ABSTRACT INTRODUCTION

Transcription

1 Virtual Assessment of Structural Health Monitoring Techniques for Wind Turines U sing Viration Data E. DI LOREZO, S. MAZATO and B. PEETERS ABSTRACT O perational Modal Analysis (OMA), also k nown a s output-only m odal anal y sis, a llows i dentifying modal paramete rs only y usin g the response measure ments o f t he s tructure i n ope rational c onditions w hen t he i nput f orces c annot e m easured. T his i nformation c an then e u sed to i mprove num erical m odels i n o rder to m onitor t he operating a nd structur al c onditions o f t he syst e m. T his i s a critica l aspe ct oth f or c ondition m onitoring a nd m aintenance o f l arge w ind t urines, particularly i n t he off-s hore s ector w here op eration a nd maintena nce repre sent a hig h pe rcentage o f total costs. T he availaility of commerci al numerical aeroelastic simulation codes simulating the response of wind turines in operation can e used as a virtual design a nd v erification tool. E ffects of desig n modif ications a nd variations i n t he e nvironmental a nd structura l c onditions c an a ll e simu lated using thes e tool s. H owever, exp erimental t est campaig ns s hould e ale t o provide accu rate a nd reliale data with which the model can e updated and e more representativ e of the r eal r esponse. T hus, t he im provement o f the se simu lation m odels i s strongly re lated t o t he i mprovement o f t he c urrent Operational M odal Analy sis ( OM A ) m odal p arameter e stimation t echniques. T he m ain i ssue f or these m ethods i s th a t, d ue t o lade r otation, f orce periodicity a nd t he presenc e o f contro l s urfaces w hich modify c ontinuously t he system confi g uration, m ost o f t he applicaility a ssumptions o f O MA a re viol a ted. I n this p aper, s ome preliminary a ssessments o n how to comin e n umerical a nd e xperimental t echniques f or S tructural H ealth M onitoring ( SHM) o f wind turines are investigated. ITRODUCTIO M odal p arameter e stimation, t o o tain n atural f requencies, m odal d ampings, m ode shapes, is a key step to characterize the dynamical ehavior of a structu r e. Operational M odal A nalysis ( OMA) i s a tech nique f or e stimating t he m odal p arameters o nly o n t he asis o f t he m easured v iration d ata w ithout a ny i nformation o n t he e xcitation forces. LMS Internatio nal, RTD Test Divisi on Interleuvenla an 68, 3001 Leuv e n, Belgium

2 This technique is very attractive for complex structures such as wind turines that are impossile to excite in an artificial way. The results can then e used for numerical model assessment, for the prediction of dynamic response and for modal parameters evolution and tracking for SHM applications [1]. OMA analyses the response of such structures to natural amient excitation, i.e. wind, rain, waves, and was successful applied to uildings, stadiums, ridges. Anyway a successful application of OMA requires that the structure and the natural forces exciting it respect certain assumptions; the structure should e linear and time invariant and the excitation forces should e distriuted randomly oth temporally and spatially [2]. Oviously, the closer is the real excitation to the assumed one, the etter the results of the modal parameter estimation will e. Even though the wind excitation can e considered as a perfect excitation oeying to the OMA assumptions, the aeroelastic phenomena, due to the rotor rotation and time-varying nature of wind turine during operation, set limitations on the applicaility of OMA to operational wind turines. TIME VARYIG ATURE O OPERATIOAL WID TURBIES While application of OMA on parked turines is straightforward, the same is not true in case of operational turines, i.e. in power production configuration. This is due to the fact that two of the key OMA assumptions are not respected y an operational wind turine. OMA requires the excitation forces to e random roadand and uncorrelated in the frequency range of interest. These assumptions are true when the turine is in parked conditions, ut the lade rotation changes the nature of the aerodynamic forces in a significant way putting several limitations to the application of OMA algorithms. It has een oserved that the aerodynamic forces acting on an operational wind turine are characterized y peaks at rotational frequency and its harmonics; these peaks can mask the modal ehavior of the turine at certain frequencies. This aspect introduces a strong influence of the aerodynamic forces in the oserved output responses making the use of OMA for identifying the dynamic characteristics of the structure more complicated than in the parked conditions. Another important assumption violated y a rotating wind turine is the time invariance. This assumption states that the structure under test must not change during the test duration, which is not the case for a wind turine in operational conditions ecause the different components move with respect to each other. Phenomena like rotation of the rotor aout its axis, pitching of the lades and yawing of the nacelle aout the tower result in the violation of the OMA assumptions. Several methods can e applied to deal with such variations. irst of all a time interval in which the wind turine is not yawing and the lades are not pitching can e taken into account. In this case the only prolem will e the rotation of the rotor aout its axis that needs to e analyzed with advanced techniques. rom a mathematical point of view, this rotation introduces time-varying terms in the equations of motion of the turine which results in time-dependent modal parameters that cannot e interpreted as the traditional modal frequencies, damping and mode shapes. One possile solution is to define a methodology that allows analyzing linear time varying systems, ut an alternative way

3 to overcome the prolem is the application of the so-called Coleman transformation or Multi-Blade Coordinate transformation (MBC). Coleman transformation The dynamics of wind turine rotor lades are expressed in rotating frames attached to the individual lades. Multi-Blade Coordinate transformation allows integrating the dynamics of individual lades expressing it in a fixed nonrotating frame. MBC offers several enefits ecause it properly models the dynamic interaction etween the nonrotating ody (tower-nacelle) and the spinning rotor. It also offers physical insight into rotor dynamics and how the rotor interacts with fixedsystem entities and it filters out all periodic terms except those which are integral multiples of Ω, where Ω is the rotor angular speed and is the numer of rotor lades [3]. The main idea ehind MBC transformation is to replace individual lade deflections y new variales which include information aout the gloal rotor ehavior and aout the instantaneous azimuth angle of each lade, thus making application of modal analysis techniques, such as OMA, possile. Consider a rotor with lades that are spaced equally around the rotor azimuth; in this case the azimuth location of the th lade is given y 2 ( 1) (1) where is the azimuth of the first lade considered as the reference lade and 0 means that the first lade is vertically up. If q is one of the rotating degree of freedom for the th lade, the MBC relates it to new degrees of freedom defined in a nonrotating fixed frame as: q 0 1 q 1 q nc 2 1 q cos n (2) q ns 2 1 q sin( n ) The physical interpretation of each one of the new coordinates depends on the degree of freedom it refers to. Anyway the new coordinates are ale to identify the cumulative ehavior of all the lades coupling the rotor with the rest of the turine. The transformation is also necessary to etter understand different kind of phenomena, such as the rotor shaft whirl ehavior that strongly depends on the collective viratory ehavior of the rotor lades.

4 Whirl modal ehavior The whirl mode ehavior has received great attention since it can lead to undesired viration levels. The analytical formulation, discussed in [4], shows that the whirlcausing shaft forces are generated y two cyclic in-plane modes of the rotor. In a single-frequency whirl mode, the rotating force vector on the shaft is a resultant of two components: a regressive force vector at a lower frequency and a progressive one at higher frequency. At the lade level, the viratory motion consists of three components: edgewise motion (in-plane), flapwise motion (out-of-plane) and torsion motion. lapwise modes are usually dominant in wind turine and they are well aerodynamically damped. Torsion modes have high frequencies and low amplitudes, so they are not interesting. inally, dominant edgewise modes must e carefully avoided since they can lead to aeroelastic instailities. In this paper, we will focus on the lade edgewise modes since they are associated to the whirl mechanism. irst of all modes are calculated in parked conditions where the rotor is not rotating and edgewise modes are well identified. Then the operational conditions in which the rotor is spinning at its nominal speed is considered. It can e seen that the spin increases the frequency due to the centrifugal stiffening. After that, MBC transformation can e applied to transform the lade coordinate into the rotor coordinates to take into account the gloal rotor ehavior and to identify the whirling modes. acelle and tower coordinates are not transformed ecause they are already in a fixed frame of reference. WID TURBIE MODEL igure 1 show the wind turine model used in the simulation. The offshore 5MW aseline wind turine has een developed y the ational Renewale Energy Laoratory (REL) to support concept studies aimed at assessing offshore wind technology. It is a conventional three-laded upwind variale-speed variale ladepitch-to-feather-controlled turine [5]. Since the main ojective is to analyze the gloal dynamic ehavior of the full-scale turine, the model has een uilt as simple as possile. igure 1: REL 5 MW SWT model (left) and Test.La geometry (right)

5 Tower: it is modeled as 5 elastic eam elements with lumped masses and hinged to the ground foundation. The total tower height is 90 m. Rotor: in the 3-laded rotor, each lade is identical and is modeled with 17 sections with specific mass, elastic and aerodynamic properties. Drivetrain: the transmission is simplified into a 1 degree-of-freedom system with a gear ratio of 97 etween the Low Speed Shaft (LSS) and the High Speed Shaft (HSS). The wind turine is modeled and simulated using the nonlinear aeroelastic code SAMCE Wind Turines (SWT) that allows the user defining oth a structural and an aerodynamic model which are then solved together to otain the coupled aero-elastic solution [6]. In order to have simulated accelerations that can e considered as those otained from tri-axial accelerometers mounted on the lades, it is necessary to consider them in the local reference frame in which the X axis is the lade axis (oriented toward the lade tip), the Y axis is aligned with the chord-line and elongs to the lade section plane (oriented toward the leading edge) and the Z axis is normal to the chord line and elongs to the lade section plane. Using this axis configuration, the edge-wise modes are descried as ending along the Y axis while flap-wise modes end the structure along the Z axis. Axial modes along the lade pitch axis can e neglected. Different locations are selected to measure the accelerations; three sensors distriuted along the tower, one sensor at the hu center and five sensors per-lade located on the pitch axis [7]. After analyzing the response of the structure in reference and ideal conditions, different possile damages can e introduced to understand how they affect the measured accelerations. In this paper we will focus on the ice formation on the lades. In the software, according to the guidelines for certification of wind turines, it is possile to introduce the ice on all the lades ut one causing a rotor unalancing situation from which several considerations can e done. Wind turine in parked conditions When the wind turine is in parked conditions, the first seconds are used to place the pitch in its parking position specified y the parking pitch angle and the rotor at the angle specified y the initial rotor angle. When the true simulation starts, the generator is disconnected (no resisting torque) and the rotor is released while the pitch remains fixed. In general, long time histories are required for confident modal parameters estimation, ut on the other hand a small sampling frequency is necessary to etter determine low frequency modes. As a good compromise etween these requirements and the computational time, the simulations last 700 seconds and the sampling frequency is set to 100 Hz. Time series are then exported to LMS Test.La in which correlations and spectra can e computed and the PolyMAX method can e applied for estimating the modal parameters. Since the interesting modes are at very low frequencies, a down-sampling to 10 Hz has een performed in the first part of the analysis [8].

6 Tale I: umerical modal parameters in parked conditions for different configurations Standard configuration Ice on all the lades Ice on all the lades ut one requency [Hz] Damping [%] requency [Hz] Damping [%] requency [Hz] Damping [%] 1 st Tower A st lap Yaw st Edge Yaw nd lap Yaw Estimated modal frequencies and damping values are shown in Tale I which shows the variation of some of these parameters with or without ice on the lades. igure 2 shows the power spectral density (PSD) for two different configurations. It shows the shift towards lower frequencies of the two peaks related to the first and second edgewise modes that are taken into account for analyzing the whirl ehavior in the following section. The frequency shift is a consequence of the mass increase when the ice is attached to the lades [9]. Wind turine in operating conditions When the wind turine is in operating conditions, the pitch, yaw and generator are managed y the controller to optimize power production and the rake is not used. While application of OMA on parked turines is straightforward, the same does not apply to operating turines. This is due, as mentioned efore, to the fact that some of the OMA assumptions are violated y a wind turine which lades are rotating. In this case accelerations of points on the lades, tower and nacelle are acquired as for the parked conditions. MBC transformation is then applied to the lade accelerations using the azimuth data while the others are left unchanged. inally, all data are fed to OMA to estimate the modal parameters in operating conditions [10] g 2 /Hz db PSD Blade2:5:+Y reference case PSD Blade2:5:+Y ice case Amplitude Hz igure 2: PSDs of accelerations measured on one point on the 2 nd lade in edgewise direction in parked conditions. Reference case (red curve) compared to the one with ice on the lade (lue curve)

7 g 2 /Hz db Amplitude PSD Blade2:5:+Y parked conditions PSD Blade2:5:+Y operating conditions Hz igure 3: PSDs of accelerations measured at the tip of the lades in parked conditions (red curve) vs. operating conditions (lue curve) igure 3 shows a PSDs comparison etween parked and operating conditions at the lade tip in the edgewise direction. The frequency shift toward higher frequencies for the first and second edgewise modes due to the centrifugal stiffening can e underlined, while the different harmonic components can e seen in the operating case. igure 4 shows the PSDs otained from the lade accelerations at the tip in the edgewise direction efore and after the MBC transformation. The whirling phenomenon is not oservale from experimental data ecause the lade responses are measured in the rotating coordinate system while whirling can only e oserved in a fixed coordinate system [11]. MBC transformation enales oservaility and identification of whirling modes transforming the lade responses into a ground coordinate system. The two whirling modes are separated y 2ω in accordance with the literature, where ω is the fundamental harmonic frequency equal to Hz g 2 /Hz db Amplitude Hz Before MBC After MBC igure 4: PSDs of accelerations measured at the tip of the lades in operating conditions efore (red curve) and after (lue curve) MBC transformation

8 COCLUSIOS The aim of this paper is to understand the applicaility of Operational Modal Analysis techniques to operational wind turines for SHM purposes. The igger limitation in applying OMA is the presence of harmonics components in the measured spectra. This limitation can e overcome y applying the Coleman transformation to experimental data otained from lades sensors. The transformation from a rotating reference system to a ground fixed reference system allows identifying the whirl modes that cannot e seen from experimental data and which represent a critical viration mode. These effects were here investigated and analyzed oth in reference as well in damaged configurations (i.e. ice accretions on the lades). While in this work the accelerations were simulated using an aeroelastic code for wind turine simulations, the same techniques will soon e applied for structural monitoring of a real wind turine. The availaility of experimental results will also lead to improvement of the associated numerical model, and the two will finally e used together in a comined SHM tool. REERECES 1. B. Peeteers, P. Guillaume, H. Van der Auweraer, B. Cauerghe, P. Veroven, and J. Leuridan, Automative and aerospace applications of LMS PolyMAX modal parameter estimation method, Proceedings of IMAC 22, International Modal Analysis Conference, Dearorn (MI), USA, January D. Tcherniak, S. Chauhan, M. Rossetti, I. ont, J. Basurko, O. Salgado, Output-only Modal Analysis on Operating Wind Turines: Application to Simulated Data, Proceedings of European Wind Energy Conference, Warsaw, Poland, April, G. Bir, Multilade Coordinate Transformation and its Application to Wind Turine Analysis, Proceedings of 2008 ASME Wind Energy Symposium, Reno, evada, USA, G. Bir, Understanding Whirl Modal Behavior of a Wind Turine Rotor, Proceedings of IMAC 21, International Modal Analysis Conference, Orlando (L), USA, January J. Jonkman, S. Butterfield, W. Musial, G. Scott, Definition of a 5-MW Reference Wind Turine for Offshore System Development, REL/TP Technical Report, USA, eruary LMS Samtech Ierica, Samtech Wind Turines V Online help, Spain, S. Manzato, D. Moccia, B. Peeters, K. Janssens, J.R. White, A Review of Harmonic Removal Methods for Improved Operational Modal Analysis of Wind Turines, Proceedings of ISMA2012, International Conference on oise and Viration Engineering, Leuven, Belgium, Septemer LMS International, LMS Test.La Rev. 12A, User Manual, Belgium, E. Di Lorenzo, S. Manzato, B. Peeters, H. Van der Auweraer, Virtual assessment of damage detection techniques for operational wind turine, Proceedings of CMMO2013, International Conference on Condition Monitoring of Machinery in on-stationary Operations, errara, Italy, 8-10 May D. Tcherniak, S. Chauhan, M.H. Hansen, Applicaility Limits of Operational Modal Analysis to Operational Wind Turines, Proceedings of 28th International Modal Analysis Conference, Jacksonville (L), USA, e M.H. Hansen, Aeroelastic Instaility Prolems for Wind Turines, Wind Energy (10), pp , 2007.