DISS. ETH NO.

18901

Aerodynamics, Modeling and Control

of an Autonomous Micro Helicopter

A dissertation submitted to
ETH ZÜRICH
for the degree of
Doctor of Sciences

presented by

Dario Martin SCHAFROTH

Maschinenbau Ingenieur
Eidgenössische Technische Hochschule Zürich, Schweiz

Born August 28, 1980

Citizen of Lützelflüh, Switzerland

accepted on the recommendation of

Prof. Roland Siegwart, Principal Advisor

Prof. Heinrich Bülthoff, Member of the Jury

Zürich, ETHZ
2010
Acknowledgments

This thesis is the result of the three and a half years I spent as a PhD stu-
dent at the Autonomous Systems Laboratory of Prof. Roland Siegwart at
the Swiss Federal Institute of Technology (ETHZ) in Zurich. This period was
a very interesting and rewarding time, where I could deepen my knowledge
in various multi-disciplinary areas, where I had different good collaborations
and met a lot of new friends.
First, I would like to thank my supervisor, Prof. Roland Siegwart for the op-
portunity to write this thesis and work in his lab and Prof. Heinrich Bülthoff
for the review of this thesis.
I also gratefully acknowledge all the partners of the muFly project, espe-
cially Slawomir Grzonka, Alexandre Pages and Kiman Velt, for the good
collaboration. My gratitude also goes to all students I supervised, especially
mentioned Florian Hänni, Michael Gloor and Peter Sax for their contribu-
tion to this work. Thanks to all my colleagues at the Autonomous Systems
Laboratory (ASL), especially Samir Bouabdallah and my former officemate
Andre Noth for the good time in our office. Special thanks to the whole
technical support team, Thomas Baumgartner, Markus Bühler, Dario Fen-
ner, Janosch Nikolic and Stefan Bertschi, who were always there to help
and solve problems. Many thanks also to my officemate and friend Stefan
Leutenegger, who offered to correct this thesis and gave many important
inputs. One of the last but most important acknowledgments goes to my
colleague and friend Christian Bermes for all his effort, support and friend-
ship during those three and a half years. Finally, I would like to thank my
family, especially my parents for having supported me during all these years,
my sister Petra, Thomas, Silas and godchild Leonie. At last I want to thank
my girlfriend Andrea for her support, her love and the wonderful time.

i
ii
Abstract

This thesis deals with two of the most important tasks during the design of a
fully autonomous micro helicopter: the rotor and control design. Both areas
are strongly affected by the extremely small size of the vehicle.
The rotors on the micro helicopter are working in the low Reynolds number
regime reducing strongly the efficiency and limiting the autonomy time of the
helicopter. In order to increase the efficiency, the influence of various blade
parameters and lift enhancing devices is investigated. This is approached
by using three different simulation methods designated for different applica-
tions: a Blade Element Momentum Theory code, a Vortex Method and CFD
simulations. Those simulation methods are validated on a coaxial test bench
capable of measuring rotors of complex shapes manufactured on a rapid pro-
totyping machine. Additionally the drive train of the helicopter is integrated
in the investigations allowing a realistic scenario as on the helicopter. The
developed methods and obtained investigation results assist researchers and
engineers in developing an efficient propulsion system for micro helicopters.
On the control side the small size leads to faster system dynamics and limits
the payload for accurate sensors and actuators. The control problem is first
addressed by the development of a nonlinear model for the coaxial helicopter
muFly. Here the goal is to derive a dynamic model which accurately reflects
the physical characteristics allowing accurate simulations and which is also
simple enough to be used for the controller design later on. In order to val-
idate the model and to find the platform depending system parameters an
identification process is carried out. The identification on the real flight data
is done using a nonlinear technique called the Covariance Matrix Adaptation
- Evolution Strategy (CMA-ES). The results obtained show the accuracy of
the nonlinear model.
Finally, the control strategy is presented including the low-level and high-
level control. The low-level controllers for attitude and heave are designed
by the H∞ method. The controllers are implemented on the helicopter and
successively tested during flight. Additionally, a high-level control approach

iii
is presented, including mission control, trajectory generation and obstacle
avoidance.
Altogether, this thesis serves as a guideline for other researchers and engi-
neers which aim to develop an autonomous micro helicopter in the size of a
small bird.

Key words: Micro helicopter design, rotor aerodynamics, low Reynolds

number, system modeling, nonlinear parameter identification, H∞ -control

iv
Kurzfassung

Die vorliegende Doktorarbeit steht im Zusammenhang mit dem europäischen

Forschungsprojekt muFly, welches das Ziel verfolgte, einen völlig autonomen
Mikrohelikopter in der Grösse und Gewicht eines kleinen Vogels zu entwick-
eln.
Der Entwicklungsprozess eines Mikrohelikopters beinhaltet viele verschiedene
Fachgebiete, wobei sich diese Arbeit hauptsächlich auf zwei der wichtig-
sten Thematiken konzentriert: Die Rotoraerodynamik und die Flugregelung.
Beide Gebiete werden stark von der geringen Grösse des Helikopters beeinflusst.
Die Rotoren eines Helikopters dieser Grösse arbeiten in den tiefen Reynold-
szahlen, wo viskose Effekte die Effizienz reduzieren und die Autonomiezeit des
Helikopters stark beschränken. Mit dem Ziel, diese Effizienz zu steigern, wird
der Einfluss von verschiedenen Rotorblattgeometrien untersucht. Die Grund-
lagen der Untersuchung bilden drei Simulationsmethoden: ein Blatt-Element-
Impuls Theorie Code (BEMT), eine Wirbelmethode und CFD Simulatio-
nen. Diese Simulationsmethoden werden mit den Messungen eines koaxialen
Rotor-Prüfstandes verifiziert. Der Einsatz einer Rapid-Prototyping Machine
zur Herstellung der Rotorblätter erlaubt eine schnelle und exakte Durch-
führung der Messungen. Um realistische Bedingungen wie auf dem Helikopter
zu erschaffen, wird zusätzlich zu den rein aerodynamischen Untersuchungen
der Einfluss des Antriebsstranges in die Simulation integriert. Die entwick-
elten Simulationsmethoden und Prüfstand, sowie die Resultate dienen als
Richtlinien für die Entwicklung eines vergleichbaren Mikrohelikopters.
Im Falle der Regelung führt die geringe Grösse des Helikopters zu einer
schnelleren Dynamik des Systems und limitiert die Nutzlast für präzise Sen-
soren und Aktuatoren. Der erste Schritt zur Flugregelung ist eine nicht-
lineare Systemmodellierung des entwickelten koaxialen Mikrohelikopters mu-
Fly. Das angestrebte Ziel der Modellierung ist ein Kompromiss zwischen
einem komplexen wahrheitsgetreuen Modell für Simulationen und einem ein-
fachen Modell, welches als Grundlage für die Regelung dient. Zur Vali-
dierung des Models und zur Identifizierung der Systemparameter wird eine

v
nichtlineare Identifizierungstechnik, die Covariance Matrix Adaptation- Evo-
lution Strategy (CMA-ES), verwendet. Die erhaltenen Resultate des Identi-
fizierungsprozesses zeigen die Richtigkeit des nichtlinearen Modells.
Als Abschluss wird die Flugregelung präsentiert, welche aus einer System-
regelung und einer Missionsregelung besteht. Zur Auslegung der Regler für
die Höhe und Lage wird die H∞ -Methode verwendet und auf dem Mikro-
prozessors des Helikopters implementiert. Die ausgelegten Regler sind erfol-
greich in Simulation und Flug getestet.
Zusätzlich zur Systemregelung wird ein Ansatz für eine Missionsregelung
präsentiert. Die Missionsregelung umfasst die Missionszustandsmaschine,
Trajektorienplanung und Obstacle Avoidance. Im Grossen und Ganzen di-
ent diese Arbeit als eine Hilfe für Forscher und Ingenieuren, welche das Ziel
verfolgen, einen autonomen Helikopter in der Grösse eines kleinen Vogels zu
entwickeln.
Schlüsselwörter: Mikro-Helikopter Desgin, Rotor Aerodynamik, Low
Reynolds Regime, Systemmodellierung, nichtlineare Parameteridentifizierung,
H∞ Regelung

vi
List of symbols

General
A, a Scalars
A, a Vectors
A, a Matrices, Tensors

Subscripts
B body-fixed frame
J inertial frame
BJ transformation from inertial frame to body-
fixed frame
bat battery
dw down
drive drivetrain
fus fuselage
gear gear
hov hover
hy heave-yaw
mot motor
react reaction
rp roll-pitch
serv servo
up up

Aerodynamics
A rotor disc area [m2 ]
c rotor chord length [m]
cD drag coefficient [-]
cL lift coefficient [-]
cQ torque coefficient [-]
cT thrust coefficient [-]

vii
D drag force [N]
FM figure of merit [-]
i gearbox reduction ratio [-]
L lift force [N]
Nb number of blades [-]
P power [W]
Q drag torque [Nm]
R rotor radius
Re Reynolds number [-]
T thrust [N]
v0 inflow velocity [m/s]
vI induced velocity [m/s]
vm mean flow velocity [m/s]
α angle of attack [-]
Γ circulation [m2 /s]
θ pitch angle [rad]
µ fluid dynamic viscosity [Ns/m2 ]
ν fluid kinematic viscosity [m2 /s]
φ inflow angle [rad]
ρ air density [kg/m3 ]

Control
A linear transformation matrix
dR motor friction [Nm s]
F external force [N]
g gravitational acceleration [m/s2 ]
i gearbox reduction ratio [-]
I body inertia [kg m2 ]
IA current [kg m2 ]
J rotor/drive train inertia [kg m2 ]
l linkage factor [-]
L loop gain [-]
M external moment [Nm]
p roll rate [rad/s]
q pitch rate [rad/s]
Q drag torque [Nm]
r yaw rate [rad/s]
R rotor radius
R rotational transformation matrix
RΩ motor terminal resistance [Ohm]

viii
T sensitivity [-]
T thrust [N]
T complementary sensitivity [-]
Tf following time [s]
u flight speed x-direction [m/s]
U voltage [V]
v flight speed y-direction [m/s]
w flight speed z-direction [m/s]
W aerodynamic body drag force [N]
x spatial direction [m]
y spatial direction [m]
z spatial direction [m]
η efficiency [-]
θ pitch angle [rad]
ΘSP swash plate angle [rad]
κM motor torque constant [Nm/A]
κE motor voltage constant [Vs/rad]
λI inflow ratio [-]
ρ air density [kg/m3 ]
φ roll angle [rad]
ψ yaw angle [rad]
ω motor rotational speed [rad/s]
Ω rotor rotational speed [rad/s]

ix
x
Acronyms

ASL Autonomous Systems Laboratory

AoA angle of attack

BC boundary condition

BLDC Brushless Direct Current

BET Blade Element Theory

BEMT Blade Element Momentum Theory

CAD Computer Aided Design

CFD Computational Fluid Dynamics

CMA-ES Covariance Matrix Adaptation - Evolution Strategy

CoG Center of Gravity

DC Direct Current

dof degree of freedom

ETHZ Eidgenössische Technische Hochschule Zürich

FM Figure of Merit

FVM Finite Volume Method

FEM Finite Element Method

GPS Global Positioning System

GUI Graphical User Interface

xi
IMU Inertial Measurement Unit
LiPo Lithium Polymer
MAV Micro Aerial Vehicle

MEMS micro-electro-mechanical systems

MIMO multiple-input multiple-output
PID Proportional Integral Derivative

PCB Printed Circuit Board

PM Panel Method
PWM Pulse Width Modulation
RC Radio Control

SLAM simultaneous localization and mapping

TPP tip path plane
UAV Unmanned Aerial Vehicle

xii
The beginning of knowledge is the dis-
covery of something we do not under-
stand.
Frank Herbert (1920 - 1986)
xiv
Contents

Acknowledgments i

Abstract iii

Kurzfassung v

List of symbols vii

Acronyms xi

1 Introduction 1
1.1 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Coaxial Helicopter . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Structure of this Work . . . . . . . . . . . . . . . . . . . . . . 8

2 The muFly Helicopter 9

2.1 Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Propulsion System . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Sensors and Actuators . . . . . . . . . . . . . . . . . . 13
2.1.3 Logic and Communication . . . . . . . . . . . . . . . . 17
2.2 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Aerodynamics 21
3.1 Rotor Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 Airfoil Aerodynamics . . . . . . . . . . . . . . . . . . . 22
3.1.2 Aerodynamics of a Rotor Blade . . . . . . . . . . . . . 25
3.1.3 Coaxial Rotor Configuration . . . . . . . . . . . . . . 28
3.1.4 Drive Train . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

xv
 3.2.1 Test Bench Setup . . . . . . . . . . . . . . . . . . . . . 31
3.2.2 Calibration and Measurement . . . . . . . . . . . . . . 34
3.3 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.1 Blade Element Momentum Theory . . . . . . . . . . . 35
3.3.2 Vortex Method . . . . . . . . . . . . . . . . . . . . . . 39
3.3.3 Computational Fluid Dynamics . . . . . . . . . . . . . 43
3.3.4 Drive Train . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4.1 Number of Blades . . . . . . . . . . . . . . . . . . . . 53
3.4.2 Blade Radius and Chord Length . . . . . . . . . . . . 53
3.4.3 Blade Profile . . . . . . . . . . . . . . . . . . . . . . . 55
3.4.4 Taper and Twist . . . . . . . . . . . . . . . . . . . . . 59
3.4.5 Tip shape . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.4.6 Lift-enhancing Devices . . . . . . . . . . . . . . . . . . 62
3.4.7 Coaxial Rotors . . . . . . . . . . . . . . . . . . . . . . 64
3.4.8 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 68

4 Dynamic Model 69
4.1 Coordinate Frames . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Rigid-Body Dynamics . . . . . . . . . . . . . . . . . . . . . . 71
4.3 External Forces and Moments . . . . . . . . . . . . . . . . . . 72
4.3.1 Rotor Forces and Moments . . . . . . . . . . . . . . . 73
4.3.2 Aerodynamic Drag . . . . . . . . . . . . . . . . . . . . 76
4.3.3 Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4 Dynamic Elements . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4.1 Swash Plate and Stabilizer Bar . . . . . . . . . . . . . 77
4.4.2 Electric Motors . . . . . . . . . . . . . . . . . . . . . . 79
4.5 Model Summary . . . . . . . . . . . . . . . . . . . . . . . . . 80

5 Parameter Identification 83
5.1 Identification on the Ground . . . . . . . . . . . . . . . . . . 83
5.2 Identification on Test Benches . . . . . . . . . . . . . . . . . . 84
5.3 Identification on Flight Data . . . . . . . . . . . . . . . . . . 86
5.3.1 Covariance Matrix Adaptation- Evolution Strategy . . 86
5.3.2 Data Generation and Application . . . . . . . . . . . . 87
5.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6 Control 95
6.1 Low-Level Control . . . . . . . . . . . . . . . . . . . . . . . . 95
6.1.1 H∞ Control . . . . . . . . . . . . . . . . . . . . . . . . 97
6.1.2 Control Structure . . . . . . . . . . . . . . . . . . . . . 98

xvi
 6.1.3 Linearization . . . . . . . . . . . . . . . . . . . . . . . 99
6.1.4 Heave-Yaw Control . . . . . . . . . . . . . . . . . . . . 99
6.1.5 Roll-Pitch Control . . . . . . . . . . . . . . . . . . . . 103
6.1.6 Position Control . . . . . . . . . . . . . . . . . . . . . 103
6.1.7 Implementation . . . . . . . . . . . . . . . . . . . . . . 106
6.1.8 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.1.9 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2 High-Level Control . . . . . . . . . . . . . . . . . . . . . . . . 112
6.2.1 Mission State Machine . . . . . . . . . . . . . . . . . . 112
6.2.2 Trajectory Generation . . . . . . . . . . . . . . . . . . 113
6.2.3 Obstacle Avoidance . . . . . . . . . . . . . . . . . . . . 114
6.2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 117

7 Conclusion 119
7.1 Review and Achievements . . . . . . . . . . . . . . . . . . . . 119
7.1.1 Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . 119
7.1.2 Modeling and Control . . . . . . . . . . . . . . . . . . 120
7.1.3 Main Achievements . . . . . . . . . . . . . . . . . . . . 121
7.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.2.1 Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . 121
7.2.2 Modeling and Control . . . . . . . . . . . . . . . . . . 122

A Nonlinear Model 123

B State-Space Models 127

Bibliography 131

Curriculum Vitae 139

xvii
xviii
List of Tables

3.1 Recommendations for the parameter ncrit . . . . . . . . . . . . 37

3.2 Comparison of the turbulence models. . . . . . . . . . . . . . 49
3.3 Performance of rotors with different blade profiles. . . . . . . 58
3.4 Influence of different pitch angles on the lower rotor. . . . . . 67

5.1 Identified parameters on the ground. . . . . . . . . . . . . . . 84

5.2 Identified parameters on the test benches. . . . . . . . . . . . 85
5.3 Identified parameters. . . . . . . . . . . . . . . . . . . . . . . 93

xix
xx
List of Figures

1.1 Different rotor configurations: a) Quadrotor, b) Axial, c) Con-

ventional main and tail rotor, d) Coaxial and e) Tandem. . . 4
1.2 Different steering concepts. a) Changing the orientation of the
air flow, b) changing the center of gravity, c) changing the tip
path plane respective to the fixed body frame. . . . . . . . . . 6
1.3 Definition of the azimuthal angle ψ on the helicopter, which
describes the rotor disc station of a blade. . . . . . . . . . . . 6

2.1 The first prototype of muFly (V1.0). . . . . . . . . . . . . . . 10

2.2 The intermediate prototype of the muFly helicopter (V1.1). . 10
2.3 The final muFly prototype (V2.0). . . . . . . . . . . . . . . . 11
2.4 Mass comparison between the muFly prototype V1.1 and the
final prototype V2.0. . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 The principle of the stabilizer bar. Due to the high inertia the
stabilizer bar lags behind the roll/pitch movement, applies a
cyclic pitch input to the rotor and creates a redress moment. 13
2.6 Inertial Measurement Unit with absolute pressure sensor. . . 13
2.7 (l) SRF10 [8] ultra sonic distance sensor. (r) Laser ring and
omnidirectional camera for eight distance measurements. . . . 15
2.8 Comparison of the image quality of the muFly downlooking
camera (l) and a state of the art webcam (r). . . . . . . . . . 15
2.9 (l) The Mighty Midget 13-6-11Y Brushless Direct Current
(BLDC) motor. (r) Walkera servo motor. . . . . . . . . . . . 16
2.10 The assembly of the final muFly prototype. . . . . . . . . . . 18
2.11 Electronics block diagram. . . . . . . . . . . . . . . . . . . . . 19
2.12 Graphical User Interface (GUI) in LabView R
for simple data
acquisition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.1 Airfoil nomenclature and forces. . . . . . . . . . . . . . . . . . 22

3.2 Effects of the Reynolds number on smooth and rough airfoils [64]. 24

xxi
3.3 Flow effects around a three dimensional rotor blade. . . . . . 25
3.4 Rotor-tip vortices mark the descent of an AH-1W Super Co-
bra [91]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Illustration of the coaxial setup. The lower rotor is working in
the downwash of the upper. . . . . . . . . . . . . . . . . . . . 28
3.6 Illustration of the interaction of the rotor wakes in coaxial
rotor configuration. . . . . . . . . . . . . . . . . . . . . . . . . 29
3.7 Influence of the gear ratio on the muFly propulsion system. . 30
3.8 Test bench for rotor blade optimization. (a) 2x Maxon EC 45
flat 30 W motor, (b) 2x optical encoders, (c) RTS 5/10 torque
sensor, (d) FGP FN 3148 force sensor. The rotor blades are
manufactured by a rapid prototyping machine. Not shown
data acquisition module NI USB 6009 and the two Maxon
Epos 24/5 motor controllers. . . . . . . . . . . . . . . . . . . 32
3.9 Motor controller current measurements versus measured drag
torque by the torque sensor. . . . . . . . . . . . . . . . . . . . 32
3.10 Rapid prototype machine manufactured blades for the rotor
test bench. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.11 Block diagram of the rotor test bench schematics. . . . . . . . 33
3.12 Side view of a blade element with constant width ∆r. . . . . 36
3.13 Momentum theory control volumes. . . . . . . . . . . . . . . . 37
3.14 Block diagramm of the Blade Element Momentum Theory
(BEMT) code. . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.15 Comparison of the BEMT simulation with the corresponding
measurements for different blade profiles. . . . . . . . . . . . 40
3.16 Lifting-line model. . . . . . . . . . . . . . . . . . . . . . . . . 40
3.17 Straight-line vortex element. . . . . . . . . . . . . . . . . . . . 41
3.18 Illustration of the prescribed rotor wakes. . . . . . . . . . . . 42
3.19 Velocity vector graph of a Vortex Method result. . . . . . . . 43
3.20 Comparison of the Vortex Method simulation with the corre-
sponding measurements for a coaxial setup. (l) Total thrust
of the upper and lower rotor. (r) Torque of the lower rotor. . 44
3.21 Single rotor Computational Fluid Dynamics (CFD) setup. . . 45
3.22 Domain (l) and mesh resolution (r) investigation results. . . . 46
3.23 CFD results for a standard NACA0012 profile. Static pressure
on the blade surfaces . . . . . . . . . . . . . . . . . . . . . . . 50
3.24 CFD results for a standard NACA0012 profile. Velocity field
in the fluid domain. . . . . . . . . . . . . . . . . . . . . . . . 51
3.25 Dependency of the thrust and torque coefficients on the rota-
tional speed. . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

xxii
3.26 Influence of the number of rotor blades on the rotor performance. 53
3.27 Influence of the rotor radius on the rotor performance. . . . . 54
3.28 Influence of the chord length on the rotor performance. . . . 55
3.29 Different airfoils of the NACA four digit series. . . . . . . . . 56
3.30 Influence of the blade thickness on the rotor performance. . . 56
3.31 Influence of the maximum camber on the rotor performance. 57
3.32 Influence of the camber position on the rotor performance. . . 58
3.33 The four best airfoil profiles found. . . . . . . . . . . . . . . . 59
3.34 The four worst airfoil profiles found. . . . . . . . . . . . . . . 59
3.35 Influence of the taper on the rotor performance. . . . . . . . . 60
3.36 Influence of the twist on the rotor performance. . . . . . . . . 60
3.37 The two tested blades with different tip shape. . . . . . . . . 61
3.38 Influence of the two different tip shapes on the rotor performance. 61
3.39 Three different tested versions for winglets on the rotor. . . . 63
3.40 Influence of winglet-like blade ends on the performance. . . . 63
3.41 Illustration of a gurney flap at the trailing edge of the airfoil. 64
3.42 Influence of the number of Gurney flaps. . . . . . . . . . . . . 64
3.43 (up) Differences in thrust and torque on coaxial rotor config-
uration and single rotor configuration. (dw) Corresponding
aerodynamic coefficients of the lower rotor. . . . . . . . . . . 65
3.44 Influences of the thrust coefficient of the lower rotor for differ-
ent working conditions. . . . . . . . . . . . . . . . . . . . . . 66
3.45 Influence of the rotor distance on the lower rotor. . . . . . . . 68

4.1 Coordinate frame convention and the forces and moments act-
ing in hover (back view). . . . . . . . . . . . . . . . . . . . . . 70
4.2 Influence of the inflow velocity on the thrust coefficient (CFD
simulation results). . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3 Visualization of the tilted thrust vector with tilt angles α and β. 75
4.4 The reaction of the modeled helicopter to an initial roll dis-
placement of 20◦ with the modeled stabilizer bar. . . . . . . 79
4.5 muFly dynamic model block diagram. . . . . . . . . . . . . . 81

5.1 Least squares result plots for the electric motor parameter
identification. . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2 The identification process block diagramm. . . . . . . . . . . 87
5.3 CMA-ES identification result plots for an actuation in yaw.
(up) Motor input signals. (mid) Ultra sonic altitude measure-
ment and model. (dw) IMU yaw rate measurement and model. 89

xxiii
xxiv ACRONYMS

5.4 CMA-ES identification result plots for an actuation in heave.

(up) Motor input signals. (mid) Ultrasonic altitude measure-
ment and model. (dw) IMU yaw rate measurement and model. 90
5.5 CMA-ES identification result plots for an actuation in roll.
(up) Servo input signals. (dw) IMU roll angle measurement
and model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.6 CMA-ES identification result plots for an actuation in pitch.
(up) Servo input signals. (dw) IMU pitch angle measurement
and model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.1 Illustration of the weighting principle for the H∞ -control design. 97

6.2 Control structure for muFly. . . . . . . . . . . . . . . . . . . . 98
6.3 Singular value plot of the heave-yaw subsystem. . . . . . . . . 100
6.4 The dual 2-degree of freedom (dof) GS/T -weighting scheme. . 101
6.5 Singular value plots of the heave-yaw plant loop gain Lu (up),
the sensitivity Su and the complementary sensitivity Tu trans-
fer function (dw). . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6 The dual 1-dof GS/T -weighting scheme. . . . . . . . . . . . . 103
6.7 Singular value plot of the roll-pitch subsystem with stabilizer
bar. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.8 (up) Singular value plot of the roll-pitch subsystem loop gain
Lu , (dw) sensitivity Su and the complementary sensitivity Tu
transfer functions. . . . . . . . . . . . . . . . . . . . . . . . . 104
6.9 Structure for position control. . . . . . . . . . . . . . . . . . . 105
6.10 Simulation result plots for a closed-loop step response in heave. 108
6.11 Measurement result plots for a step response in heave. . . . . 109
6.12 Simulation result plot for a reference following in pitch. . . . 110
6.13 Measurement result plots for a reference following in pitch.
The reference is applied by the pilot with the RC stick. . . . 111
6.14 The high-level control structure of muFly. . . . . . . . . . . . 112
6.15 The mission state machine. . . . . . . . . . . . . . . . . . . . 113
6.16 Illustration of a three-dimensional cubic spline. . . . . . . . . 115
6.17 Simulation results with trajectories for waypoint following. . . 115
6.18 The eight laser diodes seen from the top. . . . . . . . . . . . . 116
6.19 Illustration of the repulsive forces. . . . . . . . . . . . . . . . 117
6.20 High-level control simulation result. The helicopter is flying a
sample square track. . . . . . . . . . . . . . . . . . . . . . . . 118
Chapter 1

Introduction

The idea of having flying robots has been fascinating mankind for a long
time. Small autonomous flying machines even took their parts in movies
such as ’batteries not included ’, ’giant robot’ or ’wall-e’ and delighted mil-
lions of people. However, besides the entertainment in these fictive movies,
there exist real applications in different areas where an Unmanned Aerial
Vehicle (UAV) could be used. Such tasks comprise surveillance and security,
search and rescue, as well as inspection and exploration. Depending on the
task and environment different capabilities, sizes, autonomy levels and plat-
forms are necessary.
In the outdoor environment and large workspaces mostly fixed-wing systems
are used, due to the higher energy efficiency and higher velocities. Highly
developed platforms with various propulsion systems and sizes are found in
military application such as the Predator [11], the EADS Barracuda [9] or the
Ranger [16] from RUAG, but also in civilian areas such as the Sky-Sailor [72]
from the ETH Zürich. Besides these rather large vehicles there exist also
smaller airplanes belonging to the category of Micro Aerial Vehicle (MAV).
A very small and famous platform is the Black Widow [45].
In the indoor environment the use of fixed wing airplanes is strongly limited
due to restricted space and often even more importantly: the disability to
hover. Real hovering is only possible by using rotary-wing or flapping-wing
vehicles. The latter are inspired by nature and imitate the flight of birds or
insects. Believing in a higher efficiency in small scale of the insect-like loco-
motion, several groups around the globe work on flapping-wing vehicles, such
as the DelFly [7] or the MicroBat [75]. Even if there are examples showing
good flying abilities, there does not exist a flapping mechanism that produces
enough lift to carry enough sensor payload for autonomous missions yet. As

1
2 1. INTRODUCTION

long as there is not an improvement in the field of actuators (faster, more

stroke), sensors (lighter, more accurate) or sensor processing (image based
flight), this will not be the case.
This reduces the reasonable choice for autonomous indoor operation to the
range of rotary-wing vehicles and more precisely to helicopters. Possible
foreseen missions could be surveillance of exhibitions, search for survivors in
buildings threatened by collapse after an earthquake or similar scenarios. In
rescue cases in small buildings such as normal houses, the size of the MAV
and the autonomy can be of high importance to pass narrow spaces. In order
to address this scenario the European project muFly [14] was launched in
July 2006 with the goal of designing a fully autonomous micro helicopter in
the size and mass of a small bird. The muFly project consists of a consor-
tium with six equivalent partners, the ASL from Eidgenössische Technische
Hochschule Zürich (ETHZ), BeCAP at Berlin University of Technology [3],
CEDRAT Technologies in France [5], CSEM [6] from Switzerland, Depart-
ment of Computer Science at University of Freiburg [2] and XSENS Motion
Technologies [17]. The present thesis lies within the scope of this project.
Designing a fully autonomous helicopter in this small size bears various chal-
lenges and needs cutting-edge technologies. The design process consists of
many different parts and in order to reach the ambitious goals, all the indi-
vidual processes have to be thoroughly investigated. This thesis addresses
mainly two of the most important areas of the micro helicopter design: the
rotor aerodynamics and the control.1

1.1 State of the Art

The state of the art in the two mentioned areas of the thesis, the rotor aero-
dynamics and the control of helicopters, is in general very advanced. In both
areas, many years of experience in research and industry exist. However this
is mostly done on bigger platforms and the field of MAVs just established a
few years ago. Both areas are strongly affected by the small size and differ
from bigger platforms. On the aerodynamics side the rotor efficiency suffers
from the viscous effects in the low Reynolds number regime caused through
the small rotor radius and chord. On the control side the small size and low
inertias of the helicopter body strongly increase the dynamics [66] making
the helicopter more agile and the control more difficult. Additionally only
low cost actuators and sensors can be used and the computational power on
board is very limited. More insight on the different phenomena and existing
1 The system modeling is here considered as a part of the control.
1.1. STATE OF THE ART 3

works in those fields is given in the specific chapters and an overview on ex-
isting platforms and configurations is given here instead.
The design of helicopters in model size differs in many ways from the heli-
copters in full scale. One of the main differences is the rotor configuration
and the way of operation. In full scale, mostly the mechanical complexity
and the weight of the propulsion group is the limiting factor. Differently to
the model helicopters, huge and complex engines and gearboxes have to be
used to drive the rotor. This makes it often non-economic and difficult to
operate more than one main rotor. This is one of the reasons why helicopters
with one main rotor and a tail rotor for torque compensation did establish
well in the helicopter community.
In the micro helicopter size, all the vehicles are driven by electric motors
reducing the complexity and the weight of the propulsion group. Further-
more those electro motors allow to easily changing the thrust on the rotors
by changing the rotational speed, where as in full scale a constant rotational
speed is desired and the change in thrust is usually done by changing the
pitch of the rotor blades (collective pitch).
Those facts allow using many different rotor configuration in the model and
micro helicopter size. In Figure 1.1 the most common one are shown. Beside
the predominant conventional helicopter in full scale with one main rotor and
a tail rotor for torque compensation, there exist various other configuration
such as axial (single rotor with flaps for torque compensation at the fuselage),
coaxial (two counter rotating coaxial rotors), tandem (two counter rotating
rotors side by side) or quadrotor (four rotors). The choice of the rotor con-
figuration depends mainly on the maximum size and payload for the foreseen
missions.

For the most of these configurations, there exist examples of MAVs devel-
oped by different groups around the world, with the goal of taking a step to-
wards fully autonomous flying robots. An often used platform is the quadro-
tor due to its simple setup. Existing platforms are the X-4 flyer [76] from the
Australian National University, the STARMAC at Stanford University [50]
or the OS-4 [27] from the EPF Lausanne. Different highly developed systems
are commercially available for example the Hummingbird by Ascending Tech-
nologies [1]. All of those platforms are way bigger than the targeted goals
of muFly. Only the Mesicopter [54] developed by Stanford University is of
the size of muFly (even smaller), but was not capable to fly. In general it
looks like it is difficult to have a compact solution by using rotors side by side
which is also valid for the tandem configuration. In fact not a lot of tandem
MAV configuration are found an exception is found in [37].
More promising to realize a micro helicopter in the size of muFly are the con-
4 1. INTRODUCTION

Figure 1.1: Different rotor configurations: a) Quadrotor, b) Axial, c) Conventional

main and tail rotor, d) Coaxial and e) Tandem.

ventional and axial helicopters. In the toy world the smallest conventional
helicopters are the PicooZ [31] or the Pixelito with the patented steering
mechanism [78] from Van de Rostyne. However, the disadvantage of the con-
ventional helicopter is the extra device for torque compensation (tail rotor)
which does not contribute to the lift force. The same is true for the axial
configuration, where mostly flaps mounted at the fuselage are used for the
compensation. An example of such a system is the iStar [61]. In this case
the flaps add additionally extra weight and are not very efficient.
Most promising to fly in the targeted size are coaxial configurations. Existing
platforms are the MICOR [25] developed at the University of Maryland, the
CoaX [20] developed at the ETHZ or the MAVSTAR [63] developed at the
University of New South Wales. Yet again all of those examples are consid-
erably bigger than the target. The smallest existing (coaxial) helicopter is
the Picoflyer [15] with 3.3g, but of course only remote controlled.
At the moment probably the only autonomous helicopter in the targeted size
of muFly is the muFR [10] from Epson. The muFR was a great project push-
ing the state of the art much further, by the use of new actuators and other
technologies, but it was not able to do missions and only showed hovering
like capabilities. Thus the state of the art is still far away from the desire of
1.2. COAXIAL HELICOPTER 5

having autonomous MAVs in this size.

In order to choose the right concept for the muFly project an evaluation was
carried out and presented in [81]. As expected it emphasized that a coaxial
setup is the most promising, mainly due to the compactness of the coaxial
aligned rotors. This configuration allows to build the two rotors as big as
the given dimension limitation allowing a favorable setup. Of course some
losses are implied due to the operation of the lower rotor in the down wash
of the upper one, but this was rated less than reducing the rotor diameter for
example to have four rotors side by side as in the quadrotor configuration.
After the decision for the coaxial rotor configuration a closer view on the
principles of the coaxial helicopter is needed and presented in the next sec-
tion.

1.2 Coaxial Helicopter

In this section the steering principles of helicopters with coaxial rotor config-
uration are illustrated. It is important to understand the mechanisms since
it is essential for the modeling and control of the helicopter. For full control
the helicopter has to be steered around four axes: horizontal (x- and y- axis),
in heave (z-axis) and in heading (ψ).
The steering in heave and heading is usually done the same on all the model
size helicopters. Thereby the heave is controlled by simultaneously varying
the rotor speed and thus changing the net thrust. The heading is controlled
by differential speed variation of the two rotors leading to an imbalance in
the net drag torque around the rotor axis and a yawing of the fuselage. This
of course implicates a strong coupling between those two quantities and has
to be considered in the control design.
In the horizontal directions the coaxial helicopter offers several possibilities
of how to control the attitude and horizontal movement of the helicopter. In
Figure 1.2 the different principles are illustrated.
In [79] all of those principles are explained in detail and examples of ex-
isting platforms with such mechanisms are shown. Additionally, the reasons
for the choice on the muFly helicopter are discussed. It emphasized that
changing the tip path plane (TPP) with a classical swash plate steering, as
it is found on full scale helicopters, is the best choice for muFly due to the
compactness and maneuverability of this mechanism.
The functionality of changing the TPP with the help of a swash plate mech-
anism is explained in detail literature (e.g. [22,92]). The basic idea is to vary
the aerodynamic forces on the rotor blade to achieve a tilting of the TPP.
Therefore, the blade lift on the different rotor disc stations (see Figure 1.3) is
6 1. INTRODUCTION

T1 T1 T1
TPP

T2 T2
TPP T2

Mc M
Mc

Fc
G CoG G
G

Figure 1.2: Different steering concepts. a) Changing the orientation of the air
flow, b) changing the center of gravity, c) changing the tip path plane respective to
the fixed body frame.

varied cyclically with a period of one rotor revolution around a nonzero mean
lift. The lift difference between the rotor stations causes the TPP to tilt in
the desired direction, which results in a pitching and a horizontal acceleration
of the helicopter. On muFly the lift difference at the different rotor stations
is created by a change in the rotor blade pitch (named cyclic pitch) with the
help of the swash plate. As a side note it has to be mentioned that the tilting
of the TPP has a phase shift due to the inertias of the blades. This means if
the maximum lift is at the rotor station ψ = 0◦ the blade will start to flap
and reaches a maximum at an angle 0◦ < ψ ≤ 90◦ , meaning that this cyclic
pitch input will not lead in a pure nose down pitch motion of the helicopter,
it will result in a coupled pitch-roll movement.
The specialties of the coaxial aerodynamics are discussed in detail in Chap-
ter 3.
ψ=90°

ΔL
ψ=0°
ψ=180°
ΔL

ψ=270°

Figure 1.3: Definition of the azimuthal angle ψ on the helicopter, which describes
the rotor disc station of a blade.
1.3. CONTRIBUTIONS 7

1.3 Contributions
This thesis focuses on the propulsion system design, the modeling and the
control of autonomous micro helicopters. The difficulty and novel thereby
is all related to the extremely small size of the vehicle. At this point it
has to be mentioned that it is not a theoretical work on fluid mechanics or
control. In fact it is focusing on the open questions arising while designing
such a system. It addresses all the researches and engineers developing such
a system. The goal is to show possible design approaches and guidelines in
order to tackle the emerging problems. More specific this thesis contributes
to the mentioned topics in following fields

• Aerodynamics
The aerodynamics of the rotors at the desired small size differs strongly
from the aerodynamics at full scale. Since only recent advances in
battery technology made it possible to power such small helicopters,
the interest and research in this field is still not yet well established.
Therefore, there is a need for investigating the propulsion system and
blade performance in this rotor size. This thesis contributes to the
problem of rotors in the low Reynolds numbers with experimental and
numerical investigations. Thereby a coaxial rotor test bench and three
simulation approaches with different advantages and disadvantages are
presented. As another specialty, the whole drive train of the helicopter
is incorporated in the investigation, leading to a real condition as on
the final system and not only to an aerodynamic optimization. The
results of the investigations can be used as guidelines for future micro
helicopter design.

• System modeling and parameter identification

The nonlinear model developed in this thesis includes all the specialties
of the designed platform such as the stabilizer bar or the drive train. On
one hand the model has to reflect accurately the physics of the overall
system for simulations; on the other hand it has to be simple enough
to be used for the controller design. Thus all the different effects on
the system dynamics are discussed in detail and the reasons for all the
simplification are given. Together with the nonlinear parameter identi-
fication approach it contributes to the todays state of the art techniques
and delivers a good model for the controller design and simulation.

• Control
Finally this thesis also contributes to the control system theory by
8 1. INTRODUCTION

realizing model-based, multiple-input multiple-output (MIMO) H∞ -

controllers on a very small helicopter with low processing power and
low quality sensors and actuators. The successively realization on the
helicopter shows the appropriateness of the H∞ -technique and the good
quality of the identified nonlinear model. Beside the low-level control,
concepts for the high-level control on such systems with low processing
power are demonstrated.

1.4 Structure of this Work

After this brief introduction, in Chapter 2 the evolution of the muFly micro
helicopter is shown and the setup with all sensors and actuators is explained
in detail. Chapter 3 addresses the investigations on the rotor aerodynamics.
It starts with discussing the basic fluid mechanics effects around the rotor, fol-
lowed by the experimental and simulation approaches in order to investigate
the rotor performance. Finally, the investigations on different rotor setups
are shown, before starting with the dynamic modeling of the helicopter in
Chapter 4. In order to use the model for control the system parameters are
determined in a system identification process described in Chapter 5. The
low-level control is then presented in Chapter 6 together with the high-level
control for autonomous missions.
Chapter 2

The muFly Helicopter

During the muFly project three different prototypes were built. In a first
attempt an innovative design using a carbon cage for protection around the
rotors was realized which is shown in Figure 2.1. New ideas were realized
on this version such as a swash plate mechanism actuated by piezoelectric
elements (presented in [81]). Unfortunately the setup showed some stability
problems and was very difficult to fly. Furthermore the experience with this
platform showed that with todays state of the art sensors and the initially
proposed mass of 50g a fully autonomous flight is hardly reachable. Addition-
ally it emphasized that the proposed rotor diameter of 10cm is not enough
to carry the needed sensor setup.
The problems with the first prototype led to the decision of using a stan-
dard coaxial helicopter configuration with a commercially available propul-
sion system. In order to reduce the overall mass of the system, a novel
design incorporating almost all the sensors and electronics in the structure
was chosen.
During the design process of the final prototype an intermediate solution was
designed for further testing which is shown in Figure 2.2. It consists of a
structure built using a rapid prototype machine. It uses an adaptation of
the electronics of the first prototype and the sensors are similar to the final
prototype which is explained in detail in Section 2.1. Compared to the fi-
nal prototype this intermediate prototype misses the omnidirectional camera
and lasers as well as the downlooking camera. Thus it is not able to perform
obstacle avoidance or position control. Nevertheless, this platform showed
excellent flight performance and it is used for various tests in this thesis.
The final prototype is shown in Figure 2.3. As mentioned before, the
Printed Circuit Boards (PCBs) of the electronics form the structure of the

9
10 2. THE MUFLY HELICOPTER

Figure 2.1: The first prototype of muFly (V1.0).

Figure 2.2: The intermediate prototype of the muFly helicopter (V1.1).

2.1. HARDWARE 11

helicopter and help saving structural mass. The hardware and software em-
ployed on the final prototype is presented in detail in the next sections.

Figure 2.3: The final muFly prototype (V2.0).

In Figure 2.4 a mass comparison is shown between the intermediate pro-

totype V1.1 and the final prototype V2.0. With the new design the total
mass is smaller, even though there are more sensors on board. The main
difference is in the structural mass, which is about four times less with the
electronics integrated. Additionally, a lot of cables could be saved, due to in-
terfaces at the PCB connections, making the electronics more compact. The
overall mass of the final prototype is slightly more than 80g.

2.1 Hardware
This section discusses the hardware setup of the final prototype. It consists
of several electronic and mechanical components such as the sensors and
actuators or the propulsion system.

2.1.1 Propulsion System

The propulsion system on the muFly helicopter consists of the two counter
rotating rotors, the swash plate and the stabilizer bar. On the two last pro-
totypes the propulsion system of the Walkera 5G6 helicopter is used. The
12 2. THE MUFLY HELICOPTER

100
Battery
90 Actuators
Rotor system
80 Electronics
Sensors
70 Structure

60
Mass [g]

50

40

30

20

10

0
muFly V1.1 muFly V2

Figure 2.4: Mass comparison between the muFly prototype V1.1 and the final
prototype V2.0.

main reason for this choice is the easy and fast availability of spare-parts.
The rotors have a larger diameter (17.5cm) as originally foreseen in the pro-
posal since the overall mass of the helicopter is higher than expected. They
are driven by two Brushless Direct Current (BLDC) motors explained later
in this section and are able to lift a mass of approximately 100g.
The swash plate is a standard three-point swash plate with a sphere joint in
the middle and a ball bearing for the connection to the rotational part. It is
driven by two servos described later.
A specialty of the rotor system is the stabilizer bar mounted at the upper
rotor. Such stabilizer bars are found on toy helicopters and have the function
of stabilizing the roll and pitch movement of the helicopter. Without those
stabilizer bars, it is close to impossible for a pilot to steer a helicopter in this
size.
In simple words this stabilization mechanism gives cyclic pitch inputs, similar
to the swash plate, to the upper rotor in order to stabilize the helicopter in
flight. The stabilizer bar has a high inertia and lags behind a fast roll or pitch
movement of the fuselage as shown in Figure 2.5. Through a rigid connection
to the rotor, this time delay results in a cycling pitching of the rotor blades
and therefore to a tilting of the TPP, the plane the rotor blade tips are run-
ning in [92]. If the stabilizer bar is adjusted correctly the thrust vector points
into the opposite direction of the roll or pitch movement causing a redress
moment and stabilizing the helicopter. The stabilization mechanism leads to
a low-pass behavior of the roll-pitch dynamics of the helicopter.
2.1. HARDWARE 13

TUP TUP

TDW TDW

M
CoG

Figure 2.5: The principle of the stabilizer bar. Due to the high inertia the sta-
bilizer bar lags behind the roll/pitch movement, applies a cyclic pitch input to the
rotor and creates a redress moment.

2.1.2 Sensors and Actuators

In order to achieve autonomy, the helicopter has to be equipped with sensors
in order to be able to sense its attitude, altitude, position and environment.
Since the payload of the helicopter is extremely limited, the sensor selection
is restricted to the necessary minimum. Additionally, all the sensors are
selected or designed for low weight and low power consumption.

Inertial Measurement Unit

The attitude of the helicopter is measured with the help of an Inertial Mea-
surement Unit (IMU). The IMU combines the measurements of gyroscopes,
accelerometers and magnetometers in all the three space direction for an
attitude estimation of the helicopter. The IMU used on the helicopter is
especially designed for the muFly project by our partner XSENS [17] us-
ing high-end micro-electro-mechanical systems (MEMS) sensors (Figure 2.6).
Additionally, an absolute pressure sensor is integrated and fused with the ac-
celerometers in order to have an estimation of the altitude.

Figure 2.6: Inertial Measurement Unit with absolute pressure sensor.

While on the first two prototypes the IMU was working reliably with high
accuracy, some problems appeared on the final prototype: the vibrations on
14 2. THE MUFLY HELICOPTER

the Printed Circuit Board (PCB) were unforeseeable strong, leading to a

saturation of the accelerometers designed for accelerations up to 6g. This
clipping of the sensor makes it impossible to estimate a correct attitude of
the helicopter. Even on the ground the running rotors caused deviations up
to 20◦ in roll and pitch angle. This strongly limits the use of the IMU on the
final prototype.

Ultrasonic Sensor
The distance to the ground is measured with an ultrasonic range finder, the
SRF10 from Devantech [8] (Figure 2.7). The range for this sensor is from
0.06 to 6 meters in a cone of 60 degrees with a high accuracy (±0.01m). In
order to overcome problems with jumps in the measurement, the sensor is
filtered and also fused with the accelerometers in order to avoid reference
jumps while flying over an object. The implementation is based on the ideas
found in [47].

Omnidirectional Camera and Lasers

A unique sensor setup is built together with the project partner CSEM [6].
It consists of an omnidirectional camera and a ring of laser diodes as shown
in Figure 2.7. The idea is to detect the projection of the laser points on
the next object and calculate the distance by the use of triangulation. A
specialty of the camera is the polar pixel array of the Complementary Metal
Oxide Semiconductor (CMOS) image sensor [39] strongly reducing the cal-
culation power due to a simple remapping. On the downside the camera
provides only a small resolution of 128 times 64 (radial) pixels. With this
resolution it is possible to detect an obstacle between 0.3 and 1.5m with a
low accuracy.
This sensor setup leads to a structural design specification since for the tri-
angulation a minimum distance of 8cm is needed between lasers and camera.
This is the reason for the rather high fuselage of the helicopter. This sensor
is mainly used for obstacle avoidance.

Downlooking Camera
Even today indoor navigation is still a difficult task, but is crucial for au-
tonomous operations. In the indoor environment no Global Positioning System
(GPS) data is available and other techniques for navigation have to be used.
In recent research very small laser scanners are utilized for scan matching
allowing simultaneous localization and mapping (SLAM). On a platform like
2.1. HARDWARE 15

Figure 2.7: (l) SRF10 [8] ultra sonic distance sensor. (r) Laser ring and omnidi-
rectional camera for eight distance measurements.

muFly the use of such laser scanners is not possible, thus another method
has to be used. The navigation part in the muFly project is done by the
ALU-FR [2]. In order to estimate the horizontal position a 5mW downlook-
ing camera with 2.5g is mounted at the bottom of the helicopter. It tracks
features on the ground and uses the visual SLAM (V-SLAM) algorithm de-
veloped at the ALU-FR [86]. Since there is not enough onboard processing
power available, the camera images are transfered via a camera integrated
WiFi module to the ground station, processed and sent back to the helicopter
via the bluetooth module. Due to the miniature camera the resolution and
image quality is suffering strongly making, together with the vibrations of the
helicopter, the position estimation very difficult. In Figure 2.8 the difference
in the image quality between the very small 5mW camera and a standard
commercially available webcam is shown. As soon as the conditions are not
optimal the image quality suffers and good position estimation is extremely
difficult.

Figure 2.8: Comparison of the image quality of the muFly downlooking camera
(l) and a state of the art webcam (r).
16 2. THE MUFLY HELICOPTER

Drive Train and Servos

There are only four actuators mounted on muFly, two servo motors and two
BLDC motors as seen in Figure 2.9. While the original Walkera helicopter
servo motors are used to actuate the swash plate, the rotors are driven by
BLDC motors due to their higher power to weight ratio. The overall mass
on the helicopter is more than one and a half times the original weight of
the Walkera helicopter. The LRK 13-6-11Y from Mighty Midget Motors [13]
initially used on the first prototypes were redesigned for the final prototype
by the project partner CEDRAT. The BLDC motors have to be driven by
a motor controller, which controls the mean voltage supplied to the motor.
Motor controllers as well as the servo motors are steered by Pulse Width
Modulation (PWM) signals generated by one of the microprocessors.
The connection between the motors and rotors is done by gears with a ratio
of igear = 1.5.

Figure 2.9: (l) The Mighty Midget 13-6-11Y BLDC motor. (r) Walkera servo
motor.

Battery

One of the initial ideas of the muFly project was to use a miniaturized fuel
cell for the power source of the platform. But already early calculations
showed that the power to weight ratio of a fuel cell with hydrogen generator
is too low for the application on the helicopter. Thus only electric batteries
are a reasonable choice. In this field the most promising are the Lithium
Polymer (LiPo) batteries. A 13g two cell battery with 7.2V nominal voltage
and 250mAh is used on all prototypes. Notice that the battery takes a big
part of the overall mass.
2.2. SOFTWARE 17

2.1.3 Logic and Communication

Microprocessors
The heart of the muFly helicopter is the microprocessors. Two different
microprocessors are selected in order to handle the different tasks. A 16-
bit 600MHz dual core Blackfin BF561 microprocessor from Analog Devices
is used for decoding the omnidirectional camera images and calculating the
laser point distances by triangulation.
The Blackfin communicates with a 16-bit 40MHz dsPic33 from Microchip [12]
responsible for the full control of the helicopter. On the dsPic all the sensor
and input data is collected, processed and finally used as input signal for the
controllers. The dsPic also has the authority over all the actuators and runs
the high-level and low-level control.

RC Module and Bluetooth

For development, debugging and safety, the helicopter has to be steered by
a human pilot. The interface between the helicopter and the pilot is a stan-
dard 2.4GHz Radio Control (RC) device used on the Walkera helicopter.
The 2.4GHz became standard in the model airplane society due to its high
reliability.
A bluetooth module is utilized as an interface to the ground station. This
is mainly used for data acquisition and sending commands to and from the
helicopter.

Assembly
In Figure 2.10 the assembly of the helicopter parts is shown. The PCBs are
connected by the soldered electronic interface pads supplying the next PCB
with power or allowing communication between the different components.
This saves the cable connections making it more reliable and lighter. In
order to give an overview on all the electronics on board, a block diagram is
shown in Figure 2.11.

2.2 Software
Since the goal is to be as autonomous as possible, the intelligence has to be
onboard. The only part which does not run on the microprocessors is the V-
SLAM algorithm due to the high computational cost of the image processing.
However, the outcome (position data) of the V-SLAM is send back via the
18 2. THE MUFLY HELICOPTER

20cm
15
10
5
0

Figure 2.10: The assembly of the final muFly prototype.

2.2. SOFTWARE 19

IMU Pressure Walkera Servo

Power mod.
DC/DC
LDO 2x
Batt.
UART PWM

SPI
Ultrasonic dsPic 2.4Ghz RC
dsPIC33FJ256GP506
I2C master

Walkera
SRF10

I2C slave
SPI1 master
I2C SPI2 slave SPI slave

SPI

PWM
Omnidirectional Blackfin
Camera BF561 Motor mod. Motor
(+flash mem.)
SPORT
SPI master 2x
SPI slave
I/O
I/O
UART

7V
Laser Bluetooth 5V
3V3
LMX9838

Figure 2.11: Electronics block diagram.

bluetooth module to the helicopter. All the other parts run on the helicopter
itself. The firmware is programmed in the language C.
Besides the programing of the microprocessors a software part on the PC
on the ground is necessary. This software part forms the interface between
the user and the helicopter. It emphasized that LabView R
is a flexible and
fast solution for creating a GUI. The GUI used in this project is shown in
Figure 2.12. It splits and decodes the data coming over the serial bluetooth
connection and visualizes all the sensor and input data. Additionally it al-
lows the user to send mission states and target waypoints to the helicopter.
Furthermore it allows to record the data which is used for the identification
process as described in Chapter 5.
20 2. THE MUFLY HELICOPTER

Figure 2.12: Graphical User Interface (GUI) in LabView

R
for simple data acqui-
sition.
Chapter 3

Aerodynamics

Today’s unmanned MAVs are equipped with high-end electronics and up to

date algorithms for navigation and control, making them soon capable of
executing difficult missions. But all of those flying robots suffer from the
same issue: their short autonomy time. This is simply due to the insufficient
aerodynamics of those fliers. In fact, power required for the propulsion of a
micro helicopter is more than 90% of the total power consumption, making
the optimization of the electronic components insignificant. As an example
the power needed to lift muFly is around 30W, whereas the power needed for
all the electronic components is less than 3W. Those facts make the aerody-
namics a very important issue and led to the motivation of this work.
Thereby the investigations concentrate first and foremost on the rotor aero-
dynamics which holds the largest potential for optimization. More exactly
it focuses on the rotor aerodynamics in hover, the predominant operation
point the helicopter is working at. Beside the pure aerodynamics the whole
drive train is incorporated into the investigations in order to have the same
conditions as on the helicopter. Thus it tackles the problem of the low rotor
efficiency from the view of the helicopter designer.
In the first section the effects around the rotor in single and coaxial configu-
ration in the low Reynolds numbers are discussed, followed by the influence
of the drive train. Afterwards the experimental setup as well as the dif-
ferent simulation approaches is shown. Finally the investigation results are
presented.
22 3. AERODYNAMICS

3.1 Rotor Aerodynamics

In order to understand the aerodynamic effects around the helicopter rotor
the problem can be split into two parts, similar to the Prandtl approach
for fixed-wings [18]. The first part only considers the aerodynamics around
an infinitesimal airfoil, corresponding to the cross section of the blade. The
second part observes the three dimensional effects around the rotor blade.

3.1.1 Airfoil Aerodynamics

The aerodynamics around a cross section of the rotor blade is comparable
to the aerodynamics around an airfoil. These aerodynamic effects are funda-
mental fluid mechanics and can be found in various literature e.g. [18, 55]. In
this section only an overview on the most important effects is given.

Forces on an Airfoil
In Figure 3.1 an airfoil with the most important geometry denomination and
forces is shown. A two dimensional airfoil in an isothermal inviscid fluid flow
experiences forces due to the shape of the airfoil. The fluid gets accelerated
and decelerated at certain points over the surface resulting in a differential
pressure distribution. These pressure forces can be summarized and split
into a force in the mean flow direction vm , the drag force D, and a force
perpendicular to the flow, the lift force L.

Chord
c
Chord Line
Thickness
Mean Camber Line
Max Camber
vm Lift L

Drag D

Figure 3.1: Airfoil nomenclature and forces.

In the case of the inviscid flow the drag force is equal to zero, since the
pressure in flow direction is equal to the pressure against the flow. This
behavior changes in a real viscous flow, where two types of drag forces result
from the fluid viscosity. On the one hand the airfoil experiences shear stresses
in the boundary layer, named skin friction, on the other hand separation
3.1. ROTOR AERODYNAMICS 23

towards the trailing edge may occur, resulting in a ’form drag’. In order
to understand the latter phenomena a closer look at the effects around the
airfoil is needed.
All airfoils have regions of lower-than-static pressure; even symmetrical non-
lifting airfoils introduce a region of accelerated flow and the associated lower
pressure. At the trailing edge of the airfoil the higher speed has to return to
approximately free stream condition experiencing a pressure recovery through
an adverse (increasing) pressure gradient. Depending on the energy and
momentum, the flow starts to separate at a certain point from the surface.
This is caused by the adverse pressure gradient and viscous effects, resulting
in a pressure loss on the airfoil and an increased form drag.
How long the flow is able to resist the adverse pressure gradient depends on
the angle of attack (AoA) and the chord Reynolds number Re, the ratio of
inertial forces to viscous forces

vm ρc vm c
Re = = (3.1)
µ ν

with the mean flow velocity vm , the chord length c and the kinematic
fluid viscosity ν = µ/ρ (fluid property). A small Re value thus results in a
strong effect of the viscosity on the flow and vice versa.
The Reynolds number strongly affects the airfoil performance and thus the
airfoil design. The aerodynamic effects in the low Reynolds number regime
(Re < 106 ), where the model airplane and model helicopters are working in,
differs strongly from the Reynolds number regime of full-scale helicopters.
The efficiency in the low Reynolds numbers suffers strongly from the viscous
effects resulting in a much smaller efficiency of the micro helicopters compared
to the full-scale [24].
In order to understand this worse performance, the important effects of the
low Reynolds numbers on the airfoil have to be discussed. Extensive surveys
of airfoils at low Reynolds numbers can be found in [30, 40, 62, 83].

Low Reynold Number Aerodynamics

In simple words the reason for the bad performance of airfoils in low Reynolds
numbers is that transition from laminar to turbulent occurs later down-
stream. Therefore the adverse pressure gradient is impressed on the laminar
boundary layer, which is unable to resist strong pressure gradients yielding to
an early separation. Contrarily airfoils operating at high Reynolds numbers.
Here the adverse pressure gradient is impressed on the turbulent boundary
24 3. AERODYNAMICS

layer, which can resist stronger gradients due to energy mixing with bound-
ary. This effect can be seen in Figure 3.2 taken from McMasters and Hender-
son [64], where the influence of the Reynolds number on smooth and rough
airfoils is shown. At low Reynolds number the performance of the smooth
airfoils is poor due to the laminar boundary layer. At a Reynolds number
around Rec ≈ 70, 000 (named critical Reynolds number) the performance
increases more than an order of magnitude, since for Re > Rec the flow is
in transition or already turbulent, capable to withstand the adverse pressure
gradient. This effect cannot be seen on rough airfoils where the rough surface
works as ’turbulator’ leading to an early transition to turbulence.

Figure 3.2: Effects of the Reynolds number on smooth and rough airfoils [64].

Another effect occurring at low Reynolds number is the ’laminar separa-

tion bubble’ [83]. The separated laminar boundary layer rapidly undergoes
transition to a turbulent flow and makes it possible to reattach on the rear
of the airfoil as a boundary layer and forms a bubble. A separation bubble
thickens the boundary layer and thus increases the drag of the airfoil. The
drag increment can be several times the drag of the airfoil without a sep-
aration bubble and is strongly undesired. This effect occurs for Reynolds
numbers higher than 70, 000; for lower Reynolds numbers the physical chord
length is too short and a reattachment is unusual [62].
In general it can be said that in the low Reynolds number regime the design of
3.1. ROTOR AERODYNAMICS 25

airfoils differs strongly from the airfoils performing well in the high Reynolds
number of full scale planes or helicopters. This is also known as the ’scale
effect’ which shows that an airfoil with good performance at high Reynolds
number do not have (or will not) perform well at low Reynolds numbers.

Aerodynamic Coefficients
For calculations a characteristic quantity is needed, which determines the
performance of the individual airfoils in the different flows. Such quantities
are the lift and drag coefficients cL and cD defined as the quotient of the
forces and the dynamic pressure

L0 D0
cL = 1 2
, cD = 1 2
. (3.2)
2 ρcvm 2 ρcvm

These coefficients are individual for all airfoil geometry and are depending
on the Reynolds Re number and the AoA α.

3.1.2 Aerodynamics of a Rotor Blade

Until now only two-dimensional effects were considered around an infinite
blade. In contrast the real rotor blade is a three-dimensional body, and con-
sequently the flow over the finite blade is three-dimensional. In fact there is a
component in the spanwise direction resulting from the differential pressure
distribution on the lower and upper surface creating the lift on the blade.
The air is forced from the high pressure zone at the lower surface to the low
pressure zone at the upper side, leading to a spanwise flow component as
illustrated in Figure 3.3.

Top Flow

Low Pressure Bottom Flow

Tip Vortex

High Pressure

Figure 3.3: Flow effects around a three dimensional rotor blade.

26 3. AERODYNAMICS

Behind the trailing edge of the blade the air starts to form a tip vortex.
This vortex drags the surrounding air with it resulting in an induced down-
ward velocity called downwash. Due to this downwash and the rotational
movement of the rotor the vortices prolongate downwards forming a helix
called ’rotor wake’. Under certain conditions those vortices can be seen on
full scale helicopters as shown in Figure 3.4, where the air is condensing in
the low pressure region in the core of the vortices.

Figure 3.4: Rotor-tip vortices mark the descent of an AH-1W Super Cobra [91].

The existence of the rotor wake negatively affects the blade performance.
In the first instance the downwash reduces the effective AoA (especially at
the blade tip) and supplementary the vortices lead to an additional drag.
This drag is similar to the induced drag on fixed-wings. There are different
interpretations for the existence of this induced drag [18]. One theory is that
this drag force is the result of the reduced angle of attack leading to a force
component of the lift vector in the direction of the rotor movement. Another
way how to look at it is based on the energy: the vortices at the tip contain
a large amount of translational and rotational kinetic energy; this energy has
to come from somewhere. In fact this energy comes from the motor driving
the rotor.
The minimized lift and the additional drag due to the tip vortices are often
summarized as ’tip losses’. Minimizing those tip losses is one of the goals of
the aerodynamic optimization process.
3.1. ROTOR AERODYNAMICS 27

Rotor Quantities
Considering the whole rotor on the helicopter, the lift and drag forces which
are depending on the inflow angle are not the important quantities. More
of interest are the thrust force T and the drag torque Q in the direction of
the rotor axis. While the thrust is the force which finally determines the lift
force of the helicopter, the drag torque is corresponding to the mechanical
torque needed to run the rotor at a certain rotor speed. In Section 3.3.1 the
relation between the lift/drag forces on the airfoil and the thrust/torque is
shown in detail.
Similar to the lift and drag, dimensionless coefficients are defined in order to
characterize the rotor. The thrust and torque coefficients are defined as [58]

T Q
cT = , cQ = , (3.3)
ρπR4 Ω2 ρπR5 Ω2
with the air density ρ, the rotational velocity Ω and the rotor radius R.
Similar to the airfoils those coefficients are used to characterize the rotor
performance.
Another important quantity is the mechanical power needed to drive the
rotor. The mechanical power is the product of the drag torque Q and the
rotational speed Ω

Pmech = QΩ. (3.4)

The mechanical power is relevant to determine the efficiency of the rotor.

In the case of the rotor the quantification of the efficiency is not as simple as
in other areas. Looking at the aerodynamics in general, the desire is to have
a rotor which produces as much thrust as possible, by using a minimum of
power to drive the rotor. Unfortunately the ratio thrust to power T /Pmech is
not a dimensionless quantity, which led to the introduction of the Figure of
Merit (FM) by the helicopter community. The FM is the ratio between the
rotor induced power and the total power used. It is defined as

ideal induced power Pind T vi

FM = = = (3.5)
total power Pind + Pp T vi + Pp

with the induced velocity vi . The FM is mostly found in fluid mechanics

literature, but its use is problematic [29], since the induced power at con-
stant thrust can be increased by a higher induced velocity. At the same time,
28 3. AERODYNAMICS

the total power is increased, which is undesired since it lowers the thrust
to power ratio. Strictly speaking, the FM is only a reasonable choice if the
power loading is kept constant [58].
Beside the individual disadvantages of all quantities used in literature there
exists another problem considering the design of a rotor for specific appli-
cations. A rotor only designed and optimized by aerodynamics quantities
will most probably not be the optimal choice on the applied helicopter. If
the electric motor cannot drive the blade at its optimal operation point, the
aerodynamic blade optimization is pointless. It is therefore crucial to include
the motor and gearing constraints in the process of propulsion optimization.
Thus all the investigations are done on the existing drive train on the heli-
copter.

3.1.3 Coaxial Rotor Configuration

As muFly has a coaxial rotor configuration, the interaction between the two
rotors has to be considered and studied. In Figure 3.5 an abstraction of the
coaxial setup is shown.

upper rotor

lower rotor

Figure 3.5: Illustration of the coaxial setup. The lower rotor is working in the
downwash of the upper.

As illustrated the lower rotor is working in the downwash of the upper

rotor. As a consequence the performance of the lower rotor is strongly influ-
enced. But also the upper rotor is affected by the coaxial setup due to the
interaction of the two rotor wakes as illustrated in Figure 3.6, an illustration
of the results of the vortex approach (vorticity plot) discussed in Section 3.3.2.
In literature such as [34, 46] the effects of the coaxial configuration are stud-
ied for full scale helicopters. Similar to the single rotor configuration it is
3.1. ROTOR AERODYNAMICS 29

interesting to know if the found observations are true for the small scale.
In the context of the rotor design, not only the thrust and efficiency is impor-
tant, the same attention has to be given to the drag torque balance, in order
to cancel out the yawing of the helicopter. If the two rotors are optimized for
operation points with different drag torques, it will not be possible to operate
both rotors in the optimum during flight.

Figure 3.6: Illustration of the interaction of the rotor wakes in coaxial rotor
configuration.

3.1.4 Drive Train

The drive train on the muFly helicopter consists of the motors and the gears
as described in Section 2.1.2. The motors are an important part of the
propulsion system. There exist many different types of electric motors with
different advantages and drawbacks. On muFly the motors have to run on
Direct Current (DC) due to the DC power supply by the LiPo batteries.
The most common types are the brushed and brushless types, whereas the
30 3. AERODYNAMICS

brushless motors have several advantages. The most important one is the
their power to weight ratio, which is much higher than for brushed motors
as shown in [71] by testing over 2000 motors. There are not many commer-
cial available motors in the small size and weight as it is needed on muFly.
There only exist some hobbyist motors, whereby the most promising motors
were analyzed by the project partners of CEDRAT [5] and the best fitting
selected. Additionally for the final prototype improved motors were built.
Those two types of motors form the basis of the further investigations since
motor optimization is not a part of this work.
The second part of the drive train, the gears, strongly influence the perfor-
mance of the propulsion system. A wrong choice of the gear ratio lowers the
performance dramatically, since the motor runs way off the optimal point.
The influence of the gear ratio on the propulsion of muFly is shown in Fig-
ure 3.7.

ides
Tdes

ides

Figure 3.7: Influence of the gear ratio on the muFly propulsion system.

In the left plot the change in the electrical input power Pel and the me-
chanical output power Pmech are plotted. As expected the electrical power
decreases with increasing gear ratio due to the decreased torque at the motor.
The mechanical power has a peak at igear ≈ 1.5 where the product of the
rotational speed and torque is maximal and the most power is put into the
rotor system. The second plot shows different choices of the designer and
again the difficulty of the performance judgment. The gear ratio might be
chosen at a ratio around igear ≈ 1.5 (as it is chosen on muFly) in order to
produce maximal thrust or at igear ≈ 3 to run the drive train at the high-
est power efficiency ηdrive = Pmech /Pel . If the rotor is still able to produce
enough thrust another choice would be to choose igear ≈ 4 the maximum of
the thrust to electrical power ratio T /Pel .
However, if the goal is to design a new propulsion system for a micro he-
licopter none of those criteria lead to the desired goal of efficiency. Since
the electrical power decreases with higher gear ratio the optimal point for
the design is the point where just enough thrust is generated for the mis-
sions, bringing in the engineering aspect of the optimization. The idea is to
find out the thrust Tdes needed for the wanted performance of the rotor and
determine the required gear ratio. As an example the results for a desired
thrust Tdes = 0.6N is shown in Figure 3.7. The outcome is that a gear ra-
tio of igear ≈ 2 yields the desired performance and is the optimal point of
operation.

3.2 Experiments
Simulating the performance of a rotor in the low Reynolds number regime is
very difficult due to the mentioned phenomena in the last section. Thus it is
necessary to validate the simulations with experimental data. Measurements
for rotors as small as used on muFly are rare to find, since only little literature
with experimental data is available [34]. In order to overcome this difficulty
a coaxial rotor test bench is build to measure the performance of different
rotor setups in single and coaxial configuration.

3.2.1 Test Bench Setup

The rotor test bench consists of two counter rotating coaxial axes allowing
investigating single and coaxial rotor setups (Figure 3.8).
The two axes are driven by two BLDC motors and are controlled via two
MAXON EPOS motor controllers which are connected to a PC by a serial
connection. At the bottom of the mechanical structure two load cell sen-
sors are mounted, measuring the total torque and total force in rotor axis.
The sensor data is collected by a data acquisition module and transfered to
the PC. For measuring the drag torque in coaxial mode, the linear relation
between the drag torque and the current consumed by the motors is used.
Thus the torques of the individual rotors can still be determined. In Fig-
ure 3.9 the current measurement of the motor controller is plotted against
the measurement of the torque sensor to validate the linear relation.
The rotor blades are designed in Computer Aided Design (CAD) and
manufactured on a rapid prototype machine. This combination allows fast
manufacturing and provides the flexibility to generate complex shaped blades.
As an example a selection of blades manufactured for the test bench is shown
in Figure 3.10.
32 3. AERODYNAMICS

Figure 3.8: Test bench for rotor blade optimization. (a) 2x Maxon EC 45 flat 30
W motor, (b) 2x optical encoders, (c) RTS 5/10 torque sensor, (d) FGP FN 3148
force sensor. The rotor blades are manufactured by a rapid prototyping machine.
Not shown data acquisition module NI USB 6009 and the two Maxon Epos 24/5
motor controllers.

Figure 3.9: Motor controller current measurements versus measured drag torque
by the torque sensor.
3.2. EXPERIMENTS 33

Figure 3.10: Rapid prototype machine manufactured blades for the rotor test
bench.

The rotor heads on the test bench are designed such that the blades can
be mounted rapidly by changing the middle part. These parts are milled
with different angles from 10◦ to 20◦ whereby the blade pitch can be varied
precisely.
The whole test bench is controlled by a MATLAB program on the PC allow-
ing to run different measurement modes. As an overview a block diagram of
the test bench is shown in Figure 3.11.

MAXON EPOS 24/5 Motor 1

PC Force
Opt. Encoder
FN3148
MATLAB

VCP Torque
USB
MAXON EPOS 24/5 Motor 2 RTS5/10
VCP
USB Opt. Encoder
VCP
USB
LAB PWR SUPPLY

DAQ NI6009

Figure 3.11: Block diagram of the rotor test bench schematics.

3.2.2 Calibration and Measurement
Calibration of the force sensor is simple. It is done by measuring a knowing
weight several times and calculating the offset and the scale factors. More
problems are caused by the torque measurement. The torque sensor is cal-
ibrated by the manufacturer and was checked before mounting. But due to
the small dimensions of the rotors only low drag torques occur and a very
sensitive torque sensor (Qmax = 20mNm) had to be used. In case of a rotor
blade failure strong unbalances occur leading to high torques and damage
of the sensor. It is recommendable not to use the torque sensor for all the
measurements, due to the high probability of sensor loss.
Using only the motor current measurement for determining the blade torque
is not possible since the current always slightly changes depending on the
operation temperature, daily conditions and the overall friction.
Hence a blade is carefully measured with the torque sensor initially for cali-
bration and then used as a reference blade. At the beginning of a measure-
ment series the test bench is warmed up and the reference blade is used to
map the consumed current to the torques.

3.3 Simulations
Simulating the performance of rotors in the low Reynolds number regime is
very challenging. Effects like transition or laminar separation bubbles are still
hard to model and good prediction is difficult. Nevertheless it is necessary to
have a simulation tool to predict the performance of the rotors and investigate
the influence of different rotor blade setups.
There exist various approaches on how to simulate helicopter rotors and
all have their advantages and disadvantages. In this thesis three different
methods are analyzed and implemented for the rotor simulation,

• the Blade Element Momentum Theory (BEMT),

• a Vortex Method,

• and Computational Fluid Dynamics (CFD).

While only CFD simulations are able to cover all the three-dimensional
effects, they suffer from their long preprocessing and processing time. CFD
simulations are very suitable for complex geometries, but can hardly be used
for parameter studies.
3.3. SIMULATIONS 35

On the other hand a BEMT simulation is only a quasi three-dimensional

approach and not able to treat effects like tip losses or the cross flow on the
blades. However, this method is very simple, fast and robust and is suitable
for studies on blade profiles and simple shapes.
The vortex approach is settled between those two simulations. It covers par-
tially the 3D effects and models the entire velocity field around the rotor,
making it capable to simulate coaxial configurations. Yet it is not able to
cover all the three dimensional phenomena and the geometry of the blade is
limited.
All the simulations are worth investigating and the the theoretical back-
grounds are discussed in the next sections.

3.3.1 Blade Element Momentum Theory

The Blade Element Momentum Theory (BEMT) is a common quasi three-
dimensional approach for rotor simulation. It combines the Blade Element
Theory (BET) and the momentum theory to overcome the problem of the
unknown induced velocity vI . An extensive survey on this method can be
found in various literature such as [29, 58].

Blade Element Theory

The numerical BET models the rotor blade as a sum of blade elements as
shown in Figure 3.12, where every element has constant flow and constant
geometry properties. The thrust T and torque Q on the whole rotor is thus
the sum of the thrust and torque on every blade element multiplied by the
number of blades Nb .

X X
T = Nb ∆T, Q = Nb ∆Q, (3.6)

The thrust ∆T and torque ∆Q in the center of the element are a function
of the lift and drag forces ∆L and ∆D

∆T = ∆L cos φ − ∆D sin φ, (3.7)

∆Q = (∆D cos φ + ∆L sin φ) r. (3.8)

with the radial distance to the rotor hub r and the inflow angle φ.
Using the aerodynamic coefficients of an airfoil, the lift and drag on the blade
element can be written as
36 3. AERODYNAMICS

1 2
∆L = ρcL c∆rvm , (3.9)
2
1 2
∆D = ρcD c∆rvm . (3.10)
2

ΔT ΔL

vm
α
vi θ φ ΔD

Figure 3.12: Side view of a blade element with constant width ∆r.

The difficulty is to obtain the correct aerodynamic coefficients for the

airfoil sections, especially in the low Reynolds number regime. In literature
usually the coefficients are obtained from measurement data for the inves-
tigated profiles (e.g. [23, 24, 87]). In our approach the Panel Method (PM)
code XFOIL programmed by M. Drela [36] is used. This code is often found
as a design tool for model airplanes and is designed for the lower Reynolds
number regime.
The PM models the flow around an airfoil as a potential flow superimposed
by vortex panels shaping the airfoil [18]. Since potential flows are inviscid,
a two-equation integral formulation based on dissipation closure is used to
model the viscid effects. Additionally a model to describe transitional ef-
fects is utilized in the form of the en -transition criterion [83]. This model
plays an important role since it determines the performance of the XFOIL
code. A suitable value for the tune parameter ncrit depends on the ambient
disturbance level in which the airfoil operates, and mimics the effect of such
disturbances on transition. Recommendations in the XFOIL documentations
are summarized in Table 3.1.
The helicopter rotor with its strong vibration is settled at a low value for
ncrit and indeed the results are strongly depending on this parameter. Only
for low values a good correlation with the measurement could be found and
a value of ncrit = 0.1 is chosen. The dependency of the airfoil performance
on the turbulence level is nicely illustrated in [56].
3.3. SIMULATIONS 37

Table 3.1: Recommendations for the parameter ncrit .

Situation ncrit
Sailplane 12-14
Motorglider 11-13
Clean wind tunnel 10-12
Average wind tunnel 9
Dirty wind tunnel 4-8
Vibrations/Rough walls/Large free stream turbulence 0-1

In order to compute the aerodynamic coefficients the Reynolds number

Re and the AoA α have to be known (Sec. 3.1.1). This leads to the main
problem of the BET since the real AoA is not known a priori due to the
influence of the downwash. A solution to this problem is the momentum
theory described next.

Momentum Theory
The second part of the BEMT is the (numerical) momentum theory also
known as actuator disc theory. This is an integral approach where the control
volume is split into elements as shown in Figure 3.13.

Δr

Figure 3.13: Momentum theory control volumes.

Using the mass and momentum conservation equations on the different

segments a relationship between the thrust ∆T and the induced velocity vI
is obtained [58]

∆T = 2ρ∆A (v0 + vI ) vI , (3.11)

38 3. AERODYNAMICS

with the disc area ∆A, the air density ρ and the inflow velocity v0 . Using
the geometric relation between the induced velocity and the inflow angle φ
the connection to the AoA can be determined by

vI + v0
tan φ = , (3.12)
Ωr
with the rotational velocity Ω.
Combining those two theories results in the BEMT and allows to determine
the real AoA and inflow velocity and thus the real thrust and torque of the
rotor.

Implementation
The classical BEMT approach as found in literature uses different assump-
tions to obtain a closed loop solution of reduced complexity, e.g. the lift polar
is a linear function. This is only partly true especially in the low Reynold
number regime.
In order to avoid such simplifications and to obtain better solutions, in our
approach the problem of finding the correct angle of attack is solved itera-
tively. In Figure 3.14 a block diagram of the BEMT algorithm implemented
in Matlab R
is shown. At the beginning an initial inflow angle φBET is chosen
and the needed flow properties are calculated. Those are the input to the
XFOIL code which returns the aerodynamic coefficients. Using the coeffi-
cients the thrust and torque are calculated and inserted into the momentum
theory equations to calculate the inflow angle φMOM . The predicted inflow
angle φMOM is then compared with the initial inflow angle φBET . If the two
angles have the same value, both theories are fulfilled and the correct AoA
is found.

Tip Losses
A problem of the BEMT is the tip losses. The two theories do not incorporate
the reduced lift distribution close to the tips. Thus a tip loss model has to
be used to cover this effect. In our code the Prandtl’s approach is used [43].
Prandtl showed that, when accounting for the tip losses, the effective blade
radius Re , is given by
 
Re 1.386 λI
≈1− p , (3.13)
R Nb 1 + λ2I
vI
with the number of blades Nb and the inflow ratio λI = ΩR .
3.3. SIMULATIONS 39

Initial values
φBET, v, …
φBET
Calculate Flow Properties
α, Re, Ma

XFOIL
cL, cD
Calculate T and Q
T

Momentum Theory
vI, φMOM

False
|φBET –φMOM| < ε Update φBET

True

Finish

Figure 3.14: Block diagramm of the BEMT code.

Validation
Before the BEMT code can be used for investigations the code has to be val-
idated. This is done by using measurement data produced on the test bench
described in Section 3.2. In order to overcome the differences in measurement
and simulation data, the BEMT code is ’tuned’ to one rotor, in this case a
rotor with R = 0.06m, c = 0.02m, NACA0012 blade profile with no twist and
taper. After adjusting the code to one rotor the BEMT code shows good cor-
relation with the measurements as shown in Figure 3.15, where rotors with
different blade profiles are compared to the corresponding measurements.
The results indicate the accuracy and flexibility of the code.

3.3.2 Vortex Method

Another approach often used in the rotary wings society is the Vortex Method.
In [58] a detailed overview on this method is given. Similar to the BEMT
the vortex approach consists of two parts. One part is exactly the BET as
seen before, but the second part, the problem of the unknown AoA, is solved
differently.
The idea is to replace the blade elements with horseshoe vortices as shown
in Figure 3.16 forming a lifting line model. The strength of the vortices Γi
(circulation) is in relation with the lift ∆Li on the individual elements, given
by the Kutta-Joukowsky equation
40 3. AERODYNAMICS

0.8 10
naca0012 meas naca0012 meas
0.7 naca0012 BEMT naca0012 BEMT
naca4412 meas 8 naca4412 meas
0.6 naca4412 BEMT naca4412 BEMT

Torque [mNm]
naca9412 meas naca9412 meas
0.5
Thrust [N]

naca9412 BEMT 6 naca9412 BEMT

0.4

0.3 4

0.2
2
0.1

0 0
3000 4000 5000 6000 7000 8000 3000 4000 5000 6000 7000 8000
Rotational Speed [rev/min] Rotational Speed [rev/min]

Figure 3.15: Comparison of the BEMT simulation with the corresponding mea-
surements for different blade profiles.

∆Li = Γi vm ρ∆c. (3.14)

r
Γn-1 Γn

Γ4
Γ3
Γ2
Γn
Γ1 Γn-1 - Γn

Γ 2 - Γ3
Γ 1 - Γ2
-Γ1

Figure 3.16: Lifting-line model.

Since the vortices can not end in a fluid (Helmholtz-Theorem), the vortices
will prolongate in the fluid domain to the next boundary. If the dissipation
is neglected or assumed to be very small the vorticity strength will thereby
stay the same. Those vortices induce a velocity on the fluid particles in the
domain, which can be calculated using the Biot-Savart integral
3.3. SIMULATIONS 41

∞
Γ dl × r
Z
vi = . (3.15)
−∞ 4π |r|3
If the geometry of the prolongated vortices is known, the vortices can be
split in vortex filaments and the resulting induced velocity field in the domain
and especially at the blade elements can be calculated, leading to the real
AoA. For a vortex filament (3.15) can be simplified as

Γ l12 × r1
vi = (cosθ1 − cosθ2 ) · , (3.16)
4πh |l12 × r1 |
where the geometrical properties are corresponding to Figure 3.17

r1 = rP − rA , l12 = rB − rA . (3.17)

B
Γ θ2

A θ1
h

P
vy
vx
vz

Figure 3.17: Straight-line vortex element.

As seen in Section 3.1 the vortices form a rotor wake helix, caused by the
rotation and downwash of the rotor. There exist different approaches how to
define the geometry of the rotor wake, in our case two different approaches
are implemented. The first is a prescribed rotor wake by Landgrebe [57]
which is adapted in every iteration. The second works similar to a parti-
cle method, where individual points of the discretized rotor wake are shifted
corresponding to the velocity induced by the other vortex filaments. Illus-
trations of the modeled rotor wakes in single and coaxial configuration are
shown in Figure 3.18.

The overall procedure for the simulation is as follows: first a lift distribu-
tion over the blade is chosen and depending on this lift the geometry of the
rotor wake is adapted. Additionally the vortex strength on the individual
blade elements is calculated. Now the rotor wake is fully defined and the
42 3. AERODYNAMICS

Figure 3.18: Illustration of the prescribed rotor wakes.

induced velocity at the blade elements can be calculated. Using Xfoil the lift
and drag on the blade elements is calculated yielding in a new lift distribu-
tion. This procedure is repeated until a stationary solution is found and the
thrust and torque can be calculated. More information on the vortex code,
such as vortex core elimination, the equations of the Landgrebe’s rotor wake
model or further calculation details are found in [44].
The advantage of this method over the BEMT is that no tip loss model is
needed since they are modeled automatically. Additionally this method al-
lows calculating the velocity field in the whole domain which allows using
this method for coaxial rotor configurations. Of course this method has also
its disadvantage such as it is still not able to treat all the three-dimensional
effects and simulation of complex geometry is not possible. For such simula-
tions a full CFD simulation is needed.

Typical Results
In Figure 3.19 a typical result of the Vortex Method is plotted. It is a ve-
locity vector plot showing the induced velocities on a grid around the rotor.
The plot nicely illustrates the downwash in the core of the rotor and the tip
vortices at the tips of the blades. Since there is no dissipation modeled the
vortices keep the same strength in the whole domain. Additionally, the air
around the rotor is more affected than in reality leading to an unrealistic
upwards movement of the air. However, the knowledge of the velocity in the
downwash allows using this simulation method for coaxial rotor configura-
tions.
3.3. SIMULATIONS 43

Figure 3.19: Velocity vector graph of a Vortex Method result.

Validation
In Figure 3.20 the simulation results are compared to the measurements from
the test bench. In the left plot the total thrust and in the right plot the torque
of the lower rotor is compared. While the simulated thrust is reasonably cor-
related with the measurement up to high rotational speed, a larger difference
is visible in the torque. For high speeds and high angles the simulation
overestimates the torque. In general, the difficulty lies in the overlapping of
the rotor wakes what leads to stability rendering the simulation challenging.
However, the simulation results are good enough for further investigations
where the advantage is the much smaller computational and work effort than
for CFD simulations.

3.3.3 Computational Fluid Dynamics

At last a commercial Computational Fluid Dynamics (CFD) software, Ansys
CFX R
, is used to investigate the rotor aerodynamics. As common, the fluid
domain is divided into cells (mesh) and the Navier-Stokes equations [55] are
solved for the individual cells using a Finite Volume Method (FVM) or Finite
Element Method (FEM). There is a lot of theory behind those methods and
it would be beyond the subject of this thesis to go into the details, even more
as a commercially available code is used. More about the fundamentals of
44 3. AERODYNAMICS

1.8 0.035
VM Sim. 12° VM Sim. 12°
1.6 VM Sim. 14° VM Sim. 14°
0.03
VM Sim. 16° VM Sim. 16°
1.4
VM Sim. 18° VM Sim. 18°
0.025
Meas. 12°

Torque [mNm]
1.2 Meas. 12°
Meas. 14°
Thrust [N]

Meas. 14°
1 0.02
Meas. 16° Meas. 16°
Meas. 18° Meas. 18°
0.8 0.015
0.6
0.01
0.4
0.005
0.2

0 0
4000 5000 6000 7000 8000 4000 5000 6000 7000 8000
Rotational Speed [rev/min] Rotational Speed [rev/min]

Figure 3.20: Comparison of the Vortex Method simulation with the corresponding
measurements for a coaxial setup. (l) Total thrust of the upper and lower rotor.
(r) Torque of the lower rotor.

CFD can be found in [73]. Nevertheless CFD is not a black box tool and a
lot of knowledge is needed to obtain appropriate solutions. The setup of the
simulation is very important which is subject of the next subsections.

Fluid Domains and Meshes

In a first step the different fluid domains and computational meshes are
defined. The air around one rotor is divided into three fluid domains, two
rotational domains and a stationary domain as shown in Figure 3.21. Thereby
only the half of the rotor is modeled, since periodic conditions can be set
between the two rotor blades. This helps reducing mesh cells and saving
working memory.
The domain around the rotor blade is filled with a structured mesh (blade
domain) in order to half a better control on the cells in the boundary layer.
The structured mesh domain is connected to an outer rotational domain (rot
domain) and is meshed unstructured for simplification. At last the rotational
domains are integrated in a stationary domain (stat domain), where different
boundary conditions can be set.
The split into three domains has various advantages. Beside the possibility
to use diverse types of meshes, it allows to reduce the meshing effort for the
simulations. In fact the ’blade domain’ is only meshed once for every blade
parameter and than inserted into one of the six ’rot domains’ for every pitch
angle.
The domain size has to be adapted in order to minimize the influence of
the boundary condition on the flow. In a too small domain the flow is not
3.3. SIMULATIONS 45

Opening BC General
Interface

Frozen Rotor
Interface

Structured Mesh
(rot. Dom.)
Wall
a BC C
(no slip)

Unstructured
Mesh
(rot. Dom.)
Periodicity
BC

Unstructured Mesh
(stat. Dom.)

Figure 3.21: Single rotor CFD setup.

46 3. AERODYNAMICS

fully developed and the results deviate from the reality. However, a large do-
main uses more working memory and causes more computational time. Thus
it is necessary to find an appropriate domain size. This is done by running
simulations with different domain sizes and comparing the outputs.
The result for varying the domain height above the blade is shown in Fig-
ure 3.22(l). It shows that for a height more than 200mm the influence of the
boundary condition vanishes and the results converge to a constant value.
Thus the error is small if the domain height above the blade is set to this
value.
The same technique is used for all the different domain geometries and also
for the resolution of the different meshes. As a further example the result for
the structured domain around the blade is shown in Figure 3.22(r).

0.14 0.14

0.138
0.138
0.136

0.136
Thrust [N]

Thrust [N]

0.134

0.132
0.134

0.13
0.132
0.128

0.13 0.126
50 100 150 200 0 100 200 300 400 500 600
Domain Height [mm] N° of Elements in 1000

Figure 3.22: Domain (l) and mesh resolution (r) investigation results.

Boundary Conditions
The boundary conditions (BCs) used for the simulations are also shown in
Figure 3.21. As mentioned above only the have of the real fluid domain is
modeled and is connected by periodic BCs on the front surfaces. There is
no inlet or outlet BC in the hover simulations, instead opening BC are used
at the hull of the fluid domain. The opening BC allows the fluid to pass the
boundary from both sides, therefore the air is able to stream into and out of
the domain. All the parts of the rotor are modeled as non penetrating walls
with no slip condition.
Two types of fluid-fluid interfaces are used between the domains. Between the
structured and unstructured rotational domains a general interface is used
allowing the air to pass freely between the two domains.
The interface between the rotational and stationary domain is modeled as
3.3. SIMULATIONS 47

a frozen rotor interface [19]. The frozen rotor interface treats the flow from
one domain to the next by changing the frame of reference while maintaining
the relative position of the domain. This simplifies the setup and has the
advantages of being robust and using less computer resources than a tran-
sient simulation. As a disadvantage this interface model is not able to resolve
different rotor blade positions, but for the performance analysis this is not
required.

Turbulence Model and Validation

An important part of the CFD setup is the turbulence model. It is usually
not possible to resolve the fluid domain finely enough to cover all the ex-
isting length scales for a Direct Navier Stokes (DNS) simulation. Therefore
a turbulence model is needed. There exist many different turbulence mod-
els with advantages and disadvantages. Using an inappropriate turbulence
model may result in inaccurate results; therefore it is necessary to study the
effect of the turbulence models.
In the case of the low Reynolds number regime the first question arising
is, if a turbulence model is needed at all. As seen before, at low Reynolds
number the flow is laminar, which implies that no turbulence model might
be needed. Nevertheless, most probably transition to turbulence will occur,
an effect which is still not resolved properly in CFD. There exist different
turbulence models to cover transitional effects, mostly implying more com-
putational effort, thus the benefit of the use of a transition model has to
be evaluated. In order to cover the influence of the turbulence model, the
following four models are tested:

• Laminar
In this model the flow is assumed to be laminar and no turbulence
model is needed.

• k−
This is the most prominent two equation turbulence model. It solves
additional equations for the turbulence kinetic energy k and the tur-
bulence eddy dissipation . For general purpose simulations, the model
offers a good compromise in terms of accuracy and robustness. Cor-
responding to [19] there are applications for which the model may not
suit, one of them are flows with boundary separation which may or will
be the case in the rotor simulation. Hence it might be possible that
this model is not suitable for this application.
48 3. AERODYNAMICS

• k−ω
Instead of the turbulence eddy dissipation , this two equation turbu-
lence model solves a transport equation for the spatial turbulent fre-
quency ω. One of the advantages of this two equation turbulence model
is the near wall treatment for low Reynolds number computation.

• Baseline (BSL) k − ω model with transition

The BSL model is an adaptation of the standard (Wilcox-) k − ω
model in order to overcome the problem of the sensitivity to free stream
conditions. It uses the strength of the k − ω model near the walls and a
transformed k −  model in the outer region. Additionally a transition
model is used in order to better predict the transition from laminar to
turbulent.

The simulation results are compared to the experimental data obtained

on the test bench and summarized in Table 3.2. In general all the models cor-
relate well with the measurements and some of the deviation can be explained
by the slightly different setup of the experiment and simulation. Nevertheless
the evaluation shows the strong influence of the turbulence models. Even if
it was not expected, the worst performance shows the BSL turbulence model
with transition criteria, which underestimates the thrust and overestimates
the torque in all the cases. Furthermore in some cases the BSL model has
convergence problem which causes the very high torque value for 12◦ pitch
angle and 7000RPM.
The performance of the other three models is similar. The good performance
of the laminar model reflects the expectation that a turbulence model is not
absolutely necessary and the flow is close to laminar. However similar to all
the models it overestimates the torque, which is surprisingly represented the
best by the k −  turbulence model. It looks like the disadvantages of this
model do not appear in this setup. Since this model shows good correlation
in all the tested configurations, it is chosen as the standard turbulence model
for the investigations.

Computation
The CFD simulations are run on the Brutus cluster at the ETHZ [4]. This
allows running it in parallel configuration and allocating enough working
memory for the high resolution needed. In general the computations are run
on four processors for about 40 hours. This long computational time and the
costly pre- and post-processing limits the application of the CFD simulations
and an extensive parameter study with this method is hard to perform.
3.3. SIMULATIONS 49

Table 3.2: Comparison of the turbulence models.

Simulation Model Thrust[N] Torque[Nm]

12◦ , 5000RPM
Laminar 0.136 0.00174
k− 0.135 0.00153
k−ω 0.132 0.00170
BSL 0.123 0.00191
Measurement 0.151 0.0015

12◦ , 7000RPM
Laminar 0.273 0.00337
k− 0.266 0.00297
k−ω 0.261 0.00319
BSL 0.273 0.00678
Measurement 0.298 0.0027

16◦ , 5000RPM
Laminar 0.206 0.00299
k− 0.203 0.00248
k−ω 0.202 0.00288
BSL 0.184 0.00299
Measurement 0.218 0.0023

16◦ , 7000RPM
Laminar 0.399 0.00561
k− 0.400 0.00483
k−ω 0.393 0.00544
BSL 0.334 0.00630
Measurement 0.425 0.0042
50 3. AERODYNAMICS

Coaxial Setup
The coaxial setup is similar to the single setup. It differs only in the fact that
two rotational domains are used. Initially it was questionable if in the coaxial
configuration the Frozen Rotor interface is still feasible and the simulation
can still be run stationary. The investigations showed that it is feasible, again
with the limitation that only a global view is possible.

Typical Results
In Figure 3.23 typical results for the pressure distribution on a standard blade
with a pitch angle of θ = 16◦ and RP M = 7000 is displayed. Clearly visible
is the strong sucking region at the tip and leading edge of the upper blade
surface strongly contributing to the lift of the blade. On the other side of the
blade a pressure zone is visible, which increases the drag, but also contributes
to the lift of the blade. Furthermore the influence of the tip vortices is visible.
It shows that there are not only negative effects of the tip vortices like the
reduced angle of attack; there is also a positive impact along the blade edge,
where the accelerated flow from the lower to the upper side yields to a low
pressure zone on the upper surface.

Hub

Tip

Upper Side Lower Side

Figure 3.23: CFD results for a standard NACA0012 profile. Static pressure on
the blade surfaces

Figure 3.24 shows the downwash in the fluid domain, where the velocity
field in the stationary frame is displayed. Already after one chord length
downstream the ’vena contracta’ is reached and stays about the same in the
domain until the flow starts to disperse at the bottom of the domain, probably
3.3. SIMULATIONS 51

also affected by the non-perfect boundary condition at the fluid boundary.

Figure 3.24: CFD results for a standard NACA0012 profile. Velocity field in the
fluid domain.

3.3.4 Drive Train

As discussed in Section 3.1.4 the drive train is indispensable for a proper
propulsion design. Hence the influence of the drive train has to be included
into the simulations and is presented here.
First only the aerodynamics is considered and the rotors are simulated for a
certain rotor speed (7000RPM). After completion the calculated torque Q is
used for the input of the drive train part.
If the motor voltage Umot = Ubat umot is kept constant, a certain motor speed
ω and rotor speed Ω will settle depending on the rotor torque Q and the
gear ratio igear . This equilibrium can be calculated by using the stationary
solution of (4.29) presented in detail in Section 4.4.2

κM Umot − κM κE igear Ω c Q k Q Ω2
− dR Ω − 2 = 0. (3.18)
igear RΩ igear ηgear

Solving (3.18) leads to the ’real’ rotor speed on the given drive train if the
the torque coefficient cQ is constant over the considered speed range. This
condition should be reasonably fulfilled due to the definition of the coeffi-
cient. In order to check this assumption the torque and thrust coefficients
52 3. AERODYNAMICS

for different rotors and speeds are plotted in Figure 3.25.

As expected the torque coefficient does not change significantly over the wide
speed spectrum as seen in the first plot. The second plot in Figure 3.25 shows
the influence of the rotor speed on the thrust coefficient cT . Similar to the
torque coefficient the value stays more or less constant, but it is affected more
due to the change in the Reynolds number at higher speed leading to an in-
creased thrust. Anyway it is good enough to calculate the resulting thrust
as a function of the gear ratio, if the initially chosen rotor speed is not too
far away from the final one.

−3
x 10
4 θ = 12°
0.018
θ = 14°
3.5 θ = 16°
θ = 18° 0.016
3
0.014
cQ [−]

cT [−]

2.5
0.012
2 θ = 12°
θ = 14°
0.01
1.5 θ = 16°
θ = 18°
1 0.008
3000 4000 5000 6000 7000 3000 4000 5000 6000 7000
RPM [−] RPM [−]

Figure 3.25: Dependency of the thrust and torque coefficients on the rotational
speed.

3.4 Results
In this section the obtained results for the various investigated rotor config-
urations are presented. The investigations concentrate on the different blade
parameters that can be varied in order to influence the rotor performance in
hover. As mentioned in the previous sections the focus is not only on the
pure aerodynamics, in fact the whole drive train of muFly is incorporated in
the investigations in order to have a realistic scenario as on the helicopter. In
all the single rotor simulations a desired thrust of Tdes = 0.4N is chosen, with
the idea of lifting a 50g helicopter in coaxial configuration. Thereby the gear
ratio igear is adapted for every simulation in order to have the most efficient
propulsion setup.
If not mentioned differently a standard blade is chosen for the investigations
and only the considered parameter is changed. The standard blade has a
3.4. RESULTS 53

NACA4412 profile, radius R = 0.06m, chord length 0.02m and no twist or

taper. The Reynolds numbers on all the blades is below Re = 70, 000.

3.4.1 Number of Blades

An important design factor to check is the number of blades of the rotor.
Using more blades increases the thrust and reduces the needed rotational
speed and might lead to a minimized power consumption. However, the drag
torque increases. In Figure 3.26 the results for rotors with different numbers
of rotor blades are shown. The results are very interesting: up to a pitch
angle of 16◦ the two bladed rotor shows clearly the best performance, while
at higher pitch angles the three and four bladed rotors start to outperform
and even have a slightly higher efficiency. The five bladed rotor shows the
lowest efficiency over the whole pitch angle spectrum and it looks like there
exists an optimum of number of blades.
Additionally to the purely aerodynamic effects the mechanical design has to
be considered. Even if the efficiency is not the highest for the two bladed
rotor, the weight is less and the mechanical setup for cyclic pitch is easier.
Hence it is a good choice to use a two bladed rotor for the design. The same
conclusion was also found in [23].

0.065

0.06
Thrust/Electrical Power [N/W]

0.055

0.05

0.045

0.04 2 blades
3 blades
4 blades
0.035
5 blades

0.03
6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.26: Influence of the number of rotor blades on the rotor performance.

3.4.2 Blade Radius and Chord Length

The aspired specifications for the rotor diameter are very tight, originally
limited to a diameter of 10cm. Regardless the influence of the diameter is
54 3. AERODYNAMICS

investigated in order to see if it would be beneficial to extend this limitation.

In Figure 3.27 the performance of rotors with different blade radii are plotted.
As it can be seen a larger radius is clearly beneficial. This is reasonable since
the larger rotor needs less rotational speed in order to achieve the desired
thrust value which reduces the mechanical power. Additionally the larger
rotor has a lower disk loading leading to a smaller reduced angle of attack
increasing the aerodynamic efficiency. Of course the larger rotor has a slightly
increased mass and higher inertia, but with a relative difference in efficiency
of around 15% it would be beneficial to use a larger rotor.

0.08

0.075
Thrust/Electrical Power [N/W]

0.07

0.065

0.06

0.055

0.05
R50
0.045 R60
R70
0.04
R80

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.27: Influence of the rotor radius on the rotor performance.

The next parameter to choose for the rotor design is the chord length c.
Varying the chord length is, considering a constant radius, equal to changing
the aspect ratio. The results for rotors with different chord lengths are shown
in Figure 3.28. It shows that in the considered case a chord length of 0.03m
shows the best performance. The difference between the largest three chord
lengths are minimal. The plot shows that the air stream seems to separate
later for larger chords leading to a shift of the optimal pitch angle to higher
values. A reason for this is the increased Reynolds number.
Regarding the rotor design it might be more interesting to use a smaller chord
length in order to reduce the mass and inertia of the rotor and benefit from
the better dynamic performance.
3.4. RESULTS 55

0.065
c10
c15
0.06
c20
Thrust/Electrical Power [N/W]
c25
c30
0.055

0.05

0.045

0.04

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.28: Influence of the chord length on the rotor performance.

3.4.3 Blade Profile

In this section the influence of the geometry of the blade profile is investigated.
Analyzing the blade profile and determining an optimal shape is difficult.
There are hardly any limitations to the geometry design except the feasibility.
There exist thousands of different profiles with different applications and it
is hard to investigate the different airfoil features. Thus the NACA four digit
series [52] is chosen for the investigations in order to have a structure for the
airfoil features such as the maximum camber, thickness or maximum camber
position. The NACA four digit series describes the maximum camber (first
digit), the camber position (second digit) and the maximum thickness of the
profile (last two digits) in percentage of the chord length c. This allows
varying those parameters more or less independently.
In Figure 3.29 four different profiles of the NACA series are displayed. In the
upper left corner a symmetrical NACA0012 profile is shown, with a thickness
of 12% of the chord and no camber. The NACA4412 profile in the upper
right corner has a moderate camber with its maximum at 0.4c, followed
by a strongly cambered profile with the camber at the same position, the
NACA9412, and a NACA9712 with the camber at 0.7c.
Before starting with the investigation it has to be emphasized that combining
all the found trends may not lead to an optimal profile. The mutual influence
of the different parameters is strong and may lead to a different flow around
the airfoil.
56 3. AERODYNAMICS

Figure 3.29: Different airfoils of the NACA four digit series.

Thickness

The first parameter is the thickness of the profile. In Figure 3.30 the results
for profiles with thickness from 6% to 12% of the chord are shown. It is
clearly visible that a thinner profile works better in this Reynolds number
regime. The thinnest profile outperforms the thickest profile more than 10%.
The thickness of the blade is directly correllated with the adverse pressure
gradient. On a thicker blade the gradient is stronger and the flow detaches
earlier yielding a worse performance. The thickness is limited by the struc-
tural strength, thus the 6% with a maximum thickness of 1.6mm might be at
the limit for the design. But as a rule of thumb it is beneficial to have a thin
profile. This explains the acceptable performance of commercially available
blades as they are used on the final prototype of mufly.

0.065

0.06
Thrust/Electrical Power [N/W]

0.055

0.05

0.045
naca4406
naca4409
0.04 naca4412
naca4415

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.30: Influence of the blade thickness on the rotor performance.

3.4. RESULTS 57

Maximum camber
An important parameter is the maximum camber of the profile. As seen in
the previous sections the camber accelerates the flow and creates more lift on
the profile. In Figure 3.31 the influence of the maximum camber is shown. A
moderate camber of around 4% of the chord is found to be the most suitable
in this configuration. For lower camber not enough lift is produced, for higher
camber at this thickness the pressure gradient is to strong leading to a worse
performance.

0.065

0.06
Thrust/Electrical Power [N/W]

0.055

0.05

0.045
naca0012
naca2412
0.04 naca4412
naca9412

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.31: Influence of the maximum camber on the rotor performance.

Maximum camber position

Beside the value of the camber there exists also the possibility to change the
position of the camber. It is interesting to see if moving the camber to the
trailing edge of the airfoil beneficially influences the efficiency. The flow is
influenced differently by the position of the camber and thus the performance
might change strongly due to flow separation.
In the tested series with four different maximal camber position there is
no advantage or disadvantage observable. All the blades were performing
similarly.

Various airfoils
At last it is interesting to see if there is a certain geometry that performs
best. In order to check this, 247 airfoils from a data base are simulated
58 3. AERODYNAMICS

0.065

0.06
Thrust/Electrical Power [N/W]

0.055

0.05

0.045
naca4212
naca4412
0.04 naca4712
naca4912

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.32: Influence of the camber position on the rotor performance.

with the BEMT code and the maximum efficiency over the whole pitch angle
spectrum is calculated. The results are summarized in Table 3.3, where the
four best and the four worst rotors are displayed.

Table 3.3: Performance of rotors with different blade profiles.

N◦ Airfoil Thrust/El. Power [N/W] El. Power [W]
1 goe394 0.0663 6.043
2 goe342 0.0660 6.062
3 gm15sm 0.0657 6.085
4 goe142 0.0656 6.096
.. .. .. ..
. . . .
244 lnv109a hL 0.0503 7.954
245 NACA25516 0.0480 8.350
246 n642-015 0.0436 9.171
247 NACA6415 0.0415 9.628

It shows the strong influence of the blade profile. The rotor with the
NACA6415 uses over 3.5W more for the desired 0.4N of thrust than the best
performing rotor with a goe394 profile. The four most efficient rotor profiles
are shown in Figure 3.33. All the profiles have similar geometry. It looks like
thin profiles with a strong camber up to 10% perform best. This result shows
the strong coupling of the geometry parameters. While for a thin profile a
strong camber is beneficial, the camber has to be reduced if a thicker blade
is needed as it was found in Figure 3.31.
In Figure 3.34 the four worst blade profiles are shown. Here it can be
seen that the thickness of the profile is dominating and is yielding in a bad
3.4. RESULTS 59

goe394 goe342

gm15sm goe142

Figure 3.33: The four best airfoil profiles found.

performance of the rotor. Beside the thickness the geometry of the four
airfoils look quite different with different camber (even symmetrical) and
different maximum camber position. In this Reynolds number regime it is
definitely beneficial to use thin blades. However for the application on the
helicopter also the structural strength has to be considered.

NACA6415 n642-015

NACA25516 lnv109a_hL

Figure 3.34: The four worst airfoil profiles found.

3.4.4 Taper and Twist

Other features often used on helicopter rotors are twist and taper. The idea
of those two design features is to adapt the pitch angle and chord length to
the rotational velocity at the position of the blade in order to have a better
lift distribution. Therefore both quantities are reduced from the hub to the
tip.
In Figure 3.35 the influence of linear taper on the standard blade is dis-
played. All of the blades have the same projected surface and the labeling
TAPER2515 stands for a blade with a chord length chub = 0.025m at the hub
and ctip = 0.015m at the rotor tip. The results show that tapering leads to a
better performance of the blade. In fact the efficiency of the strongly tapered
60 3. AERODYNAMICS

blade TAPER3505 is more than 10% higher than for the straight blade.

0.07

0.065
Thrust/Electrical Power [N/W]

0.06

0.055

0.05

0.045
straight
taper2515
0.04 taper3010
taper3505

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.35: Influence of the taper on the rotor performance.

The influence of twist is displayed in Figure 3.36. Similar to the taper,

a strong twist is beneficial. However, the pitch angle of the blade has to be
adjusted correctly. The optimal point is shifting to smaller pitch angles as
stronger as the twist gets. This makes sense since with higher pitch angle
and stronger pitch the blade starts to stall earlier.

0.07

0.065
Thrust/Electrical Power [N/W]

0.06

0.055

0.05

0.045 straight
twist5
twist10
0.04
twist15

6 8 10 12 14 16 18 20
Pitch Angle θ [°]

Figure 3.36: Influence of the twist on the rotor performance.

3.4. RESULTS 61

3.4.5 Tip shape

Another possibility to influence the aerodynamics is blade tip design. It
is interesting to investigate if a different blade tip strongly influences the
flow and leads to a reduction of the tip losses. Two different tip shapes are
investigated: one is similar to the commercially available rotor blades of the
Walkera helicopter. The other one is a custom design where the general idea
is to taper the tip. The two tip shapes are shown in Figure 3.37.

Tip 1 Tip 2

Figure 3.37: The two tested blades with different tip shape.

In Figure 3.38 the results for the two different tip shapes compared to the
standard blade are shown. In this configuration the two tip shapes do not
positively influence the rotor efficiency. It looks like it is more efficient to use
as much blade surface as possible at the tip in order to increase the thrust
and reduce the rotational speed. A strong reduction of the tip losses due to
the different shape is not found in the two studied cases. Of course there
might exist configurations where a different tip shape increases the efficiency.
However, the results show that it is not simple to find an optimal tip shape
and thorough investigations are needed.

0.046
Tip1
0.044 Tip2
Standard
Thrust/Electrical Power [N/W]

0.042

0.04

0.038

0.036

0.034

0.032
12 13 14 15 16 17 18 19 20
Pitch Angle θ [°]

Figure 3.38: Influence of the two different tip shapes on the rotor performance.
62 3. AERODYNAMICS

3.4.6 Lift-enhancing Devices

Additionally to the previously investigated design features there exist further
possibilities to influence the rotor aerodynamics. In the area of fixed wings,
those features are often named lift-enhancing devices. In this thesis two types
of lift-enhancing devices are investigated: winglets1 and Gurney flaps.

Winglets
Winglets are often used on fixed wing airplanes with the idea of increasing
the aspect ratio without increasing the wing span, in order to reduce the
induced drag. The question is if this effect can also be used on the micro
helicopter rotor.
In Figure 3.39 three different versions of winglets are displayed and simulated
in CFD.
The results are plotted in Figure 3.40. The plot shows that in two of the
three cases the winglets yield to a better performance of the rotors. Only
the Winglet1 with winglets on small winglets on the upper and lower blade
surface does not outperform the standard blade. Additionally, the plot shows
the stronger influence of the winglets for higher pitch angle and thus higher
tip losses. While the efficiency of the standard blade starts to decrease after
a pitch angle of θ = 18◦ , the maximum efficiency of the two superior blades
with winglets is reached at more than 20◦ of pitch. Although the efficiency
is increased in this case, the application of winglets has to be tested in flight
in order to avoid other unforeseen effects.

Gurney Flaps
Another lift-enhancing device is the Gurney flap found on various objects such
as racing cars or airplanes. The Gurney flap is a very small flap, just a few
percent of the total chord, at the trailing edge of the wing virtually increasing
the camber. An illustration of a Gurney flap is shown in Figure 3.41.
The Gurney flap has a strong effect on the aerodynamics as it can be
seen in the result plot in Figure 3.42. Three blades with a NACA0012 and
different Gurney flap lengths, 1%, 3% and 5% of the chord length are shown.
In this setup the gurney flaps lead to an enhancement of the efficiency in all
cases, but there exists a maximum for a Gurney flap with a length of around
0.01c. The improvement in this setup is nearly 20%. However, as for all
1 Here the winglets are counted to the lift-enhancing devices, even if their prior goal is

to reduce the tip losses.

3.4. RESULTS 63

Winglet 1

Winglet 2

Winglet 3
Figure 3.39: Three different tested versions for winglets on the rotor.

0.048

0.046
Thrust/Electrical Power [N/W]

0.044

0.042

0.04

0.038

0.036

0.034
standard
0.032 winglet1
winglet2
0.03 winglet3
0.028
12 13 14 15 16 17 18 19 20
Pitch Angle θ [°]

Figure 3.40: Influence of winglet-like blade ends on the performance.

64 3. AERODYNAMICS

Gurney flap

Figure 3.41: Illustration of a gurney flap at the trailing edge of the airfoil.

the tested parameters the use of a Gurney flap has to be evaluated for the
individual application, since it might not be efficient in every case.

0.052

0.05
Thrust/Electrical Power [N/W]

0.048

0.046

0.044

0.042

0.04
standard
0.038 001c
003c
0.036 005c
0.034
10 11 12 13 14 15 16 17 18 19 20
Pitch Angle θ [°]

Figure 3.42: Influence of the number of Gurney flaps.

3.4.7 Coaxial Rotors

Beside the single rotor configuration it is interesting to study the influence of
the coaxial setup on the rotors. The most important question is how much
the losses are on the lower rotor and if it is possible to minimize them, for
example by changing the distance between the two rotors. But it is also
interesting to investigate, if there is an effect on the upper rotor. All those
investigations are important for the dynamic model in Chapter 4.

Influences on the Rotors

In Figure 3.43 the thrust and torque distributions on the rotors for a coaxial
and single setup are shown (NACA0012,θ = 14◦ , R = 0.06m, c = 0.02m) . As
expected the thrust of the lower rotor in the coaxial setup is much less than
the thrust of the upper rotor. The absolute difference enlarges with higher
rotational speed, where the inflow velocity gets higher. Interesting is the fact
3.4. RESULTS 65

that in this configuration the thrust of the lower rotor is about 55% of the
thrust of the upper rotor for all the rotational speeds. The same behavior is
found for the torque value, which is around 20% less on the lower rotor than
on the upper rotor. This means that the two aerodynamic coefficients cT,dw
and cQdw are close to constant.

0.35 4
Coax Down Coax Down
0.3 Coax Up 3.5 Coax Up
Single Single
0.25 3

Torque [mNm]
Thrust [N]

0.2 2.5

0.15 2

0.1 1.5

0.05 1

0 0.5
3000 4000 5000 6000 7000 3000 4000 5000 6000 7000
Rotational Speed [rev/min] Rotational Speed [rev/min]
−3 −3
x 10 x 10
12 3
RPM Up = RPM Dw RPM Up = RPM Dw
10 2.5

8
2
cQ,dw [−]
cT,dw [−]

6
1.5
4
1
2
0.5
0

0
−2
3000 4000 5000 6000 7000 3000 4000 5000 6000 7000
RPM Dw [rev/min] RPM Dw [rev/min]

Figure 3.43: (up) Differences in thrust and torque on coaxial rotor configuration
and single rotor configuration. (dw) Corresponding aerodynamic coefficients of the
lower rotor.

Another interesting fact is the slightly higher thrust and torque on the
upper rotor compared to the single setup. The rotor wake of the lower ro-
tor positively influences the performance of the upper rotor due to more
additional ’clean air’ and reduction of swirl, this is also found on full scale
helicopters [34]. However, the effect is small and an assumption that the
upper rotor is not affected by the lower one is justifiable.
Mounted on the helicopter it is necessary to have the same torque on the lower
and upper rotor in order to avoid yawing of the helicopter fuselage. Therefore
the lower rotor has to be driven at a higher rotational speed, higher pitch
angle or has to be modified. If the lower and upper rotor have the same
66 3. AERODYNAMICS

setup (as on the muFly helicopter) the rotational speed has to be increased.
Determining the needed rotational speed is not easy since the aerodynamic
coefficients are not constant anymore. In Figure 3.44 the influence on the
aerodynamic coefficients of the lower rotor for different rotational speeds of
the two rotors is displayed. As expected the downwash of the upper rotor
strongly affects the lower rotor. If the lower rotor speed is kept constant, the
thrust and torque coefficients decrease strongly with higher velocity of the
upper rotor. Even a stronger effect is visible if the upper rotor velocity is kept
constant and the coefficients are plotted against the speed of the lower rotor.
For a low rotor speed the influence of the upper rotor is strong enough to
impose a negative thrust coefficient. Only with higher velocity the influence
vanishes and the coefficients start to increase.
−3 −3
x 10 x 10
12 3
RPM Dw = 6000 RPM Dw = 6000
10 2.5

8
2
cQ,dw [−]
[−]

6
1.5
T,dw

4
c

1
2
0.5
0

0
−2
3000 4000 5000 6000 7000 3000 4000 5000 6000 7000
RPM Up [rev/min] RPM Up [rev/min]
−3 −3
x 10 x 10
12 3
RPM Up = 6000 RPM Up = 6000
10 2.5

8
2
cQ,dw [−]
cT,dw [−]

6
1.5
4
1
2
0.5
0

0
−2
3000 4000 5000 6000 7000 3000 4000 5000 6000 7000
RPM Dw [rev/min] RPM Dw [rev/min]

Figure 3.44: Influences of the thrust coefficient of the lower rotor for different
working conditions.

In order to estimate the thrust losses on the lower rotor in flight condition
an example with the above data is given. Considering a rotational speed of
RP M = 6000 of the upper rotor, a rotor speed of RP M ≈ 6650 is required
for the lower rotor in order to have equal torque. This leads to a lower
3.4. RESULTS 67

thrust of the lower rotor of around 23% compared to the upper rotor. This
is additionally coupled with a higher mechanical power consumption of the
lower rotor of around 10% due to the higher speed.
Another possibility in order to have equal torque on both rotors without
increasing the rotational speed of the lower rotor is to increase the pitch
angle of the lower rotor. In Table 3.4 the results for different pitch angles
on the lower rotor are shown. As expected less rotational speed is needed
in order to have equal torque for higher pitch angles. This leads to a lower
power consumption, but in turn also to a reduction of thrust. Therefore it
has to be decided if the goal is to have more thrust or less consumed power.
The difference in power is large for the three cases (up to 0.5W), leading to
a higher thrust to power ratio for higher pitch angle. However a stronger
twist of the lower rotor should also be considered, most probably leading in
a higher performance.

Table 3.4: Influence of different pitch angles on the lower rotor.

Angle [◦ ] Speed [rev/min] Thrust [N] Power [W] Thrust/Power [N/W]
14 6650 0.1949 2.0531 0.0950
16 5750 0.1820 1.7413 0.1042
18 5025 0.1709 1.5286 0.1118

Distance between the Rotors

On a coaxial setup it is interesting to know if the distance between the rotors

is strongly affecting the performance of the lower rotor and a better efficiency
can be achieved by separating the two rotors. Figure 3.45 displays the influ-
ence of the distance between the two rotors. Thereby the distance is varied
between 1cm and 4cm. The results show that the distance does not strongly
influence the rotor performance. This is reasonable since the downwash of
the upper rotor is fully developed after a short distance and stays constant
until the flow starts to diffuse further downstream. On a helicopter with
cyclic pitch steering a setup where the lower rotor is in the range where the
downwash is not developed yet is difficult since the blades have to be able
to flap. In the considered distances the flow is similar; the differences at
higher rotational speed are mainly due to the uncertainty of the used Vortex
Method simulation.
68 3. AERODYNAMICS

0.9 10
d10 d10
0.8 d20 9 d20
d30 d30
0.7 8
d40 d40
0.6 7

Torque [mNm]
Thrust [N]

0.5 6

0.4 5

0.3 4

0.2 3

0.1 2

0 1
4000 5000 6000 7000 8000 4000 5000 6000 7000 8000
Rotational Speed [rev/min] Rotational Speed [rev/min]

Figure 3.45: Influence of the rotor distance on the lower rotor.

3.4.8 Remarks
Different rotor parameters were investigated in this chapter. The investiga-
tions showed how difficult it is to find an optimal rotor configuration. The
main reason is the mutual influences of the parameters and the large design
space. It is close to impossible to investigate all the different configurations
and find the optimal blade. The parameter space is simply too large. How-
ever, the different trends could be shown and can be used as guidelines for
the micro helicopter rotor design.
For the use on a real helicopter additional points have to be considered such
as the behavior in forward flight, structural strength and blade flexibility.
Especially the latter is important on such a small helicopter, since hingeless
rotors are used and the flexibility of the rotor blade is needed for the steering.
Different blade materials and in-house manufacturing methods were tested
in order to emphasize appropriate rotors. Anyhow, the in-house manufactur-
ing of blades is very time consuming and good quality is difficult to achieve.
Therefore the commercially available Walkera blades are used for the model-
ing, identification and control described in the next chapters.
Chapter 4

Dynamic Model

An appropriate dynamic model is indispensable for simulations and controller

design. While there exist different other works on modeling conventional
(e.g. [65, 77]) and coaxial helicopters (e.g. [32, 38]) it is necessary to adapt
the model to the specialties of the used platform. One of those specialties
is the stabilizer bar mounted on the helicopter, which helps stabilizing the
system and strongly influences the dynamics. Further devices, such as the
electric motors and the drive train are not considered in the found works,
but are important for the designated control tasks. Therefore, a custom-
made dynamic model is developed for muFly.
The goal is to have a structured model reflecting accurately the physics of
the different subsystems for simulations, but on the other hand be as simple
as possible since it is used for the controller design later on.
The nonlinear model is based on the rigid body motions, where all the existing
external forces and moments are discussed and derived. This is followed by
defining the dynamics of the single mechanical devices such as swash plate,
electric motors or stabilizer bar to complete the model.

4.1 Coordinate Frames

Before starting to derive the dynamic system equations the coordinate frames
have to be defined. In this thesis two coordinate frames are introduced, an
inertial frame J and the body-fixed frame B. Whereas the body-fixed frame
is placed in the Center of Gravity (CoG) and moves with the helicopter, the
70 4. DYNAMIC MODEL

inertial frame is fixed to the original location of the body-fixed frame1 .

The body-fixed frame B is defined as common in aviation: the x-axis points
to the front, the y-axis over the right ’wing’ and the z-axis downwards. Sim-
ilar the inertial frame J, where the x-axis points to the geographic north N ,
the y-axis to the east E and the z-axis downwards to the earth center D. For
simplification the base vectors of the frame J are written as J e = [N, E, D]T
and the base vectors of the frame B as B e = [x, y, z]T . The definition of the
frames are shown in the schematic view of the helicopter in Figure 4.1.

QUP TUP

QDW TDW

wfus
x y
Body Frame B

z
G
N
0
E
Inertial Frame J
D

Figure 4.1: Coordinate frame convention and the forces and moments acting in
hover (back view).

Corresponding to these definitions the transformation from the inertial

frame J to the body-fixed frame B
   
x N
 y =A · E  (4.1)
BJ
z D
is given by the transformation matrix
1 Strictly speaking the frame J is an earth-fixed frame and is moving with the rotational

speed of the earth in respect to the fixed stars, but since only indoor application and short
distance are of interest this frame is assumed to be an inertial frame.
4.2. RIGID-BODY DYNAMICS 71

 
cθcψ cθsψ −sθ
ABJ =  −cφsψ + sφsθcψ cφcψ + sφsθsψ sφcθ  (4.2)
sθsψ + cφsθcψ −sφcψ + cφsθsψ cφcθ

obtained by the application of the Euler angles (cα = cos(α) and sα =

sin(α)). It is well-known that the Euler angles have a singularity for θ =
±90◦ , corresponding to a vertical pitch of the helicopter. Such extreme states
will not be required for the foreseen missions and thus the use of the Euler
angles is adequate.
Analog a correlation between the angular velocities B Ω = [p, q, r]T and the
time derivatives of the roll, pitch and yaw angles [φ̇, θ̇, ψ̇]T can be obtained

   
φ̇ p
 θ̇  = R ·  q  (4.3)
JB
ψ̇ r

using the transformation matrix


1 sφsθ sφsθ 
sθ sθ
RJB = 0 sφ −sφ  . (4.4)
0 sφ sφ
sθ sθ

4.2 Rigid-Body Dynamics

The next step is to set up the rigid body dynamics with respect to the CoG
of the helicopter. Using Newtonian Mechanics, the rigid body equation of
motion in the body-fixed frame B become

d
BF = {m B v} = m (B v̇ +B Ω ×B v) (4.5)
dt
d

BM = {B I B Ω} =B I B Ω̇ +B Ω × B I BΩ (4.6)
dt
with the body velocities B v = [u, v, w]T , the system mass m, the body
inertia tensor B I and the total external force and moment vectors B F and
2
B M . The resulting differential equation for the velocities and angular rates
in the body fixed frame are
2 In future the frame indexes are removed for simplification
72 4. DYNAMIC MODEL

     
u̇ p u
1
 v̇  = F− q × v  (4.7)
m
ẇ r w
      
ṗ p p
 q̇  = I−1 M −  q  × I  q  . (4.8)
ṙ r r

So far the equations of motion are independent of the flying platform and
can be found in various literature [65, 74, 77]. Now the platform dependent
total external force F and moment M have to be defined.

4.3 External Forces and Moments

In a stable hover position, as shown in Figure 4.1, the thrust forces from
the two rotors Tup and Tdw plus the integrated aerodynamic drag force on
the fuselage Wfus due to the down wash of the rotors equal the gravitational
force G caused by the mass of the helicopter. The moments acting on the
helicopter are the two drag torques Qup and Qdw from the counter rotating
rotors (incl. stabilizer bar), which, if unbalanced, lead to a yaw motion of
the helicopter.
In free flight, additional forces and moments result from the aerodynamic drag
due to the motion through the air, the steering moments and forces resulting
from the tilting of the tip path plane (TPP) -the plane the rotor blade tips
are running in [92]- on the lower rotor caused by a cyclic pitch input from the
swash plate [79] and on the upper rotor due to the stabilization mechanism.
At last some reaction torques occur due to the inertia of the rotors while
accelerating and decelerating, Qreact,dw and Qreact,up . Summarized, the force
and moment vectors F and M are

F = Tup + Tdw + G + Wfus , (4.9)

M = Qup + Qdw + rCup × Tup + rCdw × Tdw
+ Qreact,dw + Qreact,up + Mflap,dw + Mflap,up , (4.10)

where rCdw and rCup are the vectors from the CoG to the hub of the
lower rotor, upper rotor, respectively, and Mflap,i moments resulting from
the flapping of the blades. In a next step, the individual forces and moments
have to be determined.
4.3. EXTERNAL FORCES AND MOMENTS 73

4.3.1 Rotor Forces and Moments

The rotor thrust vector Ti and torque vector Qi we define as Ti = Ti · nT i
and Qi = Qi · nQi , with i ∈ {dw, up} for the lower and upper rotor. Hence
the vectors can be split into magnitude and direction.

Magnitude
In hover, the thrust magnitude Ti of a rotor with radius R can be defined as
seen in Section 3.1.2 as

Ti = cT i πρR4 Ω2i = cT i kT Ω2i , (4.11)

with the air density ρ, the rotational velocity Ωi and the thrust coefficient
cT i . While the air density ρ is assumed to be constant and the rotational
velocity Ωi is a function of the motor input, the thrust coefficient cT i depends
on different elements such as on the rotor setup and on the flow characteris-
tics. However, for a given rotor, the thrust coefficient can be evaluated for a
certain operation point and only possible variations from this point have to
be investigated.
During helicopter operation there exist mainly two possibilities leading to a
change of the thrust coefficient: the inflow velocity and the rotational speed.
The influence of the latter was shown in Figure 3.25 in Section 3.3.4, where
the thrust and torque coefficient of a rotor setup are plotted. Due to the
definition of those coefficients they are only little affected by the rotational
speed. Since the speed will only vary around the hover point during the mis-
sions this effect can be neglected.
The second effect is the change of the thrust coefficient due to a different
inflow velocity. This results from the movement through the air and addi-
tionally on the lower rotor due to the downwash of the upper. The effect of
the inflow velocity due to the movement in the air is simulated in CFD and
the results are displayed in Figure 4.2. In the simulation, very high climb and
descent velocities are used, in normal operation the helicopter will probably
not climb or drop faster than 0.5m/s. In this range, the change of the thrust
coefficient due to the inflow velocity is insignificant and can be assumed to
be constant.
Additionally the inflow velocity of the lower rotor is increased due to the
downwash of the upper leading to a reduction of the aerodynamic coeffi-
cients. This effect is investigated in Section 3.4.7. The results showed that
the thrust coefficient of the lower rotor is in deed strongly affected by the
upper rotor. In fact the aerodynamic coefficient are a nonlinear function of
74 4. DYNAMIC MODEL

the upper and lower rotational speed. Nevertheless, for simplicity the rota-
tional speeds of the rotors during operation are assumed to be close to the
hover point and the thrust coefficients to be constant. This assumption is
justifiable if only slow climb and descend velocities are required. Otherwise
the thrust coefficient has to be modeled as a function of the two rotor speeds.
−3
x 10
0.015 3

2.5

0.01 2

cQ [−]
cT [−]

1.5

0.005 1

0.5

0
0
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
v0 [m/s] v0 [m/s]

Figure 4.2: Influence of the inflow velocity on the thrust coefficient (CFD simu-
lation results).

The same assumptions are true for the torque coefficients cQi which are
used to describe the drag torque values Qi of the rotors and stabilizer bar

Qi = cQi πρR5 Ω2i = cQi kQ Ω2i . (4.12)

Since the helicopter controls its altitude and heading with varying the
rotor speed, the rotor has to be accelerated and decelerated causing other
torques depending on the rotor acceleration and inertia. For the case of the
lower rotor this might be negligible, but the high inertia of the stabilizer
bar, mounted on the upper rotor, causes a notable reaction torque while
accelerating the rotor. The torque vectors are assumed to act only in the
rotor axis direction with the magnitude

Qreact,i = Jrot,i Ω̇i . (4.13)

Direction
Beside the magnitude of the thrust and torques on the rotors, the direction
is important. As explained in Section 1.2 the helicopter is steered by tilting
the tip path plane (TPP) by cyclic pitch input. The direction of the thrust
4.3. EXTERNAL FORCES AND MOMENTS 75

vector ni is perpendicular to this TPP [92] and can be defined using tilt
angles around the x- and y- axis as illustrated in Figure 4.3.

Ti = Ti·nT,i

β
a
α
x
y

Figure 4.3: Visualization of the tilted thrust vector with tilt angles α and β.

Using two rotational transformations in series, the vector is expressed in

the body-fixed frame as

 
cos α sin β
nT i = sin α . (4.14)
− cos α cos β

Here it has to be mentioned that this is only true if the swash plate is
adjusted correctly to the rotor setup. If the phase shift ζ between the cyclic
pitch input and the tilting of the TPP is not compensated correctly by the
hardware setup this will result in coupled roll-pitch. This is the case for
muFly using the standard swash plate setup of the Walkera helicopter. Since
muFly is about one and a half times as heavy as the original helicopter, the
rotors have to run at much higher speed leading to another phase shift ζ.
This can be compensated in the equations by an additional rotation around
the body z-axis with an angle ∆ζ (the difference between the real phase shift
angle ζ and the initial hardware compensation angle on the swash plate) or,
as it is done on muFly, by correcting it in software on the servo input level.
The direction of the rotor torque nQi is set equal to the body z-axis, since
the torque mainly acts on the rotor axis. It has to be considered that one
rotor turns clockwise, while the other one turns counterclockwise.
76 4. DYNAMIC MODEL

   
0 0
nQup =  0  , nQdw =  0  (4.15)
1 −1

Additionally to the moments resulting from the non-alignment of the force

vectors with the CoG, further moments evolve from the flapping spring mo-
ments of the rotor blades Mflap,i . This effect of the flapping moments is well
explained in [21]. This spring moment can be simplified for one rotor as

Mx,flap = ksprg α (4.16)

My,flap = ksprg β, (4.17)

with a blade stiffness factor ksprg . Nevertheless those moments are omit-
ted in the model for simplification and compensated by a stronger tilt of the
TPP. In fact without x-y motion identification it is not possible to differ
between those two type of moments. The difference will only be visible in
long and fast forward flight.

4.3.2 Aerodynamic Drag

During operation, different aerodynamic forces and resulting moments act on
the helicopter fuselage. Drag forces due to the motion through the air occur,
but also forces caused by the downwash of the rotors. Those forces can be
modeled similarly to (3.2), as

1
Wfus,i = cD,i ρAi vi2 , (4.18)
2
with i ∈ {x, y, z}. While setting up the equations is simple, it is difficult
to find out the values for the drag coefficients cD,i and the projected surfaces
Ai . The structure of the fuselage is complex and an extensive investigation
would be necessary. In this case, it makes more sense to check the order of
magnitude of the forces by estimation.
In normal operation the helicopter will not move faster than 0.5m/s. Con-
sidering the fuselage as a cuboid with the dimensions 4x4x10cm and a drag
coefficient of cD ≈ 1.0, the maximum force is less than 0.01N and can be ne-
glected. The drag force caused by the downwash is higher due to the higher
air velocity, but not excessively since the fuselage is close to the rotor hub
where the induced velocity is smaller. Nevertheless the drag force value Wfus
4.4. DYNAMIC ELEMENTS 77

due to the downwash is considered in the model as a help for the identifica-
tion (see Chapter 5). It is assumed to be parallel to the z-axis and constant,
since the downwash is nearly the same around hover:

 
0
Wfus =  0 . (4.19)
Wfus

4.3.3 Gravity
The last force is the gravitational force G caused by the mass of the he-
licopter. This force attacks in the CoG and is always directed parallel to
the D-axis (positive direction) of the inertial frame J with the magnitude
G = mg. However, the rigid-body equations of motion are stated in the
body-fixed frame thus it has to be transformed using the transformation ma-
trix (4.2)

   
0 − sin θ
BG = ABJ ·J G = ABJ  0  = mg  cos θ sin φ  (4.20)
mg cos θ cos φ

4.4 Dynamic Elements

The last missing part in the model is the dynamics of the stabilizer bar, swash
plate and the electro motors. Those elements are responsible for the dynamic
behavior of the whole system.

4.4.1 Swash Plate and Stabilizer Bar

The swash plate is responsible for the pitch, roll and the horizontal move-
ment of the helicopter as it is explained in detail in Section 1.2. The exact
mathematical description of the cyclic pitch mechanism is rather complex.
It incorporates effects such as the phase shift or flapping of the blades. Such
complexity of the swash plate input is undesired and in fact not needed for
the model used for simulation and the controller design. For the helicopter
simulation only a description of a servo input to the tilting of the rotor TPP
is needed.
This reaction can be modeled by a first order system. Hereby, all the dy-
namics of the servos and rotor are covered by the time constant Tf,dw . The
equations for the tilting angles of the lower rotor are modeled as
78 4. DYNAMIC MODEL

1
α̇dw = (−ldw userv2 · θSPmax − αdw ),
Tf,dw
1
β̇dw = (−ldw userv1 · θSPmax − βdw ), (4.21)
Tf,dw
with the time constant Tf,dw , scaling factor ldw , maximum swash plate
tilting angle θSPmax and servo inputs userv,i .

The second dynamic element in the rotor system is the stabilizer bar
which gives cyclic pitch inputs to the upper rotor in order to stabilize the
helicopter. Its principle is explained in detail in Section 2.1.1.
The stabilizer bar following the roll/pitch movement can be modeled similar
to the swash plate as

1
η̇bar = (φ − ηbar ) ,
Tf,up
1
ζ̇bar = (θ − ζbar ) , (4.22)
Tf,up
with angles ηbar and ζbar . The tilt angles of the thrust vector in the body-
fixed frame is the difference between the two angles ηbar and ζbar scaled by
a factor lup , since we are interested in the tilt angle of the TPP and not that
of the stabilizer bar. Thus the equations for the tilting angles for the thrust
vector are

αup = lup (ηbar−φ )

βup = lup (ζbar−θ ) . (4.23)
The influence of the modeled stabilizer bar on the overall system is shown
in Figure 4.4 where the reaction of the helicopter to an initial displacement
in the roll angle is plotted. The helicopter is initially tilted 20◦ while as the
stabilizer bar is still horizontal. Immediately the system starts to recover and
after a short time period the helicopter is back in the hover position. The
dynamics is strongly influenced by the time constant of the stabilizer bar. In
the shown plot a random value of Tf,up = 0.2s is used.
At this some words on gyroscopic effects on the rotor have to be said.
One of the main idea behind changing the TPP by cyclic pitch is to avoid
high gyroscopic effects which would occur if the whole rotor would be tilted.
Therefore gyroscopic moments are neglected in the model; this would be
different on a vehicle with rigid rotors such as the quadrotor.
4.4. DYNAMIC ELEMENTS 79

20
Roll φ
Pitch θ
Yaw ψ
15

10

Angle [°]
5

−5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [s]

Figure 4.4: The reaction of the modeled helicopter to an initial roll displacement
of 20◦ with the modeled stabilizer bar.

4.4.2 Electric Motors

The motors controller used have the drawback that the motors rotational
speed is not given and motor speed control is not possible. Instead of mea-
suring the speed externally, for instance with an optical device, the motor
dynamics are included directly in the model. This additionally allows using
the model for every kind of motor without the help of any external sensor.
The dynamics equations of an electric motor can be written as [69]

dIA
L+ RΩ IA + κE ω = U, (4.24)
dt
dω
Jmot − κM IA + dR ω = −ML , (4.25)
dt
with the moment of inertia Jmot of the motor, electrical and mechanical
motor constants κE and κM , the input voltage U , the inductance L, the
armature current IA , the electrical resistance RΩ , the friction coefficient dR
and the external torque ML .
Since the inductance L is small and the change of the current usually slow
the term L dIdtA can be neglected resulting in the simplified first order system

κM U − κM κE ω
Jmot ω̇ = − dR ω − ML . (4.26)
RΩ
In the case of the helicopter, the external torque ML is the drag torque
value Q of the rotor, as shown in Eqn. (4.12), reduced by the gears. The
equations for the gearbox are
80 4. DYNAMIC MODEL

ω = Ωigear (4.27)
2
Q cQ kQ Ω
ML = = . (4.28)
igear · ηgear igear ηgear

with the gear ratio igear , the efficiency of the gear ηgear and the rotor
speed Ω.
Including the equations for the gearbox in (4.26), the final equation for the
rotor speed Ωi becomes

κM Ubat umot,i − κM κE igear Ωi cQi kQ Ω2i

Jdrive,i Ω̇i = − d R Ωi − 2 (4.29)
igear RΩ igear · ηgear

with the nominal battery voltage Ubat , the moment of inertia of the whole
drive train Jdrive and the motor input umot,i .

4.5 Model Summary

With this set of differential equations, the nonlinear model is complete. The
full model consists of 18 states and four inputs (two motors and two servos)

x = [x, y, z, u, v, w, φ, θ, ψ, p, q, r, αdw , βdw , αup , βup , Ωdw , Ωup ]T ,

u = [umot,dw , umot,up , userv1 , userv2 ]T . (4.30)

The servo and motor inputs are the PWM signals generated by the mi-
croprocessor and are scaled to

userv ∈ {−1, 1}, umot ∈ {0, 1}. (4.31)

The entire nonlinear model is shown as an overview in a block dia-

gram in Figure 4.5. The equations of the complete system are given in
the Appendix A. The nonlinear model is used for simulations and verifi-
cation of the control loop. Therefore the nonlinear plant is implemented in
Matlab R
Simulink
R
.
4.5. MODEL SUMMARY 81

Stabbar
αup ,βup
umot,up Ωup Tup p,q,r φ,θ
Motor UP Rotor UP Qup Rigid Trans- ψ
Body formations
umot,dw Ωdw Tdw u,w,v N,E,D
Motor DW Rotor DW Motion
Qdw

αdw ,βdw G Wfus

userv1
userv2
Swash Gravity Fuselage
Plate Drag

Figure 4.5: muFly dynamic model block diagram.

82 4. DYNAMIC MODEL
Chapter 5

Parameter Identification

Before the model can be used for controller design and simulations, the miss-
ing system parameters have to be identified and adjusted to the real heli-
copter. A good nonlinear model without appropriate parameters is worthless.
The identification process is a non-trivial task especially since most of the
parameters are coupled. In order to minimize the complexity of the identi-
fication on real flight data, it is necessary to measure or estimate as many
parameters as possible on the ground. The measurement on the flight data
is then used to identify the last parameters and fit the simulation data to the
sensor data, by slightly varying the beforehand estimated parameters.
The parameter identification process used for muFly can be split in three
categories:

• Identification on the ground

• Identification on test benches
• Identification on real flight data

Those three categories are the subjects of the next sections.

5.1 Identification on the Ground

The identification on the ground includes mainly mechanical parameters and
it constitutes the simplest part of the identification process. Nevertheless, it
includes important parameters such as the body inertias or the mass. While
the mass is easy to measure on a scale, the body inertias are hard to identify
and the CAD data is used as estimation. For simplification, the inertia tensor

83
84 5. PARAMETER IDENTIFICATION

I is assumed to have only diagonal elements. Such a simplification is feasible

as a result of the design of muFly.
In Table 5.1 all the parameters identified on the ground and the identification
method are shown. The values of the identified parameters are shown in
Table 5.3. Since the parameters identified in this process are only estimates
and do not necessary have to fit the real flight data, they are slightly adapted
in the identification process in Section 5.3.

Table 5.1: Identified parameters on the ground.

Parameter Description Method
m Mass Measured on scale
Ixx Inertia around x-axis From CAD data
Iyy Inertia around y-axis From CAD data
Izz Inertia around z-axis From CAD data
zdw Distance CoG lower rotor hub From CAD data
zup Distance CoG upper rotor hub From CAD data
ΘSP,max Maximal swash plate angle Measured
R Rotor radius Measured
Jdrive,dw Drive train inertia (down) Estimated
Jdrive,up Drive train inertia (up) Estimated
igear Gear ratio Measured
ηgear Gear efficiency Estimated
Ubat Battery nominal voltage Datasheet

5.2 Identification on Test Benches

There are some parameters that cannot be measured as easily as on the
ground. Those parameters are the aerodynamic coefficients of the rotors and
the motor constants. Thus the coefficients are determined using experimental
data.
The data for the motor parameter identification is generated by our partner
CEDRAT [5] on a motor test bench. The idea is to apply a constant voltage,
varying the external torque on the motor and measuring the rotational speed
and the consumed current. Knowing those quantities over the operational
range, the motor coefficient can be determined using the stationary solution
of the motor equations (4.24, 4.25) and the least-squares method [82]. Result
plots for the motor parameter identification are shown in Figure 5.1.
The aerodynamic coefficients cT and cQ which determine the thrust and
torque values of the rotors are identified on the rotor test bench introduced
in Section 3.2.1. The rotor blades used on the helicopter are mounted on
the test bench and the thrust and torques in single and coaxial setup are
5.2. IDENTIFICATION ON TEST BENCHES 85

3 8
Measurement
Measurement
7 Least−square
2.5 Least−Square
6

Torque [Nmm]
2
Current I [A]

1.5 4

3
1
2
0.5
1

0 0
0 2 4 6 8 0.6 0.8 1 1.2 1.4 1.6
Torque Q [Nmm] RPM [−] x 10
4

Figure 5.1: Least squares result plots for the electric motor parameter identifica-
tion.

measured. Thereby the upper rotor is assumed to be unaffected by the lower

rotor. Out of the measured data the coefficients for the lower and the upper
rotor are calculated using Equation (3.3).

Table 5.2: Identified parameters on the test benches.

Parameter Description Method
cT dw Thrust coefficient lower rotor Rotor test bench
cT up,rot Thrust coefficient upper rotor Rotor test bench
cQdw Torque coefficient lower rotor Rotor test bench
cQup,rot Torque coefficient upper rotor Rotor test bench
κE Electrical motor constant Motor test bench/Least Squares
κM Mechanical motor constant Motor test bench/Least Squares
dR Motor friction Motor test bench/Least Squares
RΩ Resistance Motor test bench/Least Squares

Even if it had been possible, the torque coefficient of the stabilizer bar
cQup,stab is not measured and only estimated instead. This parameter is iden-
tified later together with the fuselage drag Wfus to achieve hovering condition
with the recorded motor inputs for hovering. In other words, they are used
as tuning parameters to correct the errors from the measurements to fulfill
the equations in hover (index ’hov’)

Tup,hov + Tdw,hov = G + Wfus ,

Qup,rot,hov + Qup,stab,hov = Qdw,hov , (5.1)

whereas the drag torque of the upper rotor Qup is split in the rotor part
Qup,rot and stabilizer bar Qup,stab .
86 5. PARAMETER IDENTIFICATION

5.3 Identification on Flight Data

In addition to the parameter identification on the ground and on test benches,
identification on real flight data is needed. Primarily to identify the missing
dynamic parameters and secondarily to fit the simulation data to the sensor
data.
Only a few examples of the application of system identification techniques to
model-scale helicopters exist and the results are, compared to full-scale heli-
copters, limited [67]. There exist many different approaches for the parame-
ter identification by minimizing the difference between the measurement and
model output. An often used method is the linear prediction error method
(PEM) for example applied in [42], [68], [80]. This linear techniques identifies
the different entries of the state space matrices and is therefore an appropri-
ate choice for linear controller design. On the other hand the linearization
cancels out different effects such as cross coupling between the angular rates.
This is undesired in simulations where a model behavior as close as possible
to the real helicopter is wanted and thus a nonlinear identification technique
is needed. In this thesis a randomized search algorithm, the Covariance Ma-
trix Adaptation - Evolution Strategy (CMA-ES) [49], is used and presented
next.

5.3.1 Covariance Matrix Adaptation- Evolution Strat-

egy
Similar to quasi-Newton methods [70], the Covariance Matrix Adaptation -
Evolution Strategy (CMA-ES) is a second order approach estimating a posi-
tive definite matrix within an iterative procedure. It differs from the quasi-
Newton methods in the way that instead of the inverse Hessian matrix H −1
the covariance matrix C is estimated. This matrix is, on convex-quadratic
functions, closely related to the inverse Hessian H −1 . The key idea of the
CMA-ES is to adapt the covariance matrix in a manner that the probabil-
ity to reproduce successful mutation steps is increased. Randomized search
algorithms like the CMA-ES are regarded to be robust in a rugged search
landscape, which can comprise discontinuities or local optima. The CMA-ES
in particular is designed to tackle, additionally, ill-conditioned, non-separable,
nonlinear and non-convex problems in dimensions of three to one hundred
and thus it constitutes an adequate tool to identify the parameters using mul-
tiple sets of measurement data. The exact formulation of the optimization
algorithm is well described in [48]. For the practical application of the CMA-
ES it is important to set adequate initial conditions and feasible boundaries
5.3. IDENTIFICATION ON FLIGHT DATA 87

for the parameters. Otherwise the algorithm fits the simulation output to
the measurement data while neglecting the physics.

5.3.2 Data Generation and Application

For recording the flight data, the helicopter is steered by a pilot. In order to
cover as much frequency bandwidth as possible, a chirp signal is generated
and superimposed on the pilot input. However, the helicopter is not ade-
quately controllable in open loop by a pilot. Therefore, a simple Proportional
Integral Derivative (PID) controller is used to control attitude and help the
pilot. Effectively, the pilot controls the set point values of the PID controller.
The controller is not problematic for the parameter identification, since the
actuator signals are recorded and sent together with the sensor data using
the bluetooth module. Hence the identification of the system is independent
of the controller.
As a visualization, the identification process is shown in Figure 5.2. It shows
the reference value input r(t) of the pilot as the input of the PID controller.
The output of the controller u(t), the motor and servo inputs, is sent to the
helicopter and the nonlinear model, where a new output ŷ(t) is predicted.
This output is compared to the respective sensor measurement y(t) and the
error  is calculated and minimized by adjusting the parameters Θ using the
CMA-ES.

r e u y ε(θ)
PID muFly

Nonlinear ŷ
Model

θ
Parameter θ
Adaptation
(CMA-ES)

Minimizing ε

Figure 5.2: The identification process block diagramm.

88 5. PARAMETER IDENTIFICATION

5.3.3 Results
Due to the problems with the downlooking camera, the data for the trans-
lational motions could not be recorded and thus they are not included in
the identification process. Therefore the model is reduced by the horizontal
linear motion states, resulting in a state-space system of 14 states.
Furthermore, the identification is split into two subsystems ’heave-yaw’ and
’pitch-roll’ corresponding to the motor and servo inputs. This simplification
presumes that a pitch and roll movement during the heave-yaw data gen-
eration is negligible and vice versa for the pitch-roll identification. Those
prerequisites are fulfilled due to the controller support.
For both identification processes two datasets are generated. Concerning the
heave-yaw identification, first the lower motor input is kept constant and the
upper motor is excited, which is leading to a yaw movement of the helicopter
(Figure 5.3). In the second data set, the excitation is on the heave controller
which produces an output as presented in Figure 5.4. Those two datasets are
used simultaneously for the identification in the CMA-ES code in order to
allow a correct adjustment of the parameters since most of them act on the
yaw and heave dynamics. In Figure 5.3 and 5.4 the comparison between the
measurement and the identified model output is shown.
Similar to the heave-yaw the roll and pitch motions are coupled due to
gyroscopic effects and thus identified simultaneously. As mentioned in the
last section, here the input is on the roll and pitch PID controllers. For the
identification the data of the two gyros, the two angular speeds (p, q), and the
estimated angles (φ, θ) by the IMU are used. The servo inputs, the resulting
measured angles and the output of the identification are shown in Figure 5.5
and 5.6.
The results show a good correlation between the nonlinear model and
the experimental data, confirming the good quality of the nonlinear model.
It mainly deviates in the amplitude, which does not impose a problem in
closed-loop operation, since it is compensated by the gain of the controller.
A summary of all the identified parameters is shown in Table 5.3.
5.3. IDENTIFICATION ON FLIGHT DATA 89

1
lower rotor motor
upper rotor motor
0.95

0.9
[−]
0.85
Motor PPM

0.8

0.75

0.7

0.65

0.6
246 246.5 247 247.5 248 248.5 249 249.5 250
Time[s]

−1.3

−1.35

−1.4
Altitude z [m]

−1.45

−1.5

−1.55

−1.6 Experiment
Model
−1.65
246 246.5 247 247.5 248 248.5 249 249.5 250
Time[s]

10
Experiment
8 Model

6
Yaw rate r [rad/s]

−2

−4

−6

−8
246 246.5 247 247.5 248 248.5 249 249.5 250
Time[s]

Figure 5.3: CMA-ES identification result plots for an actuation in yaw. (up)
Motor input signals. (mid) Ultra sonic altitude measurement and model. (dw)
IMU yaw rate measurement and model.
90 5. PARAMETER IDENTIFICATION

0.95

0.9

0.85
[−]

0.8
Motor PPM

0.75

0.7

0.65

0.6 lower rotor motor

upper rotor motor
0.55
55 56 57 58 59 60 61 62 63 64 65
Time[s]

−0.8

−0.85

−0.9

−0.95
Altitude z [m]

−1

−1.05

−1.1

−1.15

−1.2 Experiment
Model
−1.25
55 56 57 58 59 60 61 62 63 64 65
Time[s]

4
Experiment
3 Model

2
Yaw rate r [rad/s]

−1

−2

−3

−4
55 56 57 58 59 60 61 62 63 64 65
Time[s]

Figure 5.4: CMA-ES identification result plots for an actuation in heave. (up)
Motor input signals. (mid) Ultrasonic altitude measurement and model. (dw) IMU
yaw rate measurement and model.
5.3. IDENTIFICATION ON FLIGHT DATA 91

0.6
Servo 1
Servo 2
0.4

0.2
Servo PPM [−]

−0.2

−0.4

−0.6

−0.8
25.5 26 26.5 27 27.5 28 28.5 29 29.5 30
Time[s]

0.15
Experiment
Model
0.1

0.05
Roll angle φ [rad]

−0.05

−0.1

−0.15

−0.2
25.5 26 26.5 27 27.5 28 28.5 29 29.5 30
Time[s]

Figure 5.5: CMA-ES identification result plots for an actuation in roll. (up) Servo
input signals. (dw) IMU roll angle measurement and model.
92 5. PARAMETER IDENTIFICATION

0.4
Servo 1
Servo 2
0.3

0.2
Servo PPM [−]

0.1

−0.1

−0.2

−0.3

−0.4
157.5 158 158.5 159 159.5 160 160.5 161 161.5 162
Time[s]

0.15
Experiment
Model
0.1
Pitch angle θ [rad]

0.05

−0.05

−0.1

−0.15

−0.2
157.5 158 158.5 159 159.5 160 160.5 161 161.5 162
Time[s]

Figure 5.6: CMA-ES identification result plots for an actuation in pitch. (up)
Servo input signals. (dw) IMU pitch angle measurement and model.
5.3. IDENTIFICATION ON FLIGHT DATA 93

Table 5.3: Identified parameters.

Parameter Description Value Unit
m Mass 0.095 kg
Ubat Battery nominal voltage 7.2 V
Ixx Inertia around x-axis 1.12e−4 kg m2
Iyy Inertia around y-axis 1.43e−4 kg m2
Izz Inertia around z-axis 2.66e−5 kg m2
zdw Distance CoG lower rotor hub −0.051 m
zup Distance CoG upper rotor hub −0.091 m
ΘSP,max Maximal swash plate angle 15 ◦

R Rotor radius 0.0875 m

cT dw Thrust coefficient lower rotor 0.0117 -
cT up Thrust coefficient upper rotor 0.0138 -
cQdw Torque coefficient lower rotor 0.0018 -
cQup Torque coefficient upper rotor 0.0025 -
Jdrive,dw Drive train inertia (down) 7.04e−6 kg m2
Jdrive,up Drive train inertia (up) 2.11e−5 kg m2
κE Electrical motor constant 0.0045 V−1 s−1
κM Mechanical motor constant 0.0035 Nm A−1
dR Motor friction 5.2107e−7 Nm s
RΩ Resistance 1.3811 Ω
igear Gear ratio 1.5 -
ηgear Gear efficiency 0.84 -
Whub Drag force on the fuselage 0.009 N
Tf,dw Following time upper rotor 0.001 s
Tf,up Following time upper rotor 0.24 s
ldw Linkage factor upper rotor 0.77 -
lup Linkage factor lower rotor 0.48 -
94 5. PARAMETER IDENTIFICATION
Chapter 6

Control

The final goal of the helicopter is to execute autonomous missions, thus the
control of the vehicle takes an important part. First a tight feedback control
is needed on the low-level in order to stabilize the system and allowing to
follow position references. Those references have to be generated by a high-
level controller depending on the state of the helicopter and the user input.
While there exist a lot of works on helicopter control, predominately full scale
helicopters, there are almost no works existing for helicopters in the size of
muFly. The dynamics of such a small system is much faster [66] compared to
the often used model helicopter with diameters around one meter or more.
Additionally the system suffers from low quality sensors and actuators as well
as from the low processing power available. Those facts limits the low-level
control and even more the high-level control. All those elements make the
control of such a helicopter a challenging task.

6.1 Low-Level Control

The low-level control is responsible for the stabilization of the attitude and
altitude of the helicopter and to follow the position references predetermined
by the high-level controller. The first question arising is which kind of control
technique to use. There exist many different methodologies and a lot of them
were tested on helicopters.
One approach is to use adaptive control where no physical model is needed.
Existing examples on helicopters are e.g. fuzzy logic adaptive control [88] or
adaptive neural networks [59]. The results seem to be satisfying, but in the
case of the helicopter the question arises if a controller based on the physical
model would not fit better than such a black box approach. Especially since

95
96 6. CONTROL

the system dynamics is already known and the model available. It seems
more reasonable that a controller based on the physics would show a better
performance.
The next decision is whether to go for a linear or a nonlinear technique.
Nonlinear approaches used on rotary wings platforms are for example back-
stepping and sliding mode control used for example on the OS-4 quadro-
tor [28]. While nonlinear controllers are in some cases indispensable, the
computational effort is typically much higher than for linear controllers. Ad-
ditionally, linearization offers well known analysis and design tools and allows
a structured controller design. Given that the the missions foreseen for mu-
Fly do not require high dynamics and the helicopter will work mainly around
the hover point, the decision was made to use a linear model-based control
approach1 . There exist various linear model-based control approaches, but
often two types of techniques are mentioned: optimal and robust control.
Optimal control deals with the problem of finding a control law for a given
system such that a certain optimality criterion is achieved. An optimal con-
trol problem includes a cost functional which is a function of state and control
variables. The most known optimal control problem is the linear quadratic
(LQ) control [60] used on a helicopter for example in [94].
The robust control aims to achieve robust performance and stability in the
presence of bounded modeling errors. The border between the optimal con-
trol and the robust control is not sharp since it can be shown that the linear
quadratic Gaussion (LQG) method is a special case of the robust H2 con-
trol [93]. The H2 method is closely related with the H∞ method, they differ
only in there minimization function. While the H2 method minimizes the
Lebesque 2-norm, the H∞ method uses the Hardy ∞-norm instead. In prac-
tice the H2 control is not found that often, unlike the H∞ method. A strong
interest exists in the H∞ -loop shaping method due to its robustness and
structured design method. In [89] a comprehensive overview on the works
on H∞ -controllers in the field of helicopters during the last 20 years is given,
showing the strength of the design method. The PhD theses [68] and [42]
showed that this control approach, extended by gain scheduling and feedback
linearization, can even be used for flying highly nonlinear acrobatic maneu-
vers with airplanes and model helicopters. A work on the control of a coax-
ial micro helicopter is [90] where PID control is combined with H∞ -control
techniques. The goal in this thesis is to use different multiple-input multiple-
output (MIMO) H∞ -controllers for an accurate control of the helicopter.

1 If needed the linear controller can be extended to nonlinear controllers by methods

such as feedback linearization or gain scheduling.

6.1. LOW-LEVEL CONTROL 97

6.1.1 H∞ Control

The theory of H∞ controllers can be found in various sources (e.g. [85]), thus
only an outline and the specific parts of the controllers used in this work is
given. For the H∞ -control design, the plant Gs to be controlled is augmented
with the dynamical weights Wi described in the frequency domain, resulting
in the transfer function Tzw . Using those weights allows to shape the con-
trol loop transfer functions, such as the sensitivity S or the complementary
sensitivity T , by minimizing the H∞ -norm of the transfer function Tzw . This
yields to the controller K. An illustration of the weighting principle is dis-
played in Figure 6.1.

zi
Wi

wi zi
Wi K Gs

Figure 6.1: Illustration of the weighting principle for the H∞ -control design.

If the H∞ -problem has a solution ||Tzw ||∞ ≤ γ with γ ≤ 1, the specifica-

tions for the weighted transfer functions are fulfilled. However, technically it
is not necessary to achieve the value γ = 1, a small value for γ is sufficient
for an asymptotically stable and robust system [41].
There exist various weighting schemes in order to shape the desired trans-
fer functions. Often used are the mixed sensitivity schemes S/KS/T and
GS/T found in [41]. Both of these schemes have their advantages and dis-
advantages. In this work the GS/T approach is applied, since the S/KS/T -
weighting scheme does not allow the system matrices to have poles in the
origin, which is the case in the helicopter model. This problem can be solved
by translating the poles slightly away from the origin, but it is more conve-
nient to use the GS/T - weighting scheme instead. As the names says this
weighting scheme shapes the product of the plant Gs with the sensitivity S
and the complementary sensitivity T .
Before starting with the design of the individual controllers, the general con-
trol structure is presented in the next section.
98 6. CONTROL

6.1.2 Control Structure

The low-level control references are the three spacial position x, y, z and the
heading angle ψ provided by the high-level controller. It would be possible to
design one controller involving the whole system model and taking care of all
references, which if designed correctly might perform well. But the layout of
such a controller is difficult and thus the controller is split into subcontrollers.
Similar to the parameter identification in Section 5.3 the control tasks are
split correlating to the system input signals. The motors are mainly respon-
sible for the heave and yaw dynamics (z and ψ), while as the servos generally
control the attitude and the horizontal movement (x and y). In the case of
the servos it makes sense to split the controller again into a fast timescale
controller taking care of the attitude and a slow timescale controller for po-
sition control2 . The main reason for this decision, beside the problems with
the position estimation on muFly, is the necessary attitude stabilization of
the helicopter which is difficult to tune if the position signal is the reference
input to the controller. It makes more sense to design a ’roll-pitch’ controller
for the attitude, with reference inputs φ and θ generated by a position con-
troller.
The control structure for full position control of muFly is shown in Figure 6.2
with the corresponding reference inputs. Since the IMU is measuring the an-
gular rates p,q and r it would be a pity not to include those measurements
into the control. These measurements are used as auxiliary sensor inputs and
are shown as side inputs in the graphic.

z
ψ Yaw-Heave
Controller umot,dw
r
umot,up
muFly
userv1
helicopter
x φ userv2
y Position θ Roll-Pitch
Controller Controller
u v p q

Figure 6.2: Control structure for muFly.

2 There is a large timescale difference in the heave and yaw dynamics as well, but this

is compensated in the control design and a split is not necessarily needed.

6.1. LOW-LEVEL CONTROL 99

6.1.3 Linearization
For the application of the H∞ -technique the nonlinear model from Chapter 4
has to be linearized around an operation point. Since the helicopter will
mainly work in and around hover, it is obvious to use the hover point as the
operation point. This choice simplifies the problem and the requirement of a
stationary point f (x0 , u0 ) = 0 is fulfilled.
Splitting the system into the heave-yaw and roll-pitch dynamics and applying
the linearization, two state space systems with following configuration evolve.

• Heave-yaw
6 states, 2 inputs and 3 outputs

 
z
 w   
    z
 ψ  umot,dw
xhy =  uhy = , yhy = ψ  (6.1)
 r  umot,up
  r
 Ωdw 
Ωup

• Roll-pitch
8 states, 2 inputs and 4 outputs
 
φ
 θ 
   
 p  φ
   
 q 
 urp = userv1 ,
 θ 
xrp = 
 αdw  yrp =  (6.2)
  userv2  p 
 βdw  q
 
 αup 
βup

The complete state-space models are shown in Appendix B. Those models

form the basis for the heave-yaw and roll-pitch controller design discussed in
the next sections. The system equations for the position control are discussed
later.

6.1.4 Heave-Yaw Control

In Figure 6.3 the singular values of the unscaled linearized heave-yaw plant Gs
are shown. The first cross over frequency is at 2.82 rad/s, whereas the second
100 6. CONTROL

frequency is very high at 575 rad/s. This shows the strong difference in the
heave and the yaw dynamics and the good scaling needed. Further the plant
shows an integrating character meaning that theoretically an integrating part
of the controller is not needed. Nevertheless the controller is designed to
have poles close to the imaginary axis in order to compensate for the battery
voltage drop during operation, requiring a higher motor input signal with
time.

SWV der Strecke Gs (−)

300

200

100
Value [dB]

−100

−200

−300 −5 −4 −3 −2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10
Frequency [rad/s]

Figure 6.3: Singular value plot of the heave-yaw subsystem.

For the heave-yaw controller a 2-dof H∞ approach [51] is used. This

means the controller K consists of two sub-controllers, a controller Kf for
feedforward reference input and a controller Kb for feedback control as shown
in Figure 6.4. The reason for the choice of a feedforward part is the desired
tight reference following for altitude. The feedforward part minimizes over-
shoots in the altitude, which is a critical aspect in the foreseen missions. On
the downside a higher order controller is obtained needing more processing
power.
The dual (y > u) 2-dof GS/T -weighting scheme of the heave-yaw controller
is shown in Figure 6.4 and has the corresponding transfer function Tzw

 T
−Wũ Tu Wd¯ Gs Su Wd¯
 Wũ Su Kf Wr̄ Tyr Wr̄ 
Tzw =  . (6.3)
 Wũ Su Kf Wr (Tyr − WM )Wr 
Wũ Su Kb Wd Te Wd

The controllers Kf and Kb are then designed using following weights

6.1. LOW-LEVEL CONTROL 101

WM

wd
Wd

wr
Wr Controller K

wr r ũ zu
Wr Kf Wũ
wd e u y zy
Wd Kb Gs

Figure 6.4: The dual 2-dof GS/T -weighting scheme.

augmented to their right dimensions

s+10
Wũ (s) = 0.001, Wd (s) = 0.1s+20

0.7s+1 s+7
Wd¯(s) = s+0.7/100 , Wr̄ (s) = 0.01s+14 (6.4)

s2 +15s+14 14
Wr (s) = 10−4 s2 +14s+0.0014 , WM (s) = s+14 .

The meaning of the weights is as follows: the weight Wũ (s) is used to
minimize the influence of the second row (can be neglected). The weight
Wd−1 (s) is a boundary for the complementary sensitivity Te (s), Wr̄−1 (s) for
the complementary sensitivity Tyr (s) and Wd¯(s) is used to shape the sen-
sitivity Su (s). The two weights WM (s) and Wr (s) are used as a reference
model for Tyr (s) and as an upper boundary for the model-matching error
respectively.
Solving the H∞ problem results in a γ = 1.38. The resulting transfer func-
tions for the loop-gain Lu , sensibility Su and complementary sensibility Tu
are shown in Figure 6.5. The cross-over frequency is at ωc = 3.26 rad/s
while the maximum sensitivity and the maximum complementary sensitivity
peaks are at Su,max = 3.7398dB and Tu,max = 3.93dB which is slightly higher
than the commonly desired 3dB.
102 6. CONTROL

Singularwertverlaeufe von Lu
300

200

100
Value [dB]

−100

−200

−300

−400 −5 −4 −3 −2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10
Frequency [rad/s]

Singularwertverlaeufe von Su
50
Sensitivity Su
Compl. Sens. Tu
0

-50

-100
Value [dB]

-150

-200

-250

-300

-350

-400 -5 -4 -3 -2 -1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10
Frequency [rad/s]

Figure 6.5: Singular value plots of the heave-yaw plant loop gain Lu (up), the
sensitivity Su and the complementary sensitivity Tu transfer function (dw).
6.1. LOW-LEVEL CONTROL 103

6.1.5 Roll-Pitch Control

The procedure for the roll-pitch control is similar to the heave-yaw control, it
differs only in the weighting scheme due to the missing feedforward controller
Kf . The 1-dof weighting scheme for the pitch-roll controller is shown in
Figure 6.6.
wv
Wv zũ
Wũ
wr y zy
Wr r e K ũ u
Gs Wy

Figure 6.6: The dual 1-dof GS/T -weighting scheme.

The corresponding transfer function Tzw is

 
−Wũ Tu Wv Wũ KSe Wr
Tzw = . (6.5)
Wy Gs Su Wv Wy Te Wr
The singular value plot of the linearized roll-pitch plant is shown in Fig-
ure 6.7. The influence of the stabilizer bar is clearly visible. It strongly
reduces the bandwidth and leads to a resonance frequency around 14.5rad/s.
This is showing the downside of the stabilizer bar: beside the positive influ-
ence of helping to stabilize the system it makes the system inertial, a negative
effect for the position control.

The analogous procedure as in the heave-yaw control leads to the loop

gain Lu , sensitivity Su and complementary sensitivity Tu in Figure 6.8. The
controller is designed to have enough bandwidth for the position control.
The resonance peak is still visible and undesired, but is not significant. The
cross-over frequency is at ωc = 36rad/s, while the maximum complementary
sensitivity peak is at Tu,max = 1.9dB. A solution to reduce the effect of the
resonance peak might also be to modify the dynamic weights Wi .

6.1.6 Position Control

The position controller is responsible to set the reference inputs to the roll-
pitch controller depending on the current position of the helicopter. For
104 6. CONTROL

SWV der Strecke Gs (−)

100

50

0
Value [dB]

−50

−100

−150 −5 −4 −3 −2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10
Frequency [rad/s]

Figure 6.7: Singular value plot of the roll-pitch subsystem with stabilizer bar.

Singularwertverlaeufe von Lu
200

150

100
Value [dB]

50

−50

−100

−150 −5 −4 −3 −2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10
Frequency [rad/s]

Figure 6.8: (up) Singular value plot of the roll-pitch subsystem loop gain Lu ,
(dw) sensitivity Su and the complementary sensitivity Tu transfer functions.
6.1. LOW-LEVEL CONTROL 105

the position controller design it is assumed that the underlying roll-pitch

controller is able to control the given angles perfectly and instantaneously,
leading to the requirement of sufficiently high bandwidth for the roll-pitch
controller. It has to be considered that the trajectory generated by the high-
level controller is defined in the inertial frame J, while as the system dynamics
equations are in the body-fixed frame B. Thus a coordinate transformation
is necessary.
In Figure 6.9 the control structure for the position control is displayed. The
controller is split into two subcontrollers one for the position in the inertial
frame Kpos and one for the velocity in the body-fixed frame Kvel . Between
those two controllers the velocity reference J vvel is transformed from the
inertial frame to the body-fixed frame using the transformation matrix (4.2).

J uref Bu ref
φref, θref
J xref ABJ Kvel
Kpos

Jx meas Bumeas

Figure 6.9: Structure for position control.

The equations for the position controller Kpos are simple. It is only the
integration of the velocities in the inertial frame. For the velocity controller
Kvel it is additionally assumed that the heave-yaw controller keeps the height
and heading constant and the horizontal force is only due to the inclination of
the helicopter. The latter is not perfectly true since a part of the force is com-
ing from the tilted TPP. However the effect is assumed to be small enough
in order to be neglected. This assumption is later verified in simulation.
With those presumptions the helicopter can be looked at as a mass point
holding its altitude. The equations are

u̇ = Fx /m, (6.6)
v̇ = Fy /m, (6.7)
Fx = −mg sin θ, (6.8)
Fy = mg cos θ sin φ. (6.9)

The linearization of these equations results in the state space system

106 6. CONTROL

 
0 −g
Avel = 0, Bvel = ,
g 0
  (6.10)
1 0
Cvel = , Dvel = 0.
0 1

For the simulation the position controller is designed using a simple P-

controller3 , while for the velocity controller a H∞ -controller similar to the
roll-pitch controller is used.
Unfortunately, the position control could not be tested on the real platform,
due to the camera problems mentioned in Chapter 2. Thus the control is
only tested in simulation, where different aspects and potentially occurring
problems such as sensor uncertainty or delay could not be treated. Never-
theless in simulation the general concept could be verified on the nonlinear
model and shown in Section 6.2.4 together with the high-level control.

6.1.7 Implementation
In order to minimize the overall development time the controller is designed
in Matlab R
. This strongly reduces the design effort and allows simulating
the controller before testing it in real flight. Those simulations are done using
the identified nonlinear model from Chapter 4. The nonlinear model allows
discovering possible unwanted effects which were not found during the linear
controller design.
After testing the controller in simulation it has to be implemented on the
target, the dsPIC microprocessor. Before translating the code into the target
language C, the controllers have to be balanced, reduced in order and dis-
cretized to ensure a good performance on the microprocessor. The controller
finally implemented on the target is 10th order for heave-yaw and 8th order
for roll-pitch.

6.1.8 Results
The results of the heave-yaw and roll-pitch control are presented by compar-
ing the closed-loop simulation with a real flight test. The performance of the
position control is presented later as a part of the high-level control.

3 This is only done for simplification, a MIMO controller might show even better per-

formance
6.1. LOW-LEVEL CONTROL 107

Heave-Yaw Control
The result of the closed-loop simulation to a step input in heave (−0.5m,
negative means upwards) on the nonlinear plant is plotted in Figure 6.10.
In this condition the helicopter can easily follow the reference step in heave
without overshoots (the same result was found for a step response in yaw).
Measurements of a corresponding flight experiment on the real system are
shown in Figure 6.11. As expected, the real system shows a slightly worse
performance due to small model deviations and noisy data. Especially the
control in yaw direction is not as tight as predicted by simulation. A reason
is the strong drift of the yaw measurement from the IMU, disturbing the
measurement data. The heave control shows a good correlation with the
prediction of the simulation, having only a small overshoot, but considering
the scale the deviation is very small (< 5cm). Both effects are hardly visible
in flight. Overall, the performance for the heave-yaw control is satisfying.
Besides the good control behavior, the plots also show the good accuracy
and appropriateness of the nonlinear model. The predicted motor inputs of
simulation and measurement do qualitatively and quantitatively correlate.

Roll-Pitch Control
In Figure 6.12 the simulation results for a reference tracking in pitch is shown
as a comparison to the measured data in Figure 6.13. Different to the heave-
yaw control, here the reference values have to be set by the pilot’s remote
control. In this measurement the real system is able to follow the reference
slightly better than the model prediction. Here a reason might be that the
nonlinear model underestimates the dynamics of the system in pitch. How-
ever, it has to be considered that the two flight scenarios are not identical.
Thus it is also hard to judge the accuracy of the nonlinear model. But qual-
itatively the servo input of the nonlinear model and the real platform do
correspond.

6.1.9 Remarks
The results show a satisfying control performance. Nevertheless, there exist
different possibilities in order to enhance the control. Besides the optimal
tuning of the presented controllers, better results could be obtained by mea-
suring the motor/rotor speed. An elegant method to estimate the motor
speed is to measure the back-electromagnetic forces (EMF). Knowing the
motor and rotor speed minimizes errors in the identification process and al-
lows to measure two additional state variables. Those additional informations
108 6. CONTROL

0.1
Reference
Nonlinear Model
0

−0.1
Altitude z[m]

−0.2

−0.3

−0.4

−0.5

−0.6

0 1 2 3 4 5 6 7 8 9 10
Time[s]

0.03
Reference
0.02 Nonlinear Model

0.01
Yaw angle ψ [rad]

−0.01

−0.02

−0.03

−0.04

−0.05

−0.06
0 1 2 3 4 5 6 7 8 9 10
Time[s]

1.6
lower rotor motor
upper rotor motor
1.4

1.2
Control Output [−]

0.8

0.6

0.4

0.2
0 1 2 3 4 5 6 7 8 9 10
Time[s]

Figure 6.10: Simulation result plots for a closed-loop step response in heave.
6.1. LOW-LEVEL CONTROL 109

−0.1
Reference
Measurement
−0.2

−0.3
Altitude z[m]

−0.4

−0.5

−0.6

−0.7

−0.8
24 25 26 27 28 29 30
Time[s]

3 Reference
Measurement
2
Yaw angle ψ [rad]

−1

−2

−3
24 25 26 27 28 29 30
Time[s]

1.8
lower rotor motor
upper rotor motor
1.6

1.4
Control Output [−]

1.2

0.8

0.6

0.4

0.2
24 25 26 27 28 29 30
Time[s]

Figure 6.11: Measurement result plots for a step response in heave.

110 6. CONTROL

0.02

−0.02
Pitch angle θ [rad]

−0.04

−0.06

−0.08

−0.1

−0.12

−0.14 Reference
Nonlinear Model
−0.16
0 1 2 3 4 5 6 7 8 9 10
Time[s]

0.6
Servo 1

0.4

0.2
Control Output [−]

−0.2

−0.4

−0.6

−0.8
0 1 2 3 4 5 6 7 8 9 10
Time[s]

Figure 6.12: Simulation result plot for a reference following in pitch.

6.1. LOW-LEVEL CONTROL 111

0.25
Reference
0.2 Measurement

0.15
Pitch angle θ [rad]

0.1

0.05

−0.05

−0.1

−0.15

−0.2
91 92 93 94 95 96
Time[s]

0.5
Servo 1
0.4

0.3
Servo Input [−]

0.2

0.1

−0.1

−0.2

−0.3
91 92 93 94 95 96
Time[s]

Figure 6.13: Measurement result plots for a reference following in pitch. The
reference is applied by the pilot with the RC stick.
112 6. CONTROL

can be used for a tighter control.

The next step would be to fly without the stabilizer bar. However, the band-
width of the servo actuators is most probably not high enough for a proper
control without the the stabilizer bar and a replacement of the servos has to
be considered.

6.2 High-Level Control

Beside a low-level control, where the system is stabilized, a high-level con-
trol is needed in order to give the helicopter the required intelligence for
fulfilling autonomous missions. Preferably the helicopter should be able to
autonomously take off, follow waypoints, avoid obstacles, map the environ-
ment, localize itself and land. As described in the previous chapters the
low payload capability and processing power narrows the capability for those
tasks to a minimum. Most of the tasks are very challenging or cannot be
accomplished on such a small system. Especially the realization of localiza-
tion and mapping of the environment remains a desire. However, in the next
sections the approach on muFly towards a fully autonomous helicopter is
presented. A general key element for reaching the goals is to keep everything
as simple as possible.

6.2.1 Mission State Machine

The structure of the high-level control is shown in Figure 6.14. Its heart
forms the mission control state machine, which gives waypoint inputs to the
trajectory generator, depending on the mission and the current state, in order
to create the reference inputs for the controller.
Low-level control

r
User Mission Trajectory y
Controller muFly
control generation

Figure 6.14: The high-level control structure of muFly.

In Figure 6.15 all the states and transition conditions of the state machine
are displayed. It starts in the ’init’ state, where the sensors are initialized
and calibrated. After initialization the user sends the command to take-
off, whereby a trajectory for the heave controller is generated. As soon as
6.2. HIGH-LEVEL CONTROL 113

the helicopter is in a confidence region the state is automatically switched

to hover and the helicopter is again ready to receive the next commands
from the user, such as a waypoint to go or the landing procedure. If the
omnidirectional camera detects an object on the route to a given waypoint,
the state switches to an obstacle avoidance procedure to prevent collision.
The ’RC’ state is only used for debugging and emergency. In this state, the
controller inputs are directly commanded by the remote control.

obstacle
signal: start
Take-off detected Obstacle

altitude reached obstacle obstacle

take-off failed obstacle detected avoided
avoided

sig: waypoint
Init Hover Waypoint
goal reached

signal: land
battery low

Landing RC

Figure 6.15: The mission state machine.

6.2.2 Trajectory Generation

An important part of the waypoint following is the generation of the ref-
erences for the position controller. Ideally, a dynamic feasible trajectory is
generated, which is tightly followed by the position controller. Additionally,
it would be beneficial to include information of the environment in the pro-
cedure. However, due to the limited processing power this is omitted and a
simpler method is implemented, the cubic spline interpolation which is used
for helicopter velocity control by Conte [35]. The idea here is not to generate
a strict trajectory for the position controller to follow, but to give reasonable
reference inputs in order to avoid steps and overshoots.
Cubic splines are used for path planning because they are continuous up to
the second derivative. Meaning that smooth curves can be generated and
discontinuities in the helicopter’s acceleration are avoided. The analytical
form of the cubic spline is:

P = At3 + Bt2 + Ct + D (6.11)

114 6. CONTROL

with time parameter t ∈ [0, 1]. The vectors A,B,C and D are computed as a
function of the boundary conditions as follows:
D = x0
C = ẋ0
B = 3(xg − x0 ) − 2ẋ0 − ẋg
A = 2(x0 − xg ) + ẋ0 + ẋg (6.12)
with boundary conditions

x0 : start coordinates,
xg : goal coordinates,
ẋ0 : normalized start velocity,
ẋg : normalized final velocity.

With these vectors, the curvature vector K and the tangent vector T are
calculated as follows:
K = T × Q × T/|T|4
T = At3 + 2Bt + C (6.13)
Q = 6At + 2B
where Q is the second order derivative. For the calculation of Np waypoints,
the trajectory has to be discretized. Therefore (6.11) is evaluated for some
discrete values of t in [0, 1]. In Figure 6.16 an example is shown for a three-
dimensional spline. The distribution in time is as important as the spatial
distribution, which ensures a smooth acceleration and deceleration.
On the helicopter the concept is to reach the waypoint as simple as possi-
ble in a straight line. Therefore, the waypoint following is split corresponding
to the controllers: first, the yaw angle and attitude is adapted and if the de-
sired point is reached the straight line trajectory to the waypoint is generated
as shown in Figure 6.17, where a waypoint following is simulated. This strat-
egy additionally supports the ideas behind the controller design to split in
heave-yaw and horizontal movements. As it can be seen in the simulation
plot using the nonlinear model, the helicopter is able to reach the waypoints
without overshoots.

6.2.3 Obstacle Avoidance

Obstacle avoidance is a crucial task on the helicopter. The idea behind
the obstacle avoidance on muFly is to use the omnidirectional camera in
6.2. HIGH-LEVEL CONTROL 115

Figure 6.16: Illustration of a three-dimensional cubic spline.

Figure 6.17: Simulation results with trajectories for waypoint following.

116 6. CONTROL

combination with the eight laser diodes allowing to obtain the distance to
the closest objects. If an obstacle is within a certain range, the obstacle
avoidance is activated.
There exist different obstacle avoidance techniques (e.g. [26], [84], [33]), but
for highly sophisticated approaches, usually a map is needed to calculate
an appropriate trajectory around the obstacle and still reaching the desired
goal. Since such a map is not available on muFly, a simple approach is needed.
The algorithm used is based on a repulsive potential field surrounding every
obstacle [53], preventing the helicopter from colliding with it.

x
7 0

y
5
2

4
3

Figure 6.18: The eight laser diodes seen from the top.

The lasers are numbered as shown in Figure 6.18. If the nearest object is
closer than the threshold distance lthresh the obstacle avoidance is activated
and the measured distances l(i) transformed in repulsive forces fi

 1 1
− lthresh
l(i) , if li < lthresh
fi = (6.14)
0, else
π 9π
αi = i + (6.15)
4 8
with i = 0, 1, 2, · · · , 7 and translated in the body-fixed coordinate system
by
 
fi · cos αi
fi = . (6.16)
fi · sin αi
The forces are summarized
6.2. HIGH-LEVEL CONTROL 117

7
X
F= fi , (6.17)
i=0

and overlaid to the position controller as

radd,x = k · Fx , (6.18)
radd,y = k · Fy , (6.19)

with a scaling factor k. In Figure 6.19 the principle of the repulsive forces is
shown.

Figure 6.19: Illustration of the repulsive forces.

6.2.4 Results
The high-level control is implemented on the dsPic microprocessor and tested
as much as possible on the ground. Unfortunately it was not possible to test
it on the flying helicopter due to the missing position reference. Nevertheless
the functionality of the state machine as well as of the waypoint following
could be tested in simulation. For this purpose the C code of the finite state
machine is translated to Matlab code and implemented into the nonlinear
Simulink model used for the low-level control simulations. The architecture
of the C and Matlab code are identical. In Figure 6.20 the simulation result
for following scenario is plotted

1. Take-off and wait 2.5 seconds if reached.

2. Fly to waypoint [−3, 0, −1]T and wait 2.5 seconds if reached,

3. Fly to waypoint [−3, 3, −2]T and wait 2.5 seconds if reached,

118 6. CONTROL

4. Fly to waypoint [0, 3, −1]T and wait 2.5 seconds if reached,

5. Fly to waypoint [0, 0, −2]T and wait 2.5 seconds if reached.
6. Land.

Figure 6.20: High-level control simulation result. The helicopter is flying a sample
square track.

The result shows the working mission state machine, which is depending
on the position and on the state of the helicopter, sends new goal waypoints
to the trajectory generator. In turn the reference input for the position and
heave-yaw controller is generated. All the control parts are working and the
helicopter reaches the desired waypoints without problems. The simulation
confirms the functionality of the control concept of muFly. Unfortunately it
could not be tested in real flight condition and the functionality could not be
verified. However, the success of the position control is strongly depending on
the quality of the sensor signals. Especially good velocity estimation imposes
some problems since the output of the V-SLAM is position data which has
to be differentiated in order to obtain the velocity.
Chapter 7

Conclusion

This chapter is divided into the review of the main contributions of this thesis
and the outlook on future work. It tries to identify possible future steps
towards an improvement of the control and further steps in the aerodynamic
investigation.

7.1 Review and Achievements

This thesis presented two of the most important tasks during the design of a
fully autonomous helicopter, the rotor and control design. The specialty in
both subjects is the extremely small size of the muFly micro helicopter.

7.1.1 Aerodynamics
Due to the small size, the rotors are working in the low Reynolds number
regime where viscous effects dominate the flow and reduce the efficiency.
Since only recent improvements in battery technology made it possible to
power such small helicopters, the research in this field is not yet well estab-
lished. Thus this thesis contributes to the state of the art in this respect,
by the investigation of rotors in this small size. One of the main achieve-
ments in this area is the implementation of the three simulation approaches,
the Blade Element Momentum Theory, a Vortex Method approach and a
Computational Fluid Dynamics setup, which all show good correlation with
the measurements and are appropriate tools for the design and investiga-
tions of micro helicopter rotors. The simulation approaches are validated on
a coaxial rotor test bench allowing to measure precisely thrust and torque.
Additionally the drive train of muFly is incorporated in the investigations

119
120 7. CONCLUSION

simulating realistic conditions as on the helicopter. This allows using the

desired efficiency quantity for the rotor design, the thrust to electrical power
ratio. Using this quantity the influence of different blade parameters and lift
enhancing devices are investigated. Beside the investigations on the single
rotor setup also the coaxial configuration is considered and possibilities for
improvements are discussed.

7.1.2 Modeling and Control

Concerning control, the small size yields a system with extremely fast dy-
namics where the low quality of the miniaturized sensors and actuators ad-
ditionally handicap the control performance. In this thesis the control of
the helicopter is approached in a structured way: first a nonlinear model is
developed, with the goal of providing an accurate model for simulations still
being simple enough to be useful for the controller design. Consequently, all
the important effects on the helicopter are discussed and possible simplifica-
tions justified. The result is an accurate model which shows good correlation
with the measurements. This validation is done together with the param-
eter identification in an identification process. Hereby as many parameters
as possible are identified beforehand in order to minimize the identification
effort on real flight data. For the identification a nonlinear technique, the
Covariance Matrix Adaptation - Evolution Strategy, is used. The nonlinear
identification has advantages over the mostly used linear identification, since
linearization cancels out many cross-coupling effects. However for an accu-
rate simulation those effects preferably are modeled, even if a linear control
approach is used. Altogether the identification shows good results and con-
tributes to the today’s state of the art in system identification.
Finally a model-based robust control approach using the H∞ - technique is
presented. Thereby the control is split according to the input signals, result-
ing in different multiple-input multiple-output (MIMO) subcontrollers. The
comparison between the simulated control and the corresponding real flight
test shows the good performance of the controllers and the nonlinear model.
The last part is the high-level control responsible for the helicopter missions.
Since only low computational power is available on the platform, simple ap-
proaches have to be used. The heart forms the mission state machine which
gives waypoint goals to the trajectory generator depending on the user input
and the current state. The trajectory generator in turn calculates appropriate
reference input for the underlying low-level controller.
7.2. OUTLOOK 121

7.1.3 Main Achievements

This thesis addresses the researchers and engineers which aim to develop a
micro helicopter in the size and mass of a small bird. Thereby it focuses on
two of the most important design aspects and shows possible approaches how
to tackle the occurring problems.
In the field of the aerodynamics three potential simulation approaches for
simulating the rotor performance are presented together with the design of a
coaxial test bench for fast and flexible rotor measurements. Those form the
tools for the conducted investigations, which can be used as guidelines for
the future design of a micro helicopter.
The control of the helicopter problem is approached in three steps. The de-
velopment of a possible nonlinear model which is simple but accurate enough
so it can simultaneously be used for nonlinear simulations and the control lay-
out. This is followed by presenting an efficient approach for nonlinear system
identification and a possibility for robust low-level control of the vehicle. Ad-
ditionally simple approaches, adjusted to the available low processing power,
for high-level mission control are presented. Those steps form a solid control
framework and can be used for further investigations and improvements on
comparable platforms.

7.2 Outlook
In this section the ideas for future work in the discussed areas are presented.

7.2.1 Aerodynamics
The rotor investigations in this thesis mainly focused on the performance
in hover state. However for an acceptable flight performance there is more
needed than only a high efficiency in hover. An important point on these
small hingeless rotors is the flapping of the blades in order to achieve the
needed tilting of the tip path plane. As a result the rotor blades need to
have some flexibility. The combination of the flexibility of the blade and
the aerodynamic forces leads to a fluid-structure interaction (FSI). It would
be interesting to perform FSI simulations, a coupling between a structural
FEM analysis and CFD, in order to investigate the effect of flapping during
operation. Furthermore this could be used to determine the effective phase
shift angle while applying a cyclic pitch input. This would be very helpful
for the design, the system modeling and the control.
On the pure aerodynamics further design features could be investigated, e.g.
122 7. CONCLUSION

turbulators or special tip shapes, or the effect of feature combination. Ad-

ditionally it would be interesting to investigate other flight conditions like
forward flight or the influence of the ground effects, which is higher pro-
nounced on the micro helicopter than in full scale. This fact was observed
during the experiments, where the influence of the ground was much longer
visible than one and a half rotor diameters distance to the ground as it is
assumed for full scale helicopters.
Finally the coaxial rotor configuration could be investigated more deeply with
the goal of optimizing the lower rotor for more efficiency.

7.2.2 Modeling and Control

In this thesis a linear technique is used for the control layout. Under cer-
tain conditions it might be advantageous to use a nonlinear technique for the
control. However, for a helicopter like muFly there is not an urgent reason
why changing to a nonlinear strategy, since highly dynamic maneuvers are
hardly possible due to the setup. Perhaps reasonable nonlinear extensions
of the existing control design are the feedback linearization mentioned or in
certain circumstances gain scheduling. For the application of the nonlinear
techniques a revision of the nonlinear model has to be conducted.
Before changing to full nonlinear control, an improvement of the linear con-
trol by optimally tuning the controllers should be considered. Additionally,
the application of a LQ-controller such as a LQG-LTR-controller or a H2 -
controller could be tested and compared to the H∞ -controller. The effort
of designing such a controller is not too large since the model state space
equations are already known.
Probably the biggest effort towards a fully autonomous micro helicopter is
needed in the high-level control and navigation. The main problem lays in
the limited processing power and in the size of the sensors, making full au-
tonomy in the size of muFly not possible yet. At this point a progress in
technology is indispensable.
On bigger platforms indoor navigation and certain autonomy were recently
demonstrated using a small laser range finder and down-looking camera.
Those results let live the hope of the existence of fully autonomous micro
helicopters in the size of muFly in future.
Appendix A

Nonlinear Model

Transformations
   
x N
 y =A · E  (A.1)
BJ
z D

 
cθcψ cθsψ −sθ
ABJ =  −cφsψ + sφsθcψ cφcψ + sφsθsψ sφcθ  (A.2)
sθsψ + cφsθcψ −sφcψ + cφsθsψ cφcθ

   
φ̇ p
 θ̇  = R ·  q  (A.3)
JB
ψ̇ r


1 sφsθ sφsθ 
sθ sθ
RJB = 0 sφ −sφ  (A.4)
0 sφ sφ
sθ sθ

Equations of Motion
     
u̇ p u
1
 v̇  = F −  q  ×  v  (A.5)
m
ẇ r w

123
124 A. NONLINEAR MODEL

      
ṗ p p
 q̇  = I−1 M −  q  × I  q  (A.6)
ṙ r r

External Forces and Moments

F = Tup + Tdw + G + Wfus , (A.7)

M = Qup + Qdw + rCup × Tup + rCdw × Tdw
+ Qreact,dw + Qreact,up (A.8)

Fx = cT dw kT Ω2dw cos βdw sin αdw − cT up kT Ω2up sin βup − mg sin θ

Fy = cT dw kT Ω2dw sin βdw + cT up kT Ω2up cos β sin αup + mg cos θ sin φ
Fz = −cT dw kT Ω2dw cos αdw cos βdw − cT up kT Ω2up cos αup cos βup
+Whub + mg cos θ cos φ (A.9)

Mx = −zdw cT dw kT Ω2dw sin βdw − zup cT up kT Ω2up cos βup sin αup
My = zdw cT dw kT Ω2dw cos βdw sin αdw − zup cT up kT Ω2up sin βup
Mz = (cQup,rot + cQup,stab )kQ Ω2up − cQdw kQ Ω2dw (A.10)

Stabilizer Bar and Swash Plate Equations

αup = lup (ηbar−φ )

βup = lup (ζbar−θ ) . (A.11)

1
η̇bar = (φ − ηbar ) ,
Tf,up
1
ζ̇bar = (θ − ζbar ) , (A.12)
Tf,up
 125

1
α̇dw = (−ldw userv2 · θSPmax − αdw ),
Tf,dw
1
β̇dw = (−ldw userv1 · θSPmax − βdw ) (A.13)
Tf,dw

Motor Equations
κM Ubat umot,dw − κM κE igear Ωdw
Jdrive,dw Ω̇dw = − dR Ωdw
igear RΩ
cQ kQ Ω2dw
− ,
i2gear · ηgear
κM Ubat umot,up − κM κE igear Ωup
Jdrive,up Ω̇up = − dR Ωup
iRΩ
cQ kQ Ω2up
− 2 (A.14)
igear · ηgear
126 A. NONLINEAR MODEL
Appendix B

State-Space Models

Heave-Yaw
 
0 1 0 0 0 0
2cT ,dw kT Ωhov,dw 2cT ,up kT Ωhov,up

 0 0 0 0 − m − m


 0 0 0 1 0 0 
Ahy =  (B.1)

 0 0 0 0 a45 a46 

 0 0 0 0 a55 0 
0 0 0 0 0 a66

dR 2cQ,dw kQ Ωhov,dw 2cQ,dw kQ Ωhov,dw κE κM

a45 = − + 2
+ (B.2)
Izz Izz ηgear igear Izz RΩ Izz
dR 2cQ,up kQ Ωhov,up 2cQ,up kQ Ωhov,up κE κM
a46 = − + − 2
− (B.3)
Izz Izz ηgear igear Izz RΩ Izz
dR 2cQ,dw kQ Ωhov,dw κE κM
a55 = − − − (B.4)
Jdrive,dw ηgear i2gear Jdrive,dw RΩ Jdrive,dw
dR 2cQ,up kQ Ωhov,up κE κM
a66 = − − 2
− (B.5)
Jdrive,up ηgear igear Jdrive,up RΩ Jdrive,up

 
0 0

 0 0 

 0 0 
Bhy = κM Ubat κM Ubat (B.6)
 
 − igear Izz RΩ igear Izz RΩ


κM Ubat

 igear Jdrive,dw RΩ 0 

κM Ubat
0 igear Jdrive,up RΩ

127
128 B. STATE-SPACE MODELS

 
1 0 0 0 0 0
Chy = 0 0 1 0 0 0  (B.7)
0 0 0 1 0 0

 
0 0
Dhy = 0 0  (B.8)
0 0

Roll-Pitch

 
0 0 1 0 0 0 0 0

 0 0 0 1 0 0 0 0 


 a31 0 0 0 0 a35 a36 0 

 0 a42 0 0 a45 0 0 a48 
Arp = 1 (B.9)
 
 0 0 0 0 − Tf,dw 0 0 0 

1

 0 0 0 0 0 − Tf,dw 0 0 

1 1
0 0 0 0 0 − Tf,up 0
 
 Tf,up 
1 1
0 Tf,up 0 0 0 0 0 − Tf,up

cT,up kT lup Ω2hov,up zup

a31 = (B.10)
Ixx
cT,dw kT Ω2hov,dw zdw
a35 = − (B.11)
Ixx
a36 = −a31 (B.12)
cT,up kT lup Ω2hov,up zup
a42 = (B.13)
Iyy
cT,dw kT Ω2hov,dw zdw
a45 = (B.14)
Iyy
a48 = −a42 (B.15)
 129

 
0 0

 0 0 


 0 0 

 0 0 
Brp = ldw ΘSP,max (B.16)
 
 Tf,dw 0 

 ldw ΘSP,max 
 0 Tf,dw

 
 0 0 
0 0

 
1 0 0 0 0 0 0 0
 0 1 0 0 0 0 0 0 
Crp =
  (B.17)
0 0 1 0 0 0 0 0 
0 0 0 1 0 0 0 0

 
0 0
 0 0 
Drp =
 0
 (B.18)
0 
0 0
130 B. STATE-SPACE MODELS
Bibliography

[1] Ascending Technologies.

http:// asctec.de , (08/18/09).

[2] Autonomous Intelligent Systems AIS, ALU Freiburg.

http:// ais.informatik.uni-freiburg.de/ , (12/19/2009).

[3] BeCAP TU Berlin.

http:// www.becap.tu-berlin.de/ , (12/19/2009).

[4] Brutus Cluster ETH Zurich.

http://www.brutus.ethz.ch, (01/09/2010).

[5] CEDRAT Technologies.

http:// www.cedrat.com , (12/19/2009).

[6] CSEM.
http:// www.csem.ch/ , (12/19/2009).

[7] DelFly Project.

http://www.delfly.nl, (11/20/2009).

[8] Devantech SRF10.

http:// www.acroname.com/ robotics/ parts/ R241-SRF10.html ,
(01/28/2009).

[9] EADS Barracuda.

http:// www.eads.net/ 1024/ en/ businet/ defence/ mas/ uav/ barracuda.html ,
(08/12/2009).

[10] Epson.
http://www.epson.co.jp/e/newsroom/news_2004_08_18.htm,
(12/19/2009).

131
132 BIBLIOGRAPHY

[11] GENERAL ATOMICS Predator.

http:// www.ga-asi.com/ products/ aircraft/ predator.php, (08/12/2009).

[12] Microchip Technologies.

http:// www.microchip.com , (01/28/2009).

[13] Mighty Midget Motors.

http:// www.microbrushless.com , (11/23/2009).

[14] Project muFly.

http:// www.mufly.org , (12/19/2009).

[15] Proxyflyer.
http://www.proxyflyer.com, (11/20/2009).

[16] RUAG Ranger.

http:// www.ruag.com/ en/ AerospaceTechnology/ Aircraft_Services/
Military_Aircraft_Services/ Unmanned_Aerial_Systems , (08/12/2009).

[17] XSENS.
http:// www.xsens.com/ , (12/19/2009).

[18] J. D. Anderson. Fundamentals of Aerodynamics. McGraw-Hill, Singa-

pore, 4th edition, 2005.

[19] Ansys. Ansys CFX Manual, 2009.

[20] C. Bermes, S. Leutenegger, S. Bouabdallah, D. Schafroth, and R. Sieg-

wart. New Design of the Steering Mechanism for a Mini Coaxial He-
licopter. In IEEE International Conference on Intelligent Robots and
Systems (IROS), Nice, France, 2008.

[21] C. Bermes, D. Schafroth, S. Bouabdallah, and R. Siegwart. Modular

Simulation Model for Coaxial Rotary Wing MAVs. In 2nd International
Symposium on Unmanned Aerial Vehicles, Reno, USA, 2009.

[22] W. Bittner. Flugmechanik der Hubschrauber. Springer, 2nd edition,

2005. (in German).

[23] F. Bohorquez and D. Pines. Hover Performance and Swashplate Design

of a Coaxial Rotary Wing Micro Air Vehicle. In American Helicopter
Society 60th Annual Forum, Baltimore, USA, 2004.
BIBLIOGRAPHY 133

[24] F. Bohorquez, F. Rankinsy, J. Baederz, and D. Pines. Hover Perfor-

mance of Rotor Blades at Low Reynolds Numbers for Rotary Wing Mi-
cro Air Vehicles. An Experimental and CFD Study. In AIAA Applied
Aerodynamics Conference, Orlando, USA, 2003.

[25] F. Bohorquez, P. Samuel, and J. Sirohi. Design, Analysis and Hover

Performance of a Rotary Wing Micro Aerial Vehicle. Journal of the
American Helicopter Society, 48:80–90, 2003.

[26] J. Borenstein and Y. Koren. The vector field histogram-fast obstacle

avoidance for mobile robots. IEEE Journal of Robotics and Automation,
1991.

[27] S. Bouabdallah. Design and Control of Quadrotors with Application to

Autonomous Flying. PhD thesis, 2007.

[28] S. Bouabdallah and R. Siegwart. Backstepping and Sliding-mode Tech-

niques Applied to an Indoor Micro Quadrotor. In IEEE International
Conference on Robotics and Automation (ICRA), pages 2247–2252,
Barcelona, Spain, 2005.

[29] A. Bramwell. Helicopter Dynamics. Butterworth Heinemann, 2nd edi-

tion, 2001.

[30] B. H. Carmichael. Low Reynolds number airfoil survey. Technical report,

NASA, 1981.

[31] N. Cartwright. PicooZ and Gyrotor. Model Helicopter World, (12):18–

22, 2006.

[32] L. Chen and P. McKerrow. Modelling the Lama Coaxial Helicopter. In

Australasian Conference on Robotics & Automation, Brisbane, Australia,
2007.

[33] H. Choset, K. M. Lynch, S. Hutchinson, G. Kantor, W. Burgard, L. E.

Kavraki, and S. Thrun. Principles of Robot Motion: Theory, Algorithms,
and Implementations. The MIT Press, 2005.

[34] C. P. Coleman. A Survey of Theoretical and Experimental Coaxial Rotor

Aerodynamic Research. Technical report, NASA, 1997.

[35] G. Conte, S. Duranti, and T. Merz. Dynamic 3d path following for

an autonomous helicopter. In 5th IFAC Symposium on Intelligent Au-
tonomous Vehicles, Lisbon, 2004.
134 BIBLIOGRAPHY

[36] M. Drela. X-FOIL, subsonic airfoil development system, 2004.

http:// raphael.mit.edu/ xfoil , (01/28/2009).

[37] A. Dzul and T. Hamel. Nonlinear control for a tandem rotor helicopter.
In IFAC World Congress, Barcelona, Spain, 2002.

[38] A. Dzul, T. Hamel, and R. Lozano. Modeling and Nonlinear Control

for a Coaxial Helicopter. In IEEE International Conference on Systems,
Man and Cybernetics, volume 6, 2002.

[39] P. Ferrat, C. Gimkiewicz, S. Neukom, Y. Zha, A. Brenzikofer, and

T. Baechler. Ultra-miniature omnidirectional camera for an autonomous
flying micro-robot. In Proceedings of SPIE, volume 7000, Strasbourg,
France, 2008.

[40] M. Gad-el Hak. Flow Control: Passive, Active, and Reactive Flow Man-
agement . Cambridge University Press, 2000.

[41] H. P. Geering. Robuste Regelung. Technical report, ETH Zurich, 2004.

(in German).

[42] M. Gerig. Modeling, Guidance, and Control of Aerobatics Maneuvers of

an Autonomous Helicopter. PhD Thesis, 2008.

[43] H. Glauert. Airplane Propellers. Aerodynamic Theory. Springer Verlag,

Berlin, Germany, 1935.

[44] M. Gloor. Vort-X: Development of a Simulation Code for a Helicopter

Rotor using a Vortex Method. PhD thesis, 2008.

[45] J. Grasmeyer and M. Keenon. Development of the Black Widow Micro

Air Vehicle. In 39th AIAA Aerospace Sciences Meeting and Exhibit,
Reno, USA, 2001.

[46] D. A. Griffiths. A Study of Dual-Rotor Interference and Ground Effect

Using a Free-Vortex Wake Model. In American Helicopter Society 58th
Annual Forum, Montreal, Canada, 2002.

[47] S. Grzonka, G. Grisetti, and W. Burgard. Towards a Navigation System

for Autonomous Indoor Flying. In IEEE International Conference on
Robotics and Automation (ICRA), Kobe, Japan, 2009.

[48] N. Hansen. The CMA Evolution Strategy: A Tutorial. Technical report,

2009.
BIBLIOGRAPHY 135

[49] N. Hansen and A. Ostermeier. Adapting Arbitrary Normal Mutation

Distributions in Evolution Strategies: The Covariance Matrix Adapta-
tion. In IEEE International Conference on Evolutionary Computation,
Nagoya, Japan, 1996.

[50] G. Hoffmann, D. G. Rajnarayan, S. L. Waslander, D. Dostal, J. S. Jang,

and C. J. Tomlin. The Stanford testbed of autonomous rotorcraft for
multi agent control (STARMAC). In Digital Avionics Systems Confer-
ence, 2004. DASC 04. The 23rd, volume 2, pages 12.E.4–121–10 Vol.2,
2004.

[51] I. M. Horowitz. Synthesis of Feedback Systems. Academic Press, New

York, 1963.

[52] E. Jacobs, K. Ward, and R. Pinkerton. The Characteristics of 78 Re-

lated Airfoil Sections from Tests in the Variable-Density Wind Tunnel.
Technical report, NASA, 1933.

[53] O. Khatib. Real-time obstacle avoidance for manipulators and mobile

robots. In IEEE International Conference on Robotics and Automation,
1985.

[54] I. Kroo and P. Kunz. Development of the Mesicopter: A Miniature

Autonomous Rotorcraft . In American Helicopter Society Vertical Lift
Aircraft Design Conference, San Francisco, 2000.

[55] P. K. Kundu and I. M. Cohen. Fluid Mechanics. Academic Press, San

Diego, 2nd edition, 2002.

[56] E. Laitone. Wind tunnel tests of wings at Reynolds numbers below 70

000. Experiments in Fluids, 23:405–409, 1997.

[57] A. J. Landgrebe. The wake geometry of a hovering rotor and its infuence
on rotor performance. Journal of the American Helicopter Society, 1972.

[58] J. Leishman. Helicopter Aerodynamics. Cambridge, 2nd edition, 2006.

[59] J. Leitner, A. Calise, and J. V. R. Prasad. Analysis of adaptive neural

networks for helicopter flight controls. Journal of Guidance, Control,
and Dynamics, 1997.

[60] F. L. Lewis and V. L. Syrmos. Optimal Control. Wiley&Sons, 2nd

edition, 1995.
136 BIBLIOGRAPHY

[61] L. Lipera, J. Colbourne, M. Tischler, M. Mansur, M. Rotkowitz, and

P. Patangui. The Micro Craft iSTAR Micro Air Vehicle: Control System
Design and Testing. In American Helicopter Society 57th Annual Forum,
Washington, DC, 2001.

[62] P. B. S. Lissaman. Low-Reynolds-Number Airfoils. Annual Review of

Fluid Mechanics, 15:223–239, 1983.

[63] L. C. Mak et al. Design and development of the Micro Aerial Vehicles
for Search, Tracking And Reconnaissance (MAVSTAR) for MAV08. In
1st US-Asian Demonstration and Assessment of Micro-aerial and Un-
manned Ground Vehicle Technology (MAV08), Agra, India, 2008.

[64] J. H. McMasters and M. L. Henderson. Low speed single element airfoil

synthesis. Technical report, 1980.

[65] B. Mettler. Identification Modeling and Characteristics of Miniature

Rotorcraft. Kluwer Academic Publishers, 2003.

[66] B. Mettler, C. Dever, and E. Feron. Scaling Effects and Dynamic Char-
acteristics of Miniature Rotorcraft. Journal of Guidance, Control, and
Dynamics, 27(3):466–478, 2004.

[67] B. Mettler, T. Kanade, and M. Tischler. System Identification Model-

ing of a Model-Scale Helicopter. Technical report, Robotics Institute,
Carnegie Mellon University, 2000.

[68] M. Moeckli. Guidance and control for aerobatic maneuvres of an un-

manned airplane. PhD Thesis, 2006.

[69] G. Mueller and B. Ponick. Elektrische Maschinen Grundlagen elek-

trischer Maschinen. VCH, Weinheim, 2006. (in German).

[70] J. Nocedal and S. Wright. Numerical Optimization. Springer, 2006.

[71] A. Noth. Design of Solar Powered Airplanes for Continuous Flight. PhD
thesis, 2008.

[72] A. Noth, R. Siegwart, and W. Engel. Autonomous Solar UAV for Sus-
tainable Flight. In Advances in Unmanned Aerial Vehicles, State of the
Art and the Road to Autonomy, pages 377–405. Springer, 2007.

[73] H. Oertel and E. Laurien. Numerische Strömungsmechanik, volume 2.

Vieweg, Braunschweig/Wiesbaden, 2003. (in German).
BIBLIOGRAPHY 137

[74] G. D. Padfield. Helicopter Flight Dynamics. Blackwell Science, Edin-

burgh, UK, 1996.

[75] T. Pornsin-Sirirak, S. Lee, H. Nassef, J. Grasmeyer, and Y. Tai. MEMS

wing technology for a battery-powered ornithopter. In The 13th An-
nual International Conference on Micro Electro Mechanical Systems,
Miyazaki, Japan, 2000.

[76] P. Pounds, R. Mahony, and J. Gresham. Towards Dynamically-

Favourable Quad-Rotor Aerial Robots. In Australasian Conference on
Robotics and Automation, Canberra, Australia, 2004.

[77] R. W. Prouty. Helicopter Performance, Stability, and Control. PWS

Engineering, Boston, 1986.

[78] A. V. d. Rostyne. Device For Steering a Helicopter, Patent WO

03080433a1, 2003.

[79] D. Schafroth, C. Bermes, S. Bouabdallah, and R. Siegwart. Micro

Helicopter Steering: Review and Design for the muFly Project. In
IEEE/ASME International Conference on Mechatronic and Embedded
Systems and Applications (MESA), Beijing, China, 2008.

[80] D. Schafroth, C. Bermes, S. Bouabdallah, and R. Siegwart. Modeling

and System Identification of the muFly Micro Helicopter. Journal of
Intelligent and Robotic Systems, 57(1-4):27–47, 2009.

[81] D. Schafroth, S. Bouabdallah, C. Bermes, and R. Siegwart. From the

Test Benches to the First Prototype of the muFly Micro Helicopter.
Journal of Intelligent and Robotic Systems, 54(1-3):245–260, 2008.

[82] H. R. Schwarz and N. Kaeckler. Numerische Mathematik. Teubner,

Stuttgart; Leipzig; Wiesbaden, 2004. (in German).

[83] W. Shyy, Y. Lian, J. Tang, D. Viiieru, and H. Liu. Aerodynamics of

Low Reynolds Number Flyers, volume 1. Cambridge University Press,
Cambridge, 2008.

[84] R. Siegwart and I. R. Nourbakhsh. Introduction to Autonomous Mobile

Robots. Bradford Book, 2004.

[85] S. Skogestad and I. Postlethwaite. Multivariable Feedback Control: Anal-

ysis and Design. John Wiley & Sons, 2005. 1121635.
138 BIBLIOGRAPHY

[86] B. Steder, G. Grisetti, C. Stachniss, and B. Wolfram. Visual SLAM for

Flying Vehicles. IEEE Transactions on Robotics, 24:1088–193, 2008.
[87] N. Tsuzuki, S. Sato, and T. Abe. Design Guidelines of Rotary Wings
in Hover for Insect-Scale Micro Air Vehicle Applications. Journal of
Aircraft, 44:252–263, 2007.

[88] R. L. Wade and G. W. Walker. Flight test results of the fuzzy logic
adaptive controller-helicopter (FLAC-H). In Society of Photo-Optical
Instrumentation Engineers (SPIE) Conference Series, 1996.
[89] D. J. Walker. Multivariable control of the longitudinal and lateral
dynamics of a fly-by-wire helicopter. Control Engineering Practice,
11(7):781–795, 2003. doi: DOI: 10.1016/S0967-0661(02)00189-2.
[90] W. Wang, G. Song, K. Nonami, M. Hirata, and O. Miyazawa. Au-
tonomous Control for Micro-Flying Robot and Small Wireless Helicopter
X.R.B. In IEEE International Conference on Intelligent Robots and Sys-
tems (IROS), Beijing, China, 2006.

[91] J. T. Watkins. AH-1W Super Cobra, U.S. Marine Corps, 1999.

[92] J. Watkinson. The Art of the Helicopter. Elsevier, 2004.
[93] M. Weilenmann. Robuste Mehrgrössen-Regelung eines Helikopters. PhD
thesis, 1994.

[94] L. Zhao and V. R. Murthy. Optimal Controller for an Autonomous Heli-

copter in Hovering and Forward Flight. In Aerospace Sciences Meeting,
Orlando, Florida, 2009. AIAA.
Curriculum Vitae

Dario Schafroth is born in Bottmingen, Switzerland

on August 28th 1980. He started his mechanical en-
gineering studies in the year 2000 at the Eidgenös-
sische Technische Hochschule Zürich (ETHZ) with
the focus on numerical fluid mechanics and control
systems. During this time he completed a six month
internship at ABB Turbosystems AG working in
the field of CFD. As a result of the internship he
received the opportunity to write a semester the-
sis at ABB with the goal of extending an existing
CFD code. His second semester thesis, written at
ETH, dealt with the control of a model airplane.
His diploma thesis was a CFD study on ’Hele-Shaw’
cells written at the University of California Santa Barbara (UCSB), receiving
his Diploma (equivalent M.Sc.) in 2006.
In summer 2006, he started his doctoral thesis in the field of autonomous
micro helicopters at the Autonomous Systems Laboratory (ASL) of ETHZ
and was working on the European project muFly. During this time he super-
vised and co-supervised over 20 student projects and was a coach of the first
focus project ’Formula Student 06/07’. He is author and co-author of several
conference and journal papers and a book chapter on aerial robotics. His cur-
rent research interests include control theory, aerodynamics and mechatronic
systems design.

139
140 B. CURRICULUM VITAE

List of Publications (only first author)

Book Chapters
• Schafroth, D., Bouabdallah, S., Bermes, C., and Siegwart, R. "From the
Test Benches to the First Prototype of the muFly Micro Helicopter" in:
Unmanned Aircraft Systems: International Symposium On Unmanned
Aerial Vehicles, UAV’08, Springer Verlag, 2009.

Journals
• Schafroth, D., Bermes, C., Bouabdallah, S., and Siegwart, R. "Mod-
eling, system identification and robust control of a coaxial micro heli-
copter", Control Engineering Practice, Special Issue on Aerial Robotics,
doi:10.1016/j.conengprac.2010.02.004, 2010
• Schafroth, D., Bermes, C., Bouabdallah, S., and Siegwart, R. "Modeling
and System Identification of the muFly Micro Helicopter" in Journal of
Intelligent and Robotic Systems, 2009
• Schafroth, D., Bouabdallah, S., Bermes, C., and Siegwart, R. "From the
Test Benches to the First Prototype of the muFly Micro Helicopter",
Journal of Intelligent and Robotic Systems, 54(1-3): 245-260, 2008.
• Schafroth, D., Goyal, N., and Meiburg, E. "Miscible displacements in
Hele-Shaw cells: Nonmonotonic viscosity profiles", European Journal of
Mechanics - B/Fluids, vol. 6, no. 3, pp. 444-453, 2007

Peer-reviewed Proceedings
• Schafroth, D., Bermes, C., Bouabdallah, S., and Siegwart,"Propulsion
System Optimization for the muFly Micro Helicopter" in 35th European
Rotorcraft Forum, Hamburg, Germany, 2009
• Schafroth, D., Bermes, C., Bouabdallah, S., and Siegwart, "Modeling
and System Identification of the muFly Micro Helicopter" in 2nd In-
ternational Symposium on Unmanned Aeriel Vehicles, UAV’09, Reno,
USA, 2009
• Schafroth, D., Bermes, C., Bouabdallah, S., and Siegwart, R. "Mi-
cro Helicopter Steering: Review and Design for the muFly Project" in
IEEE/ASME International Conference on Mechatronic and Embedded
Systems and Applications (MESA), Beijing, China, 2008