Professional Documents
Culture Documents
net/publication/317510081
CITATIONS READS
9 208
2 authors:
Some of the authors of this publication are also working on these related projects:
Florida Sea Grant project- Promoting preventive mitigations of buildings against hurricanes through enhanced risk-assessment and decision making View project
All content following this page was uploaded by Gholamreza Amirinia on 16 October 2017.
6 Hurricanes
10 Abstract
11 The objective of this paper is to investigate the effects of newly observed hurricane turbulence
12 models on offshore wind turbines by considering unsteady aerodynamic forces on the tower and
13 wind-wave-soil-structure interaction. The specific goals were analyzing the tower and blade
14 structural buffeting responses, the low cycle fatigue during different hurricane categories, and
15 extreme value of the short term responses. To achieve these goals, first, the recent observations
16 on hurricane turbulence models were discussed. Then a new formulation for addressing unsteady
17 wind forces on the tower was introduced and NREL-FAST package was modified with new
18 formulation. Results showed that recently observed turbulence models resulted in larger
19 structural responses and low cycle fatigue damage than existing models. In addition, extreme
20 value analysis of the short term results showed that the IEC 61400-3 recommendation for wind
21 turbine class I was conservative for designing the tower for wind turbine class S subjected to
22 hurricane; however, for designing the blade, IEC 61400-3 recommendations for class I
Corresponding author. Tel.: +1-850-410-6505; E-mail address: gamirinia@fsu.edu
1
24 Keywords: Offshore wind turbine, hurricane, structural responses, low cycle fatigue, IEC 61400-
25 3.
26 1. Introduction
27 Investigation of hurricanes during last decade showed that hurricanes have different turbulence
28 characteristics from regular high winds (Caracoglia and Jones, 2009; Jung and Masters, 2013; Li
29 et al., 2012; Schroeder and Smith, 2003; Yu et al., 2008). These differences in the turbulence
30 energy models affect the structures by changing the buffeting response characteristics or causing
31 low cycle fatigue. In this regard, special structures such as offshore wind turbines in hurricane
32 prone regions need to be studied more for safety and economical aspects (Amirinia and Jung,
33 2017, 2016; Amirinia et al., 2015; Gong and Chen, 2015).
34 Observations and analysis of hurricane surface winds revealed that turbulence spectrum of
35 hurricane winds differs from that of non-hurricane high winds (Balderrama et al., 2011;
36 Caracoglia and Jones, 2009; Gong and Chen, 2015; Jung and Masters, 2013; Li et al., 2012;
37 Schroeder and Smith, 2003; Yu et al., 2008). Li et al. (2012) and Caracoglia and Jones (2009)
38 showed that in the hurricane, the higher turbulence frequencies have higher level of energy;
39 however, Schroeder and Smith (2003), Yu et al. (2008), and Jung and Masters (2013) showed
40 that hurricane spectrum has higher level of energy in low frequencies. Different turbulence
41 energy models affect structures differently, while the mean wind speed and turbulence intensity
42 are identical between models. In this regard, mean responses of structures subjected to regular
43 high winds and hurricane winds are comparable, whereas, the buffeting responses are different.
44 Because of these differences in buffeting responses, conditions such as structural integrity and
45 low cycle fatigue subjected to different hurricane turbulence models should be investigated.
2
46 Various studies in recent years have been conducted to investigate hurricane effects on wind
47 turbines. Han et al. (2014) reviewed the characteristics of tropical cyclones and their probable
48 effects on the wind turbines. Guo et al. (2014) compared existing wind shear and turbulence
49 spectrum models in standards for wind turbine analysis during hurricane and showed that
50 different models result in different responses. Kim and Manuel (2014) used conventional Kaimal
51 et al. (1972) turbulence model and a coupled wind-wave approach for analyzing offshore wind
52 turbines. They explored hurricane induced loads on offshore wind turbines with consideration of
53 nacelle yaw and blade pitch control. After that, Kim et al. (2016) studied effects of hurricane Ike
54 (2008) and hurricane Sandy (2008) on an offshore wind turbine. They considered the
55 conventional Kaimal et al. (1972) wind spectrum and addressed effects of yaw misalignment and
56 blade azimuthal angle on the responses. All the mentioned studies addressed an important issue
57 in the analysis of offshore wind turbines subjected to hurricanes, however, all of them used
58 existing turbulence models for simulating the hurricane. To this end, there is a need for studying
59 the effects of recent observations of hurricane turbulence models on the wind turbines.
60 The duration of the hurricane compared to structural design life is short. Hence, large turbulent
61 forces in a short period of time are exerted on the structure. In this case an accurate low cycle
62 fatigue analysis is necessary for determining the structural integrity during hurricane. By
63 referring to the differences in turbulence spectrum models, the buffeting responses of the
64 structures subjected to the various hurricane turbulence models are important in terms of low
65 cycle fatigue analysis. Li et al. (2014) studied wind buffeting forces on fatigue analysis of the
66 wind turbine foundation. Tibaldi et al. (2015) investigated the operating wind turbine fatigue
67 based on a linear model. Lee et al. (2013) applied a numerical method to study the wake
68 turbulence impacts on wind turbine fatigue. By reviewing previous studies and new findings
3
69 about hurricane turbulence models, the low cycle fatigue analysis of the offshore wind turbines
70 should be studies carefully to investigate the effect of newly observed and presented models.
71 IEC 61400-1 (2005) and IEC 61400-3 (2009) have recommendations for different wind turbine
72 classes and load cases; however, for special events such as hurricane, they introduced wind
73 turbine class S which the design variables should be defined by the designer. In order to compare
74 the short term responses of the wind turbine subjected to hurricane with IEC 61400-1 (2005) and
75 IEC 61400-3 (2009) recommendations, an extreme value analysis is necessary. Several studies
76 has been conducted on extreme value analysis which considered various methods and approaches
77 (Kim and Manuel, 2013; Kim and Manuel, 2014; Kim et al., 2016). By considering the
78 importance of new hurricane turbulence models on buffeting responses, an accurate extreme
79 value analysis is necessary for comparing the short term results with IEC 61400-3 (2009)
80 recommendations. An accurate extreme value analysis and comparison with existing
81 recommendations also assist to consider important issues in designing wind turbines class S.
82 In this paper, the main objectives were investigating the effects of recently observed hurricane
83 models on structural responses and low cycle fatigue of offshore wind turbines. In addition, to
84 compare the short term analysis with existing IEC 61400-3 (2009) recommendations and
85 proposing extra consideration for special conditions such as hurricane, an extreme value analysis
86 was carried out. For these purposes, first, the recent observations on hurricane turbulence models
87 were discussed. Next, the buffeting wind loads on the wind turbine structure were mentioned and
88 a new formulation for addressing unsteady wind forces on the tower was introduced (Amirinia
89 and Jung, 2016). This new formulation was later used to modify NREL-FAST (Jonkman and
90 Buhl Jr, 2005) for analysis. At next step, according to recent findings about hurricane winds,
91 hurricane wind and wave fields were simulated based on the Saffir-Simpson hurricane wind
4
92 scale (Simpson and Saffir, 1974). Then, to investigate the effects of various hurricane turbulence
93 models on the wind turbine structures, the modified NREL-FAST (Jonkman and Buhl Jr, 2005)
94 in previous sections was used to analyze structure-wind-wave-soil interaction of the NREL-5
95 MW wind turbine (Jonkman et al., 2009). Finally, the structural responses and low cycle fatigue
96 analysis were presented and discussed. At the end, an extreme value analysis was carried out on
97 the short term results and extreme responses were compared with IEC 61400-3 (2009)
98 recommendations for 50-years return period load case for extra design consideration.
100 The wind turbulence spectrum represents the energy distribution in turbulent wind (Li et al.,
101 2012). The total energy of turbulent flows can be expressed as superposition of eddies (Yu et al.,
102 2008). Big eddies which represent small wave numbers or low frequencies, supply the most
103 energy content of turbulent flow; whereas, small eddies with high wave numbers in high
104 frequencies dissipate the gained energy (Kaimal et al., 1972; Tieleman, 1995; Yu et al., 2008).
105 This gain and dissipation are connected to each other with an inertial subrange where the
106 influence of viscosity is small. According to the equilibrium, the gain and dissipation of energy
/ / (1)
108 Where represents energy dissipation, is wave number, and is a universal constant. Earlier
109 studies investigated wind turbulence spectrums for non-hurricane winds (Davenport, 1961;
110 Kaimal et al., 1972; Tieleman, 1995; Von Karman, 1948). Kaimal et al. (1972), based on series
111 of experiments, showed that all spectra reduce to a limited family of curves which fit a single
5
112 universal curve in inertial subrange but spread out in low frequencies. They proposed a formula
21.6
/
(2)
1 33
114 where / represents the normalized frequency, is the frequency, is the spectral
115 density of the longitudinal velocity fluctuation at height , and is the standard deviation of
116 longitudinal velocity fluctuation. Hurricane field data observations during the last decade
117 revealed that turbulence spectrum of the hurricane boundary layer winds is different from those
118 of non-hurricane high-winds. Schroeder and Smith (2003), Yu et al. (2008), and Jung and
119 Masters (2013) showed that compared to non-hurricane spectral models, hurricane spectrums had
120 higher energy content in low frequencies. The formula derived for hurricane wind turbulence
121 spectrum (Yu et al., 2008) with high amount of energy in low frequencies was presented as:
1
(3)
∗ ∗
122 where / is the turbulence ratio, represents the friction velocity, and and are
125 On the other hand, Li et al. (2012) and Caracoglia and Jones (2009) based on series of
126 observations, presented an opposite results that higher frequencies contained larger amount of
6
127 energy rather than low frequencies. Li et al. (2012) provided spectrum models for their
16.66
/
(4)
1.72 237.24
129 Table 2 summarizes the spectrum models presented by mentioned researchers and used in this
130 paper. In addition, Figure 1 shows the difference between recent observed models and the
133
134
135 Figure 1. Comparison of hurricane spectrum models (over sea) and Kaimal spectrum
7
136 Some relevant discussion can be found in the meteorological and engineering literature to
137 understand the large variation in hurricane spectra. Wind spectra can be affected by various
138 parameters such as upstream roughness (Yu et al., 2008). It is been observed that spectral values
139 water surface and sea are higher than those of onshore regions. Also for open exposures, the
140 spectral energy contents in lower observatory anemometers are more than those in higher
141 observatory devices (Yu et al., 2008). Moreover, the low frequency range of spectrum depends
142 on the atmospheric stability (Kaimal et al., 1972; Tieleman, 1995; Yu et al., 2008), hence for
143 increased instability, the spectral peak shifts to the lower frequencies. In addition, dissipation of
144 energy commonly occurs in the high frequency range (Fiedler and Panofsky, 1970; Singer et al.,
145 1968; Teunissen, 1970). The anemometers used for observations can also affect the
146 measurements where their effects can be corrected to make the different measurements
147 comparable (Jung and Masters, 2013; Schroeder and Smith, 2003; Yu et al., 2008).
149 The effect of unsteady forces on the operating wind turbine has been investigated before
150 (Sebastian and Lackner, 2013); however, generally, the unsteady forces for parked wind turbine
151 in high-winds is neglected and only the quasi-steady method was used (Jonkman and Buhl Jr,
152 2005). Previous studies showed that in parked condition the tower play an important role in total
153 structural responses (Amirinia and Jung, 2016) and unsteady analysis of the tower in parked
154 condition resulted in more accurate and realistic responses. Figure 2 shows a schematic view of
155 aerodynamic forces on a wind turbine tower. Formulation of wind loading on the wind turbine
8
̅ ,
z, ̅ , ̅ (5)
,
,t / / (6)
158 are self-excited forces, and are longitudinal and lateral fluctuating parts of the wind, and
159 and are structural position and velocity vectors. In eq. 5 and eq. 6 it is assumed that the main
160 flow direction is perpendicular to the rotor plane and wind field is homogenous (Strømmen,
161 2010).
162
163 Figure 2. Aerodynamic forces on the wind turbine tower
164 The second term in right hand side of eq. 5, denotes wind buffeting force. In order to address
165 unsteady forces, buffeting force components can be revised by considering aerodynamic
9
166 admittance function. For instance, along-wind component of buffeting forces can be expressed as
167 eq. 7:
z, 2 ̅ ̅ (7)
2
168 where and are the frequency dependent aerodynamic admittance functions in
170 in which is the angle of attack. In the time domain, the along wind buffeting forces (for the
171 unit length of the structure) due to the wind fluctuations and can be expressed as the
z z, (8)
2
173 where and are the aerodynamic impulse functions in along wind and across wind
174 directions respectively. The aerodynamic admittance function can be addressed in the time
175 domain, by comparing Fourier transform of eq. 8 and the Fourier transform of eq. 7 (Chen et al.,
176 2000; Clobes and Peil, 2011). By addressing aerodynamic admittance function in terms of
177 rational functions, the frequency dependent unsteady buffeting forces can be computed in time
178 domain. The time domain formulation for unsteady drag force can be expressed as:
1
, , , , (9)
2
, , , , (10)
10
179 where is drag force as a function of time, and , and , represent rational function (RF)
180 coefficients. The time dependent functions , are new variables to express aerodynamic
181 phase lag and should be defined in such a way that satisfies eq. 10. In eq. 10, is scalar
182 representation of wind fluctuation introduced in eq. 6 which represents the longitudinal
184 The aerodynamic admittance function can be presented in terms of rational function (RF) as:
,
, (11)
,
185 where , and , are frequency independent coefficients and represents the level of
186 accuracy. In the current study, in the eq. 11, by using 1 (2 terms RF) and according to
187 experimental tests by Davenport (Davenport, 1977) and previous studies (Clobes and Peil, 2011)
188 for along-wind drag force responses, the coefficients were chosen as , , , 1
189 which according to Figure 3, represents more conservative condition. The NREL-FAST model
190 was later modified based on the presented procedure to analyze unsteady forces on tower in
191 parked condition. More details about the model validation and verification can be found in
11
2
193
194 Figure 3. Aerodynamic admittance function
195
197 To investigate the contribution of blades and tower under hurricane condition, the NREL 5MW
198 offshore monopile wind turbine (Jonkman et al., 2009) was selected for analysis. In order to
199 consider randomness of the wind field, 60 wind field time series for each turbulence model
200 presented in section 2 (each one 600s) were generated by NREL-TurbSim (Jonkman, 2009) with
201 frequency of 10Hz. The mean wind speeds at 10 m height were equal to 30, 40, 50,60, 70, and 80
202 m/s (each speed 10 time series) in order to cover five Siffir-Simpson hurricane categories
12
205 ASCE 7-10 vertical profile for exposure D was used for height extrapolation as eq. 11:
(11)
10
206 where is the mean hourly wind speed at 10 m height, is the height and and are
207 constants depending on the terrain. For flat terrain or sea surface and are 0.11 and 0.8
208 respectively. In addition, In order to convert hourly mean wind speed to maximum speed
209 averaged over 10-minutes, a coefficient of 1.07 was used (Durst, 1960). Turbulence intensity
210 was equal to 13% at 10 m height which corresponds with observed data by Li et al. (2012).
211 Coherence function between fluctuation parts can be expressed as eq. 12 which considers famous
(12)
213 In eq. 12, and are spatial coherence decrement and offset parameter, respectively, is the
214 distance between point and , is frequency, is coefficient for spatial coherence function and
215 and are the mean height and wind speed of points and , respectively. In this study, eq.
216 12 was used to address the spatial coherence of wind speed fluctuation where the authors
217 assumed 0 and 7.7 which represents Davenport (1961) general formula.
218 Since waves are commonly induced by wind, wind and wave climate are correlated. The
219 irregular wave climate can be represented by the significant wave height, and the spectral
13
220 peak wave period, . The JONSWAP (Joint North Sea Wave Observation Project) spectrum
221 was used to represent wind induced wave in the analysis as:
α 5 .
exp (13)
2π 4
223 generalized Phillips’ constant, is spectral width parameter, and is peak-shape parameter. By
224 considering 20 m water depth ( in the offshore wind farm and assuming 6 hours hurricane and
225 duration-limited wave field, the wind and wave field characteristics are summarized in Table 4
226 (U.S. Army Coastal Engineering Research Center, 1977). The wave length, /2 , for
227 each peak wave period, , ranges from 230 m to 613 m which represents shallow water in the
228 depth of 20 m ( / 0.5). Hence the maximum wave height will be 0.9 times of the water
229 depth (Sorensen, 2005) and the maximum wave height for the considered depth was limited to 18
230 m. In addition, the peak shape parameter followed the IEC 61400-3 (2009) annex B. The
231 nonlinear wave forces due to large wind induced wave heights were computed by the HydroDyn
232 package (Jonkman et al., 2014) through a hybrid model consist of potential flow theory and
14
235 The interaction between monopile and soil, influence the responses of offshore wind turbine. In
236 order to address the effect of soil-structure interaction, the non-linear and depth dependent p-y
237 soil model presented by Jonkman and Musial (2010) which considers sand under cyclic loading
238 condition was used. The p-y model depended on the effective sand weight, ′, angle of internal
239 friction, , pile diameter, and soil depth (Jonkman and Musial, 2010). The soil was chosen
240 layered as shown in Figure 4 with increasing internal friction and density with depth which
241 increased the soil-pile interaction participation in the dynamic analysis. A distributed spring
242 model was used for monopile flexible foundation where the stiffness constants were computed
244
245 Figure 4. Soil layer profile and simplified distributed springs models
247 For the analysis of the parked wind turbine, the following modes were considered: first and
248 second fore-aft (FA) of the tower, first side to side (SS) of the tower, first and second flapwise
15
249 (flap) of the blade, first edgewise (edge) of the blade as well as platform (monopile-soil
250 connection) six degrees of freedom in surge, sway, heave, roll, pitch and yaw. In this study, only
251 along-wind responses of the structure was focused since the spectrum Model B, only provided
252 turbulence spectrum model for longitudinal fluctuation components. In this case, since the
253 analysis purpose was along-wind responses, the yaw degree of freedom was fixed. During high-
254 wind conditions, wind turbine was parked where the blade pitch angles were set to 90°, blades
255 were put on feathered condition, and High Speed Shaft (HSS) brake was applied.
256 After applying pre-requisite condition, the structural analysis was carried out with 0.01s time
257 steps. After the simulations, in order to prevent capturing the model start up inaccuracy, the first
258 30 seconds of the simulations were omitted (Jonkman and Buhl Jr, 2005) for post-processing.
259 In post-processing of the analysis, tower and blade structural responses to the various turbulence
260 models, low-cycle fatigue analysis of the tower and blade were carried out. At the end, in order
261 to compare the results with IEC 61400-3 (2009) recommendations, an extreme value analysis
262 was carried out for long-term extrapolation of the loads and responses.
264 Since the mean wind speed and turbulence intensities in all turbulence models were identical, the
265 mean responses of all structural parts were equal; however, the buffeting responses because of
266 various levels of energy in different turbulence frequencies were different among the models.
267 The correlated wind-wave field affects the structural responses where the increase rate of wind
268 speed is not equal to increase rate of wave height and peak wave period. By increasing the wind
269 speed, the wave evolution rate gets larger in the deep water; however, sensing the bed effects
270 decreases the wave evolution rate. This issue was presented in Table 4 where by increasing the
16
271 wind speed from 30 m/s to 40 m/s and 50 m/s, the wave height increased by 3.8 m and 4.4
272 respectively, whereas by increasing the wind speed from 50 m/s to 60 m/s, the wave field start to
273 sense the bed effects and wave evolution rate decreased to 1.4 m. This trend in wind and wave
275 For all the response time series, the mean response and the mean of maximum responses were
276 computed. The comparison of mean responses with previous literature (Rose et al., 2012)
277 showed comparable values. Also the wind speeds were 10-min averaged which satisfied the 1-
278 min sustained wind speed in Saffir-Simpson categories mentioned in Table 3. Figure 5a shows
279 the mean of maximum buffeting base moments in various mean wind speeds where the slope of
280 the line increased from 30 m/s to 50 m/s, but after that by decreasing the wave evolution rate, the
281 slope decreased. The similar wave field for all cases and various turbulence models, resulted in
282 differences in base moment responses. The unsteady analysis of the Model A and Model B,
283 resulted in average 7 % smaller and 1 % larger fore-aft base moment responses than
285 Since the blade responses were mostly involved with aerodynamic forces, the change in wave
286 height did not affect the blade responses. Figure 5b shows the blade root edge-wise moments
287 where the Model A and Model B resulted in 57% smaller and 9% larger responses than
17
(a) (b)
289 Figure 5. Mean of maximum (a) tower buffeting fore-aft base moments and (b) blade root
290 edge-wise buffeting moment for different wind speeds at 10 m height
291 Figure 6a presents the tower tip buffeting displacements subjected to three different turbulence
292 models. The tower tip displacements of all models were similar; however, Model A and Model B
293 resulted in average 6% smaller and 1% larger responses than Kaimal et al. (1972) model
294 respectively. Spectral analysis of the tower tip displacement in Figure 7a depicts that Model A,
295 in low frequency responses (background responses) has higher amount of energy compared to
296 other two counterpart spectrum models. This is consistent with characteristics of spectrum model
297 A which had higher amount of energy in low frequencies compared to other two spectrums. All
298 three spectrums have similar response energy around 1st tower fore-aft frequency; however, in
299 higher frequencies, spectrum Model A, has smaller amount of energy compared to other two
300 counterparts. In addition, spectral analysis of the spectrum Model B and conventional Kaimal et
301 al. (1972) model shows that both two spectrums have similar responses energy in all frequency
302 ranges. It was the reason that average differences in tower tip displacement between spectrum
18
(a) (b)
304 Figure 6. Mean of maximum (a) tower tip buffeting displacements and (b) blade tip
305 displacements for different wind speeds at 10 m height
306 Figure 6b shows the blade tip displacement resulted from three different models where the
307 spectrum Model A and B resulted in 62% larger and 7% smaller responses respectively
308 compared to Kaimal et al. (1972) model. The spectral analysis of the blade tip responses in
309 Figure 7b shows that the 1st blade edge-wise frequency which is located in relatively high
310 frequencies is the most important resonant part of the responses. On the other hand, spectrum
311 Model A has low amount of energy in high frequencies. Hence the level of response energy
312 resulted from spectrum Model A is smaller than other two spectrums; however, spectrum Model
313 A has slightly larger response energy in background part in low frequencies. The lower amount
314 of spectral response of Model A compared two other two models, especially in the 1st blade
315 edge-wise frequency range, resulted in large difference between maximum blade tip
19
(a) (b)
317 Figure 7. (a) Tower tip displacement spectrum and (b) blade tip displacement spectrum for
318 60 m/s at 10 m height
319 5.2. Low Cycle Fatigue
320 The duration of the hurricane compared to total life of the wind turbine is short; however, since
321 the loads exerted on the structure during hurricane are very large, the cycles of loading and low
322 cycle fatigue are important. The fatigue damage on the structure can be represented as Miner’s
(14)
324 where is the damage index, is load case number, is total number of load cases, is
325 number of load cycles for case , and represents number of load cycles to failure for case .
326 Studies conducted by Coffin Jr (1954) and Manson (1965) related the number of cycles to crack
327 initiation to the amplitude of plastic strain. Together with study carried out by Basquin (1910),
328 the well-known strain-life (ε-N) model was introduced as eq. 15:
, , 2 2 (15)
20
329 where the total strain amplitude, , , is divided into elastic, , , and plastic, , , components.
330 , represents fatigue strength coefficient, is the Young modulus, 2 is the number of cycles
331 to fatigue failure (crack initiation), is fatigue ductility coefficient, and and are fatigue
332 strength and ductility exponents respectively. The first and second terms in right hand side of the
333 eq. 15, represent the strain caused by elastic and plastic parts respectively which are shown in
334 Figure 8a. The Coffin-Manson model follows a continuum mechanics approach and assumes that
335 the material is homogeneous and isotropic. Hence it can represent the low cycle fatigue in the
336 wind turbine steel tower. By applying Basquin’s exponents for the elastic strain and Coffin’s
337 exponents for plastic strain parts of the eq. 15 (Ellyin, 2012), the eq. 15 simplifies to:
3.5 . . (16)
338 The low cycle fatigue also can affect the composite fiberglass materials (Carvelli et al., 2010;
339 Karahan et al., 2011) which are used in blade structure. Conventional test-load methods and
340 simplified load spectrum methods are the most common models for fatigue analysis of the blades
341 (Freebury and Musial, 2000). In simplified load spectrum method, the S-N curves and M-N
342 curves (applied moment vs. allowable cycles to failure) can be used as:
/ (17)
/ (18)
343 where and are moment and stress amplitude in one load cycle respectively, and are
344 ultimate moment and stress of the blade, is the allowable cycles to failure, and is the slope
21
345 of the curve. Many database of fiberglass laminates showed that the slope of the curve, , range
346 from 6 to 12 (Freebury and Musial, 2000; Mandell and Samborsky, 1997). Commonly poor
347 performing fabrics and laminates with fabrication flaw such as resin and etc. have slopes close to
348 6; however, by improving the material quality, the slope would have values close to 12. Figure
349 8b show the normalized stress (or normalized moment) versus allowable number of cycles to
0
10
Elastic Strain
Plastic Strain
Total Strain
10-2
10-4
100 102 104 106
(a) Log (N )
f
351 Figure 8. (a) Coffin-Manson ε-N curve and (b) fiberglass σ-N curve
352 In this study, since the mean wind speeds and turbulence intensities of various models were
353 identical, the mean load and accordingly the mean damage index caused by fatigue were almost
354 the same; however, different turbulence spectrums resulted in different damage indices in
355 buffeting responses. By considering 2 hours of hurricane loading on the wind farm location,
356 Figure 9a presents tower base buffeting damage indices where the spectrums Model A and
357 Model B have average 53% smaller and 12% larger damage indices compared to Kaimal et al.
358 (1972) models. Figure 9b also shows the blade root damage indices for different wind speeds and
359 hurricane categories where spectrum Model A and Model B resulted in 96% smaller and 24%
360 larger damage indices compared to Kaimal et al. (1972) model. Although structural responses
22
361 caused by spectrum Model B has small difference with Kaimal et al. (1972) model, low cycle
362 fatigue analysis of the tower and blades showed that the Model B results in higher damage
(a) (b)
364 Figure 9. (a) 2 hours tower base buffeting damage indices and (b) blade root buffeting
365 damage indices for different hurricane models and hurricane categories
366 During the wind turbine parked (stand-still) condition, the blades have 90° pitch angle which
367 decreases the edge-wise moments which is the reason that the damage indices for blade edge-
368 wise root moments are very small. In this study, the spectrum Model B only represents the
369 longitudinal fluctuation components. Hence, further research is needed to address lateral
370 fluctuation components of Model B and investigate the responses such as tower side to side
371 moment, blade flap-wise moments, and damage indices correlated with lateral components.
373 IEC 61400-3 (2009) provides recommendations for extreme value analysis and long term load
374 extrapolations for different wind turbine classes; however, for special conditions such as
375 hurricane, it defines class S, where the values should be specified by the designer. Hence, for
376 better understanding the required values for class S, a comparison between IEC 61400-3 (2009)
377 recommendations and results of current study was carried out. For this purpose, the wind turbine
23
378 class I in an low turbulence offshore region (similar to recent observations (Yu et al., 2008)) was
379 selected where the reference wind speed and turbulence intensity at hub height were 50 m/s and
381 In order to conduct an extreme value analysis, there are several methods such as direct
382 integration, first order reliability method (FORM), and inverse first order reliability method
383 (IFORM) (Agarwal and Manuel, 2009; Ragan and Manuel, 2008). The direct integration method
384 which is the most straight forward method can be expressed as (Ragan and Manuel, 2008):
1 | (19)
385 where is the probability that load resulted from 10 minute simulation exceed
386 the load value of , | represents the cumulative distribution of load value in the wind
387 speed bin , and is the wind speed distribution. The most basic way to use direct
388 integration method is the method of global maxima. In this case, for each 10 minute simulation
389 the global maxima can be detected and the best distribution are fitted on the observed maxima.
390 Another common way for using the direct integration method is to use the method of peak-over
391 threshold (POT). In the POT method, a threshold is selected and the local maxima are captured.
1 | (20)
393 where is the expected number of peaks above the selected threshold. According to IEC 61400-
394 1 (2005) and IEC 61400-3 (2009), the 10 minutes averaging duration is standard for wind turbine
24
395 analysis. In this case, the probability to observe a load larger than -years load, , can be
397 In order to compare IEC 61400-3 (2009) recommendation and results of new hurricane
398 turbulence models, the extreme wind speed model (EWM) with 50-years recurrence period (50
399 m/s @ hub height) was considered which reduced eq. 21 to:
50 3.80 10 (22)
400 Choosing an appropriate threshold and distribution for exceedance probability are important for
401 extreme value analysis (Agarwal and Manuel, 2009; IEC 61400-3, 2009; Ragan and Manuel,
402 2008). The distribution representing the exceedance probability of the local maxima in each wind
403 speed should precisely represent the tail behavior. Figure 10 shows observed local maxima of
404 tower buffeting base moment with different POTs where Weibull, lognormal, and generalized
405 extreme value (GEV) distributions are fitted on the data (SD: Standard Deviation). By increasing
406 the threshold value, the POT method leads in the method of global maxima.
25
(a) (b)
(c) (d)
407 Figure 10. Fitted distributions on local maxima based on different thresholds as (a) 1.4SD,
408 (b) 2.0SD, (c) 2.5SD, and (d) 3.0SD
409 The direct integration method with a POT of 1.4 time standard deviation (IEC 61400-1, 2005)
410 was applied for extreme value analysis of the tower fore-aft base and blade root edge-wise
411 moments. In addition, three mentioned distributions for fitting local maxima in this section were
412 used for the analysis. Figure 11 shows the response extreme value analysis of the three
413 turbulence models by assuming different distributions compared to IEC 61400-3 (2009)
414 recommendations. At 50-years return period, by assuming Weibull distribution for tower
415 buffeting fore-aft moment, IEC 61400-3 (2009) recommendations are larger than resulted
416 extreme values from all turbulence models; however, IEC 61400-3 (2009) recommendations are
26
417 smaller than results of GEV assumption (Figure 11a, c, e). Hence, by assuming Weibull and
418 GEV distributions respectively, IEC 61400-3 (2009) recommendations for wind turbine class I
419 are conservative and non-conservative for designing the tower of wind turbine class S subjected
420 to hurricane.
421 Blades response extreme value analysis showed that for Kaimal et al. (1972) and Model B, all
422 three distributions resulted in larger extreme values than IEC 61400-3 (2009) recommendations
423 (Figure 11b, f); however, for Model A, Weibull and GEV distributions resulted in smaller and
424 larger values than IEC 61400-3 (2009) recommendations respectively (Figure 11d).
Exceedance Probability
Exceedance Probability
(a) (b)
Exceedance Probability
Exceedance Probability
(c) (d)
27
Exceedance Probability
Exceedance Probability
(e) (f)
425 Figure 11. Extreme value analysis of fore-aft buffeting base moment (a: Kaimal et al.
426 (1972), c: Model A, e: Model B) and blade root edge-wise moment (b: Kaimal et al. (1972),
427 d: Model A, f: Model B) using different distributions
428 The comparison of the different distributions for extreme value analysis showed that, Weibull
429 distributions has the smallest extreme values for tower and blades. Hence the IEC 61400-3
430 (2009) recommendations compared to Weibull distribution results, will be the condition for
431 proposing minimum changes in IEC 61400-3 (2009). By assuming Weibull distribution for local
432 maxima, Figure 12a shows the extreme values of tower buffeting base moment for three different
433 turbulence models where the IEC 61400-3 (2009) recommendations resulted in 30%, 37%, and
434 18% larger values than Kaimal et al. (1972), Model A and Model B respectively. For blade root
435 edge-wise moment, the IEC 61400-3 (2009) recommendations resulted in underestimating the
436 blade buffeting responses compared to conventional Kaimal et al. (1972) model and spectrum
437 Model B (Figure 12b). Table 5 shows the values of tower buffeting fore-aft base moment and
438 blade root buffeting edge-wise moment extrapolated for 50-years return period.
28
(a) (b)
439 Figure 12. Extreme value analysis of the (a) tower buffeting base moment, (b) blade root
440 buffeting edge-wise moment
441 Table 5. 50-years return period buffeting responses for different models
Response Kaimal et al. Model A Model B IEC 61400-3
(1972) (2009)
Tower Buffeting base
64.07 60.12 70.68 83.38
moment (MNm)
Blade Root edge-wise
3.54 1.17 3.78 1.67
moment (MNm)
442
443 The extreme value analysis of along-wind responses depicted that by assuming Weibull
444 distribution for local maxima, for designing the tower, the IEC 61400-3 (2009) recommendations
445 for turbine class I can be used for class S; however, for designing the blades subjected to
447 6. Conclusions
448 This paper studied the effects of newly observed hurricane turbulence models on offshore wind
449 turbines by considering unsteady aerodynamic forces on the tower and wind-wave-soil-structure
450 interaction. For this purpose, first, the recent observations and presented turbulence models were
451 discussed. Then, the Model A, with high amount of turbulence energy in low frequencies and
452 Model B, with high amount of energy in high frequencies were introduced. Next, the unsteady
453 formulation for wind turbine tower was introduced and NREL-FAST package was modified with
29
454 new formulation. Later, the new turbulence models as well as new unsteady aerodynamic
455 formulation was used for simulations. The five Saffir-Simpson hurricane categories as well as
456 wind induced wave field was generated. The goals were analyzing the tower and blade structural
457 buffeting responses, the low cycle fatigue during different hurricane categories, and extreme
458 value analysis of the responses for comparison with existing standard recommendations. At the
460 - Analysis of the tower buffeting base moments showed that spectrum Model B and
461 conventional Kaimal et al. (1972) model almost resulted in identical responses with 1%
462 difference; however, spectrum Model A resulted in almost 7% smaller responses. The
463 tower tip displacements of Model A and Model B resulted in average 6% smaller and 1%
464 larger responses than Kaimal et al. (1972) model respectively. Spectral analysis of the
465 tower tip displacement showed that Model A, in low frequency responses (background
466 responses) has higher amount of energy compared two other two spectrum models which
467 was consistent with characteristics of spectrum model A. All three spectrums had similar
468 response energy around 1st tower fore-aft frequency; however, in higher frequencies,
469 spectrum Model A, had smaller amount of energy compared to other two counterparts.
470 - For blade root edge-wise moment, since the blade responses were mostly involved with
471 aerodynamic forces rather than wave forces, the differences between models were more
472 evident. Model A and Model B resulted in 57% smaller and 9% larger responses than
473 conventional Kaimal et al. (1972) respectively. The blade tip displacement results from
474 three different models showed that the spectrum Model A and B resulted in 62% larger
475 and 7% smaller responses respectively compared to Kaimal et al. (1972) model. The
476 spectral analysis of the blade tip responses showed that the 1st blade edge-wise frequency
30
477 which is located in relatively high frequencies is the most important resonant part of the
478 responses. On the other hand, spectrum Model A has low amount of energy in high
479 frequencies. Hence the level of response energy resulted from spectrum Model A is
480 smaller than other two spectrums. The lower amount of spectral response of Model A
481 compared two other two models, especially in the 1st blade edge-wise frequency range,
482 resulted in large difference between maximum blade tip displacements of Model A
484 - Although the mean fatigue damage was identical between models, different turbulence
485 models can cause different low cycle fatigue damages due to buffeting forces. By
486 considering 2 hours of different hurricane categories on the wind farm location, the
487 spectrum Model A and Model B resulted in average 53% smaller and 12% larger damage
488 indices compared to Kaimal et al. (1972) model. In addition, spectrum Model A and
489 Model B resulted in 96% smaller and 24% larger damage indices compared to Kaimal et
490 al. (1972) model for blade root buffeting edge-wise moment.
491 - In order to compare the responses of short term analysis of the offshore wind turbine with
492 IEC 61400-3 (2009) recommendations for 50-years return period, an extreme value
493 analysis was carried out. For the extreme value analysis, the direct integration method
494 with peak over threshold (POT) approach was used. The extreme value analysis of along-
495 wind responses depicted that by assuming Weibull distribution for local maxima, for
496 designing the tower, the IEC 61400-3 (2009) recommendations for turbine class I can be
497 used for class S; however, for designing the blades subjected to hurricane, IEC 61400-3
498 (2009) should be reconsidered. On the other hand, assuming GEV distribution showed
31
499 that IEC 61400-3 (2009) recommendation for turbine class I, should be revised for
501 - The comparison of three different distributions for fitting local maxima in extreme value
502 analysis showed that, Weibull distribution has the smallest extreme values for tower and
503 blades. Hence the IEC 61400-3 (2009) recommendations compared to Weibull
504 distribution results, will be the condition for proposing minimum changes in IEC 61400-3
505 (2009). By using the Weibull distributions, the results showed that for the tower fore-aft
506 buffeting base moment, the IEC 61400-3 (2009) recommendations resulted in 30%, 37%,
507 and 18% larger values than Kaimal et al. (1972), Model A and Model B respectively;
508 however, the IEC 61400-3 (2009) recommendations resulted in underestimating the blade
509 edge-wise buffeting responses compared to conventional Kaimal et al. (1972) model and
510 spectrum Model B. Hence, for designing the tower for special events such as hurricane,
511 the IEC 61400-3 (2009) recommendations for turbine class I can be used for wind turbine
512 class S; however, for designing the blades subjected to hurricane, IEC 61400-3 (2009)
514 7. Acknowledgements
515 The research reported here is supported in part by the National Science Foundation CMMI Grant
516 1252736. Any opinions, findings, and conclusions expressed in this paper are those of the
517 authors and do not necessarily reflect the views of the sponsor.
518 8. References
519 Agarwal, P., Manuel, L., 2009. Simulation of offshore wind turbine response for long-term
520 extreme load prediction. Engineering structures 31 (10), 2236-2246.
32
521 Amirinia, G., Jung, S., 2017. Along-wind buffeting responses of wind turbines subjected to
522 hurricanes considering unsteady aerodynamics of the tower. Engineering structures 138, 337-
523 350.
524 Amirinia, G., Jung, S., 2016. Time domain analysis of unsteady aerodynamic forces on a parked
525 wind turbine tower subjected to high winds, 8th International Colloquium on Bluff Body
526 Aerodynamics and Applications., Boston, US.
527 Amirinia, G., Jung, S., Alduse, B.P., 2015. Effect of different hurricane spectrums on wind
528 turbine loads and responses, AWEA Wind Power Conference, Orlando, US,.
529 Balderrama, J., Masters, F., Gurley, K., Prevatt, D., Aponte-Bermúdez, L., Reinhold, T., Pinelli,
530 J.-P., Subramanian, C., Schiff, S., Chowdhury, A., 2011. The Florida coastal monitoring program
531 (FCMP): A review. Journal of Wind Engineering and Industrial Aerodynamics 99 (9), 979-995.
532 Basquin, O., 1910. The exponential law of endurance tests, proc. ASTM, p. 625.
533 Caracoglia, L., Jones, N.P., 2009. Analysis of full-scale wind and pressure measurements on a
534 low-rise building. Journal of Wind Engineering and Industrial Aerodynamics 97 (5), 157-173.
535 Carvelli, V., Gramellini, G., Lomov, S.V., Bogdanovich, A.E., Mungalov, D.D., Verpoest, I.,
536 2010. Fatigue behavior of non-crimp 3D orthogonal weave and multi-layer plain weave E-glass
537 reinforced composites. Composites science and technology 70 (14), 2068-2076.
538 Chen, X., Matsumoto, M., Kareem, A., 2000. Time domain flutter and buffeting response
539 analysis of bridges. Journal of Engineering Mechanics 126 (1), 7-16.
540 Clobes, M., Peil, U., 2011. Unsteady buffeting wind loads in the time domain and their effect on
541 the life-cycle prediction of guyed masts. Structure and Infrastructure Engineering 7 (1-2), 187-
542 196.
543 Coffin Jr, L.F., 1954. A study of the effects of cyclic thermal stresses on a ductile metal. trans.
544 ASME 76, 931-950.
545 Davenport, A., 1977. The prediction of the response of structures to gusty wind. Safety of
546 structures under dynamic loading 1, 257-284.
547 Davenport, A.G., 1961. The spectrum of horizontal gustiness near the ground in high winds.
548 Quarterly Journal of the Royal Meteorological Society 87 (372), 194-211.
549 Durst, C., 1960. Wind speeds over short periods of time. Meteorological Magazine 89 (1960),
550 181-186.
551 Ellyin, F., 2012. Fatigue damage, crack growth and life prediction. Springer Science & Business
552 Media.
553 Fiedler, F., Panofsky, H.A., 1970. Atmospheric scales and spectral gaps. Bulletin of the
554 American Meteorological Society 51 (12), 1114-1120.
33
555 Freebury, G., Musial, W.D., 2000. Determining equivalent damage loading for full-scale wind
556 turbine blade fatigue tests. National Renewable Energy Laboratory.
557 Gong, K., Chen, X., 2015. Improved modeling of equivalent static loads on wind turbine towers.
558 Wind and Structures 20 (5), 609-622.
559 Guo, Y., Damiani, R., Musial, W., 2014. Simulating Turbulent Wind Fields for Offshore
560 Turbines in Hurricane-Prone Regions (Poster). National Renewable Energy Laboratory (NREL),
561 Golden, CO.
562 Han, T., McCann, G., Mücke, T., Freudenreich, K., 2014. How can a wind turbine survive in
563 tropical cyclone? Renewable Energy 70, 3-10.
564 IEC 61400-1, 2005. IEC 61400-1: Wind turbines part 1: Design requirements. International
565 Electrotechnical Commission.
566 IEC 61400-3, 2009. Wind turbines – Part 3: Design requirements for offshore wind turbines.
567 International Electrotechnical Commission 3.
568 Jonkman, B.J., 2009. TurbSim user's guide: version 1.50. National Renewable Energy
569 Laboratory Golden, CO, USA.
570 Jonkman, J., Musial, W., 2010. Offshore code comparison collaboration (OC3) for IEA task 23
571 offshore wind technology and deployment. Contract 303, 275-3000.
572 Jonkman, J., Robertson, A., Hayman, G., 2014. HydroDyn user’s guide and theory manual.
573 National Renewable Energy Laboratory.
574 Jonkman, J.M., Buhl Jr, M.L., 2005. FAST user’s guide. National Renewable Energy
575 Laboratory, Golden, CO, Technical Report No. NREL/EL-500-38230.
576 Jonkman, J.M., Butterfield, S., Musial, W., Scott, G., 2009. Definition of a 5-MW reference
577 wind turbine for offshore system development. National Renewable Energy Laboratory Golden,
578 CO, USA.
579 Jung, S., Masters, F., 2013. Characterization of open and suburban boundary layer wind
580 turbulence in 2008 Hurricane Ike. Wind and Structures 17 (2), 135-162.
581 Kaimal, J., Wyngaard, J., Izumi, Y., Coté, O., 1972. Spectral characteristics of surface‐layer
582 turbulence. Quarterly Journal of the Royal Meteorological Society 98 (417), 563-589.
583 Karahan, M., Lomov, S.V., Bogdanovich, A.E., Verpoest, I., 2011. Fatigue tensile behavior of
584 carbon/epoxy composite reinforced with non-crimp 3D orthogonal woven fabric. Composites
585 science and technology 71 (16), 1961-1972.
586 Kim, E., Manuel, L., 2013. On the Extreme Rotor and Support Structure Response of an
587 Offshore Wind Turbine in an Evolving Hurricane, ASME 2013 32nd International Conference
34
588 on Ocean, Offshore and Arctic Engineering. American Society of Mechanical Engineers, pp.
589 V008T009A082-V008T009A082.
590 Kim, E., Manuel, L., 2014. Hurricane-Induced Loads on Offshore Wind Turbines with
591 Considerations for Nacelle Yaw and Blade Pitch Control. Wind Engineering 38 (4), 413-424.
592 Kim, E., Manuel, L., Curcic, M., Chen, S.S., Phillips, C., Veers, P., 2016. On the Use of Coupled
593 Wind, Wave, and Current Fields in the Simulation of Loads on Bottom-Supported Offshore
594 Wind Turbines during Hurricanes. NREL (National Renewable Energy Laboratory (NREL),
595 Golden, CO (United States)).
596 Lee, S., Churchfield, M.J., Moriarty, P.J., Jonkman, J., Michalakes, J., 2013. A numerical study
597 of atmospheric and wake turbulence impacts on wind turbine fatigue loadings. Journal of Solar
598 Energy Engineering 135 (3), 031001.
599 Li, C., Bao, G., Cao, Y., Zhou, J., Liu, M., 2014. Wind-Induced Fatigue Loading and Its Effect
600 on the Stability of Wind Power Foundations, New Frontiers in Geotechnical Engineering. ASCE,
601 pp. 78-87.
602 Li, L., Xiao, Y., Kareem, A., Song, L., Qin, P., 2012. Modeling typhoon wind power spectra near
603 sea surface based on measurements in the South China sea. Journal of Wind Engineering and
604 Industrial Aerodynamics 104, 565-576.
605 Mandell, J.F., Samborsky, D.D., 1997. DOE/MSU composite material fatigue database: test
606 methods, materials, and analysis. Sandia National Labs., Albuquerque, NM (United States).
607 Manson, S., 1965. Fatigue: a complex subject—some simple approximations. Experimental
608 mechanics 5 (7), 193-226.
609 Miner, M.A., 1945. Cumulative damage in fatigue. Journal of applied mechanics 12 (3), 159-
610 164.
611 Ragan, P., Manuel, L., 2008. Statistical extrapolation methods for estimating wind turbine
612 extreme loads. Journal of Solar Energy Engineering 130 (3), 031011.
613 Rose, S., Jaramillo, P., Small, M.J., Grossmann, I., Apt, J., 2012. Quantifying the hurricane risk
614 to offshore wind turbines. Proceedings of the National Academy of Sciences 109 (9), 3247-3252.
615 Schroeder, J.L., Smith, D.A., 2003. Hurricane Bonnie wind flow characteristics as determined
616 from WEMITE. Journal of Wind Engineering and Industrial Aerodynamics 91 (6), 767-789.
617 Sebastian, T., Lackner, M., 2013. Characterization of the unsteady aerodynamics of offshore
618 floating wind turbines. Wind Energy 16 (3), 339-352.
619 Simpson, R.H., Saffir, H., 1974. The hurricane disaster potential scale. Weatherwise 27 (8), 169.
35
620 Singer, I., Busch, N., Frizzola, J., 1968. The micrometeorology of the turbulent flow field in the
621 atmospheric boundary surface layer, Proceedings of international research seminar on wind
622 effects on buildings and structures, Ottawa, University of Toronto Press, pp. 557-594.
623 Solari, G., 1987. Turbulence modeling for gust loading. Journal of Structural Engineering 113
624 (7), 1550-1569.
625 Sorensen, R.M., 2005. Basic coastal engineering. Springer Science & Business Media.
626 Strømmen, E., 2010. Theory of bridge aerodynamics. Springer Science & Business Media.
627 Teunissen, H., 1970. Characteristics of the mean wind and turbulence in the planetary boundary
628 layer. DTIC Document.
629 Tibaldi, C., Henriksen, L., Hansen, M., Bak, C., 2015. Wind turbine fatigue damage evaluation
630 based on a linear model and a spectral method. Wind Energy.
631 Tieleman, H.W., 1995. Universality of velocity spectra. Journal of Wind Engineering and
632 Industrial Aerodynamics 56 (1), 55-69.
633 U.S. Army Coastal Engineering Research Center, 1977. Shore protection manual, 3rd ed.
634 Government Printing Office, Washington, D.C.
635 Von Karman, T., 1948. Progress in the statistical theory of turbulence. Proceedings of the
636 National Academy of Sciences of the United States of America 34 (11), 530.
637 Yu, B., Chowdhury, A.G., Masters, F.J., 2008. Hurricane wind power spectra, cospectra, and
638 integral length scales. Boundary-layer meteorology 129 (3), 411-430.
639
36