Assessment of synthetic winds through spectral modeling and validation using FAST

Transcription

1 Jornal of Physics: Conference Series PAPER OPEN ACCESS Assessment of synthetic winds throgh spectral modeling and alidation sing FAST To cite this article: A Chogle et al 2016 J. Phys.: Conf. Ser Related content - The ALICE Software Release Validation clster D Berzano and M Krzewicki - The Geant4 physics alidation repository H Wenzel, J Yarba and A Dotti - Experimental Validation for Hot Stamping Process by Using Tagchi Method Mohd Fawzi Zamri, Syh Kai Lim and Ahmad Razlan Ysoff View the article online for pdates and enhancements. This content was downloaded from IP address on 10/12/2017 at 04:25

2 Assessment of synthetic winds throgh spectral modeling and alidation sing FAST A Chogle, S T Kandkri and H G Beyer Department of Engineering Sciences, Uniersity of Agder, 4879 Grimstad, Norway abhijit.chogle@ia.no; srya.kandkri@ia.no; hans-georg.beyer@ia.no Abstract. In this paper, we analyse the simlated and measred wind data with respect to their spectral characteristics and their effect on wind trbine loads. The synthetic data is generated from a stochastic fll-field trblent wind simlator - TrbSim for netral stability conditions. We first inestigate a model for elocity spectra and, a coherence model, by comparing the model reslts with the measrements. In the second part we analyse the synthetic data ia spectra and coherence for two cases; withot and with adding coherent eents. Finally, we compare wind trbine loads calclated by sing FAST simlation of 5 MW reference wind trbine on the basis of simlated and measred data for the gien mean wind speed. 1. Introdction In order to hae proper design and life assessment of wind trbine, it is important to hae accrate prediction of wind loads. As for the precise strctral modeling of the trbine a fll 3-D representation of the incoming wind is needed. For this prpose tools for the synthesis of the respectie data sets are based on physical nderstanding of the strctre of the incoming wind flow [1, 2] are applied. These tools hae to demonstrate that they represent the flow strctre obsered in the atmospheric bondary layer (ABL) and in application reslt in a realistic dynamic response of the wind trbines. Here the tool TrbSim [3] is sbject to a respectie analysis. The TrbSim offers synthetic time series ia IEC defined spectral models [4]. It can be rn in a basic mode sing basic Forier synthesis based on the information of the freqency and coherence spectra and a mode respecting the intermittent characteristics of the wind flow by sperimposing coherent eents. In the following, two schemes Case 1: withot adding coherent eents to time-series obtained from TrbSim Case 2: with coherent eents added to the time-series, are sed for generating wind field data sets which are sbseqently passed to a dynamic wind trbine model. The reslting data sets are first analyzed for their spectral characteristics and in addition - for their increment statistics (c.f. [5]) giing an indication of non-gassian characteristics of the sets. These data are compared to the respectie characteristics of empirical data as offered by the serer of the National Renewable Energy Laboratory (NREL) Wind trbine test center. Data sed refer to tower based measrements by sonic anemometers at 100 m height aboe grond. Content from this work may be sed nder the terms of the Creatie Commons Attribtion 3.0 licence. Any frther distribtion of this work mst maintain attribtion to the athor(s) and the title of the work, jornal citation and DOI. Pblished nder licence by Ltd 1

3 The aeroelastic compter-aided engineering tool FAST [6] for horizontal axis wind trbines is then applied for simlating the copled dynamic response of a wind trbine. In order to distingish the effect of coherence, we se the NREL s National Wind Technology Center Model (NWTCUP) along with test fnction KHTEST which sperimposes one intense coherent eent in the middle of the otpt time series [4]. The analyses are performed with the objectie to assess whether sperimposition of coherence eents can sfficiently captre the spectral characteristics of real wind conditions. This is frther ealated based on their effect on wind trbine loads. The basic definitions, notation, and description of the spectral and the coherence model are proided in Section 2, along with spectral analysis of measred and synthetic data. The effects from the two cases gien aboe on wind trbine loads are shown in Section 3, along with loads calclated from measred time-series. Finally we conclde or analysis in Section 4 with discssion. 2. Spectral analysis 2.1. Model The NWTCUP model spectra are based on the smooth-terrain model (SMOOTH) deeloped by [7] and [8], and are gien, for diabatic ABL s as fsi M 2 (f) = C 1,iR(φ ɛ /φ m ) 2/3 1 + C 2,i R 5/3, (1) where for ith elocity component, f is the freqency, is the friction elocity, the ratio R = fz/(uφ m ), φ ɛ and φ m are the dimensionless dissipation rate and elocity gradient, respectiely, and U is the mean wind speed at height z. The constants C 1 and C 2 are the scaling parameters for each component and are reported in [4]. TrbSim implements the coherence fnction along with the NWTCUP spectral model, and is described as ( ) r e ( ) fr 2 Coh i (f, r) = exp a i + (b i r) 2, (2) z m U m where r is the separation distance between the points on the grid, e is the coherence exponent inpt parameter, z m and U m are the mean height and mean wind speed between the two points, respectiely. The ariables a i and b i represent the decrement and offset, respectiely for ith elocity component. Letting φ ɛ and φ m eqal to one in netral ABL limit, the model elocity spectra normalized by for the mean wind speed U = 14.8 m/s and z = 100 m, and the model coherence are shown in figre 1 and 1, respectiely. The coherences are shown for r = 50 m, z m = 75 m, U m = 14 m/s, and e = 0. For ariables a and b, we take ales referring to figres (13-15) in [4], and the approximated ales are (a, a, a w ) (12, 8, 5) and (b, b, b w ) (002, 005, 02) Obserations For the spectral analysis and load calclations, we consider the measrements from sonic anemometers measring temperatre and wind speeds at 20 Hz in three dimensions on a NREL s met-mast located in sothwest corner of the NWTC near Bolder, Colorado ( The data is selected for the mean wind speed bin of m/s measred at 88 m height by a cp anemometer. To aoid the wake effects on trblence from the wind trbines, and the met-mast itself, wind are selected for the directions , also measred at 88 m by a wind-ane. The NWTC is sitated 2

4 fs(f )/* 2 [-] w 0.6 w fz/u [-] Coh(2π f, r) [-] π fr/u m [-] Figre 1: The model; normalized elocity spectra as gien in (1), for z = 100 m and U = 14.8 m/s, and coherence gien by (2), for the mean wind speed U m = 14 m/s between two points r = 50 m apart. fs(f )/* 2 [-] fz/u [-] w Coh(k1,Δ z) [-] w k 1 Δz [-] Figre 2: The obsered; elocity spectra at z = 100 m, U 14.8 m/s, and coherence between heights 50 and 100 m, and the mean wind speed 14 m/s between the two heights. fs(f )/* 2 [-] fz/u [-] w Coh(k1,Δ z) [-] k 1 Δz [-] w Figre 3: For Case 1, from synthetic data: elocity spectra for z = 100 m, U = 14.8 m/s, and coherence for z = 50 m between the two heights. 3

5 fs(f )/* 2 [-] fz/u [-] w Coh(k1,Δ z) [-] k 1 Δz [-] w Figre 4: For Case 2, from synthetic data: elocity spectra for z = 100 m, U = 14.8 m/s, and coherence for z = 50 m between the two heights. abot 5 km to the east of the Colorado Front Range, from which the wind inflow is inflenced for the sector The sonic anemometers are installed at the heights of 15, 30, 50, 76, 100 and 131 m. We analyse three years of data from 2012 to More details abot the NWTC trbine inflow tower data, location, and instrmentation can be fond in [9]. We estimate elocity cross-spectra between two heights, as well as the power-spectra from 10-minte time series by sing basic definitions that can be fond in e.g., [1]. The spatial coherences ([4]) between two points are estimated from the cross-spectra. The calclations are carried ot for netrally stratified ABL for Obkho length ([10]) L o 500 according to [11]. The elocity spectra from the obserations at z = 100 m and for the mean wind speed U 14.8 m/s, and the obsered coherences between heights 50 and 100 m, are shown in figre 2 and 2, respectiely. One hndred and fifteen nmber of 10-minte time-series are sed to calclate the elocity spectra and are normalized with the friction elocity 0.9 m/s at 15 m Synthetic data Spectra and coherences We generate synthetic data sing TrbSim for the gien conditions; i.e., z = 100 m, U = 14.8 m/s, = 0.9, and the gradient Richardson nmber Ri = 0. The elocity spectra and the coherence for a 50 m ertical separation for Case 1, are gien in figre 3 and 3, respectiely. The spectra and cross-spectra are aeraged oer one hndred and fifteen random seeds generating corresponding 10-minte time-series. The elocity spectra and coherences for Case 2 are shown in figre 4 for the same wind conditions gien aboe for Case 1, except that the coherent eents added to the synthetic data in Case 2. It is noted that the spectra and coherences from Case 1 are nderestimated, while the inertial sb-range within the spectra are not well settled in Case 2, and also, the coherence from Case 2 are increased at smaller scales (or higher freqencies) Velocity increments As mentioned aboe, it is expected that the increments from the measred data show non-gassian characteristics (see e.g. [12]), a property that is as well gained from analyzing the series generated by Large Eddy Simlation (LES) schemes (see e.g. [5]). The templates for the coherent strctres sed here by the KHTEST scheme stem from that origin. Here we inspect the similarity of the increment statistics of the empirical and synthetic sets generated by applying the defalt setting of the TrbSim KHTEST scheme. 4

6 Figre 5: Probability distribtions of the increments, normalized by standard deiation of increments in (first row), (second row), and w (third row), for 10-minte measred time-series, Case 1: synthetic data from TrbSim withot coherent eents added, and (c) Case 2: synthetic time-series with coherent eent added. Figres in 5 gie the probability distribtions (PDF) of the increments, normalized by standard deiation of increments, on the 20 Hz scale for the three elocity components. The deiations from a Gassian distribtion are obios for all components. To be considered as eqialent the synthetic data hae to show the same characteristics. Obiosly for data sets generated by the basic TrbSim procedre withot inclsion of coherent eents there is a negatie reslt (see figre 5). The inclsion of the coherent eents in the generation procedre reslts in the distribtions of increments as gien in figre 5(c). The PDFs proe the non-gassian characteristics of this synthetic set. Howeer, it is obios that, for a proper reflection of the empirical characteristics parameters that determine the negotiation of the coherent eents hae to be tned. Ths for a general applicable synthetization scheme both, modeling of the PDFs of the increment in dependence of the flow sitation (mean wind speed, trblence intensity, thermal stratification, etc.) and the proper settings of the parameters of the scheme for the inclsion of coherent eents to approach these PDFs hae to be identified. 5

7 The Science of Making Torqe from Wind (TORQUE 2016) Measrements NWTCUP NWTCUP+COH rad FB (kn) z FB (kn) rad MB (knm) Time (sec) (c) Figre 6: Comparison of blade loads. Measred NWTCUP NWTCUP+COH 1000 rad FN (kn) rad MN (knm) Time (sec) Figre 7: Comparison of nacelle loads. 3. Wind trbine load analysis The loads on wind trbine de to arios wind conditions are stdied based on simlations sing FAST analysis tool deeloped by National Renewable Energy Laboratory. FAST is a copled simlation platform inclding wind in-flow, aerodynamics, strctral dynamics and controls for wind trbines. The 5MW reference wind trbine is simlated in onshore configration sing the synthetic data generated from TrbSim as well as the measred data. The trbine is controlled by GH Bladed style dll controller that controls the generator, collectie pitch and yaw systems [13]. In case of synthetic data, the trbine is firstly simlated sing the NWTCUP model as 6

8 Table 1: Ratio of maximm loads seen at blade-root and nacelle bearing Case (1): NWTCUP model, Case (2): NWTCUP model with sperimposed coherence and Case (3): measrement data. Ratio RootF xc1 RootF yc1 RootF zc1 RootM xc1 RootM yc1 RootM zc1 (3)/(1) (3)/(2) Y awbrf xp Y awbrf yp Y awbrf zp Y awbrm xp Y awbrm yp Y awbrm zp (3)/(1) (3)/(2) Table 2: Comparison of damage-eqialent loads Case (1): NWTCUP, Case (2): NWTCUP with sperimposed coherence and Case (3): measred data. Case RootF xc1 RootF yc1 RootF zc1 RootM xc1 RootM yc1 RootM zc1 (kn) (kn) (kn) (knm) (knm) (knm) (1) 2.385e e e e e e + 02 (2) 2.383e e e e e e + 02 (3) 2.530e e e e e e + 02 Y awbrf xp Y awbrf yp Y awbrf zp Y awbrm xp Y awbrm yp Y awbrm zp (kn) (kn) (kn) (knm) (knm) (knm) (1) 2.995e e e e e e + 03 (2) 3.096e e e e e e + 03 (3) 3.211e e e e e e + 03 described earlier, withot sperimposing the coherence eents. Frther, the coherent eents are sperimposed sing the KHTEST option and the magnitde of scales are tned. In case of the measred data, TrbSim Alpha [14] is sed to generate fll-field wind files. The mast data at heights of 15, 30, 50, 76, 100 and 131 m is spplied to the software. TrbSim Alpha calclates fll-field wind data based on the specified trblence spectra and linearly interpolates the elocity and direction profiles to match the specified time series. The effect of coherence on loads in measred data and simlated data are compared in terms of blade root loads and the nacelle bearing loads. The blade root loads corresponding to blade 1 are the radial and axial forces [RootF xc1, RootF yc1, RootF zc1 ] and corresponding moments [RootM xc1, RootM yc1, RootM zc1 ]. Similarly, the nacelle radial and axial forces [Y awbrf xp, Y awbrf yp, Y awbrf zp ] and corresponding moments are [Y awbrm xp, Y awbrm yp, Y awbrm zp ]. In each case, namely, the NWTCUP withot coherence eents, NWTCUP with coherence eents, and the measred data, 115 samples of 10-minte dration are collected and simlated with the wind trbine. For one sch simlation, the blade root loads are showcased in figre 6 and the nacelle bearing loads are shown in figre 7. The reslts from the wind trbine are frther post-processed sing NREL s MLife [15] in order to determine aggregate statistics and damage eqialent loads (DELs) as described in [16]. The damage eqialent loads are a measre of eqialent fatige damage cased by the loads taking into accont the material properties, namely the S-N cres (Wöhler exponents). 7

9 The ltimate loads are determined sing MExtremes [17]. As this is a large set of data, only specific information releant to the analysis is presented here. Across the data sets, the worst case (maximm loads) are compared for the synthetic data against the measred data. These ratios for blade root and nacelle forces and moments are calclated from the aggregate statistics of time-series data of all the simlations. In these calclations howeer, the first minte of simlation is exclded in order to accont for the initial conditions. The ratios are presented in Table 1. It can be noted that the NWTCUP with sperimposed coherence eents prodces load effects more conseratiely. It is to be noted that the scaling parameters for the coherence eents need to be tned for the gien measrement data in order to obtain a realistic representation. The damage-eqialent loads for the three cases are showcased in Table 2. These are the aggregate DEL s oer 115 test cases, that is, 50, 000 eqialent cycles. The S-N slope 10 is chosen for the composite blade and 3 for steel nacelle. Similar to the aggregate load statistics, the NWTCUP with sperimposed coherence eents reslts in a conseratie DEL s compared to its conterpart. Besides, it is eident that finer tning of the coherent eent scales are necessary for a closer fit. 4. Discssion and conclsion It is seen that the spectral model gien by (1) has -spectrm slightly higher than the obsered (figres 1 and 2), whereas the coherence model in (2) haing the elocity coherences slightly nderestimated. The model show the w-coherence smaller at low freqencies and larger at higher freqencies, which can also be seen in the obsered coherence (figres 1 and 2). Both the elocity spectra and coherences are obsered to be increased (at higher freqencies) de to the addition of coherent eents in the synthetic data in Case 2, whereas those are nderestimated in Case 1. Althogh the basic models that are sed in simlations show reasonable agreements, the reslts from the synthetic data depart from the obserations. The spectra and coherence reslts from Case 2 show that the coherence eents which are added to the time-series need to be scale-dependent. From the probability distribtions of the increments, qalitatiely the coherent eents lead to deiations from Gassian characteristics, bt not directly matching the deiations extracted from the measred sets (and conseqently will reslt in different load characteristics when applied). We analysed the wind trbine loads ia FAST simlations for two cases, with and withot coherent eents added to the synthetic data, in comparison to the loads calclations based on the measred data. The synthetic data withot sperimposed coherent eents can be seen to nderestimate the loads as opposed to the case with sperimposed spectral eents. Howeer, the scaling parameters for the coherent eents need to be tned for the gien measrement data. Similar effect was also noticed in the case of calclation of damage eqialent loads. Finally, topics for frther research we conclde are to be identified here: For general applicability, parameters of models sed (e.g., [18]) for the characterizations of the distribtions of the increment hae to be linked to basic parameters of the wind field (e.g. mean wind speed, trblence intensity, thermal stability, etc.) The procedre for the implementation of the coherent eents mst be extended to be tnable to generate the desired freqency dependent statistical characteristics ([12]), as well as wind trbine loads The stdy can be extended for non-netral ABL s indicating possible optimized scales to which the coherent eents are added to the synthetic time-series in order the maximm loads and moments to be more close to the reality, along with stdying a site-specific occrrences of different atmospheric stabilities. 8

10 Measres for testing of the eqialence of statistics of synthetic to measred winds beyond spectral power density and coherences are still incomplete. Acknowledgements This work has been fnded by Norwegian Centre for Offshore Wind Energy (NORCOWE) nder Grant /S60 from Research Concil of Norway (RCN). NORCOWE is a consortim with partners from indstry and science, hosted by Christian Michelsen Research. References [1] Chogle A, Mann J, Kelly M, Sn J, Lenschow D H and Patton E G 2012 J. Trbl. 13(36) 1 13 [2] Chogle A, Mann J, Segalini A and Dellwik E 2015 Wind Energy [3] NWTC Information Portal (TrbSim) Accessed 25-Febrary-2016 & Last modified 31-Janary-2016 [4] Jonkman B J and Kilcher L 2012 Trbsim ser s gide: Version 6.00 Technical Report NREL/TP National Renewable Energy Laboratory [5] Berg J, Natarajan A, Mann J and Patton E G 2016 Wind Energy DOI: /we.1963 [6] NWTC Information Portal (FAST) Accessed 25-Febrary-2016 & Last modified 31-Janary-2016 [7] Høstrp J 1982 J. Atmos. Sci [8] Olesen H R, Larsen S E and Høstrp J 1984 Bondary-Layer Meteorology [9] Clifton A, Schreck S, Scott G, Kelley N and Lndqist J 2013 J. Sol. Energy Eng. 135(3) [10] Kaimal J C and Finnigan J J 1994 Atmospheric bondary layer flows (New York: Oxford Uniersity Press) [11] Gryning S, Batcharoa E, Brümmer B, Jørgensen H and Larsen S 7 Bond. Layer Meteorol [12] Böttcher F, Barth S and Peinke J 7 Stoch. Eniron. Res. Ris. Assess [13] Jonkman J, Btterfield S, Msial W and Scott G 9 Definition of a 5-mw reference wind trbine for offshore system deelopment Technical Report NREL/TP National Renewable Energy Laboratory [14] 2016 NWTC Information Portal -TrbSim Alpha URL [15] 2015 NWTC Information Portal -MLife URL [16] Freebry G and Msial W 0 Determining eqialent damage loading for fll-scale wind trbine blade fatige tests Conference Paper NREL/CP National Renewable Energy Laboratory (NREL) [17] 2015 NWTC Information Portal - MExtremes URL [18] Labrich T 9 Statistical analysis and stochastic modeling of atmospheric bondary layer wind Phd thesis Bergische Uniersität 9