Sie sind auf Seite 1von 10

Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-body Dynamics http://pik.sagepub.

com/

Mode acceleration approach for rotating wind turbine blades


P J Murtagh, B Basu and B M Broderick Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-body Dynamics 2004 218: 159 DOI: 10.1243/1464419042035962 The online version of this article can be found at: http://pik.sagepub.com/content/218/3/159

Published by:
http://www.sagepublications.com

On behalf of:

Institution of Mechanical Engineers

Additional services and information for Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-body Dynamics can be found at: Email Alerts: http://pik.sagepub.com/cgi/alerts Subscriptions: http://pik.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations: http://pik.sagepub.com/content/218/3/159.refs.html

>> Version of Record - Sep 1, 2004 What is This?

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

159

Mode acceleration approach for rotating wind turbine blades


P J Murtagh, B Basu and B M Broderick Department of Civil, Structural and Environmental Engineering, Trinity College Dublin, Ireland

Abstract: This paper presents a time-domain analytical approach for evaluating the displacement response in the apping direction of rotating wind turbine blades subjected to stationary stochastic wind loading. This approach includes the effects of rotational sampling of the wind spectra on the blades. The blades are modelled as discrete multi-degree-of-freedom systems, consisting of tapered beams of rectangular hollow crosssection built using the nite element software code ANSYS. ANSYS is used to obtain the stiffness matrices of the blades, allowing the free vibration characteristics of the rotating blades to be determined by analytical formulation. A geometric stiffness matrix is included in the free vibration solution allowing for the centrifugal stiffening effects experienced by the blades to be included. Once the natural frequencies and mode shapes are obtained, the mode acceleration technique is employed to predict nodal responses to the prescribed wind loading. Nodal wind loading is obtained by generating articial drag force time-histories, which have elevated energy at frequencies corresponding to integer multiples of the blade rotational frequency, in order to replicate realistic blade loading conditions. Blade nodal displacements based on rotationally sampled and nonrotationally sampled spectra are obtained for varying rotational frequencies. Blade response is also obtained in the frequency domain and used to verify the time-domain results. Keywords: wind loading, mode acceleration method, rotationally sampled spectra, wind turbine blades

INTRODUCTION

To estimate the dynamic response of a structure, analysis through the frequency-domain is traditionally employed. For example, Madsen and Frandsen [1] investigated the failure of wind turbines in the frequency-domain, obtaining the structural response for use in lifetime prediction of the system. However, a time-domain approach may be favourable in some instances, especially with regard to nonlinearity and coupling of the structural response. This paper presents such an approach for the linear forced vibration analysis of rotating wind turbine blades in the apping direction subjected to wind loading time-histories generated from stationary stochastic wind processes. Short of possessing actual eld measurements of uctuating wind velocities, the wind engineer is tasked with the generation of articial time-histories. The generation of time-histories has long been of interest in wind and earthquake engineering, and a signicant amount of literature now exists on the topic. Three methods are generally employed; the rst is based on the Fourier transform, the
The MS was received on 11 December 2003 and was accepted after revision for publication on 23 July 2004.
Corresponding author: Department of Civil, Structural and Environmental Engineering, Trinity College Dublin, Dublin, Ireland.

second on the wavelet transform, and the third on an autoregressive moving average (ARMA) based timeseries approach. Kumar and Stathopoulos [2] simulated wind pressure time-histories with both Gaussian and nonGaussian distributions, on low building roofs, by using a fast Fourier transform (FFT) based algorithm, which made use of a previously measured pressure spectrum. Kitagawa and Nomura [3] recently used wavelet theory to generate wind velocity time-histories by assuming that eddies of varying scale and strength may be represented on the time axis by wavelets of corresponding scales. Time-histories were obtained by employing the inverse wavelet transform and the articially created wind characteristics were compared to those of natural wind, showing good agreement. Minh et al. [4] investigated the time-domain buffeting of long-span bridges by using the digital ltering ARMA method to numerically generate time-histories of wind turbulence. This method allowed for the simultaneous generation of two wind velocity component time-histories, while allowing for the inclusion of spatial correlation. Li and Kareem [5] simulated a multivariate nonstationary random process by use of spectral decomposition, incorporating an FFT-based algorithm. Hao et al. [6] generated spatially correlated ground motions using coherency data obtained from the SMART-1 earthquake array in Taiwan.
Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

K03103 # IMechE 2004

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

160

P J MURTAGH, B BASU AND B M BRODERICK

The technique of time-history simulation proposed by the authors stems from the fact that any random signal with a varying frequency content may be represented by a discrete Fourier transform (DFT) encompassing that frequency content. The Fourier coefcients associated with this DFT are obtained as randomly generated numbers with zero mean and a specic standard deviation. This standard deviation is related to the energy content within a discretized power spectral density (PSD) function. Thus, in order to randomly generate a time-history, a spectrum must rst be available. A wind turbine rotor system consists of a number exible rotating blades connected to a central hub. The blades capture the kinetic energy contained in the wind and ultimately transmute this energy into an electrical form. Connell [7] reported that a rotating blade is subjected to an atypical uctuating wind velocity spectrum, known as a rotationally sampled spectrum. Owing to the rotation of the blades, the spectral energy distribution is altered, with variance shifting from the lower frequencies to peaks located at integer multiples of the rotational frequency. Kristensen and Frandsen [8], following on from earlier work carried out by Rosenbrock [9], developed a simple model to predict the power spectrum associated with a rotating blade, and this was signicantly different to a spectrum where the rotation is not considered. Madsen and Frandsen [1] used a rotationally sampled spectrum to obtain the structural response of rotating blades, and this phenomenon has also been validated experimentally by Verholek [10], and by Hardesty et al. [11]. In order to predict the displacement response of a rotating wind turbine blade, the free vibration characteristics for blade motion in the apping direction must rst be obtained. Naguleswaran [12] presented an approach to determine the free vibration characteristics of a continuous spanwise rotating beam subjected to centrifugal stiffening. In reference [12], the mode shape equation was solved using a power series. However, the blades used in this paper are modelled as discrete systems allowing the free vibration properties to be obtained by traditional analytical means. The wellknown mode acceleration method is utilized in the forced vibration analysis. Williams [13] is credited with the rst use of the mode acceleration method and Craig [14] concedes that it has superior convergence characteristics compared to the mode-displacement method. The advantage of this method over the mode displacement technique stems from the fact that the inclusion of the higher modes are not necessary in response estimation, as the pseudostatic component of the response adequately accounts for them. Akgun [15] presented an augmented algorithm based on the mode acceleration method, which has improved convergence for computation of stresses in large models. The mode acceleration approach requires only the rst few modes to be obtained while yielding response parameters for arbitrary nodal complexity. This paper presents a method to determine the forced vibration response of rotating wind turbine blades in the time-domain subjected to randomly generated aerodynamic
Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

force time-histories, including the effects of centrifugal stiffening on the blades and rotationally sampled turbulence. The blades are modelled as multi-degree-of-freedom (MDOF) entities and forces are assumed not to be correlated at each node, due to the intensity of turbulence prevailing. Research into the forced vibration of wind turbine blades and the effects of rotational sampling has not been widely reported in the literature. Two illustrative examples are thus provided, in which the displacement response of blades rotating with frequencies of 1.57 rads21 (15 rev/min) and 3.14 rads21 (30 rev/min), subjected to rotationally sampled and nonrotationally sampled time-histories are compared. A frequency-domain analysis is also carried out to verify the results obtained using the time-domain based mode acceleration method.

2 2.1

THEORETICAL CONSIDERATIONS Formulation of mode acceleration method

The mode acceleration method is used to obtain the response of a MDOF system to arbitrary loading. The equation of motion characterizing the dynamic behaviour of a discretized system is represented by M{ (t)} C{_ (t)} K{u(t)} {P(t)} u u (1)

where [M] is the mass matrix, [C] is the damping matrix, [K] is the stiffness matrix and fu(t)g, {_ (t)} and { (t)} are u u the displacement, velocity and acceleration vectors respectively, and fP(t)g is some time-varying load vector. The blades are modelled as discrete multi-degree-of-freedom systems, consisting of tapered beams of rectangular hollow cross-section built using the nite element (FE) method. The stiffness matrices of the blades are obtained by rst evaluating the exibility matrix [Q] using the unit load method from the FE model. A unit horizontal load is placed at a node i, and the displacement at node j, is calculated, which is the fij element of the matrix [Q]. The stiffness matrix [K], is [Q]21. This technique allows the formulation of the stiffness matrix corresponding to certain nodes of interest in the full model and thus can be used to develop a reduced order model. The mass matrix at the discrete nodes of interest may be formulated as a diagonal matrix  with the ith nodal mass mi. The modal mass matrix M  FT MF, is calculated using the transformation M where F is the modal matrix consisting of n modes. The  modal damping matrix, C may be evaluated by using  the modal transformation with diagonal elements 2jj vj Mj , where jj is the modal damping of mode j and vj is the modal frequency, assuming the system to be classically   damped. The modal stiffness matrix, K is then K FT KF. The time-varying nodal force vector fP(t)g of dimension n 1 can be converted to modal force vector   {F} by using the transformation {F} FT {P(t)}.
K03103 # IMechE 2004

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

MODE ACCELERATION APPROACH FOR ROTATING WIND TURBINE BLADES

161

The equation that describes the response of the structure using the mode acceleration method is M X2zj  1 {u(t)} K {P(t)} _ Fj {hj (t)} vj j1 ! M X 1 (2) Fj {hj (t)} v2 j j1 where {u(t)} denotes the nodal displacement vector, Fj is the jth mode shape and {hj (t)} and {hj (t)} are the rst _ and second temporal derivatives of the modal coordinate. The rst term of equation (2) represents the pseudostatic response, while the second and third terms represent the dynamic response. A linear MDOF system ultimately comprises n single degree-of-freedom (SDOF) systems, allowing the total response of the MDOF entity to be obtained as the linear combination of the response of these SDOF entities. Hence, the response expressed by equation (2) may be obtained by considering a series of SDOF systems of natural frequency vj and damping zj , and obtaining solutions for {hj (t)} and {hj (t)}. The generalized (modal) coordinate _ {hj (t)}, along with its rst and second temporal derivates may then be used to evaluate {u(t)}. Nigam and Jennings [16] presented an algorithm to obtain the modal coordinate and its rst derivative, for a viscously damped SDOF oscillator subjected to a base acceleration. They assumed that the input acceleration time-history could be approximated by a piecewise linear function, thus yielding a semi-analytical process allowing the displacement and velocity response time-histories to be computed. The fundamental equation based on this algorithm is expressed as ' !& ! & ' {hi1 } c b12 b { hi } c 11 11 12 {hi1 } _ { hi } _ c21 c22 b21 b22 & ' {ai } (3) {ai1 } where the subscript i denotes the ith time instant, {h(t)} is the generalized coordinate, ai is the acceleration value at time i and c11, c12, c21, c22, b11, b12, b21, b22, are frequency and damping dependant constants, expressions for which may be found in the Appendix. The modal acceleration timehistory {aj (t)}, which is thus made up of individual acceleration values of ai , may easily be obtained for each SDOF entity as {aj (t)}  {F j (t)}  Mj (4)

2.2 Response of rotating blades in the time domain In order to predict the response of a rotating blade, its free vibration characteristics must rst be obtained. It is assumed that the individual blades are geometrically and physically similar, hence only one blade needs to be dynamically analysed. The equation of motion that describes the free vibration motion of a discrete system is M{ (t)} K{u(t)} 0 u (6)

Wind turbine blades usually have very complex geometries with mass and stiffness distributions varying along the length of the blade and about the apping and lead/lag axes. Hence, an FE software code was used to create a realistic blade geometry, in which mass and stiffness varied along the length of the blade. The FE code used for this purpose was ANSYS. A beam element, designated Beam44, was used to develop the model. This element contains six degrees-of-freedom at each node, and allows for a different cross-sectional area and second moment of area at each node, facilitating a tapering of the model. Figure 1 illustrates the model as created in ANSYS, clearly showing the number of elements present after discretization. The exibility matrix pertaining to the model was obtained using the static analysis capability of ANSYS. The inverse of this yielded the models stiffness matrix. This stiffness matrix has been further modied in this paper in order to nd the free vibration properties of the model, which includes the effects of centrifugal stiffening. The inclusion of these effects is not permitted with ANSYS. The free vibration properties of the model undergoing rotation may be obtained from   DETK0 v2 M 0 j (7)

where K0 K KG represents the modied stiffness matrix due to the geometric stiffness matrix accounting

  where {F j (t)} is the modal force time history and M j is the modal mass of mode j. The second derivative of the generalized coordinate is given by  {F j (t)} {hj (t)} 2jj vj {hj (t)} v2 {hj (t)} _ (5) j  Mj
K03103 # IMechE 2004

Fig. 1

Tapered blade as created in ANSYS

Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

162

P J MURTAGH, B BASU AND B M BRODERICK

for the effect of centrifugal stiffening, and vj is the desired natural frequency. The geometric stiffness matrix contains force contributions due to blade rotation, which are always tensile, and contributions from the self weight of the blade, which may be both tensile and compressive, depending on blade position. The geometric stiffness matrix is obtainable as N1 6 l1 6 6 N 6 1 6 6 l1 KG 6 6 . 6 . 6 . 6 6 4 0 2 N1 l1 N1 N2 l1 l2 . . . 0 3 .. . 7 7 7 7 0 7 7 7 Nm1 7 7 7 lm1 7 7 Nm1 Nm 5 lm1 lm 0

obtained as normally distributed random numbers, generated with zero mean and standard deviation si . The uctuating velocity signal is hence composed of a number of contributions from a discretized form of a continuous frequency band. Time-histories were simulated using equation (11) and a modied version of the uctuating wind velocity spectrum offered by Kaimal et al. [17] expressed as fSvv (H,f ) 200n 2 v (1 50n)5=3 (12)

(8)

Nm1 lm1

where Ni is the axial force at node i, li is the length of beam segment between the nodes i and i 1, and m is the total number of nodes. The magnitude of the tensile centrifugal axial force at node i, CTi, may be obtained from the expression CTi 0:5V2 (MB LB mi xi ) (9)

In equation (12), H is the hub elevation, f is frequency (Hz), Svv(H, f ) is the power spectral density function of the uctuating wind velocity as a function of hub elevation and frequency, v is the friction velocity (m/s), and n is known as the Monin coordinate. The latter two terms may be obtained from the expressions 1 H  v(H) v ln k z0 fH n  v(H) (13) (14)

where V is the blade rotational frequency, MB is the total mass of the blade, LB is the total length of the blade, mi is the cumulative mass at node i, and x is the distance of node i from the centre of rotation. The nodal axial force due to gravity (self weight), Gi, may be obtained from geometry and depends on the angle of the blade to the horizontal. Values of Ni are obtained from the expression Ni CTi + Gi (10)

with the sign convention that tensile forces are positive and compressive forces are negative. To perform a forced vibration analysis, the nodal forces, {P(t)} must rst be obtained. This may be facilitated by virtue of the fact that any arbitrary uctuating velocity signal v(t), with zero mean, may be represented by a DFT with a discretized version of a continuous frequency content, as v(t)
1 X k1

ak cos (vk t)

1 X k1

bk sin (vk t)

(11)

where ak and bk are the Fourier coefcients, vk is the kth discretized circular frequency (v 2pf , f is frequency in Hz) and t is the time instant. This velocity signal is generated in conjunction with a wind velocity PSD. The PSD function is conceptually divided into m frequency bands of size df. The area under the PSD function between the limits of fi and fi df is equal to the variance of the signal s2 , at the discrete i frequency fi. The Fourier coefcients in equation (11) are
Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

 with v(H) being the mean wind velocity at hub height, k is von Karmans constant (typically around 0.4), and z0 is the roughness length. This spectrum quanties the energy contained within a turbulent ow of air as a function of frequency. This spectrum, while applicable to stationary bodies, does not accurately reect the energy within a turbulent ow around rotating bodies, such as turbine blades. A modied spectrum, known as a rotationally sampled spectrum, must be used instead. Rotationally sampled spectra are used to quantify kinetic energy as a function of frequency for rotor blades within a turbulent wind ow. In order to represent this redistribution of spectral energy, the following procedure was adopted. The required redistribution of spectral energy was facilitated by identifying the specic frequencies vk , in equation (11), equalling 1V, 2V, 3V, and 4V, and then deriving their Fourier coefcients ak and bk according to specic standard deviation values. These values are obtained by virtue of a number of assumptions. First, the variance of wind velocity spectral energy at height H was calculated numerically using equation (12), and based on this value, a value of variance was allocated to each node along the length of the blade. Madsen and Frandsen [1] observed that the peaks of redistributed spectral energy in a rotationally sampled spectrum tend to become more pronounced as distance increases along the beam, away from the hub. It is therefore assumed that the variance values increase by an arbitrary value of p per cent, for each successive blade node radiating out from the hub. It is also assumed that 30 per cent of total variance values at each node are localized into peaks at 1V, 2V, 3V, and 4V. This 30 per cent energy is subsequently divided into
K03103 # IMechE 2004

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

MODE ACCELERATION APPROACH FOR ROTATING WIND TURBINE BLADES

163

four peaks, that is 15, 7.5, 4.5 and 3 per cent per peak, with the maximum energy being localized at a frequency of 1V, descending to a minimum value at a frequency of 4V. This procedure is illustrated in Fig. 2. Thus, nodal uctuating velocity time-histories vi(t), with specic energy frequency relationships were obtained using equation (11). The total nodal drag force time history Pi (t) is expressed as v Pi (t) 0:5CD Ai r i vi (t)2 (15)

Hj (v), associated with each individual mode of vibration j at a frequency v is obtainable as Hj (v) Fj {hj (v)} (17)

where hj (n) is the jth modal coordinates. The transfer function in equation (17) relating the modal force to the modal displacement of the blade is further expressed as )  F unit, j Hj (v) Fj  M j ( v2 2{jj vvj v2 ) j (

(18)

where CD is the drag coefcient, Ai is the model blade area  and r is the air density. The mean wind velocity at node i, vi , will follow a sinusoidal variation with time due to change in blade position as  vi V cos (Vt) (16)

where V is the required amplitude necessary to represent the mean wind velocity as a function of nodal position or height above the ground. The total nodal drag force time-histories may then be incorporated into the mode acceleration method through equation (2).

 where F unit,j is the jth modal force at the tip of the blade. The output may be obtained as the product of the frequencydomain representation of the modal force with the corresponding modal transfer function. An inverse Fourier transform of this frequency-domain output will give the blade response in the time-domain. 3 NUMERICAL EXAMPLES

2.3

Response of rotating blades in the frequency-domain

The response of a rotating blade may also be obtained through the frequency-domain and this response may subsequently be compared to that obtained in the timedomain. The frequency-domain approach is based on the well-own systems methodology, involving an input output relationship. The output (response) is generally related to the input (loading) by a transfer function, so called because it transforms the energy input into a system into the energy output from the system. A frequencydomain representation of the input, in this case a modal force time-history, can be obtained by use of a Fourier transform of that time-history. The transfer function,

The rotating blade considered has a length (LB) of 30 m and is tapered. The blade has a width of 1.1 m at both ends and a width of 3.0 m at its widest intermediate point. It is of rectangular hollow cross-section, with a depth or 0.4 m, a wall thickness of 0.01 m, and has an elastic modulus of 65 109 N/m2 and density of 2100 kg/m3. The rotational blade velocities considered are 0 rads21 (stationary), 1.57 rads21 (15 rev/min) and 3.14 rads21 (30 rev/min). The parameters associated with the Kaimal spectrum are k 0.4, H  60 m, v(H) 20 m=s, and z0 0.08 m. The blades are modelled as eight degree-of-freedom systems and the rst three modes are included in the response. Owing to the uncertainty surrounding levels of damping in the system, modal structural damping ratios of 1 per cent were assumed, and the presence of aerodynamic damping was not considered. However, it may be noted that the inclusion of aerodynamic damping will alter the magnitude of the response of the blade, which is not investigated in this paper. The coefcient of drag used was 2.0, and r was 1.225 kg/m3. 3.1 Free vibration analysis of the rotating blade

Fig. 2

Rotationally sampled spectrum showing distribution of energy

The reduced order model outlined in section 2.2 is primarily used to obtain these properties for a series of different blade rotational frequencies. This method allows for the inclusion of centrifugal stiffening due to blade rotation and the effects of this phenomenon are explored. The modal analysis capability of the FE code ANSYS was also employed to obtain the free vibration properties, although this method does not include the effects due to rotation. Three rotational frequencies were considered with the reduced order model, 0 rads21 (stationary), 1.57 rads21 and 3.14 rads21. Table 1 presents the values of natural frequency obtained from ANSYS and from the reduced
Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

K03103 # IMechE 2004

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

164

P J MURTAGH, B BASU AND B M BRODERICK

Table 1 Natural frequencies of rotating blade


Rotating blade natural frequencies (Hz) Mode 1 V (rads21) 0 1.57 3.14 ANSYS 0.696 ROM 0.615 0.656 0.762 Mode 2 ANSYS 3.998 ROM 3.578 3.605 3.682 Mode 3 ANSYS 10.983 ROM 9.774 9.804 9.880

order model (ROM). The two approaches may be compared at 0 rads21, with the reduced order model showing about a 10 per cent difference from the ANSYS results. It is worth noting that the ANSYS results contain about three times the degrees-of-freedom than the reduced order model, ensuring a greater level of accuracy. From Table 1, it is evident that as the rotational speed of the blades increases, their natural frequencies increase, or the blade gets stiffer.

Fig. 4 Fourier transform of articially generated velocity time-history for beam rotating with V 3.14 rads21

3.2

Forced vibration analysis of rotating blade

The mode acceleration method as previously outlined along with the frequency domain approach were used to estimate the forced vibration response of the rotating blade. Figure 3 illustrates the randomly generated total nodal drag force at the blade tip. Figure 4 presents a rotationally sampled spectrum obtained from a Fourier transform of the generated velocity time-history and clearly indicates the positions of the redistributed energy peaks. Figures 5 to 8 illustrate the response time-histories of a blade with rotational frequency of 1.57 rads21 subjected to articially generated force time-histories, obtained using the mode acceleration method and the frequency-domain method. Both methods yield almost identical response time-histories, apart from the initial portions. This is due to the presence of an initial transient component in the timedomain approach, which soon decays to zero. The excellent matching of the responses obtained by the two different methods is clearly illustrated in Figs 6 and 8 where a stretch

Fig. 5 Tip displacement for blade with V 1.57 rads21 obtained by considering rotationally sampling

Fig. 3 Total nodal drag force acting at the blade tip


Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

Fig. 6 Comparison of methods over 90110 s for V 1.57 rads21 obtained considering nonrotationally sampling
K03103 # IMechE 2004

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

MODE ACCELERATION APPROACH FOR ROTATING WIND TURBINE BLADES

165

Fig. 7

Tip displacement for blade with V 1.57 rads21 obtained considering nonrotationally sampling

Fig. 9 Tip displacement of blade with V 3.14 rads21 obtained by considering rotationally sampling

between 90 and 110 s is shown in a magnied manner. The response in Figs 5 and 6 includes the effects of rotational sampling, but that in Figs 7 and 8 does not. The nodal loading time-histories used to generate Figs 5 and 6 have the same total energy. From these gures, it may be observed that the minimum blade tip displacement occurs due to the nonrotationally sampled loading, and is approximately 1.7 m. The maximum blade tip displacement due to rotationally sample loading is approximately 2.1 m. Figures 9 and 10 illustrate the response time-histories for a blade rotating with a frequency of 3.14 rads21, subjected to articially generated force time-histories, also obtained using the mode acceleration method and the frequencydomain method. Again, both methods correlate closely after the decay of the initial transient component in the time-domain. Figure 9 includes the effects of rotational sampling and Fig. 10 was obtained without rotational sampling. Again, the nodal loading time-histories used to generate Figs 9 and 10 have the same total energy characteristics. From these two gures, it is evident that the maximum blade tip displacement again occurs due to

Fig. 10 Tip displacement of blade with V 3.14 rads21 obtained by considering nonrotationally sampling

the rotationally sampled loading, and is approximately 1.5 m. The maximum blade tip displacement due to nonrotationally sampled loading is approximately 1.0 m. 4 CONCLUSIONS

Fig. 8

Comparison of methods over 90 110 s for V 1.57 rads21 obtained considering rotationally sampling

This work presents a time-domain forced vibration analysis of rotating wind turbine blades, subjected to random wind loading. The blades are modelled as MDOF systems, modelled using the FE technique, and subsequently transformed to a reduced order model. The blade natural frequencies and mode shapes are obtained using this reduced order model, which allows for the inclusion of the effects of centrifugal stiffening due to rotation. Traditional FE software codes do not account for this phenomenon. For the case of a zero rotational frequency, blade free vibration properties obtained using the reduced order model can be compared to those obtained using the FE model. Nodal loading time-histories for the rotating blades were articially generated and included specic energyfrequency relationships, as observed from
Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

K03103 # IMechE 2004

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

166

P J MURTAGH, B BASU AND B M BRODERICK

rotationally sampled wind velocity spectra. The mode acceleration method was then used to predict nodal blade displacements. A frequency-domain based response solution was also presented and yielded almost identical displacement timehistories when compared to the time-domain results, after the initial transient component of the latter had decayed. Blade nodal displacement time-histories were obtained for similar blades with different rotational frequencies, using loading time-histories that contained rotational sampling effects and these were compared with displacements found due to loading that did not consider rotational sampling. For similar energy levels, blade displacement found considering rotational sampling was observed to be higher. The time-domain forced vibration analysis approach presented in this paper requires only the rst few modes and produces nodal responses to any desired complexity of discretization. This approach will also allow response coupling due to different types of modes to be incorporated. The blade responses obtained in this paper will primarily indicate behavioural trends. The uncertainty contained within levels of structural damping and in particular aerodynamic damping available to rotating wind turbine blades indicates that further research needs to be carried out in this regard. Also, a detailed FE analysis based on solid modelling could create a blade model based on more realistic blade geometry that could be used with the mode acceleration approach. REFERENCES
1 Madsen, P. H. and Frandsen, S. Wind-induced failure of wind turbines. Engng Structures, 1984, 6(4), 281 287. 2 Kumar, S. K. and Stathopoulos, T. Computer simulation of uctuating wind pressures on low building roofs. J. Wind Eng. Ind. Aerodyn., 1997, 69 71, 485 495. 3 Kitagawa, T. K. and Nomura, T. A wavelet-based method to generate articial wind uctuation data. J. Wind Eng. Ind. Aerodyn., 2003, 91, 943 964. 4 Minh, N. N., Miyata, T., Yamada, H. and Sanada, Y. Numerical simulation of wind turbulence and buffeting analysis of long-span bridges. J. Wind Eng. Ind. Aerodyn., 1999, 83, 301315. 5 Li, Y. and Kareem, A. Simulation of multivariate nonstationary random processes by FFT. J. Engrg. Mech., ASCE, 1991, 117(5), 1037 1058. 6 Hao, H., Oliveira, C. S. and Penzien, J. Multiple-station ground motion processing and simulation based on smart-1 array data. Nuclear Engrg. Design, 1989, 111, 293 310. 7 Connell, J. R. Turbulence spectrum observed by a fastrotating wind turbine blade, Rep. PNL-3426, 1980 (Battelle Pacic Northwest Laboratory, Richland, WA). 8 Kristensen, L. and Frandsen, S. Model for power spectra of the blade of a wind turbine measured from the moving frame of reference. J. Wind Eng. Ind. Aerodyn, 1982, 10, 249 262. 9 Rosenbrock, H. H. Vibration and stability problems in large turbines having hinged blades, Rep. C/T 113, 1955 (ERA Technology Ltd, Surrey, UK). 10 Verholek, M. G. Preliminary results of a eld experiment to characterise wind ow through a vertical plane, Rep. PNL-2518, 1998 (Battelle Pacic Northwest Laboratory, Richland, WA).
Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

11 Hardesty, R. M., Korrel, J. A. and Hall, F. F. Lidar measurement of wind velocity spectra encountered by a rotating turbine blade, NOAA Tech. Memo, 1981. 12 Naguleswaran, S. Lateral vibration of a centrifugally tensioned uniform Euler Bernoulli beam. J. Sound Vibration, 1994, 176(5), 613 624. 13 Williams, D. Dynamics loads in aeroplanes under given impulsive loads with particular reference to landing and gust loads on a ying boat, RAE Reports SMU 3309 and 3316, 1945. 14 Craig, R. R. Structural Dynamics, 1981 (John Wiley and Sons, New York). 15 Akgun, M. A. A new family of mode-superposition methods for response calculations. J. Sound Vibration, 1993, 167(2), 289 302. 16 Nigam, N. C. and Jennings, P. C. Digital calculation of response spectra from strong motion earthquake records, 1968 (Earthquake Engineering Research Laboratory Report, California Institute of Technology). 17 Kaimal, J. C., Wyngaard, J. C., Izumi, Y. and Cote, O. R. Spectral characteristics of surface-layer turbulence. J. Roy Meteorol. Soc., 1972, 98, 563 589. 18 Paz, M. Structural DynamicsTheory and Computation, 4th edn, 1997 (Chapman & Hall, London).

APPENDIX Constants associated with section 2.1 c11 exp (vj jj tinc cos (vj,D tinc ) ' vj jj sin (vj,D tinc ) vj,D exp (vj jj tinc ) c12 { sin (vj,D tinc )} vj,D c21 c22 v2 exp (vj jj tinc ) j { sin (vj,D tinc )} vj,D exp (vj jj tinc ) cos (vj,D tinc ) v j jj sin (vj,D tinc )g vj,D

(19) (20) (21)

(22)

b11

" ( ! v2 v2 j2 1 j,D j j 2 exp (vj jj tinc ) 2 vj vj,D tinc vj jj vj tinc ! 2vj,D jj vj sin (vj,D tinc ) vj,D tinc v2 j ) # 2jj vj vj,D cos (vj,D tinc ) (23) w2 j " ( ! v2 v2 j2 1 j,D j j 2 exp (vj jj tinc ) vj vj,D tinc v2 j ) ! 2vj,D jj vj sin (vj,D tinc ) cos (vj,D tinc ) v2 j # 2jj vj vj,D (vj,D tinc ) (24) v2 j
K03103 # IMechE 2004

b12

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

MODE ACCELERATION APPROACH FOR ROTATING WIND TURBINE BLADES

167

b21

1 exp ( vj jj tinc ){(vj jj v2 tinc ) j v2 vj,D tinc j sin (vj,D tinc ) (vj,D cos (vj,D tinc )} vj,D (25)

in which vj is the jth natural frequency, vj,D is the jth damped natural frequency given by q v j,D vj 1 j2 j (27)

b22

1 2 exp ( vj jj tinc ){jj vj sin (vj,D tinc ) vj vj,D tinc vj,D cos (vj,D tinc )} vj,D (26)

where jj is the jth modal damping ratio, and tinc is a uniform time increment at the ith time instant, ti.

K03103 # IMechE 2004

Proc. Instn Mech. Engrs Vol. 218 Part K: J. Multi-body Dynamics

Downloaded from pik.sagepub.com at University of Liverpool on December 5, 2011

Das könnte Ihnen auch gefallen