Air Force Institute of Technology

AFIT Scholar

Theses and Dissertations Student Graduate Works

5-1999

Close Formation Flight Control

Andrew W. Proud

Follow this and additional works at: https://scholar.afit.edu/etd

Part of the Navigation, Guidance, Control and Dynamics Commons

Recommended Citation
Proud, Andrew W., "Close Formation Flight Control" (1999). Theses and Dissertations. 5261.
https://scholar.afit.edu/etd/5261

This Thesis is brought to you for free and open access by the Student Graduate Works at AFIT Scholar. It has been
accepted for inclusion in Theses and Dissertations by an authorized administrator of AFIT Scholar. For more
information, please contact [email protected].
AFIT/GE/ENG/99M-24

CLOSE FORMATION FLIGHT CONTROL

THESIS
Andrew W. Proud
Captain, USAF

AFIT/GE/ENG/99M-24

Approved for public release; distribution unlimited

/tlfCt/U I6H
The views expressed in this thesis are those of the author and do not reflect the official
policy or position of the Department of Defense or the U. S. Government.
AFIT/GE/ENG/99M-24

CLOSE FORMATION FLIGHT CONTROL

Andrew W. Proud, B.S. Electrical Engineering

Captain, USAF

Approved:

Dr. Meir Pachter Date

Committee Chairman

Dr. John J D'Azzo Date

Committee Member
AFIT/GE/ENG/99M-24

CLOSE FORMATION FLIGHT CONTROL

THESIS

Presented to the Faculty of the School of Engineering

of the Air Force Institute of Technology

Air University

In Partial Fulfillment of the

Requirements for the Degree of

Master of Science in Electrical Engineering

Andrew W. Proud, B.S. Electrical Engineering

Captain, USAF

March, 1999

Approved for public release; distribution unlimited

 Acknowledgements

I would like to thank God and His Son, my Lord and Saviour Jesus Christ, who
has blessed me more than I deserve and made it possible for me to complete this work.
By believing in Him, I obtained eternal life. By trusting in Him, I obtained peace and
happiness. I am also grateful to my mom, Virginia, my brother Tim, and my sister
Kim for their strength and encouragement during the rough times when schoolwork kept
increasing, sleep kept decreasing and frustrations grew high. I would also like to thank my
thesis advisor, Dr. Pachter for his direction and guidance. Many times I doubted him; not
once did I prove him wrong, yet. I would also like to thank Dr. D'Azzo for his help in my
follow on assignment search and my thesis. Finally, I would like to thank my classmates;
without their help it would have been more difficult and not as much fun.

Andrew W. Proud
 Table of Contents

Page

Acknowledgements ii

List of Figures iv

List of Tables v

List of Symbols vi

Abstract vii

I. Introduction 1-1

1.1 Overview of Thesis 1-1

1.2 Background 1-1

1.3 Problem Statement 1-2

1.4 Research Objectives and Questions 1-4

1.5 Assumptions 1-4

1.6 Approach/Methodology 1-6

II. Model and Simulation Development 2-1

2.1 Aircraft/Autopilot Models 2-1

2.2 Equation of Coriolis 2-6

2.3 Formation Coordinate System . 2-8

2.4 Kinematic Equations 2-10

2.5 Formation Flight Controller 2-14

2.6 Summary 2-18

in
 Page
III. Upwash and Sidewash Effects on Formation 3-1

3.1 Upwash and Sidewash Derivations 3-1

3.1.1 Biot-Savart . 3-2

3.1.2 Average Upwash and Sidewash 3-5

3.1.3 Corrected Average Upwash and Sidewash 3-5

3.2 Calculation of Change in Lift and Drag 3-6

3.2.1 Change in Drag 3-6

3.2.2 Change in Lift 3-7

3.2.3 Change in Side Force 3-8

3.3 Modified Wing Aircraft Control System 3-9

3.3.1 Calculation of the New Formation Stability Derivatives 3-10

3.3.2 Modified Wing Aircraft Control System 3-12

3.4 Summary 3-13

IV. Control Design and Evaluation . 4-1

4.1 Complete System Model 4-1

4.2 Control Design 4-7

4.2.1 Gain Selection Method 4-8

4.2.2 Gain Selection . 4-8

4.3 Stability 4-15

4.4 Robustness 4-19

4.5 Complete Nonlinear Simulation Model 4-27

4.6 Summary 4-28

V. Performance Evaluation 5-1

5.1 Case Definition 5-2

5.2 Negative 30 Degree Heading Change 5-3

5.3 Positive 30 Degree Heading Change 5-9

IV
 Page
5.4 50 ft/sec Velocity Decrease 5-14

5.5 50 ft/sec Velocity Increase 5-19

5.6 1000 ft Altitude Decrease 5-23

5.7 1000 ft Altitude Increase . 5-28

5.8 Summary 5-32

VI. Conclusion 6-1

6.1 Conclusions 6-1

6.2 Recommendations for Further Study 6-2

6.3 Summary 6-3

Appendix A. Appendix A A-l

A.l Characteristic Data file A-l

A.2 Executable Program A-5

A.3 Linear Simulation Model A-16

A.4 Non-Linear Simulation Model A-21

Bibliography BIB-1

Vita VITA-1
 List of Figures
Figure Page

1.1. Horse Shoe Vortex 1-3

1.2. Formation Geometry . . . . . 1-5

2.1. First-Order Aircraft Models . 2-2

2.2. Comparison of First and Second-Order Responses 2-3

2.3. Second-Order Aircraft/Autopilot Models 2-4

2.4. F-16 Class Composite Flight Control System Model . 2-5

2.5. Inertial and Rotating Frames of Reference [3:Figure 1.7] 2-6

2.6. Inertial Reference Frame and Separation Distances 2-8

2.7. Wing's Rotating Reference Frame and Separation Distances 2-9

2.8. Relative Motion Diagram 2-11

2.9. FFC Structure 2-14

2.10. Dargan FFCS Block Diagram 2-16

2.11. Dargan FFCS Lateral channel . 2-18

3.1. Horse Shoe Vortex 3-1

3.2. Field Strength at Point P due to Filament A 3-2

3.3. View from above the two aircraft 3-3

3.4. Behind view showing WA field 3-4

3.5. Sideview of Wingman's wing lift rotation 3-6

4.1. Linearized System and Formation Flight Controller 4-4

4.2. 50 ft/sec Velocity Decrease Time Plot Using Dargan Gains 4-10

4.3. 50 ft/sec Velocity Decrease Lissajous Plot Using Dargan Gains .... 4-10

4.4. Negative 30 Degree Heading Change Plot Using Dargan Gains .... 4-11

4.5. Negative 30 Degree Heading Change Lissajous Plot Using Dargan Gains 4-11

vi
Figure Page
4.6. 1000 ft Altitude Increase Plot Using Dargan Gains 4-13

4.7. 1000 ft Altitude Increase Lissajous Plot Using Dargan Gains 4-14

4.8. A D z-slice at z = 0 4-20

4.9. Change in AD for a change in y evaluated at z = 0 4-20

4.10. A D y-slice at y = \b 4-22

4.11. Change in AD for a change in z evaluated at y = \b 4-22

4.12. A D at z — 0: 1st Order Approximation-Top Actual-Middle, 2nd Order

Approximation-Bottom 4-23

4.13. A Y z-slice at z = 0 4-25

4.14. Change in A Y for a change in y 4-25

4.15. A Y y-slice at y = \b 4-26

4.16. Change in A Y for a change in z 4-26

4.17. Formation Flight Control System Nonlinear Simulation Diagram . . . 4-28

5.1. -30 Degree Heading Change Without Coupling Time Plot ....... 5-6

5.2. -30 Degree Heading Change Without Coupling Lissajous Plot 5-6

5.3. -30 Degree Heading Change With Linear Coupling Time Plot 5-7

5.4. -30 Degree Heading Change With Linear Coupling Lissajous Plot ... 5-7

5.5. -30 Degree Heading Change With Non-Linear Coupling Time Plot . . 5-8

5.6. -30 Degree Heading Change With Non-Linear Coupling Lissajous Plot 5-8

5.7. 30 Degree Heading Change Without Coupling Time Plot ....... 5-11

5.8. 30 Degree Heading Change Without Coupling Lissajous Plot 5-11

5.9. 30 Degree Heading Change With Linear Coupling Time Plot 5-12

5.10. 30 Degree Heading Change With Linear Coupling Lissajous Plot . . . 5-12

5.11. 30 Degree Heading Change With Non-Linear Coupling Time Plot . . 5-13

5.12. 30 Degree Heading Change With Non-Linear Coupling Lissajous Plot 5-13

5.13. 50 ft/sec Velocity Decrease Without Coupling Time Plot 5-16

5.14. 50 ft/sec Velocity Decrease Without Coupling Lissajous Plot 5-16

Vll
Figure Page
5.15. 50 ft/sec Velocity Decrease With Linear Coupling Time Plot . . . . 5-17

5.16. 50 ft/sec Velocity Decrease With Linear Coupling Lissajous Plot . . 5-17

5.17. 50 ft/sec Velocity Decrease With Non-Linear Coupling Time Plot 5-18

5.18. 50 ft/sec Velocity Decrease With Non-Linear Coupling Lissajous Plot 5-18

5.19. 50 ft/sec Velocity Increase Without Coupling Time Plot .... 5-20

5.20. 50 ft/sec Velocity Increase Without Coupling Lissajous Plot . . 5-20

5.21. 50 ft/sec Velocity Increase With Linear Coupling Time Plot . . 5-21

5.22. 50 ft/sec Velocity Increase With Linear Coupling Lissajous Plot 5-21

5.23. 50 ft/sec Velocity Increase With Non-Linear Coupling Time Plot 5-22

5.24. 50 ft/sec Velocity Increase With Non-Linear Coupling Lissajous Plot 5-22

5.25. 1000 ft Altitude Decrease Without Coupling Time Plot .... 5-25

5.26. 1000 ft Altitude Decrease Without Coupling Lissajous Plot . . 5-25

5.27. 1000 ft Altitude Decrease With Linear Coupling Time Plot . . 5-26

5.28. 1000 ft Altitude Decrease With Linear Coupling Lissajous Plot 5-26

5.29. 1000 ft Altitude Decrease With Non-Linear Coupling Time Plot 5-27

5.30. 1000 ft Altitude Decrease With Non-Linear Coupling Lissajous Plot 5-27

5.31. 1000 ft Altitude Increase Without Coupling Time Plot .... 5-30

5.32. 1000 ft Altitude Increase Without Coupling Lissajous Plot 5-30

5.33. 1000 ft Altitude Increase With Linear Coupling Time Plot . 5-31

5.34. 1000 ft Altitude Increase With Linear Coupling Lissajous Plot 5-31

5.35. 1000 ft Altitude Increase With Non-Linear Coupling Time Plot 5-33

5.36. 1000 ft Altitude Increase With Non-Linear Coupling Lissajous Plot 5-33

A.l. Linear State Space Matlab Simulation Model A-16

A.2. Command Prefilter Simulation Model A-17

A.3. Lead Aircraft Flight Control System Model A-17

A.4. Perturbation+Nominal Block A-18

A.5. Linear Mixer Simulation Model A-19

Vlll
Figure Page
A.6. PI Controller Simulation Model A-20

A.7. Matlab Block Simulation Model A-21

A.8. Wing Aircraft Flight Control System Model A-22

A.9. Kinematic Motion Simulation Model A-23

A.10. Horshoe Vortex Disturbance Simulation Model A-24

IX
 List of Tables
Table Page

2.1. F-16 Class Composite Aircraft/Autopilot Parameters 2-4

2.2. F-16 Class Composite Aircraft/Autopilot Saturation Values 2-5

4.1. F-16 Class Aircraft Characteristic Values This data corresponds to the
following flight condition: Altitude of 45,000 ft, dynamic pressure 155.8
lb/ft2 4-6

4.2. Close Formation Stability Derivatives for y = |6 and z = 0 4-6

4.3. FFCS Gain Values 4-14

4.4. Closed Loop Eigenvalues 4-18

 List of Symbols
Symbol Page
Tfj, heading time constant 2-1

TV velocity time constant 2-1

TA altitude time constant . 2-2

Ri vector velocity in the i reference frame 2-7

Rp vector position as seen from the p reference frame 2-7

U3ip vector angular velocity of p with respect to i 2-7

Rp vector position in the p reference frame 2-7

V$L lead velocity with respect to the Wing aircraft 2-10

u>w angular velocity of Wing 2-10

R$L position of leader with respect to the Wing aircraft 2-10

V$ inertial velocity of the Wing aircraft 2-10

V™ inertial velocity of the Lead aircraft 2-10

V>JS heading error 2-12

V^ velocity of lead aircraft 2-12

VWc velocity command from PI controller 2-16

ij)Wc heading command from PI controller 2-16

hwc altitude command from PI controller 2-16

&.xE x error 2-16

AyE y error 2-16

AzE z error 2-16

KXp x channel proportional gain 2-16

KXl x channel integral gain 2-16

KYp y channel proportional gain 2-16

KYl y channel integral gain 2-16

KZp altitude proportional gain 2-16

XI
Symbol Page
Kzt altitude integral gain 2-16

ex longitudinal channel mixed error 2-17

kv velocity error signal gain 2-17

kx x separation error gain 2-17

ey lateral channel mixed error 2-17

k^ heading error signal gain 2-17

ky y separation error gain 2-17

ez vertical channel error 2-17

kz z separation error gain 2-17

W induced velocity 3-2

T vortex strength per unit length 3-2

rc distance from filament to point 3-2

b wingspan 3-2

b' corrected wingspan 3-3

y y unit vector 3-4

z z unit vector 3-4

Wuw Total Induced Upwash 3-4

x nominal separation distance in the x direction 3-4

y nominal separation distance in the y direction 3-4

z nominal separation distance in the z direction 3-4

Vsw Total Induced Sidewash 3-4

VSWavg Average Induced Sidewash 3-5

y' = I Non-dimensional y ....... 3-6

z' = | Non-dimensional z . . 3-6

ADW Change in Drag 3-6

q Dynamic pressure 3-7

S Surface Area of Wing 3-7

Xll
Symbol Page
ACJDW Non-dimensional coefficient of drag increment 3-7

ACLW Non-dimensional change in lift 3-7

AFY Change in Side Force 3-8

Vuw{y'iz') change in induced drag dimensionless expression 3-11

Osw{y',z') change in induced si deforce dimensionless expression 3-11

Xlll
AFIT/GE/ENG/99M-24

Abstract

In this research the close formation flight control problem is addressed. The forma-
tion consists of a lead and wing aircraft, where the wing flies in close formation with the
lead, such that the lead's vortices produce aerodynamic coupling effects, and a reduction in
the formation's drag is achieved. A controller, i.e., a formation-hold autopilot for the wing
aircraft, is designed such that the formation's geometry is maintained in the face of lead
aircraft maneuvers. In the formation flight control system, the wing and lead aircraft dy-
namics are coupled due to kinematic effects, and, in the case of close formations, additional
aerodynamic coupling effects are introduced. In the research these additional aerodynamic
coupling effects are properly modeled. The most significant aerodynamic coupling effect
introduced by close formation flight entails the coupling of the lateral/directional channel
into the altitude-hold autopilot channel. It is shown that formation hold autopilots de-
signed ignoring the aerodynamic coupling effects, yield satisfactory performance in close
formation flight.

xiv
 CLOSE FORMATION FLIGHT CONTROL

/. Introduction

1.1 Overview of Thesis

The introduction and review of previous research related to this subject are discussed
in Chapter I. The equations of motion describing the kinematics between the aircraft in
the formation as well as the flight control system equations for the aircraft are provided in
Chapter II. The derivation of the Lead aircraft's effects on the Wing aircraft is developed
in Chapter III. This includes derivations for upwash and sidewash effects caused by the
Lead's wing vortex on the Wing aircraft and the development of the modified Wing aircraft
flight control system. In Chapter IV the close formation flight controller along with the
complete simulation model is developed. An analysis of stability and robustness is pre-
sented. Simulations are conducted and a performance evaluation of results is presented in
Chapter V. Analysis, conclusions, and recommendations for further research are presented
in Chapter VI.

1.2 Background

Previous research has been performed by several AFIT graduate students on the
subject of formation flight control [1] - [12]. The motivation for their work was based
primarily on the needs of special operations forces (SOF). SOF missions require pilots to
fly many hours over long distances in a large formation usually in the dark at low altitudes
and at times in poor weather. These flying requirements combined with the complexities
of todays modern aircraft are very stressful and can lead to pilot saturation. A second
motivation for this research was the need to fly multiple unmanned aerial vehicles (UAV)
in a formation. These UAVs may be flying a surveillance mission or patroling a no-fly zone.
Current UAVs are limited in the number of payloads they can carry at one time. This
makes it necessary to fly multiple UAVs to obtain a complete set of data. A formation
controller offers two potential benefits. First, if used for close formations, it can reduce fuel

1-1
consumption of the formation; secondly, it can keep the aircraft in a known position with
respect to each other, making correlation of data easier for fusion from multiple sensor
payloads in multiple UAVs. In general, this could also help any formation of aircraft flying
long distances.

In previous work, the formation flight control problem was introduced and attention
was given to the kinematic coupling effects-see, e.g., [1]-[12]. Multiple versions of formation
flight controllers were designed and developed. These controllers consisted of both feedback
and feedforward designs as well as energy conserving designs. Simulations demonstrated
that these controllers work well for large formations in which the separations are more than
a wingspan. However, they were not designed for close formations, in which the lateral
separation between aircraft is less than a wingspan.

1.3 Problem Statement

An aircraft while flying causes spiral like disturbances in the air. Two disturbances
are projected backward perpendicular to it's wing and one is parallel to the wing. The
spiral disturbances are cylindrical in shape and dissipate slowly over time. The air in
the disturbances flows in a circular pattern starting at the tip of the wing, flowing up
circularly towards the body of the aircraft and continuing to circularly flow down back to
the wing tip. This leads to a clockwise flow in the disturbance created by the left wing
and a counter-clockwise flow from the disturbance created by the right wing, as observed
from behind. These disturbances created by the wing of the aircraft are often called a
horseshoe vortex, see Figure 1.1. The air flow directly between the two disturbances is
almost purely downwash. The flow of the air on the outside of the horseshoe vortex is a
mixture of up wash and sidewash. Aircraft flying into a vortex created by another aircraft
can experience substantial aerodynamic forces. The FAA has even developed a separation
criteria to reduce the danger of accidents from other aircraft flying into the wake caused
by aircraft landing and taking off [16].

Prandtl, a German scientist was one of the first to analyze the effects of these vortices
on other aircraft. It has been hypothesized that geese fly in their inverted 'V formation
to take advantage of the upwash created by the vortex from each leading goose. Another

1-2
aircraft flying in the vortex can, if positioned correctly, experience a reduction in drag and
an increase in lift [13]. Therefore, alternating aircraft in and out of the lead position can
potentially increase the range and endurance of a formation of aircraft. The formation of
the two aircraft is shown in Figure 1.2. The aircraft creating the vortex is the Lead and
the aircraft flying behind and into the vortex is the Wing. Indeed, the vortex shed by
the Lead aircraft induces an upwash on the Wing aircraft's wing, which is responsible for
reducing the Wing's induced drag.

The difficulty in taking advantage of this upwash for drag reduction by another
aircraft is the fact it requires very precise positioning with respect to the aircraft generating
the vortex. There is a small cylindrical window with a radius of about ten percent of the
wingspan in which a benefit can be achieved for the Wing aircraft [13]. However, these
vortices have substantial force and the flight control system needs to be able to reject these
disturbances and keep the aircraft stable and in the proper location.

t
Figure 1.1 Horse Shoe Vortex

1-3
 This thesis is a change in direction from the previous work on formation flight control.
The main problem focus in this research is to analyze the effects of close formation flying
on the wing aircraft's flight control system and develop a close formation flight controller
robust enough to handle these effects and maintain the close formation. This will primarily
include the need to analyze the vortex created by the Lead and determine its change in
strength based on a change in position in the formation. Secondly, the necessary modi-
fications to be applied to a formation flight control system to maintain a close formation
during flight must be determined.

1.4 Research Objectives and Questions

The primary objective of this research is to apply a formation flight controller, previ-
ously developed for large formation, to this problem in which a close formation is required.
Flight control system models previously developed will also be modified for use in this
research.

The primary question to be answered from this work is, "can a close formation flight
controller developed without due consideration of vortex effects control a close forma-
tion?". A secondary question is: "which is stronger, kinematic cross coupling or vortex
aerodynamic cross coupling?".

1.5 Assumptions

There are ten main assumptions listed in order of development. These assumptions
reduce the complexity and scope of the problem. This allows a focused study on the
research objectives. These assumptions do not negate the reasonableness of the results
achieved. An unlisted objective is to keep this research as realistic as possible. Hopefully,
the results can be applied to the development of an actual formation flight controller for
use in future aircraft and UAVs.

• Formation flight controller resides on Wing aircraft.

• Each aircraft is modeled as a point mass. This means that moment effects are not
considered.

1-4
 WING

Figure 1.2 Formation Geometry

• The aircraft flight control system is assumed to be composed of three independent

autopilots: Mach hold, heading hold, and altitude hold.

• Rotating coordinate frame is attached to the wing aircraft: x out the nose, y out the
right wing, and z down, see Figure 1.2.

• A sensor capable of measuring the separation distances as well as the velocity, heading
and altitude of the lead aircraft is attached to the wing aircraft.

• The Horseshoe vortex is modeled as two filaments originating near each wing tip and
proceeding backward for an infinite distance. The magnitude of the parallel filament
effect, that normally completes the horseshoe, is assumed to be negligible. These
filaments are shown in Figure 1.1.

• The disturbances created by the Wing on the Lead are ignored.

• The Wing aircraft is assumed to be in the most "optimal" position. "Optimal" means
that the Wing is in a position of maximum reduction in drag.

1-5
 • The wing flight control system is trimmed for straight and level flight at this "optimal"
position in the formation.

• The Lead maneuvering envelope is : +/- 30 degree heading change, +/- 50 ft/sec
velocity change, and +/- 1000 ft altitude change.

1.6 Approach/Methodology

Systems developed by previous AFIT students are used as the building blocks for this
research [11], [1], and [3]. These blocks include a Flight Control System (FCS) and Forma-
tion Flight Controller (FFC). The FCS model originally developed by Rohs [11] was a first
order model based on C-130 aircraft models provided by Lockheed. Buzogany [1] developed
a second order flight control system model through system identification techniques. This
second order model was reduced from a fourteenth order model that was developed based
on actual aircraft specifications provided by Lockheed. The complete flight control system
used in this research is composed of the first order Heading hold autopilot developed by
Rohs and the first order Mach hold autopilot and the second order Altitude hold model de-
veloped by Buzogany. These three autopilots constitute the complete flight control system
used for both the Lead and Wing aircraft. In his thesis [3], Dargan developed a FFC using
mixed feedback compensation. The controller was a three channel controller for controlling
the horizontal x and y separation distances and the vertical z separation distance in the
formation. The formation controller contained a proportional plus integral (PI) compen-
sator on each channel with a linear mixer for the inputs to the PI compensators. The
flight control system developed by Rohs and Buzogany and the formation flight controller
developed by Dargan are modified and used as the foundations for this research.

The research analysis begins with the development of a simulation using the first order
flight control system developed by Rohs and the formation flight controller developed by
Dargan. An initial simulation model is developed in Simulink [17]. This model is used
to verify the results achieved by Dargan. This ensures the flight control system model
and the formation flight controller are accurately modeled. Once confidence is gained in
the complete simulation model, the FCS models are adjusted to represent an F-16 class
aircraft. The derivation for the effects of the horseshoe vortex are developed for a close

1-6
formation. These effects are then integrated into the simulation as disturbances to the
wing and the FFC is assessed. The FFC is adjusted to obtain the best performance.

Simulations are performed using Matlab version 4.21 Simulink package by The Math-
Works [17]. Simulink allows use of simple blocks to form a frequency domain representation
of a complex system. The non-linear differential equations are integrated using a fourth
order Runge-Kutta technique and an Adams technique contained in Matlab Simulink.
Both have an error limiting capability and step size control to ensure accurate results of
integrations.

1-7
 II. Model and Simulation Development

This chapter outlines the models, kinematic equations, and basic controller used in
the nonlinear simulation. The majority of Section 2.1 has been taken from [12]. Models
and figures have been changed to reflect the use of an F-16 class aircraft instead of a C-130
class aircraft. Sections 2.2, 2.3 and 2.4 have been taken directly from [12]; however, some
figures have been modified slightly for the reasons stated above and a few paragraphs have
been added to provide additional information. Section 2.5 contains a description of the
controller developed by Dargan and walks through his controller design process. Sufficient
information is provided to enable the reader to reproduce the results of this research.

2.1 Aircraft/Autopilot Models

The development of the aircraft/autopilot models was accomplished by Rohs [11] and
Buzogany [1]. Since autopilot models were not explicitly available, models were obtained by
designing a custom autopilot system around C-130H aircraft models provided by Lockheed.
The autopilot system included Heading-hold, Mach-hold, and Altitude-hold autopilots.
Due to the inherent properties of aircraft autopilots, the models exhibited overdamped,
decoupled responses [11]. Using system identification techniques, first-order models were
developed using rate limited, decoupled, first-order differential equations. The first-order
aircraft /autopilot models are specified as:

V = --V+-Vc (2.1)
'V 'V

1 1
V> = —V + — i>c (2.2)
TV r,/,
1 1
h = h + —he (2.3)

where,

TV = heading time constant

r„ = velocity time constant

2-1
 Th = altitude time constant

The first-order aircraft/autopilot models are illustrated in Figure 2.1.

Figure 2.1 First-Order Aircraft Models

Second-order aircraft/autopilot models were developed by Buzogany [1] in order to

more accurately represent the "true" aircraft/autopilot system. The most disturbing prob-
lem with first-order models is the presence of instantaneous heading rate and vertical veloc-
ity changes. Figure 2.2 compares an overdamped second-order response with a first-order
response. While heading and altitude response was significantly improved using second-
order models, it was found that velocity response was modeled more precisely using a
first-order model with a larger time constant. Thus, the "second-order" models and time
constants are specified as:

V = (2.4)
'V 'V
1
i, = 1 \U;
—+ — 1 , 1 ,
(2.5)
T T
4>a i>J TipJi» 7
Va T1pb
1
h = T
+ —)h-
T
h. -\ 1 a,.
. (2.6)
K hJ ThaThb TKThh

2-2
 ■ First-Order Model
Second-Order Model

0123456789 10

Figure 2.2 Comparison of First and Second-Order Responses

There are multiple ways to model a second order system containing nonlinear sat-
urations caused by actuators. The models illustrated in Figure 2.3 were found to be the
best representation of a true aircraft response. Comparison of these models with models
not including saturations are almost identical for the envelope of maneuvers analyzed in
this research.

The complete flight control system is composed of Rohs' first order Heading hold
autopilot model and Buzogany's first order Mach-hold and second order Altitude-hold
autopilot models.

V = --V+-Vc (2.7)
1 1,1,
ip = —i>+— A (2.8)

h = - T— + — )h h+ he (2.9)
\ ha ThhJ ThaThb ThaThb

These models represented a C-130 class aircraft. The C-130 aircraft is a transport and
responds slowly in comparison to an F-16 class aircraft. Since Rohs' and Buzogany's
models are used in this research, the time constants and saturation values need to reflect the

2-3
 nominal
velocity

'<?■ <>

Figure 2.3 Second-Order Aircraft/Autopilot Models

Table 2.1 F-16 Class Composite Aircraft/Autopilot Parameters

Parameter Value
T$ 0.33 sec
rv 5 sec
Tha 0.307 sec
Thb 3.843 sec

performance characteristics for an F-16 class aircraft versus a C-130. First, this consisted
of simply halving the time constants of the C-130 models to speed up the system response.
This should be reasonable since an F-16 class aircraft responds much faster than C-130
aircraft. Secondly, the saturations were changed to match the maneuvering capability of
an F-16 class aircraft. In general the saturations are about two to three times greater than
those of a C-130 class aircraft. These new saturation values should also be reasonable since
the maneuvering envelope of an F-16 is much greater than that of a C-130. Tables 2.1 and
2.2 show the F-16 class aircraft time constants and saturation values used in this research
respectively. Figure 2.4 shows the complete F-16 class aircraft flight control system model
used in this research.

2-4
Table 2.2 F-16 Class Composite Aircraft/Autopilot Saturation Values
Parameter Lower Limit Upper Limit
Acceleration -10-^5
tic2 sec2
_6 £££ Qdeg
Turn Rate
Vertical Velocity -126^- 100^-
sec

nominal
velocity

^> <>

nominal
heading

-K> <)

Figure 2.4 F-16 Class Composite Flight Control System Model

2-5
2.2 Equation of Coriolis

In order to express the formation flight control problem kinematics, reference frames
must be established in the inertial and wing aircraft frames. Figure 2.5 shows the inertial
and wing aircraft frames and the angular rotation vector. The subsequent rotations caused
by maneuvering aircraft within the formation require a mathematical function relating
vectors in different, rotating frames. The Equation of Coriolis performs this function.

OJ,p

' V

/R

i3

ii

Figure 2.5 Inertial and Rotating Frames of Reference [3:Figure 1.7]

According to Blakelock [15], "The motion of an object as viewed from a reference

frame is equal to the motion as seen from the moving frame, plus the motion resulting from
the relative angular velocity of the moving frame with respect to the reference frame" [15].
This equation provides the basis for converting individual aircraft heading and velocity
into separation distances. This is an essential part of the design process. The Equation of
Coriolis is:

Ri — Rv + Wjp x Rp (2.10)

where

2-6
• Ri = the vector velocity of the point in the i reference frame

• Rp — the vector position of the point in question as seen from the p reference frame

• UiP = the vector angular velocity of the p reference frame with respect to the i
reference frame

• Rp = the vector position of a point in the p frame

2-7
2.3 Formation Coordinate System

The coordinate system used in this thesis is identical to that of Dargan [3], Buzogany
[1], and Reyna [9]. The analysis of the system kinematics uses two coordinate frames:

• Inertial base frame

• Rotating reference frame centered on wing aircraft

The base frame is an inertial North-East-Down system. For the purposes of this
research, the earth is considered to be flat and inertially fixed. The inertial reference
frame and separation distances are shown in Figure 2.6.

Xlmprtiai

Lead]

Ylriertial

Figure 2.6 Inertial Reference Frame and Separation Distances

The wing aircraft frame is centered on the wing aircraft. The x axis is in the flight
direction (i.e., aligned with the velocity vector), the y axis points out the starboard wing,
and the z axis points toward the earth. The x and y separation distances are measured in
the wing frame Figure 2.7.

2-8
Figure 2.7 Wing's Rotating Reference Frame and Separation Distances

2-9
2.4 Kinematic Equations

In order to simulate the kinematics associated with the formation flight control model,
kinematic equations must be derived. This has already been done by Dargan [3], Buzogany
[1], and Reyna [9]. Reyna's derivation is repeated here to ensure clarity (from [9], pages
3:6-10).

Using the Equation of Coriolis (Chapter II), the velocity of the lead with respect to
the wing has been found by Dargan as

V%L = VF-u%xR%L-V$+u%xK (2.11)

where the following convention is followed:

• The superscript indicates the reference frame.

• The subscript indicates the parameter described by the vector or a relation between
two parameters.

For example,

• VWL = velocity of the lead aircraft with respect to the Wing, in the Wing's reference
frame

• U™ = angular velocity of the wing aircraft in the wing's reference frame

=
• R-WL position of the Lead aircraft with respect to the Wing in the Wing's reference
frame

• V$ = inertial velocity of the Wing aircraft in its own reference frame

• Vjf = inertial velocity of the Lead aircraft in the Wing's reference frame

The development of the kinematic equations is based on the geometry defined in

Figure 2.8.

The following relationships are defined:

2-10
Figure 2.8 Relative Motion Diagram

V>£ 4>L -IpW (2.12)

0
w (2.13)
ÜJ-w 0

Rw
WL y
w (2.14)
yw

Vw
v$ 0 (2.15)
0

vL
vt 0 (2.16)
0

0
r/W
% 0 (2.17)
=
0

2-11
where,

ipE — heading error

V£ = velocity of the lead aircraft in its own reference frame

In order to solve Equation (2.11), V£ must be transformed into the wing reference
frame. A Direction Cosine Matrix (DCM), developed by Dargan [3], is used to perform
the rotation. Dargan found the DCM to be

cosipE —sintpE 0
Cf = sinipE cosipE 0 (2.18)
0 0 1

Vjf is found using by substituting (2.16) and (2.18) into the following equation

VLCOS1pE
L
v? = c?vL = VLsinipE (2.19)
0

Substituting (2.13)-(2.15), (2.17), and (2.19) into equation (2.11) yields

VLCOS1pB 0 xw Vw 0 0
V - 0 X yW ~ 0 + 0 X 0 (2.20)
WL — VLsintpE
0 ipw zw 0 4>w 0

VLCOStpE -i>wyw Vw
V™
WL VLsintpE tpwxw - 0 (2.21)
0 0 0

Separating (2.21) into scalar components yields

iff
= vLcosipE + ipwy - vw (2.22)

2-12
 yw = VLsinil)E - i>w% (2.23)
w
zLW = 0 (2.24)

Equations (2.22) - (2.24) describe the kinematics of the formation (x and y separations)
in terms of the individual aircraft's heading and velocity. The altitude (z) separations are
non-dynamic and are simply the difference between the aircraft altitudes.

2-13
2.5 Formation Flight Controller

Each aircraft is controlled through the inputs to the flight control system. These
inputs are the settings normally supplied by the pilot. The control strategy implemented
by Dargan was a two tiered structure as shown in Figure 2.9. The upper tier is for control
of the formation as a whole, and the lower tier is for control of the individual aircraft
within the formation [3]. The upper tier is used to control the guidance of the formation.
This consists of the velocity, heading, and altitude for the formation. The lower tier is
used to control the actual formation geometry. This consists of the x, y and, z separation
distances between the Lead and Wing aircraft in the formation.

Separation
Commands

Lead Aircraft/ Wing Aircraft/

Lead Commands Controller
Autopilot Models Wing Autopilot Models
Commands
^■isiii

x, y. z

Nonlinear SiiS^I^Sllfei:
Kinematics
P<.,\. H(lead) LOWKKIIIIR
COM-ICfU

UPPER TIER
CONTROL

Figure 2.9 FFC Structure

Dargan [3] separated the controller into three control channels: longitudinal, lateral
and vertical. The longitudinal or x-channel was composed of the longitudinal x separation
distance and the relative velocity between the lead and wing aircraft. The lateral or y-
channel was composed of the lateral y separation distance and the relative heading angle
between the lead and wing aircraft. The vertical or z-channel was simply the separation
distance between the lead and wing aircraft. In all channels the lead responds to the
formation level commands while the wing responds to the separation commands.

2-14
 Dargan iteratively improved the controller design in stages. First, he choose a series
of test cases to use for measuring performance between each design. This consisted of cases
for changes in the x, y, and z separation distances and cases for changes in the velocity,
heading, and altitude of the formation. Then, he analyzed the open loop response and used
this as the baseline for all subsequent designs. The open loop control response verified that
the system required some form of feedback. The formation maneuver commands could be
followed with zero steady-state errors, but the formation separation distances could not be
maintained.

Dargan then implemented state feedback for the formation commands of velocity,
heading, and altitude. A cascade compensator was not included in the feedback loop.
The longitudinal channel response for a change in the separation distance resulted in no
steady-state error. The response for a change in velocity resulted in a steady-state error in
the x separation distance. Therefore, the longitudinal channel required some compensation
due to the steady state error from a change in velocity. The lateral channel response was
similar to the response of the longitudinal channel. A change in y separation resulted in
no error. Heading change caused a steady-state tracking error in the y separation distance.
Again, some form of compensation was needed for the lateral channel as well. The vertical
channel had a steady-state error for both a change in vertical separation distance as well
as a change in altitude. Compensation was also required for this channel. Overall, state
feedback was not enough and some form of compensation was required for each channel.

Dargan next added a Proportional plus Integral (PI) compensator for control of
separation errors. This was composed of three independent PI controllers; one for each
channel. This controller resulted in a zero steady-state separation error for a change in
the separation distance and a change in formation maneuver for each channel. However,
the responses for the longitudinal and lateral channels had undesirable overshoots. Dargan
labeled this controller as the unmixed controller. The unmixed controller equations are

Vwcit) = (KXp)AxE + (KXl) f AxEdt (2.25)

Jo

4>wc(t) = {KYp)AyE + (KYl) I AyEdt (2.26)

Jo

2-15
 FORMATIOI«
MANUEVER
COMMANDS

W
LEAD
KINEMATICS
AIRCRAFT
-&

Ve

Figure 2.10 Dargan FFCS Block Diagram

hwc(t) = (KZr)AzE + {KZl) f Azh dt (2.27)

Jo

where,

• Vwc is the wing aircraft velocity command from the PI controller

• i>wc is the wing aircraft heading command from the PI controller

• hWc is the wing aircraft altitude command from the PI controller

• AXE, Aj/fi and AZE are the respective x, y, and z separation errors.

• Kx and Kx[ are the x channel proportional and integral gains

• Kyp and Kyx are the y channel proportional and integral gains

• Kzp and K%t are the wing aircraft z channel proportional and integral gains.

Lastly, Dargan developed a controller that used a mixture of separation errors and
formation errors. This was a mixed controller as opposed to the previous unmixed con-
troller. The mixed controller included a linear mixer cascaded with the PI controller as
seen in Figure 2.10. This controller worked the same way as any other PI controller. It at-
tempts to null out an error between the commanded input and the system response. The
unmixed controller attempted to null out the error between the commanded separation
distances and the actual separation distance. The linear mixer combines the separation

2-16
distance errors and the maneuver errors for each channel and feeds this mixed error signal
to the PI controller to be nulled out. The separation errors for the longitudinal and lateral
channels are replaced with the mixed error signals below

ex = kvVE + kxAxE (2.28)

eY = k^i>E + kyAyE (2.29)

ez = kzAzE (2.30)

where:

• ex is the longitudinal channel input to the PI controller

• kv is the gain on the velocity error signal, VE

• kx is the gain on the x separation error signal, AxE

• eY is the lateral channel input to the PI controller

• k^, is the gain on the heading error signal, ipE

• ky is the gain on the y separation error signal, AyE

• ez is the vertical channel input to the PI controller

• kz is the gain on the z separation error signal, AzE

The mixed control laws are

Vwc(t) = KXpex + KXl [ exdt (2.31)

Jo

i>wc{t) = KYpeY + KYl [ eYdt (2.32)

i Jo

hwc(t) = KZpez + KZl [ ezdt (2.33)

Jo

The linear mixer and PI controller are shown in Figure 2.11. The x- and y-channels
are identical except the VE term is replaced with the ij)e term and Aye replaces Axe.
The only difference between the z-channel and the x and y-channels is the z-channel does
not have an altitude error term since the z separation and the altitude difference are
synonymous.

2-17
 -& I 7®

Aye
^ ^H>>^
AzE
"a T®
LINEAR MIXER PI CONTROLLER

Figure 2.11 Dargan FFCS Lateral channel

Through iteration, Dargan determined the gains to achieve the best performance [3].
The longitudinal channel still had an overshoot/undershoot, but the magnitude of the over-
shoot/undershoot was reduced and the response was faster. The lateral channel resulted
in similar improvements. The magnitude of the undershoot/overshoot was reduced and a
faster time response was achieved. The mixed controller achieved the desired response and
concludes the controller designs developed by Dargan1.

2.6 Summary

In this chapter, the composite F-16 class flight control system for the Lead and the
Wing aircraft was developed. The large formation flight controller design process was
discussed. Finally, the mixed formation flight controller to be applied to this research was
developed.

Please note Dargan's kx and ky gains are negative for a negative feedback system model.

2-18
 III. Upwash and Sidewash Effects on Formation

This chapter outlines the method used to determine the upwash and sidewash created
by the Lead's vortex on the Wing aircraft. The upwash and sidewash are then used to
determine the new close formation stability derivatives. The new close formation stability
derivatives are then applied to the Wing FCS.

3.1 Upwash and Sidewash Derivations

An aircraft moving through the air creates vortices behind the wing. These vortices
can be seen in Figure 1.1 and are repeated here in Figure 3.1. These vortices exert aerody-
namic forces on the Wing aircraft in the formation. The formation geometry is determined
by the Lead's x, y, and z position relative to the Wing aircraft, in a rotating frame of
reference attached to the Wing as shown in Figure 1.2.
b

Figure 3.1 Horse Shoe Vortex

3-1
 Figure 3.2 Field Strength at Point P due to Filament A

3.1.1 Biot-Savart. This section outlines the method used to calculate the upwash
and sidewash created by the Lead aircraft on the Wing aircraft. The derivation of the effect
of the Lead aircraft's wing vortex on the wingman's wing is determined by using an analogy
with electric field strength produced by an electric current in electromagnetics. A vortex
filament from Fig. 3.1 is redrawn in Fig. 3.2. The angles ßi and /?2 determined by the
vortex filament segment [A' B'] and the point P are shown in Fig. 3.2. The fluid dynamics
analogue of the Biot-Savart law from electromagnetics states that the induced velocity
W from vortex filament A is given by

W (cos/?! — cos/32) (3.1)

47rr,

where the vortex strength per unit length is T, rc is the distance from the vortex filament
to the point P, and the velocity of the air wash caused by the vortex filament at point P
and out of the page is W. The unit vector $ is orthogonal to the radius vector rc. When
point A' is at -oo, the angle ßi — 0, and when point B' is at +oo, the angle ß2 = 7r.
Asssuming that the longitudinal separation in the formation is large, i.e., greater than two
wingspans, is tantamount to saying that point B' is effectively at +oo. When the actual
distance between the tail of the Leader and the nose of the Wing is greater than 2b, where
b is the wingspan, this effectively places point B' at +oo. Then the Lead aircraft's spanwise
contribution to the flow field near the Wing aircraft can be neglected. Similarly, the Wing

3-2
 WINGMAN

B
LEADMAN

Figure 3.3 View from above the two aircraft

aircraft's influence on the flow field near the Lead aircraft's wing can be neglected. These
are the main assumptions used in this research for the location of the Wing aircraft's wing
with respect to the vortices caused by the Lead aircraft's wing. The vortex goes back a
great distance before it is diffused, validating the assumption that point A' is essentially
at -oo. This reduces Eq. 3.1 to the simplified form of the Biot-Savart law,

W= (3.2)
27rrc

The wash vector, W, at the point P for the geometry of Fig. 3.2, is out of the page. This
would be the special case of no sidewash and all upwash at point P on the page.

Fig. 3.3 is a view from above the two aircraft which contains both the Lead aircraft
replaced by its horse shoe vortex approximation, with the sides of the horse shoe repre-
sented by the A and B filaments. The Wing aircraft is represented by its elliptical wing
approximation. The reduced wingspan of the Wing aircraft is represented by b'. This is
the corrected value for an elliptical lift distribution on a wing and is

(3.3)
* = T»

Fig. 3.4 shows the view from behind the two aircraft. The radius vectors, rc, from each
vortex filament to the wing, disregarding the x component due to the infinite length of the

3-3
 Figure 3.4 Behind view showing WA field

filaments, are for filaments A and B

rA = (y--^-v)y + (z- z)z (3.4)

rB = {y+ --y)y + (z- z)z (3.5)

where, y and z are the unit directional vectors. Based on Figures 3.4 and 3.5 and with a
bit of geometrical insight, it can be shown that the total induced upwash Wuw at station
(y,0) from both filaments emanating from the Lead's wing is

[(v-i- v\ [(y+bi- y\ (-z);-h'<y<h'

2TT
_ [y/(V-%-v)2 + *\ [y/(V + %-v)2 + *\
(3.6)
where x is the separation distance in the x direction, y is the separation distance in the y
direction, and z is the separation distance in the z direction. The total induced sidewash
Vsw at station (0,z) from both filaments is

[{z-zf] K* - m
VsW b y;0>z>-hz (3.7)
~2^ \J(y - iY + (-z- zy] [^(5 + *)» +(*_*)>

3-4
 3.1.2 Average Upwash and Sidewash. The average induced upwash on the Wing
aircraft's wing is thus calculated by integrating Eq. (3.6):

W,uwavg
(y-X-y) (y+hi-y) dy(-z) (3.8)
2TT6 / _y(y-bi-y)2 + z2\ y(y+bi-yy + z

After changing the variables to u — ((y - y - y)2 + z2) for the first term and similarly
u -
' — ((y + V y)2 + ^2) ^or *^e secon(
i term, and changing the limits to match with the
new variables u and u', the integral in Eq. (3.8) is evaluated as

f + z2 (y + b')2 + z
W,uwavg In In (-*) (3.9)
4irb' (y _ b')2 + Z2 y2 + z2

Similarly, averaging by integrating from the bottom of the tail at 0 to the top of the tail
at —hz yields the average sidewash, Vsw » at the vertical tail

Vsw.
f In (y -1)2 + # In (y+bi)2 + z2 (3.10)
47T/lz (y-^Y + (z + hz)2 (y2+!L)+(z + h,y

3.1.3 Corrected Average Upwash and Sidewash. A correction term, ß2, needs to
be included in both the numerator and denominator of each term contained in the natural
logarithms in Eqs. (3.9) and (3.10) in order to take into account physical viscosity effects
and to make the mathematical derivation more accurately follow the experimental data
[13],[14]. This is a dimensionless number and requires both the numerator and denominator
terms inside the logarithm to be non-dimensionalized. To make the terms non-dimensional,
both the numerators and denominators are divided by b2. The correction term, /J,2, is then
added to both numerators and denominators. The resulting equations are

</2 + z'2 + fi2 (y' + if + z'2 + »2

W,uwa
47T&
In
(y1 - \)2 + z'2 + M2
-In
y'2 + z'2 + n2 M) (3.11)

3-5
 A y Into page

Figure 3.5 Sideview of Wingman's wing lift rotation

and

Tr (</' - |)2 + *'2 + M2 (y' + |)2 + z'2 + M2

V,swa In In y (3.12)
4irh, (y' -1)2 + (* + V)2 + ^2 (y' +1)2 + (*' + ¥)2 + M2
where now the nondimensional formation geometry parameters are y' = | and z'

5.5 Calculation of Change in Lift and Drag

The upwash on the wing causes a change in the angle of attack of the wing. This
causes a rotation in the lift and drag vectors.

3.2.1 Change in Drag. In Fig. 3.5, V is the velocity of the aircraft, W is

the upwash, and V is the composite velocity of the air at the surface of the wing. The
original lift and drag vectors are represented by L and D and the rotated lift and drag are
represented by L' and D', respectively. It can be seen from Fig. 3.5 that the change in
angle of attack is
f
Aa = arctan \ftuw\\ _ \WUW\ (3.13)
V V
I
where the approximation for small angles has been applied. It can also be seen from Fig.
3.5 that consequently the change in drag, AZ>w,due to the rotation of the lift vector is

\W,uw\
AD™ = — Lw tan A« « — L w (3.14)
V

3-6
where again, the small angle approximation is used. Dividing both sides by the dynamic

pressure q and the surface area S of the wing, the non-dimensional coefficient of drag

increment ACr>w is

AD w LW\W,w\ r W,W
ACDw = — —^Lw~y (3.15)
qS qS V

Also, the vortex strength per unit length can be expressed as

yv*scLL 2S
r = pVb' JL
pV\b
_ 2
pV\b 7T&2
CLLVb=—CLLVb
KAR
(3.16)

where S = area of wing, b = wing span, V = airspeed of formation, AR = aspect ratio of

wing, and CLL = lift coefficient of the Lead aircraft. Substituting the above equation for

vortex strength and also substituting the previously derived upwash expression into the

change in drag equation, yields

AC#W - —— CLLLCL''W
,„ —- (3.17)
T2
xAR .n (y' -1)2 + z'2 + /*2 n
y'2 + Z,2 + fi2

3.2.2 Change in Lift. The change in lift, ACLw, is given by

r .
A-^Lw A
— &-otaw uw
= —^—%
\w \ (3.18)

where aw is the lift curve slope of the wing. Substitution of the upwash and vortex strength
values as before, results in

o^w_ _2_ ln y12 + z'2 + M

2
_ (y' + \ f + z'2 + M2
ACLv Ll 2 (3.19)
7rAfl 7T (2/'-f)2 + ^ + M2
,2 y« + 2'2 + [i2

This is the change in lift coefficient of the Wing from the upwash created by the Lead
aircraft.

3-7
 3.2.3 Change in Side Force. The sidewash created by the Lead also causes a
change in the force on the vertical tail. This change in side force, AFY, is

A E?
AF -o \Vsw\ -
Y = nqSvtavt-^—y (3.20)

where r\ is the aerodynamic efficiency factor at the tail, Svt is the vertical tail area, and
avt is the lift curve slope of the vertical tail. The non-dimensional change in side force
coefficient, is
\ri &vt \Vsw\ (3.21)
= r}~s~avt~y'

Inserting the sidewash expression of Eq. (3.12) back into the non-dimensional side force
Eq. (3.21) gives the side force coefficient

_Svt avt Ti
ACv »7- (3.22)
~S~~VÄ^hz'
(y'+l)2 + z"+fi2
In -In-
(y> - |)2 + (*' + V)2 + M2 (y' + |)2 + (*' + ¥)2 + ß2

Finally, substituting back in for T by using Eq. (3.16), the change in side force coefficient
becomes

1
r\SvtaviCLlb2_
ACv = (3.23)
■KA 2SK ■K

In (y' - |)2 + z'2 + »2 (y' +1)2 + z'2 + M2

In
(y' - |)2 + (z' + ^y + M2 (y' +1)2 + (z> + V)2 + M2

3-8
3.3 Modified Wing Aircraft Control System

It is envisaged that each aircraft is equipped with a flight control system that includes
three standard autopilots: Heading hold, Mach hold, and Altitude hold autopilots.The
aircraft autopilots that makeup the FCS are developed in Chapter II. The formation flight
control autopilot resides on the Wing aircraft. It is an outer-loop controller which receives
measurements of the Lead aircraft's position relative to the Wing aircraft and it drives
the reference signals of the Wing's three axes Mach hold, Heading hold, and Altitude hold
autopilots.

The formation flight control autopilot resides on the Wing aircraft. It is an outer-loop
controller which receives measurements of the Lead aircraft's position relative to the Wing
aircraft and it drives the reference signals of the Wing's three axes, Mach hold, Heading
hold, and Altitude hold autopilots.

For close formation flight, the Wing FCS needs to be modified to account for the
additional aerodynamic interactions created by the upwash and sidewash from the Lead
aircraft. The upwash causes an aerodynamic force in the x and z directions in the form of
a change in drag and change in lift as derived above. Thus, the Wing aircraft needs to be
retrimmed in pitch. The sidewash induces a force in the y direction caused by a change
in lift on the vertical tail, which requires the Wing aircraft's lateral directional control
channel to be retrimmed. This change in forces need to be trimmed out by the Wing's
FCS. Above and beyond this retrimming action, it is important to include the dynamical
change in forces caused by a perturbation Ax, Ay, and Az in the Lead's position in the
formation relative to the Wing.

The upwash and sidewash also cause changes in moments applied to the wing aircraft.
However, moment effects are not explored in this paper because of the rather rudimentary
modeling employed, where the Wing aircraft is considered a point mass.

Thus, new stability derivatives for change in Wing drag resulting from a change in
x, y, and z positions in the formation need to be determined for the Mach hold channel.
Similarly, the new stability derivatives for the change in lift due to a change in the x, y,
and z positions need to be determined for the Altitude hold channel. Finally, new stability

3-9
derivatives for the change in side force due to a change in x, y, and z positions in the
formation need to be determined for use in the Heading hold channel. The resulting Wing
FCSis

1 1
Vw = Vw + —VWc +
Tv TV

2- [ACDwxx + ACDwyy + ACDwzz]

1
/ 1 ,
Vw = Vw + —Vwc +

^y [&CYwxx + ACYwyy + ACYw,z)

I 1 , H; 1 , 1 ,
hw = — 1 nw hw H nw +
\na rhJ ThaThb rha
qS r
— [ACLwxx + ACLwyy + ACLwzz]
ill/

where: VWc is the reference signal to the Wing Mach-hold autopilot, and similarly, hWe is
the reference signal to the Altitude-hold autopilot and Vvc is the reference signal to the
Heading-hold autopilot. Vw is the Wing's velocity, if>w is the Wing's heading, hw is the
Wing's altitude, and x,y, and z are the perturbations in the Lead's position in the formation
relative to the Wing from the nominal location (x,y,z). In the Vw differential equation
the new stability derivatives for the longitudinal and the vertical perturbations, ACDwx,
ACDwy, and ACDwi, and ACLwx, ACLwy, and ACLv/t are multiplied by ff-. Multiplying
them by qS converts them back to a force, and then dividing by the mass converts them
to an acceleration, as required. The new stability derivatives for the heading differential
equation are multiplied by ^L. The additional division by the velocity is required because
the heading rate is an angular velocity. Also, the centrifugal acceleration is A = u> X V
where V is the aircraft's velocity and u is the angular velocity. Since all of these vectors
are orthogonal to each other, u = A/V. Hence, the new stability derivatives in the heading
equation, when multiplied by ^ are converted to an angular velocity, as required in the
heading equation.

3.3.1 Calculation of the New Formation Stability Derivatives. The change in

drag, lift, and side force on the Wing aircraft in a close formation have been previously

3-10
calculated. To determine the change in these forces due to a change in the x, y, and z
Lead's relative position in the formation, a linearization is performed about the nominal
Lead position (which is measured with respect to the Wing's position) in the optimal close
formation: The latter is y = jb and z = 0. This requires derivatives of the change in drag,
lift, and side force to be evaluated at these values for y and z.

First, a dimensionless expression is introduced for the change in induced drag, (Juwiy'■> %')'■

72
y" + z" +» (y' + f )2 + z'2 + M2
<ruw(y',z') In In (3.24)
(y' - f) + z'2 +ß2
2 y'2 + z'2 + n2

A similar non-dimensional function is defined for the sidewash component, aswiy'\z')

2 (y' - |)2 + z'2 + ji2 (y' + |)2 + z'2 + n2

osw(y',z') = In (3.25)
w (y> - |)2 + (z> + if)2 + ß
2
(y> + |)2 + (z> + hg-)* + ß
2

Based on these definitions, the change in lift, drag, and side force are expressed as

ACDv CLLCLWvuw(y',z') (3.26)

1
ACiv aCLwauw(y',z') (3.27)

1 r)Svtavtb
ACYw CLLo-sw(y',z') (3.28)
■KAR 2Sh

The only parts of the above equations that vary with a change in the x, y, and z position
are the auw(y',z') function for the change in drag and lift, and the aSw{y'-,z') function
for the change in side force. The derivatives of auw(y', z') and aSw(y', z') are analytically
obtained as follows: The partial derivatives for <7uw(y\z') are calculated as

do\uw i
= 0 (3.29)
dauw (3/8)TT
(3.30)
t(f) + M ][(f) + M2]
2 2 2

dauw i
= 0 (3.31)
dz' y'=^,z'=0

3-11
 and the partial derivatives for o~sw(y',z') are calculated as

d<?sw 1 „
(3.32)

»^, I(V)2
dy> '"'=^'=u [(f)2 + A*2][(|)2 + (V)3 + A*2]
3/^2
2V 6 2/
(3.33)
[(¥) + ^][(¥) + (¥)2 + /*2]
2

^| _ -(f)(V) (3.34)
Q„/ It/' —TI*'—u I7 7r\9 i ..9 i /'A.\ol 17 3» No , ..o , /fe.xol

Inserting the above derivatives into the change in drag, lift, and side force coefficients
expressions yields

ACDwx = ACLwx = ACYwx = ACDw, = ACLwz = 0

^Dw' - TTAH
LW
[(f)2 + /*2][(f )2 + M2] (3l35)

A6i- " Zl^" [(5)» + Aia][(f)a+A*a] (3,36)

l_r)Svtavth2
'■
ACy =
- I?ä;-456-
LA'
CL
(3-37)
1
L [(f) + M ] [(f)2 + (V)2+M2]
2 2
[(f)2 + M2] [(f)2 + (V)2+M2].

Ar = i vsvtavt (f)(V) .....

™ xAÄ 25 ^[(|)2 + /i2 + (^)2][(f)2 + Ai2 + (^)2] (3-38)

3.3.2 Modified Wing Aircraft Control System. Based on the above derived sta-
bility derivatives associated with the forces created by the upwash and sidewash, the new
Wing FCS is

Vw = _ Vw + —VWc + ^ACDwyy (3.39)

Ty Ty TO

hw = "( —+ —W hw +—!—hWc + ?-ACLwyy (3.40)

^ = -—V>w + — ^c + ^7[ACyWs2/ + ACyw^] (3.41)

'Vw 'Vw TOK

3-12
where ACDw , ACLw , ACYw , and ACYwt are the new close formation stability deriva-
tives.

3.4 Summary

The average upwash and sidewash effects from the Lead vortex on the Wing aircraft's
wing were developed. These effects were then used to derive the change in drag, change in
lift, and change in sideforce, AD, AL, and AY, respectively. New close formation stability
derivatives were developed and the Wing FCS was modified to account for the additional
aerodynamic forces exerted by the Lead's vortex on the Wing.

3-13
 IV. Control Design and Evaluation

The complete linear and nonlinear system models are developed in this chapter. The
method for determining the best gains for the close FFC is described. Finally, a stability
and robustness analysis is performed for the complete system using these gains.

4-1 Complete System Model

Equations (3.39) - (3.41) are adjoined to the nonlinear kinematics Equations (2.22) -
(2.24) where Equation (3.41) for ifi is inserted into the x and y separation differential equa-
tions, yielding the hybrid nonlinear six-dimensional close formation flight control system:

x = tpw -Vw + VLcos(ipL - Vv) + i'wc+ (4.1)

qS
y- ACYwy(y - y) + ACYwt(z - z)
mViw
X X
y = 4>w + VLsin(ipL -rj)W) T
iJ>Wc - (4.2)
Trpw tpw
qS
[±CyWy(v-y) + ACYwi(z-z)
mVw
Vw = -—Vw + —VWc + ^-ACDwv(y-y) (4.3)
Ty Ty Tfl

i>w = -—4>w + —i>wc + -^—[ACYwy(y-y) + ACYw,(z-z)] (4.4)

z = C (4.5)

C = -(— + —){--!—z + -^—hWo —hLc+ (4.6)

-(\Tha ThbJ ThaThb ThaThh ThaThh
qS
ACLwy(y-y)
m

where the vertical separation z = hw — h,L and the Wing and Lead aircraft are assumed
to have the same vertical dynamics.

The six states are x,y,ij}W,Vw,z, and (. The three control inputs to the respective
Wing's Heading hold, Mach hold, and Altitude hold autopilot channels are ij)Wc,VWc,
and hwc- The Lead's control inputs are viewed as a disturbance; thus, the disturbance
signals are tpL,VL, and hLc. This is the full nonlinear model used in the simulation. For
the purpose of controller design, linearization of the above hybrid system yields the linear

4-1
perturbation equations

x = ■—rßw -VW + VL+ -^-i>Wc + V^y \&CyWyy + ACYWzz (4.7)

mV
x ~-\ — x oS r
y V)ifor + VipL ipWc - x—r ACYwy + ACYwzz (4.8)
T^W J T^W mv
vw —Vw + —VWc + ?-ACDwyvy (4.9)
Ty TV m

■—i>w + —i>wc + ~ [&CYwyy + ACYwxz] (4-10)

'tpw ipw Tllv
z (4.11)

C = z + -V V + ^CLwyy - -±-hLc (4.12)

Tha ThbJ ThaThh ThaThb m ThTh.

The resulting state space representation based on the above denned states, controls,
and disturbances is

X X

Vw Vw
V\Vc vL
y y
dt
=A +B i>wc +r i>L (4.13)
4>w ipw
hwc flLc
z z
C C

where the dynamics matrix

0 -1 &±CYw,y —y-G
r
«v
0

0 L_
-^- : ^-ACD Wy 0 0 0
TV W

0 0 &A°y^ fc-?i)G S^ywtx 0

0 0 M.ACy^l/G -^- ^ACYwzl/G 0

0 0 0 0 0
0 0 1
^ACL w, 0 T
h*Thh
(4.14)

4-2
the input matrix
0

0 0

B = 0 -S-G 0 (4.15)
1
0 T
0
*W

0 0 0

0 n l

and the disturbance matrix is

1 0 0
0 0 0
0 VLG 0
r= (4.16)
0 0 0
0 0 0
n 0 T
haThb
1
J

and where the number G — 7r/180 and is used to keep consistency among units. The block
diagram for the above linearized system, Equations (4.14) - (4.16) is shown in Figure 4.1.

The dynamics A matrix shows that there is additional coupling introduced to the
system above and beyond the kinematically induced coupling. This is especially evident
in the A6i3 term which couples the y separation into the altitude hold autopilot and the
A3i5 term which couples the z separation into the heading hold autopilot. The heading
is kinematically coupled into the y separation equation and the y separation is aerody-
namically coupled into the Mach-hold autopilot, terms A3y5 and A\i3 respectively. This
leads to an indirect coupling between the z separation and the Mach-hold autopilot. There
is no coupling between the horizontal and vertical channels for a large formation. This
interaction between channels is one of the most interesting effects of close formation flight
and is discussed further in Chapter V.

4-3
 LINEAR PI
fr MIXER CONTROLLER

fr
$
FFCS

Ve

V^

Vw

fr
PLANT
(6 STATES) LEAD
(f AIRCRAFT

Figure 4.1 Linearized System and Formation Flight Controller

Typical characteristics for an F-16 class aircraft are listed in Table 4.1 for an altitude
of 45,000 ft, velocity of 0.85 Mach, and a dynamic pressure of 155.8 lbs/ft2. The close
formation stability derivatives for this flight condition and the associated characteristics
are listed in Table 4.2.

The dynamic A matrix evaluated for the typical characteristics listed in Table 4.1
and the nominal x, y, and z separations listed above, with no close formation induced
aerodynamic coupling effects, is

0 -1.0000 0 -0.3084 0 0
0 -6.0000 0 0 0 0
0 0 0 -13.6136 0 0
A= (4.17)
0 0 0 -0.7500 0 0
0 0 0 0 0 1.0000
0 0 0 0 -0.8447 -3.5118

4-4
The dynamic A matrix evaluated for the data listed in Table 4.1 and the nominal x, y,
and z separations with the close formation coupling effects included is

0 -1.0000 0.0057 -0.3084 -0.0020 0

0 -6.0000 -0.0471 0 0 0
0 0 -0.0144 -13.6136 0.0050 0
A= (4.18)
0 0 0.0138 -0.7500 0.0138 0
0 0 0 0 0 1.0000

0 0 0.4663 0 -0.8447 -3.5118

It can be seen from a comparison of Eqs. (4.17) and (4.18) that the coupling terms
introduced by the close formation are small. One notable exception is A6)3 = 0.4663. This
justifies the use of the large formation controller since this is essentially a three channel
system, although special care is required in the z-channel controller design.

4-5
Table 4.1 F-16 Class Aircraft Characteristic Values This data corresponds to the fol-
lowing flight condition: Altitude of 45,000 ft, dynamic pressure 155.8 lb/ft2

Wing Area S 300 ft2

Wing Span b 30 ft
Aspect Ratio AR 3
Lift Curve Slope a 5.3 per rad
Tail Area $vt 54.75 ft2
Tail Height K 120 in
Tail Lift Curve Slope a
vt 5.3 per rad
Aerodynamic Efficiency Factor V 0.95
Velocity Time Constant T
Vw 5 sec
Heading Time Constant T
i>w 0.75 sec
Altitude Time Constant ha 0.3075 sec
Altitude Time Constant hb 3.85 sec
Gross Mass m 776.4 slugs
Gross Weight w 25000 lb
Velocity V Mach = 0.85 or 825 ft/sec

Table 4.2 Close Formation Stability Derivatives for y = \b and z = 0

ACWy -0.000782
ACLwv -0.0077
ACW„ 0.0033
ACV„, -0.0011

4-6
4-2 Control Design

Similar to the large FFC system, note that the y and z channel states y, ipw, z, and £,
are decoupled from the x channel states, x, and Vw; which is easily seen in Equation 4.14.
Hence, one first designs a controller for the y and z channels, where the control signals
are ipWc and hWc, following which one turns to the design of the x-channel controller, viz.,
one synthesizes the control law for the VWc control signal. Furthermore, even though the
(y,z) model does not further decompose into individual y and z channels as in [8] where
the coupling is exclusively induced by the kinematics of the large FFC, the same controller
will be applied because the close formation induced coupling is weak.

The controller contains a linear mixer on the x and y channels and a standard PI
compensator on the x-, y- and z-channels. The FFC Equations 2.31 - 2.33 developed in
Chapter II are repeated below.

Vwc(t) = KXp[kvVE + kxAxE] + KXl f [kvVE + kxAxE]dt

Jo

i'wcit) = KYp[k^E + kyAyE] + KYl / [k^E + kyAyE]dt

Jo

hWc(t) = KZpkzAzE + KZl I kzAzhidt

Jo

The FFC gains have been choosen based on the three criteria listed below:

• The maximum positive and negative perturbations from nominal for a single channel
should be approximately equal for any heading maneuver within the test envelope
defined; e.g., the y separation for a -30° heading change ranges from a minimum of
20.5 ft to a maximum of 26.3 ft or -3 ft and 2.8 ft from nominal respectively. The
deviation to the right is approximately equal to the deviation to the left.

• The responses should be consistent for either a positive or a negative change in either
the heading, the velocity, or the altitude; e.g., a -30 degree heading change will cause
a maximum error of 4 ft in x and 3 ft in the y and a +30 degree heading change will

4-7
 cause a maximum error of 3.3 ft in the x and 3 ft in the y. Both maneuvers result in
errors of approximately the same magnitude.

• The Wing aircraft should not cross over the x or y axes. This criterion has the dual
effect of preventing collisions between the Lead and the Wing; also in a formation
containing both a left and right Wing aircraft.

Previous AFIT students [6],[12] used SISO techniques to determine the gains. The
method consisted of successively closing loops to create a new A matrix. After each loop
closure, the new A plant matrix is used for the next loop closure. Once the appropriate
loops were closed, the root locus was used to iteratively determine the best gains to achieve
the desired performance. Even though this method is not completely rigorous for a MIMO
system, it does give a good indication of gain magnitudes and system response.

4-2.1 Gain Selection Method. A simple systematic approach was used to itera-
tively determine the best gains. First, only the completely linearized system as shown in
Equations 4.7 - 4.12 and in Figure 4.1 is considered. With the cross coupling terms zeroed
out, Dargan's gains were chosen as the starting point [3]. The system is subjected to a
positive and negative velocity change and the horizontal channel gains are adjusted. The
system is then subject to a positive and negative heading change and the horizontal gains
are adjusted. Lastly, the system is subjected to a positive and negative altitude change
and the vertical channel gains are adjusted. The order of gain adjustment can affect the
number of iterations required to determine the best gains. The best order of gain adjust-
ment, i.e., the method which results in the fewest iterations, was to adjust the mixer gains
first, to achieve the best maneuver tracking response, and then adjust the PI compensator
gains to achieve the best formation response. Once the best gains are determined for the
complete linear system with no coupling, the next step is to add back the linear cross
coupling terms. Finally, apply these gains to the full non-linear model shown in Figure
4.17 and adjust the gains for the best response.

4-2.2 Gain Selection. The Six cases used for gain selection are: +/-50 ft/sec
velocity change, +/- 30° heading change, and +/-1000 ft change in altitude. Figures 4.2
and 4.3 show the responses of the system using the large FFC with Dargan's gains for a

4-8
50 ft/sec change in velocity. Figures 4.4 and 4.5 show the system responses for a -30°
heading change using Dargan's gains. Figures 4.6 - 4.7 show the response for a 1000 ft
increase in altitude.

The format for the time response plots and the Lissajous plots is described here.
The time plots contain six subplots. In the top three subplots, the solid line represents
the Lead aircraft and the dashed line represents the Wing aircraft. In some cases only a
solid line is visible, this is due to the overlap of the responses from the Lead and the Wing.
In the bottom three subplots, the solid line represents the nominal separation values and
the dashed line represents the actual values. Starting points in the Lissajous plots are
represented by an 'O' and the ending points are represented by an 'X'. This format is used
for each plot listed in the following sections.

As expected, Dargan's large FFCS gains resulted in poor performance for a close
formation. A velocity decrease of 50 ft/sec, Figures 4.2 and 4.3 resulted in a long delay
before the Wing's velocity responded to the Lead's velocity. This delay causes the Wing
to overtake the Lead, but since the Lead's wing and the Wing's wing overlap and both
aircraft are at the same altitude, a collision occurs. This is easily seen when the separation
in the x direction decreases to zero, crosses the y axis, and becomes negative. A -30°
heading change resulted in a maximum deviation of 4 ft (« |b) in the x direction and
a maximum deviation of 41 ft (« l\ b) in the y direction, see Figures 4.4 and 4.5. The
deviation in the y direction crossed the x axis. This would most likely have resulted in a
collision for the case of two Wingmen. The responses to an increase in altitude of 1000 ft
using Dargan's large FFCS gains is shown in Figures 4.6 - 4.7. The altitude also suffered
from a long delay before the Wing responded to the Lead's ascent. The separation in the
z direction grew to over 400 ft (« 13b). No collision would occur, but the enormous error
is still not acceptable for close formation flight. Therefore, all three maneuver responses
are unacceptable for close formation flight and require the gains to be adjusted.

4-9
 _850

fsoo
>
750
JP- _2Q_ -3fl- -40- 5Q_ 60
o
(D
3 0
'5j

4.5001 gur in -3P- jo 50 60

f 4.5
4.4999
100^ J0- _20_ 30 _4Q_ 5£L 60

£ Oh
-100
26°- JMQ_ -2P- _30 JO _50_ 60

§24
22
ÜL _20_ _J2Q_ -40- _5Q_ 60

£N o
10 20 30 40 50 60
Time (sec)

Figure 4.2 50 ft/sec Velocity Decrease Time Plot Using Dargan Gains

ou I I 1 1

60 11

40 ■

^ 20 - -
X

0- -

-20

4fl i i ' i
22.5 23 23.5 24 24.5 25
y(ft)

Figure 4.3 50 ft/sec Velocity Decrease Lissajous Plot Using Dargan Gains

4-10
 _826 1 1 1 1 1

|S25
> *s _ ^

824
) 10 20 30 40 50 60

K
oi O1 i i r
5.-20
•--40
)c10' 10 20 _J0 40 50 60
4.50011 1 1 1 I 1

r 4.5
CO

4.4999
651 )
10 20 30 40 50 60
i i

^60
X
\,'
55
501 )
10 20 30 40 50 60
1 1 1 t 1

£
>.
0 --""""" -
-50
) 10
i
20
i
30
i
_ 40
i
50i ..__ 60

& 0
N
i i i i i
-1
10 20 30 40 50 60
Time (sec)

Figure 4.4 Negative 30 Degree Heading Change Plot Using Dargan Gains

-20 -15 -10 -5

Figure 4.5 Negative 30 Degree Heading Change Lissajous Plot Using Dargan Gains

4-11
 With the system responses from Dargan's gains as the baseline, adjustment of the
x-channel was chosen first for the close FFC. Since, it is independent of the other channels
when only a velocity change occurs, it should be the easiest of the two horizontal channels
to determine the gains. Using the Dargan gain response as the baseline, the system was
subjected to a 50 ft/sec change in velocity. The velocity error gain (Ay) was increased
to get the Wing velocity to track closely with the Lead. Once, the best velocity response
was achieved the x separation error gain (kx) was increased to get the best x separation
response. Both of these gains reached a certain limit in which further increase resulted
in degraded performance. Once this point was achieved the x channel proportional gain
(KxP) was adjusted to give the best x separation response. This gain also reached a limit
in which further increase resulted in degraded performance. The x separation looked like
an elongated bell sitting on the line marking the nominal x value. This shape indicates
that the gain on the integrator was too small because the error continued to build up for
a long time and then took a long time to decay. Therefore, the x channel integration gain
(KXl) was increased. The combined effect of all gain adjustments resulted in an initial
ramp like increase in x separation error, but with a much smaller magnitude, then a quick
dropoff back to the nominal x separation distance. This was first done for an increase
in velocity. The gains were then readjusted to get the best average response for both an
increase and decrease in velocity to fulfill the second gain criterion.

These new gains provided the starting point for the next phase. This phase consisted
of inputting a heading change. This was a little trickier since heading changes affect both
the y- and x-channels simultaneously. However, applying the same approach as above
worked well. First, the heading error gain (Aty) was adjusted to track the Lead's heading
as closely as possible. The velocity and x separation responses degraded slightly from the
heading gain adjustment and required a slight re-adjustment of the velocity and the x
error gains (Ay and kx). The y error gain (ky) was then adjusted. Then the proportional
y channel gain (KYp) and finally the integral y channel gain (KYl) was adjusted. This
method worked well and was rather logical with respect to the system responses for each
step.

4-12
 _826r
|S25-
>
824L
-UL -29- 30 _40_ 50 60
-a
o
1?
s of-
(0

V[0 IfL 20 30 _4£L 50 Jß0

^4.6
(d
4.5
611 M- _2jQ_ _30_ JO 50 60

^60
X
59
261 -12- jp_ _3Q_ _40_ -SÖ- 60

§24
22
500( ) , i
10 20
i
30i 40i 50
i
60

0
cnn I I I I I
10 20 30 40 50 60
Time (sec)

Figure 4.6 1000 ft Altitude Increase Plot Using Dargan Gains

The adjustments of the z channel gains were almost trivial with respect to the x-
and y-channels. It required the adjustment of three gains, even though adjustment of the
linear mixer gain (kz) is the same as adding an equal gain to both the proportional (KzP)
and integral (Kz,) z error gains. First, the z error gain (kz) was increased to achieve a
good response. The proportional (KZp) gain was then increased to improve the response.
The result of these two adjustments was the best response and did not require any further
adjustment of the integral gain (KZ[)- SISO root locus methods would also have worked
well for determining the best gains for this channel.

The simulations were then run again using the nonlinear kinematics Equations (4.1)
- (4.7) and no cross coupling to determine if the gains determined in the previous run
needed further adjustments. Adjustments were not needed. Finally, cross coupling terms
were included; again, gain adjustments were not necessary. The final gain values are listed
in Table 4.3.

4-13
 Table 4.3 FFCS Gain Values
FFCS Numerical
Gains Value
Linear Mixer
ky 12.5
K^p 6
™x -8.0 sec~l
Ky -0.6 deg/ft
Kz 25
PI Compensator
KXP 6
KXl 0.4
KYp 11
KYl 0.9
KZP 4
KZl 0.5

Figure 4.7 1000 ft Altitude Increase Lissajous Plot Using Dargan Gains

4-14
4-3 Stability

In order to determine the stability of the closed loop system, the linearized plant and
controller must be combined together in an augmented closed loop plant. The eigenvalues
of this new system should indicate whether or not the complete linear system is stable. The
actual formation flight controller is composed of nonlinear saturation limits associated with
the FCS. For this reason, stability cannot be guaranteed, but for the operating envelope
defined, it is a good indication of system stability.

Since the FFCS consists of a linear mixer feeding into a three channel PI controller,
three more states are needed for the augmented matrix. The three augmented states are

a = kv(VL-Vw) + kx(xc-x) = ex (4.19)

ß = h (fa - V>w) + ky (yc - y) = eY (4.20)

7 = kz{zc-z) = ez (4.21)

where, a, ß, and 7 are not related to the angle of attack, the sideslip angle, or the flight
path angle. The new states are the outputs of the respective integrator for each channel's
PI controller as shown in Figure 2.11. The input commands, Equations (2.31) - (2.33) are
now solved using these new states. The resulting equations are

VWc = KXla + KXpkv(VL-Vw) + KPxkx(xc-x) (4.22)

lßWa = KYrß + KYpkftyL-^ + KYrkyfa-y) (4.23)

hWe = KZli + KZpkz(zc-z) (4.24)

Equations (4.19) - (4.21) are augmented to Equations (4.7) - (4.12) and the new
input Equations (4.22) - (4.24) are substituted into Equations (4.7) - (4.12). This creates
a new nine state system. The new inputs are xc, yc, and zc and the Lead is again treated
as a disturbance. The resulting state space representation based on the augmented states,

4-15
controls, and disturbances is

X X

vw vw
y y
ij)W tyw Xc VL
A. z = Ad z +B Vc +r i>L (4.25)
dt
C c . Ze
. tlLc
a a

ß ß
1 1

where the Act is arbitrarily broken into Aeli and Ac(2, and JBC(, and rc( matrices are,
respectively,
Ad = [AdiAd2] (4.26)
,KYpKy\
-1 y T ±(l + KYpK^)G
K
*w )
XpKx
rv -±(l + KXpKv) *fACDv
J
Wy 0
SKYP
0 0 (_* V+ T
£±)G
*v
0 0
AM = 0 0 0 0
0 0 J
£ACiv W« 0
— Kg; 0 0
0 0

0 0 0
(4.27)

3v*c?
^■y^^Y Wx 0 Uli--yG
0 0
as SKY.
r
Wz 0 ^G
KY,
^ACyWl(l/G) 0 T
Vw
Acl2 = 0 0 0 (4.28)

0 0

0 0 0 0
0 0 0

0 0

4-16
 y 0
KXpkx
TV
0 0
K k
0 X vp y
0

0 0
Br, 0 0 0 (4.29)
KzP
0 0

Kx 0 0
0 0
0 k,

1 T^w

KXpKx
TV
fy_^KyEk±\G
0 T
\ *w )
Kypk^,
0
T,= 0 0 (4.30)

0 0

Kv 0 0
0 0
0 0 0

where the number G = 7r/180 and is used to keep consistency among units.

4-17
 The evaluated Acl for the characteristics in Table 4.1 is

0 -1.0000 8.1481 -82.6581 -0.0020 0 0 1.1103 0

9.6000 -15.2000 0.0471 0 0 0 0.0800 0 0
0 0 -20.7490 196.0877 0.0050 0 0 -2.8274 0
0 0 19.8138 -201.0000 -0.0048 0 0 0 0
Ad = 0 0 0 0 0 1.0000 0 0 0
0 0 0.4663 0 -85.3131 -3.5118 0 0 0.4223
8.0000 -12.5000 0 0 0 0 0 0 0
0 0 0.6000 -6.0000 0 0 0 0 0
0 0 0 0 -25.0000 0 0 0 0
(4.31)

The eigenvalues for the above closed loop dynamics matrix Ac; are all in the left half plane
and are shown in Table 4.4. Therefore, the system is stable for the selected gains. It is also
evident from the system responses shown later in Chapter V that the system is stable.

Table 4.4 Closed Loop Eigenvalues

Eigenvalues
-0.6627
-14.47
-0.0667
-220.37
-1.69 + 9.06i
-1.69 - 9.06i
-1.29
-0.1256
-0.0818

4-18
4-4 Robustness

A robust controller is one that responds well to very dynamic plants. It was stated
in Chapter I that the window of opportunity for a reduction in drag and increase in lift
is small. Since, the change in drag, AD, and the change in lift, AL, are very complicated
3-dimensional functions, individual slices are shown for both the y and z directions. The
AD and AL, are proportional; the only difference is the magnitude of the AL is about ten
times greater than the AD. Therefore, all discussions for the AD function also apply to
the AL function.

Figure 4.8 shows a z-slice of the AD function at z = 0. Figure 4.9 shows the rate
of change in AD evaluated for z = 0 and y varied between 3.5 ft to 43.5 ft or -/+§& from
nominal, y = \b. It can be seen from these two figures that the nominal operating point
is highly dynamic. A slight movement, either to the right or to the left, can cause a large
rapid magnitude swing and even a sign change in AD. AD is almost symmetric with
respect to y = \b for small changes in y separation. The function loses it's symmetry as
the deviations from nominal exceed « -^b. This is easily seen in Figure 4.9 where the left
hand side forms a parabola and the right slowly decays to zero. Decreases in y position
greater than \b cause the Wing aircraft to move into the region in which the effect from
the Lead's vortex is downwash instead of upwash. This region is directly between the two
filaments composing the horseshoe vortex. However, the vortex never becomes a downwash
for an increase in y separation greater than the nominal, but slowly decays to zero upwash.

Figure 4.10 shows the same AD function, but sliced in the y direction at y = |6.
Figure 4.11 shows the rate of change in AD evaluated for y = \b and z varied between
-20 ft to 20 ft or -/+§& from nominal. It is obvious from these two figures that AD is also
very dynamic for changes in z, but unlike the z-slice it is perfectly symmetric. The radius
for a AD benefit is larger for the y-slice, about 55% of the wingspan.

4-19
 1.5-

0.5-

£-0.5-

-1 ■

-1.5-

-2-

-2.5-

10 15 20 25 30 35 40 45

Figure 4.8 A D z-slice at z = 0

0.08

0.06-

0.04-

0.02-

O
<

-0.02 -

-0.04 -

-0.06
10 15 20 ltf4 b 25 30 35 40 45
y=/b

Figure 4.9 Change in AD for a change in y evaluated at z = 0

4-20
 It can be seen from the z and y slices that the maximum reduction in drag is obtained
at the nominal y and z values. Deviations in the y and z positions from this nominal location
are compounded. So a combined change in y and z position in the formation results in a
larger reduction in AD than for an individual change of greater magnitude in either the y
or the z positions.

Figure 4.12 shows the actual AD function and the first and second order models for
the region of +/-j^b in which AD is roughly symmetric. It can be seen, that the first
order model truly does trivialize the effect of the upwash. A decrease in the y separation
distance causes a small decrease in the AD; an increase in the y separation causes a small
increase in the AD. However, the true response is more like a parabola. A decrease in
y separation causes a decrease in AD, but it is much faster than the linear version. An
increase in the y separation also causes a decrease in the actual AD, but the first order
model causes an increase in AD for an increase in y. The second order model is parabolic
like the actual and appears to be better than the first order model for small positive and
negative changes in y, but for larger changes the first order model is actually closer to the
true response as the second order model becomes increasingly negative. The first order
model is truly only accurate for changes in y of 0.25 ft in magnitude. It can also be seen
from Figure 4.12, that the first order model is not constant, but has a small slope. The
maximum reduction in drag is not exactly at the point y = \b or 23.562 ft, but at 23.612
ft. This shift is caused by the inclusion of the correction term ß. There is no effect on the
first and second order AD models for a change in z, because z = 0 at the nominal position.

4-21
 1.5-

0.5-

-0.5
-20 -15 -10 10 15 20 25
z(ft)

Figure 4.10 A D y-slice at y = \b

0.08

0.06

0.04

0.02

ä
„Q 0 - ■ '■ '■ :: • • • - -

-0.02

-0.04

-0.06

-20 -15 -10 -5 0 5 10 15 20 25

zsz'b

Figure 4.11 Change in AD for a change in z evaluated at y = Va

4-22
 2.5

1.5-

Q
<

0.5-

21.5 22 22.5 23 23.5 24 24.5 25 25.5 26

y(ft)

Figure 4.12 A D at z = 0: 1st Order Approximation-Top Actual-Middle, 2nd Order

Approximation-Bottom

The change in side force, AY function is also a complicated three dimensional surface.
Figure 4.13 shows a z-slice of the AY function at z = 0 and Figure 4.14 shows the change
in side force for a change in y position evaluated at z = 0. From these figures, it is evident
that the AY curve is also dynamic, but the highly dynamic region is removed from the
nominal operating point. Recall, the nominal separation distances were choosen to achieve
a maximum reduction in drag not side force. For this reason, the AY is less dynamic than
the AD near the nominal operating point. However, the function is not symmetric for
equal positive and negative changes in y location. A reduction in y separation distance
of more than \b results in a strong increase in negative side force. The side force slowly
decays to zero for increasing y separation.

Figure 4.15 shows the same AY function, but sliced in the y direction at y = \b.
The partial derivatives for AY with respect to a change in z are shown in Figure 4.16.
It is again obvious in these figures that the nominal position is removed from the highly
dynamic region. A change in z separation will have only minor impact on AY. The

4-23
magnitude changes caused by a change in z are much less severe than those caused by a
change in y.

4-24
 1 1

-2

r3 -
JO
i \

>-

-5

-6

_7 1 u
10 15 20 jt/4 b 25 30 35 40 45
y

Figure 4.13 AY z-slice at z = 0

0.4

0.3-

0.2-•

,^ 0-

-0.1 ■

-0.2-

-0.3-

-0.4
10 15 20 25 30 35 40 45
ys/b

Figure 4.14 Change in A Y for a change in y

4-25
 0.8-

0.6-

0.4-

0.2-

£ 0-
>
<
-0.2 h

-0.6-

-0.8-

-20 -15 -10 0 5 10 15 20 25

zft

Figure 4.15 AY y-slice at y = \b

0.01

0.005 -

0-

-0.005 - ■

-0.01

^-0.015 ■

-0.02 ■

-0.025 ■

-0.03 ■

-0.035 -

-0.04
-20 -15 -10 -5 0 5 10 15 20 25
z=z'b

Figure 4.16 Change in A Y for a change in z

4-26
 Overall, it was seen from Figures 4.8 - 4.16 with the exception of Figure 4.12, that the
vortex strength changes drastically from location to location and changes very quickly, but
the magnitude of the cross coupling due to the vortex is much smaller than the magnitude
of the kinematically induced coupling. This is easily seen from the linearized system
dynamics A matrix in Equation 4.18. The Aii4, A3i4, and A4|4 terms are the kinematic
induced coupled terms. The Aii3). A2i3, A3i3, A4i3, A5]3, A6i3,Aii5, A3|5, and A4|5 terms
are the aerodynamic cross coupling caused by the Lead vortex. Except for A6]3, all the
vortex terms compared to the kinematic cross coupling terms are quite small. Therefore,
the kinematic coupling has a stronger influence on the robustness of the controller than
does the vortex induced aerodynamic cross coupling. The formation flight controller was
previously shown to be robust enough to handle the kinematic coupling [3]. Hence, the
FFCS designed by Dargan is quite robust with respect to the minor disturbances caused
by the vortex, as well.

4-5 Complete Nonlinear Simulation Model

The full linear system was developed in the beginning of this chapter. This was
needed for the design of the close formation flight controller. Once the close FFC is
developed using the linear system, it needs to be tested on the actual nonlinear system.

The nonlinear system is developed by replacing the first order close formation stabil-
ity derivatives in Equations (3.39) - (3.41) with the actual coefficients for ACDw, ACLw,
and ACYw shown in Equations (3.15), (3.18), and (3.21) and adjoining them to the non-
linear kinematics Equations (2.22) - (2.24). The full nonlinear system is:

x = ■—ifar -Vw + VLcos(^L -i>w) + —i>wc + y-^ACYw (4.32)

x x öS
V = i>w + VLsin(ipL - ipw) i\)Wc-x——ACYw (4.33)
T^W r^w mvw
Vw = -—Vw + —VWc + ^-ACDw (4.34)

i>w = rfar + V%c + -KT-&CYW (4.35)

z = C (4.36)

4-27
 1 1
c = - —
Th
+—
T
c a hb
-z +
l~haThb
T
haThb
-HWc T
KThb
-HLc + —ACLv (4.37)
m

where the vertical separation z = hw - hL and the Wing and Lead aircraft are assumed
to have the same vertical dynamics and Equation (4.35) for tp is inserted into the x and
y separation differential equations as in the linear version. This full nonlinear model and
the nonlinear model with the linear close formation stability derivatives, Equations (4.1)
- (4.7) are used in the performance evaluations in Chapter V. The full nonlinear model,
Equations (4.32 - 4.37), is shown in Figure 4.17.

LINEAR
-^ MLXER
PI
CONTROLLEI
'&

Zc
FORMATION:
Yc
*G
SEPERATTONi
COMMANDS;
: Xc
*G
-*G Vjy
AL
, ^c KINEMATICS -£» HORSESHOE AD WING
FORMATION I ¥i LEAD AIRCRAFT
MANUEVER i X, AIRCRAFT \ VORTEX
V
AY
COMMANDS ■ hL

. i , i

Figure 4.17 Formation Flight Control System Nonlinear Simulation Diagram

4-6 Summary

In this chapter, the complete six state model was developed and the method for gain
selection was described. The linear closed loop system was developed and a stability anal-
ysis was conducted for the closed loop system. The aerodynamic coupling was evaluated
and shown to be weak in comparison to the coupling created by the kinematics, thereby
justifying the use the large formation flight controller. Finally, the large formation flight
controller was shown to be robust enough to handle the aerodynamic coupling created by
the vortex.

4-28
 V. Performance Evaluation

This chapter describes the complete nonlinear simulation model, including a de-
scription of the command prefilter. A description of the format for the response plots is
provided. Performance evaluations for six cases are performed: The cases evaluated are
+/- heading changes, +/- velocity changes, and +/- altitude changes. Each case is subdi-
vided into three runs. The system response for each case is discussed and comparisons are
made between cases.

The simulation model is shown in Figure 4.17. The Lead and Wing Aircraft blocks
are represented by the linear F-16 class aircraft FCS developed in Chapter II. The FFCS
block is represented by the model developed in Chapter IV. The kinematics block is
modeled using the fully non-linear Equations (2.22) - (2.24) outlined in Chapter II. The
Horseshoe Vortex block is actually composed of two models described in Chapter III. One
is the linear model developed in Equations (3.35) - (3.38). The other is the fully non-
linear version developed in Equations (3.15), (3.18), and (3.21). The coupling terms are
appropriately dimensionalized and then added as disturbances to the Wing flight control
system. Both the linear cross coupling version and the non-linear vortex version are used
in the following simulations. A command prefilter is contained in the Formation Maneuver
Command block. It processes the input step commands so they follow more closely the
commands input into an actual aircraft autopilot. This basically ramps the commands into
the aircraft FCS instead of inputting an instantaneous step change. This allows the errors
to grow gradually instead of instantly changing from zero error to 100 percent error and
prevents the immediate saturation of the aircraft flight control system. Saturation of the
FCS for long periods can result in instability. The complete Matlab Simulink simulation
model is shown in Appendix A.

The Lead and Wing aircraft are flying in a close diamond formation with x = 60ft
and y = 23.562/i or 2b and |ö, respectively and z = 0. This formation was shown
previously in Figure 1.2 in which the Lead is on the right side and in front of the Wing.
Typical F-16 class aircraft characteristics are listed in Table 4.1 and the evaluated formation
stability derivatives are listed in Table 4.2.

5-1
5.1 Case Definition

Each case consists of three runs: no cross coupling, linear cross coupling, and non-
linear cross coupling. There are cases for both positive and negative change of equal
magnitude. Comparison of the positive and negative cases show the difference between
maneuvers in which the Lead is moving away from the Wing and when the Lead is moving
toward the Wing. The positive and negative heading plots can also be used to determine
the response in the event there are two Wingmen; one on each side of the Lead.

Long distance flying usually entails flying from waypoint to waypoint until the final
destination is reached. Usually waypoints are defined by radio beacons or airports. Be-
ginning at the initial airport of takeoff, the pilot dials in the autopilots' velocity, heading,
and altitude to reach the next way point. Upon reaching the next waypoint, the pilot
enters the new velocity, heading, and altitude values into the autopilot to proceed to the
next waypoint and so on until the final destination is reached. Usually, the paths followed
are relatively strait. This means that large heading changes between waypoints will not
be encountered. The simulation cases were chosen based on this scenario and should be
representative of typical autopilot changes implemented by a pilot as he sets the autopilot
to proceed to the next waypoint.

The format for the time response plots and the Lissajous plots is the same as described
in Chapter IV and is repeated below. The time plots contain six subplots. In the top three
subplots, the solid line represents the Lead aircraft and the dashed line represents the Wing
aircraft. In some cases only a solid line is visible, this is due to the overlap of the responses
from the Lead and the Wing. In the bottom three subplots, the solid line represents the
nominal separation values and the dashed line represents the actual values. Starting points
in the Lissajous plots are represented by an 'O' and the ending points are represented by
an 'X'. This format is used for each plot listed in the following sections.

5-2
5.2 Negative SO Degree Heading Change

The formation is commanded a heading change of negative 30° in which the Lead
aircraft turns into the Wing aircraft. Figures 5.1 and 5.2 show the time and Lissajous
response plots for the linearized system without cross coupling. Even with the prefilter in
place there is a slight saturation of the Lead heading response. This is not significant and
more importantly the Lead is considered a disturbance so it truly doesn't matter whether
it is saturated or not for this analysis. The Wing does not experience any saturation in
heading response.

The Wing heading follows the Lead heading so closely that the two overlap. However,
the Wing's velocity varies to follow the heading change of the Lead. The Wing must
decrease it's velocity, since the Lead is turning into it. This causes the separation in the
x direction to initially decrease as expected. The y separation initially increases and then
undershoots the nominal and slowly increases and levels off at the nominal y separation.
The formation separation reduces to a minimum 56 ft in the x direction or 4 ft from
nominal. The closest the Lead comes to the Wing in the y direction is 20.6 ft or 3 ft from
nominal. In fact the maximum positive and negative deviations in the y direction are 3 ft,
meeting the first gain criterion. The x-channel settles within 12 seconds and the y-channel
settles within 60 seconds. The formation spacing is maintained within approximately 3 ft
(1/10 b) in both the x and y directions and zero steady state tracking errors are achieved
after the heading change is complete. The z separation stays at zero for the duration of
the maneuver.

Figures 5.3 and 5.4 are the response plots for -30° heading change with linear cross
coupling. The most obvious and most interesting difference between this run and the
previous run without cross coupling is the affect of the cross coupling on the z-channel.
Specifically, the cross coupling of the y separation into the altitude hold autopilot through
the A63 term. The linear cross coupling causes an oscillation in the vertical channel. It is
a very small oscillation with a maximum deviation of 3 x 10-3 ft and a minimum deviation
of 5 x 10~3 ft and is nulled out by the PI controller. The initial increase in y separation
causes an increase in the Wing's lift causing it to ascend above the Lead. It is apparent
from this response that the change in z separation follows the linear first order AL function

5-3
exactly. The AX first order model decreases for decreasing y separation and increases for
increasing y separation.

Figures 5.5 and 5.6 are the response plots for -30° heading change with non-linear
cross coupling. The effect of the non-linear cross coupling is very different than that of
the linear coupling. The vertical channel perturbations are less graceful and greater in
magnitude than that of the run with linear coupling. The x and y separation plots are
also different than the previous two runs.

The minimum x separation increases from -4 ft to -3.85 ft. The maximum x deviation
increases from 0 ft to 0.1 ft. Basically the entire x response plot just shifted up by 0.1
ft. This is due to the change in drag from the change in y separation. The linearized AD
function, Figure 4.12 did not accurately represent the true change in drag for changes in
y separations greater than 0.25 ft in magnitude from the nominal y separation.

The y separation also changed. The positive and negative deviations for the previous
two runs were 3 ft and -3 ft, respectively. In this run, the positive and negative deviations
are 2.3 ft and -2.2 ft, respectively. This is a change of 0.7 ft and 0.8 ft, respectively from
the previous runs. The error improved in both the positive and negative directions by
about the same amount. Recall from Figure 4.13, the nominal point rests in the middle of
the knee of the curve where the slope is changing slowly. In fact, the change in side force
magnitude is almost equal for changes in y of +/-3 ft. The increase in y separation causes
the sideforce to decrease causing the Wing to turn away from the Lead. Since, the Lead
is turning into the Wing this speeds up the Wing's heading response leading to a reduced
error in the y separation. The smaller error means the controller does not have to apply
as much correction as in the previous two runs resulting in a smaller undershoot as the
controller pulls the y separation toward nominal.

The z separation decreases initially as y increases. As the controller decreases the

y separation towards the nominal y, the z separation begins to increase. When the y
separation crosses the nominal, the z separation briefly crosses the z axis becoming slightly
positive and then quickly decreases again as y continues to decrease. The y separation then

5-4
slowly increases and levels off when it reaches the nominal y value. The z separation slowly
increases overshoots and then starts to decrease and level off at the nominal z separation.

There are four primary pieces to the z plot. The first piece is the first negative peak.
Since the Wing is sitting at the point of maximum lift in the vortex, any deviation causes a
decrease in AL. This is reversed from the linear coupling run in which the initial increase
in y separation caused an increase in z separation or an increase in AL. The crossing of the
z axis is interesting. The response in this area demonstrates how quickly the magnitude
of the AX changes for relatively small changes in separation between aircraft. Since the
changes in z separation are so small, the AD, AL, and AFY forces are dependent on the
change in y separation only.

5-5
 826

«824
>
822
40- _2p_ _SQ_ -40- _5Q_ -6.0
o
S-20
'35
°--40
gir JJL J>0_ JQ_ _40 _5Q_ 60
4.5001
| 4.5
a
4.4999
65?- 40- _2Q_ _3Q_ JO _5Q_ 60

£-60
55
30" JJL _20_ _30_ 40 _50 60

|25t
20
1°- 10- _2Q_ _30_ JO 50 60

S o

10 20 30 40 50 60
Time (sec)

Figure 5.1 -30 Degree Heading Change Without Coupling Time Plot

60.5

Figure 5.2 -30 Degree Heading Change Without Coupling Lissajous Plot

5-6
 _826
fs24
>
822
J£- _2{L _30_ _4Q_ -50- 60
(D
S.-20I-
°--40
£nr 10- _20_ 30 JO- -5P_ 60

OS

65' 10_ 20 JO JO 50_ 60

X *> _ /
55
30( -J0- J>£_ 30 _ML _50_ 60
&25
>.
20
tf10'
51 -W- 20 _3Q_ _4Q_ 50_ 60
^ 0
N
-5
10 20 30 40 50 60
Time (sec)

Figure 5.3 -30 Degree Heading Change With Linear Coupling Time Plot

60.5

Figure 5.4 -30 Degree Heading Change With Linear Coupling Lissajous Plot

5-7
 ^826r—

■824^T
822
3 0' -Q- -30- jß_ 50 60
o>
5.-20
'55
°--40 1
ft ° in -20- 30 -4L _50_ 60

4.5
I. /

65' 1SL -20- 3Q_ -4L 5Q_ 60

55
301 HL -29- -30- M- -50- 60

|25fc

20
0.021 JUL -2A- -30- _40 5Q_ 60

§ 0
I /
-0.02
10 20 30 40 50 60
Time (sec)

Figure 5.5 -30 Degree Heading Change With Non-Linear Coupling Time Plot

60.5 r -j j_ -j j_

Figure 5.6 -30 Degree Heading Change With Non-Linear Coupling Lissajous Plot

5-8
5.3 Positive 30 Degree Heading Change

The positive 30° heading change without coupling is shown in Figures 5.7 - 5.8. This
again causes a slight saturation in the Lead heading response. The two headings overlap,
however the velocity increases instead of decreasing as in the previous case. In this case,
the Lead is turning away from the Wing causing the Wing to play catch up. Since the Wing
is on the outside it needs to cover a longer distance than the Lead in order to maintain the
formation. This causes the Wing to briefly increase it's velocity. The response in the x and
y plots is exactly the opposite of the -30° heading change maneuver. Instead of decreasing
as in the -30° case, the x separation initially increases. The y separation initially decreases,
then overshoots and slowly decays to nominal. The maximum separation in the x direction
is 63.3 ft or +3.3 ft from nominal. It can be seen from the plots, that the closest the two
aircraft come to one another in the y direction is 20.6 ft and the farthest is 26.5 ft, i.e.,
approximately -3 ft and +3 ft from nominal, respectively. As expected, there is no change
in the vertical channel.

The positive 30° heading change with linear coupling is shown in Figures 5.9 - 5.10.
The primary difference between the plots with linear cross coupling versus those without
cross coupling is again the effect on the vertical channel. The cross coupling causes an
initial oscillation in the vertical channel that is almost exactly the opposite of the previous
case, -30° heading change with linear coupling. The PI controller nulls the perturbation
out after about 60 sec. The perturbation in the z-channel is again caused by the linking
term A6ß and responds exactly as expected from the AL first order model.

The non-linear coupling plots are shown in Figures 5.11 - 5.12. The x separation
error is 0.1 ft larger than in the previous two runs; 3.4 ft versus 3.3 ft. As in the -30° case
with non-linear coupling, the first order model does not accurately represent the changes
in magnitude of drag for changes in y greater than +/-0.25 ft.

In the case of the negative and positive 30° heading changes with linear cross coupling,
the results were identical except flipped in the z direction. This is not true for the non-linear
cases. The z separation responses for both a positive and negative 30° heading change are
remarkably similar. The initial decrease in y causes a decrease in z. As y begins to increase

5-9
the z separation begins to increase. The z separation briefly crosses its nominal as y crosses
its nominal and then starts to decrease again as y continues to increase. The y separation
then decreases slowly back to nominal. The z separation also increases back to nominal,
but overshoots slightly. The first z crossover is caused by both the controller attempting to
null out the error and the increasing AL caused by the shrinking error in the y separation
between the Wing and the Lead aircraft. The reason this response is almost identical to
that of the negative heading change is because the AL function is nearly symmetric with
respect to the nominal y = 23.562 ft for y separations in the range of 20.5 ft - 26.5 ft or
-/+ 3 ft as previously described and seen in Figure 4.8. The magnitude of the negative
deviation is slightly larger (0.06 ft) than in the -30° run with nonlinear coupling due to
the slight non-symmetry in AL. As in the negative heading change cases, the AD, AL,
and AFY depend on changes in y separation only.

Recall from Chapter IV, the second requirement for gain selection was to achieve
consistent results for both positive and negative maneuvers of equal magnitude. This is
best seen by comparing the Lissajous plots, Figures 5.2 and 5.8 for the -30° and +30°
runs. The positive heading change Lissajous plot is almost identical as the the negative
heading change Lissajous plot except it is rotated by 180 degrees. This is also true for
Figures 5.4 and 5.10 and and Figures 5.6 and 5.12

5-10
 828 1 1 1 1 1
^-\
-/ \ _
f V _
>
1 1 1 1 1
824
0 m 20
r
30
i
40
r 50
i
60
öS40'

S20L -
»
/
°- 0rtcio' i i

4.5001 10 20 30 40 50 60
■ 1 I I I

g
r 4.5
(0
4.4999 1 1 1 1 1

65?- M- _20_ -30- 40 _5p_ 60

g
X
60
o
30(
io 20 30 40i 50i 60

£25
20 1 i ' i

-XL J>0_ _30_ -40- 60

£ o

10 20 30 40 50 60
Time (sec)

Figure 5.7 30 Degree Heading Change Without Coupling Time Plot

63.5

Figure 5.8 30 Degree Heading Change Without Coupling Lissajous Plot

5-11
 828
^826 -/ \
I ~- - __

824
-ÜL J>0_ .SO- JO 50 60
<u
2-20-

0Ü2 1D_ 2Q_ SO. 40 _50_ 60

^4.5
id

65' -UL _20_ -30- _40_ _50_ 60

x
60
301 -UL -2P- 30 ■M. 50 60

^25h

5f*™ io 20 30 40i 50 60
/ ~~ - _ _
S 0
1 1 1 1 1 1
10 20 30 40 50 60
Time (sec)

Figure 5.9 30 Degree Heading Change With Linear Coupling Time Plot

63.5

Figure 5.10 30 Degree Heading Change With Linear Coupling Lissajous Plot

5-12
 828 'III 1
y
-\
ö826 -/ \
f
•v
-
824 1 1 1— 1 1 1
-ÜL 20 _3Q_ -40- 50 60
o
B20-
'55
*■ 0
4.51
ÖW M- J2Q_ 30 -40- _5Q_ 60
g
~4.5
CD

4.5
651 JJL 20 3Q_ -40- _5Q_ 60
g
X
60 ^
30r -M- -20- _3Q_ Jß_ 50 60

&25 h
20
0.051 JJ2- J>0_ _3Q_ -4QL _5Q_ 60
S
N
o
-0.05
10 20 30 40 50 60
Time (sec)

Figure 5.11 30 Degree Heading Change With Non-Linear Coupling Time Plot

63.5 r

Figure 5.12 30 Degree Heading Change With Non-Linear Coupling Lissajous Plot

5-13
5.4 50 ft/sec Velocity Decrease

In this case, the formation is commanded a decrease in velocity of 50 ft/sec without

cross coupling as seen in 5.13 - 5.14. The velocity of the Wing follows the Lead so closely
that the two cannot be distinguished on the velocity time plot. However, the x separation
distance immediately decreases. This shows that there is some delay between the time
the Lead starts to decrease it's velocity and the time the Wing starts to decrease it's
velocity. This is expected since the Wing does not have foreknowledge of the Lead's future
maneuvers. The FFCS is very tight and allows a maximum error of approximately 0.8
ft and settles in about 60 seconds. The y and z channels are unaffected by the velocity
change.

There is no difference between the run with cross coupling and the run with no cross
coupling. As seen in Figures 5.15 - 5.16, the response in the x separation is the same
as the previous run without cross coupling. The first order model was developed based
on the assumption that the vortex extends a long distance behind the Lead. Secondly,
the linearized kinematic equation for y separation is only affected by heading changes, not
velocity changes. Therefore, a change in x direction does not affect the upwash or sidewash.
Only a change in the y direction can cause a change in upwash and sidewash. Therefore,
the reduction in drag AD is constant for pure velocity changes. Prior to the Lead velocity
change in this run and the previous run the Wing aircraft was in steady-state, i.e., all the
forces in the x direction sum to zero, leading to a constant velocity. Therefore, the error
caused by the decrease in the Lead's velocity and input into the x-channel of the FFC is
of equal magnitude and sign and causes identical responses in both runs.

The responses to a 50 ft/sec decrease in velocity with non-linear cross coupling can
be seen in Figures 5.17 - 5.18. This is another very interesting run. The x response is
the same as in the previous two runs. However, the heading, altitude, y separation, and z
separation responses are not zero as in the previous two runs. Both the Wing's heading
and y separation plots experience an initial oscillation. The vertical channel experiences a
slight decrease in AL and causes a drop in altitude of 2 x 10~5 ft. The inital decrease in y
separation causes a slight decrease in AL that leads to an initial decrease in z separation
and then a slow return to nominal. Equations (3.15), (3.18), and (3.21) are necessary to

5-14
understand the responses from this run. The linear model assumes that the formation's
velocity is constant at 825 ft/sec. This is not significant for the heading change maneuvers,
since the Lead and Wing velocities vary only slightly. However, this is not the case for
a velocity change of 50 ft/sec. There are two reasons for this outcome. First, the Lead
velocity determines the strength of the vortex. Secondly, the Wing's velocity determines
the change in angle of attack, a, due to the upwash. Both of these directly affect the
strength of the upwash and downwash exerted on the Wing aircraft. The AL function
is proportional to the Lead's velocity and inversly proportional to the Wing's velocity, as
seen in Equation (3.18). The formation hold controller responds very quickly keeping the
velocity of the Wing almost equal to the velocity of the Lead. Since the two velocities are
almost equal, the change in angle of attack, Aa, is almost constant. It decreases the most
at the beginning of the maneuver when the error between the Lead and Wing velocities is
the greatest. This is seen in the initial decrease in z separation. AFY is affected in exactly
the same way as AD and AX. This leads to the tear drop shape shown in Figure 5.18.
Even though this response looks very dynamic, closer inspection reveals the magnitudes
of the y deviations are very small and the x separation is essentially unchanged from the
previous two runs.

5-15
 _850 T r
f 800
>
750 _i i_
0 101 ?n1 30 401 _ 50 601 70 80
15 1 1 1 1
<D
S 0
'in
^-\
0(10' 10 ?o1 30 401 501 60I 70 80
4.5001 1 I 1

~ 4.5
1
,°
1 1 1 1 1 1 1
4.4999
0 20 3,0 40 50 . 6Q 70 flfl
\
^59.5 - \
s _
59 1 1 1 1 1 1 '
2610 10 20 30 401 501 601 70 80
v
1

§24
22 1 1 1 1 1 1 1

0 10 20 30 40 50 60 70 80
1 I 1 1 1 I I

& 0
N
1 1 1- 1 1 1 r
0 10 20 30 40 50 60 70 80
Time (sec)

Figure 5.13 50 ft/sec Velocity Decrease Without Coupling Time Plot

bu 1 ri i ! ■ 1

59.9 - -

59.8 - -

59.7 - '

^59.6 -
X

59.5 "

59.4 - -

59.3

59.2 1 1 1 1

22.5 23 23.5 24 24.5 25

y(«)

Figure 5.14 50 ft/sec Velocity Decrease Without Coupling Lissajous Plot

5-16
 _ lQ W 20 3D 40 50 60 70 8p

s 0
w

4.5001 P^ IP 2p 3p 4£ 5£ 6l] zp 80
~ 4.5

4.4999
60°, IP 2p 3
i° _ _ JP_ - - -5+°- ■* - -6,Q 7
,Q JiO

£-59.5h v
59
269 IP 2p 3p 40 50 60 70 80
£24
22
Jp 2p 3£ 4p 5£ §Q 70 80

£
N
0

0 10 20 30 40 50 60 70 80
Time (sec)

Figure 5.15 50 ft/sec Velocity Decrease With Linear Coupling Time Plot

eu 1 rj I , ,

59.9 - -

59.8 - -

59.7 - -

^59.6
X
-

59.5 -

59.4 -

59.3

599 ' > i i

22.5 23 23.5 24 24.5 25
y(«)

Figure 5.16 50 ft/sec Velocity Decrease With Linear Coupling Lissajous Plot

5-17
 _850

fsoo
>
750
_/SU5£i°
An'"
ip 2p 30 4p sp ep zp ao
s 0
■55
°--5
4.5QLL0 ip 2p ap 4p sp ap zp ao
:4.5
4.5
_2p ap jjO _ _ _ §Q_ _ZQ ao
£•59.5
59
ip 2p ap 4p sp ep zp ao
£-23.562 k—^

2f™ ip 2p ap 4p sp sp 74] ao
£
N
0^

10 20 30 40 50 60 70 80
Time (sec)

Figure 5.17 50 ft/sec Velocity Decrease With Non-Linear Coupling Time Plot

23.5615 23.5616 23.5617 23.5618 23.5619 23.562 23.5621 23.5622

y (ft)

Figure 5.18 50 ft/sec Velocity Decrease With Non-Linear Coupling Lissajous Plot

5-18
5.5 50 ft/sec Velocity Increase

The formation is commanded an increase in velocity of 50 ft/sec without cross cou-

pling. This case is the exact opposite of the previous case run, 50 ft/sec decrease in velocity
with no coupling. The velocity of the Wing again follows the Lead so closely that the two
cannot be distinguished on the time plot, Figure 5.19, but the x separation distance imme-
diately increases as seen in the Lissajous plot, Figure 5.20. As in the previous case, there
is some delay between the time the Lead starts to increase velocity and the time the Wing
starts to increase velocity. The FFCS allows a maximum error of 1.4 ft, 0.6 ft greater than
the previous case and settles in about 60 seconds, Figures 5.19 and 5.20. The difference in
magnitudes between cases is caused by the rate limits for the Mach hold autopilot. The
upper limit is half of the lower limit. Therefore, the aircraft can decelerate faster than
they can accelerate. The point at which the Wing's Mach hold autopilot saturates is easily
identified by the constant slope of the x separation plot for the first 6 seconds. Except
for the affects of the non-symmetric saturation limits, the second gain criterion is again
satisfied for this case and the case of 50 ft/sec velocity decrease.

Figures 5.21 - 5.22 show the aircraft response to linear cross coupling. This is identical
to the previous run without coupling. Recall from the previous case for a velocity decrease
of 50 ft/sec; the linear first order model for lift, drag, and sideforce is not affected by
changes in x separation and velocity.

The responses for an increase in velocity of 50 ft/sec with non-linear cross coupling
is shown in Figures 5.23 - 5.24. As in the case with non-linear cross coupling for the
velocity decrease of 50 ft/sec case, all channels are affected by the velocity change. The y
separation experiences an initial small oscillation. The x separation error is the same as
the two 50 ft/sec velocity increase runs without coupling and with linear coupling. The
vertical channel experiences an initial increase in lift caused by the initial increase in Aa
caused by the Lead's velocity increasing faster than the Wing's velocity. Rotating this
Lissajous plot and comparing with the Lissajous Plot from the -50 ft/sec run, Figure 5.18
reveals the responses are very similar except in this run, the y deviation is shifted left of
the nominal. This results in the nonsymmetrical bulge in the inverted tear drop.

5-19
 900 r
;850-
800' ' ' ' 1 1 1 i—
_ iQ ip 2Q 30 40 50 fit) 70 80
0
s o
L
*t
4.5001?^ IP 2p 3p 4p 5p ep Zp 80
r 4.5
OS
4.4999
621 -JP 2p 3P 4p 5p 6p 70 80

x
60
26? IP 2p 3p 40 50 80 70
£24L
22
n Jp 2p 3£ 4p 50 80 70 80
£ 0
N
-1
10 20 30 40 50 60 70 80
Time (sec)

Figure 5.19 50 ft/sec Velocity Increase Without Coupling Time Plot

61.5

60.5-

Figure 5.20 50 ft/sec Velocity Increase Without Coupling Lissajous Plot

5-20
 _900 T r

|850h
_i i_
800
ip ^3 ap 40 5£ 6p Zp 80
<D

s 0
'55
°--1 10
4.5001'ff 10 20 30 40 50 &Q ZJ3 80
f 4.5
CO

4.4999
10 20 30 40 50 6JJ Zp 80

§61
60
26<* ip 2p ap $ 3) ep zp ao
g24
22 _i i i i_

1*
n jp 2p 3p 4p 5j3 fß Zp 80
£ 0
_i [_ _l L.
0 10 20 30 40 50 60 70 80
Time (sec)

Figure 5.21 50 ft/sec Velocity Increase With Linear Coupling Time Plot

61.5

60.5

23.5 24 25
y (ft)

Figure 5.22 50 ft/sec Velocity Increase With Linear Coupling Lissajous Plot

5-21
 -.900 -i r -i r -i r

|S50
800 _i i_
_
—.
HQÜO
I
10I 20I mI 4TL
I
50I 613I ZQI ao
O)
o M
S 0^7 ~
<ö
"5<£±°_ ip 2p 3p 4p 5£ 6p zp BO
~4.5f *-
_i i i i_ _i i_
4.5
62°- jp 2p 2p 4p 5£_ _ZQ ao
§61
X
60
23.56310 101 201 301 401 501 601 701 80

23.561 i i i i i i i

5'ffl°1_J0.._..
1
201 301 401 501 601 701 80

£ o
V 1 1 1 1 1 1 1

10 20 30 40 50 60 70 80
Time (sec)

Figure 5.23 50 ft/sec Velocity Increase With Non-Linear Coupling Time Plot

61.5

60.5

23.5612 23.5614 23.5616 23.5618 23.562 23.5622 23.5624 23.5626

y(ft)

Figure 5.24 50 ft/sec Velocity Increase With Non-Linear Coupling Lissajous Plot

5-22
5.6 1000 ft Altitude Decrease

In this case, the formation is commanded a 1000 ft decrease in altitude. Figures 5.25 -
5.26 show the response without cross coupling. The altitude channel is completely separate
from the horizontal channels when cross coupling is not included. The Wing tracks fairly
closely, but does deviate slightly from the Lead. This can be seen in the z time plot,
which shows the vertical separation grow very quickly, level off for a few seconds and then
begin to decrease. This maneuver saturates both the Lead and Wing altitude autopilots.
The PI controller experiences windup during this period of saturation. This can be seen
from the level area between 1 and 5 seconds in the z separation plot. Integrator windup
causes the feedback to be much larger than the command leading to a large error fed into
the controller even though the actual response may be close to the commanded response.
In this case, the windup only causes a delay, but in extreme circumstances windup can
cause the aircraft to become unstable. PrefHters are used to prevent this from occurring
in most cases. They slowly ramp the command into the system, allowing the error to
grow gradually and give the controller time to respond. The prefilter, described at the
beginning of this chapter, is working, but not as well as desired. The aircraft does not go
unstable, but does take a few seconds to winddown. Figure 5.26 shows there is no affect
on the horizontal separation distances from the change in altitude.

The responses for a 1000 ft decrease in altitude with linear cross coupling are shown
in Figures 5.27 - 5.28. Recall from Sections 5.3 and 5.4, a change in heading with linear
cross coupling caused a perturbation in the vertical channel. This case shows the reverse is
true also. A change in altitude, with cross coupling included, leads to a minor oscillation
in the heading. The A4i5 term directly links the z separation into the heading autopilot.
This is easily seen in the plot for the heading response. The heading is linked into the
Mach hold autopilot and the x and y separations. This causes the small initial oscillations
in the velocity and the x and y separation plots. The maximum y separation magnitude
is 2 x 10-3 ft. Under real circumstances all of these effects would be drowned out by
insignificant changes in air currents, but it is still interesting to see the simulation behave
as expected. Overall, everything settles down in about 50 seconds with the exception of
a little bit of chattering. The linear first order model AL and AD is not affected by a

5-23
change in z. All the changes in AL and AD are caused by the change in heading which is
caused by a change in sideforce, AFY, due primarily to the change in z separation, since
the y separation is extremely small. Therefore, the affect of a change in z separation on
the x-channel is only superficial and the affect on the y-channel is minor due to the simple
first order model employed.

Figures 5.29 - 5.30 show the response for a decrease of 1000 ft in altitude with non-
linear cross coupling. The inter-relationship between the z and y separations is much more
obvious for this run than for the linearized run. The comparison between this run and the
previous shows the x and y separation errors are much larger, highlighting the difference
between the full non-linear model and the model with linear aerodynamic cross coupling.
The y separation has a maximum error of 0.5 ft and a minimum error of -0.2 ft. The
maximum x separation error is also much larger in magnitude, 0.1 ft. The y deviations
are relatively small compared to the heading change cases, but larger than those from the
velocity maneuver change cases. Recall from Figure 4.13, the change in sideforce, AFy,
was not as dynamic as the change in drag for small changes in y separation. However, as
seen in Figure 4.15 large changes in z separation can cause large changes in sideforce. A z
separation of 6 ft leads to an increase in magnitude of AFy of 50%. Since AFy is negative
it pushes the Wing away from the Lead, increasing the y separation. The y separation
has a maximum magnitude of 1.1 ft. Since the nominal point sits on the knee of the curve
as seen in Figure 4.13, AFy does not vary much for the 1.1 ft change in y separation.
Therefore, in this case the effects of the change in z separation are greater than the effects
from the change in y separation.

5-24
 826 r

■55 825-
824 L
-1£- _20_ 30 40 50 60
-3 #
CD
S 0
'</)
W _2p_ -30- Jfl_ J5Q_ -60
4.51
£■
~4.4
CO

4.3
61' 10- -Q- 3Q_ -40- _5Q_ 60

^60
X
59
261 UQ_ -20- 40 50 60

g24
>.
22
101 iQ_ _2p_ 30 _40 -SSL 60

S 5
10 20 30 40 50 60
Time (sec)

Figure 5.25 1000 ft Altitude Decrease Without Coupling Time Plot

Figure 5.26 1000 ft Altitude Decrease Without Coupling Lissajous Plot

5-25
 _2p_ _3Q_ _40 50 60

_20_ _3Q_ _4Q_ 50_ 60

0 10 20 30 40 50 60
1 1 1 I I

\
1 ^ -
S-60
1 1 1 1 1

4°- _20_ 3Q_ -4L 50_ 60

23.56
101 10 _2Q_ _30 -ä0_ 50_ 60

o?
10 20 30 40 50 60
Time (sec)

Figure 5.27 1000 ft Altitude Decrease With Linear Coupling Time Plot

23.5606 23.5608 23.561 23.5612 23.5614 23.5616 23.5618 23.562 23.5622 23.5624
y (ft)

Figure 5.28 1000 ft Altitude Decrease With Linear Coupling Lissajous Plot

5-26
 _3Q_ JO -50- 60
-4L

_20_ _3Q_ JO 50 60

-30- 40 _5Q_ 60

59.8 1
251 ) P 20 3,0 40 50 60
-
'
23n 1 1 1 1 1

1<£ -J0_ JQ_ _ao_ JO -50- 60

£
N
5h'"

10 20 30 40 50 60
Time (sec)

Figure 5.29 1000 ft Altitude Decrease With Non-Linear Coupling Time Plot

60.12

Figure 5.30 1000 ft Altitude Decrease With Non-Linear Coupling Lissajous Plot

5-27
5.7 1000 ft Altitude Increase

The formation is commanded a 1000 ft increase in altitude without coupling as shown

in Figures 5.31 - 5.32. This case is almost identical to the previous except the direction
for each term is opposite. The Wing tracks fairly closely, but does deviate from the Lead.
This can be seen in the z time plot, which shows the vertical separation grows very quickly,
levels off for a few seconds and then decreases. This maneuver saturates both the Lead and
Wing altitude autopilots, but the Wing is saturated longer than the Lead. This is made
better and worse by the upper limit on the FCS altitude hold autopilot. The magnitude
of the upper limit is 1/3 less than the lower limit. The smaller limit causes the Lead
autopilot to saturate faster, thereby not allowing the Lead to move farther away from the
Wing. However, the lower saturation limit causes the vertical PI controller to windup for
a longer period of time, causing a longer delay before the controller winds down and starts
bringing the z separation back to the nominal value. This is easily seen in the z separation
in which it is constant from 1 to 7 seconds, two seconds longer than in the 1000 ft altitude
decrease case. There is no effect on the horizontal separation distances from the change
in altitude. The reasons for this were described in the last section in the case of a 1000 ft
altitude decrease.

Linear cross coupling causes some interesting results as in the previous case. The
responses for this run are shown in Figures 5.33 and 5.34. A change in altitude with
cross coupling included initially causes minor oscillations in the heading and velocity. The
changes in heading and velocity lead to slight perturbations in the x and y separation
distances. Overall, everything settles down in about 60 seconds. The x separation continues
to oscillate, but the magnitude of the oscillation is on the order of nanofeet. The effects
on the horizontal channels is again due to the coupling caused by the A4)5 term.

Figures 5.35 - 5.36 show the response for an increase of 1000 ft in altitude with non-
linear cross coupling. As in the previous -1000 ft case with nonlinear cross coupling, the
velocity and heading responses have an initial small oscillation. The y separation has a
minimum error of -1.8 ft and a maximum error of 0.5 ft. Recall from the previous case,
a z error of 6 ft resulted in an increase in si deforce of 50%. In this case, a z error of -6
ft results in a decrease of 130% causing the AFY to actually become slightly positive as

5-28
seen in Figure 4.15. This strong positive AFY pulls the Wing toward the Lead causing
the y separation to decrease. Since the AFY is stronger for a negative z change than for a
positive z change this causes the maximum magnitude in y separation to be 0.7 ft greater
between this run and the -1000 ft altitude with nonlinear coupling run. As expected, the
maximum x separation error is larger in magnitude, 0.1 ft, than the linear run. Overall,
all perturbations settle in about 60 seconds.

It is again apparent from Figures 5.25 - 5.36 that the results are consistent. The
error in the z channel is about 6 ft for both an increase and decrease in altitude. The x
and y changes also show some consistency between responses for an increase and decrease
in altitude. Therefore, these results satisfy the second gain criteria of consistent results for
positive and negative maneuver changes.

5-29
 826 r
■35 825-
>
824L
Jfl- _2Q_ 30 40_ 50 60
■a 1^
o
'a
a.
IT JLQ_ _2fl_ 40- 60
i.# -30- 4P-
r4.6-
cd
4.5L
ei9 40- -Q- 30 40- _50_ 60

x
59-
26r 40- _2fl_ -W- _4L _50_ 60

£24
22
0^ 40- ^ 43p_ 40- -5P- 60
£ -5
-10
10 20 30 40 50 60
Time (sec)

Figure 5.31 1000 ft Altitude Increase Without Coupling Time Plot

23.5 24 24.5 25
y (ft)

Figure 5.32 1000 ft Altitude Increase Without Coupling Lissajous Plot

5-30
 ö825k

HpLZZJÄ
f1T -2P- _30_ _40_ _50_ 60

s 0
'35

46
;QLlo iSL JJL _3Q_ jKL _50_ 60
-. ' '
?4.55
5
4.5
JjD_ _20_ -30- -4L _50_ 60

E-60

-UL ^D_ _30_ JJL -50 60

23.56
10 ^ 30 _ _ 4P 5Q_ 60

-10
10 20 30 40 50 60
Time (seo)

Figure 5.33 1000 ft Altitude Increase With Linear Coupling Time Plot

60

60

60

60

x
60

60

60

§0 -i 1 i i_ _i i_
23.5616 23.5618 23.562 23.5622 23.5624 23.5626 23.5628 23.563 23.5632
y (ft)

Figure 5.34 1000 ft Altitude Increase With Linear Coupling Lissajous Plot

5-31
5.8 Summary

Six cases were presented: +/-300 heading change, +/-50 ft/sec velocity change, and
+/-1000 ft altitude change. Some observations not mentioned in the cases are presented:
The FFC can maintain the formation within the 10% tolerance for y and z errors for
a Lead heading change of +/-30", +/-50 ft/sec velocity change, and +/-400 ft altitude
change. The 400 ft altitude response was not shown in the cases. The +/-300 did not
meet the 10%b tolerance for x error, but since the change in x is negligible this is not a
requirement. However, a +/-200 heading change does meet the 10%b for x, y, and z errors.
The responses were also shown to meet the gain criteria listed in Chapter IV.

5-32
 1 1 1 1 1

|S25 1 •
>
0 10 20> 30i 40i 50 60
3 0-2' i i
CD
s 0
°--0.2 1

»10* 10 20 30 40 50 60
4.61
~4.55 - y
CO

4.5
60.210 10 20 30i 40 50
r 60
"
'"*

59.8
25'0 10 1
20
I
30 1
40
1
50
1
60
s
\
>.
20 1
0
\
,° 2.0 3.Q. _ _ 40 5,0 60
£
N
-5
-10 1 ' ' ' i

10 20 30 40 50 60
Time (sec)

Figure 5.35 1000 ft Altitude Increase With Non-Linear Coupling Time Plot

60.12

24 24.5

Figure 5.36 1000 ft Altitude Increase With Non-Linear Coupling Lissajous Plot

5-33
 VI. Conclusion

6.1 Conclusions

The primary objective of this research was to determine if a large formation flight
controller could be redesigned and applied to a close formation. The close formation flight
controller designed in Chapter IV has met this objective.

The primary question to be answered by this research was to determine if a close

formation flight controller designed without due consideration of the close formation aero-
dynamic coupling can handle the additional forces created by the close formation. The
secondary question was to determine if the kinematic coupling is stronger than the aero-
dynamic cross coupling. The primary and secondary questions have also been answered
through the stability and robustness analysis of Chapter IV and the results obtained in
Chapter V.

The effects of close formation flight were derived and applied to the close formation
flight controller. It was shown that the large formation flight controller, with the gains
adjusted for a close formation, resulted in zero steady state tracking errors. The maximum
tracking error in x separation was 4 ft {K, \b), the maximum tracking error in the y
separation was 3 ft (j^b), and the maximum tracking error in the z separation was 6 ft
(|&) for the envelope defined in the assumptions with the full non-linear kinematics and
aerodynamic cross coupling included.

The criteria for gain selection was also met. The gain criteria are briefly summarized
as: similar positive and negative perturbations for a heading change maneuver, similar
results for a positive and negative maneuver change, and the Wing should not cross over
the x or y axes. Consistent positive and negative deviations from nominal for a heading
change maneuver satisfied the first gain criterion. The responses for a negative and positive
heading and velocity maneuver change of equal magnitude were consistent; satisfying the
second gain criterion. Finally, the Wing aircraft never crossed over either the x or y axes.
Therefore, no collisions occurred and the third gain criterion was satisfied.

With the exception of the A6>3 term, the kinematic coupling was generally at least
two orders of magnitude greater than any aerodynamic induced cross coupling. The close

6-1
formation flight controller developed without including aerodynamic cross coupling proved
to be stable and robust. All eigenvalues for the closed loop system were shown to be in
the left half plane. The overall responses from the system with the inclusion of the full
nonlinear kinematics and aerodynamic effects were shown to be similar to the responses
from the full nonlinear kinematics without the inclusion of the aerodynamic effects.

Finally, it was shown that the formation flight controller can maintain the formation
geometry within the required 10% tolerance for y and z errors for a Lead heading change
of +/-300, a velocity change of +/-50 ft/sec, and an altitude change of +/-400 ft. Thus
the close formation flight controller can enable aircraft to take advantage of the reduction
in induced drag brought about by the aerodynamic coupling effects.

6.2 Recommendations for Further Study

• The models employed in this research were rather rudimentary. The prefilter was
a simple first order lag and was identical for each channel. The FCS models were
based on a point mass and did not include moment effects. The FCS C-130 models
were modified to represent an F-16 class aircraft. More accurate Models should be
developed to better represent a true F-16 class aircraft.

• Moment effects caused by the vortex should be developed. The effects of linearized
and non-linear moments should be applied to the close FFC to determine if it is
robust enough or whether it needs to be modified.

• If higher precision formation flight control is desired, it may be prudent to add sensor
noise to the system and perform a stochastic based analysis using a Kaiman filter
to estimate the true location of the Lead with respect to the Wing in the formation.
This will also promote fusing sensor data from a close formation of UAVs performing
a surveillance mission in which each UAV has a different sensors.

• Implement the close formation flight controller in a digital controller.

• Incorporate anti-collision circuitry and analyze the effect.

6-2
6.3 Summary

The close formation aerodynamic coupling effects on the Wing aircraft caused by the
Lead's wing vortex have been included in the formation flight control system dynamics.
A close formation flight control system for the Wing aircraft was designed. It was shown
that formation flight control system designs, accounting for kinematic coupling effects
only, are robust and can handle the additional aerodynamic coupling effects caused by close
formation flying. The developed close formation flight controller can enable aircraft to take
advantage of benefits from the vortices created by the Lead aircraft in the formation. This
can reduce the formations fuel consumption and extend formation range and endurance.
This can also enable multiple aircraft or UAVs to fly together in a controlled close formation
during

6-3
 Appendix A. Appendix A

The F-16 characteristic data are contained in the Matlab data file, controlfile.m. The
actual program is contained in the Matlab script, runsimc.m. The script loads the data file
and then executes the simulations using the Simulink models. Two models are used the
first is the linear state space model and the second is the complete linear/nonlinear block
model. The linear simulation is used to perform initial analysis and design of the controller.
This includes the gain selection described in Chapter IV. The full block simulation model
is used to test the designed controller for stability and robustness and determine if the
gains need to be adjusted.

A.l Characteristic Data file

7, controlfile.m
'/. PLEASE NOTE

'/, Second order MODELS with saturations most accurately follow true autopilots

*/, without saturations if the inputs are limited to 30 degree magnitude heading

'/, change; 40 ft/sec magnitude velocity change; 1000 ft magnitude altitude change
'/, More thorough methods will have to be employed if values greater than these

'/, are required. They might include ramping of input or alteration of models

'/, based on better modeling methods as outlined in Russ Miller's dissertation

'/. NON-LINEAR or LINEAR

linear=l;
nonlin=-l;

'/.switch = linear

X THIS IS FIRST ORDER MODELS

*/. COMMANDS

'/, F-16 wingspan 30 ft

'/. Pi/4 X 30 = 23.6 ft

A-l
nominal_vel=825; '/, M=0.85 a=97i

command_vel=825;

nominal_heading=0;

commanded_heading=0;

nominal.alt = 45000;

commanded_alt = 46000;

x_nominal= 60; '/, 2b nominal this one does not matter as much for upwash as does

x_command= 60;

y_nominal= 23.562; '/, (Pi/4 X 132)/2 nominal this one, we also need a z component

y_command= 23.562;

z_nominal = 0;
z_command = 0;

'/. F-16 LEAD AIRCRAFT AUTOPILOT

'/, time constants

lead2_vel_time=5;

leadl_Tsi_time=l/3;

lead2_alta_time=0.3075;

lead2_altb_time=3.85;

'/, Lead Saturations

lead_accel_sat_low=-10;

lead_accel_sat_high=5;

lead_Tsi_sat_low=-6;

lead_Tsi_sat_high=6;

lead_alt_sat_low=-126;

lead_alt_sat_high=100;

*/. F-16 WING AIRCRAFT AUTOPILOT

A-2
*/, Time constants

wing2_vel_time=5;

wingl_Tsi_time=l/3; '/.for a true second order model

wing2_alta_time=0.3075;

wing2_altb_time=3.85;

'/, Wing Saturations

wing_accel_sat_low=-10;

wing_accel_sat_high=5;

wing_Tsi_sat_low=-6;

wing_Tsi_sat_high=6;

wing_alt_sat_low=-126;

wing_alt_sat_high=100;

'/, Stability Derivatives and aircraft values F-16 class

weight_max=25000;

weight_avg=21500; */, lbs half tank of fuel

weight_min=18000; '/, no fuel

AR=3; '/, aspect ratio

S=300; •/, sq ft area of wing

b=30; '/, ft wingspan

Svt=54.75; '/, sq ft

hz=10 ; •/, ft or 120 inches height of tail

aw=5.3 ; '/, per rad wing lift curve slope

avt=5.3 ; '/, per rad tail lift curve slope

eta=.95 ; '/, efficiency factor of tail

q=155.8; '/, lb/sq ft dynamic pressure

mu=0.03; */, upwash/sidewash correction factor

A-3
X DARGAN CONTROLLER

'/. LINEAR MIXER cur DAR

vel_error_gain=12.S; '/, 5

head_error_gain=6; '/, 10
x_pos_error_gain=-8; '/, 2

y_pos_error_gain=-0.6; '/, 1
z_pos_error_gain=25; '/, 0.5

•/. PI CONTROLLER

x_pro_gain=6; '/, 0.17 1.2

x_int_gain=0.4; '/, 0.02

y-Pro_gain=ll; */. 0.5

y_int_gain=0.9; */, 0.05

z_pro_gain=4.0; '/, 1.0

z_int_gain=0.5; */, 0.5

A-4
A.2 Executable Program

clear

controlfile

def_time=60;
def_cont=l;
fig_num=0;
def_par=[] ;

'/, close old simulink controller if open

parms=get_param;
if parms == def.par;
>
else

close_system;
end

'/simplify names of time constants

tv=lead2_vel_time;

tpsi=leadl_Tsi_time;

tha=lead2_alta_time;
thb=lead2_altb_time;

'/, simplify names of gains

Kv=vel_error_gain;

Kpsi=head_error_gain;

Kx=x_pos_error_gain;

Ky=y_pos_error_gain;

Kz=z_pos_error_gain;

A-5
'/, PI CONTROLLER

KXP=x_pro_gain;

KXI=x_int_gain;

KYP=y_pro_gain;

KYI=y_int_gain;

KZP=z_pro_gain;

KZI=z_int_gain;

prev=input('Do you want to close previous pictures 1 yes 0 no: ');

if prev == 1

close all;

hx=l;

hy=0;

else prev == 0

prev=0;

hx=l;

hy=l;
end

cont=input('please choose 1 for current or 0 for previous : ');

if cont == 1

cd ../dargan/current;

XI= [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0];
else cont == 0

cd ../dargan/old;

*/,XI= [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0];

XI= [0,0,0,0,60,23.562,0,0,0,0,0,0,0,0,0,0,0,0];

A-6
choice=input('please choose 1 for linear or 0 for nonlinear kinematics: ');
if choice == 1;

kswitch = linear;

else choice == 0;

kswitch = nonlin;
end

choice=input('please choose 1 for linear or 0 for nonlinear disturbance: ');

if choice == i;

dswitch = linear;

else choice == 0;
dswitch = nonlin;
end

V_dist_sw=input('Do you want a velocity disturbance, 1 yes 0 no: ');

H_dist_sw=input('Do you want a altitude disturbance, 1 yes 0 no: ');

Tsi_dist_sw=input('Do you want a heading disturbance, 1 yes 0 no: ');

end

'Calculate formation stability derivatives1

weight=weight_max;
Cl=weight/(q*S);

m=weight/32.2 ; '/.mass
bp=(pi*b)/4;

•/. TRUE CHANGE IN DRAG

A-7
'/, nominal

G=(2*Cl*nominal_vel*b)/(pi*AR);

W=G/(4*pi*bp)*(log((y_nominal~2 + z_nominal~2 + (b*mu)~2)/((y_nominal-bp)~2 +.

z_nominal~2 + (b*mu)~2))...

- log(((y_nominal+bp)~2 + z_nominal"2 + (b*mu)~2)/(y_nominal~2 +...

z_nominal~2 + (b*mu)~2)));

delta_C_D=Cl*W/nominal_vel;

D_nominal=q*S*delta_C_D/m

'/. TRUE CHANGE IN LIFT

'/, nominal

delta_C_D=aw*W/nominal_VGl;

L_nominal=q*S*delta_C_D/m

'/. TRUE CHANGE IN SIDE FORCE

'/.nominal

V=G/(4*pi*hz)*(log(((y_nominal-bp/2)~2 + z_nominal"2 + (b*mu)~2)/((y_nominal-bp/2)"2 +.

(z_nominal+hz)"2 + (b*mu)"2))...

- log(((y_nominal+bp/2)~2 + z_nominal"2 + (b*mu)~2)/((y_nominal+bp/2)~2 +...

(z_nominal+hz)~2 + (b*mu)~2)));
delta_C_Y=eta*Svt*avt*V/(S*nominal_vel);
Tsi_nominal=q*S*delta_C_Y/m

'/, velocity

delta_C_D_W_y=(1/(pi*AR))*(Cl~2)*(1/b)*((3/8)*pi/[((pi/4)~2+nnT2)*((pi/2)"2+mu"2)]);

*/, altitude

delta_C_L_W_y=aw*Cl*(3/8)*pi/ [pi*AR*b*((pi/4)~2+mu~2)*((pi/2)~2+mu~2)];

A-8
'/, heading

C_Y_W_y_first=hz"2/ [2*b~2*((pi/8)~2+mu~2)*((pi/8)~2+(hz/b)~2+mu~2)];
C_Y_W_y_sec=3*hz~2/[2*b"2*((3*pi/8)-2+mu~2)*((3*pi/8)"2+(hz/b)"2+mu"2)];

C_Y_W_y_inner=C_Y_W_y_first-C_Y_W_y_sec;

delta_C_Y_W_y=eta*Svt*avt*Cl*C_Y_W_y_inner/(pi*AR*2*S*hz);

delta_C_Y_W_z=-(eta*Svt*avt*Cl*32*hz*pi~2)/(pi*AR*2*S*pi*b~2*8~2*...

[((pi/8)-2+mu~2+(hz/b)~2)*((3*pi/8)~2+mu~2+(hz/b)"2)]);

if cont == 1
lindis=input('please choose 1 for disturb or 0 for nodisturb : ');
if lindis == 1

'/, STATE SPACE MODELS

'/, A with disturbances

A=[0 -1 [(q*S/(m*nominal_vel))*(delta_C_Y_W_y*y_nominal)] [-(y_nominal/wingl_Tsi_time)*(p

0 C-(l/wing2_vel_time)] [(q*S/m)*(delta_C_D_W_y)] 0 0 0;

0 0 [-(q*S/(m*nominal_vel))*(delta_C_Y_W_y*x_nominal)] [((x_nominal/wingl_Tsi_time - nomi

0 0 [(q*S/(m*nominal_vel))*(delta_C_Y_W_y)*(180/pi)] [-(i/wingl_Tsi_time)] [(q*S/(m*nomin
0 0 0 0 0 1;
0 0 [(q*S/m)*(delta_C_L_W_y)] 0 [-l/(wing2_alta_time*wing2_altb_time)] -[(l/wing2_alta_ti

else lindis == 0
'/, A without disturbances
A=[0 -1 0 C-(y_nominal/wingi_Tsi_time)*(pi/180)] 0 0;
0 C-(l/wing2_vel_time)] 0 0 0 0;
0 0 0 C((x_nominal/wingi_Tsi_time - nominal.vel))*(pi/180)] 0 0;
0 0 0 [-(l/wingl_Tsi_time)] 0 0;
0 0 0 0 0 1;
0 0 0 0 [-l/(wing2_alta_time*wing2_altb_time)] -[(l/wing2_alta_time) + (l/wing2_altb_time

A-9
end
end

'/, Input Matrix

bl=[0 C(y_nominal/wingl_Tsi_time)*(pi/180)] 0;
[l/wing2_vel_time] 0 0;
0 [-(x_nominal/wingl_Tsi_time)*(pi/180)] 0;
0 C(l/wingl_Tsi_time)] 0;
0 0 0;
0 0 [i/(wing2_alta_time*wing2_altb_time)]];

'/, Disturbance Matrix

b2=[l 0 0;

0 0 0;

0 nominal.vel*(pi/180) 0;
0 0 0;
0 0 0;
0 0 -Cl/(wing2_alta_time*wing2_altb_time)]];

B=[bl b2] ;

C=[l 0 0 0 0 0;
0 10 0 0 0;
0 0 10 0 0;
0 0 0 10 0;
0 0 0 0 10];

D=[0 0 0 0 0 0;
0 0 0 0 0 0;

A-10
0 0 0 0 0 0;
0 0 0 0 0 0;
0 0 0 0 0 0];

I=[0 0 0 0 0 0]';

*/, Closed Loop plant

G=pi/180;
al=[0 -1 y_nominal*((q*S*delta_C_Y_W_y/(m*nominal_vel)) - G*KYP*Ky/tpsi) -(G/tpsi)*y_nomi
(q*S/(m*nominal_vel))*(y_nominal*delta_C_Y_W_z) 0 0 y_nominal*KYI*G/tpsi 0];
a2=[-KXP*Kx/tv -(l/tv)*(i + KXP*Kv) (q*S/m)*delta_C_D_W_y 0 0 0 KXI/tv 0 0];
a3=[0 0 x_nominal*(G*KYP*Ky/tpsi - (q*S/(m*nominal_vel))*delta_C_Y_W_y) ...
G*(x_nominal/tpsi -nominal_vel + x_nominal*KYP*Kpsi/tpsi) ...
-x_nominal*(q*S/(m*nominal_vel))*delta_C_Y_W_z 0 0 -x_nominal*KYI*G/tpsi 0];
a4=[0 0 (-KYP*Ky/tpsi + (q*S/(m*nominal_vel*G))*delta_C_Y_W_y) -(l/tpsi)*(i+KYP*Kpsi) ...
(q*S/(m*nominal_vel))*delta_C_Y_W_z/G 0 0 KYI/tpsi 0];
a5=[0 0 0 0 0 1 0 0 0] ;
a6=[0 0 (q*S/m)*delta_C_L_W_y 0 -(l/(tha*thb))*(KZP*Kz + 1) -(1/tha + 1/thb) 0 0 KZI/(tha
a7= C-Kx -Kv 0 0 0 0 0 0 0];
a8=[0 0 -Ky -Kpsi 0 0 0 0 0];
a9= [0 0 0 0 -Kz 0 0 0 0] ;

Acl= [al; a2; a3; a4; a5; a6; a7; a8; a9] ;
eig(Acl)

open.system controller
cd ../../test/;

A-ll
time=input('please enter time for simulation in increments of 10 seconds :');

if time == 0;

time = def_time;

else

time = time*10;
end

'/, Set initial state values

'/,xl = [O;nominal_vel;nominal_vel;x_nominal;y_nominal;0;0]

'/,; nominal.heading; nominal_vel; nominal_heading]

'/,x2 = [O;nominal_vel;0;0;0;nominal_alt;nominal_alt];

•/.x3= □;
y.XI = [xl;x2];

max_error=le-3;
min_step=time/1000;
max_step=time/1000;

y,[t,x,y]=rk45('controller',time,XI, [max_error,min_step,max_step]);
y.[t,x,y]=rk45('controller' ,time);
yi[t,x,y]=rk45('cont_generic' ,time) ;
[t,x,y]=adams('controller',time);

y,[t,x,y]=rk45('controller' »time,XI, [max_error,min_step,max_step]);

•/, velocities of lead and wing

set(figure,'Position', [560*(hx-l),480-480*hy,560 460]);

subplot(6,1,1), plot(t,y(:,3),'-');
subplot(6,1,1), ylabeK'Vel (ft)');
hold;
subplot(6,1,1), plot(t,y(: ,5), 'c~');

A-12
'/• headings of lead and wing

subplot(6,1,2), plot(t,y(:,4),'->);

subplot(6,1,2), ylabeK'Psi (deg)');

hold;

subplot(6,1,2), plot(t,y(:,6),'c--');

*/, lead altitude

subplot(6,1,3), plot(t,y(:,12),'-');
hold

*/, wing altitude

subplot(6,1,3), plot(t,y(:,13),'c~');
subplot(6,1,3), ylabeK'alt (ft)');

'/, x seperation
subplot(6,1,4), plot(t,y(:,l),'c—');
hold;
subplot(6,1,4), line(t,x_nominal.*ones(size(t)));
subplot(6,1,4), ylabeK'x (ft)');
'/, y seperation

subplot(6,1,5), plot(t,y(:,2),'~c');
hold;

subplot(6,1,5), line(t,y.nominal.*ones(size(t)));
subplot(6,1,5), ylabeK'y (ft)');
'/, z seperation

subplot(6,1,6), plot(t,y(:,ll),'—c');
hold

subplot(6,1,6), plot(t,z_nominal.*ones(size(t)));
subplot(6,1,6), ylabeK'z (ft)');

A-13
xlabel('Time (sec)');

hx=hx+l;
last_y=size(y,1);

set (figure,'Position',[560*(hx-l),480-480*hy,560 460]);

plot(y(:,2),y(:,D);
hold on
plot(y(l,2),y(l,l),'o');
hold on
plot(y(last_y,2),y(last_y,l),'x');
ylabeK'x (ft)');
xlabeK'y (ft)');

a=x;

*/. TRUE CHANGE IN DRAG

'/, nominal
G=(2*Cl*nominal_vel*b)/(pi*AR);
W=G/(4*pi*bp)*(log((y_nominal~2 + z_nominal"2 + (b*mu)"2)/((y_nominal-bp)"2 +.

z_nominal~2 + (b*mu)"2))...

- log(((y_nominal+bp)~2 + z_nominal"2 + (b*mu)~2)/(y_nominal~2 +...

z_nominal~2 + (b*mu)"2)));
delta_C_D=Cl*W/nominal_vel;

D_w_nom=q*S*delta_C_D/m;

'/. final

G=(2*Cl*command_vel*b)/(pi*AR);

W=G/(4*pi*bp)*(log((y_command~2 + z_command~2 + (b*mu)~2)/((y_command-bp)~2 +.

z_command"2 + (b*mu)~2))...

- log(((y_command+bp)~2 + z_command~2 + (b*mu)~2)/(y_command~2 +...

A-14
z_command~2 + (b*mu)"2)));

delta_C_D=Cl*W/command_vel;

D_w_com=q*S*delta_C_D/m;

'/, change in drag

delta.diff=(D_w_com-D_w_nom)

A-15
A.3 Linear Simulation Model

The linear model, shown in Figure A.l is the State-space representation of the
system described in section 4.1 and described by Equations (4.13) - (4.16). The com-
mands input to the Lead are: commanded.velocity-nominaLvelocity, commanded_heading-
nominalJieading, and commanded_alt-nominal_alt.

The individual blocks that make up the linear model are described. The Wing&kinematic&vortex
block is the state space system described by Equations (4.13) - (4.16). The command pre-
filter block, Figure A.2 contains the prefilter discussed in Chapter V. The FFC controller
is made up of the linear mixer block described in Figure A.5 and the PI controller, Figure
A.6 described in Chapters II and IV. The Lead aircraft, Figure A.3 is composed of the
flight control system detailed in Chapter II. Finally, the Perts+nominals block, Figure
A.4 is used to add the nominal settings back to the perturbations to obtain the actual
separation distances.

m-
x_sep
X1
y_sep
—»s
r*0 lead vel
m-
z.
y2

linear_mixer r-KI]
x2

El
velocity I fc| —~-*{s]
x'.Ax+Bu J* Demu) -►I 2 | wing_vel ►J 1-*'Mux—1
heading! H
% Mu)
y - Cx+Du
\J\—bommandj
:ommand_p efilter \- PI Controllur Mux
-j-jj 'ing&kinematic&vorh -Will wingjiead
altitude
Perlurballuns
Perts+nominals
—*—HE
wing_alt
>3
leal head
y3

lead alt
SECOND_ORDER

Figure A.l Linear State Space Matlab Simulation Model

A-16
 1
n 1 / /*out_1
- \.
/
2 ]t Mux>> r^-knn ->1/s-lf'Demux ► 2
in 2 X* Mux
"\
\ To
out_2
Demux
3 ■/ -<J
in_3
■^

Vs
out 3

Figure A.2 Command Prefilter Simulation Model

F-16

ominal v<n

1
Vel c ^ >7*a 1/s 3E V lead
vel_sat_lead

<a linal heat

Psi c >*a head sat lead

1/s Ha—HPsi lead

lominal an

T|-^^>-w+-
Alt rate c
1/s
>7tS alt sat lead
1/s
Alt lead

Figure A.3 Lead Aircraft Flight Control System Model

A-17
 +
in 1

in 2 -fiL V w

in 3 *

in 4 -X&. Tsi w
*
in 5

H w
ominal m-
< nomina
linal heat'
(Miomma -1

i nomina-1

in 6

Figure A.4 Perturbation+Nominal Block

A-18
VelJ ~fc[f]—^—►fe- ->L_i_
"TTI I vel_error Xch error
Vel w

x_sep =HT-
ffil—I x-err
xe
£—'
x com

§
feu^EP
~71 I Tsi_error
N?> AS Ych error
Tsi w

_sep ■—►F1^ — ye
t^
y_com

z ze
p!!p /HT- {*> Zch error
10—' z err
z com

Figure A.5 Linear Mixer Simulation Model

A-19
 1/s
Integrator «xi

1 1
X Channel z_error1 wing_velocity_command
KXP

1/s
Integratorl «Yl

$>
Y channel ^D z error2 wing_heading_command
r\Yr

1/s
lntegrator2 «ZI

Z channel z_error wing_alt_command

KZP

Figure A.6 PI Controller Simulation Model

A-20
A.4 Non-Linear Simulation Model

The full block simulation model is shown in Figures A.7 - A.10. The prefilter, linear
mixer, PI controller, and lead aircraft blocks are the same as those used in the linear
model above. The Kinematic and Horseshoe Vortex simulation blocks contain both the
linear models and the non-linear models, Figures A.9 and A.10 respectively. The full
nonlinear model was discussed in section 4.5. The wing aircraft, Figure A.8 is the same as
the Lead except the disturbances are added to the respective autopilots they affect.

m-
x_sep

m-
y_sep x1

m-
z_sep
y2
lead vel linear mixer
x2 wlng_vel

velooltyi-

^ 5- ^ ^
♦[ül
□5-T command_prefilter|
heading:
PLController wing_alrcri
ead aircrat wing_aK
altitude —M
kinematic; ö-
lead head Horshoe_vorte>

lead alt
SECONDORDER wing_head
y3

Figure A.7 Matlab Block Simulation Model

A-21
 ominal_v< ■

0-
vel_dist

[iH*>^i-*ffl—H3 to-
vel c Vwing
vel_sat_wing

-HI]
Psidot_wing

head dist iinal_hea<

0—*£>-^i
Psi c
■+BL-tS H0
Psi_wing
iead_sat_wing

alt dist
ominai_a -i

*E>-FHI}+ffl-+& 1—=W3—^H3
Alt_wing
alt_sat_wing

^J

Figure A.8 Wing Aircraft Flight Control System Model

A-22
 non-linear -1

Demu> —- £Z*&-
Demux

HB
f(u)
L*£>E*>r>[
Tsl_E G CO! (Tsl_E

filn/Tcl c\
Sin(Tsl_E)
w
* ■

wing_alt

Figure A.9 Kinematic Motion Simulation Model

A-23
 a_C_D_V - q'S/m —i

=& —to- dswitch L_T +£>—►□

& TjE} Switch on/off_v veLdist
y sep
- Lnomlna

l-b/(pi*
to
&
lead_vel

&
Gamma
^PW »I w H
fin»« r* n
delta_C_D

q*S/m
■" ' '

non-linear -1

z_sep

■H1/U[1)F
wing_vel 1/V

I -IAJHI h q"S/m—i

—=ttZh=HT} g>-*[*>-#—K3.
Switchl 1/Q on/of(_tsi head_dis

:_nominaJ
i—i
C Y J

f(u) Svt*a\ -i

SW^HD {&-=tEH*lH
q*S/m

.nominal —i

to- FCP&
^S—H>—*s
Swltoh2 on/off_alt alt_dist

q"S/m

^Q- io- linear

Figure A. 10 Horshoe Vortex Disturbance Simulation Model

A-24
 Bibliography
1. Buzogany, Louis E., Automated Control of Aircraft in Formation Flight, MS thesis,
AFIT/GE/ENG/92D-07, School of Engineering, Air Force Institute of Technology
(AU), Wright-Patterson AFB OH, December 1992.
2. Buzogany, Louis E., M. Pachter and J.J. D'Azzo, Automated Control of Aircraft in
Formation Flight, Proceedings of the 1993 AIAA Guidance, Navigation, and Control
Conference, pp 1349 - 1369, Monterey, CA, August 1993.
3. Dargan, John L., Proportional Plus Integral Control of Aircraft for Automated Maneu-
vering Formation Flight, MS thesis, AFIT/GE/ENG/91D-14, School of Engineering,
Air Force Institute of Technology (AU), Wright-Patterson AFB OH, December 1991.
4. Dargan, John L., M. Pachter and J.J. D'Azzo, Automatic Formation Flight Control,
Proceedings of the 1992 AIAA Guidance, Navigation, and Control Conference, pp 838
- 857, Hilton Head, SC, June 1992.
5. D'Azzo, John J. and Constantine H. Houpis, Linear Control System Analysis and
Design (Fourth Edition), McGraw-Hill Book Company, 1995.
6. McCammish, Sean J., Advanced Formation Flight Control, MS thesis,
AFIT/GE/ENG/95D-16, School of Engineering, Air Force Institute of Technology
(AU), Wright-Patterson AFB OH, December 1995.
7. Miller, Rüssel B., Manual Tracking Flight Control with Amplitude and Rate Con-
strained Dynamic Actuators, PhD Dissertation, AFIT/DS/ENG/96-15, School of
Engineering, Air Force Institute of Technology (AU), Wright-Patterson AFB OH,
December 1996.
8. Pachter, M., J.L. Dargan, and J.J. D'Azzo, Automatic Formation Flight Control,
AIAA Journal of Guidance, Control, and Dynamics, Vol 17, No. 6, May 1994.
9. Reyna, Vincent P., Automation of Formation Flight Control, MS thesis,
AFIT/GE/ENG/94M-01, School of Engineering, Air Force Institute of Technology
(AU), Wright-Patterson AFB OH, March 1994.
10. Reyna, Vincent P., Formation Flight Control Automation, Proceedings of the 1994
AIAA Guidance, Navigation, and Control Conference, pp 1379-1404, Scottsdale, AZ,
August 1994.
11. Rohs, Paul R., A Fully Coupled, Automated Formation Control System for Dissimilar
Aircraft in Maneuvering, Formation Flight, MS thesis, AFIT/GE/ENG/91M-03,
School of Engineering, Air Force Institute of Technology (AU), Wright-Patterson AFB
OH, March 1991.
12. Veth, Michael J., Advanced Formation Flight Control, MS thesis,
AFIT/GE/ENG/94D-30, School of Engineering, Air Force Institute of Technology
(AU), Wright-Patterson AFB OH, December 1994.
13. Blake, W., Dieter Multhopp, Design, Performance And Modeling Considerations
For Close Formation Flight, AIAA Journal of Guidance, Control, and Dynamics,
AIAA-98-4343, August 1998.
14. Pachter, Meir, Proffessor, Electrical Engineering, Personal interviews and Notes, Air
Force Institute of Technology, Wright-Patterson AFB OH, Jan 1999.

BIB-1
15. Blakelock, John H., Automatic Control of Aircraft and Missiles (Second Edition),
John Wiley & Sons, Inc., 1991.
16. Nelson, Robert C, Flight Stability and Automatic Control (Second Edition),
McGraw-Hill, 1998.
17. Matlab Simulink: Dynamic Simulation Software , The MathWorks, Inc., 1994.
18. Park, Chil Ho, Korean Air Forces F-4 Pilot, Republic of Korea, Personal interview,
Air Force Institute of Technology, Wright-Patterson AFB OH, Oct 1998.
19. Zumwalt, Michael, U.S. Air Force C-141 Pilot, WPAFB, Personal interview, Air
Force Institute of Technology, Wright-Patterson AFB OH, Oct 1998.

BIB-2
 Form Approved
REPORT DOCUMENTATION PAGE OMB No. 0704-0188
Public reporting burden for this collection of information if estimated to average 1 hour per response, including the time for reviewing instructions, searching existing a^tasources,
gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspeaI ofthis
collection of information? including suggestions for reducing this burden, to Washington Headquarters Services Directorate for Informator, Operations^and Reports 1215 Jefferson
Davis Highway Suite 1204, Arlington/VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Proiect (0704-0188), Washington. DC 20503.

1. AÖENCV USE ÖNLV (leave blank) 2. REPORT DATK REPORT TVPE AND BATES COVERED
March 1999 Master's Thesis
4. TITLE AND SUBTITLE 5. FUNDING NUMBERS
CLOSE FORMATION FLIGHT CONTROL

6. AUTHORjS)
Andrew W. Proud
Captain, USAF

7. PERF6RMIN6 ORGANIZATION NAMEtSIAN6 APBRESStESi ö. PERFORMING ORGANIZATION

REPORT NUMBER
Air Force Institute of Technology
2950 P Street AFIT/GE/ENG/99M-24
Wright-Patterson AFB, OH 45433-6583

&. SPONSORING/MONITORING AGENCV NAMEISI AND A60AESS(ES> 1o. SPONSORINO/MONITORING

AGENCY REPORT NUMBER
Mr William Blake
AFRL/VAAD
2210 8th Street
Wright-Patterson AFB, OH 45433

11. SUPPLEMENTARY NOTES

Dr. Meir Pachter
255-3636 x4593
[email protected]
•pa
12a. DISTRIBUTION AVAILABiLITV STATEMENT 12b. DISTRIBUTION CODE
Approved for public release; Distribution Unlimited

13. ABSTRACT /Maximum 2ÖÖ words)

In this paper the close formation flight control problem is addressed. The formation consists of a lead and wing aircraft,
where the wing flies in close formation with the lead, such that the lead's vortices produce aerodynamic coupling effects,
and a reduction in the formation's drag is achieved. A controller, i.e., a formation-hold autopilot for the wing aircraft, is
designed such that the formation's geometry is maintained in the face of lead aircraft maneuvers. In the formation flight
control system, the wing and lead aircraft dynamics are coupled due to kinematic effects, and, in the case of close
formations, additional aerodynamic coupling effects are introduced. In the paper these additional aerodynamic coupling
effects are properly modeled. The most significant aerodynamic coupling effect introduced by close formation flight
entails the coupling of the lateral/directional channel into the altitude-hold autopilot channel. It is shown that formation
hold autopilots designed ignorin the aerodynamic coupling effects yield satisfactory performance in close formation
flight.

14. SUBJECT TERMS 15. NUMBER OF PAGES

Close Formation Flight Control, Formation Auto-pilot, Formation Autopilot, F-16, Flight 146
Control, Horseshoe Vortex, UAV 16. PRICE CODE

17. SEOURITV CLASSIFICATION IS. SEOURITV CLASSIFICATION 1ä. SECURTTY CLASSIFICATION äfl. LIMITATION OF AB5TRACI
OF REPORT OF THIS PAGE OF ABSTRACT

UNCLASSIFIED UNCLASSIFIED UNCLASSIFIED UL

Standard Form 538 (Rev. 5-831 (LÜI
Prescribed by ANSI Std. 239.18
Designed using Perform Pro, WHS/DIOR, Oct 94