Design_Of_The_Wind_Turbine_Airfoil_Family_Riso-A-Xx_1998

Design_Of_The_Wind_Turbine_Airfoil_Family_Riso-A-Xx_1998 -...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Design of the Wind Turbine Airfoil Family RIS AXX Kristian S. Dahl, Peter Fuglsang Ris National Laboratory, Roskilde, Denmark December 1998 Ris R1024EN Abstract A method for design of wind turbine airfoils is presented. The design method is based on direct numerical optimization of a B-spline representation of the airfoil shape. For exibility, the optimization algorithm relies on separate, stand alone tools for the analysis of aerodynamic and structural properties. The panel method based XFOIL is used during the optimization whereas the NavierStokes solver EllipSys2D is used in the evaluation of the results. The method is demonstrated by the design of an airfoil family composed of 7 airfoils ranging in thickness from 12 to 30. The design is based on Reynolds and Mach numbers representative of a 600 kW wind turbine. The airfoils are designed to have maximum lift-drag ratio until just below stall, a design lift coe cient of about 1.55 at an angle of attack of 10 and a maximum lift coe cient of 1.65. The airfoils are made insensitive to leading edge roughness by securing that transition from laminar to turbulent ow on the suction side occurs close to the leading edge for post stall angles of attack. The design method and the airfoil family provides a sound basis for further enhancing the characteristics of airfoils for wind turbines and to tailor airfoils for speci c rotor sizes and power regulation principles. The Danish Energy Agency funded the present work under the contract ENS 1363 95-0001. Ris R1024 ISBN 8755023568 ISBN 8755024874Internet ISSN 01062840 Information Service Department, Ris  1998 Contents 1 Introduction 4 2 Wind turbine airfoil characteristics 6 3 Design method 7 3.1 3.2 3.3 3.4 3.5 3.6 Design algorithm 7 Geometry description 8 Optimization algorithm 9 Sensitivity analysis 9 Flow analysis 9 Structural analysis 9 4 Design strategy 10 5 Airfoil family 12 5.1 5.2 5.3 5.4 5.5 5.6 Validity of geometry description 12 Design criteria 13 Geometric properties 14 Aerodynamic properties 16 Comparison of XFOIL and EllipSys2D predictions 20 Comparison of clean and dirty performance 23 6 Conclusion 26 References 27 Ris R1024EN 3 1 Introduction Design of tailored airfoils for wind turbine rotor blades is important for the continuing development of wind turbines. Optimization studies show that airfoils with suitable characteristics are important to further reduce the cost of the produced energy, Fuglsang and Madsen 13 . The airfoils that are currently used range from rather old NACA airfoil series originally developed for airplanes, Abbot and Doenho 1 to dedicated wind turbine airfoils. Wind turbine airfoils should di er from traditional aviation airfoils in choice of design point, o -design capabilities and structural properties. The development of wind turbine airfoils has been ongoing since the mid 1980's and a large e ort was done by Tangler and Somers 29 , who developed several airfoil families. Other airfoil designs for wind turbines can be found in Bj rk 4 , Timmer and van Rooy 30 , Hill and Garrad 17 and Chaviaropoulos et al. 6 . Most of these airfoil designs were developed by use of inverse design methods. Numerous methods for airfoil design are available and a survey of such methods and available references can be found in Henne 15 and Dulikravich 10 . In traditional inverse design, the airfoil surface ow is prescribed at speci ed operational conditions and a shape is found that will generate these surface conditions. Full-inverse methods determine the overall airfoil geometry from the overall surface pressure distribution whereas mixed-inverse methods determine parts of the airfoil contour while holding the rest unchanged. A full-inverse approach for incompressible ows is the complex mapping method originally formulated by Mangler 22 and Lighthill 20 . A method that includes a boundary layer formulation is later developed by Liebeck 19 . On basis of these methods, Eppler and Somers developed their computer code 11 , which has been used for development of numerous wind turbine airfoils, e.g., 29 . A popular mixed-inverse method is the XFOIL code by Drela 9 that uses a global Newton method. XFOIL was used for design of wind turbine airfoils by Bj rk 4 among others. Traditional inverse design methods in general have limited capabilities for multiple design points, since there is only a single target pressure distribution at a single design point. However, a method for multi-point design using an inverse method was developed by Selig and Maughmer 27 . They allow di erent segments of the airfoil shape to be determined by di erent ow constraints. Inverse design methods can not treat multidisciplinary design problems and allow only limited o -design considerations. These matters are most often taken care of manually by the designer in a cut-and-try process. Direct design methods based on numerical optimization provide basically a rational multidisciplinary design procedure where several design parameters can be improved and multiple constraints can be imposed. A general ow solver and eventually a structural code are coupled with a numerical optimization algorithm. The optimization algorithm generates an optimum airfoil shape that has desirable characteristics, as speci ed by the designer, while satisfying aerodynamic and structural constraints. Most direct design methods use gradient-based algorithms. Hicks et al. 16 used a simple feasible direction algorithm with a panel method in their design of transonic airfoils. When a more complex ow solver is used, such as in Eyi and Lee, 12 , the computational costs increase because of the sensitivity analysis, which requires a large number of analysis runs. In the case of Navier-Stokes or Euler solvers, computational costs can be reduced by the use of adjoint operator control theory, Jameson 18 . Another category of methods are evolutionary algorithms such as in Obayashi and Takanashi 25 and stochastic approaches, Aly et al. 2 . 4 Ris R1024EN They are less sensitive to local minima but have very high computational costs. Airfoil design is a multidisciplinary eld, involving aerodynamics, structural dynamics, stability and control, manufacturing and maintenance considerations. Despite available design methods, airfoil design remains to a great extent a cutand-try procedure where advanced design methods assist the designer. The purpose of the present work was to further automate the airfoil design process by developing an interdisciplinary optimization method for airfoil design, which used a numerical optimization algorithm. The method relied on a state of the art tool for analysis of the ow eld and included simple structural calculations. Attention was paid to the ability to design airfoils from scratch and a strategy for tailoring of wind turbine airfoils was developed. The design method was demonstrated by the design of an airfoil family for pitch- or stall-regulated wind turbines with a rated power around 600 kW. Ris R1024EN 5 2 Wind turbine airfoil characteristics The characteristics of an ideal wind turbine airfoil depend in principle on the speci c rotor the airfoil is intended for. But, in general, some properties can be labeled as desirable for most wind turbine airfoils. For maximum power production, the lift-drag ratio should be high for airfoils used on the outer part of the blades. In case of pitch regulation and active stall regulation, the lift-drag ratio should be high at and near the operational point. For stall regulation, the lift-drag ratio should be high in the entire operational range, i.e., angle of attack below the maximum lift coe cient. On the inboard part of the blades, the lift-drag ratio is of less importance, but the maximum lift should be high to reduce the blade area. The operational point should be close to maximum lift. This ensures high liftdrag below stall for stall regulation and in case of wind gusts for pitch regulation an autonomous stall control is build in to reduce power peaks. Good o -design characteristics are important because of the wide variation in the angle of attack during normal operation this is in contrast to aviation operating conditions. For stall regulation, the ow at maximum lift should separate from the trailing edge to have a smooth lift curve in stall which reduces the risk of stall induced vibrations in contrast to massive leading edge separation. The transition from the linear part of the lift curve to the post stall area should be well-de ned and smooth. Furthermore, the airfoil should be insensitive to double stall, Bak 3 . In natural conditions, bugs and dirt often soil wind turbine blades at the leading edge. Roughness at the leading edge will cause early transition from laminar to turbulent ow and an eventual jump in the boundary layer momentum thickness. This reduces maximum lift, lower the lift curve slope and increase the skin friction resulting in loss of power production. Especially for stall regulation, the maximum lift coe cient should be insensitive to leading edge roughness. On the inboard blade section, the airfoils should have high cross section sti ness, to limit blade weight and tip de ection. This is most easily obtained by increasing the airfoil maximum thickness at the expense of aerodynamic performance, e.g., reduced lift-drag ratio. The desirable airfoil characteristics constitute both aerodynamic and structural properties and multiple con icting characteristics are involved. High lift-drag is in contrast to high airfoil thickness. High maximum lift is in contrast to insensitivity to leading edge roughness. High lift-drag ratio at the design point is di cult to obtain together with extensive o -design requirements. But, this is exactly where numerical optimization is useful, because it can search the design space in a systematic manner and nd the best compromise between these con icting requirements. The designer of course still has to make quali ed decisions on the relative weighting of the di erent desirable properties. 6 Ris R1024EN 3 Design method The design method is based on numerical optimization. The general formulation of an optimization problem is, e.g., 31 : Minimize: F x Subject to: G x  0, j = 0; m where m + 1 is the number of constraints. The objective function, F x, is minimized by changing the design variables that compose the design vector, x. Here, the design variables are the coordinate points that describe the airfoil shape. The inequality constraints, G x, are side values for the design variables and bounds on response parameters. Equality constraints can be replaced with two inequality constraints with opposing signs. j j 3.1 Design algorithm ' & $ '  & $  The combination of numerical optimization and di erent tools for ow and structural calculations are shown in Figure 1. Initial airfoil shape Optimization algorithm - Objective function Constraints Design variables ? Interface ? - Optimum airfoil shape Flow solver Structural calc.   Target curve Figure 1. Flow chart of the design method. An airfoil shape in principle, any airfoil-like shape is input together with a de nition of the objective function, the design variables and the constraints. The optimization process is iterative and the iteration loop involves several calculations of ow and structural properties. Di erent tools carry out these tasks. An interface handles the necessary book-keeping of design variables and constraints and the calculation of sensitivity information. The interface converts the actual design vector into an airfoil shape. The ow and structural calculations are used to estimate the value of the objective function and the constraints. Multiple angles of attack can be calculated to allow o -design optimizations and the combination of ow and structural responses allows interdisciplinary optimization. When available, other calculation tools, such as calculation of aerodynamic self noise can easily be incorporated. Traditional inverse airfoil design is made possible by comparing the actual ow response with prescribed target values. Ris R1024EN 7 3.2 Geometry description A smooth airfoil shape is important for the optimization results. In principle, any physically realistic shape should be possible to allow design from scratch. The shape description should have as much geometric exibility as possible with as few design variables as possible to secure an e ective and representative search of the design space with acceptable computational costs. It is important that the geometric description does not limit the design space too much a priori. Di erent approaches can be used. Hicks et al., 1974 16 describe the airfoil thickness by a polynomial where the coe cients are design variables. Others such as 5 represent the airfoil surface by polynomials. An initial airfoil shape can be modi ed by adding smooth perturbations as in 14 where a linear combination of a set of base functions is used with weighting coe cients as design variables. However, these methods need a large number of design variables to have su ciently geometric degrees of freedom and this increases computational costs and might cause scatter in the airfoil geometry. In the present case, the airfoil shape is represented by a single B-spline curve de ned by a set of control points 7 : pu = XP N n =0 i u i;k  i where 0 u n , k + 2, k is the order of continuity, P  ;   are the coordinate points, n + 1 is the number of coordinate points, N u are in uence functions. The B-spline curve was de ned clockwise from the airfoil trailing edge and the airfoil shape was transformed into a standard x , y coordinate system with the chord along the x-axis. The B-spline curve is continuous of the k'th order and no special considerations are necessary for the airfoil nose region. B-splines, furthermore, have the advantage that k determines how large a part of the entire curve that is altered when a single control point is moved. High values of k result in a smooth curve, whereas small values of k create a more lively curve. Figure 2 shows an example with n + 1 = 12, k = 5, which were common values for the present study. Most of the control points were only allowed to move in the y direction, which limits the number of design variables to be close to n + 1. i i i i;k y  x pu Figure 2. B-spline representing the airfoil shape, n + 1 = 12, k = 5. the dots are the control points design variables. 8 Ris R1024EN 3.3 Optimization algorithm The choice of optimization algorithm is basically a choice between gradient based methods and global methods such as evolutionary type algorithms. Evolutionary algorithms are less sensitive to local minima. However they are time consuming and constraints have to be included as a penalty term on the objective function. Gradient based methods on the other hand allow multiple constraints but lack global optimality. We chose a traditional simplex optimization algorithm based on sequential linear programming with move limits in a standard bound formulation 26 . Simplex methods are search method that are simple, robust and reasonably fast. They require the gradients of the objective function and of the constraints which are provided by a sensitivity analysis. 3.4 Sensitivity analysis Adjoint operator control theory methods have recently been applied to uid ow equations 18 . This approach requires the additional solving of adjoint equations. Compared to traditional numerical nite di erences, these methods are time saving when the number of design variables is large. However, the adjoint equations have to be derived for each of the governing ow equation. We based the sensitivity analysis on numerical nite di erences. This was more time consuming, but ensured exibility in the choice of ow solver and structural calculations. 3.5 Flow analysis In principle, there are no restrictions on the choice of ow solver. Since the optimization process requires many evaluations of the objective function and the constraints before an optimum design is obtained, computational costs are high when a Navier-Stokes solver is used for each ow calculation as in 12 . In stead, we chose XFOIL 9 for the ow calculations. XFOIL is an inviscid linear-vorticity panel method with source distributions superimposed on the airfoil and its wake allowing modeling of viscous layer in uence on the potential ow. A two-equation integral boundary layer method is used to represent the viscous layer 8 . XFOIL is developed for transonic and low Reynolds number ows and is well suited for optimization because of the relative fast and robust viscous inviscid interaction scheme. For given angle of attack, Reynolds number and Mach number, XFOIL provides pressure distribution, lift and drag coe cients. In addition, numerous boundary layer parameters are calculated, e.g., displacement and momentum thickness, shape factor, skin friction, transition point location, etc. In XFOIL, transition is modeled by the e method with n = 9 as default value. n 3.6 Structural analysis Simple structural calculations were carried out on the airfoil cross section such as the airfoil thickness and mean line distributions, the airfoil maximum relative thickness, area, and area moments of inertia. Ris R1024EN 9 4 Design strategy Before turning to the speci c design of the airfoil family, we describe in general terms the design strategy followed. The design task or rather the optimization problem is de ned by the design variables, the operating conditions, the design objectives and the constraints. Design variables The design variables are chosen among the control points of the B-spline describing the airfoil shape. The control points at the trailing edge are typically xed in both the x and y directions to provide the desired trailing edge thickness. For most of the control points only the y-coordinate is a design variable to limit the number of design variables and to ensure a uniform spacing between the control points. Operational conditions The overall operational conditions are de ned by the Reynolds number based on chord and the Mach number. The Reynolds number for an airfoil section on a wind turbine blade depends on the span-wise location and on the size of the wind turbine. Since the maximum Mach number is usually around 0.2, the ow can be considered incompressible with good approximation. Design objectives To allow both aerodynamic and structural objectives and o -design objectives, the P objective function is de ned as a linear combination of objectives, F = =1 a f , where a are weight factors and f are the di erent objectives. The objectives can be both aerodynamic e.g., lift-drag ratio for one or more angles of attack and structural e.g., moment of inertia of thickness at a certain chord-wise position. The weighting of the di erent objectives is the responsibility of the designer and this has obviously great in uence on the nal design. The objective at the design angle of attack is usually given a high weight factorto secure good performance at the design point. n i i i i i Design constraints To conclude the de nition of the optimization problem, constraints are imposed on the design. To obtain the desired maximum lift coe cient and lift curve, upper and lower limits are imposed on the lift coe cient at the design angle of attack and other angles of attack, e.g., the C max-angle of attack and in the post-stall region, Figure 3. The design angle of attack, should be chosen 1-2 degrees below C max to ensure a linear C   and low drag at angles until C max . In principle, can be anywhere on the linear part of the lift curve, can even be a design variable. Depending on the desired post stall characteristics, constraints can also be added to the suction side separation point, Ssep, that should be at the trailing edge at and then move towards the leading edge just before C max. To ensure a well de ned stall, there should be a sudden movement in Ssep at C max . A smooth trailing edge stall can be speci ed with a low negative slope for Ssep   in stall, whereas an abrupt stall can be achieved with a signi cant drop in Ssep towards the leading edge at stall. L d L L L d d d L L 10 Ris R1024EN C L Figure 3. Constraints on the lift curve Insensitivity to leading edge roughness is obtained by controlling the location of the transition point on the suction side, Str   before and after C max. To increase the lift-drag ratios at the angles of attack corresponding to the design objectives, Str should in general be as far downstream as possible at and other angles of attack below stall. At C max, Str should be close to the leading edge. The ow on most of the suction side would then be turbulent because of early transition and the transition points would be equally located for both smooth and rough leading edges securing minimal di erence in C max and lift curve slopes. The transition point should remain close to the leading edge throughout the post stall region. The remaining e ect from leading edge roughness would be an increase in drag. As a structural constraint, the airfoil thickness as a function of chord-wise position is constrained to give the desired relative thickness, but also to avoid negative thickness. Other constraints can be added to the airfoil shape or the velocity distribution, the maximum suction side velocities, structural requirements or aerodynamic requirements at other angles of attack. For development of airfoil families, constraints can ensure compatibility of both the aerodynamic characteristics and of the airfoil shapes. To run an optimization, an initial airfoil shape is generated. This can in principle be an arbitrary shape that might be very di erent from the optimum shape. However, computational costs are reduced when the initial design is close to the optimum design. Side constraints are added to the design variables to ensure that they move within reasonable limits. During the optimization, the ow solver calculates the ow for all angles of attack where objectives and constraints are de ned. Typically the ow is solved at a few angles of attack before stall and at several angles of attack in stall. For a reliable optimization process, convergence problems in the ow predictions should be avoided. L d L L Ris R1024EN 11 5 Airfoil family In this section, we present the basis for and result of the design of the airfoil family. 5.1 Validity of geometry description Before presenting the results of the airfoil design, we check that the geometry description is acceptable, in the sense that it should be able to represent many di erent airfoils with a limited number of design variables. This is done by letting the design tool minimize the geometric di erence between a new design and typical wind turbine airfoils, i.e., NACA 63-418, FFA-W3-241, and DU 91-W2-250. That is, the root mean square sum of di erences in y-coordinates is minimized. The results are given in Figure 4, and they show that the geometry description can reproduce the various shapes reasonably well. NACA 63418 Initial design variables Initial design NACA 63418 Final design variables Final design a NACA 63-418 FFAW3241 Initial design variables Initial design FFAW3241 Final design variables Final design b FFA-W3-241 DU 91-W2-250 Initial design variables Initial design DU 91-W2-250 Final design variables Final design c DU 91-W2-250 Figure 4. Geometric representation of wind turbine airfoils On the basis of this exercise, we assume that the geometry description based on a B-spline is capable of generating a large part of the in nite numbers of possible airfoil shapes. 12 Ris R1024EN 5.2 Design criteria In the following, the design of 7 airfoils is described. The relative airfoil thicknesses range from 12 to 30. The design angle of attack is 10 . The objective function is the sum of lift-drag ratios at angle of attacks of 2 , 4 , 6 , 8 , and 10 . The weight factors are all the same, but in a sense the lift-drag ratio at 10 has the largest weight factor since it is the largest and the optimization algorithm would tend to optimize here. Furthermore, a high lift-drag ratio at 10 leads to high lift-drag ratios also at lower angles of attack. The constraints for the three thin airfoils are given in Table 1 and in Table 2 for the four thick airfoils. For all seven airfoils, the upper and lower limits on the lift curve are identical. The design lift at 10 is between 1.53 and 1.55 and the C max of 1.65 should be reached at about 11 . The separation point, Ssep on the suction side is xed to the trailing edge until C max is reached. Separation for a turbulent boundary layer was estimated from H 2.8 as separation criterion as in 27 , where H is the boundary layer shape factor. The constraints on the suction side transition point di er for the thin and thick airfoils. For RIS A12, RIS A15, and RIS A18, the transition point, Str is located on the rst 7 of the chord for angles of attack above the C max-angle. For the remaining thicker airfoils, the transition point is on the rst 10 of the chord. For RIS A24, RIS A27, and RIS A30, an additional constraint is that the ow on the suction side decelerates from 0:4  x=c  0:9 for = 0 . L L L Table 1. Constraints for RIS A12, RIS A15 and RIS A18. 10.0 10.5 11.0 C min 1.53 1.64 C max 1.55 1.65 Ssep min 0.999 0.999 0.999 Str max L L 11.5 12.0 12.5 13.0 13.5 1.64 1.62 1.60 1.58 1.56 1.65 1.65 1.65 1.65 1.65 ; 0.07 ; 0.07 0.07 0.07 0.07 Table 2. Constraints for RIS A21, RIS A24y, RIS A27yand RIS A 30y. 10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5 C min 1.53 1.64 1.64 1.62 1.60 1.58 1.56 C max 1.55 1.65 1.65 1.65 1.65 1.65 1.65 Ssep min 0.999 0.999 0.999 Str max 0.10 0.10 0.10 0.10 0.10 L L ; ; dvis  0 for y =0 and 0:4 x=c   0:9 In Table 3, the operational conditions are given together with selected properties of the resulting airfoil design. The operational conditions are the Reynolds numbers and the Mach numbers corresponding to a typical 600kW wind turbine. The Reynolds and Mach numbers are relatively high for the thinner airfoils in the tip region and on the mid section but lower for the thicker airfoil used in the root region. The maximum lift coe cients according to XFOIL are also given for both clean and dirty conditions i.e., rough leading edge. In calculations with rough leading Ris R1024EN 13 edge, the transition points for the suction and pressure sides were xed to 1 and 10, respectively as in 4 . We see that going from clean to dirty conditions C max drops about 10 for RIS A12 to RIS A24 and about 15 for RIS A27 and RIS A30. L Table 3. Operational conditions and selected properties the airfoil design RIS RIS RIS RIS RIS RIS RIS A12 A15 A18 A21 A24 A27 A30 t=c 12 15 18 21 24 27 30 Re  10,6 : 3:00 3:00 3:00 2:75 2:75 2:50 3 00 free transition, xed transition y M a : 0:16 0:11 0:09 0:07 0:07 0:05 0 20 C max  y 1.6511.5  1.6411.5  1.6412.0  1.6512.0  1.6512.0  1.6512.0  1.6512.0  L C max z 1.5110.0  1.5211.0  1.5311.5  1.5011.0  1.4810.5  1.3911.0  1.3711.0  L z 5.3 Geometric properties The airfoil shapes are given in Figures 5 and 6. Geometrically, RIS A18 to RIS A30 are clearly a family, whereas RIS A12 and RIS A15 do not look like their thicker relatives. The entire family is characterized by a sharp nose. For RIS A27 and RIS A30 the rear part of the suction side is slightly wavy, which it might be possible to remove if not for anything else as for aesthetic reasons with out compromising aerodynamic performance. This has not been tried in this work but it is an obvious possibility for future improvement of the design. RISØ-A-XX Figure 5. Airfoil shapes 14 Ris R1024EN RISØ-A-12 RISØ-A-15 RISØ-A-18 RISØ-A-21 RISØ-A-24 RISØ-A-27 RISØ-A-30 Figure 6. Airfoil shapes, revisited Ris R1024EN 15 5.4 Aerodynamic properties The airfoil polars C ,C , and the suction side transition point, Str are given in Figures 7 to 9 for both clean and dirty conditions. Aerodynamically, they all look a like. This is obviously due to the fact that the aerodynamic constraints are identical more-or-less. But, again, the two thinner airfoils are slightly di erent from the rest, judging from the appearance of the lift and drag curves. RIS A27 and RIS A30 are also special with their 15 drop in C max going from clean to dirty conditions. All the airfoils have a distinct and well-de ned stall with a linearly increasing C until the design angle of attack. The pressure distributions for = 8:5 , Figure 10 underline the subrelations within the family. The family consists of thin members RIS A12 and RIS  A15, intermediate members RIS A18 to RIS A24, and thick members RIS A27 and RIS A30. The pressure distributions are all characterized by a narrow suction peak that appears around 8 when the stagnation point moves slightly downstream on the leading edge part of the pressure side.. This causes the ow to accelerate around the sharp nosed leading edge of the airfoils. The suction peak is not present at low angles of attack but at higher angles of attack it eventually provokes transition from laminar to turbulent ow on the front part of the airfoils. In Figure 10, the transition points on the suction side are all around x=c = 0:3. L D L L 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 12 16 0.0 0.1 0.2 Str =c 0.5 12 16 0.0 0.1 0.2 Str =c 0.5 D 0.3 0.4 a RIS A12 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 D 0.3 0.4 b RIS A15 Figure 7. XFOIL polars. Solid line: free transition, dashed line: xed transition. 16 Ris R1024EN 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 12 16 0.0 0.1 0.2 Str =c 0.5 12 16 0.0 0.1 0.2 Str =c 0.5 12 16 0.0 0.1 0.2 Str =c 0.5 D 0.3 0.4 a RIS A18 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 D 0.3 0.4 b RIS A21 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 D 0.3 0.4 c RIS A24 Figure 8. XFOIL polars. Solid line: free transition, dashed line: xed transition. Ris R1024EN 17 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 12 16 0.0 0.1 0.2 Str =c 0.5 12 16 0.0 0.1 0.2 Str =c 0.5 D 0.3 0.4 a RIS A27 2.00 1.75 1.50 C L 1.25 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.00 0.01 0.02 0.03 0.04 0.05 0.06 -4 C 0 4 8 D 0.3 0.4 b RIS A30 Figure 9. XFOIL polars. Solid line: free transition, dashed line: xed transition. 18 Ris R1024EN -3 -3 -2.5 -2.5 -2 -1.5 CP CP -2 -1.5 -1 -0.5 -1 -0.5 0 0 0.5 0.5 1 1 1.5 0 0.2 0.4 x=c 1.5 0.6 0.8 1 0 0.2 a RIS A12 -3 0.6 0.8 1 0.8 1 0.8 1 -3 -2.5 -2 -1.5 CP -2 -1.5 CP x=c b RIS A15 -2.5 -1 -0.5 -1 -0.5 0 0 0.5 0.5 1 1 1.5 0 0.2 0.4 x=c 1.5 0.6 0.8 1 0 0.2 c RIS A18 0.4 x=c 0.6 d RIS A21 -3 -3 -2.5 -2.5 -2 -1.5 CP -2 -1.5 CP 0.4 -1 -0.5 -1 -0.5 0 0 0.5 0.5 1 1 1.5 0 0.2 0.4 x=c 1.5 0.6 0.8 1 0 0.2 e RIS A24 0.4 x=c 0.6 f RIS A27 -3 -2.5 -2 CP -1.5 -1 -0.5 0 0.5 1 1.5 0 0.2 0.4 x=c 0.6 0.8 1 g RIS A30 Figure 10. XFOIL pressure distributions for dashed line: xed transition. Ris R1024EN : . Solid line: free transition, = 85 19 5.5 Comparison of tions and EllipSys2D predic- XFOIL In the design process we have used XFOIL, which often overestimates C max and give poor results in the post stall region 21 . Therefore, as an extra check, the airfoil designs are evaluated with EllipSys2D, a Navier-Stokes solver 28 with the k , ! SST turbulence model 23 and the Michel transition criterion 24 . In Figures 11 to 17 the C and C predictions of XFOIL and EllipSys2D with both free transition and xed transition are compared. That is, athe free transition predictions of XFOIL are compared to the free transition predictions of EllipSys2D and b the xed transition predictions of XFOIL are compared to the fully turbulent predictions of EllipSys2D. By xing the transition point at the leading edge, the depositing of dirt and bugs is simulated. This comparison goes to show if we would get a completely di erent design if we used another ow solver than XFOIL. Note, that the EllipSys2D predictions are for = ,4 ; 0 ; 4 ; 8 ; 9 ; 10 ; 11 ; 12 ; 13 ; 14 ; 15 ; 16 whereas the less expensive XFOIL predictions are for from ,4 to 16 for every 0:5 . The agreement between the XFOIL and EllipSys2D results is good for the computations with xed transition dirty conditions, whereas the correspondence for free transition clean conditions is not good for RIS A24, RIS A27, and RIS A30. This is due to bad performance of the turbulence and the transition model in stall. Since the designs are based on calculations with free transition, this comparison suggests that the designs would have been identical had we used EllipSys2D instead of XFOIL for RIS A12, RIS A15, RIS A18, and perhaps RIS A21. For RIS A24,RIS A27, and RIS A30 using EllipSys2D as aerodynamic analysis tool would have given di erent but not necessarily better designs compared to the present ones. L 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) 0.12 C 0.09 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 D D L 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 a Free transition 12 16 0.00 -4 0 4 8 12 16 b Fixed transition Figure 11. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A12 L 20 D Ris R1024EN 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a Free transition 4 8 12 16 b Fixed transition Figure 12. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A15 L 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 D 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a Free transition 4 8 12 16 b Fixed transition Figure 13. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A18 L 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 D 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 a Free transition 12 16 0.00 -4 0 4 8 12 16 b Fixed transition Figure 14. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A21 L Ris R1024EN D 21 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a Free transition 4 8 12 16 b Fixed transition Figure 15. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A24 L 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 D 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a Free transition 4 8 12 16 b Fixed transition Figure 16. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A27 L 0.15 1.00 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (XFOIL) CL (EllipSys2D) CD (XFOIL) CD (EllipSys2D) D 2.00 D 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 a Free transition 12 16 0.00 -4 0 4 8 12 16 b Fixed transition Figure 17. Comparison of XFOIL and EllipSys2D predicted C and C for RIS  A30 L 22 D Ris R1024EN 5.6 Comparison of clean and dirty performance In Figures 1824 the C and C predictions with free and xed transition for both XFOIL and EllipSys2D are compared. That is, aXFOIL predictions based on free transition are compared with XFOIL predictions based on xed transition and bEllipSys2D predictions based on free transition are compared with EllipSys2D predictions based on xed transition. This comparison illustrates the decrease in C max and the increase in C going from clean to dirty conditions. Qualitatively, XFOIL and EllipSys2D give the same picture of going from clean to dirty conditions for RIS A12 to RIS A21, whereas the inadequacy of the turbulence and transition modelling in the separated region give di erent pictures for RIS A24, RIS A27, and RIS A30. L D 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) 0.12 C 0.09 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 D L D 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a XFOIL 4 8 12 16 b EllipSys2D Figure 18. Comparison of C and C predictions with free and xed transition for RIS A12 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) 0.12 C 0.09 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 D D L 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 a XFOIL 0 4 8 12 16 b EllipSys2D Figure 19. Comparison of C and C predictions with free and xed transition for RIS A15 L Ris R1024EN D 23 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a XFOIL 4 8 12 16 b EllipSys2D Figure 20. Comparison of C and C predictions with free and xed transition for RIS A18 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) 0.12 C 0.09 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 D D L 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a XFOIL 4 8 12 16 b EllipSys2D Figure 21. Comparison of C and C predictions with free and xed transition for RIS A21 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) 0.12 C 0.09 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 D D L 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 a XFOIL 0 4 8 12 16 b EllipSys2D Figure 22. Comparison of C and C predictions with free and xed transition for RIS A24 L 24 D Ris R1024EN 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) C 0.09 D 0.12 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 0 a XFOIL 4 8 12 16 b EllipSys2D Figure 23. Comparison of C and C predictions with free and xed transition for RIS A27 0.15 1.00 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) 0.12 C 0.09 C C C 1.50 L 0.12 1.00 2.00 0.09 1.50 L 0.15 CL (Free transition) CL (Fixed transition) CD (Free transition) CD (Fixed transition) D 2.00 D D L 0.50 0.06 0.50 0.06 0.00 0.03 0.00 0.03 0.00 -0.50 -0.50 -4 0 4 8 12 16 0.00 -4 a XFOIL 0 4 8 12 16 b EllipSys2D Figure 24. Comparison of C and C predictions with free and xed transition for RIS A30 L Ris R1024EN D 25 6 Conclusion A method for design of wind turbine airfoils was developed. The design method is based on direct numerical optimization of airfoil shapes described by B-splines subject to aerodynamic and structural objectives and constraints. The capabilities of the method was demonstrated by the design of a complete airfoil family composed of 7 airfoils ranging from 12 to 30 in relative thickness. The airfoils were designed for Reynolds and Mach numbers representative of a 600 kW wind turbine. Aerodynamically, the airfoils perform identically, i.e., they have high lift-drag ratio until C max is reached, a design angle of attack of 10 , a design lift between 1.53 and 1.55 and maximum lift around 1.65 at 11 according to XFOIL. Beyond =11 , the lift is constrained to lie within a band to secure a smooth post stall behavior. Moreover, for angles of attack above the maximum lift angle the transition point is located on the rst 10 of the airfoil. This constraint is put on the design to obtain insensitivity to leading edge roughness for C max. Computations with forced transition on the leading edge show as a measure of the insensitivity to leading edge roughness that the maximum lift coe cient does not drop more than 10 to 15 depending on the relative thickness. A geometrical feature of the airfoil family is the sharp nose region that rapidly accelerates the ow and generates a suction peak that eventually leads to transition close to the leading edge. The airfoil designs have been checked with the CFD code EllipSys2D and the results are in good agreement with the results of XFOIL except for the free transition computations for the thicker airfoils RIS A24,RIS A27, and RIS  A30. The discrepancies are due to the poor performance of the turbulence and transition models in post stall. But the comparison between XFOIL and EllipSys2D suggests that the design for RIS A12 to RIS A21 had been the same with the use of EllipSys2D instead of XFOIL. The present design method and the airfoil family itself provides an good basis for further improvements in the design. For some of the airfoils it should be examined if the wavy rear part of the suction side can be straightened out with out compromising the aerodynamic performance. Additionally, the geometric compatibility between the di erent airfoils might also be improved. L L 26 Ris R1024EN References 1 Abbott, I. H., and von Doenhoff, A. E. Theory of Wing Sections. Dover Publications, Inc., 1959. 2 Aly, S., Ogot, M., and Pelz, R. Stochastic Approach to Optimal Aerodynamic Shape Design . Journal of Aircraft 33, 5 Sep-Oct 1996, 956961. 3 Bak, C., Madsen, H. A., Fuglsang, P., and Rasmussen, F. Double Stall. Ris -R-1043EN, Ris National Laboratory, Denmark, 1998. 4 Bj rk, A. Airfoil design for variable RPM horizontal axis wind turbines. In EWEC'89 Glasgow, Scotland, 1989. 5 Chattopadhyay, A., Pagaldipti, N., and Chang, K. T. A Design Optimization Procedure for E cient Turbine Airfoil Design. J. Computers Math. Applic. 26, 4 1993, 2131. 6 Chaviaropoulos, P., Bouras, B., Leoutsakos, G., and Papailiou, K. D. Design of Optimized Pro les for Stall Regulated HAWTs Part 1: Design Concepts and Method Formulation. Wind Engineering 17, 6 1993, 275287. 7 de Boor, C. A Practical Guide to Splines. Springer-Verlag, New York, 1978. 8 Drela, M. XFOIL: An Analysis and Design system for Low Reynolds Number Airfoils. In Low Reynolds Number Aerodynamics 1989, vol. 54 of Springer-Verlag Lec. Notes in Eng. 9 Drela, M., and Giles, M. B. Viscous-inviscid analysis of transonic and low Reynolds number airfoils. AIAA Journal 25, 10 1987. 10 Dulikravich, G. S. Aerodynamic Shape Design and Optimization: Status and Trends. J. of Aircraft 29, 6 Nov-Dec 1992, 10201026. 11 Eppler, R., and Somers, D. M. A Computer Program for the Design and Analysis of Low-Speed Airfoils. Tech. rep., NASA TM 80210, 1980. 12 Eyi, S., Lee, K. D., Rogers, S. E., and Kwak, D. High-Lift Design Optimization Using Navier-Stokes Equations. Journal of Aircraft 33, 3 MayJune 1996, 499504. 13 Fuglsang, P., and Madsen, H. A. Numerical optimization of wind turbine rotors. In EUWEC'96 G teborg, Sweden, 1996. 14 Hager, J. O., Eyi, S., and Lee, K. D. Two-point transonic airfoil design using optimization for improved o -design performance. Journal of Aircraft 31, 5 1994. 15 Henne, P. A. E. Applied Computational Aerodynamics. American Institute of Aeronautics and Astronautics, Inc., 1989. 16 Hicks, R. M., Murman, E., and Vanderplaats, G. N. An Assessment of Airfoil Design by Numerical Optimization. NASA TM X-3092 July 1974, 31 p. 17 Hill, D. M., and Garrad, A. D. Design of Aerofoils for Wind Turbine Use. In IEA Symposium on Aerodynamics of Wind Turbines Lyngby, Denmark, 1988. Ris R1024EN 27 18 Jameson, A. Aerodynamic Design Via Control Theory. Tech. Rep. Rept. 88-64, Inst. for Computer Applications in Science and Engineering, NASA Langley, Hampton, VA., 1988. 19 Liebeck, R. H. Design of Subsonic Airfoils for High Lift. In AIAA 9th Fluid and Plasma Dynamics Conference San Diego, Californien, July 1976. 20 Lighthill, M. J. A New Method of Two-Dimensional Aerodynamic Design . Tech. Rep. R & M Report 2112, Aeronautical Research Council, London, June 1945. 21 Madsen, H. A., and Filippone, A. Implementation and Test of the XFOIL Code for Airfoil Analysis and Design. Ris -R-644EN, Ris National Laboratory, Denmark, 1995. 22 Mangler, K. W. Design of Airfoil Sections. Jahrbuch Goettingen FRG, Deutscher Luftfahrforschung, 1938. 23 Menter, F. R. Zonal Two Equation k , ! Turbulence Models for Aerodynamic Flows. AIAA Paper 93-2906 1993. 24 Michel, R. Etude de la transition sur les pro les d'aile. Tech. Rep. Report 1 1578-A, ONERA, 1952. See White F.M., Viscous uid ow, p. 442. 25 Obayashi, S., and Takanashi, S. Genetic Optimization of Target Pressure Distributions for Inverse Design methods. AIAA Journal 34, 5 May 1996. 26 Press, W. H., Flannery, B. P., Teukolsky, S. A., and Vetterling, W. T. Numerical Recipes: The Art of Scienti c Computing. Cambridge University Press, 1986. 27 Selig, M. S., and Maughmer, M. D. Multipoint Inverse Airfoil Design Method Based on conformal Mapping. AIAA Journal 30, 5 1992. 28 S rensen, N. N. General Purpose Flow Solver Applied to ow over Hill. Ris -R-827EN, Ris National Laboratory, Denmark, 1995. 29 Tangler, J. L.; Somers, D. M. Status of the Special-Purpose Airfoil Families. In WINDPOWER'87, San Fransisco 1987. 30 Timmer, W., and van Rooy, R. Thick airfoils for HAWTs. Journal of Wind Engineering and Industrial Aerodynamics 39 1992. 31 Vanderplaats, G. N. Numerical Optimization Techniques for Engineering Design: With applications. McGraw-Hill Book Company, 1984. 28 Ris R1024EN Bibliographic Data Sheet Ris R1024EN Title and authors Design of the Wind Turbine Airfoil Family RIS AXX Kristian S. Dahl, Peter Fuglsang ISBN ISSN Dept. or group Date 8755023568 01062840 Wind Energy and Atmospheric Physics Department Groups own reg. numbers Project contract No. Pages Illustrations December 1998 29 ENS 1363 95-0001 Tables 3 24 References 31 Abstract Max. 2000 char. A method for design of wind turbine airfoils is presented. The design method is based on direct numerical optimization of a B-spline representation of the airfoil shape. For exibility, the optimization algorithm relies on separate, stand alone tools for the analysis of aerodynamic and structural properties. The panel method based XFOIL is used during the optimization whereas the Navier-Stokes solver EllipSys2D is used in the evaluation of the results. The method is demonstrated by the design of an airfoil family composed of 7 airfoils ranging in thickness from 12 to 30. The design is based on Reynolds and Mach numbers representative of a 600 kW wind turbine. The airfoils are designed to have maximum lift-drag ratio until just below stall, a design lift coe cient of about 1.55 at an angle of attack of 10 and a maximum lift coe cient of 1.65. The airfoils are made insensitive to leading edge roughness by securing that transition from laminar to turbulent ow on the suction side occurs close to the leading edge for post stall angles of attack. The design method and the airfoil family provides a sound basis for further enhancing the characteristics of airfoils for wind turbines and to tailor airfoils for speci c rotor sizes and power regulation principles. Descriptors INIS EDB AERODYNAMICS; AIRFOILS; DESIGN; OPTIMIZATION; TURBINE BLADES; HORIZONTAL AXIS TURBINES; TWO-DIMENSIONAL CALCULATIONS Available on request from: Information Service Department, Ris National Laboratory Afdelingen for Informationsservice, Forskningscenter Ris  P.O. Box 49, DK4000 Roskilde, Denmark Phone +45 46 77 46 77, ext. 4004 4005 Fax +45 46 77 40 13  ...
View Full Document

This note was uploaded on 11/13/2011 for the course AEE 495 taught by Professor O.uzol during the Spring '11 term at Middle East Technical University.

Ask a homework question - tutors are online