0011 Evans2018

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Accepted Manuscript

Assessing the IEC simplified fatigue load equations for small wind turbine blades:
How simple is too simple?

S.P. Evans, D.R. Bradney, P.D. Clausen

PII: S0960-1481(18)30449-X
DOI: 10.1016/j.renene.2018.04.041
Reference: RENE 9999

To appear in: Renewable Energy

Received Date: 16 December 2017


Revised Date: 23 March 2018
Accepted Date: 10 April 2018

Please cite this article as: Evans SP, Bradney DR, Clausen PD, Assessing the IEC simplified fatigue
load equations for small wind turbine blades: How simple is too simple?, Renewable Energy (2018), doi:
10.1016/j.renene.2018.04.041.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Title:

Assessing the IEC simplified fatigue load equations for small wind turbine blades: how simple is too
simple?

Author name and affiliations

S.P. Evans1

PT
D.R. Bradney1

P.D. Clausen1

RI
1
School of Engineering, Faculty of Engineering and Built Environment, The University of Newcastle,
Callaghan NSW 2308 Australia

SC
Corresponding author:

S.P. Evans

Email: Samuel.evans@uon.edu.au
U
AN
Address: School of Engineering, Faculty of Engineering and Built Environment, The University of
Newcastle, Callaghan NSW 2308 Australia
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

1 Assessing the IEC simplified fatigue load equations for small wind turbine
2 blades: how simple is too simple?

3 S.P. Evansa,∗, D.R. Bradneya , P.D. Clausena


4
a School of Engineering, Faculty of Engineering and Built Environment, The University of Newcastle, Callaghan NSW 2308

PT
5 Australia

RI
6 Abstract
7 It is well known that wind turbine blades are fatigue critical, with much literature and methodologies

SC
8 available for assessing fatigue loading of large wind turbine blades. Little research effort has been directed
9 at assessing the fatigue life of small wind turbines which operate at higher rotational speeds and are subject
10 to highly unsteady aerodynamic loading. In this paper the simplified load model proposed in IEC 61400.2 is
11 used to determine the fatigue life of a small 5 kW wind turbine blade. This estimated life is compared to that
determined from both measured operational data and aeroelastic simulations. Fatigue life was estimated

U
12

13 by the standard at 0.09 years, compared to 9.18 years from field measurements and 3.26 years found via
aeroelastic simulations. All methods fell below the 20 year design life, with the standard over-conservative by
14
AN
15 a factor of 102 and 36 for measurements and simulations respectively. To the best of the authors’ knowledge
16 these three fatigue methods specified in the standard have not been quantitatively compared and assessed for
17 small wind turbines. Results are of importance to small wind turbine developers as they seek best practice
18 for determining blade fatigue life. Shortcomings of the IEC methodology are detailed and discussed.
M

19 Keywords:
20 IEC 61400.2, small wind turbines, fatigue loading, FAST
D
TE

21 1. Introduction

22 Wind turbines are known to be fatigue critical in that failure of components is likely to occur by the
23 accumulation of fatigue damage, as opposed to application of ultimate loads [1]. Fatigue damage in wind
EP

24 turbines has been the subject of numerous studies, and is a relatively complex field due to the nature of
25 aerodynamic loading, the complex fatigue properties and fatigue mechanisms of composite materials, and
26 difficulties in accurately measuring in-service loading [2, 3, 4]. Blades are the subject of most research
27 interest as they are critical to turbine operation, are exposed to a variety of load cases, and are typically
C

28 expensive to produce. Design drivers for small wind turbine blades are focused on reducing material and
29 manufacturing cost, while still retaining structural integrity [5]. This is necessary to ensure a competitive
levelised cost of energy (LCOE) when compared to other means of micro-generation such as solar photo-
AC

30

31 voltaic [6].
32

33 Typical methodologies for assessing fatigue life of wind turbine blades includes measuring or simulating
34 (via aeroelastic codes) in-service loading [7]. This time series load signal then undergoes analysis such as
35 rainflow counting, whereby cyclical damage is extracted from a complex time series signal [8]. These dam-
36 age cycle magnitudes can then be used as input to a fatigue model used to determine when fatigue failure
37 occurs for a given material (such as those presented in [9]). Experimental testing (both static and fatigue)

∗ Correspondingauthor
Email address: samuel.evans@uon.edu.au (S.P. Evans)

Preprint submitted to Renewable Energy April 10, 2018


ACCEPTED MANUSCRIPT

38 of manufactured blades is an important step in the product development process. For large wind turbine
39 blades, physical testing is compulsory and mandated under the relevant standard (IEC 61400.23-2014: Full-
40 scale structural testing of rotor blades), a reason which has undoubtedly driven research into developing
41 methodologies for assessing fatigue life of large blades [10].
42

43 When compared to large scale wind turbine used for commercial power generation, small wind turbines
are formally defined has having a swept area less than 200 m2 which equates to a blade length less than

PT
44

45 8 m [11]. Turbines of this class are not simply a ‘scaled down’ version of large wind turbines, as they
46 have appreciably different operating dynamics, wind regimes, and structural designs. Small wind turbines
47 operate at higher rotational speeds to ensure optimal rotor efficiency, and are exposed to highly unsteady
aerodynamic loading such as during yaw events (for a turbine controlled passively via a tail fin) or during

RI
48

49 rotor start up. Gravitational loading is typically ignored due to small blade mass and comparatively higher
50 blade stiffness [5], however gyroscopic inertial loading during rotor yaw has shown to be a governing load
51 case for free yaw turbines [12]. Small wind turbines can even experience an order of magnitude more fa-

SC
52 tigue cycles than commercial scale wind turbines, due to their higher rotational operating speeds [13]. As a
53 final consideration, small wind turbines are often sited within the built environment and exposed to highly
54 turbulent inlet wind conditions [14, 15, 16]. This has resulted in less-than-ideal capacity factors [17], and
55 a number of public structural failures, such as that detailed in [18]. Improving the structural reliability of

U
56 blades will mitigate safety hazards and increase availability factors.
57
AN
58 Methodologies for testing the fatigue life of large wind turbine blades do not necessarily translate well
59 to small wind turbines, for example, hydraulic actuators are not capable of achieving the high load rate fre-
60 quencies for physical testing [19]. Developed fatigue spectra such as the WISPER and WISPERX database
61 [20] used load measurements taken from larger scale turbines, which had rotor diameters and blade masses
M

62 higher than the 5 kW Aerogenesis turbine (subject of this study), and operational rotational speeds much
63 lower than the Aerogenesis wind turbine. In [21], an equivalent annual small wind turbine fatigue life spectra
64 was developed by directly equating wind inlet speed to an estimated blade loading state. While in theory
65 blade loading response would follow inlet wind velocity for steady state operation, sustained periods of
D

66 steady operation are rarely observed in site wind conditions. One obvious shortcoming of this approach is
67 that turbine operational dynamics which impact blade response, such as; rotor acceleration/deceleration,
TE

68 operational yaw errors, and yaw rate induced gyroscopic loading, would not be accounted for by assessing
69 the inlet wind velocity alone. These unsteady loads have been shown to be significant [22, 23]. Studies
70 such as [24] have used combined blade element and finite element approaches to predict fatigue life of small
71 wind turbine blades. Little other research detailing measured fatigue spectra for small wind turbine blades
EP

72 is available in the open literature.


73

74 With regards to component design, IEC 61400.2-2013 (henceforth referred to as ‘the standard’) allows for
75 three methods in determining design loads: loads acquired from experimental measurements, loads obtained
C

76 via aeroelastic simulations, or via the simplified load model (SLM). Each method presents various advantages
77 and disadvantages, for example; loads measured in-service would provide the most accurate load magnitudes,
however undertaking experimental measurements can be a complex and time consuming task. Using the
AC

78

79 SLM means that design loads can be deduced comparatively quickly at the ‘cost’ of high safety factors and
80 increased conservatism. It has been noted that a lack of experimental load measurements and quantitative
81 comparison to the SLM presents a barrier to small wind developers which may risk over-designing compo-
82 nents [25, 13]. Comparisons between the SLM, aeroelastic simulations, and field measurements were made
83 for small wind turbine tower loadings in [26]. It was found that the SLM does not adequately address tower
84 fatigue in this instance, as only one large damage cycle range is assumed, which is not indicative of the
85 measured variable fatigue spectra. Physical blade fatigue testing is not mandated under the standard which
86 may explain this ‘knowledge gap’ between small wind turbines and their larger counterparts with respect to
87 blade fatigue life assessment.
88

89 This study is aimed at assessing the fatigue life of a 2.5 m long small wind turbine blade as calculated by
2
ACCEPTED MANUSCRIPT

90 the SLM. Comparisons will be made to fatigue loads determined via aeroelastic simulations, and with loads
91 obtained from field measurements. A 5 kW Aerogenesis horizontal-axis wind turbine will be the subject of
92 this study, of which an aeroelastic model has been developed within FAST [27]. The lead-author has made
93 this model publicly available1 . To the best of the authors’ knowledge, these three methods of determining
94 blade fatigue life have not been quantitatively compared previously. Results of this study are expected to
95 provide guidance for small wind developers in developing light-weight and structurally sound blades, while
avoiding unnecessarily high safety factors. This is important to drive further reductions in LCOE for small

PT
96

97 wind generation.
98

RI
99 2. Methodology

100 The methodology for assessing the suitability of the SLM as applied to blade fatigue loading will be
outlined. While this study is focused on a specific blade design, this process could be followed and adapted

SC
101

102 as necessary for any blades on a turbine of this class. The Aerogenesis blade has a nominal length of 2.5 m
103 and is constructed from glass fibre reinforced polymer (GFRP). The fatigue spectrum and estimated fatigue
104 life as per the standard will first be calculated as a baseline case. This will then be compared to fatigue
105 loads determined from aeroelastic simulations and field measurements, as shown in figure 1. The measured

U
106 and simulated blade load signals will undergo rainflow counting whereby load cycle mean and range will
107 be output. The standard rainflow counting technique adopted by ASTM will be used for this study [28].
AN
108 Fatigue life will then determined using the Palmgren-Miner linear damage model (simply referred to as
109 Miner’s sum). Damage equivalent loads (DELs) will also be produced and compared for each of the three
110 methods.
111
M

112 This process could be considered akin to the product design cycle whereby simplified equations and
113 assumptions are first used for an initial approximation of design loads, with progressively more detailed
114 methods of determining loads such as aeroelastic modelling and then field testing providing improved sources
of design verification.
D

115

116
TE
C EP
AC

1 http://hdl.handle.net/1959.13/1349817 – Accessed 15/03/2018

3
ACCEPTED MANUSCRIPT

2. Simulated
3. Measured
blade loads
blade loads
(aeroelastic)

PT
Rainflow Rainflow
counting counting

RI
1. IEC 61400.2
Simulated Measured
SLM fatigue
fatigue spectra fatigue spectra
load case

SC
Comparison of

U
predicted fatigue
life
AN
Figure 1: Methodology for comparing fatigue life determined via the standard, aeroelastic simulations, and operational
measurements.
M

117 At this point it is necessary to comment on the period of time required to produce a sufficient under-
standing of load behaviour and fatigue life during operation. Naturally it is not feasible to measure loads
D

118

119 for a given 20 year design life, which means that a shorter sampling period is required. Ideally for wind
120 turbine operation, load measurements at a range of wind speeds from cut-in wind speed to the cut-out wind
TE

121 speed would be necessary for sufficient classification of operational loading. As the Aerogenesis turbine is a
122 free yaw and variable speed turbine, loads at a range of yaw errors, yaw rates, rotor speeds, and generator
123 electrical power production ranges are also required for full classification of loads.
124
EP

125 Table 1 presents summary operating statistics for both the measured and simulated data sets, whereby
126 the blade loading has been measured and simulated across a sufficient range of turbine operating parameters.
127 It is important to note that the measured data consists of approximately four hours of turbine operation, se-
128 lected to include all wind speeds, rotor speeds, and generator behaviour from below cut-in to above cut-out.
C

129 This is in comparison to the aeroelastic simulation which is based on ten-minute, Rayleigh distributed sam-
130 ples, and presents a discrepancy in sample time length. The number of cycles produced in each data file will
AC

131 be normalised and then multiplied by an equivalent annual number of cycles so a comparison can be made
132 between all data sets. Usage of simulations of these time periods to deduce fatigue loading of various wind
133 turbine components is not uncommon, as several studies have utilised ten-minute sample periods [26, 29, 16].
134

4
ACCEPTED MANUSCRIPT

Table 1: Summary statistics for both measured and simulated (FAST) data sets.
Parameter Measured Simulated (FAST)
Mean Min Max Mean Min Max
Inlet wind speed (ms−1 ) 8.20 1.0 22.0 7.5 3.5 20.0
Yaw error (◦ ) -2 -145 129 -2 -90 89
Yaw rate (◦ /sec) 0 -46 69 0 -64 58

PT
Power (W) 1,283 0 5,277 2,767 0 5,000
Rotor speed (rpm) 244 65 293 243 120 320

RI
135 2.1. IEC 61400.2 simplified fatigue load equations
136 Section 7.4 of the standard describes what is known as the simplified load model (SLM). This model
137 presents simplistic equations for determining ultimate and fatigue loads of small wind turbine components.

SC
138 The SLM allows a designer to avoid potentially complex load measurements or detailed aeroelastic simula-
139 tions at the trade-off of increased design conservatism. Key parameters used for the SLM equations include
140 the design rotational speed, design wind speed, design shaft torque, the maximum yaw rate, design power,
141 and the maximum rotational speed.

U
142

143 Load case A: normal operation, is the only load case that specifically details fatigue loading. For the
blades, this load case assumes fatigue loading to occur at a constant fatigue range for the life of the com-
AN
144

145 ponent (illustrated in figure 2). This fatigue loading is defined as occurring at a steady design wind speed
146 of 10.5 ms−1 . Fatigue loading is comprised of centrifugal effects acting in the blade axial direction (∆FzB ),
147 lead-lag bending moment (i.e. blade torque, ∆MxB ), and flapwise bending (due to thrust loading, ∆MyB ),
148 with these ranges given in equations 1 to 3. These values are said to be a peak-to-peak value which is applied
M

149 to the component for the entire design life (given as 20 years in this instance). Transient events such as
150 yawing, stopping, and starting are said to occur insufficiently as not to impact fatigue life. Turbine capacity
151 factor due to site wind resource and outages for maintenance are not considered. These peak-to-peak blade
D

152 loads are calculated via:


153
TE

2
∆FzB = 2mB rcog ωdesign (1)

Qdesign
∆MxB = + 2mB grcog (2)
B
EP

λdesign Qdesign
∆MyB = (3)
B
Where mB , is the blade mass (kg), rcog is the radial distance from the blade axis of rotation to the
C

154

155 centre of gravity (m), and B is the number of blades. Pdesign , ωdesign , and λdesign , are the power (W), rotor
156 speed (rad/s), and tip speed ratio (dimensionless) at design conditions. Other necessary parameters, design
AC

157 torque, Qdesign (Nm), and efficiency, η (dimensionless), are defined as:
158

30Pdesign
Qdesign = (4)
ηπωdesign

η = 0.6 + 0.005Pdesign (5)


159 An obvious shortcoming of this SLM fatigue load case, visualised in figure 2, is that it only considers
160 steady operation at a design speed of 10.5 ms−1 . Equations 1 to 3 do not incorporate any unsteady effects
161 such as gyroscopic loads, sustained operation at a yaw error, or complex aerodynamic phenomena such
162 as stall hysteresis which may exacerbate blade loading during transient operation. Previous experimental
5
ACCEPTED MANUSCRIPT

163 measurement campaigns detailed in [27, 30] indicate that the turbine would spend a significant period of
164 its operational life in these instances of unsteady operation. While the turbine rotor response is cyclical in
165 nature, it is highly unlikely that fatigue loading would occur as sinusoidal variations between 0.5 to 1.5 of
166 the mean steady state load at a rate of twice the rotor operating frequency for the entire operating life.
167 Unsteady effects would inherently produce a much more complex damage spectra and would likely include
168 a wider distribution of damage cycle ranges and stress ratios.

PT
169

1800

1600

RI
Flapwise blade moment (Nm)

1400

1200

SC
1000
M = 952 Nm
yB
800

600

U
400
AN
200

0
0 0.5 1 1.5 2
Operational time (s)
M

Figure 2: Sinusoidal blade flapwise fatigue loading as determined by the SLM.


D

170 In order to determine the equivalent stress, σeqB , for fatigue analysis, combination loading (i.e. total
171 beam fibre stress) is said to occur at the blade root as per:
TE

σeqB = σzB + σM B (6)


172 With the axial, σzB , and bending, σM B , components given by:
FzB
EP

σzB = (7)
AB
MxB MyB
σM B = + (8)
WxB WyB
C

173 Where AB is the cross sectional area of the blade root (m2 ), and WxB = Ixx /x and WyB = Iyy /y are
the section moduli (m3 ) in the lead-lag and flapwise direction respectively.
AC

174

175

176 For this fatigue analysis, the highest fatigue load is said to occur at the blade root, where moments are
177 at a maximum as per the cantilevered beam condition. The Aerogenesis blade root sectional properties are
178 given as; Ixx = 25.76×106 mm4 , Iyy = 644.33×103 mm4 , and AB = 1,500 mm2 . Note the second moment
179 of area is approximately 40 times stiffer in the lead-lag direction, compared to bending about the flapwise
180 direction. It is well known that blade loading in the flapwise direction is the most significant source of
181 loading when compared to either the axial centrifugal load and the in-plane lead-lag load, with these two
182 second order loads typically ignored in the design of small wind turbine blades [5]. Calculations based on
183 the SLM reveal lead-lag and centrifugal stresses to be less than 5% of the maximum flapwise fatigue stress
184 calculated to be 55.41 MPa (where the maximum flapwise bending load in this instance was found to be:

6
ACCEPTED MANUSCRIPT

185 1.5 × ∆MyB = 1,428 Nm). From this the equivalent stress, σeqB , acting at the blade root section given in
186 equation 9, can be reduced to stresses due to flapwise loading only, as per:
MyB
σeqB ≈ (9)
WyB
187 To estimate the fatigue life using Miner’s sum in equation 10, ‘damage’ is a parameter defined as a
fraction of material life that has been reduced due to the application of a fatigue cycle, ni , at a given load

PT
188

189 level, si (stress or strain), which can withstand a total of N cycles before failure. Virgin material is said to
190 have damage parameter equal to zero, whereas the material life is exhausted when damage is equal to unity,
191 and fatigue failure is said to occur. It should be noted that while damage is not a physical parameter, it

RI
192 has seen widespread use in predicting fatigue life. The standard makes use of dimensionless safety factors;
193 γf (load factor), and γm (material factor).
k
ni

SC
X
damage = (10)
i=1
N (γf γm si )
194 The number of fatigue cycles for operational life is said to be equal to:

U
Bndesign Td
n= (11)
60
AN
195 Where Td is the design life of the wind turbine in seconds. For the Aerogenesis turbine a design life of
196 20 years at a, design rotor speed ndesign = 320 rpm of is assumed. When considering the above guidance
197 from the standard, a load range of 952 Nm (determined as 36.94 MPa via equation 9) is said to occur at a
198 rate of 10.67 Hz for 20 years of operation totalling 6.73 × 109 fatigue cycles (from equation 11). This now
M

199 forms a baseline at which fatigue loading can now be calculated and compared to both the measured and
200 simulated data sets.

201 2.2. Aeroelastic simulations


D

202 Time-series aeroelastic simulations were undertaken within FAST2 using the 5 kW Aerogenesis turbine
203 model detailed in [27] to deduce the operational blade loads. A comprehensive overview of the model will
TE

204 not be provided here, instead the reader is referred to [27] for a detailed explanation and experimental ver-
205 ification. In order to execute the simulations a series of ten-minute wind periods from cut-in (3.5 ms−1 ) to
206 cut-out (20 ms−1 ) wind speeds taking the form of a Rayleigh distribution were produced within TurbSim3
207 using the Kiamal turbulence model. A mean wind speed of 7.5 ms−1 was used for the Rayleigh distribution
EP

208 at a turbulence intensity of 18%. This corresponds with that of a IEC Class III small wind turbine, for
209 which the Aerogenesis turbine was designed. Simulations of this kind may be undertaken early in the design
210 phase before site wind conditions have been fully classified.
211
C

212 The simulation was then executed in FAST with a variety of output parameters recorded such as,
213 generator power, rotor speed, blade and tower loads, structural deflections, and yaw behaviour. Blade
AC

214 flapwise bending moment is detailed in figure 3, with initial comparisons to the SLM fatigue load case
215 revealing an over estimation of load ranges for this class. This signal was used as input for rainflow counting
216 for later determination of fatigue life. Fatigue damage cycles were found to occur at a rate of 3.37 Hz, which
217 is within the once-per-cycle operating range (i.e. 1P) of this variable speed turbine.

2 https://nwtc.nrel.gov/FAST - Accessed 14/10/2017


3 https://nwtc.nrel.gov/TurbSim - Accessed 14/10/2017

7
ACCEPTED MANUSCRIPT

1500

1250 FAST simulation


SLM fatigue cycle range

Flapwise blade moment (Nm) 1000

PT
750

500

RI
250

SC
0
0 100 200 300 400 500 600
Time (s)

U
Figure 3: Flapwise bending moment simulated in FAST for IEC Class III inlet wind operating conditions.
AN
218 2.3. 5 kW small wind turbine operating measurements
219 A 5 kW Aerogenesis wind turbine has been installed on campus at the University of Newcastle, Australia,
220 and has been the subject of ongoing research work. Various instrumentation has been installed on this tur-
221 bine to facilitate experimental field work, with a detailed description of the sensors, instrumentation, and
M

222 calibration process found in [27, 30]. Relevant instrumentation includes; a cup anemometer and wind vane
223 for measuring wind speed and direction, and seven uniaxial strain gauges spaced axially along the blade for
224 determining blade aerodynamic and structural response. Measured blade strains are converted to a bending
D

225 moment at the blade root in the flapwise direction. Data is recorded at a rate of 500 Hz via the nosecone
226 data acquisition system (DAS) and written to an SD card for later retrieval. Systems are also in place to
227 measure turbine rotor speed and generator output power, which are located in a site hut approximately 20
TE

228 m from the turbine.


229

230 A series of measurement campaigns were undertaken with several hours of operational data recorded
231 (summary statistics are presented in table 1). A comprehensive range of wind speeds were captured from
EP

232 below cut-in (<3.5 ms−1 ) to above cut-out (>20 ms−1 ), allowing for analysis across the entire design oper-
233 ating range of the Aerogenesis wind turbine. While full details of experimental results can be found in [27],
234 figure 4 illustrates the complex nature of the blade loading signal. It can be observed that gyroscopic inertial
235 effects due to high, uncontrolled yaw rates contributed significantly to fatigue loadings as per [12, 22, 23].
C

236 It is clear that fatigue loading of small wind turbine blades is much more complex than simple sinusoidal
237 variations of the blade load at rated conditions.
AC

238

239 In order to understand the structural fatigue loading of the blade, rainflow counting was undertaken using
240 the measured flapwise moment data set. Outputs include the bending moment range, Mrange = 2 × Mamp ,
241 mean value of the cycle, Mmean = (Mmax + Mmin )/2, and the cycle ratio (i.e. equivalent to stress ratio in
242 the absence of material properties, R = Mmin /Mmax ). Measured fatigue cycles were found to occur at a
243 rate of 8.26 Hz, while this is less than the rate predicted by the SLM, it falls within the 2P frequency of the
244 variable speed rotor.
245

246 A histogram was output whereby the number of cycles for the rainflow counted data set was binned for
247 cycle range and cycle mean on a 50 × 50 grid in figure 5. The number of cycles were plotted on a log scale for
248 clarity. A visual inspection shows a large number of low range damage cycles, with a high variation in cycle

8
ACCEPTED MANUSCRIPT

(a) Flapwise moment (Nm)


800
Large amplitude
gyroscopic effects
600

PT
400

200
0 1 2 3 4

RI
60 20

Wind speed (ms -1 )


(b) Yaw rate ( °/s)

40 Yaw rate

SC
20 Wind speed 15
0
-20 10
-40

U
-60 5
0 1 2 3 4
AN
Time (sec)

Figure 4: Measurements of high amplitude blade loading due to gyroscopic induced yaw rate effects.
M

249 mean. This is likely to indicate the fact that many high–order cycles (harmonics of the blade rotational
250 response) occur during operation and accumulate rapidly due to the high operational frequency. Despite
D

251 the fact that these are numerous, in reality they are likely to be nothing more than minor aberrations or
252 vibrations in blade response. Also evident in the histogram are a much smaller number of high–range cycles,
with a lower variation in mean moment. These are likely due to transient yaw error and yaw rated effects
TE

253

254 such gyroscopic loading, that while occur much less frequently, exhibit much higher damage cycle ranges.
255 This measured fatigue spectra is appreciably different than that deduced by the SLM in figure 2. These
256 measurements have been in used to formulate an equivalent annual fatigue spectrum to estimate fatigue life.
257 This spectrum is presented in Appendix A.
C EP
AC

9
ACCEPTED MANUSCRIPT

104

PT
103
N cycles (log)

RI
102

SC
101

100
0

U
200 800

Be 400 600 )
AN
ndi
n nt (Nm
gm
om 400 me
ent
600 mo
ran ing
ge 800 200 b end
(Nm
) M ean
0
M

Figure 5: Histogram of rainflow counted measured blade fatigue damage cycles.


D

258 3. Results
TE

259 3.1. Comparison of fatigue spectra


260 In order to visualise and assess fatigue loading, the rainflow counted cycles from the measured and
261 simulated data sets were ranked. The number of cycles were normalised by the number of cycles in each
EP

262 data set to facilitate comparison between the measured and simulated data sets and to account for the fact
263 that different measurement periods were used. In figure 6, three plots were constructed for analysis using
264 damage cycles ranked in descending order from the simulated data, measured data, and SLM. Figure 6a is
265 produced of the ranked cycle mean value, figure 6b consists of the ranked cycle amplitude, and figure 6c is
C

266 the ranked cycle damage ratio. When assessing cycle mean and cycle amplitude for all cases it is evident
267 that the damage spectra consists of a small portion (< 10%) of high mean value cycles, a ‘plateau’ region
(10% < n < 80%) where no significant variations are observed, and a gradual reduction in value (> 80%). A
AC

268

269 visual trend is evident between the measured and simulated fatigue response, further demonstrating FASTs
270 utility in simulating fatigue loading. In this instance the mean cycle value and amplitude dictated by the
271 SLM was not exceeded by measurements - that is the SLM over-predicts loading for the entire blade life. The
272 SLM load can be considered more in-line with an ultimate load rather than a long-term fatigue load. With
273 the exception of < 2% of the simulated damage cycles, the vast majority of the damage cycle amplitudes
274 are found to be well below what is specified by the SLM. When comparing the damage cycle ratio, very few
275 cycles occur at a worse damage ratio than the SLM at R < 0.33, with the many approaching unity, R = 1.0,
276 (i.e. least damaging).
277

278 In summarising the distributions presented in figure 6; the fatigue spectra is dominated by a large number
279 of low-amplitude, low-damage ratio cycles and a small number of high amplitude, high-stress ratio cycles.
10
ACCEPTED MANUSCRIPT

PT
1500
SLM
Measured

RI
Simulated
(a) Cycle mean (Nm)

1000

SC
500

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

U
500
(b) Cycle amplitude (Nm)

AN
400

300

200
M

100

0
D

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

1
TE

0.75
(c) Cycle ratio

R=0.33
0.5
EP

0.25

0
C

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

N cycles (nomalised)
AC

Figure 6: Rainflow counted fatigue cycles for blade flapwise loading. Cycle mean (a), cycle amplitude (b), and cycle ratio (c),
ranked descending on a normalised scale for measured loading, FAST simulated loading, and the SLM.

11
ACCEPTED MANUSCRIPT

280 The simplified load model is more conservative in all cases, as measured and simulated cycles rarely exceeded
281 that specified by the standard. The constant-amplitude fatigue loading outlined in the standard does not
282 represent the distribution in fatigue cycle damage observed from measurements and predicted by aeroelastic
283 simulations.
284

3.2. Damage equivalent loads

PT
285

286 Formulation of damage equivalent loads is a common technique used to express a complex variable fatigue
287 spectra as a single simplified ‘equivalent’ load. In practice, this damage equivalent load (DEL) is a load
288 amplitude that when applied at 1 Hz for the lifetime operation of the turbine produces the same fatigue

RI
289 damage as the measured and rainflow counted spectra. Despite the limited ability of a DEL to reproduce
290 the effects of a complex spectra with variable amplitude loading, this method has been used in a number of
291 studies such as [26, 29, 31], as it provides a convenient way of comparing fatigue damage across a variety of
scenarios. A key assumption for DEL calculation is that the material fatigue life is represented by the ‘S-N’

SC
292

293 curve, whereby the number of cycles to failure, N , at a given load level, S (stress or strain), take the form
294 of a power-law relationship:

N = CS −m (12)

U
295 Where C, and m, are material parameters that are typically found experimentally (where m is also
named the Wöhler coefficient). IEC 61400.13: measurement of mechanical loads [32], provides Wöhler
AN
296

297 coefficients for the determination of DELs for a given fatigue spectra. For the case of composites (as used
298 in the Aerogenesis blade) this coefficient is given as m = 10. Equations 10 and 12 can be rearranged and
299 used to calculate this equivalent load, Req (MPa, Nm, N, etc.), as per equation 13. This equivalent load
is said to occur at a rate of 1 Hz for the turbine design life of 20 years. Using this coefficient, DELs were
M

300

301 determined from both the measured and simulated data, with the SLM equivalent DEL also calculated as
302 a baseline. The equivalent load is given by:
 m1
D

ΣRim ni

Req = (13)
neq
TE

303 Where ni is the number of cycles at load level Ri , and neq is given as 1 Hz cycles for 20 years (i.e. neq
304 = 630.72×106 cycles).
305

306 DELs were produced for measured results, aeroelastic simulation, and SLM and are presented in table
EP

307 2. When comparing the damage equivalent loads for all the load cases present, it is clearly evident that the
308 SLM significantly over-predicts all other DELs. An equivalent load amplitude of 603 Nm was found for the
309 SLM case, and is significantly higher than the measured data set which produces a DEL of 71 Nm. While
310 this is an over-prediction of 849%, it is worth noting that the DEL calculation only takes load amplitudes
C

311 into account and does not include mean load level and load ratio effects. As a secondary consideration,
312 a Wöhler coefficient equal to 10, may not necessarily be accurately representative of the GFRP material
AC

313 behaviour of the Aerogenesis blade.


314

315 3.3. Estimated fatigue life


316 Now that the fatigue damage spectra has been produced, the fatigue life of the blade can now be estimated
317 using Miner’s sum as per equation 10. In table 2 the blade fatigue life in years has been estimated using
318 Miner’s sum in conjunction with the SLM, the measured data set, and the simulated fatigue loading. The
319 material S-N curve at R = 0 was used to represent the most conservative case. These properties have been
320 obtained via coupon testing. Immediately obvious is the fact that the simplified load model predicts a fatigue
321 life of 0.09 years (which is equivalent to 34 days of operation). This is significantly over-conservative when
322 considering the Aerogenesis blade, which has been in safe (non-continuous) operation for approximately 5

12
ACCEPTED MANUSCRIPT

323 years without any obvious structural degradation or cracking. A point of concern is that this design fatigue
324 life may not be achievable for small wind developers, as producing a blade to withstand this loading for a
325 design life of 20 years could result in a significant over-design (i.e. increased shell thickness, large spar ele-
326 ments, high cost high-strength CFRP etc.) of the blade which could result in increased costs, manufacturing
327 complexity, and blade mass which has unfavourable effects for rotor inertia and turbine starting performance.
328

PT
Table 2: Comparison of calculated fatigue cycle rate, damage equivalent load, and estimated fatigue life.
Method Fatigue cycle rate DEL Fatigue life
(Hz) (Nm) (Years)

RI
SLM 10.67 603 0.09
Aeroelastic (FAST) 3.37 69 3.26
Measured 8.26 71 9.18

SC
329 When considering the measured damage spectrum, a life of 9.18 years is predicted (equal to 102 times
330 the SLM) and represents a significant gain in operational lifetime. While this lifetime is still shorter than the
331 intended design life of 20 years, in this calculation, the blade is said to experience this loading continuously

U
332 throughout the entire life and does not consider the availability of the wind resource, periods of extended
333 shut down for maintenance, or other plant availability issues which may reduce the number of damage cycles
experienced by the blade. If an optimistic capacity factor of 50% is applied, the fatigue life will approach
AN
334

335 the intended design life of 20 years.


336

337 The fatigue life predicted from the aeroelastic simulations using the IEC design wind case is equal to
338 3.26 years (36 times the SLM fatigue life). This simulated life gives a good indication of ‘mean’ fatigue
M

339 damage experienced by the blade during normal operation for a turbine of this class. From these results we
340 conclude that the simplified load model is the most conservative for fatigue loading, followed by, aeroelastic
341 simulations, with the actual operational measured data being least conservative in this case. Somewhat
D

342 concerningly, all methods predict a life of less than the nominal 20 year design period.
343
TE

344 4. Conclusions

345 In this study the fatigue loading and predicted fatigue life of a 2.5 m small wind turbine blade has
been compared using the IEC simplified load model, measured operational data, and that from aeroelastic
EP

346

347 simulations. To the best of the authors’ knowledge the IEC SLM blade fatigue case has not previously been
348 compared to experimental measurements and aeroelastic simulations. Fatigue loading of this free-yaw small
349 wind turbine consists of few high-mean, high-stress ratio fatigue cycles, in combination with many low-mean,
low-stress ratio cycles. These large cycles are due to unsteady operation and gyroscopic loads, whereby min-
C

350

351 imising erratic yaw behaviour and reducing maximum yaw rates would see a favourable reduction in the
mean stresses of these loads. The majority of the measured fatigue spectra consists of many low mean-stress
AC

352

353 and low-stress ratio cycles, likely due to effects of site turbulence, unsteady rotor aerodynamic loading, and
354 operation at moderate yaw errors. The constant damage amplitude predicted by the standard was well in
355 excess of measurements and simulations and is not physically representative of the variable damage spectra
356 that is experienced by the blade while in service.
357

358 Damage cycles were found to occur at a rate of 10.67 Hz for the SLM, 8.26 Hz for the measured data,
359 and 3.37 Hz when considering the aeroelastic simulations. A DEL amplitude of 603 Nm was found for the
360 simplified load mode – which is an over-prediction of 849% when compared to the measured DEL of 71 Nm
361 and 874% with respect to the simulated value of 69 Nm. Finally, a fatigue life of 0.09 years was predicted
362 by the SLM, well short of the 20 year design life. Fatigue life of the measured and simulated data set were
363 found at 9.18 years and 3.26 years respectively. While still short of the 20 year design life, these calculations
13
ACCEPTED MANUSCRIPT

364 do not take into account turbine availability and capacity factors.
365

366 Care should be taken when assessing predicted fatigue life via the SLM so as not to over-design compo-
367 nents. Prudent small wind manufacturers may opt to determine design loads from aeroelastic calculations
368 instead of undertaking difficult experimental measurement campaigns or risk over-designing components
369 when using the SLM. Despite its shortcomings, the authors’ are sympathetic with the intent of the SLM in
that it is formulated to be both conservative and straightforward to implement. Future work should focus

PT
370

371 on developing the SLM fatigue load case to instead incorporate the many low-amplitude damage cycles,
372 and few high-amplitude damage cycles that are observed in field measurements and aeroelastic simulations.
373 This would be more indicative of what is experienced by a blade during operation while still retaining the
simplicity and intent of the SLM. The ability to derate the turbine design life by the expected capacity

RI
374

375 factor may also improve SLM fatigue life calculations.

SC
376 Acknowledgements

377 The work in this paper was supported by an Australian Research Council Discovery Project grant
378 (DP110103938). S.P. Evans and D.R. Bradney were supported by an Australian Postgraduate Award for
379 the duration of their PhD candidature. The authors thank M. Roberts and M. Gibbs for their assistance

U
380 with turbine instrumentation and data acquisition.
AN
381 Appendix A. Annual blade fatigue spectra

382 To develop this final spectra in table A.3, the measured data set was rainflow counted and binned with
383 respect to the maximum blade root flapwise cyclical stress (σmax ) in stress increments of 10% ranging from
M

384 zero stress to the maximum recorded stress cycle. These damage cycles were also binned in stress ratios (R)
385 varying from 0 to 1.0 at increments of 0.1. The final spectra was then normalised and multiplied by the
386 number of annual cycles to produce a final equivalent annual fatigue damage spectra. It is envisaged that
D

387 persons wishing to use this spectra for testing purposes of other turbines of this class may normalise σmax
388 with respect to the maximum determined cyclical stress. This annual spectra can naturally be multiplied by
389 the desired number of operational years (i.e. 5, 10, 20, etc.), or derated by applying an appropriate capacity
TE

390 factor (30%, 40%, 50%, etc.) to account for reduced operation due to environmental wind resource factors
391 or extended shut-down for maintenance.

Table A.3: Annual fatigue damage spectra from measured data. Note: cycles displayed as n×103 for table clarity.
EP

σmax Stress ratio (R)


(MPa) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
3.70 103 18 15 41 35 65 53 59 74 0
C

7.40 26 26 59 76 141 335 597 847 2,629 2,515


11.10 12 12 29 159 562 1,403 3,535 6,318 9,206 20,373
AC

14.80 3 18 62 159 491 1,421 4,635 11,085 16,823 36,670


18.50 12 21 53 182 544 1,197 3,297 9,179 18,162 34,970
22.20 6 29 71 182 418 976 2,665 6,368 12,241 23,241
25.90 18 35 79 144 321 859 1,803 3,776 5,903 8,676
29.60 9 47 59 109 179 371 844 1,135 691 544
33.30 12 38 18 26 41 91 82 15 3 18
37.00 12 0 3 6 3 0 0 0 0 0

14
ACCEPTED MANUSCRIPT

392 References
393 [1] T. Burton, D. Sharpe, N. Jenkins, E. Bossanyi, Wind energy handbook, John Wiley & Sons, 2001.
394 [2] H. Sutherland, On the fatigue analysis of wind turbines, Tech. rep., Sandia National Labs., Albuquerque, NM (US); Sandia
395 National Labs., Livermore, CA (US) (1999).
396 [3] H. Sutherland, A summary of the fatigue properties of wind turbine materials, Wind Energy 3 (1) (2000) 1–34.
397 [4] R. Nijssen, Fatigue life prediction and strength degradation of wind turbine rotor blade composites, TU Delft, Delft
398 University of Technology, 2006.

PT
399 [5] P. Clausen, F. Reynal, D. Wood, Design, manufacture and testing of small wind turbine blades, in: Advances in Wind
400 Turbine Blade Design and Materials, Woodhead Publishing Series in Energy, Woodhead Publishing, 2013, pp. 413 – 431.
401 [6] A. Medd, D. H. Wood, Small wind turbine performance assessment for canada, in: L. Battisti, M. Ricci (Eds.), Wind
402 Energy Exploitation in Urban Environment, Springer International Publishing, Cham, 2018, pp. 155–163.
403 [7] M. Noda, R. Flay, A simulation model for wind turbine blade fatigue loads, Journal of Wind Engineering and Industrial

RI
404 Aerodynamics 83 (1) (1999) 527–540.
405 [8] S. Downing, D. Socie, Simple rainflow counting algorithms, International Journal of Fatigue 4 (1) (1982) 31–40.
406 [9] N. Post, S. Case, J. Lesko, Modeling the variable amplitude fatigue of composite materials: A review and evaluation of
407 the state of the art for spectrum loading, International Journal of Fatigue 30 (12) (2008) 2064–2086.

SC
408 [10] P. Malhotra, R. Hyers, J. Manwell, J. McGowan, A review and design study of blade testing systems for utility-scale wind
409 turbines, Renewable and Sustainable Energy Reviews 16 (1) (2012) 284–292.
410 [11] IEC 61400.2-2013 Wind turbines - design requirements for small wind turbines (2013).
411 [12] A. Wright, D. Wood, Yaw rate, rotor speed and gyroscopic loads on a small horizontal axis wind turbine, Wind Engineering
412 31 (3) (2007) 197–209.
[13] D. Wood, Small wind turbines: analysis, design, and application, Springer, 2011.

U
413
414 [14] S. Evans, P. Clausen, Modelling of turbulent wind flow using the embedded Markov chain method, Renewable Energy 81
415 (2015) 671–679.
AN
416 [15] S. Evans, A. KC, D. Bradney, T. Urmee, J. Whale, P. Clausen, The suitability of the IEC 61400-2 wind model for small wind
417 turbines operating in the built environment, Renew. Energy Environ. Sustain. 2 (2017) 31. doi:10.1051/rees/2017022.
418 [16] A. B. Tabrizi, J. Whale, T. Lyons, T. Urmee, J. Peinke, Modelling the structural loading of a small wind turbine at a
419 highly turbulent site via modifications to the kaimal turbulence spectra, Renewable Energy 105 (2017) 288–300.
420 [17] Encraft, Warwick wind trials final report[Online; accessed 19-July-2008].
URL http://www.microwindturbine.be/Rapportering_files/Warwick+Wind+Trials+Final+Report+%281%29.pdf
M

421
422 [18] Enhar, Safety and reliability of micro wind turbine installations.
423 URL http://www.enhar.com.au/userfiles/file/Press%20releases/Enhar%20bulletin%20-%20Safety%20and%
424 20reliability%20of%20micro%20and%20urban%20wind%20turbine%20installations%20in%20Australia.pdf
425 [19] J. Epaarachchi, The development and testing of a new fatigue life procedure for small composite wind turbine blades
D

426 incorporating new empirical fatigue life prediction and damage accumulation models for glass fibre reinforced plastics,
427 Ph.D. thesis, School of Engineering, The University of Newcastle, Australia (2002).
428 [20] A. Tenhave, WISPER and WISPERX: Final definition of two standardised fatigue loading sequences for wind turbine
TE

429 blades, NASA STI/Recon Technical Report N 94 (1992) 30872.


430 [21] J. Epaarachchi, P. Clausen, The development of a fatigue loading spectrum for small wind turbine blades, Journal of Wind
431 Engineering and Industrial Aerodynamics 94 (4) (2006) 207–223.
432 [22] S. Wilson, P. Clausen, Aspects of the dynamic response of a small wind turbine blade in highly turbulent flow: part 1
433 measured blade response, Wind Engineering 31 (1) (2007) 1–16.
EP

434 [23] S. Wilson, P. Clausen, Aspects of the dynamic response of a small wind turbine blade in highly turbulent flow: part 2
435 predicted blade response, Wind Engineering 31 (4) (2007) 217–231.
436 [24] Y. Jang, C. Choi, J. Lee, K. Kang, Development of fatigue life prediction method and effect of 10-minute mean wind
437 speed distribution on fatigue life of small wind turbine composite blade, Renewable Energy 79 (2015) 187 – 198, selected
438 Papers on Renewable Energy: AFORE 2013.
C

439 [25] D. Wood, Using the IEC simple load model for small wind turbines, Wind Engineering 33 (2) (2009) 139–154.
440 [26] S. Dana, R.Damiani, J. van Dam, Validation of simplified load equations through loads measurement and modeling of
441 a small horizontal-axis wind turbine tower, Tech. rep., NREL (National Renewable Energy Laboratory), Golden, CO
AC

442 (United States) (2015).


443 [27] S. Evans, Aeroelastic measurements, simulations, and fatigue predictions for small wind turbines operating in highly
444 turbulent flow, Ph.D. thesis, School of Engineering, The University of Newcastle, Australia (2017).
445 [28] ASTM E1049 - 85(2011)e1 standard practices for cycle counting in fatigue analysis.
446 [29] N. Dimitrov, A. Natarajan, J. Mann, Effects of normal and extreme turbulence spectral parameters on wind turbine loads,
447 Renewable Energy 101 (2017) 1180–1193.
448 [30] D. Bradney, Measured and predicted performance of a small wind turbine operating in unsteady flow, Ph.D. thesis, School
449 of Engineering, The University of Newcastle, Australia (2017).
450 [31] S. Lee, M. Churchfield, P. Moriarty, J. Jonkman, J. Michalakes, A numerical study of atmospheric and wake turbulence
451 impacts on wind turbine fatigue loadings, Journal of Solar Energy Engineering 135 (3) (2013) 031001.
452 [32] IEC 61400.13-2015 Wind turbines - measurement of mechanical loads (2015).

15
ACCEPTED MANUSCRIPT
Highlights:
 The IEC 61400.2 fatigue load equations are assessed for small wind turbine blades
 These simplistic equations are compared to measurements and aeroelastic simulations
 IEC fatigue life was calculated to be 102 times less than field measurements
 Use of these equations may impact manufacturability, blade weight, and cost

PT
RI
U SC
AN
M
D
TE
C EP
AC

You might also like