28th AIAA Applied Aerodynamics Conference 28 June - 1 July 2010, Chicago, Illinois

AIAA 2010-4224

Winglet Design and Optimization for UAVs

Winglets are known improve the eciency of large aircraft at high subsonic speeds, but winglet designs for smaller aircraft such as UAVs are largely unproven. Winglets improve eciency by diusing the shed wingtip vortex, which in turn reduces the drag due to lift and improves the wings lift over drag ratio. This research investigates methods for designing and optimizing winglet geometry for UAVs that operate at Reynolds numbers near 106 . The design methodology is based on the vortex lattice method. Optimized designs are tested and compared with base designs for validation and include both Whitcomb and blended winglets. Designs are validated using wind tunnel tests. The resulting methodology is then applied to existing UAV platforms for specic performance improvements.

I.

Introduction

The motivation for this research is to develop an ecient method for winglet design and optimization for UAV platforms. While research in winglets has been dominated by commercial applications, with some research applied to sailplanes, little is documented on the methods for optimizing all the parameters for a successful winglet conguration. Research details the parameters and applications for winglets but there are no tools for creating an optimized geometry that can be tailored to each individual application. Using a Matlab based Vortex Lattice Method (VLM) to calculate the aerodynamic properties of the winglet geometry, an array of geometries can be created and evaluated at specic ight conditions to determine the optimum solution. The optimization approach is based on parametric analysis. This research focuses on an open source code with low computational requirements for aerodynamic analysis using the Tornado VLM.1 Building upon that foundation, a graphical user interface was created along with algorithms for generating winglet geometries for specic air vehicle platforms. Using these tools, parametric analysis of various winglet geometries was possible for winglet optimization. These techniques however, are not necessarily limited to UAVs. Research into winglet technology for commercial aviation was pioneered by Richard Whitcomb in the mid 1970s.2 UAVs however generally operate at much lower Reynolds numbers. Research in full size aircraft reveal that winglets can provide improvements in eciency of more than 7%. For airlines this translates into millions of dollars in fuel costs. The justication for use on UAVs is dierent, however. In general there are 3 cases when winglets may be implemented on UAVs: (i) to increase eciency on existing platforms without changing span; (ii) to use for imbedded sensors; and (iii) for yaw control. For example, winglets on a UAV can provide not only greater aerodynamic eciency but can provide a platform for other functionalities. On the KillerBee and Scaneagle UAVs, the winglets are actually vertical stabilizers and are primarily designed to provide stability and control rather than enhanced aerodynamic eciency. Thus in the current paper, we will focus on the rst two scenarios.

II.

Previous Work

Whitcomb laid the groundwork upon which modern winglets are built. Whitcombs research detailed a few of the parameters that contribute to a successful wingtip device. Fig. 1 depicts the geometry that Whitcomb used for the winglets in his research. The parameters for these winglets include an upper winglet with sweep, cant, taper, and a non-symmetric airfoil with a toe-out angle. Whitcomb suggested that winglets
Graduate Associate

Research Assistant; Student Member AIAA. Professor; Associate Fellow AIAA; jdjacob@okstate.edu.

1 of 14
Copyright 2010 by the authors. Published by the AmericanAmerican Aeronauticsof Aeronautics and Astronautics Institute of Institute and Astronautics, Inc., with permission.

should have a toe-out angle. The toe angle refers to the incidence of the winglet with respect to the X-Z plane. A toe-out angle would slant the incidence of the winglet away from the main body. Fig. 2 depicts the toe-out geometry and forces. The toe-out conguration reorients what would be the lifting force for a wing in the horizontal position, it now has a component opposite to the air velocity providing a force similar to thrust, thereby decreasing the overall drag. The upper winglet is aligned with the trailing edge of the wingtip. There is also a lower winglet with sweep, cant, and taper ratio, which is aligned with the leading edge of the wingtip. Though many modern researchers are able to investigate a much broader range of geometries using computational methods, Whitcombs experiments focused on just a few winglet designs grounded in practical applications. While some (including this research) mainly look at the increased eciency in terms of increasing L/D, Whitcomb compared successful winglets against simple wing-tip extensions, which would have a similar impact on the root bending moment of the wing. After all, if the increased lift can only be used to lift the added structural weight necessary to support the winglet in the rst place, can it really be an improvement? Whitcombs research showed that winglets could improve L/D by 9% and reduce lift induced drag by nearly 20% at Reynolds numbers of 5.25106 (per foot). A wing tip extension with an equivalent impact on the root bending moment only improved L/D by 4%. Whitcombs winglets did perform better than wing-tip extensions. Even with these impressive improvements, Whitcomb realized that his winglets were undoubtedly not optimum due to the limitations of the methods used in the research.2 However, Keisuke noted that many factors contribute to the perceived improvements achieved with winglets, and suggested that parasitic drag may have been a large factor in the conclusion that winglets perform better than wingtip extensions.4 Since the winglets were very narrow compared to the wingtip extensions, they would have had less parasitic drag. Keisukes results indicate that if the root bending moment and the parasite drag are kept constant, a planar wing can be designed which is more ecient. While this research is relevant, the Reynolds numbers used in Whitcombs research are far greater than most UAVs experience during cruise conditions.

Figure 1. The winglet geometry used by Richard T. Whitcomb in his research at NASA Langley wind tunnel included an upper winglet aligned with the main wing trailing edge, and a lower winglet, aligned to the leading edge of the main wing.2

Signicant research has been performed on winglet performance, particularly during the late 70s.512 Conleys research into the toe-out angle for the winglets on the Gates Longhorn Learjet showed that small variations in the toe-out angle could help to ne-tune the performance, but ultimately, as is the case with so many aspects of airplane design, the optimum design would depend on which performance characteristic was most important.13 Importantly, it showed that tweaking the design could improve performance. For most designers, a parametric analysis of hundreds or just dozens of designs in a wind tunnel for a chance

2 of 14 American Institute of Aeronautics and Astronautics

Figure 2. Applying a toe-out angle to the winglet yields a force in the direction of the thrust.

to nd a more optimum design is simply not feasible. UAV designers especially, may be limited by time and budget constraints which require a more robust approach to nding an optimum geometry. The Learjet winglets were based on the work by Whitcomb and were simple, straight wing structures mounted vertically at the wing tip. Conley notes that the winglet should be placed aft of the main wing quarter chord so as to not superimpose the increased velocities over the inner surface of the winglet... on the high velocities over the forward region of the wing upper surface.13 Takenaka et. al. used computational models to investigate the uid dynamics and structural stresses associated with winglets and showed that wave drag develops at the wing/winglet junction when the winglet is not blended into the main wing.14 In the multidisciplinary approach the design parameters applied to a commercial aviation airplane at Mach 0.8, and a Reynolds number of about 1.2 106 . Takenaka et. al. optimized their winglet design for multiple objectives and a small number of design variables.14 Through this approach the winglets could be optimized based on minimizing block fuel and GTOW, which are indirectly aected by the performance of the winglets. Importantly, this study shows how interdependent each aspect of aircraft design is, and how winglet optimization can be useful for various aspects of design and development. Takenakas optimizations resulted in a blended winglet design with a leading edge aft of the main wing leading edge. In the design phase Takenaka notes that the dominant parameters for drag reduction are the span length and the cant angle of the winglet. Bourdin et. al. investigated the use of winglets with variable cant angle as active control surfaces. This research showed that by adjusting the cant angle of either one or both winglets aected multiple moment axes, thus providing a coupled control mechanism.15 In another study Bagwill et. al. used VLM models to investigate winglets for RPVs. Bagwill concluded that winglets increase eciency for all cases, though comparisons were made only to the wing without winglets. Bagwill also concedes that only inviscid cases were compared and that nal analysis would need to include viscous eects.16 Pfeier developed a method used to optimize winglet orientation that included both included and prole drag eects.17 While the research only addressed the winglet orientation and not the size or shape, results showed only small dierences in the root and tip incidences between an induced-drag-only solution and one with prole drag included at the high subsonic speeds examined. Other research has focused on more radical designs such as the C-wing.18 Multiple winglets mimicking the wingtip feathers of soaring birds have also been investigated and found to help in the diusion of the wingtip vortices.19 However, such optimization is based mainly on reducing drag due to lift and increasing L/D with little regard to the physical structures that would be necessary to support such a design. This previous research lays a foundation for the validity of the winglet concept, but individually do not specically account for the advantages that might be obtained on smaller platforms like UAVs. The previous research also underscores the many parameters which should be considered when determining what factors inuence an optimum design. Though the most obvious measurement for winglet performance is the reduced drag due to lift, and increased L/D, the weight of the added structure should be considered as well as the ultimate mission for the aircraft. Maughmer developed winglets for high-performance sail planes operating at Reynolds numbers from 7.0 104 to 1.0 106 .20 This is similar to the operating conditions for small to mid-size UAVs. For sailplanes, Maughmer suggests airfoils that can operate at very low Reynolds numbers. Maughmer describes the break-even point as the point at which halving the Reynolds number causes the prole drag coecient to double.20 Maughmber used the break-even point to determine the height of the winglet. Winglets for sailplanes are most benecial for competition classes where the wingspan is limited.
3 of 14 American Institute of Aeronautics and Astronautics

III.

Goals

The goals of this research are two-fold: i) develop winglet designs applicable for generic small UAVs, and ii) determine performance improvements when implementing winglet designs on UAVs. A. Develop Winglet Designs Applicable for Generic Small UAVs

Unmanned Air Vehicles (UAVs) are aircraft that can be piloted from a remote control station by a trained operator or can be programmed to y autonomously using onboard sensors, computers, and GPS. Increasingly, the dangerous and monotonous work that was once performed by manned aircraft is being turned over to UAVs. UAV design is mainly focused around the intended mission and size is not limited by the constraints of providing a habitat for a human pilot but rather by the size of the intended payload. While there are some UAVs that are very large, such as the Predator and the Global Hawk, other UAVs are much more diminutive, such as the Shadow and ScanEagle. Smaller UAVs cost less to purchase and operate, and because there is no human pilot, UAVs can be considered expendable. UAVs seem to have sparked a renaissance in aircraft design. Designers and engineers are using all their ingenuity to design UAVs for every mission imaginable, which has led to an explosion in the number of aircraft being designed and built. Because UAV size is only limited by technology and the laws of physics, engineers are free to develop unique aircraft systems for which there is no equivalent in current commercial aviation. For each unique design, there are unique challenges, with unique and often unconventional solutions. The goal for this research is to develop a methodology whereby winglet technology can be applied to a UAV platform and predict the change in performance in order to optimize the winglet geometry for the best performance. The methodology should include all the geometric parameters for dening the winglet, and include calculations for predicting the performance of each geometric conguration for comparison and analysis. B. Determine Performance Improvements When Implementing Winglet Designs On UAVs

Whitcombs criteria measured performance as reduced lift induced drag, increased L/D with equal root bending moment.2 Takenaka measured performance relative to MTOW and fuel block. Takenakas approach used CFD and FEA with analytic comparisons to determine the best performance.14 Maughmer used a modied crossover point to describe the performance as a function of weight, lift and drag.20 This helped to constrain the winglet geometries to parameters within performance limits. Maughmers cross-over point compares the change in drag, Dprof ile and Dinduced , of the wing with winglets to the wing with no winglets. There is a critical speed above which the cross-over point is exceeded and winglets are no longer benecial.20 Calculating lift and drag forces as well as lift and drag coecients help to ll in the equations used for evaluating performance. Range, endurance, and take o distance are just a few examples of the types of performance that can be evaluated, compared, and optimized. For a transcontinental airliner, range would be important. For a spy plane that needs to loiter over a geographic location, endurance would be critical. For a STOL vehicle, takeo distance is an important performance parameter. When dening performance improvements it is important to specify what type of performance is most important. For this research, performance will focus mainly on improvements to CL /CD , with consideration given to the root bending moment.

IV.
A. Winglet Design

Methodology

The focus of this research is winglets that operate at much lower Reynolds numbers. As such, generic UAV requirements are used as a baseline based on typical Group 2 and Group 3 UAVs. The specic ight conditions are for velocity of 65 knots at an altitude of 10,000 ft. For the specic application, a swept and tapered wing with a tip chord of 18 inches is used for the base conguration, this results in a Reynolds number below one million. Winglet geometries investigated include straight Whitcomb type winglets as well as blended winglets. Blended winglet designs will be the main focus for optimization because they have been shown to provide up to 60% improvement over conventional wingtips, and are prominent in wing tip design for major commercial aviation designs. Various airfoils are included in the investigation, including NACA 0009, Eppler 396, and Selig S9026. NACA airfoils help to validate and benchmark data, while other airfoils

4 of 14 American Institute of Aeronautics and Astronautics

were selected for their performance at the Reynolds numbers being investigated. Winglet parameters that will be considered include: airfoil, root chord, taper ratio, sweep, cant, semi span, twist (toe-out/in), and blend radius. Fig. 3 illustrates the many parameters that dene the winglet geometry.

Figure 3. Parameters used to dene the winglet geometry include: root chord, tip chord, sweep, height (length), radius, cant and twist.

A properly designed winglet will diuse the vortex that is shed at the tip of the wing. The winglet must produce a side force in order to be eective.2 The side forces reduce the inow above the wing at the tip, and the outow below the wing at the tip. The reduction in the inow and outow help to normalize the lift distribution along the entire span of the wing, just as a wing with a higher aspect ratio will have a more even lift distribution. A winglet should produce the same lift distribution as extending the length of the wing by 45% of the height of the winglet.3 The winglet should cause the downwash over the surface of the wing to remain constant, with no sidewash. By improving the ow of the downwash the lifting eciency of the wing is improved, and by diusing the wingtip vortex the drag is reduced. Winglets can oer a great advantage when the wing span is limited or an existing wing must be modied for extended performance. B. Vortex Lattice Method

The vortex lattice method (VLM) is the foundation for the optimization methods that were used. Computer algorithms for solving the vortex lattice method were employed to determine the tness of each design iteration. Vortex lattice theory is based on Prandtls lifting line theory. Prandtl suggested that a xed vortex lament would be subject to a force from the Kutta Jukowski theorem. To determine the lifting force on a wing, Prandtl replaced the wing with a xed vortex lament, and since a vortex cannot end in a uid, Prandtl connected a vortex at each end that extend to innity. The horseshoe vortex replaces the wing. The combination of vortices is known as a horseshoe vortex.21 The downwash created by the xed vortex along the span of the wing could be determined, where the vortex along the wing is located from -b/2 to b/2. For a more precise calculation of the lift along the span of the wing, more horseshoe vortices can be added, each with a span less than the previous. Extending the number of horseshoe vortices results in the integral form of the induced velocity along the span of the wing, then the lift can be calculated as the integral over the span of the wing. The vortex lattice theory21 begins with a basic two dimensional denition of the wing geometry then superimposes a grid on top of the wing. For each square in the grid there is a control point and a horseshoe vortex. The velocity at the control point is deduced by applying the Biot-Savart law, to each segment of the horseshoe vortex that surrounds the control point. The Biot-Savart law describes the strength of each vortex line in the horseshoe. From the strength of each vortex the velocity at the control point can be determined. The boundary condition stipulates that the ow must be parallel to the surface. These equations are placed in a matrix corresponding to their location in the lattice across the surface of the wing. Computer algorithms are able to solve these matrices quite eciently. The code used to solve the equations of the vortex lattice method in this research is Tornado.1

5 of 14 American Institute of Aeronautics and Astronautics

C.

Tornado VLM

Tornado is an open source program written in Matlab by Thomas Melin.1 Tornado is a solver for the vortex lattice method. The power of Tornado lies in its robust wing geometry denition. The parameters that can be dened for wings in Tornado include span, root chord, taper ratio, sweep, dihedral, root airfoil, tip airfoil, root twist, tip twist, x, y and z location. Tornado can also include aps for control surfaces which can be set to dierent angles. Fig. 4 illustrates as example of an aircraft dened using Tornado. Using multiple partitions for a wing allows variable sweep and taper ratio along a single wing. In this example, a wing has even been dened to represent the fuselage of the plane. Because it is open source, Tornado is easily modied to automate the geometry generation and computations for parametric analysis. Tornado is written in Matlab, so it works equally well on any computer that supports Matlab. The power of Matlab lies in its ability to solve sets of equations in matrix form. In this way it is a perfect platform for dealing with problems like vortex lattice theory, which use a matrix to represent the mathematics of the problem.

Figure 4. The wing geometry for the RQ7 Shadow 200 modeled in Tornado VLM for Matlab. On the left are the side, front, top and isometric views of the wings with partitions, panels, MAC, and reference point. The gure on the right shows the model with the collocation points associated with each panel.

D.

Pecos Bill: The Tornado Tamer

Tornado is very capable at solving the vortex lattice method, however it is not especially user friendly. In an eort to make Tornado more user friendly, the author programmed a graphical user interface using the GUIDE tool in Matlab. The graphical interface is called Pecos (short for Pecos Bill, the mythical tornado tamer of the Wild West). Pecos was designed to automate some of the basic data inputs required for geometry and state denition as well as provide an immediate feedback to the user for each change made to the geometry. Pecos links the user interface to all the functions of the Tornado program without the need to navigate various menus and commands. For the purposes of winglet research, a module was created for winglet generation. This module allows for the denition of all the parameters necessary to dene the winglet geometry. Fig. 3 illustrates all the parameters that can be dened using the winglets module in Pecos. Along with these, Pecos can also dene a dihederal angle for the root of the winglet, the root twist angle, and the airfoil for the winglet. An example is shown in Fig. 5.

V.

VLM

The main wing for the test model that would be evaluated had an Eppler 396 airfoil. The test conditions were set to an airspeed of 65 knots at an altitude of 10,000 ft. The design requirements for the winglets called for a total height xed at 1 meter. The code for generating the winglet was adjusted so the length parameter could be used as the input for setting the total height of the winglet. This code is reected in the PECOS genwinglet.m le for Matlab. A switch was dened to change the length parameter according to winglet requirements, whether the winglet design required a xed length, height, or straight section beyond the blend radius. Table V contains the winglet parameters used for each conguration during parametric

6 of 14 American Institute of Aeronautics and Astronautics

Figure 5. Sample Pecos panel showing winglet denition.

batch processing for design optimization. Fig. 6 shows the results computed by Tornado for conguration BFC03 as listed in V, with the radius varying from 0.1 to 0.9 meters and the angle of attack varying from -10 to 20 . The rst plot displays CL /CD , and the plot on the right shows the root bending moment. CL /CD was computed using the the CL as computed by Tornado, however, CD was the sum of CD as computed by Tornado plus the zero lift drag coecient. The zero lift drag coecient is an estimate of the surface drag based on the surface area of the wings. If the zero lift drag coecient was not added, CD would be zero at 0 , this would cause a premature peak for CL /CD at 0 , and would be of no value for predicting performance. Fig. 6 shows that CL /CD increases for larger radii. The root bending moment is constant for the change in radius and is inuenced mainly by the angle of attack. Fig. 7 shows the results for conguration BFC03 with the cant angle varying from 50 to 90 . The plots show that for CL /CD the best results are achieved for the lower cant angles. However, since the length is dened to maintain a constant height of 1 meter, the lower the cant angle, the longer the winglet will need to be to reach the necessary height. Therefore the optimum cant angle will be a compromise between total length (weight) and higher CL /CD . Fig. 8 shows the results for conguration BFC03 with the sweep angle varying from 0 to 40 . The values are almost invariant in relation to the sweep angle. Fig. 9 shows the results for conguration BFC03 with the cant angle varying from 60 to 90 and the blend radius varying from 0.1 to 0.9 meters. The angle of attack for this plot is set to 0 . CL /CD |max lies at the maximum radius and the minimum cant angle. The minimum root bending moment however lies at the cant angle above 70 . The optimum solution, will be a compromise. In order to prevent creating a winglet that is too large and heavy and to accommodate a straight section at the end of the winglet, the blend radius was set to 0.457 meters, or equal to the wingtip chord length and the cant angle was set to 77 . This cant angle is a compromise for optimum L/D and root bending moment as well as maintaining a shorter overall length. The sweep was set to 27 , so the winglet did not extend too far beyond the trailing edge of the wing. This sweep also aligns the sweep of the leading edge of the wing to the leading edge of the winglet. Fig. 10 shows the results for conguration BFC04. Plotting CL /CD versus the twist angle and sweep angle, there is a maximum for CL /CD is located at a twist angle of -6 and a sweep angle of 27 . The minimum value for the root bending moment is at the maximum sweep angle. The range for CL /CD is approximately 3 with an increase of 12%. The range for the root bending moment is 200 with an increase of 7%. For the nal optimized design a tip twist angle is set to -6 and the sweep angle is set to 27 .

7 of 14 American Institute of Aeronautics and Astronautics

Conguration BFC03 BFC04 Optimized

Height (m) 1.00 1.00 1.00

Root Chord 0.457 0.457 0.457

Tip Chord 0.15 0.15 0.15

Radius 0.457 0.457 0.457

Cant 90 77 77

Sweep 27 27 27

Twist 0 0 -6

Table 1. Winglet Parameters for Test Model

Figure 6. Matlab plot for conguration BFC03 with blend radius varying from 0.1 to 0.9 meters. Angle of attack varies from -10 to 20 . Left plot indicates CL /CD , right plot shows the root bending moment.

VI.

UAV Performance Evaluation

The performance improvements can be measured in terms of range and endurance. Using eqn. (1) estimates for the range can be compared. Using eqn. (2) estimates for the endurance can be compared. Table VI lists the values for the predicted performance improvements for this particular case. Using the optimized blended winglets, the maximum value for CL /CD is increased by 28%. This results in a 71% increase in endurance and the range is increased by 28%. The root bending moment at the stall angle of 15 is increased by 17%. The blended winglet also performs better than the Whitcomb winglet. The predicted performance improvements for the blended winglet are 15.61% and 33.96% better than the range and endurance respectively for the Whitcomb winglet. The root bending moment for the blended winglet is only 5% more than the Whitcomb winglet. Since the design parameters for the notional UAV require a winglet that is 1 meter high the comparison of the optimized winglet to the Whitcomb winglet is more relevant than the comparison to the conguration with no winglet. The added bending moment at the root will require more structural strength, which means the overall structure will be heavier. The performance improvements for the bended winglet are greater for the overall increase in the root bending moment when compared to the Whitcomb winglet. Fig. 11 plots the predicted performance for the model with no winglets, a Whitcomb winglet, and the optimized blended winglet. R= E= pr c 2 S pr L W0 ln c D W1 CL 1/2 1/2 (W1 W0 ) CD
3/2

(1) (2)

The three wing congurations compared in this section are illustrated in Fig. 12. Wing 1 includes the wing only, wing 2 includes the the wings, lower winglets and and a straight upper winglet. Wing 3 includes the wings, lower winglets and blended upper winglets. Table VI displays the predicted lift and drag coecients produced using the Tornado VLM. CL /CD |%1 is the percentage increase in CL /CD |max of the winglet design over the no winglet. CL /CD %2 is the percentage increase in CL /CD |max of the blended winglet design (Wing 3) over the Whitcomb winglet design (wing 2). Vortex Lattice Method (VLM) model predicts the CL /CD |max value for wing 3 is 29.46, CL /CD |max for wing 2 is 25.43, and CL /CD |max for wing 1 is 23.00. The equates to
8 of 14 American Institute of Aeronautics and Astronautics

Figure 7. Matlab plot for conguration BFC03 with cant angle varying from 50 to 90 . Angle of attack varying from -10 to 20 . Left plot indicates CL /CD , right plot shows the root bending moment.

Figure 8. Matlab plot for conguration BFC03 with sweep angle varying from 0 to 40 . Angle of attack varying from -10 to 20 . Left plot indicates CL /CD , right plot shows the root bending moment.

an 15.9% increase for wing 3 over wing 2, and 28.1% improvement over wing 1. These same congurations were tested in the Kirsten wind tunnel at the University of Washington in Seattle, Washington. Additional details on the wind tunnel tests are provided in Weierman.23 The results from the vortex lattice method can be compared to the results from the wind tunnel tests. The values measured in wind tunnel testing indicated a CL /CD |max of 14.9, 16.2 and 18.7 for wings 1, 2, and 3 respectively. The values for CL /CD |max were less for the wind tunnel measurements than those for VLM calculations. This should be expected since the drag coecients from the wind tunnel are higher. VLM predicts lower drag coecients because it does not measure viscous eects. The measured values show that wing 3 had an actual improvement over wing 2 of 11.8% and 38.7% improvement over wing 1. So, although the values diered, the trends were similar, and the performance improvement for the blended winglet over no winglet was actually better than predicted by the VLM model. Fig. 13a displays the data from the wind tunnel tests for the lift coecient for all three wing congurations vs. alpha as well as the predicted the lift coecient from Tornado. At the lower angles of attack, the values are nearly the same for the wind tunnel and VLM. At higher angles of attack, as the model approaches the stall angle, the lift curve for the wind tunnel data diverges, while the predicted values from the VLM continue. This is expected since, VLM does not predict stall or any viscous eects that predict ow separation. Fig. 13b compares the wind tunnel measured results for the lift coecient over the drag coecient vs. alpha. The curves for the VLM predictions are higher, but still follow the same trends among the the three wings ads the wind tunnel data. A higher value is expected for the VLM model because it does not model viscous drag eects, but can only approximate surface drag, therefore the drag from VLM will be lower, resulting in a higher value for CL /CD .
9 of 14 American Institute of Aeronautics and Astronautics

Figure 9. Matlab plot for conguration BFC03 with cant angle varying from 60 to 90 and the radius varying from .1 to .9 meters. Left plot indicates CL /CD , right plot shows the root bending moment.

Figure 10. Matlab plot for conguration BFC04 with tip twist angle varying from -15 to 5 . Sweep angle varying from 10 to 60 . Left plot indicates CL /CD , right plot shows the root bending moment.

Fig. 13e plots the values for the lift coecient versus the drag coecient for the three wing congurations, both the predicted values from the VLM and the measured values from the wind tunnel tests. This plot illustrates the error in the VLM for the drag coecient. The predicted values for the drag coecient are lower than measured values, this skews the plots for the VLM calculations to the left. Also, because the VLM is inviscid, it fails to predict the stalling of the wing which is evident in the wind tunnel measurements near the CL value of 1.5. If better drag predictions are available, the baseline VLM calculations can be modied. In Fig. 13d the values for CD0 from the wind tunnel results replace the estimated values for CD0 . When the corrected value for CD0 is used the plots for the predicted CL /CD are much closer to the measured values for CL /CD . Table VI lists the results for the performance parameters as calculated using Tornado and as measured in the wind tunnel. Range1 is the percentage increase in range of the winglet design over no winglet. Range2 is the percentage increase in range of the blended winglet design (Wing 3) over the Whitcomb winglet design (Wing 2) . Likewise Endurance1 is the percentage increase in endurance of the winglet design over no winglet. and Endurance2 is the percentage increase in endurance of the blended winglet design over the 3/2 Whitcomb winglet design. The endurance is a function of CL /CD and the range is a function of L/D. The range is calculated using eqn. (1), however for comparison, all variables were assumed constant except for L/D. The endurance is calculated using eqn. (2), and likewise, all variables were assumed constant except 3/2 for CL /CD , for comparison. Estimates were rst made using the predicted value for CD0 from Tornado, then the estimated value for CD0 from the wind tunnel results was substituted. The VLM model with

10 of 14 American Institute of Aeronautics and Astronautics

Optimized Winglet Increase Over No Winglet CL /CD |max RBMstall Range Endurance 28.03% 17.11% 28.03% 71.70% Optimized Winglet Increase Over Whitcomb Winglet CL /CD |max RBMstall Range Endurance 15.61% 5.16% 15.61% 33.96%
Table 2. VLM Predictions: Optimized Winglet Performance Comparison

Figure 11. Graph of VLM results for the test model with no winglet, a Whitcomb winglet and the optimized winglet for comparison of CL /CD vs. Alpha and root bending moment vs. Alpha.

CD0 |V LM predicts an increase in range of 10.75%for the Whitcomb winglet design (Wing 2) and a 28.03% increase in range for the blended winglet design (Wing 3). The VLM model predicts that the Whitcomb winglet can increase endurance by 28.17% and the blended winglet can increase the endurance by 71.7%. By substituting the value for CD0 |W T , the VLM predictions become very close to the values measured in the wind tunnel. Using CD0 |W T , the Range is increased by 14.49% using the Whitcomb winglet and 31.85% using the optimized blended winglet. The Whitcomb winglet is predicted to increase the endurance by 24.55%. Tornado predicts that the Blended winglet will increase the endurance by 22.21%. The next section of table VI lists the results from the wind tunnel test. The Whitcomb winglet (Wing 2) provides a modest increase in range of 12.4% while the blended winglet provides a 29.1% increase in range. The wind tunnel measurements show an increase of 18.8% for endurance for the Whitcomb winglet and 38.7% increase in endurance for the blended winglet design. The predictions for the increase in range is quite good, whether using CD0 |V LM or CD0 |W T . The predictions for endurance are higher than were calculated using wind tunnel measurements. The predictions improve when CD0 |V LM is replaced by CD0 |W T , yet remain considerably higher.

VII.

Conclusions

A VLM approach was developed to design and optimize winglets for UAVs. The Matlab based Tornado VLM code was used with a wrapper to provide all the input parameters to dene optimum winglet geometries. This proved to be an ecient tool for parametric analysis and winglet design optimization. The inviscid nature of the VLM provides poor results for predicting the parasite drag of each conguration. Chattots research indicates that parasitic drag does not seem to have an impact on determining the optimum geometry, but does have an impact on reducing the eciency.22 This rearms the validity of using a VLM based approach for design and optimization. Although Tornado can make estimates for the zero lift drag coecient, that value was determined to be underestimated through comparison with wind tunnel tests. In the case of these studies the value for CD0 was underestimated by 40-50% compared to experimental data. This error can lead to gross exaggeration of performance predictions since they are based on predictions of CL /CD
11 of 14 American Institute of Aeronautics and Astronautics

Figure 12. Three wings were used for comparison. Wing 1 has no winglets, wing 2 is a Whitcomb style design, and wing 3 has a straight lower winglet and a blended upper winglet.

VLM Model Predictions, CD0 |V LM Wing # CLmax CD0 |V LM CL /CD |max Wing 1 2.93 0.0143 23.00 Wing 2 3.15 0.0168 25.43 Wing 3 3.55 0.0178 29.46 VLM Model Predictions, CD0 |W T Wing # CLmax CD0 |W T CL /CD |max Wing 1 2.93 0.0322 15.55 Wing 2 3.15 0.0343 17.80 Wing 3 3.55 0.0353 20.50

CL /CD %1 N/A 10.6 28.1 CL /CD %1 N/A 10.75 28.03

CL /CD %2 N/A N/A 15.9 CL /CD %2 N/A N/A 15.61

Table 3. VLM Predictions

or CL /CD . The inviscid calculations are also error prone as the aircraft state approaches the stall angle. However, the trends predicted using VLM compare very well to experimental data at the cruise conditions. The Tornado VLM accurately predicted a more optimized geometry for the winglet design.

3/2

Acknowledgments
This work was supported in part by NexGen Aeronautics. Their support and helpful discussions with Dana Howard and Matthew Scott are appreciated. We also acknowledge helpful discussions with Prof. Andy Arena of OSU.

References
1 Melin, Thomas, A Vortex Lattice MATLAB Implementation for Linear Aerodynamic Wing Applications, Master Thesis, Royal Institute of Technology (KTH), December 2000. 2 Whitcomb, Richard T., A Design Approach and Selected Wind-Tunnel Results at High Subsonic Speeds for Wing-Tip Mounted Winglets. NASA TN D-8260, July 1976. 3 Kroo, Ilan, Drag Due to Lift: Concepts for Prediction and Reduction. Annu. Rev. Fluid Mech. 2001. 33:587-617. 4 Keisuke, Asai Theoretical Considerations in the Aerodynamic Eectiveness of Winglets, Journal of Aircraft Vol. 22, No. 7, July 1985. 5 Ishimitsu, Kichio K. Aerodynamic Design and Analysis of Winglets, AIAA Paper 76-940, Sept. 1976. 6 Montoya, Lawrence C., Stuart G. Fletcher and Peter F. Jacobs, Eect of Winglets on a First-Generation Jet Transport Wing, II - Pressure and Spanwise Load Distributions for a Semispan Model at High Subsonic Speeds, NASA TN D-8474, July 1977.

12 of 14 American Institute of Aeronautics and Astronautics

Wing Wing Wing Wing

# 1 2 3

CLmax 1.53 1.60 1.70

CD0 0.0322 0.0343 0.0353

CL /CD |max 14.86 16.24 18.66

CL /CD %1 N/A 9.03 38.67

CL /CD %2 N/A N/A 11.83

Table 4. Wind Tunnel Results

VLM Model Predictions, CD0 |V LM Wing # Range1 Range2 Endurance1 Wing 2 10.8% N/A 28.2% Wing 3 28.0% 15.6% 71.7% VLM Model Predictions, CD0 |W T Wing # Range1 Range2 Endurance1 Wing 2 14.5% N/A 24.6% Wing 3 31.9% 15.2% 52.2% Wind Tunnel Results Wing # Range1 Range2 Endurance1 Wing 2 12.4% N/A 18.8% Wing 3 29.1% 14.9% 38.7%

Endurance2 N/A 34.0% Endurance2 N/A 22.2% Endurance2 N/A 16.8%

Table 5. Increase in Range and Endurance Over Baseline

Harry H., Gregory D. Riebe and Cynthia L. Fulton, Theoretical Parametric Study of the Relative Advantages of Winglets and Wing-Tip Extensions, NASA TP 1020, 1977. 8 Reynolds, P.T., W. M. Gersten and Voorhees, C. G., Gates Learjet Model 28/29, the First Longhorn Learjet, AIAA Paper 78-1445, August 1978. 9 Marchman III, James F., David Manor, and Fred Faery, Whitcomb Winglet applications to General Aviation Aircraft, AIAA Paper 78-1478, Aug. 1978. 10 Spillman, J.J., The use of tip sails to reduce vortex drag, Aeronautical Journal, Sept. 1978, pp. 387-395. 11 Holmes, Bruce J., Cornelis P. van Dam, Philip W. Brown and Perry L. Deal, Flight Evaluation of the Eect of Winglets on Performance and Handling Qualities of a Single-Engine General Aviation Airplane, NASA TM 81892, December 1980. 12 Asai, K. Theoretical Considerations in the Aerodynamic Eectiveness of Winglets Journal of Aircraft, Vol. 22, No. 7, July 1985. 13 Conley, Norm, Winglet Toe-Out Optimization for the Gates Learjet Longhorn Wing, J. Aircraft, Vol. 17, No. 12. December 1980. 14 Takenaka, Keizo and Hatanaka, Keita, Multidisciplinary Design Exploration for a Winglet, Journal of Aircraft, Vol. 45, No. 5, September-October 2008. 15 Bourdin, P., Gatto, A., Friswell, M.I., The Application of Variable Cant Angle Winglets for Morphing Aircraft Control, AIAA Paper 2006-3660, 24th Applied Aerodynamics Conference, 5-8 June 2006, San Francisco, CA. 16 Bagwill, Tracy L., Selberg, Bruce P., An Inviscid Aerodynamic Investigation of Wing/Winglet Geometries for RPV Applications, Aerospace Atlantic Conference and Exposition, Dayton, Ohio, 18-22 April, 1994. 17 Pfeier, N. Numerical Winglet Optimization, AIAA Paper 2004-0213, 42nd AIAA Aerospace Sciences Meeting, Jan. 5-8, 2004, Reno, NV. 18 Kroo, Ilan, McMasters, J., Smith, S. C., Highly Nonplanar Lifting Systems, September 1995. 19 Smith, M. J., et. al., Performance Analysis of a Wing With Multiple Winglets, AIAA-2001-2407. 20 Maughmer, Mark D., Design of Winglets for High-Performance Sailplanes Journal of Aircraft, Vol. 40, No. 6, NovemberDecember 2003. 21 Anderson, John D. , Fundamentals of Aerodynamics, 4th edition, pp 437-443, McGraw-Hill, New York , NY 2007. 22 Chattot, Jean-Jacques, Design and Analysis of Wing/Winglet Combinations Including Viscous Eects at Low Speeds, Journal of Aircraft, 12 July 2005. 23 Weierman, J. Winglet Design and Optimization for UAVs, M.S. Thesis, Oklahoma State University, July, 2010.

7 Heyson,

13 of 14 American Institute of Aeronautics and Astronautics

(a) CL vs. : Results from wind tunnel testing and results from (b) CL /CD vs. : Results from wind tunnel testing and reVLM model. Correlation is strong in lower angles, correlation sults from VLM model. VLM does not calculate viscous drag, decreases as the wind tunnel model approaches the stall angle. so the curve for the VLM models is higher. Though the trends for three wings are similar for both the wind tunnel test and VLM model.

(c) CL vs. CD : Results from wind tunnel testing and results from VLM model. The trends between the dierent wings are similar for both methods, but the uncorrected VLM predictions are optimistic.

(d) CL /CD vs. : For the VLM plots, CD0 |V LM was replaced by CD0 |W T for CD in the calculation of CL /CD , resulting in much closer predictions.

Figure 13. Comparison between VLM predictions and wind tunnel tests for 3 wing congurations.

14 of 14 American Institute of Aeronautics and Astronautics