was evaluated. The resulting AEP estimations have the tendency to decrease
with increasing stability, as shown in figure 7.16.
The atmospheric stability strongly influences the turbulence intensity and the vertical wind gradient, as it
is presented in figure 7.17. Unstable conditions correspond to higher turbulence levels and relative low
wind gradients, while lower turbulence and high wind gradients usually occur during stable conditions.
Because the turbulence as well as the vertical wind gradient effect the power performance, the observed
dependence of the power curve on the stratification can probably be interpreted as an interference of both
influences.
7.4 Conclusions
An analytic method is introduced covering parameter identification tasks in power performance
measurements. The developed tool is based on multivariate regression analysis with a backward parameter
elimination technique in order to count for the statistically insignificant parameters. The method is applied
for the parameter identification of the power performance of wind turbines operating at different sites,
both flat and complex. The parameters chosen to comprise the model set describe the deterministic part of
the wind, the main turbulent characteristics, the turbulence length scale as well as the wind speed
distribution. The examined dependent variables were the mean power output, the standard deviation of the
power output and in the case of pitch controlled machine the response of the control system. The effect of
the above parameters was captured and quantified. It was revealed that except from the mean wind speed

other parameters, related to the deterministic as well to the stochastic wind characteristics are affecting the
power performance of the turbine and should be considered within power performance assessment
procedures.
More specifically, the following are the concluding remarks:
• air density effects, as they were captured by the parameter identification procedure, were of limited
magnitude due to the narrow variation of air density. On the other hand the capture dependency was
found to be in accordance with the theoretical linear corrections
• turbulence effects on mean power output it is mainly governed by the averaging effects that result to
higher power output at lower wind speed ranges and the reverse at higher wind speed ranges close to
power control region. On the other hand it was found that the averaging effect is not the only
mechanism that governs the turbulence effect as different sensitivities were found for the same turbine
operated at different sites. Moreover, for the case of a small sized stall regulated wind turbine the
turbulence effects, captured by the standard deviations of the three wind speed components, were
found to be adverse for all wind speed ranges.
• wind shear effect is more pronounced in large wind turbines; the effect is not systematic and depends
on the definition of the wind shear descriptor. The different effects that are captured for the same
machine at different test sites strongly indicates that the one parameter description of wind shear is
inadequate for describing the spatial variations of wind speed upwind to the rotor plane.
• wind inclination impose an adverse effect on mean power output; the effect is pronounced at lower
wind speed ranges
• wind speed higher order statistics introduce a secondary effect. Distribution skewness, that is linearly
dependent on the mean value of cubed wind speed, impose a positive effect on mean power output
that is significant for large wind turbines.
• significant influence of the atmospheric stability, vertical wind shear and the turbulence intensity on
the power performance of the 3MW wind turbine was found. The influence of such effects on the
predicted AEP is in the range of 4% and should be reflected in the recommendations for power curve
measurements for large systems.
• the results of the parameter identification indicate that wind inflow characteristics should be measured
and reported in more detail including the 3D structure of mean wind speed and turbulence. This would
allow for estimation of sensitivity factors and the optional normalization of the power performance
measurements to specified conditions

Table 7.1 Independent parameter variation for the V27 data base.
Wind speed bin U=6-8m/s U=8-11m/s U=11-13m/s U=13-16m/s
Parameter mean SDV mean SDV mean SDV mean SDV
Mean wind speed - U (m/s) 7.1 0.57 9.53 0.83 12.0 0.578 14.1 0.84
Air density - ρ (kgm-3) 1.15 0.008 1.15 0.007 1.15 0.007 1.15 0.007
Wind shear exponent - α 0.04 0.049 0.048 0.047 0.043 0.038 0.039 0.033
SDV wind speed - σu (m/s) 0.785 0.192 0.964 0.178 1.21 0.221 1.42 0.302
SDV wind direction - σϕ (deg) 6.69 2.29 5.98 1.48 5.67 1.2 5.33 0.986
Wind speed skewness - Usk -0.045 0.253 -0.046 0.268 -0.063 0.292 -0.136 0.286
Wind speed kurtosis - Uku -0.192 0.366 -0.123 0.404 -0.157 0.443 -0.22 0.346
Turbulent length scale - Lu (m) 65.29 32.31 58.37 25.37 55.2 21.8 57.6 20.0
Table 7.2 Independent parameter variation for the W110XT data base.
Air density - ρ (kgm-3) 1.19 0.027 1.19 0.029 1.18 0.020 1.19 0.018
Wind shear exponent - α 0.029 0.034 0.023 0.028 0.024 0.027 0.03 0.024
Wind inclination - θ (deg) 3.45 1.38 3.49 1.26 3.14 0.984 3.5 0.90
Lateral turbulence component σv 0.95 0.25 1.21 0.28 1.40 0.30 1.69 0.34
Vertical turbulence component σw 0.78 0.20 1.01 0.22 1.16 0.19 1.39 0.24
Wind speed skewness - Usk -0.087 0.289 -0.13 0.30 -0.255 0.295 -0.20 0.295
Wind speed kurtosis - Uku -0.054 0.53 -0.071 0.47 -0.025 0.466 -0.063 0.448
Table 7.3 Independent parameter variation for the NTK500/37 data base.
Air density - ρ (kgm-3) 1.23 0.009 1.23 0.007 1.23 0.009 1.23 0.011
Wind shear exponent - α -0.017 0.019 -0.015 0.016 -0.019 0.011 -0.024 0.009
SDV wind direction - σϕ (deg) 5.56 2.14 6.25 1.92 6.22 1.37 6.19 0.99
Wind speed skewness - Usk 0.094 0.315 -0.053 0.269 -0.026 0.26 -0.039 0.22
Wind speed kurtosis - Uku 0.388 1.22 -0.152 0.498 -0.343 0.299 -0.42 0.27

TABLE 7.4 Parameters involved in identification procedure.

Model parameter set RISO DEWI WINDTEST NEL (#A) NEL (#B) CRES
Wind speed x x x x x x
SDV wind speed x x x x x x
Wind speed min/max x x x
Wind speed skewness x
Wind speed kurtosis x
Wind shear ratio x x x x
SDV wind direction x x x x x
Power mean (norm.) x x x x x x
SDV power (norm.) x x x x x

Table 7.5 Sensitivity factors for mean power output for NORDTANK 500/37 measured at different sites.
MEAN POWER SENSITIVITY FACTOR FOR MEAN WIND SPEED (∂ P ∂U )

Wind speed RISO DEWI WINDTEST NEL (#A) NEL (#B) CRES
6m/s 41,5 39.5 35.4 33.7 35.2 39.5
8m/s 57.8 56.9 59.6 48.5 50.7 58.9
10m/s 66.7 63.2 64.1 59.7 59.2 60.3
12m/s 50.0 52.5 52.9 52.1 49.9 51.9
14m/s 39.1 35.9 38.5 39.5 31.7 21.1
16m/s - 4.3 - 18.6 11.6 4.5
18m/s - -1.0 - 0.3 -0.2 -4.5
20m/s - - - -11.7 -9.9 -
22m/s - - - -11.6 -16.8 -
MEAN POWER SENSITIVITY FACTOR FOR WIND SHEAR RATIO (∂ P ∂ wsr )
6m/s - 105.4 - .3 -22.7 -202.6
8m/s - 3.2 - -7.3 -61.5 -303.0
10m/s - 18.8 - -48.9 -124.9 -565.8
12m/s - 330.0 - 1.4 -131.4 -617.8
14m/s - 230.0 - -83.6 -154.3 -269.1
16m/s - 95.2 - -55.7 -131.3 -269.8
18m/s - 849.3 - -43.6 -112.6 -464.5
20m/s - - - -185.7 -155.1 -
22m/s - - - -210.3 110.9 -
MEAN POWER SENSITIVITY FACTOR FOR WIND DIRECTION SDV (∂ P ∂σ )U
6m/s 10.9 4.8 4.2 9.62 4.24 5.52
8m/s 7.3 6.2 -2.2 9.41 6.02 8.53
10m/s -5.1 -4.7 2.98 9.50 4.24 20.23
12m/s -17.0 4.93 5.43 -6.26 4.33 3.05
14m/s -35.1 1.41 13.36 -6.64 -5.03 14.79
16m/s - -7.2 - -4.74 -8.50 -7.08
18m/s - -9.7 - -13.98 -20.57 -9.22
20m/s - - - -27.18 -14.47 -
22m/s - - - -5.44 -14.05 -
MEAN POWER SENSITIVITY FACTOR FOR WIND DIRECTION SDV (∂ P ∂σ )ϕ
6m/s - - -0.13 -0.27 -0.34 -0.71
8m/s - - 1.02 -1.07 -1.75 -2.06
10m/s - - 0.03 -1.67 -1.19 -3.10
12m/s - - 4.06 -0.45 -2.09 -7.81
14m/s - - 3.80 -0.60 -3.37 -0.38
16m/s - - - -1.18 -0.35 -0.04
18m/s - - - 0.85 -0.91 -0.14
20m/s - - - 3.35 -6.97 -
22m/s - - - 3.78 34.46 -

Table 7.6 Sensitivity factors for power standard deviation for NORDTANK 500/37 measured at different
sites.
POWER SDV SENSITIVITY FACTOR FOR MEAN WIND SPEED (∂σ P ∂U )

6m/s - 3.46 3.93 4.36 4.71 4.56
8m/s - 4.50 4.31 4.30 5.22 4.84
10m/s - -0.89 1.89 1.90 2.47 5.19
12m/s - -2.47 -3.39 -2.52 -3.30 -5.74
14m/s - -7.78 -1.89 -6.35 -8.05 -9.70
16m/s - 0.40 - -6.09 -3.48 -3.65
18m/s - 3.09 - 1.23 1.13 1.35
20m/s - - - 2.14 1.13 -
22m/s - - - -0.58 0.45 -
POWER SDV SENSITIVITY FACTOR FOR WIND SHEAR RATIO (∂σ P ∂ wsr )
6m/s - -19.13 - -7.94 -12.77 -9.76
8m/s - -51.43 - -18.01 -22.58 -18.61
10m/s - -56.58 - -47.79 -60.16 -20.13
12m/s - 130.8 - -27.12 -37.67 68.19
14m/s - 98.2 - -7.22 16.84 85.26
16m/s - 51.1 - -5.91 7.00 15.69
18m/s - -100.1 - 9.72 24.71 -40.50
20m/s - - - 97.74 14.30 -
22m/s - - - 87.88 63.45 -
POWER SDV SENSITIVITY FACTOR FOR WIND DIRECTION SDV (∂σ P ∂σ U )
6m/s - 27.48 24.53 26.88 25.57 27.69
8m/s - 42.55 38.62 40.13 34.98 32.45
10m/s - 50.58 35.74 37.17 28.27 23.59
12m/s - 27.20 26.54 19.34 10.46 18.89
14m/s - 14.25 11.75 16.1 11.37 3.76
16m/s - 10.59 - 6.98 5.51 4.79
18m/s - 6.23 - 7.46 5.26 5.11
20m/s - - - 11.6 6.05 -
22m/s - - - 4.47 0.73 -
POWER SDV SENSITIVITY FACTOR FOR WIND DIRECTION SDV (∂σ P ∂σ ϕ )
6m/s - - 1.34 0.41 0.27 0.76
8m/s - - 2.06 0.48 1.00 1.92
10m/s - - 2.85 1.76 1.98 3.47
12m/s - - 2.56 3.62 4.45 1.44
14m/s - - 4.57 1.68 3.24 -0.05
16m/s - - 2.05 1.74 -1.17
18m/s - - 0.41 2.72 1.63
20m/s - - 1.93 3.07 -
22m/s - - 0.41 0.86 -

V27 POWER PERFORMANCE ASSESSMENT

Mean power SDV power
24 24
U=6-8m/s
U=6-8m/s
20 U=8-11m/s
20 U=8-11m/s
16 U=11-13m/s
U=11-13m/s
U=13-16m/s
16 U=13-16m/s
12
8
12
4
8
0
4 -4
-8
0
-12
-4 -16
U ñ á óu óö U s k U kr Lu U ñ á óu óö U s k U kr Lu
Mean pitch angle SDV pitch angle

40 40
U=6-8m/s
36 U=5-8m/s 36
U=8-11m/s
32 U=8-11m/s 32 U=11-13m/s
U=11-13m/s
28 28
U=13-16m/s
U=13-16m/s
24 24
20 20
16 16
12 12
8 8
4 4
0 0
-4 -4
-8 -8
U ñ á óu óö U s k U kr Lu U ñ á óu óö U s k U kr Lu
Figure 7.1 Dependence coefficient charts for V27 at Andros power performance.

W110XT POWER PERFORMANCE ASSESSMENT

24 24
U=6-8m/s
U=6-8m/s 20
U=8-11m/s
20 U=8-11m/s
16 U=11-13m/s
U=11-13m/s
U=13-16m/s
16 U=13-16m/s 12
8
12
4
0
8
-4
4 -8
-12
0
-16
-4 -20
U ñ á è óu ó v ó w U s k U ku L u U ñ á è óu ó v ó w U s k U ku L u
Figure 7.2 Dependence coefficient charts for W110XT at Lavrio power performance.
NTK500/37 POWER PERFORMANCE ASSESSMENT

28 28
U=6-8m/s
24
U=6-8m/s 24
U=8-11m/s
U=8-11m/s
20 U=11-13m/s
20 U=11-13m/s
U=13-16m/s
U=13-16m/s 16
16
12
12 8
8 4
0
4
-4
0
-8
-4 -12
-8 -16
U ñ á ó U ó ö U s k U ku Lu U ñ á ó U ó ö U s k U ku Lu
Figure 7.3 Dependence coefficient charts for NTK500/37 at Toplou power performance.

POWER PERFORMANCE ASSESSMENT

COMBINED DATA BASE FOR NTK500/37
100 200 300 400 500 600
100 200 300 400 500 600

Power (kW) - DEWI data
Power (kW) - RISO data
0
-100
-100
3 6 9 12 15 18 21 3 6 9 12 15 18 21
100 200 300 400 500 600
100 200 300 400 500 600

Power (kW) - WINDTEST data
Power (kW) - NEL (A) data

0
0
-100
-100
3 6 9 12 15 18 3 6 9 12 15 18 21 24 27 30
100 200 300 400 500 600
100 200 300 400 500 600

Power (kW) - NEL (B) data
Power (kW) - CRES data

0
0
-100
-100
3 6 9 12 15 18 21 24 27 30 3 6 9 12 15 18 21 24 27
Figure 7.4 Power output for NTK500/37 measured at different sites.

POWER PERFORMANCE ASSESSMENT

0.30
COMBINED DATA BASE FOR NTK500/37
0.30
Turbulence intensity - DEWI data
Turbulence intensity - RISO data
0.25
0.25
0.20
0.20
0.15
0.15
0.10
0.10
0.05
0.05
0.00
0.00
3 6 9 12 15 18 21 3 6 9 12 15 18 21
0.30
0.30
Turbulence intensity - WINDTEST data
Turbulence intensity - NEL (A) data

0.25
0.25
0.20
0.20
0.15
0.15
0.10
0.10
0.05
0.05
0.00
0.00
3 6 9 12 15 18 3 6 9 12 15 18 21 24 27 30
0.30
0.30
Turbulence intensity - NEL (B) data
Turbulence intensity - CRES data

0.25
0.25
0.20
0.20
0.15
0.15
0.10
0.10
0.05
0.05
0.00
0.00
3 6 9 12 15 18 21 24 27 30 3 6 9 12 15 18 21 24 27
Figure 7.5 Turbulence intensity distributions at NTK500/37 sites.

8.0 10.0 12.0 14.0

8.0
SDV wind direction - WINDTEST data
SDV wind direction - NEL (A) data

6.0
4.0
6.0
4.0
2.0
2.0
0.0
0.0
3 6 9 12 15 18 3 6 9 12 15 18 21 24 27 30
8.0 10.0 12.0 14.0
8.0 10.0 12.0 14.0

SDV wind direction - NEL (B) data
SDV wind direction - CRES data

6.0
6.0
4.0
4.0
2.0
2.0
0.0
0.0
3 6 9 12 15 18 21 24 27 30 3 6 9 12 15 18 21 24 27
Figure 7.6 Standard deviation of wind direction at NTK500/37 sites.

1.20
1.60
Wind shear ratio - NEL (A) data
Wind shear ratio - DEWI data
1.15
1.40
1.10
1.20
1.05
1.00
1.00
3 6 9 12 15 18 21 3 6 9 12 15 18 21 24 27 30
1.60
1.20
Wind shear ratio - NEL (B) data
Wind shear ratio - CRES data

1.40
1.00
1.20
1.00
0.80
3 6 9 12 15 18 21 24 27 30 3 6 9 12 15 18 21 24 27
Figure 7.7 Wind shear ratio (U35m/U18m) at NTK500/37 sites.

NTK500/37 POWER CURVES

550
500
450
400
350
Power (kW)
300 RISO
DEWI
250
WINDTEST
200 NEL (A)
NEL (B)
150
CRES
100
50
3 6 9 12 15 18 21 24 27
Wind speed (m/s)
Figure 7.8 Power curves for NTK500/37 measured at different sites.
NTK500/37 ANNUAL ENERGY PRODUCTION

(Rayleigh distribution, power curve 4-14m/s)
1400
RISO
DEWI
WINDTEST
1200
Annual energy production (MWh)
NEL (A)
NEL (B)
CRES
1000
800
600
400
4 6 8 10 12
Annual mean wind speed (m/s)
Figure 7.9 Annual energy production estimations for NTK500/37 for

different sites.

PI FOR POWER PI FOR STD OF POWER

BETA COEFFICIENT
EAST WIND DIRECTION BETA COEFFICIENT
EAST WIND DIRECTION
1 1
(6-8)
0.8 0.8
(8-11)
0.6 (6-8) 0.6
(11-13)
(8-11)
0.4 0.4 13-16)
(11-13)
0.2 0.2
(13-16)
0 0
-0.2 -0.2
-0.4 -0.4
U SDV U SDV DIR ALFA U SDV U SDV DIR ALFA
PARAMETERS PARAMETERS

BETA COEFFICIENT
WEST WIND DIRECTION BETA COEFFICIENT
WEST WIND DIRECTION
1 1
(6-8)
0.8 (6-8) 0.8
(8-11)
0.6 (8-11) 0.6
(11-13)
0.4 (11-13) 0.4
13-16)
0.2 (13-16) 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6

BETA COEFFICIENT
FLAT TERRAIN BETA COEFFICIENT
FLAT TERRAIN
1 1
(6-8) 0.8 (6-8)

0.8
(8-11) 0.6 (8-11)
0.6 (11-13) 0.4 (11-13)
0.4 (13-16) 0.2 (13-16)
0
0.2
-0.2
0
-0.4
-0.2 -0.6
Figure 7.10 Beta coefficients for power mean. Figure 7.11 Beta coefficients for power SDV.

Parameter: Turbulenzintensität 500 5

0.5 deviation in percentage
relative deviation in AEP [%]

400 4
absolute deviation in AEP

abs olute deviation in A E P
0.4 < 7.5% 300 3
Leistungsbeiwert Cp [-]
> 10%
0.3 200 2
[MWh]
100 1
0.2
0 0
0.1 -100 -1
0 -200 -2
-300 -3
-0.1
0 5 10 15
0 5 10 15 20 25
Windgeschw indigkeit v in Nabenhöhe [m/s] turbulence intensity [% ]
Figure 7.12 CP-curves of the Aeolus II for different Figure 7.13 Deviations of calculated AEP’s of the
turbulence intensities. Aeolus II for different turbulence intensities from
average value. Base: measured Weibull distribution
at hub height of the Aeolus II.
parameter: turbulence intensity parameter: w ind gradient (absolute)

300 3
3000
200 2
100 1
relative deviation in
absolute deviation in
2500
0 0
AEP [MWh]
2000
AEP [%]
14 m /s -100 -1
12 m /s -200 -2
1500 10 m /s -300 -3
8 m /s abs olute deviation in A E P
-400 -4
1000 6 m /s
-500 deviation in percentage -5
500 -600 -6
-700 -7
> 0.05
0.005
0.015
0.025
0.035
0.045
0
0 5 10 15 20
turbulence intensity [% ] w ind gradient, absolute [(m/s)/m]
Figure 7.14 Electrical power of the Aeolus II as Figure 7.15 Deviations of calculated AEP’s of the
function of turbulence intensity for different wind Aeolus II for different vertical wind gradients from
speeds. average value. Base: Weibull distribution at hub
height of the Aeolus II.
parameter: difference in pot. temperature (90m-2.5m) parameter: difference in potential temperature

T(90m) - T(2.5m)
200 4
11 0.005
150 3
10 0.0045
100 2
absolute deviation in
relative deviation in
turbulence intensity [%]
9 0.004
wind gradient, relative
50 1
8 0.0035
AEP [MWh]
turbulence
0 0
AEP [%]
[1/m]
7 intens ity 0.003

w ind gradient
-50 -1 6 0.0025
-100 -2 5 0.002
-150 AEP [MW h] -3 4 <-uns table<- neutral ->s table-> 0.0015
deviation
-200 -4 3 0.001
-250 <- uns table <- neutral -> s table ->
-5 2 0.0005
-300 -6 <-1.25K -0.5K 0.5K 1.5K 3.5K
< -1.25K -0.5 K 0.5 K 1.5 K 3.5 K difference in potential temperature [K]
difference in potential temperature [K]
Figure 7.16 Deviations of calculated AEP’s of the Figure 7.17 Turbulence intensity and wind gradient
Aeolus II for different atmospheric stability from versus difference in potential temperature. Plotted
average value. Base: Weibull distribution at hub is the relative vertical wind speed gradient, i. e. the
height of the Aeolus II. wind speed gradient normalised by the average
wind speed over the height.

8. ASSESSMENT OF POWER PERFORMANCE PRACTICES

8.1 Introduction
The assessment is focused on the recommendations of IEC 1400-12. Certain inefficiencies, related to site
calibration and uncertainty estimation procedures, were identified.
The assessment of the existing standards and recommendations focused on the following issues:
• status of wind turbine power performance assessment and verification (existing standards, EU project
results, contractual matters, market needs etc.)
• review of characteristics of complex terrain and their effect to power performance
• points that are not adequately described by the present IEC 1400-12 document, particularly in relation
to complex terrain (site calibration, uncertainty estimation etc.)
• recommendations for performing power performance assessment and verification in complex terrain.
Only IEC 1400-12 is referenced. Other previously used documents (as IEA and ECN217) have no
relevance anymore.
8.2 Review On The Characterization Of Complex Terrain

Complex terrain induces significant deterministic and stochastic variations of the wind field structure. The
main characteristics are summarized below:
Deterministic characteristics:
• vertical wind shear is by no means following flat terrain descriptions (i.e. logarithmic)
• wind profile is clearly affected by upwind terrain features inducing local speed-up regions which vary
with wind direction
• mean wind speed vector inclination follows the upwind terrain
• upwind terrain induces distortion effects in the transversal component in the vertical direction
Stochastic characteristics:
• significantly higher turbulence kinetic energy levels are encountered in complex sites
• the energy distribution mechanism differs from flat terrain; longitudinal component energy is
transferred to the vertical
• increased values of turbulence ratios i.e. σv/σu , σw/σu
• spectral density distribution of wind speed deviates from flat terrain descriptions (i.e. low frequency
spectral gap)
• smaller turbulence length scales are seen in complex sites
• non Gaussian character of wind speed distribution
The main effects that complex terrain induces to power performance are related to:
• uncertainty of wind speed measurement via cup anemometers
• uncertainty of determination of reference wind speed (i.e. large spatial variation of wind speed)
• effect of turbulence, shear deformations and wind speed distribution
• response of control systems

8.3 Critical Points On Existing Recommendations

The identified critical points regard the test site requirements, the characterization of the wind speed
sensors, the influence of mean wind and turbulence structure, site calibration and measurement analysis.
8.3.1 Test site requirements

The document gives an assessment procedure of the test site in Annex A. In this annex upwind obstacles
and upwind neighboring wind turbines at a distance of 20 equivalent diameters or more are considered as
virtually non-existing. The literature review results (Van de Borg N., 1997) indicate that this could be
non-conservative. Furthermore the annex does not take into account the hub height of the WTG under
test, which seems to be unrealistic.
Moreover, the test site assessment procedure is quite sensitive to its way of application. Different
procedures for mapping, slope computing or wind direction sectoring may lead to different assessment for
the same site.
8.3.2 Characterization of wind speed sensors

The existing standard indicates that the measurement of the wind speed must be done by using cup
anemometers with some specific characteristics. Cup anemometers present deficiencies that have an
important influence on the measure of power curve in complex terrain. The deficiencies are related to the
fact that the sensors can not detect some properties of the flow that have a proved influence on the power
performance of the machine as well as that they misread the wind speed vector. All these deviations from
the real measurement, can be included as uncertainties in power performance due to wind speed, once the
response of the sensor is experimentally investigated and the level of the disturbing factors are determined
in field. Yet, the magnitude of the resulting uncertainty indicates that the systematic usage of more
accurate methods is justified.
In complex terrain significant variations of the wind inclination angle are commonly encountered. In the
work of Glocker S. (WindTest), 1996a, 1997, an experimental research on the sensitivity of cup
anemometers to inclined flow was performed. The results, shown in figures 8.1 and 8.2, revealed the
significant deviation of the cup anemometer recording from the theoretical “cosine” response. From the
results it is clear that the anemometers may be classified according to who well their measurements match
the “cosine” or the constant response.

Figure 8.1 Sensitivity of cup anemometers to inclined flow.
Figure 8.2 Sensitivity of cup anemometers to inclined flow.

8.3.3 Influence of wind mean and turbulent structure

The power curve is defined by the IEC-document as the relation between the undisturbed wind speed at
the rotor center and the electrical power, corrected to standard air densities. The document does not take
into account the influence of the following parameters:
Turbulence
The IEC-document prescribes the measurement of the turbulence but it does not recommend any
corrections. As it was shown from the results of the parameter identification process the effect of
turbulence can be identified and quantified. The magnitude of the effect depends on the wind turbine size
and control type as well as on the specific site characteristics.
Wind shear
The power of a turbine is related to the wind speed structure all over the rotor plane rather than to the
wind speed at the rotor center. For this reason wind shear is expected to influence the power curve of a
turbine. As it was shown from the results of the parameter identification process the effect of wind shear
can be identified and quantified. Again, the magnitude of the effect depends on the wind turbine size and
control type as well as on the specific site characteristics.
Inclined flow
The IEC-document prescribes the use of a cup-anemometer and assumes implicitly that the response of a
cup-anemometer and the response of a turbine to inclined flows are identical. Recent experimental
research proved that the response of cup anemometers may deviate significantly from the expected
“cosine” response”, thus inclined flows may degrade the power curve of a wind turbine.
Since these parameters are dependent on the complexity of the test site it is expected that the power curve
of a wind turbine on a complex site may present significant variations from the power curve of the same
type turbine operated in a different site (flat or complex).
8.3.4 Site calibration

Sites that do not fulfill the requirements of the IEC (1997) can be calibrated to determine the relation of
the measured wind speed at the meteorological mast and the undisturbed wind speed at the turbine
location. The experimental site calibration is described in the IEC (1997) in such a way that it allows for
personal interpretation, which is also valid for the estimation of the uncertainties associated with the site
calibration.
The document gives the possibility to perform site calibration by numerical modeling. The results from
the literature survey suggest that numerical modeling may be not applied in a strict and systematic way.
On the other hand, great research effort is being spend on the development and validation of numerical
codes capable of dealing with atmospheric flows. Although the accuracy requirements of site calibration
are high, it is considered that the use of numerical modeling will comprise an alternative to experimental
calibration.
The major drawback of IEC (1997) is that no consideration of experimental site calibration in case of an
erected wind turbine is made.
8.3.5 Response system

In power performance verification studies it is often found that a specific behavior, such as lower power
output, double stall curves etc., of the wind turbine under test is not described properly from the wind
measurements. The monitoring and reporting of certain operational characteristics, such as yaw error or

pitch angle, would facilitate the interpretation of those phenomena.
8.3.6 Measurement analysis

The normalization procedure described in IEC (1997) regards air density. The results of the parameter
identification, in the cases where the air density variation was significant, are compatible with the
theoretical formulae. On the other hand, it was also proved that the effects of wind deterministic and
stochastic characteristics may present higher influence on the power curve determination. In that sense the
comparison of different power curves, though referenced at the same air density, may inhibit uncertainty
related to the above mentioned wind structure effects.
8.3.7 Contractual aspects

Current recommendation documents do not include references on contractual aspects. The primary point is
related to the determination of feasible magnitudes of accuracy for assessment and verification purposes.
8.4 Conclusions
The assessment of the applicability of current recommendations for power performance of wind turbine in
complex terrain is based on the characteristics of the wind structure as well as on the experience of
performing power performance measurements in those sites. The critical points of IEC (1997) are related
to test site requirements, characterization of the sensors, description of site calibration, identification of the
influence of secondary effects as well as to the data analysis procedures.


9. PRACTICAL RECOMMENDATIONS
The practical recommendations for power performance assessment in complex terrain regard the identified
inefficiencies of the IEC (1997). The recommendations regard test site requirements, characterization of
the sensors, description of site calibration, identification of the influence of secondary effects as well as to
the data analysis procedures.
9.1 Definition Of Site Complexity

The current regulations for site characterization should be kept, after the inclusion of recommendations for
mapping scaling and resolution, in order to identify a flat site in a strict way. The usage of purely
geographical characteristics is not feasible for classification of any terrain type that does not meet the
above mentioned requirements, and consequently it should be considered as complex.
9.2 Selection And Characterisation Of Instrument Configuration
9.2.1 Definition of reference wind speed

Current knowledge does not justify the selection of other than the horizontal wind component referenced
at hub height. Results from experiments regarding rotor averaged wind speed values do not justify, in
terms of accuracy and cost efficiency, this practice. In cases of large wind turbines, extrapolation
techniques may be utilized if their proved accuracy meets specified constraints.
9.2.2 Mean wind speed and turbulence structure measurements

It is recommended that wind measurements for power performance assessment and verification in
complex terrain should be more sophisticated in terms of instrumentation and methodology. Complex
terrain induced effects on power performance are associated to various deterministic and stochastic wind
parameters that should be measured and documented.
The problems of mounting effects, steady state calibration and dynamic response are extensively
investigated and guidance, supplementary to IEC (1997) can be given.
The use of 3D wind measurement devices is recommended either in order to produce directly the needed
wind speed component or in order to magnify the wind inclination angle. Given the characteristics of the
available cup and 3D measuring device characteristics the wind measurement strategy should be
straightforward in terms of accuracy. The experimental research support that turbulence and wind shear
affect the power curve. Given the fact that the above mentioned effect depends on the wind turbine and the
site it is supported that the measuring campaign should include wind shear and turbulence measurements
that are to be reported along with power curve. A typical setup that would allow for these measurements
include cup anemometers at three heights and one 3D anemometer at hub height.
9.2.3 Wind turbine control system response

The measurement and reporting of specific operational characteristics of the wind turbine would allow the
interpretation of power performance deviations from guaranteed curves.

9.3 Site Calibration

Site calibration should also be considered for the cases where the wind turbine under assessment or
verification is already erected. It is supported that alternative site calibration procedures should be
described and applied under specified ranges of accuracy. The latter should be the only restriction, thus
allowing for future improvements.
The recommendations should clarify the following:
• in cases where the site calibration can be performed without the presence of the wind turbine the use
of a temporary mast should be preferred
• procedure details, i.e. regression type, amount of data, sectoring procedure etc.
• flow distortion correction factors should be established for each wind direction sector, by regressing
the wind data generated from the above procedure (on site measurements prior to the installation of wt
or nacelle measurements and assumed wind velocity gradient), referring to the hub height wind speed
at wind turbine location, on the measured wind data from the reference mast
• alternative site calibration of a test site may well be performed by placing an anemometer above the
nacelle of the wind turbine at a separation distance carefully selected with respect to the nacelle cross
section so that the disturbance caused by the presence of it does not exceed 1% of the free stream wind
speed. This should be verified by means of physical or theoretical modeling of the flow field around
the nacelle of the wt. Any additional corrections which are applied to account for the spatial and
vertical variation of the wind speed measured at the wt location and the wind speed driving the wt
have to be reported clearly and the associated uncertainty has to be taken into account
• the application of nacelle cup anemometer calibration, on a running wind turbine, is to be applied for
identical wind turbines in sites that are in accordance with the calibration site in terms of wind field
and turbulence structure
• the application of the numerical site calibration should include specific articles that will ensure that the
physics of the wind field over complex terrain are properly simulated. The recommendations should
refer to equation set to be solved, the turbulence model, the grid independent solution acquirements as
well as the validation of the tool against typical validation cases.
9.4 Normalization Procedure

The current normalization procedure, that regards air density, is conserved. It is recommended that for
verification purposes the influence of wind parameters that affect power curve should be documented. A
practice for quantifying the wind characteristics effect on a specific wind turbine power performance in a
specific site is described as parameter identification procedure. From the existing analyses, the magnitude
of the wind characteristics effect can be assessed and guidelines for specifying the ranges within which the
power curve is to be estimated can be defined. Moreover, the results of the sensitivity analysis facilitate
the normalization procedure that will result in a power curve defined at a specific condition. The specific
condition point is defined according to the measurements, and represents the mean site condition.
9.5 Uncertainty Estimation Procedure

The uncertainty estimation procedures should be revised, keeping the already existing concepts, in order
to account for the additionally needed measurements. Specifically, for the site calibration procedures,
proper uncertainty estimation should be prescribed for all applied calibration formulae.
9.6 Power Curve Measurements Of Wind Farms

The technical background is adequate for establishing the framework of power performance assessment

and verification in complex terrain for a specific wind turbine, operating either as stand alone or within a
wind farm. Although, recommendations for evaluation techniques for complete wind farms are under
development and refinement basic guidelines are given (Hunter R., 1996).
Performance assessment of wind turbine systems may be required for a number of reasons. The interested
party might be a manufacturer, a purchaser/developer, a financier or an academic, and in each case the
scope of interest might be different. A financier for instance might wish to check that a complete
development is performing as predicted by a technical assessor during a due diligence audit, a developer
might wish to confirm that every machine in a farm is performing as wanted even when subject to wakes
from other machines, whilst a manufacturer may wish to benchmark performance in clean wind so that a
basis for setting warranties or for confirming design expectations is established.
The different requirements are not adequately covered by the existing IEC (1997) which deals only with
performance of single machines in winds which are undisturbed by adjacent turbines.
The three levels of performance assessment are summarized in the following table:
IEC 1400-12 Wind Farm Embedded Wind Wind Farm
Turbines
Application Isolated, single Individual turbines within a Complete wind farm
turbine wind farm
Nature of ‘valid’ Undisturbed by All wind directions valid, All conditions valid, no
wind during test adjacent machines or including those during which rejection criteria
obstacles machine is in the wake of
others
Loss mechanism None None Wake and on-site
included electrical losses are
reflected in the measured
performance
Typical purpose of Verification of Verification of product Verification of system
test design
Typical interested Manufacturer Developer (purchaser) Financier
commissioning party (supplier)
The aim is to define the real not idealized performance by including data from all operational conditions.
Unlike the standard method for performance determination of individual wind turbines, the proposed
method does not reject data when machines are in. the wakes of others, when the wind comes from certain
directions or when particular climatic effects are present. The method looks at the wind farm energy
delivery at the point of common coupling with the grid therefore deals with energy production net of
electrical losses in the in-site electrical collection system.
The procedure is of use to those who wish to determine or verify the overall performance of a wind farm
rather than the performance of individual machines. A wind farm developer or financier for example may
wish to apply it to verify modeling assumptions made at the wind farm planning stage.
The method does not specify accuracy requirements but does require a thorough uncertainty analysis to be
conducted to supplement the results. In figure 9.1 the comparison methodology for real and expected wind
farm performance is presented.
In a typical wind farm in complex terrain, wind speed, wind direction and density will have a spatial
dependency and what is required will be the relevant values at a reference site. This reference position
may not be one at which measurements are possible once the wind farm is operational, and correlation will
have to be established prior to the development to allow appropriate corrections to be applied. In figure

9.2 the required setup is presented. Mast M1 is a permanent meteorological mast, probably some
considerable distance away from the wind farm site which has a long term wind data archive available.
Mast M2 is a meteorological mast which is erected at the wind turbine site near to erection of the wind
farm. It has probably been used to establish a short term data archive, of say 6 months from which long
term predictions of wind climate have been made using correlations with M1 and the long term data
archive from M1. M2 is probably removed shortly before construction of the wind farm. Masts M3 and
M4 (the supplementary masts) are deployed prior to construction of the wind farm specifically as part of
the procedure described in this document, and data from them during this period are correlated with
simultaneous data from mast M2. After construction of the wind farm M3 and M4 are left on site and are
used to supply data from which estimates can be made of the wind speed, wind direction and air density
that would have been present at M2 had the wind farm not been in place.
The function served by M2 in this example (i.e. providing measurement data for the reference position
during the pre-construction, site calibration phase) may equally be served by either M3 or M4, if their
positions are deemed suitable as reference in subsequent energy yield tests. In this example, two
permanent masts M3 and M4 have been suggested. In practice any number of masts could be used, the
requirement being that at all times at least one mast must experience winds which are undisturbed by the
wind farm.
An analysis should be conducted for each supplementary meteorological mast position. Each analysis uses
data from the relevant supplementary mast and data from the reference mast. The ten minute averaged
wind speed data should be sorted according to ten degree sectors, the direction being as indicated by the
data set from the reference mast.
The data set for each directional interval can be regarded as being complete if:
• the mean wind speed of all the collected data is representative of what is expected during the
subsequent performance evaluation
• the statistical uncertainty of the inferred reference wind speed at this wind speed from the regression
factors is below 2%.
A brief overview of the calibration of test site, regarding wind speed and wind direction data, is presented
in figures 9.3 and 9.4.
The aim of the measurement procedure is to collect a database of sufficient quantity and quality to
determine to the desired level of accuracy what the power production capabilities of the wind farm are for
individual directional sectors. These power production capabilities are expressed in terms of the wind
speeds experienced and the prevailing air densities. The collected database shall cover the complete
operational range of the wind farm both in terms of wind speed and wind direction. This range should be
specified prior to initiation of the test campaign.
For each directional sector, the ten minute mean wind speeds should be sorted according to the method of
bins into 0.5m/s intervals (figure 9.5). Finally, annual energy production of the wind farm can be carried
out if reference wind speed distributions are available for each directional sector and if the probability
distribution of wind direction is also specified. For each sector the energy production should be derived
according to the procedures contained in IEC (1997). The overall energy production should be estimated
by summation of the individual directionally weighted contributions.

Figure 9.1 Comparison methodology for real and expected wind farm performance.
Figure 9.2 Correlation required for determining wind climate at a reference position.

Figure 9.3 Analysis procedure for calibration of test site.

Figure 9.4 Analysis procedure for wind direction data.
Figure 9.5 Wind farm directional power curve.


10. REFERENCES
Albers A., Hinsch C., Gabriel J., Klug H., Ronsten G., Simonssen B. (1996): “Comparison of power
performance and noise between AEOLUS II and NASUDDEN II”, Proc. EWEC’96, Goteborg Sweden,
pp.804.
Alm L.K., Nygaard T.A. (1993): “Flow over complex terrain estimated by a general purpose Navier-
Stokes solver”, Proc. EWEC’93, Travemunde, Germany, pp.591.
Antoniou I., Pedersen T.F. (1997): “Nacelle anemometer on a 1MW wind turbine”, RISO-R-941(EN).
Antoniou I., Petersen S.M. (1995): “Wind Turbine Test: Structural Loads ELKRAFT 1MW (Stall
Regulated Operation)”, Risø-I-865(EN).
Barnard J.C. (1990): “An evaluation of three models designed for siting wind turbines in areas of complex
terrain”, Report PNL—7357, 6/90.
Bergeles G. C. (1985): “Numerical calculation of turbulent flow around two dimensional hills”, J. Wind
Eng. and Ind. Aerod., Vol.2l, pp.307-320.
Bergstrom H. (1995): “Wind measurements at Lyse wind power station”, Wind Energy report WE95:2,
Department of Meteorology, Uppsala University.
Botta G, Castagna R., Borghetti I., Mantegna D. (1992): “Wind analysis on complex terrain: The case of
Acqua Spruzza”, J. of Wind Eng. and Ind. Aero., v.39, pp.357-366.
Burger, Franke J., Tetzlaff G. (1988): “Wind measurements in the wake of obstacles”, EWEC’88,
Herning.
Chien H., Meroney R.N., Sandborn V.A. (1980): “Sites for wind power installations: Physical modeling
of the wind field over Kahuku Point, Oahu, Hawaii”, Wind Energy Systems 1980, pp.75-90.
Christensen C.J., Dragt J.B. (1986): “Accuracy of power curve measurements”, RISO-M-2632,
November.
Courtney M., Hoestrup M., Jensen I. (1990): “The Hjardem escarpment experiment”, EWEC’90, Madrid,
pp.71-75.
Croba D. et al. (1996): “Improving techniques for statistical and physical modelling of wind resources in
complex terrain”, Proc. ECWEC’96, Goteborg, Sweden, pp.514.
Cuerva A. (1996a): “Multivariable analysis of power in complex terrain (MADE AE-30 wind turbine at
Monteahumada wind farm at Tarifa)”, CIEMAT/43E11-96/014.
Cuerva A. (1996b): “Multivariable analysis of power in flat terrain (MADE AE-30 wind turbine at
Cañada de la Barca wind farm at Fuerteventura)”, CIEMAT/43E11-96/013.
Cuerva A. (1996c): “Studies on the power output of a MADE AE-30 operating on complex terrain.
Annual energy production estimation and multivariable analysis”, CIEMAT/43E22-96/001.
Cuerva A. (1997): “Some preliminary results on uncertainties in AEP for the same machine operating in
complex and flat terrain (EWTS-II.IV 3rd meeting)”, CIEMAT/43E22/, 2/97.
Cuerva A., Benzdenejnykh N., Perales J.M., Sanz A. (1997): “Assessment of performances of ultrasonic
anemometers as one step ahead in wind measurements of energy production of a wind turbine”,
EWEC’97, Dublin.
Curvers A., Pedersen T.F. (1989): “Recommendations for a European Wind Turbine Standard on
Performance Determination, ECN-217, 7/89.
Dahlberg J. (1996): “Mast, boom and velocity gradient effects on cup anemometers”, FFA report FFAP-
V-009, 2/96.
Dekker J., de Groot C.M. (1986): “Measurement program for the 45m horizontal axis wind turbine”,
ECN-86-149 (EWEC’86).
Elliot D.L. (1990): “Effects of wind shear and turbulence on wind turbine power curves”, European
Community Wind Energy Conference ‘90, Madrid, pp.79 - 83.

Emeis S., Frandsen S. (1993): “Reduction of horizontal wind speed in a boundary layer with obstacles”,
Boundary Layer Meteorology, v.64, pp.297-305.
EWTS (1996a): “Wind speed measurement and site calibration”, JOU2-CT93-0387 Final report, vol.5.
EWTS (1996b): “Electrical power quality measurement procedure”, JOU2-CT93-0387 Final report, vol.7.
Foussekis D. (1996): “Characterization of WINCON 110XT wind turbine operating at CRES test station”,
MOUNTURB report, CRES.
Frandsen S. (1992): “On the wind speed reduction in the center of large clusters of wind turbines”, J. of
Wind Eng. and Ind. Aero., v.39, pp.211-265.
Frandsen S. (1994): “Uncertainties on power performance testing related to obstacles around test site”,
Windpower’94 Minneapolis, pp.523 - 531.
Frank H.P., Landberg L. (1996): “Modeling the wind climate over Ireland”, Proc.ECWEC’96, Goteborg,
Sweden, pp.631.
Fulsang P. (1996): “Separationsmønsteret for et LM19.1 Blad”, , Resultatblad AED_RB_2, Risø.
Glinou G. (1996): “Wind flow modeling: A literature review”, CRES internal report.
Glinou G., Morfiadakis E., Koulouvari M. (1996): “Analysis of full scale measurements for the
investigation of the turbulence structure acting on a rotor disk over complex terrain”, Proc. EWEC’96,
Goteborg Sweden, pp.639.
Glocker S. (1996a): “Sensitivity of cup anemometer to tilt angle, results and consequences”, Presentation
at MEASNET expert group meeting on 2-3/12/96, WINDTEST report.
Glocker S. (1996b): “Summary report on power performance of Nordtank NT500/37 and Vestas V27
data”, EWTS-II.IV WINDTEST technical note.
Glocker S. (1997): “Sensitivity to tilt angle of the reference anemometers of the MEASNET Institutes”,
Presentation at MEASNET expert group meeting on 6-7/3/97, WINDTEST report.
Hansen M., Pedersen M.B. (1990): “Influence of the meteorology mast on a cup anemometer”, Technical
University of Denmark, internal document.
Hassan U., Lindley D. (1982): “Preliminary results of a survey of the Burgar Hill, Orkney, wind turbine
site”, Wind Energy Systems 1982, Stockholm, pp.217-227.
Hauf T., Hauf G. (1982): “The turbulent wind flow over an embankment”, Boundary Layer Meteorology,
v.24, pp.357-369.
Hinsch C., Westermann D. (1996a): “Comparison of different power curve measurements at different sites
through AEP estimation for NTK500/37 and V27”, EWTS-II/4 Technical note, DEWI report, 8/96.
Hinsch C., Westermann D. (1996b): “Power Performance Measurement by means of an anemometer
located on the nacelle”, DEWI Technical Note 19/8/96.
Hojstrup J. et al. (1996): “Wind resources in the Baltic Sea”, Proc. ECWEC 96 “, Goteborg, Sweden,
pp.510.
Hunter R. (1996): “Outline procedure for performance evaluation of wind farms”, IEC TC88 WG6 report,
5/96.
Hunter R. (1996a): “Measurement summary on Nordtank NTK500/37 (Turbine A) power curve
measurement”, NEL report, version 3, 12/96.
Hunter R. (1996b): “Measurement summary on Nordtank NTK500/37 (Turbine B) power curve
measurement”, NEL report, version 2, 12/96.
Hunter R. (1996c): “Parameter identification on Nordtank NTK500/37 power performance”, NEL
technical note, 12/96.
IEA (1990): “Recommended Practices for Wind Turbine Testing and Evaluation, 1. Power Performance
Testing”, 2nd Edition, Risφ National Laboratory.
IEC (1997): “Wind turbine generator systems. Part 12: Wind turbine power performance testing”,
88/FDIS, 1400-12 Edition 1.
Jackson P.S., Hunt J.C.R. (1975): “Turbulent wind flow over a low hill” Q.J.R. Meteorol. Soc., vol.101,
pp.929-955.
Jamieson N., Anderson C.G. (1988): “Power performance measurement on the HOWDEN 33m wind
turbine”, Wind Energy Conversion ‘88 (BWEA, London, 22-24 March 1988), pp.75-79.

Kleinbaum K., Kupper L. (1985): “Applied regression analysis and other multivariate methods”, PWS-
KENT Publishing Company, Boston.
Kline J.B., Mikhail A. (1993): “Recent experience with the NOABL model in highly complex terrain”,
Windpower’93, San Francisco, pp.226-233.
Klug H. et al. (1996): “Wind speed measurement and site calibration”, Final report European Wind
Turbine Standards, JOULE-II project JOU2- CT93-0387, 2/96.
Lalas D.P. et al. (1994): “A hybrid micrositing model for wind flow simulation over complex
topographies”, Proc. ECWEC’94, Thessaloniki, Greece.
Landberg L. (1996): “Implementing prediction of power from wind farms at Utilities”, Proc.1996,
Goteborg, Sweden, pp.592, EWTS II-V reference 35.
Larsen A., Oort H. (1989): “Wind turbine wake interference: A validation study”, Report DMI--89836.
Liu et al. (1987): “Field investigation of a wake structure downwind of a VAWT in a wind farm array”,
Flow Research Report no.427, 10/87.
Magnusson M. (1994): “Full scale wake measurements at Alsvik wind farm”, EWEC’94, pp.472-477.
Mason P.J, King J.C. (1985): “Measurements and prediction of flow and turbulence over an isolated hill
of moderate slope”, Q.J.R. Meteorol. Soc., vol.111, pp.617-640.
Mason P.J., Sykes R.I. (1979), Q.J.R Meteorol. Soc., vol.105, pp.383-395.
McCarthy E., Meroney R., Neff D.E. (1993): “Elevation and vegetation considerations on wind power
availability: A wind tunnel study”, Windpower’93, San Francisco, pp.213-217.
Mengelkamp H.T. et al. (1996): “Regional and local wind climatologies over heterogeneous terrain”,
Proc. ECWEC’96, Goteborg, Sweden, pp.506.
Morfiadakis E. et al. (1997): “Assessment of power performance in complex terrain”, EWEC’97, Dublin,
10/97.
Morfiadakis E., Glinou G., Stefanatos N. (1994a): “Power performance evaluation of a wind turbine
operating in a complex terrain wind farm”, Proc. EWEC’94, Thessaloniki Greece, pp.906.
Morfiadakis E., Mouzakis F. (1996b): “Nacelle cup anemometry for WINCON 110XT operating at CRES
test site”, CRES report, 2/96.
Morfiadakis E., Mouzakis F. (1996c): “Power performance of NORDTANK NT500 operating at Toplou
site”, CRES report.
Morfiadakis E., Mouzakis F. (1996d): “Power performance of VESTAS V27 operating at Andros site”,
CRES report.
Morfiadakis E., Mouzakis F. (1996e): “Power performance of WINCON 110 XT operating at CRES site”,
CRES report.
Morfiadakis E., Mouzakis F. (1997a): “Power performance parameter identification for NTK500/37”,
CRES report, 2/97.
Morfiadakis E., Mouzakis F. (1997b): “Toplou site calibration assessment”, CRES report, 2/97.
Morfiadakis E., Mouzakis F., Dellaportas P. (1996a): ”Parameter identification of fatigue loading of a
wind turbine operating in complex terrain”, CRES report.
Mouzakis F., Bergeles G. (1991): “Numerical prediction of turbulent flow over a two-dimensional ridge”,
Int. J. for Num. Meth. in Fluids, v.12, pp.287-296.
Mouzakis F., Morfiadakis E. (1996): “Parameter identification practice”, CRES report.
Mouzakis F., Stefanatos N., Glinou G. (1997): “Assessment of site calibration techniques for complex
terrain”, EWEC’97, Dublin, 10/97.
Nielsen M.A. (1994): “Performance testing in inhomogeneous terrain”, EWEC’94 Thessaloniki, pp.911-
915.
Nierenberg A. (1993): “Detailed measurements on the effects of trees on wind speed, energy, vertical
wind shear and turbulence”, Windpower.
Pearse J.R. (1982): “Wind flow over conical hills in a simulated atmospheric boundary layer”, J. of Wind
Eng. and Ind. Aero., v.10, pp.303-313.

Pedersen M.B., Hansen K., Oye S., Brinch M., Fabian O. (1991): “Some experimental investigations on
the influence of the mounting arrangements on the accuracy of cup anemometer measurements”,
EWEC’91.
Pedersen T. (1993a): “A parametric study on terrain effects in relation to power performance testing”,
RISO-I-728(EN), 9/93.
Pedersen T. (1993b): “Recommendation for Wind Turbine Power Curve Measurements”, (ed.), Risø-I-
745(EN), 11/93.
Pedersen T. (1994): “Trends in power performance measurement standards”, Windpower’94,
Minneapolis.
Pedersen T., Petersen S.M., Volund P. (1996): “Description and measurements of the wind that drives the
power performance and mean loads of wind turbines”, EWEC’96.
Petersen E.L. et al. (1996): “Measurements and modelling in complex terrain” .
Petersen S.M. (1993): “Vindmolleafprovning: Nordtank NTK 500/37 Maling af effektkurve”, RISO-I-
731(DA), 9/93.
Petersen S.M. (1994a): “Konceptundersogelse: Nordtank NTK 500/41 Maling af effektkurve”, RISO-I-
799(DA), 12/94.
Petersen S.M. (1994b): “Konceptundersøgelse NORDTANK NTK500/41, Structurelle laster -
Tårnbøjning”, Risø-I-786(DA).
Petersen S.M. (1994c): “Vindmolleafprovning: Vestas V39 Maling af effektkurve”, RISO-I-754(DA),
2/94.
Petersen S.M., Pedersen T.F., Paulsen U.S, Morfiadakis E., Koulouvari M., Mouzakis F. (1996a):
“Comparison of performance and load characteristics of two V27-225 operating in different complex
terrain sites”, Proc. EWEC’96, Goteborg Sweden, pp.995.
Petersen S.M., Pedersen T.F., Paulsen U.S., Morfiadakis E., Koulouvari M., Mouzakis F. (1996b):
“Comparison of performance and load characteristics of two V27-225kW operating in different complex
terrain sites”, EW EC’96.
Riddle A., Warren J., Hoskin R., Musgrove P., Lindley D. (1996): “The influence of complex terrain on
the performance of windfarms”, EWEC’96, Goteborg, pp.1077.
Ronsten G., Dahlberg J. (1996): “Calibration of anemometers in FFA-L2 wind tunnel including twin
calibration and tests of sensitivity to wind velocity gradient and tilt”, FFA TN 1996-51.
Sandstrom S. (1994): “WASP A comparison between model and measurements”, EWEC’94.
Sheinman O., Rose A. (1992): “A dynamic model of the influence of turbulence on the power output of a
wind turbine”, J. of Wind Eng. and Ind. Aero., vol.39, pp.329 - 341.
Sherman C. (1978): “A mass-consistent model for wind fields in complex terrain”, J. Appl. Meteorol.,
vol.17, pp.312.
Sierputowski, Ostrowski, Cenedese (1995): “Experimental study of wind flow over the model of a valley”,
J. of Wind Eng. and Ind. Aero., v.57, pp.127-136.
Slater J.T.D. (1996): “Power Performance Measurement Techniques for Wind Farms”, , RES Document
500/RES/1117.
Smedman A., Bergstrom H., Hogstrom U. (1996): “Measured and modeled local wind field in
mountainous terrain”, EWEC’96.
Taylor G.I. (1991): “Wake measurements over complex terrain”, BWEA’91, Swansea.
Taylor P.A. (1977): “Numerical studies of neutrally stratified planetary boundary layer flow above gentle
topography. Two dimensional cases”, Bound. Layer Meteorol., vol.12, pp.47-60.
Tombrou M. et al. (1993): “Tests of prediction effectiveness of wind energy computer models in complex
terrain”, Proc. ECWEC’93, Travemunde, Germany, pp599.
Traci R.M. (1977): “Development of a wind energy site selection methodology”, US Dept. of Energy
Rept. RLO/2440-11, pp.205.
Troen I., Petersen L. (1989): “European wind atlas”, ISBN 87-550-1428-8.
Tryfonopoulos D.A. et al. (1989), Wind Engineering, vol.13, pp.324-337, 1989.

Van der Borg N. (1996): “Inventory of existing information on the issue of power performance
measurements in complex terrain”, ECN-C—96-083, 10/96.
Van der Snoek L. (1990): “Wake effects in wind farms and behind obstacles”, ECN report.
Veenhuizen S.D., Lin J.T. (1987): “Wind turbine micro-siting status”, Windpower’87.
Vermeulen F. (1982): “A wind tunnel study of the flow field over the “SWECS” test site at ECN Petten”,
ECN-report 82-014908, 11/82.
Vermeulen H. McLean I.R., Flay R. (1982), Wind Energy Systems ‘82, Stockholm, pp.195-216.
Wade J.E., Walker S.N. (1987): “The effects of local terrain on wind characteristics and turbine
performance”, Windpower’87, pp.398 - 401.
Walmsley J.L. et al. (1986), Bound. Layer Meteorol., vol.36, pp.157-186.
Walmsley J.L., Troen I., Lalas D., Mason N. (1989): “An inter-comparison of models for wind flow in
complex terrain and data from the Blashaval experiment”, EWEC’89.
Watson R. (1994): “Wind measurement and modeling in the republic of Ireland”, EWEC’94.
Wendel L., Barnard J.C., Morris V.R. (1994): “A proposed wind measurement and analysis approach for
evaluating a prospective wind plant site”, Windpower’94.
Weng W., Walmsley J.L., Karpik R.S., Xu D., Taylor P.A., Ayotte K., Salmon J.R. (1994): “Application
of the MSFD and NLMSFD models to airflow over Askervein Hill”, EWEC ‘94.


11. ACKNOWLEDGEMENTS
The participants of the project are indebted to DGXII for cofunding the project, and to the WT
manufacturers and operators for their help during the WT measurements. CRES is indebted to the GSRT
for cofunding the project and to PPC/DEME for the provision of its WTs for the Greek measurement
programs. The coordinator expresses his gratitude to the institutions that participated in this project and
the project leaders and to all members of the scientific teams participating for their commitment to the
project and their invaluable contributions.

PART 5
SITE EVALUATION
Alan Derrick (Editor) NEL

Ioannis Antoniou RISØ National Laboratory
Arno Brand ECN
Sten Frandsen RISØ National Laboratory
Georgia Glinou CRES
Thomas Pahlke DEWI
Ignacio Marti Pérez CIEMAT
Ian Ravey NEL
Bärbel Schwenk DEWI
Site Evaluation

Site Evaluation
CONTENTS OF PART 5
1. INTRODUCTION ..................................................................................................................................... 5
1.1 What is Meant by “Site Assessment/Evaluation”.................................................................................... 5
1.2 Relationship to Standards ........................................................................................................................ 5
2. SURVEY OF EUROPEAN CONSULTANTS ......................................................................................... 7

2.1 Purpose of Survey.................................................................................................................................... 7
2.2 Implementation ........................................................................................................................................ 7
2.3 Results of Assessment Technique Survey ............................................................................................... 7
2.3.1 Denmark ........................................................................................................................................... 7
2.3.2 Germany ........................................................................................................................................... 8
2.3.3 Greece ............................................................................................................................................... 8
2.3.4 Netherlands ....................................................................................................................................... 9
2.3.5 Spain ................................................................................................................................................. 9
2.3.6 United Kingdom ............................................................................................................................. 10
2.3.7 Overall ............................................................................................................................................ 11
2.4 Results of Input Data Survey................................................................................................................. 12
2.4.1 Denmark ......................................................................................................................................... 12
2.4.2 Germany ......................................................................................................................................... 13
2.4.3 Greece ............................................................................................................................................. 13
2.4.4 Netherlands ..................................................................................................................................... 13
2.4.5 Spain ............................................................................................................................................... 13
2.4.6 United Kingdom ............................................................................................................................. 14
2.4.7 Overall ............................................................................................................................................ 14
3. MEAN WIND SPEED ASSESSMENT .................................................................................................. 15

3.1 Wind Flow Modelling ........................................................................................................................... 15
3.1.1 Technology Status Review ............................................................................................................. 15
3.1.2 WAsP.............................................................................................................................................. 17
3.2 Measurement based techniques ............................................................................................................. 22
3.2.1 Status Review ................................................................................................................................. 22
3.2.2 MCP................................................................................................................................................ 24
3.3 Assessment of Uncertainty .................................................................................................................... 25
3.3.1 Definition of extended (long-distance) site calibration .................................................................. 26
3.3.2 Test parameters ............................................................................................................................... 26
3.3.3 Uncertainty components ................................................................................................................. 27
3.3.4 Additional Uncertainties due to Long-Term Climate Variability................................................... 30
3.3.5 Conclusions on Uncertainty............................................................................................................ 33
4. OTHER PARAMETERS......................................................................................................................... 34
4.1 Reference Wind Speed .......................................................................................................................... 34
4.1.1 Description of Parameter ................................................................................................................ 34
4.1.2 Methods of Assessment .................................................................................................................. 34
4.1.3 Instrument requirements ................................................................................................................. 38
4.1.4 Further Reading .............................................................................................................................. 38

Site Evaluation
4.2 Reference Turbulence Intensity ............................................................................................................. 39

4.2.3 Measurement Requirements ........................................................................................................... 41
4.2.4 Further Reading .............................................................................................................................. 43
4.3 Wind Speed Distribution ....................................................................................................................... 43
4.3.4 Further Reading .............................................................................................................................. 51
4.4 Normal Wind Profile ............................................................................................................................. 51
4.4.4 Method to Determine which Profile Gives the Higher Loading..................................................... 53
4.4.5 Further Reading .............................................................................................................................. 55
4.5 Wind Direction Distribution .................................................................................................................. 55
4.6 “Other” Parameters ................................................................................................................................ 55
4.6.1 Overview ........................................................................................................................................ 55
4.6.2 Turbulence Length Scale ................................................................................................................ 56
4.6.3 Coherence ....................................................................................................................................... 57
4.6.4 Standard Deviation Ratios .............................................................................................................. 57
4.6.5 Negative Gust ................................................................................................................................. 58
4.6.6 Instrumentation Requirements........................................................................................................ 59
4.6.7 Special Instrumentation: The Ultrasonic Anemometer................................................................... 59
4.7 Clarification of Standard........................................................................................................................ 62
5. REFERENCES ........................................................................................................................................ 64
Appendix A Assessment Of Other Parameters - Mathematical Details ...................................................... 68
Appendix B Use Of Sonic Anemometer...................................................................................................... 72

Site Evaluation
1. INTRODUCTION
1.1 What is Meant by “Site Assessment/Evaluation”
Over the ten to fifteen years that commercial wind farm development has been a reality throughout
Europe, the term “site assessment” or “site evaluation” has generally tended to be associated with the
prediction of the energy yield potential of a site prior to the installation of wind turbines. Hence,
historically, the evaluation has primarily been restricted to the assessment of the site annual average wind
speed and wind speed distribution. The method and thoroughness of the assessment has varied from
country to country depending to a certain extent on the requirements of national subsidy schemes and the
uncertainty inherent in the evaluation due for instance to the nature of the terrain and wind regime.
However, as the economic and design margins on wind turbines and wind farms become tighter, the need
to quantify and minimise risk has become more important. The site evaluation is at the core of this risk
evaluation process and consequently the site evaluation needs to consider more than just energy yield
potential if other aspects such as turbine integrity are to be assessed.
1.2 Relationship to Standards

Whether or not the prediction of the annual energy yield potential for a proposed wind farm site is a valid
subject for standards is a debatable point, however the safety, structural integrity and fitness-for-purpose
issues associated with specifying or designing specific wind turbines for a site most certainly are
important subjects for standards to address. In that context, the IEC has recently revised their wind turbine
safety standard [1] to (amongst other things) consider such issues more thoroughly.
In the terminology of the original IEC standard [1], the site evaluation contributes to the specification of
the external conditions. The external conditions are categorised into four pre-defined wind turbine classes
each of which is characterised by three basic parameters. A class I turbine is designed for more severe
external conditions than a class II turbine and so on down to class IV (an additional special class exists for
external conditions specified by the designer). The three basic parameters defining the external conditions
in the original version of the standard were:
• reference (extreme) wind speed (Vref)
• annual average wind speed (Vave)
• annual average turbulence intensity (Iave)
Of those parameters, a minimal site evaluation for the purpose of determining energy yield potential
would only yield information on the annual average wind speed. The onus to provide a more
comprehensive site evaluation is inherent in the following statement extracted from the standard [1]:
“It is the responsibility of the project engineer or the equipment owner to verify that conditions at the
actual installation site, including the increased turbulence within turbine arrays, are more benign than
those prescribed for the given turbine class”.
The three basic parameters are used to derive a number of additional external condition parameters which
allow the structure of the wind to be defined in more detail. These derived parameters may also form part
of a site evaluation. However, no guidance is given in the standard on how this or any other part of the site
evaluation should be accomplished.
A greater emphasis has been placed on site evaluation in the latest edition of the IEC safety standard [2]
with a complete chapter (Chapter 7 Assessment of External Conditions) now added. The following
statement introduces that chapter,

Site Evaluation
“It shall be assessed that the environmental, electrical and soil properties are more benign than those
assumed for the design of the WTGS. If the site conditions are more severe than those assumed, the
engineering integrity shall be demonstrated”.
As a minimum the three basic parameters have to be assessed by measurement, long term records or local
codes or standards. The three basic parameters in this edition [2] of the standard are:
• reference (extreme) wind speed (Vref)
• annual average wind speed (Vave)
• characteristic turbulence intensity (I15)
Note that the turbulence intensity parameter is different to that specified in the earlier version of the
standard.
The following additional recommendations are made in the standard:
• at least 6 months of reliable data shall be used for the evaluation.
• where seasonal effects on wind conditions are significant, the monitoring period shall include these
effects.
• particular care shall be taken in evaluating the characteristic turbulence intensity such that only wind
speeds greater than 10 m/s shall be used, low frequency trends must be removed and appropriate
anemometers, sampling rates and averaging times shall be used.
• for complex terrain, the assessment must be based on measurements.
However, except in the case of the characteristic turbulence intensity, no detailed guidance is given on
how the assessment should be carried out.
This report aims to give guidance on how such an assessment can be carried out in a consistent manner.
The task has been restricted to the assessment of wind conditions, reflecting the expertise of the
participants in the project and concentrating on that aspect which is particularly important to wind turbine
design and which is not likely to be well covered in codes associated with other fields of engineering.
Consideration to the effects of wakes from other wind turbines has not been given as this is dealt with in
another part of the EWTS 2 project.

Site Evaluation
2. SURVEY OF EUROPEAN CONSULTANTS

2.1 Purpose of Survey
All of the European nations presently involved in major commercial wind farm development were
represented on the project (Denmark, Germany, Greece, Netherlands, Spain and United Kingdom). Even
within the project participants, it was apparent that site assessments of varying extent and methodology
were in use in each country. Besides the project participants it was recognized that there were many other
organizations involved in site assessment and consequently it was decided to carry out a country-by-
country survey by questionnaire to obtain a more complete overview of site assessment techniques in use.
The first part of the data collection exercise was aimed at identifying the range of techniques that are used
in each of the countries participating in this project.
A second aspect was to investigate the sources and availability of input data for use with the site
assessment techniques. Both the measurement and modeling based techniques which are used for site
(annual average wind speed) assessment require some form of third party input data but it was expected
that the quality and range of data readily available would vary from country to country. The intended
outcome of this part of the survey was an indication of the “lowest-common-denominator” of available
input data which could be recommended for use with each assessment technique.
2.2 Implementation
To encourage a good response, most of the questionnaires were completed by telephone interview with
each organization rather than by post. In most if not all cases, only organizations involved in commercial
wind farm site assessment have been interviewed, therefore there are generally no research projects or
techniques contained within the survey results. The survey focussed on techniques for annual average
wind speed assessment as in the majority of cases this was the only parameter which was assessed in any
thorough way or even at all until now. However, some basic questions were asked to determine if other
parameters were assessed and if so what they were. Finally, the survey was carried out on the basis that
individual consultants and organizations would remain anonymous. In fact, only the person asking the
questions within each country knew who the individual organizations were as all paperwork distributed
and analyzed centrally within the project contained no indication of the responding organizations.
2.3 Results of Assessment Technique Survey

The results are first of all summarized and interpreted on a country by country basis. The overall results
are presented and summarized in tabular form at the end of this section.
2.3.1 Denmark
Five responses were received from the survey of consultants in Denmark, all of whom mainly used wind
flow models for site mean wind speed assessment. This reflects the fact that most of these consultants
primarily operate in their home market where the models are well tested and the terrain is relatively
simple. However, at least two Danish consultants operate in countries with more complex terrain and in
these situations either wind flow models or measurement based techniques are used.
All of the consultants use the RISØ WAsP wind flow model for wind speed assessment and spatial
extrapolation. Three consultants also use another Danish model called Wind Atlas System originating
from Energi- og Miljødata Ålborg. The purpose of applying the models is generally concerned with
resource (energy yield) assessment rather than structural integrity assessment. The experience of the

Site Evaluation
Danish consultants indicates that in Denmark, predictions are within 10 per cent of actual energy yields in
most situations and indeed one consultant has evidence from more than 100 sites over periods of 3 to 6
years which indicate an average prediction error of 1 per cent and a standard deviation of 5 per cent. Using
similar techniques in other countries, the same consultant assesses the prediction error to be 5 per cent. In
general, the consultants assume an uncertainty of about 10 per cent on energy yield for Danish sites rising
to 18-25 per cent for complex terrain sites.
Considering all of the consultants, the digitized representation of the terrain used with the models ranges
from a radius of 4km to 10km depending on the memory limits of the model and hence on the complexity
of the terrain height contours. Contour intervals of 2m to 5m are used in all cases and as terrain data are
available in digital format in Denmark their representation should be consistent irrespective of the
consultant.
The representation of obstacle and surface roughness in the models is more subjective. Each consultant
has a different approach to the inclusion of obstacles and what size and position of obstacle is considered
significant. However, in general roughness changes are considered to a radius of 20km (which seems
inconsistent with the terrain contour domain dimensions).
Although site specific wind measurements are not a legal or contract requirement in Denmark, significant
measured data exists on the performance statistics of existing wind farms. Therefore, it is relatively
straightforward and common practice to confirm or validate energy yield or wind speed assessments from
a proposed wind farm site on the basis of the performance of neighbouring wind farms or wind turbines.
2.3.2 Germany
Reflecting the level of growth of wind energy and legal requirements in Germany, the greatest number of
consultants and other organizations involved in site assessments are located there. Forty responses were
received on time although there were about sixty enquires made in total.
Terrain ranging from flat coastal regions, through moderately complex inland regions to complex
mountainous terrain exists in Germany and hence a broad range of assessment techniques are used.
However, the majority of consultants use a combination of modelling and measurement.
2.3.3 Greece
The responses of six developers and consultants were obtained from Greece. Additionally, a small group
of old private developers, who installed wind turbines based on visual inspection of the site and using
information from locals, were also identified. However, in these cases it was concluded that they adopted
no reasonable site assessment technique within the concept of the present survey and so they were
excluded.
The site assessment techniques used in Greece reflect the complexity of the Greek terrain (categorized as
ultra complex terrain) and the wind flow uncertainties inherent therein. Therefore, most consultants use
measurement based techniques. Half of the consultants use computational wind flow models in addition to
or instead of measurement based techniques.
Current government regulations and subsidy schemes require an energy yield assessment to be carried out
before an operation license and subsidy are granted. Hence the motivation for site assessment is primarily
financial in terms of energy revenue rather than structural integrity. Despite this, most of the consultants
do measure more parameters than just the bare minimum required to carry out an energy yield assessment
(e.g. turbulence intensity and shear are also typically determined by measurement in addition to mean
wind speed and direction).

Site Evaluation
In most cases (67 per cent), the measurements are used in conjunction with long term reference wind
speed data to carry out a long-term prediction using an empirical correlation technique commonly known
as Measure-Correlate-Predict (MCP). In such cases the on-site measurements are of duration’s ranging
from 6 months to 3 years, depending on the consultant, with most consultants typically using a minimum
of 1 year of measurements.
Unusually (at least in comparison with other countries), no two consultants use the same wind flow model,
with only one consultant using the commercially popular WAsP model and the other two each using
different in-house numerical wind flow models.
2.3.4 Netherlands
The results contain the responses of nine out of sixteen commercial consultants identified, the remainder
either not responding within the time scale or not willing to participate. In general the assessment
techniques used in the Netherlands reflect the fact that the terrain is very flat and that considerable local
experience of the wind characteristics and using the wind has been established over many decades.
Therefore, the total picture is biased towards modelling where the term “modelling” is used in a broad
sense to include local calculation codes and procedures.
A large variety of modelling based techniques are used in the Netherlands, ranging from assessments on
the basis of the production of nearby wind turbines to the application of methods which take into account
reference climatalogical data. The method which is used most is “Handbook Method III” [4], in which the
energy yield of a wind turbine at a specific site is predicted on the basis of the local distribution of wind
speed and the power curve of the wind turbine, the former being derived using formula and tables from the
handbook. The local wind speed distribution at hub height is predicted on the basis of 30 year wind data
from four reference meteorological stations which each represent a different climatological zone
(maritime, coastal, lowland and inland). Large scale differences between the target and reference site wind
regimes due to “macro” (i.e. roughness independent) wind speed, climatological differences and “meso”
roughness effects are determined. Next, local effects are taken into account in terms of local roughness,
hub height, obstacles, other single wind turbines and wind farm corrections. Finally, a stability corrected
wind speed is determined. The method is restricted to hub heights less than 80m and locations excluding
sites at dikes or offshore. Some of the consultants using this technique apparently assess as many as 125
sites per year.
Experience and specific studies have shown that the “Handbook Method III” generally over predicts
annual energy yield by 25 ± 25 per cent reflecting a possible average annual wind speed error of up to 8
percent. Much of the variability is due to the fact that the reference wind regime period and hence
prediction period is of 30 years duration which does not match well the annual average conditions used for
the comparisons. However, some organizations have modified “Handbook Method III” in some
unspecified ways to reduce the errors associated with certain components of the model.
The general feeling in the Netherlands is that, due to the flat terrain, measurements will not lead to a
reduced uncertainty compared to existing modelling techniques unless the measurements are carried out
over a long period (3 years). However, there is a trend towards more measurements particularly for large
projects where there is a trade-off between the revenue from the expected yield for the wind farm and the
cost of the measurement campaign. The motivation for site assessment in the Netherlands is connected
with accuracy of energy yield predictions and hence potential revenue rather than turbine safety are the
key issues addressed.
2.3.5 Spain
Three Spanish organizations responded to the survey. As with Greece, the terrain used for wind farms in

Site Evaluation
Spain generally ranges from complex to ultra complex, hence assessments are in all cases based on a
thorough combination of measurement and modelling.
All of the consultants carry out measurements which are more comprehensive than those required to
determine annual mean wind speed with the majority assessing shear, turbulence and gusts in addition to
mean wind speed and direction. Two out of the three consultants also use multiple measurement masts
with or without wind flow models to carry out spatial extrapolation across a proposed wind farm site.
In all three cases the long-term prediction of annual average wind speed is based on a statistical Measure-
Correlate-Predict (MCP) technique. However, there is insufficient operational experience to assess how
accurate the predictions based on this technique have been so far. Uncertainties of up to 15 per cent on
wind speed are however assumed by two of the consultants on the basis of short term validation exercises.
Where wind flow models are used for spatial extrapolation, all three organizations use WAsP and in
addition two of the organizations also use MATHEW . Although similar models are used by all three, the
method of using the models seems inconsistent from consultant to consultant. For example, each
consultant typically uses a terrain domain size of 10km x 10km, 15km x 15km and 40km x 40km.
Surprisingly, the consultant who uses the smallest domain area also uses a coarse height contour
resolution (20m to a radius of 5km and 50m thereafter), whereas at least one other consultant has access to
contour data in the range 5m to 20m.
2.3.6 United Kingdom

Nine consultants and wind farm developers involved in site assessment in the UK were contacted of which
eight responded within the time scales of the survey. A further consultant was also identified but upon
initial contact was excluded from the survey on the basis that the methods used did not constitute a formal
site or wind speed assessment, being based purely on experience. At least partly because of the perceived
uncertainties arising as a result of the relatively complex terrain which characterizes typical UK wind farm
sites, there is a formal requirement to carry out a minimum of 13 weeks of on-site wind measurements in
the UK prior to lodging an application for a wind farm electricity supply contract. Hence measurements
are used by all the UK consultants and in most cases the measurements are used in conjunction with wind
models.
The statistical Measure-Correlate-Predict (MCP) technique is used by all eight UK consultants to
determine the wind farm annual mean wind speed and wind speed distribution. In terms of energy yield
prediction, the majority of the consultants have experience which suggests that the technique is accurate to
within 5 per cent. More than half of the consultants would consider measuring for a minimum of 6-8
months and a maximum of 12 months to provide sufficient data for the correlation stage of the analysis.
One organization typically executes 24 months of measurements for this purpose. No one takes the risk of
using the formal minimum of 13 weeks of measurements. For the correlation stage of the analysis,
somewhat surprisingly, more than half of the consultants use a least squares straight line fit forced through
the origin. Other methods used are unconstrained least squares straight line fit, least squares polynomial fit
and two parameter “method-of-bins” analysis. To complete the MCP process, long-term reference wind
speed distribution data are required and without exception this originates from the UK Meteorological
Office for sites in the UK hence all consultants are consistent in this respect.
For spatial extrapolation of the wind regime across a wind farm site, more than half of the consultants use
a temporary mobile anemometer mast of typically 10m height at various locations on the proposed wind
farm site. These measurements are correlated with the longer term measurement mast on the site and the
results used to validate more comprehensive spatial extrapolations which are implemented using wind
flow models. Of the various wind models, WAsP is the most commonly used followed by the MS family
of models (MS-Micro, MS3DJH). As contour data are available in digital format, most consultants use
these data with 10m contours and a 10 x 10 km domain.

Site Evaluation
2.3.7 Overall
Referring to Figure 2.1 which is based on data extracted from the survey, as the typical terrain complexity
increases, the assessment technique trend moves away from using wind flow models to carrying out on-
site measurements with or without supporting wind flow models.
Increasing Terrain Complexity
100%
90%
80%
Technique Used (%)
70%
60% Measurements
Wind Flow Models
50%
Model & Meas.
40%
Other
30%
20%
10%
0%
DK NL D ES GB GR
Country
Figure 2.1 Prevalence of assessment techniques used per country

This situation reflects the perceived state-of-the-art of wind flow modelling in the commercial world and a
general lack of confidence in the application of such techniques in complex terrain. From the survey
results, a total of 79 consultants were identified to be using models as part of commercial site assessments
(some consultants use more than one model). The breakdown by model is illustrated in Figure 2.2.

Site Evaluation
Unspecified Model
19%
Handbook Method III

5%
MATHEW
3%
GESIMA
4%
WAsP
59%
Wind Atlas System
4%
NOABL Database
4% MS3DJH
MSMicro
1%
1%
Figure 2.2 Summary of wind flow models and calculation methods in commercial use.
As WAsP is clearly the most popular model, a particular emphasis was placed on assessing how
consistently this was being used, how reliable the results were and suggesting guidelines for a more
consistent application of the model.
2.4 Results of Input Data Survey

Various forms of input data are required by both measurement and modelling based prediction techniques.
The survey was used to investigate the sources and format of reference wind speed data and terrain
contour data available within each country.
2.4.1 Denmark
Terrain contours: maps are available at scales of 1:25000 and 1:100000. The 1:25000 maps are available
in both printed and digital formats. The height contour resolution of the printed maps is 2.5m with
additional spot heights to accurately define specific features (e.g. hill tops). However, the digital maps
have a minimum height contour resolution of 5.0m.
Reference wind data: histograms of wind speed and direction are available from a number of sites (exact
number unspecified) operated primarily by the Danish Meteorological Institute. Typically, the data
archives cover periods of 10-20 years although there are some instances where much longer records are
available (e.g. Risø has a 40-year record). According to the consultants, the quality assurance of the data is
restricted to identifying instrument failures and re-calibration of the instruments.

Site Evaluation
2.4.2 Germany
Terrain contours: most of the German consultants hand digitize the height contours from printed maps
although digitized data are also available. From the database of responses, the minimum contour interval
available seems to be 5m.
Reference wind data: DWD, the German Meteorological Office, is the major source of long term reference
data. Sometimes these data are sourced directly from DWD and in other instances the data from the
European Wind Atlas are used although, for German sites, the Wind Atlas data are in fact based on DWD
data. Typically a ten year data record is used. Although high quality instrumentation is used, there is some
doubt over the consistency of calibrations and consideration of siting effects for such instruments. There is
a lack of consistency between the techniques used by the former East and West German meteorological
institutes which further complicates general assessment of the quality of the available data.
2.4.3 Greece
Terrain contours: the most commonly available printed map scale is 1:50000. However, 1:5000 scale maps
are available to special order from the Hellenic Army Geographic Service provided they do not map
restricted areas. The 1:50000 maps have a contour resolution of 20m and the 1:5000 scale maps have a
contour resolution of 4m. The entire country is available in digitized format, however the contour
resolution is only 100m. However, the Hellenic Army Geographic Service will digitize the 1:50000 maps
at 20m resolution to special order.
Reference wind data: there are three sources of measured wind data in Greece; Hellenic National
Meteorological Service (HNMS), Public Power Corporation (PPC) and the Centre for Renewable Energy
Sources (CRES). There are 48 HNMS sites and 9 PPC sites with data records longer than 10 years,
however, the data are generally only available in processed, statistical summary form. In most cases
regular re-calibration of the instrumentation is carried out, ensuring some confidence of data quality.
2.4.4 Netherlands
Terrain contours: printed maps of 1:50000 scale are commonly available for the Netherlands. Due to the
very flat nature of the land and the assessment technique primarily used, information on contour intervals
has not been provided by any of the questionnaire recipients, although one must assume that such contours
are marked on the maps.
Reference wind data: KNMI, the Royal Netherlands Meteorological Office provide long term reference
data from seven stations with records extending for more than 7 years. Four of these sites constitute the
reference sites for the “Handbook Method III” employed by most of the consultants.
2.4.5 Spain
Terrain contours: a variety of printed maps are available with full national coverage at 1:50000 scale,
about 25 per cent coverage at 1:25000 scale and some areas where 1:10000 and even 1:5000 maps are
available. The available contour resolutions depend on the scale of the map and are respectively 20m,
10m, 5m and 2m. Digitized maps are available for some areas based on the 1:50000 scale maps but with
10m contour resolution. However such data are very expensive, so most organizations hand digitize from
printed maps.
Reference wind data: The Spanish Meteorological Office can provide long term wind speed data from a
number of sites of which there are about 15 sites with 10-15 years of data and a further 5 sites with greater
than 15 years of data available. For those sites located at airports, calibration checks are regularly carried
out, however, there is no indication of what is done at other locations. Time series (half-hour interval) data

Site Evaluation
are available along with statistical summaries of wind speed and direction.
2.4.6 United Kingdom

Terrain contours: a range of printed maps are available within the UK at scales of 1:50000, 1:25000 and
1:10000. The 1:50000 and 1:25000 scale maps have contour resolutions of 10m whereas the 1:10000
maps have contours at 5m intervals in and around towns and cities and at 10m intervals in remote areas.
Digitized terrain data based on the 1:50000 scale maps are available for the entire country with 10m
contour resolution and a height accuracy of ±3m. Terrain data based on the 1:10000 scale maps are
available to special order with contour resolutions of 5m or 10m according to the same criteria as the
printed maps. In this case the height accuracy is stated to be ±1.5m.
Reference wind data: wind speed data are available in various formats from the UK Meteorological Office
(MO). There are presently 180 sites operated by the MO throughout the UK from which measured wind
data are available. Most of these sites are concentrated around the coast, near heavily populated areas and
at military and civil airfields. Data are available to purchase in the form of hourly time series (per calendar
month) and histograms of wind speed and direction by year or month. Many of the sites have records
extending back for more than 15 years. The MO carry out checks at regular intervals to ensure data
integrity (no details available).
2.4.7 Overall
Terrain contours: all of the countries considered are abundantly described by printed maps as would be
expected. The resolution of contours available is largely dictated by the severity of the terrain in each
country; the more simple the terrain, then the higher the contour resolution that is required to represent the
undulations in the terrain and so the higher the contour resolution presented on maps. The economics of
map production mean that it is generally not viable to produce maps with high contour resolution (<5m) in
complex terrain countries. In general, maps with at least 10m contour resolution seem to be readily
available across Europe. Digital terrain data are available in most countries although generally at a lower
contour resolution than for the best printed maps and at a perceived greater cost. However, the apparent
purchase cost disadvantage must be weighed against the personnel cost, time, accuracy and consistency
constraints associated with carrying out manual digitization from printed maps.
Reference wind data: as expected, the national meteorological offices in each of the countries considered
are the main source of long-term wind speed records. In the main, the measurement masts are located at
coastal or airport sites. The measurements are generally carried out using cup anemometers at a height of
10-20m above ground level. At this height there is a real possibility of anemometer blockage from
vegetation or buildings, both of which are features which may have changed significantly over even a
relatively short period of time. In the longer term, instrumentation systems may have been replaced and
instrument characteristics may change as the instruments age, although this is accounted for in most cases
by regular re-calibration. All of these factors combine to make the reference wind data a potentially major
source of uncertainty in a site evaluation. It is recommended that this uncertainty is minimized by careful
examination of the history of a reference site and it’s instrumentation and by inspecting the reference site
via a site visit.

Site Evaluation
3. MEAN WIND SPEED ASSESSMENT

3.1 Wind Flow Modelling
3.1.1 Technology Status Review
In order to perform site assessment for wind energy applications especially in complex terrain, an accurate
description of the wind flow over the terrain is needed showing the spatial variability of wind velocity and
turbulence. Proper siting is necessary in terms of expected power output (maximum energy yielded
compared to costs) and wind velocities and turbulence are also important for the selection of the most
suitable wind turbine design in terms of fatigue resistance and strength (maximum lifetime and reliability
compared to costs).
This need for reliable prediction of the wind field over complex terrain has lead during the last decade to
the development of various models and corresponding computer codes which differ both in complexity
and accuracy. These can be divided into a number of basic categories as described in the following
paragraphs.
The kinematic or mass consistent models solve the continuity equation only, assuming that the flow is
inviscid. As a result, they are of low computational cost and easy to handle. This concept was initiated by
Sherman’s model [5], where a non divergent wind field is produced via a perturbation potential field.
Traci et al [6] have developed a mass consistent model and this was the beginning of a significant
modelling effort towards this direction (NOABL*, AIOLOS, WINDS, etc.). NOABL* is a mass-
consistent model that solves the Poisson equation for the determination of the perturbation wind speed
potential through a finite-difference scheme. The initial flow field is adjusted through the perturbation
potential to satisfy mass conservation throughout the computational domain, taking into account different
atmospheric stability conditions. However these models do not predict purely viscous effects such as
recirculation and they do not give information on turbulence. Improvement can be accomplished by
estimating the vertical velocity as solution of the turbulent momentum equation in the vertical direction
(AIOLOS-T, [7]). This method accounts for stratification and uses a zero-order model or a k-ε model for
turbulence closure.
Dynamic models solve the continuity, the momentum and the temperature equations. They vary in
complexity according to the treatment of the topography, the simplifications in the equations, the
turbulence model etc. In order to achieve low computational cost, dynamic models based on the boundary
layer equations have been developed. The concept of the linear perturbation theory of Jackson and Hunt in
two dimensions [8] and Mason and Sykes in three dimensions [9] introduced a two-layer approximation
using different perturbation schemes in the inner and outer region of the boundary layer. This concept
gave rise to the development of several analytical models (i.e. WAsP, MS3DJH), which have given
satisfactory results in geometries where no flow reversal exists (J. L. Walmsley [10]). Mason & King [11]
review different J&H oriented models and compared their results against their finite difference model
which is effectively identical to the model of Taylor [12]. At present analytical models based either on the
potential flow theory or on boundary layer parameterisations are used for predicting the wind flow over
complex terrain. The WAsP program, which is capable of generalizing a long-term meteorological data
series at a (reference) site which may then be used to estimate conditions at other (predicted) sites, is
nowadays the most extensively used practical tool applied for this purpose. The MS3DJH family of
models conserve mass but also attempt to solve momentum equations in an approximate manner by using
a constant-density atmosphere. It is assumed that the physics of the flow can be simplified by using a
mixing length closure for the turbulent stresses, which is only valid for statically stable conditions, and
that the momentum equations subjected to the thin shear layer approximation can be linearised and then

Site Evaluation
solved using perturbation techniques. In general the models belonging to this category have inherent
limitations concerning the terrain slopes and so their validity in complex terrain is not guaranteed.
To overcome the inherent limitations of the linearised flow models several researchers have recently
attempted the solution of the fully 3D turbulent Navier Stokes equations [13], [14]. In the literature several
models have been proposed for closure of the equations such as zero order models, one equation models,
two equation models, Reynolds stress models etc. To suppress the computational cost, most Navier Stokes
runs are performed on relatively coarse grids applying law of the wall type boundary conditions.
Simulations with more elaborate turbulence models, like the low Reynolds k-ε or the algebraic Reynolds
stress models, which could better estimate the wind profiles and the Reynolds stress tensor, require large
memory and computer time and this renders them cost ineffective. Validation studies have confirmed that
the k-ε is a good compromise between the simple models and the more complicated ones [15], [16].
All the previous models are unable to account for the influence that local wind systems have on the annual
wind conditions especially at a complex terrain site. One important aspect of siting wind converters is a
description of the local wind climate with high spatial resolution which is often not available from
observations in the atmosphere. The density of operational observation networks is too poor for that
purpose so that additional information is required. Therefore methods have to be applied to derive this
information from the large scale state of the atmosphere and topographic parameters by applying meso-
scale models. The meso-scale simulation models are able to simulate local wind systems produced by the
orography and/or solar radiation, because they incorporate the appropriate physics but also because of
their working domain (characteristic dimensions of time and length of the meso-scale) which is restricted
in the models mentioned previously. Only at this scale, where the typical dimension of time lies between a
few minutes and a 24 hour day and the typical length lies between 2.5km-mountain and valley winds to
2500km-tropical cyclone, do local wind system phenomena occur. All meso-scale models calculate the
wind field by numerically solving the Navier Stokes conservation equation for mass, energy, momentum
etc. The specific formulation of all equations and the approximations used depend on the specific meso-
scale model that is selected. Summarizing, meso-scale models are able to describe the wind field with a
high spatial resolution during short episodes in complex terrain. To gain a wind climatology from
simulations with these models, available statistical information about the large scale state of the
atmosphere has to be scaled down to the required resolution by applying meso-scale models, which
describe principally the modification of the large scale state by meso-scale processes induced by the
topography.
A number of meso-scale models that are used for site assessment applications are found in the literature,
i.e. the HIRLAM model by Danish Meteorological Institute, the KAMM model by University of
Karlsruhe, the GESIMA model by GKSS, the MEMO model by IEH-REG, the UK Meteorological Office
model and the meso-scale model run by the University of Uppsala [17], [18], [19], [20], [7]. The
tremendous computer resource needed for applying these models should be taken into account. Moreover,
that the grid step used in the analysis cannot always be as small as required and the fact that the models
require several basic assumptions (i.e. definition and number of episodes, assumed frequency distribution
of wind velocity etc.), lead us to believe that at the moment their results are considered to allow for a
qualitative description of wind statistics in regions where observations are not available but the accuracy
of the absolute values resulting from these procedures is not always ensured.
Recently, research has been addressed towards the combination or nesting of models with different spatial
resolutions (i.e. WAsP/local scale with KAMM or HIRLAM / meso-scale [17], [18] or WAsP with Navier
Stokes solvers) and hybrid micro-siting models [21] combining the mass consistent methodology with
principles of the Navier Stokes approach. During the JOULE I project, contract no JOUR0067 entitled
“Wind measurements and modelling in complex terrain” it was attempted to estimate the possible zones
of separating flows based on the topography input and to verify by means of the measurements obtained in
the same project. In the European Wind Atlas Vol. II (to be published shortly), the KAMM model, which

Site Evaluation
is able to calculate the resource over distances of hundreds of kilometers, was used as a supplement to the
WAsP model which in contrast can handle very detailed information on the topography in a radius of
approximately five to ten kilometers from the sites. Preliminary results for assessing the wind potential of
Ireland are quite promising [22].
In the author’s opinion the most interesting idea was to include in the WAsP package a module which uses
the Navier Stokes equations in a model based on the rapid distortion theory (RDT) for predicting the
turbulence component spectra over moderately complex terrain.
Summarizing, the current state of the art on siting appears as follows:
Evidently, depending on its background theory each model has certain possibilities and limitations.
Simple models are still in almost exclusive use. Due to their approximate character, calibration procedures
(wind tunnel studies, in situ) are often needed. In some cases, additional calculations with more advanced
(Navier-Stokes), generalized (meso-scale) or even statistical tools (i.e. MOS -Model Output Statistics
[18]) seem necessary for improving the prediction accuracies.
Navier-Stokes models require considerable computational effort which has been a serious drawback for
their widespread use.
Meso-scale models are able to simulate local wind systems produced by the orography and/or solar
radiation but they are also very complicated and therefore require tremendous computational effort.
It is worth noting that most of the models discussed have been used for resource assessment purposes
only. Their reliability in predicting the flow characteristics which are known to affect the operation of a
wind turbine in complex terrain is by no means established. For instance, although a great number of
models exist which allow, within certain limits, prediction of mean wind speed modifications due to
complex terrain, only a few are capable of providing an estimate of the modification to the turbulence
level and structure arising from complex terrain. The real dimensions of the problem can be appreciated
by considering that the greatest bottleneck associated with complex terrain is actually the identification of
the terrain parameters related to the characteristics of the wind field driving the wind turbines.
3.1.2 WAsP
3.1.2.1 Limitations
To validate the accuracy of existing flow modelling codes several factors should be considered i.e. the
sensitivity to accurate model input (i.e. factors intended to be user defined) and limitations due to the
physical assumptions in the coding.
The literature review was extended to papers and reports presenting results of case studies and sensitivity
analyses using the WAsP package in an attempt to quantify the errors and uncertainties associated with the
input data and the limitations in the physical assumptions. This revealed, in some cases, rather
contradictory results regarding the importance of the resolution of the input topography.
Sensitivity studies carried out by RAL in the framework of the JOULE project contract no JOU2-CT93-
0370 [7] for investigating the dependence of the prediction accuracy of these codes on the user defined
input and for studying their overall usefulness when extrapolating wind speeds especially in complex
terrain, revealed that the resolution of the input topography and roughness was very important to the
correctness of the wind speed estimates. More specifically it was found that when the topography and the
roughness were well defined and that when suitable wind speed initialization data sets were used, accurate
predictions could be made. The accuracy broke down over very steep slopes and in cases where local
circulations were forming. The dependence of the flow model results on some of the underlying physical
assumptions of the coding was also examined. For this purpose a model of the heat fluxes in the

Site Evaluation
atmosphere was developed to divide wind speed data sets up according to the stability of the atmosphere
at the time when the measurements were made. It was shown that the classified data sets had very different
characteristics in terms of wind speed and direction distribution, wind shear and in terms of the magnitude
and form of the correlation between data sets measured at different sites. For the case studied the errors
brought in by the stability assumptions of the flow modelling were of a similar magnitude to, and could
not easily be distinguished from, those brought about by the breakdown of the model’s representation of
the wind flow over steep slopes.
A thorough sensitivity analysis using the WAsP model for site assessment was carried out by ETSU [23]
and it was revealed that estimates of the long term mean wind speed made using the WAsP model had an
error of up to 30% while the corresponding wind power estimates, as expected, had an error of 70%.
When trying to identify the sources of error related to the physics adopted by the model, the study referred
to the fact that the WAsP model uses the Atlas file where the generalized wind data is created by forcing
the measured data to fit a standard Weibull type frequency distribution. However it was found that very
few of the observed wind climatologies can be stated to have come from a Weibull distribution, therefore
it is evident that this procedure introduces an error the magnitude of which depends on the degree of
transformation applied to the data.
Regarding the input data and additional to the points made above, it was concluded that there was no
obvious pattern found relating the magnitude of the sign of the error to terrain type. The analysis also
revealed that the impact of including obstacles was generally small and can not be guaranteed to improve
the accuracy of the prediction. Finally, it was found that the WAsP model is relatively insensitive to the
area, resolution or quality of the topographic input data, which is a rather surprising result based on the
author’s experience with the application of WAsP in complex terrain and taking into account that the
model adopts a telescoping principle. This means that the orographic model of WAsP uses an expanding
polar grid enabling a detailed description of the terrain elevations closest to the actual prediction locations
which exert the strongest influence. This is also illustrated in [26], where it is evident that the prediction
accuracy of WAsP depends heavily on the choice of the terrain resolution used.
The performance of WAsP in predicting wind speed variation across three sites in the UK with varying
complexity, is discussed in [27] where quite accurate predictions are obtained but with the standard
deviations of the differences between the observed and predicted data varying considerably, a fact which
could be related to the varying severity of the terrain from site to site. In [29] a statistical treatment of the
discrepancies between actual and WAsP predicted energy productions is performed using data from ten
wind farms. It was found that the deviations varied from a 10% under estimate to an 8% over estimate.
The difference between the forecast and actual production figures of a wind farm for a single year was up
to 14% with an 8% mean value. In [30] the WAsP model was used to estimate the energy capture at
potential wind turbine sites in Norway. Significant discrepancies are found between the data extrapolated
from short term measurements and the predicted values which could not be explained by the statistical
uncertainties generated from the use of short term measurements and by calibration errors only.
Sectorwise comparison indicated that roughness effects were exaggerated and that roughness lengths
outside the recommended intervals would have to be used to tune the model. This finding is in accordance
with the author’s experience with WAsP.
A method for estimating the uncertainty in wind energy resource among others is elaborated in [31] and it
estimates the uncertainty of the estimated wind speed due to (a) the limited data quality to be 5% on the
annual mean wind speed, due to (b) the application of the geostrophic drag law to be 5%, due to ( c) the
roughness change and choice of roughness classes to be 5% and due to (d) the incompleteness of the
model for determining the flow in complex terrain to be up to 10% for certain specified cases. According
to this the uncertainty related to complex terrain modeling is estimated to be of the order of 13-14%
assuming the error components to be independent.

Site Evaluation
WAsP prediction errors may be significant in cases where the local climate or terrain lie outside their
normal operational envelope. According to Bowen et al [24] a high level of cross correlation between
wind speeds at the reference and predicted sites is an essential but not exclusive pre-requisite for an
accurate prediction. The correlation coefficient itself does not indicate the sign or magnitude of the
prediction error. Bowen et al concluded that the sign and approximate magnitude of the prediction error
due to orography is proportional to the difference in ruggedness between the predicted and reference sites.
An approximate estimate of this error may therefore be made with a performance indicator based on site
ruggedness. One suitable indicator that they have developed is the difference in the fractional extent of the
terrain with slopes greater than a critical value between the predicted and reference sites. This indicator
(which they have called RIX for Ruggedness IndeX) is believed to provide a means of defining in
quantitative terms, the orographic limits for accurate WAsP predictions and a suitable correction if those
limits are exceeded. The use and effectiveness of this parameter is discussed further in section 3.3.
In [26] it was observed that in many sites where measurements were available at more than one height, the
values predicted from data at different heights showed poor agreement with each other, which suggests
that the wind shear profile is not being accurately modeled.
Finally, the effect of different stability (non-neutral) conditions on the predictions of WAsP are discussed
in [23], [24], [25], [26] and [28]. In [26] the discrepancies between the DTI database values (consisting of
NOABL estimated wind speeds across the UK) and predicted values from WAsP are considerably reduced
at higher wind speeds where neutral atmospheric conditions generally occur. In [28] (addendum to [23]) it
is stated that there is a separate step within WAsP which makes a correction for stability by modifying the
parameters of the Weibull distribution using functions of the mean and RMS heat fluxes over land and sea
defined in the parameter file. However, this stability correction is rather a climatological one and therefore
misses the dynamic effect that stability has on the flow. Perhaps this is the reason why in [25] it is
concluded that although the data is subdivided into stability classes and the default parameters of the
model are changed, only slight improvements in the predictions are observed.
Regarding WAsP, a number of conditions that should be fulfilled for acquiring accurate predictions have
been reported and it seems that users of the code all over the world agree upon this subject. The physical
limitations of the model are sufficiently documented but still, there are questions arising regarding the
input data.
For instance, it has been proved that the accuracy of WAsP depends on the choice of the terrain resolution
used but the problem remains, considering the memory restriction on the number of data points describing
the topography, how to choose a sufficient resolution without limiting the overall working domain and
without performing time consuming sensitivity analysis for each site. In addition to that there seems to be
no way of quantifying the effect on the user’s estimate of using coarse resolution data when this is the
only data available. (The 32-bit version of WAsP which is soon to be released will alleviate the terrain
description resolution problem).
Another point generating uncertainty in the obtained predictions is the effect of different roughness values
assigned by different observers to the same site.
The length of the reference data record used should be sufficiently long to compensate for monthly,
seasonal and even yearly variations.
Finally, the length of the data averaging time is also a parameter which should be selected with care so as
to allow a particular wind event to envelope physically the reference site and the site to be predicted.
Example Applications
A number of analyses were carried out within this project using WAsP to help confirm some of the
findings. Sites representing terrain of varying complexity for which digitized terrain data already existed

Site Evaluation
were identified and for which measured wind speeds or performance data were available for the purpose
of evaluating the prediction accuracy of WAsP. Only the case of simple, flat terrain was ignored as being
trivial in this instance. Sites in Germany, the UK and Greece were modelled with terrain ranging from
moderately complex in Germany to ultra-complex in Greece. The results of each study are summarized
below.
Germany
WASP and the European wind atlas model are widely used for prediction of annual energy yields in
Germany in both inland and coastal areas. It is known from a number of publications, that the application
of WASP to complex terrain situations can be associated with large errors, while the use in medium
complex terrain generally gives reliable results. But this is only a generalization. The results of WASP
rely very much on an exact description of surface roughness, obstacles and orography but also on reliable
meteorological input data. The quality of the meteorological data depends very much on continuos data
recording, calibrated and maintained anemometers, but it also depends strongly on the height of the met-
mast (typically only 10m in Germany) and the influence by roughness and obstacle elements, which may
vary during the measurement time (i.e. 10 years). The quality of the data and their suitability usually can
not be examined by the WASP user.
To verify the results of the European wind atlas method in moderate complex terrain, WASP was applied
to estimate the energy yield of an existing wind farm in southern Brandenburg. The wind farm was chosen
because it consists of modern wind turbines (WT’s) with a certified, measured power curve, which is an
essential input to reliable energy yield estimations.
The energy yield of the wind farm in 1995 and 1996 was compared with the energy yield of a nearby
single WT, operated over four years, to find out a value of the annual energy yield which is representative
for a several years period. The single WT was not chosen for direct comparison because of the lack of a
certified power curve. The energy yield in 1995 was chosen to be representative.
The example calculation was performed with data from different meteorological sites. The results with
data from three nearby stations (distance < 45km) were extremely bad, ranging from 42 to 48 % deviation
from the real energy yield. The farther met-station Leipzig (90km) gave better results (14% to 21%
deviation), even though the surrounding topography is not so closely related to the wind farm area as that
of the three nearby met-stations. The atlas data for Leipzig were taken from DWD and for comparison
from DEWI’s own generated data. The DEWI atlas data were generated conservatively giving a slightly
better result. But it should be noted, that this result could be random and that the Leipzig station should
not be taken generally in this area.
To help objectively assess the severity of the terrain, the RIX parameter (see section 3.3.3.2) was
determined for each combination of reference and prediction sites. The RIX was calculated to a radius of
3.5km from each site using 12, 30° direction sectors. For all sites considered, the RIX was found to be
zero; in other words, all terrain slopes were less than 30%. In fact, most of the slopes were less than 5 per
cent. It is therefore considered unlikely that the poor prediction performance was due to poor modelling of
the terrain induced flow variations. So the poor performance may have been due to poor representation of
obstacles or roughness or poor quality reference wind data. Despite the only moderately complex terrain,
as many of the reference sites analyzed are in very close proximity to forests and as the reference
anemometers are mounted at quite low heights (12m - 18m) it is possible that a combination of all of these
factors may have contributed to the poor prediction performance.
UK
Data from a complex terrain site in the North of England at which a number of meteorological masts had
been installed for site assessment purposes were analyzed. Existing WAsP input terrain data files were
modified to determine the sensitivity of the prediction performance to the precise specification of the input
terrain data. The RIX (see section 3.3.3.2) was used to characterize the terrain and an attempt was made to

Site Evaluation
derive relationships between prediction error and RIX.

Time series (hourly average) from four sites were used for the analysis, one of which was a sea level
Meteorological Office reference site. The other three sites were prediction sites, one of which was at sea
level and the other two at about 200m a.s.l. Digital map data obtained directly from the UK Ordnance
Survey were used, after conversion to WAsP format, for the analysis. The maximum available contour
resolution of 10m was used in the vicinity of the prediction sites.
The terrain in this instance was characterized by RIX values of 1-2 per cent for certain directions
indicating that slopes of greater than 30 per cent did exist. The relationship between prediction error and
RIX was found to be less consistent than that obtained by the originators of the technique [68] possibly
indicating that other factors, besides terrain ruggedness were not well represented in this model (e.g.
roughness) or that the reference site (Meteorological Office station) was subject to unusual, localized
meteorological effects. Consultation with the Met. Office personnel from this station did reveal experience
of unusual effects from certain wind directions due to the interaction of the wind with local cliffs and sand
dunes.
This analysis highlights the need to thoroughly evaluate the reference site before deciding to use it for
modelling purposes.
Greece
For this case, a site on the island of Andros was identified for which a WAsP terrain model existed.
Measured energy and wind speed data were available from a wind farm and meteorological masts on the
island from time periods both before and after the construction of the wind farm. The primary objective
was to use reference wind speed data from masts located in terrain of varying complexity to predict the
performance of individual wind farms. The first reference mast was located within the boundaries of the
wind farm before the wind farm was built. The second mast was positioned outside the wind farm and at a
distance of 1.5km from it on a well exposed site. The third reference mast was located on the edge of the
wind farm in an area with relatively steep slopes (20%) in comparison to the other reference mast sites.
These reference data sets were then used to predict the corresponding wind speeds at two heights on two
other masts at the edge of the wind farm (before the wind farm was built). In all cases, the analysts had
very good knowledge of the reference data quality, either having carried out the measurements themselves
or having good access to the organization who carried out the measurements.
The results from this first study were inconsistent. It was found possible to predict the wind speeds at one
mast with prediction errors of 10% to 36%, depending on the reference site. However, for the other mast,
the prediction error range was 2% to 11%.
A second analysis study was carried out using the same reference sites, but targeting four masts on the
boundaries of the wind farm. In this instance, the prediction error deviation was found to be 5% to 35%,
although the lowest error values generally correspond to cases where the reference mast data have been
used to do a self-prediction, such that the prediction and reference sites are the same site. It should be
noted that due to the complexity of the terrain and the memory limitations of the model, a less detailed
description of the terrain contours was used on the periphery of the wind farm. The four prediction masts
in this part of the study were located in the area with less contour detail.
Finally, predictions were carried out for the energy yields of each of the seven turbines within the wind
farm using the WAsP WECS module to apply the turbine power curves. Measured annual energy
production data for each turbine for a concurrent period were available for comparison. As it was not
possible to accurately correct the WAsP predictions to account for periods of turbine unavailability, both
the measured and predicted energy productions for each turbine were normalized with respect to one
turbine in the wind farm. So although it was found that WAsP tended to overestimate the absolute energy
production by between 7% and 22%, the relative prediction performance between individual turbines was

Site Evaluation
found to be quite impressive. In fact, referring to table 3.1, in all but one case, the relative prediction error
is 2-3%.
Table 3.1 Relative energy production (measured and WAsP prediction)
Turbine Measured WAsP
1 1.00 1.00
2 0.95 0.93
3 0.89 0.87
4 0.92 0.90
5 0.85 0.83
6 0.82 0.85
7 0.79 0.91
This particular example illustrates that even in very complex terrain, when you have good quality input
data, WAsP can be made to work very well, at least in relative terms.
3.1.2.2 Guidelines for Use

The general conclusion is that WAsP is a practical and, as indicated from the survey summarized in Figure
2.1, the most widespread tool for site assessment, particularly in simple terrain and for preliminary site
assessment in complex terrain. The literature review shows that accurate site assessment is obtained
provided that :
• both the reference and predicted sites are subject to the same weather regime
• neutral atmospheric conditions prevail
• the surrounding terrain is sufficiently gentle and smooth to ensure mostly attached flows
• the reference data are reliable
• the description of the background roughness length must be as accurate as possible
The application of WASP to moderate complex inland areas is more difficult and should ideally only be
done, if a verification with energy yield data from existing WT’s or on site wind speed measurements is
possible. This is particularly true if absolute predictions are required. However, in cases where relative
predictions across a small area (the area of a typical wind farm say) are required, the model performance
can be quite good, even in complex terrain sites. Such relative predictions would usually be carried out in
conjunction with a measurement based technique. The findings of this project suggest that uncertainties in
absolute resource assessment of at least 13-15 per cent can be expected for moderate to complex terrain
assuming that all of the above criteria are fully addressed. There are instances when the model can
perform better than this, but in most cases there will be no way of checking at the time of the assessment.
3.2 Measurement based techniques

3.2.1 Status Review
Site assessments based on measurements do not generally have the benefit of long term wind speed
measurements (10 years or more) at the actual site of the proposed wind farm. To be practical,
measurements can only be carried out at specific sites for periods of 1 - 2 years at most, hence fitting in
with the planning (commercial, technical and legislative) time scales and financial constraints. Throughout
Europe, the average wind conditions over any particular year or two will deviate substantially from the
long term conditions. The longer the averaging period (say ten years), then generally the smaller the
deviation from period to period. As wind farm developments have a planned life of 15 to 20 years, then

Site Evaluation
the average wind conditions must also be assessed for this longer period. The only way that this can be
achieved using measurements is to relate short term wind measurements at the planned wind farm site to
long term wind measurements at some reference site (e.g. national meteorological office station). There
are many ways of doing this, but the general procedure is called “Measure-Correlate-Predict” (MCP) [32].
As the name suggests, the MCP procedure comprises a measurement stage, a correlation stage and a
prediction stage. Each stage is described briefly in the following sections.
3.2.1.1 Measurement
Simultaneous wind speed measurements are required at two sites, one of which is at the site under
assessment, the other being a reference site for which a long-term wind speed data record exists. The
reference site is typically a national meteorological office station, although there may be other sources of
long-term reference wind data (refer to section 2.4 for country specific guidance). As measurement of the
reference data is generally beyond the control of the organization carrying out the site assessment, the
format of the reference wind speed time series usually dictates the minimum format of the measured data
for the prediction site. For example, the reference time series available in the UK have a minimum time
step of 1-hour. Therefore, the data measured at the reference site must also be recorded or reduced to this
format before it can be used in subsequent stages of the analysis. Although there are many reports
documenting MCP validation studies, the measurement stage is not extensively described. However, it is
generally accepted that, primarily to obtain a statistically significant directional and wind speed coverage,
the “short-term” measurement campaign should be of at least 8 months duration. Furthermore, many
researchers have investigated and highlighted the importance of careful reference site selection.
3.2.1.2 Correlation
In the correlation stage, relationships are derived between the short term measured wind speed time series
at the reference and prediction sites. Typically, the relationships are derived separately for each 30 degree
direction sector (based on the reference site wind direction). As wind vane measurements can be
somewhat erratic in light winds due to vane meandering, some analysts prefer to discard data for wind
speeds less than 3-4 m/s. However, there is evidence that this has, at best an inconsistent effect on
subsequent prediction quality [23]. The literature confirms that the application of simple linear regression
techniques with floating intercept across the complete measured wind speed range gives reasonable
correlation quality in most situations [32, 23, 7]. However, in specific cases, it may be possible to decrease
the statistical uncertainty carried through into the final stage of the analysis by using alternative regression
functions such as polynomial, or logarithmic functions. Generally, visual inspection of the best-fit
function and the raw data on a scatter plot will confirm whether or not investigation of alternative
regression functions is justified. The literature suggests that more accurate predictions can be obtained,
particularly at high wind speed sites, by carrying out an analysis of the relationships between discrete
wind speed bins at the reference and prediction sites [33]. To the author’s knowledge and based on the
evidence of the survey, this technique is not widely adopted.
In cases where the reference and prediction sites are separated by tens of kilometers, it may be possible to
marginally improve correlation quality by filtering out higher frequency components from the time series
before correlation [32] such that the remaining turbulent components are of scale more compatible with
the site separation distance. In some situations, correlation quality for all directions may be extremely
poor. This is usually indicative of an inappropriate reference site (different reference site climatology from
the prediction site or inconsistent data quality) and in such cases there is no option but to identify an
alternative reference site. However, it should be noted that, particularly for large separations between the
reference and prediction sites the flow between the sites will not be rectilinear and will be dictated by the
shape of the cyclone or anti-cyclone driving the wind and the local terrain [32]. Therefore, for a specific
reference site wind direction, there are a variety of possible concurrent wind directions at the prediction

Site Evaluation
site. Particularly in complex terrain, each slight variation in local wind direction will lead to a different
terrain induced wind speed effect and so, even if the reference and prediction sites are subjected to the
same weather systems, scatter in the wind speed correlations is inevitable. From experience with linear
regressions of wind speed data, correlation coefficients in the range 0.7 to 0.9 are typical for a well
matched pair of prediction and reference sites using 30 degree direction sectors and site separations of 20 -
100km.
3.2.1.3 Prediction
The term “prediction” (implying “future”) is somewhat of a misnomer as this stage is actually a
transposition of the measured (and hence “historic”) long-term reference wind regime from the reference
site to the “prediction” site. This aspect is discussed further in section 3.3.4. The long-term reference wind
regime is usually available in the form of a joint distribution table of wind speed and direction covering a
period of 10 - 20 years. The short term wind speed relationships are applied to the uniform wind speed bin
limits defining the distribution table on a direction sector-by-sector basis. Hence, the reference site wind
speed distribution table with uniform wind speed bins for each direction sector is transposed to a
prediction site wind speed distribution table with non-uniform wind speed bins per direction sector. By
simple arithmetic manipulation the prediction site distribution table bin contents can be re-distributed to
uniform bins. The long-term mean wind speed for the prediction site is determined from the sum of the
products of the bin frequencies and corresponding bin mean wind speeds [32].
The technique can also be adapted to “predict” directional shifts from one direction sector at the reference
site to one or more different direction sectors at the prediction site. Techniques using polynomials [32] and
discrete shifts [7] are documented in the literature.
3.2.2 MCP
3.2.2.1 Recommended Applications

MCP is an extremely versatile technique which can be used in a wide variety of situations provided that:
• validated, long-term and short-term concurrent measured reference wind speed data are available at a
reference site within the same climatalogical zone as the prediction site.
• validated prediction site wind speed measurements are available for a period of at least 8 months and
more typically 12 months or more.
Subject to these few restrictions, a successful MCP analysis is always possible. However, the question
which has to be answered on a project-by-project basis is that of whether an (relatively expensive) MCP
analysis is required or whether modelling will suffice. The answer to that question is dictated primarily by
the complexity of the terrain and the level of prediction uncertainty that is acceptable in the circumstances.
3.2.2.2 Guidelines for Use

Considering each stage of the MCP procedure in turn, the recommendations are as follows:
Measure
The reference site should satisfy the following criteria:
• located in the same climatalogical zone as the prediction site
• have time series data concurrent with the short-term measurement period at the prediction site
available
• have long-term (10-20 years) measured distributions of wind speed and direction available.
• be subject to an investigation to determine that instrumentation and measurement quality has been
consistent and is acceptable in the long-term.
• alternative reference sites should be identified for use in situations where the first-choice reference site

Site Evaluation
proves unsatisfactory.
The prediction site measurements should be carried out according to the following criteria:
• the measurement mast should be located at a position on the intended turbine or wind farm site that is
at least as well exposed to the wind as the intended wind turbines or wind farm.
• measurements should be carried out according to best anemometry practice.
• the wind speed and direction data should be measured and stored using an averaging period that is
either directly compatible with the reference site measurements or that can subsequently be reduced to
a format that is compatible with the reference site measurements.
Correlate
The choice of correlation function is quite simply that which best represents the measured data. However,
the following recommendations represent normal practice:
• it can be useful to remove particularly low wind speed data (less than 3-4 m/s) before carrying out the
correlations as erratic wind vane behaviour can distort the relationships.
• (optional) for large reference and prediction site separations, filtering out of higher frequency, short-
term effects can improve correlation quality.
• the correlations should be carried out for each of twelve, 30 degree directional sectors (as measured at
the reference site).
• a linear regression with a floating intercept gives satisfactory results in most instances. Alternative
regression functions can be tested and compared on the basis of correlation quality if time and
analytical capability are available.
• all raw data and best-fit regression functions should be visually checked on a scatter plot to ensure that
the regression function is sensible over the entire, expected wind speed range.
• (optional) as a separate task, a relationship between the simultaneous wind direction time series at the
reference and prediction sites can be determined. This can be accomplished using polynomial or
discrete (matrix) methods.
Predict
The prediction stage is simply an arithmetic transposition of a joint distribution of wind speed and
direction, carried out using the regression functions identified in the correlation stage of MCP. The only
important recommendations reiterate the points raised in the selection of the reference site. That is:
• ensure that the long-term reference period is sufficiently long (10-20 years) to have averaged out
significant year-to-year wind speed variability.
• ensure that those measurement conditions at the reference site which are likely to influence the
recorded wind speed have remained similar during the long-term reference period and the short-term
reference period. That is, check that regular anemometer calibrations have been carried out and that
blockage sources such as foliage, trees and buildings have not appeared, disappeared or changed
significantly during the long-term.
3.3 Assessment of Uncertainty

The assessment of site mean wind speed can be considered to be a form of “site calibration”, where this
term already has an accepted definition in the field of wind turbine performance measurement. It was
realized several years ago, see Frandsen, 1987 [34], and Frandsen and Christensen, 1992 [35], that site
calibration both for machine (performance) testing and for resource assessment constitutes a most severe
uncertainty, each frequently exceeding 10% and in combination often 15-20%. The potential uncertainty
when considering loads would be expected to be less, though still unacceptable. The awareness of some of
the pitfalls in site calibration has in itself reduced the uncertainty. However, it appears that a necessary
step in improving the applicability of site calibration is an extension of the concept itself in order to
include other presumed important flow characteristics than wind speed and air density. This is done
below, followed by a general discussion of the uncertainties.

Site Evaluation
3.3.1 Definition of extended (long-distance) site calibration

The objective of the site calibration is to predict flow characteristics at one geographical prediction
position (where there is or will be a wind turbine) using information/data from another position, which we
denominate reference position. Where the purpose of the site calibration is prediction of extreme wind
speeds or kinetic energy at the prediction site there will be more solid data available from the reference
position, and the site calibration effectively serves to transpose that climate information to the prediction
position (this is a measure-correlate-predict process). When the site calibration is applied for machine
testing (power curves, loads etc.) the purpose of the site calibration is to (by means of measurements at the
reference position) estimate the simultaneous flow characteristics at the prediction position.
Site calibration is defined as the action (experimental or numerical flow modelling) that links a set of flow
parameters at one geographical position to the same flow parameters at another position.
In the present context the parameters are of different importance. Thus, the mean wind speed is obviously
the most important parameter. To generalize the problem and to allow for non-linearity of that particular
parameter, we assume that the site calibration is carried out binwise (wind speed and direction at base
position: for MCP type analyses, a single wind speed bin encompassing the entire wind speed range is
usually although not exclusively used in conjunction with 30 degree direction bins. For short range site
calibrations, smaller wind speed and direction bins are typically used.). Assuming that in the bin the
relation between the parameters (excluding wind speed) measured at the base position and the wt position
is linear, the following equation to transpose the flow parameters of the base position to the wt position is
applied:
x = A b x − T + 0 x , where (1)
x = {ur , x 2 ,..., x n } and b x = { b ur , b x 2 ,..., b x n } (2)
are the parameter sets at the wt position and the base position, respectively, and
a11 a12 ... a1n 

a a ... a 
A=
21 22 2n 
(3)
.... 
 
a n1 a n 2 ... a nn 
is the partial regression coefficient tensor, where aij=aij(u,d) are functions of wind speed and direction.
The determination of A and the offset vector 0x is called site calibration. The coefficients of the tensor are
determined row-by-row by multi-regression analysis or estimation with a flow model.
If each parameter at the wt position is a function of only that same parameter at the base position the
tensor becomes diagonal, aij = 0 for i∴j.
3.3.2 Test parameters

It is presently not clear which (xi) and how many parameters (n) should be included in the test to
significantly improve the quality/accuracy of the test. Apart from wind speed (ideally the hub height 10-
minute mean U-component), it is expected that turbulence (u but possibly also v and w), flow inclination,
air density, vertical and horizontal shear and humidity are of importance. However, for MCP type
analyses, the reference data (typically from a national meteorological office) are generally restricted to 1-
hour averages of wind speed and direction at a single height and so the practical model is severely
restricted.

Site Evaluation
3.3.3 Uncertainty components

Site calibration may be performed experimentally or by application of a flow model or by a combination
of the two methods. In the practical handling of uncertainty there is good guidance in the ISO guidelines,
1993 [36], and IEC 1400-12, 1997 [37].
3.3.3.1 Experimental site calibration

Though being a matter of definition, 4 groups of uncertainties can be identified in the experimental site
calibration:
• instruments’ uncertainties - described in IEC 1400-12, 1997 [37].
• uncertainties on determination of regression coefficients - uncertainties related to the regression
analysis are outlined in annex A.
• the uncertainty on prediction when applying regression coefficients - discussed in ISO 1993, p. 76
[36].
• intrinsic uncertainties accounting for method deficiencies: the limited number of parameters, non-
linearity’s, “unexplained” variation etc.
Table 3.2 Uncertainty components

Group Site calibration uncertainties Uncertainty Parameter
1 Instruments’ uncertainty s1 x and bx
2 Regression coefficients s2 A and xo
3 Predicted value s3 bx
4 Intrinsic uncertainties Δ -
Total uncertainty on param. i tsi x
The “propagation” of uncertainties through to the final result shall not be discussed here, however IEC
1400-12, 1997 [37], gives a good description of propagation in the case that site calibration is carried out
for power performance assessment.
No detailed account of the uncertainties are offered here, merely selected, basic considerations. Should the
resulting group-uncertainties be uncorrelated1, the total uncertainty on estimate of parameter xi at the wt
site becomes
t si = s12 + s22 + s32 + ∆2 (4)
Where the experimental site calibration action is carried out to “perfection” and with, in principle, an
infinite number (m) of data sets, the lower limit of uncertainty becomes
t i s m→
→∞
s12 + ∆2 , (5)
because the uncertainties in group 2 and 3 are by nature statistical and therefore decrease with increasing
amount of data. Thus, the uncertainty has a lower limit, determined by uncertainties of instruments and
deficiencies of the site calibration model applied. Taking one step further by assuming the instruments’
uncertainties to be negligible, the resulting uncertainty has a lower limit determined by the quality of the
method itself and the number of physical quantities measured.
Discussion.
1
The validity of the statement is not obvious and should be addressed in new research.

Site Evaluation
As stated, site calibration is made for two purposes, prediction of climate or for wind turbine testing.
Typically, the difference in test circumstances for the two is distance between base- and wt position.
When aiming at climate prediction at a prospected wind turbine (farm) site, the distance varies from a few
km to tens of km, and for wind turbine performance test purposes the distance spans from about 100m to
1-2 km. The difference in separation between positions results in different needs of the extent of the data
base for determination of regression coefficients and constants (group 2), and input data (group 3).
However, group 2 and 3 uncertainties, as defined here, will decrease with increasing amount of data, and
the uncertainties may be calculated on-line from the accumulating data base to determine when group 2
and 3 uncertainties are negligible compared to group 1 and 4 uncertainties.
Instruments’ uncertainties (group 1) set limits on the integrated site calibration uncertainty. Assuming
diagonality of the site calibration tensor, it could be argued that the integrated uncertainty stemming from
instruments is the root-sum-square of instrumental uncertainties at the two positions. However, if e.g.
anemometer calibration has an uncertainty, which is biased identically for the instruments applied at both
positions, these bias errors could to some extent cancel each other. However, for an MCP type
measurement/analysis, the reference wind speed measurements will usually be carried out by a third party
(e.g. meteorological office) and so it may not be straightforward to gain access to and interpret calibration
and uncertainty information. In this situation, it is unlikely that the bias errors at the reference and
prediction sites would cancel.
An instrument uncertainty, which deserves special attention especially in complex terrain is related to the
quality of the cosine response of the cup anemometers which are typically applied, Pedersen and Paulsen,
1997 [38]. The cup anemometer measures, in principle, horizontal flow speed if the mean velocity vector
is horizontal, and if the vector is not horizontal, the horizontal component of the vector is ideally
measured. As it is, many cup anemometers have a less than ideal cosine response. Therefore, if there is a
difference in flow inclination at the two positions the site calibration results may be erroneous or directly
misleading.
Summing up, the following comments to experimental site calibration can be made:
• Instrument uncertainties are fairly well covered (IEC 1400-12, 1997 [37]), though see above comment
on cosine response. Uncertainties of the important wind speed measurement ranges from 1 to 5%, and
may be up to 15% under special conditions2. Other instruments are less uncertain, 1% or less.
• The uncertainties of regression coefficients and application of these are by nature statistical and may
in principle be eliminated by sufficient data. Quantification of uncertainties still has to be evaluated.
It is known that the site calibration procedure presently used is inadequate in cases where e.g. shear and
turbulence are different at base and wt position. Adopting the more general definition in this section, the
great challenge in experimental site calibration is the identification of the intrinsic uncertainty (group 4),
or expressed another way: the lower uncertainty limit?
3.3.3.2 Numerical site calibration

In many real life cases the experimental site calibration is ruled out as too costly and time consuming. The
alternative is numerical flow modelling, which a priori is expected to be less accurate, but far less costly.
As outlined in Derrick et al. [39], and described in section 2.3.7, a number of different models are
available and used for site evaluation. The most widely used flow model is the so-called WASP model
developed at Risø National Laboratory. Nominally, the use of the model is limited to neutrally-stable wind
flows over low, smooth hills with attached flows. Being easy to apply, the model is nevertheless used
where terrain is excessively rugged and/or the atmosphere unstable. Mortensen and Petersen, 1997 [40],
evaluate success of the model by comparison with measurements made in Portugal. A total of 3½ years of
2
Note that the uncertainty of the wind speed measurement affects also uncertainty of shear, turbulence etc.

Site Evaluation
data from 8 met stations with separations of from 2 to 16 km were available for comparison. Only mean
wind speed was considered and compared from station to station. The influence of ruggedness of terrain
and the influence of quality/quantity of the topographical input data were analyzed, with the results
presented below.
A ruggedness index (RIX) was defined as the percentage of terrain slope between two points exceeding
0.3. The results for the Portuguese sites show that a low RIX at the base (reference) position and a high
RIX at the wt (predicted) site leads to over-estimation of wind speed at the wt site, and vice versa. The
prediction error is with fair approximation linear in the difference in RIX of base- and wt positions,
respectively. A difference in RIX of 20% corresponds to a prediction error of approx. 40%. The standard
error3 is of the order 5-10%.
The importance of the size of the area covered by the flow model was investigated, the result being that
the area should be 8x8 km or larger so as not to introduce significant errors this way. Also, the influence
of height-contour line interval was analyzed, indicating that intervals of 20m or less should be used to
avoid gross errors from this source. Finally, the accuracy of the contour representation was assessed.
Referring to Figure 3.1, the accuracy was varied by removing points P2 from a contour line P1P2P3,
where the distance x from that point P2 to the resulting approximate contour line, P1P3, was less than a
certain distance. It was found that errors were minimized if the distance, x, was 50m or less.
Original
Contour Line
P1
Approximate
Contour Line x P2
P3
Figure 3.1 Graphical definition of contour accuracy.

The uncertainty of numerical site calibration for shorter separations (100m to 1-2 km) is poorly
investigated so far. However, ongoing work in another EU financed project is expected to shed light on
the issue.
Discussion
• The study referred to focuses on errors of numerical estimation of climatic information from the base-
to the wt position over distances of 2 to 16 km. An important finding of Mortensen and Petersen [40]
is that the prediction error seemingly is not a function of distance between base- and wt position.
• Further, the study points to the ruggedness of the terrain as the most important error source. For large
ruggedness - where the WASP code is really outside its operation limits - the site calibration error may
be excessive.
• The good correlation between the defined ruggedness index seems to open ground for a “ruggedness
correction”. Assuming such a correction is made the remaining standard uncertainty is of the order 5-
3
Standard error: defined as the vertical root mean square deviation from a linear fit to the individual estimates, see
annex A.

Site Evaluation
10%, being smallest for low ruggedness.
3.3.4 Additional Uncertainties due to Long-Term Climate Variability

The annual mean wind speed is the standard, minimum output from an MCP or Wasp analysis. As
previously described, the “prediction” is derived by mapping the measured wind regime from a long-term
reference site onto the prediction site using either
• relationships derived from concurrent short-term measurements at both sites (MCP) or
• a numerical wind flow model (WAsP).
Therefore, the “prediction” is in fact a historical reconstruction of what the wind regime would have been
at the prediction site over the same period in time (usually of 10 to 20 years duration) that the long-term
measurements were carried out at the reference site. Wind farm financiers are interested in the future wind
regime which will influence the wind farm economics but all that can be delivered to them is a
reconstruction based on a period in the past as we cannot predict the future. This conflict of requirements
can be dealt with in numerical terms by considering extra components of uncertainty.
Consider as an example the case where an MCP analysis has been carried out on a site in the UK based on
a long term reference data set of 10 years duration. Very long-term (i.e. 100 year) statistics show that the
average annual wind speed at sites throughout the UK does vary from decade to decade. Therefore there is
a probability that the average wind speed experienced over the next decade will differ from that
experienced over the past decade and similarly, the mean wind speed experienced over a 10-year reference
period, 1980-1989 say, may well be different to that experienced over the next ten years. This variability
can be represented in terms of the basic statistics of these long-term measured wind speeds [41].
Therefore, it is possible to give an estimate in probabilistic terms of the most likely future wind speed
based on knowledge of the historic variability.
Besides the measurement uncertainties, the uncertainties associated with the correlation and prediction
stages of the MCP or WAsP include the following:
a) the statistical uncertainty in the short term wind speed correlations (MCP) or the numerical model
uncertainties, sa.
b) the statistical uncertainty in the reference period (e.g. ten year) sample mean due to long-term climate
variability at the long-term reference site, sb.
c) the statistical uncertainty in the annual mean wind speed prediction due to year to year variability of
the reference annual mean from the reference period (e.g. ten year) mean, sc..
The following sections detail the assessment of these individual components of uncertainty.
a) Uncertainties inherent in the prediction technique (MCP or WAsP)
As applied at present, the MCP analysis technique involves determination of the relationships between the
wind speeds at the reference and prediction sites over a short-term period. In most cases a simple “least-
squares” analysis is used to determine the best fit line relating the wind speeds at the prediction site to
those at the reference site. Other regression techniques may be used for which the principles described
here still apply. Using any statistical regression technique, it is generally possible not only to derive the
“best-fit” relationship but also an uncertainty due to the “goodness” of fit [37]. For example, if we have in
simple, one-independent-variable regression
V = α + βU (6)
then the uncertainties on the regression parameters4 are
4
Wannacott and Wannacott, Regression: a second course in statistics, John Wiley & Sons, 1881.

Site Evaluation
1 σ σ
sα = and sβ = (7)
n σu n
where n is the number of data sets, σ the standard error (scatter around line) and σu is the RMS of Ui (data
normalized so that <U>=0, ensuring independence of α and β).
When the regression results are used for prediction: The standard uncertainty of the mean of V (V0) when
predicted with a (fixed) value of U (U0) is then
σ U2
smean (V0 ) = 1 + 02 (8)
n σu
If 1) σu is small (bad coverage of wind speed range), 2) U0 large (far from mean of values used for
regression analysis - the case when talking extremes), or 3) σ is large (much scatter around line, bad
correlation), then we get a poorly determined prediction of V0.. Prediction of the individual V0 is much
larger5. This uncertainty can be carried through to the prediction stage of the analysis and appears as an
uncertainty in the predicted annual average wind speed. Experience of applying the MCP technique to
many sites in the UK reveals that the value of the standard uncertainty, sa,, associated with correlation
quality is typically 0.5 - 2.0% of the predicted annual average wind speed. If this uncertainty component is
found to be much greater than this, then it suggests that the reference site is not ideal and perhaps another
reference site offering a better wind speed relationship with the prediction site should be identified.
In the case of WAsP or a more sophisticated measurement based MCP (site calibration) than is presently
possible, the evaluation of this component of uncertainty is as described in Sections 3.3.1 to 3.3.3.
b) Statistical Uncertainty Due to Long-Term Climate Variability
Determination of this component of the uncertainty requires detailed knowledge of the very long-term
wind regime (i.e. 100 years). Such data are only available for a very limited number of sites, which are
generally unlikely to be suitable for use directly as the reference site in an MCP or WAsP analysis.
However, such long term trends are, within the geographical locality of a region or country such as the
UK, not likely to be strongly influenced by the specific site location. Rather, such trends are driven by
relatively large-scale climate variations. Therefore, in the case of the UK wind climate which is dominated
by Atlantic weather systems, it is assumed that the 100 year wind speed trends observed at the few UK
sites with such records can also be applied to any specific location within the UK .
Annual mean wind speed data for 31 UK Meteorological Office sites have been reconstructed for periods
of 100 years and are described in a report commissioned by the UK Department of Energy [41, 42]. These
data can be used to estimate the decade-to-decade wind speed variability. To do this, the statistics for the
ten-year running annual average wind speed for each of the 31 sites have been extracted from [42]. For
each of the 31 sites, the statistics of interest comprise the mean ten-year average wind speed and the
standard deviation. These data are plotted in Figure 3.2.
σ U 02
5
sone (V0 ) = 1+ 2 + n
n σu

Site Evaluation
0.3
Standard Deviation (m/s)
0.2
σ ave10
i
0.1
0
0 5 10 15
V ave10
i
10-Year Average Wind Speed (m/s)
Figure 3.2 Statistics of ten-year-average wind speeds for 31 sites over a 100-year period.
The data suggest that the standard deviation of ten-year-average wind speeds at all 31 sites over 100 years
is approximately proportional to the ten-year-average wind speed. Hence, for the predicted ten-year-
average wind speed at the prediction site, the appropriate standard deviation can be estimated by
determining and applying that relationship.
From the slope of the best-fit line to the data in Figure 3.2, the standard uncertainty, sb, of the ten-year
average due to long-term climate variability is estimated to be 2.4% of the ten-year-average wind speed
for sites in the UK. The situation in other countries may however be different (e.g. a study [35] in
Denmark indicated an RMS of 8 per cent on the long-term average wind speed. Due to the nature of the
input data for all such analyses, the uncertainty involved is unavoidably high.
c) Statistical Uncertainty Due to Inter-Annual Variability
Continuing with the ten-year average example, the measured ten-year average wind speed, µ 10 , over the
actual historical reference period of ten years also has an uncertainty associated with the sample variance.
Considering the ten annual average wind speeds from which the ten-year average is composed, which
have a standard deviation σ 1 , then the standard uncertainty in the ten year mean is
σ1
sc = (9)
10
Although this component of uncertainty has been derived from measurements at the reference site, it is
assumed to apply also to the prediction site for that same ten year period.
Combined Uncertainty
Finally, it is assumed that all three components of uncertainty are uncorrelated and hence can be combined
quadratically to give the standard uncertainty, s10 (excluding measurement effects) of the predicted ten
year average wind speed.
s10 = sa2 + sb2 + sc2 (10)
Exceedance Wind Speed

It is commonly required to present the predicted mean wind speed as that value which has a 90 per cent
chance of exceedance. The 90 per cent exceedance wind speed is that wind speed defined by the lower
boundary of the 80 per cent confidence band on the distribution of ten-year-average wind speeds. By

Site Evaluation
assuming that the ten-year-average wind speeds follow a normal distribution, the coverage factor, x,
corresponding to the 80 per cent confidence band is
x = 1.28
Hence, the ten-year-average wind speed with a 90 per cent chance of exceedance, Ve90 , is defined in terms
of the central estimate, V10 , and the standard uncertainty s10 ,by:
Ve90 = V10 − xs10 (11)
In illustrating this example, a ten-year averaging period has been used, however, the principles can be
applied to other averaging periods. Furthermore, the long-term climate data presented here is only
assumed applicable to the UK. Similar data would have to be sourced and processed for countries subject
to different climates.
3.3.5 Conclusions on Uncertainty

Though some work has been done there is a fundamental lack of firm information on uncertainties related
to experimental site calibration. Basic problems are the lack of a precise definition of quantities to
measure, foremost “wind speed”. The presently used cup anemometer has several inherent problems such
as non-ideal cosine response.
An alternative definition of site calibration has been introduced as a means to reduce the inherent
uncertainties related to conventional site calibration.
A frequently used flow model has been investigated for large separations (2-16 km), showing that the
quality of the model’s prediction is strongly dependent on the ruggedness of the terrain in question, but
not the magnitude of separation. Assuming that corrections are made for ruggedness, the standard
uncertainty was of the order 5-10%, smallest for small ruggedness.

Site Evaluation
4. OTHER PARAMETERS
This section deals with the site assessment parameters other than annual average wind speed (which is
treated in other parts of the document), as these appear in the relevant standard IEC 1400-1 [2]. These
parameters include the “other” parameters describing normal wind conditions and the “additional”
parameters which are used in the various sub-models specified in the standard. For each parameter
method(s) of measurement, analysis and uncertainty calculation are given. Since in some cases the
standard leaves room for interpretation, a list of suggested changes to definitions in the standard is
presented in section 4.7.
According to the standard, normal wind conditions are specified by the three so-called basic parameters;
• Reference wind speed
• Annual average wind speed at hub height, and
• Turbulence intensity at 15 m/s;
in combination with models for
• Wind speed distribution,
• Normal wind profile, and
• Normal turbulence.
These wind conditions must be assessed at the site. Since the annual average wind speed at hub height is
related to the wind speed distribution model, and the turbulence intensity at 15 m/s is related to the normal
turbulence model, four parameters remain to be assessed. In this chapter methods to measure these four
parameters are presented: reference wind speed (section 4.1), turbulence intensity at 15 m/s (section 4.2),
annual average wind speed distribution (section 4.2.1) and normal wind profile (section 4.3.1).
The theoretical background to each parameter is presented in Appendix A.
4.1 Reference Wind Speed

4.1.1 Description of Parameter
According to the standard (IEC1400-1ed2 [2]), the reference wind speed, Vref, is the extreme 10 min
average wind speed at turbine hub height with a recurrence period of 50 years. The objective thus is to
specify methods to determine this reference wind speed from observed wind speeds, which must be 10
min average wind speeds.
Note that extreme wind speed is a stochastic parameter which must be expressed in terms of probabilities.
This means that values of the reference wind speed are probabilistic rather than exact. That is, Vref does not
occur exactly once in a period of 50 years; there is a probability of 1-(0.98)50 = 64% that Vref is exceeded
once or more in a period of 50 years. A more detailed investigation of the analysis of extreme wind speeds
is presented in the work of sub-project 1 (Load Spectra and Extreme Winds), earlier in this report.
4.1.2 Methods of Assessment

Given the definition of the reference wind speed, determining Vref, requires determining the cumulative
distribution function Fe(ue) of extreme wind speed ue. More specifically, it involves determining the
inverse dispersion factor α and the modal value β of the Gumbel distribution from observed extreme wind
speeds. There are two methods to achieve this:
By directly estimating α and β from the extremes in a time series of wind speed, and

Site Evaluation
By indirectly estimating α and β from the Weibull parameters A and k, and the number of independent
extremes M; where A, k and M are estimated from a time series of wind speed.
In this report the “direct” method is considered to be the standard Gumbel analysis. To date, various forms
and refinements of this standard analysis have been presented amongst which:
The original Gumbel method [43], which has also been presented in various forms,
The method of independent storms [44],
The Gumbel-Lieblein method [45], and
The Gumbel-Harris method [46].
In this report the “indirect” method is considered to be the generalized Gumbel analysis. It will be referred
to as the Gumbel-Bergström method [47]
Only various forms of the original Gumbel method together with the Gumbel-Bergström method will be
presented in this report.
4.1.2.1 Original Gumbel Method

The original Gumbel method was made operational by Shellard [43]. In this method it is assumed that a
series of M independent observations of extreme wind speed is available. Furthermore it is assumed that M
is large enough for the asymptotic distribution to be valid. Since how to establish a minimal value for M is
not part of the method, M usually follows from the procedure to select independent extremes. For
example, this is realized by finding the annual largest wind speeds from many-years of hourly average
wind speeds. Since the practical value of M is usually small, generally the Gumbel distribution is not
valid.
However, in the original Gumbel method the probability Fe(ue) an extreme wind speed ue is not exceeded
in one year is identified with a value
j
Fe, j = (12)
M +1
where j is the rank of the extreme when sorted in ascending order. Usually Fj is referred to as the plotting
position. Since it is assumed the distribution
Fe,j ≡ F(ue,j) = exp(- exp(- α(ue,j - β ) ) ) (13)
is valid, it follows that
( ) (
ln − ln Fe, j = −α ue, j − β ) (14)
As a consequence, the Gumbel parameters α and β can be obtained from the slope and an intercept of a
plot of ln ( - ln Fe,j ) versus ue,j. When performed graphically, this method is referred to as the “Gumbel
paper” method (which usually employs probability or recurrence period as the horizontal axis and extreme
wind speed as the vertical axis).
To deal with the effect of too low a number of independent extremes (most probably) [48], an empirically
improved plotting position was suggested [49, 50]:
j − 0.3
Fe, j = (15)
M + 0.4
This equation should give a more realistic probability of the smaller as well as the larger extremes.

Site Evaluation
The procedure to obtain the Gumbel parameters α and β on the basis of the original Gumbel method then
is:
• Collect ten-minute average wind speeds during many years;
• Select the M annual-largest ten-minute average wind speeds ue,i, where M is the number of years;
• Rank the extreme wind speeds ue,i in ascending order, so that a new series ue,j of extreme wind speeds
is obtained (smallest extreme j = 1 and largest extreme j = M);
• Determine the M quantities xj = ue,j;
• Either determine the M quantities
  j 
(a) y j = ln − ln   or
  M + 1 
  j − 0.3  
(b) y j = ln − ln  ;
  M + 0.4  
• On the basis of the xj and yj find a value for the parameters α and β by
(a) numerically determining the best linear fit to the relation y = - α x + Β, giving α and Β, and
(b) solving β from Β = αβ
The reference wind speed Vref subsequently follows from
Vref = β −
1
α
( ( )) with
ln − ln Fe,Vref Fe,Vref = 0.98 (16)
On that basis the uncertainty ∆Vref in the reference wind speed is given by:
( ( ( ))) (∆α ) + (∆β )

2 2
 ∂Vref   ∂Vref  1 2
(∆V )
2
 (∆α ) +   (∆β ) = 4 ln − ln Fe,Vref
2 2 2 2
ref = (17)
 ∂β   ∂β  α
Here ∆α and ∆β are the uncertainties in the inverse dispersion factor and the modal value, as these follow
from the best linear fit.
4.1.2.2 Gumbel-Bergström Method

In the Gumbel-Bergström method it is recognized that the Weibull distribution is the parent distribution of
the distribution of extreme wind speed [47]. This implies the relations can be employed which express the
Gumbel parameters α and β in terms of the Weibull parameters A and k, and the number of independent
observations M. In other words: If the distribution of average wind speed at a site is known in terms of the
two Weibull parameters, the distribution of extreme wind speed at that site is known as well when the
number of independent wind speeds is known. Hence, a method is provided to determine M from a time
series of average wind speed. Note this time series need not be a long-time record.
As a first step, M must be determined. Suppose now wind speed has been measured with an averaging
time Tave during a total time T. Then a time series with N=T/Tave average wind speeds is available. In
general, these N observations are dependent. In order to obtain independent observations, the time series
must be broken down into M independent sub-series. This number is related to the total time T via M =
T/TM, where TM is the time between independent observations in the time series. Equivalently, M follows
from
M = ν MT , (18)

Site Evaluation
where νM = 1/TM is the frequency of independent observations.

An average time between independent observations in a time series may be obtained from the average
time between two observations which are correlated for 50% [47]. If S(f) is the spectral density function
of the time series, it can be shown that the average frequency of independent observations can be
estimated from
f2
∫ f 2 S ( f )df
f1
ν 2M = f2
(19)
∫ S ( f )df
f1
with f1 = 1/2Tc the lower and f2=1/2Tave the upper integration limit; where Tc is a cut-off time scale.
It has however been found that νM does not depend strongly on the actual spectral density function, so that
values from a typical spectrum may be used. These values are presented in Table 4.1. The data in this table
allow convenient rules of thumb to be established. For example, if a time series is obtained with an
averaging time of 10 min, every third observation is independent from the first one.
As a second step in the Gumbel-Bergström method, the Weibull parameters A and k must be determined
from the time series with average wind speeds. To this end standard methods can be employed, which
methods will not be presented here.
Having available the parameters M, A and k, the Gumbel parameters follow from
1 A
= (ln M ) k and β = A(ln M ) k
1 1
−1
(20)
α k
It however has been reported that α’s and β’s from this equation yield a biased extreme value analysis. To
solve this problem an empirical correction was suggested: increase the value of the Weibull parameter A
as obtained from the time series by 10% [51].
Table 4.1 The frequency νM of independent wind speeds as a function of the averaging time Tave.
Tave [s] νM [Hz]
3600 2.8 10-5
600 7.3 10-4
60 1.0 10-3
10 4.6 10-3
5 7.9 10-3
3 1.0 10-2
1 2.4 10-2
The procedure to obtain the Gumbel parameters α and β on the basis of the Gumbel-Bergström method is
then:
• Collect ten-minute average wind speeds during a period T,
(a) Either determine the frequency νM of independent wind speeds from the spectrum of the observed
wind speeds, or
(b) use the value νM = 7.3 10-4 Hz from a typical spectrum;
• Determine the number M of independent wind speeds from M = νMT;
• Determine the Weibull parameters A and k from the time series with ten-minute average wind speeds
using a standard method;
• (Optional) Increase the Weibull scale parameter A by 10%;

Site Evaluation
1 A
= (ln M ) k and β = A(ln M ) k
1 1
−1
Determine the Gumbel parameters α and β from
α k
The reference wind speed Vref subsequently follows from
Vref = β −
1
α
( ( )) with
ln − ln Fe,Vref Fe,Vref = 0.98 (21)
The uncertainty in the reference wind speed is conveniently expressed in terms of the uncertainty in the
Gumbel parameters:
2 2
 ∂Vref   ∂Vref 
(∆V )
2
 (∆α ) +   (∆β )
2 2
ref = (22)
 ∂α   ∂β 
Since the Gumbel parameters now depend on the Weibull parameters A and k as well as the number M of
independent wind speeds, their uncertainties follow from
2− 2k  k  2
2 2 2 2
1  1   k   1
(∆α ) 2
= 2 (ln M )   (∆A) + 1 + ln ln M  (∆k ) +   1 −  (ln M ) (∆M )  and
2 2 −2
A  A   k   M  k 
 A2  A 
2
2
(∆β ) = (ln M ) (∆A) + 4 (ln ln M ) (∆k ) +   (ln M ) (∆M ) 
2
2 2 2 2 −2
k
(23), (24)
 k  kM  
respectively. The uncertainties ∆A and ∆k in the Weibull parameters follow from the Weibull method.
The uncertainty in the number of independent wind speeds is
∆M = T∆ν M + ν M ∆T (25)
where ∆νm is the uncertainty in the frequency of independent observations and ∆T is the uncertainty in the
total time. (Note these uncertainties are not statistical in nature.)
4.1.3 Instrument requirements

Since determining the reference wind speed basically comes down to measuring wind speeds and
averaging over an appropriate time period, instrument requirements are as described in IEC 1400-12 [37]
in relation to anemometry. However, it should be noted that the anemometer and data acquisition system
measurement range should be configured to allow measurement of wind speeds over a much greater range
than would be required for a power performance measurement. Whereas a power performance
measurement may only require measurement of wind speeds at up to just above turbine cut-out (e.g. 30
m/s), in a site assessment all wind speeds are of interest and so the upper limit of the measurement range
(depending on the site) may be 60 m/s or more. These requirements are described in section 4.3.3.
4.1.4 Further Reading

The original standard Gumbel method for extreme wind speed analysis was first presented by Shellard
[43]. Overviews of this method can be found in the books by Freris [52] and Wieringa and Rijkoort [50]
and the paper by Mayne [53]. Other standard Gumbel methods were described by Lieblein [45], Cook
[44], and Harris [46]; the latter also providing an overview of the standard Gumbel methods. The general
Gumbel method, introducing how to determine the number of independent observations from the data,

Site Evaluation
was first presented by Bergström [47]. A different approach was taken by Rijkoort [54], also described in
Wieringa and Rijkoort [50], who developed a so-called compound Weibull model allowing for analysis of
extreme winds. Since it deviates from the general “Gumbel” trend and requires a laborious data analysis,
the Rijkoort method was not presented here.
4.2 Reference Turbulence Intensity

According to the standard (section 7.2 [2]), turbulence intensity at 15 m/s I15 is the characteristic value
(mean plus standard deviation) of hub height turbulence intensity at a 10 min average wind speed of 15
m/s. The value I15 shall be determined using appropriate statistical techniques applied to measured data
obtained at wind speeds greater than 10 m/s. (Note, for example, turbulence intensity can be
overestimated when low-frequency trends exist in the wind speed data.) Where topographical or other
local effects may influence the turbulence intensity then these effects shall be represented in the data. The
characteristics of the anemometer, sampling rate and averaging time used to obtain measured data can
influence the assessment of turbulence intensity. These effects shall be considered when predicting the
turbulence intensity from measured data. The objective thus is to specify methods to determine this
characteristic turbulence intensity from wind speeds observed over periods of 10 min.

If wind speed is measured contiguously with a sampling time ∆T over a time period T, the mean µi and the
standard deviation σi of the wind speed in the ith time interval follow from:
1 1
∫ (u(t ) − µ )
ti +T ti + T
u(t )dt and
2
µ=
T ∫t = ti
σ 2i =
T t = ti
i dt (26)
The turbulence intensity Ii for the ith time interval subsequently is defined as the ratio of the standard
deviation and the mean of the wind speed for the ith time interval:
σi
Ii ≡ (27)
µi
In addition, the turbulence intensity Iu at the wind speed u is defined as
σu
Iu ≡ (28)
u
that is the ratio of the standard deviation of the wind speed at the wind speed u, and that wind speed.
(Note the difference in the definition of Iu with respect to the standard; see below.)
The turbulence intensity Iu is obtained by conditionally sampling the turbulence intensities Ii. To this end
those Ii are selected for which | µi - u | < ∆u; where ∆u is the bin width. If the corresponding number of
time intervals is M, the average turbulence intensity at u is
M
1
µ i ,u =
M
∑I
j =1
j (29)
The corresponding standard deviation σI,u of the turbulence intensity at u follows from:

Site Evaluation
M
1
∑ (I )
2
σ 2I ,u = j − µ I ,u (30)
M j =1
~
The characteristic turbulence intensity I u subsequently is defined as the sum of the mean and the standard
deviation of the turbulence intensity at the wind speed u:
~
I u ≡ µ I ,u + σ I ,u (31)
(Note the difference in notation with respect to the standard; see below.) It is straightforward to show that
~ ~
the characteristic turbulence intensity I 15 at a wind speed of 15 m/s is I u with u = 15 m/s. The bin width
∆u may be selected as 0.5 m/s in line with IEC 88/1400-12.
The preceding method is based on the definition in the standard, which states that “I15” is a characteristic
turbulence intensity (rather than the turbulence intensity). To prevent “I15” from being erroneously
interpreted as the turbulence intensity at a wind speed of 15 m/s, as could be read from the description of
symbols and units in the standard (section 1.4 [2]), in this method the symbol Iu is exclusively used for the
~
turbulence intensity at the wind speed u. Hence the symbol I u has been introduced for the characteristic
turbulence intensity. See section 4.7 for suggestions for improving the terminology used in the standard
with respect to turbulence.
The conditional sampling procedure, retaining only the Ii’s for which | µi - u | < ∆u with u = 15m/s and ∆u
~
= 0.5m/s, automatically ensures that the value I u is determined from wind speeds which are greater than
10 m/s.
In general, turbulence intensity is affected by topographical effects like orography, surface roughness and
obstacles, where obstacles can be other wind turbines. Usually this implies that turbulence intensity
depends on the wind direction. As a consequence, a sectorwise analysis must be performed resulting in
~
sectorwise values of I u . If a conservative approach is required, the largest sector value may subsequently
be chosen as the one representing the site, otherwise, the overall average value for all directions may be
adequate
~
The procedure to obtain the characteristic turbulence intensity I 15 at a wind speed of 15 m/s is then:
• Collect ten-minute average µi and standard deviation σi data of the wind speed, and ten-minute
average wind direction during a representative time period (the reference standard recommends a
minimum period of 6 months);
• Determine the ten-minute turbulence intensities Ii = σi/µi;
• For each wind direction sector (propose 30 degree sectors), select the M observations of Ii for which
14.75 m/s < µi < 15.25 m/s;
M
1
• For each sector determine the average turbulence intensity at 15 m/s from: µ I ,15 =
M
∑I
j =1
j
• For each sector determine the standard deviation in the turbulence intensity at 15 m/s from:
M
1
∑ (I )
2
σ 2I ,15 = j − µ I ,15 ;
M j =1
~
• For each sector determine the characteristic turbulence intensity at 15 m/s: I 15 ≡ µ I ,15 + σ I ,15
~
• For a conservative assessment, select the largest sectorwise I 15 as the one representing the site,

Site Evaluation
~
otherwise determine the overall average value of I 15 to represent the site condition.
4.2.3 Measurement Requirements
4.2.3.1 Anemometer
Determining turbulence characteristics requires an anemometer which has a sufficiently fast response to
measure the small-scale variations. Although the cup anemometer is not the ideal instrument for this work,
compliance with the standard for power measurement as described in IEC 88/1400-12 ([section 3.2 [37])
warrants its use.
In a turbulent wind a cup anemometer over estimates the mean wind speed. As a consequence, the
resulting turbulence intensity is underestimated. This so-called overspeeding originates from a cup
anemometer responding faster to an increase in wind speed than to a decrease of the same magnitude due
to inertial effects. On experimental as well as theoretical grounds it has been found that the over-speeding
is proportional to the turbulence intensity of the wind. If u is the mean wind speed and um is the mean
wind speed measured by the cup anemometer, in the case of negligible vertical turbulence the relative
over-speeding may be expressed as [55]:
um − u
∝ J u IU2 (32)
u
where the factor Ju depends on the characteristics of the anemometer (in the form of the distance constant)
and the wind (in the form of the inverse length scale of the energy containing eddies). It follows that the
mean wind speed can be obtained from the mean wind speed measured by the cup anemometer:
1
u≈ um (33)
1 + J u I u2
A typical expression for the factor Ju is [55]:

Lκ E
Ju = (34)
1 + Lκ E
where L is the distance constant of the cup anemometer and κE is the inverse length scale of the energy
containing eddies. These expressions can be used to derive a requirement for L on the basis of the wind
characteristics and the accepted relative over-speeding:
1  u 
=κ E I u2 − 1 (35)
L  um − u 
For example, if the expected turbulence intensity at a site is 25% and the scale of the energy containing
eddies is 10m, a cup anemometer must be used with a distance constant of less than 1.9 m if the relative
over-speeding is to be less than 1%. Ideally, a cup anemometer is employed which has a small intrinsic
over-speeding. On the basis of an earlier recommendation (section 7.4)[56], it is recommended here to
only use cup anemometers with a distance constant of less than 2 m when turbulence characteristics are to
be measured.
On the other hand, these expressions can be used in order to determine the relative over-speeding of a
given cup anemometer, which by the way must have a distance constant of less than 5 m to comply with
the standard for power measurement (section 3.2 [37]). Since this over-speeding affects the turbulence
intensity through the average wind speed, a compensated value µ’i for the average wind speed must be

Site Evaluation
employed:
1
µ ′i = µi (36)
1 + J u I i2
The compensated turbulence intensity I’i in that interval subsequently is:

σi
I i′ = (37)
µ ′i
Note the standard deviation σi of the wind speed in the interval is not compensated for the non-linear
response of the cup anemometer.
The requirements for the anemometer thus are as follows:
• A cup anemometer with a distance constant of less than 2 m is recommended;
• If a different cup anemometer is used, it must have a distance constant smaller than 5 m.
Regarding application of the data from the cup anemometer:

• For each time interval the relative over-speeding must be estimated from JuIi2, where Ii is the
turbulence intensity measured in that interval;
• The over-speeding effect on mean wind speed should then be considered in the determination of the
uncertainty in the turbulence intensity Ii . It is unlikely to be sensible to apply an over-speeding
correction in the determination of the turbulence intensity however.
These requirements ensure that the effect of cup anemometer over-speeding on the characteristic
turbulence intensity is quantified.
4.2.3.2 Data Acquisition and Pre-processing

The anemometer signals must be sampled at a rate such that all relevant wind speed variations are
recorded. Since the Kolmogorov frequency fK is the highest frequency in a turbulent wind, the sampling
rate should theoretically be 2fK.. Obviously, no cup anemometer can meet this requirement because a cup
anemometer has a limited frequency response. If fm is the cup anemometer cut-off frequency, the sampling
rate must be of the order of fm. (Note that the cup anemometer is a poor instrument for turbulence
measurement because generally fm < fK.) To comply with the standard for power measurement, the
minimal sampling rate must however be 0.5 Hz (section 3.7 [37]).
Turbulence, i.e. the variation of the wind speed, has a characteristic time scale τ. In the framework of the
standard, in a given time interval T, the characteristic parameter is the turbulence intensity, which by
definition is the ratio of the standard deviation and the average of the wind speed. By convention this
standard deviation is determined with respect to the average, which implies determining turbulence
intensity requires the average wind speed to be constant. Now in practice the average wind speed may
change over the time interval T. If so, apart from contributions with a time scale τ, the standard deviation
also contains contributions with a time scale T. In order words, in this case low-frequency events are
represented in the standard deviation and, consequently, the turbulence intensity. Since, as a consequence,
the standard deviation is too large, the resulting turbulence intensity is too large as well.
Having available the average µi and the standard deviation σi of the wind speed after an averaging time T,
there is no method to determine after the event whether the average wind speed has changed or not during
that time interval. This means one must make sure a priori that variations with a time scale of the order T
(or larger) are not recorded. This is achieved by high-pass filtering the signal from the anemometer, where
the lower cut-off frequency is of the order 1/T.

Site Evaluation
Alternatively, the data acquisition system may be designed such, that based on wind speed samples
recorded at intervals ∆T during the interval T, moving averages are determined on the basis of an interval
proportional to τ. If these differ considerably, a trend is present. This trend must subsequently be removed,
e.g. by determining the difference between the recorded samples and the moving averages. Next, mean
and variance are determined from the resulting de-trended samples.
On the other hand, one must make sure that noise, i.e. variations with a small but not relevant time scale,
are not recorded either. This is achieved by low-pass filtering the signal from the anemometer. Although
for an ideal anemometer the higher cut-off frequency is of the order 5 times the Kolmogorov frequency,
for a cup anemometer this relaxes to 5fm. Taken together, the requirement for data pre-processing is as
follows:
The signal from the anemometer must be sampled with a sampling time of 2 s or less, but ideally with
1/fm, where fm is the cup anemometer cut-off frequency;
The signal from the anemometer must be band-pass filtered for frequencies between 1/T and 5fm, where T
is the averaging time.
This ensures the relevant turbulent scales are recorded, and neither low-frequency events nor noise are
part of the data.

Determining a characteristic turbulence intensity mainly involves applying standard experimental and
processing techniques to wind speed measurements. Although these techniques are commonly applied in
wind energy, for this reason it was not possible to trace the first record of the method. Here it must
however be noted that in wind energy turbulence intensity is determined in a different way than in
meteorology, where it is obtained from the variation in time series of average wind speed and not from
considering the standard deviation with respect to this average. This means there is no direct link with the
rich literature on atmospheric turbulence.
4.3 Wind Speed Distribution

According to the standard (section 3.3.1a [2]), the wind speed distribution at the site is significant for the
wind turbine design because it determines the frequency of occurrence of the individual load conditions.
For WTGS in the standard classes, the mean value of the wind speed over a time period of 10 min shall be
assumed to be Rayleigh distributed for the purposes of design load calculations. In this case, the
probability distribution of wind speed at hub height is given by:
  V 2
P(V < Vhub ) = 1 − exp −π  hub   (38)
  2Vave  
with P(V < Vhub) the probability of the wind speed to be smaller than Vhub, Vhub the ten minute average
wind speed at hub height, and Vave the annual average wind speed at hub height. The objective thus is to
specify methods to determine the Rayleigh parameter “average wind speed Vave” by fitting a Rayleigh
distribution to an observed (or, more likely derived/predicted) annual wind speed distribution. It should be
remembered that in most cases, the Rayleigh or Weibul distributions are not good fits to real wind data in
many parts of Europe and that this fitting process is mainly required to categorize the site in terms of the
reference standard IEC 1400-1[2]. It should be checked that forcing a Weibull or Rayleigh distribution to
fit the data does not produce a distribution which is more benign to the turbine than the actual, non-

Site Evaluation
Rayleigh distribution of winds.

For ease of notation, the equivalent cumulative distribution function FR(u) is introduced:
 π  u 2 
FR (u) = 1 − exp −    (39)
 4  u  
which gives the probability a wind speed is smaller than u if the mean wind speed is u . Hence
dFR (u) π u  π  u 2 
f R (u) ≡ = exp −    (40)
du 2 u2  4  u  
is the corresponding probability density function, i.e. the probability the wind speed is u given the mean
wind speed u .

The Rayleigh distribution is basically a Weibull distribution with shape parameter k=2. This means
methods to determine the Rayleigh parameter u can be derived from the (many) methods to determine the
two Weibull parameters A and k. These methods have been grouped into five categories:
Moment fitting method,
Logarithmic method,
Maximum likelihood method,
Percentile method, and
Least-squares method.
In the following, for each of these methods first a rough description for the Weibull case and next a more
detailed description for the Rayleigh case are given. In practice, the Weibull distribution gives an accurate
wind description for moderate wind speeds only [50, 57]. This is because
Low wind speeds, less than 4 m/s, say, mainly occur under stable conditions;
High wind speeds, more than 16 m/s, say, are not very frequent and thus can not be fit accurately.
4.3.2.1 Moment Fitting Method

In Weibull parameter estimation this method employs moments of the wind speed. The first step is to
determine a value for k by deriving quantities from ratios of moments of the wind speed, such ratios,
according to the Weibull distribution, depending only on k. Subsequently, k is solved on the basis of the
observed values of wind speed. Examples of these ratios are the non-dimensional standard deviation
(standard deviation method) and the energy pattern factor (energy pattern method). As an alternative, the
energy in the wind speed distribution and the frequency of wind speeds are combined into a relation
which depends on k only (European Wind Atlas method). Then a value for A is determined, again on the
basis of a moment of the wind speed (usually the first) as obtained from the Weibull distribution and the
observations. In Rayleigh parameter estimation only the second step can be performed, so that u is
determined from the average observed wind speed:
N
1
u=
N
∑u
i =1
i (41)
where N is the number of (relevant) ten-minute records and ui is the ith (relevant) ten-minute averaged
wind speed. The uncertainty ∆u in the moment estimate of the mean wind speed is obtained from its

Site Evaluation
variance:
N
1
(∆u) ( )
2 2
= ∑ ui − u
N ( N − 1) i =1
(42)
N
( )∑ (u )
2
(Recall 1 ( N − 1) i −u is the sample variance, which is an unbiased estimate of the variance of a
i =1
random variable.) Note this uncertainty expresses variation about the mean wind speed rather than the
difference between the measured and the “real” value.
The procedure to obtain the mean wind speed u on the basis of the moment method is then:
• Collect/derive ten-minute average wind speeds ui over a representative time period;
• Select the N observations for which 4 m/s < ui < 16 m/s;
N
1
• Determine the moment estimate of u from u =
N
∑u ;
i =1
i
• Determine the uncertainty ∆u in the moment estimate of the mean wind speed from
N
1
(∆u) ( )
2 2
= ∑ ui − u .
N ( N − 1) i =1
4.3.2.2 Logarithmic method

In Weibull parameter estimation this method consists of reducing the cumulative distribution function
such that a linear relation between functions of A and k is obtained. To this end the natural logarithm is
taken twice of a rewritten cumulative distribution function. The coefficients of the resulting best linear fit
are obtained either graphically by using so-called Weibull paper or numerically by using a standard least-
squares method. Values for the two Weibull parameters are subsequently found from the slope and an
intercept of the best linear fit. This method is usually referred to as the “Weibull paper” method.
In Rayleigh parameter estimation the first step to obtain the logarithmic method is to rewrite the
cumulative distribution function, with as a result:
1 π  u 2 
= exp     (43)
1 − FR (u)  4  u  
Taking the natural logarithm twice of the left and right-hand side of this equation yields:
1 π
ln ln = ln − 2 ln u + 2 ln u (44)
1 − FR (u) 4
Since this equation is a linear relation between ln(ln(1 / (1-FR(u)))) and ln(u), with slope 2 and intercept
ln(π 4) − 2 ln u , the mean wind speed can be obtained by solving for the intercept. As a consequence, the
uncertainty ∆u in the logarithmic estimate of the mean wind speed is determined by the uncertainty in the
intercept only.
The procedure to obtain the mean wind speed u using the logarithmic method is then:
• Collect/derive ten-minute average wind speeds ui during a representative time period;
• Create the cumulative distribution F(ui) of the selected data ui;

Site Evaluation
• Determine the N quantities yi= ln(ln ( 1 / (1-F(ui) ))) and xi = ln(ui);

• From xi and yi, find a value for the parameter B by numerically determining the best linear fit to the
relation y= 2x + B;
• Determine the logarithmic estimate of u from 2 ln u = ln(π 4) − B
• Determine the uncertainty ∆u in the logarithmic estimate of the mean wind speed from,
2 ln ∆ u = ln(π 16) − B + 2 ln( B∆B) where ∆B is the uncertainty in B.
4.3.2.3 Maximum Likelihood Method

In Weibull parameter estimation A and k are determined such that the resulting probability density makes
the observed wind speeds most likely. To this end the likelihood function (i.e. the joint probability density
function) is introduced. By maximising this likelihood function (or rather its logarithm), two equations in
A and k are obtained. Subsequently, by solving these equations the maximum likelihood estimates of the
two Weibull parameters are obtained.
In Rayleigh parameter estimation the likelihood function L is the joint probability density function
N N
π ui  π  ui  2 
L = ∏ f R (ui ) = ∏ exp −    (45)
i =1 i =1 2 u2  4  u  
where ui is the ith observed wind speed. By taking the logarithm, this equation gives
π N π 1 N
ln L = N ln + ∑ ln ui − 2 N ln u − ∑u i
2
(46)
2 i=1 4 u2 i =1
Maximising ln(L),
∂ ln L
=0 (47)
∂u
subsequently yields
2N π N
−
u
+ 3 ∑u i
2
=0 (48)
2u i =1
The maximum likelihood estimate of u hence follows from
π N
∑u
2 2
u = i (49)
4N i =1
Formally, the uncertainty ∆u in the mean wind speed is given by
(∆u) ()
2
=V u (50)
where V u () is the variance of u (Recall that the mean wind speed is a stochastic quantity with a
( )
distribution too.) Using V f ( x ) = (∂f ∂x ) V ( x ) the variance of u is expressed in terms of the variance
2
of u 2

Site Evaluation
()
V u =
1V u ( ) 2
(51)
4 u2
Subsequently, by introducing the sample average µu2 of ui2 and the sample variance σ2u2 of ui2 according to
N
1 1 N
µ u2 ≡
N
∑ ui 2 and σ u2 2 ≡
i =1 N − 1 i =1
(
∑ ui 2 − µ u2 2 ) (52)
it follows that
( )  π 2 π2
N
2
V u =V
 4N
∑
i =1
ui 

=
16 N
σ u2 2 (53)
By inserting this equation into equation (51) it is found that the variance in the mean wind speed is:
2
π 2 σ u2
()
V u =
64 N u 2
(54)
so that, as a consequence, from equation (50) it follows that the uncertainty in the maximum likelihood
estimate of the mean wind speed is
π σ u2
∆u = (55)
8 N u
Here the sample standard deviation σu2 and the mean wind speed u are given by the equations (52) and
(49), respectively.
• The procedure to obtain the maximum likelihood estimate of the mean wind speed u is then:
π N
∑u
2 2
• Determine the maximum likelihood estimate of the mean wind speed u from u = i ;
4N i =1
• Determine the uncertainty ∆u in the maximum likelihood estimate of the mean wind speed from
π σ u2
∆u = , where σu2 is the sample standard deviation of ui2.
8 N u
4.3.2.4 Percentile method

In Weibull parameter estimation this method involves determining A and k by estimating two percentiles
of the observed wind speeds. (The 100pth percentile is the wind speed for which the cumulated
distribution is p.) By taking the logarithm twice of the cumulative distribution function, and employing
these percentiles and the corresponding probabilities, two equations in A and k are obtained. The estimates
of the two Weibull parameters are subsequently obtained by solving these equations.
In Rayleigh parameter estimation the probability pi of the wind speed being smaller than ui follows from
the cumulative distribution function:

Site Evaluation
 π u 2
pi = 1 − exp −  i   (56)
 4  u  
which is equivalent to
 1  π
ln ln   = ln − 2 ln u + 2 ln ui (57)
1 − pi  4
Hence, if ux is the 100xth percentile, this equation implies that
 1  π
ln ln   = ln − 2 ln u + 2 ln u x (58)
1 − p x  4
Although any percentile may be used, the 50th percentile may be preferred because it is the median of the
observed wind speeds. In this case the percentile estimate of the mean wind speed is obtained from
π
2 ln u = − ln ln 2 + ln + 2 ln umed (59)
4
where umed is the median of the observed wind speeds. Consequently, the uncertainty in the percentile
estimate of the mean wind speed is
u
∆u = ∆umed (60)
umed
where
N
1
(∆umed )2 = ∑ (ui − umed )2 (61)
N ( N − 1) i =1
gives the uncertainty ∆umed in the median wind speed as obtained from the variation about the median
wind speed.
The procedure to obtain the mean wind speed u using the percentile method is then:
• Determine the median wind speed umed from the selected wind speeds;
• Determine the percentile estimate of the mean wind speed u from
π
• 2 ln u = − ln ln 2 + ln
+ 2 ln umed .
4
• Determine the uncertainty ∆u in the percentile estimate of the mean wind speed from
N
u 1
∆umed , with (∆umed ) = ∑ (ui − umed )2
2
∆u =
umed N ( N − 1) i =1
4.3.2.5 Least-Squares Method

In Weibull parameter estimation this method involves determining A and k by minimising the sum of
squares defined by the difference between the cumulative distribution function and the observed

Site Evaluation
cumulative distribution. If the logarithms of these distributions are employed, this gives two equations in
A and k from which the two Weibull parameters are obtained.
If FR is the cumulative distribution function and F is the observed cumulative distribution, in Rayleigh
parameter estimation the sum of squares is
N
∑ ( F (u ) − F (u ))
2
R i i (62)
i =1
with ui the ith observed wind speed. Alternatively, by using the auxiliary function
G(u) = ln(1 - F(u)) (63)
it follows that an equivalent sum of squares S is
N
S = ∑ ( G R (ui ) − G(ui ))
2
(64)
i =1
Minimising S,
∂S
=0 (65)
∂u
subsequently gives
π N N
2 ∑ ui4 + ∑ ui2 ln(1 − F (ui )) = 0 (66)

4u i =1 i =1
The least-squares estimate of u hence follows from

N
2 π
∑u
i =1
i
4
u =− N
(67)
4
∑u i
2
ln(1 − F (ui ))
i =1
(Note ln(1-F(ui)) < 0 because 0 < F < 1.)
Given the variance in u from the equations (50) and (51) it follows that the uncertainty ∆u in the mean
wind speed is formally given by
(∆u)
2
=
1V u ( ) 2
(68)
4 u2
In order to derive the variance V u ( ) equation (67) is rewritten in the form

2
2 π n
u =− (69)
4d
where n and d are the auxiliary quantities

Site Evaluation
N N N N
n ≡ ∑ ni ≡ ∑ ui 4 and d ≡ ∑ d i ≡ ∑ ui 2 ln(1 − F (ui )) (70)
i =1 i =1 i =1 i =1
Equation (69) allows one to express V u ( ) in terms of these auxiliary quantities and their variances V(n)
2
and V(d):
V u ( ) 2
=
π2 1 
2  ( )
16 d 
n2 
V n + 2 V (d )
d 
(71)
Here the variances of n and d are

V(n) = Nσ2n and V(d) = Nσ2d (72)
respectively, where
1 N 1 N
σ n2 = ∑ (ni − µ n )2 with µn = ∑n i (73)
N − 1 i =1 N i =1
(sample variance σ2n and sample average µn of ni), and
1 N 1 N
σ d2 = ∑ (d i − µ d )2 with µd = ∑d i (74)
N − 1 i =1 N i =1
(sample variance σ2d and sample average µd of di); recall ni = ui4 and
di = ui2 ln (1 - F(ui)). (75)
By inserting the equation (71) into the equation (68), and using the equations (72) and (69), it is found
that the uncertainty ∆u in the least-squares estimate of the mean wind speed is given by
π N  2 n2 2
( )
2
∆u =− σ n + 2 σ d  (76)
16 nd  d 
The auxiliary quantities n and d together with their sample variances σ2n and σ2d, which appear in this
equation, are defined in the equations (70), (73) and (74).
The procedure to obtain the mean wind speed u using the least-squares method is then:
• Collect/derive ten-minute average wind speeds ui during a representative time period;
• Create the cumulative distribution F(ui) of the selected data ui;
2 π n
• Determine the least-squares mean wind speed u from u = −
4d
• Determine the uncertainty in the least-squares mean wind speed u from equation (75) and the
associated auxiliary quantities.

Determining Vave basically comes down to measuring ten-minute average wind speeds which can be used
to determine a wind speed distribution over a representative time period. If long term measurements are
not available at the site under assessment, then this will involve using one of the extrapolation or

Site Evaluation
modelling techniques described in section 3 of this report (e.g. MCP techniques or WAsP model).
According to the reference standard, the measurement time period is 6 months or longer and must include
seasonal effects if these significantly contribute to the wind conditions.
The specifications for the wind speed measurement system may be taken from the Power Performance
standard IEC 88/1400-12 [37], from which the following relevant requirements are identified:
• The meteorological mast shall be of a tube or cone or lattice type;
• The mast should be located in a position that represents the free wind flow that drives the planned
wind turbine systems,
• Wind speed measurements shall be made with a cup anemometer;
• The cup anemometer shall have a distance constant of less than 5 m, and maintains its calibration
over the duration of the measurement period.
• It is recommended that a cup anemometer shall be mounted within 2.5% of the hub height of the
planned wind turbine generator system. If this is not practical due to the size of the turbine, a series of
anemometers (at least 3) at various heights from about 10m height to the top of the mast should be
used so that an extrapolation of wind speed to the planned hub height may be carried out from
simultaneous measurements at these lower heights (see section 4.4).
• The highest anemometer should preferably be mounted on the top of a vertical circular tube standing
clear of the top of the meteorological mast or alternatively on a boom clamped to the side of the mast
pointing in the predominant wind direction. If booms are used, the vertical separation from the
anemometer to any mounting boom should be at least 7 times the boom diameter and its horizontal
separation from the mast at anemometer height should be at least 7 times the maximum mast
diameter;
• Data shall be collected continuously at a sampling rate of 0.5 Hz or faster;
• The data acquisition system shall store either sampled data or pre-processed data sets, or both;
• The pre-processed data sets shall comprise the following information on the sampled data: mean
value, standard deviation, maximum value, and minimum value;
• The total duration of each pre-processed data set shall be between 30 s and 10 min, and shall be 10
min divided by an integer number. If the data sets have a duration of less than 10 min, the adjacent
data sets shall not be separated by a time delay.

To determine the Weibull or Rayleigh parameters essentially involves application of standard estimation
techniques. For this reason it was difficult to trace the various methods to a genuine founder, apart from
the percentile method which was first reported by Dubey [58]. Operational reviews of Weibull parameter
estimation methods are available in the books by Wieringa and Rijkoort [50] and Lysen [59], and the
report by Stevens [60]. The European Wind Atlas method, basically a moment fitting method, was, as its
name suggests, first published in the European Wind Atlas [57].
4.4 Normal Wind Profile

As described in the standard ([2] section 3.3.1) the wind profile V(z) denotes the average wind speed as a
function of height z above the ground. For WTGS designs in the standard classes, the normal wind speed
profile shall be assumed to be given by the power law:
α
 z 
V (z ) = Vhub   (77)
 z hub 

Site Evaluation
where Vhub is the ten minute average wind speed at hub height, zhub is the hub height, and where the power
law exponent α shall be assumed to be 0.2. The assumed wind profile is used to define the average wind
shear across the rotor swept area. Application of this model might lead to underestimating the wind shear
(and subsequently the loading), in particular when the power law exponent at the site is larger than 0.2 or
when the power law is not a good model anyway. This implies that in practice, the normal wind profile
should be compared to the observed profile, and, if needed, a more realistic profile must be fitted to the
observations. The objective thus is to specify methods to
Determine the power law exponent from wind speed measured at various heights, and
Decide when use of the power law results in too low a loading.
Note that wind speed profiles during stable atmospheric conditions could lead to too low an estimate of
the wind shear, so that one should consider taking into account diurnal variation as well.
4.4.2.1 General Power Law

In a turbulent boundary layer over a smooth surface and under stable conditions, the wind speed u at
height z above the surface may be described by the general power law
n
 z
u(z ) = u∞   (78)
δ 
where u∞ is the free-stream velocity, δ is the boundary layer thickness, and n is the power law exponent
{Hinze}. If the wind shear is to be determined from observed wind speeds at height zref, the general power
law gives
n
 z 
u(z ) = u zref   (79)
 z ref 
As a consequence, for wind shear evaluation, only the power law exponent needs to be determined. By
taking the logarithm of the general power law, it reduces to
ln(u) = ln(u∞) + nln(z) - nln(δ) (80)
This implies the power law exponent n can be obtained from the slope of the curve with ln(u) versus ln(z).
As a consequence, the uncertainty in the power law exponent follows from the uncertainty in this slope.
The procedure to obtain the exponent n of the general power law is then:
Collect ten-minute average wind speeds ui at M heights zj over a representative time period;
Select the N observations for which 4 m/s < ui < 16 m/s;
Determine the quantities xj = ln(zj) and yij = ln(ui(zj));
Based on xj and yij, find a value for the parameters A and B by numerically determining the best linear fit
to the relation y= Ax + B;
Determine n from n = A;
Determine the uncertainty ∆n from ∆n = ∆A, where ∆A is the uncertainty in A.
4.4.2.2 Logarithmic Law

In a turbulent boundary layer over a rough surface and under stable conditions, the wind speed u at height
z above the surface may be described by the logarithmic law

Site Evaluation
u∗ z
u(z ) = ln (81)
κ z0
where u* is the friction velocity, κ is the Von Karman constant, and z0 is the roughness length [61]. If the
wind shear is to be determined from observed wind speeds at height zref, the logarithmic law gives
( ) lnlnz z −−lnlnz z
u(z ) = u z ref 0
(82)
ref 0
which implies that, for wind shear evaluation, only the roughness length needs to be determined.
Rewriting the power law gives
u∗ u
u(z ) = ln z − ∗ ln z 0 (83)
κ κ
This equation reveals that the roughness length z0 can be obtained from the root of the curve of u versus
ln(z). (In addition, given the value κ = 0.4, the friction velocity u* can be obtained from the slope of this
curve.) Consequently, the uncertainty in the roughness length is obtained from the uncertainty in this root.
The procedure to obtain the roughness length z0 of the logarithmic law is then:
Collect ten-minute average wind speeds ui at M heights zj during a representative time period;
Select the N observations for which 4 m/s < ui < 16 m/s;
Determine the quantities xj = ln(zj) and yij = ui(zj);
Based on xj and yij, find a value for the parameters A and B by numerically determining the best linear fit
to the relation y = Ax + B;
Determine z0 from ln(z0)= - B/A;
Using the uncertainties ∆A in A and ∆B in B, determine the uncertainty ∆z0 from
2 ln(∆z0) = - 2 ln(A) - 2B /A + ln((B∆A/A)2 + (∆B)2).

Determining the parameters in wind profiles basically comes down to measuring wind speeds at different
heights. Although the minimum requirement is two anemometers, the requirement of one anemometer to
be within 5% of hub height (see IEC 88/1400-12 (ed.1)[37]) warrants three anemometers to be employed.
Other relevant requirements as described in IEC 88/1400-12 (ed.1) [37], are already described in section
4.3.3.
4.4.4 Method to Determine which Profile Gives the Higher Loading
4.4.4.1 Relation between Loading and Wind Profile

In the following it is assumed that the loading increases with the wind shear. This means that a high
loading originates from a large vertical wind speed difference over the rotor plane. If the hub height is H
and the rotor radius is R, this wind speed difference is
δu(H,R) ≡ u (H + R) - u (H - R). (84)
In the following δu(H,R) will be determined for wind profiles according to the general power law and the
logarithmic law.

Site Evaluation
4.4.4.2 Wind Speed Difference - General Power Law

In the case of the general power law (equation (78)), if the wind speed at hub height is u(H), the wind
speed at the height H + R is
n
 H + R
u( H + R) = u( H ) 
 H 
n
 R
= u( H ) 1 +  (85)
 H
 R 1  R
2
1  R
3

= u( H )1 + n + n(n − 1)  + n(n − 1)(n − 2)  + h. o. t 
 H 2  H 6  H 
Similarly, the wind speed at the height H - R is
 R 1  R
2
1  R
3

u( H − R) = u( H )1 − n + n(n − 1)  − n(n − 1)(n − 2)  + h. o. t  (86)
 H 2  H 6  H 
As a consequence, the vertical wind speed difference over the rotor plane is
R  1  R 
2
δu ≈ 2n u( H )1 + (n − 1)(n − 2)   (87)

H  6  H  
This means that if a general power law is used in order to describe the wind profile, the wind speed
difference δu can be estimated by using equation (87).
4.4.4.3 Wind Speed Difference - Logarithmic Law

In the case of the logarithmic law (equation (81)), if the wind speed at hub height is u(H), the wind speed
at the height H + R is
ln( H + R) − ln z 0
u( H + R) = u( H )
ln H − ln z 0
  R 
 ln 1 +  
 H
= u( H )1 +  (87)
 ln H − ln z 0 
 
 R 1 R
2
1 R
3

 +   +   − h. o. t 
H 2  H 3 H
= u( H )1 + 
 ln H − ln z 0 
 
 
Similarly, the wind speed at the height H - R is

Site Evaluation
 R 1 R
2
1 R
3

 +   +   − h. o. t 
H 2  H 3 H
u( H − R) ≈ u( H )1 −  (88)
 ln H − ln z 0 
 
 
As a consequence, the vertical wind speed difference over the rotor plane is
2 R  1 R2
δu ≈ u( H )1 +    (89)
ln H − ln z 0 H  3  H  
This implies that if a logarithmic law describes the wind profile, using equation (89) the wind speed
difference δu can be estimated.
4.4.4.4 Profile with the Higher Loading

In order to establish which profile gives the higher loading, the general power law and the logarithmic law
are fitted to wind speeds which are measured at different heights. This yields values for the power law
exponent n (see section 4.4.2.1) and the roughness length z0 (see section 4.4.2.2). This subsequently
allows one to determine the corresponding wind speed differences δupow from the equation (87) and δulog
from the equation (89). From a comparison of the values of δupow and δulog it is straightforward to decide
which value is the higher, and, consequently, which profile gives the higher loading.

Determining the parameters of either a power law or a logarithmic law basically requires application of
standard estimation techniques. For this reason no effort was put into finding a founder of these methods.
4.5 Wind Direction Distribution

The wind direction distribution prediction must be based on modelling (e.g. WAsP) or extrapolation
techniques (e.g. MCP) as described in section 3. It is therefore not considered further here.
4.6 “Other” Parameters

4.6.1 Overview
In the course of the EWTS II project, sub-project 1, “Load spectra and extreme wind conditions”
additional site assessment parameters were identified. These parameters, which are used in the various
models which are specified in the standard (section 3.3.1[2]), include:
• Wind direction distribution
• Wind speed distribution
• Roughness length
• Wind shear
• Turbulence length scales
• Coherence
• Standard deviation ratios
• Negative gust

Site Evaluation
Wind direction distribution and wind speed distribution have been treated in the first part of sub-project 5.
Wind speed distribution is also covered in section 4.3 of this report. Furthermore, in section 4.4 roughness
length and wind shear are presented.
This chapter deals with the other “additional” parameters. Since, apart from negative gust, these
parameters originate from the theory of series, a review of the basics of series analysis is presented in
Appendix A Assessment Of Other Parameters - Mathematical Details. On the basis of this theory
turbulence length scales (section 4.6.2), coherence (section 4.6.3), and standard deviation ratios (section
4.6.4) are introduced subsequent to this chapter. In addition, negative gust is described (section 4.6.5).
Finally, the requirements for measurement of these parameters are presented (section 4.6.6 and 4.6.7).
4.6.2 Turbulence Length Scale

By definition (p. 417 [62]), integral length scale is the integral of the correlation coefficient over all
separation distances. This means a spatial description of the wind speed field is required, or, in other
words, many measuring stations along a line are needed. Since this is difficult to achieve, even in a
laboratory situation, the practical approach has been to consider two stations only but to vary the
separation distance (section 6.4 [63], section 1.7 [61]). Accordingly, the wind speed components ui in the
positions x and x + r are
ui [ x ] and u j [ x + r ] (90)
where r is the distance between the two positions. From equation (A16) the corresponding covariance
function is
( )(
Rij [r ] = ∑ u j [ x ] − u j ui [ x + r ] − u i ) (91)
By using equation (A18) to determine the corresponding correlation function rij®, from the definition it
follows that the integral length scale is
1+ N
Lij = ∑ r [r ]
r =1− N
ij (92)
Note that there are nine integral length scales because there are three wind speed components. As it is un-
practical to consider two measuring stations and to vary the separation distance in a field situation, an
alternative approach is used (section 6.4 [63], section 1.7 [61]): an integral “length” scale Li is obtained by
using the integral time scale Tii and the average wind speed u i via
Li = u i Tii (93)
Here the integral time scale Tij is the integral of the correlation coefficient over all time lapses:
1+ N
Tij = ∑ r [τ ]
r =1− N
ij (94)
Hence one measuring station and a temporal description of the wind speed are used. This process is
described in more detail by way of an example in Appendix B, Use of sonic anemometer.. Since there are
three average wind speed components, in this context there are three integral length scales: longitudinal
L1, transversal L2 and vertical L3. Although integral/turbulence length scale is not defined in the standard
[2], it is assumed here that integral length scale is equivalent to turbulence length scale.

Site Evaluation
4.6.3 Coherence
By definition (p. 418 [62]), coherence is the absolute value of the normalized cross-spectrum for two wind
speeds measured at points separated in space. This warrants a two-point spatial description of the wind
speed field. If the two wind speed time series which are measured at a distance R apart are
ui[x] and uj[x + R] , (95)
their covariance is
( )( )
Rij [ R] = ∑ u j [ x ] − u j [ x ] ui [ x + r ] − u i [ x + R]
n
(96)
where the summation and the averages are over all elements n in the time series. On basis of equation
(A19) the corresponding spectral density is
1
hij [w] = ∑ R [R]exp(− jwk )
ij (97)
2π k
Accordingly, the coherence is:
hij [w]
cij [w] = (98)
hii [w]h jj [w]
This means there are six cross-coherences: | c12[w] | and | c21[w] |, | c13[w] | and |c31[w] |, and | c23[w] |
and | c32[w] | ; where the pairs are equal if the wind speed field is homogeneous, i.e. does not depend on
the actual position. Note the three auto-coherences | c11 | = | c22 | = | c33 | ≡ 1. In the framework of the
standard [appendix B]{iec1400-1ed2}, coherency function (is) defined by the complex magnitude of the
cross-spectral density of the longitudinal wind velocity components at two spatially separated points
divided by the auto-spectrum function. This suggests, according to the standard, the coherence is | c11 |;
which can not be the case because | c11 | ≡ 1. A solution might be to use an alternative definition of the
coherency function in the sense that the factor in the denominator of equation (A20) is hii[0], hjj[0] rather
than hii [f], hjj[f], so that the coherence is:
hij [w]
cij [w] = (99)
hii [0]h jj [0]
Since in this case | c11[w] | = h11[w]/ h11[0] ≠ 1, coherence is worthwhile. However, from neither the
standard nor the literature is it clear which definition of the coherency function is employed. Equation
(A20), however, is the common one in series analyses.
4.6.4 Standard Deviation Ratios

By definition, wind speed standard deviation σ is the square-root of the averaged squared difference
between the wind speed realizations and the average wind speed. This means a temporal wind speed
description ui[n] in a measuring position is required.
From the description of the covariance function in equation (A16) it follows that the standard deviation σi
of the ith wind speed component is given by

Site Evaluation
Nt
1 1
∑ (u [n ] − u )
2
σi 2
= i t i = Rii [τ = 0] (100)
Nt n =1 Nt
so that
Rii [0]
σi = (101)
Nt
Hence the ratio of the standard deviation of a lateral or vertical and the longitudinal wind speed
component is:
σ2 R22 [0] σ3 R33 [0]

= and = (102), (103)
σ1 R11 [0] σ1 R11 [0]
These equations show that these ratios are conveniently obtained from the auto-covariance functions of
the time series with wind speed components. Although the standard does not contain a definition of wind
speed standard deviation (appendix B [2]), equations (102) and (103) are in agreement with the standard’s
definition of turbulence intensity.
4.6.5 Negative Gust

According to the standard (section 1.3 [2]), a gust is a temporary change in the wind speed which may be
characterized by its rise-time, its magnitude and its duration, where the change may either be an increase
or a decrease. Since in general “gust” is associated with an increasing wind speed, the term “negative
gust” was introduced to indicate a decreasing wind speed. It has been observed that usually a negative gust
occurs before a gust [64], where the amplitude of the negative gust may be as large as 25% of the one of
the gust. Therefore, the combined effect on turbine loads of a gust preceded by a negative gust could be
much more severe than that of the gust alone.
Consider the time series ui[n] with the ith wind speed component. Gusts and negative gusts are detected by
identifying extremes in the wind speed, e.g. by marking zeros in the derivative of the series. The wind
speed extremes so determined constitute a series ue,i[m] with alternating minima and maxima. There is an
associated series te,i[m] with the time instants when the extremes in the ith component occur. The
magnitude δUi of wind speed change in the ith component is the difference between the wind speed in two
subsequent extremes:
δUi = ue,i[m] - ue,i[m-1] . (104)
Likewise, the “rise time” δTi is
δTi = te,i[m] - te,i[m-1] . (105)
The change is a decrease if a minimum is preceded by a maximum; otherwise it is an increase. If an
increase is sufficiently strong, the wind speed change is a gust characterized by the gust magnitude Ug,i:
Ug,i = ue,i[m] - ue,i [m-1] > Hg,i (106)
where Hg,i is the selection level. On the other hand, the wind speed change is a negative gust if a decrease
is sufficiently strong:
Ung,i = ue,i[m-1] - ue,i[m] > Hng,i (107)
where Ung,i is the negative-gust magnitude and Hng,i is the selection level. In both cases the “rise time” of

Site Evaluation
the (negative) gust is:

T(n)g,i = te,i[m] - te,i[m-1]. (108)
Note gusts and negative gusts may be observed in the three velocity components independently, defining
longitudinal, lateral and vertical (negative) gusts. If a longitudinal gust is preceded by a longitudinal
negative gust, an operating gust in the sense of the standard occurs. The duration of this operating gust is
the sum of the “rise time” of the gust and the one of the negative gust.
4.6.6 Instrumentation Requirements

The three dimensional characteristics of most of the “other” parameters require three-dimensional wind
velocity measuring equipment. The cup-anemometer does not meet this specification. In addition, the
mechanical characteristics of the cup anemometer limit the response of the instrument and so higher
frequency spectral information is lost. Section 4.6.7 discusses in more detail the special instrumentation
requirements of these parameters.
4.6.7 Special Instrumentation: The Ultrasonic Anemometer
4.6.7.1 Principle of operation of the sonic anemometer

Sonic anemometers make use of the propagation of sound to measure the wind velocity. An acoustic pulse
is fired alternatively from two given sensors (functioning both as transmitters and receivers) and travels
along a given path between the sensors. It can be proved that the time it takes for the pulse to reach the
sensor depends on the wind speed component along that path. This is schematically shown in Figure 4.1
[65].

Site Evaluation
S
Sn
Sl
ct2
Slt1
ct1
St2 Sn t 2
S nt1 St1
α α
T1 T2 Slt2
Ultrasonic sound pulses are transmitted back and forth sequentially along the same path
between the transducers T1 and T2 and the elapsed times t1 and t2 are measured along the
same path.
a = arcsin(Sn/c)
S n= wind speed component perpendicular to the sound path
Sl = wind speed component parallel to the sound path
c = speed of sound
S= wind speed
l l
= c ⋅ cos α + S l = c ⋅ cos α − S l
t1 t2
1 c ⋅ cos α + S l 1 c ⋅ cos α − S l
= =
t1 l t2 l
1 1 S l 1 1
− = 2⋅ l Sl = ⋅ − 
t1 t2 l 2  t1 t2 
Figure 4.1 Principle of operation of the sonic anemometer (Courtesy Ref. [65])
In reality a sonic anemometer consists, as a rule, of three pairs of sensors arranged in a three dimensional
array. The form of the array depends on the manufacturer and can either have the form of an orthogonal
co-ordinate system with individual paths for each sensor pair or, as in some other types of instruments,
Figure 4.2, the sensor pairs are oriented at 120° azimuth intervals while being inclined at 45° to the
horizontal such that the sensor paths intersect. In this way sonic anemometers can measure the three
velocity components and their turbulence characteristics.

Site Evaluation
Figure 4.2 The Gill sonic anemometer
4.6.7.2 Cup- and sonic anemometers, merits and drawbacks

The most widely used instrument for the measurement of the wind speed is the cup anemometer. It is
cheaper than the sonic anemometer, more robust, and its behaviour is better known and understood.
Furthermore well established procedures exist for its calibration and mounting, such that measurements
carried out with cup anemometers can be accepted between measuring institutes -provided these
procedures are observed. Together with the mean value of the wind speed, the cup anemometer measures
also the turbulence characteristics of the wind speed but its response capabilities are limited. The response
of the cup anemometer (see section 4.2.3.1) is known to be a function of the mean wind speed, the degree
of turbulence and the inclination of the instrument relative to the velocity vector. Performance
deterioration with time is a known problem for cup anemometers and therefore they should periodically be
calibrated in order to ensure that their response satisfies certain pre-defined limits.
Compared to the cup anemometer, the sonic anemometer can produce all three components of the velocity
vector and their associated fluctuations as well as higher order moments. Additionally some sonic
anemometers also determine the ambient temperature. The sonic anemometer has a faster response than
the cup anemometer as it has no moving mechanical parts. It’s response depends on the instrument’s built-
in electronics while a physical limitation on its response is imposed by the length of the acoustic paths,
[66]. Among its drawbacks has been (and still is) the relatively high purchase price which however has
become more reasonable in recent years.
There are a number of potential error sources associated with the use of the sonic anemometer. A detailed
presentation of these can be found in [65, 67, 68]. The most important of them are:

Site Evaluation
• inaccuracies of the probe geometry during manufacturing, as a result of which, the actual geometric
characteristics of the sensor array (i.e. positions and angles) are not as intended. To compensate for
this, some of manufacturers individually calibrate their instruments and equip them with individual
correction look-up tables.
• temperature sensitivity of the instrument’s electronics which results in thermal drift and contamination
of the measured values.
• flow distortion due to the design of the sensor array and the supporting struts of the instrument’s head.
To account for the above errors many manufacturers supply the sonic anemometers with instrument
specific calibration tables for the on line correction of the results. The correctness or adequacy of these
tables has not been documented by relevant investigations. It should be mentioned that no
investigations exist showing to what extent the shape of the sonic anemometer body influences the
measured wind characteristics.
• distortion of the sonic head geometry from wind loads during high wind speeds due to lack of stiffness
of the construction of the sensor array.
• lack of commonly agreed procedures for the mounting of the sonic anemometers, either for boom- or
for top mounting. Furthermore no agreed procedures exist as to the orientation of the sensor array so
the users generally follow the manufacturers’ instructions. For top mounting it is anticipated that rules
corresponding to the placement of cup anemometers should be sufficient.
As a result of the aforementioned potential sources of error, and since no generally accepted procedures
exist for its mounting or calibration, it is suggested that the sonic anemometer should mainly be used for
the determination of the turbulent characteristics of the atmospheric flows and the relative sizes of the
mean wind vector components. For the measurement of the absolute mean wind speed values the cup
anemometer should be preferred.
4.7 Clarification of Standard

The preceding text has presented methods to measure and calculate the site assessment parameters which
appear in the standard IEC 1400-1 [2]. Since in some cases the standard leaves room for interpretation, or
even gives rise to confusion, a list of recommendations is presented.
• The definition of reference wind speed does not specify the meaning of recurrence period. Since
several exist, it is recommended to include the appropriate description of recurrence period in the
definition of reference wind speed: the reference wind speed Vref is the extreme 10 min average wind
speed at turbine hub height with a recurrence period of 50 years, where this recurrence period is the
inverse annual probability that Vref is exceeded.
• It is recommended to include in the list of symbols and units that recurrence period Tr is the inverse
annual probability an extreme wind speed is exceeded.
• From the description in the list of symbols and units, it can be erroneously concluded that turbulence
intensity I15 is the turbulence intensity at a wind speed of 15 m/s. Since this is not correct (I15 is a
characteristic turbulence intensity at 15 m/s) it is recommended to remove I15 and its description from
the list of symbols and units, and to introduce a specific notation for the characteristic turbulence
~
intensity at a wind speed of 15 m/s, for example I 15 , throughout the standard. Furthermore, it is
~
recommended to include in the list of symbols and units that I 15 is the characteristic value, i.e. the
sum of the average and the standard deviation, of the turbulence intensity at a wind speed of 15 m/s.
• It is recommended to include in the standard a definition of turbulence length scale. Assuming that in
the context of the standard turbulence length scale is equivalent to integral length scale, it is
recommended that a turbulence length scale is the integral time scale of a wind speed component
multiplied by the corresponding average of the wind speed component, where this integral time scale
is the integral over all time lapses of the auto-correlation function of a time series of the wind speed

Site Evaluation
component.
• Since in the standard coherence is not unambiguously defined, it is recommended to include a
definition which complies with the “engineering rules” for coherence that are specified in the
standard.
• A definition of standard deviation ratio must be included: Standard deviation ratio is the standard
deviation in the lateral or vertical wind speed component divided by the standard deviation in the
longitudinal component.
• Since it has been observed that usually an increase in the wind speed is preceded by a decrease, it is
recommended to change the definition of a gust accordingly: A gust is a significant temporary
increase in the wind speed. It is usually preceded by a negative gust, that being a temporary decrease
in the wind speed. A gust and the preceding negative gust constitute an operating gust. A (negative)
gust is characterized by its rise-time and magnitude. The duration of an operating gust is the sum of
the rise times of the gust and the negative gust.

Site Evaluation
5. REFERENCES
[1] IEC 1400-1, Wind Turbine Generator Systems - Part 1: Safety Requirements; First Edition 1994-12
[2] IEC 1400-1, Wind Turbine Generator Systems - Part 1: Safety Requirements; 88/69/CD (Committee
Draft), Edition 2, May 1996.
[3] Butterfield S, Holley, W, et al, Report on 88/69/CD - Wind Turbine Generator Systems Part 1: Safety
Requirements, 2. edition.
[4] Verheij F.J, Snoek L.v.d, (ed.), Handboek energie-opbrengsten van windturbines, TNO Rapport 88-
145, 1989 (in Dutch).
[5] Sherman C: A mass-consistent model for wind fields in complex terrain, J. Appl. Meteorol., 17, 312.,
1978.
[6] Traci R.M: Development of a wind energy site selection methodology, US Dept. of Energy Rept.
RLO/2440-11, 205pp., 1977.
[7] Croba D. et al: Improving techniques for statistical and physical modelling of wind resources in
complex terrain, Proc. ECWEC’96, Goteborg, Sweden, pp.514, 1996.
[8] Jackson P.S., Hunt J.C.R: Turbulent wind flow over a low hill, Q.J.R. Meteorol. Soc., 101, 929-955.,
1975.
[9] Mason P.J., Sykes R.I: Q. J.R Meteorol. Soc., 105, 383-395, 1979.
[10] Walmsley J.L., et al: Bound. Layer Meteorol., 36, 157-186, 1986.
[11] Mason P.J, King J.C: Measurements and prediction of flow and turbulence over an isolated hill of
moderate slope, Q. J. R. Meteorol. Soc., 111, pp.617-640, 1985.
[12] Taylor P.A: Numerical studies of neutrally stratified planetary boundary layer flow above gentle
topography. I: Two dimensional cases, Bound. Layer Meteorol., 12, pp.37-60, 1977.
[13] Tryfonopoulos D..A. , et al: Wind Engineering, 13, 324-337, 1989.
[14] Alm L.K., Nygaard T.A: Flow over complex terrain estimated by a general purpose Navier-Stokes
solver, Proc. EWEC’93, Travemunde, Germany, pp.591, 1993.
[15] Tombrou M., et al: Tests of prediction effectiveness of wind energy computer models in complex
terrain, Proc. ECWEC’93, Travemunde , Germany, pp599, 1993.
[16] Bergeles G. C: Numerical calculation of turbulent flow around two-dimensional hills, J. Wind Eng.
and Ind. Aerod., Vol.21, pp.307-321., 1985.
[17] Petersen E. L., et al: Measurements and modelling in complex terrain, Proc. 1996, Goteborg, Sweden,
pp.580, 1996.
[18] Landberg L: Implementing prediction of power from wind farms at Utilities, Proc. 1996, Goteborg,
Sweden, pp.592, 1996.
[19] Mengelkamp H.T., et al: Regional and local wind climatologies over heterogeneous terrain, Proc.
ECWEC’96, Goteborg, Sweden, pp.506, 1996
[20] Hojstrup J. et al: Wind resources in the Baltic Sea, Proc. ECWEC’96, Goteborg, Sweden, pp.510,
1996.
[21] Lalas D. P., et al: A hybrid micro-siting model for wind flow simulation over complex topographies,
Proc. ECWEC’94, Thessaloniki, Greece, pp, 1994.
[22] Frank H.P., Landberg L.: Modelling the wind climate over Ireland, Proc. ECWEC’96, Goteborg,
Sweden, pp.631,1996.
[23] Hollis D. et al: Techniques for estimating the local wind climatology, ETSU W/11/00400/REP, 1995.
[24] Bowen A.J., Mortensen N.G: Exploring the limits of WAsP, the Wind Atlas Analysis and Application
Program, Proc.ECWEC’96, Goteborg, Sweden, pp.584., 1996.
[25] Barthelmie L.J., et al: Application of the WAsP model to determine the wind resource in non-neutral
conditions in coastal areas, Proc. ECWEC’96, Goteborg, Sweden, pp.627, 1996.
[26] Brocklehurst F.K., et al, ETSU document

Site Evaluation
[27] Hannah P., et al: Comparison of wind speed modelling techniques at existing and potential wind farm
sites across the UK, Proc. Wind Energy Conversion 1995, Warwick, UK, July, 1995.
[28] Addendum to [23]
[29] Pedersen A.M.J, et al: Reliability of wind farm production forecasts
[30] Nygaard T.A: Estimating expected energy capture at potential wind turbine sites in Norway, Proc.
EWEC’91, Amsterdam, Netherlands, pp.787., 1991.
[31] Frandsen S., Christensen C.J: Accuracy of estimation of energy production from wind power plants ,
Wind Engineering, Vol.16, No. 5, pp.257, 1992.
[32] Derrick A., Development of the Measure-Correlate-Predict strategy for site assessment, Proc. EC
Wind Energy Conf. 1993, Lubeck-Travemunde, Germany, pp681-685, 1993.
[33] Mortimer, A., A new correlation/prediction method for potential wind farm sites, Proc. BWEA Conf.
1994, Stirling, Scotland, pp349, 1994.
[34] Frandsen, S., On Uncertainties in Power Performance measurements. Sixth ASME Wind Energy
Symposium, Dallas, Texas, USA, February 15-19, 1987.
[35] Frandsen, S. and J.C. Christensen, Accuracy of estimation of energy production from wind power
plants, Jour. Wind Engineering vol. 16 No. 5, 1992.
[36] Guide to expression of uncertainty in measurement, International Organisation for Standardisation
(ISO), 1993.
[37] Wind turbine generator systems, Part 12: Wind turbine power performance testing; Final draft
international standard, International Electrotechnical Commission (IEC), 1400-12 Ed. 1, FDIS 88, 1997.
[38] Pedersen, T.F. and U.S. Paulsen, A procedure for classification of cup anemometers, EWEC’97,
Dublin, Ireland, October, 1997.
[39] Derrick, A. et al., A unified approach to the evaluation of the site specific wind characteristics for use
in both energy and load modelling of potential wind turbine development site, EWEC’97, Dublin, Ireland,
October, 1997.
[40] Mortensen, N.G. and E.L. Petersen, Influence of topographical input data on the accuracy of wind
flow modelling in complex terrain, EWEC’97, Dublin, Ireland, October, 1997.
[41] Palutikof, J.P, et al, The impact of climate variability on the UK wind resource, ETSU WN 6029-P1,
1993.
[42] Palutikof, J.P, et al, The impact of climate variability on the UK wind resource, ETSU WN 6029-P2,
1993
[43] H.C. Shellard, 1965, The estimation of design wind speed, In: Proc. First Int. Conf. on Wind Effects
on Buildings and Structures, London, 1965.
[44] N.J. Cook, Towards better estimation of extreme winds, J. Wind Eng. Ind. Aerodyn. 9, p. 295. , 1982.
[45] J. Lieblein, Efficient methods of extreme-value methodology, Report NBSIR 74-602, National Bureau
of Standards, 1974.
[46] R.I. Harris, Gumbel re-visited - A new look at extreme value statistics applied to wind speeds, J.
Wind Eng. Ind. Aerodyn. 59, p. 1, 1996.
[47] H. Bergström, Distribution of extreme wind speed, Wind Energy Report WE92:2, Department of
Meteorology, Uppsala University, 1992.
[48] C. Cunnane, Unbiased plotting positions - A review, J. Hydrol. 37, p. 205, 1978.
[49] A. Benard and E.C. Bos-Levenbach, The plotting of observations on probability paper, Statistica 7, p.
163, 1953.
[50] J. Wieringa and P.J. Rijkoort, Windklimaat van Nederland, Staatsdrukkerij, 1983.
[51] A.G. Davenport, The dependence of wind loads on meteorological parameters, Proc. Second Int.
Conf. on Wind Effects on Buildings and Structures, 1967, Ottowa, 1967.
[52] L.L. Freris, Wind energy conversion systems, Prentice Hall, 1990.
[53] J.R. Mayne, The estimation of extreme winds, J. Ind. Aerodyn. {\bf 5}, p. 109, 1979.
[54] P.J. Rijkoort, A compound Weibull model for the description of surface wind velocity distributions,
Koninklijk Nederlands Meteorologisch Instituut, Report W.R. 83-13, 1983.

Site Evaluation
[55] N.E. Busch, O. Christensen, L. Kristensen, L. Lading and S.E. Larsen, Meteorological field
instrumentation - Wind speed and direction by means of cups, vanes, propellers and lasers, Report Risø-
R-400, Risø National Laboratory, 1979.
[56] C.J. Christensen and J.B. Dragt (ed.), Accuracy of power curve measurements, Report Risø-M-2632,
Risø National Laboratory,1986.
[57] I. Troen and E.L. Petersen, European Wind Atlas, Commission of the European Communities, 1989.
[58] S.D. Dubey, Some percentile estimators for Weibull parameters, Technometrics 9, p. 119, 1967.
[59] E.H. Lysen, Introduction to wind energy (2nd edition), Publication CWD 82-1, Consultancy services
Wind energy Developing countries, 1983.
[60] M. Stevens, De bruikbaarheid van Weibullfuncties bij de bepaling van de opbrengst van
windenergiesystemen, Report R-370-A, Dept. Applied Physics, Eindhoven University of Technology,
1979.
[61] J.O. Hinze, Turbulence (2nd ed.), McGraw-Hill, 1975.
[62] D.A. Spera (ed.), 1994, Wind turbine technology, ASME Press.
[63] H. Tennekes and J.L. Lumley, A first course in turbulence, MIT Press, 1972.
[64] C.P. Butterfield, C. Stork, W.E. Holley, P.H. Madsen and P.H. Jensen, External condition models
from the International Electrotechnical Commission `Safety of Wind Turbine Standards (1400-1)’, Proc.
16th ASME Wind Energy Symposium, 1997, Reno, 1997.
[65] Mortensenn N.G. (1994), “Flow response characteristics of the Kaijo Denki omni-directional sonic
anemometer (TR-61B)”, Risø-R-704(EN).
[66] Kaimal,J.C., Finnigan,J.J. (1994) “Atmospheric boundary layer flows-their structure and
measurement”, Oxford Univ. Press.
[67] Mortensen, N.G. (1994), “Wind measurements for wind energy applications - a review”, 16th BWEA
Conference Stirling 15-17 June 1994, Scotland, pp 353-360.
[68] Mortensen, N.G. and Højstrup Jørgen, (1995), “The Solent sonic-response and associated errors”,
Ninth Symposium on Meteorological Observations and Instrumentation, Charlotte, NC.

Site Evaluation

Site Evaluation
Appendix A Assessment Of Other Parameters - Mathematical

Details
A.1 Extreme Wind Speeds: Distribution
In practice, wind speeds observed in a time interval are dependent rather than independent quantities. This
hampers a proper statistical analysis of wind speed, with as a consequence a poor description of events
which are not very frequent like high wind speeds. The situation improves when statistically independent
observations of extreme wind speed are used in order to obtain a distribution of extreme wind speed.
If a quantity x has a cumulative distribution function F(x), the exact cumulative distribution function Fi(x)
of M independent observations of that quantity is FM(x). If M is sufficiently large, Fi(x) converges towards
an asymptotic cumulative distribution function Fa(x). On the other hand, for M = 1 the distribution of
independent observation is equal to the parent distribution F(x).
Now suppose a record with observations is broken down into M sub-records, and that the extreme is
selected from each sub-record. If the sub-records are large enough, the extremes xe obtained from the sub-
records are independent. In this case the cumulative distribution function of the independent observations
is equivalent to the cumulative distribution function of the independent extremes:
Fe(xe) ≡ Fi(xe) = FM(x). (A1)
In the case of wind speed, the general parent cumulative distribution function is the Weibull distribution
[A1]
  uk
FW ( u) = 1 − exp −   (A2)
  A  
where FW is the probability a wind speed is smaller than u given the scale parameter A and the shape
parameter k. As a consequence, the distribution of M independent extreme wind speeds ue is:
M
   u  k 
Fe ,w (u E ) = FWM (u) =  1 − exp −   (A3)
   A  
This implies the probability an independent extreme wind speed ue is larger than a given value u is
M
   u  k 
Ge,w ( ue ) ≡ 1 − Fe,w ( ue ) = 1 − 1 − exp −  
 (A4)
   A  
If M is large, this probability converges towards the asymptotic distribution

Ge,w ( ue ) ≈ 1 − exp( − exp( −α ( u − β ))) (A5)
with
1 A
= (ln M ) k andβ = A(ln M ) k
1 1
−1
(A6)
α k

Site Evaluation
the dispersion factor and the modal value, respectively [A2]. These equations show that the distribution of
independent extreme wind speeds can be obtained from the Weibull parameters A and k, and the number
M of independent extreme wind speeds. The cumulative distribution function
F(x) = exp( - exp(- α (x - β) ) ) (A7)
is the Gumbel distribution [A3], also known as the Fisher-Tippett Type 1 distribution [A4]. To facilitate
calculations (see section 4.1.2.1),
Ge,w (ue ) ≈ 1 − exp( − exp(−α ⋅ u + B)) with B ≡ α ⋅ β (A8)
is introduced here as an equivalent extreme winds distribution

A.2 Extreme Winds: Recurrence Period
If Fe(ue) is the probability the extreme wind speed ue during a standard observation period, Tobs of one
year, is not exceeded, the corresponding probability of exceedance is 1 - Fe(ue).
If the random variable T denotes the time between consecutive occurrences of independent exceeding
( )
events then the mean value of T : T ue , is by definition the return period.
The return period can be related to the probability of exceedance. If an independent exceedance occurs
every 50 years on the average, then the probability that such an event occurs in any given year is: 1/50 =
( )
0.02 or 2 percent. Thus the return period T ue and the probability of exceedance 1 - Fe(ue) are related
as:
T (ue ) 1
= (A9)
Tobs 1 − F (ue )
The IEC 1401-1 standard uses the term recurrence period. In the literature, as far as we know, there is
however no distinction made between return period and recurrence period. In any case the IEC 1400-1
standard does not describe the meaning of recurrence period, although the reference wind speed is defined
as the 10-minute average wind speed with a recurrence period of 50 years. It is therefore recommended
that the definition of recurrence period is included in list of symbols and units of the IEC standard.
The use of the term ‘return period’ can lead to dangerous misunderstanding if one is not aware of the fact
( )
that T ue is only the average value of the length of the time interval between consecutive exceedances.
( )
After the occurrence of an exceedance there is a big chance that before T ue has past, there is at least
one more exceedance. Rather than choosing a return period it is better to state which risk, r, one considers
to be acceptable, that ue is exceeded within Tp standard observation periods (Tobs) of one year. The
( )
corresponding return period T ue then follows from the relation:
TP / Tobs
 T 
1 − r =  1 − obs  (A10)
 T (ue ) 
A.3 Series Analysis

When measuring wind speed, it is usually sampled and stored in the form of a series. The samples are
obtained either in a given position during a time period or in various positions at a given time instant.
Hence, a series with the ith component of the wind speed is

Site Evaluation
ui [n] (i = 1,2,3; n = 1,2,... N ) (A15)
where N is the number of observations. In the case wind speed is measured at a given position during a
time period, n corresponds to time t [s]: n → nt = t. This is referred to as the temporal description of the
wind speed field. On the other hand, if wind speed is measured in various positions at a give time instant,
n corresponds to position x [m]: n → nx = x. This is referred to as the spatial description.
By definition, the covariance function of the series ui[n] and uj[n] is:
N
( )(
Rij [k ] = ∑ u j [n] − u j ui [n + k ] − u i
n =1
) (k = 1 − N .....,1 + N ) (A16)
with
N
1
ui =
N
∑ u [n]
i =1
i (A17)
the average of the ith component. If n corresponds to time, k corresponds to time shift τ [s]: k → kt = τ.
Likewise, if n corresponds to position, k corresponds to separation distance r [m]: k → kx = r. The
correlation function
Rij [k ]
rif [k ] = (A18)
Rii[0]R jj [0]
is the non-dimensional covariance function. The spectral density function is obtained by Fourier
transforming the covariance function:
1+ N
1
hij [ f ] = ∑ R [k ]exp(− jfk )
ij (A19)
2π k =1− N
If k corresponds to time shift, f corresponds to frequency ω [1/s]: f → ft = ω. Alternatively, if k

corresponds to separation distance, f corresponds to wave number w [1/m]: f → fx = w. The coherency
function
hij [ f ]
cij [ f ] = (A20)
hii [ f ]h jj [ f ]
is the non-dimensional spectral density function.
A4 References
[A1] W. Weibull, 1951, A statistical distribution function with a wide applicability, J. Appl. Mech., p.
293.
[A2] H. Bergström, 1992, Distribution of extreme wind speed, Wind Energy Report WE92:2, Department
of Meteorology, Uppsala University.
[A3] E.J. Gumbel, 1958, Statistics of extremes, Columbia University Press.
[A4] R.A. Fisher and L.H.C. Tippett, 1928, Limiting forms of the frequency distribution of the largest or
smallest member of a sample, Proc. Cambridge Philos. Soc., p. 180.
[A5] H.A. Panofsky and J.A. Dutton, 1984, Atmospheric turbulence, John Wiley & Sons.

Site Evaluation
[A6] J. Wieringa and P.J. Rijkoort, 1983, Windklimaat van Nederland, Staatsdrukkerij.
[A7] L.L. Freris, 1990, Wind energy conversion systems, Prentice Hall.
[A8] E. Castillo, 1988, Extreme value theory in engineering, Academic Press.

Site Evaluation
Appendix B Use Of Sonic Anemometer

B.1 Spectral characteristics of turbulence
Before proceeding with the analysis of data from a sonic anemometer, a brief presentation is given of the
power spectral density function and its relation to the turbulent kinetic energy of the wind speed
components. As a result, in the following we shall limit our discussion to the variations of the three
velocity components (u,v,w).
When considering the energy spectrum of the boundary layer turbulence, three distinct regions can be
identified [B1,B2]:
• The energy containing range where energy is produced by buoyancy and shear.
• The inertial sub-range where energy is neither produced or dissipated, but simply transported from
bigger to smaller size eddies. In the inertial sub-range the spectrum expression has the form
E x ( k ) = a1 ε k where x is one of the u, v or w, a1 is the Kolmogorov constant, ε the dissipation rate
2/3 −5 / 3
of the turbulent kinetic energy and k the wavenumber. Kolmogorov was the first to conceive the idea
of the inertial sub-range and to prove that the spectrum should have a slope of -5/3. According to IEC
1400-1 document “Safety of Wind Turbine Generator systems”, the inertial sub-range at 10m/s
extends roughly from 0.02Hz to 2 kHz.
• The dissipation range where kinetic energy is converted to internal energy.
energy containing inertial dissipation

range subrange range
Energy Spectral Density E(k)
-5/3
Wave number (k)
Figure B1 Spectral characteristics of the boundary layer turbulence

For the determination of the turbulence characteristics of an atmospheric flow, simultaneous
measurements at a number of spatial locations are required. As this is rather difficult and has a high cost,
use of the Taylor’s frozen turbulence hypothesis is made. According to Taylor, temporal and spatial
measurements are equivalent since turbulence eddies do not change substantially while being convected
by the mean wind speed. The equivalence between the temporal and spatial characteristics of the
turbulence parameters allows their study by using one instrument (e.g. sonic anemometer) at one location.
Assuming a stationary random process we can define an autocorrelation function:
R(τ ) = x '( t ) x ' ( t + τ ) (B1)

Site Evaluation
where τ is the time delay and x′ the variation of the corresponding velocity component (u,v,w).
The Fourier transform of this function is:
∞
∫ Φ(ω )e
iωτ
R(τ ) = 1
2 dω (B2)
−∞
where Φ(ω ) is the energy spectrum density (even function) and ω is the frequency expressed in
radians/second.
Now for τ = 0 , ⇒ R(0) is equal to twice the turbulent kinetic energy and equation (B2) becomes:
∞ ∞ ∞
σ x = R( 0) = ∫ Φ(ω )dω = ∫ S ( f )df = ∫ fS ( f )d ln f

2
(B3)
0 0 0
where S ( f ) = 2π ⋅ Φ(ω ) a new spectral density function, ω = 2π f is the frequency in radians per
second and f is the frequency in Hertz.
All expressions of equation (B3) are equivalent and their integral is seen to be the variance of the
corresponding turbulent velocity component. The function fS(f) is as a rule adopted for the description of
the atmospheric spectra. Since the variance is finite, the function becomes practically zero at very high or
very low frequencies.
B.2 Analysis stages for the calculation of the turbulent parameters
In the following the procedure for the calculation of the standard deviation, the power spectral densities
and the integral length scales is presented. This is done for the Kaimal spectral model. A corresponding
method can be used when the von Karman formulation is used. The power spectral density function can be
approximated by an analytical expression. In the IEC 1400-1 document “Safety of Wind Turbine
Generator systems”, the power spectral densities are, in non-dimensional form, given by the equation:
fS x ( f ) 4 fLx / Vhub
= , where: (B4)
σ x 2
(1 + 6 fLx / Vhub )5 / 3
f the frequency in Hz
x the corresponding velocity component
Sx the velocity component spectrum
σx the standard deviation of the x velocity component
Lx the integral length scale of the x velocity component
The following steps are taken to calculate the power spectral density and the corresponding length scales:
Mounting of the instrument
The sonic anemometer is mounted so that the sensor array is oriented towards the wind direction that will
be analysed (in this way minimum influence is experienced from the presence of the struts).
The vertical positioning of the instrument is checked with the help of a level at two orthogonal wind
directions.
Data acquisition
The length of the time series used for the calculation of the turbulent parameters depends entirely on the
application or on agreed procedures.
However, it is normal practice to use ten minutes time series for the calculation of the standard deviation.
For the calculation of the spectral density function and the length scales a “common practice” [B3], is to
divide a ten-minutes time series into eight non-overlapping segments, calculate the spectra for each

Site Evaluation
segment and average over the eight segments. Finally, to smooth out any large variations in the values of
the final spectrum, a moving average can been used. However, whether or not the above practice is
followed is down to individual judgement..
The choice of the record length T influences automatically the value of the lowest resolvable frequency
(1/T) and the width of the frequency band in the spectrum:
1 1
∆f = = (B5)
T N∆t
with N: the number of data points and ∆t: the time between two consecutive samples
After the FFT calculation, the number of points included in the spectrum is N/2 and the bandwidth of the
spectrum is (1/2∆t) Hz.
The choice of the sampling rate depends as a rule on the bandwidth wished to be covered. As a rule of
thumb the sampling frequency should be at least twice the highest frequency that needs to be investigated
(Nyquist criterion).
Finally attention is needed regarding the filtering of the data so that the results do not get contaminated
from the “aliasing” process. During this process frequencies higher than (1/2∆t) Hz are folded back on the
spectrum over the lower frequencies and contaminate the results. Anti-aliasing filter should therefore be
employed in the measurement system.
Data reduction
The time series chosen for the analysis should not contain large variations or trends in the magnitude of
the wind velocity or wind direction. The existence of trends in the data violates the stationarity assumption
and, as will be seen later, introduces uncertainties in the calculation of the length scales.
The output of the sonic anemometer is given as three velocities in an orthogonal co-ordinate system, fixed
to the anemometer. Assuming this is a right-hand system (which is not always the case), a two-step
transformation is applied (Figure B2), where the “fixed to the anemometer” co-ordinate system (x1,y1,z1)
is translated to the (x3,y3,z3) system. Thus the resultant wind speed u3 is aligned to the x3 axis and the
mean values of the two other components are equal to zero (v3=w3=0).

Site Evaluation
z1-axis
z3-axis
w1
w3
w2
x3-axis
u3
x1-axis
u1
ϕ v3
v2
θ
v3-axis
u2
v1
y1-axis
u3 cosθcosϕ sinθcosϕ sinϕ u1
v3 -sinθ cosθ 0 v1
=
w3 -cosθsinϕ -sinθsinϕ cosϕ w1
Figure B2 Transformation of the fixed-to-the-anemometer co-ordinate system

Calculate and subtract the mean value (either of the whole time series or of each segment) of the u3
velocity and find the corresponding standard deviations for all three velocity components. The difference
between subtracting the mean value and de-trending (by using a linear or higher order approximation) is
that it affects the standard deviation and variance of the resulting series. This is illustrated in Figure B3
where, for the time series (top), the standard deviation is 1.79m/s for the time series in the middle (mean
value removed) and 1.39m/s for the bottom one (linear de-trend) while both have zero mean values.
Figure B3 From top to bottom: a) original time series, b) mean value subtracted (std. dev=1.79m/s), c)
linear de-trending (std. dev.=1.39m/s)

Site Evaluation
The difference in the value of the standard deviation influences both the lower spectral frequencies and the
values of the length scales. This is also predicted from equation (B4), where a higher standard deviation is
connected to higher length scale values.
fS(f) (m/s)^2 1.0
0.1
Trend removal
Mean Removal
U=15.7m/s stdv=1.32m/s l=40m
U=15.7m/s stdv=1.74m/s l=89m
0.0
0.00 0.01 0.10 1.00 10.00
f (Hz)
Figure B4 Different ways of removing the mean velocity value affect both the lower spectral frequencies
and the magnitude of the length scale
Further evidence is presented in the example of Figure B4 where the spectra of the time series of a wind
velocity with a mean value of 15.7m/s are shown both after removing the mean value and by linear de-
trend. As a result of the different standard deviations, the length scales differ also. In Figure B4 the power
spectral density curves as derived from equation (B4), and corresponding to the two different standard
deviation values and length scales, are plotted.
10.0
9
8
7
Variance (calculated-experimental)(m^2/s^2)
6
5
4
1.0
9
8
7
6
5
4
2
1 10 100 1000 10000
Length scale (m)
Figure B5 Calculating the length scale from the least square fit

Site Evaluation
Use an FFT transformation to derive the power spectral density of the time series and a Rectangular
window for data filtering.. With the help of equation (B4) fit the linear part of the analytical spectrum to
the experimental data (linear part) using a least squares method. This “best-fit” procedure then gives the
magnitude of the length scale. The results of this procedure are seen in Figure B5 where the correct length
scale is given by the minimum of the curve.
The frequency interval where fitting takes place depends on the response characteristics of the sonic
anemometer but as a minimum, a band from 0.25Hz and covering one frequency decade is recommended.
It must however be emphasized that there are really no clear rules on this topic and care is needed since
the choice of the fitting interval interacts with the fitting procedure and this results in length scales of
different size. Thus it really depends on the person carrying out the data analysis to look at the data and
take a sound decision.
Experience has shown that the number of observations plotted against the logarithm of the length scale
(for a large number of observations) is normally distributed. This means that the individual length scale
calculated using the above described procedure is not necessarily representative and that a large number of
measurements are required before the “true” length scale is determined.
B3 References
[B1] Kaimal,J.C., Finnigan,J.J. (1994) “Atmosperic Boundary Layer Flows-Their Structure and
Measurement”, Oxford Univ. Press.
[B2] Sorbjan, Z., (1989), “Structure of the Atmospheric Boundary Layer”, Prentice Hall.
[B3] “MOUNTURB-Data Analysis Benchmark Test”, C.R.E.S. (1996).

Ewts Ii

Uploaded by

Copyright:

Available Formats

Ewts Ii

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ewts Ii

Uploaded by

Copyright:

Available Formats

European Wind Turbine Standards II

ECN Solar & Wind Energy

Realisation: BetaText, Bergen, The Netherlands

II European Wind Turbine Standards II

European Wind Turbine Standards II III

PARTICIPANTS OF THE EWTS-II PROJECT

CRES, Centre for Renewable Energy Sources (Greece)

IV European Wind Turbine Standards II

2. quantification of failure probabilities;

3. integration of blade tests in design;

4. power performance in complex terrain;

This report presents the results of these investigations.

Jos Dekker and Jan Pierik,

European Wind Turbine Standards II V

EXECUTIVE SUMMARY ……………………………………………………… IX

2. COMPLEX TERRAIN AND FATIGUE LOADING …………………………… Part 1 Sub B.1

3. EXTREME WIND CLIMATE EVENTS ……………………………………… Part 1 Sub C.1

4. QUANTIFICATION OF FAILURE PROBABILITIES ………………………… Part 2.1

5. INTEGRATION OF BLADE TEST IN DESIGN ……………………………… Part 3.1

6. POWER PERFORMANCE IN COMPLEX TERRAIN ………………………… Part 4.1

7. SITE EVALUATION …………………………………………………………… Part 5.1

European Wind Turbine Standards II VII

VIII European Wind Turbine Standards II

1. LOAD SPECTRA AND EXTREME WIND CONDITIONS

1.2 Wind farms-Wind field and turbine loading

European Wind Turbine Standards II IX

1.3 Complex terrain and fatigue loading

X European Wind Turbine Standards II

1.4 Extreme wind climate events

European Wind Turbine Standards II XI

XII European Wind Turbine Standards II

2. QUANTIFICATION OF FAILURE PROBABILITIES

2.1 Target values for structural reliability

European Wind Turbine Standards II XIII

2.2 Guidelines for data collection and parameter estimation

In addition to these reports, the following is recommended:

XIV European Wind Turbine Standards II

2.3 Risks to the public

2.4 Final recommendations

European Wind Turbine Standards II XV

3. INTEGRATION OF BLADE TEST IN DESIGN

3.2 Technical description

XVI European Wind Turbine Standards II

4. POWER PERFORMANCE IN COMPLEX TERRAIN

European Wind Turbine Standards II XVII

The major achievements of the project are:

4.2 Technical description

XVIII European Wind Turbine Standards II

Table 1.1. Overview of experimental and analytical research within EWTS-II.IV.

Assessment of power curve evaluation using nacelle anemometers.

European Wind Turbine Standards II XIX

An analytic method is introduced covering parameter identification tasks in power performance

The discussed critical points regarded the following issues:

XX European Wind Turbine Standards II

Drafting of a status report providing a comprehensive overview of the up to date available

European Wind Turbine Standards II XXI

The approach taken may be summarised as follows:

5.1 Survey of European Consultants

5.3 Mean Wind Speed Prediction

The wake turbulence intensity, Iwt

S D Hub CT Stability Amount Va Ia Mea. Est. Rel.

Sr = 20:8 + (S ; 2)0:8 for S 2: (2.4.1.7)

S D Hub CT Stability Amount Va Ia Mea. Est. Rel.

S D Hub CT Stability Amount Va Lua Mea. Est. Rel.