Journal of Wind Engineering and Industrial Aerodynamics 96 (2008) 2217–2227

A peak factor for non-Gaussian response analysis of wind turbine tower
Luong Van Binha,Ã, Takeshi Ishiharab, Pham Van Phuca, Yozo Fujinoa

Department of Civil Engineering, The University of Tokyo, 7-3-1, Hongo, Bunkyo-ku, Tokyo 113-0033, Japan b Institute of Engineering, Innovation School of Engineering, The University of Tokyo, 2-11-16, Yayoi, Bunnkyo-ku, Tokyo 113-8656, Japan Available online 28 March 2008

Abstract Equivalent static wind load evaluation formulas considering the dynamic effects based on peak factor were proposed to estimate the design wind load on the wind turbine tower in complex terrain. The non-linear part of wind pressure was considered to estimate the mean wind loads. The peak factor based on a non-Gaussian assumption was derived to estimate the non-linearity of wind load, especially in the high turbulence intensity. The formula of the peak factor is simplified to a function of the third order moment (skewness) considering the spatial correlation of wind velocity, the resonance response and the background response. The proposed methods showed favorable agreements with dynamic wind response analysis by FEM. r 2008 Elsevier Ltd. All rights reserved.
Keywords: Wind turbine tower; Non-Gaussian response; Peak factor; Skewness; Spatial correlation; Resonance

1. Introduction Wind load on wind turbine is usually evaluated either by finite element model (FEM) or by equivalent static method. While FEM simulation is commonly used in turbine design, equivalent static method is used widely in design of lower and other support structures. Equivalent static method is adopted in many design codes (recommendations for loads on
ÃCorresponding author. Tel.: +81 3 5841 1145; fax: +81 3 5841 1147.

E-mail address: (L.V. Binh). 0167-6105/$ - see front matter r 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.jweia.2008.02.019

3. The formula is verified by FEM using the wind turbine investigated by Ishihara et al. Equivalent static method for wind turbine To illustrate this method.V.ARTICLE IN PRESS 2218 L. (2002). Wind load on the rotor is calculated and transferred to the tower as shear force and bending moment at top of the tower. only the first mode is considered. Wind turbine model In this study the model of an elastic tower and a rigid rotor. Besides. Binh et al. the non-linear part of wind pressure is neglected. Using the moment-based Hermite transformation method and the definition of peak factor proposed by Nishijima et al. it should be noted here that in the case of seismic load. Aerodyn. which considers the non-linear part of wind pressure. 1992). if a structure is under high turbulence intensity. A wind turbine is characterized by a low structural damping and a heavy head. Kareem and Zhao (1994) proposed a formula for peak factor. wind turbines exposed to high wind turbulence in areas with complex terrain like Japan can exhibit strong non-Gaussian responses. In formulas of mean wind load. The mean wind load. standard deviation and peak factor of fluctuating wind load proposed in codes. let us start with a simple model of wind turbine response € þ Cx _ þ Kx ¼ F tot . is used to implement the theoretical formula of mean. shown in Fig. The effect of higher modes is negligibly small. meanwhile. since contribution of the non-linear part of wind pressure is large and the response is non-Gaussian. is derived. Wind Eng. Ishikawa (2004). Parameters of the formula of the peak factor are determined by the results of FEM simulation. However. 1b. However. Ind. considering spatial correlation is essential. Mx (1) . is proposed. Since in wind load of wind turbine tower the effect of the first mode is dominant. where the power spectrum is high in high frequency regions. The Danish Society of Engineers and The Federation of Engineers. and with the rapid increase of wind turbine size. / J. Wind velocity and turbulence intensity at the hub of the wind turbine are used as representative for that of the whole rotor. which takes into account both the spatial correlation of wind velocity and resonance response. this formula neglects resonance response due to high damping ratio of the transmission line. Therefore. The non-Gaussian peak factor. especially when spatial correlation of wind velocity is considered. pointed out that Kareem’s formula gives conservative results. mean wind load and peak factor may be underestimated. standard deviation and the peak factor of wind load on a tower base. This study proposes a formula of the maximum wind load on wind turbines in complex terrain. 2. (2005). which results in significant resonant response. which can be applied to non-Gaussian process and confirmed its validity in the case of a single degree of freedom system through numerical simulation. 96 (2008) 2217–2227 buildings. 1993. This method uses a coefficient called the peak factor proposed by Davenport (1964) to account for fluctuating wind load. Architectural Institute of Japan. Danish standard DS472. because of the low power in high frequency region of the spectrum of wind load. the effect of higher modes is not negligible. Ishikawa (2004) proposed a formula for peak factor which considers both nonGaussianity and spatial correlation of wind load on transmission line.

(a) Wind turbine. Cf is the aerodynamic force coefficient. where Iu is the turbulence intensity. 96 (2008) 2217–2227 2219 WIND WIND XF FD ML FL yF MD Blade φ Nacelle WIND Fig. the bending momentbased peak factor is adopted. and 2 2 2 1 F tot ¼ 1 2 rC f S ðU þ uÞ ¼ 2 rC f S ðU þ 2Uu þ u Þ. Mean wind load From (2) the mean wind force and bending moment can be derived: 2 2 2 2 1 F tot ¼ 1 2 rC f S ðU þ su Þ ¼ 2 rC f SU ð1 þ I u Þ. A study by Kareem and Zhou (2003) proved that the bending moment-based peak factor can yield more reliable results than displacement-based peak factor. This means the term wind load should be interpreted as a bending moment. C is the damping matrix and K is the stiffness matrix. (c) Wind direction and wind load. 1. c(x) is the characteristic size of the element at position x and R denotes all over wind turbine. (2) where r is the density of air. S is the considered area. Wind turbine model and wind direction definition.V. Aerodyn. 3. Binh et al. (b) Simplified model. because the mean value of displacement may be zero.1. Ind. Therefore.ARTICLE IN PRESS L. U is the mean wind velocity and u is the fluctuating wind velocity. in this study. where M is the mass matrix. Wind Eng. / J. . (3) (4) Z M¼ R 2 1 2 rC f ðxÞU ð1 þ I2 u ÞcðxÞx dx.

From (2) and (3) fluctuating wind force can be calculated: 2 2 2 2 2 2 1 1 1 Ft ¼ 1 2 rC f S ðU þ uÞ À 2 rC f SU ð1 þ I u Þ ¼ rC f SUu þ 2 rC f Su À 2rC f SU I u .ARTICLE IN PRESS 2220 L.x2 . If the considering point is on the rotor then l is the distance from top of the tower to the tower base. x2 about the tower base.nÞ u m ðx1 Þm1 ðx2 Þcðx1 Þcðx2 Þ dx1 dx2 0 0 Cou R1  . (9) where r12 is the cross correlation of wind velocity at x1 and x2. Aerodyn. Wind Eng.V. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sMB 2I u K SMB þ K 0SMB . M x vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi uR R R R ðx1 . The resonant part of standard deviation which considers only the first mode of tower can be derived from modal analysis. Standard deviation Standard deviation of fluctuating wind load consists of a background part sMB and a resonant part sM1: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 (5) sM ¼ s2 MB þ sM 1 . x1. Coð is the normalized co-spectrum of wind u velocity and m1 is the first mode shape. Ru is the normalized power spectrum of wind. l2 are the bending lever arms of elements at x. Ind. Binh et al. respectively. as follows: pffiffiffi pI u lM 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sM 1 pffiffiffi ¼ (10) Ru ðn1 ÞK Sx1 ðn1 Þ. / J. ¼u t R 0 m1 ðrÞcðrÞ dr RR 0 K Sx1 (11) lM 1 mðrÞm1 ðrÞr dr ¼ RR m1 0 cðrÞr dr Z 0 R c ð r Þ m 1 ð r Þ dr . (7) ¼ M 1 þ I2 u R R K SMB ¼ 2 1 2 1 R R ð2rC f U 1 Þð2rC f U 2 Þr12 cðx1 Þcðx2 Þl 1 l 2 ÀR 1 Á2 2 R 2rC f U cðxÞl dx dx1 dx2 . the background standard deviation of wind load can be calculated by dividing the bending moment caused by Ft to mean bending moment M. l1. (8) K 0SMB 1 ¼ I2 2 u R R 2 1 2 2 1 R R ð2rC f U 1 Þð2rC f U 2 Þr12 Â cðx1 Þcðx2 Þl 1 l 2 dx1 dx2 ÀR 1 Á2 2 R ð2rC f U ÞcðxÞl dx . (12) where x is the structural damping ratio. The bending lever arm l is the distance from the considering point to tower base if that point is on the tower. l. n1 is x1 .2. Calculation of the integrals in (8) and (9) is implemented by a computer program which uses two lists of wind turbine elements to consider all available correlations. (6) Therefore. . 96 (2008) 2217–2227 3.x2 . The number of lists becomes three and four for three-fold or four-fold integrals.nÞ the first modal frequency of the structure.

a4 are the third. / J.0) and the expression of peak factor becomes npffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi o g¼k 2 ln n0Y T þ h3 ð2 ln n0Y T À 1Þ . From formula of a3 and a4 derived by Ishikawa (2004). a4 is then assumed to be equal to the value of a Gaussian process (i. (2002) in which the peak factor of a process is the value that the process up-crosses once on average in a certain time T: nqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi o g¼k 2 ln n0y T þ h3 ð2 ln n0y T À 1Þ þ h4 ½ð2 ln n0y T Þ3=2 À 3 2 ln n0y T Š . the skewness of fluctuating wind load is necessary to calculate the peak factor g. 1 þ ð a2 3 = 9Þ 1 k ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . n0 y is the zero up-crossing number in T of the non-Gaussian process Y0 and ny is the zero up-crossing number in T of a Gaussian process Y which can be calculated by (14). T is the estimated time interval (normally T ¼ 600 s).. Ishikawa (2004) derived a formula for the peak factor using the definition of Nishijima et al. and n is the frequency variable. Peak factor model In order to take the non-linear component of wind load into account. the effect of the forth order part a4 is neglected since it is negligibly small compared to that of the second and third order from the order analysis of turbulence intensity Iu.ARTICLE IN PRESS L. 1 þ ð a2 3 =18Þ (18) In this model. Assuming that wind response of structure is a Gaussian process.V. 6 1 n0Y ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ny . (17) h3 ¼ a3 . A model for skewness of wind load on transmission line. Peak factor A widely adopted model in codes is the peak factor model proposed by Davenport (1964). 3. Binh et al. 2 lnðnT Þ sR ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 2 0 n S u ðnÞ dn R . h4 ¼ . (15) 1 n0Y ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ny . Wind Eng.e. proposed by Ishikawa . 96 (2008) 2217–2227 2221 3. Su is the power spectrum of wind velocity. the formula is given as g¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0:5772 2 lnðnT Þ þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . 2 1 þ 4h2 3 þ 18h4 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 þ ð3ða4 À 3ÞÞ=2 À 1 a3 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . 4. Aerodyn. n¼ 1 2 sMB þ s2 M1 0 S M ðnÞ dn (14) where n is the zero up-crossing number in a unit of time of a Gaussian process.3. 2 1 þ 2h2 3 þ 6h4 (16) where a3. h3 ¼ 18 4 þ 2 1 þ ð3ða4 À 3ÞÞ=2 1 k ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . respectively. forth order moments of wind load. SM is the power spectrum of wind load. Ind. n0 ¼ 1 0 S u ðnÞ dn (13) sR ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 2 2 2 2 n2 0 sMB þ n1 sM 1 0 n S M ðnÞ dn R ¼ .

This is a function of the resonance–background ratio Rd of 0 standard deviation denoted by f(Rd). they cannot be applied directly to wind turbines. Ind. a function of resonance response is introduced into (19). In this study. the Newmark method Subspace iteration procedure Beam element Total Lagrangian formulation Quasi-steady aerodynamic theory Rayleigh damping Table 2 Main characteristics of the wind turbine studied Name Rated power Regulation Number of blades Rotor diameter Hub height Description 400 kW Stall 3 31 m 36 m . Aerodyn. It is noted that these formulas do not consider resonance response. is as follows: a3 ¼ 3I u ar1 þ I 3 u ar 2 ðK SMB þ K 0SMB Þ3=2 RL RL RL ar 1 ¼ 0 0 . / J. Wind Eng. respectively. 96 (2008) 2217–2227 (2004). Therefore. (19) 2 1 2 1 2 1 0 ð2rC f U 1 Þð2rC f U 2 Þð2rC f U 3 Þr12 r23 cðx1 Þcðx2 Þcðx3 Þx1 x2 x3 dx1 dx2 R 3 L 1 2 0 ð2rC f U ÞcðxÞx dx dx3 (20) RL RL RL a r2 ¼ 0 0 2 1 2 1 2 1 0 ð2rC f U 1 Þð2rC f U 2 Þð2rC f U 3 Þr12 r23 r13 cðx1 Þcðx2 Þcðx3 Þx1 x2 x3 R 3 L 1 2 ð r C U Þ c ð x Þ x d x f 0 2 dx1 dx2 dx3 . Binh et al.V. (22) Table 1 Description of the FEM code Name Dynamic analysis Eigenvalue analysis Element type Formulation Aerodynamic force Damping Description Direct numerical integration.ARTICLE IN PRESS 2222 L. Since I 3 u and K SMB are negligibly small compared to Iu and KSMB. (21) where L is the length of the transmission line. the expression of a3 in (19) becomes a3 ¼ f ðRd Þ Â 3I u ar1 ðK SMB Þ3=2 .

Ind. Wind Eng. Wind series at all nodes are generated by a correlation matrix and wind load derived from these series is used in the equation of motion. . (2005).74° North 5 SYMBOL Wind direction Wind velocity Strain gauges Temperature correction Accelerator X direction Accelerator Y direction Yof f1=300mm Yof f2=400mm 35000 1 1 6 2 2 2 E N S W y x X Y Fig.V.27° 31000 X Y Yof f2 8700 X Y Yof f1 100 1000 22750 ‡@ 3650 ‡A ‡C ‡B 6.ARTICLE IN PRESS L. developed by Ishihara et al. sMB (23) The FEM code. is described in Table 1. 96 (2008) 2217–2227 2223 Rd ¼ sM 1 . 2 with 21. 2. The main idea is using an aerodynamically and structurally modeled beam element to model wind turbine tower and blades. The FEM program is used to simulate the response of the wind turbine model described in Table 2 and Fig. / J. Binh et al. Configuration of wind turbine. Aerodyn.

005 to 0. a¼ 2 Rd ðK SMB Þ3=2 a3 (25) In order to determine a. Wind Eng. It is also noticed that skewness increases when damping ratio increases.. when Rd ¼ 0). since the damping ratio x is À2 proportional to Rd . FEM wind response simulations of wind turbine of different Rd (i.V. Finally. Therefore the following form of f(Rd) is proposed and a can be derived: f ðRd Þ ¼ aR 2 d 1 .2 Turbulence Intensity 0.6 Skewness 0.3 is proposed. Results in Fig. turbulence intensity and damping. different damping ratio x) were carried out to calculate skewness a3. Since skewness decreases when Rd increases. The design wind speed at hub is 50 m/s. Aerodyn.e.4 0. it can be assumed that the skewness a3 and the damping ratio x have a linear relation. the peak factor decreases when Rd increases. the peak factor decreases when the . Binh et al. Relation of skewness. þ1 (24) " # 1 3I u ar 1 1 À1 . different structural damping. the peak factor decreases when skewness decreases. 3 show that skewness and turbulence intensity have a linear relationship. From the result in Fig.1 0. It is also noticed that f(Rd) should become 1 if there is no resonance (i.e. The formula of skewness a3 becomes a3 ¼ 1 3I u ar1 . The power law for wind shear and turbulence intensity of different terrain categories described in Architectural Institute of Japan’s (1993) recommendations for loads on buildings is adopted.2 0 0 0. Therefore.e. f(Rd) is supposed to be proportional 2 to the damping ratio x (i.. / J.. 96 (2008) 2217–2227 0. which confirms the validity of formula (22). 3.ARTICLE IN PRESS 2224 L.3 Fig. Ind. Â 3=2 1 :3 R 2 þ 1 ð K SMB Þ d (26) In this model of skewness.8 skew (ξ = 0. Other parameters are calculated from theoretical formula. 4 the conservative value a ¼ 1. a is calculated by formula (25).8%) skew (ξ = 2%) skew (ξ = 4%) skew (ξ = 6%) skew (background) 0.01. Since the damping ratio of wind turbine x varies in a narrow range from 0. Because Rd increases when the resonant load increases. proportional to RÀ d ).

96 (2008) 2217–2227 2225 4 α (observed) α (propose) 3 alpha 2 1 0 0 1 Rd Fig. Resonant Coefficient a.3 resonant load increases. Since in codes. (1991) which states that the peak factor decreases when the correlation of peaks increases. 5. which means the formulas can be used to estimate wind load on wind turbine towers in complex terrain. Figs. 4.e. 2 with the same wind conditions described in Section 4. Aerodyn.2 Turbulence Intensity Fig. / J. 0.. this load case is investigated.V. the inflow angle is zero).1 0. because an increase of the resonant load means that peaks occur in a certain manner and the correlation of peaks increases. . the largest wind load is considered to be drag force when wind flows from in front of wind turbine (i.ARTICLE IN PRESS L. 2 8000 FEM Bending moment (kNm) This Study 6000 Linear 4000 2000 0. Ind. 5–8 are examples of how these results strongly correlate with the FEM simulation in both low and high turbulence intensity. Comparison of mean load. 5. Verification of proposed model The proposed formulas are used to calculate design wind load on the wind turbine tower described in Table 2 and Fig. This model agrees well with the study by Kitada et al. Binh et al. Wind Eng.

0. / J.2 Turbulence Intensity Fig.3 1.1 0.V.5 3 0. Comparison of maximum load.3 .2 Turbulence Intensity Fig.5 FEM This Study peak factor 4 Linear 3. 6. 7. 0. 8. Comparison of peak factor. 96 (2008) 2217–2227 4000 FEM This study sdev (kNm) 2000 0 0. Wind Eng. 0. Ind.6x104 Bending moment (kNm) FEM This Study Linear 1. Binh et al.3 4.2 Turbulence Intensity Fig.ARTICLE IN PRESS 2226 L. Comparison of standard deviation.1 0.2x104 8000 0. Aerodyn.1 0.

.ARTICLE IN PRESS L. K. References Architectural Institute of Japan (AIJ). Grant no. (3) The formulas have been verified using FEM simulation of a stall-regulated wind turbine. 247–257. Proposed skewness formula. Concluding remarks In this study.. 2005. Gust loading factor—past. present and future. 2002. Doctoral Thesis. September 12–14. Hattori. 1992. 96 (2008) 2217–2227 2227 6. Ishihara. T. J. Aerodyn. Analysis of non-Gaussian surge response of tension leg platforms under wind loads. K. M. Ishihara.. . consists of a theoretical background part and an empirical turbulent part.. The Danish Society of Engineers and The Federation of Engineers.. / J. Nucl. In: Proceedings of the Institute of Civil Engineering.G. which considers the non-linear part of wind pressure. 187–196. Trans. J. Aerodyn. Constr. 2004. turbulence intensity and the resonance–background ratio of wind load. Seoul. Mekaru.. P. Loads and safety of wind turbine construction Danish standard DS472. The followings were obtained: (1) A formula of mean wind load. Ishikawa.. Wind Eng.. Kanda.. Zhou. Nishijima. 1993. Stochastic seismic floor response analysis method for various damping system. A.. 1994. A. 18360212 to Dr. Recommendations for loads on buildings.V. 91. H. A study on wind load estimation method considering dynamic effect for overhead transmission lines. pp. pp. Kareem. Fujino. Kanda. the representative of the research group. ASME 116 (August). Ind. Ind. Struct. J.. 1301–1328. Takahara. (2) A formula of peak factor was proposed to consider a non-Gaussian response of a wind turbine tower by introducing skewness. Eng.. J. Davenport. 599–612. Phuc. Kareem. Y.. which considers the spatial correlation of wind velocity.. 1964. Binh et al.V. Waseda University. A. T. K. T. Wind Eng.. Ogata. Y. 128.. Zhao. Choi. Field test and full dynamic simulation on a stall regulated wind turbine. Eng. Acknowledgment The support for this study was provided by the Ministry of Culture and Education of Japan. 2003. 79–84 (in Japanese). Estimation of peak factor for non-Gaussian wind pressure.. was proposed to evaluate wind load in region of high turbulence intensity. the calculated load’s error is limited to less than five percent. Kitada. Y. an equivalent static method to evaluate wind load on wind turbines has been studied. 2005. Japan (in Japanese). AIJ 557 (July). Note on the distribution of the largest value of a random function with application to gust loading. T. J. Korea. 1991. In: Proceedings of the Sixth Asia-Pacific Conference on Wind Engineering APCWE-VI. Especially in regions of high turbulence.. Des.

Sign up to vote on this title
UsefulNot useful