Vasco Coelho - Master

AIR FORCE ACADEMY
Aerodynamic Detailed Design of an Unmanned Aerial

Vehicle with VTOL Capabilities
Vasco Luís Martins Ferreira Coelho

ALF/ENGAER 139424-L
Thesis to obtain the Master of Science Degree in
Military and Aeronautical Sciences - Aeronautical

Engineering
Examination Committee
Chairperson: BGEN/EngEl 119923-E Rui Fernando da Costa Ferreira
Supervisor: MAJ/EngAer 131603-G Joao Vítor Aguiar Vieira Caetano
Dr. Frederico José Prata Rente Reis Afonso
Member of the Committee: Prof. Dr. Afzal Suleman
Sintra, December 2021

ii
”Research is what I’m doing when I don’t know what I’m doing”
Wernher von Braun
iii
iv
Acknowledgments
First of all, I have to express my sincere gratitude to Dr. Frederico Afonso for joining this project and
for all the technical guidance, patience, and readiness shown, and to MAJ João Caetano for the critical
and constructive spirit transmitted which made me overcome the adversities encountered throughout
the period it took me to complete this dissertation.
I would like to recognize the remaining coordinators of the project, Dr. André Marta and MAJ Luı́s
Félix for the contributions and inputs given at the early design stages which guided the project in the
right direction. In this area, I also thank MAJ Tiago Oliveira and SAJ Paulo Mendes for the help given
when required.
To my comrades and friends Bernardo Alves, Pedro Silva and Paulo Sá, six years ago we started
this journey together and we now end it together. The course and this final project could not be done
without the friendship, mutual help and cooperation shown by all of you. The effort and dedication put
into the conceptual phase of this project greatly eased my work in this dissertation, and for that, I can
not thank you enough. Furthermore, I have to mention my second family - Distintos - for the fabulous
experiences we have had during the past years.
Finally, I would like to dedicate this work to my parents, brother and Marta. Their understanding and
support during my academic journey helped me to complete this milestone of my life.
v
vi
Resumo
Esta tese está integrada num projeto de desenvolvimento de um veı́culo aéreo não tripulado ca-
paz de efetuar descolagem e aterragem vertical, e tendo hidrogénio como principal fonte de energia
utilizando para tal uma célula de combustı́vel. A dissertação foca-se nas fases de desenvolvimento
preliminar e detalhada no que diz respeito a estudos aerodinâmicos e desempenho em voo.
A fase preliminar abrange a conceção da asa e da cauda, recorrendo ao software XFLR5, em
conjunto com uma estimativa da resistência aerodinâmica total da aeronave, recorrendo a expressões
semi-empı́ricas.
Para a análise detalhada, foi utilizado o software de mecânica de fluidos computacional Fluent. A
escolha do modelo de turbulência SST, em conjunto com o modelo de transição γ−Reθ , é validada pelas
simulação 2D do perfil SG6042, apresentando resultados consistentes com os dados experimentais.
A polar aerodinâmica da asa é obtida através da simulações 3D da mesma para vários ângulos de
ataque. Por forma a melhorar as propriedades aerodinâmicas da asa, foi aplicada torção à ponta da
asa, movendo a região inicial da perda da ponta da asa para a raiz. O impacto do sistema de propulsão
vertical na resistência aerodinâmica em voo cruzeiro é avaliado através da realização de testes em
túnel de vento e simulações em Fluent. Simulações de toda a aeronave concluem que, dependendo do
alinhamento dos rotores, a resistência aerodinâmica da aeronave varia entre 16.32 e 19.22 N para voo
cruzeiro, resultando num tempo total de voo entre 3H05 e 3H25.
Palavras-chave: Aerodinâmica, Célula de Combustı́vel, Descolagem e Aterragem Vertical,

Hidrogénio, Mecânica de Fluidos Computacional, Veı́culo Aéreo Não Tripulado.
vii
viii
Abstract
This thesis is part of a project to design an unmanned aerial vehicle capable of performing vertical
take-off and landing, and having hydrogen as its main energy source by using a fuel cell. The present
dissertation is focused on the preliminary and detailed design phases regarding aerodynamics and flight
performance studies.
The preliminary phase encompasses the wing and tail design, with the aid of XFLR5, together with
an estimate of the total aircraft drag by resorting to semi-empirical expressions. A longitudinal static
stability analysis is conducted, and the unmanned aerial vehicle characteristics are presented after the
preliminary phase of the project.
For the detailed analysis, Fluent was chosen as the computational fluid dynamics software to be
used. 2D simulation over the SG6042 wing airfoil validated the choice of the SST turbulence model,
coupled with the γ − Reθ transition model, as the results were consistent with experimental data. The
drag polar of the wing is obtained by simulating the 3D wing at various angles of attack. To enhance the
wing aerodynamic properties, twist was given to the wingtip, moving the stall region from the wingtip to
the root. The impact of the vertical propulsion system on the drag at cruise is assessed by performing
wind tunnel tests and simulations on Fluent. Simulations of the entire aircraft conclude that, depending
on the stopping position of the rotors, the drag of the aircraft varies between 16.32 and 19.22 N for
cruise, which results in a total flight time between 3H05 and 3H25.
Keywords: Aerodynamics, Computational Fluid Dynamics, Fuel Cell, Hydrogen, Unmanned

Aerial Vehicle, Vertical Take-Off and Landing.
ix
x
Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Resumo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
1 Introduction 1
1.1 Topic Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Initial Design Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.1 Objectives & Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Project Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Literature and Technology Review 7

2.1 Aerodynamic Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Boundary Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Design Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Airframe Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Wing Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 Tail Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Low-fidelity Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 XFOIL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.2 Vortex Lattice Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Computational Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.1 Physical Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.2 Domain Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.3 Discretization of Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
xi
3 Conceptual Design – Performance Considerations 27
3.1 Initial Drag Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Flight Speed Adjustments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4 Preliminary Design 31
4.1 Wing Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1.1 Wing Planform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.1.2 Airfoil Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2 Tail Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 Drag Build-Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3.1 Lift Producing Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3.2 Fuselage and other Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3.3 VTOL Propulsion System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.4 Preliminary Characteristics of the UAV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4.1 Static Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4.2 Performance Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5 CFD Validation 45
5.1 Experimental Data and XFOIL Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Fluent Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2.1 Physical Modeling and Grid Generation . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.2 2D Mesh Convergence Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.2.3 Turbulence Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 2D CFD Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6 Detailed Design – Aerodynamic Performance 55

6.1 Analysis of 3D Wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.1.1 Domain Characterization and Mesh Studies . . . . . . . . . . . . . . . . . . . . . . 55
6.1.2 Wing Aerodynamic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.1.3 Parametric Studies to Enhance Wing Performance . . . . . . . . . . . . . . . . . . 60
6.2 VTOL System wind tunnel Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.3 Complete Aircraft Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.3.1 VTOL System Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7 Conclusions 75
7.1 Achievements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Bibliography 79
A Drag from Wind Tunnel testing of the VTOL Propulsion System 85
xii
List of Tables
1.1 Design proposal presented by CIAFA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1 Database of UAV with similar designs as the one projected. Adapted from Coelho (2019). 28
4.1 Aerodynamic related inputs from conceptual design. Adapted from Alves et al. (2021). . . 31
4.2 First iteration wing dimensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 2D Airfoils results from XFOIL. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 3D Wing results from XFLR5, for each airfoil. . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 wind tunnel data obtained by Mendes. Adapted from Mendes (2021). . . . . . . . . . . . 39
4.6 UAV general characteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.1 2D 91k mesh parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.2 Comparison between experimental, XFOIL and Fluent results for AoA of 0 and 4 degrees. 49
5.3 Comparison between experimental, XFOIL and Fluent results for AoA of 12 degrees. . . . 50
6.1 Properties of the meshes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

6.2 Wind tunnel testing conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Inflation layer properties for each component. . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.4 Properties of the UAV meshes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.5 Lift and drag coefficients from Fluent for the UAV plus wind tunnel data for the VTOL system. 69
6.6 Lift and drag coefficients regarding the full aircraft simulated in Fluent. . . . . . . . . . . . 70
6.7 Aerodynamic properties of the UAV. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.8 Performance update of the UAV. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
A.1 Drag mean values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

A.2 Standard deviation values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
A.3 Lift mean values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
A.4 Lift standard deviation values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
xiii
xiv
List of Figures
1.1 Market study of possible hybrid UAV configurations. . . . . . . . . . . . . . . . . . . . . . 2

1.2 Ion Tiger UAV. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Laminar separation bubble. Retrieved from Mueller and DeLaurier (2003). . . . . . . . . . 9
2.2 Tail configurations for Double Boom Displacement. Adapted from Gundlach (2012). . . . 13
2.3 Vortex lattice method implementations. Retrieved from Peerlings (2018). . . . . . . . . . . 16
2.4 Structure of a CFD simulation system. Adapted from Hirsch (2009); Moukalled, Mangani,
and Darwish (2015). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Different topologies for structured grids. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 Different topologies for unstructured grids. . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7 Velocity profile on a turbulent boundary layer. Adapted from Brederode (2018). . . . . . . 24
2.8 Near-wall treatments. Retrieved from Ansys Inc. (2013). . . . . . . . . . . . . . . . . . . . 24
2.9 Control volume construction in 2D unstructured meshes. Retrieved from Versteeg and
Malalasekera (1995). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1 UAV mission profile. Adapted from Alves et al. (2021). . . . . . . . . . . . . . . . . . . . . 27

3.2 Power curve with important flight speeds. . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1 Wing Baseline Geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

4.2 SG6042 airfoil. Adapted from Selig (1997). . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Nose and main gear of similar MTOM aircraft in CIAFA. . . . . . . . . . . . . . . . . . . . 39
4.4 VTOL propulsion system components. Retrieved from T-Motor (2021). . . . . . . . . . . . 40
4.5 Preliminary CAD model of the UAV. Retrieved from Alves et al. (2021). . . . . . . . . . . . 43
5.1 2D Mesh with the detail on the inflation layer. . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.2 2D Mesh convergence study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Aerodynamic polars for the SG6042 airfoil. . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.4 Intermittency factor for an AoA of -4◦ . Black circles indicate the location of the LSB. . . . 52
◦
5.5 Turbulent kinetic energy for an AoA of -4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.6 XFOIL – Skin friction coefficient distribution for an AoA of -4◦ . . . . . . . . . . . . . . . . . 53
◦
5.7 Velocity contour plot for an AoA of 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.1 3D Wing mesh convergence study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
xv
6.2 Lift to drag ratio at various AoA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 3D Wing lift and drag coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.4 3D Wing lift and drag coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
◦
6.5 Reverse flow region (red) and high turbulent kinetic energy regions (blue) for an AoA of 14 . 60
6.6 Velocity contour plots at two different wing sections for an AoA of 14◦ . . . . . . . . . . . . 60
6.7 Streamlines over the upper surface of the wing at the beginning of the stall region. . . . . 61
6.8 Examples of possible wingtip configurations. Retrieved from Gudmundsson (2014). . . . . 62
6.9 Boom with VTOL system to be tested in the wind tunnel. . . . . . . . . . . . . . . . . . . . 63
6.10 Stopping positions considered for the front rotor. . . . . . . . . . . . . . . . . . . . . . . . 64
6.11 Mean values of the drag force obtained in each test, with the curve-fitting equation. . . . . 65
6.12 CAD model of the UAV and the domain considered for Fluent. . . . . . . . . . . . . . . . . 67
6.13 UAV Mesh convergence study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.14 Pressure distribution over the wing at an AoA of 0◦ . . . . . . . . . . . . . . . . . . . . . . . 71
6.15 Velocity contour plot at the symmetry plane z=0m . . . . . . . . . . . . . . . . . . . . . . . 72
6.16 Velocity contour plot at the booms z=0.72m . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.17 Aerodynamic graphs for the UAV with rotors parallel to booms. . . . . . . . . . . . . . . . 73
xvi
Abbreviations
AFA Academia da Força Aérea
AoA Angle of Attack
BL Boundary Layer
CAD Computer Aided Design
CAE Computational Aided Engineering
CFD Computational Fluid Dynamics
CG Center of Gravity
CIAFA Centro de Investigação da Academia da Força Aérea
DNS Direct Numerical Simulation
ESC Electronic Speed Controller
FAP Força Aérea Portuguesa
FDM Finite Difference Method
FEM Finite Element Method
FVM Finite Volume Method
ICE Internal Combustion Engine
LES Large Eddy Simulator
LLT Lifting Line Theory
LRN Low Reynolds Numbers
LSB Laminar Separation Bubble
MTOM Maximum Take-Off Mass
PITVANT Projeto de Investigação e Tecnologia em Veı́culos Aéreos Não-Tripulados
RANS Reynolds Average Navier-Stokes
xvii
RoC Rate of Climb
RPM Rotations per Minute
SA Spalart-Allmaras
SIMPLE Semi-Implicit Method for Pressure Linked Equations
SST Shear-Stress Transport
TE Trailing Edge
UAV Unmanned Aerial Vehicle
UIUC University of Illinois at Urbana-Champaign
VLM Vortex Lattice Method
VTOL Vertical Take-Off and Landing
xviii
Nomenclature
Greek symbols
α Angle of Attack.
β0.7r Rotor Incidence Angle (at 70% radius).
δ Maximum Boundary Layer Thickness.
δ∗ Boundary Layer Displacement Thickness.
Dissipation Rate of Turbulent Kinetic Energy.
η Efficiency.
Γ Dihedral Angle.
γ Intermittency.
λ Taper Ratio.
µ Molecular Viscosity Coefficient.
µt Eddy Viscosity.
νt Kinematic Eddy Viscosity.
ω Specific Dissipation Rate of Turbulent Energy.
φ Rotor Stopping Position.
ρ Density.
τw Wall Shear Stress.
ϕ0.25c Sweep Angle (at 25% Chord).
Roman symbols
AR Aspect Ratio.
C Volume Coefficient.
c Chord.
xix
D Drag Force.
e Oswald Efficiency Factor.
∗
EH2 Hydrogen Specific Energy Density.
eN Amplification ratio.
ef Finite Wing Efficiency Factor.
FF Form Factor.
K Induced Drag Factor.
k Turbulence Kinetic Energy
L Lift Force.
l Characteristic Length.
l/d Fineness Ratio.
mH2 cr /mH2 tot Hydrogen Mass Percentage Available for Cruise.
mH2 tot Total Hydrogen Mass.
Ma Mach Number.
Ne Number of Engines over the Wing.
P Power.
p Pressure.
PP P S Power Required by the Payload and Primary Systems.
Q Interference Factor.
q Dynamic Pressure.
R Ideal Gas Air Constant.
Re Reynolds Number.
Reθ Momentum-Thickness Reynolds Number.
S Surface Area.
Sµ Sutherland Temperature Constant.
Swet Wet Area.
T Temperature.
t Time.
xx
t/c Airfoil Thickness-to-Chord Ratio.
U Longitudinal Velocity.
uτ Wall Friction Velocity.
W Weight.
yB Boom Spanwise Position.
CD 3D Drag Coefficient.
Cd 2D Drag Coefficient.
Cf Skin-Friction Coefficient.
CL 3D Lift Coefficient.
Cl 2D Lift Coefficient.
Cp Pressure Coefficient.
CD0 Parasite Drag Component.
CDi Induced Drag Component.
Cli Airfoil Ideal Lift Coefficient.
CMα Pitching Moment Derivative with Respect to Angle of Attack.
Subscripts
∞ Free-stream Condition.
i, j, k Computational Indexes.
x, y, z Cartesian Components.
cr Cruise Conditions.
HT Relative to the Horizontal Tail.
max Maximum Conditions.
pr Relative to the Propeller.
ref Reference Condition.
T Relative to the Tail.
VT Relative to the Vertical Tail.
Superscripts
’ Fluctuating Component.
+ Dimensionless Unit.
xxi
xxii
Chapter 1
Introduction
1.1 Topic Overview
The idea of flying has pursued mankind since ancient times with the famous tale of Icarus from the
Greek ancient civilization. Although most known primordial flight projects and pioneers (e.g., Leonardo
da Vinci 1452-1519 and Montgolfier 1783) are related to manned piloted devices, unmanned vehicles
have had a parallel yet more discrete and delayed evolution over the years. Unpiloted aerial systems are
known by many nomenclatures and acronyms, like Unmanned Aerial System, Remotely Piloted Aircraft,
and Unmanned Aerial Vehicle (UAV), among others (Prisacariu, 2017). The latter will be the preferred
term throughout the presented thesis.
Early contributions to autonomous mechanics go back to the IV century BC with Archytas of Tarantas
creating a mechanical bird and China documenting the idea of vertical flight. However, only during the
XIX century, the first vertical flight machines were created using boilers and steam to provide power to
small embryonic propellers, with the credits attributed to Phillips in 1840 and d’Amécurt in 1860. The
first report of the usage of unmanned combat air vehicles dates to 1849, when the Austrians attacked
Venice with unmanned balloons loaded with timed bombs (Prisacariu, 2017). Ever since that time, UAV
development has increased and accelerated during military necessity events such as World War I and
II and the Cold War, with most superpowers now possessing a wide range of UAVs at their disposal to
conduct various missions such as surveillance, communication relay links, ship decoys, and detection
of biological, chemical, or nuclear materials (Mueller & DeLaurier, 2003). Nowadays, some missions
are being assigned solely to unmanned aerial platforms, formerly performed by combat piloted aircraft.
One example is the ”drone strike” performed by the United States in Afghanistan on August 28th, 2021,
which successfully completed its assigned mission without putting the lives of any US military personnel
at risk (BBC, 2021).
Similar to piloted aircraft, the demand for new and improved capabilities reached the UAV platforms,
including the ability to perform Vertical Take-Off and Landing (VTOL), a characteristic historically at-
tributed to pure rotary-wing aircraft. The ability to take-off vertically without a runway increases the
flexibility of the platform, and being able to hover and perform slow lateral and longitudinal movement
1
is preferred and sometimes a must for emergency response or precise transport missions, for example.
Airspeed constraints, especially in forward flight conditions, are known limitations of rotary-wing aircraft,
therefore hybrid platforms that include both rotary and fixed-wing aspects have been developed and are
desired. The fuel efficiency and propulsion system efficiency are the main weaknesses of these hybrid
platforms, not being quite as good as their traditional fixed or rotary-wing counterparts (Mrusek, 2021).
While there are several possible configurations for hybrid platforms, they can be divided into two
categories: the first one uses the same propulsion system in both vertical and horizontal flight by either
tilting the propulsion system – Tiltrotor (Figure. 1.1a), rotation of the wing – Tiltwing (Figure. 1.1b), or
even by rotating the entire aircraft – Tailsitter (Figure. 1.1c). The second category encompasses the
designs which have the vertical flight propulsion system segregated from the horizontal flight, commonly
named Lift+Cruise configurations or transitional aircraft (Figure. 1.1d). This latter configuration tackles
some of the drawbacks presented before since each propulsion system can be optimized for a single
flight condition and does not require complex tilt mechanisms, however, both propulsion systems add up
to an increase of the total aircraft weight (Serrano, 2018).
(a) Tiltrotor (Park et al., 2013) (b) Tiltwing (Defense Turkey, (c) Tailsitter (Li et al., 2018) (d) Lift + Cruise (UAVision,
2017) 2020)
Figure 1.1: Market study of possible hybrid UAV configurations.
An UAV with a similar configuration to the one depicted in Figure. 1.1d is the Top Engineering Falcon
5 UAV. It has the traditional gasoline Internal Combustion Engine (ICE) to provide power to the propeller
and has a maximum flight time of five hours. An interesting fact about this UAV is that the ICE can be
replaced by an electric motor and the aircraft becomes fully electric, requiring batteries instead of a fuel
tank. This modification increases the UAV weight and reduces the endurance to two hours, a reduction
in the total flight time of 60% (Group, 2014). Additionally, a fuel cell, fed by liquid hydrogen, can replace
the ICE on the same UAV, making it a greener alternative to fossil fuels. Nevertheless, this modification
also comes with a reduction in the total endurance to three hours in total (Group, 2014).
The use of hydrogen in aviation dates back to the golden age of space exploration and is now cited
as one of the contenders to replace fossil fuels in the aviation industry (Airbus, 2020b). One way of
using hydrogen as an energy source is with a fuel cell which converts oxygen and hydrogen into water
and electric energy. The energy in 1 kg of hydrogen gas at 350 bar (the usual pressure of the UAV’s
on-board cylinders) is the same as the energy in 2.8 kg of gasoline, making a fuel cell coupled with an
electric motor two to times more efficient than a gasoline ICE (Intelligent Energy, 2021; U.S. Department
of Energy, 2021). Moreover, fuel cells tend to be more efficient power sources when compared to
conventional batteries for steady flight conditions (Gadalla & Zafar, 2016). However, two main drawbacks
are pointed out to this new technology: there are no fuel cells with a specific power high enough for its
2
usage on larger scale projects (Ng & Datta, 2019), with studies from 2018 reporting a maximum power
density of 1.6 kW/kg, but with suppliers predicting this value to grow by a factor of five in the near future
(Intelligent Energy, 2021; Kadyk, Winnefeld, Hanke-Rauschenbach, & Krewer, 2018); the weight of the
fuel storage unit is substantial as it must safely contain high-pressure gases or liquids. For instance, the
storage tank of the Ion Tiger UAV – Figure. 1.2a – composes one-third of the total aircraft weight, and if
combined with the fuel cell system and the fuel itself, it adds up to 46% of the UAV weight (Figure. 1.2b)
(US Naval Research Laboratory, 2018).
(a) Ion Tiger prototype (US Naval Research Laboratory, (b) Mass breakdown of Ion Tiger. Adapted from US Naval Research
2018) Laboratory (2018)
Figure 1.2: Ion Tiger UAV.
While two examples were put forward in this chapter, their characteristics should not be blindly ex-
trapolated as general conclusions; each aircraft has its own properties and requirements, and extensive
studies into each specific case should be performed to understand which is the best solution.
1.2 Motivation
The Portuguese Air Force (FAP) is an organization focused on its mission of cooperating towards
the military defense of the Portuguese Republic and its airspace, including missions of public interest
to satisfy the population’s needs. To accomplish these objectives, one of its cornerstones is related to
sustainability and efficiency, and the innovation, research, and development of new technologies (Força
Aérea, 2021). The strategic vision of FAP regarding unmanned aerial vehicles is presented in MFA 500-
12, published under the title of ”Strategic Vision for Autonomous Unmanned Aircraft Systems”, defining
the development and operation of UAVs as a complement to accomplish its mission (Morgado, Santos,
& Caetano, 2017).
The Air Force Academy Research Center (CIAFA) is the primary source of innovation in aeronautics
within the organization, conducting aeronautical research and development while also operating small
UAVs on missions. Its activities began in 2006, and in 2009 it was approved for a major project called
PITVANT (Projeto de Investigação e Tecnologia em Veı́culos Aéreos Não-Tripulados), with the goal of
implementing a research, development, and innovation methodology centered on the development of
UAVs within the FAP for maritime surveillance and search and rescue missions. Between other projects
3
with external partners and financing, such as PERSEUS and SEAGULL, by 2017, CIAFA had accumu-
lated over 750 flight hours on Class I UAVs (less than 150 kg (Joint Air Power Competence Centre -
NATO, 2010)) and managed to implement a successful collaboration between academic research (small
scale projects and scientific dissertations) and the operation of the developed technology (Morgado et
al., 2017).
With this innovation scope in mind and paired with the development of emerging energy sources on
the aeronautical market, their implementation needs to be studied and analyzed. While ICE-powered
UAVs are preferred for long-duration missions, fossil fuels pose a serious environmental and economic
problem because they are the primary cause of the increase in CO2 presence in the atmosphere
(Boukoberine, Zhou, & Benbouzid, 2019). Global warming awareness has grown and zero-emission
policies including various sectors of society are being applied to tackle this problem. In the aviation
sector, full decarbonization is expected to occur within the second half of the XXI century, in agreement
with the goals defined in the Paris Agreement (United Nations Climate Change, 2018).
Battery-powered UAVs may pose as an alternative to ICE-based ones. They have several advan-
tages, including lower noise and thermal signatures, and no pollutant emissions, however, their low
energy density and long charging time prevent them from being used on large-scale projects (Airbus,
2020a). Another runner-up for replacing fossil fuel usage in UAVs is by using hydrogen as its fuel source.
While hydrogen is traditionally obtained by using electricity from fossil fuels in the electrolysis process to
separate the hydrogen from the oxygen, thus giving origin to the term ”grey hydrogen”, ”green hydrogen”
production, which uses electricity from renewable sources, will tend to increase over the coming years,
making hydrogen a fully zero-emission fuel source (Airbus, 2020b).
Fomented by this constant innovation pursued by the Air Force and the emerging use of hydrogen
in the aeronautical field, a design proposal for a small Class-I UAV has been made, which led to the
present project and dissertation.
1.3 Initial Design Requirements
The design proposal presents some general characteristics required for the UAV, including the ability
to perform Vertical Take-Off and Landing and having a fuel cell as its primary source of energy. Some of
the main performance and operational requirements are listed in Table 1.1.
1.3.1 Objectives & Deliverables
This thesis is part of an embryonic project proposed by CIAFA to develop an UAV powered by a fuel
cell that uses hydrogen as its fuel source. Additionally, it should carry surveillance and photography
equipment as payload and be able to perform VTOL to increase its flexibility.
Although a design project encompasses various areas like aerodynamics, propulsion, structures,
control, optimization, among others, the objective of the current dissertation is to obtain the aerody-
namic and flight performance characteristics of the aircraft, both in the preliminary and detail stages
4
Table 1.1: Design proposal presented by CIAFA.
Requirement Value Description

MTOM < 25 kg –
Payload 2 kg Includes a daylight camera
Endurance >2h Preferable over 3 hours
Cruise Speed 35-45 kts –
Stall Speed < 25 kts Without flaps
Maximum Speed 70 kts Level flight
Ceiling 15 000 ft –
Takeoff & Landing - VTOL, fully autonomous
Maximum Takeoff Altitude 10 000 ft Above mean sea level
Engine Type - Electric
of the design process. After an initial phase in which the design team came up with an initial concept
for the UAV and some baseline characteristics were determined, the author focus his studies on the
aerodynamic domain.
To perform the objective proposed for the current dissertation, the author makes a division into the
following sections:
• Theoretical background:
Review of some fundamental concepts regarding general aerodynamic definitions, design features
for the lifting surfaces and computational methods to be applied in the detailed design phase.
• Conceptual design:
While the conceptual design is part of a common project of four MSc proponents, initial perfor-
mance considerations proposed by the author at this phase are presented.
• Preliminary design:
Definition of the wing and tail geometry, together with predictions concerning the flight performance
of the aircraft and longitudinal static stability.
• Detailed design:
Application of the Computational Fluid Dynamics (CFD) methodology prescribed to validate the
models to be used; verification of the aerodynamic properties of the UAV using high-fidelity soft-
ware tools.
1.3.2 Project Outline
Based on design proposal, an initial concept was developed by the four MSc proponents which cul-
minated in a baseline solution. The considerations and methodology followed in the conceptual design
step, along with the characteristics of the UAV at the end of this design phase are available in a paired
revised conference article through reference (Alves et al., 2021). With the conceptual stage concluded,
5
Alves started working on optimizing the initial concept (Alves, 2021) and Silva, Sá and the author ex-
tended the initial design into its preliminary and detail design phase in the propulsion, structural and
aerodynamic areas, respectively (P. Silva, 2021; Sá, 2021).
It is important to note that after the conceptual design phase, the baseline solution is said to be
’locked’ meaning that changes proposed by each member of the design team during the preliminary and
detailed design are not considered for the remaining dissertations. During the course of the preliminary
and design phase, the author mentions the remaining areas whenever some minor inputs are required
for his work.
1.4 Thesis Outline
This thesis is divided into eight main chapters. Chapter 1 provides an introduction for this thesis,
together with the motivation and the design requirements of the project.
Chapter 2 gathers information over some aerodynamic concepts and design properties of wings
and tails, along with theoretical background on the software to be used further into the thesis and the
computational methods to be employed.
In chapter 3, insight on some aerodynamic and performance considerations done on the conceptual
design phase is given, which were fundamental on the process to obtain the baseline solution.
Chapter 4 encompasses the preliminary design phase where the wing and tail geometry is defined
and quick, low-fidelity software tools are employed to obtain aerodynamic data. The work done by all the
project members in the preliminary design step is put together and the characteristics and performance
of the UAV are updated.
The methodology to perform CFD simulations is tested and validated in chapter 5, providing impor-
tant feedback to be considered in later simulations.
In chapter 6, the CFD methodology is applied to a 3D model of the wing and the whole aircraft.
Improvements on the baseline solution are discussed, together with the impact of the VTOL propulsion
system during cruise. The aerodynamic properties of the UAV, obtained using high-fidelity software, are
presented at the end of the chapter
The dissertation is completed with a general overview of the project in conjunction with conclusions
taken from the various stages of the design process. Recommendations for future work and development
of the project are made in order to bring the designed UAV closer to production.
6
Chapter 2
Literature and Technology Review
To fulfill the objectives of the present dissertation, the influence of the airframe configurations and
wing geometry parameters on the performance of the aircraft is revised. As a result, simple aerody-
namics considerations like force decomposition, types of drag, boundary layer, are considered known
by the reader and thus are briefly mentioned by the author at the beginning of this chapter. The compu-
tational software used throughout the dissertation is mentioned, along with the theoretical background,
methodology, and models they apply.
2.1 Aerodynamic Considerations
The aerodynamic force acting on a plane is the sum of all the forces on the airframe caused by
the moving airstream, which creates a pressure and a shear stress distribution over the body surface
(Anderson, 2010). These forces can be split into a lift component, perpendicular to the free stream
direction, and a drag component aligned with the free stream (Lowry, 1999). For a steady level powered
flight, lift is equal to the aircraft’s weight and drag is equal to the thrust produced (Roskam & Lan, 2016).
2.1.1 Drag
Drag is a force caused by the disruption of the airflow by the airframe (Federal Aviation Administration,
2016). Drag force is usually divided into two components: the parasite drag, which is non-dependent on
the lift production, and the induced part, defined as the component ”induced” by the lift production, as
the name suggests (Roskam & Lan, 2016).
Parasite drag can be further detailed into form, skin friction, and interference drag. Form drag, or
pressure drag, occurs when airflow deviates from its original path around the aircraft’s shape (Federal
Aviation Administration, 2016); if flow separation occurs, this drag component contributes significantly to
parasite drag (Anderson, 2012). Skin-friction drag is related to the shear forces acting in the boundary
layer normal to the surface due to the viscosity of the flow (Gundlach, 2012). According to Gundlach
(2012), interference drag is caused by the interference of two or more aerodynamic bodies that are not
necessarily in direct contact (the tail being in the wake of the wing, for example).
7
Induced drag occurs due to the downwash produced by the wingtip vortices. Anderson (2010)
presents an in-depth explanation for the induced drag origin and phenomena. When considering only
the wings, Equation (2.1) illustrates the relation between the induced drag coefficient CDi and the lift
coefficient CL , where AR is the aspect ratio of the wing and ef represents the span efficiency factor of a
finite wing which varies between 0.9 and 1.0, being closer to the unit the closer the wing is to an elliptical
one (Anderson, 2010).
CL2
CDi wing = (2.1)
πARef
Wave drag is an additional component that influences both parasite and induced drag and occurs at
flight speeds higher than the critical Mach number due to the presence of compressibility effects such as
shock waves (Raymer, 1992). For low subsonic speeds, compressibility effects are negligible and thus,
not relevant to the scope of the current thesis.
Equation (2.2), also known as the aircraft drag polar represents the total aircraft drag coefficient as
the sum of the aircraft parasite drag coefficient at zero lift – CD0 and the drag coefficient due to lift, with
the latter being given by the same expression as Equation (2.1), however, when considering the entire
aircraft, e is denominated the Oswald efficiency factor and its value typically ranges from 0.7 to 0.85
depending on the aircraft design (Anderson, 2010). The parameters which multiply the square of the lift
coefficient are commonly grouped using the K notation and represent the induced drag factor (Sadrey,
2013).
CL2
CD = CD0 + KCL2 = CD0 + (2.2)
πARe
2.1.2 Boundary Layer
The boundary layer (BL) is the thin region of flow in contact with a surface, where the flow is retarded
due to the influence of shearing forces on the surface, originating a velocity profile from zero at the
surface due to the no-slip condition, to the freestream velocity where the influence of the shear forces
does not impact the flow (Anderson, 2010). The no-slip condition occurs between a solid surface and
a fluid, and depends on the viscosity of the latter, as such, a dimensionless parameter denominated
Reynolds number – Re is used to define the relative influence of viscosity when compared to the inertial
forces of the fluid – Equation (2.3).
ρU l
Re = (2.3)
µ
Reynolds number accounts for the density of the air – ρ, the freestream velocity – U , a characteristic
length – l, and the dynamic viscosity of the fluid – µ. For wings, the mean aerodynamic chord is
chosen as the characteristic length. Small UAVs have lower Reynolds numbers than commercial aircraft
because of their lower flight speeds and smaller size, typically less than 5 × 105 , defined as the Low
Reynolds Number (LRN) regime, according to Gundlach (2012).
8
The BL can be divided into three regions: an initial laminar region characterized by a highly ordered
motion of the flow; a transition region where the boundary layer becomes unstable and the presence
of Tollmien-Schlichting waves induces the transition to turbulent flow, originating a third region marked
by a chaotic and irregular nature of the flow, governed by chaotic fluctuations over an average flow field
(Brederode, 2018; Schlichting, 1979).
Under LRN, laminar flow occurs over the airfoil even after the minimum pressure point. Due to the
adverse pressure gradient downstream of this location, the flow separates from the surface, forming a
shear layer. At Reynolds numbers over 5 × 104 transition occurs and, if the adverse pressure gradient
is not too large, the flow can recover sufficient energy through entertainment to reattach to the airfoil
surface and redevelop a turbulent BL. The region between the separation and the reattachment, visible
in Figure 2.1 between points S and R, is denominated as the Laminar Separation Bubble (LSB) or
transitional separation bubble since it induces flow transition (Mueller & DeLaurier, 2003).
Figure 2.1: Laminar separation bubble. Retrieved from Mueller and DeLaurier (2003).
Focusing the study on Reynolds above 7 × 104 which excludes most of the Micro UAVs, the LSB
is present and occupies a significant portion of the airfoil. As the angle of attack increases, the reat-
tachment point gets closer to the trailing edge, increasing the size of the bubble. When the necessary
pressure for the reattachment is not present, the bubble bursts, causing the airfoil to stall. For higher
Reynolds number (over 2×105 ), still within the LRN scope, the LSB is shorter and fully induces transition
to turbulent flow; a short bubble reduces pressure drag while turbulent flow increases skin-friction drag
when compared to laminar flow (Mueller & DeLaurier, 2003).
2.2 Design Features
2.2.1 Airframe Configuration
Airframe configurations for UAVs are vast and even more when compared with crewed aircraft since
the commercial risk in trying innovative solutions is less for unmanned aircraft and there is no need to
accommodate space for the flight crew. Furthermore, UAV airframes are usually much smaller than
9
human-piloted aircraft (Austin, 2010).
Having the conceptual design in mind, the author chose to narrow the airframe configurations study
to the ones which can be applied into a Lift+Cruise configuration, not focusing on traditional VTOL
configurations such as single rotors, the common quad-copter or any other hybrid configurations (e.g.
tilt-rotor, tilt-wing, tail sitter) since these configurations were analyzed in the initial design phase with the
concept generation and they are described in detail on reference (Alves et al., 2021).
One decision that should be made is to select the number of wings. While biplane and triplane con-
figurations were common in the early days of aviation, one wing configurations became the standard in
aviation (Sadrey, 2013). However, when considering the location and the presence of control surfaces,
aircraft configurations can be categorized into ’tailplane aft’, ’tailplane forward’, ’three surfaces’ or ’tail-
less’ (Austin, 2010) A summary of the main characteristics of each configuration is presented, retrieved
from references (Austin, 2010; Gundlach, 2012; Raymer, 1992; Sadrey, 2013).
• Tailplane aft:
– The center of gravity (CG) of the aircraft is located ahead of the wing center of lift;
– The tail is located after the wing and has the elevator and rudder (or a combination of both);
– Allows for a single-boom or twin-boom tail design;
– The tail produces a negative lift contribution to trim the aircraft.
• Tailplane forward (canards):
– Canards are ahead of the wing;
– The CG is located ahead of the wing, therefore, canards create positive lift to trim the aircraft;
– The wake of the canards interfere with the freestream of the wing, compromising the lift gen-
eration of the latter.
• Three surfaces:
– Is a combination of the previous configurations, by having a tail and canards;
– Has a higher margin for the CG location;
– Extra surfaces increase the parasite drag, aircraft weight and complexity.
• Tailless:
– Does not possess neither a tail, nor canards;
– The ailerons at the wing are responsible for both the lateral and longitudinal control;
– The longitudinal control is compromised due to the ailerons short arm;
– May require a vertical fin for aid in yaw control.
The main wing can be categorized as high, mid, or low, depending on its vertical placement. Each
configuration has its advantages and disadvantages, as such, some remarks concerning its application
10
in UAVs are considered. High wings allow the placement of the fuselage closer to the ground, although
this limits the integration of external payload at the bottom of the fuselage, like cameras, for instance. If
the wing is positioned over the fuselage, the wing spar can be a continuous beam, structurally favorable
at the cost of an increase in the frontal area and, consequently, the overall drag. A low wing layout has
increased maneuverability yet lower ground clearance. To increase lateral stability, low wings commonly
have dihedral associated to them. A mid-wing is a compromise between the previous configurations
and it presents the lowest interference drag. Nevertheless, the wing spar has to be cut in two sections
if payload is to be carried on the fuselage at the wing location and the wing root intersection with the
fuselage has to be reinforced, increasing the structure weight (Sadrey, 2013).
2.2.2 Wing Geometry
The wing geometry has an impact on how much drag this structure produces in flight. As seen
in Equation (2.2), the aspect ratio of the wing has a significant impact on the induced drag. For the
same wing area, the higher the AR, the higher the wingspan is, reducing the strength of wingtip vortices
and thus decreasing the induced drag (Raymer, 1992). However, high AR wings are heavier since
the structure needs to be stiffened as the wing weight bending moments gets larger, increasing the
root bending moment, therefore, requiring a stronger wing root. From a control perspective, on a high
AR wing, the ailerons have a longer arm providing more lateral control, however, the mass moment
of inertia around the longitudinal axis is higher, decreasing the maneuverability of the aircraft in roll
(Sadrey, 2013).
The general planform of the wing impacts its lift distribution. The minimum induced drag occurs
when the lift distribution is elliptical, which can be achieved with an elliptical wing planform. While this
geometry is desirable, its difficulty in construction prevents its general application in the aviation industry
(Raymer, 1992).
A rectangular wing has the simplest geometry. While it is easy to manufacture and has reasonably
good performance, it is aerodynamically inefficient by producing induced drag due to the excessive
chord length towards the wingtip, as such stall occurs primarily over the root and extends towards the
tip, giving room for control at stall conditions due to the location of the ailerons being near the tip.
Taper ratio of a wing is defined as the fraction between the root chord and the tip chord. When
taper is applied to rectangular wings, it reduces the tip chord and approximates the lift distribution to the
desired elliptical one. Raymer (1992) states: ”A taper ratio of 0.45 (... ) produces a lift distribution very
close to the elliptical ideal. This results in drag due to lift less than 1 % higher than the ideal, elliptical
wing.”. From a structural point of view, applying taper to a wing is also beneficial as the aerodynamic
loads on the tip are diminished when compared to an untapered case, resulting in a lower root bending
moment. However, with the reduction of the tip chord, these wing sections tend to stall first, which is
critical from a control perspective. As a result, taper ratio can be restricted due to the requirement for an
adequate chord at the tip for roll control with the ailerons (Howe, 2000). The delta wing is a special case
in which taper ratio is equal to 0. This shape is mainly used in high-subsonic or supersonic aircraft, and
11
its advantages at these flight speeds are coupled with the introduction of sweep (Sadrey, 2013).
The considerations stated above are applied and used for general aviation, and the principles behind
them are valid to small UAVs. However, when conducting a market study on similar UAV designs to the
one projected, three main planforms have risen: the basic rectangular wing, an elliptical leading edge
with a straight trailing edge, and a multi-panel tapered wing. It is relevant to note that for the latter, the
tapered sections present higher values (less taper) than the one presented as the ideal to prevent the
chord at the tip from being too small.
Sweep angle, more specifically, the leading edge sweep, is the angle between the wing leading edge
and the span-wise axis of the aircraft. The main reason to introduce sweep on the wing leading edge
is to increase performance at speeds higher than transonic to delay compressibility effects. For the
Mach number (Ma) considered, these effects do not occur, and Sadrey suggests that for M a below 0.3,
there should be no leading-edge sweep since sweep increases the effective chord length, reducing the
effective thickness-to-chord ratio which, for low speeds, is not ideal (Sadrey, 2013). Sweep angle also
influences the performance of the wing control surfaces and the general aesthetics of the plane.
To reduce the tip stall effects from tapered wings and to approximate the lift distribution to the desired
elliptical one, one solution is to introduce twist on the wing. This twist can be geometric in nature, in
which the tip has a lower incidence than the root, or aerodynamic when there is a difference in the airfoil
shape along the wingspan (Sadrey, 2013). When geometric twist is applied, the leading edge of the tip
has a lower incidence by around 2◦ to 6◦ than the root and the twist evolution along the wingspan is
usually linear. One drawback of applying geometric twist is the difficulty in the manufacturing process:
the main spar may also need to be twisted, which is problematic if composite materials are used in its
construction, since composite fibers’ strength is very sensitive to fiber alignment. Aerodynamic twist
does not have this problem, as the twist arises from the difference in the stall angle of each of the
airfoils chosen for the root and the tip – the tip airfoil has a higher 2D stall angle and is usually thinner
(Gudmundsson, 2014).
Dihedral is the angle the wing makes relative to the ground plane when viewing the airplane from the
front. Dihedral can be positive if the tip is higher than the root or negative (anhedral) if the opposite is true
(Gudmundsson, 2014). Dihedral is applied to improve the lateral stability of the aircraft (the tendency of
the aircraft to return to an equilibrium position after a disturbance occurs which makes the aircraft roll) at
the cost of a small reduction on the wing effective area which impacts the lift generation (Howe, 2000).
Most low-wing configurations have dihedral to guarantee ground clearance for any nacelles or propellers
mounted on the wing (Sadrey, 2013).
2.2.3 Tail Configuration
Looking further into the conventional airframe configuration, two main groups of tail configurations
arise: single-boom and twin-boom tails, as stated in sub-section 2.2.1. Each group has numerous
layouts, however, the author will focus on the twin-boom arrangements. Although they are typically
heavier, twin-boom tails are vastly used in conjunction with pusher configurations as they reduce the
12
overall length of the aircraft (Raymer, 1992).
As referred in Gundlach (2012), there are four main tail configurations for twin-boom design, as seen
in the Figure 2.2: Conventional, H-tail, U-tail and Inverted V-tail.
Figure 2.2: Tail configurations for Double Boom Displacement. Adapted from Gundlach (2012).
A twin-boom conventional tail has the horizontal stabilizer on the same plane as the booms and
two vertical fins, each one on top of one boom. The vertical fins act as endplates, increasing the
effectiveness of the horizontal stabilizer (Gudmundsson, 2014). On an H tail, the vertical stabilizers
extend below the horizontal stabilizer plan, making the overall aesthetics more appealing at the cost of
lower ground clearance for take-off.
The horizontal stabilizer is in the wake of the propeller in both of the previous designs, increasing
the drag produced by it (Gudmundsson, 2014). One solution is to allocate the horizontal stabilizer on
top of the fins on the conventional tail, originating the twin-boom T arrangement. While this removes the
influence of the propeller wake and the wing downwash on the tail, due to the position of the horizontal
tail, the bending moments created by it must be transferred through the main spar of the fins, causing
the tail to be structurally more robust and heavier. Moreover, this configuration is subjected to deep stall,
a critical condition that occurs when the tail is in the wing wake at stall or post-stall conditions, reducing
the capability of the elevators (Sadrey, 2013).
The inverted V-tail, also known as A-tail, has the same advantage as the T-tail by preventing the wake
of the propeller and the downwash from the wing to interfere with the tail, providing less drag. This layout
also delivers the lowest interference drag when compared with any other tail (Raymer, 1992; Stettmaier,
1999), provides good stability in crosswind conditions (for this scenario, the tail will provide a pitch-up
movement). Furthermore, the cost of construction is lower since the parts are symmetric and it uses
fewer materials, making it lighter (Fathi, 2019). Nevertheless, this configuration implies a more complex
control actuation process since the rudder and elevator control inputs are combined. These new control
surfaces are named ”ruddervators” (Raymer, 1992).
2.3 Low-fidelity Software
During the design process, numerical methods and computational simulations can be used at the
preliminary phase to corroborate the results from the semi-empirical and historical trends given in the
13
literature and provide further information. The tools used at this still early phase of the design should be
simple and have acceptable reliability while not being time-consuming and computationally heavy.
XFLR5 is an open-source analysis tool for Low-Reynolds-Number airfoils, wings, and planes that
is widely used in academia. XFLR5 uses XFOIL for 2D direct and inverse analysis, and for 3D wing
simulations, several methods from Lifting Line Theory (LLT), 3D Panel Method, or Vortex Lattice Method
(VLM) can be applied at the choice of the user (Deperrois, 2009).
OpenVSP, created in 2012 by NASA, is an open-source parametric aircraft geometry software with
an integrated and simplified design tool for wings and fuselages, allowing the user to create a 3D model
of an aircraft with ease. It incorporates a parasitic drag tool, in which the user can choose the empirical
formulations to use from a wide database, and also a solver for 3D analysis, similar to XFLR5, resorting
to VLM or 3D Panels Method (NASA Open Source Agreement, 2021).
While both tools are capable of conducting preliminary aerodynamic analysis, XFLR5 will be used at
the early stages of the design process to obtain data regarding the lifting surfaces (wing and tail). The
parasite drag tool of OpenVSP is favored when conducting the drag build-up at the preliminary phase
since it incorporates models to predict the drag of the remaining aircraft components.
2.3.1 XFOIL
XFOIL is an interactive program for the design and analysis of subsonic, isolated airfoils capable of
performing inviscid or viscous analysis while accounting for eventual transition and/or separation of the
boundary layer and outputting pressure distribution and drag polars for the required Reynolds and Mach
numbers set as user inputs (Drela & Youngren, 2001).
XFOIL inviscid formulation uses a simple linear-vorticity model in which a freestream flow, a vortex
sheet on the airfoil surface, and a source sheet on the airfoil surface and wake are superimposed. The
airfoil contour and its wake are discretized into flat panels (number of panels being user-specified), with
each one having a distribution of the properties stated before. The trailing edge (TE) of the airfoil is
closed with a vertical panel and the Kutta condition is applied at the TE to close the system of linear
equations. The freestream angle influences the inviscid vortex strengths, providing an inviscid solution
for each angle of attack (Drela, 1989; Peerlings, 2018).
Viscosity is introduced by employing von Kármán’s momentum integral equation and kinetic energy
shape parameter equation and discretizing them over the same panels using two-point central differ-
ences (Drela, 1989). These equations incorporate BL properties like the kinetic energy shape parameter,
density shape parameter, and skin friction coefficient which have different approximations depending on
the flow regime on the BL. For laminar flow, approximations of Falkner-Skan are used, while for turbulent
flow, various models presented in reference (Swafford, 1983) are applied.
The transition from laminar flow to turbulent is predicted according to eN criteria. This method has
been used since 1956 and models the growth of Tollmien-Schlichting waves in the boundary layer which
are the precursor of the transition. These unstable waves are perturbations of various frequencies, with
the most unstable one being characterized by an amplification factor related to its growth rate, the N -
14
factor. When the amplitudes of the 2D Tollmien-Schlichting waves surpass a certain threshold, transition
occurs. The critical amplification factor (at the transition point) Ncrit typically has a value between 6 and
12 depending on disturbance levels (Wu et al., 2019) and, for airfoils, a value of 9 for Ncrit is considered
the standard (Drela & Youngren, 2001).
The viscous/inviscid coupling is achieved via the surface respiration model, which replaces the
source strength presented in the inviscid formulation with the local gradient of mass deflected, related to
the boundary layer thickness (Drela, 1989). This approach to the coupling reduces the number of equa-
tions to be solved by a Newton non-linear system method and allows a reduction of the computational
time to one-third when compared to the full system (Drela, 1989). The low computational effort, together
with the capability to accurately account for separation, transition, and bubble losses, makes XFOIL the
chosen tool for initial 2D aerodynamic analysis.
2.3.2 Vortex Lattice Method
From the three types of 3D solver methods XFLR5 has to offer, for a wing + tail analysis, only VLM
is available, therefore this method is the focus of the literature review. VLM background is based on the
theoretical model of a system of vortices that have the same impact as the wing on the flow and sustain
a force equivalent to the wing lift (Houghton, Carpenter, Collicott, & Valentine, 2013). According to the
Kutta-Joukowski theorem, a bound vortex filament can replace a finite wing and translate its lift force.
Moreover, due to the Helmholtz theorem, two free vortices trailing downstream from the wingtips should
also be considered. This vortex system is called a horseshoe vortex due to its shape (Anderson, 2010).
In the VLM formulation, the wing surface is split into quadrilateral panels and a horseshoe vortex
is placed at one-quarter chord of each panel. Applying the Biot-Savart law and considering the flow
tangency condition at control points at 75% of each panel chord, a system of linear equations is set up
with the horseshoe vortex strength of each panel as the unknown, which, after solved, can be used to
compute lift and induced drag (Peerlings, 2018). The VLM incorporates viscosity by interpolating the lift
coefficient from the airfoil 2D analysis done by XFOIL, therefore, results near the stall condition should
not be considered (Dantsker & Vahora, 2018).
XFLR5 allows the user to choose between ring vortices instead of the classic horseshoe ones on
the VLM formulation. Ring vortices form a close loop with an additional vortex line after the end of
each panel and only the panels located after the trailing edge extend downwards to infinity (horseshoe
vortices). Figure 2.3 illustrates the difference between the two formulations. Deperrois (2009) states that
considering ring vortices is simpler from the programming point of view, nonetheless, both formulations
present almost identical results.
Although VLM does not consider thickness in its analysis, it provides results for wings with sweep
and dihedral (Peerlings, 2018). For a configuration with multiple surfaces, it takes into consideration the
interaction of the wake produced by a forehand surface onto another one downstream of it. Moreover,
VLM is capable of estimating stability and control derivatives (Gundlach, 2012), useful for an initial
stability analysis done in the preliminary design phase.
15
(a) Horseshoe Vortices (b) Ring Vortices
Figure 2.3: Vortex lattice method implementations. Retrieved from Peerlings (2018).
2.4 Computational Fluid Dynamics
Computational Fluid Dynamics is the standard simulation tool in the Computational Aided Engineer-
ing (CAE) domain for design, analysis, performance determination, and investigation of engineering
systems involving fluid flows. Initially used by the aeronautics and aerospace industry, it has grown in
applications and become an essential tool for other design-intensive industries such as the automo-
tive, power generation, chemical, nuclear, and marine industries (Moukalled et al., 2015). Due to the
complexity of the fluid governing equations which need to be solved, the development and evolution of
CFD software is linked with the technological advances of the past three decades, namely the increase
in memory capacity and more affordable computer speed, leading to a steady reduction in the cost of
simulations when compared to traditional experimental testing (Hirsch, 2009). In the aeronautical de-
sign process, at the preliminary and detailed phases, CFD tools are employed to determine the entire
flow field around the aircraft at any conditions specified by the user, unlike wind tunnel testing, which is
usually done at a lower scale and some similarity analysis has to be employed to simulate the real-world
flight conditions (Raymer, 1992).
To develop a CFD simulation, Hirsch (2009) and Moukalled et al. (2015) present a systematic step-
by-step process that is coherent with the schematic presented by the CFD software of choice. Figure
2.4 represents a diagram with the methodology defined by these two authors, which serves as the basis
for the current chapter.
After the identification of the real-world problem to be simulated, the first step is to define the domain
of interest and develop a Computer Aided Design (CAD) model of it. For bounded flows or internal
aerodynamic problems, the region of interest is easily identified; for external aerodynamic flow analysis,
the domain needs to be carefully thought out: the distance between the geometry to be modeled and
the far-field should be large enough to prevent the boundary conditions applied at the outer bounds of
the domain from interfering with the flow next to it and affecting the quality of the obtained results (Ideal
16
Figure 2.4: Structure of a CFD simulation system. Adapted from Hirsch (2009); Moukalled et al. (2015).
Simulations, 2020). In parallel with the domain modeling, the physical phenomenon is mathematically
postulated, and some approximations must be made in some cases (Moukalled et al., 2015). Section
2.4.1 presents some insight into the physical modeling process.
With the governing equations and the computational domain set, the geometrical and mathematical
models which are continuous are translated into discrete and numerical counterparts, through the dis-
cretization process. The geometry is discretized into a set of points representing the real space, forming
a grid or a mesh (Hirsch, 2009). Similarly, the governing equations are discretized into a set of algebraic
equations and applied to each element of the mesh created (Moukalled et al., 2015). Both the domain
and equations’ discretization processes are further detailed in sections 2.4.2 and 2.4.3, respectively.
The system of algebraic equations from the previous step must be solved to obtain the discrete val-
ues of the variables to be calculated. Direct or iterative methods may be used. The former is based on
inverting the matrix of coefficients of the system of algebraic equations, which requires large computa-
tional and storage requirements for problems with hundreds of thousands of cells, this being rarely used
for complex CFD simulations. Iterative methods rely on an initial value and a guess-and-correct proce-
dure to obtain more accurate solutions, repeatedly solving the set of equations. This iterative process
continues until a stopping criterion is met, related to the difference between the ongoing solution and the
previous iteration. Whenever this difference decreases with the number of iterations, the process is said
to be converging, which is a must to guarantee that the solution is acceptable (Moukalled et al., 2015).
17
Software Solutions
To perform CFD simulations, both commercial and open source software packages are available for
use.
Star-CCM+ is a commercial software from Siemens which has an integrated environment including
all phases of the CFD process, from developing the CAD model to creating the mesh and modeling
the physics. Its post-processing capabilities are remarkable and it is recognized as a reference for
aeronautical applications (Siemens, 2021).
The Ansys package contains a wide variety of analysis systems in the engineering domain, from
material and structural analysis to fluid simulations, easing the coupling of these two domains if required.
Fluent and CFX are the Ansys reference analysis systems for fluid simulations, both having an integrated
environment similar to Star-CCM+, each one with its applications in the aeronautical field. (Ansys Inc.,
2021). Due to its accurate results, versatility and the option to use a free student version to familiarize
and solve simple problems and test cases, Ansys has become a staple tool in the CAE domain.
OpenFOAM has been the leading free open source CFD software since 2004, largely used for in-
dustrial applications and in academia. It is capable of dealing with an array of problems, from heat
transfer and turbulence to acoustics and chemical reactions. It also has meshing, pre-processing, and
post-processing tools embedded with the CFD package (OpenCFD Ltd, 2021).
Since the Air Force Academy (AFA) possesses the academic license for the Ansys 2019 package,
without restrictions on the meshing number of elements/nodes, it is the software of choice for the pre-
sented thesis regarding high-fidelity CFD simulations. Between Fluent and CFX, Fluent is chosen due
to its advanced physics modeling in aeronautical applications to freestream flows and user-friendly in-
terface, while the latter is recommended for turbomachinery applications (Ansys Inc., 2021).
2.4.1 Physical Modeling
The governing equations in fluid dynamics are known as the Navier-Stokes equations, expressing the
conservation of mass, momentum and energy through a system of five fully coupled time-dependent par-
tial differential and nonlinear equations. For an incompressible flow, the energy equation becomes de-
coupled from the remaining Navier-Stokes equations whenever viscosity is assumed to be independent
of the temperature, a simplification valid for low-speed cases where the fluid is considered isothermal.
While laminar flows are well solved using computational methods without any additional information,
turbulent flows are chaotic and involve 3D vorticity fluctuations (Hirsch, 2009).
One way to model turbulence is to assume the presence of eddies of different sizes in which the
energy they carry is dependent on their dimensions. Larger eddies break up and transfer their energy
into smaller ones until the smallest eddy size is reached and their molecular viscosity dissipates the
turbulent kinetic energy as heat. This turbulence modeling approach is applied in Direct Numerical
Simulation (DNS), however, it requires a very small time step and a mesh fine enough to capture these
tiny eddies, resulting in a computational cost that is extremely high and still not viable to solve industrial
problems with the current computational power and memory (Moukalled et al., 2015). To reduce the
18
large computational cost associated with a direct solution of the Navier-Stokes equations, Large Eddy
Simulation (LES) is an approach similar to DNS which models the larger-scale eddies to directly simulate
the turbulent fluctuations (Hirsch, 2009).
The most widely used approach is the Reynolds Averaged Navier-Stokes (RANS). In this model, a
statistical averaging, presented in Equation (2.4), is applied to approximate random fluctuations, decom-
posing the flow variables into a time-mean value component – ξ and fluctuating one – ξ 0 , substituting
in the original Navier-Stokes equations, and time-averaging the obtained equations (Moukalled et al.,
2015).
ξ(xi , t) = ξ(xi ) + ξ 0 (xi , t) (2.4)
The RANS equations for continuity (2.5a) and momentum (2.5b) are written in tensor notation, with p
denoting the pressure term. Averaging the Navier Stokes introduces additional unknowns to the momen-
tum equation with the term ρu0i u0j , known as the Reynolds stress tensor. As such, RANS equations are
not a closed system of equations and require additional equations in an attempt to derive the Reynolds
stress tensor in terms of the mean quantities, denoted as turbulence modeling (Moukalled et al., 2015).
∂(ρui )
= 0, (2.5a)
∂xi

∂(ρui ) ∂ 0 0
∂p ∂ ∂ui ∂uj
+ ρui uj + ρui uj = − + µ + (2.5b)
∂t ∂xj ∂xi ∂xj ∂xj ∂xi
The turbulence models can be grouped according to the number of equations they add to the RANS
system. Three turbulence models, widely used in commercial CFD software, are presented together with
their applications and limitations. Additionally, to accommodate the transition from laminar to turbulent
flow, transition models are also applied and two of them, available in Fluent, are mentioned.
Spalart-Allmaras Model
Created by P. R. Spalart and S. R. Allmaras in 1992, the Spalart-Allmaras (SA) one-equation model
resorts to a transport equation model for the kinematic eddy viscosity. This model was developed to im-
prove the existing algebraic (zero-equation models) and one-equation models since they treat the BL as
a single module, presenting discrepancies when separation occurs or multiple shear layers are present
(Spalart & Allmaras, 1992). Being a one-equation model, it is simpler and requires less computational
effort than the two-equation models.
The SA model was designed initially for aerospace applications for wall-bounded flows like airfoils,
nonetheless, its usage has extended to other fields like turbomachinery applications. This model was
validated for the RAE 2822 airfoil, presenting a lift force coefficient consistent with experimental results,
giving good results for boundary layers subjected to adverse pressure gradients. Despite the authors’
improvements, the presence of massive separations or free vortices appears to be limitations of the SA
model (Spalart & Allmaras, 1992).
19
k − Model
The turbulence kinetic energy k and turbulent energy dissipation rate are computed in this two-
equation turbulence model. Developed by B.E. Launder and D.B. Spalding in 1974, its applications
are varied, being capable of modeling with acceptable accuracy free shear layer flows, heat transfer
problems, and flows in pipes, among others, presented in reference (Launder & Spalding, 1974).
Because the k − model is based on the assumption that the flow is fully turbulent and that the
effects of molecular viscosity are negligible, it is not recommended for use in LRN applications where
the laminar portion of the flow and transition have a significant impact, as such, it is not widely used in
external aerodynamics simulations (Ansys Inc., 2013).
SST k − ω Model
The Standard k − ω model, first proposed by Kolmogorov in 1942, is a two-equations model which
considers the specific dissipation rate ω. Wilcox elaborated on the formulation, simplifying it by consid-
ering ω as a ratio of to k. Wilcox’s formulation presents valid results for flows close to walls in the
boundary layer, particularly under adverse pressure gradients (Wilcox, 2006).
A modification to the turbulent viscosity formulation to account for the transport effects of the principal
turbulent shear stress gave origin to the Shear-Stress Transport (SST) which uses the k − ω turbulence
model in the inner regions of the boundary layer and the standard k − model in the outer regions and
in free-shear flows (Ansys Inc., 2013).
The SST k − ω transport equations are presented below,

Dρk ∂ui ∂ ∂k
= τij − β ∗ ρωk + (µ + σk µt ) (2.6a)
Dt ∂xj ∂xj ∂xj

Dρω γk ∂ui ∂ ∂ω 1 ∂k ∂ω
= τij − βρω 2 + (µ + σω µt ) + 2ρ(1 − F1 )σω2 (2.6b)
Dt νt ∂xj ∂xj ∂xj ω ∂xj ∂xj
in which τij represents the wall shear stress tensor, β, β ∗ , γk , σk , σω and σω2 are constants, µt is the
eddy viscosity, νt is the kinematic eddy viscosity and F1 is a blending function that goes from zero away
from the surface to one close to the surface. The reference values for the constants are presented in
references (Ansys Inc., 2013; Menter, 1994).
By coupling the two-equation turbulence models presented, SST presents good results in either free
shear layer flows and in adverse pressure gradient flows on boundary layers, as such this model is widely
used in aeronautical applications and has become a reference in this field of CFD (Menter, 1994).
Transition k − kl − ω
In a complement to the k − ω turbulence model, in 2008, D. Walters and D. Cokljat introduced a third
equation to predict the magnitude of low-frequency velocity fluctuations in the pre-transitional boundary
layer that have been identified as the precursors to transition. This model improved on the k − ω model
for laminar-turbulent flows, providing accurate data for some test cases like turbulent channel flows, flat
plate with pressure gradient, and airfoils, being able to predict the transition location within approximately
10–20% for most of the cases (Walters & Cokljat, 2008).
20
γ − Reθ Transition Model
Traditional transition predictions are not able to reliably capture the influence of pressure gradients
and separation among other phenomena, as such, in 2006, F. R. Menter and R. B. Langtry proposed
a new transition method based on local variables to be compatible with the existing turbulence models.
This model postulates two transport equations: one for intermittency γ and one for the transition onset
criteria related to the momentum-thickness Reynolds number Reθ (Menter et al., 2006).
The transport equation for intermittency (2.7a) is used to trigger transition locally, and coupled with
the SST k − ω base model, is capable of accurately accounting for transition caused by the LSB, as
described in section 2.1.2. The second equation (2.7b), related to Reθ , captures the non-local effects of
freestream turbulence intensity and pressure gradient at the boundary layer edge. References (Menter
et al., 2006) and (Langtry et al., 2006) present further information about the model formulation and
validation through test cases. For the reasons stated above, this model is also named SST Transition
and encompasses the two equations of the SST turbulence model and the two transport equations
related to the transition (Ansys Inc., 2013).

∂(ργ) ∂(ρuj γ) ∂ µt ∂γ
+ = Pγ1 − Eγ1 + Pγ2 − Eγ2 + µ+ (2.7a)
∂t ∂xj ∂xj σγ ∂xj
" #
∂(ρRe ˜ θt )
˜ θt ) ∂(ρuj Re ∂ ˜ θt
∂ Re
+ = Pθt + σθt (µ + µt ) (2.7b)
∂t ∂xj ∂xj ∂xj
In Equation (2.7a), Pγ1 and Eγ1 are transition sources, and Pγ2 and Eγ2 are destruction sources.
These parameters depend on the length of the transition region and the vorticity magnitude, respec-
˜ θt to match the local
tively. Pθt is a source term in Equation (2.7b) that forces the transported scalar Re
value of Reθt , obtained from an empirical correlation outside the boundary layer, and σθt and σγ are
constants. The formulation of these parameters is fully described in (Menter et al., 2006), together with
the constants.
2.4.2 Domain Discretization
The computational domain discretization into grid cells to properly define the physical domain is
denominated as meshing or grid generation. The mesh influences the accuracy, convergence, and
speed of the simulation, therefore, an adequate mesh is essential to provide adequate resolution for
the CFD simulation to be run (Moukalled et al., 2015). The grids can have various types and shapes,
each with its generation method. The nomenclature for the types of grids varies between the references
consulted, as such Hirsch (Hirsch, 2009) will be the base reference to be followed, classifying grids as
being structured or unstructured, with various sub-types in each category based on their topology.
A structured grid has its grid points at the intersection of co-ordinates lines, with each interior grid
point having a fixed number of neighboring grid points (Versteeg & Malalasekera, 1995). Structured
grids possess only quadrilateral (4 faces) elements in 2D meshes and hexagons (6 faces) in 3D, making
it possible to attribute a set of indices (i, j, k) to each computational cell, and each neighbor has its own
21
indices differing by plus or minus one (Moukalled et al., 2015). While structured grids are more efficient
in terms of accuracy, processor and memory requirements, they are harder to implement on complex
geometries.
Cartesian grids present the gridlines aligned with the Cartesian coordinate system and are the ideal
topology from an accuracy point of view. Whenever a local refinement is required, the mesh spacing is
reduced locally, forming a denser cell region, resulting in a non-uniform Cartesian grid; or the grid can
be sub-divided into smaller cells to adapt to the geometry, forming a quadtree/octree grid (Figure 2.5a).
To deal with curvilinear geometries, body-fitted structured grids are employed, with their configuration
being related to the letter they resemble the most (H-type, C-type Figure 2.5b, O-type, for example).
To increase the flexibility of structured grid generation, a combination of techniques can be applied to
various domain zones, forming multi-block grids or even overlapping independent grids in a technique
called ”chimera” (Figure 2.5c), the latter being a powerful tool since no extra care is needed for the mesh
mating of the different grid topologies, though the overlapping grid zone must be accurately interpolated
(Hirsch, 2009).
(a) Quadtree Grid (Versteeg & (b) Body-fitted C-type Grid (Versteeg & (c) Chimera Grid (Hirsch, 2009)
Malalasekera, 1995) Malalasekera, 1995)
Figure 2.5: Different topologies for structured grids.
For arbitrary geometries and when using automatic mesh generation software, structured grids are
nearly impossible to be generated, therefore, unstructured grids and flow solvers for this mesh type have
gained wide acceptance. Unstructured grids offer more flexibility and local refinements are easier to im-
plement without affecting the surroundings (Hirsch, 2009). The elements can be triangular/tetrahedral
or quadrilateral/hexahedral for a 2D/3D space domain, respectively, and their enumeration is done se-
quentially and no indication of the local connectivity or position can be obtained based on their indices
(Moukalled et al., 2015; Versteeg & Malalasekera, 1995). Various methods for generating an unstruc-
tured grid can be applied, however, grid generation techniques are beyond the scope of the presented
thesis.
Fully triangular/tetrahedral grids (Figure 2.6a), while not ideal for calculating boundary layer velocity
profiles, are easily generated by automatic meshers. At these locations, quadrilaterals for 2D and prisms
for 3D with proper aspect ratios are more indicated, forming a hybrid unstructured mesh (Figure 2.6b)
(Hirsch, 2009). Although hexahedral cells require less memory and have increased accuracy, their
generation might be harder in regions with high curvature and sharp angles, generating cells with poor
orthogonal quality (Moukalled et al., 2015).
22
(a) Triangular Cells Grid (Versteeg & Malalasekera, 1995) (b) Hybrid Grid (Versteeg & Malalasekera, 1995)
Figure 2.6: Different topologies for unstructured grids.
Near-Wall Treatment
The no-slip condition at the wall affects the velocity profile in its vicinity in addition to the turbulence.
Close to the wall, turbulent fluctuations are reduced and viscous forces become dominant. Experimen-
tal results demonstrated that the near-wall region can be subdivided into sub-layers according to the
prevalence of these phenomena (Ansys Inc., 2013). These layers exibit different dimensionless velocity
profiles U + along its dimensionless height y + , and this evolution is commonly displayed on a logarithmic
scale (Figure 2.7). The dimensionless variables are obtained considering the velocity of the flow and the
wall friction velocity uτ , which is related to the velocity gradient near the wall, expressed in terms of the
wall shear stress τw (Brederode, 2018).
r
U∞ τw
U+ = , uτ = (2.8a)
uτ ρ
uτ y
y+ = (2.8b)
ν
Closer to the wall on the linear sub-layer, the flow is practically laminar, and viscosity effects pre-
dominate, with an exponential evolution of the velocity profile until y + = 5. The buffer layer occurs at
5 < y+ < 30 − 50, where viscous and turbulent stresses have similar magnitudes, and then, turbulence
overshadows viscosity effects in the log-law region or wall layer. At around 15% of the boundary layer
displacement thickness – δ ∗ , at the end of the inner layer, a deviation from the low-law velocity profile
occurs, with the velocity tending to its freestream value U∞ at the end of the boundary layer (Brederode,
2018).
Due to these physical phenomena and variations in the inner boundary layer, some turbulence mod-
els are not valid in this region (k − for instance), as a result, semi-empirical wall-functions are used to
resolve this boundary layer portion. Using wall functions reduces the need for a very fine refinement at
the boundary layer since only one layer with y + ≈ 0.15δ ∗ is required to capture the whole inner layer,
thus saving computational resources (Figure 2.8). Flows characterized by high Reynolds numbers are
well solved using wall functions, and the y + target value to be chosen for the first layer depends on the
23
Figure 2.7: Velocity profile on a turbulent boundary layer. Adapted from Brederode (2018).
wall functions to be used: for standard wall functions, y + should be between 30 and 300, guaranteeing
that the first cell layer contains the whole inner layer (Ansys Inc., 2013).
If separation is bound to occur or in the LRN regime, turbulence models which are valid in the
viscosity-affected region should be employed. These require a much finer grid to properly resolve the
viscous sub-layer (Figure 2.8), having a first grid cell with y + value up to 5, as proven by simulations
done using the Spalart-Allmaras and k − ω models (Kalitzin, Medic, Iaccarino, & Durbin, 2005). When
transition models are used, the first layer must be fine enough to capture the laminar and transitional
boundary layers correctly, requiring a y + value lower than 1. Additionally, if laminar separation bubbles
are present, special care should be given to the streamwise width of the cells, to properly represent the
recirculating flow (Langtry et al., 2006; Menter et al., 2006).
Figure 2.8: Near-wall treatments. Retrieved from Ansys Inc. (2013).
2.4.3 Discretization of Equations
Hirsch defines three methods for discretization of the space derivatives: the Finite Difference Method
(FDM), the oldest and most traditional method, only applicable in structured grids; the Finite Element
Method (FEM), widely used in structural analysis and also applicable in CFD with a few modifications
(Hirsch, 2009); and the Finite Volume Method (FVM) which transforms the partial differential equations
representing the governing laws of physics into discrete algebraic equations over finite volumes. FVM
24
can be applied to unstructured grids and is the most widely used method in CFD software (Moukalled et
al., 2015). In the FVM, after the mesh is generated, a local control volume is associated with each mesh
point and the integral conservation law is applied to it. Using the Gauss Divergence theorem transforms
the control volume integrals into face fluxes which are further discretized according to the numerical
schemes chosen (Moukalled et al., 2015).
The definition of the control volume in unstructured meshes has two variants: cell-centered control
volumes (Figure 2.9a), with the nodes placed at the centroid of the control volumes, making the control
volumes the same as the grid cells; and vertex-centered control volumes (Figure 2.9b) in which the flow
variables are stored at the vertices of the grid cells and a ”new” control volume is created by linking
the centroids of the grid cells or using the centroids of the grid cell faces. Although the cell-centered
method is the most common (OpenFOAM, StarCCM+ and Fluent use this method), being second-order
accurate and requiring less storage space since the control volumes and the grid cells are identical, non-
conjunctional elements impact the accuracy of the results and non-orthogonality affects the robustness
of the mesh. The vertex-centered method is used in CFX and accurately solves the face fluxes while
having a lower order of accuracy. Moreover, control volumes near the grid boundaries have their centroid
inside the boundary, thus complicating the discretization of the face fluxes with the walls in those cells
and increasing the discretization error (Moukalled et al., 2015).
(a) Cell-Centered Control Volumes (b) Vertex-Centered Control Volumes
Figure 2.9: Control volume construction in 2D unstructured meshes. Retrieved from Versteeg and
Malalasekera (1995).
Numerical schemes are used to discretize both the temporal and spatial terms. Regarding the spatial
discretization schemes, Fluent always uses second-order central-differences for computing the diffusion
terms while convective terms are discretized using upwind schemes with the order of accuracy being
defined by the user. While first-order upwind schemes are robust, the face value is assumed to be
equal to the cell center value of the upstream cell, making it inaccurate at evaluating steep gradients. A
second-order upwind scheme is more accurate since it takes into consideration the cell center value and
the gradient of the upstream cell, nevertheless, this scheme may introduce undesirable non-physical
oscillations. For a transient/unsteady simulation, temporal discretization is necessary, and implicit or
explicit time integration is supported by Fluent with various orders of accuracy (Ansys Inc., 2013).
CFD methods can be classified into two categories according to the solver type, being density-based
and pressure-based. Density-based methods were traditionally applied to transonic and supersonic
25
cases, while pressure-based solvers were initially restricted to low Mach numbers, yet improvements
allowed their application to all flow regimes (Moukalled et al., 2015). Due to the flow regimes in which
the UAV will operate, the pressure-based solver will be used throughout the presented work.
The pressure-based solver can solve the flow problem in either a segregated or coupled manner.
Segregated or uncoupled algorithms have as their basis a Semi-Implicit Method for Pressure Linked
Equations (SIMPLE) in which the solution is found iteratively by generating pressure and velocity fields
that consecutively satisfy the momentum and continuity equations (Moukalled et al., 2015). By solv-
ing each equation separately and applying corrections to guarantee mass conservation, segregated
methods converge faster and require less computational power. Fluent allows the user to choose the
SIMPLE algorithm or other modifications that account for skewness and/or neighbor corrections. The
coupled method solves the continuity and momentum equation together, being more robust and accurate
than the segregated method while being computationally heavier. Compressible flow problems require
this method to present accurate and reliable results (Ansys Inc., 2013).
Boundary conditions must be described for the boundary elements which have faces on the boundary
of the domain. There are a variety of boundary conditions that can be applied and they depend on the
physics of the problem: for a fluid flow simulation over a surface, a velocity inlet upstream and pressure
outlet downstream are defined at the edges of the fluid domain, representing the far-field condition and
undisturbed flow; wall conditions are set where the fluid domain is in contact with a solid surface and
the no-slip condition is specified; and symmetry conditions can be applied in 3D simulation when the
physical geometry of interest presents a mirror symmetry (Ansys Inc., 2013).
26
Chapter 3
Conceptual Design – Performance

Considerations
The conceptual design is the initial design phase which encompasses considerations from different
areas of the aeronautical spectrum and gives a general overview of the aircraft itself. As stated at the
beginning of the dissertation, the conceptual design phase developed by the design team along with the
methodology followed and the final solution which spurred the current thesis is published and can be
consulted through reference (Alves et al., 2021).
Based on the design proposal and mission requirements, a main mission profile was defined in the
conceptual phase, illustrated in Figure 3.1. The UAV would take-off vertically (1-2) and begin its transition
to horizontal flight (2-3). After the VTOL system is turned off, the aircraft initiates its conventional climb
phase (3-5), divided into two segments, and starts the cruise/loiter mission (5-8) upon reaching an
altitude of 5000 f t. At the end of the cruise segment, the aircraft glides (8-9) and performs a landing
circuit (9-10), after which the VTOL propulsion system is activated again to perform the vertical descent
and landing (10-11). The vertical climb and descent phases are powered by batteries dedicated to these
VTOL systems, while the forward flight segments are powered by the hydrogen fuel cell, with the aid of
an additional battery to support the required power for climb and other peak power requirements. A
secondary mission is also defined with take-off at a high altitude (6000 f t) and a cruise altitude of 7500
f t, with a smaller total flight time (Alves et al., 2021).
Figure 3.1: UAV mission profile. Adapted from Alves et al. (2021).
27
From a performance and aerodynamics point of view, flight mechanics equations were used in the
conceptual phase where the aerodynamic parameters, CD0 and K, are applied. These two parameters
can be initially estimated based on similar aircraft (market study) or some semi-empirical formulations.
3.1 Initial Drag Estimation
For an initial estimation based on commercial and general aviation, Corke suggests a value for CD0
ranging from 0.01 to 0.02 (Corke, 2003). From the market study conducted, information about the drag
coefficient is not widely available on the technical sheets consulted as this information is too detailed
and can be seen as classified. As such, the author based its estimates on a fixed-wing propeller UAVs
database created by Coelho (2019) in his dissertation. An adaptation of Coelho’s database was created
– Table 3.1, narrowing the thirty-eight UAVs into those with a design similar to the one proposed in
the initial design phase (Boom Mounted A-Tail and After Fuselage Mounted Engine) and with a MTOM
(Maximum Take-Off Mass) of the same order of magnitude.
Table 3.1: Database of UAV with similar designs as the one projected. Adapted from Coelho (2019).
UAV M T OM [kg] Ucr [m/s] Sw [m2 ] AR CD0 CHT

Aerosonde 4 15 25.72 0.567 14.75 0.014 1.23
Ogassa 36 25.72 1.06 16.64 0.022 1.02
PD-1 40 26.4 1.17 13.68 0.024 1.16
Penguin B 21.5 22 0.79 13.78 0.031 0.90
It is relevant to mention that none of the aircraft listed on the database have VTOL capabilities.
Vertical flight motors and their respective propellers introduce additional drag and its contribution is
relevant. Wind tunnel testing of a VTOL system, similar to the one idealized for the project, in forward
flight conditions showed that the parasite drag from the vertical propellers is considerable and its relative
position to the freestream has a big impact, increasing drag up to five times (for U = 25m/s) when the
propeller is perpendicular to the airflow when compared with the optimal condition in which the propeller
is parallel to the booms (Wang & Chan, 2017). These results demonstrated the importance of controlling
the propeller’s stopping position after its operation on the vertical flight segments.
From the values presented, the range proposed by Corke is too optimistic if applied to the design in
question and even more when considering the additional drag stated above, as such CD0 is considered
to be 0.04 as an initial estimation.
The induced drag factor on the drag polar depends on the wing aspect ratio and the Oswald efficiency
factor as stated in Equation (2.2). In an attempt to narrow the range of values given to the Oswald factor,
Raymer (1992) suggests two empirical expressions for computing the Oswald efficiency factor: one of
them should only be used for straight-wing aircraft and aspect ratios lower than 25, and the other for
swept wings with a leading-edge sweep angle higher than 30◦ . Other authors such as McCormick,
DATCOM and Jenkinson use additional parameters and diagrams to give better estimates, however, this
process is time-consuming and some of the data is still unknown at this stage of the design process (Niţă
28
& Scholz, 2012). Howe’s proposed equation - Equation (3.1), strikes a good balance between accuracy
and input data required and takes into account the taper ratio of the wing – λ, sweep angle at 25% of the
chord – ϕ0.25c , the airfoil thickness to chord ratio – t/c, flight Mach number, and the number of engines
over the wing – Ne . This approach delivers reliable results with deviations of 10% despite having the
influence of the taper ratio decoupled with the sweep angle (Niţă & Scholz, 2012).
1
e= (3.1)
0.142+0.005·[1+1.5·(λ−0.6)2 ]·AR·(10·t/c)0.33 0.1·(3·Ne +1)
(1 + 0.12(M a)2 ) 1+ cos(ϕ0.25c )2 + (4+AR)0.8
At this stage, wing geometry parameters are yet to be defined. Nevertheless, at first glance, when
using typical values for these variables, the Oswald efficiency values obtained using Equation (3.1) are
in the range presented in subsection 2.1, as such, e = 0.75 is considered as an initial approximation.
3.2 Flight Speed Adjustments
As discussed in the conceptual phase of the project, the design point had little margin for error (Alves
et al., 2021), with the possibility that poor initial estimations or small deviations could compromise the
project since the power required for level flight cannot surpass the fuel cell’s maximum power output,
which for the fuel cell considered is 800 W . Hence, all the initial considerations were revised and a new
iteration was done with more accuracy and inputs from the different specific areas.
From the performance point of view, the initial flight speeds considered for level flight were purely
based on the range of values given on the mission proposal without much care if they were the most
efficient ones. To lower the power demand during cruise, two solutions rose based on the Power (P ) to
Weight (W ) ratio, computed using Equation (3.2), with ηpr being the propeller efficiency (Raymer, 1992).
" #
ρU 3 CD0 2K W

P 1 S
= W
+ (3.2)
W min ηpr 2 S ρU
• Reduce flight speed: with the remaining parameters fixed, the higher the flight speed, the higher
the power required (for the range of speeds in the design proposal). This evidence is supported
by the power curve plot (Figure 3.2) in which the minimum flight speed of the design proposal (35
3/2
kts) is higher than the speed of minimum power for flying ((CL /CD )max );
• Increase wing loading: for the range of speeds considered initially and with the remaining param-
eters set, the minimum power loading for level flight is achieved for higher wing loading values.
With these tendencies in mind, the objective would be to apply both solutions simultaneously and
check if the final results were better than the initial ones. Reducing the flight speed is easily done since
the initial values were closer to the upper bound of the flight speed interval given in the design proposal,
nonetheless, for safety and operational purposes, the minimum flight speed should be at least 1.2 to 1.3
times higher than the stall one (Gundlach, 2012).
The maximum wing loading is set according to the constraints described in the design point of the
conceptual design (Alves et al., 2021), most notably the stall speed and the maximum ceiling. The stall
29
condition, given by Equation (3.3), was initially the critical condition, so either the stall speed or the
maximum lift coefficient - (CLmax ) must rise to increase the wing loading value. A substantial increase
in the maximum lift coefficient can only be obtained using high-lift devices such as flaps (Sadrey, 2013),
nevertheless, this solution would not change the wing loading for level flight conditions when the high lift
devices are not in use. The strategy is then to increase the stall speed of the UAV.
The wing loading is not exclusively given by the stall condition: if the flight speed decreases, the
critical condition for the wing loading might become the flight ceiling – Equation (3.4). To tackle this
problem, the maximum ceiling condition is discarded, and after the analysis is done, one would check
what the new maximum ceiling is and if it is feasible for the high altitude mission mentioned.
r
W 1 W 1 2 CD0
= ρUs2 CLmax (3.3) = ρceiling Ucr (3.4)
S stall 2 S ceiling 2 K
Figure 3.2 represents the power curve, given by Equation (3.2) when considering a stall speed of
28 kts. From the vertical lines, it is possible to check that the UAV cannot operate at its maximum
aerodynamic efficiency flight speed – (CL /CD )max . Therefore, there is no reason to consider different
speeds for the cruise and the main mission segments. This graph has the same speed relationship
when the stall speed is changed between 25 and 32 kts which is around the desired values.
Figure 3.2: Power curve with important flight speeds.
Since the goal of this analysis was to decrease the power at level flight, an optimization study was
conducted with the stall speed and cruise speed as design variables to minimize the power. The op-
timization problem is beyond the scope of the presented thesis and is described in detail in reference
(Alves, 2021). From this study, the stall speed was set to 28 kts and the level flight to 38 kts, the latter
being increased compared with the results obtained to give an extra margin for operation. With these
changes, a reduction in the fuel cell power output to 670 W is achieved, well below its maximum. With
the wing loading set by the stall speed constraint and the flight speed defined, the maximum ceiling is
inferred to be 10000 f t, above the maximum altitude set for any of the two mission profiles.
30
Chapter 4
Preliminary Design
With the conceptual design finished, the preliminary design of the UAV is conducted, focused on the
aerodynamics field. This design phase includes the wing and tail design as well as a more accurate
estimation of the total UAV drag, to better predict the aircraft performance and serve as input for the
propulsion system study. At the end of the preliminary design phase, the work developed in the struc-
tural and propulsion areas by the other members of the project is put together with the aerodynamic
considerations, and an update on the UAV characteristics is done.
4.1 Wing Design
The wing should be designed in such a way that it can provide lift throughout all the flight regimes
described in the mission profile while producing minimum drag. From the enhanced conceptual design
phase with the motors already chosen, the wing surface area and its aspect ratio is set. The initial design
phase parameters relevant for the aerodynamic study are summed up in Table 4.1.
Table 4.1: Aerodynamic related inputs from conceptual design. Adapted from Alves et al. (2021).
Parameter Value
Wing Area 1.372 m2
Aspect Ratio 11.66
MTOM 21.61 kg
Level Flight Speed 38 kts
Level flight Altitude 5000 f t
For the desired flight speed and altitude defined for level flight, and based on the results achieved
in the initial phase (wingspan, wing area, and aspect ratio), one could calculate the expected Mach
number and Reynolds number. At cruise, Mach number is 0.056, which is considered incompressible
flow. Reynolds number requires a characteristic length to be computed, which for a wing corresponds to
the mean chord c.
31
4.1.1 Wing Planform
Considering the characteristics provided in subsection 2.2.2, the wing has different geometric pa-
rameters which influence its performance. With the double boom configuration chosen and bearing in
mind the construction methods which are thought to be used, the wing can be divided into three panels:
the middle one, which goes from one boom to the other having its midsection coincident with the aircraft
symmetry plane, and the tip panels after the booms. For simplicity and structural rigidity, the mid panel
has a constant chord and no sweep (straight-rectangular shape). The exact span location of the booms
is still unknown and it is dependent on the tail geometry and its span. As an initial hypothesis, the
rectangular central panel is initially assumed to have 0.75 m of semi-span yB . If the booms are located
further away from the symmetry plane than the assumed position, this distance has to be rethought and
thus, the wing geometry redone.
To the tip panels, taper is applied to reduce the induced drag. Although the recommended taper ratio
is around 0.4 to 0.45, as stated in subsection 2.2.2, for an initial geometry, taper ratio is set at 0.55 to
avoid having a tip chord too small and prevent tip stall problems and structural difficulties (Sadrey, 2013).
Leading edge sweep is set to 2◦ for aesthetic purposes, and neither dihedral nor twist are considered at
this stage.
The total wingspan is easily obtained and it does not depend on the multi-panel configuration chosen.
Basic geometric expressions to compute the root chord – cr and the tip chord – ct become more complex
so that the different properties of each panel are thought out. Table 4.2 sums up the expressions used,
which were adapted from Cook (2013) for a cranked wing (two panels along the semi span of the wing,
each with its own taper; λi representing the taper ratio of the interior panel).
Table 4.2: First iteration wing dimensions.
Property Formulation Value [m]

√
Wingspan b = AR · S 4.000
2·S
Root chord cr = 2·yB ·(1−λ)+b·(λ+λi ) 0.399
Tip chord c =c ·λ
t r
h i
0.220
yB
(λ2i +λi λ+λ2 )+ (b/2)−y (1+λi +λ2i )
2
Mean chord c= 3 cr
h
yB
Bi 0.353
(λi +λ)+ (b/2)−y (1+λi )
B
Figure 4.1 represents the initial geometry considered for the wing, with the leading edge being on the
top part of the figure. With the mean aerodynamic chord computed and considering the Standard Atmo-
sphere model with an offset of +20 ◦ C (National Oceanic Atmospheric Administration, 1976), Reynolds
is around 3.7 × 105 for cruise conditions.
Figure 4.1: Wing Baseline Geometry.
32
4.1.2 Airfoil Selection
The airfoil section is responsible for generating the optimal pressure distribution on the upper and
lower wing surfaces to generate the required lift without producing excessive drag. The airfoil can be
designed according to the needs or chosen from the available sources. Designing an airfoil is a complex
and time-consuming process based on optimization techniques coupled with CFD aerodynamic software
packages. Additionally, the airfoil design is a design project in itself, and it is just a step in the whole
aircraft design process (Sadrey, 2013). For these reasons, the author opted for the second option:
choosing an existing airfoil.
The University of Illinois at Urbana-Champaign (UIUC) Aerodynamics Research Laboratory pro-
duced a database (UIUC Department of Aerospace Engineering, 2020) with over 1600 airfoils to choose
from among Clark, Eppler, Wortmann, Gottingen, Martin Hepperle, NACA, Selig, Selig-Ashok, Selig-
Donovan, Selig-Giguere, and others.
To start the selection process, the lift coefficient at cruise must be computed. For level flight, lift is
equal to the aircraft’s weight, and CL is given by Equation (4.1). Most aircraft use fossil fuel as their
energy source and due to fuel consumption, they become lighter throughout the flight, reflected in a
decrease of the cruise lift coefficient, therefore, an average value for CL is usually assumed at this
stage. Because the weight of hydrogen is negligible when compared to the total aircraft weight, the lift
coefficient for cruise flight is assumed to be constant and equal to CLdesign = 0.821.
2W
CL = 2S
(4.1)
ρcr Vcr
The value obtained for the cruise lift coefficient might seem high when compared with the typical
ones for commercial aircraft, however, gliders and small UAVs tend to have high lift coefficients due to
their relatively low cruise speed. Recalling the data from table 3.1, the previous statement is confirmed,
as CLcr for those UAV ranges between 0.78 and 1.12. In light of this, the airfoil choice was narrowed to
airfoils with high lift-to-drag ratios or known to have good performance at low Reynolds numbers.
XFOIL tool from XFLR5 v6.48 was used to analyze the airfoils. A 2D direct analysis was performed
for a Reynolds number of 3.7 × 105 , Mach number equal to 0.056 and eN with Ncrit = 9 as the transition
criteria. All airfoils are normalized using 250 panels and the study was done with the angle of attack
(AoA represented by α) varying between -15◦ and 25◦ . This 2D analysis revealed that around 20 airfoils
3/2
were suitable candidates, having a Clmax over 1.35, good Cl /Cd and Cl /Cd values for the wing airfoil
ideal lift coefficient – Cli and not an exaggerated Cd0 . The wing airfoil ideal lift coefficient is a 2D
approximation of the 3D cruise lift coefficient obtained through Equation (4.2), in which the contribution
to the generation of lift from all surfaces besides the wing is removed and the finite wing consideration
when going from 3D to 2D is taken into account.
CLcr
Cli = (4.2)
0.95 · 0.9
Since the wing initial geometry is already defined, 3D analysis can be conducted and their results
can be used for the final choice. A fixed lift analysis was chosen, which requires 2D polars for a wide
33
range of Reynolds numbers. The wing was discretized into panels with a cosine distribution and the
solution was obtained using the VLM considering ring vortices.
From the 20 airfoils tested in the 3D analysis, those who presented a CLmax under 1.3 were excluded
since this was the value considered for the stall condition at the conceptual design phase. For the
remaining, the decision criteria are based on the aerodynamic efficiency and the maximization of the
3/2
endurance criteria for cruise, CL /CD . Table 4.3 and 4.4 list the top five airfoils, along with their 2D and
3D properties, respectively.
Table 4.3: 2D Airfoils results from XFOIL.
Airfoil Clark Y N-22 SD7062 SG6042 SG6043

Clmax 1.42 1.55 1.62 1.51 1.66
Clα [/◦ ] 0.0840 0.0906 0.0818 0.0744 0.0769
α0l [◦ ] -3.7 -5.9 -4.2 -4.5 -6.2
αstall [◦ ] 13.2 11.2 15.6 15.8 15.4
Cd0 0.0119 0.0122 0.0127 0.0156 -
Cdmin 0.0074 0.0077 0.0088 0.0073 0.0079
(t/c)max [%] 11.71 12.39 14.00 10.00 10.02
x(t/c)max [%] 28.42 27.62 27.03 34.43 32.13
Table 4.4: 3D Wing results from XFLR5, for each airfoil.
Airfoil Clark Y N-22 SD7062 SG6042 SG6043

CLmax 1.35 1.48 1.58 1.46 1.62
αStall [◦ ] 12.0 10.6 14.0 11.8 11.4
CD0 0.0078 0.0080 0.0086 0.0081 0.0121
For cruise conditions - CL = 0.821
αtrim [◦ ] 5.86 3.60 5.20 4.37 2.13
CD 0.0277 0.0282 0.0285 0.0263 0.0271
CL /CD 29.65 29.16 28.79 31.18 30.28
3/2
CL /CD 26.86 26.42 26.09 28.25 27.44
From the analysis results, the SG6042 airfoil was chosen (Figure 4.2), proving to have good aero-
dynamic efficiency at cruise while presenting a low base drag coefficient. Without incorporating the tail,
the wing should be positioned with an incidence of 4.5◦ to provide the necessary lift at cruise conditions.
Figure 4.2: SG6042 airfoil. Adapted from Selig (1997).
34
4.2 Tail Design
After defining the wing initial design, the tail should be sized. The selected design for the tail is the
twin-boom inverted V, supported by the literature background provided in subsection 2.2.3. The chosen
layout does not have the traditional horizontal and vertical stabilizer as two different surfaces, therefore
the sizing procedure is done to an equivalent horizontal and vertical tail area. Literature related to the
design of an inverted V tail is scarce, consequently, the author assumes that the design steps and
considerations that should be foreseen are similar to those of a single boom V-tail. For this tail layout,
the dihedral angle (Γ) is the angle between the stabilizer and the horizontal plane – Equation (4.3).
Additionally, the tail arm length (lHT ), which corresponds to the distance between the tail aerodynamic
center and the wing aerodynamic center, is the same for both the equivalent vertical and horizontal tail
(Gundlach, 2012).

SV T
ΓV −tail = arctan (4.3)
SHT
The initial sizing of the equivalent horizontal tail is based on the horizontal tail volume coefficient
(CHT ), which relates parameters of the horizontal tail with the wing (Sadrey, 2013).
SHT · lHT
CHT = (4.4)
S · cw
Empirical values are usually used as an initial estimate for the tail volume coefficient, however, the
literature gives no information on typical values for UAVs. The database from Coelho (Coelho, 2019)
contains information about this coefficient, hence, it is considered as a reference. Analyzing the exhibited
values, they tend to be above the ordinary ones generally used, as such, the horizontal tail coefficient is
set to 0.8, lower than the ones presented.
The tail arm length, is another unknown at this point. This parameter influences the dimension of
the booms and the tail area. Sadrey (Sadrey, 2013) proposes the calculation of the optimum tail arm to
minimize the wet area and, consequently, drag. For this approach, the wet area of the booms should
be given as a function of the tail arm length. Without the boom geometry defined, specifically the boom
cross-section, this method becomes impracticable.
A simpler solution proposed by Corke (2003) is to give a reasonable value for the tail arm length and
infer the tail surface area from it. To choose a proper value for the tail arm length, a ratio between this

lT
variable and the UAV total length was considered – ltotal . From the database and analysis made by
Coelho (2019), this ratio is between 0.4 and 0.6 for UAVs with an aft fuselage mounted engine (pusher
configuration). Looking from a lateral perspective on a twin-boom aircraft, its total length can be divided
into segments according to Equation (4.5):
ltotal = lf uselage − (0.75 · cW + lT E to f us−end ) + lT + 0.75 · cT (4.5)
An embryonic design of the main fuselage was conducted, considering all the components the main
fuselage has to carry and their allocation. The details of this design are depicted in the structural analysis
35
oriented dissertation, yielding a total length for the main fuselage of 1.4 m (Sá, 2021).
(lT E to f us−end ) is the distance between the trailing edge of the wing and the rear of the main fuse-
lage. On a twin-boom configuration, the wing is located close to the rear of the main fuselage, therefore,
its length is much smaller compared with the remainder of the fuselage. At this stage, the tail mean
chord is yet to be determined, however, this value will be lower than the mean chord of the wing. With
these two variables still unknown, they are ignored as a first approximation, and the tail arm is fixed at
1.5m.
Applying Equation 4.4 gives an equivalent horizontal tail area of 0.258m2 . Equation 4.3 defines the
dihedral angle, which is set to 45◦ , similar to the UAVs presented in the database. This dihedral angle
defines an equivalent vertical tail area equal to the horizontal one.
p
2
At first glance, the theoretical area for the complete tail is ST = SHT + SV2 T due to simple geometric
analysis, making a V-tail area 70.7% of a conventional one. Nonetheless, according to Pursee and
Campbell (1944); Stettmaier (1999), the tail has to be larger to have the same efficiency, as such this
area reduction does not exist and, consequently, ST = SHT + SV T .
From the market study conducted for this tail configuration, it was concluded that each tail half has
an aspect ratio of 4 which is in agreement with the bibliography for conventional tails. Moreover, neither
sweep nor taper are usually applied on inverted V tails. Considering this aspect ratio for each half, the
tail has a projected aspect ratio of 5.66 when viewed from above, resulting in a projected semi-span
of 0.716 m and thus making the wing geometry initially considered feasible. Since each tail panel is
rectangular, the tail chord is constant yielding cT = 0.255 m.
The airfoil selection for the tail has different criteria than the wing. Concerning the tail, the airfoil
should be symmetrical to either create positive or negative lift and have a low base drag (Sadrey, 2013).
Because the tail airfoil should have a lower thickness to chord ratio than the wing, NACA 0008 is selected.
At this stage, the lift force produced by the tail to trim the aircraft cannot be computed since the location
of the CG is yet unknown.
4.3 Drag Build-Up
The total aircraft drag is the sum of each component contribution. Every component exposed to the
airflow produces parasite drag, and the wing is the main generator of lift induced drag (Anderson, 2012).
4.3.1 Lift Producing Surfaces
Regarding the wing, both drag components can be calculated using empirical formulations. For the
induced drag components, Howe’s approximation for the Oswald efficiency factor – Equation (3.1) can
be applied for the wing geometry defined, yielding e = 0.7435, which is very close to the initial estimate,
making CDi = 0.0247 for cruise. With the tail having no incidence, no lift is expected to be produced
and, therefore, no contribution to the induced drag coefficient.
For the parasite drag, Hoerner proposes using Equation (4.6) which makes an analogy with the flat
36
plate. Swet represents the area of the component exposed to the airflow, Q the interference factor,
Cf the flat plate average skin coefficient and F F the form factor. The interference factor accounts for
the interaction between two surfaces which increases the boundary layer thickness near the interaction
region, causing additional drag. The skin friction coefficient is highly dependent on the boundary layer
structure, whether it is laminar or turbulent. Form Factor makes the flat plate correction to the actual
geometry and takes into account the pressure drag caused by flow separation (Raymer, 1992).
Swet
CD0 = · Q · Cf · F F (4.6)
Sref
According to Roskam and Lan (2016), Qwing = 1 for a high-wing, and Qtail = 1.08 for a double boom
tail configuration. Concerning the skin friction coefficient, it is influenced by the location of the transition
from laminar to turbulent flow. The transition point depends on the angle of attack, which varies during
flight. Raymer (1992) suggests assuming the transition to occur near the maximum thickness point of
the airfoil. With the transition point defined, Equation (4.7) can be applied to give a better estimate of the
skin friction coefficient. The partial skin friction coefficients in this formula are calculated using the wing
characteristic Reynolds number times the percentage of laminar flow. For laminar flow, Blasius solution
is applied and for turbulent flow, the author opts to choose Schlichting formulation for the skin-friction
parameter. Since XFLR5 provides information on the transition location for the 3D wing, this information
will be considered for a better estimate. At an incidence of 4.5◦ , transition occurs at around 65% of the
chord on the upper surface, much later than Raymer suggests. XFLR5 predicts fully laminar flow in the
tail, with transition occurring very close to the trailing edge at high tail AoA, implying that 90% of the tail
is always in the laminar regime.
Cf = Cf (100%T urb) − (Cf (%P artialT urb) · %Lam) + (Cf (%P artialLam) · %Lam) (4.7)
Several authors suggest different estimations for the form factor. Hoerner (1965) introduces a simple
empirical expression – Equation (4.8) – as a function of the wing thickness-to-chord ratio, valid for
incompressible flow.
F F = 1 + 2 (t/c) + 60 (t/c)4 (4.8)
Modeling the wing and the tail on OpenVSP and considering the empirical expression displayed
above, yields CD0 wing = 0.00785 and CD0 tail = 0.00265. The results obtained in OpenVSP are similar
to the ones obtained using XFLR5, with differences below 5%.
4.3.2 Fuselage and other Structures
Equation (4.6) is also valid for slender bodies, thus, it can be applied to the main central body and
the tail booms. For these bodies, the Q = 1. Skin friction and the wet area are computed by dividing
the surface into segments of constant length and considering each segment Reynolds number, which
uses its position from the start of the fuselage as the characteristic length. Without information on the
37
transition location, the boundary layer is assumed fully turbulent as the worst-case scenario.
The form factor is given as a function of the fineness ratio, corresponding to the length to diameter
ratio of the body. Following the same reference, Hoerner (1965) proposes Equation (4.9) to calculate
the form factor of streamlined bodies.
1.5 7
FF = 1 + 1.5
+ (4.9)
(l/d) (l/d)3
Similar to the lifting surfaces, the fuselage and the booms were added to the model on OpenVSP,
resulting in a CD0 booms = 0.00181 for both booms and CD0 body = 0.00336.
The drag build-up must also account for other structures in contact with the external airflow, such as
the landing gear (struts and wheels) and the daylight camera. The contribution of these components
is estimated at the early stages by using the characteristic drag coefficient based on the component’s
shape and form, and its frontal surface area Sf – Equation (4.10). In Cdcomponent the lower case subscript
d does not stand for 2D coefficients, instead, it is related with the data gathered from the literature,
which is the result of experimental testing. Because the experimental results do not distinguish between
parasite and induced drag (Hoerner, 1965), the nomenclature used with these empirical coefficients
does not include the subscript 0 . This change is used in order to distinguish it from CD which already
accounts for the surface area of the component as seen on Equation (4.10).
Sfcomponent
CDcomponent = Cdcomponent (4.10)
Sref
The landing gear has a tricycle gear configuration in which the main gear is located after the CG,
supporting most of the aircraft’s weight. The nose gear is composed of a ”double” wheel and a single
strut, while the main gear has two wheels equidistant to the symmetry plane and an arc-shaped carbon
fiber strut (Sadrey, 2013). At this stage of the design phase, the dimensions of the landing gear are still
unknown and hard to get, as such, UAVs at CIAFA with a similar MTOM serve as a reference for both
landing gear (strut and wheels): the nose gear is based on Alpha Extended 1 – Figure 4.3a – while the
main gear is from a real manufactured piece designed for a 30 kg UAV – Figure 4.3b. For non fairing
struts, Cdstrut = 1 while for wheels, Cdwheel = 0.3 (Hoerner, 1965; Sadrey, 2013). With the frontal area
measured, the nose gear strut and its ”double” wheel yields CDN oseGear = 0.00256, and the main gear
CDM ainGear = 0.00296
The camera has a belly pan-tilt integration on the fuselage, making the optical camera sphere fully
in contact with the airflow, therefore the external surface can be approximated as a sphere and its
parasite drag contribution can be estimated by Equation (4.10) (Gundlach, 2012). Considering the
TASE 150 model as reference, with a diameter of 0.112 m (UTC Aerospace Technology, 2015), and a
Cdsphere = 0.41 for a sphere under Reynolds number below 4 × 105 , the camera drag coefficient yields
CDCamera = 0.00294.
38
(a) Nose Gear (b) Main Gear
Figure 4.3: Nose and main gear of similar MTOM aircraft in CIAFA.
4.3.3 VTOL Propulsion System
As stated in section 3.1, the VTOL propulsion system has a big impact on the total aircraft drag at
cruise. For the drag build-up process, only the components in contact with the freestream are important,
namely the motor and the rotor. For the sake of clarity, the author uses the word ”rotor” to mention the
propeller which provides vertical thrust, with the term ”propeller” used solely for the one which provides
forward thrust, like on a conventional propeller-driven airplane.
Empirical expressions for the motor consider a blunt body and for the propeller an equivalent flat
plate. These expressions failed to accurately predict the parasitic drag on a similar Lift+Cruise UAV
design project, as such they should be used with care and validated with experimental data (Wang &
Chan, 2017).
In a previous design project done by the AFA, Mendes (Mendes, 2021) conducted a wind tunnel
experiment to obtain the drag of a P-60 motor by T-Motor with a 22x6.6 rotor attached to it, subjected
to atmospheric conditions with a freestream flow velocity of 20 m/s. Table 4.5 summarizes the obtained
results. It is important to note that the data is exclusively related to the motor and rotor since the scale
bench was calibrated to not include the drag of the structure supporting the components. Furthermore,
the rotor remained stationary during the whole experiment.
Table 4.5: wind tunnel data obtained by Mendes. Adapted from Mendes (2021).
Single motor and rotor Rear and forward motor and rotor
Drag with rotor parallel to boom 0.808 N 1.30 N
Drag with rotor perpendicular to boom 1.52 N 2.54 N
Looking at the results, the drag created by these components can almost double depending on the
stopping position of the rotor, reinforcing the necessity and the advantage of creating a mechanism to
stop the rotor when they are parallel to the boom if it does not create more drag than the worst-case
scenario. Although the tested conditions are not the same as the cruise conditions of this project, neither
are the components tested, they are very similar. Additionally, the experiment with both rear and forward
VTOL systems installed has an increase by around 65% in the drag force when compared to each
individual contribution, which is less than two sets separated, indicating that the wake of the front VTOL
39
system interacts with the rear components, reducing the drag produced by the latter.
Without additional data from wind tunnel tests with different components and freestream conditions,
these results cannot be used and scale them to the project requirements. Nevertheless, the author can
compare them with empirical expressions and subsequently apply these expressions to the project’s
conditions and components.
Recalling Equation (4.10), Hoerner defines the Cdmotor = 1.2 for cylindrical bodies without any fairing,
which is the case of the motor. The frontal area corresponds to the length times the diameter of the motor
(Hoerner, 1965).
Regarding the rotor, Hoerner (1965) considers Cdpropeller = 0.1 + cos2 β0.7r , where β0.7r is the angle
between the chord at 0.7 of the blade radius and the rotation plane. Since this expression is applied
to stopped propellers providing forward thrust, β0.7r = 0◦ corresponds to the blades with their surface
almost fully perpendicular to the incoming free stream, thus creating more drag as predicted by the
previous expression. When the rotating plane is horizontal, which is the case for the rotor on a fully VTOL
system at hover, β0.7r = 90◦ would have the same interaction between the blades and the freestream
as the previous case. To keep the same definition for β0.7r , the expression applied to rotors should be
changed to Cdrotor = 0.1 + sin2 β0.7r . The surface area to be considered is cumbersome to compute due
to the shape of the rotor blades, as such a simplification is used which multiplies the rotor diameter with
the chord at 0.7 of the blade radius.
Using the geometry data from the motor and rotor from the supplier (T-Motor, 2021) of the compo-
nents tested by Mendes (2021), and calculating the drag force using Equation (4.10) yields a DM otor =
0.6796N and DRotor = 0.7852N . When summing up these contributions and not considering any in-
terference of these components with the structure, the drag produced by a single VTOL system set is
1.465N . Compared to the case where a single rotor is perpendicular to the boom, the empirical expres-
sions are off by less than 5% which validates the use of empirical expressions to get, at least, an order of
magnitude of the parasitic drag produced by the VTOL system. The ideal stopping position of the rotors
cannot be computed using the previously mentioned expressions since a conventional propeller never
has the same relative position to the freestream as this position for the rotor. A flat plate analogy to the
rotor at this position was discarded as it heavily underestimated the drag produced by this component.
(a) V605 KV210 VTOL motor (b) V22x7.4R VTOL rotor
Figure 4.4: VTOL propulsion system components. Retrieved from T-Motor (2021).
40
For the V605 motor to be used (Figure 4.4a), T-Motor (2021) provides the dimensions of the motor in
question, yielding a CDM otor = 0.00217. The rotor – Figure 4.4b – has a 16◦ for β0.7r , resulting in Cdrotor =
0.17598. With a chord of 41.2mm at 70% of the radius and knowing the rotor diameter, CDRotor =
0.00295 when it is perpendicular to the freestream. A single VTOL set with the rotor perpendicular to
the freestream presents CDSingleV T OL = 0.00512. To assess the drag of both front and rear VTOL sets
and for both booms, the percentages obtained from the experimental data mentioned at the beginning
of the subsection are applied, yielding a CDperpendicularV T OL = 0.01690 as the worst case scenario and
approximately half as the best case scenario, with CDparallelV T OL = 0.00845.
4.4 Preliminary Characteristics of the UAV
With the preliminary phase finished, the author’s aerodynamics considerations are put together with
the remaining areas covered by the design team and the UAV general characteristics are updated. Since
a lot of thought was put into the initial conceptual phase and a conservative approach was taken when
faced with uncertainties, there were no noteworthy changes after the preliminary design phase, which
validates the concept created. In each respective area, Sá (2021) developed the UAV V-n diagram,
conducted an in-depth weight analysis with all the components to be used, calculating the CG, as well
as estimates for each structural element, taking into account the the properties of the materials to be
used in a future construction phase. Additionally, an initial wing structure was thought out for the wing
described in section 4.1. Regarding the propulsion systems, P. Silva (2021) justified the choice of the
V605 motor with the 22x7.4 T-Motor rotor to be used in the VTOL system based on the supplier’s data
as well as the forward motor with possible propellers to be applied.
4.4.1 Static Stability Analysis
With the CG location defined and an update to the total weight of the UAV, XFLR5 is used to conduct
a simple longitudinal static stability analysis. Because of the increased MTOM, CLcruise = 0.827 and the
wing incidence must be increased to 4.9◦ .
With the wing and tail implemented in XFLR5, the analysis showed that the moment around the
CG is small at cruise conditions, easily nullified by a small upward deflection of both ruddervators.
The static margin provided by XFLR5 is around 50% , which is considered excessive even for low-
maneuvrable aircraft. This effect is due to having a ”heavy tail” design and a high initial tail volume
coefficient. Nevertheless, the value considered was slightly lower than the ones provided in Table 3.1.
While a stability analysis is not the primary focus of the current dissertation, it is important to guarantee
that the designed aircraft is stable and not too rigid and hard to maneuver.
In an attempt to understand the incongruities between the results obtained and the similar UAVs
over which the project is based, Ogassa and Penguin-B from Table 3.1 were modeled in XFLR5 with the
information available (UAV Factory, 2021; UAVision, 2020, 2021). The results showed that for a similar
modeling approach as the one used for the tail design, the static margin of these UAVs was also over
41
50%. Since these UAVs are real-world and operable aircraft, the author considers that following the
trends and properties of these UAVs outweighs the results obtained from the low fidelity software (re-
garding the static margin) and no reduction to the tail arm nor to the tail size is considered. Additionally,
a higher static margin than usual allows more flexibility for the location of the CG, and since the static
margin is a percentage of the wing’s mean chord, this CG location range is translated into just a few
centimeters.
Another basic concern related to the tail and longitudinal control surface is the ability to trim the
aircraft in cruise conditions and in more aggressive attitudes such as stall. The ruddervators occupy
around 60% of each half-tail span, and the hinge is located at 80% of the tail chord. At stall conditions,
trim is achieved by deflecting both ruddervators upwards by 8◦ .
4.4.2 Performance Update
From the drag build-up procedure in section 4.3, the UAV yields a parasitic drag coefficient between
0.03258 and 0.04103 depending on the stopping position of the rotors, which corroborates the good
initial estimate of 0.04 given at the early stages of the design phase. Recalling the information featured
in subsection 4.3.1 regarding the induced drag component, e = 0.7435 when considering the wing
structure defined.
With the previous two parameters computed and assuming a drag polar in the form of Equation
(2.2), the thrust – T and power required for cruise and climb are given by Equation (4.11) and Equation
(4.12), in which RoC represents the rate of climb of the aircraft (Raymer, 1992). The obtained values are
summarized in Table 4.6. It is important to note that the power required during cruise, which includes the
required shaft power, efficiency of the forward motor and ESC (Electronic Speed Controller), ηmotor and
ηESC , and power consumption by the payload and primary systems (PP P S = 80 W), is still below the 800
W fuel cell maximum output which does not jeopardize the entire project. Additionally, the propulsion
system study done by Silva verified that the forward motor and propeller chosen can output the required
thrust for the conventional climb phase at an acceptable range of RPM (rotations per minute) and below
the motor’s 100% throttle (P. Silva, 2021).
RoC W ρU 2 SCD0 2KW 2 TU

T = + + (4.11) P = (4.12)
U 2 ρU 2 S ηpr
From the conceptual design phase, it was determined that around 70% of the total hydrogen in the
tank is available to be used during cruise – (mH2 cr /mH2 tot ), with the remaining being consumed during
climb (11.9%), descent and landing circuit (6.2%), battery recharge (6.8%), and a small percentage kept
as reserve (5%). Additionally, for the flight speeds defined for the VTOL segments, climb, descent and
landing circuit phase, they sum up to a total of 50 min of flight time, with the cruise segment varying
according to the hydrogen mass available and power requirement. The cruise flight time, in hours, is
obtain through Equation 4.13 in which ηF C is the fuel cell efficiency at cruise (Raymer, 1992).
42
ηF C ∗
tcr = Pcr
· EH2 · mH2 tot · (mH2 cr /mH2 tot ) (4.13)
ηmotor ηESC + PP P S
mH2 tot = 0.1485 kg ηmotor = 0.90 ηESC = 0.95 (P. Silva, 2021)
∗
ηF C = 0.52 EH2 = 33.3 W h/kg (IntelligentEnergy, 2021)
Updating the Oswald efficiency factor and the parasite drag coefficient into the design tool created in
the conceptual phase resulted in an estimated total flight time of 3h25 min, above the minimum required
in the design proposal. With the update on the UAV characteristics done, the preliminary design phase
of the project is finished. Figure 4.5 displays a CAD model done in OpenVSP with the UAV designed at
this stage of the design process.
Table 4.6: UAV general characteristics.
Weights Wing Tail

MTOM 21.77 kg Aspect Ratio 11.66 Total Area 0.516 m2
Structural 6.11 kg Area 1.372 m2 Dihedral 45◦
Propulsion 2.62 kg Span 4m Arm 1.4 m
H2 System 4.48 kg Taper at the Tip 0.55 Chord 0.255 m
Energy 2.91 kg Mean Chord 0.353 m Airfoil NACA0008
CG Location 875 mm from nose Airfoil SG6042
VTOL Prop System Forward Prop System Performance
Motor V605 Motor AT5220 Stall Speed 28 kts
Rotor T-Motor 22x7.4 Propeller APC 20x10 Max ceiling 10000f t
P Hover 858.07 W ηpr 0.65 Cruise Speed 38 kts
P max 3182.62 W P cruise 513.6 W Climb Speed 36 kts
RPM Hover 4734 P climb 1149 W Climb Angle 6.2◦
RPM max 7036 T cruise 17.08 N Cruise Altitude 5000f t
T climb 40.34 N Flight time 3h25
Figure 4.5: Preliminary CAD model of the UAV. Retrieved from Alves et al. (2021).
43
44
Chapter 5
CFD Validation
To validate the turbulence model to be used for the 3D wing and the aircraft, a 2D analysis was
performed considering the airfoil chosen at the preliminary design phase – SG6042. Initially, a mesh
convergence study is done and simulations applying the various turbulence models are run. Results
from experimental data and the low-fidelity software are compared, together with considerations featured
in the literature review.
5.1 Experimental Data and XFOIL Results
Together with its database, UIUC, along with Selig (1997), contains data related to the wind tunnel
testing of the chosen airfoil at various Reynolds numbers, ranging from 1 × 105 to 5 × 105 . The results for
a Re = 4 × 105 were used since its the closest regime to the author’s case. To replicate the experimental
results with the highest level of detail as possible, the simulation conditions for the low-fidelity software
and for Fluent are the same as the testing one: the airfoil chord is 12 in – 0.3048 m, a freestream
velocity of 19.41 m/s, air has Sea-level properties according to the Standard Atmosphere, and turbulent
intensity is kept around 0.1%, important for simulations using Fluent as this is a required inlet condition.
The conditions are specified on the experimental data in Selig (1997). The experimental data regarding
the lift and drag coefficients was interpolated for integer AoA between -4◦ and 12◦ to be compared
afterward with the software simulations.
For the low-fidelity software, XFOIL was employed. A 2D direct analysis was conducted for the test
Reynolds Number, a Mach number equal to 0.057, and considering eN as the transition criteria with
Ncrit = 9. The airfoil was discretized into 250 panels and tested for the same range of AoA as the
experimental data.
5.2 Fluent Considerations
Ansys enables the user to create a Fluent project from scratch and divides it into consecutive steps.
These steps are in agreement with the schematic presented in Figure 2.4, where a CAD model is created
45
together with the fluid domain, followed by the mesh generation, and culminating in the CFD simulation
on Fluent where the model equations, boundary conditions, and solver properties are set. To avoid
compatibility issues, the in-built Ansys CAD software - DesignModeler and automatic mesh generator
were used.
The methodology to be used for CFD simulations is determined by several factors, including the
application cases, flow regimes, and objects of study. Verı́ssimo (2016) features a set of Best Practice
Guidelines for CFD applied to UAVs. In his study, he validated the methodology to be employed for
UAVs in the LRN regime, which is the case of the presented dissertation. Although Verı́ssimo’s guide
was designed for Star-CCM+, most of the methodology can be replicated in Ansys Fluent and the Fluent
User Manual (Ansys Inc., 2013) serves as confirmation and additional insight.
5.2.1 Physical Modeling and Grid Generation
Following the methodology stated before, a 2D CAD model was produced containing the airfoil and
domain surrounding it. The SG6042 airfoil coordinates, retrieved from Selig (1997), were imported to the
CAD software, however, the airfoil’s sharp trailing edge was creating issues during the meshing process:
the inflation at the trailing edge was being poorly generated, creating elements with unacceptable values
for their orthogonal quality. To address this problem, the trailing edge was slightly cut and replaced with a
straight vertical one with a thickness of less than 1 mm. The global control volume was built according to
the airfoil chord length, creating a C-shape being at least 60 chords away from the airfoil in all directions.
The wind tunnel test conditions are replicated in the CFD simulation. For the freestream velocity
considered, incompressible flow is assumed, which makes the constant density assumption valid, thus
disregarding the use of the energy equation. The turbulence models to be considered are the ones
stated in subsection 2.4.1 with transition criteria, if applicable.
To simulate various AoA in the CFD software, two possibilities can be considered: rotating the airfoil
to the desired AoA or introducing a vertical component to the freestream velocity. The former option
better simulates the real conditions in terms of freestream orientation, although it requires an update to
the CAD model and consequently a new mesh generation. The latter requires changes in the direction
of the velocity inlet and pressure outlet freestream and in the force direction to be read for computing
the lift and drag coefficients. From the two options, the second one is preferred since no additional
computational time is required in re-meshing and the changes in the simulation and report definitions
are easily parameterized. One additional concern when choosing this AoA variation method is to also
consider the top and bottom horizontal boundaries as velocity inlets.
Apart from the freestream intensity (velocity magnitude) and direction, the turbulence intensity and
viscosity ratio need to be set. Turbulence intensity is set to 0.1%, the same as the wind tunnel, and
viscosity ratio to 1 according to Verı́ssimo’s Best Practices (Verı́ssimo, 2016) for both velocity inlet and
pressure outlet. The no-slip condition is set to the airfoil surface and the fluid domain properties are
defined according to the wind tunnel conditions in section 5.1.
For the mesh generation, the author considers three topics of relevance: general properties, local
46
refinements, and wall treatment. The considered mesh is unstructured with triangular cells with an
inflation layer around the airfoil. The cell base size is set, limiting the maximum cell dimensions far away
from the airfoil. A slow growth rate of 1.05 is applied and the smoothing is set to high to improve the
quality of the elements at the cost of a slight increase in the mesh generation time (Ansys Inc., 2010).
To refine the region near the airfoil, a discrete number of divisions were set on the upper and lower
surfaces of the airfoil. A bias type was applied to this edge sizing, making the regions near the leading
edge and the trailing edge smaller than in the middle. To be concise in the mesh convergence study,
only the number of divisions was changed, keeping the bias factor constant at 20 – ratio between the
smallest edge division and the largest (Ansys Inc., 2010). Furthermore, a circular region centered on
the trailing edge of the airfoil with a radius of 0.5 m (1.64 times bigger than the chord) was refined to
capture the pressure gradients which are larger in the vicinity of the airfoil, as well as the major portion
of the airfoil wake and separation if it occurs.
Special care is taken into consideration in the region near the airfoil where the boundary layer is
developed. To capture the BL evolution with accuracy, a structured grid is applied, denominated as the
inflation layer in Ansys Fluent.
The inflation layer thickness should be high enough to capture the whole BL, as such Prandtl’s
expression – Equation (5.1) for a fully turbulent boundary layer is applied to give an estimate of the
maximum boundary layer thickness – δ and thus the total height of the inflation layer (Schlichting, 1979).
Although the boundary layer is expected to have a considerable portion of laminar flow, applying Equa-
tion (5.1) assumes the highest thickness possible, guaranteeing that all the BL phenomena are captured.
δ = 0.37Re−0.2 c (5.1)
As stated in subsection 2.4.2, the first layer height of the inflation layer – y1 – needs to be properly set
depending on the turbulence model to be applied. Equation (5.2) defines y1 according to the target y +
and considers the friction velocity as a function of the friction coefficient, given by an empirical expression
from Prandtl-Schlichting valid for Reynolds numbers below 109 (Schlichting, 1979).
+
r
ytarget µ 1
y1 = , with uτ = 2
Cf U∞ and Cf = [2 log10 (Re) − 0.65]−2.3 (5.2)
ρuτ 2
To fully define the inflation layer in Ansys with the properties stated above, a first layer height control
is used where the y1 value, the total number of layers and the growth rate must be specified. The growth
rate defines the thickness of the next layer relative to the previous one, and it is set to 1.2, the default
value (Ansys Inc., 2010). The inflation layer height follows a geometric series where the first term and
the growth ratio are set, with the number of layers inferred, as the sum of all the layers must be close to
the total boundary layer thickness previously calculated.
Given the wind tunnel test conditions, the maximum thickness of the boundary layer is estimated to
be 1 × 10−2 m. Since all the turbulence models applied are capable of modeling the viscous sublayer
+
except the k − model, ytarget = 1. However after some simulations, the leading edge of the airfoil was
presenting values for y + around 2, therefore the first layer height was reduced by half from the value
47
obtained through Equation (5.2) and y1 was set to 8 × 10−6 m; 30 layers were considered to capture the
+
whole boundary layer. For the k − model, y1 = 8 × 10−4 m for a ytarget = 50 and 7 layers were defined.
Figure 5.1 shows the airfoil surrounding mesh where the circular region with a smaller refinement is
visible, together with a detail of the 30 layer inflation adjacent to the airfoil surface, in which it is also
possible to visualize the effect of the bias factor applied to the airfoil division.
Figure 5.1: 2D Mesh with the detail on the inflation layer.
5.2.2 2D Mesh Convergence Study
With the local refinements considered in the mesh generation, a grid convergence study was con-
ducted. A coarser mesh is generated first, and for subsequent grids, a percentage ratio is defined and
applied to keep the base cell size, the local refinement near the airfoil, and the number of elements in
the upper and lower surface concise across all grids generated, while the trailing edge divisions and the
inflation layer were kept constant to guarantee that y + < 1 for the first layer. Four meshes were created,
the coarser having 57k and the finer 213k elements. This study was performed for an AoA of 12◦ near
stall conditions and considering the recommended turbulence model – the Transition SST, since this
model is the most complex one and requires more computational effort.
The airfoil 2D lift and drag coefficient were the variables of interest for the mesh convergence study,
and the results are plotted in Figure 5.2. The solution is said to be converged whenever the residuals
drop by at least 3 decimal cases and no significant variations occur between iterations on the variables
of interest. Both lift and drag coefficients vary less than 0.5% from the mesh with 92k elements to the
one with 146k, thus making the 92k elements mesh the one to be used in future 2D analysis. Details of
the mesh to be considered for future simulations are shown on Table 5.1.
Table 5.1: 2D 91k mesh parameters.
Parameter Value
Base Cell Size 3.75 m
Number of divisions of the upper/lower surface 200
Number of divisions of the trailing edge 7
Sphere of influence Cell Size 7.5 × 10−3 m
48
(a) Cl evolution (b) Cd evolution
Figure 5.2: 2D Mesh convergence study.
For the mesh to be used with the turbulence models that require near-wall functions, the k − model,
a similar grid convergence study was conducted. The conclusions are similar to the previous case, with
the chosen mesh having 86k elements with the same refinements and properties listed in Table 5.1.
5.2.3 Turbulence Model Validation
The turbulence models considered for the 2D simulations are described in subsection 2.4.1. The k−
Realizable model is used, which is a modification of the base model recommended for boundary layers
under strong adverse pressure gradients, separation, and recirculation. For the SST model, Fluent
incorporates a low Reynolds number correction which will be applied (Ansys Inc., 2013). Simulations for
an AoA of 0◦ and 4◦ (Table 5.2) were run and compared to the experimental UIUC data and to XFOIL,
as well as for the highest AoA with experimental data available – 12◦ (Table 5.3).
Table 5.2: Comparison between experimental, XFOIL and Fluent results for AoA of 0 and 4 degrees.
AoA = 0◦ AoA = 4◦
Cl Cd % Cl % Cd Cl Cd % Cl % Cd
Experimental 0.4923 0.0077 0.9122 0.0091
XFOIL 0.5155 0.0073 4.73% -5.16% 0.9397 0.0082 3.01% -9.44%
SA 0.4938 0.0133 0.31% 74.07% 0.9180 0.0158 0.64% 74.21%
k − Realizable 0.4889 0.0153 -0.68% 99.38% 0.9091 0.0182 -0.33% 100.97%
SST Low Re 0.5018 0.0093 1.94% 21.97% 0.9220 0.0117 1.08% 29.18%
k − kl − ω 0.5142 0.0075 4.46% -1.81% 0.9618 0.0093 5.44% 3.03%
Transition SST 0.4883 0.0076 -0.82% -0.77% 0.9326 0.0093 2.24% 2.67%
For these 2D simulations, a pressure-based segregated solver was considered, with the SIMPLE
scheme. For the models without transition modeling (one and two-equation models), a steady, second-
order spatial discretization numerical scheme resulted in a relatively quick convergence with all residuals
becoming smaller than the 1 × 10−6 criteria defined. Nevertheless, for the Transition SST and k − kl − ω
models, second-order schemes were resulting in a poor convergence for the continuity and k turbulence
49
Table 5.3: Comparison between experimental, XFOIL and Fluent results for AoA of 12 degrees.
AoA = 12◦
Cl Cd % Cl % Cd
Experimental 1.4504 0.0341
XFOIL 1.4500 0.0292 -0.03% -14.47%
SA 1.6215 0.0316 11.79% -7.43%
k − Realizable 1.5718 0.0402 8.37% 17.74%
SST Low Re 1.4854 0.0581 2.41% 70.32%
k − kl − ω 1.6219 0.0397 11.82% 16.46%
Transition SST 1.4867 0.0361 2.50% 5.89%
equations, so, first-order upwind schemes were used for these equations. This modification managed
to reduce the residuals to the criteria, while compromising the accuracy of the results, as the drag
coefficient values were twice as high when compared with the experimental ones . This variation was
expected to occur due to the low order schemes adopted as stated in 2.4.2. Resorting again to second-
order schemes for all equations and using the previous solution as initial conditions, the lift and drag
coefficient were presenting oscillations around a value much closer to the experimental one. A transient
(unsteady) simulation was then run, which resulted in all equations converging.
The one-equation SA turbulence model over-predicts drag by up to 75% for 0◦ and 4◦ AoA. This
method does not model transition, therefore it assumes turbulent flow right from the airfoil leading edge,
contrary to what XFOIL and the transition models predict, which demonstrate that for these small angles
of attack, a considerable portion of the upper surface is in the laminar regime, having a lower drag
contribution than turbulent flow.
The k − Realizable presents a substantial discrepancy in the drag coefficient values, up to twice
as much as the experimental ones for small angles of attack. This corroborates the necessity of using
a model capable of accurately predicting the physics of the viscous sub-layer without resorting to wall
functions.
The results of k − ω SST model were very similar to the SA, with the drag prediction significantly
improving after the Low Reynolds correction was added to it, with errors dropping to around 30%, still
significantly high. For an AoA of 12◦ , the drag coefficient exhibits a considerable discrepancy com-
pared to the UIUC data. One reason might be due to the difficulty in providing accurate results when
considerable separation occurs and no real transition prediction.
Transition models feature much closer to the experimental ones for the simulated AoA than the previ-
ous one and two-equation models. Both of them manage to capture the LSB which occurs at the upper
surface of the airfoil, inducing the transition to turbulent flow; however, no noteworthy changes can be
observed either in the location of the LSB nor in the length of it. The k − kl − ω model exhibits drag val-
ues higher than the Transition SST model and this tendency is more significant with higher AoA. Studies
from Aftab, Rafie, Razak, and Ahmad (2016) on a NACA4415 at Low Reynolds support this trends, in
which the k − kl − ω predicted the location of the LSB but not the turbulent reattachment; additionally,
the k − kl − ω model presented convergence problems and required a higher computational effort than
50
the Transition SST.
By incorporating an accurate transition model and featuring the most concordant results with the
experimental ones, this study validates the selection of the SST transition method as the go-to model for
airfoil flow analysis under a LRN regime.
5.3 2D CFD Simulations
With the physical model set, the solution method resorts to a SIMPLE scheme using steady-state
first order spacial numerical schemes initially for rapid convergence and to approximate the solution to
the real one, followed by a transient simulation with second-order temporal and spacial schemes for
more accurate results.
For a transient simulation using the pressure-based solver, a time-step must be defined. The time
step should be carefully selected: if it is too small, the simulation takes an excessive amount of time
to be solved, while a time step too large does not capture the small changes in the flow and does not
converge in a couple of iterations. The time step should be defined to capture a single particle around
20 to 50 times in the region of interest, and convergence should be achieved in 20 to 30 iterations per
time step. For the inlet velocity prescribed and the tested airfoil chord, a time step of 0.5×10−3 s captures
a particle around 30 times on the airfoil vicinity. For this time step, the solution converges in around 40
iterations, with all residuals below 10−6 .
Figure 5.3 plots the lift and drag polars for the available data from UIUC, XFOIL results for integer
AoA inside the range of available experimental data, as well as the results from Fluent simulations only
for even numbers of AoA, due to time constraints. XFOIL exhibits errors below 7% for the lift coefficient
when compared to the experimental data, while the drag coefficient is always under-predicted by around
10%, reaching 15% for high AoA. These results validate the use of XFOIL as a quick and reliable
software to compute the lift coefficient of an airfoil at low Reynolds and an order of magnitude for the
drag coefficient.
(a) Cl vs alpha curve (b) Cd vs alpha curve
Figure 5.3: Aerodynamic polars for the SG6042 airfoil.
51
The lift coefficient data from the Fluent simulation using the Transition SST model is concordant with
the experimental values, having discrepancies under 4% for the AoA simulated. For the drag coefficient,
Fluent presents higher values than the experimental ones, with a tendency of increasing this discrepancy
the higher the AoA is, up to 15%. This behavior is not concerning from a design standpoint, and it is
more conservative than the values of low-fidelity software.
The intermittency factor included in the Transition SST model is used to turn on the production term
of the turbulent kinetic energy downstream of the transition point. The model sets the intermittency to
one on the freestream and to zero on the fluid region in contact with the airfoil (Menter et al., 2006). On
the outer boundary layer, a small increase in the intermittency indicates the beginning of the LSB, while a
unit value for the intermitency downstream, confirms that transition has occurred and the boundary layer
is fully turbulent. As a result, this parameter is useful for predicting the location of the LSB in conjunction
with the velocity contour plots.
While using γ − Reθ transition model, a LSB was developed on the upper surface of the airfoil, being
located closer to the trailing edge for low AoA and moving forward for increasing angles of attack. For
negative AoA, a small bubble was also present near the leading edge of the airfoil at the lower surface,
which is visible on Figure 5.4. The turbulent kinetic energy after the LSB becomes considerable (larger
than one), confirming the presence of transition at these locations (Figure 5.5).
Figure 5.4: Intermittency factor for an AoA of -4◦ . Black circles indicate the location of the LSB.
Figure 5.5: Turbulent kinetic energy for an AoA of -4◦ .
XFOIL is adapted to be used in the LRN regime where the LSB phenomenon is common. Without
data regarding the fluid surrounding the airfoil, the LSB location can be inferred through the wall shear
stress distribution along the airfoil surface, since XFOIL gives a negative value for the wall shear stress
over regions with reverse flow tangent to the airfoil (McArthur, 2007). Figure 5.6 plots the skin friction
52
coefficient along the airfoil and confirms that XFOIL predicts the presence of the LSB in the same region
as Fluent, both for the upper and lower bubbles, at an AoA of -4◦ . In Fluent, the skin-friction coefficient
was always positive, so, at the reverse flow region, it was converted to its symmetric value to be coherent
with the XFOIL data.
Figure 5.6: XFOIL – Skin friction coefficient distribution for an AoA of -4◦ .
Analyzing Figure 5.6, the skin friction coefficient along the upper surface is practically coincident
until 90% of the chord, after which Fluent presents a sudden increase (absolute value) until around
95% chord. Looking at the velocity vector plot at this location, the reverse flow vortex is developed at
this position, which has a higher adverse velocity gradient near the airfoil than the upstream ”dead air”
region, as depicted in Figure 2.1. For the lower surface, both XFOIL and Fluent have the same behavior,
with XFOIL predicting higher skin friction at the lower bubble near the leading edge while being lower
than Fluent throughout the majority of the surface. The general trend is concordant with the typical skin
friction plots, having a lower value in the laminar region, increasing at the transition, and having a small
decay over the turbulent region.
At an AoA of 6◦ , separation starts to occur, confirmed by the presence of a blue region of null velocity
over the upper surface of the airfoil at the trailing edge, bigger than the expected boundary layer total
thickness. The separation region increases in size until 12◦ , the final AoA tested and comparable with
UIUC data. At this attitude, the separation region occupies around 40% of the airfoil, however, according
to XFOIL data, the lift coefficient increases up to 15◦ , at which the airfoil stalls. To validate these results,
a Fluent analysis for 13, 14, 15, and 16◦ was conducted and, in fact, after 15◦ the lift coefficient starts
decreasing. Figure 5.7 shows a velocity contour plot at an AoA of 15◦ where large separation is visible
and vortex shedding is very active.
53
Figure 5.7: Velocity contour plot for an AoA of 15◦ .
The discrepancies between both software results and the experimental data can have multiple rea-
sons. In XFOIL, a set of points of the airfoil is defined and line segments are created between them.
To prevent sharp angles, at the region of bigger curvature (leading edge and trailing edge), more points
are defined. Despite this improvement, the final airfoil shape is different from the real one and it can
be one reason for the different outcomes. Additionally, the transition criteria eN was applied with the
default calibration, and no studies were performed by varying the Ncrit parameter. Although giving this
parameter the value of nine, the default for the wind tunnel conditions (Drela, 1989), small disturbances
in the inlet conditions might change the indicated value for this variable.
Fluent uses the set of points imported and creates a spline curve from it, creating a smooth airfoil.
Although using splines is better than using XFOIL since it does not create angles, the final shape might
be slightly different from the real airfoil, thus, producing different results. Furthermore, the inlet conditions
defined might be a little different from the wind tunnel conditions, namely some viscosity parameters
of atmospheric conditions. Studies on different airfoils with variable turbulence intensities showed that
increasing the freestream turbulence intensity resulted in earlier transition and reattachment, contributing
to an overall decrease in LSB length (Istvan, Kurelek, & Yarusevych, 2018); resorting to CFD tools, the
same conclusions were achieved by N. Silva (2014), corroborating the impact of the turbulence intensity
at the inlet on the obtained results. The turbulence viscosity ratio was set according to the best practices
provided, without any information of the experimental value; nevertheless; its influence is negligible,
either in the results or at the location and size of the LSB, according to Verı́ssimo (2016). The grid
convergence study was performed for the transition and turbulence model applied, for a specific angle
of attack. Although the methodology used allowed using the same grid for all the spectrum of AoA
studied by only changing the freestream orientation, the grid convergence should be done for every AoA
simulated, a very time-consuming process. A similar study was done for an AoA of 0◦ with conclusions
close to the ones described in subsection 5.2.2, which strengthens the robustness and accuracy of the
considered grid for the 2D analysis conducted.
54
Chapter 6
Detailed Design – Aerodynamic

Performance
With the turbulence and transition models validated in 2D, a full 3D analysis can be conducted to
aerodynamically characterize the aircraft. Since XFLR5 is not recommended for a full plane analysis
(with fuselage), the wing is modeled and simulated in Fluent and compared with XFLR5.
6.1 Analysis of 3D Wing
In the preliminary design phase, the wing geometry was defined according to the recommendations
featured in the theoretical background and validated using the low-fidelity software XFLR5. At this stage,
CFD methods are employed to corroborate the previous results and verify if the wing geometry is suitable
and meets up with the project requirements.
6.1.1 Domain Characterization and Mesh Studies
A 3D model of the wing was created, similar to the 2D analysis previously presented. Since Fluent
has a symmetry boundary condition option, only half of the wing is considered to greatly reduce the
computational time of the simulations. The fluid domain consists of a semi-sphere upstream of the
wing and a cylinder downstream, with the boundaries at least 60 chords away from the wing, which
guarantees that over the span-wise direction, the fluid domain boundary is at least ten wingspans apart
from the wing (Verı́ssimo, 2016). This creates a C shape on the symmetry boundary, similar to the airfoil
simulation. A second control volume was created around the wing to capture the wake and possible
vortex shedding up to five chords away from the trailing edge.
For the mesh generation, tetrahedral cells are chosen and the same principles considered before are
applied to the 3D case: the base cell size is set and all the refinements done are a fraction of this base
value, except the inflation layer, which is kept constant. Since the number of elements for 3D simulations
is much bigger than in 2D cases, the same level of refinement is not possible to translate to the wing
55
since the computational time and effort would be impracticable. A growth rate of 1.25 is set for the main
domain and 1.15 for the second control volume, closer to the wing, to better predict the phenomena in
this region. The bias factor applied on the airfoil surface for the 2D case is not possible to replicate in 3D,
as the automatic mesh generation software does not allow a progressive surface sizing in the chordwise
direction. As a result, both the upper and lower surfaces were divided into 2 regions each: one from the
leading edge to around 10% of the chord to apply a smaller surface cell size to better account for the
high curvature of this wing region, and the remaining one, which is more smooth and does not require
such a small cell size. The wing has a fifth surface corresponding to the finite trailing edge, which was
created for the same reasons as explained in the 2D case.
+
Since the same turbulence and transition model are applied, ytarget < 1 is a conditions for the models
to be valid. To account for higher values of y + expected, 0.5 is considered the target for the calculation,
yielding y1 = 1.01 × 10−5 m, a total of 30 layers with a growth rate of 1.2 to capture the whole boundary
layer. The BL is predicted to have at most a thickness of around 1 × 10−2 m, according to Equation (5.1).
To select the most appropriate mesh, five meshes were created with the base cell size and the refine-
ments varying in the same proportion. Based on the lift and drag coefficients and on the computational
time, the most appropriate mesh is chosen for further simulations. Each mesh simulation was run using
first-order and second-order spatial schemes for a steady case and also for a second-order transient
temporal discretization.
(a) Cl evolution (b) Cd evolution
Figure 6.1: 3D Wing mesh convergence study.
The plots in Figure 6.1 represent the evolution of CL and CD for the five meshes. Looking at the
results obtained, the first order steady simulations exhibits a higher lift and drag coefficient for each of
the five simulations run when compared to the second order, both steady and transient, even though
all of them tend to converge as the number of elements increases. XFLR5 predicts CL = 0.8347 and
CD = 0.0270 for the 4.5◦ AoA simulated in this study. Although first order simulations are much quicker
than the remaining ones, they present a drag coefficient over 40% higher than XFLR5 even for the most
fine mesh considered, discarding them from being used in further studies. The difference between the
results obtained from second order steady and transient solutions is almost negligible, with the transient
one taking over 6 times more time than the second order steady one. This analysis justifies the choice
56
of running second order steady simulations, however, if for high AoA, poor residual convergence occur
or oscillations on CL and CD , a transient solution might be necessary.
Assuming the results of the most refined mesh as reference, the deviation of both lift and drag
coefficients were computed to choose the most appropriate mesh. The ”intermediate” mesh, having
5.13M cells, has a deviation of less than 0.5% for CL and 1% for CD , featuring a good compromise
between result accuracy and computational time, and is thus used for further analysis.
For 3D flow analysis, Fluent can transform a tetrahedral mesh into a polygonal one. A polygonal
mesh can have up to three to five times fewer cells with the same level of refinement, however in the
presence of inflation layers, the cell count reduction is not so pronounced (Ansys Inc., 2013). While
the mesh has fewer cells in total, which would decrease the computational time, each cell has more
faces and neighbors, requiring more gradients to be computed, thus having the opposite effect. Overall,
after transforming the mesh to polygons, the minimum skewness has increased, which is a preliminary
indicator of a better mesh quality.
A similar convergence study was performed using the five previous meshes converted to polyhedral
elements. The results show the same tendency as the tetrahedral ones for an AoA of 4.5◦ with the
”intermediate” polygonal mesh having a good compromise between accuracy and processing time, nev-
ertheless, for conditions near stall, the behavior is slightly different: an in-depth analysis was conducted
comparing the original tetrahedral mesh to its polygonal transformation (3.52M cell polygon mesh), and
to a polygonal mesh with a number of elements close to the original one. This was achieved by slightly
refining the previous mesh, which gave origin to a 5.01M cell polygonal mesh. The general characteris-
tics of each mesh are given in Table 6.1.
Table 6.1: Properties of the meshes.
Cell Geometry Nr. of Cells Nr. of Faces

Tetrahedral 5.13M 11.87M
Polyhedral 3.52M 13.53M
Polyhedral 5.01M 19.24M
Figure 6.2: Lift to drag ratio at various AoA.
The lift-to-drag ratio obtained for various AoA by the three meshes in question is plotted in Figure
6.2. At low AoA, both polyhedral meshes exhibit similar results, corroborating the conclusions from the
polyhedral convergence study that the 3.52M poly mesh was suitable for simulations at 4.5◦ . However,
the polyhedral meshes present higher aerodynamic efficiency than the tetrahedral mesh up to 8◦ .
Analyzing the results near stall and at an AoA above 11◦ , the tendency stated before is inverted:
57
the 5.13M tetra mesh has a higher L/D ratio than both polyhedral meshes, and the 3.52M poly mesh
exhibits a considerable difference from the 5.01M poly mesh, indicating that for high AoA, an increase
in the mesh refinement has a substantial impact on the results obtained.
The characteristics of each mesh, displayed in Table 6.1, confirm the increase in the number of
faces and the decrease in the number of cells compared to its tetrahedral counterpart (5.13M tetra to
3.52M poly). No noteworthy differences were encountered in the computational time between these two
meshes, while the third one took more time. From all the analysis done, the 5.12M tetrahedral mesh
will be used for the wing analysis since, for cruise conditions, this mesh is more conservative than the
others, and without experimental data to confirm which one is more accurate, being conservative is the
safest option. Furthermore, while the solution time is similar, the extra step of converting the mesh to
polyhedral takes some time and no significant gains are obtained by doing it.
6.1.2 Wing Aerodynamic Analysis
With the mesh defined, a comparison of the lift and drag polars is done between XFLR5 and Fluent,
illustrated in Figure 6.3. For low AoA, Fluent and XFLR5 provide similar outcomes, which justifies using
the latter in the initial design stages. XFLR5 presents slightly higher lift and lower drag values, with the
differences becoming more pronounced for higher AoA. As expected, XFLR5 does not provide results
beyond stall, and even while approaching stall, special care should be taken (Dantsker & Vahora, 2018).
After 6◦ , Fluent expresses a noticeable decrease in the lift coefficient, having a maximum value of 1.310
at 12.5◦ , a stall later than the 11.5 ◦
depicted by XFLR5. For the drag curve, XFLR5 having higher
values than Fluent for the same angle of attack can be misleading: if the comparison is made between
both stall conditions, Fluent has a higher drag coefficient than XFLR5 as the transition and turbulence
model from Fluent is more accurate than the 2D viscosity extrapolation to 3D from XFLR5 (Menter et al.,
2006; Peerlings, 2018).
(a) Cl vs AoA evolution (b) Cd vs AoA evolution
Figure 6.3: 3D Wing lift and drag coefficients
In the 2D analysis previously done, the laminar separation bubble was present and induced the
transition from laminar to turbulent flow. Depending on the angle of attack, this bubble varies in its
58
location and thickness. To predict the location of the LSB on the 3D wing model, velocity contour
plots with a negative value in the freestream direction were generated; this phenomenon occurs in two
situations: in the presence of an adverse pressure gradient and in the recirculation inside the LSB.
Additionally, from Figure 5.5, the turbulent kinetic energy has a sudden increase after transition and
starts decaying after a short length, therefore it can also be useful to locate the transition region. Figure
6.4a depicts the reverse flow region in red and high turbulent kinetic energy regions in blue, indicating
that the flow is laminar until around 70% of the chord and turbulent after the LSB located in that region.
The pressure coefficient – Cp – of three sections of the wing, plotted in Figure 6.4b, supports these
findings: all three sections have the same pressure coefficient behavior, and a sudden decrease in Cp
is visible in the same region where the LSB is predicted.
(a) Reverse flow region (red) and high turbulent kinetic energy re- (b) Pressure coefficient plots of wing sections at z=0.5m,
gions (blue) for an AoA of 4.5◦ . z=1m and z=1.5m.
Figure 6.4: 3D Wing lift and drag coefficients
As the angle of attack increases, the LSB moves to the leading edge and a higher portion of the
wing upper surface becomes turbulent. As the wing attitude approaches stall conditions, separation
starts to occur at the trailing edge. Figure 6.5 is similar to the previous case, for an AoA of 14◦ . Two
distinct phenomena are visible: near the leading edge, the presence of the LSB similar to Figure 6.4a,
and at the trailing edge, a reverse flow region and an increase in the turbulent kinetic energy caused
by the separation of the flow from the upper surface. At some wing sections, only one continuous blue
region is noticeable as the LSB bursts and separation occurs right from the leading edge. The velocity
contour plots over wing sections at 0.5m and 1.5m in the spanwise direction are shown in Figure 6.6a
and 6.6b, respectively. In the first one, although not clear, the existence of a blue region near the leading
edge indicates the LSB, and turbulent flow is developed until separation around 75% of the chord. At
z = 1.5m, the flow never reattaches after the LSB and separation occurs immediately. These burst
phenomena are chaotic and, for the same angle of attack, the regions where they occur are random
and change over time, resulting in oscillations in the lift and drag coefficients even after the solution has
converged. As the angle of attack further increases, more regions on the upper surface have flow fully
detached and the lift coefficient further decreases.
59
Figure 6.5: Reverse flow region (red) and high turbulent kinetic energy regions (blue) for an AoA of 14◦ .
(a) z = 0.5m (b) z = 1.5m
Figure 6.6: Velocity contour plots at two different wing sections for an AoA of 14◦ .
6.1.3 Parametric Studies to Enhance Wing Performance
Geometric twist at the tip of the wing

With the wing design in the preliminary phase fully characterized, the author conducts some para-
metric studies in order to improve its aerodynamic characteristics. As stated in section 2.2.2, adding
twist to a wing can delay stall conditions at the tip and approximate the lift distribution to the ideal ellip-
tical one, thus reducing the induced drag component. In the preliminary phase, Sá (2021) idealized a
wing structure for the outer wing panels that has a single spar located around 30% of the chord; this po-
sition is defined as the center of rotation to avoid twisting the spar. Moreover, on the interior wing panel
(rectangular part), twist will not be applied to avoid further modifications to the wing-boom connection,
and the twist variation on the outer panel will be linear.
The values provided in subsection 2.2.2 were retrieved from Sadrey (2013) and are relative to aircraft
much heavier than the UAV designed. Without further information on the geometric twist angles of Class
I UAVs, a simple and quick analysis using XFLR5 is conducted to check which is the most beneficial
twist angle to be applied.
For twist angles smaller than 2◦ , both lift and drag exhibited little variance when compared to the
original geometry. Angles steeper than 5◦ present a reduction in both the lift and drag coefficients. For
this case, the wing AoA has to be higher to produce the same lift force. Comparing attitudes with the
same lift coefficient, the drag force was higher for the twisted wing than in original geometry. The author
60
defines the incidence angle of the tip as being 3.5◦ lower than the incidence angle of the inner wing
panel, to perform further studies using CFD tools.
Simulations for various angles of attack were performed and the results showed little to no difference
in the aerodynamic polar up to the maximum lift coefficient of the baseline wing. Nevertheless, the AoA
has to be between 0.75◦ and 1◦ higher on the twisted wing to match the CL produced by the baseline
model. Due to this, the angle of attack at which the twisted wing stalls is 13.5◦ , with its maximum lift
coefficient being the same as the original wing. The major difference is in the wing portion, which begins
to enter in stall first. In the baseline wing geometry, the outer panel of the wing was the one entering in
stall first, a phenomenon not desired as this is where the control surfaces are located, which lose their
effectiveness in stall conditions. Figure 6.7a represents the streamlines of the baseline geometry at an
AoA of 14◦ , an attitude in the stall region which corroborates the conclusions derived from Figure 6.5. In
the proposed wing geometry with twist, stall occurs initially at the root of the wing and propagates to the
tip, as seen in Figure 6.7b, allowing the aircraft to recover and keep control over roll motion.
(a) Original wing at an AoA of 14◦ (b) Wing with twist at an AoA of 15◦
Figure 6.7: Streamlines over the upper surface of the wing at the beginning of the stall region.
Wingtip configurations
Another improvement that can be made to the wing is to modify its tip geometry and configuration
to reduce the effects of wingtip vortices and, consequently, reduce the induced drag. Induced drag de-
creases as the aspect ratio of the wing increases, however, wingspan constraints or structural difficulties
may impose a maximum aspect ratio (Raymer, 1992). To tackle this problem, choosing a wingtip con-
figuration that reduces the strength of the vortices is beneficial and leads to an increase in the effective
AR (Gudmundsson, 2014).
One option is to simply reshape the tip of the wing in such a way that these effects are diminished,
reducing the induced drag. A Hoerner wingtip is a simple solution adopted by most of the UAVs in
CIAFA. This configuration has a sharp finishing on the tip, obtained by performing a flat cut on the wing,
as seen in Figure 6.8a. The leading edge has a round shape to prevent the upper surface from stalling
at low angles of attack (Gudmundsson, 2014).
Applying the Hoerner tip on the baseline wing resulted in a decrease of just 1.5% on the drag co-
efficient, with the lift generation not being affected. While a reduction in the drag is always beneficial,
61
this reduction is quite small and may be misleading. Although the same mesh properties were used for
this study, it is important to note that during the mesh convergence study, this grid had a deviation in the
order of 1% when compared to the most refined one. The results may just reflect a small change in the
grid and not actual changes in the wingtip vortices, as no remarkable differences were observed in the
vorticity distribution nor in the wingtip vortex strength in the CFD post-processing plots and results.
Other wingtip configurations are more effective at reducing the induced drag. Two types are common
in general aviation: raked tips, which are a prolongation of the wing on the same plane as the wing –
Figure 6.8b, and winglets, which also extend the wing vertically (up, down, or a combination of both) –
Figure 6.8c. Both of these configurations increase the effective wingspan of the aircraft and have a larger
wet area due to the added structure. For these configurations to be viable, the gains in reducing the
induced drag component need to outweigh the increase in the skin-friction component, and optimization
studies on the geometry of each design must be thought out. The raked wingtip can vary in span length,
taper, sweep, and dihedral, and the winglet can also have its toe angle vary.
(a) Hoerner wingtip (b) Raked tip (c) Winglet
Figure 6.8: Examples of possible wingtip configurations. Retrieved from Gudmundsson (2014).
Preliminary studies done on XFLR5 showed a small gain (up to 4%) in the aerodynamic efficiency of
the wing. Nevertheless, when trying to keep the wingspan fixed at 4 m, both of these options could not be
applied, and even when the wing was slightly shortened to introduce these wingtip devices (while keep-
ing the total wing area), these alterations performed worse than the original planform. If the wingspan
constraint was violated and the wingtip devices study was carried out, the additional structure also
comes with the drawback of increasing the MTOM of the UAV, and trade-off studies should be carried
out to check the real effectiveness of the wingtip configuration.
While advantages were observed with these parametric studies, especially with the introduction of
geometric twist, their influence on the remaining areas of the project is unknown and has not been
studied, namely on the wing structure. As a result, the author will continue the detailed analysis with
the design fixed on the preliminary phase, which is also the one considered by Sá (2021) and P. Silva
(2021) in the detail design in the structural and propulsion areas, respectively.
62
6.2 VTOL System wind tunnel Testing
Due to the complexity of the rotors’ geometry, calculating the parasitic drag produced by these com-
ponents during cruise poses some challenges, as seen in the preliminary phase. A viable alternative is
to perform wind tunnel tests to determine the impact of this sub-system on the total aircraft drag.
To proceed with the experiment, a physical assembly was created with the VTOL system mounted
on a boom, as illustrated in Figure 6.9. The distance between the front and rear rotors is 1.19 m, and
because both systems are the same, the center of vertical thrust is the midpoint between the rotors
and coincides with the CG of the aircraft. To connect the motors to the booms, a support was created
in Solidworks and 3D-printed, and the Electronic Speed Controller (ESC) of each motor was attached
under their support. To nullify the torque from the rotors, diagonally opposite rotors rotate in the same
direction (Khan, 2014). As a result, the rotors used in the assembly are symmetrically equal: the front
rotor rotates clockwise while the rear one rotates counterclockwise, as visible in Figure 6.9 by looking at
the leading edge of each rotor. On the other boom, the front and rear rotors swap positions.
Figure 6.9: Boom with VTOL system to be tested in the wind tunnel.
The test conditions should be as similar as possible to the operation ones, as such, cruise velocity
is replicated in the wind tunnel. The atmospheric conditions (static pressure, relative humidity, and
temperature) are given by a precise in-room climate measurement device, while a pitot tube located
near the testing area outputs the dynamic pressure – q. Applying the perfect gas law - Equation (6.1),
dynamic pressure equation – Equation (6.2) and Sutherland’s law – Equation (6.3) gives the air density,
freestream velocity, and dynamic viscosity, respectively. Table 6.2 sums up the conditions for the tests
conducted. To remove the influence of the boom and the balance arm, the drag of these components
without the VTOL system was obtained for the same freestream conditions, and the data presented
throughout this section is already subtracted by the results from this control test.
63
p = ρRT , with R = 287 J/kgK (6.1)
1 2
q= ρU (6.2)
2
3/2
T Tref + Sµ
µ = µref , with µref = 1.718×10−5 kg/(ms) , Tref = 273.15 K , Sµ = 110.4 K
Tref T + Sµ
(6.3)
Table 6.2: Wind tunnel testing conditions.
Data Calculations
◦
Temperature 23.9 C Density 1.181 kg/m3
Climate Sensor
Static Pressure 1006.7 kP a Velocity 19.69 m/s
Pitot Tube Dynamic Pressure 229 Pa Dynamic Viscosity 1.832E-05 P a.s
As the vertical climb phase end and some horizontal velocity is achieved, the rotors will stop rotating
in an arbitrary position and their stopping position is independent from each other, for this reason, the
tests were executed for a combination of possible stopping positions for the rotor. For the sake of
clarity, the angle of the rotor φ is measured relative to the spanwise direction (rotor perpendicular to
the freestream) and positive angles in the clockwise direction. Increments of 45◦ were applied, which
translates into four different angles tested for each rotor (for a two-blade propeller, after 180◦ the position
is the same), resulting in a total of sixteen different configurations evaluated, enough data to understand
the general trend of the results. Figure 6.10 illustrates the positions considered for the front rotor.
Although the rear rotor rotates in the opposite direction, its position definition is the same as the front
rotor, following the birds-eye view from Figure 6.10.
(a) φ = −π/4 rad (b) φ = 0 rad (c) φ = π/4 rad (d) φ = π/2 rad
Figure 6.10: Stopping positions considered for the front rotor.
Each test ran for around 30 seconds, and data regarding the force and moments around the three
axis was obtained. The data provided by the balance and the testing software already outputs the mean
value and the standard deviation for the test duration, thus easing the data processing step. Figure 6.11
depicts the mean drag values obtained for each combination of rotor position, along with a first-order
two-variable polynomial-trigonometric curve fitting equation. The curve-fitting equation expression is
given by Equation (6.4), in which z corresponds to the drag force and φ1 and φ2 , the front and rear rotor
position angle, respectively. Annex A provides in greater detail the results obtained in a table with the
64
mean values, along with the standard deviation, together with the values obtained in the vertical direction
– lift generation. Additionally, the formulation of the curve-fitting equation is also described in Annex A.
Figure 6.11: Mean values of the drag force obtained in each test, with the curve-fitting equation.
z = a + c1 cos(2φ1 + f1 ) + d1 sin(2φ1 + g1 ) + c2 cos(2φ2 + f2 ) + d2 sin(2φ2 + g2 ) (6.4)
a = 2.453 c1 = 0.2373 c2 = −0.7637 d1 = 0.615 d2 = −0.6392
f1 = 3.57 f2 = 1.962 g1 = 1.801 g2 = −0.4765
The curve-fitting equation from Equation 6.4 has a R2 = 0.91138, having a good correlation with
the experimental data, following the trends of the drag values obtained and providing a reasonable
estimation for the drag produced by these components. Because the number of data collected is small,
a second order curve-fitting provided a perfect correlation, nevertheless, the graph and values presented
for intermediate, not tested angles (φ = ± π/3 rad,φ = ± (2π)/3 rad, for example) were not physically
correct. Furthermore, the standard deviation values, while not significantly high, indicate that some
uncertainty occurs in the data gathered, which gives more margin for slight changes in the real drag
values and might benefit the curve-fitting equation.
Analyzing the results obtained, it confirms that, for the spectrum of positions tested, the combination
which provides the least drag is when the rotors are aligned with the booms – φ = π/2 rad, while the
worst case corresponds to φ = 0 rad, with the rotor spanwise direction perpendicular to the freestream,
having a drag force over 230% when compared to the previous case.
The results obtained also confirm the conclusions stated by Mendes (2021): the front rotor has a
greater impact on the drag produced than the rear one. This can be deduced by comparing the results
from Table A.1 when the front rotor is at 0 rad and the rear at π/2 rad, with the opposite case: the first
case produces a higher drag than the second. One explanation for this phenomenon is that the rear
65
VTOL set is located in the wake produced by the front set, reducing the drag produced by the former.
Another inquisitive phenomenon that occurred is that the drag is higher whenever the leading edge
of the rotor is ahead of the motor (π/4 rad for the front and −π/4 rad for the rear) than when it is behind.
When decomposing the drag into viscous and induced components, the viscous part should remain
relatively similar for both cases as they have the same surface in contact with the airflow, therefore it is
the induced component which varies. Analyzing the force in the vertical direction, presented in Table
A.3, it follows the same tendency as the drag force, corroborating the hypothesis stated before.
Although the test conditions were defined to resemble the cruise conditions, it is impossible to fully
replicate them in the wind tunnel, and thus, the drag forces measured should be transformed into their
coefficient form. Considering the wind tunnel density and freestream velocity, the drag coefficient of the
full VTOL system (4 motors and 4 rotors) varies between 0.00891 and 0.0204, with the reference area
being the wing. When comparing these results with the ones computed in the preliminary design phase
(subsection 4.3.3), the lower bound is slightly lower by less than 5% while the upper bound is significantly
higher by 20%. These variations come as no surprise since empirical estimations for propellers were
applied and the T-Motor 22x7.4 rotors were specially designed for vertical flight operations, which makes
their geometry different from a standard propeller and their characteristics at 70% of its radius might not
accurately represent the rotor properties; for example, incidence angle of the rotor at 40% of its radius
is higher than the β0.7r used, as well as the chord at this location, if the properties at this location were
used in the preliminary phase, the drag at φ = 0 rad would have increase and thus approximate the
results to the experimental ones Furthermore, one hypothesis for the higher discrepancy in the upper
bound can be related to the lift production at this position, which, consequently, causes an induced drag,
not predicted by the drag formulation considered in the preliminary phase.
6.3 Complete Aircraft Analysis
To define the aerodynamic properties of the UAV, a CAD model of the full aircraft (only the right
half since the symmetry boundary condition is applied) was created to serve as the basis for Fluent
simulations. While the preliminary design indicated that the wing would produce the required lift for
cruise at an angle of 4.5◦ , it is expected that this angle will have to be higher when performing the
simulations in Fluent for two reasons: Figure 6.3a demonstrated that, for a 3D wing model, Fluent
features a lower lift coefficient than XFLR5 for all AoA; when the tail was added to XFLR5, a slight
reduction in CL occurred, a tendency which is prone to be more severe if more components are added
to Fluent. As a result, the wing is set with an incidence of 5.5◦ for the initial UAV CAD model.
Due to the complexity of some components and the connections between them, some simplifications
had to be made. In the mesh generation process, small angles between two surfaces pose some prob-
lems due to interference between the inflation layers of these surfaces. At these locations, a smoother
transition was achieved by creating a chamfer of fillet on the problematic edges. Figure 6.12a displays
the CAD model created in Ansys DesignModeler.
The front gear from Figure 4.3a has small holes, a gap for the spring that acts as a shock-absorber,
66
and a finite distance between the wheels and the gear strut. Modeling this geometry with all these
details requires excessive refinements in the mesh generation, increasing the number of elements and
the computational time without gaining noticeable accuracy in the final result (Verı́ssimo, 2016). The
decision is to simply model the strut and add the wheels attached to it, without any gaps in between.
The same approach was taken for the main landing gear.
The domain where the inlet and outlet boundary conditions are applied is similar to the one used in
section 6.1, forming a C-shape at the symmetry plane with the boundaries over 60 chords away from the
wing of the UAV. Additionally, a block is created, containing the entire aircraft in order to apply a smaller
refinement to the fluid zone closer to the UAV during the meshing process. Both the domain and this
refinement block can be seen in Figure 6.12b.
(a) CAD model of half of the UAV (b) C-shape domain and refinement region near the UAV
Figure 6.12: CAD model of the UAV and the domain considered for Fluent.
The considerations taken for the 3D wing mesh generation process are used for the full aircraft: at
the trailing edge of the wing and tail, a smaller region is created to apply a higher refinement and better
capture the high curvature of the surface, as well as over the trailing edges. Furthermore, at the fuselage
and booms, the surfaces around the connections with other components have a smaller refinement than
the remainder of the component surface to control the stitching of the mesh between components and
create a smoother transition between parts with high discrepancies in the surface cell sizes. Some of
these regions are visible in Figure 6.12a like on the fuselage near the landing gear and camera, and at
the back of the boom.
The base cell size to be used on each surface was chosen according to the general dimensions of
each component in order to divide each one into a similar number of partitions, with the exception of the
wing and tail, which have a smaller cell size to better estimate the production of lift by these parts. To
define the inflation layer properties, Equation (5.1) and (5.2) are applied. Each component has its own
characteristic length, as such, the inflation layer, first layer, and total height are not constant for the entire
aircraft. Table 6.3 indicates the parameters to be applied for the inflation layer to each component. The
+
growth rate of the inflation layer is kept at 1.2, similar to the previous simulations, and ytarget = 0.5 to
have a safety margin in order to guarantee that after the mesh is generated, this value is always smaller
than the unit.
With a baseline mesh defined, a grid independence study was performed. A coarser grid was cre-
67
Table 6.3: Inflation layer properties for each component.
Component Characteristic Length (m) y1 (m) δ (m) Nr. of layers

Wing 0.353 1.02 × 10−5 1.01 × 10−2 30
Tail 0.255 9.99 × 10−6 7.68 × 10−3 28
Fuselage 1.400 2.93 × 10−5 3.04 × 10−2 30
Booms 2.341 3.34 × 10−5 4.58 × 10−2 31
Camera 0.112 9.43 × 10−6 4.02 × 10−3 25
Front Landing Gear Strut 0.015 8.16 × 10−6 8.06 × 10−4 17
Front Landing Gear Wheels 0.060 9.01 × 10−6 2.44 × 10−3 22
Rear Landing Gear Strut 0.035 8.67 × 10−6 1.59 × 10−3 20
Rear Landing Gear Wheels 0.089 9.27 × 10−6 3.35 × 10−3 24
ated, as well as four finer grids. Figure 6.13 shows the lift and drag coefficient obtained for each mesh.
It is important to note that the meshes created have their base cell sizes and surface sizes as a fixed
percentage of the baseline mesh. Due to mesh stitching errors and non-conformities that happened in
the mesh generation process, this percentage does not follow any pattern since some desired cell re-
finements could not be performed. Nevertheless, both aerodynamic coefficients tend to converge as the
number of cells in each mesh increases. The computational time required to obtain a solution was also
a decisive factor for the mesh selection: Table 6.4 indicates an approximation of the required time to ob-
tain a converged solution, with the time exponentially increasing as the mesh elements become smaller.
For reference, the hardware used to perform the CFD simulations had a 6-core 2.20 GHz CPU and a
16 GB RAM. The most refined mesh took around three full days to generate results, and due to time
constraints and computer processing limitations, no finer meshes were possible to be generated and
solved. A compromise between accuracy and computational effort was made, with the author choosing
the fourth mesh to be used in further studies, having a total of 15.02M cells, requiring around one day
to be solved. In terms of accuracy, in comparison with the finer mesh, it has a lower CL value with a
difference of under 0.25% and a higher CD with a discrepancy of under 2%. Although the computational
time might be seen as relatively high, this was the ’lightest’ mesh that had an error low enough to be
acceptable.
The selected mesh yields CL = 0.8076 and CD = 0.05175. These results showed that at 0◦ of AoA
with a wing incidence of 5.5◦ , the UAV does not produce the required lift for cruise, as a result, the wing
incidence is set to 5.75◦ providing CL = 0.8214 while having a CD = 0.05441. Table 6.5 compares the
results from this analysis with the drag build up done on the preliminary phase. The first line of the
table is related solely to the CFD results from the UAV model without the VTOL system incorporated, the
other two lines add the drag coefficient from the wind tunnel tests to the previous CFD result; the drag
coefficient values related to the preliminary phase were computed using Equation (2.2) using the values
obtained at the end of chapter 4, with K constant and CD0 varying according to each case.
Analyzing solely the UAV without the VTOL set reveals that the preliminary drag estimation is smaller
by around 10% compared to the CFD results. Although the empirical expressions used in the drag
build-up phase were carefully chosen according to the shape of the components analyzed, most of the
68
Table 6.4: Properties of the UAV
meshes.
Nr. of Cells Solver Time (h)

5.23 M 5.5
7.05 M 7.5
10.28 M 12.5
15.02 M 23
16.84 M 30.5
22.63 M 72
Figure 6.13: UAV Mesh convergence study.
Table 6.5: Lift and drag coefficients from Fluent for the UAV plus wind tunnel data for the VTOL system.
CFD + Tunnel results (case A) Preliminary results

CD Difference (%)
CL CD CD
UAV - No VTOL 0.8214 0.05441 0.04891 -10.12%

UAV - VTOL Parallel 0.8214 0.06332 0.06006 -5.16%
UAV - VTOL Perpendicular 0.8292 0.07481 0.06627 -11.41%
coefficients mentioned in the literature were obtained from experimental data of components used in
general aviation, those being larger scale components subjected to higher flow speeds (Hoerner, 1965).
Additionally, the preliminary drag build-up fails to account for the interference drag, which occurs due to
the physical connections between structures and boundary layer and flow interference. When adding the
VTOL system, the drag predictions follow the same tendency as presented in section 4.3.3, as expected.
6.3.1 VTOL System Integration
Without a CAD model of the rotors or additional information provided by the supplier, adding these
components to the UAV model poses some difficulties. Silva created a CAD model of the T-Motor
22x7.4 rotor by molding the rotor in wax, after the creation of a negative mold in plaster. The details
of the rotor CAD construction are depicted in his dissertation (P. Silva, 2021). This rotor CAD model
was used, and the motors were approximated as cylinders with the dimensions from T-Motor (2021); a
smooth connection between the motors and the booms was also generated. Regarding the mesh, a
fine refinement was made on the rotor blades to capture its high curvature and an inflation layer was
created. Following the methodology for the inflation layer settings, the rotor characteristic length changes
according to the rotor position relative to the freestream. Since only the maximum and minimum drag
positions will be considered for the full UAV simulation, the characteristic length of the rotors corresponds
to their chord at 70% radius and their diameter, respectively.
69
With the integration of the VTOL system into the CAD model, the converged mesh reached around
19.4M cells, taking around 42 to 48 hours to solve. The results of the simulation are displayed in Table
6.6 with the CFD UAV model having a wing incidence of 5.75◦ and 0◦ AoA. For the preliminary results
presented, the consideration stated in the previous table is valid: for the CL obtained in Fluent, Equation
2.2 outputs the CD value.
Table 6.6: Lift and drag coefficients regarding the full aircraft simulated in Fluent.
CFD results (case B) Preliminary results

CD Difference (%)
CL CD CD
UAV - VTOL Parallel 0.8112 0.06241 0.05944 -4.76%

UAV - VTOL Perpendicular 0.8169 0.07390 0.06553 -11.32%
Although the results from the CFD simulation fail to achieve the required lift for cruise (CLcr = 0.8269),
a small AoA is considered for cruise conditions, which avoids changing the wing incidence once again
and remaking the whole CAD model and mesh. Further analysis confirms this small angle is at most
0.15◦ , to obtain the required lift for cruise with little change to the drag coefficient (around 1 to 1.5%
higher drag due to an increase in the induced component for both cases). Once again, the CFD model
exhibits a higher drag coefficient than what the preliminary studies indicated. These results were ex-
pected since the full UAV model simulated in Fluent is able to account for the interference drag. Addi-
tionally, the drag difference is larger when the rotors are perpendicular, confirming once again the lack
of reliability in the empirical expressions used for the rotors in the preliminary phase.
A comparison between these results and the ones obtained using the UAV in Fluent (without VTOL
system) plus wind tunnel data brings some interesting conclusions. For simplicity the first case will be
denominated case A (UAV model without VTOL in CFD + wind tunnel data for VTOL) and the latter
case B (complete aircraft in CFD). For the same wing incidence and angle of attack, case B presents a
lower lift coefficient. This can be explained by the fact that a portion of the wing is under the influence
of the front rotors’ wake (Figure 6.16), disturbing the airflow reaching the leading edge of the wing and
reducing the local lift production in that region (visible in Figure 6.14 with a reduction in the pressure
difference on the region above the boom and behind the front rotor). While this influence was expected
to be larger whenever the rotors were perpendicular to the airflow, the wind tunnel tests showed that the
rotors also produced lift when they are at this position, thus balancing the wing’s lift losses.
Although the drag coefficient is not directly comparable since the lift coefficient is marginally different,
this difference is so small that some nuances can be deduced. Overall, CD is slightly higher in case A.
One hypothesis for this behavior might be related to the fact that the VTOL system on the wind tunnel
assembly was located at the end of the boom, while in the CAD model, the boom goes beyond each
motor, ending in a dome ahead of the front rotor, and at the back, the boom extends up to the tail
location. This change reduces the wet area of the motor and the supports on the CAD model, thus
reducing the skin-friction drag component. Moreover, in the wind tunnel assembly, the front support had
a flat area perpendicular to the freestream, which is the least aerodynamic shape possible and increases
the pressure drag downstream of this structure (Hoerner, 1965). Furthermore, if a small increase in the
70
Figure 6.14: Pressure distribution over the wing at an AoA of 0◦ .
AoA is considered for case B to match the lift coefficient of both cases, CD will also increase, narrowing
the gap between the two cases.
Non-slender components have a big impact on the total drag, as seen in Figure 6.15 which is a
velocity plot of the flow at the symmetry plane. The optical camera’s and the front landing gear strut’s
wake significantly contribute to the form drag component. Moreover, in the same picture, it is visible
that, on the rear of the fuselage, some detachment and vortex shedding are occurring; in reality, this
phenomena would not occur, at least in these proportions, since the propeller located at this position
would provide an active suction of the flow and prevent the BL from detaching due to the curvature of
the fuselage rear. Figure 6.16 is a velocity contour plot, similar to Figure 6.15, for a plane parallel to the
symmetry plane, located at the booms and rotor. In this figure, the wake produced by the front rotor is
visible and extends over the wing and up to the rear rotor, confirming the hypothesis stated before.
To study the behavior of the aircraft at different attitudes, simulations using the full UAV CAD model
were performed for different AoA to obtain the drag polar. In this study, only the case in which the rotors
are parallel to the booms is considered.
The evolution of UAV lift and pitch moment coefficient with angle of attack is depicted in Figure 6.17a.
The lift curve slope exhibits the typical linear behavior up to an AoA of 4◦ and reaches a maximum value
of 1.31 at 7◦ , after which, stall is bound to occur. This CLmax is smaller than the 1.35 value predicted in
the conceptual phase, which affects the wing loading constraint at stall. To have the same wing loading
as initial thought and considering this updated value for CLmax , stall speed needs to increase from 28
kts to 28.4 kts, according to Equation (3.3). This poses no noteworthy changes since the increment
in the stall speed is small and there is still enough margin for the flight speeds defined in the mission
profile. The moment coefficient around the CG, confirms that the UAV is naturally stable by having a
71
Figure 6.15: Velocity contour plot at the symmetry plane z=0m.
Figure 6.16: Velocity contour plot at the booms z=0.72m.
negative value for CMα . The angle of attack at which the moment coefficient is null is slightly higher than
the one required for cruise, yet, this difference is so small that a null moment coefficient at cruise can be
achieved with a slight deflection of both ruddervators upward, as stated in subsection 4.4.1.
The data retrieved to construct the drag polar, represented in Figure 6.17, is the same as the one in
the previous figure, in which each scatter mark stands for a different AoA, ranging from -6◦ to 7.5◦ . It is
important to note that the minimum drag value can not be seen on the polar since this value occurs at
an angle of attack smaller than the lowest AoA simulated in CFD. This is due to the fact that the wing
incidence was set to 5.75◦ , meaning that on cruise, the wing is effectively operating at an angle of around
6◦ and based on the XFLR5 results, the wing minimum drag was obtained at an angle below -3◦ which
would correspond to an AoA near -9◦ . The drag polar has the typical parabola shape up to CL values of
1.2, with the drag significantly increasing after this point. The maximum lift coefficient corresponds to the
beginning of the stall region, after which the lift coefficient decreases and drag increases. The physical
phenomena related to stall can be seen in Figure 6.6, in which regions of separated flow dominate the
upper wing surface, increasing the form drag produced by the wing.
72
(a) CL alpha and CM alpha curves (b) Drag polar
Figure 6.17: Aerodynamic graphs for the UAV with rotors parallel to booms.
Table 6.7 sums up the conclusions taken from the aerodynamic analysis of the UAV designed with
this data being extracted from the graphs in Figure 6.17, which is related to the best case (rotors parallel
to freestream). Correcting the value of Table 6.6 to match the cruise lift coefficient, CDcruise = 0.07445
for the perpendicular case.
Table 6.7: Aerodynamic properties of the UAV.

Table 6.8: Performance update of the UAV.
CLmax 1.31
AoA for CLmax 7◦
Stall Speed 28.40 kts
Cruise lift coefficient 0.8269
Thrust required for cruise 16.32 – 19.22 N
Cruise AoA 0.15◦
Fuel cell output required for cruise1 654.0 – 756.0 W
Cruise drag coefficient 0.06321
Thrust required for climb 39.21 - 41.94 N
Maximum L/D 13.39
Power output required for climb12 1387 – 1471 W
AoA of maximum L/D 1◦
Total flight time 3h05 – 3h25
Maximum L3/2 /D 13.58
AoA of maximum L3/2 /D 4◦
Table 6.8 uses the results from the final CFD simulations to update the flight performance data
calculated at the end of the preliminary phase. For the thrust, power, and endurance, both the lower and
upper bounds are shown to exhibit the impact of the rotors’ stopping position. These results reinforce,
once again, the necessity of thinking about a stopping mechanism for the rotor in the least drag position,
as this may increase the endurance of the UAV by up to 20 minutes (10% more flight time).
1 This value includes motor required power, motor and ESC efficiencies and avionics and camera power consumption
2 800 W come from the fuel cell and the remaining from batteries attached to the fuel cell
73
74
Chapter 7
Conclusions
7.1 Achievements
At the beginning of this document, the objectives of the project in which this thesis takes part were
defined, consisting of the design of a small UAV powered by a fuel cell and capable of performing VTOL,
with the author focusing its study on aerodynamics and flight performance. This objective was fulfilled
by following a step-by-step approach from the conceptual phase up to the detailed design.
In the initial phase, the author’s contributions to aerodynamic parameters like the Oswald efficiency
factor and the parasite drag coefficient allowed the project members to come up with a viable solution for
the design process. Since the output of the detailed phases of all the project members was in agreement
with the initial design, the conceptual design can be considered a success as it guided the project on
the right track.
The previous two parameters were target of an exhaustive analysis in the preliminary phase, with the
Oswald efficiency factor taking into account the wing geometry defined at this stage, and the parasite
drag the influence of each aircraft component in contact with the airflow. These studies demonstrated
that most of the aircraft’s drag at cruise is due to the parasite component, with the VTOL propulsion
system contributing to this component by around 25 to 40% of the total parasite drag, depending on the
rotor’s stopping position after the vertical take-off segment.
Before starting the detailed design using CFD tools, the turbulence model and mesh general prop-
erties were validated by comparing the results of 2D simulations of the wing’s airfoil with experimental
data. The SST k − ω turbulence model, in conjunction with the γ − Reθ , was found to be the most ap-
propriate for the case, accurately predicting the location of the laminar separation bubble and transition,
as well as the separation phenomena at high angles of attack.
When conducting CFD simulation of the 3D wing model, the results were similar to XFLR5 for AoA
lower than 6◦ with a reduction in lift and an increase in drag for attitudes above these angles. These
simulations allowed the author to understand the post-stall behavior of the wing, which fomented the
application of twist to the wing. This modification managed to slightly reduce the drag produced by the
wing at high AoA and changed the stall behavior by making the root stall before the tips, as pretended.
75
Regarding the VTOL system, the preliminary drag estimates of these components present discrep-
ancies of up to 20% when compared to the wind tunnel data gathered. These differences might be
related to the rotor geometry being different from typical propellers and the properties at 70% radius, not
accurately reflecting the geometry of the rotor. For the UAV without this system modeled, CFD simula-
tions resulted in a drag coefficient 10% higher in contrast with the drag build-up, as CFD tools are able
to account for the interference drag.
The entire aircraft modeling and subsequent mesh generation posed some challenges due to the
UAV complexity and intricate geometries, which resulted in failed grids or grids of poor quality. Some
simplifications and smooth transitions between components had to be made to obtain results in a timely
manner. Additionally, the hardware used to run Fluent was not the most optimal which highly influenced
the time it took to obtain results and limited their accuracy and precision, as finer meshes were unfeasible
to generate and solve. Nevertheless, the results, obtained in Fluent, are similar to the previous case,
which also confirms the impact of the rotor’s stopping position: when the UAV has all rotors perpendicular
to the freestream, it produces 15% more drag in total than when all rotors are parallel to the freestream.
This impacts the total flight time of the UAV by reducing it from 3h25 to 3h05.
The main objective of the present dissertation has been achieved, with the baseline aircraft aerody-
namic properties being computed with high-fidelity tools. The presence of a VTOL system in a conven-
tional configuration is unusual, and the current work makes an important and valuable contribution in
understanding the interference of this system in the cruise flight segment for small UAVs.
7.2 Future Work
The present dissertation featured the aerodynamic characterization of the baseline solution achieved
after the preliminary phase. While simple studies regarding modifications to the wing were conducted
with promising outcomes, due to temporal and computational limitations, their impact on the total aircraft
could not be assessed. Applying wingtip devices may reduce the induced drag component, however,
this addition must be balanced with the increase in the skin-friction drag and additional weight. The
author suggests using a dedicated optimization tool to improve the aerodynamic efficiency of the wing
and study its impact on the whole aircraft.
Although a preliminary longitudinal static stability analysis was performed, it should be extended to
incorporate lateral stability with the definition of the aileron characteristics, together with the dynamic
stability for both motions. CFD simulations could be run to obtain the stability derivatives of the UAV.
During the course of the dissertation, it became evident how much the rotor stopping position impacts
the drag at cruise conditions. A stopping mechanism ought to be thought to block the rotor in a position
where it is aligned with the boom. A mechanical device such as a rod or a rotating arm, which would
act only after the operation of the VTOL system ceased, may be a solution, nevertheless, its drag and
weight impact must be assessed, as well as its effectiveness.
The aerodynamic characterization of the UAV was performed using CFD tools, with its definitions
validated from the 2D analysis performed. Nevertheless, wind tunnel testing of a scale model would
76
allow a comparison between the computational and experimental data.
Regarding the state of the project, a coupling between the aerodynamics studies and the remaining
areas of the project (P. Silva, 2021; Sá, 2021) should be conceived to end up with a UAV design ready for
a prototype to be manufactured and tested. Other areas that were not studied by the project team, such
as flight control, should also be explored. Additionally, the multidisciplinary optimization studies resulted
in a second generation concept with improved endurance and slight changes in the UAV characteristics
(Alves, 2021), and these properties ought to be confirmed using high-fidelity software.
77
78
Bibliography
Aftab, S., Rafie, A., Razak, N., & Ahmad, K. (2016, 04). Turbulence model selection for low reynolds
number flows. PloS one, 11. doi: 10.1371/journal.pone.0153755
Airbus. (2020a). How hydrogen could be a game-changer for future aircraft. Retrieved April 28,
2021, from https://www.airbus.com/newsroom/stories/Is-this-the-next-clean-energy-to
-power-aviation.html
Airbus. (2020b). Hydrogen in aviation: how close is it? Retrieved April 28,
2021, from https://www.airbus.com/newsroom/stories/hydrogen-aviation-understanding
-challenges-to-widespread-adoption.html
Alves, B. (2021). Multidisciplinary optimisation of an unmanned aerial vehicle with a fuel cell powered
energy system (Unpublished master’s thesis). Academia da Força Aérea.
Alves, B., Coelho, V., Silva, P., Marta, A., Afonso, F., Sá, P., . . . Caetano, J. (2021, July). Design of a
hydrogen powered small electric fixed-wing uav with vtol capability. In A. C. Marta & A. Suleman
(Eds.), International conference on multidisciplinary design optimization of aerospace systems
(p. 290-304). Lisbon, Portugal: Aerobest 2021. (ISBN: 978-989-99424-8-6)
Anderson, J. D. (2010). Fundamentals of aerodynamics (5th ed.). McGraw-Hill. Hardcover.
Anderson, J. D. (2012). Introduction to flight. McGraw-Hill.
Ansys Inc. (2010, 11). Ansys meshing user’s guide, release 13.0 [Computer software manual].
Ansys Inc. (2013, 11). Ansys fluent user’s guide, release 15.0 [Computer software manual].
Ansys Inc. (2021). Ansys fluids computational fluid dynamics (cfd) simulation. Retrieved June 18, 2021,
from https://www.ansys.com/products/fluids
Austin, R. (2010). Unmanned aircraft systems: Uavs design, development and deployment. John Wiley
and Sons.
BBC. (2021). Afghanistan: Us says drone strike killed is-k planner. Retrieved September 07, 2021, from
https://www.bbc.com/news/world-asia-58364743
Boukoberine, M. N., Zhou, Z., & Benbouzid, M. (2019). A critical review on unmanned aerial vehicles
power supply and energy management: Solutions, strategies, and prospects. Applied Energy ,
255, 113823. doi: 10.1016/j.apenergy.2019.113823
Brederode, V. (2018). Aerodinâmica incompressı́vel: Fundamentos (2nd ed.). IST Press.
Coelho, D. (2019). Conceptualization and application of unmanned aerial vehicles design methodology
(Master’s thesis, Instituto Superior Técnico). Retrieved from https://fenix.tecnico.ulisboa
79
.pt/downloadFile/1407770020546924/IST MSc Thesis 78423.pdf
Cook, M. V. (2013). Flight dynamic principles: A linear systems approach to aircraft stability and control
(3rd ed.). Elsevier.
Corke, T. C. (2003). Design of aircraft. New Jersey: Pearson Education, Inc.
Dantsker, O., & Vahora, M. (2018, 01). Comparison of aerodynamic characterization methods for design
of unmanned aerial vehicles. 2018 AIAA Aerospace Sciences Meeting. doi: 10.2514/6.2018-0272
Defense Turkey. (2017). Bna tiltrotor uav is flight ready. Retrieved September 12, 2021, from https://
www.defenceturkey.com/en/content/bna-tiltrotor-uav-is-flight-ready-4146
Deperrois, A. (2009). Xflr5 - analysis of foils and wings operating at low reynolds numbers [Computer
software manual].
Drela, M. (1989). An analysis and design system for low reynolds number airfoils. In Conference on low
reynolds number airfoil aerodynamics.
Drela, M., & Youngren, H. (2001). Xfoil 6.9 user primer [Computer software manual].
Eubank, R., & Speckman, P. L. (1990). Curve fitting by polynomial-trigonometric regression. Biometrika,
77 , 1-9. doi: 10.1093/biomet/77.1.1
Fathi, M. K. (2019). Wild area aerial protective inspector, proposal for 2018-2019 undergraduate indi-
vidual aircraft linear infrastructure inspection uas (Tech. Rep.). Tehran, Iran: Sharif University of
Technology.
Federal Aviation Administration. (2016). Pilot’s handbook of aeronautical knowledge. Aviation Supplies
& Academics, Inc.
Força Aérea. (2021). Força Aérea - Missão. Retrieved March 26, 2021, from https://www.emfa.pt/
p-709-mission
Gadalla, M., & Zafar, S. (2016). Analysis of a hydrogen fuel cell-pv power system for small uav. Interna-
tional Journal of Hydrogen Energy, 41(15), 6422–6432. doi: 10.1016/j.ijhydene.2016.02.129
Group, T. E. (2014). Falcon-v fixed wing vtol (vertical take off and landing) uav (unmanned aerial vehicle)
aircraft. Retrieved March 11, 2021, from http://www.top-enggroup.com/product.htm
Gudmundsson, S. (2014). General aviation aircraft design: Applied methods and procedures.
Butterworth-Heinemann.
Gundlach, J. (2012). Designing unmanned aircraft systems: A comprehensive approach. American
Institute of Aeronautics and Astronautics, Inc.
Hirsch, C. (2009). Numerical computation of internal and external flows: Fundamentals of computational
fluid dynamics (2nd ed.). Elsevier.
Hoerner, S. F. (1965). Fluid-dynamic drag. Hoerner fluid dynamics.
Houghton, E., Carpenter, P., Collicott, S. H., & Valentine, D. T. (2013). Aerodynamics for engineering
students (6th ed.). Elseview.
Howe, D. (2000). Aircraft conceptual design synthesis. Professional Engineering.
Ideal Simulations. (2020). Cfd computational domain. Retrieved July 04, 2021, from https://www
.idealsimulations.com/resources/cfd-computational-domain/#
Intelligent Energy. (2021). Our uav products. Charnwood Building, Holywell Park, Ashby Road. Re-
80
trieved from https://www.intelligent-energy.com/our-products/uavs/
Istvan, M. S., Kurelek, J. W., & Yarusevych, S. (2018). Turbulence intensity effects on laminar separation
bubbles formed over an airfoil. AIAA Journal, 56(4), 1335-1347. doi: 10.2514/1.J056453
Joint Air Power Competence Centre - NATO. (2010, 1). Strategic concept of employment for unmanned
aircraft systems in NATO.
Kadyk, T., Winnefeld, C., Hanke-Rauschenbach, R., & Krewer, U. (2018). Analysis and design of fuel
cell systems for aviation. Energies, 11(2), 375. doi: 10.3390/en11020375
Kalitzin, G., Medic, G., Iaccarino, G., & Durbin, P. (2005). Near-wall behavior of rans turbulence models
and implications for wall functions. Journal of Computational Physics, 204(1), 265-291. doi: 10
.1016/j.jcp.2004.10.018
Khan, M. (2014). Quadcopter flight dynamics. International journal of scientific & technology research,
3(8), 130–135.
Langtry, R., Menter, F., Likki, S., Suzen, Y., Huang, P., & Völker, S. (2006, 07). A correlation-based
transition model using local variables—part ii: Test cases and industrial applications. ASME J.
Turbomach, 128. doi: 10.1115/1.2184353
Launder, B., & Spalding, D. (1974). The numerical computation of turbulent flows. Computer Methods
in Applied Mechanics and Engineering, 3(2), 269-289. doi: 10.1016/0045-7825(74)90029-2
Li, B., Zhou, W., Sun, J., Wen, C.-Y., & Chen, C.-K. (2018). Development of model predictive controller
for a tail-sitter vtol uav in hover flight. Sensors, 18(9), 2859. doi: 10.3390/s18092859
Lowry, J. T. (1999). Performance of light aircraft. American Institute of Aeronautics and Astronautics,
Inc.
Mathworks. (2004, 10). Curve fitting toolbox for use with matlab [Computer software manual].
McArthur, J. (2007). Aerodynamics of wings at low reynolds numbers (Doctoral dissertation, University
of Southern California). Retrieved from https://www.kimerius.com/app/download/5784130381/
Aerodynamics+of+wings+at+low+Reynolds+numbers.pdf
Mendes, P. (2021). Design of a fixed-wing tilt-rotor quadcopter class i mini unmanned aircraft - propulsion
system design and prototype manufacturings (Unpublished master’s thesis). Academia da Força
Aérea.
Menter, F. (1994). Two-equation eddy-viscosity turbulence models for engineering applications. AIAA
Journal, 32, 1598-1605. doi: 10.2514/3.12149
Menter, F., Langtry, R., Likki, S., Suzen, Y., Huang, P., & Völker, S. (2006, 07). A correlation-based
transition model using local variables—part i: Model formulation. ASME J. Turbomach, 128. doi:
10.1115/1.2184352
Morgado, J., Santos, A., & Caetano, J. (2017). Portuguese air force research, development and inno-
vation centre (CIDIFA): Rd&i in the area of autonomous unmanned aerial systems. Instituto da
Defesa Nacional.
Moukalled, F., Mangani, L., & Darwish, M. (2015). The finite volume method in computational fluid
dynamics: An advanced introduction with openfoam and matlab (1st ed.). Springer Publishing
Company, Incorporated.
81
Mrusek, B. (2021). Vertical takeoff and landing (vtol) small unmanned aircraft systems (suas). Trends
in Technical & Scientific Research, 5(1). doi: 1019081/TTSR.2021.05.555651
Mueller, T. J., & DeLaurier, J. D. (2003). Aerodynamics of small vehicles. Annual Review of Fluid
Mechanics, 35(1), 89-111. doi: 10.1146/annurev.fluid.35.101101.161102
NASA Open Source Agreement. (2021). Openvsp. Retrieved May 21, 2021, from http://openvsp.org/
wiki/doku.php
National Oceanic Atmospheric Administration. (1976). U.s. standard atmosphere (Tech. Rep.). Wash-
ington D.C., USA: NASA.
Ng, W., & Datta, A. (2019). Hydrogen fuel cells and batteries for electric-vertical takeoff and landing
aircraft. Journal of Aircraft, 56(5), 1765–1782. doi: 10.2514/1.C035218
Niţă, M., & Scholz, D. (2012). Estimating the Oswald Factor from Basic Aircraft Geometrical Parameters.
In German aerospace congress 2012.
OpenCFD Ltd. (2021). Openfoam. Retrieved June 18, 2021, from https://www.openfoam.com/
Park, S., Bae, J., Kim, Y., & Kim, S. (2013). Fault tolerant flight control system for the tilt-rotor uav.
Journal of the Franklin Institute, 350(9), 2535–2559. doi: 10.1016/j.jfranklin.2013.01.014
Peerlings, B. (2018). A review of aerodynamic flow models, solution methods and solvers – and their
applicability to aircraft conceptual design (Tech. Rep.). Delft, Netherlands: Delft University of
Technology.
Prisacariu, V. (2017). The history and the evolution of uavs from the beginning till the 70 s. In Journal
of defense resources management. Air Force Academy, Brasov, Romania.
Pursee, P. E., & Campbell, J. P. (1944). Experimental verification of a simplified vee tail theory and
analysis of available data on complete models with vee tails (Tech. Rep. No. 823). UNT Libraries
Government Documents Department: NACA.
Raymer, D. P. (1992). Aircraft design: A conceptual approach. American Institute of Aeronautics and
Astronautics, Inc.
Roskam, J., & Lan, C.-T. E. (2016). Airplane aerodynamics and performance. Design, Analysis and
Research Corporation.
Sadrey, M. H. (2013). Aircraft design - a systems engineering approach. John Wiley & Sons, Ltd.
Schlichting, H. (1979). Boundary-layer theory (7th ed ed.). McGraw-Hill.
Selig, M. (1997). Summary of low speed airfoil data vol. 3. SoarTech Publications.
Serrano, A. R. (2018). Design methodology for hybrid (vtol+ fixed wing) unmanned aerial vehicles.
Aeronautics and Aerospace Open Access Journal, 2(3), 165–176. doi: 10.15406/aaoaj.2018.02
.00047
Siemens. (2021). Simcenter star-ccm+. Retrieved June 18, 2021, from https://www.plm.automation
.siemens.com/global/en/products/simcenter/STAR-CCM.html
Silva, N. (2014). Parametric design, aerodynamic analysis and parametric optimization of a solar uav
(Master’s thesis, Instituto Superior Técnico). Retrieved from https://fenix.tecnico.ulisboa
.pt/downloadFile/395146454348/Thesis.pdf
Silva, P. (2021). Projeto detalhado do sistema propulsivo para um veı́culo aéreo não tripulado com uma
82
célula de combustı́vel de hidrogénio (Unpublished master’s thesis). Academia da Força Aérea.
Sinha, P. (2013, 12). Multivariate polynomial regression in data mining: Methodology, problems and
solutions. International Journal of Scientific and Engineering Research, 4.
Spalart, P., & Allmaras, S. (1992). A one-equation turbulence model for aerodynamic flows. AIAA, 439.
doi: 10.2514/6.1992-439
Stettmaier, H. (1999, November). V-tails for aeromodels.
Swafford, T. (1983). Analytical approximation of two-dimensional separated turbulent boundary-layer
velocity profiles. AIAA Journal, 21, 923-926. doi: 10.2514/3.8177
Sá, P. (2021). Projeto detalhado estrutural de veı́culo aéreo não tripulado com uma célula de com-
bustı́vel de hidrogénio (Unpublished master’s thesis). Academia da Força Aérea.
T-Motor. (2021). T-motor website. Retrieved June 11, 2021, from https://store-en.tmotor.com/
UAV Factory. (2021). Penguin B UAV. Retrieved April 30, 2021, from https://uavfactory.com/en/
penguin-b-uav
UAVision. (2020). OGS42VN aircraft flight manual. (Unpublished confidential document)
UAVision. (2021). Ogassa ogs42v. Retrieved May 04, 2021, from https://www.uavision.com/ogassa
-ogs42v
UIUC Department of Aerospace Engineering. (2020). UIUC Airfoil Coordinates Database. Retrieved
April 14, 2021, from https://m-selig.ae.illinois.edu/ads/coord database.html
United Nations Climate Change. (2018). Contribution of the global aviation sector to achieving paris
agreement climate objectives. Retrieved October 03, 2021, from https://unfccc.int/sites/
default/files/resource/156 CAN%20ICSA%20Aviation%20TD%20submission.pdf
U.S. Department of Energy. (2021). Alternative fuels data center - hydrogen basics. Retrieved November
04, 2021, from https://afdc.energy.gov/fuels/hydrogen basics.html
US Naval Research Laboratory. (2018, April). The hybrid tiger: A long endurance solar/fuel cell/soaring
unmanned aerial vehicle. In Doe annual merit review, inter-agency session.
UTC Aerospace Technology. (2015, March). Cloud cap technology TASE150 - datasheet.
Versteeg, H. K., & Malalasekera, W. (1995). An introduction to computational fluid dynamics - the finite
volume method. Addison-Wesley-Longman.
Verı́ssimo, R. (2016). Best practice guidelines in cfd external aerodynamics: Applied to unmanned aerial
vehicles at cruise conditions (Unpublished master’s thesis). Academia da Força Aérea.
Walters, D. K., & Cokljat, D. (2008, 10). A Three-Equation Eddy-Viscosity Model for Reynolds-Averaged
Navier–Stokes Simulations of Transitional Flow. Journal of Fluids Engineering, 130(12). doi:
10.1115/1.2979230
Wang, A., & Chan, T. (2017, 11). Estimation of drag for a quadplane hybrid unmanned aerial vehicle. In
2017 aiaa student conference region vii-au.
Wilcox, D. (2006). Turbulence modeling for cfd. DCW Industries.
Wu, M., Han, Z., Nie, H., Song, W., Clainche, S. L., & Ferrer, E. (2019). A transition prediction method
for flow over airfoils based on high-order dynamic mode decomposition. Chinese Journal of Aero-
nautics, 32(11), 2408-2421. doi: 10.1016/j.cja.2019.03.020
83
84
Appendix A
Drag from Wind Tunnel testing of the

VTOL Propulsion System
The balance software outputs the mean value of the forces and moments and the standard deviation
from a set of values obtained during the the test session. Table A.1 displays the mean force in the
freestream direction produced by the VTOL system which was obtained by subtracting the values from
the control test with only the boom and the balance, to the ones with the full assembly, and Table A.2
contains the standard deviation calculated by square rooting the sum of the squares of the standard
deviations from the control and each test case.
Table A.1: Drag mean values Table A.2: Standard deviation values
Rear Rotor Angle (rad) Standard Rear Rotor Angle (rad)

Mean Value (N)
−π/4 0 π/4 π/2 Deviation (N) −π/4 0 π/4 π/2
Front −π/4 2.56 2.89 2.49 2.12 Front −π/4 0.142 0.127 0.135 0.142
Rotor 0 2.91 3.21 2.82 2.44 Rotor 0 0.120 0.150 0.120 0.127
Angle π/4 2.84 3.18 2.76 2.39 Angle π/4 0.135 0.142 0.135 0.127
(rad) π/2 1.81 2.14 1.72 1.4 (rad) π/2 0.158 0.150 0.120 0.135
The force values in the vertical direction (positive values indicate generation of lift) are displayed in
Table A.3 and A.4 From the combinations tested, the one which presented the lowest value was the one
with both rotors at π/2 rad. Since the weight of the tested components remained the same throughout
the experiment, higher values than this case can only mean that there was a higher lift generation. As a
result, this case was assumed to be the reference by having zero lift while all the remaining cases had a
positive lift generation.
Table A.3: Lift mean values Table A.4: Lift standard deviation values
Rear Rotor Angle (rad) Standard Rear Rotor Angle (rad)

Mean Value (N)
−π/4 0 π/4 π/2 Deviation (N) −π/4 0 π/4 π/2
Front −π/4 1.97 2.06 0.98 0.85 Front −π/4 0.184 0.198 0.166 0.161
Rotor 0 2.07 2.00 0.86 0.92 Rotor 0 0.161 0.191 0.166 0.172
Angle π/4 2.88 3.3 1.87 1.85 Angle π/4 0.184 0.178 0.166 0.166
(rad) π/2 1.19 1.31 0.17 0.00 (rad) π/2 0.184 0.191 0.166 0.184
85
Whenever the data does not present a simple linear regression, a polynomial regression can be
applied for single variable or multi variable cases (Sinha, 2013). For the problem in question, the stopping
position of the front rotor is independent from the rear one, and each one has a different impact on the
total drag produced. A second order multiple polynomial regression can be expressed as:
y = β0 + β1 x1 + β2 x2 + β11 x21 + β22 x22 + β12 x1 x2 + (A.1)
,with β1 , β2 the linear effect parameters, β11 , β22 the quadratic effect parameters, β12 the interaction effect
parameters and the error component relative to the higher order terms (Sinha, 2013).
With both independent variables being angles and producing results with a period of π as the rotors
have two blades, it comes to mind using a trigonometric function to express the drag as a function of
both rotor angles. A single variable polynomial-trigonometric regression is given by:
d
X λ
X
y = b0 + bj x j + [cj cos(jx) + sj sin(jx)] (A.2)
j=1 j=1
, where the bj , cj and sj are obtained by regression on the y data (Eubank & Speckman, 1990). Since
the period of the function is known, this regression can be slightly altered and updated to give better
estimates by introducing an horizontal offset to the trigonometric terms.
Due to the low number of data gathered, a second or higher order regression produces unrealistic
results as the drag predictions for untested cases are not physically reliable: for instance, it is expected
the drag to always increase while going from 0 to π/2 rad, however a second order trigonometric function
completely adapts to all points considered, with all data being a local minimum/maximums, which is
unrealistic for drag when φ = ±π/4 rad. As a result a first order multi-variable polynomial-trigonometric
function is employed.
With all the considerations done, the final general expression for the curve-fitting expression to be
applied is in the form:
z = a + b1 2φ1 + b2 2φ2 + c1 cos(2φ1 + f1 ) + d1 sin(2φ1 + g1 ) + c2 cos(2φ2 + f2 ) + d2 sin(2φ2 + g2 ) (A.3)
, with all the constant to be computed based on the z data. To get the regression for the drag values,
the MATLAB curve-fitting toolbox was used and the coefficients were obtained (Mathworks, 2004). The
terms b1 and b2 were in the order of 10−7 which is negligible when compared to the remaining constants,
therefore, they were removed from the curve-fitting expression without compromising the final result
since the R2 remained the same. The final expression is then retrieved and expressed in Equation (6.4).
86

Vasco Coelho - Master

Uploaded by

Copyright:

Available Formats

Vasco Coelho - Master

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vasco Coelho - Master

Uploaded by

Copyright:

Available Formats

AIR FORCE ACADEMY

Aerodynamic Detailed Design of an Unmanned Aerial

Vasco Luís Martins Ferreira Coelho

Thesis to obtain the Master of Science Degree in

Military and Aeronautical Sciences - Aeronautical

Sintra, December 2021

Palavras-chave: Aerodinâmica, Célula de Combustı́vel, Descolagem e Aterragem Vertical,

Keywords: Aerodynamics, Computational Fluid Dynamics, Fuel Cell, Hydrogen, Unmanned

2 Literature and Technology Review 7

6 Detailed Design – Aerodynamic Performance 55

A Drag from Wind Tunnel testing of the VTOL Propulsion System 85

1.1 Design proposal presented by CIAFA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

5.1 2D 91k mesh parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

6.1 Properties of the meshes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

A.1 Drag mean values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

1.1 Market study of possible hybrid UAV configurations. . . . . . . . . . . . . . . . . . . . . . 2

3.1 UAV mission profile. Adapted from Alves et al. (2021). . . . . . . . . . . . . . . . . . . . . 27

4.1 Wing Baseline Geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

5.1 2D Mesh with the detail on the inflation layer. . . . . . . . . . . . . . . . . . . . . . . . . . 48

6.1 3D Wing mesh convergence study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

AFA Academia da Força Aérea

AoA Angle of Attack

CAD Computer Aided Design

CAE Computational Aided Engineering

CFD Computational Fluid Dynamics

CIAFA Centro de Investigação da Academia da Força Aérea

DNS Direct Numerical Simulation

ESC Electronic Speed Controller

FAP Força Aérea Portuguesa

FDM Finite Difference Method

FEM Finite Element Method

FVM Finite Volume Method

ICE Internal Combustion Engine

LES Large Eddy Simulator

LLT Lifting Line Theory

LRN Low Reynolds Numbers

LSB Laminar Separation Bubble

MTOM Maximum Take-Off Mass

PITVANT Projeto de Investigação e Tecnologia em Veı́culos Aéreos Não-Tripulados

RANS Reynolds Average Navier-Stokes

RPM Rotations per Minute

SIMPLE Semi-Implicit Method for Pressure Linked Equations

SST Shear-Stress Transport

UAV Unmanned Aerial Vehicle

UIUC University of Illinois at Urbana-Champaign

VLM Vortex Lattice Method

VTOL Vertical Take-Off and Landing

β0.7r Rotor Incidence Angle (at 70% radius).

δ Maximum Boundary Layer Thickness.

δ∗ Boundary Layer Displacement Thickness.

 Dissipation Rate of Turbulent Kinetic Energy.

µ Molecular Viscosity Coefficient.

νt Kinematic Eddy Viscosity.

ω Specific Dissipation Rate of Turbulent Energy.

φ Rotor Stopping Position.

τw Wall Shear Stress.

ϕ0.25c Sweep Angle (at 25% Chord).

e Oswald Efficiency Factor.

ef Finite Wing Efficiency Factor.

Dissipation Rate of Turbulent Kinetic Energy.