ROBUST CONTROLLER DESIGN FOR A FIXED WING UAV

A THESIS SUBMITTED TO
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES
OF
MIDDLE EAST TECHNICAL UNIVERSITY

BY

ANNA PRACH

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR
THE DEGREE OF MASTER OF SCIENCE
IN
AEROSPACE ENGINEERING

SEPTEMBER 2009
 Approval of the thesis:

ROBUST CONTROLLER DESIGN FOR A FIXED WING UAV

submitted by ANNA PRACH in partial fulfillment of the requirements for the

degree of Master of Science in Aerospace Engineering Department, Middle East
Technical University by,

Prof. Dr. Canan Özgen _______________

Dean, Gradaute School of Natural and Applied Sciences

Prof. Dr. Ismail H. Tuncer _______________

Head of Department, Aerospace Engineering

Prof. Dr. Nafiz Alemdaroğlu _______________

Supervisor, Aerospace Engineering Dept., METU

Examining Committee Members:

Assoc. Prof. Dr. Serkan Özgen _______________

Aerospace Engineering Dept., METU

Prof. Dr. Nafiz Alemdaroğlu _______________

Aerospace Engineering Dept., METU

Assoc. Prof. Dr. Altan Kayran _______________

Aerospace Engineering Dept., METU

Prof. Dr Kemal Leblebicioğlu _______________

Electrical and Electronics Engineering Dept., METU

Dr. Volkan Nalbantoğlu _______________

ASELSAN

Date: _______________
I hereby declare that all information in this document has been obtained and
presented in accordance with academic rules and ethical conduct. I also declare
that, as required by these rules and conduct, I have fully cited and referenced
all material and results that are not original to this work.

Name, Last name: Anna Prach

Signature :

iii
 ABSTRACT

ROBUST CONTROLLER DESIGN FOR A FIXED WING UAV

Prach, Anna
M.S., Department of Aerospace Engineering
Supervisor: Prof. Nafiz Alemdaroğlu

September 2009, 148 pages

This study describes the design and implementation of the pitch and roll autopilots
for a fixed wing unmanned vehicle. A Tactical Unmanned Aerial Vehicle (TUAV),
which is designed at the Middle East Technical University (METU), is used as a
platform. This work combines development of the classical and robust controllers,
which are used for the pitch and roll autopilots. One of the important steps in the
thesis is development of the non-linear dynamic model of the UAV, which is
developed in MATLAB/Simulink environment. Two different strategies of the
controller design imply development of the PID and controllers. Simulation
results illustrate the performances of the designed controllers. Simulation is
performed for the nominal model of the UAV and for the model that includes
uncertainties and sensor noises.

Keywords: PID Controller, Robust Controller, Autopilot

iv
 ÖZ

SABİT KANATLI BİR İHA İÇİN GÜRBÜZ KONTROLCÜ TASARIMI

Prach, Anna
Yüksek Lisans, Havacılık ve Uzay Mühendisliği
Tez Yöneticisi: [Link] Alemdaroğlu

Eylül 2009, 148 sayfa

Bu tez sabit kanatlı bir Insansız Hava Aracının (IHA) yunuslama ve yalpalama
otopilotu tasarımının tasarım ve uygulanmasını anlatmaktadır. ODTU’de tasarlanan
bir taktik İHA (TİHA) uygulama platformu olarak kullanılmaktadır. Bu çalışma
yunuslama ve yalpalama harekteleri için klasik ve gürbüz kontrolcü tasarımlarını
içermektedir. Bu tezde aynı zamanda IHA’nın doğrusal olmayan dinamik modeli
MATLAB/SIMULINK ortamında geliştirilmiştir. Bu kapsamda PID ve

kontrolcüleri tasarlanmıştır. Benzetim sonuçları tasarlanan kontrolcülerin

performansını göstermektedir. Benzetimler hem nominal IHA modeli hem de
belirsizlik ve sensör gürültüsü içeren modeller üzerinde denenmiştir.

Anahtar Kelimeler: PID Kontrolü, Gürbüz Kontrol, Otopilot

v
In Living Memory of M. J. …

vi
 ACKNOWLEDGMENTS

I would like to express gratitude to my advisor, Prof. Dr. Nafiz Alemdaroğlu, for his
valuable support and encouragement. Thanks to him I got an opportunity to start my
master program at this university. I am very thankful to Dr. Volkan Nalbantoğlu, for
his continuous guidance and supervision throughout this work. I want to express
special gratitude to Dr. Ilkay Yavrucuk for his valuable suggestions and help that
enabled me to complete my work successfully.
Lastly, I want to express my love and gratitude to my family for understanding and
love through the duration of my studies.

vii
 TABLE OF CONTENTS

ABSTRACT................................................................................................................iv

ÖZ.................................................................................................................................v

ACKNOWLEDGMENTS..........................................................................................vii

TABLE OF CONTENTS...........................................................................................vii

LIST OF TABLES………………………………………………………………….xiv

LIST OF FIGURES………………………………………………………………....xv

LIST OF SYMBOLS………………………………………………………………xxi

CHAPTERS

1. INTRODUCTION........................................................................................1

2. 6-DOF MATHEMATICAL MODELING...................................................5

2.1 Platform...............................................................................................5

2.2 Reference Frames and Coordinate Systems.........................................8

2.3 6-DOF Equations of Motion................................................................9

2.4 Aerodynamic Forces and Moments...................................................10

2.4.1 Longitudinal Aerodynamic Forces and Moments...................11

2.4.2 Lateral-Directional Aerodynamic Forces and Moments.........14

2.5 Mass-Inertia Model............................................................................17

2.6 Propulsion Model...............................................................................17

2.7 Actuators Models...............................................................................19

viii
 2.8 Atmosphere Model.............................................................................20

2.9 Gravity Model....................................................................................20

2.10 Non-Linear Simulation......................................................................21

2.11 Linearization of Equations of Motion and State Space

Representation....................................................................................22

2.11.1 Longitudinal Dynamics...........................................................23

2.11.2 Lateral Dynamics....................................................................25

2.12 Open-Loop Simulation.......................................................................28

3. CLASSICAL CONTROLLER DESIGN....................................................33

3.1 PID Controller Background...............................................................34

3.1.1 PID Controller Tuning Rules..................................................37

3.2 Control Problem Definition...............................................................38

3.3 Sensors...............................................................................................39

3.4 Classical Controller Design for METU TUAV: Pitch Attitude

Controller...........................................................................................40

3.4.1 Controller Configuration........................................................41

3.4.2 Determining Controller Parameters and Controller

Performance............................................................................42

3.5 Classical Controller Design for METU TUAV: Roll Autopilot........47

3.5.1 Controller Configuration.........................................................47

3.5.2 Determining Controller Parameters and Controller

Performance............................................................................48

ix
 3.6 Command Filter Implementation for a Control System....................51

3.6.1 Command Filter......................................................................51

3.6.2 Pitch Autopilot Performance with Command Filter

Implementation...................................................................................52

3.6.3 Roll Autopilot Performance with Command Filter

Implementation...................................................................................55

4. ROBUST CONTROLLER DESIGN..........................................................58

4.1 Historical Review...............................................................................58

4.2 Theoretical Background.....................................................................59

4.2.1 Problem Formulation..............................................................59

4.2.2 Signals and Systems Norms....................................................60

4.2.3 Modeling Uncertainty.............................................................62

[Link] Additive Uncertainties...................................................63

[Link] Multiplicative Uncertainties...........................................63

[Link] Parametric Uncertainties................................................64

4.2.4 Linear Fractional Transformation...........................................65

4.2.5 Transfer Functions of a Closed Loop System.........................67

4.2.6 Stabilization and Performance Requirements (SISO

Systems)…………………………………………….……….69

[Link] Nominal Stability (NS)..................................................69

[Link] Nominal Performance (NP)...........................................70

[Link] Robust Stability (RS).....................................................70

x
 [Link] Robust Performance (RP)..............................................71

4.2.7 Stabilization and Performance Requirements (MIMO

Systems).................................................................................72

[Link] Nominal Stability (NS)..................................................73

[Link] Nominal Performance (NP)...........................................73

[Link] Robust Stability (RS).....................................................73

[Link] Robust Performance (RP)..............................................74

4.3 Design........................................................................................74

4.4 Analyses……………………........................................................76

4.4.1 Definition and Properties of Structured Singular Value.........76

4.4.2 Robust Stability Condition.....................................................79

4.4.3 Robust Performance Condition...............................................79

4.5 Robust Controller Design for METU TUAV....................................82

4.5.1 Choice of Weight Functions for Robust Controller................82

4.5.2 Robust Controller Design for Pitch Attitude Autopilot..........84

[Link] Performance of the Robust Controller for Pitch Attitude

Autopilot........................................................................88

[Link] Robust Performance Analysis………………….……..90

[Link] Robust Performance Analysis: Connection with

Analysis….....................................................................92

4.5.3 Robust Controller Design for Roll Autopilot.........................95

xi
 [Link] Performance of the Robust Controller for Roll

Autopilot........................................................................98

[Link] Robust Performance Analysis………………………..101

[Link] Robust Performance Analysis: Connection with

Analysis…………………………………….………...102

4.6 Command Filter Implementation for a Robust Control

System..............................................................................................103

4.6.1 Robust Pitch Autopilot Performance with Command Filter

Implementation.....................................................................103

4.6.2 Robust Roll Autopilot Performance with Command Filter

Implementation.....................................................................107

5. COMPARISON OF PID AND ROBUST CONTROLLER.....................110

5.1 Performance Issues for the Nominal Model....................................110

5.1.1 Pitch Autopilot......................................................................110

5.1.2 Roll Autopilot.......................................................................112

5.2 Performance Issues for the Nominal Model with a Command Filter

Implementation...............................................................................115

5.2.1 Pitch Autopilot......................................................................115

5.2.2 Roll Autopilot.......................................................................117

5.3 Performance Comparison of PID and Robust Controllers for

Uncertain Plant.................................................................................119

xii
 5.4 Performance Comparison of PID and Robust Controllers with a

Sensor Noise Presence.....................................................................123

5.4.1 Pitch Autopilot...................................................................123

5.4.2 Roll Autopilot.......................................................................152

6. CONCLUSIONS......................................................................................126

REFERENCES.........................................................................................................130

APPENDICES

A AERODYNAMIC COEFFICIENTS AND DERIVATIVES...............................135

B LINEARIZATION OF EQUATIONS OF MOTION...........................................143

xiii
 LIST OF TABLES

TABLES

Table 2.1 Technical Specifications of the METU Tactical UAV System............ 7

Table 2.2 Mass and Inertia Properties of the METU TUAV................................ 17

Table 2.3 Longitudinal Modes Characteristics..................................................... 25

Table 2.4 Dutch-roll Mode Characteristics........................................................... 27

Table 2.5 Roll and Spiral Mode Characteristics................................................... 27

Table 4.1 Lower and Upper Bounds and Critical Frequency................................ 95

Table 4.2 Lower and Upper Bounds and Critical Frequency................................ 103

Table 5.1 Performance Comparison of Controllers for Pitch Autopilot............... 112

Table 5.2 Performance Comparison of Controllers for Roll Autopilot................ 114

xiv
 LIST OF FIGURES

FIGURES

Figure 2.1 METU Tactical UAV............................................................................ 7

Figure 2.2 Longitudinal Forces and Moment in Stability Axes.............................. 12

Figure 2.3 Lateral-Directional Force and Moments in Stability Axes.................... 15

Figure 2.4 Limbach L 275 E................................................................................... 18

Figure 2.5 Power, Torque and SFC vs. RPM of Limbach L 275 E........................ 19

Figure 2.6 Simulink Block Diagram of the Nonlinear System............................... 22

Figure 2.7 System Response to Elevator Input: x-body Velocity........................... 29

Figure 2.8 System Response to Elevator Input: z-body Velocity........................... 29

Figure 2.9 System Response to Elevator Input: Pitch Angle.................................. 30

Figure 2.10 System Response to Elevator Input: Pitch Rate.................................. 30

Figure 2.11 System Response to Ailerons Input: y-body Velocity...................... 31

Figure 2.12 System Response to Ailerons Input: Roll Angle............................... 31

Figure 2.13 System Response to Ailerons Input: Roll Rate................................. 32

Figure 2.14 System Response to Ailerons Input: Yaw Rate................................... 32

Figure 2.15 System Response to Ailerons Step Input............................................. 33

Figure 3.1 Standard Feedback Configuration......................................................... 35

Figure 3.2 Block Diagram of a PID Controller....................................................... 36

xv
Figure 3.3 Basic Controller Configuration............................................................. 41

Figure 3.4 S-shaped Response Curve..................................................................... 43

Figure 3.5 Bode Plot of the Open-Loop System..................................................... 44

Figure 3.6 System Response to a Unit Step Input: Pitch Angle............................. 45

Figure 3.7 Elevator Deflection for a Unit Step Input.............................................. 46

Figure 3.8 Elevator Deflection Rate for a Unit Step Input..................................... 46

Figure 3.9 Bode Plot of the Open-Loop System..................................................... 49

Figure 3.10 System Response to a Unit Step Input: Roll Angle............................. 49

Figure 3.11 Ailerons Deflection for a Unit Step Input........................................... 50

Figure 3.12 Ailerons Deflection Rate for a Unit Step Input................................... 50

Figure 3.13 Command Filter................................................................................... 51

Figure 3.14 Basic Controller Configuration with Command Filter........................ 52

Figure 3.15 Pitch Angle Command: Nominal vs. Filtered..................................... 53

Figure 3.16 System Response to Pitch Angle Command....................................... 54

Figure 3.17 Elevator Deflection.............................................................................. 54

Figure 3.18 Elevator Deflection Rate..................................................................... 55

Figure 3.19 Roll Angle Command: Nominal vs. Filtered....................................... 56

Figure 3.20 System Response to a Command Roll Angle...................................... 56

Figure 3.21 Ailerons Deflection............................................................................. 57

Figure 3.22 Ailerons Deflection Rate..................................................................... 57

Figure 4.1 Block Diagram....................................................................................... 60

xvi
Figure 4.2 Schematic Block Diagram of an LTI System........................................ 60

Figure 4.3 Additive Uncertainty Configuration...................................................... 63

Figure 4.4 Multiplicative Uncertainty Configuration............................................. 64

Figure 4.5 Typical Behavior of Multiplicative Uncertainty................................... 65

Figure 4.6 Lower Linear Fractional Transformation.............................................. 66

Figure 4.7 Upper Liner Fractional Transformation................................................ 67

Figure 4.8 Feedback Configuration........................................................................ 67

Figure 4.9 System Block Diagram.......................................................................... 69

Figure 4.10 Structure................................................................................ 71

Figure 4.11 Plant Controller Configuration Set Up................................................ 73

Figure 4.12 Standard configuration with diagonal structure of .......... 77

Figure 4.13 Modified Structure................................................................ 78

Figure 4.14 Generalized Plant................................................................................. 80

Figure 4.15 Generalized Plant with Controller....................................................... 80

Figure 4.16 Uncertain Plant.................................................................................... 81

Figure 4.17 Uncertain Plant with block........................................................... 81

Figure 4.18 Block Diagram of Closed-loop System.............................................. 83

Figure 4.19 Bode Plot of Uncertain Plants............................................................. 85

Figure 4.20 Bode Plot of ................................................................................ 86

Figure 4.21 Bode plot of .................................................................................. 87

Figure 4.22 Bode Plot of .................................................................................. 88

xvii
Figure 4.23 Step Input Response of the Nominal Plant: Pitch Angle..................... 89

Figure 4.24 Elevator Deflection.............................................................................. 89

Figure 4.25 Elevator Deflection Rate..................................................................... 90

Figure 4.26 Generic Tradeoff Between Uncertainty Level and Performance........ 91

Figure 4.27 Bounds for Robust Performance................................................... 93

Figure 4.28 Bounds for Robust Performance................................................... 94

Figure 4.29 Bode Plot of Uncertain Plants............................................................. 96

Figure 4.30 Bode Plot of ................................................................................. 97

Figure 4.31 Bode Plot of ................................................................................ 98

Figure 4.32 Step Input Response of the Nominal Plant: Roll Angle...................... 99

Figure 4.33 Ailerons Deflection............................................................................. 100

Figure 4.34 Ailerons Deflection Rate..................................................................... 100

Figure 4.35 Bounds for Robust Performance................................................... 101

Figure 4.36 Bounds for Robust Performance................................................... 102

Figure 4.37 Basic Controller Configuration with Command Filter........................ 104

Figure 4.38 Pitch Angle Command........................................................................ 105

Figure 4.39 System Response to a Pitch Angle Command................................... 105

Figure 4.40 Elevator Deflection............................................................................. 106

Figure 4.41 Elevator Deflection Rate..................................................................... 106

Figure 4.42 Roll Angle Command.......................................................................... 107

Figure 4.43 System Response to a Roll Command................................................. 108

xviii
Figure 4.44 Ailerons Deflection............................................................................. 108

Figure 4.45 Ailerons Deflection Rate..................................................................... 109

Figure 5.1 Pitch Angle Response to a Unit Step Input........................................... 111

Figure 5.2 Elevator Deflection for a Unit Step Input............................................. 111

Figure 5.3 Elevator Deflection Rate for a Unit Step Input..................................... 112

Figure 5.4 Roll Angle Response to a Unit Step Input............................................. 113

Figure 5.5 Ailerons Deflection for a Unit Step Input............................................. 114

Figure 5.6 Ailerons Deflection Rate for a Unit Step Input..................................... 115

Figure 5.7 Pitch Angle Response............................................................................ 116

Figure 5.8 Elevator Deflection for the Command Pitch Angle.............................. 116

Figure 5.9 Elevator Deflection Rate for the Command Pitch Angle...................... 117

Figure 5.10 Roll Angle Response........................................................................... 118

Figure 5.11 Ailerons Deflection for the Command Roll Angle.............................. 118

Figure 5.12 Ailerons Deflection Rate for the Command Roll Angle..................... 119

Figure 5.13 Pitch Angle Step Response of Uncertain Model................................. 120

Figure 5.14 Elevator Deflection of Uncertain Model............................................. 120

Figure 5.15 Elevator Deflection Rate of Uncertain Model..................................... 121

Figure 5.16 Roll Angle Step Response of Uncertain Model.................................. 121

Figure 5.17 Ailerons Deflection of Uncertain Model............................................. 122

Figure 5.18 Ailerons Deflection Rate of Uncertain Model.................................... 122

Figure 5.19 Pitch Angle Response ......................................................................... 124

xix
Figure 5.20 Elevator Deflection Response ............................................................ 124

Figure 5.21 Elevator Deflection Rate Response .................................................... 125

Figure 5.22 Roll Angle Response .......................................................................... 126

Figure 5.23 Ailerons Deflection Response ............................................................ 126

Figure 5.24 Ailerons Deflection Rate Response .................................................... 127

Figure A-1. vs. ............................................................................................ 136

Figure A-2. vs. ............................................................................................ 136

Figure A-3. vs. for Different Elevator Positions........................................ 137

Figure A-4. vs. ............................................................................................ 138

Figure A-5. vs. for Different Rudder Positions.......................................... 138

Figure A-6. vs. ............................................................................................. 139

Figure A-7. vs. for Different Ailerons Positions......................................... 140

Figure A-8. vs. ............................................................................................ 141

Figure A-9. vs. for Different Ailerons Positions........................................ 141

Figure A-10. vs. for Different Rudder Positions........................................ 142

xx
 LIST OF SYMBOLS

ROMAN SYMBOL GREEK SYMBOLS

Drag coefficient Angle of attack

Lift coefficient Sideslip angle

Rolling moment coefficient Roll angle

Pitching moment coefficient Yaw angle

Yawing moment coefficient Density

Power coefficient Pitch angle

Thrust coefficient Ailerons deflection angle

FC Fuel consumption Elevator deflection angle

I Inertia Rudder deflection angle

J Advanced ratio Angular velocity

m Mass
n Revolution per second ABBREVIATION DEFINITIONS
p Body roll rate LFT Linear Fractional Transformations
q Body pitch rate MIMO Multi Input Multi Output
r Body yaw rate NP Nominal Performance
u Body velocity in x-direction NS Nominal Stability
v Body velocity in y-direction RP Robust Performance
w Body velocity in z-direction RS Robust Stability
X A/C position in north direction RC Robust Controller
Y A/C position in east direction SISO Single Input Single Output
Z A/C altitude PIDC Proportional-Derivative- Integral
Controller

xxi
 CHAPTER 1

INTRODUCTION

Unmanned Aerial Vehicles (UAVs) are one of the fastest growing sectors of the
world’s aerospace industry. High demand for UAVs in the commercial industry
arises by taking into account the advantages of UAVs over manned aircraft: low
manufacturing and operational costs of the systems; elimination of the aircrew and
related life support systems allows UAVs to be smaller than manned aircraft [1] and
eliminates the risk of human lifes in difficult and dirty missions [2]. Removal of
aircrew and their support systems means that UAVs are capable of flying longer,
higher and faster without endangering lives. Furthermore, they promise better cost-
effectiveness and greater utility than manned aircraft [3]. Some of the military and
civil applications of UAVs are listed below [2].
Military applications: reconnaissance surveillance and target acquisition (RSTA),
surveillance for peacetime and combat Synthetic Aperture Radar (SAR), deception
operations, maritime operations (naval fire support, over the horizon targeting, anti
ship missile defense, ship classification), UCAV (Unmanned combat air vehicle)
applications, meteorology missions, route and landing reconnaissance support,
electronic warfare (EW) and SIGINT (SIGnals INTelligence), radio and data relay.
Civil applications: security, search and rescue, wild fire suppression, topographical
material and agriculture, communications relay, disaster and emergency
management, industrial applications, etc.
Unmanned Aerial Vehicles rely on two basic approaches in implementing unmаnned
flight-autonomy аnd pilot-in-the-looр - which rely predominantly on microрroсessor

1
and communication technologies, respectively. Both are used in differing lеvеls in
UАV’s fields, together they compensate the absence of an onboard pilot and thus
enable unmanned flight [4]. Autonomy of UAVs requires efficient and accurate
control systems. One of the most important steps in control system design is the
development of the aircraft model. The full model of an aircraft includes models of
its subsystems such as: aerodynamics, propulsion, actuators. Development of the
dynamic model of a UAV is of primary importance. For an accurate dynamic model
it is necessary to have accurate data of UAVs’ aerodynamics, which can be obtained
by various means such as wind tunnel tests, CFD analyses or calculated by empirical
formulas [5]. Physically an aircraft is a complicated system, which includes variation
of its aerodynamic coefficients, sensor noise, and limited information of the system.
Controllers designed with PID methods are unable to take into account errors in the
system modeling and signal uncertainties. In cases, when it is impossible to obtain an
accurate aerodynamic data for dynamical modeling of a UAV or there are
simplifications used in the modelling (linearization of nonlinear dynamics), classical
controller, implemented to the actual physical system, might not show the desired
performance. Robust control theory provides methods for designing controllers that
would produce accurate and fast response in the face of uncertainties or disturbances
in the plant model.
The main objective of this thesis is the development, implementation and comparison
of performances for pitch attitude and roll autopilots for the simulation of the METU
TUAV, designed by means of classical and robust control theories. The problem for
a controller design for the METU TUAV is the absence of an accurate data of the
UAV’s aerodynamics. In [6] aerodynamic coefficients and derivatives are obtained
by empirical formulas. [7] provides the aerodynamic data obtained by wind tunnel
test for a very similar UAV SCAUT, designed and developed at Politecnico Di
Torino, which has similar geometrical properties to the METU TUAV. The purpose
of this thesis is the comparison of performances of the PID controller and robust
controller. Simulations are performed for the nominal model of the UAV and the

2
models, which include uncertainties and sensor noise. Uncertainties are assumed to
be present in the aerodynamic coefficients of the UAV.
In the literature, many different approaches can be seen related to the autonomous
control of UAVs. In [8] Andrievsky B.R [Link]. propose an algorithm of a combined
adaptive controller, which uses both adaptation methods based on the parameter
identification and variable-structure control technique. In [9] authors introduce flight
control system is based on PID controller and uses gain scheduling algorithm based
on the airspeed to improve performance of the controller. In [10] PID controller is
used for autonomous landing for the METU TUAV. Model inversion is used in
landing algorithm. S. Kurnaz [Link]. [11] introduce fuzzy logic based control system
for an autonomous UAV system. Robust nonlinear controller design is employed in
[12] by M. Sadraey [Link]. [13] by Etkin B. describes design of a robust flight control
system for a mini-UAV using coupled stability derivatives.
This thesis is organized as follows: Chapter 2 includes the modelling of the UAV, its
mass-inertia properties. Dynamic model is based on the 6-DOF equations of motion.
Aerodynamic data is used from the reference [7]. Mass-inertia properties are taken
from [6]. UAV model also includes models of actuators, gravity and atmosphere.
Engine model is developed using the experimentally obtained data. Chapter 3
examines the basics of classical control theory, defines requirements to the control
system. It also covers the design of PID controllers for the pitch and roll autopilots
and simulation by a linear environment for a nominal model; command filter
implementation to the control systems. Theoretical background to the robust control
theory, definition of uncertainties, design techniques are reviewed in Chapter 4.

This chapter also contains the design of robust controllers for pitch and roll

autopilots, their simulation for a nominal model. Chapter 5 includes comparison of

PID and robust controllers performances by a nonlinear simulation. Simulation is
performed for both the nominal model and the model with uncertainties and sensor
noise. In Chapter 6 conclusions and recommendations for future work are presented.
The outline of the thesis can be expressed as follows:

3
• Development of the nonlinear dynamic model of the METU TUAV, its
linearization and investigation of the open-loop response.
• Design and implementation of pitch and roll autopilots based on PID
controllers.
• Design and implementation of pitch and roll autopilots, which use the
controllers based on robust technique.

• Performance comparison of PID and robust controllers.

4
 CHAPTER 2

6-DOF MATHEMATICAL MODELING

In this chapter development of a 6-DOF nonlinear model of the METU TUAV is

presented. The 6-DOF nonlinear model includes the aerodynamics, mass-inertia,
propulsion, atmospheric and gravity models. The chapter covers linearization of the
nonlinear model, decoupling of airplane dynamics into longitudinal and lateral
dynamics, their state space representations as well as the analysis of the open-loop
responses.

2.1 Platform

The platform used in this thesis is the METU tactical UAV system which is
developed with the initial funding provided from the State Planning Office (DPT) in
2004. The UAV is designed and manufactured in the Aerospace Engineering
Department of Middle East Technical University and is practically near its
completion. The first flight is foreseen within 2009 or early 2010. The UAV is
completely manufactured from composite materials, using fiber glass, carbon fiber
and Kevlar clothes as composite materials. The aircraft has a fuselage and a high
wing configuration and a twin boom T tail configuration. The booms are also
manufactured from composite material using filament winding technique. The basic
geometrical dimensions and the technical characteristics of the tactical UAV is given
in the following Table 2.1. The maximum take off weight of the UAV is estimated as
120 kg. However, this may be slightly less than this predicted value. The final weight
will be definite when the aircraft is fully completed. The total payload capacity of the

5
UAV is about 30 kg and is sufficient to carry a daylight and an IR camera mounted
in a gimbal system. Besides these Electro Optical (EO) payload system envisaged at
this stage other useful payload systems can easily be integrated into the UAV since
there is enough payload volume within the fuselage. The aerodynamical control
surfaces of the aircraft includes left and right ailerons, left and right flaps, left and
right rudders and an elevator. All these control surfaces are actuated by means of
servomotors with sufficient torque capacity. The take off and landing of the UAV is
foreseen to be conventional at this stage. However, in the near future this capability
will be augmented by incorporating catapult launch and parachute recovery systems
into the UAV. The UAV is equipped with steerable front and fixed main landing
gears having hydraulic breaking capability. All of the landing wheels are flight
qualified. The aircraft is a pusher configuration with the engine mounted at the rear
of the fuselage in between the tail booms. The engine used is gasoline powered
Limbach L 275 E engine producing 21 HP of output. It is a two stroke cycle opposed
twin cylinder engine (boxer engine) where each cylinder is operated by separate
carburators. Due to this fact each cylinder operates individually and in case of failure
of one of the cylinders the other cylinder produces enough power to propel the
aircraft. The preliminary ground tests of the engine and the propeller system shows
that a total thrust of about 40 kg is obtained from the engine at normal ground level
operations. This much thrust is more than enough for flying the UAV. The aircraft
can be operated remotely line of sight by radio control or by auto pilot.
Characteristics of the METU TUAV system are the following:
• Conventional take off and landing from runway
• Autonomous Flight
• 2 stroke cycle, 2 opposed cylinders, dual carburator IC engine delivering 21 HP
• Electro-optical and infrared camera inside a Gimbal system
• Real time video downlink to ground station
• Fully composite structure

6
 Figure 2.1 METU Tactical UAV

Table 2.1 Technical Specifications of the METU Tactical UAV System

Wing Span: 4.3 m

Length: 3m
Width of Fuselage: 0.3 m
Empty Weight: 90 kg
Maximum Take-off Weight: 120 kg
Maximum Payload Weight: 30 kg
Maximum Speed: 300 km/h
Stall Speed: 65 km/h
Cruise Speed: 145 km/h
Range: 600 km
Maximum Endurance: 5 hr
Operation Altitude: 3000 m
Payload: FLIR Camera
Propulsion: 21HP Two Cylinder Gasoline Engine

7
2.2. Reference Frames and Coordinate systems

The following frames and systems are of primary importance for the modeling of the
vehicle dynamics.
Inertial reference frame is fixed or in uniform rectilinear translation, relative to the
distant stars; in it Newton’s second law is valid for the motion of a particle [13]. If
the rotation of the Earth is neglected and the surface is assumed to be flat an Earth-
surface reference frame is considered as an inertial frame, with origin near the
vehicle if possible and with axis directed vertically down. x- and y-axis is the
local horizontal plane, x-axis points north and y-axis points east. Vehicle’s position
is calculated with respect to the inertial frame.
Navigation frame (north-east-down frame), also called as vehicle-carried frame, is
used the reference frame in which the origin is attached to the vehicle (usually at the
center of gravity), z-axis points down along the local gravity vector, x-axis is
directed north and y-axis is directed east.
If the aircraft is considered as a rigid body, it is represented by a frame, the so-called
body frame. Although this frame is not used as a reference, it is nevertheless
important [14]. The body coordinate system is used for deriving the equations of
motion of the vehicle. The origin of this coordinate system is at the center of gravity
of the airplane, x-axis points through the nose of the vehicle and lies with the
downward pointing z-axis in the plane of symmetry, y-axis, out the right wing,
completes the coordinate system.
Transformation from navigation frame to the body frame is composed of three
transformations, which are represented by so-called Euler angles: yaw, pitch and roll,
or , and . The first rotation is around the z-axis of the navigation frame by
angle , second rotation is around the y-axis of an intermediate frame, obtained
after the first rotation by the pitch angle and the last rotation is around the x-axis

of the coordinate system obtained after second rotation by angle . This

transformation is also called the direction cosine matrix.

8
The transformation matrix:

Wind coordinate system is frequently used for defining aerodynamic forces and
moments acting on the aircraft. x-axis is parallel to the free-stream velocity vector.
Transformation from the wind axis to body is composed of two rotations: by angle
around the z-axis of the wind coordinate system and rotation by angle
around y-axis such that it coincides with the body axis. Angles and are called
angle of attack and sideslip angle. The transformation matrix from the wind axis to
body:

2.3 6-DOF Equations of Motion

The airplane nonlinear equations of motion are developed by applying Newton’s

second law and law of conservation of linear momentum. The equations are written
in a body-fixed axis system, which moves with the airplane and which has

its origin at the airplane center of mass. The vehicle is assumed to be a rigid body
with six degrees of freedom. Three translational degrees describe the motion of the
center of mass and three attitude degrees describe the orientation of the vehicle.
Plane XZ is a plane of symmetry of the airplane, therefore .

The scalar force equations of motion in the airplane body-fixed axis system [15]:
Force along X:
(2.1a)

9
Force along Y:
(2.1b)
Force along Z:
(2.1c)
The scalar moment equations in the airplane body-fixed axis system:
Rolling moment about :

(2.2a)

Pitching moment about :

(2.2b)

Yawing moment about :

(2.2c)

Kinematic equations:
Roll rate about :

(2.3a)
Pitch rate about :

(2.3b)
Yaw rate about :

(2.3c)
In scalar force equations of motion forces X, Y and Z include aerodynamic forces
and propulsion forces.

2.4 Aerodynamic Forces and Moments

Aerodynamic forces and moments can be determined in two ways: by experimental

methods (flight test or wind tunnel test), by computational and/or empirical methods.

10
Experimental methods have a great advantage of allowing rather accurate predictions
of full-scale airplane aerodynamic behavior over a wide range of flight conditions,
including nonlinear effects. A disadvantage of experimental methods is that their cost
is very high, both in time and in money. Therefore, in most preliminary design and
parametric design studies theoretical and empirical methods are used [15].
The main emphasis is on the so-called build-up method for modeling aerodynamic
and thrust forces and moments. In this method, the airplane is assumed to be built up
from a number of components. The total forces and moments, which act on the
airplane, are then assumed to follow from summing the forces and moments, which
act on these components. In these terms, aerodynamic coefficients can be expressed
as follows:
(2.4)

The number of components, which should be used, depends on the airplane

configuration and on the level of accuracy desired. The stability axis system is used
for modeling aerodynamic forces and moments.

2.4.1 Longitudinal Aerodynamic Forces and Moments

Figure 2.2 illustrates the longitudinal aerodynamic forces and moments that act on
the airplane in a steady state flight condition. In the stability axis system, these forces
and moments are written as:
Longitudinal force in x-direction:
(2.5)

Longitudinal force in z-direction:

(2.6)

Airplane pitching moment:

(2.7)

Aerodynamic longitudinal forces and moment are non-dimensionalized in the

following way:

11
 (2.8)

(2.9)

(2.10)

where: is the total airplane drag coefficient; is the total airplane lift

coefficient. Total airplane drag and lift coefficients are the functions of the angle of
attack, , pitching rate, , elevator and flap deflections, , , Mach number and

Reynolds number. is the total airplane aerodynamic pitching moment coefficient

which in the steady state flight depends on the angle of attack ( ), dynamic
pressure( ), control surface deflections ( , ), Mach number and Reynolds

number, and moment reference center (usually the center of gravity) location.

Figure 2.2 Longitudinal Forces and Moment in Stability Axes

12
For an airplane equipped with an elevator and flaps, the drag coefficient is expressed
in terms of the first order Taylor series [15]:
(2.11)

where - the value of airplane drag coefficient for zero angle of attack, zero

deflections of control surfaces.

- the change of airplane drag due to change in airplane angle of

attack, .
- the change in airplane drag due to elevator deflection .

- the change in airplane drag due to flap deflection .

Similarly, the lift coefficient, , can be expressed in terms of a first order Taylor
series as follows:
(2.12)
where:
- the value of airplane lift coefficient for zero angle of attack, zero control

surfaces deflections.
- the change of airplane lift due to a change in airplane angle of

attack, .
- the change in airplane lift due to change in elevator angle.

- the change in airplane lift due to change in flap angle.

The aerodynamic pitching moment coefficient is expressed in the form of the

first order Taylor series as:

(2.13)

and denote change in airplane pitching moment due to change in

deflection of control surfaces.

The derivative is the change in airplane aerodynamic pitching moment

coefficient due to a change in angle of attack, it is also called the static longitudinal

13
stability derivative. It is of major importance to airplane stability and controls [15].
Another important derivative, which effects airplane stability is , also called the

pitch-damping derivative.
The steady state model for the aerodynamic force in the stability x-axis direction is:
(2.14)

The steady state model for the aerodynamic force in the stability z-axis direction is:
(2.15)

The steady state model of the aerodynamic pitching moment in stability axes is:
2.16)

It should be noted that the stability y-axis coincide with the body-fixed y-axis,
therefore .

2.4.2 Lateral-Directional Aerodynamic Forces and Moments

Lateral-directional force and moments that act on the airplane are shown in Figure
2.3.
In stability axis system, aerodynamic lateral-directional force and moments are
written as follows:
Aerodynamic rolling moment:
(2.17)

Aerodynamic side force:

(2.18)

Aerodynamic yawing moment:

(2.19)

For steady state flight conditions these force and moments are non-dimensionalized
as:
(2.20)

14
 (2.21)

(2.22)

where is aerodynamic rolling moment coefficient of an airplane; is airplane

aerodynamic side-force coefficient; is the airplane aerodynamic yawing

coefficient, which in a steady state depends on:

• angle of sideslip,
• angle of attack,
• dynamic pressure,
• moment reference center (usually the center of gravity) location
• deflection of lateral control surfaces
• deflection of directional control surfaces
• Mach number and Reynolds number

Figure 2.3 Lateral-Directional Force and Moments in Stability Axes

15
In terms of the first order Taylor series, these coefficients can be expressed as
follows:
(2.23)

(2.24)

(2.25)

The coefficients and derivatives are defined in a similar way as for the longitudinal
stability coefficients and derivatives.
The derivative is called the airplane effective dihedral; this derivative plays a

major role in determining airplane stability. The control power derivative is a

dominant factor on the bank angle maneuverability of airplanes. The magnitude of a

cross-control derivative should preferably be close to zero.

The coefficient tends to be zero for symmetrical airplanes. The derivative

is an important derivative in dutch-roll dynamics. It is also important in flight path

control. The control derivative is normally negligible.

For symmetrical airplanes coefficient tends to be equal to zero. The derivative

is an important derivative in dutch-roll and spiral dynamics, it is referred to as a

static directional stability derivative [15].

The steady state models for airplane aerodynamic rolling moment, side force and
yawing moment are:
(2.26)

2.27)

(2.28)

The values of the aerodynamics coefficients and derivatives are taken from the
reference [7], which contains wind tunnel test data, performed for analog airplane

16
SCAUT at Politecnico di Torino, Italy. Gaphs for the aerodynamic coefficients and
derivatives may be found in the Appendix A.

2.5 Mass- Inertia model

The mass inertia model contains the mass of the airplane and its tensor of inertia.
The mass is lumped and concentrated at the center of gravity (c.g.) location and the
inertia tensor is calculated with respect to the c.g.
Table 2.2. includes maximum take-off gross weight, fuel weight and moments of
inertia of the MTUAV [6].

Table 2.2. Mass and Inertia Properties of the METU TUAV

105 15 37.58 34.12 67.04 -6.91

The change of the gross weight and c.g. location of the MTUAV is neglected for
simulation purposes. Change in the c.g. location is very small, and its position moves
towards the nose of the airplane during the flight, increasing longitudinal stability of
the aircraft. Moments of inertia are considered to be constant for all time of
simulation.

2. 6 Propulsion Model

17
The propulsion model of the METU TUAV is represented by the engine model and
propeller model. METU TUAV uses two-cylinder engine Limbach L275 E shown in
the Figure 2.4
Characteristics of the engine are obtained from [16] and presented in the Figure 2.5.
Characteristics include required power ( ), torque and specific fuel consumption
(SFC) dependence on RPM. The data are shown for RPM variation from 4000 to
7200. It is assumed that the dependence of required power and SFC on RPM for
RPM is less than 4000 is linear. Fuel consumption (FC) can be found as follows:
(2.29)

Figure 2.4 Limbach L 275 E

Propeller is represented by the following characteristics:

the thrust coefficient given by

(2.30)

where T is the thrust of the propeller;

18
power coefficient

(2.31)

advanced ratio

(2.32)

Thrust and power coefficients and geometric properties of METU TUAV propeller
sections can be found in [6].

Figure 2.5 Power, Torque and SFC vs. RPM of Limbach L 275 E

2.7 Actuator Models

19
The actuators servo are assumed to be a first order servo with a transfer function

(2.33)

Dynamics of control surfaces (elevator, ailerons and rudder) is assumed faster than
dynamics of a UAV, the bandwidth T has been chosen equal to 30 rad/s. The inputs
to the actuators are subjected to the strict limits on the range of variations that can be
achieved. These limitations include physical limitations on the actuators themselves
and also ensure the system of being damaged [55]. Elevator deflection is constrained
up to , rudder and ailerons deflection is limited to .

2.8 Atmosphere model

Atmosphere model is used for developing more accurate aerodynamic model of the
vehicle. The main parameter that has an influence on the aerodynamic and thrust
forces and moments is air density , which depends on altitude. The altitude range
for MTUAV is 10 km. Change in air temperature, pressure and viscosity has a
secondary effect upon the aerodynamic and thrust forces and moments.
For simulating atmosphere effects Aerospace blockset of MATLAB/Simulink
provides a COESA Atmosphere Model block. The COESA Atmosphere Model block
implements the mathematical representation of the 1976 Committee on Extension to
the Standard Atmosphere (COESA) United States standard lower atmospheric values
for absolute temperature, pressure, density, and speed of sound for the input
geopotential altitude. Below the geopotential altitude of 0 m and above the
geopotential altitude of 84852 m temperature values are extrapolated linearly and
pressure values are extrapolated logarithmically. Density and speed of sound are
calculated using a perfect gas relationship. More information can be found in [17].

2.9 Gravity Model

20
The Earth model blocks available in the Aerospace Blockset. WGS84 Gravity Model
block implements the mathematical representation of the geocentric equipotential
ellipsoid of the World Geodetic System (WGS84). The block output is the Earth's
gravity at a specific location.
Use of the WGS84 Close Approximation model should be limited to a geodetic
height of 20000 m. Below this height, it gives results with high precision. More
specific information about the Earth gravity model is present in [20].

2.10 Non-Linear Simulation

6-DoF nonlinear model of METU TUAV is developed in MATLAB/Simulink, its

Simulink block diagram is illustrated in the Figure 2.6. The model consists of several
blocks:
• The 6DoF (Euler Angles) block implements an Euler angle representation of
six-degrees-of-freedom equations of motion of a vehicle
• Block which contains aerodynamic coefficients and derivatives of the METU
TUAV
• Environment model block, which contains blocks with atmosphere model and
gravity model
• Block, which computes the aerodynamic forces and moments using the
aerodynamic coefficients, dynamic pressure, center of gravity, and center of
pressure.
• Block that is used to calculate angle of attack, sideslip angle, and Mach
number.
For the 6DoF equations of motion block, it is assumed that the applied forces are
acting at the center of gravity of the body, and that the mass and inertia are constant.
Aerodynamic coefficients and derivatives were taken from [6] and [7]. Atmospheric
model provides data for atmospheric values for absolute temperature, pressure,

21
density, and speed of sound for the given altitude. Gravity model provides the Earth's
gravity information.

Figure 2.6 Simulink Block Diagram of the Nonlinear System

2.11 Linearization of Equations of Motion and State Space

Representation

Using the small disturbance theory, the nonlinear equations of motion of the airplane
are linearized for small disturbances around an equilibrium point of the airplane.

22
The steady-level-flight condition is chosen for the trim analysis. Under certain
assumptions, equations of motion can be uncoupled and airplane longitudinal and
lateral dynamics can be analyzed separately. Existence of pure longitudinal motion
requires three approximations and assumptions: the flat-Earth approximation,
existence of a plane of symmetry and absence of rotor gyroscopic effects. Existence
of the uncoupled lateral motions besides the restrictions above also requires
neglecting all aerodynamic cross-coupling terms (which may not be strictly equal to
zero) [14].
Linearization of the nonlinear model of the METU TUAV has been done using the
Simulink Control Design tools. Linearization is the approximation of a nonlinear
system as a linear system, based on the assumption that the system is almost linear
within a certain range of operation. Linearization is performed around certain
equilibrium point (or trim condition) of the nonlinear system.
Trim condition used for linearization task is given below:
velocity in x-direction in body axis , velocity in z-direction in body

axis , pitch angle , .

2.11.1 Longitudinal Dynamics

Consider the state space representation of the longitudinal dynamics. The states are
, the input is elevator deflection and the outputs are

. Expression for a state matrix and input matrix are the following:

23
For a trim condition described above, longitudinal state matrix and input

matrix are given below:

Roots of the characteristic equation of the airplane (eigenvalues) are equal to:

The pair of roots of the characteristic equation that are located more far away from
the imaginary axis and have a greater absolute value is referred to as a short period
mode. The other pair of roots of the characteristic equation, with a smaller imaginary
part that are located closer to the imaginary axis, defines a phugoid mode [18]. The
phugoid or long-period mode is the one in which there is a large-amplitude variation
of air-speed, pitch angle, and altitude, but almost no angle-of-attack variation. In
practice airplane is always under the action of disturbances caused by different
factors (pilot, control system etc.). The phugoid oscillation is really a slow
interchange of kinetic energy (velocity) and potential energy (height) about some

24
equilibrium energy level as the aircraft attempts to re-establish the equilibrium level-
flight condition from which it had been disturbed. The short-period mode is a very
fast, usually heavily damped, oscillation with a period of a few seconds. This motion
might be, for example, a rapid pitching of the aircraft about the center of gravity. The
period is so short that the speed does not have time to change, so the oscillation is
essentially an angle-of-attack variation [15]. Characteristics of the longitudinal
modes are given in Table 2.3.

Table 2.3 Longitudinal Modes Characteristics

Mode Root Location Natural Period Damping Time to Half

Name Frequency (s) ratio Amplitude
(rad/s) (s)
Short 9.855 0.637 0.288 0.252
Period
Phugoid 0.262 23.981 0.076 34.65

2.11.2 Lateral Dynamics

States, which correspond to the lateral dynamics of an aircraft, are: side velocity in
body axis ( ), roll rate ( ), yaw rate ( ) and roll angle ( ). To control the lateral
motion ailerons and rudder are used. State and control matrices, which describe
lateral motion of an aircraft are as given below:

25
where

For a trim condition, lateral state and input matrices and are obtained as
follows:

The eigenvalues of the lateral state matrix are equal to:

26
The lateral motion of aircraft is described by 3 dynamic modes: roll mode, dutch-roll
mode and spiral mode.
The first two roots of the characteristic equation is a complex conjugate pair. This
pair is referred to as a dutch-roll mode, which is a coupled roll and yaw motion. This
mode is usually poorly damped.
The real root with a larger magnitude defines the roll mode. The roll mode consists
of almost pure rolling motion and is generally well damped. The real root which is
closer to the origin corresponds to the spiral mode, which is usually slow and
unstable. In order to reduce the impact of the spiral mode aircraft should have
sufficiently large wing dihedral, which increases roll stability. The lateral modes
characteristics are shown in the Table 2.4 and 2.5.

Table 2.4 Dutch-roll Mode Characteristics

Mode Root Location Natural Period Damping Time to Half

Name Frequency (s) ratio Amplitude
(rad/s) (s)
Dutch-roll 3.216 1.953 0.076 2.826

Table 2.5 Roll and Spiral Mode Characteristics

Mode Name Root Location Time Constant (s) Time to Half

Amplitude (s)

Roll -6.0467 0.165 0.114

Spiral -0.0092 108.695 75.0

27
2.12 Open-Loop Simulation

Open-loop simulation shows response of the system to the deflection of control

surfaces. As it has been mentioned before, elevator is used as a control to
longitudinal dynamics, ailerons and rudder are used as controls to lateral motion of
an aircraft. Open-loop response simulation is performed for both linearized and
nonlinear models of the UAV in comparison purposes.
Giving the elevator input of 5 degrees for the time interval between 10 and 15
seconds system’s open-loop response represented by body x and z velocities, pitch
angle and pitch angular rate is shown in Figures 2.7 – 2.10. System’s response to the
ailerons deflection of 1 degree for five seconds starting at 10th second in terms of
body y velocity, roll angle, roll and yaw angular rates is shown in Figures 2.11 –
2.14. Change of the UAV position in time for the aileron one degree step input at 5th
second, simulated for a nonlinear model is shown in the Figure 2.15.
As it is seen from the plots on the figures below, that the open-loop response
characteristics are not satisfactory: it takes more than 200 seconds for a system to
reach its longitudinal trim condition with quite large overshoot. From the response of
the system to change in ailerons position, it is seen that though analysis of the linear
model shows that the spiral mode is stable (because the eigenvalue, which
corresponds to the spiral mode, is negative), nonlinear simulation shows that the
spiral mode is unstable. This difference may occur because of the numerical
algorithm, which was used for the linearization.

28
Figure 2.7 System Response to Elevator Input: x-body Velocity

Figure 2.8 System Response to Elevator Input: z-body Velocity

29
Figure 2.9 System Response to Elevator Input: Pitch Angle

Figure 2.10 System Response to Elevator Input: Pitch Rate

30
Figure 2.11 System Response to Ailerons Input: y-body Velocity

Figure 2.12 System Response to Ailerons Input: Roll Angle

31
Figure 2.13 System Response to Ailerons Input: Roll Rate

Figure 2.14 System Response to Ailerons Input: Yaw Rate

32
Figure 2.15 System Response to Ailerons Step Input

33
 CHAPTER 3

CLASSICAL CONTROLLER DESIGN

In this chapter, classical approach to controller design is presented. Pitch attitude

controller and roll autopilot design for METU TUAV using a PID controller is
performed.

3.1 PID Controller Background

Classical controller design techniques are based on a frequency domain analysis of

system dynamics. A closed-loop controller uses feedback to control states or outputs
of a dynamical system.
Feedback control system, in general, consists of system to be controlled, which
includes actuators, sensors to measure system’s output, and a controller (see Figure
3.1).
The main reason to use feedback is to minimize the effect of “uncertainty” [21]. The
term “uncertainty” includes the uncertainties of system modeling, disturbances and
noise. More detailed description of uncertainty and its sources will be given further.
Feedback controllers have many advantages over the open-loop controllers:
• improve system response characteristics
• show good performance in the presence of model uncertainties
• noise and disturbance rejection
• unstable processes can be stabilized

34
 Measured System System
Reference + error input output
Controller System

-
13

Measured output Sensor

Figure 3.1 Standard Feedback Configuration

A common closed-loop controller architecture is the PID controller. PID controller

includes three terms in parallel: proportional (P), integral (I) and derivative (D).
Because of the simple design, PID controllers are widely used in industry. Transfer
function of a PID controller is the following [22]:

(3.1)

where , and are controller gains.

Block diagram of a PID controller is shown in the Figure 3.2 .

Some authors (for example, in references [23], [36], [37]) use representation of a PID
controller in terms of proportional gain , integral time , and derivative time

. In these terms, the PID algorithm can be described as:

(3.2)

where u is the control signal and e is the control error (difference between the set-
point and output).

35
 Input Error + Output
+ +
Process

- +
+

Figure 3.2 Block Diagram of a PID Controller

Thus, controller parameters that must be determined are , , and . PID

controller combines both PI and PD controllers and therefore includes the advantages
of both. Since PI controller performs its action in a low-frequency region and PD
controller – in a high-frequency region, PID controller can be used for design of
control systems in which it is required to improve characteristics of both transient
and steady state response [23].
Proportional term reduces the rise time and steady-state error (but never eliminates
it); it also attenuates noise at high frequencies. However, increasing the controller
gain makes system unstable. Adding an integral term adds a zero at the origin, which
makes the controller to have an infinite gain at zero frequency. The presence of a
pole at the origin of the complex plane allows the reduction to zero of the steady-
state error when a step reference signal is applied or a step load disturbance occurs.
Derivative term improves the closed-loop stability: reduces overshoot and settling
time of the system response, it has no effect on the steady-state error. Disadvantage

36
of a derivative term is its sensitivity to noise in the error term, so if noise amplitude
and the derivative gain are large system becomes unstable. More detailed
information about PID controllers is introduces in [25], [26], [27].
Considering representation of a PID controller as it is shown in (3.1), if denotes
the input to PID controller, then its output is a linear combination of , its
integral, and its derivative.
Presence of pure derivation action makes the controller transfer function not proper
and therefore it cannot be implemented in practice. The difficulty in designing PID
controller is in physical implementation of a derivative term. Very often, in practice,
proper approximation of a derivative term is used:

(3.3)

where is a small positive number [21].

Usually it is sufficient to use only PI controller, which is done in many industrial
controllers. PI controller shows sufficient performance for all processes where the
dynamics are essentially of the first order or if the process requirements are not too
high. PID controllers are used for systems with dynamics of second order or when a
tight control of a high-order system is required. [36] and [37] cover principles of PID
controllers, implementation details, when to use PID controllers, design of PID
controllers, etc.

3.1.1 PID Controller Tuning Rules

There are different techniques for determining gains of PID (proportional–integral–

derivative) controllers. Tuning the gains of a PID controller can be done using a
design experience or by various design techniques for determining parameters for the
controller. [23] gives Ziegler-Nichols rules for tuning PID controllers. [32] presents
Ziegler-Nichols Types of Tuning Rule for PID controllers. An Internal Model
Control method (IMC) is described in [28]. This method is used for defining the

37
structure of a controller and it introduces a tuning technique for tuning of the
controller’s gains, so-called IMC-PID tunings. [30] provides the most comprehensive
summary of PID controller tuning rules to date. Tuning of the controller’s parameters
is performed after specifying the structure of the controller as well as the form of the
model used to represent the dynamic response of the controlled variables. Although
tuning rules that provide the best performance often vary from case to case, modern
rules such as IMC and direct synthesis tend to provide better results than traditional
techniques such as Ziegler Nichols or Cohen-Coon tuning. [31] covers PID controller
design for various systems, evaluates several existing PID tuning rules from the
robustness and fragility point of view. The range of system parameters is given for
each tuning rule such that the system remains stable when the PID parameters
deviate from the nominal value within the given range. This reference also provides
several examples to illustrate the design of PID control to satisfy time domain
performance specifications. [32] gives basics and fundamentals of PID control,
online model-free methods, with emphasis on the iterative feedback tuning (IFT)
method, nonparametric methods for PID tuning through the concept of relay
feedback, a well-known technique that experimentally locates critical points in the
frequency response that can then be used for controller tuning, it also covers the
topics of using of relay feedback for multivariable systems, parametric design
methods that assume that a transfer function model is available at the onset of the
design procedure. [36] and [37] include Ziegler-Nichols classical methods of PID
controllers design and their extensions; pole placement and other methods; auto-
tuning techniques. Reference [36] provides some adoptive techniques, which include
auto-tuning, gain scheduling and adaptation.

3.2 Control Problem Definition

Solving the control problem generally involves:

• Choosing sensors to measure the plant outputs

38
 • Choosing actuators to drive the plant
• Developing mathematical models for plant, actuators and sensors
• Define control criteria
• Designing the controller, which is based on the developed models and control
criteria
• Check controller performance by simulation.
Control criteria usually include (see reference [22]):
• Reducing the effects of the disturbance of the output
• Steady-state errors
• Transient response characteristics
• Sensitivity to parameter changes in the plant.
Mathematical models of plant and actuators has been given in a previous chapter.
The following section describes sensors required for a controller synthesis and give
definition to a command filter.
The following requirements for the system response must be considered and satisfied
pending the controller design for METU TUAV:
• overshoot: < 10%
• settling time: < 3 sec
• rise time: < 1 sec
• steady state error: < 2%

3.3 Sensors

Onboard sensors are necessary to measure the states of the aircraft. The
measurements are performed with respect to certain reference frames (inertial frame,
Earth frame, body frame, etc.). Dynamics of inertial sensors, like accelerometers and
gyros, with a high bandwidth is usually neglected. These sensors are modeled as
simple gains [24].
The following sensors are available on the board of METU TUAV:

39
 • Inertial Navigation System (INS) is used to obtain the position and
orientation of a vehicle. INS is an autonomous system. Inertial measurement
unit (IMU) typically includes three orthogonal accelerometers, and three
orthogonal gyros (usually rate-gyros). Inertial measurement unit gives the
information about linear accelerations and angular turning rates about the
airplane’s body axis. Euler angles are obtained by integrating body angular
rates. Position of the vehicle is calculated by integrating linear accelerations.
INS outputs also provide “steering commands to the autopilot to steer the
airplane through predetermined waypoints”. However, output of INS tends to
drift during the integration process: small errors in the measurement of
acceleration and angular velocity grow in time with integration. This leads to
significantly larger errors in position and orientation. This problem can be
eliminated by fusing INS with GPS. More detailed information about
characteristics and basic principles of INS can be obtained from [29].
• Global Positioning System (GPS), which gives accurate three-dimensional
position and velocity information. However, GPS is not autonomous system;
it is subjected to jamming. Integrating INS and GPS combines advantages of
both systems and eliminates the disadvantages. Therefore, data, obtained in
such a way, are reliable for navigation purposes.
• Airspeed indicator. The airspeed indicator or airspeed gauge is used in an
aircraft to display the aircraft’s airspeed.
• Altimeter is an instrument used to measure the altitude of an aircraft with
respect to a fixed level. Barometric altimeter (or pressure altimeter) uses an
aneroid barometer measures the atmospheric pressure from a static port
outside the aircraft.

3.4 Classical Controller Design for METU TUAV: Pitch

Attitude Controller

40
The purpose of the pitch attitude autopilot is to follow the desired reference input and
maintain the pitch orientation of the aircraft after it has reached desired value. For
design purposes, linearized longitudinal dynamic equations are used for the
controller synthesis and simulation. Later performance of the controller is checked
for the 6-DoF nonlinear model of the airplane. Response of the control system
through the nonlinear simulation with comparisons to a robust controller response is
given in Chapter 5.

3.4.1 Controller Configuration

The basic block diagram for the closed-loop system of the pitch autopilot is shown in
Figure 3.3. Command input is first introduced into the closed-loop. A pitch angle

error is the input to the controller. It represents the difference between the system
output (measured pitch angle ) and the commanded pitch angle .

Command
Input Error A/C Output
PID Dynamics
Controller
-

Figure 3.3 Basic Controller Configuration

41
Aircraft dynamics includes linearized longitudinal dynamics of the METU TUAV
and actuator model. Elevator deflection is used as a control surface.
Pitch angle to elevator deflection transfer function is obtained as:

(3.4)

The transfer function of a PID controller:

(3.5)

where controller parameters (proportional gain, integral time and derivative time)
must be determined.
Negative sign indicates that a positive deflection of a control surface creates a
negative pitching moment therefore, negative pitch angle.

3.4.2 Determining Controller Parameters and Controller Performance

For determining the parameters of a PID controller, which is used in pitch attitude
control system, Ziegler-Nichols rule, based on step response of the plant, presented
in [3] is used. This method can be applied in case if its response to the unit step looks
like S-shaped curve, which is shown in the Figure 3.4.

42
 Figure 3.4 S-shaped Response Curve

The response curve can be characterized by two parameters: delay time L and time
constant T, which are obtained by drawing a tangent line at the inflection point and
defining cross points as it is shown in Figure 3.4. Reference [3] gives formulas for
calculating , and based on values for L and T. It should be noted, that a PID

controller tuned by Ziegler-Nichols rules will have a response with 10%-60%

maximum overshoot. However, by experimental tuning it is possible to adjust
parameters of the controller such that it would exhibit desired responses. Therefore,
Ziegler- Nichols tuning rules are very useful in order to find an initial point to start
tuning of a PID controller.
The plant is described by the transfer function (3.4). Applying the method explained
above, define the parameters of the S-shaped step response as: , , the
following values of the controller parameters are obtained: , ,

. However, the simulation does not show the desired response to the

pitch command. Having knowledge of how each parameter effects the response of
the closed-loop system, by manual tuning it is possible to adjust the values of the

43
parameters of the PID controller such, that the response curve would have a desired
shape. The following values of the parameters were obtained: , ,

Then the controller transfer function can be written as:

(3.6)

It should be noted, that tuning of the PID controller’s coefficients is performed for
the nominal plant, however presence of a sensor noise is also taken into account in
simulation.
Phase and gain margins are illustrated in the Bode plot of the open-loop system in
Figure 3.5.

Figure 3.5 Bode Plot of the Open-Loop System

44
 Figure 3.6 System Response to a Unit Step Input: Pitch Angle

Response of a closed-loop system to a unit step input is illustrated in Figure 3.6 and
has the following characteristics:
• overshoot: 10%
• settling time: < 3 sec
• rise time: < 1 sec
• steady state error: < 0.2%.
Elevator deflection and deflection rate for a unit step input is shown in Figures 3.7,
3.8.

45
 Figure 3.7 Elevator Deflection for a Unit Step Input

Figure 3.8 Elevator Deflection Rate for a Unit Step Input

46
From Figure 3.8 it is seen that the rate of elevator deflection is high at initial time of
the input, which is not desired in the physical system. This problem is solved by
using a command filter. Architecture of the command filter and its implementation
are described in section 3.6.

3.5 Classical Controller Design for METU TUAV: Roll

Autopilot

The basic function of the roll autopilot is to bring the airplane to a desired roll
orientation. Ailerons are used as a control surfaces. Design of the PID controller and
simulation is done using a linearized lateral dynamics of the METU TUAV. In
Chapter 5 performance of the PID controller is given on the nonlinear model in
comparison with nonlinear simulation of Robust controller.

3.5.1 Controller Configuration

Configuration of a controller, which is used for roll autopilot, is chosen to be similar

to those used in pitch attitude control, as it is given in Figure 3.3.
Roll angle command is used as an input to the system. Similarly, to the pitch attitude
controller, the error, which represents difference between the command roll angle
and system response , is given to the input of a PID controller. Ailerons deflection
signal is an output of the controller. Aircraft dynamics is represented by the lateral
linearized dynamics of the METU TUAV and actuator model. Lateral linearized
dynamics of the MTU TUAV is given as a transfer function from ailerons deflection
to roll angle of the airplane:

(3.7)

The transfer function of a PID controller is given by:

47
 (3.8)

Controller parameters are subjected to be determined such that response of a closed-

loop system would satisfy specified requirements.

3.5.2 Determining Controller Parameters and Controller Performance

The same method as it is described in section 3.4.2 is used to determine parameters

of the PID controller used for roll autopilot. Consider the same S-shape of the desired
step response as for the pitch controller. PID parameters, which correspond to this
response are: , , . As simulation shows, the response

of the closed-loop system is not satisfactory. Using these values as a start point,
manual tuning is performed to make a step response’s characteristics correspond to
the desired ones; noise attenuation is taken into account while tuning. The following
values of controller parameters are chosen: , , .

Transfer function of the PID controller:

(3.9)

Figure 3.9 shows the Bode plot of the open-loop system, which provides information
about phase and gain margins and stability of the closed loop system.
Response of a closed-loop system in terms of roll angle, position and rate of ailerons
is illustrated in the Figures 3.10, 3.11, 3.12.
Response of the closed-loop system has characteristics listed below:
• overshoot: 7%
• settling time: < 3 sec
• rise time: < 1 sec
• no steady state error

48
Figure 3.9 Bode Plot of the Open-Loop System

49
Figure 3.10 System Response to a Unit Step Input Roll Angle

Figure 3.11 Ailerons Deflection for a Unit Step Input

Figure 3.12 Ailerons Deflection Rate for a Unit Step Input

50
3.6 Command Filter Implementation for a Control System

3.6.1 Command Filter

Command filters are used in order to shape the command signal so its dynamics
matches the dynamics of the vehicle. If the command input is a step function (for
example pitch angle for a pitch attitude autopilot), this means that the rate of change
of the input (pitch rate) will be infinite. This might cause saturation of control
surfaces. The purpose of command filter is to smooth the input signal such an
airplane would be able to follow it (see Figure 3.13).

Figure 3.13 Command Filter

Command filters are low-pass filters. For the purposes of this thesis, second order
command filter is used. The transfer function for a second order low pass filter is

(3.11)

Command filters are frequently used in control systems and are well described in
literature. In [33] a time varying bandwidth (TVB) command shaping filter, which is

51
used to improve the tracking transient performance, is introduced. TVB filter limits
the actuator deflection rate to achievable values. Reference [34] explaines the usage
of a command filter in order to improve a system’s step response: command filter
removes high frequency components from a command input, therefore reducing the
overshoot of the response. Command filter does not cause a destabilizing action upon
a control system as it is located outside of the loop. [35] shows the way of using a
pre-filter in order to eliminate the effect of the zero of a closed-loop transfer function
on the step response. In reference [10] 2nd order command filter is employed to the
autolanding control system designed for the METU TUAV.
Figure 3.14 gives a block diagram of a closed-loop control system with a command
filter implemented.

3.6.2 Pitch Autopilot Performance with a Command Filter

Implementation

Consider the block diagram of the control system as it is shown in Figure 3.14.
Aircraft dynamics includes a state space representation of the linearized longitudinal
dynamics of the UAV. Command input to a closed-loop system is given in Figure
3.15.

Filtered
Command Command
Input Error A/C Output
Command Input PID Dynamics
Filter Controller
-

52
 Figure 3.14 Basic Controller Configuration with Command Filter
Nominal command for angle of attack is shaped by passing through a command
filter. Filtered command represents the desired change in pitch angle, which should
be provided by a PID controller in a closed-loop configuration. Response of the
closed-loop system that includes linear dynamics of the UAV to a given input is
shown in Figure 3.16. Control surface deflection and its rate are presented in Figure
3.17 and Figure 3.18 respectively. From the plots above it is seen that implementing
a command filter to the control system improves its response. Shaping the input
command decreases the control surface efforts required to bring an airplane to a
desired pitch attitude.

Figure 3.15 Pitch Angle Command: Nominal vs. Filtered

53
Figure 3.16 System Response to Pitch Angle Command

Figure 3.17 Elevator Deflection

54
 Figure 3.18 Elevator Deflection Rate

3.6.3 Roll Autopilot Performance with a Command Filter

Implementation

Control system configuration is given in Figure 3.11. Aircraft dynamics consists of

the state space model of the linearized lateral dynamics of the METU TUAV. In
order to analyze performance of the roll autopilot with an implemented command
filter, consider the roll angle input command as it is given in the Figure 3.19. Actual
command to the control system is a filtered output of the command filter.
Performance of the PID controller, represented by a roll angle response, is given in
Figure 3.20. Deflection of the control surfaces is shown in Figure 3.21. Figure 3.22
shows rate of the ailerons deflection.
From the simulation results, it is seen that adding a command filter to the control
system sufficiently reduces angle of ailerons deflection and the rate of deflection
making it correspond to the UAV’s physical capabilities.

55
Figure 3.19 Roll Angle Command: Nominal vs. Filtered

Figure 3.20 System Response to a Command Roll Angle

56
 Figure 3.21 Ailerons Deflection

Figure 3.22 Ailerons Deflection Rate

57
 CHAPTER 4

ROBUST CONTROLLER DESIGN

Robust control design focuses on the implications of model uncertainty. The

controller should provide stability of the system in the face of uncertainties.
This chapter gives brief historical and theoretical backgrounds of the robust control
theory. Techniques of robust control, such as design and -analysis, are

described.

4.1 Historical Review

Research in robust control theory has been one of the most active areas of modern
control theory since the late 70s. The first important works were made by Zames [38]
and Zames and Francis [39] in the late 1970s and early 1980s. Later this theory
became known as optimal control theory. The main idea was to find design

techniques that would provide stability of a closed-loop system under the system
uncertainties. Zames formulated three fundamental problems in robust control:
1) the uncertain disturbance attenuation problem, know as the standard problem;

2) a plant and disturbance attenuation problem, known as the optimal robust

disturbance attenuation problem;
3) and a filtering of plant uncertainty problem in which he posed the question of how
much feedback can reduce uncertainty, when the plant lies in some set of uncertainty.

58
The works of Athans [40], Safanov [41], Grimble [42] and others show how
uncertainty can be modelled and give the concept of the norm and -synthesis

theory. In 1987 B. A. Francis and John C. Doyle gave a modified solution to the
general rational MIMO optimal problem but suffered from the high order of the

Riccati equations [43], [44]. State space methods for solving the problem for

MIMO systems were firstly introduced in [45]. The connection between

control problem and classical control was made in [46], in which loop shaping

design techniques were introduced. More detail and different perspectives can be
found in the books by Zhou, Doyle, and Glover (see references [47] ,[48]).

4.2 Theoretical Background

This section introduces brief theoretical information about systems uncertainties,

their modeling, stability and performance specifications, and robust controller design
tools.

4.2.1 Problem Formulation

Consider a control system, block diagram of which is given in the Figure 4.1
In the Fig. 4.1 P is a generalized plant derived from the nominal plant but includes
weighting functions. P is also assumed Finite Dimensional Linear Time Invariant
(FDLTI) system; is a controller; is an external input signal, which includes the
reference signal, disturbances and noise; represents control signal; is a signal
that contains measured variables, and represents the error signals. It is assumed
that state space models of P and are available and that their realizations are
assumed stabilizable and detectable [47]. The goal of robust control design is to
minimize the norm of the transfer function from to under the constraint of

internal stability. The definition of internal stability will be given in further section.

59
 z w
P

y u
K

Figure 4.1 Block Diagram

4.2.2 Signals and Systems norms

Norms are used to characterize the size of signals or systems. The main purpose of
system norms computation is to get proper measures to compare the performance of
a control system, which can be based on the measure of the error signals, system
gains, etc.
Consider the block diagram of an LTI system in the Figure 4.2.

u(t) y(t)
G

Figure 4.2 Schematic Block Diagram of an LTI System

Following norms can be defined for scalar signals:

60
 norm: (4.1)

norm: (4.2)

norm: (4.3)

norm: (4.4)

Another important measure for signals is root mean square value (RMS). It reflects a
signal’s eventual average size:

(4.5)

Norms of systems are usually determined in terms of norms of input and output
signals (induced norms). For a stable SISO linear system the norm:

(4.6)

Physical interpretation of norm: -norm of a transfer function measures the

RMS response of its output when it is driven by a white noise input.

norm of the system provides a measure of a worst-case system gain. norm

of a stable SISO system is defined as:

(4.7)

norm is a peak value in the Bode magnitude plot of the system. It provides a
bound on the system gain.
For a MIMO system is a rational transfer function matrix
. norm is defined as:

61
 norm:

(4.8)

where is the maximum singular value of the matrix .

norm of the system describes the maximum energy gain and is computed by the

peak value of the largest_singular_value of the frequency response matrix over the
whole_frequency range [49].

4.2.3 Modeling Uncertainty

The term “uncertainty” means that the mathematical model of the plant differs from
the actual physical model. It can be caused by few reasons: disturbance signals and
dynamic perturbations. Disturbance signals usually include input and output
disturbances (gust on the A/C), sensor noise, actuators noise. Dynamic perturbations
include the differences between the mathematical model and the actual system:
missing high frequency dynamics, system parameters errors or their variation in time,
etc.
Taking into account the model uncertainty it is assumed that the dynamic behaviour
of the plant is described by a set of possible LTI models is a set of all perturbed
plants. Consider the following notation for plant models:
: nominal plant model with no uncertainties;
: particular perturbed plant.

Assume that all the uncertainties (such as missing high frequency dynamics), that
may occur in different parts of the system, are lumped into one block , which is
represented by any stable transfer function with norm less than unity.

62
[Link] Additive Uncertainties

A block diagram that represents an additive uncertainty configuration is shown in

Figure 4.3 below.

+
+

Figure 4.3 Additive Uncertainty Configuration

Perturbed plant is defined as:

(4.9)

is a stable transfer function, which defines the magnitude of the uncertainty in all
frequency range. block represents a stable transfer function, which satisfies
.

[Link] Multiplicative Uncertainties

A block diagram that represents a multiplicative uncertainty configuration is given as

in the Figure 4.4.

63
Perturbed plant with a multiplicative uncertainty can be defined as:
(4.10)

Figure 4.4 Multiplicative Uncertainty Configuration

As for an additive uncertainty, for multiplicative uncertainty is a stable

weighting transfer function, which defines the uncertainty magnitude at each

frequency:

(4.11)

Weighting function for a multiplicative uncertainty typically has the properties

illustrated in Figure 4.5 (see reference [47]).

[Link] Parametric Uncertainties

For parametric representation of uncertainty it is assumed that the structure of the

plant is known but some of the parameters are uncertain. Parameters of a physical
system, which is modeled, determine the coefficients of the nominal plant . The

64
structure of the perturbed plant is known: the whole set of possible perturbed

plants have the same structure as the nominal plant , but some of the
parameters are uncertain [21]. These uncertainties will affect the closed-loop system
performance at low frequencies.

Figure 4.5 Typical Behavior of Multiplicative Uncertainty

4.2.4 Linear Fractional Transformations

Consider the general framework of a control problem as shown in Figure 4.6.

Suppose, that that plant is partitioned as follows:

(4.12)

65
 (4.13)

Figure 4.6 Lower Linear Fractional Transformation

Using it is possible to eliminate and :

, if is invertible.
When the inverse exists, then
(4.14)

is called a lower linear fractional transformation (LLFT), which is used to

indicate the incorporation of a controller into a system.

In a similar way, consider a configuration shown in the Figure 4.7.
Upper linear fractional transformation of and can be defined as:
(4.15)

66
 P

Figure 4.7 Upper Linear Fractional Transformation

4.2.5 Transfer Functions of a Closed Loop System

Consider the following feedback configuration as it is shown in Figure 4.8.

K P
-

Figure 4.8 Feedback Configuration

67
 represents the plant to be controlled, block represents a controller,
represents a reference command, represents the error signal, represents
controller’s output, and represents disturbances and noise, and represents
system’s output.
Transfer function of a closed-loop system for a given feedback configuration is
defined as:

By easy manipulations, it can be derived that

(4.16)

is so-called Sensitivity Function, is a Complementary Sensitivity Function.

is a transfer function from the reference command input to system’s output, is
a transfer function from disturbances to the output. It should be emphasized that
.
From the relation (4.16) above it is obvious, that for a good reference tracking
( ) complementary sensitivity function should be as much closer to unity. For a
good disturbance rejection sensitivity function has to be zero. Noise rejection
requires being zero, which contradicts to . Therefore, it is impossible to
have a good tracking and disturbance/noise rejection at the same time. However, this
problem can be solved by frequency analysis of a reference signal and noise. Usually
command inputs are signals with a low frequency and noise is assumed as a high
frequency signal. Therefore, for a low frequency range should be large for a good
tracking and small for disturbances attenuation. At high frequencies the main
purpose is noise rejection, therefore is small.
Using the definition of a complementary sensitivity function:

(4.17)
From relation (4.17) it is obvious that for a good tracking system gain has to be very
large.

68
4.2.6 Stabilization and Performance Requirements (SISO
Systems)

This section provides stabilization and performance requirements in robust design

and introduces a ‘small-gain theorem’, which is of the great importance in

optimization techniques. The system to be considered is shown in the Figure 4.9.

+ +
+
K P
-

Figure 4.9 System Block Diagram

[Link] Nominal Stability (NS)

NS implies that the system is stable with no model uncertainty. An interconnected

system is internally stable if the subsystems of all input-output pairs are
asymptotically stable [12]. To guarantee NS of a system its closed-loop transfer
function should have all the poles in LHP and no unstable zero/pole cancellation.
More detailed explanation of internal stability can be found in [48], [49].

69
[Link] Nominal Performance (NP)

NP means that the system satisfies the performance specifications with no model
uncertainty. NP implies that the norm of a transfer function from disturbances to
error must be small. For norm notation:

(4.18)

where is any stable transfer function which defines the magnitude of uncertainty.

is a transfer function from disturbances to errors.

[Link] Robust Stability (RS)

The definition of RS is the following: the system is stable for all perturbed plants
about the nominal model up to the worst case of uncertainty. So if the controller K
internally stabilizes each of the perturbed plant including the nominal, robust
stability is guaranteed.
If the norm is used, condition for robust stability is the following:

, (4.19)

Small gain theorem is the basis for derivation of a robust stability test. Consider the
block diagram from the Figure 4.9 in a modified form. By isolating the block and
re-drawing the feedback connection, the block diagram of a feedback configuration
is obtained as shown in Figure 4.10.
Theorem: if and are stable linear systems, the closed-loop system is internally
stable if and only if for all .

Consider the transfer function from disturbances to errors:

Applying Nyquist stability criteria: should not encircle the -1 point on a

complex plane for all :

70
 .

The worst case for the condition above is when . Then the condition for RS
can be written in the form: .

e d

Figure 4.10 Structure

In terms of norm: if ( is a small positive number) the condition for

RS is:

(4.20)

[Link] Robust Performance (RP)

RP implies that the performance specifications are satisfied for every perturbed plant
about the nominal model up to the worst case uncertainty. In other words RR
requires the NP condition to be satisfied for all possible perturbed plants.

71
For internally stable nominal feedback system NP condition is .

Considering the case with multiplicative uncertainty. The perturbed plant

, then is perturbed to .

The RP condition becomes

and , , .

The theorem below gives a test for a RP:

“A necessary and sufficient condition for robust performance is
” (4.21)

Proof of (4.21) can be found in [50].

4.2.7 Stabilization and Performance Requirements (MIMO

Systems)

Similar approach as for SISO systems is used for MIMO case. Conception of LFT is
used to derive stability and performance requirements for MIMO system.
For a general case hence both uncertainties and disturbances act upon the system the
following configuration as shown in Figure 4.11 is considered.
In the Figure 4.11, plant and controller are combined into block , which
represents the lower linear fractional transformation: .

The transfer function from disturbances to errors: .

represents the upper linear fractional transformation of with :

(4.22)

72
 w z w z
d
P e d
M e

u y
K

Figure 4.11 Plant Controller Configuration Set Up

[Link] Nominal Stability (NS)

Condition for NS for a MMO system comes from the criterion for SISO system.
Transfer functions from all inputs to all outputs must satisfy NS conditions for SISO
case.

[Link] Nominal Performance (NP)

NP requires minimization of a transfer function from disturbances to errors with no

system uncertainties. Therefore, from relation (4.22) in terms of norm condition

for NP is:
(4.23)

[Link] Robust Stability (RS)

73
A criterion for RS is derived using the small gain theorem: if then

(4.24)

(4.24) gives a condition for RS in terms of norm.

[Link] Robust Performance (RP)

RP requires that the must be small even in the presence of . Minimizing of

the norm:

if for all with , then the closed-

loop system satisfies RP criterion.

Using the small gain theorem, the “sufficient condition” for RP is to have
(4.25)

4.3 Design

The main aim of robust design is to find a controller for the system, such that the
closed-loop system is robust.
Optimal control problem statement is the following: “find all the admissible

controllers such that is minimized”. However, for MIMO systems, the

procedure of finding the optimal controller is very complicated andthere is no

closed form solution. Therefore, instead of the problem statement given above,
another definition of control problem is used: “Given , find all admissible

controllers , if there is any, such that ”. This problem is called

suboptimal control problem.

74
Consider the system block diagram of which is shown in Fig. 4-1, where controller K
and plant G are assumed real rational and proper.
Consider the following realization of the transfer function matrix:

(4.26)

(4.27)

According to (4.26) system equations can be written in the form:

(4.28)

Theorem above gives necessary and sufficient conditions for existence of a controller
.
According to the theorem, if three conditions are satisfied there exist a controller
such that .

The required conditions are:

i) there exists - a solution to ARE

(4.29)

ii) there exists - a solution to ARE

(4.30)

iii) (4.31)

75
When these conditions are satisfied, the controller is given by:

(4.32)

where

;
The following assumptions should be made for the derivation of the results of the
theorem:
1) is stabilizable

2) is detectable

3)

4) .

More detailed description of the solution of problem for a general case can be

found in [45], another approach for solution derivation is present in [51].

4.4 Analysis

4.4.1 Definition and Properties of Structured Singular Value ( )

Consider the block diagram as it is shown in the Figure 4.12. The uncertainty block
in the structure connection usually has a block diagonal structure:
.

76
 e d

Figure 4.12 Standard configuration with diagonal structure of

The sufficient condition for RS, as defined previously, is:

, .
Define a scaling matrix as
(4.33)

where is a scalar and is an identity matrix, which has the same dimension as

the uncertainty .

Modified structure is shown in the Figure 4.13.

Including scaling matrix D to the structure does not change the overall
system, its stability properties remain the same.
Then condition for RS can be written in the form:
if
, (4.34)
then closed-loop system is robustly stable.
Less conservative condition for RS can be obtained by minimizing at each frequency
the scaled singular value:
, (4.35)

where D is a set of block-diagonal matrices compatible with :

77
 (4.36)
The Structured Singular Value (SSV) , , was first suggested by Doyle in 1982 as a
way of analyzing systems with uncertainties. The SSV is defined as follows

(4.37)

If there is no that makes singular, is defined as zero.

Figure 4.13 Modified Structure

SSV is a function, which provides a generalization of the largest singular value . It

should be noted, that depends on both and structure of .

Computation of a SSV is very difficult procedure; therefore, upper and lower bounds
on are used:
(4.38)

78
Still, the gap between spectral radius and largest singular value may be
arbitrary large, that is not acceptable. For this reason, it is necessary to define the
bounds by scaling in a way that will not affect but will affect and . In

particular, the bounds maybe tightened in a following way:

Define two subsets in :
(4.39)

(4.40)

Then the upper and lower bounds from (4.37) can be tightened to
(4.41)

More detailed information about the SSV and its bounds can be found in [48], [52].

4.4.2 Robust Stability Condition

Consider and are stable linear systems. Then the closed-loop system
is stable for all allowed uncertainties with if and only if
, (4.42)

4.4.3 Robust Performance Condition

Consider a generalized plant as it is shown by block diagram in the Figure 4.14

79
 P

Figure 4.14 Generalized Plant

Let be the plant after controller has been connected (see Figure 4.15)

Figure 4.15 Generalized Plant with Controller

Channel is related to uncertainty, channel is related to performance.

Both of them should be minimized.
The set of uncertain plants is given as it is shown in the Figure 4.16
Assuming that uncertainties block has a diagonal structure condition for robust
performance can be turned into condition for robust stability by adding a fictitious
performance block (Figure 4.17):

Denote the modified uncertainties block as

80
 (4.43)

The condition for RP can be written as follows:

(4.44)

Figure 4.16 Uncertain Plant

Figure 4.17 Uncertain Plant with block

81
4.5 Robust Controller Design for METU TUAV

The task is to design a pitch and roll displacement autopilots that will bring the
airplane to a desired pitch and roll orientations. The control inputs to the system are
elevator deflection for a pitch controller, and ailerons deflection for a roll controller.
Design of the robust controller involves the solutions of Algebraic Riccati Equations
(ARE) as it had been described in section 4.3. Therefore, design of controllers for
METU TUAV is performed using MATLAB Robust Control Toolbox. The toolbox
enables to compute a stabilizing controller K for a given plant P. For design

purposes plant P should be partitioned in terms of the number of the disturbances,

the control inputs, errors, and the output measurements provided to the controller.
Partitioning of the plant and system interconnection is also performed with
MATLAB tools. Next step is to describe the uncertainties and selection of proper
weight functions, which would represent the uncertainties and performance
characteristics. The command “hinfsyn” solves the problem of robust controller
synthesis using the procedure given in section 4.3.
Design and simulation of the robust controller for pitch and roll autopilots is
performed using the linearized dynamics of the METU TUAV. Simulation through
the nonlinear environment is presented in Chapter 5, which gives the comparative
analysis of classical and robust controllers’ performances.

4.5.1. Choice of Weight Functions for Robust Controller

Consider the the block diagram of a closed-loop system as it is shown in Figure 4.18.
It is composed of weight function for multiplicative uncertainty of the actual

model, weight function for performance of the controller, for sensor noise,

a nominal model G, model error and controller K. It is assumed that the error is

82
not known, except that it satisfies the condition . For - synthesis control

system from Figure 4.18 can be expressed to a general plant shown in Figure 4.1.
Remind that the generalized plant P contains what is usually called the plant in a
control problem plus all weighting functions. The signal w contains all external
inputs, including disturbances, sensor noise, and commands, the output z is an error
signal, y is the measured variables, and u is the control input. The desired controller
K satisfies the nominal performance and robust performance.

+
ref output
K Act +
G
-
+

Figure 4.18 Block Diagram of Closed-loop System

First it is necessary to determine weight functions and in order to design the

controller K. It is known, high suppression ratio to the disturbance is required only in

the low-frequency range since only the low-order modes are strongly excited. On the
other hand, the error of the nominal system model is small in the low-frequency
range, but it can be large in the high-order modes since the high-order modes are

83
neglected. Weighting transfer function represents the multiplicative uncertainty,

it must be chosen such that the effect of the uncertainties of system’s parameters will
be represented in the best way. For the purposes of this work it is assumed that main
source of multiplicative uncertainty is in change of the parameters of the system. It
can be found from the Bode plots of the perturbed plants such that the Bode plot of
‘covers’ the Bode magnitude plots of all the plants with different combinations

of uncertainties. Weighting transfer function must be chosen such that the norm

, where S is the sensitivity function of the system. Having some

knowledge about the frequency characteristics of the Sensitivity function, weighting

transfer function is chosen as a low-pass filter.

4.5.2 Robust Controller Design for Pitch Attitude Autopilot

Block diagram of the closed-loop system is shown in Figure 4.18. Plant

represents a transfer function from the system’s input (elevator deflection) to the
system’s output (pitch angle) obtained from the linearized longitudinal dynamics of
the METU TUAV:

(4.45)

Reference input is a pitch input command to the system. System’s output is a

measured pitch angle.
As it was mentioned above, weighting function reflects system’s multiplicative
uncertainty. Consider the variation in system’s aerodynamics. For longitudinal
motion the most important aerodynamic derivatives that effect stability are static
longitudinal stability derivative and pitch-damping derivative . Therefore,

in order to determine a weighting function variations in and are

considered. 1% and 5% of uncertainty with respect to the nominal value of the

84
parameter is assumed. (4.11) gives an expression for weighting function that

characterizes multiplicative uncertainty. The Bode plot of is

shown in Figure 4.19.

Figure 4.19 Bode Plot of Uncertain Plants

According to this plot, had been chosen as

(4.46)

Its Bode plot is given in Figure 4.20.

85
 is chosen such that it would reflect frequency characteristics of the input signal

and satisfies the condition . had been obtained as:

(4.47)

Its frequency characteristics are shown in a Bode plot in Figure 4.21.

Figure 4.20 Bode Plot of

86
Weighting function represents effect of the sensor noise on the system’s output.

Assume 0.1% noise at low frequencies and 1% noise at high frequencies. Transfer
function of is obtained as:

(4.48)

Frequency characteristics of the weighting function are given in Figure 4.22.

Figure 4.21 Bode plot of

87
 Figure 4.22 Bode Plot of

[Link] Performance of the Robust Controller for Pitch Attitude Autopilot

Robust controller is designed using MATLAB “hinf” command, which computes the
controller. As a result the state-space model of the controller is obtained such
that the controller has 1 output, 1 input, and 8 states. Pitch angle response of the
nominal plant to the 1deg step command input, which is done by a linear simulation,
are given below in Figure 4.23. Deflection of the control surface that brings pitch
angle to the required value and deflection rate are illustrated in Figures 4.24, 4.25.
Step response characteristics of the closed-loop system with a nominal plant are the
following:
• overshoot: 2.89 %
• settling time: < 2.5 sec
• rise time: < 0.5 sec
• steady state error: < 0.4 %

88
Figure 4.23 Step Input Response of the Nominal Plant: Pitch Angle

Figure 4.24 Elevator Deflection

89
 Figure 4.25 Elevator Deflection Rate

[Link] Robust Performance Analysis

The test for robust performance is performed by “robustperf” command in

MATLAB, which gives information about the robust performance margin of the
uncertain system. If the input system is represented as uncertain state space model
(“uss” in MATLAB script), then an appropriate frequency grid is automatically
generated, and the robust performance analysis is performed on that frequency grid.
The performance of a nominally-stable uncertain system model is measured by the
input/output gain, which, in general, degrades (increases) for specific values of its
uncertain elements. Moreover, the maximum possible degradation increases as the
uncertain elements are allowed to further deviate from their nominal values. A
typical tradeoff curve between allowable deviation of uncertain elements from their
nominal values and the worst-case system gain is shown in Figure 4.26 (see
reference [53]).

90
 Figure 4.26 Generic Tradeoff Between Uncertainty Level and Performance

Robust performance analysis involves determining where the system performance

degradation curve crosses the dashed-line, the y=1/x hyperbola. Robust performance
margin, computed by “robustperf” command, gives the measure of the level of
the input/output gain degradation. The value of robust performance margin greater
than 1 implies that the input/output gain ( norm) of the uncertain system remains

less than 1 for all values of the the uncertain systems. A performance robustness
margin less than 1 means that certain values of the uncertain elements, within their
specified ranges, lead to an input/output gain greater than 1 ( see references [53],
[54]). “robustperf” provides the information about upper ([Link])
and lower ([Link]) bounds on performance margin and frequency at
which the minimum robust performance margin occurs, associated with upper bound
on performance margin; structure of values of uncertain elements associated with
[Link]. Another output arguments of “robustperf” command are

91
“Frequency vector” associated with analysis, cell array (same length as “Frequency
vector”), upper and lower bounds from “mussv”, structure of compressed data from
“mussv”. “mussv” is used to compute upper and lower bounds of the Structured
Singular Value.
Using the “robustperf” command for the closed-loop system shown in the Figure
4.18, for which plant is represented by the transfer function (4.45), the following
results are obtained:
Upper Bound: 0.9434
Lower Bound: 0.9434
Critical Frequency: 0.4921
“perfmargunc” is a structure of values of uncertain elements associated with the
hyperbola crossing. By substituting the values into the closed-loop system it can be
verified that this collection of values causes the closed-loop system norm to be
greater than or equal to the reciprocal of the performance margin upper bound:
perfmargunc = 1.0567
1/([Link]) = 1.0600.
Finally, plot of the bounds from “mussv” is illustrated in Figure 4.27.
The peak value is the reciprocal of the performance margin, and the frequency at
which the peak occurs is the critical frequency.

[Link] Robust Performance Analysis: Connection with Analysis

Consider the -analysis technique, described in the section 4.4. Extract the
( ) decomposition, and call “mussv” command on the appropriate channels of
M. In order to perform robust performance analysis a fictitious uncertain element
must be generated. This fictitious element is often referred to as the "performance
block." It should be a complex matrix-valued uncertain element with nominal value
of 0, and norm-bounded by 1.

92
 Figure 4.27 Bounds for Robust Performance

The block diagram of the uncertain plant with performance block is shown in

Figure 4.17. In MATLAB is created as “complexm” object. This element is

"wrapped" around the input/output channels of the system under consideration, and
then a robust stability analysis is performed. Since the rows of uncertain elements
correspond to the columns of M, and vice-versa, the row dimension of the
performance block should be the column (input) dimension of closed-loop system.
Generating the performance block (denoted as “PerfBlock” in the code) and closing
the input/output channels of the closed-loop system with it a modified closed-loop
system is created by “lft” command. The modified closed-loop system has
following characteristics: 13 States, 0 Outputs, 0 Inputs, Continuous System.
It should be noted, that the modified closed-loop system has 0 inputs and 0 outputs
(this is expected), but has dependence on the original uncertain elements, as well as
the new performance block.

93
“lftdata” command is used to separate the uncertain system into a certain system M,
in feedback with a normalized block diagonal uncertain matrix (call it “NDelta”).
The structure of the block-diagonal matrix is described by another output argument
(“BlkStruct”), which will be used as the block structure argument to “mussv”. “frd”
command is used to compute a frequency response of M. The same frequency range
as in the “robustperf” analysis. “mussv” bounds for M are generated and their
plot is shown in Figure 4.28. It should be noted that the plot is identical to the plot
obtained from the “robustperf” analysis.
The performance margin is the reciprocal of the structured singular value. Therefore
upper bounds from “mussv” become lower bounds on the performance margin.
Making these conversions the bounds and frequency associated with the upper bound
of the performance margin are obtained.

Figure 4.28 Bounds for Robust Performance

94
For comparison compare the lower and upper bounds from “robustperf” and
“mussv” as well as the corresponding critical frequencies are given in Table 4.1.

Table 4.1 Lower and Upper Bounds and Critical Frequency

“robustperf” “mussv”
Lower bound 0.934 0.934
Upper bound 0.934 0.934
Critical frequency 0.4921 0.4921

4.5.3 Robust Controller Design for Roll Autopilot

Consider the block diagram of a feedback control system as it is shown in the Figure
4.18. Plant G represents a transfer function from ailerons deflection to roll angle,
which is described by the following expression:

(4.49)

Reference input to the closed-loop system is a command roll angle. Output of the
closed-loop system is the measured value of the roll angle.
Procedure of determining weighting functions required for robust controller design is
similar to the procedure described above for the pitch autopilot. In order to derive
, consider 1% and 5% of uncertainty of system’s parameters that mostly effect

airplane stability. Airplane effective dihedral is of the primary importance for

lateral airplane stability. Another important aerodynamic derivative is roll-

95
damping . Recall (4.11) for the expression for weighting function . Magnitude

frequency characteristics of is shown in Figure 4.29.

Figure 4.29 Bode Plot of Uncertain Plants

is chosen as given below and its Bode plot is in Figure 4.30 :

(4.50)

Weighting function had been chosen in a similarly to the previous section and is

equal to:

96
 (4.51)

Its frequency characteristics are illustrated in Figure 4.31.

Weighting function is chosen the same as for the pitch controller (see Figure
4.22).

Figure 4.30 Bode Plot of

97
 Figure 4.31 Bode Plot of

[Link] Performance of the Robust Controller for Roll Autopilot

As a result of the design procedure, the controller with 1 output, 1 input, and 8 states
is obtained. To estimate performance of the robust controller used for roll autopilot
consider response of the nominal plant to a step input of 1 deg, which is represented
in Figure 4.32. Simulation is performed for a linear environment. Ailerons
deflection, which corresponds to system’s response, and deflection rate are shown in
Figures 4.33, 4.34.
Step response characteristics of the closed-loop system with a nominal plant are the
following:
• overshoot: 2.55%
• settling time: 1.02 sec

98
• rise time: < 0.5 sec
• steady state error: 0.2%

Figure 4.32 Step Input Response of the Nominal Plant: Roll Angle

99
 Figure 4.33 Ailerons Deflection

Figure 4.34 Ailerons Deflection Rate

100
[Link] Robust Performance Analysis

Robust performance analyses is performed in the same way as it as described in the

section [Link] for the pitch autopilot.
Applying the “robustperf” command for closed-loop system shown in Figure
4.18, where plant is represented by the transfer function (4.49) the following results
are obtained:
Upper Bound: 0.9974
Lower Bound: 0.9974
Critical Frequency: 0.5460
perfmargunc = 1.0008
1/[Link] = 1.0026.

Figure 4.35 Bounds for Robust Performance

101
Figure 4.35 illustrates the plot of the bounds from “mussv”. The peak value is the
reciprocal of the performance margin, and the frequency at which the peak occurs is
the critical frequency.

[Link] Robust Performance Analysis: Connection with Analysis

Using “mussv” algorithm, described in details in section [Link], the robust

performance is checked and compared to the results obtained in the previous chapter.
Modified closed-loop system block has the following characteristics: 13 States, 0
Outputs, 0 Inputs, Continuous System.
Plot of the “mussv” bounds is shown in Figure 3.36, which coincides with the one
given in Figure 4.35.

Figure 4.36 Bounds for Robust Performance

102
 Table 4.2 Lower and Upper Bounds and Critical Frequency

“robustperf” “mussv”
Lower bound 0.9974 0.9974
Upper bound 0.9974 0.9974
Critical frequency 0.5460 0.5460

The lower and upper bounds from “robustperf” and “mussv” as well as the
corresponding critical frequencies are given below for comparison in Table 4.2.

4.6 Command Filter Implementation for a Robust Control System

From Figures 4.25 and 4.34 is seen that deflection rate of the control surfaces is too
high for realization in the actual physical system. Saturation of the control surfaces
and will make controller performance worse. Using a command filter in the control
system architecture helps to avoid this problem. Detailed information and structure
of the command filter used in simulation is given in section 3.5 of Chapter_3.
The block diagram of the control system with a command filter implementation is
shown in Figure 4.37 below. For simulation of the controllers performances
linearized models of longitudinal and lateral dynamics are used. Simulation using
nonlinear UAV model in comparison to a PID controller performance is presented in
Chapter 5.

4.6.1 Robust Pitch Autopilot Performance with a Command

Filter Implementation

Consider the block diagram illustrated in Figure 4.37 .

103
 Filtered
Command Command
Input Error Output
Input
Command Robust G
Filter Controller
-

Figure 4.37 Basic Controller Configuration with Command Filter

The plant G as in Figure 4.37 represents a transfer function from elevator deflection
to pitch angle obtained from the linearized longitudinal dynamics of the METU
TUAV. Assume the input to the control system is as it is shown in Figure 4.38.
Response of the control system in terms of pitch angle to the given input is shown in
Figure 4.39. Elevator deflection and deflection rate are illustrated in Figures 4.40,
4.41.
From the simulation results it is seen that shaping the input command with a
command filter sufficiently decreases the rates of actuators making them to
correspond to the physical performance capabilities.

104
 Figure 4.38 Pitch Angle Command

Figure 4.39 System Response to a Pitch Angle Command

105
 Figure 4.40 Elevator Deflection

Figure 4.41 Elevator Deflection Rate

106
4.6.2 Robust Roll Autopilot Performance with a Command
Filter Implementation

For the roll autopilot he plant G in Figure 4.18 consistes of the transfer function
from ailerons deflection to roll angle obtained from the linearized lateral dynamics of
the METU TUAV. Consider the following command input to the closed-loop system
as it is shown in Figure 4.42. Figures 4.43, 4.44 and 4.45 give the roll angle response,
ailerons deflection and rate of ailerons deflection, respectively.

Figure 4.42 Roll Angle Command

107
Figure 4.43 System Response to a Roll Command

Figure 4.44 Ailerons Deflection

108
 Figure 4.45 Ailerons Deflection Rate

Simulation results give similar conclusions, as made for the pitch autopilot:
implementing a command filter to the control system improves its performance in
terms of actuators behaviour, mainly by decreasing their rates.

109
 CHAPTER 5

COMPARISON OF PID AND ROBUST CONTROLLERS

This chapter includes comparison of the controller designed using classical control
theory, and the controller designed by methods of robust control theory. Comparison
analysis of controllers’ performances is performed by nonlinear simulation.

5.1 Performance Issues for the Nominal Model

In this section controllers’ performances are compared through the nonlinear

simulation for a unit step input command. Simulation is performed using a nominal
model of the UAV, i.e. no parameter uncertainties is assumed to be present in a
system’s model.

5.1.1 Pitch Autopilot

Consider the response of the classical and robust controllers designed for a pitch
autopilot to a unit a step pitch angle command, which is represented by a pitch angle
response given in Figure 5.1 below.
Numerical values of performance characteristics of the PID and robust controllers
are listed in Table 5.1.
Change in elevator position and deflection rate for PID and robust controller for a
step pitch command is illustrated in Figures 5.2 and 5.3 respectively.

110
Figure 5.1 Pitch Angle Response to a Unit Step Input

Figure 5.2 Elevator Deflection for a Unit Step Input

111
 Figure 5.3 Elevator Deflection Rate for a Unit Step Input

Table 5.1 Performance Comparison of Controllers for Pitch Autopilot

Characteristic Required Result with CCT Result with RCT

Overshoot < 10 % 10 % 6%
Settling time, s <3 <2 <3
Rise time, s <1 < 0.3 < 0.5
Steady state error <3% < 0.4 % < 1%

5.1.2 Roll Autopilot

112
Consider response of the classical and robust controllers for roll autopilot to a unit
step roll angle command in terms of roll angle, ailerons deflection and ailerons
deflection rate, given in Figures 5.4 – 5.6, respectively.

Figure 5.4 Roll Angle Response to a Unit Step Input

Numerical values of performance characteristics of the PID and robust controllers are
listed in Table 5.2. Figure 5.4 shows that characteristics of the roll angle response for
the robust and PID controllers are satisfactory for the design requirements. However,
the actuators rate is high for the robust controller at initial moment of the input (see
Figures 5.6), which contradicts the performance capabilities of the physical system.
This problem is solved by implementing a command filter to the control system.

113
 Table 5.2 Performance Comparison of Controllers for Roll Autopilot

Characteristic Required Result with CCT Result with RCT

Overshoot < 10 % 6.4 % 6.8 %
Settling time, s <3 <2.5 3
Rise time, s <1 < 0.3 < 0.5
Steady state error <3% 0 0.6 %

Figure 5.5 Ailerons Deflection for a Unit Step Input

114
 Figure 5.6 Ailerons Deflection Rate for a Unit Step Input

5.2 Performance Issues for the Nominal Model with a Command

Filter Implementation

This chapter gives the simulation results of the PID and robust controllers for a
nonlinear model of the METU TUAV using a command filter implemented to the
control systems.

5.2.1 Pitch Autopilot

Response to the filtered command input for the PID and the robust controllers
designed for a pitch autopilot in terms of pitch angle, elevator position and elevator
deflection rate is shown in Figures 5.7 – 5.9.

115
 Figure 5.7 Pitch Angle Response

Figure 5.8 Elevator Deflection for the Command Pitch Angle

116
 Figure 5.9 Elevator Deflection Rate for the Command Pitch Angle

5.2.1 Roll Autopilot

Consider response to the filtered command of the PID and robust controllers
designed for the roll autopilot. Figures 5.10 – 5.12 illustrate change in roll angle,
ailerons position and rate of change, respectively.

117
 Figure 5.10 Roll Angle Response

Figure 5.11 Ailerons Deflection for the Command Roll Angle

118
 Figure 5.12 Ailerons Deflection Rate for the Command Roll Angle

5.3 Performance Comparison of PID and Robust Controllers for

Uncertain Plant

Consider the nonlinear model of the METU TUAV, which has uncertainties due to
variation in aerodynamic derivatives. As it have been mentioned in the previous
chapters, the most important aerodynamic derivatives that effect stability of the
vehicle are static longitudinal stability derivative and pitch-damping derivative

(for longitudinal dynamics), airplane effective dihedral and roll-

damping (for the lateral dynamics). Simulation is performed assuming 5% of

uncertainty with respect to the nominal value of the parameter. Simulations are
performed for pitch and roll angle inputs as they are shown in the Figures 5.13 and
5.16 for pitch and roll autopilots, respectively. The results of simulations are
illustrated in the figures below.

119
Figure 5.13 Pitch Angle Step Response of Uncertain Model

Figure 5.14 Elevator Deflection of Uncertain Model

120
Figure 5.15 Elevator Deflection Rate of Uncertain Model

Figure 5.16 Roll Angle Step Response of Uncertain Model

121
 Figure 5.17 Ailerons Deflection of Uncertain Model

Figure 5.18 Ailerons Deflection Rate of Uncertain Model

122
Simulations results illustrate, that both PID and robust controllers, designed for pitch
and roll autopilots, perform equally well at tracking the input signal for the nominal
plant, and for the plant, which has the uncertainties in aerodynamic parameters. Such
results are expected from the robust controller, which takes into account model
uncertainties in design algorithm. However, it is seen that the PID controller is also
able to handle with system uncertainties.

5.4 Performance Comparison of PID and Robust Controllers with a

Sensor Noise Presence

Consider the case when there is a noise, present in the system. Assume, that the
sensors are the sources of the noise (sensor noise). The simulation for comparison of
the classical and robust controllers is performed for the nonlinear nominal model.

5.4.1. Pitch Autopilot

In order to compare performances of the PID and the robust controllers, designed for
a pitch attitude autopilot, consider responses of the PID and the robust control
systems. Simulation includes responses of the classical and robust control systems to
the input, which is shaped by the command filter as it is given in the Figure. Assume
presence of the sensor noise with a frequency of 100 Hz and standart deviation of
0.5% of the output nominal value.

123
 Figure 5.19 Pitch Angle Response

Figure 5.20 Elevator Deflection Response

124
 Figure 5.21 Elevator Deflection Rate Response

Simulation results show that both PID and robust controllers show good performance
in tracking the input command. However, the PID controller is more sensitive to the
noise than the robust controller, it considers noise as a reference input and tries to
follow it. This is clearly seen from the position of the control surface, and its rate.
Minimization of the actuator’s efforts as well as taking sensor noise into account by
algorithm is performed systematically during the design of the controller.

5.4.2. Roll Autopilot

As for the pitch autopilot, consider responses of the control systems, which use PID
and robust controllers, to the input command shown in the Figure 3.19. Simulation is
performed for the presence of sensor noise with a frequency of 100 Hz and standart
deviation of 0.5% of the output nominal value.

125
 Figure 5.22 Roll Angle Response

Figure 5.23 Ailerons Deflection Response

126
 Figure 5.24 Ailerons Deflection Rate Response

According to the obtained results, which are shown in the figures above, robust
controller shows satisfactory performance in terms of roll angle response, ailerons
position and ailerons deflection rate. PID controller illustrates good input tracking,
however it is more sensitive to the sensor noise than the robust controller, which is
reflected in actuator’s behaviour: ailerons deflection and its rate for the PID
controller exceeds the corresponding values for the robust controller.

127
 CHAPTER 6

CONCLUSION

As a result of this work, the pitch and roll autopilots are designed for the METU
TUAV. Two different techniques are used for the autopilots design. PID controller is
developed by methods of the classical control theory and robust controller is
developed by the design technique. Performances of both types of controllers

for the pitch and roll autopilots are compared by simulations.

Nonlinear 6 DOF dynamic model of the UAV is developed by means of
MATLAB/Simulink. The UAV model includes aerodynamics, propulsion, mass-
inertia and environment models, and can be modified in terms of aerodynamic and
geometric properties of the vehicle. Linearization of the nonlinear model around the
specified trim condition is performed using the MATLAB Control Design Toolbox.
Characteristics of the UAV dynamics, referred to its longitudinal and lateral
dynamics, are established from the linear model and the open-loop response
simulations are performed.
Purposes of the pitch and roll autopilots are to follow the input commands. The
reason of designing controllers by different control algorithms is to compare which
type of the controller would provide better performance when the plant (actual
physical model) differs from the nominal model used for controller design. These
differences may occur as a result of linearizing the nonlinear dynamics of the UAV;
inaccurate aerodynamic model, which does not reflect the aerodynamic properties of
the UAV truly; neglecting presence of the sensor noise during the design of the
controller. Designing a PID controller, it is impossible to take into account the

128
uncertainties, which are present in the model parameters. The design technique,

which is used for the robust controller design, consideres the uncertainties of the
model’s parameters and the disturbances, which act upon the system. Two
approaches are used to establish robust performance of the robust closed-loop
system. The “robuststab” function gives the answer to the basic robustness
question: “is the closed-loop system stable or not?”. analysis is the mathematical
tool used to find the robust stability and robust performence margins. The “mussv”
function is used directly to compute the structured singular value , as a function of
frequency.
Simulation results provide comparison of the controllers performances. Simulations
are performed for the nominal model, for model that includes uncertainties due to
variation in aerodynamic derivatives, and for the case when sensor noise is taken into
account. It should be noted, that when sensor noise is present response of the
controller designed by technique is better than response of the control system
with a PID controller. The most efficient way to decrease sensitivity of the PID
controller to the noise is decreasing the derivative time . However, this effects the
response of the system by increasing overshoot and settling time. Therefore, a trade-
off decision should be made between system’s performance and its noise rejection
capabilities. Implementation of the command filter to the control systems has a
positive effect on their performances.
The following suggestions can be made for future work:
• implementation of the model into the open source code Flight Gear for a real-
time simulation
• verification of the dynamic model by the flight tests of the METU TUAV

129
 REFERENCES

[1]. Barak J. Carlson, “Past UAV Program Failures and Implications for Current
UAV Programs”, Maxwell Air Force Base, Maxwell Air Force Base, AL: Air
Command and Staff College; Air University Rep. No.037/2001-04.
[2]. Zak Sarris, “Survey of UAV Applications in Civil Markets”, Technical
University of Crete, 2001
[3]. Koldaev A., Sokut S., “Russia: Military Conservatism in Relation to Industry
UAV Initiatives”, UAV Systems, The Global Perspectives 2006/2007, p.100
– 104.
[4]. “UAV Control Systems Hunger for Compute Density”, COTS Journal,
January 2006.
[5]. Roskam, Jan, “Airplane Design”, Roskam Aviation and Engineering Corp.,
1985.
[6]. Kargın, Volkan, “Design of an Autonomous Landing Control Algorithm for a
Fixed Wing UAV”, Master Thesis, Ankara : METU, 2007.
[7]. “Caratteristiche Aerodinamiche Modello SCAUT”, Politecnico di Torino,
2001.
[8]. Andrievsky B.R., A.L. Fradkov, “Combined Adaptive Controller for UAV
Guidance”, Proc. European Control Conference ECC 2003, University of
Cambridge, UK.
[9]. Fu Xu, Zhou Zhaoying, Xiong Wei, “MEMS-Based Low-Cost Flight Control System
for Small UAVs”, Tsinghua Science and Technology, vol.13, #5, 2008.
[10]. [Link], [Link],”Autolanding Strategies for a Fixed wing UAV Under
adverse Atmospheric Conditions”, American Institute of Aeronautics and
Astronautics, Guidance, Navigation and Control Conference, Honolulu, HI,
USA, August 2008.

130
[11]. S. Kurnaz, O. Cetin, O. Kaynak, “Fuzzy Logic Based Approach to Design of
Flight Control and Navigation Tasks for Autonomous Unmanned Aerial
Vehicles”, Journal of Intelligent and Robotic Systems, 2008.
[12]. M. Sadraey, R. Colgren, “Robust Nonlinear Controller Design for a
Complete UAV Mission”, AIAA Guidance, Navigation, and Control
Conference and Exhibit, Keystone, Colorado, Aug. 21-24, 2006.
[13]. Etkin, Bernard, “Dynamics of Atmospheric Flight”, New York, Wiley 1972.
[14]. Peter H. Zipfel, “Modeling and simulation of aerospace vehicle dynamics”,
AIAA, 2007.
[15]. Jan Roskam, “Airplane Flight Dynamics and Automatic Flight Controls”,
Lawrence, DARcorporation, 1995.
[16]. Limbach Flugmotoren website, “[Link] as accurate of 15
July, 2009.
[17]. U.S. Standard Atmosphere, 1976, U.S. Government Printing Office,
Washington, D.C., 1976.
[18]. А. Aslanyan, “Automatic Control Flight Control Systems of Vehicles”, Kiev,
1984.
[19]. Kotwani, K., Sane, S. K., Arya H. and Sudhakar, K., “Performance Mapping
of Mini Aerial Vehicle Propellers”, Indian Institute of Technology, Bombay,
India, 2004.
[20]. National Imagery and Mapping Agency, Technical Report 8350.2, Third
Edition, Department of Defence World Geodetic System, 1984.
[21]. Hitay Özbay, “Introduction to feedback control theory”, CRC Press, 2000.
[22]. Charles L. Phillips, Royce D. Harbor, “Feedback control systems”, Prentice
Hall, 1991.
[23]. Ogata, Katsuhiko, “Modern Control Engineering”, Prentice Hall, 2002.
[24]. Peter H. Zipfel, “Modeling and Simulation of Aerospace Vehicle Dynamics”,
AIAA, 2007.

131
[25]. Sellers, David, "An Overview of Proportional plus Integral plus Derivative
Control and Suggestions for Its Successful Application and Implementation",
Retrieved on 2007-05-05.
[26]. Kok Kiong Tan, Wang Qing-Guo, Hang Chang Chieh, “Advances in PID
Control”, Springer, 1999.
[27]. Antonio Visioli, “Practical PID Control”, Springer, 2006.
[28]. Morari, Manfred, “Robust Process Control”, Prentice Hall, 1989.
[29]. “Avionics Fundamentals”, Jeppesen and United Airlines, 2006.
[30]. Aidan O’Dwyer, “Handbook of PI and PID Controller Tuning Rules”,
Imperial College Press, London, 2003.
[31]. Guillermo J. Silva, Aniruddha Datta, S.P. Bhattacharyya, “PID Controllers
for Time-Delay Systems”, Birkhauser, Boston, 2005.
[32]. Cheng-Ching Yu, “Autotuning of PID Controllers: A Relay Feedback
Approach”, Springer, 2006.
[33]. Mickle, M. C ., Zhu, J. J., “Simulation Results for Missile Autopilot Design
Based on Extended Mean Assignment”, System Theory, Proceedings of the
Twenty-Eight Southern Symposium, 1996.
[34]. George Ellis, “Observers in Control Systems: A Practical Guide”, Academic
Press, 2002.
[35]. Richard C. Dorf, Robert H. Bishop, “Modern Control System”, Pearson
Prentice Hall, 2005.
[36]. Karl J. Astrom, Tore Hagglund, “Advanced PID Control”, ISA-The
Instrumentation, Systems, and Automation Society, 2006.
[37]. Karl J. Astrom, Tore Hagglund, “Automatic Tuning of PID Controllers”,
Instrument Society of America, 1988.
[38]. G. Zames, “Feedback and Optimal Sensitivity: Model Reference
Transformations, Multiplicative Seminorms, and Approximate Inverses”,
IEEE Trans. Automat. Contr., vol. AC-26, 1981.

132
[39]. G. Zames and B.A. Francis, “Feedback, Minimax Sensitivity, and Optimal
Robustness”, IEEE Transactions on Automatic Control, AC-28:585–600,
1983.
[40]. Athans, M., “The Role and Use of the Stochastic Linear-Quadratic-Gaussian
Problem in Control System Design”, IEEE Transactions on Automatic
Control AC-16, 1971.
[41]. Safanov, M.G., “Stability and Robustness of Multivariable Feedback
Systems”, M.I.T. Press, Cambridge, 1980.
[42]. Grimble, M.J. and Johnson, M.A., “Optimal Control and Stochastic
Estimation: Theory and Application”, Vols 1 and 2, John Wiley & Sons,
Chichester, UK, 1988
[43]. B. A. Francis, “A Course in H-infinity Control Theory”, Lecture Notes in
Control and Information Sciences, vol. 88, 1987.
[44]. B. A. Francis and J. C. Doyle, “Linear control theory with an H-infinity
optimality criterion”, SIAM J. Control Opt., vol. 25, 1987.
[45]. J. C. Doyle, K. Glover, P.P Khargonekar, B. A. Francis, “State-space
Solutions to Standard and Control Problems”, AC-34:831–847,
1989.
[46]. D.C. MacFarlane, K. Glover, “A Loop Shaping Design Procedure Using
-synthesis”, AC-37:759–769, 1992.
[47]. K. Zhou, J.C. Doyle, “Essentials of Robust Control”, Prentice Hall, 1998.
[48]. K. Zhou, J.C. Doyle, K. Glover, “Robust and Optimal Control”, Prentice
Hall, 1996.
[49]. D.-W. Gu, [Link]. Petkov, M.M. Konstantinov, “Robust Control Design with
MATLAB”, 2005.
[50]. J. C. Doyle, B. Francis, A. Tannenbaum, “Feedback Control Theory”,
Macmillan Publishing Co., 1990

133
[51]. S. Ariki, “A New System Invariant and an Algebraic Proof of the Standard
Problem”, Conference Proceeding, 35th Conference on Decision and
Control, 1996
[52]. Alok Sinha, “Linear Systems : Optimal and Robust Control”, Boca Raton,
FL: CRC Press, 2007.
[53]. The MathWorks website, “[Link] as accurate of July
28, 2009.
[54]. “Robust Control Toolbox for Use with MATLAB”, User’s Guide, The
MathWorks, Inc., 2001.
[55]. William S. Levine, “The Control Handbook”, CRC Press, IEEE Press, 1996.

134
 APPENDIX A

AERODYNAMIC COEFFICIENTS AND DERIVATIVES

This section contains values of the aerodynamic coefficients and derivatives, which
are used for the aerodynamic model of the UAV. These data were obtained by the
wind tunel tests for the SCAUT UAV at the Politecnico di Torino, Italy. METU
TUAV is very similar to SCAUT UAV by geometrical and mass-inertia properties.
Therefore, these data are assumed to be suitable for the purposes of the thesis.
= 1.46

= 3.9

= 0.4275

=0

=0

= 0.0741

135
Figure A-1. vs.

Figure A-2. vs.

136
= -5.2

= -13.3

Figure A-3. vs. for Different Elevator Positions

= -0.03

= 0.13

= 0.019

137
 Figure A-4. vs.

Figure A-5. vs. for Different Rudder Positions

138
= -0.87

= 0.13

Figure A-6. vs.

139
 Figure A-7. vs. for Different Ailerons Positions

= -0.03

= -0.099

140
 Figure A-8. vs.

Figure A-9. vs. for Different Ailerons Positions

141
Figure A-10. vs. for Different Rudder Positions

142
 APPENDIX B

LINEARIZATION OF EQUATIONS OF MOTION

Linearization of the equations of motion is performed using the small disturbance

theory.
The scalar force and moment equations of motion in the airplane body-fixed axis
system:
(B-1.a)
(B-1.b)
(B-1.c)

(B-2.a)

(B-2.b)

(B-2.c)

Kinematic equations:
(B-3.a)

(B-3.b)

(B-3.c)
Assume the small perturbations on the states, forces and moments. The steady-state
values are denoted by subscript e (for equilibrium) and changes from them by the
prefix .
Motion variables:

143
 (B-4)

Forces:

(B-5)

Moments:

(B-6)

Carrying out these perturbation substitutions into the general airplane equations of
motion (1.1) and (1.2) result in:

(B-7.a)

(B-7.b)

(B-7.c)

(B-8.a)

(B-8.b)

(B-8.c)

144
Following approximations are applied:
, , ;

Force and moment equations can be written in the following form:

(B-9.a)

(B-9.b)

(B-9.c)

(B-10.a)

(B-10.b)

145
 (B-10.c)

The thin underlined terms in the equations above represent the general steady-state
equations of motion and assuming that they are inherently satisfied they can be
eliminated from without loss of generality. The double underlined terms are referred
to non-linear terms, which are assumed to be negligible compared with the linear
terms, therefore can be eliminated from the equations. The result is the following
equations:
(B-11.a)

(B-11.b)

(B-11.c)

(B-12.a)

(B-12.b)

(B-12.c)

Obtained equations represent coupled linearized equations of motion of an airplane.

Forces and moments are represented in terms of the stability derivatives. For
example, force in x-axis in terms of stability derivatives can be written as follows:

where , , ,

146
Equations of motion can be uncoupled and airplane longitudinal and lateral dynamics
under certain assumptions. States that are referred to the longitudinal dynamics are
, the input is elevator deflection . States, referred to the lateral

airplane’s dynamics are: , inputs are ailerons and rudder deflections:

. In these notations it is assumed that equilibrium values for angular rates

and body y-velocity are equal to zero, therefore:

.
State and input matrices for the longitudinal dynamics are:

State and input matrices that describe lateral dynamics are the following:

147
where

148