Sie sind auf Seite 1von 10

ASSESSMENT OF WIND TURBINE FOR SITE-SPECIFIC CONDITIONS

USING PROBABILISTIC METHODS


Enrique Gmez de las Heras1,*, Roberto Gutirrez1, Elena Azagra1, John Dalsgaard Srensen2
(1) Gamesa Innovation and Technology, Madrid, Spain
(2) Aalborg University, Aalborg, Denmark
*Corresponding author:
Address: Ramrez de Arellano, 37, 28043 Madrid, Spain
Phone: +34 91 5107330
E-mail: egdelasheras@gamesacorp.com

Abstract
This paper describes a new approach to assess the structural integrity of wind turbines for sitespecific conditions using probabilistic methods, taking into account the particular uncertainties
associated to each site. This new approach intends to improve the site suitability analysis of
wind turbines, helping to the decision making during the site assessment phase of wind farm
designs.
First, the design equation for the failure mode of interest is defined, where the loads associated
to the site-specific wind conditions are compared with the design limits of the structural
component. A limit state equation is defined making the loads and resistance depending on a
set of stochastic variables representing the uncertainties. In this paper, special focus is put on
the uncertainties related to the assessment of wind data, which is the main input for the sitespecific load assessment, and can be very dependent on the site. The uncertainties on the wind
properties depend on issues like the available wind data, the quality of the measurement
sensors, the type of terrain or the accuracy of the engineering models for horizontal and vertical
spatial extrapolation.
An example is included showing two fictitious sites with different uncertainty scenarios, which
would result in a different verdict depending whether a traditional deterministic approach or a
probabilistic approach is used to assess the structural integrity of the wind turbine.

1. INTRODUCTION AND BACKGROUND


Every time a wind turbine is intended to be installed at a site it has to be verified that the sitespecific conditions do not compromise the structural integrity of the turbine. The standard
IEC6100-1 [1] states that this assessment may be done by demonstrating that all the wind
conditions are no more severe than those assumed for the design or by comparing the loads for
the specific site conditions with those calculated during the design. Both options represent a
Deterministic approach since they are based on the comparison of characteristic site-specific
values (based on wind variables) with respect to the characteristic design values, usually
defined based on standard Classes. However, the evaluation of the wind conditions is subjected
to many sources of uncertainties, which can be different for each site. A deterministic approach
can not deal with the effect of those uncertainties, even though it is quite intuitive that larger the
uncertainties the larger the risk that the site-suitability assessment is not trustworthy.
Happily, probabilistic methods can be used for the evaluation of structural reliability considering
the eventual uncertainties in the variables of interest. This paper presents a proposal on how to
apply those methods to the site-suitability assessment of wind turbines depending on the
uncertainties in the characterization of wind conditions at each site. The final goal is comparing
the reliability level of the site of interest (with its specific uncertainties) with respect to a
reference site with some reference uncertainties. An example is presented where the outcome
from a traditional deterministic approach is compared with the results from a probabilistic
approach.

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

The technical background for probabilistic techniques and structural reliability methods is widely
described in many references (as [2] and [3]), and it is not the purpose of this paper to describe
them in detail. References [4], [5], [6], [7], [8] and [9] provide some examples of application of
these methods to wind turbines. Just for a general overview, the general workflow during a
reliability analysis is sketched in Figure 1.
Identify
failure
modes

Define
Design
equation

Define
loads
model

Fatigue
Ultimate
Buckling

Define Limit
State
Equation

Stress vs loads
Analytical expressions
for resistance and loads

(depending on
variables and
uncertainties)

Define
stochastic
variables
(uncertainties)
Variables
described as
statistical
distributions

Reliability
calculations

Numerical
techniques
(FORM,
SORM,
Simulation)

Results

Probability of
failure (Pf)
Reliability
Sensitivities

Figure 1 Workflow of a reliability analysis


First, once the failure mode of interest has been identified, a limit state equation is defined as an
analytical expression linking the loads S and the resistance R. A failure occurs whenever the
loads S exceed the resistance R of the structure, so the probability of failure becomes
Pf = P(R S 0)
The Limit State Equation is denoted Z(X), where the loads are defined as a function of some
stochastic variables XS and the resistance as a function of some stochastic variables XR

Z ( X ) = R( X R ) S(X S )

(1)

Each Xi represents a different source of uncertainty of the loads or strength and is defined by a
statistical distribution. Then the probability of failure can be obtained by solving the reliability
problem by means of several techniques, as FORM, SORM or simulation techniques (see [3]):

Pf = ( ) = Pr {R(X R ) S(X S )}

(2)

where is the reliability index and ( ) is the standard Normal distribution function. Assuming
that the turbine itself does not change from site to site (except the foundation or some
components provided by different suppliers) the only stochastic variables that change from site
to site are actually the ones associated to the wind conditions. Those wind conditions are
usually estimated during the site assessment process prior to the construction of the wind farm,
but that process is subjected to many sources of uncertainties, like the quality and quantity of
wind data, the type of terrain, the models used, etc. So, the stochastic variables associated to
the wind conditions can be quite different from site to site, leading to a different reliability level.
This paper addresses this issue, where the main goal is comparing the reliability level of the site
of interest (with its specific uncertainties) with respect to a reference site with some reference
uncertainties.
Finally, for some load events (loadcases) the actual probability of failure will not depend only on
the uncertainties on the wind conditions but also on the probability of some external events, like
electrical network losses or turbine malfunctions. The likelihood of those events can modify the
actual probability of structural failure and should also be taken into account.

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

2. DESIGN EQUATION
The first step to perform a structural reliability analysis is defining the design equation that links
the external design loads with the design resistance of the structural component. The design
parameters (thickness, geometries, etc) are adjusted to make sure that the design loads does
not exceed the design strength. The design loads and design strength are obtained by applying
some safety factors to the characteristic values of loads and strengths, obtained by standard
engineering procedures.
A different design equation must be defined for each failure mode of interest, for instance,
fatigue failure, ultimate failure, buckling failure, etc. Several examples of design equations for
wind turbine components can be found in [5], [6] and [9].
A general example of a design equation is:
char
srf c cinf s char f = 0
m

(3)

where
char is the characteristic strength of the material. In case of a lifetime fatigue failure mode, that
would be the maximum allowable lifetime equivalent fatigue stress (range) of the material. In
case of an ultimate failure mode, char would represent the characteristic ultimate strength.

schar is the characteristic load. For instance, the external bending moment applied to the
structure. In case of a fatigue failure mode, the schar would represent the characteristic lifetime
fatigue equivalent design load (range). In case of an ultimate failure mode, schar would represent
the characteristic ultimate design load.
cinf is the influence coefficient, which is used to convert from loads to stresses, assuming a
linear relationship. cinf depends on the geometry and stiffness of the structure.

m and f are the safety factors associated to the material strength and loads, respectively.
srf is the Stress Reserve Factor. If the component has not been designed to the limit, the srf will
be larger than 1, reflecting the extra safety margin. This can happen, for instance, if the design
driver is the fatigue load but the design equation is dealing with an ultimate failure mode.
For a site-specific wind turbine suitability assessment, the characteristic load is a function of the
site-specific wind conditions (schar,site). Then, for a given design (char and cinf), it must be verified
that the resulting site specific stress reserve factor (srfsite) is equal or larger than 1.
3. LOAD MODEL
The load term in equation (3) is a function of external conditions (e.g. wind conditions) and the
wind turbine design (aerodynamics, structure and control and safety system). So, for a given
turbine design, the load model is a function that provides the loads as a function of the wind
variables (wind speed distribution, turbulence, wind shear, etc). For a site-specific assessment it
is necessary that the load model effectively capture the dependency of the loads with regard to
the wind characteristics at the site. The resulting load model will be different whether it is
representing a fatigue load or an ultimate load. In the latter case, it can even be different
depending on the specific loadcase of interest. Examples of load models for fatigue or ultimate
analysis can be found in [5], [6] and [7].
In the present work, a response surface (two degrees polynomial) is used to express the loads
as a function of wind variables. The coefficients of the polynomial are obtained by running
several thousands of aeroelastic simulations of the wind turbine at different combinations of
external wind conditions, and fitting the response surface to the resulting loads with the software
TM
Statistica [10]. Different loads models have been defined for different load scenarios:

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

Fatigue loads
A function providing the 20 years fatigue damage equivalent load, including normal
production, idling and transient events. The loads are defined as a function of air density
(), wind shear exponent (), characteristic turbulence intensity at 15m/s (ti15), annual
average wind speed (vave) and Weibull shape factor (k). It is assumed that the long-term
wind speed is defined by a Weibull distribution.
(4)

sfatigue (xwind ) = ffatigue ( , ,ti15 ,v ave ,k)




Ultimate loads
Since the ultimate loads depend on the loadcase of interest, different response surfaces
have been defined for different representative loadcases from IEC standard [1]:
- Power production. The characteristic extreme load value is calculated as the
mean value of the maximum loads over several 10-minute turbulent realizations
at the most critical wind speeds, multiplied by an extrapolation factor to obtain
the ultimate load with a recurrence period of 50 years.
- Power production plus occurrence of fault. The maximum load for
representative turbine faults is obtained at the most critical wind speeds.
- Parked in extreme winds. The characteristic load value is calculated as the
mean value of the maximum loads over several turbulent realizations (seeds) in
idling conditions, for several wind directions.
The response function for the ultimate loads is then defined for each of the
aforementioned loadcases as a function of air density (), wind shear exponent (),
upflow angle () and characteristic turbulence intensity at 15m/s (ti15). For the parked
loadcase the loads model also includes the extreme wind speed (vextreme). So, for each
ultimate load case LCj, a loads response surface is defined as:

sUlt LCj (xwind ) = fUlt LCj ( , , ,ti15 ,v extreme )

(5)

4. PROBABILISTIC MODELS AND LIMIT STATE EQUATION


Once the limit state equation and the loads model have been defined, a probabilistic model is
created to include the uncertainties in the resistance and in the loads. In the probabilistic model
all identified sources of uncertainties are defined as stochastic variables, expressed by
probability distributions. The uncertainties can be physical or epistemic (measurement,
statistical, model), see [3].
5.1 Probabilistic loads model
The probabilistic loads model has to reflect the uncertainties in the site-specific wind
characteristics and other more general uncertainties.
For the case of fatigue loads, a probabilistic loads model can be defined as:

Sfatigue = ffatigue ( , , ti15 ,v ave , k) X lfc X rose X seeds X NTM X WindModel X aero X dyn
144
42444
3 1444444442444444443
Site specific

General uncertaint ies

stochastic variables
1444
424444
3

(6)

Response surface for fatigue loads

In a similar way, a probabilistic model for the ultimate loads of each loadcase LCj is defined as:

SUlt LCj = K Tyear extrapol fUlt LCj ( , , , ti15 ,v extreme ) X seeds X azi X aero X dyn
144424443 1444
144244
3
424444
3
Extrapolat ion factor

Site specific

variables
1444stochastic
44244
444
3

General uncertaint ies

(7)

Response surface for extreme loads

For a site-suitability assessment the uncertainties in the characterization of main wind


properties should be considered (, , , ti15, vave, k, vextreme). For the fatigue loads, those

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

variables are related to the annual mean values, whereas for the ultimate loads, those variables
are associated to 10-min period. More details about how to define these uncertainties are given
in next section.
Additional stochastic variables Xi are defined as multiplicative factors to take into account other
more general sources of uncertainties in the loads: Xlfc models the low frequency variability of
the wind speed (physical, model); Xrose models the variability of wind direction during the year
(physical, model); Xseeds models the limited amount of simulations used to obtain the loads
(statistical); XNTM models the assumption that the NTM turbulence model ([1]) is representative
of the whole turbulence distribution (model); Xwind model models the specific turbulence model
used in the simulations (model); Xazi models the variability of some ultimate loads with respect to
the rotor azimuth; Xaero models the aerodynamic models used in the aeroelastic simulations
(model); Xdyn models the structural dynamics models used in aeroelastic simulations (model).
The stochastic variables Xi are usually defined as Normal or Log-Normal distributions, with
mean value equal to 1 (except if there is some bias effect) and a coefficient of variation usually
between 5 and 20%. Some of these general uncertainties can also be dependant on the
loadcase. For instance the uncertainty associated to the aerodynamic model would be larger for
the loadcases with large yaw misalignment or airfoils in stall conditions. Some information about
the definition of the uncertainties in the loads can be found in [5], [6], [7] and [8].
5.2 Limit State Equation
From the design equation, a Limit State Equation is defined for each failure mode of interest.
For instance, for the failure mode representing the lifetime fatigue failure, the limit state equation
can be defined as:
1

T m
(8)
g fatigue = X X SIZE X LS Cinf Sfatigue XCAL X B
TL
where m is the Whoeler exponent for the fatigue analysis and Sfatigue represents the probabilistic
model for the fatigue loads (usually for 20years lifetime), as defined in equation (6). The former
expression would be used to calculate the fatigue structural reliability in the whole lifetime (TL,
usually 20 years). The analysis could also be used to estimate the annual probability of failure
for a reduced period of time T.
For an ultimate failure mode:

gUlt LCj = X X SIZE Cinf SUlt LCj XCAL

(9)

where SUlt LCjj represents the probabilistic model for the ultimate loads for the loadcase LCj, as
defined in equation (7).
The additional stochastic variables defined in the Limit State equation are: XCAL models the
conversion method from loads to stresses; XB models the assumption of fatigue SN linear curve;
X models the uncertainty in material strength limit (fatigue or ultimate); XSIZE models the
influence of size effects on material strength; XLS models the load sequence effects in fatigue;
Cinf models the uncertainty in the stiffness or geometry. Information about the definition of some
of these uncertainties can be found in [5], [6], [7] and [8].

5. DEFINITION OF SITE-SPECIFIC STOCHASTIC VARIABLES


As explained above, the probabilistic load models from expressions (6) and (7) are dependant
on stochastic wind variables associated to the site-specific wind conditions. The assessment of
the wind conditions at each site is subjected to many sources of uncertainties, which can differ
from site to site, so the stochastic wind variables are also site-dependant.
The main uncertainty sources related to the assessment of wind conditions at a specific site are
described in Table 1:

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

Table 1 Uncertainty sources related to the assessment of wind conditions


Uncertainty source

Type of
uncertainty

Factors affecting the uncertainty

Uncertainty in the wind measurements:


Anemometers

Measurement

Tower Effects
Mounting Effects

Measurement
Measurement

Type of anemometer (Sonic, 1st Class Cup)


Calibration (Calibrated 1st, Not Calibrated)
Position of the anemometers at met mast
Positioning of the boom

Uncertainty in the estimation of long-term wind resource:


Available wind data at the site
Statistical
Long-term estimation
Available wind data at the reference mast
and model
Long-term extrapolation models (MCP models)
Statistical
Available wind data
Extreme winds
and model
Extreme wind models
Uncertainty due to time variations of wind properties:
Statistical
Data availability
Inter-Annual variations
and physical
Physical variations
Statistical
Data availability
Seasonal variations
and physical
Physical variations
Uncertainty in the models for spatial extrapolation:
Fluid dynamic models
Topographic Effects
Quality of map
Model
(horizontal extrapolation)
Type of terrain and Roughness
Distance from turbine to mast
Fluid dynamic models
Wind Shear Model
Type of terrain
Model
(vertical extrapolation)
Roughness
Heights of measurements at the met
Each of the factors on the third column of Table 1 induces some uncertainty in the estimated
wind properties. These uncertainties are expressed as a coefficient of variation (COV) on the
wind variables of interest. The estimation of the COV can be done in several ways, depending
on the type of uncertainty:
 Physical uncertainties: By measuring a big enough data set to quantify the physical
variations.
 Epistemic uncertainties:
- Statistical: From a large data set, extract many reduced sets and evaluate the
variability of the results depending on the size of the reduced data set.
- Measurement: Through calibration of the measurement sensors.
- Model: Comparing the results from the applied models with respect to more
advanced models or real data.
Table 2 provides the definition of the main stochastic wind variables.
Table 2 Site-specific stochastic variables
Wind variable

Name

Air density
Wind shear exponent
Turbulence intensity
Weibull shape factor
Annual mean wind speed
Annual extreme wind speed

Charcteristic
value
1.225 kg/m3

ti15
k
vave

0.2
18 %
2
8.5 m/s

vextreme

34 m/s

th

Char.
Distribution
COV
quantile
0.5
LN
0.5
N
0.9
LN
(see Table 3)
0.5
LN
0.5
LN
0.5
Gumbel

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

The COV from several sources of uncertainties from Table 1 could be combined by using the
Law of propagation of uncertainty (see [11])
N

COV =

COV

(10)

i =1

Alternatively, a multi-level uncertainty model can be defined, where the parameters of some
statistical distributions are defined as well as stochastic variables. In this case, the stochastic
model for the wind variables has been defined as a three-level model.
 Level 1: Seasonal variations. A distribution representing how the 10-min mean value
varies along the different days within one year.
 Level 2: Inter-annual variations. A distribution representing how the annual mean value
changes from year to year.
 Level 3: Epistemic uncertainties. Representing the uncertainties due to the engineering
models, the availability of data and the quality of the measurement sensors.
This means that the average value of the Level 1 distribution (the annual mean value), is
defined by another distribution (Level 2), representing how the annual mean value varies from
year to year. For the fatigue assessment, only Level 2 and 3 have been considered, since it is
assumed that the fatigue loads only depend on annual mean values, and not on 10-min values.
Table 3 provides the estimated COVs for the wind variables for three site-specific uncertainty
scenarios (a Best case, an Intermediate case and a Worst case), assuming a hub at 100m.
Table 3 Definition of site-specific uncertainty scenarios
Uncertainty source
Best case
Intermediate case
Worst case
1st Class Sonic
Anemos on mast top
140 and 40m
3 years
MCP with 50 years
Near
Curve lines every 10m
Flat and low roughness

Anemometers
Met mast mounting
Measurements heights
Available wind data
Long-term correction
Distance to Met mast
Quality of map
Terrain complexity

Air density
Wind shear exponent
Turbulence intensity
Weibull shape factor
Annual mean wind speed
Extreme wind speed

Level
1
0.05
0.35
0.12
-

Level
2
0
0.02
0.02
0.01

0.01
0.10

Calibrated Cup
No mast distortion
90m and 60m
1year
MCP with 5 years
Medium
High resol. satellite map
Complex

Not calibrated Cup


Mast distortion
40m and 20m
10months
No MCP
Far
Curves further than 20m
Very complex

Coefficient of variation (COV)


Level Level Level Level Level
3
1
2
3
1
0.01
0.05
0.01
0.01
0.05
0.05
0.35
0.10
0.18
0.35
0.02
0.12
0.02
0.05
0.12
0.04
0.01
0.17
0.04
0.10

0.01
0.10

0.17
0.15

Level
2
0.01
0.15
0.02
0.01

Level
3
0.02
0.23
0.10
0.30

0.01
0.10

0.30
0.20

6. RELIABILITY RESULTS
The assessment of the structural integrity is done by solving the reliability problems defined by
Limit State Equations previously defined. The reliability problem can be solved by applying
different techniques (see [3]), like the First Order Reliability Method (FORM), the Second Order
Reliability Method (SORM) or Simulation Techniques (Crude Monte Carlo, Importance
Sampling,). The main outcome from this analysis is the probability of failure in the reference
time for each of the limit states considered. Alternatively, the Reliability Index is also used.
Equation (8) is used to calculate the reliability of the fatigue limit state in the time of interest T,
and equation (9) for the ultimate limit state for each loadcase LCj.

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

Figure 2 shows a comparison of calculated and empirical probabilities of failures, together with
the target reliability defined in some Standards and Guidelines ([11], [13] and [12]). The
empirical data has been obtained form statistics of turbines at field ([14] and [15]). The
calculations have been performed for the shaft and tower of a representative 2MW pitch
regulated turbine, using SORM. The results for the fatigue limit state and for the most critical
ultimate limit state are shown for the wind conditions of Table 2 and a Intermediate Uncertainty
Case of Table 3. The program Strurel [16] has been used for the reliability calculations.
Blades

Nacelle
(shaft, hub,)

Shaft

Gearbox

Tower

Failures per turbine per year

1.0E+00
1.0E-01
Target
reliability

1.0E-02
1.0E-03

ISO2394 [11]
JCSS [12]

1.0E-04

EN1990 [13]

1.0E-05
LWK [14]
Windstats Danmark [14]
Calculated (fatigue failure in 20 years)

Windstats Deutchland [14]


Handboek Risicozonering Windturbines [15]
Calculated (ultimate, worst loadcase )

Figure 2 Experimental and calculated probabilities of failure


Figure 2 has to be considered just as a general overview of the probabilities of failure since
failure data from field comes from a wide variety of sites and wind conditions, and for less than
20 years. It has to be noted that the calculated probabilities of failure in fatigue are for 20 years
period so they cannot be compared directly with the others (annual probabilities).
It is also important to remark that real failures can be provoked by human mistakes,
manufacturing flaws or installation faults, which can not be predicted by the reliability methods.
So, it can be concluded that probabilistic techniques should be used only to compare different
reliability scenarios, but any connection with real data should be done with caution.
Combination of results from different loadcases
Assuming independent failure modes, the total probability of failure of a component could be
obtained by adding the probability of failure of all failure modes:
P(componen t failure) = P(fatigue failure) + P(ultimate failure) + P(buckling ) + ...
(11)
For ultimate limit states the contribution from all possible loadcases LCj should be considered:
P(ultimate failure ) = P(failure LC1) + P(failure LC2) + ... =
P(failure LCj) (12)

where

P(failuredlc LCj) is the probability of having failure AND load case LCj, so:
P(failure LCj) = P(failure | LCj) P(LCj)
(13)
P(failure | LCj) is the probability of failure given the loadcase LCj. This is actually the
outcome of the reliability analysis done for equation (9) for the loadcase LCj.
P(LCj) is the annual probability of occurrence of the loadcase LCj. This is related to the
probability of the wind conditions, turbine states or other events associated to the
loadcase.

This approach would give an upper bound of the total probability since there will always be
some correlation between the failure modes. Ideally a Series System model should be used.

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

7. EXAMPLE OF APPLICATION FOR SITE SUITABILITY ASSESSMENT


An example of the application of the described methods (Probabilistic approach) to the
assessment of wind turbines for specific sites is presented in the following.
Lets assume that it is necessary to asses the suitability of a turbine for two different fictitious
sites, Site 1 and Site 2, assuming that the turbine is designed to the limit with respect to some
reference conditions, see Table 4.
Table 4 Example of site specific conditions for turbine assessment
Wind variable

Site 1

Reference

Air density
Wind shear exponent
Turbulence intensity I15
Average wind speed

1.217 kg/m
0.2
17%
8.5 m/s

50 year extreme wind speed

39.1 m/s

42.5 m/s

45.0 m/s

Uncertainty scenario
(see Table 3)

Worst case

Intermediate
case

Best case

1.225 kg/m
0.2
18%
8.5 m/s

Site 2
1.236 kg/m
0.2
18.3%
8.5 m/s

Traditional
(deterministic)
approach

Probabilistic
approach

For the traditional approach described in the IEC standard [1] only the characteristic values of
wind conditions are used, while for the Probabilistic Approach the Uncertainty Scenario is also
needed. Site 1 has less severe characteristic wind conditions, but larger uncertainties, while
Site 2 has more severe characteristic wind conditions, but smaller uncertainties.
With the traditional approach, the turbine would be suitable for Site 1, but it wouldnt be suitable
for Site 2, since the (characteristic) wind conditions for Site 2 are more severe than those
assumed as reference. Table 5 shows the relative differences in the characteristic load for two
representative structural components, as it would proceed according to the traditional approach.
Table 5 Example of suitability assessment with traditional approach
Site 1
Reference
Site 2
Main Shaft fatigue bending loads
-3%
0
+2%
Tower bottom extreme loads
-15%
0
+13%
Wind Turbine Suitability

WT is suitable

WT is suitable

WT is NOT suitable

However, if the turbine suitability assessment is done based on a probabilistic approach the
indicator of interest is not the characteristic loads, but the reliability index , which depends on
the uncertainty scenario. Larger means less probability of failure, so larger reliability. Table 6
shows the resulting values for the same two components and scenarios as above.
Table 6 Example of suitability assessment with probabilistic approach
Site 1
Reference
Site 2
- Shaft fatigue failure
2.349
2.350
2.351
- Tower ultimate failure
2.697
3.457
3.480
Wind Turbine Suitability

WT is NOT suitable

WT is suitable

WT is suitable

Using a probabilistic approach the turbine would be suitable for the Site 2, but it wouldnt be
suitable for the Site 1. This is actually a different conclusion than the traditional approach.
This example shows a case where one site is OK and the other is not OK for all the wind
parameters by a deterministic approach. But another case could be one with a mixture: some
parameters are OK and others not. Here a probabilistic approach could also help in considering
all parameters and uncertainties in a rational way. So the better the wind data quality the more
chances of a turbine to be suitable for a site, given a target reliability level. Of course, it can also
happen the other way around.

th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

Another interesting outcome from the reliability analysis is the sensitivity of the reliability index
to the different parameters of the stochastic variables. For fatigue failure modes, the
uncertainties associated to the wind variables are usually not so important (as also concluded in
[5]) but the uncertainties associated to the damage models and fatigue strength (Miner rule, SN
curves) are the more relevant. For ultimate failure modes, the uncertainties associated to the
wind variables can be very relevant, especially in the loadcases dealing with extreme winds.
8. CONCLUSIONS
In the present paper is described a new approach to assess the structural integrity of wind
turbines for site-specific conditions using probabilistic methods, as an alternative to the
traditional approach described in the wind turbine standard IEC61400-1. The new approach is
based on the structural reliability methods, where the uncertainties in the different variables
associated to the load and structural strength can be taken into account. These methods can be
very useful if used in a relative basis to compare the reliability level of different scenarios.
For the particular problem of the site-suitability analysis of wind turbines the uncertainties in the
wind conditions are addressed and the coefficient of variation for the main wind variables is
presented for different uncertainty scenarios. Some typical limit state equations are shown and
a model for fatigue and extreme loads is presented, focusing on the dependency of the loads
with respect to relevant wind variables. A proposal on how to combine the probability of failure
from different loadcases is also presented.
With the new approach, the Reliability Index is taken as indicator of the wind turbine suitability.
For the site of interest, the site-specific is compared with a target value, which is defined
based on some reference wind conditions and uncertainties scenario.
A numerical example is included where the suitability of a turbine is assessed for two fictitious
sites based on the traditional approach and the new probabilistic approach. The results show
that the probabilistic approach can give a more comprehensive verdict, improving the quality of
the site assessment process of wind turbines.
REFERENCES
[1] IEC 61400-1. Wind turbines Part 1: design requirements. 3rd ed. IEC.2005.
[2] Thoft-Christensen P, Baker MJ. Structural Reliability Theory and its applications. SpringerVerlag, Berlin, 1982.
[3] Srensen, JD. Notes in Structural Reliability Theory and Risk Analysis. Aalborg Univ. 2004.
[4] H.Braam et al. Probabilistic Design Tool, Prodeto. ECN. 1999, Report No.: ECN-C--99-023.
[5] Veldkamp D, Chances in wind energy - A probabilistic approach to wind turbine fatigue
design (Ph.D thesis). Delft University of Technology; 2006.
[6] Tarp-Johansen NJ, Madsen PH, Frandsen S. Partial safety factors for extreme load effects.
Roskilde, Denmark: Ris National Laboratory; 2002. Report No.: Ris-R-1319(EN).
[7] Srensen JD, et al. Effective turbulence models and fatigue reliability in wind farms.
Probabilistic Engineering Mechanics (2008), doi:10.1016/j.probengmech.2008.01.009.
[8] Stensgaard Toft H, Srensen JD. Reliability-based design of wind turbine blades. Structural
Safety (2011), doi:10.1016/j.strusafe.2011.05.003.
[9] Tarp-Johansen NJ, Srensen JD, Madsen PH. Experience with Acceptance Criteria for
Offshore Wind Turbines in Extreme Loading. Proceedings of JCSS Workshop on Reliability
Based Code Calibration, Zurich, 2002.
[10] StatisticaTM Users Manual. Statsoft. 2011
[11] ISO 2394. General principles on reliability of structures. ISO. 1998.
[12] Probabilistic Model Code. Joint Committee on Structural Safety (JCSS). 2001.
[13] EN 1990. Eurocode Basis of structural design; 2002.
[14] Tavner P, Spinato F, van Bussel GJW, Koutoulakos E. Reliability of Different Wind Turbine
Concepts with Relevance to Offshore Application, EWEA Conference 2008; Brussels. 2008
[15] Rademakers L. et al . Handboek Risicozonering Windturbines. 2nd ed. NOVEM. 2005.
[16] Strurel: Comrel and Sysrel Users Manual. RCP Consult. 2008
The present document, its content, its annexes and/or amendments (the Document) has been drawn up by GAMESA
CORPORACIN TECNOLGICA, S.A. (Gamesa) for information purposes only, and contains private and confidential
information regarding Gamesa and its subsidiaries (the Company), directed exclusively to its addressee. Therefore it
must not be disclosed, published or distributed, partially or totally, without the prior written consent of Gamesa, and in
any case expressly indicating the fact that Gamesa is the owner of all the intellectual property. All the content of the
Document, whether it is texts, images, brands, trademarks, combination of colours or any other element, its structure
and design, the selection and way of presenting the information, are protected by intellectual and industrial property
rights owned by Gamesa, that the addressee of the Document must respect. In particular (notwithstanding the general
confidentiality obligation), the addressee shall not reproduce (except for private use), copy, transform, distribute or
publish to any other third party, any of the information, totally or partially.
th

EWEA 2013 Event 4-7 February 2013 Vienna (Austria)

10

Das könnte Ihnen auch gefallen