Parametric Design, Aerodynamic Analysis and Parametric

Optimization of a Solar UAV

Nuno Antonio Silvanuno.antonio.silva@ist.utl.pt

Instituto Superior Tecnico, Lisboa, Portugal

May 2014

Abstract

The aim of this thesis was to develop and implement a computational process to enable theswift design of different UAV configurations and their aerodynamic analysis. To this end, a CADtool using scripts was adopted to define the UAV external shape which was later imported into aCFD tool to generate suitable meshes. The test case consisted of a Long Endurance Electric UAV(LEEUAV), that was aerodynamically analyzed and parametrically optimized. While performing theaerodynamic analyses, turbulence models Spalart-Allmaras and k − ω SST, the later used in tandemwith the γ − Reθ transition model, were employed and their predictions compared with experimentaldata. Only the k − ω SST turbulence model and the γ − Reθ transition model were employed in thedetailed aerodynamic simulations. During cruise, the baseline LEEUAV presents a lift-to-drag ratioof 14.01, stall speed of 6.21 m/s and maximum cruise speed of 29.3 m/s. To enhance the baselinecruise performance, several parametric optimization and sensitivity studies were performed where itsnose, wing and fuselage shapes were modified. With the nose shape modification proposal, the adversepressure gradients that previously existed in that surface were reduced. With rounded wingtips, thewing long laminar separation bubbles were predicted to decrease. With the proposed wing root fairing,a region of separated flow that formed beginning at 2o disappeared thus reducing the aircraft drag.With the proposed wing washout, aileron control effectiveness was extended to angles of attack up to10o.Keywords: Solar UAV, UAV Airframe, Long Endurance, Computational Fluid Dynamics, ParametricModeling, Sensitivity Analysis.

1. Introduction

This thesis aims to make use of the current com-puter capabilities of both commercial and opensource software packages by developing a designtool that enables the swift design of complex Un-manned Aerial Vehicles (UAVs) geometries throughthe use of Computed-Aided Design (CAD) tools butalso provides rapid and reliable aerodynamic datato validate specific UAV configurations given keymission objectives. This tool was developed withthe objective of being in between the conceptualand detailed UAV design phases so that after de-signing a specific UAV configuration the engineerwould send that model into a Computational FluidDynamics (CFD) software and collect specific aero-dynamic data.

This thesis takes part in a project proposal by thename of Long Endurance Electric UAV (LEEUAV)whose goals are the development of a low cost, smallfootprint, easy to build and maintain multirole elec-tric UAV, capable of being deployed from short air-fields and highly flexible so that it is able to adapt to

different civilian surveillance missions, thus not be-ing restricted to specific mission profiles. Althoughsuch a project covers numerous areas of expertise,this thesis focuses on completing a cycle that beginswith the definition of a UAV geometric parameters,building that UAV geometry using custom madescripts using a CAD program, sending that geom-etry to a CFD software where high-fidelity aerody-namic analyses are performed and, finally, collectingand analyzing that aerodynamic data.

In this thesis, the LEEUAV will be the object ofthe geometric modeling, aerodynamic analyses andparametric optimization.

2. Parametric Design

FreeCAD [1] was the selected CAD tool for theLEEUAV design. This program proved to be veryefficient in the parametric modeling of commonUAV geometries. This software is customizable andallows the parametric modeling of complex three-dimensional geometries. It is open source and mostof its functions are accessible using Python scripts

1

and if not, the user can develop new ones. It canalso read and write open file formats such as STEP,IGES and STL which can then be used by the CFDsoftware.

The design scripts were developed in such a waythat the geometric design of a general UAV con-figuration can be divided into five main parts: theloading of the FreeCAD different modules requiredto run its native functions and the definition ofspecifically developed functions, the statement ofa UAV geometry parameters required for the com-plete definition of its airframe components, the testof those parameters to ensure their relative dimen-sions are valid, the building and assembly of thedifferent components and the export of the gener-ated mesh and object files later used by the CFDsoftware. The design scripts divide the geometry infour different but dependent geometries, namely itswing, fuselage, tail and nose. After building eachcomponent, they are assembled into a single entityso that the subtraction between the UAV and thecomputational control volume used in the aerody-namic analyses is easily performed.

The LEEUAV building process starts with theleft half wing, defined by four sections. At eachsection one airfoil is specified, together with five ad-ditional variables which are the airfoil chord, wingsweep, twist and dihedral angles, and each sectionspanwise location. The airfoil used in all sections,was especially developed for this project using an in-house gradient based shape optimization tool cou-pled to the aerodynamic analysis tool XFOIL [2].After building and assembling the wing, the fuse-lage is built section by section, until the final ge-ometry is fused together into a single entity. Thefuselage was built using a total of ten different cross-sections placed along its longitudinal axis to guar-antee a smooth geometric transition between them.The LEEUAV tail has a characteristic inverted cru-ciform shape. Much of the tail building process isidentical to that of the wing. The tail has a totalof five sections which are defined by the locationof different airfoils and their angles. The real nosegeometry is comprised of a spinner, propeller andelectric motor. However, the flow going around suchgeometries would probably cause unsteady phenom-ena that would require the use of unsteady schemesin the aerodynamic simulations. To avoid that, asimple ellipsoid geometry was produced to emulatethe nose geometry. The control volumes employedin the aerodynamic simulations and used to refinespecific areas of the aircraft meshes were developedusing the design script. A total of five parallelepi-pedic control volumes were defined: two control vol-umes that cover the entire wing, another controlvolume that englobes the aircraft geometry com-prised between the nose and half the tail boom, a

fourth control volume that englobes the entire air-craft and a fifth that serves as the aerodynamic sim-ulation control volume.

After building every LEEUAV component andfusing them all together, these are exported in threedifferent file formats: STEP, IGES and STL files.All file types are easily imported into the CFD soft-ware.

The LEEUAV dimensions and views are providedin Table 1 and Figure 1.

Table 1: LEEUAV dimensions. [3]

Wingspan 4.5 mLength 2.37 m

Wing root chord 0.350 mWing tip chord 0.25 m

Wing planform area 1.5 m2

Wing aspect ratio 13.5Tailplan chord 0.213 mTailplan span 0.85 m

Tailplan planform area 0.181 m2

Empennage root chord 0.258 mEmpennage tip chord 0.364 m

Empennage planform area 0.077 m2

Figure 1: LEEUAV views. [3]

3. Turbulence and Transition Models Valida-tion

When designing a specific airplane configuration, anengineer might use the classical theories of aerody-namics in a pre-design phase to get a grasp of theforces, moments and general flow physics the air-plane will experience. However, in most cases, CFDsimulations and possibly wind-tunnel tests for val-idating and experimenting with new configurationswill be performed at a later stage. In the presentwork, CD-adapco’s Star-CCM+ [4] was employed inall the in-depth aerodynamic analyses. This soft-ware does all the pre-processing, including mesh

2

generation, processing and post-processing. BesidesStar-CCM+, XFOIL was also employed in caseswhere no wind-tunnel data was available.

Before performing the aerodynamic simulationson the entire LEEUAV, several mesh and aerody-namic model studies were carried out on simplergeometries. Because there was no wind-tunnel datafor the LEEUAV airfoil, there was the need to findwell studied airfoils subject to flows at Reynoldsnumbers similar to those of the LEEUAV duringits cruise flight phase. Two airfoils, MA409 andCAL1215J, that fit this project objectives were se-lected. All data concerning these airfoils was ob-tained from the University of Illinois at Urbana-Champaign (UIUC) Aerodynamics Research Lab-oratory in its low-turbulence subsonic wind-tunnel[5]. Two turbulence and transition models were em-ployed in the validation tests: the Spalart-Allmaras(SA) [6] turbulence model, that was sporadicallyemployed with Turbulence Suppression model, andthe k−ω SST turbulence model [7] that was alwaysused with the γ −Reθ [8] [9] transition model.

The LEEUAV was designed to fly at Reynoldsnumbers regimes of about 1.6×105 and Mach num-bers of approximately 0.02, corresponding to airproperties at 1,000 m above sea level (ν = 1.5810−5

m2s−1, a=336.4 m/s), airplane mean chord of 0.35m and design cruise speed of 7.59 m/s.

Both airfoils were subjected to two Reynoldsflow regimes and these were Re1 = 2 × 105 andRe2 = 3 × 105. A comparison between the airfoilsimulations and experimental data proved that bothturbulence models at either Reynolds number canaccurately determine the lift coefficient. However,that is not the case when comparing the drag coef-ficient data. In the majority of the simulations, theSA turbulence model fails to provide acceptable re-sults, or at least, ones as good as the k − ω SSTturbulence model. The SA turbulence model dragpredictions exhibit decreasing discrepancies as theangle of attack increases. A consequence of usingthis turbulence model without a transition model isthat the flow is turbulent starting near the airfoilsleading edge. When testing an airfoil using a wind-tunnel, as the angle of attack increases so does thevelocity peak that also moves closer to the airfoilleading edge. As the velocity peak increases so doesthe adverse pressure gradient that follows and it isat that region that the flow under the right con-ditions becomes turbulent over the airfoil surface.This means that the turbulent drag contributionincreases as the angle of attack increases since nat-ural transition occurs closer to the airfoil’s leadingedge. This way, the increasing SA turbulence modeldrag prediction accuracy at higher angles of attackmight just be the result of the model’s turbulent ki-netic energy generation in areas where there should

not be any generation at all coupled with the airfoilupper surface transition location gradually movingin the direction of the leading edge. No transitionmodel is available in Star-CCM+ to use with theSA turbulence model. There is however the Tur-bulence Suppression. While using this model theobtained Cl and Cd values are much closer to theUIUC data. However, as good as it may be, it ishardly a transition model since it is up to the userto define where turbulence is to be suppressed.

In all simulations using the k−ω SST turbulencemodel together with the γ −Reθ transition model,the laminar flow that first develops at the airfoilssurfaces grows unstable until at a certain point ittransitions into a turbulent boundary layer. At lowReynolds numbers, and when the free-stream turbu-lence intensity is low, the flow starts as laminar andafter the velocity peak, when the adverse pressuregradient is strongest, the laminar boundary layerseparates. If the conditions are adequate, the flowthen experiences laminar-turbulent transition andthe boundary layer re-attaches forming a laminarseparation bubble (LSB).

4. LEEUAV Airfoil Aerodynamic Perfor-mance

Following the validation tests, a series of aerody-namic simulations using XFOIL and Star-CCM+were performed on the LEEUAV airfoil at the de-sign Reynolds numbers. In these studies, only thek−ω SST turbulence model used together with theγ − Reθ transition model were employed. UsingXFOIL, a comparison study of where natural tran-sition is predicted to occur is established. Figure 2compares the predicted natural transition locations.

Figure 2: LEEUAV airfoil natural transition curves.

The mechanism in which transition occurs is thesame as in the previous airfoil studies. Transitionoccurs gradually on the airfoil suction side. A lam-

3

inar separation bubble forms near the trailing edgeand as the angle of attack increases that bubbledecreases in length and moves closer to the lead-ing edge. This gradual transition does not occurat the airfoil pressure side where no laminar sepa-ration bubbles are detected until angles of attackbelow 0o when the flow becomes turbulent at theleading edge.

Figure 3 shows the intermittency factor and re-gions of reversed flow at an angle of attack of 5o.Close to the surface, in the viscous sublayer, theintermittency factor equals zero. There is a regionon the outer boundary layer where the intermit-tency is approximately 0.5 which is caused by thelaminar separation which occurs upstream of there-attachment point. This flow strip disappears inthe downstream direction as the turbulent bound-ary layer thickness increases.

Figure 3: LEEUAV airfoil intermittency factor andregions of reversed flow, AoA=5o.

The LEEUAV has a solar powered assistedpropulsion and as such its batteries may rechargeas it flies by using the energy surplus generated byits solar panels. These solar panels are placed ontop of its wing, which can be a source of problemsby adding discontinuities to its surface which canproduce effects similar to those of boundary layertrips. Two studies reproducing the effects of havingoffset surfaces on top of the LEEUAV airfoil wereperformed, simulating the effect of having solar pan-els on top of the wing. With the two-dimensionalresults one can infer what might happen at a three-dimensional level. These studies consisted of havingtwo panel configurations: in configuration 1 a panelis placed on top of the original airfoil, thus increas-ing its thickness and adding two steps to the airfoilupper surface; in configuration 2 a tilted panel isadded in order to have the panel trailing edge over-lapping with the airfoil surface while maintainingthe previous leading edge step. The solar panelsused in these studies have a thickness of 2 mm andlength of 256 mm.

Figure 4 shows the turbulent kinetic energy gen-eration and regions of reversed flow around solarpanel configuration 1 at an angle of attack of 5o.The simulations show that for both solar panel con-figurations the leading edge step promotes the for-mation of a laminar separation bubble in conse-

Figure 4: Solar panel configuration 1, turbulent ki-netic energy (J/kg) and regions of reversed flow,AoA=5o.

quence of the sharp edge the flow encounters in itsway, resulting in flow transition at that location.The flow behavior at the airfoils lower surfaces isthe same as in the original airfoil shape. In config-uration 1 the aft step promotes the formation andlocking of a recirculation bubble.

Figure 5: LEEUAV airfoil, Cl and Cm,c/4 versusAoA.

Figure 6: LEEUAV airfoil, Cl versus Cd.

4

Both solar panel configurations promote the gen-eration and fixation of a laminar separation bubbleat their leading edge step. The implications are theforced transition at that location with the inherentdrag penalties due to turbulence. Figures 5 and 6show the comparison between the lift, quarter-chordpoint moment and drag coefficients of the clean air-foil and both solar panel configurations. For thetested configurations, placing a solar panel on topof the airfoil produces enormous penalties in boththe lift and drag coefficients at any angle of attack.In terms of the moment coefficient, there is no sig-nificant difference between the three airfoil configu-rations. The solar panels that will eventually makepart of the LEEUAV ought to be installed insideits wing, or if not, there is the need to implementa smooth surface transition between the airfoil andsolar panels. The LEEUAV clean airfoil configura-tion shows good aerodynamic performance. Thisis a relatively thin airfoil and as such its Cl vsAoA curve is very characteristic. Looking at theextremes of the range of angles of attack in Fig-ure 5, and recalling the airfoil transition curves, thehigh angle of attack lift loss is somewhat gradual incomparison with the lower angle of attack abruptlift slope variation at -5o. The lift curve slope vari-ation is rather abrupt at low angles of attack giventhe airfoil leading edge shape and the laminar sepa-ration bubble that develops there and later bursts.The aerodynamic performance data for the cleanairfoil is listed in Table 2.

Table 2: LEEUAV airfoil aerodynamic performancetable.

Clmax 1.56αClmax 11o

α0lift -4o

Clα 6.15 rad−1

Cdmin 0.0129Cd,0 0.0207

Cl/Cdmax 61.03αCl/Cdmax 3o

C3/2l /Cdmax 66.8

αC

3/2l /Cdmax

5o

5. LEEUAV Aerodynamic PerformanceThe physical models and settings employed in thefull three-dimensional LEEUAV analyses are thesame as those used in the two-dimensional analy-ses. A mesh with approximately 7 million cells wasused in the three-dimensional aerodynamic simula-tions.

In the previous section it was understood thatthere are laminar separation bubbles present on the

airfoil surfaces and, depending on the angle of at-tack, those bubbles change their location, lengthand thickness. The flow behavior and its intrin-sic consequences found on the LEEUAV airfoil arepresent on its wing. The flow undergoes naturaltransition on the wing surfaces thus creating thewell known laminar separation bubbles. Althoughat low angles of attack the bubbles thickness arerelatively small, the wing effective shape does notremain the same which in turn affects the pressurefield in its vicinity. Throughout the simulations,and depending on the angle of attack, there arelong laminar separation bubbles located on the wingupper and lower surfaces. The wing lower surfaceLSB is only predicted to occur at negative anglesof attack and can be quite extensive. Its presenceis first detected at -4o and is located generally overthe wing chord latter half. Its width is not the samethroughout the wingspan. The bubble’s width nearthe wing root is approximately 30%c, and graduallyincreases until it remains approximately constantand equal to 60%c near the wing tip. Its thicknessis approximately one or two millimeters over thewingspan. As the angle of attack decreases to -6o,the bubble locates itself at the wing leading edgeand its width is approximately half the wing chord.At this angle of attack the bubble width can beconsidered approximately constant over the entirewingspan and its thickness ranges from one to fourmillimeters. At -8o, the LSB width decreases evenfurther and remains near the wing leading edge. At-10o the flow is no longer attached to the wing andthere are large regions of separated flow.

The wing upper surface presents laminar sepa-ration bubbles at every simulated angle of attackand every bubble occupies approximately 90% ofthe wingspan. No laminar separation bubbles aredetected on the wing area on top of the fuselageand close to the wingtip. Below -6o there is onlyone small LSB located at the wing trailing edge.At -6o, the LSB is uniformly distributed over thewing upper surface and its width is approximately30%c and is located at the trailing edge. As theangle of attack increases the bubble moves closer tothe leading edge which is consistent with the air-foil simulations. In the angle of attack range of -6o

to 0o the bubble width decreases to values of ap-proximately 15%c. Figure 7 shows the wing uppersurface bubble at 0o. In the angle of attack range of2o to 4o the bubble naturally keeps moving closerto the leading edge but its width now increases tovalues of approximately 45%c. From then on, asthe angle of attack increases even further, the bub-ble width gradually decreases while remaining closeto the leading edge. It is also in this angle of at-tack range, located at the wing root and startingat the wing leading edge, that a region of separated

5

flow starts to develop. As the angle of attack in-creases the flow region where the flow is separatedgrows even further. Figure 8 shows this region ofseparated flow at an angle of attack of 8o. After 8o,when the lift coefficient reaches its maximum value,the leading edge bubble bursts at several locationsalong the wingspan and consequently the flow is nolonger attached to the wing surface which in turnproduces massive lift loss. The regions where sepa-rated flow occurs start at the leading edge which, interms of lift loss, present what is typically known asleading edge loss. Figure 9 evidences the regions ofseparated flow at 10o. That figure also shows thatnear the wing surface where the ailerons will be lo-cated the flow is also separated. This means that athigh angles of attack, when the flow is separated inmost of the wing surface, there is no way of control-ling the LEEUAV roll actuating surfaces. Depend-ing on the angle of attack, the wing upper surfacebubble thickness presents different values. In gen-eral, away from the wing root region and while thebubbles have not burst yet, its thickness ranges be-tween one and six millimeters.

Figure 7: Spanwise upper laminar separation bub-ble, AoA=0o.

Figure 8: Wing upper surface flow separation,AoA=8o.

Figure 10 gives an idea of the transition processat an angle of attack of 0o by showing in red the iso-surface where reversed flow is detected and in bluethe isosurface of the turbulent kinetic energy whichgives an idea where the flow is expected to be tur-

Figure 9: Wing upper surface flow separation,AoA=10o.

Figure 10: Isosurfaces of reversed flow (red) andturbulent kinetic energy (blue), AoA=0o.

bulent. The flow first becomes turbulent after goingaround the LEEUAV nose where it encounters ad-verse pressure gradients. Afterwards, the turbulentboundary layer grows and propagates over the fuse-lage, boom and tail. This figure shows that the hor-izontal and vertical stabilizers inner surfaces are inthe turbulent wake region being shed from the fuse-lage. The transition model predicts bypass transi-tion in these surfaces as a consequence of the highlocal freestream turbulence intensity from the wake.Outside the wake, the local freestream turbulenceintensity is low and as a result the transition modelpredicts natural transition. These results are in linewith the obtained results from reference [10]. As itwould be expected, the flow becomes turbulent afterthe wing separation bubbles.

Figures 11 and 12 show two contour plots of theskin friction coefficient in the wing and tail sur-faces. Figure 11 shows an increase in the skin fric-tion coefficient after the laminar separation bubbleand Figure 12 shows an increase in the skin fric-tion coefficient on the inner tail surfaces. The skinfriction increment verified on the wing is expectedas a consequence of natural transition and the tailevidences bypass transition due to the fuselage tur-bulent wake. On the horizontal and vertical stabi-lizers, outside the turbulent wake, the skin frictioncoefficient gradually increases which leads to believe

6

natural transition is predicted in those surfaces.

Figure 11: Wing upper surface skin friction coeffi-cient contour plot with reversed flow isosurface inred, AoA=0o.

Figure 12: Boom and tail surface skin friction coef-ficient contour plot with reversed flow isosurface inred, AoA=0o.

Figure 13: LEEUAV CL and CM,c/4 versus AoA.

Figure 14: LEEUAV CL versus CD.

Figures 13 and 14 show the lift, quarter-chordmoment and drag coefficients at the simulated an-gles of attack. A total of eleven angles of attackwere simulated from -10o to 10o, with 2o incre-ments. The moment coefficient, obtained at thequarter-chord point, evidences that the LEEUAV,as it was designed, is naturally stable. At positiveangles of attack it produces a pitch-down momentwhereas at negative angles of attack its tendencyis the opposite. Table 3 summarizes the LEEUAVaerodynamic performance data. The LEEUAVcruise stall speed was determined to be 6.21 m/s,maximum cruise speed 29.3 m/s and maximum lift-to-drag ratio 14.01.

Table 3: LEEUAV aerodynamic performance table.

CLmax 1.51αCLmax 8o

α0lift -9.5o

CLα 5.59 rad−1

CDmin 0.0527CD,0 0.06

CL/CDmax 14.01αCL/CDmax 0o

C3/2L /CDmax 15.06

αC

3/2L /CDmax

2o

As a matter of curiosity, the Spalart-Allmarasturbulence model was employed in the LEEUAVaerodynamic simulations an its results comparedwith the k − ω SST model predictions. In general,the lift data obtained with the SA model is in ac-cordance with the k − ω SST model for angles ofattack smaller than 6o. Although at high angles ofattack the SA model lift predictions fail to detect

7

lift loss, at low angles of attack, its results are verysimilar to the ones obtained with the k − ω SSTmodel. As for the drag coefficient, like it was thecase when analyzing the airfoils, the SA turbulencemodel fails to produce aerodynamic data that is inline with what the k−ω SST model presents. Gen-erally the percentile discrepancy between the twomodels is in the range of 15% to 25%. Althoughthe SA model predictions fail to provide accuratelift data at high angles of attack and the computeddrag data is underpredicted when compared to thek − ω SST model, it provided swift simulations re-quiring only 36% of the time.

6. Parametric Optimization Studies

In the course of studying the flow around theLEEUAV at the different angles of attack, it was un-derstood that some geometric characteristics shouldbe subject to parametric studies and further aero-dynamic considerations. As such, a few shape re-designs were proposed.

6.1. Rounded Wing Tips

The wingtip geometry, as it was originally designed,is truncated rather than having a rounded surface.The purpose of designing a rounded wingtip was tocreate a smooth surface without discontinuities thatenabled the easier flow passage from the wing pres-sure side to its suction side [11]. Comparing the twoconfigurations, it is understood that the recircula-tion bubbles located at the wingtip truncated re-gion are now non-existent. Furthermore, with thenew geometry and at 0o, there is a reduction inthe upper LSB of approximately three centimeters.With the rounded wingtip, the wingtip vortex ef-fectively moves in the fuselage direction thus prov-ing the beneficial effects in reducing the wing LSB.In terms of lift and drag coefficients there are nosignificant gains although the wing drag coefficientdecreased very slightly.

6.2. Wing Root Fairing

The LEEUAV wing root region was not designedto present a particularly smooth transition betweenthe wing and fuselage surfaces. Not having asmooth geometric transition between those surfacesproved to generate undesired aerodynamic effects.One of those was the development, beginning at 2o,of a region of separated flow on the wing upper sur-face. As the angle of attack increased so did thatregion of separated flow. The wing root region iscritical due to the intersection of two bodies withtheir own pressure distribution fields and boundarylayers. If no geometric transition is employed, thedrag build-up could, in the right conditions, dra-matically increase in those areas. With the pro-posed wing root fairing, apart from the wing up-per surface laminar separation bubble, the regions

of reversed flow are now considerably smaller andtheir thickness is comparable to that of the LSB,if not smaller. Figure 15 shows the regions of re-versed flow at 4o for the proposed wing root fairingdesign. With the proposed design, the lift coeffi-cient is smaller although very similar to that of theoriginal configuration. Due to the mitigation of theregion of reversed flow in the wing root region, thereis a decrease in the drag coefficient.

Figure 15: LEEUAV with wing root fairing and iso-surfaces of reversed flow in the wing root region,AoA=4o.

6.3. Nose ShapeThe original nose design does not promote a smoothgeometric transition with the fuselage. The originalgeometry, with its small length and angular gen-eratrix produces significant adverse pressure gradi-ents that promote flow transition. A solution to theoriginal design was achieved by increasing the noselength by twenty millimeters and decreasing its gen-eratrix curvature so that the geometric transitionbetween the nose and fuselage became smoother.Figure 16 shows the nose pressure coefficient as afunction of position, at an angle of attack of 0o.The proposed nose design evidences lower pressuregradients and pressure peaks. The proposed designalmost does not present adverse pressure gradientsat the simulated angle of attack, thus proving to bean improvement to the original design.

6.4. Wing WashoutWing washout consists in reducing the lift distribu-tion on the wing surface in the wingspan direction.There are a number of ways to adjust the lift distri-bution over the wing and one of those is ensured byreducing the wing section incidence angle from thewing root to the wingtip. The consequence is that,at high angles of attack, the flow that is expectedto be detached in the wing root region is not at

8

Figure 16: LEEUAV nose Cp curves comparison,AoA=0o.

the wing tip region. This way, controlling the air-craft trough the ailerons is still possible. In the newwing design proposal, the incidence angle nearingthe wingtip is gradually reduced in 4o. ComparingFigures 9 and 17, at an angle of attack of 10o, theeffects of the wing washout proved to be very pos-itive with the wing region where the ailerons arelocated not evidencing boundary layer detachment.There was an expected decrease in the lift and dragcoefficients.

Figure 17: LEEUAV with wing washout isosurfacesof reversed flow, AoA=10.

7. ConclusionsThe objectives of this thesis consisted in the devel-opment of a CAD tool that enabled the swift para-metric design of generic UAV geometries and theiraerodynamic evaluation using high-fidelity models.Furthermore, one UAV in particular, the LEEUAV,would be built using the developed CAD tool andaerodynamically evaluated and parametrically op-timized.

Using FreeCAD as the CAD software package andthe developed scripts, it was possible to not onlydesign generic UAV geometries, but also to designa particular aircraft configuration, the LEEUAV.

Considering the aerodynamic simulations, priorto using specific physics, meshing and turbulencemodels, first there was the need to perform val-idation tests using reliable sources. During thevalidation phase two turbulence models were em-ployed in the aerodynamic simulations where itwas determined that only turbulence model k − ωSST used together with transition model γ − Reθprovided valuable lift and drag data. During theLEEUAV airfoil aerodynamic simulations, two so-lar panel configurations were simulated in order tounderstand their effects when installed on the wingupper surface. From the performed simulations itwas understood that both configurations forced flowtransition at the panel’s leading edge which in turncontributed to lift reduction and drag increase. Theconclusion was that any solar panel installed on thewing must be allocated in such a way that no surfacediscontinuities are present. At a three-dimensionallevel, several aerodynamic simulations were per-formed having in consideration the LEEUAV cruiseflight phase. The selected mesh and physics modelsallowed to capture aerodynamic phenomena suchas the spanwise laminar separation bubbles alongthe wing, regions of separated flow beginning at thewing leading edge, the effects of the interaction be-tween the fuselage and wing boundary layers andthe flow bypass transition on the tail surfaces.

Turbulence model SA was also employed in thetwo- and three-dimensional aerodynamic analysesand the conclusions are that the lift coefficient isaccurately predicted as long as the angles of attackdo not increase to the point where flow separation ispresent in a significant way. The drag predictionsdiffer between the two turbulence models becauseno transition prediction model was employed withthe SA turbulence model. In terms of the requiredCPU time for the residuals to converge, the SAmodel is clearly superior. In conclusion, at low an-gles of attack and when one can manage to over- orunder-predict an aircraft drag coefficient and whentime is of the essence, the use of turbulence modelSA is recommended. However, if there is the needto better predict the drag coefficient and associatedphenomena, then turbulence model k−ω SST usedwith transition model γ−Reθ should be employed.

Some geometric improvements were proposed tothe baseline LEEUAV shape which consisted inmodification of the wing, fuselage and nose geome-tries. All design proposals proved to enhance theLEEUAV flight performance at some level and fur-ther studies regarding their adaptation are recom-mended.

9

In terms of future work, there are a few conceptsthat might be interesting to think about. Only thecruise flight phase was considered in the LEEUAVdesign and three-dimensional analyses. However,during the entire flight profile a number of otherflight conditions are encountered. It would be in-teresting to simulate the effects of crosswinds whichare a common occurrence during takeoffs and land-ings, to simulate the effects of coordinated turns orto simulate the effects of control surfaces like theailerons, rudder and elevators. Since the LEEUAVwill require a landing gear it is also recommended toperform studies regarding its impact in the globalaerodynamic performance. The LEEUAV will havea propeller in the nose region and its effects on theflow were not addressed. It would be interesting todo so since the propeller is transmitting energy tothe fluid which should delay natural transition inthe fuselage region.

Acknowledgements

The author would like to thank Professor AndreCalado Marta and Dr. Jose Manuel da Silva ChavesRibeiro Pereira for their patience, guidance andsupport throughout the development of this the-sis, and LASEF for the use of its installations andresources.

References

[1] FreeCAD. https://www.freecadweb.org/,2014.

[2] XFOIL. https://www.web.mit.edu/drela/public/web/xfoil/.

[3] A. C. Marta and P. Gamboa. Conceptual andPreliminary Design Summary Long EnduranceElectric UAV. 2013.

[4] CD adapco STAR-CCM+. https://www.cd-adapco.com/products/star-ccm/, 2014.

[5] Gregory A. Williamson, Bryan D. McGrana-han, Benjamin A. Broughton, Robert W. De-ters, John B. Brandt, and Michael S. Selig.Summary of Low-Speed Airfoil Data, volume 5.2012.

[6] P. R. Spalart and S. R. Allmaras. A One-Equation Turbulence Model for AerodynamicFlows. AIAA, 1992.

[7] F. R. Menter. Two-Equation Eddy-ViscosityTurbulence Models for Engineering Applica-tions. AIAA Journal, Volume 32(Number 8),August 1994.

[8] F.R. Menter, R. B. Langtry, S. R. Likki,Y. B. Suzen, P. G. Huang, and S. Vlker. A

Correlation-Based Transition Model Using Lo-cal Variables - Part I: Model Formulation.Journal of Turbumachinery, 128, July 2006.

[9] R. B. Langtry, F.R. Menter, S. R. Likki,Y. B. Suzen, P. G. Huang, and S. Volker. ACorrelation-Based Transition Model Using Lo-cal Variables - Part II: Test Cases and Indus-trial Applications. Journal of Turbumachinery,128, July 2006.

[10] R. B. Langtry and F. R. Menter. Transi-tion Modeling for General CFD Applicationsin Aeronautics. AIAA, 2005.

[11] Sighard F. Hoerner and Henry V. Borst.Fluid-Dynamic Lift Practical Information onAerodynamic and Hydrodynamic Lift. Mrs.Liselotte A. Hoerner, 1985.

10