Alberts Accurate 2012

Accurate Autonomous Landing of a Fixed-Wing
Unmanned Aerial Vehicle
by
Frederik Nicolaas Alberts
Thesis presented in partial fulfilment of the requirements for the degree of
Master of Science in Engineering
in the Faculty of Engineering at Stellenbosch University
Supervisor:
Prof T. Jones
Department Electrical and Electronic Engineering
December 2012
Stellenbosch University http://scholar.sun.ac.za
Declaration
By submitting this thesis electronically, I declare that the entirety of the work contained
therein is my own, original work, that I am the sole author thereof (save to the extent
explicitly otherwise stated), that reproduction and publication thereof by Stellenbosch
University will not infringe any third party rights and that I have not previously in its
entirety or in part submitted it for obtaining any qualification.
December 2012
Copyright © 2012 Stellenbosch University

All rights reserved
Abstract
This thesis presents the analysis, design, simulation and practical implementation of a
control system to achieve an accurate autonomous landing of a fixed-wing unmanned
aerial vehicle in the presence of wind gust atmospheric disturbances.
Controllers which incorporate the concept of direct-lift control were designed based on
a study of the longitudinal dynamics of the UAV constructed as a testbed. Direct-lift
control offers the prospect of an improvement in the precision with which aircraft height
and vertical velocity can be controlled by utilising actuators which generate lift dir-
ectly, instead of the conventional method whereby the moment produced by an actuator
results in lift being indirectly generated. Two normal specific acceleration controllers
were designed. The first being a conventional moment-based controller, and the second
a direct-lift-augmented controller. The moment-based controller makes use of the air-
craft’s elevator while the direct-lift augmented controller in addition makes use of the
flaps of the aircraft which serve as the direct-lift actuator.
Controllers were also designed to regulate the airspeed, altitude, climb rate, and roll
angle of the aircraft as well as damp the Dutch roll mode. A guidance controller was im-
plemented to allow for the following of waypoints. A landing procedure and methodology
was developed which includes the circuit and landing approach paths and the concept of
a glide path offset to calibrate the touchdown point of a landing.
All controllers and the landing procedure were tested in a hardware-in-the-loop simula-
tion environment as well as practically in a series of flight tests. Five fully autonomous
landings were performed, three of these using the conventional NSA controller, and the
final two the direct-lift-augmented NSA controller.
The results obtained during the landing flight tests show that the project goal of a landing
within five meters along the runway and three meters across the runway was achieved
in both normal wind conditions as well as in conditions where wind gusts prevailed. The
flight tests also showed that the direct-lift-augmented NSA controller appears to achieve
a more accurate landing than the conventional NSA controller, especially in the presence
of greater wind disturbances. The direct-lift augmented NSA controller also exhibited
less pitch angle rotation during landing.
iii
Opsomming
Hierdie tesis verteenwoordig die analise, ontwerp, simulasie en praktiese implemente-

ring van ’n beheerstelsel wat ten doel het om ’n akkurate en outonome landing van ’n
onbemande vastevlerk vliegtuig in rukwind atmosferiese toestande te bewerkstellig.
Gegrond op ’n studie van die longitudinale dinamika van die vliegtuig wat as proeftuig
gebruik is, is beheerders ontwerp wat die beginsel van direkte-lig insluit. Direkte-lig
beheer hou die potensiaal in om die vliegtuig se hoogte en vertikale snelheid akkuraat
te beheer deur gebruik te maak van aktueerders wat lig direk genereer in teenstelling
met die konvensionele metode waar die moment van die aktueerder indirek lig gene-
reer. Twee normaal-versnellings beheerders is ontwerp. Die eerste is ’n konvensionele
moment-gebaseerde beheerder wat gebruik maak van die hys-aktueerder van die vlieg-
tuig, en die tweede is ’n direkte-lig-bygestaande beheerder wat addisioneel gebruik maak
van die flappe van die vliegtuig wat as die direkte-lig aktueerder dien.
Vedere beheerders is ontwerp wat die lugspoed, hoogte, klimkoers, en rolhoek van die
vliegtuig reguleer asook die “Dutch roll” gedrag afklam. ’n Leiding-beheerder wat die
volg van vliegbakens hanteer, is ingestel. Die landingsprosedure en -metodologie is ont-
wikkel wat die landingspad sowel as die sweef-pad bepaal en wat terselfdertyd ’n metode
daarstel om die posisie van die landingspunt te kalibreer.
Die beheerders en landingsprosedure is in ’n hardeware-in-die-lus omgewing gesimu-

leer en deur middel van ’n reeks proefvlugte getoets. Vyf ten volle outonome landings is
uitgevoer waarvan drie van die konvensionele normaal-versnellings beheerder gebruik
gemaak het, en die laaste twee die direkte-lig-bygestaande normaal-versnellings beheer-
der.
Die vlugtoetsuitslae bevestig dat die navorsingsdoel om ’n landing binne vyf meter in
lyn met en drie meter dwarsoor die landingstrook te bewerkstellig, behaal is. Hier-
die akkuraatheid is verkry in beide goeie atmosferiese toestande sowel as toestande
met rukwinde. Volgens die vlugtoetse blyk dit dat die direkte-lig-bygestaande normaal-
versnellings beheerder ’n meer akkurate landing kan bewerkstellig as die konvensionele
normaal-versnellings beheerder, veral dan in toestande met rukwinde. Die direkte-lig-
bygestaande normaal-versnellings beheerder het ook ’n laer hei-hoek rotasie tydens die
landing vertoon.
iv
Contents
Abstract iii
Opsomming iv
List of Figures ix
List of Tables xiii
Nomenclature xiv
Acknowledgements xviii
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Research Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Testbed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Aircraft Model Description 6

2.1 Axis System Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.1 Inertial and Runway Axes . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Body Axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Wind and Stability Axes . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.4 Navigation Axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Actuator Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Six Degree of Freedom Equations of Motion . . . . . . . . . . . . . . . . . . 11
2.3.1 Point Mass Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1.1 Point Mass Dynamics . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1.2 Attitude and Attitude Dynamics . . . . . . . . . . . . . . . . 14
2.3.2 Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.2.1 Rigid Body Rotational Dynamics . . . . . . . . . . . . . . . . 14
2.3.3 Forces and Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3.1 Aerodynamic Forces and Moments . . . . . . . . . . . . . . . 17
v
CONTENTS vi
2.3.3.2 Thrust Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3.3.3 Gravitational Forces and Moments . . . . . . . . . . . . . . . 20
2.4 Simplifying and Decoupling the Model . . . . . . . . . . . . . . . . . . . . . 20
3 Longitudinal Analysis and Control 23

3.1 Direct-Lift Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Normal Specific Acceleration Controller . . . . . . . . . . . . . . . . . . . . 25
3.2.1 Overview and Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.2 Investigation of the Elevator and Flap Actuators . . . . . . . . . . . . 28
3.2.3 Investigation of the Natural NSA Dynamics . . . . . . . . . . . . . . . 29
3.2.3.1 NSA Response to Elevator Input . . . . . . . . . . . . . . . . 31
3.2.3.2 NSA Response to Flaps Input . . . . . . . . . . . . . . . . . . 34
3.2.4 NSA Controller Design - Elevator Actuator . . . . . . . . . . . . . . . 37
3.2.4.1 Pole/Zero Placement . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.4.2 HIL Simulation and Practical Flight Test Results . . . . . . . 43
3.2.5 NSA Controller Design - Flaps Actuator . . . . . . . . . . . . . . . . . 44
3.2.5.1 Pole/Zero Placement . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.6 NSA Controller Design - Elevator and Flaps Combined . . . . . . . . 49
3.2.6.1 HIL Simulation and Practical Flight Test Results . . . . . . . 52
3.3 Climb Rate Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.2 Pole Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.3 HIL Simulation and Practical Flight Test Results . . . . . . . . . . . . 59
3.4 Altitude Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.4.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4.2 Pole Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5 Airspeed Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.2 Pole Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Lateral Analysis and Control 71

4.1 Decoupling the Lateral Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2 Roll Angle Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.2 Pole Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Dutch Roll Damper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.2 Pole Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
CONTENTS vii

4.4 Aircraft Guidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5 Landing 85
5.1 Landing Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.1.1 Standard Aircraft Landing Procedure . . . . . . . . . . . . . . . . . . 86
5.1.2 Modified Aircraft Landing Procedure . . . . . . . . . . . . . . . . . . 87
5.2 Landing Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3 Longitudinal Landing Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.4 Landing State Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.5 Landing Flight Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.5.1 Landing Test 1 - 28 November Flight 1 . . . . . . . . . . . . . . . . . 97
6 Conclusion 114
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4 Recommendations for Further Research . . . . . . . . . . . . . . . . . . . . 117
A Additional Mathematical Principles [1] 119

A.1 Transforming the Derivative of a Vector in a Rotating Reference Frame . . 119
A.2 Cross Product Transformation Matrix . . . . . . . . . . . . . . . . . . . . . . 119
A.3 Direction Cosine Matrix (DCM) . . . . . . . . . . . . . . . . . . . . . . . . . . 120
A.4 Moment of Inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
A.5 Small Angle Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
B UAV System Description 122

B.1 Airframe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
B.1.1 Propulsion Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
B.2 Avionics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
B.3 Ground Station Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
B.4 Hardware-in-the-Loop Simulation Environment . . . . . . . . . . . . . . . . 127
B.5 Actuator and Controller Limits . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.6 Aircraft Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
B.6.1 Standard Flight Conditions . . . . . . . . . . . . . . . . . . . . . . . . 129
B.6.2 Aircraft Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
CONTENTS viii
B.6.3 Aerodynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

B.6.4 Dimensional Stability and Control Derivative Notation . . . . . . . . 131
C Detailed Derivations 133

C.1 Normal Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
C.1.1 Characteristic equation for the poles . . . . . . . . . . . . . . . . . . 134
C.1.2 Characteristic equation for the zeros . . . . . . . . . . . . . . . . . . 134
C.2 NSA Controller Design - Elevator Actuator . . . . . . . . . . . . . . . . . . . 138
C.3 NSA Controller Design - Flaps actuator . . . . . . . . . . . . . . . . . . . . . 142
C.4 Glide Path Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
D Additional Flight Test Information 155

D.1 Helderberg Radio Flyer’s Airfield . . . . . . . . . . . . . . . . . . . . . . . . 155
D.2 Windspeed data - 29 November . . . . . . . . . . . . . . . . . . . . . . . . . 156
D.3 Flight Test Cards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
D.3.1 Flight Test 1 - 25 July . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
D.3.2 Flight Test 2 - 10 August . . . . . . . . . . . . . . . . . . . . . . . . . 158
D.3.3 Flight Test 3 - 7 September . . . . . . . . . . . . . . . . . . . . . . . . 161
D.3.4 Flight Test 4 - 20 September . . . . . . . . . . . . . . . . . . . . . . . 164
D.3.5 Flight Test 5 - 25 November . . . . . . . . . . . . . . . . . . . . . . . 166
Bibliography 170
List of Figures
1.1 UAV photo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 Thesis chapter outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Inertial and runway axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1
2.2 Body axis system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2
2.3 Wind and stability axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2
2.4 Navigation axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3
2.5 Actuator definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Block diagram overview of the six degree of freedom equations of motion [2] . 11
3.1 Longitudinal control system overview . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2 Block diagram of NSA controller strategy . . . . . . . . . . . . . . . . . . . . . . 28
4
3.3 Longitudinal actuator configuration of the aircraft . . . . . . . . . . . . . . . . 28
3.4 Pole-zero map and step response of the NSA to elevator input transfer function 32
3.5 Step response of the full and simplified NSA to elevator input transfer functions 33
3.6 Pole-zero map and step response of the NSA to flaps input transfer function . . 35
3.7 Pole-zero map and step response of the NSA to flaps input transfer function
where the flaps moment coefficient is of opposite direction . . . . . . . . . . . 36
3.8 Pole-zero map and step response of the NSA to flaps input transfer function
where the flaps moment is cancelled out . . . . . . . . . . . . . . . . . . . . . . 37
3.9 Pole-zero map and step response of the NSA to flaps input transfer function for
a range of moment gearings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.10 Pole-zero map of the closed-loop elevator-based NSA controller . . . . . . . . . 41
3.11 NSA step response to a reference NSA with corresponding elevator deflection
for the elevator-based NSA controller . . . . . . . . . . . . . . . . . . . . . . . . 42
3.12 Pole-zero map of the closed-loop elevator-based NSA controller for a change in
the control gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.13 NSA response to a reference along with corresponding elevator deflection for
the elevator-based NSA controller during HIL and flight tests . . . . . . . . . . 44
3.14 Pole-zero map of the closed-loop flaps-based NSA controller . . . . . . . . . . . 47
3.15 NSA step responses to a reference NSA with corresponding flaps deflection of
the flaps-based NSA controller for a range of feedback gains . . . . . . . . . . 47
3.16 Simulink block diagram of the direct-lift-augmented NSA controller . . . . . . 50
ix
LIST OF FIGURES x
3.17 Step response of the combined elevator and flaps NSA controller for a range
of flaps feedback gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.18 NSA step response with corresponding pitch rate response of the combined
elevator and flaps NSA controller for a range of flaps to elevator gearing ratios 52
3.19 NSA reference following in the presence of an NSA disturbance . . . . . . . . 53
3.20 Climb rate and corresponding NSA response for the conventional and direct-
lift-augmented NSA controllers during HIL and flight tests . . . . . . . . . . . . 54
3.21 High-pass filtered NSA and NSA reference signals during HIL and flight tests . 55
3.22 Step response of the combined elevator and flaps NSA controller for a range
of flaps to elevator gearing ratios . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.23 Pole-zero map and step response of the climb rate controller for a range of
feedback gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.24 Climb rate response and corresponding NSA response for the conventional and
direct-lift-augmented NSA controllers during practical flight tests . . . . . . . 61
3.25 Climb rate step response and corresponding NSA response for a series of climb
rate feedback gains during practical flight tests . . . . . . . . . . . . . . . . . . 61
3.26 Pole-zero map and step response of the altitude controller for a range of feed-
back gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.27 Altitude step and corresponding climb rate response for the conventional and
direct-lift-augmented NSA controllers during HIL and flight tests . . . . . . . . 65
3.28 Altitude response and roll angle during a looped flight path . . . . . . . . . . . 66
3.29 Pole-zero map and step response of the airspeed controller . . . . . . . . . . . 69
3.30 Airspeed step with corresponding throttle and altitude responses during a
flight test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.31 Airspeed with corresponding throttle and altitude responses for sloped altitude
path during a flight test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.1 Longitudinal Control System Overview Block Diagram . . . . . . . . . . . . . . 71

4.2 Pole-zero map and step response of the roll angle controller . . . . . . . . . . . 75
4.3 Roll angle response with corresponding roll rate response and aileron deflec-
tion angle during HIL simulation and flight tests . . . . . . . . . . . . . . . . . . 77
4.4 Roll angle step response with corresponding roll rate response and aileron
deflection angle during flight tests . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5 Root locus design of the Dutch roll damper at the aircraft’s trim velocity . . . . 79
4.6 Yaw rate response and rudder deflection angle during HIL simulation and flight
tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.7 Yaw rate response to safety pilot rudder perturbations for Dutch roll damper
disabled and enabled during flight test . . . . . . . . . . . . . . . . . . . . . . . 81
4.8 Guidance logic [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.9 Conversion of lateral and normal acceleration in navigation axes to wind axes
NSA and roll angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
LIST OF FIGURES xi
4.10 Flight path and corresponding roll angle, navigation axis LSA and wind axis
NSA reference commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.1 Standard rectangular airfield traffic pattern . . . . . . . . . . . . . . . . . . . . 86

5.2 Standard landing path geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 Effect of an altitude error in the landing path on the touchdown point . . . . . 88
5.4 Landing path geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.5 Landing circuit geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6 Landing circuit in runway axis system during HIL simulation and flight test . . 92
5.7 Glide path reference offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.8 Landing longitudinal path for a mock landing during HIL simulation and flight
test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.9 Control System Overview Block Diagram . . . . . . . . . . . . . . . . . . . . . . 95
5.10 Landing state machine overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.11 Landing Test 1 - Altitude, airspeed and circuit flight path for the entire duration
of the flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.12 Landing Test 1 - Altitude during landing leg of the mock landings, and all flight
paths across the runway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.13 Landing Test 1 - Altitude, flight path and roll angle during landing leg . . . . . 99
5.14 Landing Test 1 - Climb rate and NSA, and pitch rate, pitch angle and elevator
deflection during landing leg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.15 Landing Test 1 - Altitude during touchdown . . . . . . . . . . . . . . . . . . . . 100
of the flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
of the flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
of the flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
LIST OF FIGURES xii
of the flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.36 Landing Test 3 and 4 - Pitch angle, pitch rate and airspeed during landing leg 113
B.1 Aircraft photo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

B.2 Avionics overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
B.3 Ground station graphical user interface . . . . . . . . . . . . . . . . . . . . . . . 126
B.4 HIL overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
B.5 HIL graphical display . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.6 Aircraft geometry in AVL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
C.1 Control system block diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
D.1 Satellite photo of Helderberg Radio Flyers (HRF) airfield. Source: Google Earth155
D.2 Wind speed during Landing Test 3,4 and 5 . . . . . . . . . . . . . . . . . . . . . 156
D.3 Flight test card - Flight Test 1 - 25 July . . . . . . . . . . . . . . . . . . . . . . . 157
D.4 Flight test card - Flight Test 2 - 10 August . . . . . . . . . . . . . . . . . . . . . 158
D.5 Flight test card - Flight Test 3 - 7 September . . . . . . . . . . . . . . . . . . . . 161
D.6 Flight test card - Flight Test 4 - 20 September . . . . . . . . . . . . . . . . . . . 164
D.6 Flight test card - Flight Test 4 - 20 September . . . . . . . . . . . . . . . . . . . 165
D.7 Flight test card - Flight Test 5 - 25 November . . . . . . . . . . . . . . . . . . . 166
List of Tables
3.1 Longitudinal stability and control derivatives . . . . . . . . . . . . . . . . . . . 29
B.1 Actuator and controller limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

B.2 Wing geometric data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
B.3 Longitudinal stability derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 131
B.4 Lateral stability derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
B.5 Control derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
B.6 Dimensional stability and control derivatives (forces) [1] . . . . . . . . . . . . . 132
B.7 Dimensional stability and control derivatives (moments) [1] . . . . . . . . . . . 132
xiii
Nomenclature
Abbreviations and Acronyms
AVL Athena Vortex Lattice
CAN Controller Area Network
DC Direct Current
DCM Direction Cosine Matrix
DGPS Differential Global Positioning System
DLC Direct-Lift Control
GPS Global Positioning System
HIL Hardware-In-the-Loop
IMU Inertial Measurement Unit
LQR Linear Quadratic Regulator
LSA Lateral Specific Acceleration
MIMO Multi-Input, Multi-Output
NMP Non-Minimum Phase
NP Neutral Point
NSA Normal Specific Acceleration
PC Personal computer
OBC On-Board Computer
RC Remote Controlled
RF Radio Frequency
RTK Real Time Kinematic
SU Stellenbosch University
UAV Unmanned Aerial Vehicle
xiv
NOMENCLATURE xv
Vector Notation
XAB
C Denotes vector X with data about axis system A relative to axis system B,
coordinated in axis system C
Coordinate Vectors
Px ,Py ,Pz Position vector components in the X ,Y and Z axes
V̄ ,α,β Airspeed magnitude, angle of attack, angle of sideslip
ψ ,θ,φ Yaw, pitch and roll angles
P ,Q,R Roll, pitch and yaw rates
L,M ,N Rolling, pitching and yawing moments
X ,Y ,Z Roll, pitch and yaw rates
P ,Q,R X ,Y and Z -axis forces
A,B ,C Specific accelerations in the X ,Y and Z -axis directions
Vectors and Matrices
DCM Direction cosine matrix
F,M Force and moment vectors
G Gravitational acceleration vector
H Angular momentum vector
I Moment of inertia matrix
P,V,A Position, velocity and acceleration vectors
S Cross product matrix
ω Angular rate vector
Landing Symbols
h Altitude above touchdown point
ΨR Runway heading angle
γ Glide path angle
W Px,y Waypoint location

NOMENCLATURE xvi
Aircraft Modelling Symbols
A Aspect ratio
b Wing span
c̄ Mean aerodynamic chord
δA ,δE ,δF ,δR Aileron, elevator, flaps, rudder deflection angles
e Oswald efficiency factor
g Gravitational acceleration
Ix ,Iy ,Iz Moments of inertia about the X ,Y and Z axes
m Mass
q Dynamic pressure
ρ Air density
S Reference wing area
Tc ,T Commanded and actual thrust
τT Engine time constant
V̄ Airspeed magnitude
CD Aerodynamic drag coefficient
Cy Aerodynamic side force coefficient
CL Aerodynamic lift coefficient
Cl Aerodynamic roll moment coefficient
Cm Aerodynamic pitch moment coefficient
Cn Aerodynamic yaw moment coefficient
System Dynamics
ζ Damping ratio
ωn Natural frequency
p Pole location
z Zero location
NOMENCLATURE xvii
Subscripts and Superscripts
I,R,B,W,N Inertial, runway, body, wind, and navigation axis systems
A Aerodynamic forces and moments
G Gravitational forces and moments
T Engine forces and moments

Acknowledgements
I would like to thank the following people and organisations:
• Prof Thomas Jones for your support, guidance and advice throughout the project.
• Dr Iain Peddle and Mr Japie Engelbrecht, for your input and guidance during our
research group meetings.
• Armscor, The Defence Research and Development Board (DRDB) and the South
African Navy for providing the funding that made this project possible.
• AM de Jager, Anton Runhaar and Lionel Basson, for all your assistance during the
flight tests.
• The technical staff at the EE Department, including Wessel Kroukamp and Lincoln
Saunders for providing technical assistance with various aspects of the project.
• Micheal Basson for doing an excellent job as the safety pilot of the aircraft.
• All my friends in the ESL for creating a fun and enjoyable working atmosphere.
• My family for all their support and motivation during the project.
• My girlfriend, Sarah Boyd, for believing in me.
xviii
Chapter 1
Introduction
1.1 Background
The Centre of Expertise in Autonomous Systems within the Department of Electrical

and Electronic Engineering at Stellenbosch University performs ongoing research in the
field of autonomous systems. Active research is being conducted on topics related to the
control and automation of commercial aircraft, terrestrial and underwater vehicles, and
unmanned aircraft.
Previous research projects on fixed-wing unmanned aircraft focused on issues such as:
Autonomous Flight of a Model Aircraft [4]; Autonomous Take-Off and Landing of a Fixed-
Wing Unmanned Aerial Vehicle [5]; Autonomous Aerobatic Flight of a Fixed Wing Un-
manned Aerial Vehicle [6]; Acceleration Based Manoeuvre Flight Control System for Un-
manned Aerial Vehicles [2]; Die Presisie Landing van ’n Onbemande Vliegtuig “The Preci-
sion Landing of an Unmanned Aircraft” [7]; Flight Control System for a Variable Stability
Blended-Wing-Body Unmanned Aerial Vehicle [8]; Aggressive Flight Control Techniques
for a Fixed-Wing Unmanned Aerial Vehicle [1]; Advanced Take-off and Flight Control
Algorithms for Fixed Wing Unmanned Aerial Vehicles [9]; and Fault Tolerant Adaptive
Control of an Unmanned Aerial Vehicle [10].
This project is a further progression in this area of research. In particular, it is a continu-

ation of research into the autonomous landing of a fixed-wing unmanned aerial vehicle
(UAV). Ultimately, there is a need for a high degree of accuracy in autonomous landings
under various flight conditions to expand the operational capabilities of UAV’s. Landing
must for example be possible in operational circumstances that necessitate the land-
ing of a UAV on the deck of a ship or on a short or improvised runway in a land-based
application. This project aims to contribute to the achievement of this goal.
Research in autonomous landings was started by [5], where the successful take-off and
landing of a fixed-wing UAV was demonstrated. However, the sensors used to determine
the aircraft’s position limited the accuracy of the landings to the extent that a runway
width of at least 20 m was required to guarantee a landing. In an attempt to address this
1
CHAPTER 1. INTRODUCTION 2
limitation, [7] successfully developed an infra-red vision-based system to improve the ac-
curacy with which the aircraft’s position, velocity and attitude states can be determined.
However, no landing flight tests were conducted.
The aim of this research project is to build upon the knowledge base of the Centre of
Expertise in the field of autonomous fixed-wing aircraft landings. In particular, it will
endeavour to further develop the ability to accurately land a UAV even under adverse
atmospheric conditions, and also demonstrate this with practical flight tests.
1.2 Overview
The successful landing of a UAV presents two challenges: the accurate control of the
aircraft in the presence of atmospheric disturbances and the accurate determination of,
in particular, the position and velocity states of the aircraft. In this project, the focus
will be on the first part of the landing challenge: developing a control system which can
achieve an accurate landing in the presence of atmospheric disturbances. Wind gusts
and crosswinds are the two major atmospheric disturbances that impact on the landing.
This research will be limited to addressing the disturbances caused by wind gusts.
In this thesis an advanced GPS system is used to address the challenge of accurately
determining the position and velocity of the aircraft. This system is an implementation
of Real Time Kinematic (RTK) differential GPS (DGPS) by the Autonomous Systems Re-
search Group at Stellenbosch University (SU Research Group), and it is accurate to the
order of a few centimeters. It involves having one GPS unit on the ground, and another
on board of the aircraft. The GPS units measure and compare the phase of the satellite
signals which allow the relative position of the two units to be calculated to a high degree
of precision [11].
The primary goal of this thesis is to practically demonstrate an accurate autonomous

landing, and the research focusses on the development of a landing control system which
can achieve accurate aircraft control in the presence of atmospheric disturbances. Pro-
gress has been made in the development of aircraft controllers since those used by [5].
These enhanced controllers were developed by [2] and have been utilised practically by
[1] and [8]. They function on the acceleration level of the aircraft, and will be used as
the basis for the controller design in this project. The research also explores the concept
of direct-lift control, which offers the prospect of an improvement in the precision with
which aircraft height and vertical velocity can be controlled in the presence of wind gusts
[12].
For the purpose of this project, an accurate autonomous landing is defined as achieving
a landing touchdown within five meters along the runway and two meters across the
runway from the intended touchdown point.
1.3 Research Outcomes
This research project requires the following milestones to be attained in order to suc-
cessfully achieve an accurate autonomous landing of a fixed-wing UAV,
Aircraft:
• Construction of a UAV to serve as the testbed for the various controllers de-
veloped. This involves modifying and strengthening a model aircraft, incorpor-
ating an electric motor system as well as assembling and integrating sensors
and avionics. These include,
– Inertial Measurement Unit (IMU) providing 3-axis acceleration and rota-
tion measurements
– Magnetometer
– Differential pressure sensor for calculating airspeed and pressure altitude
– RTK DGPS system
• Calibration of all sensors and actuators.
Modelling and controllers:
• Determination of the aircraft model. This involves,

– Determining the stability and control derivatives through the use of Athena
Vortex Lattice (AVL).
– Conducting a static thrust test of the aircraft’s motor and propeller.
– Finding the moment of inertia through the double pendulum method.
• Analysis of the dynamics of the aircraft in order to formulate a control strategy.
This also involves an in-depth study of the longitudinal dynamics of the aircraft
in order to design aircraft controllers which incorporate the concept of direct-
lift control.
• Design of two normal specific acceleration controllers. The first being a con-
ventional elevator-based NSA controller, and the second a direct-lift-augmented
NSA controller which makes use of both the elevator and the flaps of the air-
craft.
• Design of controllers to regulate the airspeed, altitude, climb rate, and roll
angle of the aircraft, as well as a Dutch roll damper.
• Implementation of a guidance controller to allow for the following of waypo-
ints.
Landing:
• Developing a landing procedure and methodology to accomplish an accurate

landing. This includes the circuit and landing approach paths and the concept
of a glide path offset to calibrate the touchdown point of a landing.
Simulation and flight tests:
• Implementation of all controllers and the landing procedure on an OBC.
• Testing of all controllers in a hardware-in-the-loop (HIL) simulation environ-

ment as well as practically in a series of flight tests in order to verify the
controller design and the aircraft model.
• Testing of the landing procedure in a HIL simulation environment and practic-

ally in a series of flight tests.
• Conducting practical autonomous landings using the conventional NSA con-

troller as well as the direct-lift augmented NSA controller.
1.4 Testbed
The UAV constructed to act as the testbed for this project is shown in Figure 1.1 and the
details are contained in Appendix B.
Figure 1.1 – UAV photo

1.5 Thesis Outline
The description of the research conducted unfolds in six chapters as depicted in Figure
1.2.
Figure 1.2 – Thesis chapter outline
Chapter 2 describes the definitions required to model the aircraft and its environment.
This comprises the axis system and actuator definitions. The six degree of freedom equa-
tions of motion are then developed in such a way that they can be split into an outer- and
an inner-loop model. The outer-loop model consists of the slower, aircraft independent,
point mass dynamics, while the inner-loop model consists of the fast aircraft-specific dy-
namics. The chapter concludes by simplifying and decoupling the model such that the
control systems of Chapters 3 and 4 can be designed.
Chapter 3 addresses longitudinal analysis and control. During landing, it is intuitive

that the control of the aircraft occurs primarily in its longitudinal axis. Therefore, in
order to land accurately, it is essential that the longitudinal motion of the aircraft is
controlled accurately. The analysis of the longitudinal dynamics of the aircraft and the
subsequent design of appropriate controllers focus on this prerequisite. This chapter
therefore explores the concept of direct-lift control (DLC) and how it could be applied to
this thesis. This is followed by the design, HIL simulation, and practical flight test results
of the normal specific acceleration (NSA), climb rate, altitude and airspeed controllers.
Chapter 4 provides an analysis of the lateral dynamics of the aircraft, followed by the
design, HIL simulation and practical flight test results of the roll angle controller and
the Dutch roll damper. This is followed with by outline, HIL simulation and practical
flight test results of the guidance controller.
The procedure followed during a landing is presented in Chapter 5. The flight and des-
cent paths are investigated, followed by an overview of the landing state machine. The
flight test results of several autonomous landings where both the conventional and the
direct-lift-augmented NSA controllers were utilised are presented and analysed.
The concluding Chapter 6 presents a summary of the research, the results obtained and
the contributions made by this project to the knowledgebase of the SU Research Group.
Recommendations for further research are also made.
Chapter 2
Aircraft Model Description
This chapter describes the aircraft model presented by [2], the core feature of which is
the ultimate design of attitude-independent inner-loop acceleration controllers. These
controllers reduce the aircraft to a point mass with a steerable acceleration vector
as viewed from an outer-loop guidance controller’s perspective. It was demonstrated
that, for normal flight conditions, the inner-loop dynamics decouple and become linear,
thereby allowing for the design of inner-loop axial, normal and lateral specific accelera-
tion controllers. Similar controllers have been practically implemented by [1; 8] in which
certain limitations of this strategy were encountered and discussed. These limitations
have been taken into account in the design of the controllers of Chapters 3 and 4. This
thesis follows the same derivation of the aircraft model as that presented by [2] with
the addition of a flaps actuator, the use of which is covered in the controller designs of
Chapter 3.
This chapter furthermore describes the axis system and actuator definitions required to
model the aircraft and its environment. The six degree of freedom equations of motion
are then developed in such a way that they can be split into an outer- and an inner-loop
model. The outer-loop model consists of the slower, aircraft independent, point mass
dynamics, while the inner-loop model consists of the fast, aircraft-specific dynamics.
The chapter concludes by simplifying and decoupling the model such that the control
systems of Chapters 3 and 4 can be designed.
2.1 Axis System Definitions
Several appropriate reference frames are required to formulate the equations of motion
of the aircraft. The aircraft model presented by [2] requires that the motion of the
aircraft be split into the motion of a gross point mass with a superimposed rotational
motion relative to the commonly used wind axis system. This requirement is met by
defining a frame of reference relative to inertial space, as well as two aircraft fixed axis
6
CHAPTER 2. AIRCRAFT MODEL DESCRIPTION 7
systems. The motion of the aircraft is thus captured completely by the inertial reference
frame and the aircraft fixed reference frames rotating relative to the inertial frame.
2.1.1 Inertial and Runway Axes
An inertial axis system needs to be defined so that the aircraft’s position, velocity and
attitude can be described relative to a fixed reference frame. The inertial axis system is
defined as right-hand orthogonal with the horizontal plane tangential to the surface of
the earth and the origin at some convenient location such as the middle of the runway.
In order for Newton’s law of motion to be applied, it must be assumed that the earth is
flat and non-rotating. This assumption is valid in the case of this project where the flight
will involve short distances and durations, and where the sensors used are incapable of
registering the earth’s motion [13].
For the purpose of this project a runway axis system is also defined and will be used in
Chapter 5 where the aircraft landing geometry is explored. It is essentially the same as
the inertial axis system, except for a rotation angle (ΨR ) around the ZI -axis to account
for cases where the runway and inertial axes do not line up (see Figure 2.1).
Figure 2.1 – Inertial and runway axes

2.1.2 Body Axes
The body axis system is defined as right-hand orthogonal with its origin fixed at the
centre of gravity of the aircraft. The XB -axis points towards the nose of the aircraft and
is parallel to the horizontal fuselage line [13]. The YB -axis points out along the right
wing and the ZB -axis points down. The body axis system is depicted in Figure 2.2 along
with the associated notations and positive directions for the aircraft’s body axis forces
and moments.
Figure 2.2 – Body axis system1
2.1.3 Wind and Stability Axes
Similar to the body axis system, the wind and stability axis systems are defined as right-
hand orthogonal with their origins fixed at the center of gravity of the aircraft. The
stability axis system is reached by rotating the body axes around the YB -axis through the
angle of attack (α). The XS -axis therefore points towards the direction of the oncoming
free-stream velocity vector projected into the XB ZB -plane, whereas the ZW -axis lies in
the aircraft’s plane of symmetry. The wind axis system is a further extension of the
stability axis system. It is reached by rotating the stability axes around the ZW -axis
through the angle of sideslip (β ). The XW -axis therefore points towards the direction
of the oncoming free-stream velocity vector. The wind and stability axis systems are
illustrated in Figure 2.3.
1
Public Domain Image adapted. Source: Wikimedia.org
Figure 2.3 – Wind and stability axes2
2.1.4 Navigation Axes
Similar to the other aircraft-fixed axis systems, the navigation axis system is defined as
right-hand orthogonal with its origin fixed at the centre of gravity of the aircraft. The ZN -
axis is always parallel to the ZI -axis of the inertial axis system, and the XN -axis remains
in the aircraft’s plane of symmetry. The navigation axis system is illustrated in Figure
2.4
Figure 2.4 – Navigation axes2
2
2.2 Actuator Definitions
The aircraft used in this project is equipped with five conventional actuators. Four of
these are aerodynamic actuators, namely the elevator, rudder, aileron and flaps. The
final actuator is the throttle.
The flaps actuator is of special significance in this thesis. It is usually fairly inactive
in the sense that the flaps are typically set to specific deflection angles during certain
phases of flight, such as take-off or landing. The flaps of a conventional aircraft typically
have a very low rate of deflection when compared to its other aerodynamic actuators.
By contrast, in this project the flaps will be used actively. This is done to improve upon
the longitudinal path following performance of the aircraft, and is explored further in
Chapter 3. Using the flaps in this manner is possible due to the particular aircraft used
in this project, which is equipped with flaps that can be actuated at the same rate as the
other aerodynamic actuators. The flaps can also deflect upwards instead of the usual
downwards-only movement.
The aerodynamic actuators are defined such that a negative control surface deflection
creates a positive moment around the corresponding axis as shown in Figure 2.5. Estab-
lishing the correct deflection direction is intuitive for the elevator, aileron and rudder.
However, for the flaps it is somewhat more complicated since they are located much
closer to the center of gravity of the aircraft than the other aerodynamic actuators. This
is investigated further in §3.2.2.
Figure 2.5 – Actuator definitions3
3
2.3 Six Degree of Freedom Equations of Motion
This section presents the six degree of freedom equations of motion for a rigid body in
the form described by [2]. This form works on the basis that the motion of the body can
be seen as the superposition of the body’s slower point mass motion (kinematics) and
its faster rigid body rotational motion (kinetics). The point mass motion of the body is
described by the position and attitude of the wind axis system over time and the rigid
body motion is described by the attitude of the body axis system with respect to the wind
axis system.
A block diagram overview of the six degree of freedom equations of motion is shown in
Figure 2.6. The point mass kinematics depicted on the right side of the diagram describe
the gross motion of the aircraft. They are aircraft independent and are driven by the
specific accelerations (AW , BW and CW ) as well as the roll rate (PW ). The rigid body
rotational dynamics depicted on the left side of the diagram contain the aircraft-specific
parameters, and therefore all of the aircraft-specific uncertainty [2]. It can be seen that
the rigid body rotational dynamics describe how the various forces and moments acting
on the aircraft translate into specific accelerations and roll rate. These accelerations and
roll rate then propagate over time into velocity, position and attitude in the kinematic
equations.
Figure 2.6 – Block diagram overview of the six degree of freedom equations of motion [2]
As demonstrated by [2], the effect of wind axis attitude parameters coupling into the
rigid body rotational dynamics through the gravitational acceleration vector (GW ) can
be cancelled via dynamic inversion. It was also demonstrated that the velocity magnitude
and air density feedback couplings of the point mass kinematics back into the rigid body
rotational dynamics can be ignored due to the timescale separation which exists between
these parameters and the rigid body rotational dynamics.
It was proposed by [2] that a set of high-bandwidth controllers be designed around the
rigid body rotational dynamics which would encapsulate all of the aircraft specific uncer-
tainty. Since these controllers operate at the acceleration and angular rate level, they
would reject disturbances before they appear in the slower dynamics as error in posi-
tion, velocity and attitude. In this thesis, a similar approach is followed for some of the
controller designs of Chapters 3 and 4.
2.3.1 Point Mass Kinematics
Kinematics describe the motion of a rigid body, in this case the aircraft’s position, velo-
city, acceleration and attitude, in three dimensional space. The aircraft is represented
by its wind axis system which translates and rotates relative to the inertial axis system.
The point mass kinematics are aircraft independent and are driven by the specific ac-
celerations (AW , BW and CW ) acting on the aircraft as well as the roll rate (PW ). These
inputs originate from the rigid body rotational dynamics of §2.3.2.
2.3.1.1 Point Mass Dynamics
The point mass dynamics describe the kinematic relationship which exists between the
position, velocity and acceleration of the aircraft’s center of mass with respect to the
inertial axis system. The aircraft is represented by its wind axis system since the origin
of the wind axis system coincides with the aircraft’s center of mass. It is assumed that
the aircraft is a rigid body with a fixed center of mass. The kinematic relationship can
therefore be written as,
d WI
P = VW I (2.3.1)
dt I
and,
d WI
V = AW I (2.3.2)
dt I
where PW I , VW I and AW I , are the position, velocity and acceleration vectors of the wind
axis system with respect to the inertial axis system. If it is assumed that the aircraft’s
mass remains constant, then by Newton’s second law, the applied resultant force vector
(F) can be written as,
F = mAW I (2.3.3)
Substituting the equation above into Equation 2.3.2 yields,
d WI 1
V = F (2.3.4)
dt I m
For the control architecture presented by [2] it is more intuitive to work with a velocity
magnitude and the attitude of the wind axis system when describing the velocity vector.
The inertially coordinated derivative of the velocity vector of Equation 2.3.2 is now con-
verted into a derivative with respect to wind axes by making use of Equation A.1.1 and
substituting in Equation 2.3.4,
d WI 1
V = −ω W I × VW I + FW (2.3.5)
dt W m
where ω W I is the angular velocity of the wind axis system with respect to the inertial axis
system. The above equation is now rewritten in matrix format with all terms coordinated
in the wind axis system. The cross product can be simplified by making use of Equation
A.2.2,
˙     
V̄ 0 −RW QW V̄ XW
  1 
 0  =  RW 0 −PW   0  +  YW  (2.3.6)
   
m
0 −QW PW 0 0 ZW
where V̄ is the magnitude of the aircraft’s velocity vector, XW , YW and ZW are the com-
ponents of the force vector applied to the aircraft, and PW , QW and RW are the roll,
pitch and yaw rates of the wind axis system with respect to the inertial axis system. The
equation above can be split into the dynamic equation for the velocity magnitude in wind
axes,
1
V̄˙ = XW (2.3.7)
m
and two algebraic constraint equations,
" # " #
RW 1 YW
= (2.3.8)
QW mV̄ −ZW
The velocity dynamics of Equation 2.3.5 are coordinated in the wind axis system. In order
to arrive at the position dynamics, the velocity vector must therefore be transformed
back to the inertial axis system. This is accomplished by making use of the direction
cosine matrix (DCM) defined in Equation A.3.5. The position dynamics of Equation 2.3.1
are rewritten as,
T W I
ṖW I
= DCMW I VW

I (2.3.9)
After simplification, the position dynamics become,

   
Ṗx cos (ψW ) cos (θW )
Ṗy  =  sin (ψW ) cos (θW )  V̄ (2.3.10)
   
Ṗz − sin (θW )
where Px , Py and Pz are the position coordinates of the wind axis system in inertial space,
also often denoted as N (north), E (east) and D (down), and ψW and θW are the yaw and
pitch angles of the wind axis system with respect to the inertial axis system.
2.3.1.2 Attitude and Attitude Dynamics
Attitude Amongst the various methods available for describing the attitude of one axis
system relative to another, Euler angles are chosen for their simplicity and intuitive
nature. Euler angles have a disadvantage over other methods such as Quaternions in
the sense that they exhibit a singularity under certain orientations [14]. However, by
selecting the appropriate Euler angle sequence, the singularity can be positioned such
that it will never be encountered during conventional flight. To this end, the Euler 3-2-1
angle sequence will be used.
Euler angles use three angles and a specific order of rotation to describe the attitude
of the wind axis system with respect to the inertial axis system. The two axis systems
are initially aligned. The wind axis system is then rotated positively about the ZW -axis
through the yaw angle (ψ ), followed by a positive rotation about the YW -axis through the
pitch angle (θ), and finally a positive rotation about the XW -axis through the roll angle
(φ).
Attitude Dynamics The following dynamic equation describes the relation between
the angular rates of the wind axis system, and the time derivatives of the Euler angles
which represent the attitude of the wind axis system with respect to the inertial axis
system [2],
     
φ̇ 1 sin φ tan θ cos φ tan θ PW
π
 θ̇  = 0 cos φ − sin φ  QW  |θ| 6= (2.3.11)
     
2
ψ̇ (321) 0 sin φ sec θ cos φ sec θ (321) RW
where the subscript (321) denotes the Euler 3-2-1 angle sequence, and PW , QW and
RW are the roll, pitch and yaw rates of the wind axis system. It can be seen that the
Euler 3-2-1 sequence exhibits a singularity at such a large pitch angle that it will not be
encountered during conventional flight.
2.3.2 Kinetics
Kinetics describe how the various forces and moments acting on a body translate into
linear and angular accelerations. These accelerations then propagate over time into
velocity, position and attitude in the kinematic equations of §2.3.1.
2.3.2.1 Rigid Body Rotational Dynamics
The rigid body rotational dynamics of the aircraft describe the rotational motion of the
body axis system relative to the wind axis system as a result of the moments applied
to the aircraft. These dynamics together with the point mass dynamics of §2.3.1.1 then
completely describe the six degree of freedom motion of the aircraft.
The kinetic relationship between an applied external moment (M) and the aircraft’s an-
gular momentum (H) is given by [15],
d
M= H (2.3.12)
dt I
where, Z
rdmB × ω BI × rdmB dm

H= (2.3.13)
v
where rdmB is the position vector of a mass element dm relative to the center of mass
of the rigid body with volume v , and ω BI is the angular velocity of the body axis system
with respect to the inertial axis system.
As argued by [2], the angular momentum vector takes on its simplest form when coordin-
ated in body axes. Because the aircraft is regarded as a rigid body, the moment arm to all
mass elements will remain fixed and independent of other translation or rotation motion
variables. By making use of Equation A.1.1, Equation 2.3.12 can be rewritten as,
d
M= H + ω BI × H (2.3.14)
dt B
Combining Equation 2.3.13 and Equation 2.3.14 and rearranging as shown by [2], the
dynamics of the body axis system angular velocity with respect to the inertial axis system
can be found,

ω̇BBI = I−1
B MB − SωBBI IB ωBBI (2.3.15)
where IB is the moment of inertia matrix referenced in body axes as shown in Equation
A.4.2, and SωBI implements the cross product as given by Equation A.2.2.
B
The orientation of the wind axis system relative to the body axis system is defined by
two rotations as shown in §2.1.3. The first is a negative rotation around the YB -axis
through the angle of attack (α), followed by a positive rotation around the new ZW -axis
(see Figure 2.3) through the angle of sideslip (β ). The angular velocity vector of the wind
axis system with respect to the body axis system can therefore be written as the sum of
the angular rates (α̇ and β̇ ) about their respective unit vectors,
ω W B = −α̇jB + β̇kW (2.3.16)
The angular velocity of the wind axis system with respect to the inertial axis system
(ω W I ) can be thought of as the sum of the angular velocity of the body axis system with
respect to the inertial axis system (ω BI ) and the angular velocity of the wind axis system
with respect to the body axis system (ω W B ),
ω W I = ω BI + ω W B (2.3.17)
Substituting Equation 2.3.16 into Equation 2.3.17 and analysing in the body axis system
gives,
ωBBI = α̇jB W WI
B − β̇kB + ωB (2.3.18)
This equation can be rewritten by making use of the DCM of Equation A.3.3 to convert
the vectors to the desired axis systems,
BW W
ωBBI = α̇jB
B − β̇DCM kW + DCMBW ωW
WI
(2.3.19)
Expanding and rearranging the equation above yields,

      
P 0 sin α " # cos α cos β − cos α sin β − sin α PW
 α̇
Q = 1 0  +  sin β cos β 0  QW  (2.3.20)
     
β̇
R 0 − cos α sin α cos β − sin α sin β cos α RW
where P , Q and R are the roll, pitch and yaw rates of the body axis system, respectively,
with respect to the inertial axis system. The equations are now rearranged so as to
make α̇, β̇ and PW the subject of the formula, while the two algebraic constraints from
Equation 2.3.8 (for QW and RW ) are substituted in. The resultant equation is combined
with the angular velocity dynamics of Equation 2.3.15 to form the complete rigid body
rotational dynamics,
 
" # " # P " #" #
α̇ − cos α tan β 1 − sin α tan β   1 sec β 0 ZW
= Q + (2.3.21)
β̇ sin α 0 − cos α mV̄ 0 1 YW
R
       
Ṗ L 0 −R Q P
−1 
Q̇ = IB M  −  R 0 −P  IB Q (2.3.22)
      
Ṙ N −Q P 0 R
with constraint,
 
i P
" #
h i h 1 h i Z
W
PW = cos α sec β 0 sin α sec β Q + − tan β 0 (2.3.23)
 
mV̄ YW
R
The rigid body rotational dynamics above maintain the attitude of the body axis system
with respect to the wind axis system over time (α and β ), as a function of the components
of the applied moment vector in the body axis system (L, M , N ) and the lateral and
normal force vector coordinates in wind axes (YW and ZW ).
2.3.3 Forces and Moments
The forces and moments acting on the aircraft encompass all of the aerodynamic, thrust
and gravitational effects on the aircraft. The sum of the different forces and moments
acting on the aircraft result in linear and angular accelerations which propagate into
velocity, position and attitude over time through the rigid body rotational dynamics and
point mass kinematics of §2.3.1.1 and §2.3.2.1.
The equations of motion derived in §2.3.1.1 and §2.3.2.1 require the force vector coordin-
ated in the wind axis system and the moment vector coordinated in body axis system. The
total forces and moments acting on the aircraft can therefore be written as,
FW = FA + FT + FG (2.3.24)
MB = MA + MT + MG (2.3.25)
where FW is the total force vector coordinated in the wind axis system, MB is the total
moment vector coordinated in the body axis system and the superscripts A, T and G
denote the aerodynamic, thrust and gravitational components of the force and moment
vectors.
2.3.3.1 Aerodynamic Forces and Moments
The aerodynamic forces and moments are those created due to the movement of air
across the aircraft’s various surfaces. These forces and moments are modelled in the
wind axis system [14],
 A  
XW − CD
 A
 YW  = qS  Cy  (2.3.26)
 
A
ZW −CL
 A    
LW b 0 0 Cl
 A
MW  = qS 0 c̄ 0 Cm  (2.3.27)
  
A
NW 0 0 b Cn
where,
1
q = ρV̄ 2 (2.3.28)
2
Here, q is the dynamic pressure, S the wing area, b the wing span, c̄ the mean aero-
dynamic chord, ρ the air density and V̄ the airspeed magnitude. The dimensionless
coefficients CD , Cy and CL are the drag, lift and side-force coefficients, respectively, with
Cl , Cm and Cn the dimensionless roll, pitch and yaw moment coefficients, respectively.
It follows from the equations above that the aerodynamic forces and moments are pro-
portional to the dynamic pressure, which in turn is a function of the air density and the
airspeed magnitude. The aerodynamic forces are the product of dynamic pressure, wing
area and the dimensionless force coefficients. Similarly, the aerodynamic moments are
products of dynamic pressure, wing area, the dimensionless moment coefficients, as well
as a moment arm. This moment arm is the mean aerodynamic chord in the case of the
pitching moment, and the wing span in the case of the roll and yaw moments.
The equations of motion derived in §2.3.2.1 require the moment vector to be coordinated
in the body axis system. Transforming the aerodynamic moment vector from the wind
axis system to the body axis system can be accomplished by making use of the DCM of
Equation A.3.3,
   A 
LA h i LW
 A BW  A
M  = DCM MW  (2.3.29)
NA A
NW
where LA , M A and N A are the aerodynamic moment vector coordinates in the body axis
system.
The dimensionless aerodynamic force and moment coefficients are defined by using di-
mensionless stability and control derivatives. The stability and control derivatives de-
scribe a change in a force or moment due to a change in a normalised motion variable
or actuator deflection. The aerodynamic force and moment coefficients can therefore be
defined as the sum of the various aircraft motion variables and actuator deflections as
given in [16],
CL2
CD = CD0 +
A
π e
 
(2.3.30)
α
" # " # " # β 

b b

Cy 0 0 Cyβ 2V̄a CyP 0 C
2V̄a yR  
 
= + c̄ P  (2.3.31)
CL CL0 CLα 0 0 2V̄a
C L Q
0
Q
 
R
 
" # δA 
CyδA 0 CyδR 0 δE 
+  
0 CLδE 0 CLδF  δR 

δF
 
α
     b b
 
Cl 0 0 Clβ 2V̄a ClP 0 C
2V̄a lR
β 
c̄  
Cm  = Cm0  + Cmα 0 0 C 0 P  (2.3.32)
    
2V̄a m Q   
b b
Cn 0 0 Cnβ 2V̄a CnP 0 C Q
 
2V̄a nR  
R
 
  δA
ClδA 0 C l δR 0  
 δE 
+ 0 CmδE 0 CmδF  
 
δ 
R
CnδA 0 CnδR 0
 
δF
where e is the Oswald efficiency factor, A is the aspect ratio of the wing, C L0 is the static
lift coefficient and Cm0 is the static moment coefficient. The following notation is used
for the non-dimensional stability and control derivatives,
∂CA
CAB = (2.3.33)
∂B 0
with,
B 0 = nB (2.3.34)
where n is the applicable normalising coefficient of B . The normalising coefficient for

c̄
incident and control deflection angles is unity, for the pitch rate it is 2V̄
and for the roll
b
and yaw angles it is 2V̄
.
To arrive at the model above, it is assumed that the added mass effects [14] as well as
the effects of downwash lag [17] of the main wing on the horizontal tail are negligible for
control design purposes. It is also assumed that the aircraft will fly at small incidence
angles, therefore the stability and control derivatives are considered to be independent
of the rigid body rotational dynamics.
The stability and control derivatives specific to the aircraft used in this project were
obtained by modelling the aircraft in the vortex lattice program Athena Vortex Lattice
(AVL), the results of which are given in Appendix B.6.3.
2.3.3.2 Thrust Model
The aircraft used in this project is equipped with a brushless DC electrical motor and
propeller (see Appendix B.1.1). This propulsion source can be modelled by a simple first
order lag (low pass filter) from thrust command to thrust output [8]. This accounts for
the significantly band-limited response of most propulsion sources. The dynamic effect
of airspeed on the thrust output is ignored since its effect is often negligible for control
purposes [2]. The thrust model is therefore given as,
1 1
Ṫ = − T + Tc (2.3.35)
τT τT
where T is the thrust in Newton, Tc is the thrust command and τT is the time constant of
the thrust dynamics. The time constant τT is determined experimentally by commanding
a step of the aircraft’s throttle and then analysing the resulting thrust [1].
The motor of the aircraft used in this project is mounted such that the thrust vector acts
through the center of mass and along the XB -axis. Therefore, the motor creates a force
along the XB -axis equal to the thrust output, while all other thrust forces and moments
are zero,
 T  
X T
 T  
Y  =  0  (2.3.36)
ZT 0
 T  
L 0
 T  
M  = 0 (2.3.37)
NT 0
The equations of motion derived in §2.3.1.1 and §2.3.2.1 require the force vector coordin-
ated in the wind axis system, therefore the body axis thrust vector is coordinated into
the wind axis system as follows,

 T  
XW cos α cos β
 T 
 YW  = − cos α sin β  T (2.3.38)

T
ZW − sin α
2.3.3.3 Gravitational Forces and Moments
The gravitational field is assumed to be uniform and acting in the direction of the ZI -axis
of the inertial axis system,
 G  
XI 0
 G  
 YI  =  0  (2.3.39)
ZIG mg
where g is the gravitational force per unit mass (m). The moments produced by the
gravitational force are all zero since it acts through the center of mass of the aircraft.
The equations of motion derived in §2.3.1.1 and §2.3.2.1 require the force vector co-
ordinated in the wind axis system, therefore the inertial axis system gravitational vector
is coordinated into the wind axis system by means of the DCM of Equation A.3.5,
 G  
XW h i 0
 G WI 
 YW  = DCM  0  (2.3.40)

G
ZW mg
2.4 Simplifying and Decoupling the Model
According to [2] the rigid body rotational dynamics of §2.3.2.1 can be simplified by ig-
noring the inertial cross coupling terms of Equation 2.3.22. These cross coupling terms
are negligible during normal flight where the very high angular velocities required to
see the effect of the cross coupling terms are not encountered. The rigid body rotational
dynamics of Equations 2.3.21 and 2.3.22 therefore become,
 
" # " # P " #" #
α̇ − cos α tan β 1 − sin α tan β   1 sec β 0 ZW
= Q + (2.4.1)
β̇ sin α 0 − cos α mV̄ 0 1 YW
R
  1  
Ṗ Ix
0 0 L
1
Q̇ =  0 Iy 0  M  (2.4.2)
    
Ṙ 0 0 I1z N
The aerodynamic, thrust and gravitational forces and moments of §2.3.3 are now summed
according to Equations 2.3.24 and 2.3.25 and become,
       
XW − CD cos α cos β − sin θ
 YW  = qS  Cy  + − cos α sin β  T +  cos θ sin φ  mg (2.4.3)
       
ZW −CL − sin α cos θ cos φ

    
L h i b 0 0 Cl
BW
M  = DCM qS 0 c̄ 0 Cm  (2.4.4)
    
N 0 0 b Cn
The dimensionless aerodynamic force and moment coefficients of Equations 2.3.30 to
2.3.32 are now substituted into the equations above, and the result is in turn substituted
into the simplified rigid body rotational dynamics of Equations 2.4.1 and 2.4.2. The
assumption is made that the wind axis system moments can be used without conversion
to the body axis system since, for small incidence angles, the inaccuracy added by the
assumption is far less than the inherent uncertainty in the aerodynamic model [2]. These
steps yield,
   1 1
 
α̇ − mV̄
sec β L̄α 1− mV̄
sec β L̄Q 0 − cos α tan β − sin α tan β α
1 1
Q̇  Mα MQ 0 0 0  Q
    
   Iy Iy  
 β̇  =  1 1 1
   0 0 mV̄
Ȳ β sin α + mV̄
Ȳ P − cos α + mV̄
Ȳ R
 β 
 
1 1 1
Ṗ   0 0 L L L  P 
    
Ix β Ix P Ix R
1 1 1
Ṙ 0 0 N
Iz β
N
Iz P Iz R
N R
  
− m1V̄ sec β sin α − m1V̄ sec β L̄δE 0 0 − m1V̄ sec β L̄δF T
1 1
0 Mδ E 0 0 Mδ F  δE 
  

 Iy Iy  
1
+  − mV̄ cos α sin β 1 1   δA 
 0 Ȳ
mV̄ δA
Ȳ
mV̄ δR
0  
1 1
0 0 L L 0   δR 
  
Ix δ A Ix δ R

1 1
0 0 N
Iz δ A
N
Iz δR
0 δF
 
− m1V̄ sec βqSCL0 + m1V̄ sec β cos θ cos φmg
1
qSc̄Cm0
 

 I y


+ 1 (2.4.5)
 mV̄
cos θ sin φmg 

0
 
 
0
where use has been made of the dimensional derivative notation as outlined in Appendix
B.6.4 in order to simplify the representation of the equation. Several assumptions are
now made to further simplify and decouple the model [2]:
a) The standard trigonometric small angle assumptions are made with regard to the two
incidence angles α and β . The assumption is also made that the product of small
angles is negligible (see Appendix A.5).
b) It is assumed that during normal flight, the angle of sideslip is negligibly small as long
as the control system enforces coordinated turns. Therefore, in the dynamic equation
for the angle of attack, the angle of sideslip is regarded as zero, thereby ignoring the
nonlinear couplings to the roll and yaw rates.
c) The zero angle of attack lift and pitching moment coefficients CL0 and Cm0 are re-
moved from the model under the assumption that the controllers will remove any
steady state errors caused by their omission.
d) The coupling of the thrust into the normal and lateral dynamics is ignored. This as-
sumption is made on the basis of the large timescale separation which exists between
the throttle dynamics, and the normal and lateral dynamics.
Taking into account these simplifications, the aircraft’s dynamics can be decoupled into
axial, normal and lateral dynamics,
Axial
1 1
Ṫ = − T+ TC (2.4.6)
τT τT

1 qS
AW = T + − CD − g sin θ (2.4.7)
m m
Normal
" # " #" # " #" # " #
1 1 g
α̇ − L̄
mV̄ α
1− mV̄
L̄Q α − m1V̄ L̄δE − m1V̄ L̄δF δE cos θ cos φ
= 1 1
+ 1 1
+ V̄
Q̇ Iy
Mα Iy
MQ Q Iy
M δ E Iy
M δ F δ F 0
(2.4.8)
" # " #
h i h i α h i δ
1 1 1 E
CW = − L̄
m α
− m L̄Q + − L̄
m δE
− m1 L̄δF (2.4.9)
Q δF
Lateral
   1 1 1
   1 1
 g 
β̇ Ȳ
mV̄ β
Ȳ
mV̄ P
−1 + mV̄
ȲR β Ȳ
mV̄ δA
Ȳ
mV̄ δR
" # cos θ sin φ
V̄
   1 1 1    1 1  δA
Ṗ  =  Ix Lβ L L  P  +  Ix LδA L + 0
 
Ix P Ix R Ix δ R  
1 1 1 1 1
δR
Ṙ N
Iz β
N
Iz P
N
Iz R
R N
Iz δA
N
Iz δ R
0
(2.4.10)
 
" # " Ȳ # β " Ȳ ȲδR
#" #
β ȲP ȲR δA
BW m m m m m
δA
= P  + (2.4.11)
 
PW 0 1 0 0 0 δR
R
Chapter 3
Longitudinal Analysis and Control
During landing, it is intuitive that the control of the aircraft occurs primarily in its lon-
gitudinal axis. Therefore, in order to land accurately, it is essential that the longitudinal
motion of the aircraft is controlled accurately. The analysis of the longitudinal dynamics
of the aircraft and the subsequent design of appropriate controllers will focus on this
point of departure.
During landing the aircraft must essentially follow an altitude reference path which
guides it down to the runway. The strategy to achieve this longitudinal control involves
designing two sets of controllers. The first set is a series of controllers based around the
longitudinal aerodynamic control surfaces (elevator δE and flaps δF ), and the second is a
controller based around the throttle actuator (TC ). The two sets of controllers are shown
highlighted in Figure 3.1 where the first set consists of an altitude (h), climb rate (ḣ) and
normal specific acceleration (CW ) controller, and the second consists of an airspeed (V̄ )
controller.
Figure 3.1 – Longitudinal control system overview
23
CHAPTER 3. LONGITUDINAL ANALYSIS AND CONTROL 24
The normal specific acceleration (NSA) controller is designed around the linear, de-
coupled normal dynamics of Equations 2.4.8 and 2.4.9 as proposed by [2]. This control-
ler possesses a high bandwidth and encapsulates all of the aircraft specific uncertainty.
Since the NSA controller operates at the acceleration level it has the benefit of rejecting
disturbances before they appear in the slower dynamics as an error in climb rate and
altitude. This controller is capable of using the aircraft’s elevator and flaps in order to
create a commanded NSA, and receives its input from the climb rate controller. The
climb rate controller creates a commanded climb rate by commanding the NSA control-
ler, and receives its input from the altitude controller. In turn, the altitude controller
creates a commanded altitude by commanding the climb rate controller. The airspeed
controller is independent of the other controllers, and makes use of the aircraft’s throttle
in order to create a commanded airspeed. Both the airspeed and the altitude controllers
receive their input from the landing state machine as discussed in Chapter 5.
As mentioned in §2.2, the flaps actuator is of special significance in this thesis. The
mechanism through which it is used is known as Direct-Lift Control (DLC). DLC offers
the prospect of an improvement in the precision with which aircraft height and vertical
velocity can be controlled [12]. It does so by addressing some of the limitations faced by
traditional longitudinal motion control, as discussed in §3.1.
This chapter explores the concept of DLC and how it could be applied to this thesis. This
is followed by the design, HIL simulation, and practical flight test results of the NSA,
climb rate, altitude and airspeed controllers.
3.1 Direct-Lift Control
If it is assumed that the airspeed of a conventional fixed-wing aircraft remains con-

stant, then longitudinal control of the aircraft is essentially accomplished by varying
the amount of lift produced by the aircraft as a whole. For a conventional aircraft, this
is realised by means of a rear elevator. An elevator deflection by itself produces some
change in lift of the tailplane, however, this lift is relatively small and in the opposite
sense to what is ultimately intended. This lift does however act with a relatively large
moment arm with respect to the centre of gravity of the aircraft, thereby causing a sig-
nificant pitching moment. This pitching moment can change the angle of attack of the
aircraft and thereby change the amount of lift produced by the aircraft. Conventional
elevator-based longitudinal control is therefore accomplished by α-generated lift pro-
duced primarily by the wing of the aircraft. This is also known as pitch-moment-based
longitudinal control.
Another mechanism by which longitudinal control can be accomplished is DLC. DLC util-
ises a control mechanism by which lift is generated without, or largely without, a signific-
ant change of aircraft incidence, and ideally should not generate pitching moment [12].
The flaps are an example of an actuator with inherent direct-lift properties. Operating
the flaps causes a change in the camber of the wing, which results in an instantaneous
change in the lift of the wing. Since the flaps are located close to the centre of gravity of
the aircraft, the corresponding moment produced is relatively small when compared to
actuators such as the elevator.
An important difference between pitch-moment-based and direct-lift-based longitudinal

control is the transient response to a control input. In the case of pitch-moment-based
control, the transient response is largely defined by the short period mode of the aircraft.
There is in essence a second order lag imposed between a control input and achieving
the desired output. This lag is proportional to the period of the short period mode. It is
not unreasonable to imagine that this lag could have an adverse effect on any attempts
to control the longitudinal motion aircraft very precisely. In the case of an idealised pure
direct-lift control, [12] argues that the response is in essence only defined by the amount
of lift commanded by the actuator. In this case the aircraft can simply be considered a
mass with an applied force. This force causes a normal acceleration, which in turn causes
a vertical velocity and finally a height change. The prospect of controlling the aircraft’s
height and vertical velocity independent of any pitch rate dynamics holds great promise
for improvements in the precision with which these parameters can be controlled. Such
improvements could impact directly on the accuracy with which the landing altitude
reference path is followed, and consequently an improvement in the accuracy with which
the landing touchdown point can be reached.
The design and implementation of a practically feasible DLC system is more complicated
than the idealised scenario described, would suggest. The complexity arises from the
fact that a practical direct-lift control surface, such as the flaps, also generates a pitch
moment in addition to lift. These adverse effects are evaluated further during the design
of the NSA controller in §3.2.
3.2 Normal Specific Acceleration Controller
This section describes the design, simulation and practical testing of a NSA controller
based on the principles of DLC as outlined in §3.1. The NSA of the aircraft acts normal
to the wing along the ZW -axis of the wind axis system. The NSA controller is designed
around the linear, decoupled normal dynamics of Equations 2.4.8 and 2.4.9 as proposed
by [2]. This controller possesses a high bandwidth and encapsulates all of the aircraft
specific uncertainty. Since the NSA controller operates at the acceleration level it has
the benefit of rejecting disturbances before they appear in the slower dynamics as an
error in climb rate and altitude. The input to the NSA controller stems from the climb
rate controller of §3.3, which creates a commanded climb rate by commanding the NSA
controller. The NSA controller then makes use of the aircraft’s elevator and flaps to
create the commanded NSA.
Previous implementations of this type of NSA controller [2; 1] made use of only the
aircraft’s elevator as the control surface. As discussed in §3.1, it is expected that with
the addition of the flaps as a control surface, a direct-lift NSA controller can be designed
which will show improved performance over the conventional NSA controller.
It must be noted that the controller design is not specific to the use of flaps as the direct-
lift actuator. In general, any actuator or actuator combination which causes a change
in lift force with a minimal change in pitch moment can be used. These could include
spoilers and hybrid actuators such as spoilerons or flaperons (where the ailerons can
move both differentially and in tandem).
3.2.1 Overview and Strategy
Several approaches to implement a direct-lift control system are discussed by [12], the
most promising of which is a method of using DLC systems with relatively restricted
authority. This involves combining traditional pitch-moment-based longitudinal control
with direct-lift-based longitudinal control in order to form a hybrid longitudinal motion
controller. The direct-lift control part of this hybrid controller only has a limited author-
ity, whereas the pitch moment control part can respond to commands beyond the point
where the direct-lift control authority is exhausted.
It is suggested by [12] that the most efficient use of a limited amount of direct-lift would
be to superimpose it in transient form upon conventional moment-based control. This
essentially means that the NSA controller signals must be filtered such that the direct-
lift portion of the controller will only respond to higher frequency signals, whereas the
moment-based portion of the controller will respond to lower frequency and steady state
signals. In such a configuration the long term longitudinal response of the aircraft will
remain as without direct-lift control, only the initial response will be affected by the
direct-lift.
To illustrate this control concept, consider a step input to the NSA controller. The fil-
ter passes the higher frequency part of this signal to the direct-lift portion of the NSA
controller. The direct-lift portion of the controller then allows the aircraft to achieve an
immediate normal acceleration. Assuming that the direct-lift actuator possesses suffi-
cient control authority, the commanded acceleration will be achieved. The effect of the
filter is then to return the direct-lift actuator to its neutral position since the control
signal now no longer contains any higher frequency components. Simultaneously, the
lower frequency part of the control signal is passed to the moment-based portion of the
controller. The moment-based actuator then simply holds the steady state acceleration
demand. In the absence of the direct-lift portion of the controller, the nature of the re-
sponse is in essence a second order lag which is proportional to the period of the short
period mode. The effect of the direct-lift is thus to speed up the transient response to
a control demand. This is also the case when the direct-lift actuator does not possess
sufficient control authority to achieve the commanded acceleration. In such a case, the
response is a combination between the direct-lift and moment-based control responses.
An additional advantage of a direct-lift-augmented NSA controller is that, for small nor-

mal acceleration demands, there is less pitch angle variation than for a conventional
moment-based controller. This attribute is especially useful in the presence of disturb-
ances such as wind gusts, as the disturbance can be counteracted without affecting the
pitch angle (and therefore the flight path angle) of the aircraft [18]. This results in
improved altitude path following ability when compared to moment-based longitudinal
control [19].
For a DLC system with relatively restricted authority, the strategy proposed by [12] is to
use the moment-based control actuator to cancel out any unwanted moment produced
by the DLC actuator. The NSA controller design presented makes use of the elevator
as the pitch-moment-based control actuator, and the flaps as the direct-lift-based control
actuator. Therefore, by using the elevator to cancel out the moment produced by the
flaps, the flaps can be approximated as a pure direct-lift actuator.
In general, making use of two control actuators simultaneously to control the same para-
meter leads to the inherent problem of proper control allocation. However, for the
strategy proposed in this section, this is overcome through the frequency separation
between the elevator- and flaps-based portions of the controller. This frequency separa-
tion is the result of a complementary filter pair on the NSA controller signals and allows
the two portions of the controller to be designed independently, before simply being
superimposed.
Based on the preceding discussion, the NSA controller can be presented conceptu-
ally as shown in Figure 3.2. Both the commanded and the actual NSA signals are
passed through a complementary filter pair. The low frequency parts of signals enter
the elevator-based portion of the NSA controller, while the high-frequency parts enter
the flaps-based portion. The resultant elevator and flaps deflections are then inputs to
the NSA dynamics of the aircraft. The true NSA of the aircraft is again passed through
the complementary filter pair before being fed back to the controllers. The gain KM rep-
resents the gearing required between the flaps and the elevator in order for the elevator
to cancel out any unwanted moment produced by the flaps.
Figure 3.2 – Block diagram of NSA controller strategy
3.2.2 Investigation of the Elevator and Flap Actuators
In the control strategy of §3.2.1 it is stated that the aircraft’s elevator is used as a
moment-based longitudinal actuator and the flaps as a direct-lift-based longitudinal ac-
tuator. In this section these actuators are studied further to confirm their suitability for
this controller design.
The longitudinal actuator configuration is shown in Figure 3.3 where LW , LT and LTotal
are the lift forces of the wing, tailplane and the total lift force of the aircraft respectively,
acW and acT the aerodynamic centre of the wing and tailplane respectively and NP the
neutral point of the aircraft.
Figure 3.3 – Longitudinal actuator configuration of the aircraft1
To gauge the relative effects of the longitudinal actuators, it is necessary to determine

the dimensionless stability and control derivatives of the aircraft. This is accomplished
by modelling and analysing the aircraft in the vortex lattice program AVL (see Appendix
B.6.3). The stability and control derivatives obtained describe a change in a force or
1
moment due to a change in a normalised motion variable or actuator deflection. The

derivatives of interest in the design of the NSA controller are contained in Tables B.3
and B.5, and summarised in Table 3.1 for convenience.
Lift Force Value Pitch Moment Value

CLα 4.808411 Cm α -0.664939
CLQ 7.812170 Cm Q -7.438796
CLδE 0.456085 CmδE -0.957351
CLδF 1.078236 CmδF 0.172092
Table 3.1 – Longitudinal stability and control derivatives
From the relative values of the derivatives, it can be seen that for an equivalent de-
flection angle, the flaps generate more than twice the amount of lift than the elevator.
Similarly it can be seen that moment generated by a flap deflection is more than five
times less than that of the elevator. It can also be seen that the lift generated directly by
an elevator deflection is in the opposite direction to what is intended. These attributes
are in agreement with those of direct-lift-based and moment-based longitudinal control
actuators as discussed in §3.1. It can also be safely assumed that the elevator possesses
sufficient margin with which to counter any moments produced by a flaps deflection as
discussed in §3.2.1.
It was shown in [4; 6; 8] that the stability and control derivatives generated by AVL are
sufficiently accurate to enable the design of the aircraft’s controllers. However, there is
some uncertainty as to the direction of the flaps moment coefficient. With reference to
Figure 3.3, the coefficients in Table 3.1 show that a positive flaps deflection will increase
the lift of the aircraft whilst also generating a positive pitching moment. However, the
actual pitching moment produced depends on how the change in lift and drag caused by
the flaps deflection is positioned around the centre of gravity of the aircraft. It is there-
fore possible that the direction of the flaps moment coefficient can change for a different
aircraft mass distribution. An additional complication is the difference in downwash on
the tailplane between a small and a large flaps deflection. It is thus necessary to confirm
the nature of the flaps moment coefficient through practical flight tests.
3.2.3 Investigation of the Natural NSA Dynamics
In this section the natural NSA dynamics are studied in order to gain further insight into
the behaviour of the aircraft in respect of both elevator and flaps control deflections.
This insight will be required in the controller designs of §3.2.4 and §3.2.5.
The decoupled model of the NSA dynamics from Equations 2.4.8 and 2.4.9 are restated
below for convenience,
" # " #" # " #" # " #
g
α̇ − m1V̄ L̄α 1 − m1V̄ L̄Q α − m1V̄ L̄δE − m1V̄ L̄δF δE cos θ cos φ
= 1 1
+ 1 1
+ V̄
Q̇ Iy
M α Iy
M Q Q Iy
M δE Iy
M δF δF 0
(3.2.1)
" # " #" # " #" #
CW − m1 L̄α − m1 L̄Q α − m1 L̄δE − m1 L̄δF δE
= + (3.2.2)
Q 0 1 Q 0 0 δF
The effect of gravitational acceleration in the system can be removed by means of direct
feedback linearisation as shown by [2]. However, provided that the flight path angle re-
mains small, the gravitational acceleration effect can be considered static [8]. The gravit-
ational acceleration term can thus be ignored under the assumption that the longitudinal
controllers will remove any steady state error caused by its omission. An assumption is
also made that,
L̄Q
1 (3.2.3)
mV̄
since its effect is negligibly small for most conventional aircraft [14]. After these simpli-
fications the normal dynamics are given by,
" # " #" # " #" #
α̇ − m1V̄ L̄α 1 α − m1V̄ L̄δE − m1V̄ L̄δF δE
= 1 1
+ 1 1
(3.2.4)
Q̇ Iy
M α Iy
M Q Q Iy
Mδ E Iy
Mδ F δF
" # " #" # " #" #
= + (3.2.5)
Q 0 1 Q 0 0 δF
To analyse the natural NSA behavior, the transfer function matrix is found as shown in
Appendix C.1, and can be written as,
1
G(s) = N(s) (3.2.6)
∆(s)
where N(s) is a polynomial matrix whose elements consist of the response transfer func-
tion numerators, and where ∆(s) is the characteristic polynomial common to all transfer
functions. For the NSA dynamics as shown in Equations 3.2.4 and 3.2.5 four transfer
functions exist,
 
CW CW
NδE (s) NδF (s)
1 
G(s) =   (3.2.7)
∆(s)  Q Q

NδE (s) NδF (s)
where NBA (s) is the response transfer function numerator of state variable A to control
input B . Calculating the characteristic equation and the response transfer function nu-
merators yields,

2 L̄α MQ L̄α MQ Mα
∆(s) = s + − s− + (3.2.8)
mV̄ Iy mV̄ Iy Iy
and,
 
L̄δ L̄Q MQ Mδ L̄α Mα Mδ L̄δ L̄Q MQ Mδ L̄α Mα Mδ
− m
E
s2 − Iy L̄Q
− L̄δ
E
s− Iy L̄α
− L̄δ
E
− m
F
s2 − Iy L̄Q
− L̄δ
F
s− Iy L̄α
− L̄δ
F

E E F F
N(s) =  Mδ

L̄δ

Mδ

L̄δ

 E L̄α MαE F L̄α Mα
F

Iy
s+ mV̄
− mV̄ MδE Iy
s+ mV̄
− mV̄ MδF
(3.2.9)
It follows from Equation 3.2.8 that the normal dynamic poles are not affected by the lift
due to pitch rate (L̄Q ) or either of the control deflections (δE and δF ). It can also be
seen in Equation 3.2.9 that the response transfer functions for the elevator and flaps are
identical except for the respective control deflection lift and pitching moment derivat-
ives. The NSA response to both the elevator and flaps control input is further looked at
in the following sections.
3.2.3.1 NSA Response to Elevator Input
The transfer function for the NSA response to an elevator input can be determined from
Equations 3.2.7 to 3.2.9, and is given by,
h i
L̄ L̄Q MQ MδE L̄α Mα MδE
NδCEW (s) − mδE 2
s − Iy L̄Q − L̄δ s− Iy L̄α
− L̄δE
GC
δE (s) =
W
= E (3.2.10)
∆(s) M
s2 + mL̄αV̄ − IyQ s − mL̄αV̄
MQ
+ Mα
Iy Iy
To gain a better appreciation for the nature of the transfer function, the values for the
aerodynamic derivatives, standard flight conditions and aircraft parameters (see Ap-
pendix B.6) are substituted into the above equation. Thus,
−9.65s2 + 85.88s + 8794 −9.65(s − 34.96)(s + 26.06)

GC
δE (s) =
W
2
= (3.2.11)
s + 13.04s + 106 (s + 6.52 + 7.97i)(s + 6.52 − 7.97i)
C
The location of the poles and zeros of GδEW (s) as well as the NSA response to an elevator
step input can now be determined as shown in Figures 3.4a and 3.4b. The damped pole
pair depicted in Figure 3.4a represents the short period mode of the aircraft. The short
period mode is a damped pitch oscillation of the aircraft which is excited whenever the
aircraft is disturbed from its pitch equilibrium state. It manifests itself as a classical
second order oscillation in which the principal variables are angle of attack (α), pitch
rate (Q) and pitch attitude (θ) [13]. The airspeed magnitude remains approximately
constant in the timescale of this mode due to the effects of inertia and momentum.
The right half plane zero depicted in Figure 3.4a indicates the non-minimum phase
(NMP) nature of the NSA response to an elevator input. The NMP nature results from
the small amount of lift produced by an elevator deflection where the lift is in the oppos-
ite sense to what is ultimately intended. The effect of this portion of the lift can be seen
in Figure 3.4b as the small initial dip in NSA which occurs before the dominant portion
of the response. As discussed in §3.1, the dominant portion of the response results from
the pitch moment created by that lift. This moment can change the angle of attack of
10 1.6
8 1.4
6
1.2
4
1
2
0.8
0
0.6
−2
0.4
−4
0.2
−6
−8 0
−10 −0.2
−40 −30 −20 −10 0 10 20 30 40 0 0.2 0.4 0.6 0.8 1
(a) Pole-zero map (b) Step response
Figure 3.4 – Pole-zero map and step response of the NSA to elevator input transfer function
the aircraft and thereby change the amount of lift produced by the aircraft. As shown
by [2], the NMP nature of the NSA dynamics can be ignored if the natural frequency of
both the open- and the closed-loop dominant poles of the system is at least three times
slower than that of the NMP zero. Figure 3.4a shows an approximate NMP zero location
of,
ZN M P ≈ 35 rad/s (3.2.12)
which is more than three times faster than the open-loop poles. If it is assumed that
the controller design of §3.2.4 will ensure that the closed-loop poles also meet the above
requirement, then the NMP zero can be ignored by assuming that,
CLδE = 0 (3.2.13)
This assumption reinforces the notion expressed in §3.2.1 that the elevator is used as a
pitch moment based actuator. In general, as the airspeed increases, the frequency of the
short period mode poles increases while their damping increases slightly. The two zeros
move further away from the origin as the airspeed increases, and the NMP zero remains
sufficiently far from the poles across the conventional airspeed range of the aircraft in
order for the assumption to remain valid.
A left half plane zero is also depicted in Figure 3.4a. This zero is located at a much higher
frequency than the poles of the NSA dynamics. As such, its effects will be negligible
due to the presence of other high-frequency poles, such as those from servo lag, which
remain unmodelled. This zero represents the lift generated by the aircraft due to the
induced angle of attack created when the aircraft is experiencing a pitch rate. It can
therefore be ignored by assuming that,
CLQ = 0 (3.2.14)
After the simplifications above, the NSA dynamic equations for an elevator input are
given by,
" # " #" # " #
α̇ − m1V̄ L̄α 1 α 0
= 1 1
+ 1 δE (3.2.15)
Q̇ I
M α I
M Q Q Iy
M δ E
" # " y #y" # " #
1
CW − m L̄α 0 α 0
= + δE (3.2.16)
Q 0 1 Q 0
In order to compare these simplified dynamics with the full dynamics of Equations 3.2.4
and 3.2.5, the transfer function for the NSA response to an elevator input of the simpli-
fied dynamics is determined,
M L̄α
NδCW (s) − IδyEm
GC
δE (s)
W
= E = (3.2.17)
∆(s) s2 + mL̄αV̄ −
MQ
s − L̄α MQ
+ Mα
Iy mV̄ Iy Iy
Substituting in the values for the aerodynamic derivatives, standard flight conditions and
aircraft parameters (see Appendix B.6) gives,
9414 9414
GC
δE (s) =
W
= (3.2.18)
s2 + 13.04s + 106 (s + 6.52 + 7.97i)(s + 6.52 − 7.97i)
The NSA response to an elevator step input for both the full and the simplified dynamics
can now be determined as shown in Figure 3.5. It can be seen that there is only a slight
difference in the open-loop step response between the full and simplified dynamics. The
most significant difference is the absence of the NMP effect in the response of the sim-
plified dynamics. The simplified dynamics of Equations 3.2.15 and 3.2.16 are therefore
used in the elevator-based portion of NSA controller design in §3.2.4.
1.8
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0 Full Model
Simplified Model
−0.2
0 0.2 0.4 0.6 0.8 1
Figure 3.5 – Step response of the full and simplified NSA to elevator input transfer functions
3.2.3.2 NSA Response to Flaps Input
The transfer function for the NSA response to a flaps input can be determined from
Equations 3.2.7 to 3.2.9, and is given by,
h i
L̄ L̄Q MQ MδF L̄α Mα M δF
NδCFW (s) − mδF s −2
Iy
− L̄Q
s−L̄δF Iy L̄α
−L̄δF
GC
δF (s) =
W
= (3.2.19)
∆(s) M
s2 + mL̄αV̄ − IyQ s − mL̄αV̄
MQ
+ Mα
Iy Iy
In order to gain a better appreciation for the nature of the transfer function, the values
for the aerodynamic derivatives, standard flight conditions and aircraft parameters (see
Appendix B.6) are substituted into the above equation. Thus,
−22.81s2 − 196.8s − 3158 −22.81(s + 4.31 + 10.95i)(s + 4.31 − 10.95i)

GC
δF (s) =
W
2
=
s + 13.04s + 106 (s + 6.52 + 7.97i)(s + 6.52 − 7.97i)
(3.2.20)
C
The pole-zero map as well as the NSA response to a flaps step input of GδFW (s) are shown
C
in Figures 3.6a and 3.6b. It can be seen in Figure 3.6a that the transfer function, GδFW (s),
displays a pair of complex poles as well as a pair of complex zeros, and that these are
located in the same general area on the pole-zero map. As discussed in §3.2.3.1, the
damped pole pair represent the short period mode of the aircraft. From Figure 3.6b it
can be seen that the NSA response to a flap step input consists of a quick initial step in
NSA followed by a slower rise to some final value. This means that, for a downwards
flap deflection, the aircraft will experience an instantaneous upwards NSA followed by
an upwards pitch rate which then causes the NSA to rise even more.
To understand the origin of this response, the NSA dynamics of Equations 3.2.4 and 3.2.5
must be considered for a step flap deflection. It can be seen in the output equation that
this flap deflection causes an instantaneous increase to NSA due to the lift produced by
a flap deflection term L̄δF . This is the direct-lift effect as described in §3.1 and can be
seen in Figure 3.6b as the initial jump in NSA. The L̄δF term also affects the angle of
attack directly. This angle of attack effect is relatively small, and its effect on the NSA is
in the opposite sense to that intended. It can be seen in Figure 3.6b as the small bump in
the NSA response directly after the direct-lift part of the response. The angle of attack
and pitch rate both affect one another and, in the absence of the moment produced by
the flaps, will cause the pitch rate to continue to rise in the opposite direction to that
intended and eventually settle at a constant value. The moment produced by the flaps,
as reflected by the MδF term, is the largest contributor to pitch rate. This pitch rate
causes a change in the angle of attack of the aircraft, and this then reflects as a change
in the NSA. This effect on the NSA is in the same direction as that intended and can be
seen in Figure 3.6b as the slow rise in NSA following the small bump in the response. At
a certain point, the moment due angle of attack and pitch rate, as reflected by the MQ
and Mα terms, becomes high enough to oppose further increase in the pitch rate. This
results in the response settling at a constant pitch rate and NSA. The effect of the lift
due to pitch rate term, L̄Q is negligibly small as was the case for the NSA response to
elevator input of §3.2.3.1.
15 0.1
0
10
−0.1
5
−0.2
0
−0.3
−5
−0.4
−10
−0.5
−15
−15 −10 −5 0 5 10 15 0 0.2 0.4 0.6 0.8 1
Figure 3.6 – Pole-zero map and step response of the NSA to flaps input transfer function
As stated in §3.2.2 the direction of the flaps moment coefficient CMδF can change for a
different aircraft configuration. This means that, for a downwards flap deflection, the
aircraft will experience an instantaneous upwards NSA followed by a downwards pitch
rate. This is in contrast to the scenario discussed previously, where a downwards flap
deflection causes the aircraft to pitch upwards. In order to illustrate the consequences of
this, the values for the aerodynamic derivatives, standard flight conditions and aircraft
parameters are again substituted into Equation 3.2.19. Thus,
−22.81s2 − 140.3s + 226 −22.81(s + 7.48)(s − 1.33)

GC
δF (s) =
W
2
= (3.2.21)
s + 13.04s + 106 (s + 6.52 + 7.97i)(s + 6.52 − 7.97i)
C
C
in Figures 3.7a and 3.7b. It can be seen in Figure 3.7a that the transfer function GδFW (s)
now possesses two real zeros instead of a pair of complex zeros as in Figure 3.6a. From
Figure 3.7b it can be seen that the NSA response to a flap step input consists of a quick
initial step in NSA followed by a slower rise to some final value which is in the opposite
direction to that intended. The previous discussion regarding the response of Equation
3.2.20 holds here as well. The only difference is that the moment produced by the flap
deflection is now in the opposite direction. This moment then causes the steady state
NSA response to be in the opposite direction than intended.
As discussed in the NSA controller strategy of §3.2.1, the NSA controller makes use of
the elevator to cancel out any moment produced by the flaps. If it is assumed that the
flap moment is cancelled out perfectly, then the NSA dynamic equations for a flaps input
15 0.1
0
10
−0.1
5
−0.2
0
−0.3
−5
−0.4
−10
−0.5
−15
−15 −10 −5 0 5 10 15 0 0.2 0.4 0.6 0.8 1
where the flaps moment coefficient is of opposite direction
become,
" # " #" # " #
α̇ − m1V̄ L̄α 1 α − m1V̄ L̄δF
= 1 1
+ δF (3.2.22)
Q̇ Iy
Mα Iy
MQ Q 0
" # " #" # " #
CW − m1 L̄α 0 α − m1 L̄δF
= + δF (3.2.23)
Q 0 1 Q 0
The transfer function for the NSA response to a flaps input of this simplified model is
given by,
h i
L̄δF M
NδCFW (s) − m
s2 − IyQ s − MIyα
GC
δF (s) =
W
= (3.2.24)
∆(s) s2 + L̄α M M
− IyQ s − mL̄αV̄ IyQ + Mα
mV̄ Iy
To gain a better appreciation for the nature of the transfer function, the values for the
aerodynamic derivatives, standard flight conditions and aircraft parameters (see Ap-
pendix B.6) are substituted into the above equation. Thus,
−22.81s2 − 168.6s − 1466 −22.81(s + 3.70 + 7.11i)(s + 3.70 − 7.11i)

GC
δF (s) =
W
= (3.2.25)
s2 + 13.04s + 106 (s + 6.52 + 7.97i)(s + 6.52 − 7.97i)
C
in Figures 3.8a and 3.8b. It can be seen in Figure 3.8b that the NSA response to a flap
input now consists of two parts, the first being the initial jump in NSA caused by the
direct-lift component of the flaps deflection. The second part is the result of the effect
of the flap deflection on the angle of attack. The angle of attack and pitch rate interact
in a way which lowers the angle of attack, eventually settling to some constant angle of
attack and pitch rate. The effect of this reduction in angle of attack is then to reduce
the NSA. It must be noted that if the moment produced by a flaps deflection is perfectly
cancelled out by the elevator, then the responses are identical for either direction of
15 0.1
0
10
−0.1
5
−0.2
0
−0.3
−5
−0.4
−10
−0.5
−15
−15 −10 −5 0 5 10 15 0 0.2 0.4 0.6 0.8 1
where the flaps moment is cancelled out
flaps moment coefficient CMδF . It is therefore possible to design a direct-lift controller

regardless of this coefficient’s direction.
As discussed in §3.2.1 the NSA controller strategy involves superimposing the flaps-
based direct-lift portion of the controller in transient form upon the elevator-based mo-
ment portion of the controller. The direct-lift portion of the controller only acts on higher
frequency NSA commands and therefore the long term response of the aircraft is entirely
conventional. This means that the direct-lift portion of the controller can be optimised to
simply yield an improved control response. One such optimisation involves changing the
gearing between the flaps and the elevator. This means that the elevator does not cancel
out the moment produced by the flaps perfectly, but rather increases or decreases the
net moment acting on the aircraft. In this way the response can be tuned to obtain a sat-
isfactory result. This is demonstrated in Figures 3.9a and 3.9b where the pole-zero map
and the NSA response to a flaps step input are shown for different moment gearings.
It can be seen in Figure 3.9b that increasing the gearing makes the steady state NSA
value shift upwards. This allows a satisfactory response to be found through simulation
as shown in §3.2.6.
3.2.4 NSA Controller Design - Elevator Actuator
This section presents the design of the elevator-based portion of the NSA controller. This
controller uses the aircraft’s elevator to create a commanded wind axes NSA through
the principles of moment-based longitudinal control as outlined in §3.1. As discussed
in section §3.2.1, this controller will be superimposed with the direct-lift based NSA
controller of §3.2.5. The detailed derivation of this controller is given in Appendix C.2.
15 0.1
0
10
−0.1
5
−0.2
0
−0.3
−5
−0.4
−10
−0.5
−15
−15 −10 −5 0 5 10 15 0 0.2 0.4 0.6 0.8 1
Figure 3.9 – Pole-zero map and step response of the NSA to flaps input transfer function for
a range of moment gearings
The simplified NSA dynamic equations from §3.2.3.1 are restated below,
" # " #" # " #
α̇ − m1V̄ L̄α 1 α 0
= 1 1
+ 1 δE (3.2.26)
Q̇ Iy
Mα Iy
MQ Q Iy
Mδ E
" # " #" # " #
1
CW − m L̄α 0 α 0
= + δE (3.2.27)
Q 0 1 Q 0
To remove steady state errors on NSA, the state vector is augmented with an integrator
xI which obeys the differential equation,
ref
ẋI = CW − CW (3.2.28)
ref
where CW is the reference NSA input. The augmented NSA dynamics are now given by,
        
ẋI 0 − m1 L̄α 0 xI 0 1
  ref
 α̇  = 0 − m1V̄ L̄α 1   α  +  0  δE − 0 CW (3.2.29)
      
Q̇ 0 I1y Mα 1
Iy
MQ Q 1
Iy
Mδ E 0
 
" # " # xI " #
CW 0 − m1 L̄α 0   0
= α + δE (3.2.30)
Q 0 0 1 0
Q
An elevator control law is defined which uses feedback from the integrator and pitch
rate states as well as feed-forward from the reference input. The integrator state feed-
back will ensure that any steady state errors are removed from the closed-loop system,
whereas the pitch rate feedback will allow the damping of the NSA dynamics to be se-
lected. The feed-forward from the reference input will allow a closed-loop zero to be
placed. This control strategy was proposed by [1] in response to difficulties that were en-
countered when using feedback directly from NSA. It was found that using NSA feedback
without filtering it first made the controller very sensitive to noise on the NSA measure-
ment. It was also found that attempts to increase the system’s natural frequency can
result in undesirable pole placements, due to unmodelled effects such as delays. The
elevator control law is thus given by,
 
h i xI
ref
δE = − KIδE 0 KQδE  α  + N̄CδE CW (3.2.31)
 
Q
The control law is now substituted into the dynamics and the closed-loop system be-
comes,
      
ẋI 0 − m1 L̄α 0 xI 1
 ref
 α̇  =  0 − m1V̄ L̄α 1 α −  0  CW
     
Q̇ − I1y MδE KIδE 1

Iy
Mα 1
Iy
1
MQ − Iy MδE KQδE Q 1
− Iy MδE N̄CδE
(3.2.32)
 
" # " # xI
CW 0 − m1 L̄α 0  
= α (3.2.33)
Q 0 0 1
Q
Calculating the closed-loop characteristic equation gives,

L̄α
3 MQ Mδ E 2 L̄α MQ L̄α MδE Mα
p(s) = s + − + KQδE s + − + KQδE − s (3.2.34)
mV̄ Iy Iy mV̄ Iy mV̄ Iy Iy
L̄α MδE
− K IδE
mIy
In order to place the closed-loop poles a desired characteristic equation for the NSA
dynamics is defined as,
αc (s) = (s2 + 2ζωn s + ωn2 )(s + a) (3.2.35)
where the complex pole pair corresponds to the short period mode and the single real
pole to the closed-loop integrator. Expanding Equation 3.2.35 equation gives,
αc (s) = s3 + (2ζωn + a)s2 + (2ζωn a + ωn2 )s + ωn2 a (3.2.36)
Equating the coefficients of Equations 3.2.34 and 3.2.36 results in the following two
equations for KQδE ,

Iy L̄α MQ
KQδE = 2ζωn + a − + (3.2.37)
MδE mV̄ Iy
and,

mV̄ Iy 2 L̄α MQ Mα
KQδE = 2ζωn a + ωn + + (3.2.38)
L̄α MδE mV̄ Iy Iy
If the damping of the short period mode (ζ ) and the location of the closed-loop integrator
pole (a) are selected, the resulting natural frequency can be solved by setting Equations
3.2.37 and 3.2.38 equal to one another yielding,
2
L̄α Mα L̄α L̄α
ωn2 + ωn 2ζa − 2ζ + + −a =0 (3.2.39)
mV̄ Iy mV̄ mV̄
Solving for positive values of ωn the closed-loop natural frequency of the short period
mode is given by,
s
Mα L̄α
ωn = ζη + (ζη)2 − − η (3.2.40)
Iy mV̄
where,
L̄α
η= −a (3.2.41)
mV̄
The natural frequency of the short period mode (ωn ), along with the previously selec-
ted damping of the short period mode (ζ ) and the location of the closed-loop integrator
pole (a) can now be used in conjunction with either of the pitch rate equations, Equa-
tions 3.2.37 and 3.2.38, to determine the pitch rate feedback gain KQδE . The integrator
feedback gain can be solved by equating the coefficients of the final terms of Equations
3.2.34 and 3.2.36 and is given by,
ωn2 mIy
KIδE = −a (3.2.42)
L̄α MδE
The integrator’s response can be removed from the reference input by placing a feed-
forward zero as follows,
KIδE
N̄CδE = − (3.2.43)
zf
where zf is the location of the desired zero.
3.2.4.1 Pole/Zero Placement
Placement of the closed-loop poles and zero essentially involves selecting the damping
ratio of the short period mode (ζ ), the location of the close loop integrator pole (a) and
the location of feed-forward zero (zf ). The pole/zero placements which follow are for the
aircraft parameters and standard flight conditions as outlined in Appendix B.6.
From Figure 3.4a it can be seen that the open-loop poles of the short period mode have
the following attributes,
ωnol ≈ 10.3 rad/s ζol ≈ 0.63 (3.2.44)
The closed-loop poles are now chosen to have a somewhat higher damping ratio of,
ζcl = 0.707 (3.2.45)
The integrator pole is placed at a frequency lower than that of the short period mode
poles,
a = −4 rad/s (3.2.46)
The feed-forward zero is placed at a frequency twice that of the integrator pole to avoid
excessive feed-forward whilst still removing some of the integrator dynamics from the
reference input,
zf = −8 rad/s (3.2.47)
The results of the pole/zero placement are shown in Figure 3.10. It can be seen that
the natural frequency of the short period mode poles has decreased slightly whilst the
damping has increased. The assumption made in §3.2.3.1 that the NMP nature of the
NSA dynamics can be ignored if the natural frequency of both the open- and the closed-
loop dominant poles of the system is at least three times slower than that of the NMP
zero remains valid.
10
−2
−4
−6
−8
−10
−20 −18 −16 −14 −12 −10 −8 −6 −4 −2 0
Figure 3.10 – Pole-zero map of the closed-loop elevator-based NSA controller
The NSA response to a reference step input can be seen in Figure 3.11a. The response
is well damped, and has a rise time of,
tr63.2% = 0.3 s (3.2.48)
The corresponding elevator deflection for this step input is shown in Figure 3.11b. It can
be seen that for a NSA step of 1 m/s2 the elevator remains well within the limits defined
in Appendix B.5. The effect of the feed-forward term in the control law can also be seen
clearly as the immediate jump in elevator deflection following the NSA command.
0.8
1
0.7
0.6
0.8
0.5
0.6
0.4
0.4 0.3
0.2
0.2
0.1
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
(a) Step response (b) Elevator deflection
Figure 3.11 – NSA step response to a reference NSA with corresponding elevator deflection
for the elevator-based NSA controller
During HIL simulation and practical flight tests the control gains are often tuned slightly
in order to achieve a satisfactory aircraft response. Figures 3.12a to 3.12c show what
effect varying the control gains has on the closed-loop poles and zeros of the NSA con-
troller. It can be seen in Figure 3.12a that increasing the pitch rate feedback gain KQδE
increases the damping and natural frequency of the short period mode poles, while it
decreases the frequency of the closed-loop integrator pole. This gain therefore has the
effect of slowing down the NSA step response and increasing the damping. Figure 3.12b
shows that increasing the integrator feedback gain KIδE decreases the damping and nat-
ural frequency of the closed-loop short period mode poles and increases the frequency of
the closed-loop integrator pole and feed-forward zero. This gain therefore has the effect
of speeding up the NSA step response while increasing the overshoot. It can be seen in
Figure 3.12c that increasing the feed-forward gain N̄CδE , decreases the frequency of the
feed-forward zero. This has the effect of speeding up the NSA response.
10 10
8 8
6 6
4 4
2 2
0 0
−2 −2
−4 −4
−6 −6
−8 −8
−10 −10
−20 −18 −16 −14 −12 −10 −8 −6 −4 −2 0 −20 −18 −16 −14 −12 −10 −8 −6 −4 −2 0
(a) Pitch rate feedback gain (b) Integrator feedback gain
10
−2
−4
−6
−8
−10
−20 −18 −16 −14 −12 −10 −8 −6 −4 −2 0
(c) Feed-forward gain
Figure 3.12 – Pole-zero map of the closed-loop elevator-based NSA controller for a change
in the control gains
3.2.4.2 HIL Simulation and Practical Flight Test Results
The NSA controller was implemented on an OBC and tested in the HIL simulation envir-
onment as well as practically in a series of flight tests. The HIL simulation environment,
aircraft, avionics and other hardware used in this project are outlined in Appendix B,
only the results are presented in this section. The HIL tests allow the controller to be
fully tested before practical flight tests are attempted. Ideally, the practical flight test
results should closely resemble those achieved in the HIL simulation.
Figures 3.13a and 3.13b show typical HIL and practical flight test results of the NSA
controller. The responses shown were selected since they give an approximate indication
of a step response. It is clear that the NSA signal is significantly more noisy in the
practical flight test than the HIL simulation. This is most likely due to the presence of
wind gusts and other such disturbances, and it is possible to adjust the HIL simulation
wind disturbance parameters to reflect a similar level of disturbance. Due to the high
noise level it is difficult to gauge an accurate rise time for the NSA in the practical flight
test. A better comparison between the HIL simulation and flight tests can be drawn by
examining the climbrate controller as shown in §3.3. The rise time of the HIL simulation
and practical flight tests appear to be very slightly slower than that of the pure simulated
NSA response. The difference can be explained due to uncertainties and unmodelled
effects, however the design can be considered a success.
−9 −4
−6
−10
−8
−10
−11
−12
−12 −14
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
−3 5
−4
−5 0
−6
−7 −5
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
(a) HIL (b) Flight test
Figure 3.13 – NSA response to a reference along with corresponding elevator deflection for
the elevator-based NSA controller during HIL and flight tests
3.2.5 NSA Controller Design - Flaps Actuator
This section presents the design of the flaps-based portion of the NSA controller. This
controller uses the aircraft’s flaps to create a commanded wind axes NSA through the
principles of direct-lift longitudinal control as outlined in §3.1. As discussed in section
§3.2.1, this controller will be superimposed with the moment-based NSA controller of
§3.2.4. The detailed derivation of this controller is given in Appendix C.3.
After the simplifications in §3.2.3.2 the NSA dynamic equations for a flaps input are given
by,
" # " #" # " #
α̇ − m1V̄ L̄α 1 α − m1V̄ L̄δF
= 1 1
+ 1
δF (3.2.49)
Q̇ Iy
Mα Iy
MQ Q Iy
Mδ F
" # " #" # " #
= + δF (3.2.50)
Q 0 1 Q 0
A flaps control law is now defined which uses proportional feedback from the NSA,
ref
δF = −KPδF (CW − CW ) (3.2.51)
Substituting the NSA output equation, CW , from Equation 3.2.50 into Equation 3.2.51
yields,
" # !
K P δF h
1
i α
ref
δF = − L̄δF
− L̄
m α
0 − CW (3.2.52)
1 − KPδF Q
m
The control law is now substituted into the dynamics and the closed-loop system be-
comes,
L̄ L̄α
" # " # " # " L̄ #
α̇ − mL̄αV̄ − ∆ mδF2 V̄ 1 α − mδV̄F ∆ ref
= M MδF L̄α MQ
+ Mδ CW (3.2.53)
Q̇ Iy
α
+ ∆ Iy m Iy
Q Iy
F
∆
" # " L̄ L̄α
# " # " L̄
#
CW − L̄mα − ∆ δmF 2 0 α − mδF ∆ ref
= + CW (3.2.54)
Q 0 1 Q 0
where,
 
K P δF
∆= L̄δF
 (3.2.55)
1 − KPδF m
The transfer function for the NSA response to a reference NSA input can be determined
from Equations 3.2.53 to 3.2.55,
h ih i
L̄ M M
− mδF ∆ s2 − IyQ s + L̄L̄δα IδyF − MIyα
GC W
C ref
(s) =
L̄δF L̄α
F
L̄ L̄α

M L̄α

L̄α MQ MQ L̄α MQ
W
s2 + s mV̄
+ ∆ m2 V̄ − Iy − Iy mV̄ + Iy ∆ mδF2 V̄ + − MIyα − ∆ IδyFm
(3.2.56)
It can be seen in the transfer function of Equation 3.2.56 that the feedback gain KPδF
does not have an effect on the location of the closed-loop zeros but does have an effect on
the location of the closed-loop poles and on the steady state gain of the closed-loop sys-
tem. In order to place the closed-loop poles a desired characteristic equation is defined
as,
αc (s) = (s2 + 2ζωn s + ωn2 ) (3.2.57)
Equating the characteristic equation (the denominator of the transfer function of Equa-
tion 3.2.56) with the desired characteristic equation allows for two equations for ∆ to be
derived,
m2 V̄

L̄α MQ
∆= 2ζωn − + (3.2.58)
L̄δF L̄α mV̄ Iy
and,

ω 2 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄
∆=− n (3.2.59)
MQ L̄δF L̄α + MδF L̄α mV̄
By substituting Equation 3.2.58 into Equation 3.2.59 an equation for the natural fre-
quency ωn can be determined,
L̄δF Iy2 ωn2 + 2ζωn MQ L̄δF Iy + MδF Iy mV̄ + MQ2 L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy = 0

(3.2.60)
Solving for the positive values of ωn the closed-loop natural frequency is given by,
q
(ζη)2 − L̄δF Iy2 MQ2 L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy

−ζη +
ωn = (3.2.61)
L̄δF Iy2
where,
η = MQ L̄δF Iy + MδF Iy mV̄ (3.2.62)
Substituting Equations 3.2.61 and 3.2.62 into Equations 3.2.58 and 3.2.59 respectively,
allows the derivation of two equations for the feedback gain KPδF ,
m L̄
KPδF = − α (3.2.63)
L̄δF L̄δF V̄ 2ζωn +
MQ
Iy
and,
ωn2 Iy mV̄ + MQ L̄α + Mα mV̄
KPδF = − (3.2.64)
MδF L̄α V̄ − L̄δF ωn2 Iy V̄ − L̄δF Mα V̄
3.2.5.1 Pole/Zero Placement
The direct-lift portion of the NSA controller only acts on higher frequency NSA com-
ponents, therefore the long term response of the aircraft is entirely conventional and
dependent on the elevator-based portion of the NSA controller of §3.2.4. The pole/zero
placement strategy in this section is therefore not focused on obtaining a specific NSA
response but rather to obtain a response which will yield an improvement in the over-
all NSA controller. Such improvement is explored in §3.2.6. The pole/zero placements
which follow are for the aircraft parameters and standard flight conditions as outlined in
Appendix B.6.
The transfer function for an NSA response to a reference NSA input is restated for con-
venience,
h ih i
L̄ M M
GC W
C ref
(s) =
L̄δF L̄α
F
L̄δF L̄α

MδF L̄α

L̄α MQ MQ L̄α MQ Mα
W
s2 +s mV̄
+ ∆ m2 V̄ − Iy − Iy mV̄ + Iy ∆ m2 V̄ + − Iy − ∆ Iy m
(3.2.65)
where,
 
K P δF
∆= L̄δF
 (3.2.66)
1 − KPδF m
As seen in the transfer function of Equation 3.2.65 the feedback gain KPδF does not have
an effect on the location of the closed-loop zeros. It is only possible to place the poles
and change the steady state gain of the system, both of which are interlinked.
To gain further insight into the behaviour of the system, it is useful to look at the pole-
zero map and step response of the closed-loop transfer function for a range of feedback
gains. Figure 3.14 shows the location of the poles and zeros for several different values
of the feedback gain KPδF . It can be seen that as the gain increases the poles move
towards the zeros. The location of the zeros remains unchanged. Therefore, as the
feedback gain increases, the response will become more like a pure gain as the poles
and zeros begin to cancel each other out.
15
10
−5
−10
−15
−15 −10 −5 0 5 10 15
Figure 3.14 – Pole-zero map of the closed-loop flaps-based NSA controller
The insight gained in the pole-zero map of Figure 3.14 is again reflected in the step
response of the transfer function as shown in Figure 3.15a. It can be seen that as the
feedback gain increases, the steady state NSA value increases and the transient compon-
ent of the response becomes smaller. The flap deflection angle corresponding to these
step responses is shown in Figure 3.15b. Here it can be seen that as the gain increases,
the required flaps deflection angle increases, and the flaps exhibit increasing damped
oscillating behaviour.
1
−0.5
0.8
−1
0.6
−1.5
0.4
−2
0.2 −2.5
0 −3
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) Step response (b) Flaps deflection
Figure 3.15 – NSA step responses to a reference NSA with corresponding flaps deflection of
the flaps-based NSA controller for a range of feedback gains
From the preceding analysis it follows that a high feedback gain will result in a good
response with little transient components and a small steady state error. However, this
could result in a required flap deflection which is beyond the physical limits of the air-
craft. Placement of the closed-loop poles is therefore a trade-off between response mag-
nitude and control effort required. For the aircraft used in this project the limits of the
flaps deflection angle are (see Appendix B.5),
−5 ◦ ≥ δF ≤ 5 ◦ (3.2.67)
Ideally, the feedback gain selected should be the largest possible where the reference
NSA input will not cause the controller to exceed the deflection angle limits of the flaps.
If a reasonable assumption is made regarding the maximum NSA demand expected dur-
ing flight, the feedback gain can be selected such that the flaps deflection limits are
not exceeded for such a control demand. This NSA demand will need to be established
through flight tests and HIL simulation, but a reasonable assumption is made that,
ref
−3 m/s2 ≥ CW ≤ 3 m/s2 (3.2.68)
These requirements result in a feedback gain which is used during simulation of the
complete NSA controller of §3.2.6. These simulations and subsequent flight tests show
that a lower feedback gain resulted in a more satisfactory overall response. Therefore,
this gain has been retroactively applied to be the result of the analysis of the rest of this
section. For a damping ratio of,
ζcl = 0.523 (3.2.69)
the closed-loop natural frequency can be calculated using Equations 3.2.61 and 3.2.62
as,
ωncl ≈ 10.78 rad/s (3.2.70)
The feedback gain can be calculated from either Equation 3.2.63 or Equation 3.2.64
giving,
KPδF = −0.02 (3.2.71)
This gain, along with the values for the aerodynamic derivatives, standard flight condi-
tions and aircraft parameters (see Appendix B.6) are substituted into the transfer func-
tion of Equation 3.2.65 giving,
0.3133s2 − 7.39s − 138.4 0.3133(s + 3.6949 + 11.1712i)(s + 3.6949 − 11.1712i)

GC W
C ref
(s) = 2
=
W s + 11.27s + 116.2 (s + 5.6354 + 9.1887i)(s + 5.6354 − 9.1887i)
(3.2.72)
The pole-zero map and step response of this transfer function are also shown in Figures
3.14 and 3.15. It can be seen in Figure 3.15a that the response consists of two parts.
The first is the initial jump in NSA up to the value of the steady state gain of the transfer
function, followed by a transient resulting in an initial slight decrease followed by a slight

increase in NSA as the response settles. The flap deflection angle corresponding to this
step responses is shown in Figure 3.15b where it can be seen that the deflection angle
remains well within the limits of Equation 3.2.67.
It is apparent that the step response for this pole placement only achieves about a third
of the reference NSA. This is usually unacceptable from a control design perspective,
however, since the design consideration is to improve the overall NSA response, it proves
to be a successful design as shown in §3.2.6.
3.2.6 NSA Controller Design - Elevator and Flaps Combined
This section presents the design of the overall NSA controller where both the elev-
ator and the flaps are used as actuators. As discussed in the NSA controller strategy
of §3.2.1, the flaps-based portion of the NSA controller must be superimposed on the
elevator-based portion of the NSA controller. To overcome the problem of proper con-
trol allocation, and in order to superimpose the flaps-based NSA controller in transient
form, the two NSA controllers are separated in frequency. This frequency separation is
achieved by a complementary filter pair on the NSA controller signals. The controller
configuration is shown diagrammatically in Figure 3.2.
In this design there are several parameters of interest namely the cut-off frequency of
the complementary filter pair ωC , the gain of the flaps-based portion of the NSA con-
troller KPδF and the gearing between the flaps and the elevator KM . The parameters of
the elevator-based portion of the NSA controller are considered fixed and as shown in
§3.2.4. In order to determine which combination of these parameters will yield the best
result, the controller is simulated in Matlab Simulink as shown in Figure 3.16. This pro-
cess is largely iterative, and is focused on the step response and disturbance rejection
characteristics of the controller.
During the HIL and subsequent flight tests of this controller it was found that the follow-
ing set of parameters lead to an acceptable controller response,
ωC = 4 rad/s KPδF = −0.02 KM = 0.2 (3.2.73)
These gains are somewhat lower than what the simulation suggests is feasible. However,
the parameters of Equation 3.2.73 still lead to an improvement over the conventional
elevator-based NSA controller of §3.2.4. The cause of this discrepancy is discussed in
§3.2.6.1. Each of these parameters is now investigated in order to illustrate their effect
on the response of the NSA controller.
Complementary Filter The complementary filter pair consists of first order low-pass
filters on the NSA reference and feedback signals of the elevator-based portion of the
controller, and first order high-pass filters on the NSA signals of the flaps-based portion
c
1
-
s
1
s
Figure 3.16 – Simulink block diagram of the direct-lift-augmented NSA controller
of the controller. These filters have the effect that the elevator will respond to NSA sig-
nals below the cut-off frequency and the flaps to NSA signals above the cut-off frequency
of the filter, with some overlap in between.
During simulation it was found that, for relatively low values of flaps feedback gain, ap-
plying a low-pass filter on the NSA signals of the elevator-based portion of the controller
leads to weakened response characteristics. This is due to the limited amount of direct-
lift available at low flaps feedback gains which means that only a part of the NSA demand
is achieved as shown in Figure 3.15. Under these conditions limiting the frequency range
in which the elevator-based portion of the controller acts, results in degraded perform-
ance since the flaps do not have sufficient control authority to compensate for the loss
of higher frequency elevator control. If the flaps feedback gain were made sufficiently
high, then enough direct-lift is available for the filter to be used without degrading per-
formance. Unfortunately practical limitations exist which prohibit the flaps gain from
being made high enough as discussed in §3.2.6.1. Alternatively, placing the cut-off fre-
quency of the filter much higher also eliminates this problem, however this then also
means that disturbances available to the flaps-based controller to be rejected are far
fewer, therefore under-utilising its disturbance rejection capabilities.
To overcome this limitation, the NSA signals to the elevator-based portion of the con-
troller are not filtered. This can be considered an acceptable deviation from the control
strategy of §3.2.1 since the dominant closed-loop pole of the elevator-based controller ef-
fectively acts as a low-pass filter. In light of this, the cut-off frequency for the high-pass
filters is chosen to coincide with that of the dominant closed-loop pole of the elevator-
based NSA controller as shown in Figure 3.10. Omitting the low-pass filter means that
there will now be a greater overlap between the elevator and flaps portions of the con-
troller. This will result in situations where the flaps and the elevator portions of the con-
troller act on the same signal which raises the question of interference between them.
However, it was found during simulations that since the flaps only produce a part of the
NSA demand there is never a situation where it overpowers the elevator portion of the
controller. The flaps serve to simply augment the elevator at higher frequencies.
Flaps Feedback Gain The flaps-based portion of the controller makes use of propor-
tional feedback from the high-pass filtered NSA reference and feedback signals. In order
to illustrate the effect of the flaps feedback gain KPδF the step response for a range of
feedback gains is shown in Figure 3.17. Here the cut-off frequency is as in Equation
3.2.73 and there is no gearing between the flaps and the elevator. It can be seen that as
the gain increases, the initial step in NSA due to the direct-lift component of the flaps
controller becomes more pronounced. This results in faster rise times, however it comes
at the cost of greatly increased settling time due to oscillation induced in the response,
as well as greater flaps deflection demand. It is possible to design a controller with
the higher feedback gains which functions in simulation, however, practical limitations
encountered during flight tests indicate that the gain be chosen as,
KPδF = −0.02 (3.2.74)
0.8
0.6
0.4
0.2
0 0.5 1 1.5 2 2.5 3
Figure 3.17 – Step response of the combined elevator and flaps NSA controller for a range
of flaps feedback gains
Flaps to elevator gearing The gain KM represents the gearing required between the
flaps and the elevator in order for the elevator to cancel out any unwanted moment
produced by the flaps. As stated in §3.2.3.2 this gearing can be selected such that the
elevator does not cancel out the moment produced by the flaps perfectly, but rather
increases or decreases the net moment acting on the aircraft. In this way the response
can be tuned to obtain a satisfactory result. In order to illustrate this effect, the NSA
step response is shown for a range of gearing ratios. Here the cut-off frequency and
flaps feedback gain are as Equation 3.2.73.
1
−0.5
0.8 −1
−1.5
0.6
−2
0.4
−2.5
0.2 −3
−3.5
0
−4
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
(a) NSA step response (b) Pitch rate response
Figure 3.18 – NSA step response with corresponding pitch rate response of the combined
elevator and flaps NSA controller for a range of flaps to elevator gearing ratios
It can be seen in Figure 3.18b that for a gearing ratio of 0.2 the pitch rate response of the
elevator-based controller is nearly identical to that of the elevator and flaps combined
controller. This shows that the moment produced by the flaps deflection has been mostly
cancelled out by the gearing to the elevator. Examination of the dimensionless pitch
moment derivative of the elevator and flaps as given by Table 3.1 shows that the ratio
between these derivatives is, as expected, also approximately 0.2 . It can be seen that
for a gearing ratio of −0.2 the elevator is adding to the moment produced by the flap
deflection. This has the effect of speeding up the pitch rate as well as the NSA response
up to a point. HIL simulation and practical flight tests indicate that the gearing ratio be
chosen as,
KM = 0.2 (3.2.75)
Disturbance rejection With the controller parameters now fixed, the disturbance re-
jection capabilities of the controller are demonstrated. Figure 3.19 shows the controller
attempting to follow a constant NSA reference in the presence of a disturbance. It can
be seen that the controller, making use of both flaps and elevator, follows the reference
signal more closely. The disturbance rejection can be improved greatly by increasing the
flaps feedback gain, however certain practical limitations to this arise as discussed in
§3.2.6.1.
3.2.6.1 HIL Simulation and Practical Flight Test Results
The NSA controller was implemented on an OBC and tested in the HIL simulation envir-
onment as well as practically in a series of flight tests. The HIL simulation environment,
aircraft, avionics and other hardware used in this project are outlined in Appendix B,
only the results are presented in this section. The HIL tests allow the controller to be
0.6
0.4
0.2
−0.2
−0.4
−0.6
0 0.5 1 1.5 2 2.5 3
Figure 3.19 – NSA reference following in the presence of an NSA disturbance
The amount of noise and disturbances on the NSA signal make it difficult to evaluate the
controller’s performance directly. It is therefore useful to simultaneously examine the
climb rate performance of the aircraft since the NSA controller has a direct effect on how
well the climb rate controller functions. This allows a comparison to be made between
the conventional and direct-lift-augmented NSA controllers. Figures 3.20a and 3.20b
show the climb rate and corresponding NSA response of the conventional and direct-lift-
augmented NSA controllers during HIL simulation, and Figures 3.20c and 3.20d show
the same during practical flight tests. Firstly, it can be seen that in general the responses
of the HIL simulation correspond very well with those obtained during flight tests, thus
confirming the validity of the aircraft model used to design the controllers. The exception
to this is the NSA signals where it can be seen that during the flight tests the NSA signal
is significantly more noisy than during the HIL simulation. It is however possible to
obtain similar NSA signals in the HIL simulation environment by adjusting the model’s
noise parameters. The origin of this noise is examined further in the subsequent analysis
of the direct-lift-augmented NSA controller.
Secondly, a comparison can be made between the conventional and direct-lift-augmented

NSA controllers by examining the flight test plots since the differences are more appar-
ent than in the HIL simulation. The data shown are of two separate flight tests where
the reference climb rates are very similar. Unfortunately, atmospheric conditions dif-
fer between the two flight tests, such that an exact comparison cannot be made but is
is still possible to gauge the general performance of the two controllers. It must be
noted that the atmospheric conditions during the conventional NSA controller test were
more favourable than during the direct-lift-augmented NSA controller test. It can be
seen in Figure 3.20c that the direct-lift-augmented NSA controller shows a somewhat
faster climb rate response than the conventional NSA controller. The corresponding
NSA response shown in Figure 3.20d does not reveal much other than the presence of
significant disturbances.
2
−5
0
−10
−2
−15
−4
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
2
−5
0
−10
−2
−15
−4
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
(a) HIL climb rate response (b) HIL NSA response
2
−5
0
−10
−2
−15
−4
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
2
−5
0
−10
−2
−15
−4
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
(c) Flight test climb rate response (d) Flight test NSA response
Figure 3.20 – Climb rate and corresponding NSA response for the conventional and direct-
lift-augmented NSA controllers during HIL and flight tests
The flaps-based portion of the NSA controller makes use of high-pass filtered NSA and
reference NSA signals as shown in Figure 3.16. These signals are shown in Figure 3.21
for the same HIL simulation and flight tests as in Figure 3.20. It can be seen in the upper
plot of Figure 3.21a that in the HIL simulation there is a slight difference in the high-
pass filtered NSA signals of the conventional and direct-lift-augmented NSA controllers
after the climb rate step time. The direct-lift-augmented NSA controller responds faster
than the conventional NSA controller as can be seen in the initial jump in NSA after the
climb rate step. This response is similar in form to that seen during the controller design
as depicted in Figure 3.18a and is the result of the direct-lift component of the flaps
deflection.
It can be seen in the upper plot of Figure 3.21b that during the flight tests, the mag-
nitude of the high-pass filtered NSA signal is on average slightly less for the direct-
lift-augmented NSA controller than for the conventional NSA controller. This is to be
expected since one of the functions of the flaps is to counter high-frequency NSA dis-
turbances. It must be noted that there is an improvement over the conventional NSA
controller in spite of the fact that the atmospheric conditions were less favourable dur-
ing the direct-lift-augmented NSA controller test.
The lower plot in Figures 3.21a and 3.21b show the high-pass filtered NSA reference
signal which originates from the climb rate controller of §3.3. The error between this
signal and that of the high-pass filtered NSA signal in the upper plot of Figures 3.21a
and 3.21b, is used in conjunction with the flaps feedback gain in order to determine the
required flaps deflection. As expected, this signal remains fairly small for most climb
rate controller demands, however during the climb rate step it can be seen that an
immediate NSA demand is generated which subsequently dies down due the effect of
the high-pass filter. The direct-lift portion of the NSA controller therefore receives a
large input which it can immediately respond to due to the direct-lift component of the
flaps deflection. This allows the direct-lift-augmented NSA controller to achieve a faster
response as demonstrated by the faster climb rate step shown in Figure 3.20c.
5 5
0 0
−5 −5
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
5 5
0 0
−5 −5
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
(a) HIL NSA signals (b) Flight test NSA signals
Figure 3.21 – High-pass filtered NSA and NSA reference signals during HIL and flight tests
Figures 3.22a and 3.22b show the error between the high-pass filtered NSA and NSA
reference signals, and the corresponding actuator deflections for the HIL simulation,
and Figures 3.22c and 3.22d show the same during practical flight tests. In these figure
it can again be seen that there is a large amount of noise on the NSA signal since the
error fluctuates considerably. It can also be seen that the maximum error is greater for
the conventional NSA controller than for the direct-lift-augmented NSA controller. The
actuator deflection plots show that both the elevator and the flaps remain well within
their respective deflection limits. The elevator deflection for the direct-lift augmented
NSA controller shows small fluctuations on top of its main response. These are the
component of the elevator deflection which is geared from the flaps in order to cancel
out the moment produced by the flaps.
5
5
4
3 0
2
−5
1
0 2 4 6 8 10 12 14
0
−1
5
−2
−3 0
−4
−5
−5
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
(a) HIL High-pass filtered NSA signals error (b) HIL elevator and flaps deflection
5
5
4
3 0
2
−5
1
0 2 4 6 8 10 12 14
0
−1
5
−2
−3 0
−4
−5
−5
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
(c) Flight test High-pass filtered NSA signals (d) Flight test elevator and flaps deflection
error
Figure 3.22 – Step response of the combined elevator and flaps NSA controller for a range
of flaps to elevator gearing ratios
As discussed previously, there is a fairly large amount of noise on the NSA signal. This
noise is far greater than the measurement noise of the accelerometers used in this pro-
ject (see Appendix B). The origin of this noise is therefore most likely a combination
of atmospheric disturbances and unmodelled effects such as engine noise. True atmo-
spheric disturbances should be attenuated by the flaps-based NSA controller, and this
effect is seen in the results of this section. However, it was expected that this effect
would be greater. A possible explanation for this is that some of the excess noise is not
aerodynamic in origin, therefore when the flaps attempt to counteract this disturbance
they might instead be adding to the true disturbance. This is also suggested by the fact
that the flaps feedback gain used had to be much lower than the simulation suggested in
order to achieve a stable controller. It might therefore be necessary to revise the control
law of the flaps-based portion of the NSA controller to mitigate this effect. One such
option would be to augment a fast integrator on the NSA signal similar to that of the
elevator-based NSA controller of §3.2.4. The time constraints of this project do not allow
such a controller to be tested but initial simulations show that the strategy does hold
promise and that further investigation is warranted.
3.3 Climb Rate Controller
This section describes the design, simulation and practical testing of a climb rate con-
troller. This controller regulates the climb rate of the aircraft by making use of either
the conventional NSA controller of §3.2.4 or the direct-lift-augmented NSA controller of
§3.2.6. The input to the climb rate controller stems from the altitude controller of §3.4,
which creates a commanded altitude by commanding the climb rate controller. The con-
troller is designed in the navigation axis system as shown in Figure 2.4. It is therefore
assumed that the NSA vector of the aircraft is always pointing up normal to the flat earth.
This assumption is valid as long as the flight path angle remains small and the effect of
the aircraft’s roll orientation is compensated for by the guidance controller as shown in
§4.4.
It must be noted that the climb rate controller is designed independent of the airspeed
dynamics. This is despite the coupling which exists between the airspeed and climb rate
dynamics. Typically, this coupling is addressed by designing a multi-input-multi-output
(MIMO) airspeed and climb rate controller. Linear Quadratic Regulator (LQR) theory is
often used for this purpose as demonstrated by [8]. In such a controller strategy, the
airspeed and climb rate commands are both achieved by a combination of the throttle
and longitudinal aerodynamic actuators. To illustrate this, consider an airspeed step; the
throttle command will increase to achieve the greater airspeed demand, and the elevator
actuator will cause a downwards climb rate which also serves to increase the airspeed.
The relative amounts of actuation used is determined by the weighting matrices of the
LQR design. It is therefore clear that applying such a controller to this project will make
the relationship between the NSA controller and the climb rate of the aircraft become
significantly less clear. This is especially important since the conventional moment-based
NSA controller must be adequately compared to the direct-lift-augmented NSA control-
ler. The climb rate and airspeed controllers will therefore be designed completely separ-
ately from one another under the assumption that the airspeed is maintained sufficiently
well by the airspeed controller. As shown in the airspeed controller design of §3.5, the
gravitational coupling into the airspeed dynamics is also compensated for, which further
aids in separating the airspeed and climb rate controllers.
3.3.1 Design
The basic model for the aircraft’s climb rate dynamics is given by,
h ih i h i
V̇h = 0 Vh + 1 CN (3.3.1)
where Vh is the climb rate and CN the normal specific acceleration of the aircraft in the
navigation axis system.
The dominant closed-loop pole of the NSA controller is located at 4 rad/s as shown in
§3.10. If the climb rate dynamics are to be timescale separated from the normal dynam-
ics, then by the generally held design rule, the climb rate dynamics bandwidth should
be at least five times less than that of the normal dynamics. Such a pole placement will
lead to an unacceptably slow climb rate response. If however the dynamics of the NSA
controller are taken into account during the design of the climb rate controller, then
the frequency separation requirement falls away and the climb rate controller poles can
be placed at a somewhat higher frequency. To accomplish this the NSA controller can
be approximated with a simple first order lag and the climb rate dynamics augmented
accordingly,
" #  " #  
V̇h 0 1 Vh 0
= 1  +  1  CNref (3.3.2)
V̈h 0 − V̇h
τNSA τNSA
" #
h i V
h
Vh = 1 0 (3.3.3)
V̇h
where τNSA is the time constant of the dominant closed-loop pole of the NSA controller
ref ref
and CN is the component of commanded NSA in the navigation axis system. CN can
be considered a virtual actuator of this system and it is assumed that any command is
instantly met. Approximating the NSA controller with a first order lag is only truly valid
for the conventional moment-based NSA controller since for the direct-lift-augmented
NSA controller a part of the NSA is generated almost instantaneously as shown in Fig-
ure 3.18a. This means that the climb rate response for the direct-lift-augmented NSA
controller will be somewhat faster than the design in this section suggests.
A control law is now defined which uses proportional feedback from the climb rate,

CNref = −KPCR Vh − Vhref (3.3.4)
" # !
h i V
h
= −KPCR 1 0 − Vhref
V̇h
ref
where Vh is the reference climb rate command. The control law is now substituted into
the dynamics and the closed-loop system becomes,
" # " #" # " #
V̇h 0 1 Vh 0
= + Vhref (3.3.5)
V̈h −KPCR aNSA −aNSA V̇h KPCR aNSA
where,
1
aNSA = (3.3.6)
τNSA
is the location of the closed-loop integrator pole of the NSA dynamics as described in
§3.2.4.
p(s) = s2 + aNSA s + KPCR aNSA (3.3.7)

In order to place the closed-loop poles, a desired characteristic equation for the climb
rate dynamics is defined as,
αc (s) = s2 + 2ζωn s + ωn2 (3.3.8)
By equating the coefficients of Equations 3.3.7 and 3.3.8 the control gain can be calcu-
lated as,
aNSA
KPCR = (3.3.9)
4ζ 2
and the resulting natural frequency as,
aNSA
ωn = (3.3.10)
2ζ
3.3.2 Pole Placement
Placement of the closed-loop poles essentially involves selecting the damping ratio (ζ ) of
the climb rate dynamics. The location of the closed-loop integrator pole of the NSA con-
troller (aNSA ) was determined in §3.2.4.1 and is therefore fixed for this design. The climb
rate feedback gain can then be calculated from Equation 3.3.9. The pole placements
Appendix B.6.
The damping ratio is selected so that there is no overshoot in the response. Thus,
ζcl = 1 (3.3.11)
From Equations 3.3.9 and 3.3.10 the climb rate feedback gain and resulting natural
frequency of the climb rate dynamics are then calculated to be,
KPCR = 1 ωn = 2 rad/s (3.3.12)
During HIL simulation and flight tests the control gains are often tuned slightly in order
to achieve a satisfactory aircraft response. Figures 3.23a and 3.23b show what effect
varying the feedback gain has on the closed-loop poles and zeros as well as on the step
response of the climb rate controller. It can be seen that when the feedback gain is
increased, the system’s poles move towards one another, eventually becoming a complex
pole pair. Increasing the climb rate feedback gain therefore increases the frequency
and decreases the damping of the dominant closed-loop pole(s). This has the effect of
speeding up the climb rate response, while also increasing the overshoot.
3.3.3 HIL Simulation and Practical Flight Test Results
The climb rate controller was implemented on an OBC and tested in the HIL simulation
environment as well as practically in a series of flight tests. The HIL simulation environ-
ment, aircraft, avionics and other hardware used in this project are outlined in Appendix
2.5
2 1
1.5
0.8
1
0.5
0.6
0
−0.5
0.4
−1
−1.5
0.2
−2
−2.5 0
−5 −4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3.23 – Pole-zero map and step response of the climb rate controller for a range of
feedback gains
B, only the results are presented in this section. The HIL tests allow the controller to be
With reference to Figure 3.20 it has already been show in the HIL and practical flight
tests of the NSA controller in §3.2.6.1 that the responses of the climb rate controller
during HIL simulation correspond very well with those obtained during flight tests. Fig-
ures 3.20c and 3.20d are presented again as Figures 3.24a and 3.24b so that a more
detailed comparison can be made between the climb rate controller making use of a
conventional NSA controller, and a direct-lift-augmented NSA controller. The practical
fight test data shown in Figure 3.24 are of two separate flight tests where the reference
climb rates are very similar. Unfortunately, atmospheric conditions differ between the
two flight tests, and as such an exact comparison cannot be made but is is still possible
to gauge the general performance of the two NSA controllers. It must be noted that the
atmospheric conditions during the conventional NSA controller test were more favour-
able than during the direct-lift-augmented NSA controller test. It can be seen that the
direct-lift-augmented NSA controller shows a somewhat faster climb rate response than
the conventional NSA controller or the pole placement design suggests. This is due to
the direct-lift component present in the augmented NSA controller, and it demonstrates
one of the advantages that the direct-lift-augmented NSA controller holds over the con-
ventional NSA controller. The corresponding NSA response shown in Figure 3.24b does
not reveal much other than the presence of significant disturbances which are discussed
in §3.2.6.1
Figure 3.25 shows the effect of the climb rate feedback gain on the response during
practical flight tests. The conventional NSA controller is used and its response is shown
in Figure 3.25b. The results are very similar to those of the the pole placement design.
This controller design can thus be considered a success.
2
−5
0
−10
−2
−15
−4
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
2
−5
0
−10
−2
−15
−4
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
(a) Climb rate response (b) NSA response
Figure 3.24 – Climb rate response and corresponding NSA response for the conventional
and direct-lift-augmented NSA controllers during practical flight tests
5 −5
4 −10
0 2 4 6 8 10
3
−5
2
−10
1
0 2 4 6 8 10
0
−5
−1 −10
−2 0 2 4 6 8 10
0 2 4 6 8 10
(a) Step response (b) NSA response
Figure 3.25 – Climb rate step response and corresponding NSA response for a series of
climb rate feedback gains during practical flight tests
3.4 Altitude Controller
This section describes the design, simulation and practical testing of an altitude control-
ler. This controller regulates the altitude of the aircraft by making use of the climb rate
controller of §3.3. The input to the altitude controller stems from the aircraft landing
state machine of §5.4. The controller is designed in the inertial axis system as shown in
Figure 2.1. It is therefore assumed that the climb rate vector of the aircraft is always
normal to the flat earth. This assumption is valid as long as the flight path angle remains
small and the effect of the aircraft’s roll orientation is compensated for by the guidance
controller as shown in §4.4.
3.4.1 Design
The basic model for the aircraft’s altitude dynamics is given by,
h ih i h i
ḣ = 0 h + 1 Vh (3.4.1)
where h is the altitude and Vh the climb rate of the aircraft in the inertial and navigation
axis systems respectively.
The dominant closed-loop pole of the climb rate controller of §3.3 is located at 2 rad/s as
shown in Equation 3.3.12. If the altitude dynamics are to be timescale separated from
the climb rate dynamics, then by the generally held design rule, the altitude dynamics
bandwidth should be at least five times less than that of the climb rate dynamics. This
pole placement will lead to an unacceptably slow altitude response. If however, the
dynamics of the climb rate controller are taken into account, the frequency separation
requirement falls away and the altitude controller poles can be placed at a somewhat
higher frequency. To accomplish this the climb rate dynamics can be approximated with
a simple first order lag and the altitude dynamics augmented accordingly,
" #  " #  
ḣ 0 1 h 0
= 1  +  1  Vhref (3.4.2)
V̇h 0 − Vh
τCR τCR
" #
h i h
h= 1 0 (3.4.3)
Vh
where τCR is the time constant of the dominant closed-loop climb rate dynamics pole and
Vhref is the commanded climbrate in the navigation axis system. Vhref can be considered
a virtual actuator of this system and it is assumed that any command is instantly met.
In this project accurate altitude following is of paramount importance. An argument

can therefore be made that the control law of the altitude controller should contain
integrator feedback in order to eliminate steady state errors which will arise when a
ramp command, such as the landing approach path, is followed. However, the control
law defined makes use of only proportional feedback from the climb rate which means
that the landing approach path will be followed with a constant altitude offset. The
reasoning behind this decision is discussed in §5.3. Thus,
Vhref = −KPh h − href

(3.4.4)
" # !
h i h
= −KPh 1 0 − href
Vh
where href is the reference altitude command. The control law is now substituted into
the dynamics and the closed-loop system becomes,
" # " #" # " #
ḣ 0 1 h 0
= + href (3.4.5)
V̇h −KPh aCR −aCR Vh KPh aCR
where,
1
aCR = (3.4.6)
τCR
is the location of the closed-loop integrator pole of the climb rate dynamics as given by
Equation 3.3.10.
p(s) = s2 + aCR s + KPh aCR (3.4.7)
In order to place the closed-loop poles, a desired characteristic equation for the altitude
αc (s) = s2 + 2ζωn s + ωn2 (3.4.8)
By equating the coefficients of Equations 3.4.7 and 3.4.8 the control gain can be calcu-
lated as,
aCR
KPh = (3.4.9)
4ζ 2
aCR
ωn = (3.4.10)
2ζ
Placement of the closed-loop poles essentially involves selecting the damping ratio (ζ )
of the altitude dynamics. The location of the closed-loop integrator pole of the climb
rate controller (aCR ) was determined in §3.3 and is therefore fixed for this design. The
altitude feedback gain can then be calculated from Equation 3.4.9. The pole placements
Appendix B.6.
The damping ratio is selected so that there is no overshoot in the response. Thus,
ζcl = 1 (3.4.11)
From Equations 3.4.9 and 3.4.10 the altitude feedback gain and resulting natural fre-
quency of the altitude dynamics are then calculated to be,
KPh = 0.5 ωn = 1 rad/s (3.4.12)
During HIL simulation and flight tests the control gains are often tuned slightly in order
to achieve a satisfactory aircraft response. Figures 3.26a and 3.26b show what effect
varying the feedback gain has on the closed-loop poles and zeros as well as on the step
response of the altitude controller. It can be seen that when the feedback gain is in-
creased, the systems poles move towards one another, eventually becoming a complex
pole pair. Increasing the altitude feedback gain therefore increases frequency and de-
creases the damping of the dominant closed-loop pole(s). This has the effect of speeding
up the altitude response, while also increasing the overshoot.
1 1
0.8
0.5
0.6
0
0.4
−0.5
0.2
−1
0
−2.5 −2 −1.5 −1 −0.5 0 0 1 2 3 4 5 6 7 8
Figure 3.26 – Pole-zero map and step response of the altitude controller for a range of
feedback gains
The altitude controller was implemented on an OBC and tested in the HIL simulation
Figures 3.27a and 3.27b show the altitude step and corresponding climb rate response
of the aircraft for both the conventional and the direct-lift-augmented NSA controllers
during HIL simulation, and Figures 3.27c and 3.27d show the same for practical flight
tests. Firstly, it can be seen that the HIL simulation and practical flight test altitude
steps correspond well with one another with the exception that the HIL simulation ex-
hibits somewhat greater overshoot. The practical fight test data shown in Figure 3.27c
is of two separate flight tests. Unfortunately, atmospheric conditions differ between the
two flight tests, and as such an exact comparison cannot be made but it is still possible
to gauge the general performance of the two NSA controllers. It can be seen that while
the rise time is very similar, the altitude controller which makes use of the direct-lift-
augmented NSA controller follows the reference more closely than the controller which
makes use of the conventional NSA controller. This demonstrates the improved dis-
turbance rejection capabilities of the direct-lift-augmented NSA controller compared to
the conventional NSA controller. The comparisons between the climb rate controllers
driven by the conventional and the direct-lift-augmented NSA controllers are discussed
in §3.3.3.
A good measure of the altitude controller’s performance is its ability to maintain a con-
stant altitude during different aircraft roll angles. This is demonstrated in Figures 3.28a
60 2
0
55
−2
50
−4
0 2 4 6 8 10 12 14 16 18
45
40
0
35
−2
30 −4
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
(a) HIL altitude response (b) HIL climb rate response
60 2
0
55
−2
50
−4
0 2 4 6 8 10 12 14 16 18
45
40
0
35
−2
30 −4
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
(c) Flight test altitude response (d) Flight test climb rate response
Figure 3.27 – Altitude step and corresponding climb rate response for the conventional and
direct-lift-augmented NSA controllers during HIL and flight tests
and 3.28b where the altitude response of the aircraft is shown whilst the aircraft is flying
a circuit. It can be seen that even during high roll angles the altitude is maintained with
only a slight decrease in precision. This is of importance since any deviation in altitude
during a turn will affect the landing approach path as discussed in Chapter ??.
56
150
54
52 100
50
50
48
46 0
44
0 10 20 30 40 50 60 70 80 −50
−100
10
0 −150
−10
−200
−20
−30 −250
−40
−300
−50
0 10 20 30 40 50 60 70 80 −300 −200 −100 0 100 200
(a) Altitude and roll angle (b) Flight path
Figure 3.28 – Altitude response and roll angle during a looped flight path
3.5 Airspeed Controller
This section describes the design, simulation and practical testing of an airspeed con-
troller. This controller regulates the airspeed of the aircraft by means of the throttle
actuator. The input to the airspeed controller stems from the aircraft landing state ma-
chine of §5.4. The controller is designed in the aircraft’s wind axis system as shown in
Figure 2.3 and will make use of the measurements from a pitot tube to determine the
aircraft’s airspeed. As discussed in §3.3, the airspeed controller is designed independent
of the climb rate dynamics despite the coupling which exists.
3.5.1 Design
The simplified and decoupled model of the aircraft’s axial dynamics from Equations 2.4.6
and 2.4.7 is restated for convenience,

1 1
Ṫ = − T+ TC (3.5.1)
τT τT

1 qS
AW = T + − CD − g sin θ (3.5.2)
m m
where the standard trigonometric small angle assumptions have been made with regard
to the two incidence angles α and β . The dynamics for the velocity magnitude of the
aircraft’s wind axes can be written as,
" # " #" # " # " # " #
Ṫ − τ1 0 T 1
τT 0 0
= T + Tc + + (3.5.3)
V̄˙ 1
m
0 V̄ 0 − qS
m D
C −g sin θ
The drag term above is treated as an unmodelled disturbance [2] and can be removed by
augmenting an integrator to the system in order to reject this and any other disturbances
to the airspeed in the steady state. The integrator is defined as,
ĖV = V̄ − V̄ ref (3.5.4)

where V̄ ref is the reference airspeed in wind axes. After simplifying the dynamics and
augmenting the integrator the system becomes,
          
ĖV 0 0 1 EV 0 0 −1
   1
   1   ref
 Ṫ  = 0 − τT 0  T  +  τ T  TC +  0  +  0  V (3.5.5)
     
V̄˙ 0 1
m
0 V̄ 0 −g sin θ 0
A throttle control law is defined which uses feedback from the integrator and airspeed
states as well as a feed-forward term which will be used to cancel out the effect of
gravitational coupling. The gravitational coupling effect must be minimised in particular
when the aircraft descends steeply during a landing approach in order to prevent a rapid
increase in airspeed. Thus,
TC = −KV V̄ − KE EV + Tg (3.5.6)
where Tg is the throttle command which removes the effect of gravitational coupling. The
control law is now substituted into the dynamics and the closed-loop system becomes,
          
ĖV 0 0 1 EV 0 0 −1
   1 1
   1 
− τ1 KV 
 ref
− τT KE − τT
 Ṫ  =    T  +  τ  Tg +  0  +  0 V (3.5.7)
    
T T
V̄˙ 0 1
m
0 V̄ 0 −g sin(θ) 0
The gravitational coupling into the airspeed dynamics can be cancelled out by letting,
Tg = gmτT sin θ (3.5.8)
Thus the closed-loop system becomes,

      
ĖV 0 0 1 EV −1
   1 1
  
− τ1 KV 
 ref
− τT KE − τT
 Ṫ  =  T  T  +  0 V (3.5.9)
V̄˙ 0 1
m
0 V̄ 0

1 2 KV KE
p(s) = s3 + s + s+ (3.5.10)
τT mτT mτT
In order to place the closed-loop poles a desired characteristic equation for the airspeed
αc (s) = (s2 + 2ζωn s + ωn2 )(s + a) (3.5.11)
By equating the coefficients of Equations 3.5.10 and 3.5.11 control gains can be calcu-
lated as,
(1 + (4ζ 2 − 1)aτT )(1 − aτT )
KV = m (3.5.12)
4ζ 2 τT
2
1 − aτT
KE = mτT a (3.5.13)
2ζτT
1 − aτT
ωn = (3.5.14)
2ζτT
The throttle actuator is prone to integrator windup especially during steep climbs and
descents. Integrator anti-windup can be implemented by placing a limit on the integrator
state whenever the controller commands a thrust which is outside the physical range of
the aircraft’s motor. This integrator state is obtained by rewriting the control law of
Equation 3.5.6,
Tg − KV V̄ − TC
EV = (3.5.15)
KE
It must be noted that several assumptions are made regarding the thrust produced by
the motor. Firstly, it is assumed that the motor’s thrust is commanded, when in reality
it is the voltage to the motor which is controlled by sending a pulse-width-modulated
(PWM) signal to the speed controller. This signal is based on a fraction of the maximum
thrust of the motor. There is therefore a difference between the thrust commanded and
the true thrust produced, however, this is compensated for by the integrator on airspeed.
Secondly, it has been assumed that the effect of airspeed on the thrust is constant. This
means that the maximum dynamic thrust of the aircraft was fixed at a value somewhat
lower than the maximum static thrust determined by a static thrust test. Again, any
errors will be compensated for by the integrator on airspeed.
Placement of the closed-loop poles essentially involves selecting the damping ratio (ζ )
and the location of the closed-loop integrator pole (a) of the airspeed dynamics. The
feedback gains (KV and KE ) can then be calculated from Equations 3.5.12 and 3.5.13.
The value of the thrust time constant (τT ) as well as the maximum thrust were determined
through tests (see Appendix B.6.2). The pole placements which follow are for the aircraft
parameters and standard flight conditions as outlined in Appendix B.6.
The damping ratio and closed-loop integrator location are selected such that there is
little overshoot in the response and that the throttle command remains within reasonable
bounds for a typical airspeed command. Thus,
ζcl = 0.8 a = 0.5 rad/s (3.5.16)
The pole placement and step response of the airspeed controller are shown in Figure
3.29.
1
1.5
1
0.8
0.5
0.6
0
−0.5 0.4
−1
0.2
−1.5
−2 0
−4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0 2 4 6 8 10
Figure 3.29 – Pole-zero map and step response of the airspeed controller
The airspeed controller was implemented on an OBC and tested in the HIL simulation
During HIL simulations the airspeed of the aircraft is maintained well, however the
throttle actuator behaviour differs significantly from that of the practical flight tests.
This is most likely due to modelling inaccuracies of the drag coefficient term but is ulti-
mately of little consequence. The HIL results are thus omitted from this section.
Figure 3.30a shows the airspeed step response of the aircraft during a flight test, and
Figure 3.30b shows the corresponding throttle and altitude responses. The airspeed
of the aircraft as measured by the pressure sensors of a pitot tube, the velocity of the
aircraft as measured inertially by the GPS as well as the velocity determined by the
estimator of the aircraft, are shown in Figure 3.30a. If the aircraft is flying straight and
level, and there is zero wind, then the inertially measured aircraft velocity should closely
match the true airspeed measured by the pitot tube. The climb rate of the aircraft will
have some effect on the inertially measured velocity, and when this is accounted for,
the remaining difference is due to wind disturbances. It can be seen that the airspeed
follows the reference well and that the rise time of the response resembles that of the
pole placement design.
In this project, the ability of the airspeed controller to maintain the airspeed during
the descent phase of the landing is of paramount importance. Figure 3.31a shows the
airspeed of the aircraft for a flight path circuit similar to that of Figure 3.28b where
the aircraft descends and ascends a sloped path simulating a landing approach. The
20 30
19
20
18
10
17
16 0
0 5 10 15 20 25 30 35 40
15
14 56
54
13
52
12 50
48
11
46
10 44
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
(a) Step response (b) Throttle actuator and altitude
Figure 3.30 – Airspeed step with corresponding throttle and altitude responses during a
flight test
altitude of the aircraft during this manoeuvre is shown in Figure 3.31b along with the
corresponding throttle actuator. It must be noted that this flight test plot was selected
specifically because it shows a very high degree of wind disturbance. It can be seen that
in spite of the presence of wind disturbances the airspeed is well maintained during the
descent and ascent.
20
30
19
20
18
17 10
16
0
0 20 40 60 80 100
15
14 70
13 60
12 50
11 40
10 30
0 20 40 60 80 100 0 20 40 60 80 100
(a) Airspeed response (b) Throttle actuator and altitude
Figure 3.31 – Airspeed with corresponding throttle and altitude responses for sloped altitude
path during a flight test
Chapter 4
Lateral Analysis and Control
During landing, the flight path of the aircraft in the inertial axis system must be con-
trolled such that it coincides with that of the runway. In order to accomplish this, several
controllers which operate in the lateral axis system of the aircraft are required. These
controllers are presented in this chapter, while the specifics of the landing procedure are
discussed in Chapter 5. This chapter also provides an analysis of the lateral dynamics of
the aircraft, followed by the design, HIL simulation and practical flight test results of the
roll angle controller and the Dutch roll damper. The chapter ends with an outline, HIL
simulation and practical flight test results of the guidance controller.
As highlighted in Figure 4.1 the lateral flight control system consists of a roll angle
controller, Dutch roll damper and a guidance controller.
Figure 4.1 – Longitudinal Control System Overview Block Diagram
The roll angle controller and Dutch roll damper are designed around the linear, de-
coupled lateral dynamics of Equations 2.4.10 and 2.4.11. The roll angle controller is
capable of using the aircraft’s ailerons in order to create a commanded wind axes roll
angle, and receives its input from the guidance controller. The Dutch roll damper makes
71
CHAPTER 4. LATERAL ANALYSIS AND CONTROL 72
use of the aircraft’s rudder in order to damp the Dutch roll mode of the aircraft. The
guidance controller receives its input from the landing state machine as a path to be fol-
lowed, and does so by commanding a lateral acceleration in the navigation axis system
which, in combination with the commanded normal acceleration in the navigation axis
system, is then converted to a wind axis system NSA and roll angle.
The analysis of the lateral dynamics of the aircraft, and the subsequent design of ap-
propriate controllers has already been extensively covered by [2; 1] and the controllers
presented in this section are based directly on that work. The lateral control system in
[2; 1] also made use of a lateral specific acceleration (LSA) controller which enables the
aircraft to fly coordinated turns. Such a controller might be useful during manouvres
such as crosswind landings, but since these fall outside the scope of this project an LSA
controller was not implemented. The guidance controller implemented was developed
by [3].
4.1 Decoupling the Lateral Dynamics
For convenience, the simplified lateral dynamics of Equations 2.4.10 and 2.4.11 are
shown again,
Lateral Dynamics
   1 1 1
   1 1
 g 
β̇ Ȳ
mV̄ β
Ȳ
mV̄ P
−1 + mV̄
ȲR β Ȳ
mV̄ δA
Ȳ
mV̄ δR
" #
V̄
cos θ sin φ
   1 1 1    1 1
δA
Ṗ  =  Ix Lβ L L  P  +  Ix LδA L + 0
  
Ix P Ix R Ix δ R  
1 1 1 1 1
δR
Ṙ N
Iz β
N
Iz P
N
Iz R
R N
Iz δA Iz
N δ R
0
(4.1.1)
 
" # " Ȳ # β " Ȳ ȲδR
#" #
β ȲP ȲR δA
BW m m m   m m
δA
= P
  + (4.1.2)
PW 0 1 0 0 0 δR
R
According to [2], these dynamics can be decoupled into roll mode and directional dy-
namics. This then allows for the design of separate roll angle and lateral acceleration
controllers. However, since a lateral acceleration controller is not implemented in this
project, the directional dynamics will only be used in the design of the Dutch roll damper.
The lateral dynamics are simplified by assuming that the lateral force generated by an ail-
eron deflection and roll rates is negligible. It is also assumed that the coupling between
the directional and the roll rate dynamics is weak. The lateral decoupling assumption
holds for most conventional aircraft such as the one used in this project. A detailed list
of conditions under which the lateral decoupling assumption can be made is presented
in [2]. These assumptions allow for separate controllers to be designed for the roll angle
and Dutch roll mode where any aileron and rudder couplings into the directional and roll
dynamics respectively will be considered disturbances.
The decoupled models are given by,
Roll Dynamics
h i
1 1
Ṗ = LP P + L δ δA (4.1.3)
Ix Ix A
PW = P (4.1.4)
Directional Dynamics
1 1 1
   
"g
−1
" # " # #
β̇ Ȳ β + Ȳ R β Ȳδ R cos θ sin φ
=  m1V̄ mV̄  +  m1V̄  δR + V̄ (4.1.5)
  
Ṙ 1 
R
Nβ NR Nδ R 0
Iz Iz Iz
" #
Ȳβ ȲR β ȲδR
BW = + δR (4.1.6)
m m R m
4.2 Roll Angle Controller
The design, simulation and practical testing of a roll angle controller which is based on
the work of [2; 1] is presented in this section. This controller regulates the roll angle of
the aircraft by making use of the ailerons. The input to the roll angle controller stems
from the guidance controller of §4.4.
4.2.1 Design
The decoupled equations for the roll dynamics of Equations 4.1.3 and 4.1.4 are restated
for convenience,

1 1
Ṗ = LP P + Lδ δA (4.2.1)
Ix Ix A
PW = P (4.2.2)
In order to remove any steady state errors such as those arising due to the cross coupling
effects from the directional dynamics, and to counter other disturbances, an integrator
is augmented to the system as follows,
ĖP = P − P ref (4.2.3)
where P ref is the reference roll rate. The system now becomes,
 "
#  "
−
" # #
ĖP 0 1 E 1 0 P ref
=  1  P + 1  (4.2.4)
Ṗ 0 LP P 0 LδA δA
Ix Ix
With an appropriate aileron control law, the derivation thus far allows for the design of a
roll rate controller. In order to control the roll angle, the dynamics must however again
be expanded to include the roll angle error,
φ̇E = P ref − P (4.2.5)
The system thus becomes,

      
ĖP 0 1 0 E − 1 0
 P 
" #
   1 1  P ref
0 Ix LP 0  P  +  0
 Ṗ  =  Lδ  (4.2.6)
 
Ix A  δA
φ̇E 0 −1 0 φE 0 0
Two control laws are now defined. The first is an aileron control law which will control
the roll rate segment of the controller by making use of feedback from the roll rate and
roll rate integrator states. The second control law is for the roll rate reference and makes
use of feedback from the roll angle error. Thus,
δA = −KP P − KE EP (4.2.7)
P ref = Kφ φE (4.2.8)
The control laws are now substituted into the dynamics and the closed-loop system be-
comes,
    
ĖP 0 1 −Kφ E
 L δ LP LδA  P
 Ṗ  = − A KE − KP 0  P  (4.2.9)
   
Ix Ix Ix
φ̇E 0 −1 0 φE

3 L δA LP 2 L δA L δA
p(s) = s + KP − s + KE s + Kφ KE (4.2.10)
Ix Ix Ix Ix
In order to place the closed-loop poles a desired characteristic equation for the roll
αc (s) = s3 + α2 s2 + α1 s + α0 (4.2.11)
By equating the coefficients of Equations 4.2.10 and 4.2.11 the control gains can be
calculated as,

Ix LP
KP = α2 + (4.2.12)
L δA Ix
Ix
KE = α1 (4.2.13)
L δA
α0
Kφ = (4.2.14)
α1
The reference roll angle of the controller enters the system as follows,
φE = φref − φ (4.2.15)
The pole placement strategy involves selecting the location of three closed-loop poles
and then determining the coefficients of the resulting desired characteristic equation.
This then allows the control gains to be determined. The three poles in question are the
roll rate, roll rate integrator and the error angle pole. The pole placements which follow
are for the aircraft parameters and standard flight conditions as outlined in Appendix
B.6.
The roll rate pole stems from the roll rate dynamics of Equation 4.2.1 and is fixed as,
LP
p1 = (4.2.16)
Ix
≈ −8 rad/s
The location of the roll rate integrator and error angle poles are selected as,
1
p2 = p1 = −4 rad/s (4.2.17)
2
and,
1
p3 = p2 = −2 rad/s (4.2.18)
2
respectively.
The coefficients of the desired characteristic equation of Equation 4.2.11 can now be de-
termined and the results substituted into Equations 4.2.12 to 4.2.14 in order to calculate
the controller gains. The resulting pole-zero map and step response of the controller are
shown in Figures 4.2a and 4.2b.
5 11
4 10
3
9
8
2
7
1
6
0
5
−1
4
−2
3
−3
2
−4 1
−5 0
−10 −9 −8 −7 −6 −5 −4 −3 −2 −1 0 0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 4.2 – Pole-zero map and step response of the roll angle controller
The roll angle controller was implemented on an OBC and tested in the HIL simulation
Figures 4.3a and 4.3b show the roll angle response with the corresponding roll rate re-
sponse and aileron deflection angle during HIL simulation, and Figures 4.3c and 4.3d
show the same for a practical flight test. Firstly, it can be seen that the HIL simulation
and practical flight test roll angle responses correspond well with one another, with the
exception that the flight test exhibits somewhat greater disturbances. The cause of these
disturbances can be seen in the plot of the roll rate response where there is significant
disturbances on the roll rate. It must be noted that the flight test data shown was se-
lected specifically since atmospheric conditions were unfavourable during the test. The
origin of this disturbance is therefore mostly aerodynamic in nature. This is confirmed
by the roll angle step response shown in Figure 4.4 where it can be seen that the roll
rate disturbances are significantly less than those of Equation 4.3d. This demonstrates
that the controller continues to function in the presence of significant atmospheric dis-
turbances.
Figures 4.4a and 4.4b show the roll angle response with the corresponding roll rate
response and aileron deflection angle during a flight test with favourable atmospheric
conditions. It can be seen that the roll angle step response corresponds well with the
response obtained in the pole placement of the controller as shown in Figure 4.2b. It
can also be seen that the aileron deflection angle remains well within the limits defined
in Appendix B.5. This controller design can therefore be considered a success.
20 30
20
10
10
0
−10
0 −20
−30
0 5 10 15 20 25 30
−10
−20 5
0
−30
−5
−40
0 5 10 15 20 25 30 0 5 10 15 20 25 30
(a) HIL roll angle response (b) HIL roll rate and aileron deflection
20 30
20
10
10
0
−10
0 −20
−30
0 5 10 15 20 25 30
−10
−20 5
0
−30
−5
−40
0 5 10 15 20 25 30 0 5 10 15 20 25 30
(c) Flight test roll angle response (d) Flight test roll rate and aileron deflection
Figure 4.3 – Roll angle response with corresponding roll rate response and aileron deflection
angle during HIL simulation and flight tests
5 30
20
10
0 0
−10
−20
−5 −30
0 2 4 6 8 10
−10 5
−15 0
−20 −5
0 2 4 6 8 10 0 2 4 6 8 10
(a) Roll angle step response (b) Roll rate and aileron deflection
Figure 4.4 – Roll angle step response with corresponding roll rate response and aileron
deflection angle during flight tests
4.3 Dutch Roll Damper
This section describes the design, simulation and practical testing of a Dutch roll damper
as presented by [1]. The Dutch roll mode of the aircraft is a lightly damped oscillation
in yaw, which couples into roll. This controller damps the Dutch roll mode poles of the
aircraft by means of the rudder actuator in order to improve the overall flying qualities
of the aircraft.
4.3.1 Design
The control strategy involves utilising a high pass filter on the yaw rate feedback signal
to the rudder. The transfer function of this filter is given by,
KDR s
D(s) = (4.3.1)
s + ωf
where KDR is the feedback gain and ωf the filter’s cut-off frequency. This filter has
the effect of countering higher frequency Dutch roll effects while filtering out any low
frequency effects. This ensures that the controller does not attempt to actively counter
constant turn rates.
The decoupled equation for the directional dynamics of Equation 4.1.5 is restated below
for convenience,
1 1 1
   
Ȳβ −1 +
" # " #
β̇ ȲR β  mV̄ ȲδR 
=  m1V̄ mV̄  +  δR (4.3.2)

Ṙ 1 
R
 1
Nβ NR Nδ R
Iz Iz Iz
where the gravitational coupling term has been ignored [2]. Calculating the open-loop
characteristic equation gives,

21 1 1 1 1
p(s) = s + − NR − Ȳβ s + Ȳβ NR + Nβ − Nβ ȲR (4.3.3)
Iz mV̄ Iz mV̄ mV̄
The frequency of the open-loop Dutch roll mode poles is therefore given by,
s
1 1 1
ωDR = Ȳβ NR + Nβ − Nβ ȲR (4.3.4)
Iz mV̄ mV̄
Since the poles and zeros of the lateral dynamics scale uniformly with airspeed, [1]
presented the following general equation for the Dutch roll damper which will function
over the entire airspeed range of the aircraft,
0
KDR s
D(s) = (4.3.5)
ηd ωDR (s + ηd ωDR )
0
where KDR is the normalised feedback gain, ωDR is the frequency of the open-loop Dutch
roll mode poles and ηd is the fraction of this frequency where the filter will cut off.
The pole placements in this section are for the aircraft parameters and standard flight
conditions as outlined in Appendix B.6.
The cut-off frequency of the filter is chosen as,
1
ωf = ηd ωDR where, ηd = (4.3.6)
3
This ensures that the Dutch roll mode response lies within the filter’s passband whilst
also filtering out low frequency effects such as constant turn rates.
The feedback gain of the filter at the aircraft’s trim velocity must now be determined. In
order to accomplish this, the root locus for a variation in the filter feedback gain of the
closed-loop system is obtained as shown in Figure 4.5.
5
0.92 0.85 0.76 0.62 0.44 0.22
4
0.965
2
0.992
14 12 10 8 6 4 2
0
−1
0.992
−2
−3
0.965
−4
0.92 0.85 0.76 0.62 0.44 0.22

−5
−14 −12 −10 −8 −6 −4 −2 0
Figure 4.5 – Root locus design of the Dutch roll damper at the aircraft’s trim velocity
It can be seen that the open-loop damping ratio of the Dutch roll mode poles is given by,
ζol = 0.14 (4.3.7)
which is very lightly damped. Selecting the feedback gain as,
KDR = −0.214 (4.3.8)
results in the closed-loop poles shown, and yields a closed-loop damping ratio of,
ζcl = 0.57 (4.3.9)
The filter’s transfer function at trim velocity is therefore given by,
−0.214s
Dtrim (s) = (4.3.10)
s + 1.4
The normalised feedback gain can now be calculated as,
0
KDR = KDR (ωDR ηd ) (4.3.11)
= −0.3 (4.3.12)
0
During implementation of this controller, substituting the gain KDR from Equation 4.3.11
alongside ωDR from Equation 4.3.4 as well as ηd from Equation 4.3.6 into Equation 4.3.5
leads to the transfer function of the Dutch roll damper which will function at the current
airspeed of the aircraft.
The Dutch roll damper was implemented on an OBC and tested in the HIL simulation
Figures 4.6a and 4.6b show the yaw rate and rudder deflections during a HIL simulation
and flight test respectively. These responses are for the same set of HIL and flight test
results shown in Figure 4.3 where the aircraft is undergoing large roll angle variations.
It can be seen that the HIL simulation and flight test results are somewhat similar, but
that there is significantly more disturbances in the case of the flight test.
20 20
10 10
0 0
−10 −10
−20 −20
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
3 3
2 2
1 1
0 0
−1 −1
−2 −2
−3 −3
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
(a) Dutch roll damper during HIL (b) Dutch roll damper during flight test
Figure 4.6 – Yaw rate response and rudder deflection angle during HIL simulation and flight
tests
Figures 4.7a and 4.7b show the yaw rate and rudder deflections during a flight test where
the Dutch roll damper is disabled and enabled respectively. During these tests, the safety
pilot manually generated a series of rudder deflections in order to demonstrate the effect
of the Dutch roll damper on the rudder actuator as well as to excite the Dutch roll mode
dynamics of the aircraft. Figure 4.7b shows the rudder deflections commanded by the
safety pilot, the rudder deflection commanded by the Dutch roll damper and the total
rudder deflection. It can be seen that as the yaw rate increases so does the Dutch roll
damper rudder command. This command is in the opposite direction to that of the safety
pilot thereby resulting in a smaller total rudder deflection. This demonstrates the ability
of the Dutch roll damper to oppose high yaw rates. Upon completion of the final safety
pilot rudder step the excited Dutch roll mode of the aircraft can be observed. It can be
seen that with the Dutch roll damper enabled, the yaw rate oscillations are significantly
more damped than with the Dutch roll damper disabled.
100 100
50 50
0 0
−50 −50
−100 −100
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
20 20
10 10
0 0
−10 −10
−20 −20
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
(a) Dutch roll damper disabled (b) Dutch roll damper enabled
Figure 4.7 – Yaw rate response to safety pilot rudder perturbations for Dutch roll damper
disabled and enabled during flight test
4.4 Aircraft Guidance
Guidance Controller The purpose of the guidance controller is to control the aircraft’s
flight path in the XY-plane of the inertial axis system. The guidance controller used was
developed by [3] and only the outline is presented in this section.
The guidance controller receives a desired flight path from the landing state machine,
ref
and then commands a lateral acceleration BN in order to move the aircraft towards the
path. The flight path is generated between waypoints as shown in Figure 5.5.
The guidance logic consists of two parts, first a reference point is selected on the desired
path at a certain distance (L1 ) ahead of the aircraft. Secondly a lateral acceleration
command is generated by using the aircraft’s airspeed, the length (L1 ) as well as the
angle (η ) between the velocity vector of the aircraft and the reference length. This is
depicted in Figure 4.8 and the lateral acceleration command is generated by,
ref V̄ 2
−BN =2 sin η (4.4.1)
L1
Figure 4.8 – Guidance logic [3]
The guidance controller has the effect that the aircraft will tend to align its velocity
direction with the direction of the reference point length (L1 ). When the aircraft is
far away from the desired path, the vehicle rotates its velocity vector to approach the
desired path at a large angle, and when the vehicle is close to the desired path, the
vehicle rotates its velocity vector to approach the desired path at a small angle. This
results in a smooth overall guidance path.
Linearisation of the nonlinear guidance logic leads to a differential equation for a con-
troller which regulates the cross track error to zero. The poles of this controller are
given by [7],
V̄ V̄
s=− ± i (4.4.2)
L1 L1
The guidance controller is thus linearly approximated as a critically damped pole pair
with natural frequency,
√ V̄
ωn = 2 rad/s (4.4.3)
L1
The design choice for the guidance controller is selecting an appropriate length (L1 ) for
the distance between the aircraft and the reference point. A shorter length will result in
a more aggressive controller. For the aircraft parameters and standard flight conditions
as outlined in Appendix B.6 it was found during HIL simulation and practical flight tests
that a satisfactory response is obtained for a length of,
L1 = 100 m (4.4.4)
which results in a natural frequency of,
ωn = 0.25 rad/s (4.4.5)
NSA and Roll Angle Command Calculation The lateral acceleration command gen-
ref
erated by the guidance controller (BN ) and the normal acceleration command gener-
ref
ated by the climb rate controller (CN ) are both in the navigation axis system and cannot
Figure 4.9 – Conversion of lateral and normal acceleration in navigation axes to wind axes
NSA and roll angle
be controlled directly. These commands must therefore be converted to a NSA reference

ref
command (CW ) and a roll angle (φref ) as shown in Figure 4.9.
The roll angle reference command is therefore given by,

!
ref
BN
φref = arctan − (4.4.6)
CNref
where it is assumed that the roll angle remains within the controller limits as defined in
Appendix B.5. The NSA reference command is given by,
r 2 2
ref
CW =− CNref ref
+ BN (4.4.7)
It must be noted that since the guidance controller lacks an integrator term, the roll
angle must be exactly achieved as is the case for the roll angle controller of §4.2. How-
ever, a practical constraint emerges in that during mounting, the sensors might not be
exactly aligned with the axes of the aircraft. In particular, the roll angle determined by
the estimator could have an offset relative to the actual roll angle of the aircraft. This
will in turn result in the guidance controller following a straight path with a constant
offset which is proportional to the error in roll angle. This problem is overcome via calib-
ration; adding an appropriate offset to the roll angle measurement such that it matches
the actual roll angle of the aircraft. This roll angle offset can also be exploited during
landing since it allows small corrections to be made to the flight path of the aircraft as is
flies along the runway.
The guidance controller was implemented on an OBC and tested in the HIL simulation
Figure 4.10a shows the flight path of the aircraft and Figure 4.10b the corresponding
roll angle, navigation axis LSA and wind axis NSA reference commands during HIL sim-
ulation, whilst Figures 4.10c and 4.10d show the same during a practical flight test.
It can be seen that the flight test results compare well with those obtained in the HIL
simulation. It can also be seen that the aircraft’s path only converges on the reference
path after approximately 250 m. This is of importance during landing since sufficient dis-
tance must be allowed for in the landing procedure to allow the guidance controller to
settle. Decreasing the guidance length (L1 ) will decrease the distance required for the
controller to settle, however, this comes at the cost of potentially higher roll angles at
the touchdown point since the controller will react more aggressively to any path error.
0
20
−50 0
−20
−100
0 5 10 15 20
−150 5
0
−200
−5
−250 0 5 10 15 20
−5
−300
−10
−350
−15
−150 −100 −50 0 50 100 150 200 250 0 5 10 15 20
(a) HIL flight path (b) HIL controller reference commands
0
20
−50 0
−20
−100
0 5 10 15 20
−150 5
0
−200
−5
−250 0 5 10 15 20
−5
−300
−10
−350
−15
−150 −100 −50 0 50 100 150 200 250 0 5 10 15 20
(c) Flight test flight path (d) Flight test controller reference com-
mands
Figure 4.10 – Flight path and corresponding roll angle, navigation axis LSA and wind axis
NSA reference commands
Chapter 5
Landing
The controllers developed in Chapters 3 and 4 allow for the control of the longitudinal
and lateral motion of the aircraft respectively. This chapter shows how this control is
utilised to accomplish an aircraft landing.
The landing control of the aircraft can be split into three broad categories. Firstly, the
flight path of the aircraft must be controlled to coincide with the runway. This is accom-
plished by means of the guidance controller of §4.4. Secondly, the altitude of the aircraft
must be controlled such that the aircraft descends along a specific path onto the runway.
This control is accomplished by means of the altitude controller of §3.4. Thirdly, the air-
speed of the aircraft must be maintained throughout the landing. This is accomplished
be means of the airspeed controller of §3.5.
For an accurate landing to be performed, the controllers must strictly regulate their
respective parameters. The biggest threat to this is the presence of wind disturbances
of which crosswinds and wind gusts are of greatest concern. Crosswind landings fall
outside the scope of this project and have been discussed in [5]. The focus falls on wind
gusts, the impact of which is more pronounced during landing due to the small size and
weight of the aircraft used in this project. The direct-lift-augmented NSA controller of
§3.2.6 was designed to provide improved rejection of these disturbances when compared
to the conventional NSA controller of §3.2.4. These two controllers will be tested and
compared.
The procedure followed during a landing is discussed in this chapter. The flight and
descent paths are thereafter investigated, followed by an overview of the landing state
machine. The chapter ends with the flight test results of several autonomous landings
where both the conventional and the direct-lift-augmented NSA controllers were utilised.
85
CHAPTER 5. LANDING 86
5.1 Landing Procedure
5.1.1 Standard Aircraft Landing Procedure
Whilst developing a suitable landing procedure for this project, it is useful to cross-
reference the procedures used by piloted aircraft at small airfields. Such a standard
aircraft landing procedure is presented in [20] and consists of several phases. These
phases are discussed briefly with reference to the standard rectangular airfield traffic
pattern illustrated in Figure 5.1, and its corresponding landing path geometry as illus-
trated in Figure 5.2.
Figure 5.1 – Standard rectangular airfield traffic pattern
Figure 5.2 – Standard landing path geometry
Enter and maintain circuit The purpose of the landing circuit is to allow the aircraft’s
position, airspeed and altitude to settle into a controlled state such that only small
corrections are required as the aircraft transitions into the final phase of the land-
ing sequence. The aircraft enters the landing circuit on the downwind leg. It then
follows the circuit at a constant altitude until permission to land is granted by the
air traffic control authority. Once this permission is obtained, the airplane will
commence the actual landing when it reaches the subsequent base leg.
Base leg When the aircraft enters the base leg for a landing approach its airspeed is
reduced to the recommended landing airspeed. This is typically 1.4 times the stall
speed of the aircraft. The aircraft then flies perpendicular to the runway, and at the
appropriate distance from the runway centreline turns onto the final approach leg.
During this turn the aircraft must not exceed a shallow to medium roll angle since
the stall speed of the aircraft increases at higher roll angles. After completion of
the turn, the aircraft’s flight path must be aligned with the centreline of the runway.
Final approach Assuming that there is no crosswind, the longitudinal axis of the air-
craft should be kept aligned with the centreline of the runway during the final
approach. The final approach leg consists of a descending flight path which origin-
ates from the base-to-final turn, and extends down towards the intended touchdown
point on the runway. This point should be within the first one-third of the runway.
Precise control of the aircraft is essential during the final approach.
Flare The flare is a smooth transition from the approach attitude to a landing attitude.
During the flare, the descending flight path of the final approach is gradually roun-
ded out until the flight path of the aircraft is parallel with the runway. This sheds
much of the remaining kinetic energy of the aircraft. At this point the aircraft
touches down on the runway.
Touchdown Ideally, the aircraft should touch down gently on the main landing gear
first, and at the lowest controllable airspeed.
Rollout After touchdown, control of the aircraft must be maintained whilst it deceler-
ates to the normal taxi speed. The landing is now complete.
Go-arounds If at any point during the landing procedure the aircraft’s airspeed, posi-
tion or attitude deviate significantly from the normal state, or if any landing con-
dition is not satisfactory, the landing is rejected and a go-around initiated. The
aircraft then re-enters the circuit, and the landing is attempted again.
5.1.2 Modified Aircraft Landing Procedure
The standard aircraft landing procedure described in §5.1.1 is now modified to more
effectively accommodate the project goals and the aircraft used as a testbed. One of the
primary goals of this thesis is an accurate autonomous landing and the adaptation of the
standard landing procedure will focus on achieving this outcome.
The standard rectangular landing circuit will be used as is, only the path of the final
approach leg is modified in order to minimise any potential path following errors which
may arise. These errors can have a significant effect on the accuracy with which the
aircraft touches down, as illustrated in Figure 5.3 where the intended and the actual
straight approach paths are shown with a constant offset error.
Figure 5.3 – Effect of an altitude error in the landing path on the touchdown point
The angle of the approach path γ is typically around 3 ◦ [19]. This means that a small
error in height will cause a large error between the actual and the intended landing
point. With reference to Figure 5.3, this is expressed mathematically as,
herror
derror = (5.1.1)
tan γ
Figure 5.3 also shows the effect of a height error when a flare is performed. Any error
between the intended and the actual approach paths is now of even greater consequence
due to the fact that the angle of the flight path approaches 0 ◦ towards the end of the
flare. This means that for even a small error in height, the aircraft could potentially
travel a great distance along the runway before touching down. This is usually of little
consequence during normal landings. However, if an accurate landing is desired, the
flare manouvre presents a significant drawback.
Another significant drawback of the flare manouvre is that, due to the nonlinear nature
of the wing when it is close to stall, the precise trajectory of the flare is unknown [7].
This means that the relationship between the flare height hflare and the distance dflare
cannot be accurately determined. This is aggravated further when the effects of wind
disturbances are taken into account. Therefore, to aid in the landing accuracy of the
aircraft, the flare is omitted from the landing procedure, and the aircraft is simply flown
along the straight path of the final approach leg. This omission comes at the cost of
the gentle touchdown which would otherwise have occurred with the use of the flare,
and increases the likelihood of undercarriage damage or bouncing of the aircraft during
landing.
The standard landing path is further modified to accommodate the specific layout of
the airfield where the practical flight tests are conducted (see Appendix D.1). Firstly,
the glide path is split into two phases of descent as shown in Figure 5.4. The first
descent phase is at a steeper angle than the last phase so as to provide sufficient obstacle
clearance during the landing. This intermediate slope also means that the aircraft flies
at a higher altitude when it is still far away from the runway than it would have done
without this slope. This provides an additional altitude margin for the safety pilot to
be able to observe, and act to recover the aircraft from, any anomalies during flight.
Secondly, the base leg, and those that precede it, is set at a higher altitude than the start
of the final approach leg. This is again done to reduce the risks associated with low level
flight, especially at the large distance between the aircraft and the safety pilot.
Figure 5.4 – Landing path geometry
The runway circuit is defined by four waypoints as shown in Figure 5.5. These waypoints
are also set to allow for the specific layout of the airfield where flight tests are conducted.
The waypoint locations allow for the avoidance of obstacles, and also allow for sufficient
distance for the guidance controller to settle once the aircraft has turned from the base
onto the final leg.
Figure 5.5 – Landing circuit geometry

It must be noted that the landing approach followed is very similar to that employed by an
aircraft landing on the deck of an aircraft carrier. An outline of the specific procedure is
presented by [19]. In essence, such a landing also makes use of a straight approach path
with no flare so as to ensure the accuracy of the touchdown point. When the aircraft
touches down, it is captured by an arrestor cable by means of a hook attached to the
rear of the aircraft. This cable then brings the aircraft to a stop. This means that the
aircraft does not perform any runway control. If a similar arrestor mechanism is used
in this project, the need for runway control also falls away and the entire landing could
be conducted autonomously. Although such a mechanism falls outside the scope of this
project, the effect of using an arrestor mechanism is in essence simulated during the
practical flight tests where the safety pilot performs the same function by stopping the
aircraft after touchdown has been achieved by the autopilot.
After the modifications to the standard landing procedure, the landing circuit is as shown
in Figure 5.5, and the corresponding landing path geometry is as shown in Figure 5.4.
The specific details contained in these figures are discussed in the relevant sections
which follow.
5.2 Landing Circuit
The landing circuit is the horizontal path followed by the aircraft during landing. The
procedure followed is best explained in the context of a practical flight test. During a
practical flight test, the safety pilot controls the aircraft during take-off, and positions it
in the entry area region as shown in Figure 5.5 at a moderate altitude. Upon enabling
the autopilot, the guidance controller will steer the aircraft towards the path between
waypoints 1 and 2, while the altitude controller will move the aircraft towards the circuit
altitude. The airspeed controller will maintain the aircraft’s airspeed at the circuit air-
speed. When the aircraft is sufficiently close to waypoint 2, the guidance controller will
transition to the base leg causing the aircraft to turn. The same occurs for every waypo-
int reached, and the aircraft will maintain this guidance circuit until a landing command
is issued. When a landing command is received, the same circuit is flown, however, in-
stead of following a constant altitude, the longitudinal landing path controller will cause
the aircraft to descend towards the runway as discussed in §5.3. This altitude control
is completely independent of the landing circuit control. Upon touchdown the aircraft’s
controllers are disabled, and the safety pilot brings the aircraft to a stop.
As shown in Figure 5.5 the landing circuit is defined in the runway axis system of §2.1.1.
It can be seen that the runway axis system is rotated from the inertial axis system by an
angle ΨR . The waypoints are converted from the runway to the inertial axis system on
the groundstation before being sent to the OBC.
The values for the waypoints used are,
W P1North = 220 m W P1East = −180 m (5.2.1)

W P2North = −350 m W P2East = −180 m (5.2.2)
W P3North = −350 m W P3East = 0 m (5.2.3)
W P4North = 220 m W P4East = 0 m (5.2.4)
and the angle between the runway and the inertial axis system is given by,
ΨR = 19 ◦ (5.2.5)
The HIL simulation and practical flight test results of all of the controllers involved in
executing the landing circuit are shown in Chapters 3 and 4. In order to confirm the
proper execution of the landing circuit, the waypoints were entered into the OBC, and
the circuit flown in the HIL simulation environment as well as practically in a series of
flight tests. The HIL simulation environment, aircraft, avionics and other hardware used
in this project are outlined in Appendix B, only the results are presented in this section.
The HIL tests allow the landing circuit controllers to be fully tested before practical flight
tests are attempted. Ideally, the practical flight test results should closely resemble those
achieved in the HIL simulation.
Figures 5.6a and 5.6b show the landing circuit of the aircraft during a HIL simulation
and a practical flight test respectively. In these tests, the autopilot is enabled at the point
indicated. The aircraft then flies the entire circuit, and on the second pass descends and
lands. It can be seen that the flight test result corresponds well with that obtained during
HIL simulation, and that the path of the aircraft is aligned closely with the centreline of
the runway at the touchdown point.
300 300
250 250
200 200
150 150
100 100
50 50
0 0
−50 −50
−100 −100
−300 −200 −100 0 100 200 −300 −200 −100 0 100 200
(a) HIL landing circuit (b) Flight test landing circuit
Figure 5.6 – Landing circuit in runway axis system during HIL simulation and flight test
5.3 Longitudinal Landing Path
The longitudinal landing path is the altitude path followed by the aircraft as it descends
during the landing. This path is shown in Figure 5.4, and the path variables are defined
as,
Base leg height (hbase ) - the height of the base leg above the touchdown point
Final approach leg height (happ ) - the height of the final approach leg above the touch-
down point, this is also the height of the start of the glide path
Slope transition point height (htp ) - the height of the slope transition point above the
touchdown point
Glide path start distance (dgp ) - the distance between the start of the glide path, and the
touchdown point
Slope transition point distance (dtp ) - the distance between the slope transition point,
and the touchdown point
Initial glide path angle (γ2 ) - the angle of the glide path directly following the glide path
start point
Final glide path angle (γ1 ) - the angle of the glide path directly following the slope
transition point
The values of the longitudinal path variables are,
γ1 = 3 ◦ (5.3.1)
γ2 = 7 ◦ (5.3.2)
dtp = 70 m (5.3.3)
dgp = 200 m (5.3.4)
The final glide path angle (γ1 ) was selected to provide a relatively gentle touchdown. This
angle could be increased which would allow a more accurate landing due the reduction
in the relationship between a height error and the actual and intended touchdown points
as shown in Figure 5.3. However, the landing gear of the aircraft would first need to be
strengthened to absorb the increased impact force which will result.
During a practical flight test, the aircraft is positioned on the runway at the point where
touchdown during the landing is intended. The aircraft’s sensors are then zeroed, and
the estimator started (see Appendix B). The touchdown point therefore serves as the
reference point for the landing circuit and longitudinal landing paths, and the altitude
commands are calculated based on the horizontal distance between the aircraft and this
point.
During the design of the altitude controller of §3.4 it was noted that the altitude control
law intentionally lacks integrator feedback, and makes use of only proportional feedback
from the climb rate. This results in the altitude controller tracking a ramp command
with a steady state error which means that the landing approach path will be followed
with a constant altitude offset. The motivation behind this control law design choice
is twofold. Firstly, the total duration of the glide path is not very long, therefore an
integrator term on the altitude might not have sufficient time to settle the altitude to the
reference value before touchdown occurs. Secondly, an integrator will have almost no
effect on minimising random disturbances such as wind gusts. The practical implication
of this control decision is that, for the aircraft to land at a specific point on the runway,
the commanded touchdown point must be positioned below the runway by the same
distance as the altitude offset. This is illustrated in Figure 5.7.
Figure 5.7 – Glide path reference offset
The altitude control is a type one system, therefore the steady state error to a ramp input
r(t) = At is given by,
A
eSS = (5.3.5)
KVec
where KVec is the velocity error constant. As shown in Appendix C.4,
KVec = KPh τCR (5.3.6)
KPh is the altitude feedback gain. The actual steady state error is greater than Equa-
tion 5.3.5 suggests, due to inaccuracies in the model. It is therefore best determined
experimentally during flight tests.
Figure 5.7 reveals an additional implication of the glide path offset. It is evident that the
actual touchdown position of the aircraft can be directly affected by simply changing the
glide path offset. An increase in the glide path offset will result in the aircraft touch-
ing down ahead of the reference touchdown point, whereas a decrease will result in the
aircraft touching down beyond the reference point. Alternatively, it means that any con-
stant error acting on the aircraft and affecting the accuracy of the touchdown point can
be compensated for by an appropriate glide path offset. Such errors arise mainly from
constant wind, ground effect and aircraft configuration changes. Therefore, to achieve
improved landing accuracy, a runway flyby can be performed with a mock landing at a
moderate altitude. The glide path offset can then be determined and the reference up-
dated accordingly. This is shown in Figures 5.8a and 5.8b where the actual and reference
glide paths are shown for both a HIL simulation and a flight test, and where the glide
path offset at the mock touchdown point is shown. Next, a landing is performed. The
error in touchdown position can then be used to calculate the required change in offset
in order to compensate for any additional sources of error such as ground effect. In this
manner, the accuracy of the touchdown point can be improved.
30 30
25 25
20 20
15 15
10 10
5 5
0 0
−200 −150 −100 −50 0 50 −200 −150 −100 −50 0 50
(a) HIL landing offset (b) Flight test landing offset
Figure 5.8 – Landing longitudinal path for a mock landing during HIL simulation and flight
test
5.4 Landing State Machine
The landing state machine is responsible for the overall control of the aircraft to ac-
complish a landing. It does so by providing the reference commands of the airspeed,
altitude and guidance controller as shown in Figure 5.9. These references are set on the
groundstation and can be updated during flight. During a landing, the altitude reference
is calculated on board the aircraft and is based on the longitudinal landing path shown in
Figure 5.4, whereas the guidance controller’s reference is based on the landing circuit
shown in Figure 5.5. The airspeed reference is typically fixed for the duration of the
landing.
Figure 5.9 – Control System Overview Block Diagram
The logic of the landing state machine is depicted in Figure 5.10. The aircraft is posi-
tioned in the air by the safety pilot in the region shown in Figure 5.5. When the autopilot
is armed, the state machine is entered, and the airspeed and altitude references sent to
their respective controllers. The first two waypoints are sent to the guidance control-
ler and the aircraft begins following the reference path. The landing state machine will
update the waypoints when the aircraft is sufficiently close to its current destination way-
point. In this manner the circuit is flown at a constant airspeed and altitude. When the
landing command is issued, the airspeed and circuit are maintained as before, however,
the altitude reference now becomes dependent on the current position of the aircraft.
When the aircraft reaches the final approach leg, the altitude command is calculated
based on the glide path offset, the glide path angles (γ1 ,γ2 ) and the horizontal distance
that the aircraft is away from the glide path origin. The aircraft therefore follows the
longitudinal landing path shown in Figure 5.4. If at any point the position, airspeed or
attitude of the aircraft is outside a set of safety limits, the landing is aborted and the
safety pilot assumes control of the aircraft. The altitude descends along the glide path
until the aircraft touches down. When a touchdown is registered, the controllers are
disabled and the safety pilot takes control, guiding the aircraft down the runway and
bringing it to a stop.
Figure 5.10 – Landing state machine overview
5.5 Landing Flight Test Results
In addition to all the controller verification flight tests presented in Chapters 3 and 4,
five fully autonomous landings were performed over two days. Three of these made use
of the conventional NSA controller of §3.2.4 and the final two landings were performed
using the direct-lift-augmented NSA controller of §3.2.6. The results of these flight tests
are presented in the sections which follow. Only the plots with direct bearing on the
landing circuit and landing path are shown and evaluated.
The procedure for all flight tests is as follows,
1. The aircraft is placed on the runway and the OBC is initialised from a laptop running
the ground station software.
2. The correct functioning of all sensors and actuators is confirmed, the DGPS is al-
lowed time to lock and the estimator is started.
3. The waypoint locations, altitude and airspeed references, as well as the controller
gains are uploaded to the OBC.
4. The safety pilot manually takes off and positions the aircraft in a predetermined
area in the air.
5. The autopilot is enabled on the ground station, and the safety pilot flips the autopi-
lot switch on the RC transmitter. This enables the autopilot and the aircraft begins
to follow the waypoints at a constant airspeed and altitude.
6. When the aircraft enters the landing approach leg it begins to follow an altitude
reference which will result in a mock landing. This mock landing serves to con-
firm the correct functioning of the landing controllers, and also to confirm that the
heading and path of the aircraft coincide acceptably with the runway.
7. If adjustments to the heading or path are required, the relevant parameters are
updated and another mock landing performed. If the heading and path are accept-
able, an actual landing command is issued.
8. The aircraft descends along the glide path and impacts the runway.
9. At, or shortly after touchdown, the safety pilot retakes control of the aircraft, guid-
ing it down the runway until it stops.
Further details of the tests are contained in the flight test cards shown in Appendix D.3.
5.5.1 Landing Test 1 - 28 November Flight 1
During this flight test the conventional elevator-based NSA controller was used. As this
was the first landing attempt, the goal was merely to achieve a landing regardless of the
accuracy thereof.
The aircraft performed three mock landings with the glide path origin set 5 meters above
the runway. As the aircraft passed the glide path origin, the reference altitude began to
increase at the same slope as during the approach. This caused the aircraft to ascend
back to the circuit reference altitude. The actual landing was performed with the glide
path origin set 1.3 meters below the runway. This is an intentionally conservative set-
point which will result in the aircraft overshooting the touchdown point, but touching
down at a lower climb rate and higher pitch angle than it would otherwise. This is the
result of the glide path reference increasing again as the origin is passed.
Figure 5.11a shows the altitude and airspeed of the aircraft for the duration of the flight
test. The three mock landings can be seen in the altitude plot. It can also be seen
that the airspeed was regulated well throughout the flight, including during the steep
descent phases of the final approach legs. Figure 5.12a shows the altitude of the aircraft
during the landing approach legs of the mock landings where it can be seen that the
paths coincide reasonably well with one another.
The circuit path flown by the aircraft is shown in Figure 5.11b. During each of the mock
landings the aircraft’s path along the runway was observed and adjustments made to the
runway angle and roll angle offset (see §5.2 and §4.4 respectively). It can be seen that
each successive lap of the circuit is flown with the path more aligned with the runway.
This is again illustrated in Figure 5.12b where the flight paths across the runway are
shown. It can be seen that the path of the third mock landing corresponded sufficiently
well with the runway that an actual landing could be performed.
50
300
40
30 250
20
200
10
0 150
0 50 100 150 200 250 300

100
25 50
20 0
15
−50
10
−100
5
0 −150
0 50 100 150 200 250 300 −300 −200 −100 0 100 200
(a) Altitude and airspeed (b) Circuit flight path
Figure 5.11 – Landing Test 1 - Altitude, airspeed and circuit flight path for the entire duration
of the flight
35
10
30
25 5
20
0
15
10 −5
5
−10
0
−200 −150 −100 −50 0 50 −60 −40 −20 0 20 40 60 80 100
(a) Altitude during mock landings (b) Flight paths across runway
Figure 5.12 – Landing Test 1 - Altitude during landing leg of the mock landings, and all flight
paths across the runway
The altitude of the aircraft during the actual landing leg is shown in Figure 5.13a. It can
be seen that the aircraft followed the steeper reference glide path well with a constant
offset. When the aircraft reached a lower altitude, the slope of its flight path is reduced
somewhat. This is likely caused by the ground effect1 which causes an increase in lift
1
An increase in lift and decrease in drag due to the interaction of the wingtip vortices with the ground.
Takes effect when the wing is within one wingspan length of the ground [5].
when the aircraft is close to the ground. The flight path of the aircraft during the actual
landing, and the corresponding roll angle of the aircraft is shown in Figure 5.13b. It can
be seen that the aircraft touched down close to the centre of the runway, but that the
heading of the path is not aligned with the centreline of the runway. The roll angle plot
indicates that the wings of the aircraft were not completely level at the touchdown point.
25
10
5
20
0
−5
15
−10
−200 −150 −100 −50 0 50 100
10
20
5
10
0
0
−10
−5 −20
−200 −150 −100 −50 0 50 0 5 10 15
(a) Altitude during landing leg (b) Flight path and roll angle during landing
leg
Figure 5.13 – Landing Test 1 - Altitude, flight path and roll angle during landing leg
Figure 5.14a shows the climb rate and NSA of the aircraft during the final approach leg.
The touchdown point is clearly visible in the NSA plot as the large spike in acceleration
when the aircraft touches down. The pitch rate, pitch angle and elevator deflection are
shown in Figure 5.14b. It can be seen that at the touchdown point the pitch angle is very
slightly negative. This is one of the consequences of the straight line landing path, since
the aircraft does not flare in order for its rear landing gear to strike the runway first.
In practice this angle is small enough to not have any negative effect on the landing.
However, should this angle be problematic, it can be increased by reducing the landing
airspeed which in turn would cause the aircraft to fly at a higher angle of attack.
Figure 5.15 shows the altitude of the aircraft in the moments before and after touch-
down. It can be seen that the aircraft touches down approximately 17 meters from the
desired point at the glide path origin. This was expected due to the conservative value
of the glide path offset. If the glide path offset is increased, the aircraft will touch down
closer the intended touchdown point.
This flight test confirms the functioning of the landing controllers and state machine,
and resulted in a successful landing. The subsequent flight tests will focus on improving
the accuracy of the landing.
4 20
2
0
0
−20
−2 0 5 10 15
−4 10
0 5 10 15
0
−10
−5
0 5 10 15
−10 5
0
−15
−5
0 5 10 15 0 5 10 15
(a) Climb rate and NSA during landing leg (b) Pitch rate, pitch angle and elevator de-
flection during landing leg
Figure 5.14 – Landing Test 1 - Climb rate and NSA, and pitch rate, pitch angle and elevator
deflection during landing leg
−1
−2
−3
−4
−5
−25 −20 −15 −10 −5 0 5 10 15 20 25
Figure 5.15 – Landing Test 1 - Altitude during touchdown
During this flight test the conventional elevator-based NSA controller was again used.
This test was conducted with the parameters as in Landing Test 1.
Only one mock landing was required since the runway heading parameters had been
established in Landing Test 1. This mock landing was performed with the glide path
origin set 5 meters above the runway. The actual landing was performed with the glide
path origin set 1.5 meters below the runway. This is a slightly more aggressive setpoint
which should result in an improved landing accuracy when compared to Landing Test
1. However, atmospheric conditions during this landing were less favourable and some
wind disturbances prevailed.
test. The mock landing can be seen in the altitude plot. It can also be seen that the
airspeed was regulated well throughout the flight, including during the steep descent
phases of the final approach legs. When compared to the airspeed plot of Landing Test 1
as shown in Figure 5.11a, the impact of wind gusts can be seen. Figure 5.17a shows the
altitude of the aircraft during the landing approach leg of the mock landing.
The circuit path flown by the aircraft is shown in Figure 5.16b. It can be seen that
the flight path corresponds well with the reference path. This is again illustrated in
Figure 5.17b where the flight paths of the mock and actual landing across the runway
are shown. It can the seen that the path of the mock landing corresponded sufficiently
well with the runway that an actual landing could be performed.
50
40 300
30 250
20
200
10
0 150
0 50 100 150
100
25 50
20
0
15
−50
10
5 −100
0
0 50 100 150 −300 −200 −100 0 100 200
of the flight
35
10
30
25 5
20
0
15
10 −5
5
−10
0
−200 −150 −100 −50 0 50 −60 −40 −20 0 20 40 60 80 100
(a) Altitude during mock landings (b) Flight path across runway
offset. However, the effect of the increased wind disturbance is evident when compared
to Landing Test 1 shown in Figure 5.13a. The flattening out of the aircraft’s path caused
by the ground effect is again evident. The flight path of the aircraft during the actual
landing, and the corresponding roll angle of the aircraft is shown in Figure 5.18b. It can
be seen that the aircraft touched down close to the centre of the runway, but that the
heading of the path is not aligned with the centreline of the runway. The roll angle plot
indicates that the wings of the aircraft were not perfectly level at the touchdown point.
This is caused by the presence of the wind disturbances as can be seen in the larger roll
angle fluctuations when compared to those in Landing Test 1 shown in Figure 5.13b.
25
10
5
20
0
15 −5
−10
−200 −150 −100 −50 0 50 100
10
20
5
10
0
0
−10
−5 −20
−200 −150 −100 −50 0 50 0 2 4 6 8 10 12 14 16 18
leg
slightly negative. The effect of the wind disturbance can be clearly seen when comparing
these plots to those of Landing Test 1 shown in Figures 5.14a and 5.14b.
Figure 5.20 shows the altitude of the aircraft in the moments before and after touch-
down. It can be seen that the aircraft touches down approximately 22 meters from the
desired point at the glide path origin. It was expected that this point would be closer to
the desired touchdown point when compared to Landing Test 1, however, the presence
of significant wind disturbances meant that it fared slightly worse.
This flight test confirms that the controllers function sufficiently well in the presence of
wind disturbances in order for a landing to be conducted successfully.
4 20
2
0
0
−20
−2 0 5 10 15
−4 10
0 5 10 15
0
−10
−5
0 5 10 15
−10 5
0
−15
−5
0 5 10 15 0 5 10 15
−1
−2
−3
−4
−5
−40 −30 −20 −10 0 10 20 30 40
During this flight test the conventional elevator-based NSA controller was once again
used. This test was conducted using the parameters of Landing Test 2 and is in essence
a repeat of that test under more favourable wind conditions.
The aircraft performed two mock landings with the glide path origin set 5 meters above
the runway. The actual landing was performed with the glide path origin set 1.5 meters
below the runway. This is the same as during Landing Test 2 and should result in an
improved landing accuracy when compared to Landing Test 1, which did not occur in
Landing Test 2 due to the presence of atmospheric disturbances.
test. The two mock landings can be seen in the altitude plot. It can also be seen that the
phases of the final approach legs. Figure 5.22a shows the altitude of the aircraft during
the landing approach legs of the mock landings where it can be seen that the paths
coincide well with one another.
The circuit path flown by the aircraft is shown in Figure 5.21b. During each of the mock
landings the aircraft’s path along the runway was observed and adjustments made to
the runway angle and roll angle offset. It can be seen that each successive lap of the
circuit is flown with the path more aligned with the runway. This is again illustrated in
Figure 5.22b where the flight paths across the runway are shown. It can the seen that
the path of the second mock landing corresponded sufficiently well with the runway that
an actual landing could be performed.
50
40 300
30 250
20
200
10
0 150
0 50 100 150 200 250
100
25 50
20
0
15
−50
10
5 −100
0
0 50 100 150 200 250 −300 −200 −100 0 100 200
of the flight
35
10
30
25 5
20
0
15
10 −5
5
−10
0
−200 −150 −100 −50 0 50 −60 −40 −20 0 20 40 60 80 100
The altitude of the aircraft during the actual landing leg is shown in Figure 5.23a. It
can be seen that the aircraft followed the steeper reference glide path well with a con-
stant offset. The flattening out of the aircraft’s path caused by the ground effect is again
noticeable. The flight path of the aircraft during the actual landing, and the correspond-
ing roll angle of the aircraft is shown in Figure 5.23b. It can be seen that the aircraft
touched down close to the centre of the runway, and that the heading of the path is well
aligned with the centreline of the runway. The roll angle plot indicates that the wings of
the aircraft are nearly level at the touchdown point.
25
10
20 5
0
−5
15
−10
−200 −150 −100 −50 0 50 100

10
20
5
10
0
0
−10
−5 −20
−200 −150 −100 −50 0 50 0 5 10 15 20
leg
slightly negative. It is further evident that there is significantly less wind disturbance
during this test than during Landing Test 2 as shown in Figures 5.19a and 5.19b.
Figure 5.25 shows the altitude of the aircraft in the moments before and after touchdown
occurs. It can be seen that the aircraft touches down approximately 12.5 meters from
the desired point at the glide path origin. This point is closer to the desired touchdown
point when compared to Landing Test 1. This was accomplished by increasing the glide
path offset from 1.3 meters to 1.5 meters.
This flight test thus confirms that the touchdown point can be controlled by changing
the glide path offset. Subsequent flight tests attempt to achieve an even more accurate
landing by further increasing the glide path offset.
4 20
2
0
0
−20
0 5 10 15
−2
−4 10
0 5 10 15
0
−10
−5
0 5 10 15
−10 5
0
−15
−5
0 5 10 15 0 5 10 15
−1
−2
−3
−4
−5
−25 −20 −15 −10 −5 0 5 10 15 20 25
The direct-lift-augmented NSA controller was used during this flight test. This test was
conducted using the parameters of Landing Test 3 and is intended to allow a direct
comparison between the conventional and the direct-lift-augmented NSA controllers. It
must be noted that the atmospheric conditions during this flight test were less favourable
than during Landing Test 3. The wind conditions as measured by a small weather station
at the airfield are shown in Figure D.2.
The aircraft performed two mock landings with the glide path origin firstly set at 20
meters, and then at 5 meters above the runway. The higher initial mock landing was to
provide an additional safety margin since the direct-lift-augmented NSA controller had
not been tested before during a landing. The actual landing was performed with the glide
path origin again set 1.5 meters below the runway. This is the same as during Landing
Test 3 and a comparison between the conventional and the direct-lift-augmented NSA
controllers can therefore be made.
test. The two mock landings can be seen in the altitude plot. It can also be seen that the
phases of the final approach legs. Figure 5.27a shows the altitude of the aircraft during
the landing approach legs of the mock landings.
The circuit path flown by the aircraft is shown in Figure 5.26b. The adjustments to the
runway angle and roll angle offset were determined during Landing Test 3, and only a
small adjustment to the runway angle was required. This is again illustrated in Figure
5.27b where the flight paths across the runway are shown. It can be seen that the paths
for the mock and actual landings fall very close to one another and are well aligned with
the centreline of the runway.
50
300
40
30 250
20
200
10
0 150
0 50 100 150 200
100
25 50
20
0
15
−50
10
5 −100
0
0 50 100 150 200 −300 −200 −100 0 100 200
of the flight
offset. Unlike the previous landings, the flattening out of the aircraft’s path caused by
the ground effect is less pronounced. This is likely due to the smaller pitch rotation
performed by the aircraft due to the utilisation of the direct-lift-augmented NSA control-
ler. A full comparison of the pitch behavior of the conventional and direct-lift-augmented
NSA controllers is made in §5.5.5. The flight path of the aircraft during the actual land-
ing and the corresponding roll angle is reflected in Figure 5.28b. It can be seen that the
aircraft touched down nearly in the centre of the runway, and that the heading of the
path is well aligned with the centreline of the runway. The roll angle plot indicates that
the wings of the aircraft are nearly level at the touchdown point.
The touchdown point is clearly visible in the NSA plot as the large spike in acceleration as
45
10
40
35
5
30
25
0
20
15
−5
10
5
−10
0
−200 −150 −100 −50 0 50 −60 −40 −20 0 20 40 60 80 100
25
10
20 5
0
−5
15
−10
−200 −150 −100 −50 0 50 100

10
20
5
10
0
0
−10
−5 −20
−200 −150 −100 −50 0 50 0 5 10 15
leg
the aircraft touches down. The pitch rate, pitch angle, and elevator and flaps deflection
are shown in Figure 5.29b. It can be seen that at the touchdown point the pitch angle is
very slightly positive. This is in contrast to the previous Landing Tests where the pitch
angle is slightly negative at touchdown. This is possibly a consequence of using the
direct-lift-augmented NSA controller and is discussed further in §5.5.5.
occurs. It can be seen that the aircraft touches down approximately 5 meters from the
desired point at the glide path origin. This point is closer to the desired touchdown point
than that achieved in Landing Test 3.
This flight test shows that, for the same parameters, the direct-lift-augmented NSA con-
troller appears to achieve a more accurate landing than the conventional NSA controller.
However, the uncertainty about the exact atmospheric conditions during the tests means
4 20
2
0
0
−20
−2 0 5 10 15
−4 10
0 5 10 15
0
−10
−5
0 5 10 15
−10 5
0
−15
−5
0 5 10 15 0 5 10 15
−1
−2
−3
−4
−5
−25 −20 −15 −10 −5 0 5 10 15 20 25
that it cannot be stated with absolute certainty. Such an assertion would require more
landings than the time constraints of this project allow for.
The direct-lift-augmented NSA controller was again used during this flight test. During
this test, the glide path offset is increased in order to achieve an even more accurate
landing. It must be noted that the atmospheric conditions during this flight test were
significantly less favourable than during any of the other Landing Tests (see Figure D.2).
The aircraft performed two mock landings with the glide path origin firstly set at 20
meters, and then at 5 meters above the runway. The actual landing was performed with
the glide path origin set 2 meters below the runway. This is greater than for any of the
previous Landing Tests and should result in the most accurate landing.
test. The two mock landings can be seen in the altitude plot. The airspeed plot shows the
presence of significant wind gusts as evidenced by the large fluctuations of the estimator
velocity when compared to the airspeed measured by the pitot tube sensor. Figure 5.32a
shows the altitude of the aircraft during the landing approach legs of the mock landings.
The circuit path flown by the aircraft is shown in Figure 5.31b. The adjustments to the
runway angle and roll angle offset were determined during Landing Test 4. Figure 5.32b
shows the flight paths across the runway. It can be seen seen that at the touchdown
point the paths of the mock and actual landings fall very close to one another and are
well aligned with the centreline of the runway. However, the path of mock landing 1
diverged from the others after the glide path origin was passed. This was likely caused
by a strong wind gust.
50
40 300
30
250
20
10 200
0 150
0 50 100 150 200 250
100
25 50
20
0
15
−50
10
−100
5
−150
0
0 50 100 150 200 250 −300 −200 −100 0 100 200
of the flight
offset. Similar to Landing Test 4, the flattening out of the aircraft’s path caused by the
ground effect is less pronounced. The flight path of the aircraft during the actual landing,
and the corresponding roll angle of the aircraft is shown in Figure 5.33b. It can be seen
that the aircraft touched down nearly in the centre of the runway, and that the heading
of the path is well aligned with the centreline of the runway. The roll angle plot indicates
that the wings of the aircraft were not level at the touchdown point which resulted in a
sharp roll angle in the opposite direction as the aircraft touched down.
The touchdown point is clearly visible in the NSA plot as the large spike in acceleration as
the aircraft touches down. The pitch rate, pitch angle, and elevator and flaps deflection
are shown in Figure 5.34b. It can be seen that at the touchdown point the pitch angle is
45
10
40
35
5
30
25
0
20
15
−5
10
5
−10
0
−200 −150 −100 −50 0 50 −60 −40 −20 0 20 40 60 80 100
25
15
10
20 5
0
−5
15
−10
−15
−200 −150 −100 −50 0 50 100
10
20
5
10
0
0
−10
−5 −20
−200 −150 −100 −50 0 50 0 5 10 15
leg
very slightly negative. This is in contrast with Landing Test 4 where the pitch angle was
slightly positive. This is likely due to different wind conditions during the glide path.
occurs. It can be seen that the aircraft touches down approximately 3 meters from the
desired point at the glide path origin. This is closer to the desired touchdown point than
that achieved in Flight Test 4, but is ahead of the glide slope origin instead of behind.
This means that the glide path offset was larger than necessary and must be reduced
slightly in order to achieve a more accurate landing.
This flight test confirms that the direct-lift-augmented NSA controller continues to func-
tion even in the presence of significant wind disturbances. This test achieved the greatest
accuracy, and it is clear that with the appropriate glide path offset, further improvements
are possible.
4 20
2
0
0
−20
0 5 10 15
−2
−4 10
0 5 10 15
0
−10
−5
0 5 10 15
−10 5
0
−15
−5
0 5 10 15 0 5 10 15
−1
−2
−3
−4
−5
−25 −20 −15 −10 −5 0 5 10 15 20 25
Whilst further tests would be desirable, the landings which were achieved in both normal
wind conditions as well as in conditions where wind gusts prevail can, for the purposes
of this thesis, be considered a success. Further improvements in accuracy are possible,
however, at some point the random nature of wind disturbances will prevent any further
repeatable landing improvements.
The flight test also show that the direct-lift-augmented NSA controller appears to achieve
a more accurate landing than the conventional NSA controller. The wind measurements
reflected in Figure D.2 also strongly suggest that the direct-lift-augmented NSA control-
ler offers superior performance in the presence of wind disturbances. However, this
cannot be stated with absolute certainty since the respective atmospheric conditions
during the tests cannot be truly quantified and compared. To be confident of such an
assertion would require more landings than the time constraints of this project allow for.
In addition to the landing accuracy achieved, the conventional and direct-lift-augmented

NSA controllers can also be compared by examining the pitch response of the aircraft
during landing. This is of importance since part of the motivation behind employing a
direct-lift type controller is to minimise aircraft high speed pitch rotation during land-
ing. Such pitch changes can greatly influence airspeed control, and can also lead to
dangerous situations such as striking the runway with the nose wheel first. To make this
comparison, the pitch angle, pitch rate and airspeed of the aircraft for Landing Test 3
and 4 are shown in Figures 5.36a and 5.36b. These Landing Tests were selected since
they allow the comparison of the NSA controllers under the most similar conditions.
Landing Test 3 utilised the conventional NSA controller, and Landing Test 4 utilised the
direct-lift-augmented NSA controller. It must be noted that the atmospheric conditions
during Landing Test 4 were less favourable than during Landing Test 3. This is shown in
the windspeed plot of Figure D.2. It can be seen in Figure 5.36a that the conventional
NSA controller exhibited far greater pitch angle changes than the direct-lift-augmented
NSA controller. This is again shown in the upper plot of Figure 5.36b where it can be
seen that the conventional NSA controller showed greater pitch rates. It can be seen in
Figure 5.36b that the airspeed regulation for the conventional and direct-lift NSA con-
trollers exhibited similar variations. These results suggest that the direct-lift augmented
NSA controller provides improved landing performance over the conventional NSA con-
troller.
15 20
10
10
0
5 −10
−20
0 2 4 6 8 10 12
0
22
−5 20
18
16
−10
14
12
−15 10
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) Pitch angle during landing leg (b) Pitch rate and airspeed landing leg
Figure 5.36 – Landing Test 3 and 4 - Pitch angle, pitch rate and airspeed during landing leg
Chapter 6
Conclusion
This thesis presents the design, practical implementation and testing of a control sys-
tem to achieve an accurate autonomous landing of a fixed-wing UAV in the presence of
wind gust atmospheric disturbances. This chapter presents a summary of the work done
and highlights the results achieved. It concludes with a list of contributions made to
the research group at the Centre of Expertise in Autonomous Systems at Stellenbosch
University, as well as recommendations for further research.
6.1 Summary
During the research project the following milestones were attained leading to the suc-
cessful execution of accurate autonomous landings of a fixed-wing UAV.
Aircraft:
• The construction of a UAV to serve as the testbed for the various controllers
developed. This involved modifying and strengthening a model aircraft, in-
corporating an electric motor system as well as assembling and integrating
sensors and avionics. These include,
– IMU providing 3-axis acceleration and rotation measurements

– Magnetometer
– Pressure sensors for calculating airspeed and pressure altitude
– RTK DGPS system for providing accurate position and velocity state meas-
urements
All sensors and actuators of the aircraft were also calibrated.
114
CHAPTER 6. CONCLUSION 115
Modelling and controllers:
• The aircraft model was determined on the basis of the model presented by [2].
This also involved,
– Determining the stability and control derivatives through the use of AVL.
– Conducting a static thrust test of the aircraft’s motor and propeller.
– Finding the moment of inertia through the double pendulum method.
• The dynamics of the aircraft were analysed in order to formulate a control

strategy. This also involved an in-depth study of the longitudinal dynamics
of the aircraft in order to design aircraft controllers which incorporate the
concept of direct-lift control.
• Controllers were designed to regulate the airspeed, altitude, climb rate, nor-
mal specific acceleration and roll angle of the aircraft as well as damp the
Dutch roll mode.
• Two normal specific acceleration controllers were designed. The first being
a conventional elevator-based NSA controller, and the second a direct-lift-
augmented NSA controller which makes use of both the elevator and the flaps
of the aircraft.
• A guidance controller was implemented to allow for the following of waypoints.
Landing:
• A landing procedure and methodology was developed. This includes the circuit
and landing approach paths and the concept of a glide path offset to calibrate
the touchdown point of a landing.
Simulation and flight tests:
• All controllers and the landing procedure were implemented on an OBC.
• All controllers were tested in a HIL simulation environment as well as practic-

ally in a series of flight tests in order to verify the controller design and the
aircraft model.
• The landing procedure was tested in a HIL simulation environment and prac-
tically in a series of flight tests.
• Five fully autonomous landings were performed. Three of these using the
conventional NSA controller, and the final two the direct-lift-augmented NSA
controller.
6.2 Results
The results of the landing flight tests show that the project goal of a landing within five
meters along the runway and three meters across the runway was achieved in both nor-
mal wind conditions as well as in conditions where wind gusts prevailed. The flight tests
show that the direct-lift-augmented NSA controller appears to achieve a more accurate
landing than the conventional NSA controller, especially in the presence of greater wind
disturbances. However, the inability to quantify and directly compare the respective at-
mospheric conditions during the tests means that this cannot be stated with absolute
certainty. Such an assertion would require more landings than the time constraints of
this project allowed for.
In addition to the landing accuracy achieved, the conventional and direct-lift-augmented

NSA controllers were also compared by examining the pitch response of the aircraft
during landing. It was seen that the conventional NSA controller exhibited far greater
pitch angle changes than the direct-lift-augmented NSA controller. This suggests that
the direct-lift augmented NSA controller provides improved landing performance over
the conventional NSA controller.
6.3 Contributions
The work completed for this thesis offers the following contributions to the research
group at the Centre of Expertise in Autonomous Systems at Stellenbosch University,
• The UAV which was constructed performed well, and can serve as a reliable testbed
for future research. Through practical flight tests, it was demonstrated that the
aircraft model is accurate, and that the controllers function well.
• The groundwork has been done for further research into the accurate autonomous
landing of fixed-wing UAV’s which could include aspects such as crosswind land-
ings.
• The concept of direct-lift control has been explored and successfully implemented.
This technique holds promise for future projects where accurate longitudinal con-
trol and the rejection of wind disturbances are required.
• The landing procedure and strategy has been shown to work in practice and offers
scope for further development.
• The controllers which were developed are modular and can be used in other air-
craft.
• The DGPS system has been demonstrated to function well on a fixed-wing aircraft.
6.4 Recommendations for Further Research
Aircraft:
• The use of an electric motor as a propulsion source was found to have great
benefits when compared to the traditional liquid fuel powered engines. The
benefits include simple and reliable operation, as well as improved airspeed
regulation. It is recommended that future projects follow the same approach.
Control system:
• The path following ability of the aircraft was shown to work well, but there is
room for improvement. A conservative approach was followed in the design of
the guidance controller in order to avoid high roll angles. This resulted in a
fairly long settling time. Relaxing this requirement and adding an integrator
term to the guidance controller could result in an improvement. However, care
would need to be taken to avoid integrator wind-up.
Direct-lift control:
– The direct-lift-augmented NSA controller can be improved further. This
could involve revising the control law of the flaps-based portion of the
NSA controller to mitigate the effect of noise on the NSA signal. This
could be accomplished by augmenting a fast integrator on the NSA
signal similar to that of the elevator-based NSA controller of §3.2.4.
Alternatively, direct-lift control can also be attempted by using climb
rate feedback.
– A pure direct-lift controller can be designed which primarily makes use
of the flaps to achieve longitudinal control This could open up interest-
ing new horizons for future UAV’s.
Landing:
• In order to achieve greater landing accuracy, the angle of the glide path can be
increased. This will reduce the effect that a height error has on the distance
between the intended and actual touchdown point. However, for this to be
implemented practically, the landing gear of the aircraft would need to be
modified to absorb the additional landing forces.
Further expansion of operational capability:
• The vision-based system for determining the aircraft’s position, developed by

[7], could be revisited in order to eliminate the need for a communication link
between the aircraft and the ground as is the case with the DGPS system used
in this project.
• The following controllers can be considered:

– A runway controller to steer the aircraft after touchdown.

– Controllers to enable cross-wind landing: This could consist of a lateral
specific acceleration controller which will aid in maintaining the flight path
of the aircraft, a de-crab controller which would align the centreline of the
aircraft with that of the runway just before touchdown, and a controller to
level out the wings just before touchdown.
• Implementation of an arrestor cable mechanism to capture the aircraft after

touchdown which would allow for landing on a short runway.
• Landing on a moving platform such as the deck of a ship: This would require
estimation of the ship’s motion. Research into this field has been done by
[21; 22], and could serve as a basis for further expansion.
Appendix A
Additional Mathematical Principles

[1]
A.1 Transforming the Derivative of a Vector in a

Rotating Reference Frame
The following equation can be used to transform the derivative of a vector from one axis
system to another axis system where the axis systems rotate relative to one another [17],
d d
R = R + ω BA × R (A.1.1)
dt A dt B
where ω BA is the angular velocity of the rotation of B relative to A.
A.2 Cross Product Transformation Matrix
The cross product of two vectors coordinated in the same axis system can be written as,
JA × KA = SJA KA (A.2.1)
where the matrix SJA is given by,

 
0 −ZA YA
SJA =  ZA 0 −XA  (A.2.2)
 
−YA XA 0
where XA , YA and ZA are the components of J coordinated in axis system A.
119
APPENDIX A. ADDITIONAL MATHEMATICAL PRINCIPLES [1] 120
A.3 Direction Cosine Matrix (DCM)
The direction cosine matrix can be used to transform a vector from one axis system into
another as follows,
h i
VB = DCMBA VA (A.3.1)
where the rows of the DCMBA matrix consist of the unit vectors of axis system B , co-
ordinated in axis system A.
Since it consists of three orthogonal unit vectors, the DCM it is an orthogonal matrix
which implies that its inverse is given by its transpose. Therefore, the reverse transform
from axis system B to A is given by,
h i-1 h iT
VA = DCMBA VB = DCMBA VB (A.3.2)
The following conversions are used in this thesis:
Wind to Body Axes:

 
cos α cos β − cos α sin β − sin α
DCMBW =  sin β cos β 0  (A.3.3)
 
sin α cos β − sin α sin β cos α
Body to Wind Axes:

h iT
DCMW B = DCMBW
 
cos α cos β sin β sin α cos β
= − cos α sin β cos β − sin α sin β  (A.3.4)
 
− sin α 0 cos α
Inertial to Wind Axes:

The DCM can be written in terms of the Euler 3-2-1 sequence used in this project,
 
cos ψ cos θ sin ψ cos θ − sin θ
DCMW I = cos ψ sin θ sin φ − sin ψ cos φ sin ψ sin θ sin φ + cos ψ cos φ cos θ sin φ 
 
cos ψ sin θ cos φ + sin ψ sin φ sin ψ sin θ cos φ − cos ψ sin φ cos θ cos φ (321)
(A.3.5)
APPENDIX A. ADDITIONAL MATHEMATICAL PRINCIPLES [1] 121
A.4 Moment of Inertia
For an object with a fixed mass distribution, the moment of inertia referenced to its
centre of mass and coordinated in its body axis system is given by,
 
Ix −Ixy −Ixz
IB = −Ixy Iy −Iyz  (A.4.1)
 
−Ixz −Iyz Iz
For most conventional aircraft this simplifies to,

 
Ix 0 0
IB =  0 Iy 0  (A.4.2)
 
0 0 Iz
A.5 Small Angle Approximations
The standard small angle approximations are given by,
cos α ≈ 1 (A.5.1)
sin α ≈ α (A.5.2)
αβ ≈ 0 (A.5.3)
Appendix B
UAV System Description
This appendix describes the UAV system which was constructed/utilised to serve as the
testbed for this project. The system consists of the airframe, propulsion source, avionics,
ground station and a HIL simulation environment.
B.1 Airframe
The airframe is a standard high-wing type model aircraft with a wingspan of just un-
der 2 m as shown in Figure D.1. It was modified to accommodate the avionics and
strengthened to withstand higher than normal landing impact forces.
Figure B.1 – Aircraft photo
B.1.1 Propulsion Source
The aircraft is powered by a Hyperion ZS4025-10 brushless DC outrunner motor equipped

with a standard 14 × 7 electric motor propeller. The motor is driven by a Hyperion Ti-
tan 90 A electronic speed controller (ESC), which is in turn powered by a 18.5 V 5-cell
122
APPENDIX B. UAV SYSTEM DESCRIPTION 123
5000 mAh 20C lithium-polymer (LiPo) battery. This allows for a safe flight duration of 5
minutes .
B.2 Avionics
The avionics used in this project have been developed within the SU Research Group and
used in previous research projects [8; 10]. Figure B.2 depicts the avionics diagrammat-
ically.
Figure B.2 – Avionics overview
The avionics consist of the following,
OBC
The OBC is based around two dsPIC30F6014 microcontrollers:
• PIC A - Communicates with the ground station laptop via a MaxStream RF

module, this involves sending telemetry and receiving commands; samples the
sensors via the CAN bus; monitors the voltage level of the avionics and backup
batteries; receives safety pilot commands from, and sends actuator commands
to the servo board; executes the controller algorithms; logs sensor, control-
ler, actuator and other flight data on the SD card; runs an extended Kalman
filter (EKF) which provides kinematic state estimation by using sensor inform-
ation from the IMU, GPS and magnetometer to estimate the aircraft’s position,
velocity and attitude [6].
• PIC B - Performs the calculations related to the RTK DGPS and sends the result
to PIC A.
IMU
The IMU is based around the Analog Devices ADIS16350AMLZ sensor. It measures
the specific acceleration and rotation rates of the aircraft in the body axis system,
and sends these values to the OBC via the CAN bus.
GPS
The GPS module used is a NovAtel OEMV-1G. This module is used in conjunction
with another such module on the ground in order a create an RTK DGPS system
with an accuracy on the order of a few centimeters.
Pressure and magnetometer board

The pressure and magnetometer board contains one static and one differential
pressure sensor which are connected to a pitot tube mounted in the aircraft’s wing
via a series of tubes. These sensors allow the static and dynamic pressure to be
determined which allows the calculation of the airspeed and pressure altitude of
the aircraft. The board also contains a three-axis magnetometer that measures the
earth’s magnetic field vector in the aircraft’s body axes. This measurement aids in
determining the aircraft’s attitude.
Receiver
The receiver used is a Spektrum AR 9100 which has a dual redundant power sup-
ply and four satellite receivers mounted in different orientations throughout the
aircraft. This ensures a reliable link with the safety pilot remote.
Servo board
The servo board acts as the intermediary between the OBC, safety pilot RC receiver
and the actuators of the aircraft. It samples the safety pilot’s commands and sends
them to the OBC via the CAN bus. It also sends the servo commands to the air-
craft’s actuators. These will either be the safety pilot commands or the autopilot
commands received from the OBC. The autopilot can be armed from a switch on
the safety pilot remote. The servos are powered by the servo board. Both the servo
board and the receiver have a backup power supply to ensure that the safety pilot
can control the aircraft even in the event of a complete failure of the rest of the
avionics.
Batteries
The avionics is powered by a 11.1 V 3-cell 1800 mAh 35C LiPo battery. Backup power
is provided by a 5 V 700 mAh nickel-cadmium (NiCad) battery which can power the
servo board and the receiver if power from the main battery is lost.
B.3 Ground Station Software
The ground station software used in this project is based on the standard ground station
software developed within the SU Research Group. It runs on a laptop, and communic-
ates with the aircraft via a MaxStream RF module. This RF module is contained in an
enclosure and connects to the laptop via a serial connection.
The functions of the ground station are split into several tabs described briefly below,
Main page
Side panels on the main page provide the following: start and reset the OBC; start
and stop data logging on the SD card; initialise and disable the estimator; enable
and disable the autopilot; toggle the avionics to HIL simulation mode; display the
command history and any errors; display the status of the OBC and a summary of
the sensor data.
Controller tab
This tab allows the controllers to be enabled and disabled, and the controller gains
and references to be set and updated (see Figure B.3a).
Navigation tab
This tab allows the waypoints, runway heading angle and origin of the glide path to
be set. The waypoints are entered in the runway axis system. The ground station
then converts these to the inertial axis system before the values are sent to the
OBC. (see Figure B.3b).
Estimator tab
This tab allows the kinematic state estimator used on this aircraft to be initialised
and configured. The current state estimates are also displayed.
Sensors tab
This tab shows the current measurements of all the aircraft’s sensors, and allows
the aircraft’s sensors to be zeroed.
Actuator tab
This tab shows the current safety pilot commands as well as the commands sent to
the servo board from the OBC.
Actuator Setup tab

This tab is used during the calibration of the aircraft’s actuators and allows the
range of motion and the neutral point of all servos to be set.
The ground station also stores all telemetry data on the hard drive of the laptop.
(a) Controllers tab
(b) Waypoints tab
Figure B.3 – Ground station graphical user interface

B.4 Hardware-in-the-Loop Simulation Environment
The HIL simulation environment allows the avionics and controllers to be tested as if the
aircraft is actually flying. This is accomplished by running a non-linear aircraft simu-
lator in Matlab Simulink on a PC [6]. This simulator generates simulated sensor signals
and sends these to the OBC. The OBC then uses these sensor signals in its controller
algorithms and generates actuator signals. These actuator signals are fed back into the
simulator and the aircraft model responds accordingly. This allows the controllers and
avionics to be fully tested before practical flight tests are attempted. Sensor noise and
wind disturbances can also be set in the simulation which allows the robustness and
accuracy of the controllers to be tested. The components of the HIL simulation environ-
ment are depicted in Figure B.4.
The HIL simulation environment also contains a graphical display of the aircraft and its
environment as shown in Figure B.5. This allows the flight to be visualised which aids
greatly in controller design and evaluation.
Figure B.4 – HIL overview

Figure B.5 – HIL graphical display
B.5 Actuator and Controller Limits
To ensure safe and controlled operation of the aircraft, limits are placed on the deflection
angles of the actuators, as well as on the controller commands as shown in Table B.1.
Parameter Limit
2 ref
NSA controller −15 m/s ≥ CW ≤ −5 m/s2
Climb rate controller −3.5 m/s ≥ Vhref ≤ 3.5 m/s
Airspeed 0 m/s ≥ V̄ ref ≤ 30 m/s
Roll angle controller −40 ◦ /s ≥ P ref ≤ 40 ◦ /s
Roll angle controller −30 ◦ ≥ φref ≤ 30 ◦
Throttle 0 N ≥ TC ≤ 33 N
Elevator −7.5 ◦ ≥ δE ≤ 7.5 ◦
Flaps −5 ◦ ≥ δF ≤ 5 ◦
Ailerons −7.5 ◦ ≥ δA ≤ 7.5 ◦
Rudder −7.5 ◦ ≥ δR ≤ 7.5 ◦
Table B.1 – Actuator and controller limits
B.6 Aircraft Model
B.6.1 Standard Flight Conditions
The standard flight conditions are defined in this section. These values are used in the
design of the controllers.
Air Density The density of air is considered a constant due to the small altitude ranges
involved in this project. The airfield where practical flight tests are conducted is nearly
at sea level. Therefore,
ρ = 1.225 kg/m3 (B.6.1)
Gravitational Acceleration Through similar arguments as for air density, the gravit-
ational acceleration at sea level is used,
g = 9.81 m/s2 (B.6.2)
Trim Airspeed The aircraft’s trim airspeed has been selected as,
V̄trim = 18 m/s (B.6.3)
B.6.2 Aircraft Parameters
This section lists the physical parameters of the aircraft.
Mass
The total flight mass of the aircraft is,
m = 6.35 kg (B.6.4)
Moment of Inertia
The moment of inertia was determined by means of the double pendulum method which
is summarised by [4]. It involves suspending the aircraft by two equally long strings such
that the strings are parallel to the moment of inertia axis of concern. The aircraft is then
rotated slightly about this axis and released. It will begin to oscillate and this period is
timed. The moment of inertia is then calculated as,
mgd2 2
I= T (B.6.5)
4π 2 l
where m is the mass of the aircraft, d the distance between each string and the axis of
concern, l is the length of the string and T is the period of oscillation. Thus,
 
0.722 0 0
IB =  0 0.514 0  kgm2 (B.6.6)
 
0 0 0.925
Engine Parameters
The maximum static thrust of the aircraft’s motor was determined through a static thrust
test as,
Tmax = 33 N (B.6.7)
The engine can be approximated by a first order lag model with a time constant of,
τT = 0.25 s (B.6.8)
B.6.3 Aerodynamic Model
The aerodynamic model of the aircraft is given by its stability and control derivatives.
These were obtained by modelling the aircraft in the vortex lattice program AVL as
shown in Figure B.6. During this process, the wing’s span (b), mean aerodynamic chord
(c̄) and area (S ) are used as non-dimensionalising coefficients. These parameters are
given in Table B.2.
Figure B.6 – Aircraft geometry in AVL
Parameter Value
Wing span (b) 1.918 m
Mean aerodynamic chord c̄ 0.37 m
Wing Reference Area (S ) 0.677 m2
Table B.2 – Wing geometric data
The longitudinal and lateral stability derivatives are presented in Tables B.3 and B.4
respectively, and the control derivatives in Table B.5.
Angle of Attack (α) Value Pitch Rate (Q) Value

CLα 4.808411 CLQ 7.812170
Cm α -0.664939 Cm Q -7.438796
Table B.3 – Longitudinal stability derivatives
Angle of Sideslip (β ) Value Roll Rate (P ) Value Yaw Rate (R) Value
C yβ -0.167475 C yP 0.137007 C yR 0.113738
Cl β -0.062813 ClP -0.429110 Cl R 0.141714
Cnβ 0.058436 CnP -0.030226 CnR -0.063085
Table B.4 – Lateral stability derivatives
Elevator (δE ) Value Flaps (δF ) Value

CLδE 0.456085 CLδF 1.078236
CmδE -0.957351 CmδF 0.172092
Aileron (δA ) Value Rudder (δR ) Value
C yδ A 0.003319 C yδ R 0.093545
ClδA -0.264734 C l δR 0.002171
CnδA 0.008785 CnδR -0.041144
Table B.5 – Control derivatives
B.6.4 Dimensional Stability and Control Derivative Notation
The dimensional stability and control derivative notation used in this thesis is defined as
follows,
∂CA
Ax = qSl n (B.6.9)
∂B 0
with,
B 0 = nB (B.6.10)
where the length l is given by c̄ for pitch moment derivatives, b for roll and yaw moment
derivatives and unity for force derivatives. The normalising coefficient n for incident and
c̄
control deflection angles is unity, for the pitch rate it is 2V̄
and for the roll and yaw angles
b
it is 2V̄
[1].
The dimensional stability and control derivatives for the aerodynamic forces and mo-
ments is presented in Tables B.6 and B.7 respectively.
Due to Lift Forces Sideslip Forces

Angle of attack (α) L̄α = qSCLα
Angle of sideslip (β ) Ȳβ = qSCyβ
Roll Rate (P ) ȲP = qS 2V̄b a CyP
Pitch Rate (Q) L̄Q = qS 2V̄c̄ a CLQ
Yaw Rate (R) ȲR = qS 2V̄b a CyR
Elevator Deflection (δE ) L̄δE = qSCLδE
Flaps Deflection (δF ) L̄δF = qSCLδF
Aileron Deflection (δA ) ȲδA = qSCyδA
Rudder Deflection (δR ) ȲδR = qSCyδR
Table B.6 – Dimensional stability and control derivatives (forces) [1]
Due to Roll moments Pitch moments Yaw moments

Angle of attack (α) Mα = qSc̄Cmα
Angle of sideslip (β ) Lβ = qSbClβ Nβ = qSbCnβ
Roll Rate (P ) LP = qSb 2V̄b a ClP NP = qSb 2V̄b a CnP
Pitch Rate (Q) MQ = qSc̄ 2V̄c̄ a CmQ
Yaw Rate (R) LR = qSb 2V̄b a ClR NR = qSb 2V̄b a CnR
Elevator Deflection (δE ) MδE = qSc̄CmδE
Flaps Deflection (δF ) MδE = qSc̄CmδE
Aileron Deflection (δA ) LδA = qSbClδA NδA = qSbCnδA
Rudder Deflection (δR ) LδR = qSbClδR NδR = qSbCnδR
Table B.7 – Dimensional stability and control derivatives (moments) [1]
Appendix C
Detailed Derivations
C.1 Normal Dynamics
This section presents the detailed derivation of the transfer function matrix of the normal
dynamics. The simplified normal dynamics of Equations 3.2.4 and 3.2.5 are restated for
convenience,
" # " #" # " #" #
α̇ − m1V̄ L̄α 1 α − m1V̄ L̄δE − m1V̄ L̄δF δE
= 1 1
+ 1 1
(C.1.1)
Q̇ Iy
Mα Iy
MQ Q Iy
Mδ E Iy
Mδ F δF
" # " #" # " #" #
= + (C.1.2)
Q 0 1 Q 0 0 δF
The normal dynamics can thus be written as,

# "
δE
ẋ = Ax + B (C.1.3)
δF
" # " #
CW δE
= Cx + D (C.1.4)
Q δF
where,
" #
α
x= (C.1.5)
Q
" # " #
− m1V̄ L̄α 1 − m1V̄ L̄δE − m1V̄ L̄δF
A= 1 1
B= 1 1
(C.1.6)
Iy
Mα Iy
MQ Iy
MδE Iy
Mδ F
" # " #
− m1 L̄α − m1 L̄Q − m1 L̄δE − m1 L̄δF
C= D= (C.1.7)
0 1 0 0
133
APPENDIX C. DETAILED DERIVATIONS

C.1.1 Characteristic equation for the poles
The poles of the system are the roots of the characteristic equation,
p(s) = det(sI − A) (C.1.8)
thus,
" #!
s + m1V̄ L̄α −1
p(s) = det (C.1.9)
− I1y Mα s − I1y MQ

=s + − s− +
mV̄ Iy mV̄ Iy Iy
C.1.2 Characteristic equation for the zeros
The zeros of the system are the roots of the equation,
z(s) = Cadj(sI − A)B + D det(sI − A) (C.1.10)
134

solving the first part of the equation,
" # 1
" #
−m 1
L̄α − m 1
L̄Q s − Iy MQ 1 − m1V̄ L̄δE − m1V̄ L̄δF
Cadj(sI − A)B = 1
 1 1 (C.1.11)
0 1 Iy
Mα s + m1V̄ L̄α Iy
MδE Iy
MδF
 "
1 1
L̄α I1 MQ − m 1
L̄Q I1 Mα − m 1 1 1
L̄Q m1V̄ L̄α
#
−m L̄α s + m
y y
L̄α − m L̄Q s − m − m1V̄ L̄δE − m1V̄ L̄δF
= 
1
 1 1
Iy
Mα s + m1V̄ L̄α Iy
MδE Iy
MδF
  
M L̄ L̄ L̄ L̄δ L̄δ
− L̄α s + L̄mα I Q − mQ M α
− L̄mα − mQ s − mQ m L̄α
V̄  − mV̄
E
− mV̄F
= m Mα
y Iy
L̄α M δE M δF

Iy
s+ m V̄ Iy Iy
 L̄ Mδ L̄δ Mδ

M L̄ L̄ L̄ M L̄ L̄ L̄
− mV̄E (− L̄mα s + L̄mα I Q − mQ M
δ
Iy
α
) + I E (− L̄mα − mQ s − mQ m L̄α
V̄
) − mV̄F (− L̄mα s + L̄mα I Q − mQ M α
Iy
) + I F (− L̄mα − mQ s − mQ m L̄α
)
V̄ 
y y y y
= 
L̄δ Mδ L̄δ Mδ
− mV̄E M Iy
α
+ I E (s + m L̄α
V̄
) − mV̄F M Iy
α
+ I F (s + m L̄α
V̄
)
y y
 L̄ L̄ L̄ M M M L̄ L̄ L̄ Mδ Mδ L̄ Mδ

δE L̄α M L̄ L̄ L̄ M L̄ L̄Q L̄α
s − mV̄E L̄mα I Q + mV̄E mQ M
δ δ
− I E L̄mα − I E mQ s − I E mQ m
δ δ δ L̄α δF L̄α
s − mV̄F L̄mα I Q + mV̄F mQ M
δ δ
α α
− I F L̄mα − I F mQ s − I F
=  mV̄ m y
L̄δ
Iy
Mδ
y
Mδ
y y V̄ mV̄ m y
L̄δ
Iy
Mδ
y
Mδ
y y m mV̄ 
− mV̄E M Iy
α
+ I E s+ I E m L̄α
V̄
− mV̄F M I
α
+ I F s+ I F m L̄α
V̄
y y y y y
 
L̄δ Mδ Mδ Mδ L̄δ Mδ Mδ Mδ
L̄α E L̄Q L̄α MQ L̄Q Mα E L̄α E L̄Q L̄α L̄α F L̄Q L̄α MQ L̄Q Mα E L̄Q L̄α
 m
E
mV̄
− I L̄ )s + (− mV̄ I + mV̄ I − I L̄ − I L̄ mV̄ m
F
mV̄
− I L̄ )s + (− mV̄ I + mV̄ I − I F L̄L̄α − Iy L̄δ mV̄
y δE y y y δE y δE y δF y y y δF
= F

L̄δ Mδ Mδ L̄δ M Mδ Mδ

− mV̄E M I
α
+ I
E
s + I
E L̄α
mV̄
− F
mV̄ I
α
+ I
F
s + I
F L̄α
mV̄
y y y y y y
now solve for the second part of the equation,

" "# #!
1
−m L̄δE 1
−m L̄δF s + m1V̄ L̄α −1
D det(sI − A) = det (C.1.12)
0 0 − I1 Mα s − I1 MQ
y y
" #
1 1
−m L̄δE − m L̄δF 2 L̄α MQ L̄α MQ Mα
= s + − s− +
0 0 mV̄ Iy mV̄ Iy Iy
 i
L̄δ
h M
M
i L̄ h MQ

L̄α L̄α L̄α MQ
+M
δF L̄α Mα
m
E
−s2 − m V̄
− IQ s + m V̄ I
Q
I
α
m
−s2 − m V̄
− Iy
s+ mV̄ Iy
+ Iy
=  y y y 
0 0
135

Combining the result of the two parts,
z(s) = Cadj(sI − A)B + Ddet(sI − A) (C.1.13)

 " !   " ! 
L̄δE L̄α MδE L̄Q L̄α MQ L̄Q Mα MδE L̄α L̄δF L̄α MδF L̄Q L̄α MQ L̄Q Mα MδF L̄α
 − s+ − + −   − s+ − + − 
 m mV̄ Iy L̄δE mV̄ Iy mV̄ Iy Iy L̄δE   m mV̄ Iy L̄δF mV̄ Iy mV̄ Iy Iy L̄δF 
 ! #   ! #  
 M δ L̄Q L̄α L̄ α M Q L̄α M Q M α
  MδF L̄Q L̄α L̄α M Q L̄α M Q M α

− E
− s2 − − s+ + − 2
−s − − s+ +
   
 
 Iy L̄δE mV̄ mV̄ Iy mV̄ Iy Iy Iy L̄δF mV̄ mV̄ Iy mV̄ Iy Iy 
=     

 
     
 − L̄δE Mα + MδE s + MδE L̄α  L̄δ M Mδ Mδ
 
L̄
   
 − F α + F
s + F α  
mV̄ Iy Iy Iy mV̄  mV̄ Iy Iy Iy mV̄ 
   
     
 ! !
L̄δ L̄Q MQ Mδ Mδ L̄Q
L̄δ

L̄Q

MQ Mδ

Mδ

L̄Q

L̄α Mα L̄α Mα
 m
E
−s2 + Iy L̄Q
− L̄δ
E
s+ Iy L̄α
− L̄δ
E
1+ mV̄ m
F
−s2 + Iy L̄Q
− L̄δ
F
s+ Iy L̄α
− L̄δ
F
1+ mV̄

E
!E F
!F
 
= 
L̄δ Mδ Mδ L̄δ Mδ Mδ
 
E Mα E E L̄α F Mα F L̄α
− + s+ − + F
s+
 
mV̄ Iy Iy Iy mV̄ mV̄ Iy Iy Iy mV̄
 ! !
L̄δ L̄Q MQ Mδ L̄α Mα Mδ L̄δ L̄Q MQ Mδ L̄α Mα Mδ
 m
E
−s2 + Iy L̄Q
− L̄δ
E
s+ Iy L̄α
− L̄δ
E
m
F
−s2 + Iy L̄Q
− L̄δ
F
s+ Iy L̄α
− L̄δ
F

 E E F F 
= ! ! 
Mδ L̄δ Mδ L̄δ
 
E L̄α Mα
E F L̄α MαF
s+ − s+ −
 
Iy mV̄ mV̄ MδE Iy mV̄ mV̄ MδF
where it is again assumed that,
LQ
1 (C.1.14)
mV̄
If only the location of the zeros is desired the characteristic equation for the zeros becomes,
 
L̄Q MQ MδE L̄α Mα M δE L̄Q MQ M δF L̄α Mα MδF
s2 − Iy L̄Q
− L̄δE
s− Iy L̄α
− L̄δE
s2 − Iy L̄Q
− L̄δF
s− Iy L̄α
− L̄δF
z(s) =  L̄δE Mα L̄δF Mα
 (C.1.15)
L̄α L̄α
s+ mV̄
− mV̄ MδE
s+ mV̄
− mV̄ MδF
136

The transfer function matrix can be written as,
1
G(s) = N(s) (C.1.16)
∆(s)
where N(s) is a polynomial matrix whose elements consist of the response transfer function numerators, and where ∆(s) is the
characteristic polynomial common to all transfer functions. For the NSA dynamics as shown in Equations 3.2.4 and 3.2.5 four transfer
functions exist,
 
CW CW
NδE (s) NδF (s)
1 
G(s) =   (C.1.17)
∆(s)  Q Q

NδE (s) NδF (s)
where NBA (s) is the response transfer function numerator of state variable A to control input B . From the preceding derivations,

∆(s) = s + − s− + (C.1.18)
mV̄ Iy mV̄ Iy Iy
and,
 h i h i
L̄δE L̄Q MQ M MδE L̄ L̄Q MQ M M δF
− m s2 − Iy L̄
− L̄δδE s − L̄Iyα Mα
− − mδF 2
s − Iy
− L̄δδF s − L̄Iyα Mα
−
N(s) =  h Q E i L̄α L̄δE L̄
h Q F i L̄α L̄δF  (C.1.19)
MδE L̄α L̄ MδF L̄
Iy
s + mV̄
− mδV̄E MMδα Iy
s + L̄α
mV̄
− mδV̄F MMδα
E F
137
APPENDIX C. DETAILED DERIVATIONS 138
C.2 NSA Controller Design - Elevator Actuator
This section presents the detailed derivation of the elevator-based NSA controller. The
simplified NSA dynamic equations form §3.2.3.1 are restated below,
" # " #" # " #
α̇ − m1V̄ L̄α 1 α 0
= 1 1
+ 1 δE (C.2.1)
Q̇ Iy
Mα Iy
MQ Q Iy
Mδ E
" # " #" # " #
CW − m1 L̄α 0 α 0
= + δE (C.2.2)
Q 0 1 Q 0
ẋ = Ax + BδE (C.2.3)
" #
CW
= Cx + DδE (C.2.4)
Q
where,
" #
α
x= (C.2.5)
Q
" # " #
− m1V̄ L̄α 1 0
A= 1 1
B= 1 (C.2.6)
Iy
Mα Iy
MQ Iy
Mδ E
" # " #
− m1 L̄α 0 0
C= D= (C.2.7)
0 1 0
To remove steady state errors on NSA, the state vector is augmented with an integrator
xI which obeys the differential equation,
ref
ẋI = CW − CW (C.2.8)
ref
ẋI = C0 x + D0 δE − CW (C.2.9)
ref
where CW is the reference NSA input, and C0 and D0 are the first rows of C and D
respectively. The augmented NSA dynamics are now given by,
" # " #" # " # " #

ẋI 0 C0 xI D0 1 ref
= + δE − CW (C.2.10)
ẋ 0 A x B 0
" # " #
CW h i x
I
= 0 C + DδE (C.2.11)
Q x
An elevator control law is defined which uses feedback from the integrator and pitch
rate states as well as feed-forward from the reference input. Thus,
 
h i xI
ref
δE = − KIδE 0 KQδE  α  + N̄CδE CW (C.2.12)
 
The control law is now substituted into the dynamics,

   1          
ẋI 0 −m L̄α 0 xI 0 h i xI 0 1
  0 − 1 L̄ ref   ref
 α̇  =  mV̄ α
1   α  +  0  − KIδ
    
0 −KQδ

 α

 +  0  N̄C CW
 
− 0 CW (C.2.13)
E E
1 1 1 1
Q̇ 0 Iy
Mα Iy
MQ Q Iy
MδE Q Iy
M δE 0
   1       
ẋI 0 −m L̄α 0 xI 0 0 0 xI 1
  0 − 1 L̄  ref
 α̇  =  1  α +  0 0 0 α −  0  CW
     
mV̄ α
1 1 1 1 1
Q̇ 0 Iy
Mα Iy
MQ Q − I MδE KIδ 0 − I MδE KQδ Q − I MδE N̄C
y E y E y
   1    
ẋI 0 −m L̄α 0 xI 1
− m1V̄ L̄α  ref
 α̇  =  0 1 α −  0  CW
     
Q̇ − I1 MδE KIδ I
1
M α I
1
M Q − I
1
M δ E
KQδ Q − I
1
M δ E
N̄C
y E y y y E y
The closed loop system is therefore given by,

      
ẋI 0 − m1 L̄α 0 xI 1
 ref
 α̇  =  0 − m1V̄ L̄α 1 α −  0  CW
     
Q̇ − I1y MδE KIδE 1

Iy
Mα 1
Iy
MQ − I1y MδE KQδE Q − I1y MδE N̄CδE
(C.2.14)
 
" # " # xI
CW 0 − m1 L̄α 0  
= α (C.2.15)
Q 0 0 1
Q
Calculating the closed loop characteristic equation,
p(s) = det(sI − A) (C.2.16)

 
1
s m L̄α 0
 
= det 
 0 s + m1V̄ L̄α −1 

1
Iy MδE KIδE − I1y Mα
s − ( I1y MQ − I1y MδE KQδE )

1 1 1
= s3 − − L̄α + MQ − MδE KQδE s2
mV̄ Iy Iy

1 1 1 1
+ (− L̄α )( MQ − MδE KQδE ) − (1)( Mα ) ) s
mV̄ Iy Iy Iy
1 1
− (− L̄α )(1)(− MδE KIδE )
m Iy

3 L̄α M Q Mδ E 2 L̄α MQ L̄α MδE Mα L̄α MδE
=s + − + KQδE s + − + KQδE − s− KIδE
mV̄ Iy Iy mV̄ Iy mV̄ Iy Iy mIy
To place the closed loop poles, a desired characteristic equation for the NSA dynamics
is defined as,
αc (s) = s3 + α2 s2 + α1 s + α0 (C.2.17)
When equating the coefficients of the actual and desired characteristic equations the
following results are obtained,
L̄α MQ Mδ E
α2 = − + KQδE (C.2.18)
mV̄ Iy Iy
L̄α MQ L̄α MδE Mα
α1 = − + KQδE − (C.2.19)
mV̄ Iy mV̄ Iy Iy
L̄α MδE
α0 = − KIδE (C.2.20)
mIy
The desired characteristic equation can be rewritten as,
αc (s) = (s2 + 2ζωn s + ωn2 )(s + a) (C.2.21)
where the complex pole pair corresponds to the short period mode and the single real
pole to the closed-loop integrator. Expanding the equation gives,
αc (s) = s3 + (2ζωn + a)s2 + (2ζωn a + ωn2 )s + ωn2 a (C.2.22)
If the damping of the short period mode (ζ ) as well as the integrator pole location are
selected, then the resulting natural frequency can be solved for. This is accomplished
by solving for the pitch rate feedback gain KQ in the first and second order terms of the
characteristic equation,
Iy L̄α MQ
K Q δE = (2ζωn + a − + ) (C.2.23)
Mδ E mV̄ Iy
and,
mV̄ Iy L̄α MQ Mα
K Q δE = (2ζωn a + ωn2 + + ) (C.2.24)
L̄α MδE mV̄ Iy Iy
Setting these equal to one another yields an equation for the natural frequency,
Iy L̄α MQ mV̄ Iy 2 L̄α MQ Mα

(2ζωn + a − + )= (2ζωn a + ωn + + ) (C.2.25)
MδE mV̄ Iy L̄α MδE mV̄ Iy Iy
mV̄ Iy L̄α L̄α L̄α L̄α MQ L̄α mV̄ Iy 2 L̄α MQ Mα
(2ζωn +a − + )= (2ζωn a + ωn + + )
L̄α MδE mV̄ mV̄ mV̄ mV̄ Iy mV̄ L̄α MδE mV̄ Iy Iy
L̄α L̄α L̄α L̄α MQ L̄α 2 L̄α MQ Mα
2ζωn +a − + = 2ζωn a + ωn + +
mV̄ mV̄ mV̄ mV̄ Iy mV̄ mV̄ Iy Iy
2 L̄α MQ Mα L̄α L̄α L̄α L̄α MQ L̄α
0 = 2ζωn a + ωn + + − 2ζωn −a + −
mV̄ Iy Iy mV̄ mV̄ mV̄ mV̄ Iy mV̄
2
2 L̄α M α L̄α L̄α
0 = ωn + ωn 2ζa − 2ζ + + −a
mV̄ Iy mV̄ mV̄
Solving for positive values of ωn the closed loop natural frequency of the short period
mode is given by,
s
Mα L̄α
ωn = ζη + (ζη)2 − − η (C.2.26)
Iy mV̄
where,
L̄α
η= −a (C.2.27)
mV̄
The natural frequency of the short period mode (ωn ), along with the previously selected
damping of the short period mode (ζ ) and the location of the closed-loop integrator pole
(a) can now be used in conjunction with either of the pitch rate equations to determine
the pitch rate feedback gain KQδE . The integrator feedback gain can be solved by equat-
ing the coefficients of the final terms of actual and desired characteristic equations and
is given by,
ωn2 mIy
KIδE = −a (C.2.28)
L̄α MδE
The integrator’s response can be removed from the reference input by placing a feed-
forward zero as follows,
KIδE
N̄CδE = − (C.2.29)
zf
where zf is the location of the desired zero.

C.3 NSA Controller Design - Flaps actuator
This section presents the detailed derivation of the flaps-based NSA controller. The
simplified NSA dynamic equations from §3.2.3.2 are restated below,
" # " #" # " #
α̇ − m1V̄ L̄α 1 α − m1V̄ L̄δF
= 1 1
+ 1
δF (C.3.1)
Q̇ Iy
Mα Iy
MQ Q Iy
Mδ F
" # " #" # " #
= + δF (C.3.2)
Q 0 1 Q 0
ẋ = Ax + BδF (C.3.3)
" #
CW
= Cx + DδF (C.3.4)
Q
where,
" #
α
x= (C.3.5)
Q
" # " #
− m1V̄ L̄α 1 − m1V̄ L̄δF
A= 1 1
B= 1
(C.3.6)
Iy
Mα Iy
MQ Iy
Mδ F
" # " #
− m1 L̄α 0 − m1 L̄δF
C= D= (C.3.7)
0 1 0
A flaps control law is now defined which uses proportional feedback from the NSA,
δF = −KPδF (y − s) (C.3.8)
ref
= −KPδF (CW − CW )

ref
= −KPδF C0 x + D0 u − CW
ref
where CW is the reference NSA input, and C0 and D0 are the first rows of C and D
respectively. Thus,
" # !
h i α h i
ref
δF = −KPδF − m1 L̄α 0 + − m1 L̄δF δF − CW (C.3.9)
Q
" #
h i α h i
ref
δF = −KPδF − m1 L̄α 0 − KPδF − m1 L̄δF δF + KPδF CW
Q
" #
h i h i α
1 ref
δF + KPδF − L̄
m δF
δF = −KPδF − m1 L̄α 0 + KPδF CW
Q
" #
KPδF h
1
i α KPδF CWR
δF =− L̄
− m
L̄ α 0 + L̄
1 − KPδF mδF Q 1 − KPδF mδF
The control law is now substituted into the dynamic equation,

ref
" # " #" # " # " # " #
α̇ − m1V̄ L̄α 1 α KPδF − m1V̄ L̄δF h 1 i α KPδF CW − m1V̄ L̄δF
= 1 1
− L̄δF 1 − m L̄α 0 + L̄δF 1
Q̇ Iy Mα Iy MQ Q 1 − KPδF m Iy MδF Q 1 − KPδF m Iy MδF
ref
" # " #" # " # " # " #
1 KPδF 1 KPδF CW 1
CW −m L̄α 0 α −m L̄δF h 1 i α −m L̄δF
= − L̄δF
− m L̄α 0 + L̄δF
Q 0 1 Q 1 − KPδF m 0 Q 1 − KPδF m 0
" # " #" # " L̄δ L̄α #" # ref

" #
α̇ − m1V̄ L̄α 1 α KPδF F
2
m V̄
0 α K Pδ
C W − 1
mV̄
L̄δF
= 1 1
− L̄δF M L̄ + F
L̄δF 1
Q̇ Iy Mα Iy MQ Q 1 − KPδF m − IδyFm α 0 Q 1 − KPδF m Iy MδF
ref
" # " #" # " L̄ L̄ #" # " #
1 KPδF δF α
KPδF CW 1
CW −m L̄α 0 α m 2 0 α −m L̄δF
= − L̄δF
+ L̄δF
Q 0 1 Q 1 − KPδF m 0 0 Q 1 − KPδF m 0
 !   !
KPδ L̄δF L̄α L̄δF KPδ
" # − L̄α − F
1 " #  − F
 mV̄ L̄δ m2 V̄ mV̄ L̄δ 
α̇ 1−KPδ m
F  α  1−KPδ m
F
!  C ref

= F !
 Q +
  F
 W (C.3.10)
Q̇ 
 Mα KPδ MδF L̄α MQ   MδF KPδ
Iy +
F F 
L̄δ
F Iy m Iy Iy L̄δ
F
1−KPδ m 1−KPδ m
F F
 !
" # " #" # " L̄ #" # L̄δF KPδ
L̄α δF L̄α
CW − 0 α KPδF 0 α −
F
 ref
+ m
m m2 L̄δ
= − L̄δF
1−KPδ m
F
 CW (C.3.11)
Q 0 1 Q 1 − KPδF 0 0 Q F
m 0
let,
 
KPδF
∆= L̄δF
 (C.3.12)
1 − KPδF m
Now,
L̄ L̄α
" # " # " # " L̄ #
α̇ − mL̄αV̄ − ∆ mδF2 V̄ 1 α − mδV̄F ∆ ref
= M MδF L̄α MQ
+ Mδ CW (C.3.13)
Q̇ Iy
α
+ ∆ Iy m Iy
Q Iy
F
∆
" # " L̄δF L̄α
# " # " L̄δF
#
L̄α
CW − m − ∆ m2 0 α − m ∆ ref
= + CW (C.3.14)
Q 0 1 Q 0
The transfer function for the NSA response to a reference NSA input is now determined.
Calculating the closed-loop characteristic equation,
p(s) = det(sI − A) (C.3.15)

thus,
L̄ L̄α
" #!
L̄α
s+ mV̄
+ ∆ mδF2 V̄ −1
p(s) = det M L̄α M
(C.3.16)
− MIyα − ∆ IδyFm s − IyQ

L̄α L̄δF L̄α MQ Mα MδF L̄α
= s+ +∆ 2 s− − − −∆ (−1)
mV̄ m V̄ Iy Iy Iy m

2 MQ L̄α L̄δF L̄α MQ L̄α L̄δF L̄α
=s −s +s +∆ 2 − +∆ 2
Iy mV̄ m V̄ Iy mV̄ m V̄

Mα Mδ L̄α
+ +∆ F (−1)
Iy Iy m

2 L̄α L̄δF L̄α MQ MQ L̄α MQ L̄δF L̄α Mα MδF L̄α
=s +s +∆ 2 − − + ∆ 2 + − −∆
mV̄ m V̄ Iy Iy mV̄ Iy m V̄ Iy Iy m

The zeros of the system are the roots of the equation,
z(s) = Cadj(sI − A)B + D det(sI − A) (C.3.17)
solving the first part,
M
" L̄ L̄α
#" # " L̄ #
− L̄mα − ∆ δmF 2 0 s − IyQ 1 − mδV̄F ∆
Cadj(sI − A)B = M L̄α L̄ L̄α MδF (C.3.18)
0 1 MIyα + ∆ IδyFm s + mL̄αV̄ + ∆ mδF2 V̄ Iy
∆
 "
L̄ L̄α M L̄ L̄α L̄δF
#
(− L̄mα − ∆ δmF 2 )(s − IyQ ) − L̄mα − ∆ δmF 2
=  −MmV̄ ∆
Mα M L̄α L̄ L̄α δF
Iy
+ ∆ IδyFm s + mL̄αV̄ + ∆ mδF2 V̄ Iy
∆
h ih i h ih i
L̄ L̄α M L̄ L̄ L̄α MδF
(− L̄mα − ∆ δmF 2 )(s − IyQ ) − mδV̄F ∆ + − L̄mα − ∆ δmF 2 Iy
∆
= h
M L̄α
ih
L̄
i h
L̄ L̄α
ih
M
i 
Mα δ δ L̄α δ δ
Iy
+ ∆ F
Iy m
− mV̄
F
∆ + s + mV̄
+ ∆ F
m V̄2 Iy
F
∆
now solve for,

" L̄δF
#
− m
∆ 2 L̄α L̄δF L̄α MQ MQ L̄α MQ L̄δF L̄α Mα MδF L̄α
D det(sI − A) = s +s +∆ 2 − − + ∆ 2 + − −∆ (C.3.19)
0 mV̄ m V̄ Iy Iy mV̄ Iy m V̄ Iy Iy m
145

combining the results of the two parts above,
z(s) = Cadj(sI − A)B + Ddet(sI − A) (C.3.20)
 h i 
L̄α L̄δ L̄α MQ L̄δ L̄α L̄δ L̄α Mδ 
L̄δ L̄δ L̄α L̄δ L̄α Mδ L̄α
 (− m − ∆ m2 )(s − Iy ) − mV̄ ∆ + − m − ∆ m2
F F F F
Iy
∆ − F
∆ s 2+s L̄α
+ ∆ F
−
MQ
−
MQ L̄α
+
MQ
∆ F
+ − Mα
− ∆ F
2 2

= h + m mV̄ m V̄ Iy Iy mV̄ Iy m V̄ Iy Iy m 
Mδ L̄α L̄δ L̄δ L̄α Mδ
i
 Mα L̄α 
Iy
+ ∆ I Fm − mV̄F ∆ + s + m V̄
+ ∆ mF2 V̄ I
F
∆ 0
y y
 h 
L̄δ L̄α L̄δ L̄α MQ L̄δ L̄δ L̄α Mδ
i
L̄α L̄α MQ
− m s+ m I −∆ m F
2 s+∆ m F
2 Iy
− mV̄F ∆ + − L̄mα − ∆ m F
2 Iy
F
∆
 y 
 
 L̄δ 2 L̄ L̄δ L̄α M Q M Q L̄ M Q L̄δ L̄α M Mδ L̄α 
 + − mF ∆ s + s mα V̄
+ ∆ mF2 V̄ − I − I mα V̄
+ I ∆ mF2 V̄ + − I α − ∆ I Fm 
=

y y y y y 

 
 
M L̄ L̄ L̄ L̄
hM i
δF α δF α
 Mα δF L̄α δF

Iy
+ ∆ Iy m
− mV̄
∆ + s + mV̄
+ ∆ 2
m V̄ Iy
∆
 
L̄δ L̄α L̄α MQ L̄δ L̄α L̄δ L̄α MQ 1 L̄α L̄δ L̄α m MδF
− m
F
∆ − m
s + m I
− ∆ F
m 2 s + ∆ F
m 2 I V̄
+ − m
− ∆ F
m 2 − L̄δ I
y
y F y
 
 
 2+s L̄α L̄δ L̄α MQ MQ L̄α MQ L̄δ L̄α Mα Mδ L̄α 
F F F
 + s mV̄
+ ∆ m2 V̄
− Iy
− Iy mV̄
+ Iy
∆ m2 V̄
+ − Iy
− ∆ Iy m

=



 
 h 
Mδ L̄α L̄δ L̄δ L̄α Mδ
 i 
Mα F L̄α
Iy
+ ∆ Iy m
− mV̄
F
∆ + s + mV̄
+ ∆ F
m2 V̄ Iy
F
∆
 
L̄δ L̄α L̄α MQ L̄δ L̄α L̄δ L̄α MQ L̄α MδF L̄α Mδ 2+ L̄α L̄δ L̄α MQ MQ L̄α MQ L̄δ L̄α Mα Mδ L̄α
F F F F F F F
− m
∆ − mV
s + mV Iy
− ∆ m V2 s + ∆ m V2 Iy
+ L̄δ Iy
+ ∆ mIy
+ s mV̄
s + ∆ 2
m V̄
s − Iy
s − Iy mV̄
+ Iy
∆ 2
m V̄
+ − Iy
− ∆ Iy m
F
 
 
=




 L̄δ Mα L̄δF M δF L̄α M δF M δF L̄α MδF L̄δF L̄α

F
− mV̄ ∆ I − mV̄ ∆∆ I m + I ∆s + I ∆ mV̄ + I ∆∆ m2 V̄
y y y y y
 
L̄δ L̄α MδF
M
 − m ∆
F
L̄δ Iy
+ s2 + − I Q s + − M Iy
α
F y 
 
=



h
L̄δ Mδ Mδ
 i 
Mα L̄α
F
− mV̄ ∆ I F
+ I ∆s + I ∆ mV̄ F
y y y
 
L̄δ 2 − MQ s + L̄α MδF Mα
F
− m
∆ s Iy L̄δ Iy
− Iy
F
 
 
=




Mδ L̄δ Mα
 h i 
F F L̄α
I
∆ s − mV̄ + mV̄
y
146
The transfer function for a NSA response to a reference NSA input can therefore be
written as,
h ih i
L̄ M M
GC W
C ref
(s) =
L̄δF L̄α
F
L̄ L̄α

M L̄α

L̄α MQ MQ L̄α MQ
W
s2 + s mV̄
+ ∆ m2 V̄ − Iy − Iy mV̄ + Iy ∆ mδF2 V̄ + − MIyα − ∆ IδyFm
(C.3.21)
The desired characteristic equation is,
αc (s) = s2 + α1 s + α0 (C.3.22)
equating the coefficients gives,
L̄α L̄δ L̄α MQ

α1 = + ∆ F2 − (C.3.23)
mV̄ m V̄ Iy
MQ L̄α MQ L̄δF L̄α Mα Mδ L̄α
α0 = − − ∆ 2 − −∆ F
Iy mV̄ Iy m V̄ Iy Iy m
thus,
m2 V̄

L̄α MQ
∆= α1 − + (C.3.24)
and,
MQ L̄α MQ L̄δF L̄α Mα Mδ L̄α

α0 = − − ∆ 2 − −∆ F (C.3.25)
Iy mV̄ Iy m V̄ Iy Iy m
MQ L̄δF L̄α Mδ L̄α MQ L̄α Mα
∆ 2 +∆ F = −α0 − −
Iy m V̄ Iy m Iy mV̄ Iy

MQ L̄δF L̄α MδF L̄α MQ L̄α Mα
∆ + = − α0 + +
Iy m2 V̄ Iy m Iy mV̄ Iy

MQ L̄α Mα
α0 + Iy mV̄ + Iy
∆ = −
MQ L̄δF L̄α MδF L̄α
2
Iy m V̄
+ Iy m

α0 Iy m V̄ + MQ L̄α m + Mα m2 V̄
2
∆=−

Equating the above two equations of ∆ gives,

m2 V̄ α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄

L̄α MQ
α1 − + =− (C.3.26)
L̄δF L̄α mV̄ Iy MQ L̄δF L̄α + MδF L̄α mV̄

L̄δF L̄α α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄

L̄α MQ
α1 − + =− 2
mV̄ Iy m V̄ MQ L̄δF L̄α + MδF L̄α mV̄

L̄α MQ L̄δF α0 Iy mV̄ + MQ L̄α + Mα mV̄
α1 − + =−
mV̄ Iy mV̄ MQ L̄δF + MδF mV̄
L̄α MQ L̄δ
α1 MQ L̄δF + MδF mV̄ − MQ L̄δF + MδF mV̄ + MQ L̄δF + MδF mV̄ = − F α0 Iy mV̄ + MQ L̄α + Mα mV̄ (C.3.27)
mV̄ Iy mV̄
L̄α MQ L̄δ MQ L̄α
α1 MQ L̄δF + MδF mV̄ − MQ L̄δF + MδF mV̄ + MQ L̄δF + MδF mV̄ = −L̄δF α0 Iy − F − L̄δF Mα
mV̄ Iy mV̄
L̄α L̄α MQ L̄δ MQ L̄α
α1 MQ L̄δF + MδF mV̄ − MQ L̄δF − MδF mV̄ + MQ L̄δF + MδF mV̄ = −L̄δF α0 Iy − F − L̄δF Mα
mV̄ mV̄ Iy mV̄
MQ
α1 MQ L̄δF + MδF mV̄ − L̄α MδF + MQ L̄δF + MδF mV̄ = −L̄δF α0 Iy − L̄δF Mα
Iy
MQ
α1 MQ L̄δF + MδF mV̄ + MQ L̄δF + MδF mV̄ − L̄α MδF + L̄δF Mα = −L̄δF α0 Iy
Iy
MQ L̄δF Iy + MδF Iy mV̄ + MQ2 L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy = −L̄δF α0 Iy2

α1
148
The characteristic equation can also be written as,
αc (s) = (s2 + 2ζωn s + ωn2 ) (C.3.28)
where the complex pole pair corresponds to the short period mode. Equating the coeffi-
cients allows an equation for the natural frequency ωn to be determined,
2
L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy = −L̄δF Iy2 ωn2

2ζωn MQ L̄δF Iy + MδF Iy mV̄ + MQ
(C.3.29)
L̄δF Iy2 ωn2 + 2ζωn MQ L̄δF Iy + MδF Iy mV̄ + MQ

2

L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy = 0
Now solve for the positive values of ωn ,

√
−b + b2 − 4ac
x= (C.3.30)
2a
2
b2 = 4ζ 2 MQ L̄δF Iy + MδF Iy mV̄ (C.3.31)
4ac = 4L̄δF Iy2 MQ

2

L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy (C.3.32)
2
b2 − 4ac = 4ζ 2 MQ L̄δF Iy + MδF Iy mV̄ − 4L̄δF Iy2 MQ
2

L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy
(C.3.33)

therefore,
q 2
−2ζ MQ L̄δF Iy + MδF Iy mV̄ + 4ζ 2 MQ L̄δF Iy + MδF Iy mV̄ − 4L̄δF Iy2 MQ2 L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy
ωn = (C.3.34)
2L̄δF Iy2
q 2
−ζ MQ L̄δF Iy + MδF Iy mV̄ + ζ 2 MQ L̄δF Iy + MδF Iy mV̄ − L̄δF Iy2 MQ2 L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy
=
L̄δF Iy2
q
(ζη)2 − L̄δF Iy2 MQ2 L̄δF + MQ MδF mV̄ − L̄α MδF Iy + L̄δF Mα Iy

−ζη +
=
L̄δF Iy2
where,
η = MQ L̄δF Iy + MδF Iy mV̄ (C.3.35)
150
Substituting the equation for ωn into the two equations for ∆ allows the derivation of two
different but equivalent equations for the feedback gain KPδF .
For the first ∆ equation containing α1 ,

m2 V̄

L̄α MQ
∆= α1 − + (C.3.36)
where,
 
KPδF
∆= L̄δF
 (C.3.37)
1 − KPδF m
now,
 
KPδF 2

 = m V̄ L̄α MQ

L̄δF
α1 − + (C.3.38)
1 − KPδF L̄δF L̄α mV̄ Iy
m
m2 V̄

L̄α MQ L̄δF
KPδF = α1 − + 1 − KPδF
L̄δF L̄α mV̄ Iy m
2

m V̄ L̄α MQ mV̄ L̄α MQ
KPδF = α1 − + − KPδF α1 − +
L̄δF L̄α mV̄ Iy L̄α mV̄ Iy
2

mV̄ L̄α MQ m V̄ L̄α MQ
KPδF α1 − + + KPδF = α1 − +
L̄α mV̄ Iy L̄δF L̄α mV̄ Iy

m2 V̄ L̄α MQ
L̄δF L̄α
α1 − mV̄ + Iy
KPδF =
mV̄ L̄α MQ
L̄α
α 1 − mV̄
+ Iy +1

L̄α m2 V̄ L̄α MQ
mV̄ L̄δF L̄α
α 1 − mV̄
+ Iy
KPδF =
L̄α mV̄ L̄α M L̄α
mV̄ L̄α
α1 − m V̄
+ IyQ + m V̄

m L̄α M
L̄
α1 − m V̄
+ IyQ
δF
KPδF = M
L̄α L̄α
α1 − m V̄
+ IyQ + m V̄

m L̄α MQ
L̄
α1 − mV̄
+ Iy
δF
KPδF = MQ
α1 + Iy

m MQ L̄α
L̄δF
α1 + Iy − L̄δF V̄
KPδF = MQ
α1 + Iy

M
m α1 + IyQ − L̄V̄α
KPδF =
M
L̄δF α1 + IyQ

M
m α1 + IyQ L̄α
KPδF = − V̄
M MQ
L̄δF α1 + IyQ L̄δF α1 + Iy
m L̄
KPδF = − α
L̄δF MQ
L̄δF V̄ α1 + Iy
m L̄
KPδF = − α
L̄δF MQ
L̄δF V̄ 2ζωn + Iy

For the second equation for ∆, containing α0 ,

α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄
∆=− (C.3.39)
where,
 
KPδF
∆= L̄δF
 (C.3.40)
1 − K P δF m
now,
 

KPδF
 =− (C.3.41)
L̄δF M L̄ L̄ + M L̄ mV̄
1 − KPδF m Q δ F α δ F α

KPδF

L̄δ
=− (1 − KPδF F )
1 MQ L̄δF L̄α + MδF L̄α mV̄ m
L̄δ α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄ α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄

KPδF − KPδF F =−
m MQ L̄δF L̄α + MδF L̄α mV̄ MQ L̄δF L̄α + MδF L̄α mV̄

2 2

m MQ L̄δF L̄α + MδF L̄α mV̄ KPδF − KPδF L̄δF α0 Iy m V̄ + MQ L̄α m + Mα m V̄ = − m MQ L̄δF L̄α + MδF L̄α mV̄
KPδF m MQ L̄δF L̄α + MδF L̄α mV̄ − L̄δF α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄ = − α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄ m

α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄ m

KPδF = −
m MQ L̄δF L̄α + MδF L̄α mV̄ − L̄δF α0 Iy m2 V̄ + MQ L̄α m + Mα m2 V̄
KPδF = −
MQ L̄δF L̄α + MδF L̄α mV̄ − L̄δF α0 Iy mV̄ + MQ L̄α + Mα mV̄
KPδF = −
MδF L̄α mV̄ − L̄δF α0 Iy mV̄ − L̄δF Mα mV̄
α0 Iy mV̄ + MQ L̄α + Mα mV̄
KPδF =−
MδF L̄α V̄ − L̄δF α0 Iy V̄ − L̄δF Mα V̄
MQ L̄α
α0 + Iy mV̄
+M
Iy
α
KPδF = − M
δF L̄α L̄ F α0 L̄ Mα
152
Iy m − δm − δIFy m
In summary
The natural frequency of the closed-loop poles is given by,

q
(ζη)2 − L̄δF Iy2 MQ2 L̄δF + MQ MδF mV̄

−ζη +
ωn = (C.3.42)
L̄δF Iy2
where,
η = MQ L̄δF Iy + MδF Iy mV̄ (C.3.43)
The flaps feedback gain is give by either,
m L̄
KPδF = − α (C.3.44)
L̄δF L̄δF V̄ 2ζωn +
MQ
Iy
or,
ωn2 Iy mV̄ + MQ L̄α + Mα mV̄
KPδF = − (C.3.45)
MδF L̄α V̄ − L̄δF ωn2 Iy V̄ − L̄δF Mα V̄
Pole placement therefore involves choosing a damping ratio ζ , then calculating the res-
ultant natural frequency ωn and the gain KPδF .
C.4 Glide Path Offset
Figure C.1 – Control system block diagram
With reference to Figure C.1 the steady state altitude error is found as follows,
E(s) 1
= (C.4.1)
R(S) F (s)P (s)C(s)
1
= (C.4.2)
τCR 1
1+ KP
τCR s + 1 s h
For a ramp input r(t) = At, the final value theorem states,

E(s) A
eSS = lim e(t) = lim s (C.4.3)
t→∞ s→0 R(s) s2
 
1 1
= lim s (C.4.4)

s→0 τCR 1 s2

1+ KPh
τCR s + 1 s
A
= (C.4.5)
τCR KPh
The altitude control is a type one system, therefore the steady state error to a ramp input
is given by,
A
eSS = (C.4.6)
KVec
where KVec is the velocity error constant. Thus,
KVec = KPh τCR (C.4.7)
KPh is the altitude feedback gain.
Appendix D
Additional Flight Test Information
D.1 Helderberg Radio Flyer’s Airfield
Figure D.1 – Satellite photo of Helderberg Radio Flyers (HRF) airfield. Source: Google Earth
155
APPENDIX D. ADDITIONAL FLIGHT TEST INFORMATION 156
D.2 Windspeed data - 29 November
Figure D.2 – Wind speed during Landing Test 3,4 and 5

D.3 Flight Test Cards
D.3.1 Flight Test 1 - 25 July
TEST CARD: Fixed Wing ATOL Notes:
UAV: Phoenix
Test Name:
Flight Test 1
Maiden flight test of the airframe
Test No: Airframe Flt No: Location: Date:

1 n/a HRF 25 July 2011
Test Coordinator: Pilot: Ground Station: Test Leader:

Lionel Basson Michael Basson Nico Alberts Nico Alberts
Configuration and Status:
This is the maiden flight test of the airframe and will be conducted exclusively under
the control of the safety pilot. It serves to verify the airframe from a performance,
stability and control perspective.
Restrictions:
Maximum flight time 5 minutes.
No excessive manoeuvres
ESL Test Card Page 1 of 2
(a) Page 1 of 2
Take-Off # 1
Test # Test Goals Pilot Information Expected Behaviour
1
Verify the airframe from a performance, stability Safety Pilot uses best judgement to evaluate the airframe from The aircraft will fly as per safety pilot commands
and control perspective. a performance, stability and control perspective.
If any adjustments are required, the aircraft should be landed

and the control surfaces or weight distribution adjusted on the
ground.
(b) Page 2 of 2
Figure D.3 – Flight test card - Flight Test 1 - 25 July

D.3.2 Flight Test 2 - 10 August
UAV: Phoenix
Test Name:
Flight Test 2
Controller Steps, Airspeed, Climb Rate, Altitude, Roll
Angle and Dutch Roll Damper

2 n/a HRF 10 August 2011


Changes since Flight Test 1 – 25 July flight:
None
The previous flight test confirmed the proper functioning of the aircraft, this flight test
will commence testing of the controllers
This flight test makes use of the uBlox GPS module. Allowance for the future
integration of the NovAtel DGPS unit and antenna has been made through the use of
two dummy weights.
Restrictions:
(a) Page 1 of 6
Take-Off # 1
1 During this test the airspeed controller is tested. Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
at the start of the FAR long leg with the aircraft straight and on the markings on the runway
level at around 60 m altitude
Safety Pilot takes off and engages autopilot on the
Safety Pilot maintains control over all actuators except for the transmitter at the start of the FAR long leg with the
Airspeed Controller Gains set at: throttle aircraft straight and level at around 60 m altitude
a = 0.5
zeta = 0.8 Safety Pilot continues flying a relaxed circuit at reasonably The aircraft will fly as per safety pilot commands except
constant altitude. that the airspeed is regulated by the controller
Safety Pilot monitors aircraft for any strange behaviour and Airspeed Controller will be given several references:
must retake control if airspeed or altitude become too low/high. Leg 1:
20 m/s
Leg 2:
16 m/s
Test Leader will inform Safety Pilot that test is complete. 13 m/s
Safety Pilot disarms autopilot on the transmitter and takes full Leg 3:
control of aircraft. 16 m/s
If aircraft airspeed is judged too high/low Safety Pilot

disengages autopilot and stabilises aircraft.
Land Aircraft: Soft Landing should consist of a glide followed

by a flare and touchdown.
Upon completion the Safety Pilot will be informed to

retake control of the aircraft and land as close as
possible to the take-off point.
(b) Page 2 of 6
Figure D.4 – Flight test card - Flight Test 2 - 10 August

Take-Off # 2
2 During this test the airspeed and climb rate Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
controllers are tested concurrently. at the start of the FAR long leg with the aircraft straight and on the markings on the runway
Safety Pilot maintains control over all actuators except for transmitter at the start of the FAR long leg with the
throttle and elevator. aircraft straight and level at around 60 m altitude
NSA Controller Gains set at:
Safety Pilot maintains level wings as much as possible for the
a=4 duration of the leg. Before reaching the end of the leg Safety The aircraft will fly as per safety pilot commands except
zeta = 0.707 Pilot disarms the autopilot. Safety Pilot is now in full control to that the airspeed and climb rate is regulated by the
Nbar = 2 make the turn. When the next leg is entered, the autopilot is controllers during the long legs
armed once more.
ClimbRate Controller Gains set at: The following references will be set:
Airspeed: 15 m/s throughout
Leg1: Kp = -0.5 Safety Pilot monitors aircraft for any strange behaviour and Leg 1:
must retake control if airspeed or altitude become too low. Climb Rate: 0 m/s +2m/s
Leg 2:
Look out for any pitch oscillation. Any visible oscillation will Climb Rate: -2m/s 0m/s
Airspeed Controller Gains set at: likely worsen and the Safety Pilot must retake control
immediately.
a = 0.5
zeta = 0.8 Test Leader will inform Safety Pilot that test is complete.
Safety Pilot disarms autopilot on the transmitter and takes full Upon completion the Safety Pilot will be informed to
control of aircraft. retake control of the aircraft and land as close as

(c) Page 3 of 6
Take-Off # 3
3 During this test the airspeed and altitude Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
zeta = 0.707 Pilot disarms the autopilot. Safety Pilot is now in full control to that the airspeed and altitude is regulated by the
armed once more.
ClimbRate Controller Gains set at:
Kp = -0.5 Safety Pilot monitors aircraft for any strange behaviour and The following references will be set:
must retake control if airspeed or altitude become too low. Airspeed: 15 m/s throughout
Leg 1:
Altitude controller gain set at: Look out for any pitch oscillation. Any visible oscillation will Altitude Steps: + 25 m
likely worsen and the Safety Pilot must retake control Leg 2:
Kp = 0.5 immediately. Altitude Steps: - 25 m
Test Leader will inform Safety Pilot that test is complete.

Airspeed Controller Gains set at: Safety Pilot disarms autopilot on the transmitter and takes full
control of aircraft.
a = 0.5
zeta = 0.8

Land Aircraft: Soft Landing should consist of a glide followed possible to the take-off point.
(d) Page 4 of 6

Take-Off # 4
4 During this test the roll angle controller is tested Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
ailerons. aircraft straight and level at around 60 m altitude
Safety Pilot maintains altitude as much as possible for the

duration of the leg. Before reaching the end of the leg Safety The aircraft will fly as per safety pilot commands except
Pilot disarms the autopilot. Safety Pilot is now in full control to that the roll angle is regulated by the controllers during
make the turn. When the next leg is entered, the autopilot is the long legs
Roll Angle Controller Gains armed once more.
Leg1: Roll Angle Pole = 4 .

Error Angle Pole = 2 The following references will be set:
If the aircraft’s roll angle is too high Safety Pilot disengages Leg 1:
autopilot and stabilises aircraft Roll angle: 0degrees +15 degrees
Look out for any roll oscillation. Any visible oscillation will Leg 2:
likely worsen and the Safety Pilot must retake control Roll angle: 0degrees -15 degrees
immediately

Safety Pilot disarms autopilot on the transmitter and takes full

(e) Page 5 of 6
Take-Off # 5
5 During this test the Dutch Roll Damper is tested Safety Pilot takes off and enters a relaxed circuit at a Before takeoff the aircraft must be positioned accurately
moderate altitude. on the markings on the runway
Safety Pilot maintains control over all actuators. However, the

rudder is also under control of the Dutch Roll Damper.
DRD Gains set at: Test Leader will arm the DRD controller and inform Safety
Pilot. Safety Pilot engages autopilot on the transmitter at the
DRD Pole = 0.333 start of a long leg with the aircraft straight and level.
DRD Gain = 0.3
Safety Pilot maintains level flight as much as possible for the
duration of the leg. The aircraft will fly as per the safety pilot commands and
attempt to dampen the dutch roll mode when the
Perform these tests on the closer leg so that the response can autopilot is armed.
be visually seen.
1st Pass:
Controller disarmed:
Input rudder doublet to excite the Dutch roll, pilot should
remember the range of the doublet that he used.
2nd Pass:
Arm controller:
Input rudder doublet to excite the Dutch roll, one should be
able to visually see the damped response
Upon completion the Safety Pilot will be informed to retake

control of the aircraft and land as close as possible to the take-
off point.
Land Aircraft:
Soft Landing should consist of a glide followed by a flare and
touchdown.
Flaps should be in the middle position.
(f) Page 6 of 6

D.3.3 Flight Test 3 - 7 September
UAV: Phoenix
Test Name:
Flight Test 3
Controller Steps, Navigation and glide path, Natural
Elevator and Flaps response

3 n/a HRF 7 September 2011


Changes since Flight Test 3 – 10 August flight:
None
Restrictions:
(a) Page 1 of 6
Take-Off # 1
1 During this test the airspeed and climb rate Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
zeta = 0.707 Pilot disarms the autopilot. Safety Pilot is now in full control to that the airspeed and climb rate is regulated by the
armed once more.
ClimbRate Controller Gains set at: The following references will be set:
Airspeed: 18 m/s throughout
Leg1: Kp = -0.5 Safety Pilot monitors aircraft for any strange behaviour and Leg 1:
Leg2: Kp = -0.5 must retake control if airspeed or altitude become too low. Climb Rate: 0 m/s +3m/s(for 2 seconds) 0 m/s
Leg3: Kp = -0.8 Leg 2:
Leg4: Kp = -1.0 Look out for any pitch oscillation. Any visible oscillation will Climb Rate: 0m/s -3m/s(for 2 seconds) 0m/s
immediately. Climb Rate: 0m/s +3m/s(for 2 seconds) 0m/s
Leg 4:
Airspeed Controller Gains set at: Test Leader will inform Safety Pilot that test is complete. Climb Rate: 0m/s -3m/s(for 2 seconds) 0m/s
a = 0.5 control of aircraft.
zeta = 0.8
Land Aircraft: Soft Landing should consist of a glide followed Upon completion the Safety Pilot will be informed to
by a flare and touchdown. retake control of the aircraft and land as close as
(b) Page 2 of 6
Figure D.5 – Flight test card - Flight Test 3 - 7 September

Take-Off # 2
2 During this test the airspeed and altitude Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
zeta = 0.707 Pilot disarms the autopilot. Safety Pilot is now in full control to that the airspeed and altitude is regulated by the
armed once more.
ClimbRate Controller Gains set at highest stable
from previous test:
Safety Pilot monitors aircraft for any strange behaviour and The following references will be set:
Kp = -1.0 must retake control if airspeed or altitude become too low. Airspeed: 18 m/s throughout
Leg 1:
Look out for any pitch oscillation. Any visible oscillation will Altitude Command: 70 m 60 m
Altitude Controller Gain set at:
immediately. Altitude Steps: 60 m 70 m
Kp = 0.5
Airspeed Controller Gains set at: control of aircraft.
a = 0.5
zeta = 0.8
Land Aircraft: Soft Landing should consist of a glide followed possible to the take-off point.
(c) Page 3 of 6
Take-Off # 3
3 During this test the roll angle controller is tested Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
ailerons. aircraft straight and level at around 60 m altitude
Safety Pilot maintains altitude as much as possible for the

duration of the leg. Before reaching the end of the leg Safety The aircraft will fly as per safety pilot commands except
Pilot disarms the autopilot. Safety Pilot is now in full control to that the roll angle is regulated by the controllers during
make the turn. When the next leg is entered, the autopilot is the long legs
Roll Angle Controller Gains armed once more.
Leg1: Roll Angle Pole = 4 .

Error Angle Pole = 2 The following references will be set:
If the aircraft’s roll angle is too high Safety Pilot disengages Leg 1:
autopilot and stabilises aircraft Roll angle: 0degrees +15 degrees(for 3 seconds)
0degrees
Look out for any roll oscillation. Any visible oscillation will repeat.
likely worsen and the Safety Pilot must retake control
immediately

control of aircraft. Upon completion the Safety Pilot will be informed to

(d) Page 4 of 6

Take-Off # 4
4 During this test the navigation controller is tested Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
alongside a glide path at the start of the FAR long leg with the aircraft straight and on the markings on the runway
Test Leader will arm the Navigation controller and inform transmitter at the start of the FAR long leg with the
Safety Pilot. Safety Pilot engages autopilot on the transmitter aircraft straight and level at around 60 m altitude
at the start of the FAR long leg with the aircraft straight and
level at around 60 m altitude.
The aircraft will fly the landing circuit, and on the second
NSA Controller Gains set at: Safety Pilot does not have control over any actuators. pass descend along the glide path.
a=4 Safety Pilot monitors aircraft for any strange behaviour and
zeta = 0.707 must retake control if airspeed or altitude become too low. The Aircraft will be allowed to make one full circuit; the
Nbar = 2 The controller can only bank at 30 degrees, if a steeper bank glideslope controller will then be activated.
angle is judged to be occurring Safety Pilot must retake
ClimbRate Controller Gains set : control. Setpoints:
Kp = -1.0 Safety Pilot will be informed if the Navigation Controller Airspeed: 18 m/s
becomes disengaged for whatever reason. If this occurs the
aircraft will keep flying straight and level indefinitely. Safety Altitude: 60 m
Pilot must retake control immediately.
Roll Angle Controller Gains Glide path height: 30m
The approximate location of the waypoints will be discussed
Leg1: Roll Angle Pole = 4 prior to flight. If the aircraft fails to turn when expected safety
Error Angle Pole = 2 pilot must retake control.
The Aircraft will be allowed to make one full circuit; the Upon completion the Safety Pilot will be informed to
glideslope controller will then be activated. No action is retake control of the aircraft and land as close as
required of the Safety Pilot for the activation. possible to the take-off point.

off point.
Land Aircraft:
touchdown.
(e) Page 5 of 6
Take-Off # 5
5 During this test the natural response of the Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
elevator and flaps is invetigated at the start of the FAR long leg with the aircraft straight and on the markings on the runway
Safety Pilot maintains control over all actuators except for the transmitter at the start of the FAR long leg with the
throtle. aircraft straight and level at around 60 m altitude
Only the airspeed controller is activated and references

NSA Controller Gains set at: between 15 – 18 m/s are set.
The aircraft will fly as per the safety pilot commands and
a=4 Flaps Response: attempt to maintain a constant airspeed
zeta = 0.707
Nbar = 2 Safety Pilot flies straight, level and trimmed. Setpoints:
Toggle Flaps to Middle position. This is a relatively small flaps
ClimbRate Controller Gains set : deflection so no elevator mixing will be done. The aircraft Airspeed: 18 m/s
must be allowed to pitch naturally (no change in elevator
Kp = -1.0 input). After 2 seconds, or if the pitch angle becomes
excessive the flaps may be disengaged.
If successfully executed, carry on with the next test. Otherwise
repeat.
Roll Angle Controller Gains Upon completion the Safety Pilot will be informed to
Leg1: Roll Angle Pole = 4 Elevator Response: possible to the take-off point.
Error Angle Pole = 2
Safety Pilot flies straight, level and trimmed.
“Toggle” Elevator. The aircraft must be allowed to pitch
naturally After 2 seconds, or if the pitch angle becomes
excessive the elevator may be returned

off point.
Land Aircraft:
touchdown.
(f) Page 6 of 6

D.3.4 Flight Test 4 - 20 September
UAV: Phoenix
Test Name:
Flight Test 4
Navigation Circuit , climbrate steps and glide path with
DLC

4 n/a HRF 20 September 2011


Changes since Flight Test 3 – 7 September 2011 flight:
None
Restrictions:
(a) Page 1 of 4
Take-Off # 1
1 During this test the navigation controller is tested Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
for both estimators. at the start of the FAR long leg with the aircraft straight and on the markings on the runway
Safety Pilot does not have control over any actuators. transmitter at the start of the FAR long leg with the
aircraft straight and level at around 60 m altitude
NSA Controller Gains set at: Waypoint Navigation:
Test Leader will arm the Navigation controller and inform
a=4 Safety Pilot. Safety Pilot engages autopilot on the transmitter
zeta = 0.707 at the start of the FAR long leg with the aircraft straight and
Nbar = 2 level at around 60 m altitude. It is preferable for the controller The Aircraft will be allowed to fly one and a half
to be engaged at a higher altitude instead of too low. circuits. At a constant airspeed and altitude. Safety
Pilot must land as the aircraft approaches him for the
DLC Gains set at: Safety Pilot monitors aircraft for any strange behaviour and second time.
Not used must retake control if airspeed or altitude become too low.
The controller can only bank at 30 degrees, if a steeper bank
ClimbRate control.
Kp = -1.0 Setpoints:
Safety Pilot will be informed if the Navigation Controller
becomes disengaged for whatever reason. If this occurs the Airspeed: 15 m/s
Roll Angle Controller Gains aircraft will keep flying straight and level indefinitely. Safety
Pilot must retake control immediately. Altitude: 50 m
Roll Angle Pole = 4
Error Angle Pole = 2 The approximate location of the waypoints will be discussed
prior to flight. If the aircraft fails to turn when expected safety
pilot must retake control.
The Aircraft will be allowed to fly one and a half circuits. Upon completion the Safety Pilot will be informed to
Safety Pilot must land as the aircraft approaches him for the retake control of the aircraft and land as close as
second time. possible to the take-off point.
The aircraft must be reset on the runway ready for takeoff.

Test Leader will then reconfigure the estimator to the triad
method and inform Safety Pilot when ready.
Safety Pilot takes off as does exactly the same as previously.


(b) Page 2 of 4

Take-Off # 2
2 Before takeoff the aircraft must be positioned accurately on
During this test several climb rate steps will be Safety Pilot takes off and engages autopilot on the the markings on the runway
performed with the DLC controller transmitter at the start of the FAR long leg with the
aircraft straight and level at around 60 m altitude Safety Pilot takes off and engages autopilot on the
transmitter at the start of the FAR long leg with the
Safety Pilot maintains control over all actuators except aircraft straight and level at around 60 m altitude
for throttle, elevator and flaps.
a=4
zeta = 0.707 Safety Pilot maintains level wings as much as possible The following references will be set:
Nbar = 2 for the duration of the leg. Before reaching the end of Airspeed: 15 m/s throughout
the leg Safety Pilot disarms the autopilot. Safety Pilot is Leg 1:
DLC Gains set at: now in full control to make the turn. When the next leg Climb Rate: 0 m/s +2m/s(for 2 seconds) 0 m/s
is entered, the autopilot is armed once more Leg 2:
Leg1: Kp = 0 CutoffFreq= 4 CCF = 0 Climb Rate: 0m/s -2m/s(for 2 seconds) 0m/s
Leg2: Kp = 0.2 CutoffFreq= 4 CCF = 0 Leg 3:
Leg3: Kp = 0.5 CutoffFreq= 4 CCF = 0 Climb Rate: 0m/s +2m/s(for 2 seconds) 0m/s
Leg4: Kp = 0.7 CutoffFreq= 4 CCF = 0 Leg 4:
If aircraft airspeed or altitude is judged too high/low Climb Rate: 0m/s -2m/s(for 2 seconds) 0m/s
Leg4: Kp = 0.7 CutoffFreq= 4 CCF = 0.1 Safety Pilot disengages autopilot and stabilises aircraft. etc
Leg6: Kp = 0.7 CutoffFreq= 4 CCF = 0.15
Leg7: Kp = 0.7 CutoffFreq= 4 CCF = 0.2 Look out for any pitch oscillation. Any visible
Leg8: Kp = 0.7 CutoffFreq= 5 CCF = 0.2 oscillation will likely worsen and the Safety Pilot must
retake control immediately.
Kp = -1.0 Test Leader will inform Safety Pilot that test is
complete. Safety Pilot disarms autopilot on the Upon completion the Safety Pilot will be informed to retake
transmitter and takes full control of aircraft. control of the aircraft and land as close as possible to the
Altitude Controller Gains set at: take-off point.
Kp = 0.5
Land Aircraft:
Soft Landing should consist of a glide followed by a
flare and touchdown
(c) Page 3 of 4
Take-Off # 3
3 During this test the navigation controller is Safety Pilot takes off and engages autopilot on the transmitter Before takeoff the aircraft must be positioned accurately
enabled, and a glide path followed on the at the start of the FAR long leg with the aircraft straight and on the markings on the runway
runaway leg level at around 60 m altitude
Safety Pilot does not have control over any actuators. transmitter at the start of the FAR long leg with the
The aircraft will fly the runway circuit.
During the runway leg, the aircraft will descend from 60 m
a=4 down to 35 m when it reaches the middle of the runway.
zeta = 0.707 The aircraft will fly the runway circuit.
Nbar = 2 Safety Pilot must take control if any severe climb rate is
observed. The aircraft should be descending at around 2 m/s. During the runway leg, the aircraft will descend from
60 m down to 35 m when it reaches the middle of the
DLC Gains set at: After the middle of the runway is passed, the aircraft will begin runway.
Not used to climb as well as turn left at the next waypoint. Safety pilot
must take control if the airspeed drops significantly.
ClimbRate The Aircraft will be allowed to fly the circuit until the test time
Kp = -1.0 runs out Setpoints:
Airspeed: 15 m/s
Safety Pilot monitors aircraft for any strange behaviour and
Roll Angle Controller Gains
must retake control if airspeed or altitude become too low. Altitude: 60 m
The controller can only bank at 30 degrees, if a steeper bank
Roll Angle Pole = 4
Error Angle Pole = 2
control.

becomes disengaged for whatever reason. If this occurs the Upon completion the Safety Pilot will be informed to
aircraft will keep flying straight and level indefinitely. Safety retake control of the aircraft and land as close as
Pilot must retake control immediately. possible to the take-off point.
The approximate location of the waypoints will be discussed

prior to flight. If the aircraft fails to turn when expected safety
pilot must retake control.


(d) Page 4 of 4

D.3.5 Flight Test 5 - 25 November
TEST CARD: Fixed Wing ATOL using DGPS Notes:
UAV: Phoenix
Test Name:
Flight Test 5
DGPS confirmation through manual and automated flight

5 n/a HRF 25 November 2011


Changes since Flight Test 4 – 20 September 2011 flight:
The uBlox GPS has been disabled. The DGPS dummy weights were removed and
replaced with a NovAtel OEMV1-G rover and NovAtel antenna. The total mass, and
balance of the aircraft are as before.
The introduction of the differential GPS has introduced a large number of changes to
the way GPS is processed by the avionics. It has been tested in practical ground tests
as well as HIL simulations. This flight test will confirm that the DGPS is functioning
properly and that the EKF performs well. Automated flights will then be conducted.
Restrictions:
(a) Page 1 of 4
Take-Off # 1
1 This flight will be RC only since it is the first flight with The aircraft will be under safety pilot control for the Before takeoff the aircraft must be positioned accurately on
the NovAtel DGPS system. entire test. the markings on the runway
The main goals are to confirm: No aggressive manoeuvres.

that the airframe is still in good flying Since no AP will be armed during this test the behaviour is
condition Pilot should take off and fly a relaxed circuit at dictated by the safety pilot.
a moderate airspeed.
avionics subsystems are fully operational in
flight conditions After 5 minutes of flight the test should be After landing the aircraft must once again be positioned
concluded with a landing as close as possible accurately on the markings on the runway to confirm the
that the DGPS system maintains the L1 to the take-off point. functioning of the estimator
integer solution for the entire flight envelope.
that the EKF is functional in flight conditions.
(b) Page 2 of 4
Figure D.7 – Flight test card - Flight Test 5 - 25 November

Take-Off # 2
2 The Navigation controllers will be armed if the Before takeoff the aircraft must be positioned accurately on
Safety Pilot takes off and engages autopilot on the the markings on the runway
NovAtel system performed reliably in the previous transmitter at the start of the FAR long leg with the
test: aircraft straight and level at around 60 m altitude Safety Pilot takes off and engages autopilot on the
Had continuous L1 Int fix transmitter at the start of the FAR long leg with the
Maintained well bounded sigmas (2cm for Safety Pilot does not have control over any actuators. aircraft straight and level at around 60 m altitude
lat/lon, 4cm for altitude).
Safety Pilot monitors aircraft for any strange behaviour After the autopilot has been armed the aircraft will begin to
and must retake control if airspeed or altitude become fly a navigation circuit at an altitude of 50 m and an
The aircraft will fly the navigation loop until the test too low. airspeed of 18 m/s
time runs out
The controller can only bank at 30 degrees, if a During this time the following steps will be performed.
During this time several airspeed and altitude steps steeper bank angle is judged to be occurring Safety Airspeed: 18 m/s 15m/s 18 m/s
will be performed. Pilot must retake control.
Altitude: 55m 50m 55m 50m
becomes disengaged for whatever reason. If this The Direct-Lift flaps controller will now be engaged and the
occurs the aircraft will keep flying straight and level following steps performed: Pay close attention to aircraft
indefinitely. Safety Pilot must retake control pitch behaviour at this point
immediately. Altitude: 55m 50m 55m 50m
The approximate location of the waypoints will be More steps will be performed should time remain.
NSA Controller Gains set at: discussed prior to flight. If the aircraft fails to turn when
a=4 expected safety pilot must retake control. For all runway legs, note the aircraft’s path offset from the
zeta = 0.707 middle of the runway and inform test leader
Nbar = 2 When the Direct-Lift flaps controller is armed the pitch
behaviour of the aircraft must be closely observed. Upon completion the Safety Pilot will be informed to retake
DLC Gains set at: Safety Pilot must retake control immediately if any control of the aircraft and land as close as possible to the
Kp = 0.02 oscillations are seen. take-off point.
CutoffFreq. = 4
CCF = 0.2 After landing the aircraft must once again be positioned
accurately on the markings on the runway to confirm the
ClimbRate Controller Gains set at: functioning of the estimator
Kp = -1.0
Altitude Controller Gains set at:

Kp = 0.5
(c) Page 3 of 4
Take-Off # 3
3 The Navigation + Landing Path controllers will be Before takeoff the aircraft must be positioned accurately on
armed if the : Safety Pilot takes off and engages autopilot on the the markings on the runway
aircraft straight and level at around 60 m altitude Safety Pilot takes off and engages autopilot on the
NovAtel system performed reliably in the previous
test:
Safety Pilot does not have control over any actuators. aircraft straight and level at around 60 m altitude
Had continuous L1 Int fix
Maintained well bounded sigmas (2cm for
Safety Pilot monitors aircraft for any strange behaviour After the autopilot has been armed the aircraft will begin to
lat/lon, 4cm for altitude). and must retake control if airspeed or altitude become fly a navigation circuit at an altitude of 50 m and an
too low. airspeed of 18 m/s
Analysis of the previous flight’s data showed proper
operation of the DLC controller
The controller can only bank at 30 degrees, if a During each runway leg the aircraft will descend down a
steeper bank angle is judged to be occurring Safety glide slope of 7 degrees initially and 3 degrees finally until
Pilot must retake control. it reaches an altitude of around 20 m. This will occur in the
The aircraft will fly the navigation loop until the test vicinity of the takeoff position markings.
time runs out Safety Pilot will be informed if the Navigation Controller
becomes disengaged for whatever reason. If this
During this time mock landing paths will be flown on occurs the aircraft will keep flying straight and level GlidePath origin set as: --20 m
the runway leg of the circuit. indefinitely. Safety Pilot must retake control
immediately.
NSA Controller Gains set at: For all runway legs, note the aircraft’s path offset from the
a=4 The approximate location of the waypoints will be middle of the runway and inform test leader
zeta = 0.707 discussed prior to flight. If the aircraft fails to turn when
Nbar = 2 expected safety pilot must retake control. Circuits will be flown until test time expires.
DLC Gains set at: Upon completion the Safety Pilot will be informed to retake
Kp = 0.02 control of the aircraft and land as close as possible to the
CutoffFreq. = 4 take-off point.
CCF = 0.2
After landing the aircraft must once again be positioned
ClimbRate Controller Gains set at: accurately on the markings on the runway to confirm the
Kp = -1.0 functioning of the estimator
Altitude Controller Gains set at:

Kp = 0.5
(d) Page 4 of 4

UAV: Phoenix
Test Name:
Flight Test 6
Landing Approach flights with the possibility of full landing



Changes since Flight Test 5 – November 25 2011 flight:
None
Restrictions:
(a) Page 1 of 2
Take-Off # 1
1 The Navigation + Landing controllers will be armed if Safety Pilot takes off and engages autopilot on the Before takeoff the aircraft must be positioned accurately on
the transmitter at the start of the FAR long leg with the the markings on the runway
NovAtel system performed reliably in the previous Safety Pilot takes off and engages autopilot on the
test: Safety Pilot does not have control over any actuators. transmitter at the start of the FAR long leg with the
Had continuous L1 Int fix aircraft straight and level at around 60 m altitude
Maintained well bounded sigmas (2cm for Safety Pilot monitors aircraft for any strange behaviour
lat/lon, 4cm for altitude). and must retake control if airspeed or altitude become After the autopilot has been armed the aircraft will begin to
too low. fly a navigation circuit at an altitude of 40 m and an
Analysis of the previous flight’s data showed good airspeed of 18 m/s
glide path following. The controller can only bank at 30 degrees, if a
steeper bank angle is judged to be occurring Safety During each runway leg the aircraft will descend down a
Pilot must retake control. glide slope of 7 degrees initially and 3 degrees finally until
The aircraft will fly the navigation loop until the test it reaches an altitude of around 10 m. This will occur in the
time runs out Safety Pilot will be informed if the Navigation Controller vicinity of the takeoff position markings.
becomes disengaged for whatever reason. If this
During this time mock landing paths will be flown on occurs the aircraft will keep flying straight and level GlidePath origin set as: --10 m
the runway leg of the circuit. indefinitely. Safety Pilot must retake control
immediately. If adjustments need to be made based on Safety Pilot
The height of the mock landings will be adjusted feedback the circuit is flown again and re-checked.
downward for a low level flyby. During this flyby the The approximate location of the waypoints will be
pitch angle, descent rate, roll angle, offset from the discussed prior to flight. If the aircraft fails to turn when
middle of the runway and environmental factors will expected safety pilot must retake control. If the feedback from the Safety Pilot is good, the approach
be judged by the Safety Pilot. height is lowered for the next circuit.
The height of the mock landings will be adjusted
A landing will only be attempted if the Safety Pilot downward for a low level flyby. During this flyby the GlidePath origin set as: --5 m
feels comfortable with the above factors. pitch angle, descent rate, roll angle, offset from the
middle of the runway and environmental factors must
NSA Controller Gains set at: be judged by the Safety Pilot. A landing will only be IF the Safety Pilot feels comfortable with a landing attempt:
a=4 attempted if the Safety Pilot feels comfortable with the
zeta = 0.707 above factors. During a landing the aircraft will NOT GlidePath origin set as: +1.3 m and +1.5m
Nbar = 2 flare, the Safety Pilot should therefore judge the above
parameters accordingly.
DLC Gains set at: During a landing the aircraft will NOT flare, it will strike the
DLC not enabled runway close to the takeoff markings.
During landing, the aircraft will strike the runway at the
takeoff markings.
ClimbRate Controller Gains set at: The instant that the aircraft strikes the runway the
Kp = -1.0 Safety Pilot must retake control.
The instant that the aircraft strikes the runway the
Safety Pilot must retake control. The Safety Pilot should then either complete the landing, or
Altitude Controller Gains set at: take off again if landing is not feasible.
Kp = 0.5 The Safety Pilot should then either complete the
landing, or take off again if landing is not feasible.
(b) Page 2 of 2

UAV: Phoenix
Test Name:
Flight Test 7
Autonomous Landing


No changes since Flight Test 6 – 28 November 2011.
Restrictions:
(a) Page 1 of 2
Take-Off # 1
1 During the previous flight test a successful Safety Pilot takes off and engages autopilot on the Before takeoff the aircraft must be positioned accurately on
autonomous landing was performed. During this test transmitter at the start of the FAR long leg with the the markings on the runway
several more such landings will be attempted. aircraft straight and level at around 60 m altitude
The aircraft will fly the navigation loop and perform Safety Pilot does not have control over any actuators. transmitter at the start of the FAR long leg with the
mock landing flybys until either the test time runs out, aircraft straight and level at around 60 m altitude
or an autonomous landing is performed. Safety Pilot monitors aircraft for any strange behaviour
and must retake control if airspeed or altitude become After the autopilot has been armed the aircraft will begin to
The height of the mock landings will be adjusted too low. fly a navigation circuit at an altitude of 40 m and an
downward for a low level flyby. During this flyby the airspeed of either 18 m/s or 16 m/s
pitch angle, descent rate, roll angle, offset from the The controller can only bank at 30 degrees, if a
middle of the runway and environmental factors will steeper bank angle is judged to be occurring Safety During each runway leg the aircraft will descend down a
be judged by the Safety Pilot. Pilot must retake control. glide slope of 7 degrees initially and 3 degrees finally until
it reaches the GlidePath origin height. This will occur in
A landing will only be attempted if the Safety Pilot Safety Pilot will be informed if the Navigation Controller the vicinity of the takeoff position markings.
feels comfortable with the above factors. becomes disengaged for whatever reason. If this
occurs the aircraft will keep flying straight and level The GlidePath origin set as: --20 m or -- 5 m
indefinitely. Safety Pilot must retake control
immediately. The aircraft will perform a low level flyby of the runway.
a=4
zeta = 0.707
The approximate location of the waypoints will be If adjustments need to be made based on Safety Pilot
Nbar = 2
discussed prior to flight. If the aircraft fails to turn when feedback the circuit is flown again and re-checked.
expected safety pilot must retake control.
DLC Gains set at:
Kp = 0.02
The height of the mock landings will be adjusted IF the Safety Pilot feels comfortable with a landing attempt:
CutoffFreq. = 4
CCF = 0.2 downward for a low level flyby. During this flyby the
pitch angle, descent rate, roll angle, offset from the GlidePath origin set as: +1.5 m
middle of the runway and environmental factors must This will cause the aircraft to land approximately 20m from
be judged by the Safety Pilot. A landing will only be the takeoff markings
Kp = -1.0
attempted if the Safety Pilot feels comfortable with the
above factors. GlidePath origin set as: +2 m
Altitude Controller Gains set at: This will cause the aircraft to strike the runway close to the
Kp = 0.5 takeoff markings.
.
Safety Pilot must retake control.
Safety Pilot must retake control.
The Safety Pilot should then either complete the The Safety Pilot should then either complete the landing, or
landing, or take off again if landing is not feasible. take off again if landing is not feasible.
(b) Page 2 of 2

Bibliography
[1] Gaum, D.R.: Aggressive Flight Control Techniques for a Fixed-Wing Unmanned Aer-
ial Vehicle. Master’s thesis, Stellenbosch University, 2009.
[2] Peddle, I.K.: Acceleration Based Manoeuvre Flight Control System for Unmanned
Aerial Vehicles. Ph.D. thesis, Stellenbosch University, 2008.
[3] Park, S.: Avionics and Control System Development for Mid-Air Rendezvous of Two
Unmanned Aerial Vehicles. Ph.D. thesis, Massachusetts Institute of Technology,
2004.
[4] Peddle, I.K.: Autonomous Flight of a Model Aircraft. Master’s thesis, Stellenbosch
University, 2005.
[5] Roos, J.-C.: Autonomous Take-Off and Landing of a Fixed Wing Unmanned Aerial
Vehicle. Master’s thesis, Stellenbosch University, 2007.
[6] Hough, W.: Autonomous Aerobatic Flight of a Fixed Wing Unmanned Aerial Vehicle.
Master’s thesis, Stellenbosch University, 2007.
[7] Visser, B.J.: Die Presisie Landing van ’n Onbemande Vliegtuig “The Precision Land-
ing of an Unmanned Aircraft”. Master’s thesis, Stellenbosch University, 2008.
[8] Blaauw, D.: Flight Control System for a Variable Stability Blended-Wing-Body Un-
manned Aerial Vehicle. Master’s thesis, Stellenbosch University, 2009.
[9] de Hart, R.D.: Advanced Take-Off and Flight Control Algorithms for Fixed Wing
Unmanned Aerial Vehicles. Master’s thesis, Stellenbosch University, 2010.
[10] Basson, W.A.: Fault Tolerant Adaptive Control of an Unmanned Aerial Vehicle. Mas-
ter’s thesis, Stellenbosch University, 2011.
[11] Jeffrey, C.: An Introduction to GNSS GPS, GLONASS, Galileo and other Global Nav-
igation Satellite Systems. First edition edn. NovAtel Inc., 2010.
[12] Pinsker, W.J.G.: The control characteristics of aircraft employing direct-lift control.
Tech. Rep., Aerodynamics Dept., R.A.E., Bedford, 1970.
[13] Cook, M.V.: Flight Dynamics Principles. 2nd edn. Elsevier Ltd., 2007.
170
BIBLIOGRAPHY 171
[14] Etkin, B. and Reid, L.D.: Dynamics of Flight Stability and Control. 3rd edn. John
Wiley & Sons, Inc., 1996.
[15] Zipfel, P.H.: Modeling and Simulation of Aerospace Vehicle Dynamics. American
Institute of Aeronautics and Astronautics Inc., 2000.
[16] Koen, M.: Modelling and simulation of an rpv for flight control system design pur-
poses. Department of Electronic and Computer Engineering, University of Pretoria,
2006.
[17] Blakelock, J.: Automatic Control of Aircraft and Missiles. 2nd edn. JohnWiley &
Sons, 1991.
[18] Gerrits, M.: Direct Lift Control for the Cessna Citation II. Master’s thesis, Eind-
hoven University of Technology, 1994.
[19] Fitzgerald, P.: Flight Control System Design for Autonomous UAV Landing. Ph.D.
thesis, Cranfield University, 2004.
[20] Airplane Flying Handbook. U.S. Department of Transportation, Federal Aviation

Adminstration, Airman Testing Standards Branch, 2004.
[21] Smit, P.E.: Development of a 3-DOF Motion Simulation Platform. Master’s thesis,
Stellenbosch University, 2010.
[22] de Jager, A.M.: The design and implementation of vision-based autonomous rotor-
craft landing. Master’s thesis, Stellenbosch University, 2011.

Alberts Accurate 2012

Uploaded by

Copyright:

Available Formats

Alberts Accurate 2012

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Alberts Accurate 2012

Uploaded by

Copyright:

Available Formats

Accurate Autonomous Landing of a Fixed-Wing

Unmanned Aerial Vehicle

Frederik Nicolaas Alberts

Thesis presented in partial fulfilment of the requirements for the degree of

Master of Science in Engineering

in the Faculty of Engineering at Stellenbosch University

Department Electrical and Electronic Engineering

Copyright © 2012 Stellenbosch University

Hierdie tesis verteenwoordig die analise, ontwerp, simulasie en praktiese implemente-

Die beheerders en landingsprosedure is in ’n hardeware-in-die-lus omgewing gesimu-

List of Tables xiii

2 Aircraft Model Description 6

2.3.3.2 Thrust Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Longitudinal Analysis and Control 23

4 Lateral Analysis and Control 71

4.3.3 HIL Simulation and Practical Flight Test Results . . . . . . . . . . . . 80

A Additional Mathematical Principles [1] 119

B UAV System Description 122

B.6.3 Aerodynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

C Detailed Derivations 133

D Additional Flight Test Information 155

1.1 UAV photo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Inertial and runway axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.1 Longitudinal control system overview . . . . . . . . . . . . . . . . . . . . . . . . 23

4.1 Longitudinal Control System Overview Block Diagram . . . . . . . . . . . . . . 71

5.1 Standard rectangular airfield traffic pattern . . . . . . . . . . . . . . . . . . . . 86

B.1 Aircraft photo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

C.1 Control system block diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

3.1 Longitudinal stability and control derivatives . . . . . . . . . . . . . . . . . . . 29

B.1 Actuator and controller limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

Abbreviations and Acronyms

AVL Athena Vortex Lattice

CAN Controller Area Network

DCM Direction Cosine Matrix

DGPS Differential Global Positioning System

DLC Direct-Lift Control

GPS Global Positioning System

IMU Inertial Measurement Unit

LQR Linear Quadratic Regulator

LSA Lateral Specific Acceleration

MIMO Multi-Input, Multi-Output

NMP Non-Minimum Phase

NSA Normal Specific Acceleration

OBC On-Board Computer

RTK Real Time Kinematic

UAV Unmanned Aerial Vehicle

Px ,Py ,Pz Position vector components in the X ,Y and Z axes

V̄ ,α,β Airspeed magnitude, angle of attack, angle of sideslip

ψ ,θ,φ Yaw, pitch and roll angles

P ,Q,R Roll, pitch and yaw rates

L,M ,N Rolling, pitching and yawing moments

X ,Y ,Z Roll, pitch and yaw rates

P ,Q,R X ,Y and Z -axis forces

A,B ,C Specific accelerations in the X ,Y and Z -axis directions

Vectors and Matrices

DCM Direction cosine matrix

F,M Force and moment vectors

G Gravitational acceleration vector

H Angular momentum vector