American Institute of Aeronautics and Astronautics092407

1

Aerodynamic Analysis of the Unmanned Aerial Vehicle forEcological Conservation

Pedro J. Boschetti1, Elsa M. Cárdenas2 and Andrea Amerio3

Universidad Simón Bolívar, Sede del Litoral, Caracas, 1080, Venezuela

The present work has as objective to do a complete aerodynamic analysis of theUnmanned Aerial Vehicle for Ecological Conservation. The panel method code PAN AIR isused to compute the inviscid flowfield. The viscous effects in drag are estimated by theclassic Hoerner method, and the maximum lift coefficient via the classic Valarezo and Chinmethod. The numerical aerodynamic forces of the complete airplane are compared toexperimental data for validation. The spanload and wing pressure distribution are estimatedfor four configurations: wing, wing-body, wing-body-tail, and wing-body-tail with wingtwist. The sources of induced drag for all configurations are achieved graphically via Trefftzplane. All the data were estimated at cruise flight, Reynolds number equal to 1.413×106 andMach 0.15.

NomenclatureB = fuselageCDp = viscous drag coefficient or parasite drag coefficientCLmax = maximum lift coefficientCLo = lift coefficient at zero angle of attackCLα = lift slopeCMo = pitching moment coefficient at zero angle of attackCMα = pitching moment coefficient slopeCp = pressure coefficientCpl = pressure coefficient at lower side of panelCpu = pressure coefficient at upper side of panelc = chorde = Oswald efficiency factoren = two–dimensional free transition criterionk = lift dependent drag factor or induced drag factor, k = (π·eRA)-1 (L/D)max = maximum lift–drag ration = fixed parameter for two–dimensional free transition criterionRA = wing aspect ratioT = tail assembly and twin-boomt = twistW = wingxTrefftz = distance between the gravity center to the Trefftz Planey = transversal axis∆Cp = pressure coefficient difference

I. IntroductionINCE 1917 in Venezuela, the Lake of Maracaibo has been a petroleum extraction zone. The continuous oilleakages from extraction towers and transport pipelines have negatively affected its delicate ecosystem for the

last 90 years. Because early detection of the oil leakages helps to minimize the ecological and economical damage,

1 Assistant Professor, Department of Industrial Technology, Valle de Sartenejas, 89000, AIAA Member.2 Assistant Professor, Department of Industrial Technology, Valle de Sartenejas, 89000, AIAA Member.3 Aggregate Professor, Department of Industrial Technology, Valle de Sartenejas, 89000, AIAA Member.

S

47th AIAA Aerospace Sciences Meeting Including The New Horizons Forum and Aerospace Exposition5 - 8 January 2009, Orlando, Florida

AIAA 2009-1480

Copyright © 2009 by The Authors. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

American Institute of Aeronautics and Astronautics092407

2

Petróleos de Venezuela S. A. (PDVSA), a petroleum company, carries out daily patrols using manned helicopters.These can operate only in daylight and under good climatic conditions, and their activity is relatively expensive. In2002, the design of an unmanned aerial vehicle for ecological conservation (ANCE, for its Spanish acronym) wasinitiated as a joint project between the Universidad Nacional Experimental Politécnica de la Fuerza Armada(UNEFA) and the Universidad Simón Bolívar (USB).1

The ANCE design presents a small, twin-boom, pusher–propeller airplane with a maximum takeoff mass of182.055 kg, capable of carrying 40 kg of payload in a high-technology camera to find oil leakages during daylight orat night. The propeller is powered by a 26-kW two-stroke engine with two pistons. The wingspan of the vehicle is5.18 m, with a rectangular straight wing with no twist, or dihedral of 3.13 m2 of surface area, and a wing aspect ratioof 8.57. The wing section is a NACA 4415 airfoil along the whole wingspan. It is expected that the ANCE will havea cruise speed of 41.65 m/s at 2438 m above sea level for a wing Reynolds number of 1.413×106.1,2 Figure 1 showsan isometric view of the airplane design.

Early wind-tunnel tests helped in the drag cleanup process by drag and lift estimation,3,4 although pressuredistribution and spanload were not measured.

Panel methods for aerodynamic applications have been used since the 1960 decade for industry, research andacademy. They have been successfully applied to compute potential flow in complex and complete aircraftconfigurations.5 Although a panel method could not predict the viscous effects by itself, like lift loss and flowseparation, several methods are available that couple a panel code with boundary layer estimation6 to predict liftloss7 and maximum lift coefficient.8,9 For this reason, when the angle of the attack approaches stall, the discrepancybetween experimental data and numerical resultsincreases. The advantage of the panel methods is thatthey only require a surface panel representation of thegeometry, instead of a complete flowfield used by finitedifference, finite element and volume elementmethods.10

The present work has as objective to do a completeaerodynamic analysis of the ANCE. Panel method PANAIR is used to compute the inviscid flowfield. Theviscous effects in drag are estimated by the classicHoerner method,11 and the maximum lift coefficient viathe classic Valarezo and Chin method.8 The numericalaerodynamic forces of the complete airplane arecompared to experimental data for validation.3,4 Thespanload and wing pressure distribution are estimatedfor four configurations: wing, wing-body, wing-body-tail, and wing-body-tail with wing twist.12,13 The sourcesof induced drag for all configurations are achievedgraphically via Trefftz plane.

II. Aerodynamic Analysis Method

A. Panel Method AnalysisA panel method code PAN AIR, version A502i,14 was used to perform the inviscid aerodynamic analysis of the

four configurations in study. This is a high–order panel method code capable of solving a variety of boundary valueproblems in steady subsonic or supersonic inviscid flow by the classic three–dimensional Prandtl–Glauert equationfor linearized compressible flow. The configuration is represented by a distribution of linear source and quadraticdouble singularities, each of which is a solution of Prandtl–Glauert equation. The singularity strength parameters aredetermined by solving the appropriate boundary condition equations. Once these are known, the velocity andpotential fields are computed. The pressure field can then be calculated from an appropriate pressure–velocityrelationship, and forces and moments calculated by pressure integration.15–17 In order to improve the accuracy of theinduced drag calculation, the version A502i can calculate it by Trefftz plane analysis.14,18

The PAN AIR pilot code and later versions have been used to compute aerodynamic forces and moments, andpressure distribution on arbitrary configurations.16,17 A complete discussion of the method may be found in Ref. 19.

For the symmetric flight conditions considered in this paper, the half wing and wing-body configurationsrepresented consisted of 1551 and 2388 panels, respectively. The wing-body-tail geometries with and without wingtwist have 4204 panels. The wing-body-tail geometries with and without wing twist were divided in twenty-seven

Figure 1. ANCE representation created with CADtools.

American Institute of Aeronautics and Astronautics092407

3

surface networks. Twenty-two are defined as indirect condition on impermeable thick surface (for lifting surfaces).Four are defined as direct condition on impermeable thick surface (for non lifting surface), and one as base surfacecondition. The landing gear and the camera were not included in the paneled geometry, due to the fact that thecontribution of these components to inviscid forces and moments is assumed negligible. There are seventeen wakesurface networks to perform the Kutta condition (zero vorticity at trailing-edges and body bases).18 The wing-bodygeometry was divided in eight surface networks and seven wake surface networks. Five surface networks aredefined as indirect condition on impermeable thick surface; two are defined as direct condition on impermeablethick surface and one as base surface condition. The wing only geometry has five surface networks defined asindirect condition on impermeable thick surface and four wake surface networks.

Figures 2-5 show the paneled geometry of the wing, wing-body and wing-body-tail with and without wing twistused in the analysis, respectively. Lift and induced drag coefficients were obtained at 0.15 Mach number, at differentangles of attack for the paneled ANCE model.

B. Viscous DragThe classic technique presented in Ref. 11,20 was applied to predict the viscous drag coefficient of the complete

airplane geometry. A combination of analytical and empirical data was used to calculate the drag contributions dueto skin friction, component interference, flow separation, and surface imperfections.

The analysis included the skin friction of wing, tail, fuselage, and booms; empirical data of drag contribution oflanding gear and camera; and interference drag between the wing and fuselage, between tail surfaces, and betweenthe main and nose gears and the fuselage. For a complete review of the viscous drag breakdown, see Ref. 4.

C. Maximum Lift Coefficient PredictionThe classic Valarezo and Chin method or Pressure difference rule was applied to estimate the maximum lift

coefficient of the complete airplane.8 The pressure difference rule is a stripwise analysis of an inviscid flowfield todetermine an estimate of the maximum lift coefficient of the entire wing. It assumes that all of the viscous effects aregenerally local two-dimensional natures. It is based on the examination of wind tunnel data which indicates that, at agiven flow condition (Reynolds number and Mach number combination), there is a certain pressure differencebetween the suction peak of the airfoil and its trailing-edge at the maximum lift condition. Thus, at a given flowcondition, there is a pressure difference that indicates when maximum lift is attained.21 Although this method is

Figure 4. Wing–body–tail configuration.Figure 2. Wing configuration.

Figure 5. Wing–body–tail with wing twistconfiguration.

Figure 3. Wing–body configuration.

American Institute of Aeronautics and Astronautics092407

4

based on two-dimensional data, it may be used for lifting three–dimensional configuration and it has been validatedfor those cases.8,22

In this paper, the pressure difference at the maximum lift coefficient was obtained by a two-dimensional panelmethod coupled with a boundary-layer XFOIL23 version 6.94. This is an open-source program created for the designand analysis of isolated airfoils. The code uses a simple linear-vorticity stream function panel method for theinviscid formulation, and it incorporates a Karman–Tsien compressibility correction. The boundary-layer freetransition occurs when an en criterion is achieved.23

The NACA 4415 airfoil was tested at a Reynolds number of cruise flight (1.413×106) at angles of attack from-12 to 22 deg in steady, incompressible, and viscous flow with free transition criteria at n = 9 and 140 panels aroundthe section. No roughness effects were considered on the surface. The lift curve was obtained and the ∆Cp atmaximum lift coefficient was extracted.

III. Results and Discussion

A. The Complete AirplaneTo obtain the complete airplane lift and drag coefficients presented in this work, the flowfield was divided in two

regions. The induced drag and lift coefficients were computed using the panel code PAN AIR. The minimum dragmay be assumed equal to the total viscous drag, and it was estimated by viscous drag build-up. The total drag isobtained when the viscous drag is added to the induced drag. All the data obtained were achieved assuming steady,incompressible, and subsonic flow at wing Reynolds numbers equal to 1.413×106 and Mach 0.15.1,2

The experimental data presented in Ref. 3,4 was used to validate the numerical data achieved in this paper. Thesedata were obtained in a subsonic, closed–throat, closed–circuit, and unpressurized wind tunnel at nine differentReynolds numbers using a scale model of the ANCE with no wing twist. Buoyancy, blockage, and tare andinterferences corrections were applied to adapt the wind-tunnel data to the flight condition.4 The scale–effectcorrections on lift and drag were made via the Jacobs method24 and the extrapolation method,25 respectively, toadapt the wind-tunnel data to the cruise Reynolds number.

Figures 6–8 and Table 1 show comparisons between the current aerodynamic analysis results and thoseestimated from wind-tunnel tests. It is observed that the aerodynamic analysis agrees fairly well with the wind-tunnel data excluding the stall.

The Oswald efficiency factor (or the induced drag factor) and the lift curve slope estimated are in excellentagreement with the experimental data. The experimental minimum drag coefficient and the lift–drag ratio are ingood agreement with the values estimated by the aerodynamic analysis.

A difference of 6.42% is observed between the lift coefficient at zero angle of attack predicted by the panelmethod and that one achieved via the wind-tunnel test. It may be due viscous lift loss. The lift of a wing is less thanthe value computed on the basis of potential flow because of the presence of a boundary layer on the surface.7

The lift–drag ratio of the complete airplane with wing twist is greater than that one with no twist. The maximumlift coefficient of the airplane with twist is 3.55% lower than CLmax with no twist. The wing twist increments (L/D)max

in 1.51%, representing an improvement of the aerodynamic characteristics.The maximum lift coefficient estimated is quite different from the wind-tunnel data. The actual airfoil simulation

used turbulent modeling without roughness effects, and the wind-tunnel model had a standard roughness. This maybe the cause of the differences between both values. Roughness on the wing leading edge affects the stallcharacteristics of an aircraft. Ref. 21 shows a comparison of predicted and experimentally measured maximum liftcoefficient with and without roughness for cruise flight of M100 ONERA wing-body configuration. It is observedthat the maximum lift coefficient computed with roughness is less than the values estimated with no roughness, andthis agrees fairly well with experimental results.

B. Comparison between Computational ConfigurationsThe purpose of the four computational configurations was to determine the contribution and influence of each

mean component on the complete airplane. Figures 9-11 and Table 2 show the lift coefficient curve, the drag polarcurve, and the pitching moment curve for the four computational configurations. The wing configuration results arein a good relation with lift line theory estimation, where CLα, = 0.08677/deg and e = 0.935 for a wing withRA = 8.57.26 The wing configuration produces the highest lift curve slope, whereas the wing-body, the lowest slope.The WBT and WBTt induced similar lift curve slope. This is close to the WB lift curve slope.

American Institute of Aeronautics and Astronautics092407

5

Figure 8. Lift–drag ratio as a function of liftcoefficient for the ANCE in unpowered condition.

Figure 9 shows the value of lift coefficient at zeroangle of attack of the wing configuration is the highest.The WB CLo is larger than that one of WBT, and this ishigher than that one of the WBTt configuration. Figures12-13 show the spanwise lift distribution for allconfigurations at angles of attack equal to 0 and 14 deg,respectively. The spanwise lift distribution for the WBTtconfiguration is the smallest, and for the W configurationis the largest. The WBT configuration spanwise lift

distribution is smaller than that of the WB configuration. The twin booms and the fuselage produce a reduction ofwing lift in the WB, WBT and WBTt configurations. The wing twist in the WBTt configuration changes the spanwiselift distribution and reduces de lift for a specific angle of attack respect to the WBT configuration. The spanwise liftdistribution is analoguous to CLo for all configurations.

Table 1. Comparison of aerodynamic analysisresults and wind-tunnel data for the ANCE.

Aerodynamic analysisWindtunnel No twist Twist

k 0.0509 0.0508 0.0493e 0.7282 0.7296 0.752

(L/D)max 13.588 13.377 13.579CLα, 1/deg 0.0743 0.0742 0.074

CLo 0.4458 0.474 0.312CDp 0.0266 0.0275 0.0275

CLmax 1.4236 1.6965 1.636

Figure 6. Comparison of aerodynamic analysisresults for lift curve and wind-tunnel data for theANCE in unpowered condition.

Figure 7. Comparison of drag polar curve obtainedby aerodynamic analysis and experimental data forthe ANCE in unpowered condition.

American Institute of Aeronautics and Astronautics092407

6

Figure 10 shows the polar drag curve for allconfigurations. The wing configuration produces thehighest Oswald efficiency factor. The WB, WBT andWBTt configurations induced similar e. However, theWBTt configuration Oswald efficiency factor is largerthan that one of the WBT configuration. The minimumdrag coefficient of the W and WB configurations arezero, but the WBT and WBTt configurations minimumdrag coefficient is 0.0014. PAN AIR computes onlyinviscid forces, and the minimum drag present in theconfigurations with tail has to be induced drag. Whenthe wing does not produce lift, the angle of attack of thetail is -2.37 deg, and it produces lift and vortex drag.Figures 14-15 show a comparison between the W andWB configurations off-body pressure distribution Trefftzplane and the WBT and WBTt configurations at angle ofattack equal to 0 and -4 deg, respectively. The W andWB configurations produced only one vortex at angle ofattack equal to zero, and the WBT and WBTt producetwo vortexes, one behind the wing tip, and another onebehind the tail assembly tip. The WBTt configurationwing vortex is smaller than that one of the WBTconfiguration. As it is shown in Fig. 15, at an angle ofattack equal to -4 deg, it could be observed that thevortex created for the tail assembly in WBT and WBTt isgreater than that one generated for the wing. The tailassembly is an important source of induced drag.

Figure 11 shows pitching moment coefficient as afunction of angle of attack for all configurations. It isobserved that pitching moment slopes for allconfigurations are negative, and that for theconfigurations with tail the CMα absolute value is higherthan that with no tail.

The pressure distributions at four wing stations were computed to study the effect of twin boom, fuselage andwing twist on the lift generation. Figures 16–20 show the pressure distribution for all configurations at y equal to

Figure 10. Lift coefficient as a function of induceddrag coefficient for all configurations.

Figure 11. Pitching moment coefficient as afunction of angle of attack for all configurations.

Figure 9. Lift coefficient as a function of angle ofattack for all configurations.

Table 2. PAN AIR results for all configurations.W WB WBT WBTt

CL,α, 1/deg 0.0874 0.0733 0.0742 0.074CLo 0.6159 0.5146 0.4744 0.3124e 0.9266 0.7534 0.7296 0.7518

CMα, 1/deg -0.0038 -0.0023 -0.0806 -0.077CMo -0.1254 -0.1085 -0.2246 -0.1292

American Institute of Aeronautics and Astronautics092407

7

0.73, 0.815, 1.56 and 2.08 m, respectively, for angles of attack 0 and 14 deg. The boom is located between station0.73 and 0.815 m in the WBT and WBTt configurations, and the twist starts at station 0.815 m. It is observed that fory equal to 0.73 and 0.815 m, the pressure distributions for the W and WB configurations are nearly coincident, andfor the WBT and WBTt configurations, the pressure distributions for the W and WB configurations are nearlycoincident, and for the WBT and WBTt configurations, the pressure distributions agree with each other, and differentfrom those for the W and WB configurations. The WBTt configuration pressure distribution does not match with W,WB and WBT configuration pressure distributions.

IV. ConclusionsAn aerodynamic analysis procedure was applied to compute and estimate the aerodynamic characteristics of the

ANCE. The high order panel code PAN AIR was used to compute the inviscid flowfield, a viscous drag was built toestimate the viscous drag, and the pressure difference rule was applied to calculate the maximum lift coefficient ofthe ANCE. PAN AIR was used to simulate the potential flowfield around four configurations: wing, wing–body,wing–body–tail and wing–body–tail with wing twist. The aerodynamic analysis results for the complete airplanewith no wing twist are in good agreement with experimental data. The wing twist improves the aerodynamiccharacteristics of the aircraft. It increases the lift–drag ratio 1.51 %.

Figure 12. Spanload for α = 0 deg.

Figure 14. Pressure coefficient distribution on theTrefftz Plane (xTrefftz = 5 m) for all configurations atangle of attack equal to 0 deg.

Figure 13. Spanload for α = 14 deg.

Figure 15. Pressure coefficient distribution on theTrefftz Plane (xTrefftz = 5 m) for all configurations atangle of attack equal to -4 deg.

American Institute of Aeronautics and Astronautics092407

8

Figure 16. Pressure distribution for allconfigurations at y = 0.73 m, α = a) 0 and b) 14 deg.

Figure 17. Pressure distribution for allconfigurations at y = 0.815 m, α = a) 0 and b) 14 deg.

Figure 18. Pressure distribution for allconfigurations at y = 1.56 m, α = a) 0 and b) 14 deg.

Figure 19. Pressure distribution for allconfigurations at y = 2.08 m, α = a) 0 and b) 14 deg.

American Institute of Aeronautics and Astronautics092407

9

The comparison of the four configurations demonstrates that the boom and the fuselage reduce the lift andreduce the Oswald efficiency factor. It represents a downgrade of the aerodynamic characteristics. The tail assemblyis an important source of induced drag, but it is necessary to keep the longitudinal stability.

AcknowledgmentsThe authors wish to acknowledge the financial support of the Direction of Investigation, Universidad Simón

Bolívar, Sede del Litoral, and FUNDACITE Aragua, Maracay, both in Venezuela.

References1Boschetti, P., and Cárdenas, E., “Diseño de un Avión No Tripulado de Conservación Ecológica”, B.S. Thesis, Department of

Aeronautical Engineer, Universidad Nacional Experimental de la Fuerza Armada, Maracay, Venezuela, 2003.2Cárdenas, E., Boschetti, P., Amerio, A., and Velásquez, C., “Design of an Unmanned Aerial Vehicle for Ecological

Conservation,” AIAA Paper 2005-7056, Sep. 2005.3Boschetti, P. J., Cárdenas, E. M., and Amerio, A., “Drag Clean-Up Process of Unmanned Airplane for Ecological

Conservation,” Aerotecnica Missile e Spazio, Vol. 85, No. 2, 2006, pp.53–62.4Boschetti, P., “Reducción de Resistencia Aerodinámica en el Avión No Tripulada de Conservación Ecológica,” M.S. Thesis,

Coordination of Mechanical and Civil Engineer, Universidad Simón Bolívar, Caracas, Venezuela, 2006.5Erickson, L. L, “Panel Methods – An Introduction,” NASA TP-2995. 1990.6McLean, J. D., and Randall, J. L., “Computer Program to Calculate Three-Dimensional Boundary Layer Flows Over Wings

With Wall Mass Transfer,” NASA CR-3123, 1979.7Crabbe, R. S. “Simple Numerical Method to Compute Viscous Lift Loss of Wings,” Journal of Aircraft, Vol. 35, No. 1,

1998, pp. 27–32.8Valarezo, W. O., and Chin, V. D., “Method for Prediction of Wing Maximum Lift,” Journal of Aircraft, Vol. 31, No. 1,

1994, pp. 103–109.9Dam (van), C.P., Vander Kam, J. C., and Paris, J. K., “Design-Oriented High-Lift Methodology for General Aviation and

Civil Transport Aircraft,” Journal of Aircraft, Vol. 38, No. 6, 2001, pp. 1076–1084.10Tronger, L., and Selvy, G., “Computation of the Aerodynamic Characteristics of a Subsonic Transport,” Journal of Aircraft,

Vol. 35, No. 2, 1998, pp.183–190.11Hoerner, S. F., Fluid – Dynamic Drag, published by the Author, Bakersfield, CA. 1965, Chaps. VI, VIII, XIII, XIV.12Boschetti, P. J., Cárdenas, E. M., and Amerio, A., “Increasing the Lift–Drag Ratio of an Unmanned Aerial Vehicle Using

Local Twist,” Journal of Aircraft, Vol. 45, No. 1, 2008, pp.10–15. DOI: 10.2514/1.33353.13Boschetti, P. J., Amerio, A., and Cárdenas, E. M., “Aerodynamic Performance as a Function of Local Twist in an

Unmanned Airplane,” AIAA Paper 2009-1481, Jan. 2009.14Saaris, G. R., “A501i User’s Manual – PAN AIR Technology Program for Solving Problems of Potential Flow about

Arbitrary Configurations:,” The Boeing Company, Report D6 – 54703, Seattle, Washington, Feb.1992.15Derbyshire T. y Sidwell K. W. “PAN AIR Summary Document (Version 1.0),” NASA CR-3250, Apr. 198216Tinoco, E. N., Johnson, F. T., and Freeman, L. M. “Application of a Higher Order Panel Method to Realistic Supersonic

Configurations,” Journal of Aircraft, Vol. 17, No. 1, 1980, pp. 38–44.17Cenko, A., Tinoco, E. N., Dyer, R.D., and DeJongh, J. “PAN AIR Application to Weapons and Separation,” Journal of

Aircraft, Vol. 18, No. 2, 1981, pp. 128–134.18Katz, J., and Plotkin, A., Low-Speed Aerodynamics, 2nd ed., Cambridge University Press, New York City, NY, 2001, pp.

201–204, 210–213, 358.19Epton, M. A. and Magnus, A. E. “PAN AIR- A Computer Program for Predicting Subsonic or Supersonic Linear Potential

Flows About Arbitrary Configurations Using a Higher Order Panel Method Volume I – Theory document (Version 3.0),” NASACR-3251. 1990.

20Torenbeek, E., “Prediction of the airplane polar at subcritical speeds in the route configuration,” Synthesis of SubsonicAirplane Design, Deflt University Press, Rotterdam, Nederland, 1976, pp. 487–524.

21Cebeci, T., Shao, J. P., Kafyeke, F., and Laurendeau, E., Computational Fluid Dynamics for Engineers, Horizonspublishing and Springer, Long Beach, CA, 2005, pp. 10–19.

22Syms, G. F., “Low Order Method for Predicting Aerodynamic Performance Degradation Due to Ground Icing,” Journal ofAircraft, Vol. 39, No. 1, 2002, pp. 59–64.

23Drela, M., and Youngren, H., “XFOIL 6.9 User Primer,” Massachusetts Institute of Technology, Cambridge, MA, 2001.24Jacobs, E. N., and Sherman, A., “Airfoil Section Characteristics as Affected by Variations of the Reynolds Number,”

NACA R-586, Jun. 1936.25Rae, W. H., and Pope, A., Low-Speed Wind Tunnel Testing, 2nd ed, Wiley Interscience Publication, New York City, NY,

1984, pp. 457–464.26Thwaites, B., Incompressible Aerodynamics, Dover Publication, Inc., New York City, NY, 1960, pp. 335-337.