Aerodynamic Design and Optimization of LR MUAV
Aerodynamic Design and Optimization of LR MUAV
Aerodynamic Design and Optimization of LR MUAV
12-02-2010
12-02-2010
The undersigned hereby certify that they have read and recommend to the Faculty of
Aerospace Engineering for acceptance a thesis entitled “Aerodynamic Design and
Optimization of a Long Range Mini-UAV” by Dennis Trips B.Sc. in partial
fulfillment of the requirements for the degree of Master of Science.
Dated: 12-02-2010
Supervisor:
MSc. M.H. Straathof
Second supervisor:
MSc. ing. B.D.W. Remes
Reader:
Ir. W.A. Timmer
Acknowledgements
First of all I want to thank my supervisor Michiel Straathof for always providing assistance
whenever needed and for dealing with the tedious task of correcting this report. I also
wish to address my gratitude towards the people of the MAVLab. Bart Remes and Rick
Ruijsink who provided a lot of guidance and experience during the practical parts of this
assignment. Christophe de Wagter for helping out with the autopilot and for flying the
mini-UAV in manual mode. Guido de Croon for helping in organizing the EMAV2009
and providing ideas on optimization tools.
I also wish to thank Bart Bruggeman, Vinodh Gunaseelan, Remco Zwinderman, Tamara
Doornbos and Ritesh Sharma for helping out with building the models, balance and
aircraft. Without their help this thesis would not have been possible.
Thanks also goes out to the guys in “de put” for the great moments spent together in our
office and to Sejla Lagumdzija for designing the front cover of this thesis.
Last but not least I want to thank my parents for always trying to help the best they can
and for giving me the opportunity to become an aerospace engineer.
I wish to dedicate this thesis to my brother, Benny Trips.
v
Abstract
vii
Contents
Acknowledgements v
Abstract vi
Nomenclature xix
1 Introduction 1
2 Assignment Description 3
2.1 Thesis Project Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Theory 5
3.1 Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.1 Lift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.2 Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.1.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 The Optimization Algorithm . . . . . . . . . . . . . . . . . . . . . 34
3.2.2 Variables and Constraints . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.3 The Optimization Objective . . . . . . . . . . . . . . . . . . . . . . 38
ix
x Contents
6 Results 125
6.1 Optimization Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.1.1 Optimization with Tight Constraints and the Prototype as an Initial
Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.1.2 Optimization with Tight Constraints and an Inferior Initial Point . 131
6.1.3 Optimization with Relaxed Constraints . . . . . . . . . . . . . . . 134
6.2 The Competition Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.2.1 Geometric and Structural Details . . . . . . . . . . . . . . . . . . . 137
6.2.2 Aircraft Components . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.2.3 Aerodynamics and Performance Details . . . . . . . . . . . . . . . 142
D 3-Views 161
xii Contents
List of Figures
3.1 Values of Kv plotted against the aspect ratio corrected for the mach number
(figure is extracted from reference [16] page 28). . . . . . . . . . . . . . . . 7
3.2 Lift coefficient comparison between equation 3.1 and wind tunnel data for
a thick straight wing at different aspect ratios (figure is extracted from [16]
page 48). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Comparison between a potential method, equation 3.1 and Fluent for a 20◦
sweep, A = 3 wing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4 Comparison between a thick airfoil, turbulent, viscous, lift curve and a thin
airfoil inviscid lift curve. Both calculations are performed with Xfoil. . . . 9
3.5 Overview of the different drag contributions and the defenition of CDref
and CLref . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.6 Visualization of the angles and forces present in the quasi-3D method to
combine a inviscid lift distribution with viscous airfoil data. (L’ = local
lift force from lift distribution, l = viscous airfoil lift force (at αef f ), D′ =
local total (profile and induced) drag force, d = airfoil profile drag force
(at αef f )). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.7 Definition of the forces and angles used to determine the inviscid downwash
angle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.8 Illustration of a laminar separation bubble on an airfoil (picture extracted
from ref. [7] page 74). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.9 Pressure arrows for an airfoil at Re = 3000000 (no bubble) and for Re =
500000 (bubble present). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.10 Edge velocity and momentum loss jumps at reattachment (figure copied
from ref. [12] page 725). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.11 Illustration of a typical pressure gradient for Low Reynolds number airfoils. 20
3.12 Effect of the transition location on bubble size and momentum loss (figure
copied from ref. [12] page 726). . . . . . . . . . . . . . . . . . . . . . . . . 21
3.13 Friction drag coefficient for plane surfaces plotted against the Reynolds
number (page 2-6 reference [15]). . . . . . . . . . . . . . . . . . . . . . . . 22
xiii
xiv List of Figures
3.14 Parasitic interference drag in the corner between a wing and a fuselage for
different longitudinal locations of the wing/vertical tail (page 8-15 reference
[15]). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.15 Illustration of forces and moments involved in trimming an aircraft (L =
lift, D = drag, M = aerodynamic moment, W = aircraft weight). . . . . . 30
3.16 Evolution from a normal configuration to a swept tailless aircraft. . . . . . 32
3.17 Forces and moments contributing to the wing moment coefficient. . . . . . 32
6.1 Comparison between the initial wing geometry and the resulting geometry
after the optimization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2 Lift over drag ratio for different velocities, maintaining straight, level and
trimmed flight. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.3 Aerodynamic coefficients for the optimized wing at 19.1m/s. . . . . . . . 130
6.4 Comaprison between optimized wing from section 6.1.1(red) and the opti-
mization of this section (green). . . . . . . . . . . . . . . . . . . . . . . . . 132
6.5 Aerodynamic coefficients for the optimized wing at 19.3m/s. . . . . . . . 133
6.6 Comaprison between optimized wing from section 6.1.1 (upper) and the
optimization of this section (lower). . . . . . . . . . . . . . . . . . . . . . . 135
6.7 Aerodynamic coefficients for the optimized wing. . . . . . . . . . . . . . . 136
6.8 polars for the S5010 airfoil at a Reynolds number of 2 · 105 . . . . . . . . . 137
6.9 General overview of the electric layout for the EMAV2009 competition
aircraft. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.10 Autopilot build into the prototype aircraft. . . . . . . . . . . . . . . . . . 139
6.11 Range plotted against the battery capacity(ranging from 0-200 Ah), for
a fixed aircraft shape with a varying weight with respect to the battery
capacity, and a 5W subsystem power consumption. . . . . . . . . . . . . . 141
6.12 Range plotted against the battery capacity(ranging from 0-15 Ah), for
a fixed aircraft shape with a varying weight with respect to the battery
capacity, and a 5W subsystem power consumption. . . . . . . . . . . . . . 142
6.13 Subsystem locations in the aircraft. . . . . . . . . . . . . . . . . . . . . . . 144
6.14 Aerodynamic coefficients for the TRIP50 aircraft at 20m/s. . . . . . . . . 145
3.1 Comparison between calculated and experimental drag coefficients for stream-
lined bodies of revolution (drag coefficients are based on the fuselage wetted
area). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Comparison between experiments/CFD and the calculated aerodynamic
center using equation 3.32 and 3.36. . . . . . . . . . . . . . . . . . . . . . 29
3.3 Flying plank airfoils with corresponding moment coefficients and zero lift
angles of attack. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1 Comparison of (CL /CD )max and its corresponding lift coefficient, between
CFD and the quasi-3D method. . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2 Comparison of (CL /CD )max and its corresponding lift coefficient, between
experimental results and the quasi-3D method. . . . . . . . . . . . . . . . 74
5.1 Overview of components involved in setting the pitch angle and their like-
lihood of causing errors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.2 Wing geometries of the two prototypes. . . . . . . . . . . . . . . . . . . . 114
C.1 Force range and resolution information for the Nano 17 SI-25-0.25. . . . . 159
C.2 Calculated stiffness information for the Nano 17 SI-25-0.25. . . . . . . . . 159
xvii
xviii List of Tables
Nomenclature
Latin Symbols
A Aspect ratio [−]
Atun Wind tunnel aspect ratio [−]
b Wing Span [m]
b∗ Fuselage width at largest cross-section [m]
C Battery Capacity [Ah]
c Chord length [m]
C′ Wind tunnel cross-sectional area of the test section [m2 ]
cavg Average geometric wing chord [m]
CD Aircraft drag coefficient [−]
C Db Induced interference drag (referred to Scovered ) [−]
C Dc Corrected drag coefficient [−]
C Di Induced drag coefficient [−]
C Dp Profile drag coefficient [−]
C Ds Secondary drag coefficient [−]
CDf us Fuselage drag coefficient [−]
CDinttail Fuselage-tail interference drag [−]
CDint Interference drag coefficient [−]
CDref Reference drag coefficient [−]
CDstrip Drag coefficient calculated with the strip method [−]
CDtail Tail drag coefficient [−]
CDtotw Total wing drag coefficient [−]
xix
xx Nomenclature
Greek Symbols
α Angle of attack [◦ ]
α0 Zero lift angle of attack [◦ ]
Nomenclature xxiii
Abbreviations
DAR Design integration and operation of Aircraft and Rotorcraft
EMAV European Micro Aerial Vehicle competition
LTT Low speed Low Turbulence wind Tunnel
xxiv Nomenclature
Introduction
In the last decade a lot of UAV’s have been developed for military applications. These
UAV’s are mostly very large and have performance capabilities equal to or better than
manned aircraft. However a trend is observed towards smaller, mini-UAV’s, that serve the
purpose of being an “eye in the sky” for information gathering and reconnaissance mis-
sions. These applications might also serve civil purposes, e.g. assisting fire-fighter crews,
power line inspections and filming sports events. Miniaturization of electronics and major
leaps in battery technology have pushed the development of mini-UAV’s towards practi-
cal applications. However the development of knowledge on designing high performance
aircraft in this field has still a long way to go. Up until now, the aerodynamics for mini-
UAV’s is hardly understood as remote controlled toy planes were the only beneficiaries
of this knowledge. The available computational aerodynamics tools, including advanced
CFD methods, are incapable of predicting the low Reynolds number(< 2.5 · 105 ) flow
phenomena on wings. The best tool available at this moment is Xfoil which provides a
reasonable approximation of the low Reynolds number effects on airfoils.
This thesis will focus on the aerodynamic design of a mini-UAV that will compete in
the EMAV2009 (European Micro Aerial Vehicle Competition 2009) endurance mission.
The mission consists out of several phases but the focus lies on flying as far as possible.
To obtain this goal an aerodynamic tool will have to be developed for the analysis of the
aerodynamic forces acting on the wing. These forces are required for obtaining information
on the aerodynamic efficiency of the aircraft and to determine its range capabilities.
An optimization algorithm will be added on top of this analysis tool to establish the
best possible preliminary design. It was decided that a tailless aircraft configuration
would be the starting point of the design, putting the focus on the wing design. This
requires that the neutral point of the aircraft and the moment acting on it will have to be
predicted very accurately to establish stable trimmed flight. Literature on wind tunnel
force measurements and advanced CFD methods will be used to validate the data.
Wind tunnel tests are required to validate the results obtained with the wing optimiza-
tion tool. Two models will be tested in the low-speed, low-turbulence wind tunnel of
the TU Delft. The small forces acting on mini-UAV’s require a new balance for force
1
2 Introduction
measurements. Especially the drag force is really small, but it has to be determined very
accurately to establish a validation for the aerodynamic efficiency.
Several DAR-exercises will be performed on the practical side of the thesis work. A sum-
mary of these exercises will be presented throughout this document. Those assignments
consist of: milling molds, building wind tunnel models, designing and building a new
wind tunnel balance and designing the propulsion system for the competition aircraft.
In chapter 2 a detailed description of the assignment is presented and a project goal
is formulated. This is followed by chapter 3 with an in dept theoretical investigation to
establish the tools required for the creation of the aerodynamic analysis program. Next, a
short introduction into the theory behind the sequential quadratic programming algorithm
is presented. This algorithm will be used to perform the optimization. Several objective
functions will be formulated for the design of electrically powered aircraft. The final
implementation of this knowledge is presented in chapter 4 together with an extensive
validation of the developed aerodynamics analysis tool. Details on the preparations for
and the results of the wind tunnel experiments will be presented in chapter 5. This is
followed by an analysis of the results obtained with the optimization algorithm, in chapter
6. Finally, in chapter 7, the conclusions and recommendations will be formulated.
Chapter 2
Assignment Description
In this chapter, the thesis assignment is summarized, and slightly adapted from the orig-
inal one. A project goal will be formulated which will function as a guideline for this
thesis. The original assignment description can be found in appendix A.
The main goal of the thesis work is to find an optimal aerodynamic shape for a mini-UAV
that will participate in the EMAV2009 competition. The requirements of this competition
put certain restrictions on the maximum size of the aircraft and, for safety reasons, on its
momentum. The aircraft should fit, in all configurations, in a sphere of 50 cm diameter
and the maximum momentum should be lower than 20 kgs m (for detailed competition
rules, see ref [2]). The objective of the competition is to fly as far as possible resulting in
a maximization of the lift over drag ratio.
The proposed use of FLUENT is partially discarded due to its inability to change the
geometry and grid in an automated manner, its inability to predict the location of the
transition point, and its rather high calculation times. Instead the option of combining a
3D vortex lattice method with a 2D panel method with an added boundary layer will be
studied. However, validation of this method can be done by comparing it with FLUENT
results and experimental data.
For the optimization it is possible to use different options. As there are metaheuristic
algorithms, gradient based optimizers, and the adjoint-equation method. The adjoint
equation method is an elaborate method and will probably take longer than a single
thesis subject to program.
While optimizing for range it has to take into account that a stable aircraft should be
created with sufficient internal space for the electronics. This is to ensure a feasible design
with flight characteristics that can be handled by the autopilot.
All this can be summarized in the following project goal.
3
4 Assignment Description
Project Goal
Design of a mini-UAV for the EMAV 2009 outdoor competition using a novel
aerodynamic optimization tool, with the objective to win the competition.
Chapter 3
Theory
In this chapter an explanation will be presented on the aerodynamics that one can expect
when working with a tailless mini-UAV’s. This is then followed by an explanation on the
theory behind the sequential quadratic optimization tool, used to obtain the best aerody-
namic shape. A lot of the information in this chapter is gathered from different sources
and is synthesized here to provide background information for subsequent chapters.
3.1 Aerodynamics
The focus of this section is on low Reynolds number, low aspect ratio (LAR) wing aero-
dynamics. The reason to focus on this aerodynamic field lies in the requirements and
logical limits set forward for mini-UAV’s. Low Reynolds numbers are the result of the
low speeds in combination with the small wing chords, while the LAR (< 4) is the result of
a combination of different factors, being: high wing loadings due to high aircraft weight,
stall speed requirements and limits on the wing span.
3.1.1 Lift
Low aspect ratio wings generate lift due to two effects. As for all wings, lift is generated
by the circulation around the wing, also referred to as linear lift because of its linear
nature with respect to the angle of attack. A second non-linear lift is due to vortex lift.
This vortex lift can be produced on the leading edge of the wing or at the side edges of
the wing. The leading edge vortex is generated when the wing is highly swept and when
it has a sharp leading edge. The side edge vortex is present on all wings but its effect is
only noticed in the wake, while in the case of low aspect ratios (≤ 3.5), the vortex will
have its effect on the wake and the wing. The prediction of side vortex flow characteristics
is presented in reference [17].
5
6 Theory
For a tailless aircraft configuration the assumption is made that the only part of the
aircraft generating lift is the main wing. This lift is thus composed of the above mentioned
components and one way of approximating it is by using:
This equation was created by Polhamus in reference [22]. The potential lift coefficient,
CLp , is dependent on the value of Kp which can be determined by a potential flow cal-
culation at a single angle of attack. However, for this report the determination of the
potential lift, at every angle of attack, will be calculated using a vortex lattice method,
AVL. The reason for this is that in a latter stage not only the lift coefficient is required,
but also the potential lift distribution.
The vortex lift coefficient depends on the angle of attack and the value of Kv . This
vortex parameter is the sum of the vortex parameter for the side edge and the leading
edge vortex. To generate a leading edge vortex it is required to have separated flow over
the leading edge which reattaches aft on the wing. However in the case of mini-UAV’s
the airfoils will have significant leading edge thickness, which prevents flow separation
on the leading edge. This implies that the leading edge vortex is not expected on the
mini-UAV wing and will thus not be included in the calculations. Whenever referred
to the vortex parameter, Kv , in this report, it will only be due to the side edge vortex
(in literature referred to as Kv,se ). The value of Kv for straight, untapered wings can be
determined using figure 3.1 (this figure is extracted from reference [16] page 28). For swept
tapered wings the calculation of Kv is much more complicated as can be seen in reference
[16]. A way to efficiently implement this has not yet been found and thus remains to be
determined. The simplification of using the value of Kv for straight wings and linearly
scaling it for the taper ratio, is acceptable as the effect of the side edge vortex is already
small for the aspect ratios under consideration.
In several reports, e.g. reference [16] and [20], this method of estimating the vortex lift
effect has been compared to CFD and experimental results. It proved to work really
well for different low aspect ratio wings with or without leading edge vortex. One of
the comparisons can be seen in figure 3.2, in which results from equation 3.1 are plotted
together with wind tunnel results for a straight wing with a 12.6% thick Göttingen 409
airfoil for different aspect ratios.
In figure 3.3 a comparison is made, for a wing with an aspect ratio equal to 3 and
20◦ sweep, between a potential flow method (vortex lattice), potential flow with vortex
correction and a Fluent calculation. From this figure it can be seen that the difference
between the potential flow and Fluent increases with increasing angle of attack, while
with the vortex correction the difference is drastically reduced. When assuming Fluent
is accurate, this illustrates that the vortex lift correction shows an increased accuracy
in the estimation of the lift curve slope. The remaining difference can have different
causes as there are, the inviscid and thin wing assumptions for the vortex lattice method,
and maybe a small error in the vortex parameter, Kv , for swept wings. The thin wing
assumption causes the vortex lattice method to wrongly estimate the zero lift angle of
3.1 Aerodynamics 7
Figure 3.1: Values of Kv plotted against the aspect ratio corrected for the mach number
(figure is extracted from reference [16] page 28).
attack by about 0.3◦ , this explains the constant difference between the curves. In chapter
4 a correction for this difference will be presented. The remaining difference in the lift
curve slope is rather small and is also caused by the assumptions of a thin wing in inviscid
flow. This is clearly demonstrated when we compare airfoils in Xfoil submitted to the
same conditions and assumption as in Fluent and AVL. In figure 3.4 the lift coefficient of
the airfoil used in the wings for figure 3.3, is plotted against the angle of attack. The lift
coefficient is plotted for the case of a thick airfoil with forced turbulent flow over it at a
Reynolds number of 200000, illustrating the Fluent case. The other graph is for the same
airfoil but with its thickness to chord ratio reduced to 1% (originally 10%), simulating
a thin airfoil, and the calculation was performed for a invicid flow, this compares to the
AVL calculation. Two things become clear when looking at figure 3.4 and illustrate the
differences found between Fluent and AVL. The zero lift angle of attack for the thin
airfoil is 0.36◦ lower than for the thick airfoil in viscous flow, almost exactly the same
difference as mentioned earlier in the 3D-case. The lift curve slope for the thick airfoil is
higher as for the thin airfoil, this is also what was expected by looking at figure 3.3. So
the differences in figure 3.3, remaining after the vortex lift correction, are caused by the
assumptions made in the vortex lattice method.
8 Theory
Figure 3.2: Lift coefficient comparison between equation 3.1 and wind tunnel data for a
thick straight wing at different aspect ratios (figure is extracted from [16] page
48).
Figure 3.3: Comparison between a potential method, equation 3.1 and Fluent for a 20◦
sweep, A = 3 wing.
3.1 Aerodynamics 9
Figure 3.4: Comparison between a thick airfoil, turbulent, viscous, lift curve and a thin
airfoil inviscid lift curve. Both calculations are performed with Xfoil.
10 Theory
3.1.2 Drag
In this section an overview will be presented on the aerodynamic drag of tailless mini-
UAV’s. The performance estimation, which is of major importance for the optimization
(see section 3.2.3), will largely depend on the correct calculation of the drag. As an
introduction, some attention will be spend on how all drag components are defined in
further discussions. As there are a lot of different ways in which drag is subdivided, it is
important to create an insight on how they are used in this document. This is followed by a
discussion on the drag phenomena for each component, wing, tail and fuselage separately.
The last section discusses the drag arising when those components are joint together.
Drag: Overview
In an aircraft drag polar it is customary to distinguish two different types of drag, a lift
independent part also referred to as zero lift drag and a lift dependent part. However it is
mostly unclear what they are composed of and how they are influenced. Lets start with
a breakdown of the overall drag:
X
D = CD qS = CDs qSs + CDp qS + CDi qS,
⇓
P
D C D s Ss
CD = = + C Dp + C Di
qS S
C2
P
C D s Ss
= + C Dp + pCL2 + k L
S CL =0 πA
CL2
= C D0 + (3.2)
πAe
The CDs represent the secondary drag coefficients for e.g. the fuselage, vertical tail. These
are assumed independent of lift and are referred to there own reference areas, Ss . In this
derivation they are joint under CD0 , which is composed of drag from several components,
and by definition is independent of the lift. Examples of components contributing to this
part of the drag are the fuselage and the tail assuming small angles of attack. Those
components often also present a variation with the angle of attack (lift coefficient), how-
ever for the optimization they have an inclination with respect to the wing such that they
produce the least amount of drag while cruising.
The profile drag is the total drag of the wing minus the induced drag. This drag can be
subdivided into airfoil friction drag and pressure drag. Friction drag is the component
of the forces tangential to the airfoil’s surface, acting in the direction of the flow. The
pressure drag is the component of the forces normal to the surface, in the direction of
the flow. These components only exist if the aerodynamic calculations on the airfoil
incorporate viscous effects, potential flow calculations on airfoils (2D) produce lift but no
drag (d’Alembert’s paradox). A part of this drag is constant and will be added to CD0 ,
while the other part varies with the lift and is incorporated in the second term of equation
3.2.
3.1 Aerodynamics 11
This second term is often referred to as “induced drag”, however this is a unfortunate
name as it incorporates more than just the induced drag. It would be better to refer to
it as the lift dependent drag. The Oswald facter, e, combines the induced drag factor, k,
and the profile drag factor, p:
1
e= (3.3)
k + pπA
The only problem with equation 3.2 is the fact that the minimum drag is found at zero
lift. This is however not the case for most of the aircraft with cambered airfoils. The
profile drag for positively cambered airfoils has its lowest drag at a positive lift coefficient.
For the three dimensional case, at low lift coefficients, the variation in the profile drag
dominates over the trend in the induced drag resulting in a slight shift of the minimum
drag point to positive lift coefficients (see figure 3.5). As mentioned in reference [29], this
can be taken into account in equation 3.2 by defining a CDref and a CLref . The definition
of this CDref and CLref can be seen in figure 3.5.
(CL − CLref )2
CD = CDref + (3.4)
πAe
Figure 3.5: Overview of the different drag contributions and the defenition of CDref and
CLref .
However, the method used in this report for calculating the wing drag will take the profile
drag into account in a different way. The profile drag, and its variation with the lift, will
be directly obtained from airfoil data, reducing the problem to the following,
CD = CDf us + CDtail + CDp (CL ) + CDi (CL ) + CDint (3.5)
12 Theory
In which CDf us and CDtail are invariant with the lift. The last term CDint represents the
interference drag which is due to the interference between the different components. The
induced drag, CDi , is calculated using,
CL2
CDi = k (3.6)
πA
In this equation k is a amplification factor used for non-elliptical lift distributions.
Equation 3.5 avoids problems like working with reference lift and drag coefficients and it
separates the lift dependent drag components.
In this section the profile drag and the induced drag, both dependent on the lift, will
be treated. The profile drag is dominant at lower lift coefficients and, at low Reynolds
numbers, drag induced by the transitional separation bubble and other viscous effects
have a large contribution to the overall drag of the wing. The drag caused by these effects
is also highly variable with the lift coefficient (or angle of attack), as for example the
transition point and the transitional separation bubble changes location and size with
changing angle of attack. To capture these effects a different approach to estimating the
profile drag will be explained in the following section.
Figure 3.6: Visualization of the angles and forces present in the quasi-3D method to combine
a inviscid lift distribution with viscous airfoil data. (L’ = local lift force from
lift distribution, l = viscous airfoil lift force (at αef f ), D′ = local total (profile
and induced) drag force, d = airfoil profile drag force (at αef f )).
2. Set αi = 0 and d = 0
6. Determine αef f and d from values of l, Reef f , Vef f and airfoil experimental/Xfoil
data
7. Set αi = α − αef f + ǫ (ǫ is the local geometric twist angle and is defined negative
when reducing the angle of attack)
9. D′ = d · cos(αi ) + l · sin(αi )
These steps should be repeated for every strip and the sum of all the strip-drag forces gives
the three dimensional wing drag force. This drag force is composed of the profile drag
force and the induced drag. The induced drag part is overestimated due to a difference
14 Theory
between the down-wash angle used to calculate the lift and the induced drag, the reason
for this is not completely clear but is also mentioned, but not clarified, in reference [15]
on page 7-2.
A downside of this method is its inability to calculate beyond the maximum lift coefficient
of the airfoil even if this information is available. This issue can be solved when the down-
wash angle distribution is known or when it can be calculated from the lift distribution.
An added advantage of using the downwash distribution is the fact that the lift estimate,
obtained by the vortex lattice method, can be corrected for viscous and thickness effects.
Several methods to determine this downwash distribution were tested but the method ap-
plied in the lifting line theory was the easiest to implement, fast to calculate and seemed
to work very well for all wings (even swept wings). The mathematical procedure is the
following. Making the same assumptions as used in the vortex lattice method, one can
approximate the lift coefficient of an airfoil by,
Rewriting this provides information about the angle of attack, required for a certain lift
coefficient,
Cl
αef f = + α0 (3.8)
C lα
This equation can be used in combination with the lift distribution, obtained from the
vortex lattice method, to determine the local effective angle of attack. From this effective
angle of attack it is easy to obtain the downwash angle using:
Here αef f is determined with equation 3.8, ǫ is the local twist angle, defined positive in
the same direction as the wing angle of attack, α, and y represents the spanwise location
on the wing.
The procedure becomes a bit more complicated as the required two dimensional lift co-
efficient, used to determine the downwash angle, also depends on the downwash angle.
This is clearly illustrated in figure 3.7 , l = L′ / cos(αi ), however to determine the effective
angle of attack the forces must be converted into coefficients.
3.1 Aerodynamics 15
Figure 3.7: Definition of the forces and angles used to determine the inviscid downwash
angle.
L′
l =
cos(αi )
⇓
1 2 ′ CL′ 21 ρV∞
2 S′
Cl ρVef f S =
2 cos(αi )
⇓
2
CL′ V∞
2
Cl Vef f =
cos(αi )
V∞
⇓ Vef f =
cos(αi )
Cl CL′
=
(cos(αi ))2 cos(αi )
⇓
Cl = CL′ cos(αi ) (3.10)
Equation 3.10 clearly shows the dependency of Cl on αi and thus requires a converging
iterative loop to uniquely define the effective angle of attack and the the downwash angle.
This iterative loop looks like this,
1. Run AVL for the wing at hand and extract the lift coefficient distribution
16 Theory
2. Set αi = 0
6. Set αi = α − αef f + ǫ
9. Determine Cl and Cd from values of αef f , Reef f and airfoil experimental or viscous
Xfoil data
10. L′ = 12 Cl ρVef
2 S ′ · cos(α ) + 1 C ρV 2 S ′ · sin(α )
f i 2 d ef f i
11. D′ = 12 Cd ρVef
2 S ′ · cos(α ) + 1 C ρV 2 S ′ · sin(α )
f i 2 l ef f i
This procedure is repeated for all strips and the forces are then summed to obtain the
wing lift and drag forces and coefficients. The drag estimation incorporates, like in the
other method, the profile and induced drag, however the problem with an overestimated
induced drag remained. Changes to the method to improve on this induced drag problem
will be explained in section 4.2.1.
For low aspect ratio wings an extra drag component is present, the vortex drag. This
drag is due to the tip vortex, which for low aspect ratio wings (A ≤ 3.5), also acts on
the wing itself. This is due to the side edge suction force acting in the normal direction,
which has components in the lift and drag direction. In reference [17] an equation has
been suggested to empirically estimate the vortex drag component:
• AVL:
– 3 dimensional
3.1 Aerodynamics 17
• Xfoil:
– 2 dimensional
– Incompressible (Karman-Tsien compressibility correction implemented), irro-
tational flow
– Viscous (fully-coupled viscous/inviscid interaction method, en transition crite-
rion)
– Airfoil with thickness
Airfoil Drag1
Low Reynolds number (< 1 million) airfoil design is dominated by minimizing the tran-
sitional separation bubble losses, which can cause up to 50% more drag. To be able to
minimize these losses one must know what the nature of this transitional bubble is.
The bubble is a product of a very stable laminar boundary layer caused by the low
Reynolds numbers. The laminar boundary layer is so stable that, even after it separates
from the airfoil surface, it takes some time to lose its stability and turn turbulent. The
bubble is thus a sequence of events starting with the separation of the laminar boundary
layer. This laminar boundary layer rapidly loses stability, turns unstable and transitions
some time after separation. After the transition the boundary layer is turbulent and able
to reattach. This is clearly illustrated in figure 3.8.
An interesting question at this point is how this separation bubble is able to cause drag,
and how can it be minimized? There are two explanations for the same drag increase.
The first takes a look at the pressure distribution over the bubble while the second one
focuses on the flow itself.
The transitional separation bubble contributes to the profile drag due to its position and
shape. The bubble is always located at the beginning of the pressure recovery region
which is on the aft facing part of the airfoil. Due to its convex shape, a low pressure
is established over the bubble. Having this low pressure perpendicular to an aft facing
surface means that the component of the resulting force pulling back on the airfoil, is
larger. Figure 3.9 provides a graphical presentation of this. The large pressure arrows, in
1
This section is mainly based on information found in reference [12] and [8]
18 Theory
Figure 3.8: Illustration of a laminar separation bubble on an airfoil (picture extracted from
ref. [7] page 74).
the red circle, represent the case where the airfoil is experiencing a Reynolds number of
500000 with a bubble present. The small arrows are the pressure arrows on the surface of
the same airfoil but at a Reynolds number of 3000000 (no bubble present). It is obvious
that the aft pointing component is larger for the larger arrows. Thus there is a stronger
aft pulling force on the airfoil in the lower Reynolds number case.
Figure 3.9: Pressure arrows for an airfoil at Re = 3000000 (no bubble) and for Re = 500000
(bubble present).
2 The second explanation is more mathematically based and takes a look at what happens
inside the bubble. The drag which one wants to minimize during airfoil design is profile
drag. The profile drag is equal to the boundary layer momentum loss, ρu2e θ, far behind
the airfoil. The von Kármán integral momentum equation represents the streamwise
development of ρu2e θ in the boundary layer.
1 d(ρu2e θ) Cf H due
= − (3.13)
ρu2e θ dξ 2θ ue dξ
2
This information was found in reference [12] but is repeated in this report for completeness and clarity
reasons
3.1 Aerodynamics 19
This equation shows that pressure recovery in a separated flow has a strong negative effect
on the drag, due to the large shape factor (in separated flow H > 10). Both terms on
the right hand side of the equation are positive in the pressure recovery region ( du
dξ < 0).
e
This means that the change in the momentum loss must also be positive due to the fact
that the term ρu12 θ is positive. A small change in the velocity will be amplified by the
e
large shape factor which results in a large jump in the momentum losses. This is clearly
shown in figure 3.10. It can be seen from this figure that the edge velocity and the
momentum loss gradients at the transition-reattachment region are so steep that one can
consider the changes as discontinuous jumps. Integrating the von Kármán equation over
the transition-reattachment region becomes possible if the skin friction term is neglected,
and this is permitted in the separation bubble. This then results in the following relation
between the ue and ρu2e θ jumps:
∆(ρu2e θ) ∆ue
= −H (3.14)
ρu2e θ ue
,with the values of ρu2e θ, ue and H averaged over the transition-reattachment region.
Equation 3.14 can be rewritten as a function of the absolute jumps in ue and ρu2e θ:
Figure 3.10: Edge velocity and momentum loss jumps at reattachment (figure copied from
ref. [12] page 725).
This equation provides a handle to minimize the drag caused by a separation bubble. The
increase in drag, A(ρu2e θ), over the turbulent reattachment region is thus proportional to
the average mass defect, ρue δ ∗ , and the edge velocity jump, ∆ue . So keeping the mass
defect small means that the loss is small. This can be achieved by keeping the distance,
20 Theory
at the transition position, between the free shear layer and the wall as small as possible,
because the stagnant fluid under the shear layer must mix out with the turbulent boundary
layer for reattachment to occur. This mixing is also the physical mechanism that converts
the work done by the airfoil on the airflow, via the bubble pressure drag, into heat. The
total drag increase, due to the presence of the transitional separation bubble on the airfoil,
will likely be larger than the bubble momentum loss jump. This is due to the momentum
loss jump being invariably amplified by the airfoil’s pressure recovery region.
To keep the drag as low as possible one has to minimize the distance between the free
shear layer and the wall at the transition point. But how can this be controlled? The
answer can be found in the positioning of the transition point. Forcing the location of the
transition farther upstream, by destabilizing the laminar boundary layer at the leading
edge, makes the bubble shorter and the momentum loss lower. This destabilizing can
be done by inducing an adverse pressure gradient close to the leading edge of the airfoil
(see figure 3.11). But one has to be careful not to eliminate the complete bubble by
positioning the transition point in front of the bubble because this would increase the
drag again (see figure 3.12). This increase in drag is invoked by the higher friction drag
generated by the turbulent boundary layer. Finding the optimal location of the transition
point is very important, this location is mostly just aft of the laminar separation point.
The optimal location also changes with the angle of attack. The goal of low drag airfoil
design is thus: Obtain transition as close as possible to that optimum point over as much
of the operational range, of the airfoil, as possible.
Figure 3.11: Illustration of a typical pressure gradient for Low Reynolds number airfoils.
Apart from the aerodynamic considerations, which for low Reynolds numbers lead to very
thin airfoils, one always has to keep in mind that the airfoils have to provide space in the
wing to locate the structure.
3.1 Aerodynamics 21
Figure 3.12: Effect of the transition location on bubble size and momentum loss (figure
copied from ref. [12] page 726).
22 Theory
Fuselage Drag
l∗ l∗
= 2 · (3.16)
d∗eq b ∗ + h∗
,in which b∗ is the width and h∗ the height of the shape at the largest cross-section.
This thickness ratio can then be used in equation 3.17 to determine the minimum drag
coefficient of the body.
CDf us d∗ d∗
= 1 + 1.5 · ( ∗ )3/2 + 7 · ( ∗ )3 (3.17)
Cff us l l
This equation is empirically developed and is presented in reference [15] by S.F. Hoerner.
The only variable remaining to be determined in equation 3.17, is the friction coefficient.
As stated on page 2-7 of reference [15], it is possible to use the friction drag coefficient
for an equivalent flat plate (plane surface with the same Reynolds number). In figure
3.13 the friction drag coefficient is plotted against the Reynolds number for experiments
and approximation methods. One approximation method for laminar flow and several for
turbulent flow are given in reference [15].
Figure 3.13: Friction drag coefficient for plane surfaces plotted against the Reynolds number
(page 2-6 reference [15]).
In these equations it is important to get the right value for the Reynolds number. However
in some mini-UAV’s the propeller is in front of the fuselage increasing the velocity and
thus the Reynolds number over it. To calculate this increase in the velocity it is necessary
to resort to propulsion theory which states that the propulsive efficiency is relate to the
difference between the velocity at infinity and the velocity behind the propeller. This can
be seen in:
2
ηj = (3.20)
2 + ∆V
V∞
Rewriting this provides the velocity increase,
2
∆V = − 2 · V∞ (3.21)
ηj
Based on a given propulsive efficiency, the Reynolds number corresponding to the fuselage
can be calculated,
l∗ · (V∞ + ∆V )
Ref us = (3.22)
ν
,and used in equation 3.18 and 3.19.
The total friction drag of a body is then determined using a linear function between both
friction coefficients with the location of the transition point, xtr , as its variable.
xtr xtr
Cf = Cflam · ( ) + Cfturb · (1 − ∗ ) (3.23)
l∗ l
To use this equation xtr must be provided by the user. An educated guess will probably
provide a sufficiently accurate result. The real location of the transition point strongly
depends on the shape and thus the pressure distribution over the fuselage. Combining
the result for the friction coefficient with equation 3.17 gives an estimate for the fuselage
drag coefficient.
The best method for the determination of the turbulent flow friction coefficient was se-
lected based on comparison with measurements given in reference [32]. This comparison
can be seen in table 3.1. Most micro aerial vehicles have fuselages with a thickness ra-
tio between 5 and 8, and have a Reynolds number between 4 · 105 and 1.5 · 106 . These
Reynolds numbers take into account that a propeller might increase the velocity over the
fuselage.
In table 3.1 one can clearly see that the calculation method developed by Young in
reference [32] works extremely well. The Hoerner method3 shows larger differences, but
when looking at cases with lower Reynolds numbers (< 2·106 ), it is clear that this method
performs equally good or even better (see the case for Re = 1173000). The Young method
is overall better but is computationally much more complicated as it requires the pressure
distribution over a body of revolution. This method might be an interesting topic to look
at in the future.
3
Hoerner provides two equations (see reference [15]), one calculates the drag coefficient based on the
fuselage wetted area and the other on the largest cross-section. The equation used in this report is based
on the largest cross section. Comparison in table 3.1 is based on the wetted area equation as the data
from literature was only available for drag coefficients based on the wetted area. However both equations
presented by Hoerner are expected to have the same accuracy.
24 Theory
Table 3.1: Comparison between calculated and experimental drag coefficients for stream-
lined bodies of revolution (drag coefficients are based on the fuselage wetted
area).
l∗ xtr
Re d∗ l∗ CD CD Young CD Hoerner
measured Young -error [%] Hoerner -error [%]
840000 4.95 0.1 0.0053 0.0053 0 0.005225 -1.41
1173000 6.67 0.1 0.0045 0.0047 4.44 0.004525 0.57
1260000 5.49 0 0.005 0.005 0 0.005014 0.29
1580000 5.43 0 0.0045 0.0048 6.67 0.004807 6.81
1580000 3.00 0 0.006 0.0062 3.33 0.006398 6.63
2040000 5 0.6 0.0022 0.00236 7.27 0.002537 15.32
3060000 5 0.52 0.0022 0.00244 10.91 0.00255 15.89
2090000 5 0.22 0.00435 0.00409 -5.98 0.003876 -10.90
3130000 5 0.17 0.00415 0.00388 -6.51 0.003733 -10.05
2070000 5 0.33 0.00364 0.00361 -0.82 0.003491 -4.1
3110000 5 0.28 0.00359 0.00351 -2.23 0.003361 -6.38
2050000 5 0.15 0.0044 0.00437 -0.68 0.004142 -5.87
3070000 5 0.1 0.00421 0.00414 -1.66 0.003987 -5.29
11610000 5.88 0.065 0.0028 0.003 7.14 0.003079 9.98
3160000 3.00 0 0.005 0.0052 4 0.005597 11.94
3160000 3.00 0.04 0.0052 0.0051 -1.92 0.005419 4.21
The fuselage drag coefficient uses the largest cross-sectional area of the fuselage as a
reference area, but when used to compute the aircraft drag coefficient this must be with
respect to the wing area. This conversion is simple and can be done using the following
equation:
′ Sf us
CD = CDf us · (3.24)
f us
Sref
Interference Drag
Interference drag is the drag that arises when two solid bodies touch each other or when
they are positioned very close to each other such that their boundary layers interfere.
This drag can thus be seen as the difference between the total drag of the configuration
and the sum of the drags of the separate components constituting the configuration. It is
hard to calculate these effects but they do depend on the configuration and the position of
the components and might thus have a significant effect on the optimization. In reference
[15] an extensive discussion is presented on this topic but little tools are presented that
estimate the effects, the ones presented are only proven for higher Reynolds numbers.
However in this section the same tools will be used and it is assumed that they apply on
the flow regime under consideration.
The first thing to do is to figure out which components have influence on each other. For
flying wings, these are the wing and the fuselage and secondly the vertical tail and the
3.1 Aerodynamics 25
fuselage. Interference drag exists of two components, the parasitic interference, which is
caused by interaction of the boundary layers at CL = 0, and the induced interference
drag, due to changes in the lift.
The parasitic interference drag is plotted in figure 3.14 for different longitudinal locations
of the wing on the fuselage. It can be seen that there is an increase in drag and that it
changes for different locations of the wing, however the wing location will be determined
by the stability margin. This will not be a variable for the optimization, but rather a
given increase in the drag. This effect is however reduced as the wing and the fuselage
both loose wetted area, and thus friction drag, when put together. This is the reason for
assuming that when the wing and fuselage are joined, both effects cancel each other.
However when looking at the vertical tail, this does not lose wetted area and as this surface
does not provide side forces in straight horizontal flight parasitic interference drag will be
the only form of interference drag acting on it. From figure 3.14 CDc for the vertical tail
is taken equal to 0.0015, it can be seen that the effect is small for wings located aft on
the fuselage. To convert this drag coefficient to the wing area one can use the following
formula
c2
CDinttail = CDc · (3.25)
Sref
,in which c is the root chord length of the vertical tail.
Figure 3.14: Parasitic interference drag in the corner between a wing and a fuselage for
different longitudinal locations of the wing/vertical tail (page 8-15 reference
[15]).
The induced interference drag is caused by changes in the lift distribution on the wing
due to the presence of the fuselage. Reference [15] explains this in great detail. In this
report the focus will be on the parts coming into play for mini-UAV’s.
The presence of the fuselage at lift coefficients different from zero causes an increase in the
angle of attack at the roots of the wing, this is refered to as alpha-flow. At the same time
a part of the circulation of the wing roots is transfered to the fuselage and the fuselage
itself also generates lift due to separatation at the rear of the body. These effects might
increase the lift curve slope and thus cause an increase in drag due to lift. These effects
26 Theory
are however hard to predict and for cruise the body will be attached to the wing at an
incidence such that it has a zero angle of attack avoiding this kind of interference drag.
A second effect is that the body displacement causes the average velocity at the wing root
to increase. For a low wing configuration this increases the velocity on the upper surface
of the wing while in a high wing configuration this will be on the lower surface. This has
its effect on the lift which will be higher in the low wing configuration and lower in the
high wing configuration. For a fuselage with a diameter equal to 10% of the span this
effect is in the order of CLb1 = +0.1 for a low wing configuration and CLb1 = −0.1 for
high wing configurations. This CLb1 is based on the wing area covered by the fuselage
width. This means that the effect has to be converted to the reference area using:
Scovered
∆CL = ∆CLb1 · (3.26)
Sref
A linear variation of CLb1 will be applied for wings with a vertical location different from
the high or low wing configuration, and for small changes in percentages of d∗ /b.
The incidence angle of the wing with respect to the fuselage has also its effect on the lift
and thus on the induced interference drag. This effect is best described with equation
3.27.
In this equation ∆α represents the incidence angle, which is defined as the angle between
the wings zero lift line and the fuselage axis (in degrees). Equation 3.28 gives a relation
between the changes in the lift coefficient and the interference drag.
Scovered
∆CD = CDb · (3.29)
Sref
The previously mentioned changes in the lift coefficient will also be taken into account
for the lift but were not discussed in chapter 3.1.1 because of there strong relation to the
interference drag.
3.1 Aerodynamics 27
3.1.3 Stability
The way of treating stability for a mini-UAV is no different from large aircraft. The
autopilot in mini-UAV’s could provide active stability but the autopilots developed in the
MAVlab work better with a inherently stable aircraft. Having an unstable aircraft would
also require a high frequency of control surface deflections which stresses the servos, this
in turn shortens their life span and consumes a lot of useful energy.
The focus of this section will be on the pitch stability, as this is critical for tailless aircraft
configurations. The directional stability will be treated by giving an estimate for the
vertical tail size. Sweep, which is required for trimming the aircraft as will be explained
later in this section, will also increase the directional stability. Roll stability will be
omitted as experience showed that the roll controllability is very high for mini-UAV’s and
can thus be easily compensated for by the autopilot.
Dynamic stability is hard to determine as the values for the stability derivatives are badly
estimated using the vortex lattice method, AVL. A better way to approach this problem
is to optimize the design and determine these values in a wind tunnel or with advanced
CFD-methods and then make small changes to the design to ensure dynamic stability.
Pitch Stability
The static longitudinal stability is determined by the stability margin. This margin is
defined by the difference between the aircraft neutral point and the location of the center of
gravity. For a tailless aircraft this simplifies to the location of the wing aerodynamic center
with respect to the aircrafts center of gravity. Equation 3.30 provides a mathematical
description of this stability margin.
xn − xcg
σ= (3.30)
cref
In this equation the location of the center of gravity, xcg , is determined by the location of
the components inside the aircraft. When the stability margin is positive, the aircraft will
be stable. The aircraft is neutrally stable when the location of the center of gravity is the
same as the neutral point, which explains its name. A negative value results in an unstable
aircraft. The reference chord length is taken equal to the wing mean aerodynamic chord,
and can be calculated with,
b/2
2
Z
cmac = c2 (y)dy (3.31)
Sref 0
The longitudinal location, when taken as the reference point4 through which the aerody-
namic forces act, in which the pitching moment of the complete aircraft is independent of
the angle of attack.
For a tailless aircraft configuration the determination of the neutral point location is the
same problem as determining the aerodynamic center of the wing (when the fuselage
influence is assumed to be small). Two methods to determine this aerodynamic center
were tested and compared to experimental results. A first method is purely based on the
geometry of the wing and was obtained from reference [21].
A1+2·λ croot
xac = cmac · · tan(Λ0.25 ) + (3.32)
6 1+λ 4
A second method was also obtained from reference [21], and this one uses aerodynamic
coefficients to determine the aerodynamic center. When assuming that the aerodynamic
center is located in the plane of the wing, then one can write the pitching moment around
the wing origin (leading edge of the wing root) as:
xac
CMorigin = CMacw − (CL · cos(α) + CD · sin(α)) · (3.33)
cref
Rewriting equation 3.33 by ignoring the drag effects and assuming small angles of attack
gives:
xac
CMorigin = CMacw − CL · (3.34)
cref
Differentiating this equation with respect to the angle of attack results in,
xac
CMoriginα = 0 − CLα · (3.35)
cref
, which can then be written as
CMoriginα
xac = −cref · (3.36)
C Lα
, in which cref is equal to cmac . The values of CMoriginα and CLα can be found using AVL.
To validate these methods, a comparison between calculated aerodynamic center and
experimentally determined aerodynamic center is represented in table 3.2. The data in
this table was obtained from reference [21] and [5]. Reference [21], referred to as Phillips
in table 3.2, used CFD data combined with equation 3.36 to determine the aerodynamic
center, while the data from reference [5], referred to as NACA in table 3.2, was obtained
from wind tunnel measurements.
The NACA wings all have the same cambered airfoils and varying sweep and twist. The
two numbers following the 24 define the sweep angle in degrees and the last two define
4
Remember that the real point through which the aerodynamic forces act, is the point were there is
no aerodynamic moment. This point is known as the center of pressure which changes location with the
angle of attack
3.1 Aerodynamics 29
Table 3.2: Comparison between experiments/CFD and the calculated aerodynamic center
using equation 3.32 and 3.36.
the washout angle (twist) of the wing, e.g. NACA243085 has 30◦ quarter chord sweep
and −8.5◦ twist. The wings from reference [21] are designated Phillips (after the author)
with the quarter chord sweep angle behind it. These wings did not have any twist and
symmetric airfoils were assumed. From the comparisons it is clear that the second method
(eq. 3.36) is better in estimating the location of the aerodynamic center than the first
method (eq. 3.32), except for the 50◦ sweep angle. When compared to the experimental
results, the error is around ±0.5% for swept wings while for unswept wings the estimate
has an error of 3.8%. The larger error when compared to the Phillips wings with higher
sweep angles are probably due to errors in the CFD calculation because when comparing
the 30◦ sweep from the NACA report it appears to work fine. Another cause might be the
assumed symmetric airfoil, however this is unlikely to be the problem as airfoils should
only have a minor influence on the location of the aerodynamic center.
For the stability of flying wings it is important to validate the impact of this error on the
stability margin. That is why a ∆σ was presented in table 3.2, the calculation of this
value is the following.
xaccomputed − xcg
∆σ = − 0.05 (3.37)
cmac
Assuming an original stability margin of 5%, relative to the experimental/CFD aerody-
namic center, and having the mean aerodynamic chord gives the location of the center
of gravity, xcg , by rewriting equation 3.30. This value of xcg is then used to determine
the stability margin with respect to the calculated aerodynamic center. The difference
between the calculated stability margin and the original stability margin (= 5%) is rep-
resented in the table.
It is clear from table 3.2 that this error is quite high for the first method but is acceptably
low for the second method. The first method also has the tendency to always overestimate
the stability margin, while this is not the case for the second method.The positive values
of ∆σ represent overestimations of the stability and are thus relatively dangerous as they
can cause instability in the airplane while on paper everything looks stable. The second
method seems to provide no unstable aircraft as long as the design stability margin is
30 Theory
higher than 2% and when the sweep angle is lower or equal to 30◦ . The experiments
show no problems up to 30◦ while for the CFD problems start at this sweep angle, but
as mentioned before this might be due to calculation problems in CFD or an error in the
assumed airfoil. From this discussion it is clear that equation 3.36 is the one that will
be used for estimating the aerodynamic center and thus the neutral point location of the
tailless aircraft configuration.
Trimming an aircraft is to ensure that the aircraft flies level at the desired speed, the
cruise speed. This speed is determined by the design lift coefficient, corresponding to
(L/D)max , when considering straight horizontal flight. The trick about trimming the
aircraft is quite simple, get the right amount of aerodynamic moment to ensure that it
cancels the weight induced moment around the aerodynamic center. A very clear visual
presentation of this is given in figure 3.15. For our purpose the thrust is assumed to be in
line with the center of gravity and is conveniently left out of the picture. Mathematically
this translate into:
M − W · (xac − xcg ) = 0 (3.38)
There are two ways to get this aerodynamic moment right. The first is by simply
Figure 3.15: Illustration of forces and moments involved in trimming an aircraft (L = lift,
D = drag, M = aerodynamic moment, W = aircraft weight).
deflecting the elevator control surfaces, this is a good approach when considering non-
cruise phases of flight. The second way is by giving the horizontal tail surface, for normal
configurations, a decalage angle. For flying wings one has to distinguish between swept
and unswept wings. For swept wings one can change the lift distribution by applying
twist (geometric and aerodynamic), while for unswept wings this has no effect as there
3.1 Aerodynamics 31
is no arm with respect to the aerodynamic center and it is thus obtained by using the
appropriate airfoils (with the correct moment coefficient).
However the difficulty lies in getting a good estimate for the moment coefficient around
the aerodynamic center. For flying wings this is a bit tricky due to the fact that the
moment coefficient of the airfoils have a major impact on the overall moment coefficient
around the aerodynamic center, while for a normal configuration this is dominated by
the lift forces of main and horizontal tail wing. These large forces are easier to determine
accurately.
For swept wings the moment coefficient is not only determined by the airfoil moment
coefficient but also by the lift distribution. This makes it possible for a swept wing to
use aerodynamic and geometric twist to trim the aircraft. A way to understand this
better is to make a comparison between a swept wing tailless aircraft and a normal
configuration as presented in figure 3.16 from reference [19]. In this figure it is clear how
a normal configuration can change into a swept wing tailless aircraft. It also shows that
the function of the horizontal tail is now performed by the wingtips, and thus showing
the similarities between the decalage angle for a normal configuration and the twist for a
swept wing tailless aircraft.
The next step is of course to determine the wing moment coefficient around the aerody-
namic center. Once this moment coefficient has been calculated it is possible to determine
32 Theory
the required amount of twist such that the aircraft is trimmed, for a certain stability mar-
gin, at the cruise speed without deflecting the elevators.
The method to calculate the wing moment coefficient was developed by the author and
a small correction was made from information in reference [10]. From figure 3.17 it can
be seen that the moment coefficient consists of two parts: one is due to the integration
over the span of Cmac and the other one is due to the integration of the lift distribution
times its arm to the aerodynamic center over the wing. A strip method will be used to
calculate these integrals.
Figure 3.17: Forces and moments contributing to the wing moment coefficient.
3.1 Aerodynamics 33
, in which
N
X Cmaci · ci · (ci · ∆yi )
C M1 =2·f · (3.40)
Sref · cref
i=1
and
N
X CLi · (ci · ∆yi ) · (xacw − xaci )
C M2 = 2 · (3.41)
Sref · cref
i=1
The equations are multiplied by two to obtain the moment for the complete wing, as the
equations are developed for half wing span. All subscripts i define the strip number going
from 1 to N . In equation 3.40 an empirical correction factor f was used to correct for
planform effects (see reference [10]), and can be found using:
2·A
f= · cos(Λ0.5 ) (3.42)
2·A+1
For this method AVL is used to determine the lift distribution, while Xfoil is used to
determine the airfoil average moment coefficient around its aerodynamic center. This
averaging is done over the attached flow α-range of the airfoils.
Directional Stability
Directional stability is handled by the vertical tailplane. A simple equation was used to
estimate the required vertical tail surface. This equation is often used in the design of
RC-planes and has proven to work well.
Sv · (xacv − xcg )
= 0.2 ∼ 0.4 (3.43)
S·b
In this equation xacv is the location of the vertical tail aerodynamic center. For more
directional stability it is better to be close to 0.4 however 0.3 is usually sufficient.
34 Theory
3.2 Optimization
In this section an introduction will be provided into the algorithm used for the optimiza-
tion. The algorithm is required to handle non-linear behaviour as the objective and some
of the constraints will be non-linear. Based on a comparison between two candidate al-
gorithms one is selected and explained. This is followed by an insight into the variables
used for the optimization and the constraints imposed on them. The last part will provide
an insight into the different objective functions for range and endurance of electrically
powered aircraft.
The objective function and some of the constraints applied to the optimization are non-
linear and thus require a nonlinear optimization algorithm. Several algorithms are avail-
able to tackle these problems. However for this research the two algorithms of interest are
the sequential quadratic programming algorithm, and the genetic algorithm. These two
options are selected because they exist in Matlab and provide the capability to handle all
types of constraints. They are also state of the art algorithms when dealing with complex
optimization problems.
Algorithm Selection
The selection is based on their capabilities and limitations in handling the objective
function and its constraints. An overview of both algorithms is presented to create a base
for the selection process.
• Genetic algorithm
– Unlikely to get stuck in local minima, when the settings for the algorithm are
correct
– Able to handle strong nonlinear objective functions
– Inefficient use of objective function evaluations
– Can handle all types of constraints (nonlinear, equality, inequality, bounds)
From this overview it is clear that the differences are in the efficiency of the algorithm
and the capability to handle local minima. The conclusion is thus that it is preferred to
use the sequential quadratic programming algorithm when it is proven that the objective
function together with its constraints, contain no local minima.
3.2 Optimization 35
For a sequential quadratic program, the necessary and sufficient condition for a nonlinear
minimization problem, is that the objective function and the constraints are convex. De-
termining the convexity of the objective function is impossible, as the objective function
for performance optimizations, is dependent on 5 variables and the calculation of the
aerodynamic forces is based on a numerical approach. The linear constraints are always
convex but the nonlinear constraints are also dependent on the aerodynamic calculations
imposing the same problem as for the objective function. The conclusion of this is that
the use of sequential quadratic programming is not guaranteed to find a global minimum.
However the efficiency of this algorithm is much larger and is thus preferred. One way of
testing if it could still be used is by applying it with different initial points and then deter-
mining if it always finds the same result. Another option is using the genetic algorithm
to find a global optimum and then verifying if the sequential quadratic programming
algorithm results in the same solution to the problem.
The sequential quadratic programming algorithm exists of two steps in which a quasi-
Newton method and a line search are used iteratively to find the best possible solution.
The quasi-Newton method is used to locally, around the current location, approximate
the real problem with a quadratic function. The line connecting the solution to this
quadratic problem and the current location, is then used to determine the new-current
location, and is referred to as the line search.
Karush-Khun-Tucker Equations
The Karush-Khun-Tucker (KKT) equations are used to formulate the necessary condi-
tions for a general constraint optimization problem. These conditions are also sufficient
conditions if the objective function and the constraint functions are convex. A general
optimization problem,
min f (x) (3.44)
x
subject to,
Gi (x) = 0, i = 1, ..., me
Gi (x) ≤ 0, i = me + 1, ..., m
the Lagrange multipliers which are used to scale the gradients of the constraints such that
they balance with the objective gradients. The last two equations implicitly ensure that
only the active constraints are included by setting the Lagrange multipliers of non-active
constraints equal to 0. Further details about the Karush-Kuhn-Tucker conditions can be
found in [18] and [1].
Line Search
The solution, dk , of the above mentioned quadratic subproblem provides a direction for
the line search which will provide a new current location. In equation 3.47 one can see a
linear function with a single variable ξ, which is changed such that the required decrease
in the merit function 3.48 is obtained.
xk+1 = xk + ξdk (3.47)
me
X m
X
ψ(x) = f (x) + ri · gi (x) + ri · max[0, gi (x)] (3.48)
i=1 i=me +1
One of the first steps in setting up an optimization tool is to determine the variables. For
the problem at hand two sets of variables are needed, geometric variables which define the
wing geometry and variables that define the flow around the aircraft, like there are the
velocity and the altitude. After the definition of the variables it is important to recognize
and define the linear and non-linear constraints.
Variables
As already mentioned, there are two sets of variables. The first set, the geometric vari-
ables, is used to unambiguously define the wing geometry. The approach will be to use
multiple linear wing sections. This is the same convention as used in AVL and makes it
possible to define the geometry with a limited amount of variables. The assumption is
made that the wing is symmetric with respect to the aircraft longitudinal plane, and the
variables thus define one side of the wing while the other side is mirrored. The variables
to define a trapezoidal wing section are:
b
• Section span, 2
• Taper ratio, λ
In this list the twist is missing, this is because of the fact that this variable depends on
all other variables. The twist for a tailless aircraft is varied such that the wing generates
the required amount of moment around the aerodynamic center to trim the aircraft at
its design lift coefficient. When defining multiple wing sections these variables become
vectors. A more detailed insight into these variables will be provided in section 4.1.1.
Up until this point one variable has not been discussed, namely the airfoil. At each
wing section an airfoilhas to be defined, however defining multiple airfoil options for each
section causes discontinues jumps in the objective domain. These discontinues jumps in
the domain will cause problems for a sequential quadratic optimizer. The solution to this
problem will be discussed in chapter 4.
Now that the geometry has been defined the only variables left for discussion are the flow
variables. The two variables for the flow are the aircraft velocity and the altitude. The
velocity is used to determine the design lift coefficient and the required moment coefficient,
which in turn relates to the wing twist. The altitude is important to determine values like
the air-density and the dynamic viscosity. However the altitude has been left out of the
picture because of two reasons. First of all, mini-UAV’s are limited to a maximum altitude
of 300 meters, to avoid interference with normal aircraft traffic. This altitude limitation
also has its effect on the second reason. The temperature, which strongly influences the
38 Theory
air density and dynamic viscosity, at these low altitudes changes drastically with time of
the day, season, location on the earth, altitude, etc. . It would therefore not be practical
to take this into account during an optimization. However it would be wise to take it into
account during mission planning.
Constraints
To limit the search field of the optimizer and to ensure a feasible design it is important
to apply different kinds of constraints. A large distinction must be made between linear
and non-linear constraints. In the wing optimizater these two types of constraints will
be used. The only linear constraints are inequality constraints, two for each variable, to
define a lower and upper bound for the variables.
The non-linear constraints serve the main purpose of providing a feasible design and
ensuring the quality of the aerodynamic calculation. All non-linear constraints used will
be inequality constraints, to set an upper limit to the stall speed, ensure that the twist
is not too large, set an upper bound on the design lift coefficient and to apply a lower
bound on the Reynolds number. The first three ensure a feasible design and the bound
on the Reynolds number is to ensure the credibility of Xfoil-results.
In this section the value used to optimize for will be discussed and further referred to as the
objective. In this section some interesting findings will be discussed for different situations.
First of all the fact that the aircraft under consideration are using electric power, needs
some extra attention. Apart from this it is interesting to observe the effect the subsystems,
excluding the propulsive subsystem, have on the definition of the objective.
A distiction will be made in this section between the case in which energy from the bat-
teries is used only for propulsion and the case where energy is drained from the batteries
to supply all subsystems.
In this derivation C represents the electrical capacity and Iel represents the electrical
current, while R and V are respectively the range and velocity. Integrating equation 3.49
will result in the overall range flown for a certain capacity consumption.
Z C2
V∞
R = − dC
C1 Iel
⇓
Z C1
V∞
R = dC (3.50)
C2 Iel
In this part C1 is the start capacity and C2 is the end capacity. The next step is to get the
electrical current out of the picture by looking at the power required and the electrical
power.
Pr = D · V ∞ , Pr = ηtot Pel
⇓
DV∞
Pel =
ηtot
⇓ Pel = Uel · Iel
DV∞
Iel = (3.51)
ηtot Uel
,in which ηtot represtents the total power efficiency from battery to propeller. Substi-
tuting equation 3.51 into equation 3.50 results in,
Z C1
ηtot Uel
R = dC
C2 D
⇓
ηtot Uel C1
Z
R = dC
D C2
⇓
ηtot Uel
R = (C1 − C2 ) (3.52)
D
In this part the assumption is made that the electrical tension remains constant over
time which is true for most of the discharge time except for the beginning and end of
the discharge phase. The last step in the derivation is to rewrite the drag by using
D = CD
CL · W , resulting in the electrical range equation.
CL 1
R = ηtot Uel (C1 − C2 ) (3.53)
CD W
The capacity in this equation has the units “Ampère second”, however in real life this
unit is mostly expressed in “Ampère Hour”, changing equation 3.53 into,
CL 1
R = 3600 · ηtot Uel (C1 − C2 ) (3.54)
CD W
From this equation it is clear that the aerodynamic objective, to maximize range, is the
maximization of the lift over drag ratio, the same as for fuel powered propeller aircraft.
40 Theory
This maximization of the lift over drag ratio can be simplified to an other ratio when
optimizing the wing geometry with aspect ratios larger than 3.5, resulting in a new ob-
jective. The constraint on the aspect ratio is due to the non-linear behavior invoked by
the influence of the tip-vortices on the lift and drag forces, rendering the conclusion from
equation 3.55 invalid. The maximum lift over drag for a certain geometry can be found
when the induced drag equals the zero lift drag, as can be seen in the following derivation:
CL
d C D
= 0
dCL
⇓
!
CL
d C2
L
CD0 + πAe πAe(CD0 πAe + CL2 ) − 2CL2 πAe
=
dCL (CD0 πAe + CL2 )2
CL2
CD0 (πAe)2 − CL2 πAe C D0 − πAe
= = =0
(CD0 πAe + CL2 )2 2 2
CD0 πAe+CL
πAe
⇓
C Di = C D0
⇓
C D = C D0 + C Di = 2 · C D0 (3.55)
From equation 3.55 it can be deduced that the maximum lift over drag ratio is found
when the drag is equal to twice the zero lift drag and the lift corresponding to this drag
is found to be, p
CL = CD0 πAe (3.56)
Substituting this into the maximum lift over drag ratio results in,
p √ s
CD0 πAe
CL π Ae
= = (3.57)
CD max 2 · C D0 2 C D0
From equation 3.57 it becomes clear that the optimization objective, for an aircraft with
aspect ratio higher than 3.5 and when energy is only used for propulsion, changes to the
maximization of CAeD0
. The constraint of the energy consumption is based on the validity
of maximization of lift over drag which becomes invalid when other subsystems start to
drain energy from the same source.
Some interesting sidemarks can be made by looking at this new objective. At first glance
it might be obvious to maximize the aspect ratio, however when there is a constraint on
the span, as is the case for most planes, the chord will decrease and thus the surface area.
This increases CD0 due to indirect consequences as there are:
• The decrease in surface area increases the relative impact of the fuselage & tail drag
coefficient (see equation 3.24).
• The reduced size of the wing will result in components, like batteries, being moved
from the wing to the fuselage increasing its size and thus its drag.
3.2 Optimization 41
• The reduction in chord results in a slightly higher speed, to maintian the lift co-
CL
efficient corresponding to ( C D
)max . This increase in the speed, however, is not
compensating the reduction in chord length thus reducing the Reynolds number.
This might drastically increase the profile drag when Reynolds numbers become
lower than 2 · 105 .
the velocity must be as high as possible to minimize the second term. This means that
the denominator will become the new objective to be minimized. In this equation the
electrical tension is assumed to be constant as in equation 3.54. The conversion of units
for the battery capacity also holds for equation 3.59 resulting in.
C1 − C2
R = 3600 Ielsub
(3.60)
CD W
CL ηtot Uel + V∞
42 Theory
The electrical endurance equation will be derived here for completeness reasons, however
it will not be used in this project, but will be interesting to future work. The distinction
made, in the range equation, between the case for which power was consumed by the
propulsive system only or by all subsystems, will not be interesting for the endurance
case. The endurance will only be reduced by the consumption of energy by the other
systems, and has no further influence on the speed nor the aerodynamic objective. The
equation to calculate the endurance in the case were all systems consume energy from the
same source will be given at the end of this section.
The derivation of the electrical endurance equation is the following,
Z t2
E = dt
t1
dC
⇓ Iel = −
dt
Z C2
dC
E = −
C1 Iel
⇓
Z C1
dC
E = (3.61)
C2 Iel
Using equation 3.51 it is possible to get the electrical current out of the picture.
C1
ηtot Uel
Z
E = dC
C2 DV∞
⇓
ηtot Uel
E = (C1 − C2 )
DV∞
s
CD 2W
⇓ D= W, V∞ =
CL ρSCL
ηtot Uel
E = q (C1 − C2 )
CD 2W
CL W ρSCL
⇓
s r
CL3 Sρ
E = 2 ηtot Uel (C1 − C2 ) (3.62)
CD 2W 3
C3
From this equation it is obvious that the maximization of C L2 remains the endurance
D
objective, even when dealing with electrically powered propeller aircraft.
The units for the capacity must be adapted to Ampère hour, resulting in,
s r
CL3 Sρ
E = 3600 · 2 ηtot Uel (C1 − C2 ) (3.63)
CD 2W 3
3.2 Optimization 43
To include the power consumed by the non propulsive subsystems one can use the same
technique, as for the range equation, of distinguishing between electrical current used by
the propulsive system and by the other systems. The resulting formula is the following,
ηtot Uel
E = 3600 · r (C1 − C2 ) (3.64)
2
CD 2W 3
3
CL Sρ + ηtot Uel Ielsub
2
CD
From this equation it is obvious that the only ratio influenced by the aerodynamics is 3
CL
3
CL
which has to be minimized or, as before, 2
CD
has to be maximized.
C3
As for the range, the C L2 changes into a new objective function when optimizing the wing
D
geometry. The only constraint for this change is that the aspect ratio of the wing is larger
than 3.5. The non-linearities in the forces, due to the tip-vortices at lower aspect ratios,
prevents the derivation of equation 3.65. The same technique as in equation 3.55 will be
used to state,
CL
d C D
= 0
dCL
⇓
C Di = 3CD0 (3.65)
3
CL
Substituting this into the ratio, 2
CD
, results into,
3 3 p
CL3 (3π) 2 (Ae)3
(3CD0 πAe) 2
2 = = (3.66)
CD max
(4 · CD0 )2 16 (CD0 )2
Thus to optimize the wing geometry for endurance, including the case√were the subsystems
(Ae)3
consume power from the same source as the propulsion, the ratio (CD0 )2
, must be used
as the objective function.
44 Theory
Chapter 4
The construction of the optimization program will be the topic for this chapter. The
theory explained before will be implemented in a program which has the purpose to find
the best wing for a certain aircraft weight. This program exist out of an optimization
algorithm that uses a newly developed program to determine the objective value. This
new program is a quasi-3D method for determining the aerodynamics of a wing.
Section 4.1 will provide an insight into the program setup with a detailed description
of every step taken during the optimization process. This is followed by section 4.2,
with an overview of the differences between the theory and the actual implementation.
An extensive validation of the new quasi-3D method will be provided in section 4.3, by
comparing it to CFD results and experimental results. The wings for the validation
process were carefully selected to provide a broad validation basis. At the end of the
chapter, in section 4.4, the reasons for developing this quasi-3D method are repeated and
checked against its performance.
The idea behind an optimization is to start with an initial design which is iteratively
changed to obtain the final design. The iterative loop for the wing optimization can be
seen in figure 4.1. This flow diagram will be used as a guideline through the following
discussion, every block in the diagram will correspond to a subsection. Figure 4.2 provides
an insight in all the functions used in the program and shows there relation to each other.
From this figure it becomes clear that all aerodynamic data, two dimensional or three
dimensional, are obtained from external programs, like Xfoil and AVL. The airfoil data
obtained from Xfoil is stored in a structured way, with the idea to reduce calculation
time. The way this data is stored will be explained in 4.1.7.
45
46 Implementation of the Theory
The program initialization serves the purpose to provide the information needed by the
program to start the optimization. Several inputs are required to define some constant
design values and to define the boundaries and initial values of the optimization variables.
An overview of these parameters can be found below.
Fixed Parameters
The fixed parameters are used to define some specific aircraft details, that are constant
during the optimization.
• General parameters
• Wing parameters
– Wing transition point for the upper and lower surface at each section: This
is used to force transition at a certain chord wise location on the wing. The
values vary from 0 to 1. Set to 0, transition is forced at the leading edge and
1 is used for free transition. This parameter is a vector of length equal to
the number of wing sections. In between the sections the forced transition is
linearly varied from one section to the other.
• Fuselage parameters
4.1 Program Setup 47
Figure 4.2: Overview of the program structure, in which it is visualized how functions relate
to each other.
– Fuselage cross-section: This parameter can have three values, ’circular’, ’rect-
angular’ or ’none’, according to the fuselage cross-sectional shape.
– Fuselage length: The fuselage length is used to determine the thickness ratio
and the fuselage Reynolds number
– Propeller efficiency: The velocity increase over the fuselage, due to the pro-
peller, can be determined using this efficiency. This velocity increase is needed
to establish the fuselage Reynolds number which in turn effects the fuselage
48 Implementation of the Theory
drag. When the propeller is not in front of the fuselage this value must be set
to 1. The parameter range is ]0, 1].
– Fuselage transition point: This value is used to determine the fuselage friction
drag. It is hard to determine an accurate value but an educated guess will
provide sufficient accuracy. It is advised to set this value to zero when a prop
is present in front of the fuselage. The values range is [0, 1].
– Fuselage diameter [m]: When the fuselage cross-section is set to ’circular’ it is
required to provide the fuselage diameter at the largest cross-section.
– Fuselage height [m]: When the fuselage cross-section is set to ’rectangular’ one
has to provide the height at the largest cross-section.
– Fuselage width [m]: When the fuselage cross-section is set to ’rectangular’ one
has to provide the width at the largest cross-section.
• Tail parameters
Optimization Variables
The optimization variables require constraints and initial values before starting the opti-
mization. The variables were already discussed in chapter 3.2.2 however a more in dept
discussion will be presented to avoid any ambiguities.
The only flow variable, the airspeed, requires no further explanation, this is a single value
defining the flow velocity at infinity. This variable is bound by an upper and a lower limit,
and, like all variables, requires an initial value. There are also three nonlinear constraint
acting on this variable as will be explained later on.
The geometric variables can be seen in figure 4.3 in which an example is presented of a
wing with three sections (two surfaces). Most of the variables, to define the geometry for
this example, will be vectors with two elements, except for the root chord.
All geometric variables require an upper and a lower bound and an initial value to start
the optimization. The vectors of the bounds for the sweep and taper must have the same
length as for the variables themselves such that different bounds can be applied to them
for every surface or section. The upper and lower bounds for the span only requires one
value. The sum of all section spans will then be limited by these bounds. This is to enable
the user to specify a minimum and maximum total wing span. The upper and lower bound
vector, for the root chord, naturally contain only one element constraining its domain.
4.1 Program Setup 49
Figure 4.3: Example of the geometric variables for a wing with three sections (the subscripts
define the vector element).
As mentioned earlier, the bounds are implemented as linear inequality constraints. This
is because inequality constraints are incorporated into the optimization problem while
bounds have to be checked which decreases the efficiency.
Apart from the bounds, the initial values of the variables need to be provided. It is
obvious that these initial values are contained within the specified bounds and that they
preferably also meet the non-linear constraints. A good initial guess is not necessary but
will most likely shorten the time it takes to find the optimum.
The other geometric variables, not represented in the figure, are the twist, dihedral and
the airfoils. The twist will be explained in the following section as it is dependent on
the other variables. For now, the dihedral is not considered as a variable, as it mainly
influences the lateral stability and not so much the range performance. As mentioned
earlier, from experience it was noted that the aileron efficiency of these small aircraft is
really high providing the autopilot with great controllability in the lateral direction. This
was the reason to only consider wings without dihedral.
The airfoils have to be defined at all sections. The variables defined this far have been
continuous variables, meaning that they can have any value within the constraints. The
airfoils, however, are implemented as discrete variables, which means that different options
must be specified which all have to be evaluated separately. Having a lot of different airfoil
combinations will increase the calculation time dramatically. The user has to provide all
50 Implementation of the Theory
airfoil combinations that he/she wants to evaluate. Every airfoil combination then requires
a separate optimization of the continuous variables, as having the airfoils changed in the
same process as the continuous variables will result in an objective domain with local
minima. As mentioned before, the sequential quadratic algorithm does not handle local
minima to well. The best airfoil combination is selected at the end of all optimizations.
This however is probably not the fastest way of optimizing and thus requires further
investigation.
Once the optimization is initialized, the iterative loop of looking for the best wing can
be entered. The first step in this loop is determining the wing twist such that a trimmed
straight horizontal flight is achieved. This in itself requires an iterative loop which, re-
gretfully, increases the optimization time drastically1 . In figure 4.4 an overview of this
iterative process can be seen.
Figure 4.4: Ilustration of the iterative loop to determine the required twist.
The determination of the required moment coefficient is done with the help of equation
1
For normal configurations (wing-fuselage-tail-configurations)this iterative process can be skipped by
fixing the twist and thus reducing the calculation time.
4.1 Program Setup 51
3.38. From the figure it is also clear that the moment coefficient around the aerodynamic
center needs to be determined at least 3 times. The longitudinal location of the aero-
dynamic center needs to be established every time the moment coefficient is evaluated.
This involves evaluating equation 3.36 which requires a double quasi-3D aerodynamic
evaluation of the wing to determine the CMα and CLα . Resulting in at least 6 quasi-3D
aerodynamic calculations only to determine the twist, drastically increasing the calcula-
tion time.
At this point the complete wing geometry is defined, and thus can the process of deter-
mining the objective value start.
The first part in determining the objective value, is estimating the drag at the lift coeffi-
cient for straight horizontal flight at the requested velocity. This drag is the sum of the
wing, fuselage and tail drag. The aircraft is first trimmed at the required angle of attack
such that the lift equals the weight. This is done using the quasi-3D method, which also
provides information on the wing drag.
The tail drag is determined using a very simplified version of the quasi-3D method. The
vertical tail, in a non-slipping flight, is not producing lift so a good approximation is to
determine the profile drag of the vertical tail airfoil at the Reynolds number corresponding
to the tail mean aerodynamic chord. The following step is to determine the fuselage drag
which is easily obtained using equations 3.16 to 3.24.
Once these values are in place it is only a matter of substituting them into the equation
suiting the aircraft purpose and its power usage, as explained in chapter 3.2.3.
Termination Criteria
Once the optimization is running it must know when it is close enough to the optimum
to stop. This is done by the termination criteria. There are two types of termination
criteria, the once that limit the number of iterations or function evaluations and the ones
that put constraints on the variation of certain values.
The limitations on the number of iterations and function evaluations are set such that they
are almost never the termination criteria. This is because they can stop the optimization
before the actual optimum is found. However they become active in case the optimization
is taking to long or a optimum is hard to find.
The other criteria take into account the variation of the objective function, the variation of
the variables and the tolerance on the non-linear constraint violation. Ones these criteria
are met the optimization will break out of the iterative loop and assume that the obtained
result is the optimum. These termination criteria are better for determining the end of
the optimization as they are directly linked to the result.
52 Implementation of the Theory
Non-linear Constraints
Apart from the bounds on the variables, some non-linear constraints are imposed on the
variables. These non-linear constraints are:
The lower limit on the Reynolds number is to ensure that the results obtained from Xfoil
are still correct. A major problem for Xfoil is to predict bubble bursting which arises
at very low Reynolds numbers. This bubble bursting can be seen in figure 4.5, were
experimental results are compared to Xfoil. Ones the adverse pressure gradient, increases
with increasing angle of attack, even the turbulent boundary layer is not able to reattach
to the surface any more. This results in a large separated flow region over the airfoil and
appears as if the laminar separation bubble suddenly bursts. This effect can be seen in
the force coefficients as a sudden drop in the lift coefficient and a strong increase in the
drag (see figure 4.5). However predicting this bubble burst is very hard and is a result of
a combination of the adverse pressure gradient and the Reynolds number. Limiting the
Reynolds number to a lower bound should, at least partially, avoid this problem. The
lowest Reynolds number for the wing is calculated using,
ctip · V∞
Remin = (4.1)
ν
A limit on the stall speed has more to do with the practical side of using mini-UAV’s than
with the performance itself. Without this limit the optimizer might look for high aspect
ratio wings to reduce the induced drag, however together with a limit on the span this
might result in a very small wing surface area. This in case of flying wings, with their
reflexed (read less cambered) airfoils and thus lower CLmax values, will result in rather
high stall speeds. High stall speeds imply that the operator can’t throw the airplane
anymore but must use a launching system for the take-off. Launching systems add weight
to the overall system and, when kept small, put a very large force on the aircraft. However
controlling this stall speed limit provides the user with the option to optimize the design
with a certain take-off in mind. The stall speed is determined using
s
2·W
Vstall = (4.2)
ρ S CLmax
In this equation CLmax has to be supplied by the user. In the future the quasi-3D method
must be validated for its capability to determine the maximum lift coefficient. At this
point it can not be assured that the maximum lift coefficient obtained with this method
is correct as its stall is predicted from Xfoil airfoil data, which become less accurate when
4.1 Program Setup 53
Figure 4.5: Effects of bubble bursting and the comparison with Xfoil results (figure obtained
from reference [8]).
dealing with separated flow. Next to that, the spreading of the separated flow over the
wing is a three dimensional phenomena not predicted with the vortex lattice method.
Further investigation is required and might result into a good prediction of the maximum
lift coefficient.
The maximum twist constraint and the design lift coefficient is to avoid that the optimizer
starts looking for a wing that is just not feasible and that it does not wast time in
calculating the aerodynamics for them.
When the termination criteria are not met the algorithm starts with establishing a new
geometry and speed for the following iteration. This is part of the constraint optimization
algorithm available in Matlab. As explained earlier it resorts to the sequential quadratic
programming algorithm to establish a new search direction and uses the line search to
establish the distance it will travel in that direction.
Once the new direction and distance is established, a new geometry and speed are de-
termined. Those values will then be used to repeat the twisting of the wing, objective
function evaluation and checking the termination criteria. This process keeps repeating
itself until the termination criteria are met.
54 Implementation of the Theory
The Xfoil calculations on the airfoils form the backbone of the complete quasi-3D method.
As this data will be accessed quite a lot it is of importance that the retrieval of the required
data is fast. Based on this argument it was decided to store the data in a structured way
rather than running Xfoil for every strip and for every wing as this would be to slow and
inefficient.
Storage
Storing the data requires a structured set-up such that the data is easy to find and
retrieve when needed. The database is contained under one single folder in which the
Matlab structures, with the airfoil name as filename, are saved. Such a structure is set
up in layers, as is presented in figure 4.6.
From this figure it is clear that storage of the viscous data is always with a Reynolds
number step of 2 · 104 and with a forced transition step of 0.05 (ranging from 0.05 to 1).
Forced transition is only applied if the transition is aft of the forced transition location
and will be free once it moves in front of the specified location, for higher angles of attack.
When running the quasi-3D method an automatic check of the database will be performed
to establish that the required data is present. The database will automatically be created
or expanded when the requested data is not present. The way this checking algorithm
works can be seen in figure 4.7. From this figure it can be seen that the lowest Reynolds
number available in the database will be 6 · 104 .
Figure 4.7 also shows that when the airfoil is used for the first time a basic database is
created. This includes the basic information of the airfoil and a start of the aerodynamic
database. The basic airfoil and aerodynamic information includes:
• Viscous aerodynamic data from Reynolds number 6 · 104 to 1.3 · 106 with steps of
2 · 104 and free transition
• Viscous aerodynamic data from Reynolds number 6 · 104 to 1.3 · 106 with steps of
2 · 104 and transition forced at 5% chord on the lower and upper surfaces
The inviscid data is saved for angles of attack ranging from −10◦ to 20◦ in steps of 0.1◦ ,
while the viscous data is stored from −5◦ to 20◦ in steps of 0.1◦ . The viscous data is
prone to Xfoil not being able to converge for certain angles of attack. These angles are
automatically skipped as they are not passed to Matlab. This database will be expanded
automatically once different Reynolds numbers or different tripping locations are required.
4.1 Program Setup 55
Figure 4.6: Overview of how the airfoil aerodynamic data is stored in a structured way.
Retrieval
The retrieval of the airfoil data is preceded by an evaluation of the database, as can be
seen in figure 4.7. This evaluation is necessary to ensure that the required data is present.
56 Implementation of the Theory
Figure 4.7: Overview of how the airfoil aerodynamic data is created and retrieved.
Once ensured that the data is present a search algorithm starts looking for the required
data. For that it needs information on several values to establish the location in the
database, these values are:
1. Airfoil name
2. Inviscid or viscous
To clarify how the retrieval works an example of a viscous forced transition data retrieval
will be shown. Assume that the data required is the following
• Alpha-search: α = 0.32
The algorithm knows to start looking in the “ viscous-forced transition” database, as one
of the forced transition values is lower than one. It then looks for the data surrounding the
requested Reynolds number, in this case data with Reynolds numbers 2 · 105 and 2.2 · 105
(remember stored data is available for ∆Re = 2·104 ). Next, it determines the data closest
to the forced transition point locations, or for the example, xtr up = 1 and xtr low = 0.8
(remember stored data is available for ∆xtr = 0.05). This leads directly to two vectors
with angle of attack information. If these vectors are empty it will automatically request
Xfoil to generate the airfoil aerodynamic data, after which it is checked and stored at
this location in the database. Then the two angles of attack surrounding the requested
angle of attack are selected, α = 0.3 and α = 0.4. This provides 4 data locations, two
for each Reynolds number. The last step is to linearly interpolate the corresponding
aerodynamic coefficients, Cl and Cd , for the right angle of attack and for the requested
Reynolds number (see figure 4.8 for the case of a Cl -interpolation)
The interpolation is only performed for the Reynolds number and angle of attack, while
the forced transition is located to the closest value. Interpolating for the forced transition
point is a recommended improvement for future versions of the program. It was omitted
as the closest forced transition was good enough for this research.
58 Implementation of the Theory
Xfoil Settings
The amount of panels defined in the airfoil files, that will be loaded into Xfoil, might differ
substantially. To remove this discrepancy an option in Xfoil is called upon to rearrange
the airfoil paneling before calculating the aerodynamics. With this option the number
of panels and its distribution can be changed. The number of panels for the quasi-3D
calculations is set to 240 while the remaining options, for defining the distribution, are
left at their default values. A clear illustration of this panel rearrangement can be seen
in figure 4.9. It is clearly visible that the leading edge has a way higher panel density and
that it also increases a bit towards the trailing edge.
Figure 4.9: Comparison between original paneling (upper airfoil) with 77 panels and the
rearranged paneling with 240 elements (lower airfoil).
Another important feature in Xfoil is the prediction of the transition point on the airfoil,
which is dependent on the ambient turbulence level. The theory behind this is rather
complicated, however more information can be found in reference [7]. In Xfoil the ambient
turbulence can be controlled with the critical amplification ratio, ncrit . The higher this
value the lower the ambient turbulence. This ncrit is left at the default value of 9, which
corresponds to a turbulence level of 0.07%. The value has an rather significant impact
on the drag when changed between the extreem of high ambient turbulence (ncrit = 6,
turbulence level = 0.245%) and a very low one (ncrit = 14, turbulence level = 0.009%) as
can be seen in figure 4.10.
The normal procedure is to test the airfoil in a wind tunnel and based on those results
a convenient ncrit is selected. However the wind tunnel turbulence level (LST turbulence
level at 20m/s is 0.015% see reference [3]) is different from the atmospheric turbulence
level, which in turn depends on the weather when considering low altitudes. As this
research deals with the optimization of low flying mini-UAV’s, an intermediate value of
ncrit = 9 is selected for the optimizer. On a clear day without too much wind this value
will probably be too low, but at the same time too high for a very gusty day.
4.2 Differences between the Quasi-3D Method and the Theory 59
AVL Settings
The only setting of importance for the vortex lattice method is the grid layout. The grid
is defined for half the wing and AVL is notified of the fact that it is symmetric. The
wing, independently of the number of sections, will have 40 elements in spanwise and
20 in chordwise direction. Resulting in a total of 800 elements for half the wing. The
distribution of these panels is such that they are more concentrated towards the edges and
the root, as the pressure gradients are higher at these locations. However the density for
the tip and the leading edge are the highest. A typical grid layout for the vortex lattice
method can be seen in figure 4.11.
Two empirical corrections have been made to the quasi-3D method such that better results
are obtained. A correction is introduced for the problem with the induced drag estimation
discussed in section 3.1.2. While the other one is to better estimate the zero lift angle of
attack (see section 3.1.1)
60 Implementation of the Theory
As mentioned earlier, the induced drag is overestimated by the quasi-3D method. This
problem arises as there might be a difference between the induced angle of attack required
for the determination of the correct lift coefficient and the one for tilting the lift vector
backward. In reference [15] it is mentioned that there is a difference in these angles,
however it is not explained why there is a difference nor how to calculate it.
The solution to this problem proved to be simpler than expected. Using the generally
accepted formula for estimating the induced drag:
CL2
C Di = k (4.3)
πA
In this formula the value of k is kept constant at 1 which is not entirely correct but proves
to work rather well as will be shown in the validation of the code. However this definitely
needs more attention as it is not what the author had in mind for calculating the induced
drag.
To clarify, the quasi-3D method is used to establish the profile drag and its variation with
the lift coefficient, while the induced drag is calculated using equation 4.3.
The vortex lattice method proved to underestimate the lift coefficient. This was mainly
due to the difference in the zero lift angle of attack, as explained earlier, caused by the
assumptions made in the vortex lattice method. A method to correct for this, that seemed
to work extremely well, was to shift the zero lift angle of attack of the wing towards the
zero lift angle of attack of the viscous airfoil data. Without twist in the wing the zero lift
angle of the wing and the airfoil should be the same. This is due to the fact that there is
no lift and thus no downwash angle present.
4.2 Differences between the Quasi-3D Method and the Theory 61
Equation 3.8, used to determine the effective angle of attack, was accordingly changed to
use the α0 and Clα from the viscous airfoil data. This equation then became:
Cl
αef f = + α0visc (4.4)
Clαvisc
The improvement suggested here can be seen in figure 4.12 and 4.13. The results pre-
sented in figure 4.12 are found using the quasi-3D method with equation 3.8 used to de-
termine the effective angle of attack. Figure 4.13 shows results for the quasi-3D method
with the above mentioned improvements, resulting in equation 4.4 to determine the ef-
fective angle of attack. The difference in the lift curve of figure 4.12(A) have completely
disappeared in figure 4.13(A) while the estimation of the maximum lift over drag ratio
maintained its accuracy.
The validation of the quasi-3D method is performed using CFD results and wind tunnel
data. The CFD calculations are obtained during this research and their purpose is to
define the gradient in the maximum lift over drag ratio for wings with varying sweep and
taper. The validation with wind tunnel data was done by comparing the results obtained
with the quasi-3D method to wind tunnel data found in literature (reference [5] and [6]).
The selection of the wing shapes for the CFD calculations were such that they comple-
ment the wind tunnel results found in literature. Those wind tunnel results are however
obtained at a high Reynolds number (Re = 3.2 · 106 ) when considering mini-UAV’s,
and have a higher than expected aspect ratio. However validation for different Reynolds
numbers and aspect ratios only proves the capabilities and limits of the quasi-3D method.
The CFD-results for the validation were obtained using Fluent 6.3 and Gambit was used
to create the grid.
The first step in obtaining CFD results, is the tedious work of generating a grid and
providing the appropriate Fluent settings. In reference [20] an extensive investigation
was presented into a good grid size and on the preferred Fluent settings, for low Reynolds
number wings. The results obtained with CFD assume a fully turbulent flow which is
approximated in the quasi-3D method by using airfoil data with a forced transition point
at 5% of the chord2 .
The Grid
The dependency of the results on the grid were determined in reference [20] and will not
be investigated further. Having a similar grid and the same Fluent settings as reference
[20] was used to ensure good results.
The domain around the wing is a C-grid with the dimensions shown in figure 4.14. The
grid itself is constructed out of ∼ 1·106 hexahedral elements with a boundary layer around
the wing and the grid density decreases towards the outside of the domain. The grid is
refined towards the leading and trailing edge of the wing and the tip, clearly illustrated in
figure 4.15. This is to capture the large pressure gradients at those edges. Hexahedral cells
were selected as they can have a higher aspect ratio in the direction of the flow reducing
numerical diffusion. However this results in a higher density of the cells at positions were
it is not necessarily needed, which might increase the calculation time.
The boundary conditions for the C-grid are rather straight forward. The front side is
defined as a velocity inlet for which only the velocity magnitude and the direction must
2
Forced transition at 0% of the chord is not possible in Xfoil. 5% is the closest to fully turbulent in
the quasi-3D method
64 Implementation of the Theory
Figure 4.15: Refinement of the grid around the wing. The denser grid to capture the
boundary layer is clearly visible in the left picture. The refinement of the grid
near the edges (leading edge , trailing edge and the tip) can be seen in the top
view of the wing in the right picture.
be specified. The magnitude corresponds to the free stream velocity and the direction
is defined by the angle of attack. The outlet is a pressure outlet in which the gauge
pressure is set to zero (thus setting the boundary condition equal to the static pressure)
and the direction is specified by the angle of attack. The symmetry-plane only requires
the indication of being a symmetry-plane. The opposite side, from the symmetry plane
was also defined as a symmetry plane. However defining it as a pressure outlet would
have been better, but this has no influence on the results as it is far away from the wing.
The wing itself is defined as a wall with a no slip condition, with the roughness height set
to zero and the roughness constant equal to 0.5. This ensures a smooth surface.
The settings used in reference [20] are copied for the CFD-calculations. A steady, coupled,
pressure based solver was selected with an absolute velocity formulation and a Green-
4.3 Quantitative Validation of the Quasi-3D Method 65
Gauss node based gradient option. Steady flow was selected due to the time independence
of the flow. A pressure based solver was used because this solver is better in handling
incompressible or mildly-compressible flows. As the problem at hand only involves very
low speeds (< 35m/s at sea level), compressibility effects are not expected. Under the
pressure based solver one can specify the option of a segregated or a coupled algorithm
for the pressure-velocity calculation. The coupled algorithm has the advantage of better
convergence compared to the segregated algorithm, but this comes at the cost of more
memory usage. The Courant number for this coupled algorithm is set to 10 and serves the
purpose of stabilizing the calculation. This is a rather small Courant number, meaning
that the calculation does not require a lot of implicit relaxation to establish a converging
calculation.
Another option for the calculation is the velocity formulation. The purpose of this option
is to minimize the numerical diffusion by selecting a relative or absolute velocity formula-
tion. For the problem at hand the absolute formulation was selected as most of the flow
in the domain is not rotating. The Green-Gauss node based gradient option was used
because of its increased accuracy.
A second-order discretization was selected for the pressure, as this provides less numerical
discretization errors, in the form of numerical diffusion, and thus better results for hexa-
hedral grids. This however decreases the convergence but is compensated for by selecting
the coupled algorithm. For the momentum and the turbulent viscosity, the discretization
scheme was set to second order upwind both with an explicit relaxation factor of 0.75, to
stabilize the calculation. In case of unstable calculations this number can be lowered to
0.5.
The Spalart-Allmaras model was used for the turbulence model. This is a RANS-based
turbulence model with a single equation linear eddy viscosity model. It was because it
has several advantages for the calculations under consideration:
The turbulence specification method for the boundary conditions was set to the turbulence
viscosity ratio, whose value was set to 40. The value of this ratio has a small effect on the
drag and no effect on the lift. The difference in the drag coefficient between a viscosity
ratio of 1 and 40, was only 0.001. Results obtained with a turbulence viscosity ratio of
40 appeared to be better compared to reference [20] and was thus selected.
Comparison
Several single tapered wings with carefully selected shapes were processed using Fluent.
The names of the wings relate back to there shape, e.g. wing201000, refers to a wing with
20 degree sweep, a taper ratio of 1 and a twist of 0 degrees. So the first two numbers
refer to the sweep angle, the following to define the taper ratio times 10, and the last two
are the linear twist angle, between the root chord and the tip chord, times 10.
The wing shapes tested using CFD are:
66 Implementation of the Theory
• wing001000: Λ = 0◦ , λ = 1, δ = 0◦ , A = 3
• wing000500: Λ = 0◦ , λ = 0.5, δ = 0◦ , A = 3
• wing000300: Λ = 0◦ , λ = 0.3, δ = 0◦ , A = 3
• wing201000: Λ = 20◦ , λ = 1, δ = 0◦ , A = 3
• wing401000: Λ = 40◦ , λ = 1, δ = 0◦ , A = 3
See appendix B for detailed geometry descriptions. The CFD calculations were performed
at a constant velocity of 20m/s, corresponding to a root Reynolds number of 232000. The
aspect ratio was taken constant such that only the viscous effects on the drag could be
observed. The induced drag would change significantly with changing taper due to the
change in aspect ratio. The span was reduced to compensate for the loss in surface area,
while the root chord remained the same.
The first quasi-3D validation is its capability of estimating the maximum lift over drag
ratio and its corresponding lift coefficient.
Table 4.1: Comparison of (CL /CD )max and its corresponding lift coefficient, between CFD
and the quasi-3D method.
From table 4.1 it is clear that the absolute value of the maximum lift over drag ratio,
is accurate within 4%. The error increases with decreasing taper ratio, but the gradient
of the maximum lift over drag ratio with respect to the taper is pointing in the correct
direction (see figure 4.16). The gradient for the quasi-3D method is not as large as for
CL
the CFD-calculations. The decrease in the ( C D
)max with decreasing taper ratio is caused
by the decrease in the overall wing Reynolds number. This causes the profile drag to
increase as the boundary layer around the wing thickens.
CL
The absolute values of ( C D
)max for wings with varying sweep and no taper show results
closely resembling the CFD-results. The gradient however proofs to be a problem, as
CL
they are opposite to each other. The CFD-results suggest an increasing ( C D
)max with
increasing sweep while this is not the case for the quasi-3D calculations (see figure 4.17).
The gradient in both cases is very small. Dealing with flying wings will nevertheless require
some sweep to establish the correct moment coefficient through twisting. Future research
might focus on this to improve the gradient such that non-flying wing optimizations are
not effected too much by this problem.
In the following sections results for every wing under consideration, will be discussed in
detail.
4.3 Quantitative Validation of the Quasi-3D Method 67
CL
Figure 4.16: Variation of ( C D
)max against the taper ratio.
CL
Figure 4.17: Variation of ( C D
)max against the sweep.
68 Implementation of the Theory
Wing001000
There is a close fit between the CFD and the quasi-3D results as can be seen in figure
4.18. At α = 0, there is a small difference in the lift(∆CL = 0.0067) and in the drag
(∆CD = −0.0011). The differences in the lift and drag graphs are clearly visible in the
lift over drag ratio of graph (C). The overestimated lift and the underestimated drag
amplify each other in the lower regions of the graph. However the underestimated lift
gradient and the overestimated drag gradient, in graph (A) and (B), causes the lift over
drag ratio to be slightly underestimated for α > 7.5. From graph D it can be seen that
CL
the induced drag is slightly overestimated. The ( C D
)max however is very close to the one
predicted by Fluent and is located at about the same lift coefficient (see plot (D) and
table 4.1)
4.3 Quantitative Validation of the Quasi-3D Method 69
Wing000500
This wing shows the same difference as for wing001000 concerning the lift curve, namely
the lift curve slope is less steep and the lift is a bit higher at α = 0 for the quasi-3D
calculation. The drag curve in figure 4.19(B) shows that at α = 0 the drag coefficient is
exactly the same while the gradient is a bit underestimated. These differences cause the
lift to drag ratio in graph (C) to be overestimated in the complete alpha range. The drag
however seems to increase nicely with the lift coefficient (see graph (D)).
Wing000300
The estimated lift curve slope appears to be the same as for the CFD results. The
dissimilarity in lift coefficient at α = 0 is almost negligible (∆CL = 0.003). The trend of
a increasing underestimation of the drag curve slope with decreasing taper ratios is also
observed for this wing, as can be seen in figure 4.20(B) and (D). At the same time, the
drag at α = 0 is also a bit to low (∆CD = 8 · 10−4 ). These two differences cause the lift
over drag ratio to be overestimated with a growing error towards larger angles of attack.
From graph (D) it can be seen that the induced drag is underestimated. The maximum
lift over drag ratio also appears to be estimated a slightly higher lift coefficient, as can be
observed in table 4.1 and graph (E).
Wing201000
As for wing001000 the lift gradient is underestimated, but the lift coefficient at α = 0 is
only slightly overestimated (∆CL = 0.0027). The drag slope in figure (B) seems to follow
the CFD results quite nicely however when looking at plot (D) it can be seen that the
induced drag is slightly overestimated. The lift over drag curve in graph (C) shows an
CL
extremely nice fit at and around the maximum ( C D
)max .The CL location of this maximum
CL
( CD )max is very close, as can be seen in graph (E) and table 4.1.
Wing401000
As for the other wings with taper ratio 1, the lift curve slope is underestimated. With
increasing sweep a decreasing trend in the lift coefficient at α0 , with respect to the one
obtained from CFD, can be observed. The drag curve slope in graph (B) is too low which
is a consequence of the lower lift at the same angle of attack. The induced drag however
shows a nice fit in figure (D). The overall underestimation of the lift and the overestimation
of the drag in the lower α-range cause an underestimation of the lift over drag ratio. The
error decreases at higher angles of attack (see graph (C)) as both the lift and the drag
CL
are underestimated at these angles. Graph (E) shows a more constant difference in ( C D
),
CL
and it can be seen that the CL location of ( CD )max is predicted accurately.
The wind tunnel measurements for validation were retrieved from old NACA reports,
CL
reference [5] and [6]. Again several wings were selected to define the gradient of ( C D
)max
with respect to the sweep and to establish the accuracy for higher aspect ratios. The
wings all have aspect ratios higher than the once defined for the CFD-validation. Like in
the previous section, the name of the wing reveals information about the geometry:
• NACA2400: Λ = 0◦ , λ = 0.5, δ = 0◦ , A = 6
• NACA3-10-18: Λ = 0◦ , λ = 13 , δ = 0◦ , A = 10
The first four wings start their name number with 24 which refers to the airfoil used,
namely a NACA2415 for the root and a NACA2409 for the tip. The taper ratio of these
wings remains unchanged, while the sweep (referred to by the following number(s)) and
twist (referred to by the last number(s)) are the only geometric variables. The first
number in the NACA3-10-18, refers to the inverse of the taper ratio while the second
number indicates the aspect ratio. The last number refers to the root airfoil thickness.
This root airfoil is a NACA23018 and the tip airfoil for this wing is the NACA23009.
Detailed dimensions of these wings can be found in appendix B.
The NACA wind tunnel results date back to the end of the thirties of the previous century,
during which the subsonic NACA tunnels had a high turbulence level. This is why, like
for the comparison to CFD, the boundary layer is forced to transition at 5% of the airfoil
chord. The velocity for the quasi-3D simulation was set at 272m/s such that the root
Reynolds number was about 3.2 · 106 , the same as for the wind tunnel data.
Lets first take a look at the performance of the quasi-3D method in estimating the
(CL /CD )max and its corresponding CL .
Table 4.2: Comparison of (CL /CD )max and its corresponding lift coefficient, between ex-
perimental results and the quasi-3D method.
It is clear, from table 4.2, that the errors for (CL /CD )max are rather small (< 3.5%),
however for the swept cases they are biggger than when compared to CFD. Another
interesting point is that the (CL /CD )max -gradient, against the sweep, is very close to
4.3 Quantitative Validation of the Quasi-3D Method 75
zero for the quasi-3D method, while the experimental gradient compares closely to the
CFD results. The comparison of the quasi-3D results to the experimental results of
(CL /CD )max for varying sweep can be seen in figure 4.23.
Table 4.2 also shows that the quasi-3D method works fine for higher aspect ratios and
wings with a significant twist. The error remains in the same order of magnitude.
CL
Figure 4.23: Variation of ( C D
)max against the sweep.
In subsequent sections, results for all wings under consideration will be discussed in more
detail.
76 Implementation of the Theory
NACA2400
The lift curve, for this wing, has a slightly lower zero lift angle of attack (∆α0 = 0.80◦ )
which expresses itself trough the complete curve. In figure 4.24(B) it can be seen that
there is a shift in the minimum drag angle of attack. This is probably due to the shift
observed in the lift coefficient as the difference is not seen in graph (D). In this same graph
it can be seen that the induced drag is slightly underestimated. The shift in lift and drag,
with respect to the angle of attack, is even better seen in plot (C) which is also shifted
towards lower angles of attack. However when the lift over drag ratio is plotted against
the lift coefficient, in plot (E), a nice fit is observed for the lower lift coefficients. Due to
the underestimated induced drag, a slight overestimation is observed for the higher lift
coefficients.
4.3 Quantitative Validation of the Quasi-3D Method 77
NACA24150
The lift curve in figure 4.25(A) nicely fits the wind tunnel results only a minor shift in α0
can be observed (∆α0 = −0.21◦ ). The difference in the CD-alpha curve is composed of a
difference in the zero lift drag of ∆CD0 = 8 · 10−4 , and an underestimated induced drag,
as can be seen in graph (B) and (D). The lift over drag curves in figure (C) and (E) fit
very nicely over the complete domain with only a minor underestimation of (CL /CD )max ,
as can also be seen in table 4.2.
NACA24300
The same conclusions as for NACA24150 can be drawn for this wing. A nice fit for the lift
curve slope, but a slight shift in the zero lift angle of attack can be observed in graph (A).
The drag is slightly overestimated at the point of no lift and the induced drag is slightly
underestimated (see graph (B) and (D)). The lift over drag graphs in plot (C) and (E)
are a bit too low at their maximum point and are slightly overestimated at higher angles
of attack/lift coefficients.
NACA243085
This wing shows a very good fit in all graphs up to about α = 13◦ , after which the
lift curve from the wind tunnel results shows a decrease in the lift curve slope. This is
probably due to partially separated flow on the wing. The induced drag for this wing is
nicely estimated as can be seen in graph (D). This shows that the aerodynamics for wings
with a large amount of twist are nicely estimated using the quasi-3D method.
NACA3-10-18
Up until this point, the wings used in the validation had a rather low aspect ratio (CFD
validation A = 3, Experimental validation A = 6). The NACA3-10-18 was selected
because of its higher aspect ratio (A = 10). The results for this wing, as seen in figure 4.28,
show no larger errors as for the other wings. The only difference with respect to the other
NACA-wings, is that the maximum lift over drag ratio is slightly overestimated instead
of underestimated. The induced drag for this wing also shows a slight underestimation
as can be seen in graph (D).
Some general remarks after the comparison of the quasi-3D method to CFD and wind
tunnel results are in place.
It was noted that the maximum lift over drag ratio is estimated within a bound of 4%
accuracy. This is very acceptable knowing that a small difference in the hard to determine
drag coefficient, makes a large difference in the lift over drag ratio.
CL
In general it can be seen that the error in ( C D
)max is increasing with decreasing taper
ratio. This is mostly caused by an under prediction of the induced drag. However the
gradient of the maximum lift over drag ratio against the taper is estimated rather good
and are unlikely to produce problems when using the optimizer.
The error for the sweep has the tendency to increase for increasing sweep, but remains
within a bound of 3.5% even for a sweep of 40◦ . When looking at the CFD-comparison
the induced drag seems to be overestimated, while when compared to wind tunnel results,
the induced drag is underestimated by the quasi-3D method. This difference might be
caused by the difference in aspect ratio. The wings used for the CFD-validation have an
aspect ratio of three, requiring a correction for tip-vortex effects on the wing. This is not
the case for the NACA-wings.
CL
A point of concern is the gradient of ( C D
)max against the sweep. The quasi-3D method
does not match the gradients of the CFD and wind tunnel Results, which in turn compare
very well to each other. This will be a problem when optimizing with an optimization
algorithm that relies on gradient information. The tailless aircraft under consideration
are not as prone to this problem as they require a combination of sweep and twist to trim
the aircraft. The higher the sweep the less twist required, which will push the gradient
into the correct direction. On top of this the gradient observed in the validation data is
rather small, as can be seen in figure 4.17 and 4.23.
A final remark is the fact that the accuracy of the quasi-3D method is not affected by
larger aspect ratios nor by a lot of twist.
The reasons for developing the quasi-3D method, as mentioned in the beginning of this
thesis, are:
• Having the ability to predict the viscous effects at low Reynolds numbers.
Now it is interesting to verify if these requirements are met, and to see if and how much
they have improved.
4.4 Qualitative Comparison of the Quasi-3D Method to Fluent 83
The comparison made here will be between the time it takes for Fluent and the quasi-
3D method to present their results. Both calculations were performed on a HP dv 9700
Notebook PC, with a Intel(R) Core(TM)2 Duo CPU T7250 Processor, 3GB RAM, and
a Vista 32-bit operating system.
The time for an aerodynamic evaluation will be defined as follows: The time passed
between the start of the aerodynamic calculation and the presentation of the aerodynamic
coefficients, for a single angle of attack.
This definition excludes the construction and validation of the computational domain in
the form of a mesh, and the time it takes to get the settings right in the tool used for the
aerodynamic evaluation.
A Fluent evaluation with the grid and settings described in section 4.3.1, consumes on
average 2 hours and 15 minutes. For the quasi-3D method this is about 1 minute.
A single objective evaluation, during the optimization, takes at least 9 aerodynamic eval-
uations. These are required to get trimmed and level flight. The optimization requires
about 200 objective evaluations. This results in an optimization time of at least:
This illustrates the fact that Fluent is very impractical to use for the optimization of
the EMAV2009 aircraft, it would just require too much time. However one has to keep
in mind that when new airfoils are used in the quasi-3D method an addition of about
3 hours is required to create the basic airfoil database. The optimizer is developed for
low Reynolds numbers, whenever high Reynolds numbers are used the data needs to be
added to the database increasing the time required to evaluate the aerodynamics with
the quasi-3D method.
The reduction in time is definitely illustrated. It can even be stated that the quasi-3D
method makes an optimization a practical option.
Low Reynolds number effects in this report mainly refer to the laminar separation bubble
and its influence on the aerodynamic forces. As the maximum lift over drag ratio is mostly
obtained at lower lift coefficients one does not have to bother too much about viscous
effects, like large areas of separated flow, that normally occur in the near stall region.
Using Fluent the easiest and fastest way to account for viscous effects is using RANS-
models. They however do not predict the location of the transition point nor the effects
of the laminar separation bubble. It would not be correct to state that the quasi-3D fully
captures these effects, as they are influenced by the 3-dimensional flow, but it does at
least take them into consideration through the viscous airfoil data.
84 Implementation of the Theory
For Fluent the computational domain consists out of a volume mesh. This requires that
a volume is defined around the wing which is then cut into smaller volumes creating the
mesh. Producing a good mesh is important to obtain good results, requiring a denser
mesh in certain parts of the volume. This, however, requires a lot of time and experience.
Doing this automatically is rather difficult and does not ensure good meshes.
The quasi-3D method requires two meshes, one for the vortex lattice method and one for
the airfoil calculations in Xfoil. The advantage of these meshes is the fact that they are
surface meshes meaning that only the surface of the wing or airfoil needs to be meshed.
Refining these meshes in the critical areas does not require a lot of experience. The
construction of these meshes only need a limited amount of input.Changing these variables
and validating the results, provides a feel for what kind of meshes work well and what
inputs should be avoided. This experience can then be translated into a program that
creates these meshes automatically while ensuring good results.
Chapter 5
The second part of the thesis work was to design, build and analyze the TU Delft com-
petitor in the EMAV2009 endurance competition. This required a lot of practical work
for which the author was assisted by a group of students performing their DAR-exercise
on different parts of this practical work. The subjects of these different DAR-exercises
are:
• Design and construction of the master molds for the EMAV2009 mini-UAV
• Building the production molds and the prototype model for the EMAV2009 mini-
UAV
• Designing and building a wind tunnel force balance for mini-UAV wind tunnel force
measurements
• Design of the propulsive system and selection of the energy source for the EMAV2009
mini-UAV
These exercises were closely monitored, such that they could neatly fit into this research.
All these topics will be addressed in this chapter1 to a level of detail suited for this
report. Some later stage changes in the designs, produced during these exercises, will
be explained. Reference will be made to the DAR-exercise reports for more detailed
information on their topic.
Wind tunnel experiments were performed in the scope of analyzing the aircraft for the
EMAV2009 and to validate the quasi-3D method. Two different kinds of tests are required
for this purpose. They consist out of aerodynamic force measurements and boundary layer
visualizations. These visualizations are required to see and understand what is happening
inside the boundary layer at low Reynolds numbers. The wind tunnel experimental setup
will be discussed first in section 5.1 after which the results of the experiments will be
presented in section 5.2.
1
Except for the design of the propulsion system which is explained in section 6.2.2
85
86 Wind Tunnel Experiments
Two wind tunnel campaigns were performed for this research. During these campaigns
the two prototypes for the competition aircraft were tested. One of them had a manually
designed wing while the second one was created with a first version of the optimization
program.
The experiments were performed in the low speed low turbulence wind tunnel of the TU
Delft. This facility is equipped with a six degree balance system. The resolution of this
system is very high, but not high enough to measure the small forces experienced by
mini-UAV’s. This was the reason for designing and building a new balance dedicated to
measurements on mini-UAV’s.
The wind tunnel models were created at the MAVLab using the same techniques as for
the competition aircraft, such that experience was gained in building mini-UAV’s before
building the competition aircraft.
The TU Delft low speed low turbulence Tunnel (LTT) is a closed-throat single-return
atmospheric wind tunnel. It achieves the low free stream turbulence levels trough a large
contraction ratio of 17.8. Its turbulence level is only 0.015% at 20m/s and 0.07% at
75m/s. The top speed of this tunnel is about 120m/s which it obtaines with its six
bladed fan and a 535 kW DC motor.
2
Information obtained from reference [3]
5.1 Wind Tunnel Measurement Setup 87
The test section of the tunnel is interchangeable such that test preparations can be per-
formed while the tunnel is running. The size of these test sections is 1.8m wide, 1.25m
high and 2.6m long.
The measurement equipment, available at the LTT, consists of an electronic pressure
scanner, a electronically read multi manometer, a six component balance and an infra-red
visualization system. This infra-red system is the only one of interest to this work as
it will be used for boundary layer visualizations. Force measurements will be performed
using a newly created balance system explained in the following section.
5.1.2 Balance
As mentioned before, a new force measurement system was required for mini-UAV’s as
they experience far smaller forces than the average model tested in the LTT. The most
critical of all forces is the drag force. The LTT force balance has a resolution of about
4 ∼ 5gr which is not nearly high enough to accurately measure the lowest drag force on
the prototype aircraft, which is about 27gr. This was the reason to resort to a new force
measurement system.
Hardware
The design and construction of the new force balance was performed during a DAR-
exercise. The details of this design are presented in reference [33].
The balance hardware consists of two different parts, one is the force sensor and the other
one is the yaw-pitch system. An of-the-shelf force sensor was selected based on several
criteria.
• Maximum force and moment range
• Resolution
• Size
• Flexibility to use for several types of measurments
These criteria were all checked and it was decided to use the 6 axis “ATI Nano 17 SI-25-
0.25” force transducer. This sensor has a resolution of 1/160N = 0.637gr for the 3 forces.
This is considerably better for accurate drag and lift measurements than the available
LTT-balance. A picture of the complete sensor system can be seen in figure 5.2. This
figure shows the sensor, signal conditioner and the data acquisition card. Details on the
Nano 17 sensor and its selection process can be found in reference [14] and [33]. A data
sheet on the sensor can be found in appendix C.
The yaw-pitch system can be seen in figure 5.3. The purpose of this system is to set
the angle of attack and the yaw angle during the wind tunnel measurements. Two digital
servos were used to set the angles. However, this system has no feedback loop between the
servo setting and the yaw/pitch-angle and thus requires a calibration before the tunnel
tests start. During the calibration a table has to be created that couples the servo setting
to a certain angle.
A pictures of the finished balance can be seen in figure 5.4.
88 Wind Tunnel Experiments
Figure 5.2: Nano 17 force/torque sensor system with data logging card.
Figure 5.4: Finished yaw-pitch system with the sensor connected on top.
in the yaw-pitch system, the sensor was found to produce very good results. Setting the
angles correctly proved to be hard due to several causes, summarized here:
• There was some mechanical play combined with friction on the connections between
the servos and the rotating head of the balance.
• The digital servo potentiometer, which produces the feedback between the servo
input and output, was located on the second to last gear. This results in the play
between the last two gears, not being eliminated.
• No clear pitch and yaw angle indication was present nor a zero angle indication.
• During the wind tunnel tests the forces put on the system cause strain and thus
misalignment with respect to the calibration (performed without forces on it).
• At high angles of attack the pitch servo was not strong enough.
• Construction of the balance was rushed in the final stages resulting in less accurate
finishing.
The electronic-board connecting the computer to the yaw and pitch servo is a stripped
down versions of an autopilot, which in essence is developed for controlling servos. The
link between the computer and the servos is the only part that remained on it. The
attitude sensors and GPS-receiver were removed as they served no purpose in the balance.
Software
Two software tools were developed for the wind tunnel measurements. One is used to
log all data and to control the yaw-pitch servos on the balance, while the other one is
used for post processing the measurements. At this point, real-time evaluation of the
coefficients is not possible as both programs are not yet linked to each other, this is a
recommendation for future work.
• Atmospheric information
Some of the data are automatically extracted while others have to be manually provided
by the user. An improvement for the future would be the automatic readout of the wind
tunnel settings and atmospheric information, which at this point have to be manually
entered for every measurement point.
• A zero-run file. This is a file that contains sensor measurements for all angles, pitch
and yaw, with the wind tunnel off. Subtracting these measurements from the “wind
tunnel on” measurements ensures that the weight effects are taken removed and
thus only aerodynamic forces are presented.
• The sensor pitch offset. This is the angle between the sensor horizontal and the zero
angle of attack of the aircraft. (see figure 5.6)
• Servo calibration information. This information relates the servo input, saved in
the wind tunnel measurement file, to an angle of attack and yaw angle.
The first step is to read the measurement and zero-run file, followed by averaging of the
available data at each measurement point. The servo inputs for the zero-run file and the
measurement file are then converted into angles. These angles are used to subtract the
corresponding zero-run data from the measurement data. Once this is finished the data
must be rotated from the sensor reference system, which is only rotated over αof f set with
92 Wind Tunnel Experiments
respect to the aircraft body reference frame, towards the aerodynamic reference frame.
Figure 5.7 visualizes the rotation between the two references systems. In matrix form this
can be formulated as:
L sin(α − αof f set ) 0 cos(α − αof f set ) Fx
D = − cos(α − αof f set ) cos(β) sin(β) sin(α − αof f set ) cos(β) Fy (5.1)
S∗ − cos(α − αof f set ) sin(β) − cos(β) sin(α − αof f set ) sin(β) Fz
N sin(α − αof f set ) 0 cos(α − αof f set ) Tx
L∗ = − cos(α − αof f set ) cos(β) sin(β) sin(α − αof f set ) cos(β) Ty (5.2)
M − cos(α − αof f set ) sin(β) − cos(β) sin(α − αof f set ) sin(β) Tz
These aerodynamic forces are then converted into coefficients based on the velocity, wing
surface and the atmospheric density.
These coefficients need to be corrected for solid and wake blockage, the 3D lift interference
for small wings and the buoyancy effect. These corrections are however expected to be
small as the model and the forces acting on it are small. Information for the wind tunnel
corrections was obtained from reference [23],[31],[30] and [13].
The wind tunnelblockage creates an increase in the dynamic pressure as the mass flow
through the tunnel must be constant while the presence of the model, balance and their
wakes blocks part of the cross section. To calculate this effect one must determine the
value of:
ǫSB = ∆V /V∞ = ǫSBw + ǫSBf us + ǫSBbal + ǫSBwake (5.3)
The velocity increase due to the wing and the balance can be calculated using the following
equations (obtained from [23]):
K1 τ1 Vwing
ǫSBw = 3 (5.4)
C′ 2
K1 τ1 Vbal
ǫSBbal = 3 (5.5)
C′ 2
In these equation K1 and τ1 can be obtained from figure 5.8 and figure 5.9. The C ′
represents the wind tunnel cross-sectional area in the test section and Vwing and Vbal ,
are the wing volume and balance volume. The equation for the velocity increase for the
fuselage is the following:
K3 τ1 Vf us
ǫSBf us = 3 (5.6)
C′ 2
5.1 Wind Tunnel Measurement Setup 93
Figure 5.7: Visualization of the rotation between the sensor reference system and the aero-
dynamic reference system.
The wind tunnel cross-sectional area is once again represented by C ′ and K1 and τ1 are
obtained from figure 5.8 and 5.9, respectively. The fuselage volume is represented by Vf us .
The wake blockage is the only value that still needs to be determined. For this equation
94 Wind Tunnel Experiments
Figure 5.8: Values of K1 and K3 for a number of bodies (figure obtained from reference
[23]).
Figure 5.9: Values of τ1 for a number of tunnel types. Use b/B = 0 for bodies of revolution
(figure obtained from reference [23]).
2 1 S
ǫSBwake = C D0 + (2.8 − 0.068 · A)(CD − CDi − CD0 ) (5.7)
4C ′ 2 C′
In this equation CD represents the uncorrected measured value of the drag coefficient.
The CDi and CD0 in this equation are determined by approximating the attached flow
5.1 Wind Tunnel Measurement Setup 95
part of the measured data with a full second order polynomial and can be calculated
using equations 5.8 and 5.9. The A and S are the wing aspect ratio and the wing surface
respectively. As before, C is the cross-sectional area of the wind tunnel test section.
b 2
CDi = a · (CL + ) (5.8)
2a
b2
C D0 = c − (5.9)
4a
In these equations a, b and c represent the coefficients of the second order polynomial
approximation of the attached flow CD (CD = a · CL2 + b · CL + c).
The velocity increases are then corrected for compressibility effects using:
ǫSBw + ǫSBf us + ǫSBbal ǫSBwake
ǫSB = + (5.10)
β ′3 β ′2
qc = q(1 + 2ǫSB )
CDc1 = CD (1 − 2ǫSB )
CLc1 = CL (1 − 2ǫSB )
CSc∗ = CS ∗ (1 − 2ǫSB )
1
CNc1 = CN (1 − 2ǫSB )
CL∗c = CL∗ (1 − 2ǫSB )
1
CMc1 = CM (1 − 2ǫSB )
The second correction is for the 3D lift interference with the wind tunnel walls. The flow
field around the aircraft has changed due to the presence of the wind tunnel walls. This
can be modeled by mirroring the horseshoe vortices with respect to the wind tunnel walls
and evaluating their effect on the flow around the model. A visual representation of this
mirroring can be seen in figure 5.10.
This correction only has effect on the angle of attack and the aircraft moment coefficient
and is modeled using equation 5.12 and 5.13 which are obtained from reference [30].
cmac δ1 S
∆α = δ0 + · ′ · CL (5.12)
2β ′ h C
96 Wind Tunnel Experiments
Figure 5.10: Principle of modeling the lift interference effect trough mirroring horseshoe
vortices with respect to the wind tunnel walls (figure obtained from reference
[30]).
2 !
cmac δ1 cavg 1 2 cavg S
∆CM = · + (A tan(Λ0.5 )) 2 − · C Lα · ′ (5.13)
16β ′ h cmac 3 cmac C
In these equations:
• CLα is the lift coefficient gradient against the angle of attack for the solid blockage
corrected lift curve slope
n=∞ m=∞
Atun X X m2 A2tun − n2
δ0 = (−1)n # 2 , excluding n = m = 0 (5.14)
8π n=−∞ m=−∞ m2 A2 + n2 tun
5.1 Wind Tunnel Measurement Setup 97
n=∞ m=∞
Atun X X m2 A2tun − 2n2
δ1 = (−1)n # 5 , excluding n = m = 0 (5.15)
8π n=−∞ m=−∞ m2 A2 + n2 2
tun
In these equations Atun represents the wind tunnel aspect ratio, defined as the tunnel
width over hight ratio. Equation 5.14 and 5.15 are obtained from reference [13].
These last two equations are the mathematical representation of the vortex mirroring and
are obtained from reference [13], which should be consulted for further details.
The last correction that needs to be applied is the buoyancy effect, which only results in
a change of the drag coefficient. This difference can be calculated using,
′2 CL
∆CD = −(ǫSBw + ǫSBf us + ǫSBbal )(1 + 0.4M ) CD − (5.16)
πA
, in which CL and CD are the measured lift and drag coefficient corrected for the solid
blockage.
Applying the lift interference and buoyancy correction to the coefficients (which are al-
ready corrected for the solid blockage) and the angle of attack gives,
αc = α + ∆α
C D c2 = CDc1 + ∆CD
C M c2 = CMc1 + ∆CM
After the corrections have been applied the measured forces have been properly converted
into aerodynamic coefficients.
The model construction was divided over two DAR-exercises, one with the focus on the
creation of the master molds while the other looked at the construction of the production
molds and the wind tunnel model.3 The reports on these DAR-exercises, reference [9]
and [11], were written such that they can be used as manuals for future projects. It
was decided to use the same production technique for the wind tunnel models as for the
airworthy aircraft. This decision was based on:
1. Knowledge and experience could be gained from construction of the wind tunnel
models before producing the competition aircraft.
2. Costs would be lower if the molds for the competition aircraft could also be used
for the wind tunnel models.
3. Using the same technique makes the wind tunnel results more representative.
3
A master mold is used to make production molds. This way the expensive milled master mold does
not get damaged during the production process.
98 Wind Tunnel Experiments
Four molds were required for the construction of the prototype, two master molds for the
wing and two production molds for the fuselage. These molds were milled out of Sieca
block. This material is especially suited for making molds trough a milling process. The
creation of these molds required several steps:
Figure 5.11: Wing, upper and lower surface, production molds in the process of being pre-
pared for production.
5.1 Wind Tunnel Measurement Setup 99
The fuselage and tail were joint together in one mold but required two steps to make. The
first one was the construction of the tail which consisted of 3 layers, an outside 50gr/m2
glass fiber layer, followed by a 1mm thick Rohacell core and the inside was finished with
another layer of glass fibers. Once the tail was ready the fuselage could be created in 1 step
using the technique of inflating a bladder inside the mold. The fuselage was constructed
from 110gr/m2 Kevlar skin with a second layer of the same material as reinforcements in
the nose and in lengthwise direction of the fuselage.
For further details on the wind tunnel model construction, see reference [11].
The construction of the fuselage would not change throughout the project as it proved to
be a fast and easy production which did not require improvements. The wing construction
however did change. Several reasons for this are:
• Several components needed to be fitted inside the wing which was hard with a shell
structure
• The outgassing of the coating created trapped air between the coating and the
Kevlar
The first two problems were solved by having a foam core in the wing and the last two
problems were solved by removing the coating. The outer shells would be constructed
the same way as before, however this time without the Rohacell core in between the fiber
layers. Then a cnc-hotwire cut foam core would be inserted in each of the shells (see figure
5.12). The core would then be sanded flush with the production molds resulting in a wing
that was exactly cut in a upper and lower half. Holes for the servo system, batteries and
spar are then cut in the foam core enabling the components to be nicely fixed in the wing.
This is followed by gluing the upper and lower half of the wing together with a paste of
epoxy and glass bubbles.
Figure 5.12: Foam core, of which two are needed during the construction of the wing, and
the CNC-hotwire foam cutter.
The total weight of the wing, without the components inside, lowered from 205gr for
the first wind tunnel model to 85gr for the competition model(see figure 5.13). This
illustrates the gain in experience with building lightweight constructions. However future
100 Wind Tunnel Experiments
research into this topic is required as building mini-UAV’s requires very light and delicate
fibers, and very thin structures. The building process explained here is a first step into
this research. The author strongly suggests a close cooperation between the MAVLab,
the structures department and the materials department, to obtain better construction
methods and thus better mini-UAV’s.
During both wind tunnel campaigns the same tests plan was followed. This test plan
involved aerodynamic force measurements and boundary layer visualizations. A detailed
overview of the tests is presented next.
Force measurments
The force measurements were performed to analyze the aerodynamic efficiency and the
stability of the aircraft.
The first set of tests:
• Configuration:
– No turbulator strips
– No propeller mounted
– Elevons in the neutral position
These test have to provide insight in the effects caused by different Reynolds numbers.
Expected differences are the maximum lift coefficient and changes in the drag curve.
The second set of tests:
• Configuration:
• Velocity: 20m/s
Changing the locations and shapes of the turbulator strips will most likely change the
drag curve. The situation that provides the best lift over drag ratio will be used for the
following tests.
The third set of tests:
• Configuration:
During these tests the motor will be set at different power settings. This is to provide an
insight in the power required to maintain straight horizontal flight, and to determine the
efficiency of the propulsion system at different speeds.
The fourth set of tests:
• Configuration:
• Velocity: 20m/s
During these test the elvons will be deflected symmetrically at different upward and
downward angles. This is to determine the elevator efficiency, CMδe .
The fifth set of tests:
• Configuration:
– Turbulator strips added at the best locations
– No propeller mounted
– Asymmetric elevon deflection
• Angle of attack range: 0◦ → stall → −5◦ → 0◦
• Side slip angle range: 0◦
• Velocity: 20m/s
Asymmetric elevon deflections provide information about the aileron efficiency of the
aircraft. It is interesting to investigate this as it is noticed that mini-UAV’s are extremely
responsive to aileron deflections.
The sixth set of tests:
• Configuration:
– Turbulator strips added at the best locations
– No propeller mounted
– Elevons in the neutral position
• Angle of attack range: 3◦ → 8◦
• Side slip angle range: 0◦ → 12◦ → 0◦
• Velocity: 20m/s (15m/s, 25m/s)
Setting the aircraft under different side slip angles provides information about the yaw-
stability of the aircraft. It might be interesting to perform these tests at low speeds to
investigate if the yaw-stability is maintained at low speeds.
Reserve set of tests:
• Configuration:
– Turbulator strips added at the best locations
– Propeller mounted and motor on
– Symmetric and Asymmetric elevon deflections
• Angle of attack range: 3◦ → 8◦
• Side slip angle range: 0◦
• Velocity: 20m/s
These test can be performed when there is still some testing time left. This provides
information on the effect of the propeller on the elevon effectiveness.
5.1 Wind Tunnel Measurement Setup 103
During the first wind tunnel campaign the boundary layer visualizations were limited
to three cases. All of them are performed at the angle of attack corresponding to the
maximum lift over drag ratio and at a velocity of 20m/s. The following situations were
visualized:
The first two boundary visualizations are to investigate the effect of the propeller flow on
the boundary layer. The third one was to investigate if the turbulator strips were effective
in tripping the boundary layer.
During the second wind tunnel campaign the visualizations were expanded to investigate
more situations. The same tests as in the first campaign were performed to establish
differences between the two wing shapes. The added visualizations were to establish
trends in the boundary layer, like the effects of different angles of attack and changes due
to elevon deflections. The added visualizations are:
First set of visualizations:
• Configuration:
– No propeller mounted
– No turbulator strips
– Elevons neutral
• Angle of attack range: 0◦ , 5◦ , 10◦ , 20◦
• Side slip angle range: 0◦
• Velocity: 20m/s
• Visualizations on upper and lower wing surface
• Configuration:
– No propeller mounted
– No turbulator strips
– Elevons 11◦ up
• Angle of attack range: 5◦
• Side slip angle range: 0◦
• Velocity: 20m/s
• Visualizations on upper and lower wing surface
104 Wind Tunnel Experiments
In this section the results obtained during the wind tunnel testing will be presented.
All the tests mentioned in the test plan were performed. However after a thorough
investigation of the force measurements it had to be concluded that some mistakes had
occurred during the measurements. It was discovered that these problems were related
to the yaw-pitch system of the balance. It was decided to use the results from the force
measurements to give a clear analysis of the problems that have presented them selfs such
that they will not occur again. This however implies that the results can not be used for
the intended validation of the quasi-3D tool and the optimizer.
The boundary layer visualization on the other hand are very nice and present some really
clear illustrations of low Reynolds number effects on the boundary layer. Comparisons
and trends observed with these visualizations can be found in section 5.2.2.
The force measurements showed signs that they might be wrong. These signs will be
explained in the first part of this section. After that an overview will be presented of
possible causes that contributed to the errors. The last part will be to show how sensitive
the results are to these mentioned causes.
slightly negative (CL = −0.024), the wind tunnel results however suggest a rather large
lift coefficient (CL = 0.067) at that angle.
There is also a large difference in the lift curve slope. This difference however was not
used as an argument to treat the results as incorrect. A possible cause for the difference
was found and will be discussed later on in this chapter.
Figure 5.14: Lift coefficient against the angle of attack for the S5010 airfoil.
This argument is purely based on a trend discovered in the data while deflecting the
elevons up. Having a elevon up deflection it can be expected that the lift will shift down
and the moment coefficient shift up, at the same angle of attack. This was the case,
however the minimum drag was lower than the ones without elevon deflection. This is
illustrated in figure 5.16.
On top of that the maximum lift over drag and increased to 18 for a 7.5◦ elevon deflection
(see figure 5.17). The only logical explanation at the time was that the lift distribution
improved to a more elliptical one, improving the induced drag. A good validation of the
maximum lift over drag ratio was to check it against the maximum theoretical lift over
drag ratio.
Figure 5.15: Lift coefficient against the angle of attack for the Prototype wing. Experimen-
tal data compared to CFD and the quasi-3D method.
Figure 5.16: Drag polar for elevons in neutral position and for a deflection of 7.5◦ up.
up).
√ s √ r
CL π Ae π 3.07
= = = 13.23 (5.17)
CD max 2 C D0 2 0.01378
5.2 Wind Tunnel Results 107
Figure 5.17: Drag polar for elevons in neutral position and for a deflection of 7.5◦ up.
It was thus discovered that the theoretical maximum lift over drag ratio was exceeded by
38%. This is impossible and required further attention.
The sensor used has a high accuracy and was assumed not to be the cause of the mea-
surement problems. The yaw-pitch system on the other hand already proved to posses
some imperfections. It was concluded that something might be wrong in setting the an-
gles. Another aspect is the fact that the balance is a rather large object in the tunnel
compared to the aircraft. Flows over the balance cover might also have contributed to
the problem.
108 Wind Tunnel Experiments
The software used to control the servo has a higher resolution than the servo digital input,
thats why it is unlikely to have a negative effect. The same is true for the electronics
used to control the servo, as it was developed for autopilot purposes which require high
accuracy.
The servo accuracy is degraded by the fact that the potentiometer is positioned on the
second to last gear and thus does not take into account the play between the last two
gears. The holding force of the servo might also cause problems at the higher angles of
attack as the aerodynamic moment might be higher than the maximum servo force. The
servo mount is unlikely to have any effect as it is a quite sturdy mount together with a
rather small moment-arm acting on it. A picture of the servo mount can be seen in figure
5.18.
The linkage between the servo and the balance head is shown in figure 5.3. Two issues
are involved with this system, namely play in the linkages between different components
and strain due to the forces put on the system. The linkage has some weak spots which
are likely to flex under high stresses. Play in the axis on which the balance head rotates,
to set the pitch angle, has been removed before the first measurements and is unlikely to
cause problems any more. The rotational stiffness of the sensor is known and is presented
in the sensor data sheet which can be seen in appendix C.
The pitch offset between the sensor and the model can be seen in figure 5.6. If this angle
is not entirely correct, the rotation of the forces and moments from the measurement
reference frame to the aerodynamic reference frame will be wrong. This can be seen in
equation 5.1 and 5.2. This error is likely to be present because only a small error in this
angle has a rather large effect on the drag coefficient. In case of a wind tunnel model with
structural flexibility around the connection to the sensor, results in this angle to change
with the angles of attack. This however was assumed not to be the case for the models
under consideration in this report.
So summarizing these issues in angles gives:
5.2 Wind Tunnel Results 109
• ∆αplay
• ∆αmaxHF
• ∆αof f set
The ∆αstrain is a collection of the changes in pitch angle due to the forces and moments
acting on the different components. The ∆αsensor is easy to model using:
−Tyaero
∆αsensor = ; (5.18)
Kt y
In which Kty is obtained from the sensor data sheet in appendix C and Tyaero is the
measured moment around the y-axis with the zero-run measurement subtracted. The
other components in the ∆αstrain are harder to determine. It is possible to determine a
similar value as the Kty in equation 5.18 but than for the complete mechanical system
from servo gears to balance head. However a better solution to the problem would be
to measure the angle at the head of the balance or even better, creating a feedback loop
between the sensor head and the servo input.
The ∆αplay and ∆αmaxHF are discontinues errors. The play in the mechanical connections
is pushed into the same direction as the force acting on it. A constant change in the
pitch angle can be noticed once the force switches from one direction to the other. The
∆αmaxHF is the change in the pitch angle once the force acting on the servo exceeds
the maximum holding force. The servo will keep the pitch angle fixed at the angle were
the maximum holding force balances the aerodynamic forces and moments. An angle
measurement at the head of the balance or a feedback loop will also bring solution to
these discrepancies.
110 Wind Tunnel Experiments
The pitch offset has to be measured before the wind tunnel tests and should be very
accurate as a small difference in this angle results in rather large errors in the coefficients
as will be shown in section 5.2.1.
A detailed investigation, probably involving wind tunnel testing, will be required to state
something useful concerning this subject.
Effect of Strain
Lets try to define a correction for the angle change due to strain. Figure 5.19 shows all
the forces involved in producing deformations in the model, sensor and yaw-pitch system.
The force and moment due to the weight of the model were taken into account during the
calibration of the angles and can thus be emitted from the problem. With the remaining
forces, moments and dimensions equation 5.19 can be formulated. The Tyaero and the
Fzaero are shown, in figure 5.19, in positive direction.
1
∆αstrain = (Tyaero + Fzaero · x1 ) (5.19)
Kttot
In this equation Tyaero and Fzaero are the measured aerodynamic moment around the y-
axis and the measured aerodynamic force in the z-direction, respectively. The torsional
stiffness of the complete system, Kttot , can be estimated with the values previously defined
in section 5.2.1. Assuming a ∆α = 1.3◦ was observed and the change in aerodynamic
force and moment due to the elevon deflection are established at ∆Fzaero = −0.7299 and
5.2 Wind Tunnel Results 111
Figure 5.19: Forces acting on the balance head causing strain in the system.
∆Tyaero = −0.1256. The value of x1 was measured to be 1.35 · 10−2 m. Then Kttot is
defined as,
(∆Tyaero + ∆Fzaero · x1 ) Nm
Kttot = = −0.104 [ ] (5.20)
∆αmeasured deg
This creates a large change in the aerodynamic coefficients as can be observed in figures
5.20 to 5.22. The observations from these figures can be summarized in:
• The maximum lift over drag ratio has changed drastically due to the previous men-
tioned changes.
Conclusion: The effect of strain in the system is unforgiving when it is not modeled
correctly.
Figure 5.20: Changes in the lift coefficient due to a strain correction on the results.
Figure 5.21: Changes in the drag polar due to a strain correction on the results.
due to a remaining difference in the angle of attack, as was also observed in section 5.2.1.
It is unlikely to be aerodynamic hysteresis as this only occurs after stall and disappears
completely when the angle of attack is decreased below a certain threshold.
5.2 Wind Tunnel Results 113
Figure 5.22: Changes in the lift over drag ratio due to a strain correction on the results.
Figure 5.23: Drag polar for angles of attack going from 0◦ → 22◦ → −8◦ → 0◦ .
During some of the measurements it was noted that the servo was unable to set the angle
of attack above a certain value. When this is not taken care of, it might lead to thinking
that the stall angle of attack is reached, as the input to the servo increases, thus believing
to be at higher angle of attack, while the lift coefficient remains constant.
114 Wind Tunnel Experiments
Figure 5.24: Effect of the pitch-offset angle on the lift over drag ratio.
During each of the two wind tunnel campaigns a different model, Prototype A and B,
was tested. The fuselage was kept the same for both models while the wing geometry
changed. A detailed geometric description of both wings is presented in table 5.2. A
3-view of each model can be found in appendix D.
In this section several boundary layer visualizations will be compared to each other. This
is to provide an insight into the flow phenomena on finite wings submitted to low Reynolds
numbers. They will help to establish the best inputs for the optimization tool.
The visualizations presented in this section are rather unique. No visualizations could be
5.2 Wind Tunnel Results 115
found in literature that showed a boundary layer on a low aspect ratio wing at these low
Reynolds numbers (∼ 2 · 105 ).
The angles of attack mentioned in this section are all angles at which the aircraft was set
without any of the previous mentioned corrections. Whenever referring to left and right
in this section, it is assumed to be from the readers point of view.
A picture of the boundary layer for each prototype can be seen in figure 5.26. These
pictures are taken when both aircraft were under an angle of attack of 5 degrees without
any turbulator strips, and without the propeller.
In these pictures a top-down view is seen from the upper surface of the wing. The leading
edge is pointed to the upside of the page.
To start this explanation some typical features of a low Reynolds number boundary layer
will be addressed. At the leading edge of the wing, the boundary layer starts as a laminar
layer which is represented by the darker area above the bright yellow area. As explained
earlier, the laminar boundary layer separates from the wing surface after which it desta-
bilizes and finally transitions into a turbulent boundary layer. This turbulent boundary
layer reattaches to the wing leaving behind a so called laminar separation bubble. This
bubble is clearly visual as the bright yellow region on the wing. After this region the
boundary layer is turbulent up until the trailing edge. These are typical features which
can also be seen in an Xfoil-calculated pressure distribution over the airfoil (See figure
5.27).
116 Wind Tunnel Experiments
Figure 5.26: Boundary layer visualization for prototype A (upper) and prototype B (lower)
at α = 5◦ .
Figure 5.27: Boundary layer features on the S5010 airfoil at α = 5◦ and Re = 2 · 105 .
Looking back at figure 5.26 some interesting features and difference between the models
can be noted.
Starting at the root of the wing a clear difference can be seen between the laminar
5.2 Wind Tunnel Results 117
separation bubbles of models A and B. In model A (the upper picture) the laminar
separation bubble appears to have a local decrease in its chord wise length and even
disappears close to the fuselage. The decrease in size is probably due to a local change in
the pressure distribution due to the presence of the fuselage. The disappearance of the
bubble has probably its origin in the fact that the boundary layer over the fuselage is
turbulent and interacts with the boundary layer of the wing.
Model B shows some different features close to the root. Here it appears as if the bubble
starts at the interface between the wing leading edge and the fuselage and reattaches at
about 50% of the wing chord. A strong backward movement of the laminar separation
is noticed up until 20% af the span after which it remains more or less parallel with the
leading edge.
Another noticeable difference can be seen at the tip of the wing. For model A it is not
very clear for the right wing tip, however the left wingtip shows sufficient detail. It can
be seen that the bubble only extents to about 95% of the wing span for model A and up
to about 98% for model B. This has probably to do with the shape of the tip and how the
tip vortex curls around it. The tip vortex introduces a lot of diturbances in the boundary
layer making it turbulent and thus removing the laminar boundary layer. A change to
the tip shape had been introduced after reading in reference [15] about the influence the
tip shape has on the effective aspect ratio.
Model A has a rounded tip while model B has a straight wing tip. An illustration of
these wing tips and their effects on the tip vortex can be seen in figure 5.28. The rounded
tip keeps the vortex closer to the wing while the straight tip sheds it further reducing its
effect on the wing.
Figure 5.28: Illustration of a rounded tip and a straight tip with there tip vortex.
A difference can also be noted between the left and the right wing of model A. This was
actually the first aircraft produced and due to the lack of experience in building, it suffered
from a slight indentation over quite a large portion of the left wing. This distorted the
airfoil shape and is believed to have caused the difference between both sides.
The laminar separation bubble on model B also seems to increase in size, on both sides,
at about 75% of the wing span. However no clear explanation can be given for this
phenomenon which is not present in model A. It might have something to do with the
increased sweep of this model.
118 Wind Tunnel Experiments
Adding a sufficiently thick turbulator, that does not protrude through the boundary layer,
at the correct location, will force the boundary layer to trip before laminar separation
can occur. In figure 5.29 a turbulator strip is added to the wing of model A at 5% of
the chord length. The wing is still at a 5 degree angle of attack, and it is clearly visible
that the laminar separation bubble has disappeared. This shows that the laminar flow
has transitioned before seperating from the surface. This however does not directly imply
that the drag is lower. The pressure drag originaly present over the separation bubble
has been removed, but in stead the friction drag due to the turbulent boundary layer has
increased.
Figure 5.29: Boundary layer visualization for prototype A at α = 5◦ and a turbulator strip
at 5% of the chord length.
In figure 5.30 a different type of turbulator was used at 30% of the chord length of
model A. The wing, still at an angle of 5 degrees, shows that the bubble has not been
eliminated but encapsulated the turbulator reducing its effectiveness. It can be seen that
in the inboard 50% of the span there appears to be some turbulent flow right after the
turbulator, but it is not enough to trip the boundary layer further downstream.
Figure 5.30: Boundary layer visualization for prototype A at α = 5◦ , with a turbulator strip
encapsulated by the bubble.
5.2 Wind Tunnel Results 119
Figure 5.31: Boundary layer visualization of upper and lower wing surface for prototype B
at different angles of attack.
5.2 Wind Tunnel Results 121
The effect of the propeller on the boundary layer is not exactly clear (see figure 5.32).
The large laminar separation bubble has disappeared in the span region covered by the
propeller. Focusing on this region it can be seen that close to the leading edge there is
still some laminar flow present, this is followed by a highly swept bright line, connecting
the interface between the wing leading edge and the fuselage to the front side of the
laminar separation bubble. This line might be a vortex emanating from the wing-fuselage
interface, or it is a small laminar separation bubble. This bubble is then reduced in
size due to the higher Reynolds number produced by the velocity increase induced by
the propeller. Both the vortex and the separation bubble would explain why the large
separation bubble disappears in this portion of the wing.
Figure 5.32: Boundary layer comparison between a propeller off and a propeller/motor on
case for prototype B (α = 5◦ ).
The elevons in figure 5.33 are deflected upward by 11◦ . This means that on the upper sur-
face there will be a strong concave region, introducing a high pressure in this region. This
in turn causes a stronger adverse pressure gradient which causes the laminar boundary
122 Wind Tunnel Experiments
layer to separate earlier pushing the separation bubble closer to the leading edge. This
is clearly visual on the left side of figure 5.33, especially the part close to the tip, which
had an increased bubble size, is pushed forward.
For the lower surface the opposite is true. A sharp convex corner is created by the upward
deflection of the elevon, producing a sudden decrease in the pressure which is then followed
by a strong adverse pressure gradient tripping the boundary layer. Looking at the right
side of figure 5.33 it can be seen that the laminar separation, that originally extended into
the wake, is now followed by transition creating a closed bubble over the elevon. Close to
the edge of the elevon, parallel with the flow, no bubble is visual which probably has to
do with vortices emanating from these edges.
Figure 5.33: Boundary layer comparison for the prototype B wing with the elevons neutral
and 11◦ (α = 5◦ ).
The intended validation of the quasi-3D method with wind tunnel force measurements
could not be performed. This was due to the high level of uncertainty present in the
data. The uncertainty was introduced by several factors like, the lack of accuracy in the
angle of attack, due to several problems with the yaw-pitch system of the balance, and
the undetermined effects the balance has on the flow field around the aircraft.
5.3 Conclusions Concerning the Wind Tunnel Experiments 123
The boundary layer visualizations on the other hand provided insight in the low Reynolds
number effects on the wing. This helps in getting the right settings for the optimizer such
that the best approximation of the lift over drag ratio can be established. It also provided
proof of the fact that laminar flow is present and how it is influenced by several factors
like the propeller flow and elevon deflections.
124 Wind Tunnel Experiments
Chapter 6
Results
In this chapter the results obtained with the wing-optimizer will be discussed and the final
design for the competition aircraft will be presented. Three different optimizations were
performed to establish the best possible design for the EMAV2009 long range competition
aircraft. Two of these optimizations were used to establish the best possible wing assuming
the same fuselage as used for the prototypes. For the third optimization this constraint
was removed to see how much could be gained with a different fuselage. Results of these
optimizations can be seen in section 6.1.
A detailed description of the final design for the competition aircraft can be found in
section 6.2. The geometry and subsystems will be discussed followed by a performance
analysis of its range capabilities.
Three different optimizations were performed of which the results will be presented here.
The first optimization used the prototype aircraft as an initial point for the variables.
The boundaries and constraints on the variables were set such that the final result would
provide sufficient space for the batteries in the wing and that the same fuselage as the
prototype could be used.
The second optimization was set up exactly the same way as the first optimization, with
the sole difference that it used a bad wing, in an aerodynamic sense, as a initial starting
point. This was to provide insight in the capability of the optimizer of finding the same
final result.
The third optimization was provided a lot more freedom, by increasing the range between
the upper and lower bounds on the variables. It started with a reasonably good initial
wing and would have to find its way to the best possible wing in the domain enclosed by
the constraints. This was to show the maximum attainable aerodynamic efficiency and
how it related to the result obtained by the previous two optimizations. It thus provided
125
126 Results
insight on what to do with the fuselage. Changing to a new fuselage would require new
molds and a lot of work, compared to having the prototype fuselage and accepting a small
loss in the aerodynamic efficiency.
CL (x, c)
min f (x, c) = −
x CD (x, c)
In which x represents the vector with geometric variables for the optimization, containing:
• x(1) = b/2
• x(2) = croot
• x(3) = Λ
• x(4) = λ
• x(5) = V∞
The vector c contains all constant values to define the fuselage, tail, aircraft weight and
certain flow aspects on the wing. The values of the constants for this , and the following
optimizations are:
• Forced transition on lower surface (for root and tip section) = [0.8 0.8]%c
• Forced transition on upper surface (for root and tip section) = [1 1]%c
• Number of tails = 1
6.1 Optimization Results 127
Linear constraints:
Non-linear constraints:
Constraint on the Reynolds number:
In this equation W represents the aircraft weight and the maximum lift coefficient is set
equal to 0.95.
Constraint on the twist angle:
|ǫ(x, c)| ≤ 10
The twist is a intricate function dependent on variables and constants as explained in
section 4.1.2.
Constraint on the design lift coefficient:
2W
≤ 1.4
(x(2) + x(2) · x(4)) · x(1) · 1.225 · (x(5))2
Initial point:
x0 (1) = 0.235
x0 (2) = 0.17
x0 (3) = 23.274
x0 (4) = 0.8
x0 (5) = 20
128 Results
Optimized Geometry
After 29 hours, 13 iteration and 139 objective function evaluations, the optimization was
terminated. The reason for the termination was the magnitude of the step size in the
search direction, which had decreased below 1 · 10−10 .
The optimized result, x∗ , is:
The twist corresponding to this x∗ , airfoil and stability margin, is −6.56◦ . The S5010
airfoil was selected, before the optimization, for both the root and tip airfoil. In figure
6.1 the difference can be seen between the initial wing and the resulting new geometry.
Figure 6.1: Comparison between the initial wing geometry and the resulting geometry after
the optimization.
From figure 6.1 and x∗ it is clear that the optimizer tried to get the highest possible
aspect ratio. This was accomplished by having the span at the upper bound, while the
root chord and the taper ratio were forced to the lower bound. This increased the aspect
ratio from 3.07 to 3.71, with the most likely result of reducing the induced drag. When
evaluating the non-linear constraints it is noted that non of them are actually active. The
stall speed limit is getting close with 11.89m/s.
6.1 Optimization Results 129
Range Performance
The lift and drag coefficients for the optimized wing were calculated after the optimization,
and can be seen in figure 6.3. The maximum lift over drag is 12.76 at a lift coefficient
of 0.457. The optimizer however established a value of 12.65 to be the optimum at a lift
coefficient of 0.37(V∞ = 19.1m/s). This however does not mean that the optimizer has
not found the optimum. To go up to the CL = 0.457 it requires less speed to remain
in a straight level flight. Changing this requires the twist to decrease further and the
Reynolds number lowers, resulting in more profile drag. Calculating the lift over drag
ratio by changing the velocity, but remaining in straight level and trimmed flight, gives
the plot as can be seen in figure 6.2. From this figure it is clear that the maximum lift
over drag ratio is obtained at 19m/s as was obtained with the optimizer.
Figure 6.2: Lift over drag ratio for different velocities, maintaining straight, level and
trimmed flight.
The total distance that can be flown, using the propulsive system as described in section
6.2.2 and a battery capacity of 4.27Ah, can be calculated using equation 3.60. The
subsystems are assumed to consume about 5W which corresponds to a subsystem current
of 0.71A. This results in a total range of:
R = 103.8 km
For the starting point of the optimization this was 100.4km, which was obtained with
an lift over drag ratio of 11.91 and a velocity of 20m/s. This means an improvement of
about 3.4%
130 Results
The mathematical description of this problem only differs, compared to the one in section
6.1.1, in the initial point, which is changed into:
x0 (1) = 0.15
x0 (2) = 0.17
x0 (3) = 10
x0 (4) = 1
x0 (5) = 35
Optimized Geometry
First of all it is interesting to see how the result of this optimization compares to the one
previously obtained from a better starting point. The first difference that could be noted
was the number of iterations and function evaluations needed to obtain the optimum.
This optimization required 24 iterations, 11 more than the first optimization, and 213
function evaluations. This resulted in a total running time of about 47 hours, which is
a 60% increase in the calculation time. From this it can be concluded that starting at a
good initial point might safe quite some time.
The result of this optimization matches closely with the results obtained from the first
optimization. The optimized x is the following:
The difference between this result and the previous optimization is in the sweep angle
and the velocity. The sweep angle only differs by about 1.5◦ and is most probably caused
by the very small gradient in the lift over drag with respect to the sweep angle, as was
observed in section 4.3. The very small velocity difference probably is related to the
discrepancy in the sweep. It can be concluded that the optimization works well even
when the initial estimate is far from the optimal point. The twist for trimmed flight was
determined to be −6.87◦ , and the airfoil, for the root and tip, was the S5010.
The geometry comparison between the result obtained in the previous optimization and
in this optimization can be seen in figure 6.4, which clearly illustrates the strong similarity
between them.
132 Results
Figure 6.4: Comaprison between optimized wing from section 6.1.1(red) and the optimiza-
tion of this section (green).
Range Performance
The optimizer established that the maximum lift over drag ratio is 12.651, which is only
6· 10−3 higher than the first optimization, showing the extremely small gradient involving
1
the sweep (0.004 deg ). The calculated aerodynamic coefficients can be seen in figure 6.5,
CL
and once again the ( C D
)max is higher than found by the optimizer. However the same
explanation as in section 6.1.1 applies to this wing.
The change in the maximum range between the initial wing and the optimized result is
rather impressive for this optimization. The starting wing had a range of 76km while the
optimized wing attained a flying distance of:
R = 104.2 km
The gain in range over the optimization is 28.2km, which is an increase of 37%.
6.1 Optimization Results 133
For this problem the mathematical description remains the same as in section 6.1.1. The
only differences are the linear constraints and the initial point. These are transformed
into:
Linear constraints:
0.05 ≤ x(1) , x(1) ≤ 0.235
0.03 ≤ x(2) , x(2) ≤ 0.5
0 ≤ x(3) , x(3) ≤ 40
0.2 ≤ x(4) , x(4) ≤ 1
10 ≤ x(5) , x(5) ≤ 35
Initial point:
x0 (1) = 0.235
x0 (2) = 0.2
x0 (3) = 10
x0 (4) = 0.5
x0 (5) = 25
Optimized Geometry
After 50 hours, 27 iterations and 228 function evaluations the optimization terminated
with the following result:
x∗ (1) = b/2 = 0.235[m]
x∗ (2) = croot = 0.1946[m]
x∗ (3) = Λ = 31.683[◦ ]
x∗ (4) = λ = 0.28[−]
x∗ (5) = V∞ = 19.236[m/s]
The visual presentation of this result can be seen in figure 6.6. In this case it is clear
that the optimizer tried to establish the highest possible aspect ratio within the non-
linear constraints. The linear constraint on the span and the non-linear constraint on the
stall speed, together establish the maximum aspect ratio that can be achieved. For this
optimization this is:
b2max b2max ρVstall
2 C
max Lmax
Amax = = = 3.77
Smin 2·W
The aspect ratio for the optimized wing is exactly the same. This implies that the stall
speed constraint is active. The other nonlinear constraints are not active however the
twist is −9.83◦ which is very close to the maximum 10◦ . This value is rather high and
such a result must thus be treated with care. The airfoil used at the root and tip is once
again the S5010.
6.1 Optimization Results 135
Figure 6.6: Comaprison between optimized wing from section 6.1.1 (upper) and the opti-
mization of this section (lower).
Range Performance
The lift over drag ratio for this wing improved form 10.9 for the initial wing to 13.0 for
the optimized design. This lift over drag ratio is established at a lift coefficient of 0.369.
However the maximum lift over drag ratio in figure 6.7 is 13.13 at a lift coefficient of
0.416. Like before, reaching this lift coefficient requires more twist and thus more drag
reducing the lift over drag ratio again.
The total distance this airplane can fly with a battery capacity of 4.27Ah and subsystems
consuming the same energy as before, is:
R = 106.2 km
With respect to the first optimization result this is only 2.4km difference and with respect
to the prototype, used as initial point for the first optimization, this is a difference of
5.8km.
Based on this it can be concluded that creating a new design and new molds is to expensive
for an increase in range of only 6%. This range might even be compromised as it is not
certain that the same amount of battery capacity still fits in the wing. A better option
is to look at the batteries and see if something can be gained in increasing their capacity
for the prototype aircraft. This topic will be discussed in section 6.2.2.
136 Results
The aircraft consists out of a single wing, a fuselage and a tail. A three-view of the
TRIP50 can be seen in appendix D.
Wing
The wing has a single tapered shape with a S5010 airfoil at the root and the tip of the
wing. The S5010 was selected based on the fact that it is designed for low Reynolds
numbers, it has a low moment coefficient, and it has a thickness to chord ratio of 10%
providing sufficient space in the wing. The Xfoil results for this airfoil at a Reynolds
number of 2 · 105 can be seen in figure 6.8.
Figure 6.8: polars for the S5010 airfoil at a Reynolds number of 2 · 105 .
The TRIP50 wing planform is defined as described in table 6.1, and can be seen in the
upper drawing of figure 6.1.
138 Results
The wing is made out of an upper and a lower shell. These shells are made from an outer
layer of 50gr/m2 glass-fibers and an inner layer of 36gr/m2 Kevlar-fibers. The fibers have
a 0-90-orientation but are put under 45◦ with respect to the trailing edge of the wing to
provide torsional stiffness. A low-density polystyrene foam core is located in between
those shells, to provide form-rigidity to the shells and to facilitate locating components in
the wing. Parallel to the leading edge, at 75mm from it, a spar of carbon fiber is located
to take the bending loads and to act as part of the torsion box.
The fuselage and the tail form a single construction element. The fuselage is 250mm long
and has a circular cross-section with a diameter of 43mm at the thickest point. A detailed
description of the fuselage geometry can be found in reference [9]. The tail is a single
tapered surface with a naca0012 airfoil. The planform details of the tail are presented in
table 6.2.
Table 6.2: Tail planform of the TRIP50.
The fuselage-tail structure is build up out of 50gr/m2 glass fibers for the tail and 110gr/m2
Kevlar for the fuselage itself. The tail is constructed out of two layers of glass fiber with
a Rohacell core in between to provide form-rigidity. The fuselage is mainly constructed
out of a single layer of Kevlar however at certain points, like the nose and the edges just
above the wing, an extra layer is added to withstand hard landings.
In this section a general outline of the electronics will be presented and a visual overview
can be seen in figure 6.9. However not much attention will be given to the autopilot nor
on the other electrical components, except for the propulsion system as this is important
for the range performance.
The autopilot was especially designed for this competition by the TU Delft MAVLab. The
autopilot uses infra-red sensors to establish its attitude and GPS for its position. A picture
of the autopilot inside the aircraft can be seen in figure 6.10. The communication with
the ground station is established using a 2.4GHz Xbee transceiver. For safety purposes
6.2 The Competition Aircraft 139
Figure 6.9: General overview of the electric layout for the EMAV2009 competition aircraft.
it was required to be able to manually override the autopilot which required a 35MHz
receiver. The “Penta Profi” was selected as the receiver based on its weight and size.
Three servos were required, two of which controlled the elevons and one for the paint ball
drop mechanism. The original idea was to use “RC MART Lexi” servos for the elevons
and drop mechanism as they only weigh 2.5gr. However they proved not strong enough
140 Results
for the elevons and were thus changed for 4.7gr “Diamond D47” servos.
See reference [14] for more details on the design, selection and technical specifications.
This combination for the propulsive system was applied in the prototype. However during
the first test flights the pilot was forced to land as the motor controller stopped working
when applying full throttle. This was the reason to change the “TMM1210-3 PL auto”
with a “YGE-12” speed controller. Another problem was the propeller breaking at every
landing. The Graupner type was then switched for a “4.75 × 4.75 ACP”, which proved
to survive 2 to 3 landings.
6.2 The Competition Aircraft 141
Figure 6.11: Range plotted against the battery capacity(ranging from 0-200 Ah), for a fixed
aircraft shape with a varying weight with respect to the battery capacity, and
a 5W subsystem power consumption.
This figure shows that the range keeps increasing towards an asymptote. Thus it could
be concluded that the largest battery that still fits in the aircraft, and who’s weight does
not compromise the stall speed, must be used. That was the reason to start looking
for a battery that meets those requirements, which resulted in a 8Ah Hellpower Lithium
polymer battery. This change in battery increased the range from 113.5 km to 170.8 km
(see figure 6.12), taking into account the above mentioned assumptions. The flight time
increased to 2h9min at a velocity of 22m/s.
Weight
Details on the weight for each component are presented in table 6.3, from which it can
also be seen that the aircraft weight is 575gr.
142 Results
Figure 6.12: Range plotted against the battery capacity(ranging from 0-15 Ah), for a fixed
aircraft shape with a varying weight with respect to the battery capacity, and
a 5W subsystem power consumption.
The original design weight of the aircraft was 500gr but due to the increased battery size
the weight increased with 75gr. This required a small increase of the velocity to maintain
the lift coefficient that corresponded to the maximum lift over drag ratio, which meant
that the elevons had to be deflected down a small amount to maintain straight, level and
trimmed flight.
Layout
The location of all components in the aircraft can be seen in figure 6.13. They were
positioned such that they located the center of gravity exactly as required by the stability
margin and the location of the neutral point. The location of the nuetral point and
the center of gravity are illustrated in the 3-view in appendix D, from which it can be
deduced that the stability margin is 2.7%. This seems low but a small margin provides
the autopilot with a lot of pitch control, meaning that a small deflection of the elevons
has a large effect on the pitch angle. A small stability margin makes the airplane more
sensitive to gusts however the wing loading of the TRIP50 is rather high for a mini-UAV,
78.4N/m2 , making it less sensitive to gusts.
The aerodynamic coefficients for the TRIP50 can be seen in figure 6.14 for a velocity of
20m/s. The drag for the fuselage and the tail are taken into account by adding a constant
drag coefficient to the polar. These constants are:
CDf us = 0.002
CDtail = 0.0016
6.2 The Competition Aircraft 143
wing structure 85 gr
fuselage structure 13 gr
batteries 354 gr
motor 35 gr
speed controler 6.8 gr
propeller 5 gr
reciever 4.5 gr
auto-pilot 36 gr
xbee-transciever 3.5 gr
elevon-servos 9.6 gr
servo torsion rods 8 gr
paintball servo 2.95 gr
paintball tube 1.5 gr
cables 10 gr
TOTAL 574.85 gr
Both these drag coefficients are with respect to the wing surface area.
The maximum lift over drag coefficient is equal to 12.31 and is obtained at a flight speed
of 19.88m/s (CL = 0.324). The moment coefficient at the maximum lift over drag ratio
is equal to 4.4 · 10−3 in stead of the required 8.75 · 10−3 . This is not sufficient to trim the
aircraft which has a twist of −3◦ instead of the required −7.6◦ . This can be compensated
for by deflection of the elevons up generating a higher positive moment. This however
comes at the cost of increased drag and thus lowering the lift over drag ratio, which is
also the case for increasing the twist. Twisting the wing to −7.6◦ lowers the lift over drag
ratio to 11.76. Taking this value for the lift over drag ratio, assuming the subsystems use
5W (which translates to a current of 0.71A at a tension of 7V ) and a battery capacity of
8Ah gives a range of:
R = 166.7 km
7.1 Conclusions
At the start of this thesis the following project goal was stated:
Design of a mini-UAV for the EMAV 2009 outdoor competition using a novel
aerodynamic optimization tool, with the objective to win the competition.
The EMAV 2009 outdoor competition has 3 phases, consisting of flying as far as possible
towards a goal, dropping a paint ball, and flying the same distance back. The focus of
the competition is mainly on the range performance, that is why this thesis expands on
this topic.
To reach the above stated goal an in dept theoretical analysis of the aerodynamics in-
volved was performed. The conclusion of this was that a quasi-3D method would yield
the best results considering requirements on computational effort, automatic creation of
the computational domain and capability to handle low Reynolds numbers(< 2.5 · 105 ).
These requirements were set forward by the fact that this analysis tool was goning to
be implemented in an optimization environment. The quasi-3D method uses a vortex
lattice method to simulate the three dimensional aerodynamics and Xfoil results are put
on top of the vortex lattice results to obtain information on the profile drag. The exact
implementation of this method is presented and a detailed validation is performed against
a RANS-CFD method and wind tunnel results obtained from literature. The validations
proved that the accuracy of the maximum lift over drag ratio remained within 4%. The
wings used for validation varied in sweep, taper ratio, aspect ratio and twist.
The quasi-3D method was then submitted to an optimization algorithm that would be
used to established the best wing shape for long range flight. The variables for the opti-
mization would consist out of 4 geometric variables and 1 flow variable. The 4 geometric
variables are the wing span, root chord, quarter chord sweep angle and the taper ratio.
The velocity is the only flow variable. Apart from these variables several constants are
147
148 Conclusions and Recommendations
required. The most important ones are the weight of the aircraft and the stability margin.
The other constants are used to establish the fuselage and tail drag, and the values for the
constraints. Based on these inputs two values are determined in the objective function,
one is the twist angle required to maintain trimmed straight horizontal flight, and the
second one is the objective value. The objective functions for range performance of elec-
trically powered propeller aircraft are presented. The lift over drag ratio is taken as the
objective for the optimizations presented in this thesis however a better approximation
can be attained when the power consumed by the non-propulsive subsystems is taken into
account. The domain of the optimization variables was limited by linear and non-linear
constraints. These constraints served the purpose to limit the search domain such that a
valid design could be obtained.
The optimization algorithm was selected based on its efficiency, ability to find a global
minimum in the objective domain and ability to handle linear and non-linear constraints.
This resulted in selecting the sequential quadratic programming algorithm as the best
option. The requirement for this algorithm to find a global minimum could not be con-
firmed, due to the intricate way in which the objective value is calculated. It was however
verified that the same final result was obtained for different initial points.
Two wind tunnel campaigns were performed to establish the performance of the opti-
mization tool and the quasi-3D method. The required data from the wind tunnel tests,
to obtain this goal, were force measurements and boundary layer visualizations. The low
speed low turbulence wind tunnel of the TU Delft was selected for the measurements.
The required tools for the boundary layer visualizations were available at this facility.
The resolution for the force balance, available in the wind tunnel, was too low for the
small drag forces on mini-UAV’s. This thus required the construction of a new balance
with a higher resolution. This resulted in selecting a very accurate force transducer (ATI
nano17) for which a yaw-pitch system was build to be able to set the required pitch angles
and side slip angles during the measurements.
The force measurements appeared to have major errors in them due to problems with
the yaw-pitch system of the balance. A detailed description of the problems with the
yaw-pitch system are presented together with an overview of the effects they have on
determining the aerodynamic coefficients. The boundary layer visualizations on the other
hand proved to be vary useful in understanding the phenomena in the boundary layer
at low Reynolds numbers and different angles of attack. They also provided information
on what happens on the wing when the propeller is blowing air over it, or what happens
when the elevons are deflected.
This thesis work created several spin-offs in the form of 4 DAR-exercises. These exercises
focused on producing wind tunnel models and their molds, designing and building the
wind tunnel balance, and designing the propulsion system for the competition aircraft.
The help obtained from these exercises made the practical execution of this research
possible.
In the final stages of the thesis work three different optimizations were performed to estab-
lish which design would be selected to become the competition aircraft. The conclusion of
this was that the first prototype would be the best option. The only change to it was the
battery size which increased the range by 66%. The range this aircraft can fly is 166.7km.
7.2 Recommendations 149
It could however not be established if the above mentioned goal was obtained. The
day of the competition was very rainy, resulting in wet vegetation which had the same
temperature as the sky. The autopilot needs a temperature difference between the sky
and the ground to establish its attitude with the infra-red sensors. All competitors had
the same problem resulting in the competition being canceled.
7.2 Recommendations
A summary of the recommendations will be presented on which future work might expand.
Quasi-3D method:
• Improving the determination of the induced downwash angle such that the induced
drag can be determined using the quasi-3D technique.
• Improving the estimation of the vortex parameter Kv for swept and tapered wings.
• Improving the lift over drag estimation for swept wings such that its gradient is
better estimated.
• Implement the interpolation of forced transition points for the airfoil data.
• Improve and validate the determination of the maximum lift coefficient by the quasi-
3D method.
Optimization tool:
• The aerodynamic coefficients could be evaluated real time during the wind tunnel
measurements.
• The windtunnel settings and the atmospheric data are logged automatically.
150 Conclusions and Recommendations
Bibliography
[1] Matlab help files on: Constraint optimization, sequential quadratic programming,
August 2007. Matlab Version 7.5.0.342(R2007b).
[7] Ir. L.L.M. Boermans. Aerodynamic design of aircraft and advanced transportation
systems. In Part I. Delft University Copy Machine, 1998.
[8] Ir. L.L.M. Boermans. Aerodynamic design of aircraft and advanced transportation
systems. In Part II: The Real Case. Delft University Copy Machine, 1998.
[9] B.A.P Bruggeman. Designing and constructing the molds for emav2009 uav. DAR-
exercise Report.
[10] P.D. Chappell. Wing pitching moment at zero lift at subcritical mach numbers. Esdu
87001, ESDU international, 1991.
[11] T.M.T. Doornbos. Building the half meter uav for emav2009. DAR-exercise Report.
[12] Mark Drela. Low-reynolds-number airfoil design for the m.i.t. daedalus prototype:
A case study. Journal of Aircraft, 25(8):724–732, 1988.
[13] B.F.R. Ewald (Editor). Wind tunnel wall corrections. Agard-ag-336, AGARD, Oc-
tober 1998.
151
152 BIBLIOGRAPHY
[14] Vinodh Kumar Gunaseelan. Design of propulsive system for micro aerial vehicle and
wind tunnel measurement system. DAR-exercise Report.
[15] S.F. Hoerner. Fluid-Dynamic Drag. Hoerner Fluid Dynamics, june 1965.
[17] John E. Lamar. Prediction of vortex flow characteristics of wings at subsonic and
supersonic speeds. Journal of Aircraft, 13(7):490–494, 1976.
[19] Karl Nickel and Michael Wohlfahrt. Tailless Aircraft in Theory & Practice. American
Institute of Aeronautics and Astronautics, 1994. ISBN 0-340-61402-1.
[21] W.F. Phillips and D.F. Hunsaker. Estimating the subsonic aerodynamic center and
moment components for swept wings. Journal of Aircraft, 45(3):1033–1043, 2008.
[23] Alan Pope. Wind-Tunnel Testing. John Wiley & Sons, Inc., New York and Chapman
& Hall, Limited, London, second edition, 1954.
[24] Singiresu S. Rao. Engineering Optimization: Theory and Practice. John Wiley &
Sons, fourth edition, 2009. ISBN 978-0-470-18352-6.
[27] Michael Selig, Andy Broeren, and Philippe Giguere. Volume 3. In Summary of
Low-Speed Airfoil Data. SoarTech Publications, 1997. ISBN 0-9646747-3-4.
[28] Michael Selig, James Guglielmo, Andy Broeren, and Philippe Giguere. Volume 1. In
Summary of Low-Speed Airfoil Data. SoarTech Publications, 1995. ISBN 0-9646747-
1-8.
[29] E. Torenbeek. Synthesis of Subsonic Airplane Design. Delft University Press, 1982.
ISBN 90-247-2724-3.
[30] L.L.M. Veldhuis. Lecture notes for the low speed wind tunnel test 3. Course code:
“AE3-193P”.
[31] L.L.M. Veldhuis. Source code for the tu-delft ltt wind tunnel balance-measurment
program. “program w3d” and “subroutine correc”.
BIBLIOGRAPHY 153
[32] A.D. Young. The calculation of the total and skin friction drags of bodies of revo-
lution at zero incidence. Reports and memoranda no. 1874, Aeronautical Research
Committee, April 1939.
[33] Remco Zwinderman. Designing and production of a wind tunnel force measuring
system for lightweight uav’s. DAR-exercise Report.
154 BIBLIOGRAPHY
Appendix A
155
156 Original Assignment Formulation
Appendix B
157
Wing Geometry Details
Table C.1: Force range and resolution information for the Nano 17 SI-25-0.25.
Range
Fx , Fy Fz Tx , Ty Tz
25 N 35 N 250 N mm 250 N mm
Resolution
Fx , Fy Fz Tx , Ty Tz
1/160 N 1/160 N 1/32 N mm 1/32 N mm
159
ATI Nano 17 SI-25-0.25 Data Sheet
6.957±0.013
120° MOUNTING 14.50 TOOL ADAPTER
1.085±0.013 (2) 2.0 5.411±0.013 (2) 2.0
ISO H7 Slip Fit ADAPTER Sensing Reference
ISO H7 Slip Fit
10° 3.0 Deep Frame Origin
+X 3.0 Deep
(B.C. 12.5) +X
(B.C. 12.5)
6.248±0.013
Nano17-A Cable Exit (3) M2-.40 ISO TAP FOR
(Alternate) INTERFACE 3.5 DEEP
6.153±0.013 (SEE NOTE 6)
120° B.C. 12.5
8.291±0.013 9.373±0.013
+Y 17 EQUALLY SPACED
-Z +Z
NOTES:
1. TOOL AND MOUNTING ADAPTER MADE OF EITHER
ALUMINUM OR SST (CUSTOMER SPECIFIED).
2. TOOL AND MOUNTING ADAPTER HAVE M2 TAPS AND 2.0MM
DOWEL PIN HOLES FOR INTERFACING.
3. CABLE CAN BE FACTORY INSTALLED ON SIDE OR THE TOP. ALTERNATE CABLE EXIT
4. CONNECTOR (NOT SHOWN) HAS 17MM DIAMETER AND IS (Nano17-A Cable Exit) ISOMETRIC VIEW
67.5MM LONG.
NOTES: UNLESS OTHERWISE SPECIFIED
5. WARNING-- DO NOT LOOSEN OR REMOVE INTERFACE PLATES 1 0 3 1 G o o d w o r t h D r iv e , A p e x , N C 2 7 5 3 9 , U S A
DUE TO POTENTIAL DAMAGE DO NOT SCALE DRAWING. DRAWN IN SOLIDWORKS.
Tel: + 1.919.772.0115 Email: [email protected]
ALL DIMENSIONS ARE IN MILLIMETERS.
6. DO NOT EXCEED INTERFACE DEPTH MAY CAUSE DAMAGE Fax: + 1.919.772.8259 www.ati-ia.com
ISO 9001 Registered Company
3-Views
161
162 3-Views