See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2363835

Parametric Models For Estimating Wind Turbine Fatigue Loads For Design [4]

Article in Journal of Solar Energy Engineering · November 2001

DOI: 10.1115/1.1409555 · Source: CiteSeer

CITATIONS READS
42 535

3 authors:

Lance Manuel Paul S. Veers

University of Texas at Austin National Renewable Energy Laboratory
240 PUBLICATIONS 3,344 CITATIONS 106 PUBLICATIONS 2,919 CITATIONS

SEE PROFILE SEE PROFILE

Steven R. Winterstein
Probability-Based Engineering
111 PUBLICATIONS 3,194 CITATIONS

SEE PROFILE

All content following this page was uploaded by Lance Manuel on 17 February 2013.

The user has requested enhancement of the downloaded file.

 AIAA-2001-0047

PARAMETRIC MODELS FOR ESTIMATING WIND TURBINE

FATIGUE LOADS FOR DESIGN
Lance Manuel1
Paul S. Veers2
Steven R. Winterstein3
1
Department of Civil Engineering, University of Texas at Austin, Austin, TX 78712
2
Sandia National Laboratories, Wind Energy Technology Department, Albuquerque, NM 87185-0708
3
Department of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305-4020

ABSTRACT INTRODUCTION

International standards for wind turbine certification Design constraints for wind turbine structures fall into
depend on finding long-term fatigue load distributions either extreme load or fatigue categories. In the case of
that are conservative with respect to the state of extreme load design drivers, the load estimation
knowledge for a given system. Statistical models of problem is limited to finding a single maximum load
loads for fatigue application are described and level against which to assess the structural strength.
demonstrated using flap and edge blade-bending data For design against fatigue, however, loads must be
from a commercial turbine in complex terrain. defined over all input conditions and then summed over
Distributions of rainflow-counted range data for each the distribution of input conditions weighted by the
ten-minute segment are characterized by parameters relative frequency of occurrence. While this might
related to their first three statistical moments (mean, seem to be a more daunting task, it is in many ways
coefficient of variation, and skewness). Quadratic quite similar to the extreme load problem, as can be
Weibull distribution functions based on these three seen by comparing with Fitzwater and Winterstein1. In
moments are shown to match the measured load both cases, the loads must be determined as functions of
distributions if the non-damaging low-amplitude ranges wind speed (or other climatic conditions).
are first eliminated. The moments are mapped to the
wind conditions with a two-dimensional regression over Parametric models define the response, statistically,
ten-minute average wind speed and turbulence with respect to input conditions. Such models fit
intensity. With this mapping, the short-term analytical distribution functions to the measured or
distribution of ranges is known for any combination of simulated data. The parameters of these distribution
average wind speed and turbulence intensity. The long- functions can be useful in defining the response/loads
term distribution of ranges is determined by integrating as a function of the input conditions. The end result,
over the annual distribution of input conditions. First, then, is a full statistical definition of the loads over all
we study long-term loads derived by integration over input conditions.
wind speed distribution alone, using standard-specified
turbulence levels. Next, we perform this integration In the most prevalent alternative to parametric
over both wind speed and turbulence distribution for the modeling, an empirical distribution of loads (i.e., a
example site. Results are compared between standard- histogram describing frequency of occurrence of the
driven and site-driven load estimates. Finally, using modeled response quantity) is used to define the turbine
statistics based on the regression of the statistical response at the conditions of the measurement or
moments over the input conditions, the uncertainty (due simulation. When using simulations, a ten-minute time
to the limited data set) in the long-term load distribution series is generated at specified environmental
is represented by 95% confidence bounds on predicted conditions using an aeroelastic analysis code. The time
loads. series is rainflow-counted and the number of ranges in
specified intervals is summarized in histograms. The
histograms serve as empirical distributions that are
taken to be representative of the response of the turbine
at those particular conditions. The full lifetime
distribution is then obtained by summing the
This paper is declared a work of the U.S. Government and is not distributions after weighting by the frequency of
subject to copyright protection in the United States. Sandia is a occurrence of the wind speed associated with each
multiprogram laboratory operated by Sandia Corporation, a Lockheed simulated data segment included in a histogram
Martin company, for the U.S. Department of Energy under contract
DE-AC04-94AL85000.
interval. In the case of measured data, a similar

1
approach has been described by McCoy et al.2 but with limited duration of the sample, whether based on
an innovative weighting scheme to account for the simulation or field measurements. Only then can
irregular input conditions of field measurements. design margins be trimmed to the point of least cost
while still maintaining sufficient margins to keep
The empirical approach uses only the measured or reliability levels high.
simulated data without any fitting of distributions or
extrapolation to higher values that would be seen if The approach to load modeling is not uniform across
more data were obtained. One of the disadvantages of the wind community by any measure. This lack of
using a purely empirical approach is, therefore, that the commonality in approach was reflected in the working
loading distribution may not be representative. Perhaps group that produced IEC’s Mechanical Load
a subtler shortcoming is that the uncertainty in the loads Measurement Technical Specification6. No consensus
is almost impossible to characterize. could be obtained on how to use measured loads to
either create or substantiate a fatigue load spectrum at
With regard to uncertainties in loads and how they the conditions specified in the Safety Standard5. All
might be dealt with in design, one might expect that that is offered are several examples of differing
these uncertainties could be covered by the use of approaches in an annex of the specifications6.
standard specifications of characteristic loads (derived
from a specified high turbulence level) and safety Here, we present a methodology for using measured or
factors. However, current standard load definitions use simulated loads to produce a long-term fatigue-load
safety factors that do not depend on the relative spectrum at specified environmental conditions and at
uncertainty in the load estimates. Either the margins desired confidence levels. To illustrate, example
are larger than they need to be when load estimates are measurements of the two blade-root moments (flap and
reasonably well established (i.e., exhibit low edge) from a commercial turbine in complex terrain are
uncertainty), or they fail to cover the uncertainty when used. The ten-minute distributions of rainflow ranges
load estimates are based on limited data (i.e., large are modeled by a quadratic Weibull distribution
uncertainty cases). function based on three statistical moments of the
ranges (mean, coefficient of variation, and skewness).
Parametric load distribution models offer significant The moments are mapped to the wind conditions by a
advantages over empirical models; they provide a two-dimensional regression over ten-minute average
means to (1) extrapolate to higher, less frequent load wind speed and turbulence intensity. Thus, the “short-
levels, (2) map the response to the input conditions, and term” distribution of ranges may be predicted for any
(3) calculate load uncertainty. For example, Ronold et combination of average wind speed and turbulence
al.3 have published a complete analysis of the intensity. The “long-term” distribution of ranges is,
uncertainty in a wind turbine blade fatigue life then, easily obtained by integrating over the annual
calculation. They used a parametric definition of the distribution of input conditions. Results are compared
fatigue loads, matching the first three moments of the between standard-driven and site-driven load estimates.
wind turbine cyclic loading distribution to a quadratic Finally, using statistics based on the regression of the
(transformed by a squaring operation) Weibull statistical moments over the input conditions, the
distribution. uncertainty (due to the limited data set) in the long-term
load distribution is represented by 95% confidence
Veers and Winterstein4 described a parametric bounds on predicted loads.
approach, quite similar to that employed by Ronold et
al.3, that can be used with either simulations or IEC LOAD CASES
measurements, and have shown how it may be used in
an uncertainty evaluation. Although Reference 4 The loads specified by IEC 61400-1 Wind Turbine
describes how to use the statistical model to estimate Generator Safety Requirements for design must be
the complete load spectrum, it does not indicate how defined for a specified combination of mean wind speed
these models compare with the design standards5. It is and turbulence intensity known as the Normal
critical that the load distributions generated by any Turbulence Model5. The standard provides an equation
statistical methodology be adaptable for use in existing for the standard deviation of the ten-minute wind speed,
design standards. Moreover, it is arguably even more σ 1, that depends on the hub-height wind speed and two
important that the load model provide insight into how parameters, I15 and a.
the design standards might be improved in future
revisions. The standards should require an accurate σ 1 = I 15 (15m/s + aVhub ) /(a + 1) (1)
reflection of the load distribution with sufficient
conservatism to cover the uncertainties caused by the

2
Equation 1 is based on wind speed standard deviation conditions because of the limitations of the test site. In
data gathered from around the world and aggregated that case, the parametric approach provides a method to
into a common data set. The equation was created to be interpolate to a specified turbulence level using all of
“broadly representative of sites with reasonable the data collected (thus adding to the confidence of the
international marketing interest,”8 and does not interpolation), or to extrapolate beyond the limits of the
represent any single site. σ 1 is intended to represent a measurements. In either case, the parametric approach
characteristic value of wind-speed standard deviation. simplifies the generation of fatigue loads to Standard
Certification guidelines are provided for high (A) and specifications.
moderate (B) turbulence sites. I15 defines the
characteristic value of the turbulence intensity at an EXAMPLE DATA SET
average wind speed of 15 m/s, and a is a slope
parameter when σ 1 is plotted versus hub-height wind An example data set taken from the copious
speed. The values of these parameters for each measurements of the MOUNTURB program7 is used to
category are shown in Table 1. illustrate the parametric modeling process. The data are
comprised of 101 ten-minute samples of rainflow-
counted flap-wise and edge-wise bending-moment
A B
CATEGORY ranges at the blade root. The test turbine is a WINCON
(HIGH) (MODERATE)
110XT, a 110kW stall-regulated machine operated by
I15 0.18 0.16
CRES (the Centre for Renewable Energy Systems,
a 2 3 Pikermi, Greece) at their Lavrio test site. The terrain is
Table 1: Parameters for IEC turbulence categories. characterized as complex.

The Category B moderate turbulence specification is The original time series of the loads and winds were not
intended to roughly envelope (i.e., be higher than) the available for further analysis; thus, only the rainflow-
mean plus one sigma level of turbulence for all the counted ranges were employed. The number of cycle
collected data above 15 m/s. Similarly, Category A counts was tallied in 50 bins ranging from zero to the
envelopes all collected values of turbulence intensity maximum range in each sample. A single ten-minute
(with the exception of one southern California site) for sample is categorized by the mean wind speed and the
mean wind speeds above 15 m/s and is above the raw turbulence intensity at hub height. The average
overall mean plus two sigma level in high winds8. wind speeds are limited to the range between 15 and 19
Clearly, the IEC Normal Turbulence Model is intended m/s and thus reflect response in high wind operation.
to be conservative for all but the most turbulent sites. Turbulence intensities cover a wide range of operating
conditions as can be seen in Figure 1. The measured
It is a relatively straightforward matter to create a loads are summarized by frequency of occurrence in
loading distribution that meets the standard criteria Figure 2a for flap moment ranges and in Figure 2c for
when using an aeroelastic simulation code. Input winds edge moment. Plots showing exceedance counts for
can be generated for any combination of wind speed specified flap and edge loads are shown in Figures 2b
and turbulence intensity. Representative loadings can, and 2d, respectively.
in theory, be generated by simulating repeatedly until
sufficient data are produced to drive the uncertainty to 0.24
an arbitrarily small level. Practically, however, it
would be beneficial to generate a loading distribution
Turbulence Intensity

0.20
with small, or at least known, uncertainty from a
smaller data set. This is where the parametric approach
provides significant value. By means of regression of 0.16

load statistics (e.g., moments) over the entire range of

wind speeds and turbulence levels, the uncertainty in 0.12
the values of the parameters defining the short-term
distributions at any specified turbulence condition can 0.08
be estimated. 15 16 17 18 19
Wind speed (m/s)
In the case of measured loads, it may be simply Figure 1 Wind speed and turbulence intensity values
impossible to gather data at the specified turbulence for the 101 10-minute data samples.

3
 1000 1000

100 100
No. of occurrences

No. of occurrences
10 10

1 1

0.1 0.1
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Flap-wise bending moment range (kN-m) Edge-wise bending moment range (kN-m)

Figure 2a Histogram of flap-wise bending moment Figure 2c Histogram of edge-wise bending moment
ranges for 101 10-minute data sets. ranges for 101 10-minute data sets.
10000 10000

1000 1000
No. of exceedances of r

No. of exceedances of r
100 100

10 10

1 1

0.1 0.1
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Flap-wise bending moment range, r (kN-m) Edge-wise bending moment range, r (kN-m)

Figure 2b Cumulative counts of flap-wise bending Figure 2d Cumulative counts of flap-wise bending
moment ranges for 101 10-minute data sets. moment ranges for 101 10-minute data sets.

SHORT-TERM ANALYSIS
µ3 =
[
E (r − r ) 3 ], (4)
The FITS9 software for fitting probability distribution σr 3

models to empirical data was used to analyze each ten-

minute sample. FITS calculates the central moments of where E[ ] is the expectation (or averaging) operator.
the data and estimates the best fit distribution model to The first moment is the mean or average range, a
match a user-specified set of moments (e.g., the user measure of central tendency. The second moment is the
can request a distribution model fit based on two Coefficient of Variation (COV), which is the standard
moments or one based on three moments). FITS is, deviation divided by the mean, a measure of the
therefore, a tool for examining the fit of a probability dispersion or spread in the distribution. The third
model to the short-term response, conditional on wind moment is the coefficient of skewness, which provides
speed and turbulence level. information on the tail behavior of the distribution.
Load range data are often well fit by a Weibull
For purposes of the present discussion, the first three distribution, a slight distortion of the Weibull
moments µi, (i=1,2,3) of the rainflow-range amplitudes, distribution is used to exactly match the first three
r, are defined here as: statistical moments10.

µ 1 = E[ r ] = r , (2) To illustrate the fit of the quadratic Weibull distribution

to a ten-minutes sample, one of the 101 samples shown
µ2 =
σr
r
; σ r 2 = E (r − r ) 2 ,[ ] (3) in Figure 2 is studied. This data sample is taken from
the middle of the measured wind conditions; V = 17 m/s
and I = 0.18. The data are plotted on a Weibull scale
for the flap loads in Figure 3 and for the edge loads in
Figure 4. The vertical scale is transformed from the

4
Transformed CDF 10.0 10.0

Transformed CDF
1.0 1.0

0.1 0.1
1 10 100 1 10 100
Flap bending moment range (kN-m) Edge bending moment range (kN-m)

(a) Weibull scale plot. (a) Weibull scale plot.

10.0 10.0
Transformed CDF

Transformed CDF
1.0 1.0

0.1 0.1
1 10 100 1 10 100
Flap bending moment range - 11.5 (kN-m) Edge bending moment range - 32 (kN-m)

(b) Weibull scale plot (truncation at 11.5 kN-m). (b) Weibull scale plot (truncation at 32.0 kN-m)
10000 1000
Data Set with V = 17 m/s, I = 0.18
Quadratic Weibull Fit (with shift = 11.5 kN-m)
1000
Number of exceedances

Number of exceedances

100

100
10
10

1
1
Data set with V = 17 m/s, I = 0.18
Quadratic Weibull Fit (with shift = 32 kN-m)
0.1 0.1
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Flap bending moment range (kN-m) Edge bending moment ranges (kN-m)

(c) Exceedance plot format (truncation at 11.5 kN-m). (c) Exceedance plot format (truncation at 32.0 kN-m).

Figure 3 Quadratic Weibull model fits to data on flap- Figure 4 Quadratic Weibull model fits to data on edge-
bending moment ranges (V = 17.0 m/s, I = bending moment ranges (V = 17.0 m/s, I =
0.18). 0.18).

Cumulative Distribution Function (CDF) as (-ln(1- in Figure 3a and especially Figure 4a, the data have a
CDF)) so that a Weibull distribution will be a straight kinked appearance which the smooth probability
line on a log-log plot. (Recall that the CDF is the distribution, in spite of the quadratic distortion, can not
complement of the traditional exceedance diagram; match. Closer examination of the data reveals that the
exceedance = 1-CDF.) kink is due to a very large number of cycles at
relatively low amplitudes.
Figures 3a and 4a show attempts to fit the entire flap
and edge data with quadratic Weibull models. As seen

5
The proliferation of small amplitude cycles seen in reorient the reader back to the original summaries of the
Figure 2 produces a distribution difficult to duplicate data shown in Figure 2. They also serve to illustrate
with a simple analytical form, but these small cycles how the analytical distribution functions may be used to
produce relatively little damage. By truncating the extrapolate to less frequent, higher amplitude loads.
loads at a threshold, the kink in the data can be
eliminated without significantly reducing the damage. The low amplitude cycles (that make distribution fits
In the edge case, there are obviously a great number of difficult as described in the preceding) can only be
cycles of smaller amplitude than the dominant gravity discarded if they produce an insignificant amount of
load at about 32 kN-m. The flap loads have a less damage. The damage unaccounted for due to the
distinctive kink at around 10-13 kN-m (11.5 kN-m was truncation of rainflow range data at 11.5 kN-m for the
used as the filtering threshold). Figures 3b and 4b are flap loads is represented in Figure 5a, and due to a
similar to Figures 3a and 4a but include only a subset of truncation at 32 kN-m for the edge loads in Figure 5b.
the data and can be thought of as applying a “shift” to All 101 ten-minute data segments are represented in
all loads that effectively discards the smallest cycles. Figures 5a and 5b. Lost damage is plotted for three
Clearly, the fits of the quadratic Weibull are improved fatigue exponents, b, representing typical values of
dramatically. Thus, the short-term data are well wind turbine materials ranging from b=3 in welded
modeled by a quadratic Weibull distribution that steel up to b=9, more characteristic of fiberglass
preserves the first three central moments of the composites. In no case does the truncation remove
truncated rainflow ranges. more that 12% of the damage, and that only when b=3.
In almost all cases for the higher b values, the lost
Figures 3c and 4c show the same data as do Figures 3b damage is less than 3%. In both flap and edge bending
and 4b but with the axes in the more common cases, over 80% of the rainflow-counted ranges are
exceedance plot format. These plots are included to removed by truncating the data sets. Our findings that
discarding so much of the data does not lead to grossly
0
unconservative estimates of damage is not unexpected
since it has long been known that eliminating most of
Percent difference in damage

-3 the small amplitude cycles has a negligible effect on

due to shift of flap data

damage11.
-6 REGRESSION ANALYSIS

Because a good match can be obtained to the short-term

-9 b=3
distribution of rainflow ranges given the first three
b=6
b=9
moments and a fixed data truncation, it is sufficient to
-12 know the moments over all the operating conditions in
15.0 15.5 16.0 16.5 17.0 17.5 18.0 18.5 19.0 order to fully define the turbine fatigue loading.
Wind speed (m/s) Regression of the moments over ten-minute average
(a) Flap-wise bending wind speed and turbulence intensity can achieve the
desired result and assist in understanding the loading
0 dependence on and sensitivity to both turbulence and
wind speed. Results from a regression analysis can also
provide information on the uncertainty of the loads.
Percent difference in damage

-3
due to shift of edge data

The moments presented in the following figures, µ2 and

-6 µ3 describe the COV and skewness, respectively, of the
shifted range r’=r-rt. More precisely, by eliminating all
ranges below the truncation level rt, we obtain the
-9 b=3 shifted values r’=r-rt of the remaining ranges and
b=6 consider models based on statistics of rt. This is done
b=9
-12
to conform with the quadratic Weibull model, which
15.0 15.5 16.0 16.5 17.0 17.5 18.0 18.5 19.0 generally assigns probability to all outcomes r’ > 0.
Wind speed (m/s) For example, the second moment is
(b) Edge-wise bending
σr
µ2 = . (5)
Figure 5 Effect on damage estimation of shift in r − rt
blade bending moment range data.

6
 PARAMETER FLAP EDGE
COEFFICIENT FLAP EDGE
R2 – Mean, µ1 0.51 0.76
a1 21.49 40.02 a1 (σ) 0.108 0.050
Mean

b1 0.808 0.359 b1 (σ, t statistic) (0.081, 9.9) (0.020, 17.5)

c1 0.202 0.039 c1 (σ, t statistic) (0.032, 6.4) (0.008, 4.9)
a2 0.722 0.635
R2 – COV, µ2
COV

0.05 0.43
b2 0.031 -0.573
a2 (σ) 0.003 0.003
c2 0.080 0.063
a3 0.963 0.980 b2 (σ, t statistic) (0.076, 0.3) (0.080, 7.1)
Skew-

c2 (σ, t statistic)
Ness

b3 -1.260 -0.468 (0.030, 2.7) (0.031, 2.0)

c3 0.033 0.132 R2 – Skewness, µ3 0.17 0.05
Reference values: Vref =17.1m/s; Iref =0.145 a3 (σ) 0.018 0.016
b3 (σ, t statistic) (0.301, 4.2) (0.266, 1.8)
Table 2: Coefficients from regression analysis. c3 (σ, t statistic) (0.117, 2.8) (0.104, 1.3)
Avg. Cycle Rate 1.75 Hz 1.38 Hz
Because these definitions of µ2 and µ3 are
dimensionless,, they are insensitive to the choice of Table 3: Regression parameter summary.
truncation level. For example, if the loads are
exponentially distributed, µ2=1 and µ3=2 regardless of of ci, the less the importance of I in the estimate of the
the truncation level. The mean value, µ1, presented is ith moment.
the same as r , the mean with respect to zero of the
ranges retained after eliminating the small-amplitude Table 3 summarizes the regression uncertainties in
cycles. terms of the widely used R2 and t statistics. A high R2
value, approaching unity, indicates that a large
As in Reference 4, the first three moments (µi, i=1-3) percentage of the data variation is explained by the
are fitted to a power law function of wind speed, V, and regression. In contrast, a low R2 value suggests the
turbulence intensity, I. presence of other influences, not included in the
regression model, that induce the scatter in the data.
bi ci Note that the goal here is not to predict the moment
 V   I 
µ i = ai     (6) statistics in a single 10-minute history, but rather the
 Vref   I ref  long-run average of such 10-minute samples over the
   
entire turbine lifetime.
The reference wind speed, Vref, and reference
The t statistic, which is the estimated coefficient
turbulence intensity, Iref, are determined from the
divided by the standard deviation of the estimate,
geometric mean values of the data4. For the Lavrio data
indicates whether a particular coefficient is statistically
set, Vref=17.1m/s and Iref=0.145. The calculated
significant. A t value less than about two would
regression coefficients are shown in Table 2.
indicate that the coefficient is not significantly different
The regression results for the flap bending moments are from zero at about the 95% confidence level. Since the
shown in Figure 6 and for the edge bending moments in leading coefficients, ai, are estimates of the moments at
Figure 7. Mean, COV, and skewness are plotted in the the reference conditions, they are always significantly
parts (a), (b), and (c), respectively, of the figures. In all different from zero, and t is not reported for them.
cases, the regression line uses the reference value for However, the t values of bi and ci indicate whether any
turbulence intensity. The circles correspond to the functional variation with respect to wind speed or
measured data from the 101 samples. The solid turbulence intensity, respectively, is supported by the
symbols (squares) show the regression prediction using data.
the measured wind speed and turbulence intensity for
Examination of Tables 2 and 3 suggests that the mean
each ten-minute data set. A large spread in solid
load range is strongly related to both wind speed and
symbols about the regression line indicates a sizeable
turbulence, although the relation to turbulence has small
dependence on turbulence level (e.g., Figure 6a’s mean
exponents (0.202 and 0.039). The only higher moment
flap range) while a small variation in the solid symbols
relationships that have high t statistics are the edge
indicates that the turbulence has little effect on that
COV relation to wind speed and the flap skewness
particular moment (e.g., Figure 6c’s flap skewness).
relation to wind speed. The overall low exponents and t
This sensitivity can also be inferred from the magnitude
statistics for the higher moments indicate that the
of the ci coefficients in Table 2. The smaller the value
distribution shapes are relatively constant over all input

7
 25 44

Mean Edge BM Range (kN-m)

Mean Flap BM Range (kN-m)

23
42

21
40
19

Measured 10-minute data sets 38 Measured 10-minute data sets

17
Regression prediction at measured V and I Regression
Regression at I-ref Regression at I-ref
15 36
15 16 17 18 19 15 16 17 18 19
Wind speed (m/s) Wind speed (m/s)
(a) Mean (a) Mean
0.9 0.8

COV of Edge BM Range

COV of Flap BM Range

0.8 0.7

0.7 0.6

0.6 Measured 10-minute data sets 0.5 Measured 10-minute data sets
Regression prediction at measured V and I Regression
Regression at I-ref Regression at I-ref
0.5 0.4
15 16 17 18 19 15 16 17 18 19
Wind speed (m/s) Wind speed (m/s)
(b) Coefficient of variation (COV) (b) Coefficient of variation (COV)
1.6 1.6
Skewness of Edge BM Range
Skewness of Flap BM Range

1.2 1.2

0.8 0.8

0.4 0.4
Measured 10-minute data sets Measured 10-minute data sets
Regression prediction at measured V and I Regression
Regression at I-ref Regression at I-ref
0.0 0.0
15 16 17 18 19 15 16 17 18 19
Wind speed (m/s) Wind speed (m/s)

(c) Coefficient of skewness (c) Coefficient of skewness

Figure 6 Regression results for flap-wise bending Figure 7 Regression results of edge-wise bending
moment range. moment range.

conditions. The variation seen in Figures 6 and 7 The Lavrio data set used in this example is limited to a
beyond that indicated by the solid symbols is sample- range of wind speeds from 15 to 19 m/s. The long-term
to-sample variation not indicative of a systematic analysis in the next section will, for example purposes
relationship with the independent variables, V and I. only, assume the regression trends found in high winds
Part of this remaining variation will be irreducible, a apply to all wind speeds. In an actual application, the
natural outcome of random processes, but some could data from a particular turbine will need to be examined
possibly be reduced with regression over better over the entire range of damaging wind speeds. It
turbulence descriptors than the simple turbulence might be amenable to regression fits that run all the way
intensity. from cut-in to cut-out. More likely, the wind speed
range will have to be partitioned into divisions over

8
 which the response moments are well behaved enough designer. The turbulence intensity is a deterministic
to be fit with simple regression. For example, it is function of wind speed given by Eq. 1. A lifetime load
likely that the response will have different distribution must sum all the short-term distributions at
characteristics above and below rated wind speed. In each wind speed and associated turbulence intensity
that case, the analysis presented here would have to be weighting them by the annual wind speed distribution.
repeated for each wind-speed division before This can be written as
proceeding with the long-term analysis in the next
section. It may also be the case that the response in low F (r ) = ∫ F (r | V , I ) f (V )dV , (7)
winds has an insignificant contribution to the fatigue
damage and the analysis can safely deal with only high where F(r) is the long-term distribution of stress ranges,
wind response. The individual application will r, and F(r|V,I) is the short-term distribution of stress
determine the constraints. ranges conditional on the ten-minute average wind
speed, V, and the specified turbulence intensity, I. f(V)
LONG-TERM ANALYSIS
is the wind speed probability density function (PDF).
The long-term distribution of fatigue loads is obtained The integration is carried out over all damaging wind
by integrating the short-term distributions (for loads speeds. The distribution functions of r can be either the
conditional on wind conditions) over the specified CDF or the exceedence (1-CDF). However, the
distribution of wind conditions. Current IEC standards integration must be over the probability density
specify a Rayleigh distribution of wind speed with the function for wind speed, f(V).
annual average depending on class. Class I sites have a
Any environmental conditions can be used with Eq. 7
10 m/s average and Class II sites have a 8.5 m/s
once the response moments have been defined with
average. Wind-speed classes defined as “Special” are
respect to the turbulence levels and wind speeds. This
also allowed with conditions that may be defined by the
has been accomplished by the regression of the
1.E+0
moments over V and I and by determining the short-
V = 17 m/s
term distributions, F(r|V,I), from the moments. As
Probability of exceedance of r

1.E-2 examples we will calculate the long-term distributions

for two standard-driven and two site-driven
environments.
1.E-4
Figures 9a and 9b show the long-term distributions of
flap and edge loads respectively for the IEC Class I
1.E-6 Reference Intensity
IEC Class A Intensity
wind speed distribution (Rayleigh 10m/s) and for both
IEC Class B Intensity turbulence Categories A and B. Both of these standard
1.E-8 environments define the turbulence level as a function
0 20 40 60 80 of wind speed by Eq. 1.
Flap bending moment range, r

(a) Flap-wise bending The specification of a fixed turbulence intensity

functionally related to the wind speed is somewhat
1.E+0 artificial; measurements indicate that the turbulence
V = 17 m/s intensity varies over a range of values for each ten-
minute sample (see Figure 1). A more realistic
Probability of exceedance of r

1.E-2
representation than Eq. 7 for the long-term distribution
might be to include turbulence intensity as a random
1.E-4
variable by integrating over both wind speed and
turbulence intensity as follows:

1.E-6 Reference Intensity F (r ) = ∫∫ F (r | V , I ) f (V , I )dIdV (8)

IEC Class A Intensity
IEC Class B Intensity
1.E-8 Figures 9a and 9b show the long-term flap and edge
20 40 60 80 distributions derived by integrating over both wind
Edge bending moment range, r speed and turbulence assuming wind speed and
(b) Edge-wise bending turbulence variations are independent. (Independence
implies that the joint PDF, f(V,I) in Eq. 8 is simply the
Figure 8 Distribution of bending moment ranges product of the individual marginal PDFs, f(V) and f(I).)
conditional on wind speed and turbulence
intensity.

9
 1E+0 0.24
Data
I = 2.5/V (regression)
regression mean + 1 std dev
regression mean - 1 std dev
IEC-A
Probability of exceedance

Turbulence Intensity
1E-2 0.20 IEC-B

1E-4 0.16
IEC-A Turbulence Intensity
IEC-B Turbulence Intensity
1E-6 0.12
I assumed normal from data
Average I at each wind speed
1E-8 0.08
0 20 40 60 80 15 16 17 18 19
Flap bending moment range (kN-m) Wind speed (m/s)
(a) Flap-wise bending Figure 10 Lavrio site turbulence intensity as a function
1E+0 of wind speed, regression fits, and IEC
Category A and B definitions.
Probability of exceedance

1E-2
Discussion

The Lavrio site’s mean-plus-one-sigma turbulence

1E-4
intensity at 15 m/s wind speed is quite similar to the
IEC-A Turbulence Intensity
IEC standard specification of 16% (Class B) to 18%
IEC-B Turbulence Intensity
1E-6 (Class A). This similarity in turbulence levels is
I assumed normal random variable
evident throughout the high wind range as shown in
Average I at each wind speed Figure 10. The differences between the distributions in
1E-8 Figures 9a and 9b therefore provide an indication of the
0 20 40 60 80
Edge bending moment range (kN-m)
conservatism built into the IEC load cases relative to a
fairly turbulent site.
(b) Edge-wise bending
Within the context of standards development, it may be
Figure 9 Long-term distribution of edge-wise bending
reasonable to argue for lower turbulence specifications
moment ranges (Rayleigh distributed wind
if differences as seen above can be shown to be
speed with mean = 10 m/s).
significant and consistent. However, because the
standards are based on past experience and industry
The wind speed is a 10 m/s-average Rayleigh consensus rather than objective risk-based analysis, it
distribution as prescribed for IEC Class I sites. The may be dangerous to remove conservatism from one
turbulence is assumed to be normally distributed with area without also checking elsewhere to insure that this
mean, I , defined by I ≅ 2.5 / V , and standard conservatism isn’t covering for an unknown lack of
deviation equal to 0.025, based on a best fit to the data conservatism elsewhere in the design process.
of turbulence vs. wind speed shown in Figure 10.
In general, the current standards give a load calculation
Also plotted in Fig. 9 is the result of assuming the “recipe” that is meant to meet some specific reliability
turbulence at the Lavrio site is defined by the average criteria. If these current reliability levels are deemed
value at each wind speed (the line with the black circles adequate on average (over various cases), one cannot
in Figure 10). This simpler assumption allows the use reduce conservatism in turbulence specification without
of Eq. 7. The comparison indicates that the integration adjusting the recipe to compensate elsewhere; e.g.,
over all turbulence levels, which is the most realistic through use of a higher load factor. Note that this
reflection of the measurements, produces a much lower alternative procedure – unbiased turbulence with higher
load spectrum. The simplified alternative, i.e., fixing load factor – may result in more uniform reliability
the turbulence at the mean level, leads to a more across a range of cases. In contrast, current standards
conservative result, although the conservatism may lead to potential over-design of machines that are
diminishes at very high load levels for the flap bending particularly sensitive to turbulence, and under-design in
moment ranges. However, these results may be turbulence-insensitive cases.
sensitive to the choice of a normal distribution for
turbulence intensity.

10
 1E+0
ESTIMATING UNCERTAINTY IN LONG-TERM
LOADS

Probability of exceedance
1E-2
To review, the parametric load modeling proposed here
proceeds by (1) modeling loads by their statistical
moments µi (i=1,2,3) and (2) modeling each moment µi 1E-4
as a parametric function of V and I (Eq. 6). The
Deterministic Estimate
moment-based model in step (1) is in principle
1E-6
independent of the turbine characteristics (although the
optimal choice among such models may be somewhat 95% Confidence Level

case-dependent). Hence, in this parametric approach, 1E-8

the turbine characteristics are reflected solely through 0 20 40 60 80
the moment relations in Eq. 6; specifically, the 9 Flap bending moment range (kN-m)
coefficients ai, bi, ci (i=1,2,3). For clarity, we organize (a) Flap-wise banding
these here into a vector,
= {a1 , b1 , c1 , a 2 , b2 , c 2 , a 3 , b3 , c 3 } . 1E+0
Simpler 2-moment models would require only 6
coefficients.

Probability of exceedance
1E-2

The preceding section has shown one benefit of this

parametric model. Because it permits load statistics to
1E-4
be estimated for arbitrary V and I, the results can be
weighted to form the long-term loads distribution as in Deterministic Estimate
Eqs. 7-8 (and Figs. 9a-b). Symbolically, we rewrite Eq. 1E-6
7 here, noting explicitly its dependence on the vector θ. 95% Confidence Level

F (r | ) = ∫ F (r | V , I , ) f (V )dV (9) 1E-8

0 20 40 60 80
Edge bending moment range (kN-m)
(Eq. 8 can be rewritten analogously.) The foregoing (b) Edge-wise Bending
results (Figs 9a-b) have used our best estimates for the
entries of θ; i.e., the mean values of each entry in θ. Figure 11 95% Confidence levels on the exceedance
These are the values of ai, bi, and ci cited in Table 2. probability of fatigue loads for the Lavrio
site with turbulence set to the average value
A further advantage of the parametric model lies in its for each wind speed.
usefulness in estimating the effects of statistical
uncertainty. To clarify, it is useful to distinguish Figures 11a and 11b show the 95% confidence level on
between the various terms in Eq. 9. The quantities V the exceedance probability, 1-F(r), which result from
and I are “random variables;” that is, their future the simulation procedure described above. Each of the
outcomes will show an intrinsic randomness that cannot 9 coefficients in Eq. 9 were generated as statistically
be reduced by additional study of past wind conditions. independent, normally distributed random variables,
In contrast, the 9 coefficients in θ are in principle fixed with means and standard deviations given by Tables 2
(under the model’s assumptions). We may, however, and 3, respectively. (Correlation among these variables
be uncertain as to their values due to limited response can also be included; however, this was not done here.)
data. This “uncertainty” (as opposed to “randomness”) All of these results adopt the site-specific mean
can be reduced through additional sampling. The turbulence model; i.e., the results labeled “Average I at
consequence of having only limited data can be each wind speed” in Figs. 9a-b. These results from
reflected through 95% confidence levels, for example, Figs. 9a-b are repeated in Figs. 11a-b, and referred to
on the exceedance probability 1-F(r). These are there as "deterministic" results. Also shown are 95%
conceptually straightforward to establish by simulation. confidence results; i.e., probability levels below which
Assuming the entries of θ are each normally distributed, 95% of the simulations fall.
for example, one may (1) simulate multiple outcomes
of θ; (2) estimate F(r) for each θ as in Eq. 9; and (3) The increase in probability, over the deterministic
sort the resulting F(r) values (at each fixed r value) to results in order to achieve 95% confidence, is found to
establish confidence bands; e.g., in which 95% of the be relatively modest. This reflects the benefit of having
values lie. as many as 101 10-minute samples. If the same mean

11
 trends had resulted from fewer samples, the resulting Energy Symp., at the AIAA Aerospace Sciences
95% confidence results would be correspondingly Mtg., Reno, Nevada, AIAA-2001-0046, January
higher than the mean results. Note also that, at least for 2001.
flap-wise loads, the conservatism induced by the IEC 2. McCoy, T. J., Malcolm, D. J., and Griffin, D. A.,
turbulence models exceeds that required to cover our “An Approach to the Development of Turbine
statistical loads uncertainty, based on the data at hand. Loads in Accordance with IEC 1400-1 and ISO
Of course, as noted earlier, this IEC conservatism may 2394,” A collection of the 1999 ASME Wind
be desirable to cover other sources of uncertainty. Energy Symp., at the AIAA Aerospace Sciences
Finally, we caution again that these long-term load Mtg., Reno, Nevada, AIAA-99-0020, January
results are intended for example purposes only; 1999, pp. 1-9.
accurate numerical values would require data across a 3. Ronold, K. O., Wedel-Heinen, J. and Christensen,
broader range of wind speeds. C. J., “Reliability-based fatigue design of wind-
turbine rotor blades,” Elsevier, Engineering.
SUMMARY Structures 21, 1999, pp. 1101-1114.
4. Veers, P. S. and Winterstein, S. R., “Application of
Fatigue load spectra are generated for arbitrary site Measured Loads to Wind Turbine Fatigue and
conditions (wind speed and turbulence intensity Reliability Analysis,” Journal of Solar Energy
distributions) by using parametric models to fit the Engineering, Trans. of the ASME, Vol. 120, No. 4,
short term load spectrum to the first three moments of November 1998.
the truncated rainflow range distributions and 5. IEC/TC88, 61400-1 Wind Turbine Generator
regressing the moments over wind speed and turbulence Systems – Part 1: Safety Requirements,
intensity. The spectra are generated to specified IEC International Electrotechnical Commission (IEC),
conditions for wind speed Class and turbulence Geneva, Switzerland, 1998.
Category. The spectra are also generated for as- 6. IEC/TC88, Draft IEC 61400-13 TS, Ed. 1: Wind
measured scatter in the turbulence levels across all wind turbine generator systems – Part 13: Measurement
speeds. The comparison of the two approaches reveals of mechanical loads, 88/120/CDV, International
the level of conservatism that results from assumed Electrotechnical Commission (IEC), Geneva,
high turbulence levels written into the current Switzerland, 1999.
standards. The selected confidence level can be 7. MOUNTURB, Load and Power Measurement
calculated using the statistics from regression analysis. Program on Wind Turbines Operating in Complex
Since the confidence interval depends on the Mountainous Regions, Volumes. I - III, Editor P.
uncertainty in the load characterization, it could provide Chaviaropoulos, Coord. A. N. Fragoulis, CRES,
a better margin of safety on the loads than can be RISO, ECN, NTUA-FS, published by CRES,
accomplished with an inflated turbulence level. The Pikermi, Greece, Nov. 1996.
parametric approach presented here illustrates how 8. Butterfield, S., Holley, B., Madsen, P. H. and
statistically based standards may be able to reflect the Stork, C., Report on 88/69/CD – Wind Turbine
uncertainty in the loading definition caused by finite- Generator Systems Part 1: Safety Requirements, 2.
length data records. Edition, National Renewable Energy Laboratory,
Golden, CO, Oct. 1997.
ACKNOWLEDGMENTS 9. Manuel, L., Kashef, T. and Winterstein, S. R.,
Moment-Based Probability Modelling and Extreme
The authors would like to extend a special debt of
Response Estimation The Fits Routine, Version 1.2,
gratitude to the Centre for Renewable Energy Sources
SAND99-2985, Sandia National Laboratories,
(CRES), Pikermi, Greece and especially to Fragiskos
Albuquerque, NM, Nov. 1999.
Mouzakis who found time in a busy schedule to supply
10. Lange, C. H., Probabilistic Fatigue Methodology
the portion of the MOUNTURB data set used in the
and Wind Turbine Reliability, SAND96-1246,
examples. The authors also gratefully acknowledge the
Sandia National Laboratories, Albuquerque, NM,
efforts of LeRoy Fitzwater, of Stanford University, who
May 1996.
provided the uncertainty analysis described in the final
11. Nelson, D. V., and Fuchs, H. O., “Predictions of
section of this paper.
Cumulative Fatigue Damage using Condensed
REFERENCES Load Histories,” Fatigue under Complex Loading:
Analysis and Experiments, Advances in
1. Fitzwater, L. M. and Winterstein, S. R., “Predicting Engineering, Vol. 6, Ed. R. M. Wetzel, SAE,
Design Wind Turbine Loads from Limited Data: Warrendale, PA, 1977, pp. 163-187.
Comparing Random Process and Random Peak
Models,” A collection of the 2001 ASME Wind

12

View publication stats