Numerical and Experimental Investigation of Wind Turbine Wakes X. Huang∗1 , S. Vey†2 , M. Meinke1 , W. Schröder1 , G. Pechlivanoglou2 , C. N. Nayeri2 , and C. O. Paschereit2 1 2 Institute of Aerodynamics, RWTH Aachen University, Wüllnerstr. 5a, 52062 Aachen, Germany Institute of Fluid Dynamics and Technical Acoustics, TU Berlin, Müller-Breslau-Str. 8, 10623 Berlin Numerical and experimental investigations of the flow characteristics around the wind turbine blade and the wake flow field are performed. The numerical method is based on a large-eddy simulation formulated in a rotating reference frame on a structured bodyfitted grid to predict the flow field around a complete rotating wind turbine blade at an average blade Reynolds number of 300,000. Periodic boundary conditions are applied in the circumferential direction, such that the flow over only one of the three blades of the wind turbine is computed. The numerical results show a flow separation on the suction side on the blade with a subsequent transition to turbulence and flow reattachment. Wind tunnel measurements are performed on a 3 m diameter wind turbine installed in the TU Berlin large wind tunnel. Initial experimental data were measured using the data acquisition system installed inside the rotating hub of the wind turbine. First results of the inflow angle and blade root strain gauge measurements at varying inflow angles are presented. I. Introduction To efficiently exploit available installation space and to reduce maintenance and installation costs, wind turbines are clustered in large wind farms. Therefore, a substantial number of wind turbines are located in the wake of upstream wind turbines. The velocity deficit in the wake will lead to a reduction of the power production and the high turbulence intensity will increase dynamic loads on the downstream wind turbine. Depending on the location and wind direction, the power loss of one downstream wind turbine can reach 40%. If the data are averaged over the various wind directions, approx. 8% power reduction is observed for the onshore and 12% for the offshore installations [1]. The wake of a wind turbine is usually divided into two regions, the near wake and the far wake. In the near wake up to one or two diameters downstream of the rotor, the influence of the rotor geometry can be directly identified in the form of the tip and root vortex. The tip and root vortex can cause large velocity gradients and turbulence intensities and even form shear layers in the case of high tip-speed ratios. In the far wake, the influence of the detailed rotor geometry is no more detectable, but the tip and root or hub vortices still have a significant influence on the turbulence intensity. A better understanding of the tip and root vortex dynamics can help to accelerate the recovery of the velocity deficit and decrease the overall turbulence intensity in the far wake. To investigate the development of the wake downstream of a turbine blade including the vortex system, numerical simulations and wind tunnel experiments have been conducted. Numerical methods, based on Reynolds averaged Navier-Stokes (RANS) equations including the k-ω turbulence model, have been widely used for wind turbine simulations [2, 11, 18] because of its robustness and accuracy for most turbulent flows. It is well known, however, that RANS solutions for the tip and root vortex are strongly dependent on the turbulence models. Compared to RANS, large-eddy simulation methods resolve most of the turbulence energy containing structures and the modeling approach only focuses on the subgrid scale of the turbulent spectrum. ∗ PhD Student, Institute of Aerodynamics, RWTH Aachen University, Wüllnerstr. 5a, 52062 Aachen, x.huang@aia.rwthaachen.de, Tel.: +49 241 80 90404 † Dr.-Ing, Institute of Fluid Dynamics and Technical Acoustics, TU Berlin, Müller-Breslau-Str. 8, 10623 Berlin, stefan.vey@tu-berlin.de, Tel.: +49 30 314 22693 1 of 12 American Institute of Aeronautics and Astronautics LES has been proven to be a successful approach to simulate unsteady flows around airfoils and there are applications of LES coupled with actuator line techniques to wind turbine wake simulations [5, 6, 10, 17]. In this paper, the turbulent flow around the wind turbine blade and hub at a Reynolds number of Rec = 300, 000 based on the mean chord will be investigated by LES. The flow field around the blade and the near wake is predicted on a block-structured body-fitted mesh, to ensure an adequate resolution of the boundary layer on the blade. The emphasis of the investigation is put on the accurate prediction of the tip and hub vortices in the wake which cannot be accurately captured by RANS methods. The experimental data are obtained on the Berlin Research Turbine (BeRT). The BeRT is a load control research turbine that is equipped with sensor and actuator systems installed in the rotating system. Chordwise and spanwise pressure distributions, blade root bending moment, tip acceleration, and inflow angle and velocity are captured at a sample rate of 10 kHz. The dynamic flow effects on the rotor blade can be studied under yawed inflow angle of −30◦ ≤ Θyaw ≤ 30◦ . II. A. Methodology Governing Equations The Navier-Stokes equations are transformed into a rotating frame of reference, where the relative velocity → − − − vr is defined by the difference of the absolute velocity → va and the rotational velocity → ve , → − − − vr = → va − → ve (1) For simplicity, the x-coordinate direction is assumed to be the rotation axis, hence, the Navier-Stokes equations in the relative frame of reference for the unsteady compressible flow can be written as I Z Z ∂Q dV + F · ndA = SdV (2) ∂t A V V T where n is the normal vector of the surface dA and Q = [ρ, ρu, ρE] is the vector of conservative variables. The quantities ρ, u, E denote the density, the velocity vector in the relative frame and the relative total energy. The relative total energy is defined as E =e+ → − → − − vr 2 ve 2 u2 + v 2 + w2 ω2 → r2 − =e+ − , 2 2 2 2 (3) where ω is the rotation speed and r is the radius of the rotation axis. The flux vector F contains an inviscid Fi and a viscous part Fv ρu 0 1 (4) F = Fi + Fv = ρuU + p · n + . τ Re ρHU τu + q The Reynolds number is defined by the fluid properties at rest state and p is the static pressure. The relative total enthalpy in a steadily rotating frame of reference is defined as − → − p → vr 2 ve 2 + − ρ 2 2 (5) U = nx u + ny v + nz w, (6) H =e+ U is the contravariant velocity where nx , ny , nz represent components of the normal vector of the surface A. The viscous stress τ can be expressed by 2 T τ = η (∇ · u) I − 2η ∇u + (∇u) , (7) 3 where η is the dynamic viscosity computed by Sutherland’s law. The vector of heat conduction q is q=− k ∇T Pr (γ − 1) . 2 of 12 American Institute of Aeronautics and Astronautics (8) The quantity k is the thermal conductivity and the Prandtl number is defined as Pr = η∞ cp /k∞ , where cp is the specific heat at constant pressure and γ the ratio of specific heats. The source term S contains the Coriolis and centrifugal force S= 0 ρfx ρω(yω + 2w) + ρfy ρω(zω − 2v) + ρfz ρf vr + q̇h (9) with f denoting the body force and q̇h being the time rate of heat transfer. The system of equations is closed by the equation of state for a perfect gas p e= . (10) ρ (γ − 1) B. Numerical Methods The Navier-Stokes equations are spatially filtered assuming an implicit grid filter and discretized using a finite-volume technique on a structured body-fitted grid. The monotonic integrated large-eddy simulation (MILES) approach proposed in [3] is used, where the dissipative part of the truncation error is assumed to mimic the effect of the dissipation of the unresolved subgrid scales. Various difference stencils are used to discretize the convective and viscous fluxes. The convective fluxes are formulated by a low dissipation variant of the AUSM scheme [8] and a MUSCL interpolation at second-order accuracy. The discretization of the viscous fluxes is performed by a second-order accurate central difference scheme [9]. An explicit five-step Runge-Kutta formulation also at second-order accuracy is used for the temporal integration. This solution method has been extensively validated for various turbulent flow problems, see e.g. [7, 13, 14]. III. Computational Setup and Grid Strategy The wind turbine geometry is identical to that used in the wind tunnel of TU Berlin. The rotor uses the downscaled geometry from the previous European project MEXICO. The Clark-Y profile is used for the blade from the root to the tip. Experimental measurements are performed in the wind tunnel of TU Berlin with a downscaled rotor diameter of approx. 3m and the hub height of approx. 2.1m. The Reynolds number, based on the freestream velocity (6.4m/s) and radius of the rotor (1.5m), is 640,000 and the tip-speed ratio is equal to 4.6, which matches the experimental value. Figure 1. Domain size and boundary conditions for the LES of the flow around the wind turbine blade 3 of 12 American Institute of Aeronautics and Astronautics Fig. 1 shows the geometric setup and boundary conditions in the CFD model of the wind turbine blade. To reduce the computational effort, the simulation domain around the wind turbine will be restricted to a 120 degree section in the circumferential direction such that the flow around one blade has to be resolved. Periodic boundary conditions are used in the circumferential direction as shown in the Fig. 1. The computational domain has a length of 10R, where R is the radius of the rotor, upstream and downstream of the turbine blade to avoid an influence of the approximate boundary conditions. At the inflow boundary the mean inflow velocity is prescribed. Figure 2. Mesh and topology near the blade Figure 3. Mesh and topology for the periodic planes in the circumferential direction A structured body-fitted grid with 127 million mesh points around the blade and hub was created with a commercial mesh generation software. The mesh resolution required by the LES of wall-bounded flows (∆y + ≈ 2, ∆z + ≈ 30, ∆x+ ≈ 50) is reached in the blade-tip region. The grid system consists of 48 blocks covering one blade and one third of the hub. To obtain a better grid orthogonality near the blade surface, a C-type topology is used along the span. Due to the periodic boundary conditions, the number of cells and the distributions upstream and downstream of the blade must be identical such that the special topology depicted in Fig. 2 was designed to meet the requirements. Besides, additional blocks, indicated in blue in Fig. 3, are inserted as buffer zone between the blade blocks and periodic surfaces, which can reduce the mesh skewness near the blade and improve the mesh resolution of the near wake. 4 of 12 American Institute of Aeronautics and Astronautics IV. Numerical Results In Fig. 4, the contours of the second invariant of the rate of strain tensor is used to illustrate the vortical structures around the wind turbine blade. A stable tip vortex and turbulent structures in the boundary layer and the wake can be seen. The axial velocity distribution in the wake flow on a vertical plane through the blade is shown in Fig. 5. The higher velocity in the blade tip region and the tip vortex are evident. The wake deficit downstream of the wind turbine blade is visible in the lower values of the axial velocity. (a) Pressure side (b) Suction side Figure 4. Q-Criterion of vortical and turbulent structures on the pressure side (a) and suction side (b) of the blade. Figure 5. Axial velocity contours on a cut plane parallel to the rotation axis and through the turbine blade. The instantaneous and mean pressure contours on the suction and pressure surfaces of the blade are shown in Fig. 6. A plateau and sudden decline of the static pressure is visible on the suction side of the blade indicating flow separation. In Fig. 6, I, II and III denote the slices at 63%, 80%, and 90% span. Figs. 7 and 8 show the averaged chordwise pressure coefficient and the skin-friction coefficient against φ at 63%, 80%, and 90% span of the blade (I, II, and III as shown in Fig.6). The quantity φ is defined as the arcsin (L/r), where L is the distance between the point on the blade and XY plane and r the distance to the X-axis. The pressure coefficient and skin-friction coefficient are computed by 5 of 12 American Institute of Aeronautics and Astronautics (b) (a) Figure 6. (a) Instantaneous pressure contours on the suction and pressure surface; (b) time-averaged pressure contours on the suction and pressure surface. (a) (b) (c) Figure 7. Pressure coefficient around the blade at 63%(a), 80%(b), and 90%(c) span of the turbine blade (a) (b) (c) Figure 8. Skin-friction coefficient around the blade at 63%(a), 80%(b), and 90%(c) span of the turbine blade Cp = (P − P∞ ) + (ωr)2 ]) 1 2 2 ρ[U∞ 6 of 12 American Institute of Aeronautics and Astronautics (11) (a) (b) (c) Figure 9. Mach number contours on the slices at 63%, 80%, and 90% span of the turbine blade (a) (b) (c) Figure 10. Contours of turbulent kinetic energy on the slices at 63%, 80%, and 90% span of the turbine blade Cf = 1 2 2 ρ[U∞ τw + (ωr)2 ]) (12) In Figs. 7 and 8, the pressure distribution on the suction side also show the plateau, while the skin-friction coefficient reaches a negative value at the same position showing the flow to separate and reattach further downstream. The instantaneous contours of the relative velocity magnitude and the turbulent kinetic energy around the blade section at 63%, 80%, and 90% span is shown in Fig. 9 and 10. It can be seen that the separated flow region is small and a turbulent boundary layer develops downstream of the reattachment. V. Research Turbine and Experimental Setup A 3 m diameter modular load control research wind turbine was designed, built, and installed in the TU Berlin large wind tunnel wind energy measurement section. The ”Berlin Research Turbine”, BeRT is the first German load control research turbine. A further unique feature for this turbine size is the installation of data acquisition and real time load control processor within the rotating system. With this setup a multitude of sensor signals can be read simultaneously at a sampling frequency of up to 10 kHz. The rotor blades can be easily interchanged so that various load control principles can be tested experimentally. The setup in the wind energy measurement box (WOX) is illustrated in Fig. 11. The existing wind tunnel’s settling chamber was extended in two stages (WOX 1 and WOX 2, respectively) by a total of 5 m. This new test section has a cross-section of 4.2 m x 4.2 m. It is mainly used for measurements incorporating the BeRT and features a 1.1 m diameter turntable with eccentric mounting for the turbine tower so that yaw movements do not interfere with blade tip – side wall distance symmetry. The turbine can be set to yaw angles in the range −30◦ ≤ Θyaw ≤ 30◦ . The baseline turbines power curve is shown in Fig. 12. Maximum power output is reached at a tip speed ratio of TSR=4. At a maximum inflow velocity of U∞ = 8 m/s the blade Reynolds number is in the range 250 000 ≤ Re ≤ 350 000. Note that the measurements were performed with only WOX 1 installed. This means that the wind tunnel nozzle is located 0.5 rotor diameters downstream of the rotor-plane and therefore 7 of 12 American Institute of Aeronautics and Astronautics an interaction with the wake-development of the turbine cannot be ruled out. WOX 2 is now installed, so that future measurements will be performed with the nozzle further downstream so that the interaction between wake and nozzle will be reduced. The blade is equipped with a Clark Y airfoil along the full span – a cylindrical root section is omitted. This airfoil was chosen because of its suitability for flaps, while still being relatively insensitive to Reynolds number effects. Furthermore, it provides the necessary thickness to house the required sensors and actuators inside the blade. WOX 1 WOX 2 Figure 12. Power curve of BeRT. Figure 11. Setup in wind tunnel. The initial set of three baseline blades was used in the development and validation of the data acquisition and load control system. The baseline blades were therefore equipped with stick-on pressure sensors, threehole probes, blade-root strain gauges, and a tip accelerometer. These sensors are representative of the sensors that will be integrated into the active blade. With these preliminary measurements some problems could be identified and fixed at an early-on stage, such as an insufficient stiffness of the turbine foundations. The fully sensor- and actuator-equipped rotor blade has just been commissioned and measurements will be performed in the near future. A picture of the BeRT with the sensor- and actuator-equipped rotor-blade is shown in Fig. 13. Figure 13. BeRT with smart rotorblade. VI. A. Figure 14. View of BeRT with flow tufts. Measurement Techniques and Results Quantitative Tuft Flow Visualization The BeRT presented the unique opportunity for the development of a novel measurement technique where a quantitative tuft flow visualization technique was synchronized with time resolved pressure and vibration 8 of 12 American Institute of Aeronautics and Astronautics measurements. With this technique, an arbitrary measured variable (e.g. pressure) can be linked to an instantaneous surface flow field on the rotor blades. It is possible to capture the complete rotor in one image, with all three blades equipped with flow tufts and image registration markers. A dedicated in-house code then distinguishes between the three blades and computes instantaneous tuft-angles for each tuft in each image. The resulting data-set is then processed to characterize the flow regime at an individual tuft location. In combination with e.g. the relative angle of attack reading from the 3-hole probe and the root strain gauge a deeper insight into the response of the turbine to dynamic inflow conditions can be gained. B. Variation of Yaw Angle Some interesting observations can be made from the development of the measured variables under various yawed inflow cases. An example of which is shown for the strain gauge (top) and surface pressure (bottom) signals in Fig. 15. The surface pressure signal was measured using a stick-on surface pressure port at x/c=75% and a radial position of 75% blade length. It shows the expected behavior, namely that signal amplitude is increased for larger deviations from Θyaw = 0◦ . In contrast to the pressure signals, the strain gauge signal shows no material change in signal amplitude when the yaw angle is altered. This behavior was not expected and the source was identified to be a natural frequency overlap caused by an insufficient stiffness of the wind turbine foundation. This causes turbine vibrations which are directly transmitted to the turbine blade root. Once the source of the problem was identified the base plate was replaced with a stiffer construction that incorporates a central stiffening element. Initial qualitative vibration tests were promising and the improvement will be quantified in the next measurement campaign. yaw angle -30° -15° 0° 5° 10° 15° 30° blade root strain gage 10% c pressure 75% radial position Figure 15. Signals from blade root strain gauge and 10 % chord, 75 % span stick-on surface pressure port. Note the change in pressure signal amplitude over yaw angle while the strain gauge amplitude remains the same. C. Three-Hole Probe Measurements One of the tested load control schemes requires the accurate and time resolved measurement of the relative blade section inflow angle. This is accomplished using three hole probes mounted to the respective blade section. Shown in Fig. 16 is the orientation of the probe itself and the pressure signal from the three probe holes for various yaw-cases. These pressure signals will be used directly as an input to a machine learned control scheme (MLC). In the future, a calibration function will be applied and both the relative inflow speed and direction (i.e. the angle of attack) can be derived from the raw signals. Furthermore, the number of required pressure transducers can be reduced from currently three to two per three hole probe. This is accomplished by measuring against the central three hole probe tube. D. Results from Quantitative Tuft Flow Visualization Technique It was already mentioned above that a quantitative tuft flow visualization technique was combined and synchronized with the time resolved measurement system. The setup is shown in Fig. 17. The turbine was set to a yaw angle of 30◦ and operated at a tip speed ratio of T SR = 4. All three blades were equipped with flow tufts and image registration markers (see also the photo in Fig. 14). A high resolution digital camera was mounted downstream of the turbine in the high speed test section of the wind tunnel. With the high camera-resolution it was possible to capture the whole rotor plane in a single image. 9 of 12 American Institute of Aeronautics and Astronautics U 8 p10 p04 p09 p10 p04 p09 Figure 16. Raw signals from three hole probe. Signals are used as an input to the AFC control loop. 0° flow tufts 8 U Θ yaw strain gage flash detector ω 90° 270° 2 kW flash-lamp camera 90° Figure 17. Experimental setup for tuft flow visualization measurements. ΔΦ ΔΦ ΔΦ 0° 270° Figure 18. Surface flow patterns obtained with phase averaged quantitative tuft flow visualization. A high-energy, high-speed flash unit provided the necessary lighting. The light pulse was detected by the flash detector, whose signal was recorded simultaneously with the other measured variables by the data acquisition system. In a post processing step the individual images can then be associated with the instantaneous measured variables. The raw tuft images were then post-processed using the in-house FlowViz code. The resulting per-blade vector fields were summarized to yield per-tuft statistics. These can then be further processed using tools and techniques known from PIV data post-processing such as phase- and ensemble-averaging and POD. This technique has already been applied to utility scale wind turbines [15], and full scale test cars and trucks. The FlowViz-technique is under continuous development because surprisingly the evaluation of wind tunnel FlowViz-data poses the larger effort compared to the free-field measurements. The reason is the better lighting conditions in the free field measurements. It is a non trivial task to create acceptable lighting conditions in a 4.2 m by 4.2 m cross section with black walls and a rotating object with curved surfaces. Some preliminary results of this exciting new measurement possibility are shown in Fig. 18. The blade’s vector fields were phase averaged for Θwing =270◦ , 0◦ , and 90◦ . The top part of Fig. 18 shows the average 10 of 12 American Institute of Aeronautics and Astronautics Figure 19. Mean surface flow angle over phase angle. Data extracted from full flowviz-dataset. deviation ∆Φ between the chord-wise and the local flow vectors. The lower half of the figure shows the streamlines visualized using a line integral convolution (lic) [4] with a visual notion of the streamline deviation. The sequence shows the periodic behavior of the crossflows of the BeRT under Θyaw = 30◦ . The data is again summarized in Fig. 19. Here, each data point represents the mean flow direction of a complete blade at a certain rotor azimuthal angle Θwing . In the azimuthal rotor angle range of −200◦ ≤ Θwing ≤ −125◦ the turbine’s tower is blocking sight. This data can now be correlated to the pressure or strain gauge signals shown in Fig. 15. VII. Conclusion and Outlook This study presents a high fidelity flow simulation for a wind turbine blade. The approach uses a large-eddy simulation solver to investigate the aerodynamic characteristics around wind turbine blade. The turbulence around the blade and the tip vortex in the near wake is well captured by the LES. For the numerical part, the future work will focus on a detailed comparison of the numerical and experimental data and the analysis of the stability of the tip vortex system in the wind turbine wake [12, 16]. Two oscillating flaps will be installed close to the wind turbine blade tip and the mesh deformation method will be used to mimic the motion of the flaps. The influence of the perturbation frequency of the flaps on the stability of the vortex system as well as the aerodynamic loads on the blade will be investigated. Initial results from an experimental wind tunnel study on the Berlin Research Turbine (BeRT) were presented for varying yaw angles. A quantitative tuft flow visualization technique was conducted and the variations in surface flow patterns on the rotor blades under a yaw angle of Θyaw = 30◦ were shown. In future experiments active trailing edge flaps will be used in an attempt to excite tip vortex instabilities. The tip vortex structure and dynamics will then be studied using stereoscopic particle image velocimetry. Acknowledgement This numerical and experimental research is conducted in the DFG funded project Active control of wind turbine wakes PAK 780. All computations were performed at the High Performance Computing Center Stuttgart (HLRS). 11 of 12 American Institute of Aeronautics and Astronautics References 1 R. Barthelmie and L. Jensen. Evaluation of wind farm efficiency and wind turbine wakes at the nysted offshore wind farm. Wind Energy, 13(6):573–586, 2010. 2 A. Bechmann and N. Sorensen. Cfd simulation of the mexico rotor wake. In European wind energy conference and exhibition, Marseille, 2009. 3 J. P. Boris, F. F. Grinstein, E. S. Oran, and R. L. Kolbe. New insights into large eddy simulation. Fluid Dynamics Research, 10:199–228. 4 B. Cabral and L.C. Leedom. Imaging vector fields using line integral convolution. In Proceedings of ACM SIGGRAPH ’93, pages 263–270, 1993. 5 S. Ivanell. Numerical analysis of the tip and root vortex position in the wake of a wind turbine. 2007. Journal of Physics: conference series, 75-012035. 6 A. Jimenez. Application of a les technique to characterize the wake deflection of a wind turbine in yaw. Wind Energy, 13:559–572, 2010. 7 M. Konopka, M. Meinke, and W. Schrder. Large-eddy simulation of shock-cooling-film interaction at helium and hydrogen injection. Phys.Fluids, 25(10),106101, 2013. 8 M. S. Liou and C. J. Steffen-Jr. A new flux splitting scheme. J. Comput. Phys., 107:23–39, 1993. 9 M. Meinke, W. Schröder, E. Krause, and Th. Rister. A comparison of second and sixth-order methods for large-eddy simulations. Comp. Fluids, 31:695–718, 2002. 10 J Meyers and C Meneveau. Large eddy simulations of large wind-turbine arrays in the atmospheric boundary layer. In 48th AIAA Aerospace Sciences Meeting, 2010. AIAA 2010-827. 11 N. Sorensen and M.O.L. Hansen. Rotor performance predictions using navier-stockes method. In 36th AIAA Aerospace Sciences Meeting and Exhibit, 1998. AIAA-98-0025. 12 V. L. Okulov and J. N. Sorensen. Stability of helical tip vortices in a rotor far wake. J. Fluid Mech., 576:1–25, 2007. 13 P. Renze, W. Schröder, and M. Meinke. Large-eddy simulation of film cooling flows at density gradients. International Journal of Heat and Fluid Flow, 29:18–34, 2008. 14 V. Statnikov, T. Sayadi, M. Meinke, P. Schmid, and W. Schrder. Analysis of pressure perturbation sources on a generic space launcher after-body in supersonic flow using zonal rans/les and dynamic mode decomposition. Phys.Fluids, 27,106101, 2015. 15 S. Vey, H.M. Lang, C.N. Nayeri, C.O. Paschereit, and G. Pechlivanoglou. Extracting quantitative data from tuft flow visualizations on utility scale wind turbines. In The Science of Making Torque from Wind 2014, Journal of Physics: Conference Series 524, 2014. 012011. 16 S.E. Widnall. The stability of a helical filament. J. Fluid Mech., 54(4):641–663, 1973. 17 Y Wu and Porte-Agel F. Large-eddy simulation of wind turbine wakes: evaluation of turbine parameterization. BoundaryLayer Meteorology 2011, 138. 18 F. Zahle and N. Sorensen. Wind turbine rotor-tower interaction using an incompressible overset grid method. Wind Energy, 12:594–619, 2009. 12 of 12 American Institute of Aeronautics and Astronautics

1/--страниц