development of a neurofuzzy control system for the guidance of air to air missiles (master...
TRANSCRIPT
i
DEVELOPMENT OF A NEUROFUZZY CONTROL SYSTEM FOR
THE GUIDANCE OF AIR TO AIR MISSILES
By
Eng. Ahmed Momtaz Ahmed
A Thesis submitted to
The Faculty of Engineering, Cairo University
In partial fulfillment of the degree of
MASTER OF SCIENCE
in
Mechanical Design and Production
Under the Supervision of
Prof. Galal A. Hassaan
Professor of System Dynamics
Dr. Yasser F. Zeyada
Assistant Professor of System Dynamics
Department of Mechanical Design and Production Faculty of Engineering, Cairo University
Giza, Egypt
2004
ii
ACKNOWLEDGEMENT
The author would like to acknowledge with great thanks Prof. Galal A. Hassaan,
Department of Mechanical Design and Production, Cairo University, for his supervision,
sincere help and valuable advice.
The author would like also to thank Dr. Yasser F. Zeyada, for the great efforts he
offered to prepare this thesis.
Thanks also to the staff members of the Department of Mechanical Design and Production,
Cairo University who offered their help and support.
iii
SUMMARY
Recently, neurofuzzy modeling techniques have been successfully applied to modeling
complex systems, where traditional approaches hardly can reach satisfactory results due
to lack of sufficient domain knowledge. Much research has been done on applications of
neural networks (NNs) for identification and control of dynamic systems. According to
their structures, the NNs can be mainly classified as feedforward neural networks and
recurrent neural networks. It is well known that a feedforward neural network is
capable of approximating any continuous functions closely. However, the feedforward
neural network is a static mapping. Without the aid of tapped delays the feedforward
neural network is unable to represent a dynamic mapping. Although much research has
used the feedfoward neural network with tapped delays to deal with dynamical problem,
the feedforward neural network requires a large number of neurons to represent a
dynamic response in the time domain. Moreover, the weight updates of the feedforward
neural network do not utilize the internal information of the neural network and the
function approximation is sensitive to the training data. On the other hand, recurrent
neural networks (RNNs) have superior capabilities than the feedforward neural
networks, such as dynamic and ability to store information for later use. Since recurrent
neuron has an internal feedback loop, the RNN is a dynamic mapping and demonstrates
good control performance in the presence of uncertainties, which is usually composed of
unpredictable plant parameter variations, external force disturbance, unmodeled and
nonlinear dynamics, in practical application of air-to-air missile.
In recent years, the concept of incorporating fuzzy logic into a neural network has been
grown into a popular research topic. In contrast to the pure neural network or fuzzy
system, the fuzzy neural network (FNN) possesses both their advantages. It combines the
capability of fuzzy reasoning in handling uncertain information and the capability of
artificial neural networks in learning from processes. In missiles controlling branch it is
needed to minimize the output performance errors until fulfilling the required
specifications such as the minimum locking time and the maximum fuse distance.
Nowadays we have different types of missiles using different types of control strategy
iv
some of them uses reference frame unit as they controlled by a remote base or an
aircraft. Consequently controlling such missiles by reference angles ,, with respect
to fixed frame, other type of control strategy doesn’t use reference frame unit (fire and
forget types) such as all heatseeking, active radar missiles. In this paper, both types of
control strategy are discussed. A technique for the preliminary design of a control
system is presented using neurofuzzy systems for a diverse data range of a highly
nonlinear MIMO 5 DOF model (multi-input multi-output 5 degree of freedom AIM 9R
air to air missile). This system is composed of 3-body axes velocities and rotating in both
pitch and yaw direction only as the rolling motion is prevented due to gyroscopic
stabilizers (anti roll system) attached at the rear fins. The model has cross coupling
effects between the longitudinal and lateral motions due to the coupled equations of
motion. MIMO model was divided into two separated SISO models (single-input single-
output) taking into consideration the nonlinear cross coupling effects, then generating
the neural network controller for each separated SISO model (single-input single-output
model), blending the output performance using a 25 rule base Mamdani type fuzzy logic
controller. A Mamdani fuzzy logic controller can deal with the system uncertainty due
to cross coupling effect and affecting gust in the whole system when instantaneously
applying both inputs to the model. The main goal is to fulfill certain specifications within
its limited ranges such as proximity fuse distance and the target locking time during the
whole mission of the missile consequently keeping the error between actual and desired
performance according these tolerances. The first obvious error is due to the AOA
(angle of attack of the missile) and SSA (side slipping angle), resulting in error in the
desired trajectory itself as the (camera) detector is attached to the body frontal area of
the missile that is inclined to the actual body velocity vector. A second error is due to the
error between the desired and actual trajectory of the missile and this is actually due to
the whole system performance (AIM_9R dynamic model and controllers). To minimize
the second error over different regimes of flight, a hierarchical design of multi NN
controllers using a TSK(Takagi Suguno Kang) fuzzy fusion system classifier is
recommended which compute the weight of each controller according to the flight
regime parameters such as (altitude, velocity, required track rate [target-seeker
distance]). Fortunately both types of performance errors are damping each other to
some extent, this will lead to minimize the net output error of the actual performance.
v
The model was established and simulated through (MATLAB-SIMULINK), and the
final performance was examined using FLC fuzzy logic controllers with and without
NNC neural network controllers.
vi
LIST OF CONTENTS
ACKNOWLEDGMENT i
SUMMARY ii
LIST OF CONTENTS vi
ABBREVIATIONS viii
NOMENCLATURE ix
LIST OF FIGURES xi
LIST OF TABLES xv
Chapter 1 GENERAL BACKGROUND ON AIR TO AIR
MISSILES
1.1 Historical Background 1
1.2 Overview 8
1.3 AIM Missiles 13
Chapter 2 LITERATURE REVIEW
2.1 Introduction 32
2.2 Conventional Guidance and Control Design 34
2.3 Neural Net-based Guidance and Control Design 37
2.4 Fuzzy Logic-Based Guidance and Control Design 43
2.5 Gain-Scheduling Guidance and Control Design 49
2.6 Guidance Laws 53
2.7 Advantages over conventional designs 55
2.8 Problem Formulation 56
2.9 Previous Work 56
2.10 Thesis Layout 56
vii
Chapter 3 MODELING OF AIR TO AIR MISSILE
3.1 Introduction 58
3.2 The Rigid –Body Equations 58
3.3 Evaluation of the Angular Momentum 61
3.4 Euler’s Equations of Motion 62
3.5 Orientation and Position of the Missile 63
3.6 The Flight Path Determination 65
3.7 The Orientation of the Missile 66
3.8 The Applied Forces 68
3.9 AIM 9R Model Characteristics 73
3.9.1 Model verification 73
3.9.2 Cross effect representation 74
3.9.3 Gust effect 75
3.10 Summary 76
Chapter 4 NEURAL NETWORK FOR MISSILE GUIDANCE
4.1 Introduction 77
4.2 Theory and Examples 80
4.2.1 Single-input neuron 80
4.2.2 Multiple-input neuron 80
4.2.3 A layer of neurons 81
4.2.4 Multiple layer of neurons 82
4.3 Neural Network Controllers 83
viii
4.3.1 Control with inverse models 83
4.3.2 Control by feedfoward models 86
4.3.3 Control by feedback linearization 87
4.4 Summary 91
Chapter 5 MISSILE GUIDANCE USING FUZZY LOGIC
CONTROL
5.1 Introduction 92
5.2 Applying Fuzzy Inference System for Control 92
5.2.1 Designing membership functions (Fuzzification process) 93
5.2.2 Rule derivation 95
4.2.3 Fu 5.2.3 Fuzzy rule-based inference 97
5.2.4 Response pattern of a simple generic fuzzy control system 100
5.3 Fuzzy Logic Control Application 102
5.4 Comparison between Different Fuzzy Logic Controllers and PD Controller 103
5.5 Low pass Filter 104
5.6 The Effect of Using Neural Network in the First Time Period 106
5.7 Fuzzy Logic Controller Output Signal 107
5.8 Neurofuzzy Structure 107
5.9 The Effect of Flight Regimes on Neurofuzzy Controller Performance 108
5.10 Composite Fuzzy Guidance Law 108
5.11 Summary 120
Chapter 6 A HYBRID NEUROFUZZY CONTROL SYSTEM FOR
MISSILE GUIDANCE
ix
6.1 Introduction 121
6.2 Methodology 122
Chapter 7 DISCUSSIONS AND CONCLUSIONS
7.1 Discussions 128
7.2 Conclusions 130
7.3 Recommendation for Future Work 131
References 132
Appendix A Simulink models
x
ABBREVIATIONS
AAM Air to Air Missile
AFD Arm/Fire Device
AGM Air to Ground Missile
AIM Air Intercept Missile
AOA Angle of Attack
ASM Air to Surface Missile
CLOS Command to Line of Sight
DOF Degree Of Freedom
DSMAC Digital Scene Matching Area Correlation
F.O.R Frame Of Reference (at earth)
FLC Fuzzy Logic Controller
FNN Fuzzy Neural Network
GBU Ground Bomb Unit
GPS Global Positioning System
IR Infra Red
MIMO Multi Input Multi Output
NNC Neural Network Controller
OOLD Out-Of-Line Device
RF Radio Frequency
RNN Recurrent Neural Network
SISO Single Input Single Output
SSA Side Slipping Angle
TASM Tactical Air to Surface Missile
TDD Target Detection Device
TERCOM Inertial and Terrain Contour Matching
TIVS Thermally Initiated Venting System
TSK Tagaki Suguno Kang
USAF United States Air Force
WGU Wing Ground Unit
xi
NOMENCLATURE
ac Acceleration vector of missile mass center (m/s2)
(acx, acy, acz) Scalar components of ac in X, Y, Z directions respectively (m/s2)
A, B, C Moments of inertia about (x, y, z) axes respectively (kg/m2)
CL Lift coefficient
CD Drag coefficient
CN Side force coefficient
Cm Pitching moment coefficient
Cn Yawing moment coefficient
D Product of inertia yz (kg/m2)
E Product of inertia xz (kg/m2)
J Product of inertia xy (kg/m2)
ea, ee, er Mass eccentricities of control surfaces (m)
F Resultant external force vector (N)
(Fx,Fy,Fz) Scalar components of F in X, Y, Z directions respectively (N)
(Fa, Fe, Fr) Generalized control forces for aileron, elevator and rudder deflection
respectively (N)
f Generalized force in Lagrange’s equation (N)
G Resultant external moment vector (N.m)
h Angular momentum vector of the missile (kg.m2/s)
h’ Angular momentum vector of spinning rotor inside the missile (kg.m2/s)
hx, hy, hz Scalar components of h in X, Y, Z directions respectively (kg.m2/s)
h'x, h’y, h’z Scalar components of h’ in X, Y, Z directions respectively (kg.m2/s)
Ha, He, Hr Hinge moments on aileron, elevator and rudder respectively (N.m)
Ia, Ie, Ir Effective moments of inertia of control systems (kg/m2)
L, M, N Scalar components of G (N.m)
ma, me, mr Masses of aileron, elevator and rudder respectively (kg)
P, Q, R Scalar components of ω (rad/s)
xii
Sw Wing surface area (m2)
Vc Absolute velocity of the missile (m/s)
U, V, O Scalar components of Vc in X, Y, Z direction (m/s)
X, Y, Z Missile body axes components in X, Y, Z directions respectively
x’, y’, z’ Fixed frame of reference axes components in X, Y, Z directions respectively
α Angle of attack (rad)
β Side angle (rad)
δa, δe, δr Deflection angles of aileron, elevator and rudder respectively (rad)
ρ Air density (kg/m3)
Ψ, Θ, Φ Euler equation angles (rad)
ω Absolute angular velocity of the missile (rad/s)
xiii
LIST OF FIGURES
Fig. (1.1) Henschel Hs 293 air-to-ship, wireless guided, gliding bomb A 1
Fig. (1.2) Henschel Hs 293 air-to-ship television guided gliding bomb A 2
Fig (1.3) The Ruhrstahl SD 1400 'Fritz X' air-to-ship, wireless guided 6
Fig. (1.4) The Ruhrstahl SD 1400 'Fritz X' air-to-ship, wireless guided 7
Fig. (1.5) Types of ASM missiles 9
Fig. (1.6) Russian AAM 10
Fig. (1.7) TERCOM, DSMAC missiles guidance technique 11
Fig. (1.8) Tomahawk missile 13
Fig. (1.9) Chinese nuclear missiles DONG FENG/JULANG SERIES 13
Fig. (1.10) Missile components 14
Fig. (1.11) Active homing system 15
Fig. (1.12) Semiactive homing system 16
Fig. (1.13) Comparison between AIM 7 and AIM 120 guidance technique 23
Fig. (1.14) AIM 7 air to air missile 26
Fig. (2.1) Supervisory control scheme 38
Fig. (2.2) Hybrid control scheme 39
Fig. (2.3) Model reference control scheme 40
Fig. (2.4) Internal model control scheme 41
Fig. (2.5) Adaptive control scheme 41
Fig. (2.6) Predictive control scheme 42
Fig. (2.7) Model reference control of coupled lateral dynamics 43
Fig. (2.8) Fuzzy PD controller 45
Fig. (2.9) Typical adaptive fuzzy control scheme 46
Fig. (2.10) Fuzzy sliding mode control scheme 46
Fig. (2.11) Fuzzy model-following control scheme 47
Fig. (2.13) A fuzzy terminal guidance system 49
Fig. (2.14) Conventional gain-scheduling control scheme 50
Fig. (2.15) Fuzzy gain-scheduling control scheme 51
xiv
Fig. (2.16) Neural network gain-scheduling PID control scheme 52
Fig. (2.17) Neural-fuzzy gain-scheduling control scheme 53
Fig. (2.18) Integrated control model for the guidance of AIM 9R missile 57
Fig. (3.1) Missile body axes 58
Fig. (3.2) Missile orientation 64
Fig. (3.3) The applied forces 67
Fig. (3.4) Missile model algorithm 71
Fig. (3.5) AIM 9 Nonlinear Model 72
Fig. (3.6) Pitch angle step response for 5 DOF and 3 DOF models 73
Fig. (3.7) Pitch angle dynamic response for 3D, 2D models 73
Fig. (3.8) Cross coupling effect (yawpitch) 74
Fig. (3.9) Mean value of cross coupling effect versus input frequencies 74
Fig. (3.10) Standard deviation of cross coupling effect 75
Fig. (3.11) Gust effect (pitchyaw) 75
Fig. (4.1) Single Input neuron 80
Fig. (4.2) Multiple Input Neuron 81
Fig. (4.3) Layer of (S) Neurons 82
Fig. (4.4) Direct inverse control 84
Fig. (4.5) Direct Inverse Control at steady state flight 84
Fig. (4.6) Direct Inverse Control at climb angle =90 deg 85
Fig. (4.7) Direct Inverse Control at initial velocity =700 m/s 85
Fig. (4.8) Control by Feedforward 86
Fig. (4.9) Control by Feedforward at steady state flight 86
Fig. (4.10) Control by feedback linearization 87
Fig. (4.11) Control by Feedback Linearization at steady state flight 88
Fig. (4.12) Pitch motion using NN controller with combined sine wave input 89
Fig. )4.13) Integrated square error using NN controller with combined sine wave
input
89
Fig. (4.14) Absolute velocity of the missile using NNC with combined sine wave 90
xv
input
Fig. (4.15) Angle of attack using NNC with combined sine wave input 90
Fig. (4.16) Elevator deflection angle using NNC with combined sine wave input 91
Fig. (5.1) The structure of a Fuzzy Logic Control System 93
Fig. (5.2) Triangular membership function 94
Fig. (5.3) Membership function for input (e) 94
Fig. (5.4) Membership function for input (ce) 94
Fig. (5.5) Membership function for output (u) 95
Fig. (5.6) Fuzzy matching process 97
Fig. (5.7) Fuzzy inference process 97
Fig. (5.8) Fuzzy combination process 98
Fig. (5.9) Defuzzification process 99
Fig. (5.10) Rule base calculations 99
Fig. (5.11) Architecture of the generic fuzzy control system 100
Fig. (5.12) Step response of a simple generic fuzzy controller 101
Fig. (5.13) Fuzzy inference system control action 102
Fig. (5.14) Comparison between fuzzy logic and NN controllers 102
Fig. (5.15) Effect of normalized and denormalized factors 103
Fig. (5.16) Comparison between PD and FL controllers with combined sine wave
input
103
Fig. (5.17) Integral square error for PD and FL controllers 104
Fig. (5.18) Dynamic response of a low pass filter 104
Fig. (5.19) Integrated square error for different fuzzy logic controllers using low
pass filter
105
Fig. (5.20) Error rate using filter 105
Fig. (5.21) Error rate using delay only 106
Fig. (5.22) Effect of using NNC for the first 5 seconds 106
Fig. (5.23) Error, error rate and control signal 107
Fig. (5.24) Neurofuzzy structure 107
Fig. (5.25) Two types of fuzzy logic controllers output performance 108
xvi
Fig. (5.26) Composite guidance law 109
Fig. (5.27) Defuzzification process 109
Fig. (5.28) Fusion Fuzzy system output K1 110
Fig. (5.29) Fusion Fuzzy system output K2 110
Fig. (5.30) Fusion Fuzzy system output K3 110
Fig. (5.31) Comparison between FL, PD controllers in pitch motion with combined
sine wave
111
Fig. (5.32) Integrated square error for PD and FL controllers with combined sine
wave
111
Fig. (5.33) Angle of attack using FLC with combined sine wave input 112
Fig. (5.34) Elevator deflection angle using FLC with combined sine wave input 112
Fig. (5.35) Absolute velocity of the missile using FLC with combined sine wave
input
113
Fig. (5.36) Comparison between fuzzy logic controller and PD controller for .3 rad
step input
114
Fig. (5.37) Comparison between FLC and PD controller for a step response 114
Fig. (5.38) Elevator deflection angle for .3 rad step input using FLC 115
Fig. (5.39) Angle of attack for .3 rad step input using FLC 115
Fig. (5.40) Absolute velocity for .3 rad step input using FLC 116
Fig. (5.41) Path angle response using PD, FL controllers with ramp input 117
Fig. (5.42) Comparison between integrated square error for PD, FL controllers with
ramp input
117
Fig. (5.43) Angle of attack using FLC with ramp input 118
Fig. (5.44) Missile velocity using PD, FL controllers with ramp input 118
Fig. (5.45) Elevator deflection angle using PD, FL controllers with ramp input 119
Fig. (6.1) AIM 9 Model with 2 separate controllers 121
Fig. (6.2) Fig. (5.2) NeuroFuzzy structure 122
Fig. (6.3) Target location transformation from fixed frame of reference coordinate
to missile body axes coordinate
123
Fig. (6.4) Actual trajectories for target and missile 124
Fig. (6.5) 3D Plot for missile trajectory in both X, Y axes with respect to time 124
Fig. (6.6) 3D Plot for missile trajectory in both X, Z axes with respect to time 125
Fig. (6.7) Calculation of Maximum fuse distance, and the best triggering point 126
xvii
Fig. (6.8) Error in Z direction due to AOA, control system error 127
Fig. (6.9) Error in Y direction due to SSA, control system error 127
LIST OF TABLES
Table (1.1) Tomahawk specification 12
Table (1.2) Major types (J, L, M, P, R) of AIM 9 missiles 20
Table (1.3) Major types (B, D, E, G, H) of AIM 9 missiles 21
Table (1.4) Data specifications of AIM 7 25
Table (5.1) Control rules for simple generic fuzzy controller 101
Table (7.1) ISE for PD, NN, FL controllers 129
Table (7.2) Comparison between NNC and FLC for the first 5_seconds 129
xviii
ABBRIVIATIONS
A/C Aircraft
AAM Air to Air Missile
AFD Arm/Fire Device
AGM Air to Ground Missile
AIM Air Intercept Missile
AOA Angle of Attack
ASM Air to Surface Missile
C/L Center Line
DOF Degree Of Freedom
DSMAC Digital Scene Matching Area Correlation
F.O.R Frame Of Reference
F.O.R Frame of Reference (at earth)
FLC Fuzzy Logic Controller
FNN Feedforward Neural Network
GBU Ground Bomb Unit
GPS Globe Positioning System
IR Infra Red
MIMO Multi Input Multi Output
NNC Neural Network Controller
OOLD Out-Of-Line Device
RF Radio Frequency
RNN Recurrent Neural Network
SISO Single Input Single Output
SSA Side Slipping Angle
TDD Target Detection Device
TERCOM Inertial and Terrain Contour Matching
TIVS Thermally Initiated Venting System
TSK Tagaki Suguno
WGU Wing Ground Unit
1
Chapter (1)
GENERAL BACKGROUND ON AIR TO AIR MISSILE
1.1 Historical Background
During World War 2 there have been many ideas about guided weapons such as
guided bombs instead of direct projectiles or bombs. The beginning ideas that were to
evolve into the HenschelHs 293 [1] appeared in as early as 1939 {Fig. (1.1)}. In 1940, an
experimental model having the shape of a glider was built. The goal was to develop a
remote-controlled air-to-surface missile against ships. Development proceeded even
though no suitable rocket motors were available. The experimental model used a
standard SC 500 bomb with extra wings and tail unit but no rudder. Finally a
propulsion system was developed, and the liquid rocket was fitted under the main
missile body. An 18-channel radio system was used for control.
Fig. (1.1) HenschelHs 293 air-to-ship, wireless guided, gliding bomb, model A
The missile was designed to be carried under a parent bomber. Warm exhaust air from
the aircraft engines was channeled to the missile to prevent it from freezing up at high
altitudes. Once dropped the Hs 293 would fall for some 90m (295ft) before the rocket
achieved maximum thrust. The parent bomber would continue to fly a pre_designated
course parallel with the target. The Aircraft could visually track the missile with the aid
of red guidance flare in the tail, and control the projectile using a small control box with
a joystick. The actual flight path resembled a series of arcs as corrections were received
and followed.
2
The main weakness of the Hs 293A was that the parent bomber had to fly a steady, level
path. Evasive moves to avoid anti-aircraft fire were impossible, even though the Hs 293
outranged most ship-borne anti-aircraft guns. An improved Hs 293D with a television
camera installed in the head of the missile as aiming system was planned but the war
concluded before it could be realized. Also, the problem of icing was never resolved and
thus further propulsion units were designed. The war ended before these plans left the
experimental stage.
The Hs 293 was based on a normal 500 kg (1,102 lb) bomb with wings and fins added
and an engine suspended from the main body. The Hs 293V-4 and C-1 were guided by
radio, like the Fritz-X, but after the Germans found that the Allies were capable of
interfering with the control signals to the missile wire guidance was adopted.
The C-3, C-4, and A-0 versions of the Hs293 relied on wire guidance. During the
bomb's fall, two wired coils on the wing tips unwound, so maintaining the link with the
launch plane and allowing the transmission of electrical pulses for guidance.
A model D {Fig. (1.2)} was also built, which transmitted television images of the target to
the controller. Model H was supplied with an acoustic/magnetic detector to attack
bomber formations.
Fig. (1.2) HenschelHs 293 air-to-ship television_guided gliding bomb, model D (3views)
3
The guidance operators always sat on the starboard right side of the cabin, and
therefore it was always necessary to attack a target from the port side and in the same
movement direction. Another version from Henschel guided bomb was HenschelHs 294
which used the television guidance from launch plane.
In 1941,Professor Wagner conceived and designed the Schmetterling anti-aircraft
missile and submitted it for appraisal and approval to the RLM
(Reichsluftfahrtministerium, phew!) or the State Ministry of Aviation, which rejected
the design because its defensive purpose was considered weak and still unnecessary.
Official support finally came in 1943, when Allied strategic bombing had accelerated the
war to a higher stage, and the Reich was feeling the pain.
The Henschel company, which was also developing several other types of missiles, was
given the go-ahead and development contract to begin production in February 1945.
Experiments and production were interrupted by Allied bombers and there were only
59 experimental launchings, 34 of them failed for different reasons. Schmetterling was a
bomb-shaped projectile with wings and fins, and two rocket boosters were attached to
the outside for take-off.
The external rockets were powered by solid fuel, while the internal motor was driven by
a fuel composed of hydrocarbon and nitric acid. It could accelerate up to 200 m/s and
had an effective range of approximately 8 km with a ceiling of 1000m. A ground
controller would guide the missile through radio signals and track its course visually.
The warhead contained 20 kg of high explosive and was armed with a proximity fuse
called "Fox."
The lion's share of post-World War II literature concerning German guided missiles has
concentrated on the V-1 "Buzz Bomb" and the V-2 stratospheric rocket. However, the
guidance systems of these weapons were so rudimentary that a successful launch was
one which hit a target as vast as London. German science developed several other, more
sophisticated guidance systems for weapons which were used successfully in combat
before either the V-1 or V-2.
4
These first true guided missiles, by current definition, are virtually unknown today.
They were Henschel'sHs 293A and Ruhrstahl's SD1400X, known as "Fritz X." Both
missiles were air-launched from mother bombers and radio-controlled primarily against
maritime targets. It is interesting to note that of the fifteen battleships lost to airpower
(seven in the open sea) one of these, the 41,650-ton Italian flagship, Roma, was sunk by a
Fritz X. Later, the British battleship, Warspite, was hit by a Fritz X and put out of
action for six months. Fritz Xs also sunk the Royal Navy light cruiser, Spartan, heavily
damaged the cruiser USS Savannah, and hit the cruiser USS Philadelphia.
The Henschel program received a great boost when the RLM assigned Prof. Dr. Herbert
A. Wagner to assume control of their missile program. The brilliant Wagner had earlier
been dismissed by Junkers Aircraft as the company found it impossible to reconcile his
intermittent and desultory research methods with its ideas of series production. After
his dismissal, he carried out independent research on an anti-ship ricochet bomb.
In January 1940, Wagner joined Henschel's already distinguished missile research team.
This team of specialists included: ReinhardLahde and Otto Bohlmann (aeronautics),
Wilfried Hell, from Schwarz Propeller Werke (engineering), Josef Schwarzmann
(electronics), and Dr. Hinrici (design study). Theodor Sturm of
StassfurterRundfunkGessellschaft worked closely with the Henschel team in developing
and adapting radio systems.
The first Henschel design, designated Hs 293V-1, was a standard glider bomb bearing
slight resemblance to its Schwarz forerunner. The Hs 293V-1 did not leave the drawing
board.
The second design, the Hs 293V-2 (later FZ21) was built as a model in February 1940
and used in control tests at Karlshagen. It, too, was unarmed and powerless. Note that,
unlike German aircraft prototype 'V' numbers where only one example was built for
each V number, the Henschel missile V numbers referred to design alone and many
examples could have been built.
Further studies led to a new design in July 1940, the Hs 293V-3. The powerless
prototype was test launched from minimum release altitudes of 3,280 feet. Since the
5
RLM was primarily interested in an anti-ship missile, it was found that a pure glide
bomb lacked sufficient speed and weight to penetrate warship hulls.
The first successful test launching occurred two days later against a small barn as a
target. Subsequent tests ended in failure usually for two main reasons. First, the original
white tracking flares were difficult to follow and were replaced with red flares.
Secondly, the electrical system often failed due to design failure particularly with tubes
which were replaced with relays. In 1941, drop tests used a Heinkel He 177A-0 bomber
and later two He 177A-1s were used in the first successful tests.
The A-0 design led to the finalization of the production design, the Hs 293A-1, which
consisted of three pre-production models (V-2, V-3, A-0). One thousand nine hundred
were produced for the test and modification programs which continued into early 1944
(1,700 were A-0s). Early Hs 293 production was carried out at HenschelWerke 111,
Spandau-Hakenfelde but was transferred to Berlin-Schoenfeld in August 1943.
In early tests, bomb-aimers had difficulty following the bomb in flight due to its high
velocity. Also, the electromagnetically activated spoilers tended to jam. Pneumatic
power was tested but was found to develop problems in low temperatures causing the
electromagnetic system to be retained. It was not until early 1942, when a high-speed
wind tunnel had been constructed, that Dr. Kramer's control spoilers were modified and
that a velocity-reducing tail-mounted air brake was perfected.
As issued the Fritz X was the standard SD 100X armor-piercing (AP) bomb with four
centrally-mounted standard cruciform wings giving the bomb aerodynamic pivotal
points for control. The total span across the wings was 4 feet 5 inches. The tail assembly
consisted of four fins similar to those of an orthodox bomb, but were surrounded by the
air brake ring which contained the control spoilers and operating solenoids. The dive
brake was fitted around the tail fins to limit the terminal velocity of the bomb to 600
miles per hour. The dive brake was fabricated of metal but a section was electrically
insulated to serve as a radio antenna. The fuselage and tail measured 10 feet 8* inches
and had a diameter of 1 foot 10 inches.
6
The sudden but arranged capitulation of the Italian fleet to the Allies on 9 September
1943 spurred the Germans to take quick actions against their former ally. The battleship
Italia was damaged and Roma sunk as victims of a new type of air-to-surface weapon,
the Ruhrstahl/Kramer X-1 [1] as shown in Fig. (1.3) or Fritz X. It was a free-falling
bomb guided by the parent aircraft. Usually it was dropped at an altitude of about
6,000m (19,685ft); by the time of detonation it would have had gained a velocity close to
that of sound.
The first tests were conducted in Germany in 1942, and were moved to Italy later. In
Italy, pneumatic power was tested to substitute for the electromagnetic actuation of the
spoilers. However, variations in temperature in different parts of the atmosphere posed
obstacles and the idea had to be dropped. The Allied advance to Italy forced the
Germans to press the Fritz X into increased use. The cruiser USS Savannah was
attacked successfully alongside several naval transports. During a night attack two
British cruisers collided in utmost confusion. Seven days later the Germans scored hits
on the battleship HMS Warspite, which had to be towed to Malta.
Fig. (1.3) The Ruhrstahl SD 1400 'Fritz X' air-to-ship, wireless guided, gliding bomb
German wire-guided missiles belong generally to the X-series. The X-4air-to-air missile
shown in Fig. (1.4) was the first of the series to be developed in June 1943. It was a
finned and winged projectile with a wingspan of 6ft 6in, and powered by a rocket using
a combination of two fuels (T and C stoff). The wire, coiled around spools, was stored in
the wing tips and carried guidance signals to control spoiler tabs and ailerons on the
7
wings to alter direction. The warhead composed of 20 kg of explosive and a acoustic
proximity fuse called "Kranich."
Fig. (1.4) Restored Ruhrstahl X-4 Air-to-Air missile
It was certainly natural and reasonable that a ground-to-ground version of the X-4 [1]
was also developed, as an anti-tank weapon, in addition the successful Panzerfaust. This
new cousin missile was called the X-7. The X-7 was navigated and stabilized by a
gyroscope and guided by wire, like the X-4, but it contained a hollow charge warhead
armed with an impact fuse which could penetrate over 200mm of armor within 1000
meters. It was driven by a two-stage, solid-fuel rocket motor and weighed 10 kg.
In early 1943, work began on the X-4 air-to-air wire-guided missile by Dr. Kramer at
Ruhrstahl. The missile received a development order in the summer of 1943 and was
given the number 8-344 by the RLM, and was developed to give fighters a chance to
down the ever increasing number of Allied bombers from outside of their defensive gun
range.
By the early sixties the sidewinder waqs the principal heatseeker in western service it
was used by USAF, USN. Its early combat record was not spectacular, as the seeker
performance limitations were exacerbated by the poor reliability of the tube electronics
and the inexperience of its users. Kill probabilities were in the tens of percent, very
sensitive to how well the launch aircraft was positioned. Designed to intercept lumbering
bombers, the AIM was ill suited to knife-fights with MIG-17s at low level.
Nevertheless, no less than 28 MIGs were killed for 175 launches between 1965 and 1968,
by USAF F-4C/D aircraft, an aggregate (kill probability) of 16 %.
8
In 1973, Ford began production of an enhanced AIM-9J-L, later registered the AIM-
9N. The November model employed a similar configuration to the AIM-9B, but the main
printed circuits were redesigned to improve seeker performance.
In 1982 AIM-9L was produced which is essentially an improved AIM-9L. The Mile
has improved background rejection, counter-countermeasures capability and a low
smoke motor to reduce the visual signature of the inbound weapon.
1.2 Overview
Missiles are classified according to many specifications such as range, warhead
weight, guidance technique and mission type. Many types of missiles will be shown and
discussed briefly in the following pages. But the focus of this thesis will be the air to air
missiles. Air-to-air missiles have many guidance techniques such as active, semi-active
radar, laser and heatseeking missiles. These types of guidance technique are classified
according to the range of each missile. For example, radar homing missiles will have
long range than the other heatseeking ones according to the capability of the detecting
element itself. In this thesis, the air-to-air heatseeking missiles (SIDEWINDER AIM 9R)
will be discussed. This missile has a large number of produced units all over the world
about 21000 piece that from the USAF inventory. Table (1.1) gives the specifications of
this missile for many versions.
1.2.1 Missiles types
There are many common types of missiles such as AAM, ASM, SAM, TASM cruise
and ballistic missiles. The next section will discuss each type briefly. Each type of
missiles is classified in accordance with guidance techniques, target-missile range,
propulsion and type of the missile warhead. The basic type will be discussed in this
research is the AAM or AIM missiles.
1.2.1.1 ASM missiles
Figure (1.5) shows different types of ASM missiles.
9
Fig (1.5) Types of ASM Russian missiles
1.2.1.2 AAM missiles
Figure (1.6) shows different types of AAM missiles.
10
Fig (1.6) Russian AAM
1.2.1.3 Cruise missiles
Tomahawk is an all-weather submarine or ship-launched land-attack cruise missile.
After launch, a solid propellant propels the missile until a small turbofan engine takes
over for the cruise portion of flight. Tomahawk is a highly survivable weapon. Radar
detection is difficult because of the missile's small cross-section and low altitude flight.
11
Similarly, infrared detection is difficult because the turbofan engine emits little heat.
Systems include Global Positioning System (GPS) receiver; an upgrade of the optical
Digital Scene Matching Area Correlation (DSMAC) system; Time of Arrival (TOA)
control, and improved 402 turbo engines.
The Tomahawk land-attack cruise missile has been used to attack a variety of fixed
targets, including air defense and communications sites, often in high-threat
environments. The land attack version of Tomahawk has inertial and terrain contour
matching (TERCOM) radar guidance {Fig. (1.7)}. The TERCOM radar uses a stored
map reference to compare with the actual terrain to determine the missile's position. If
necessary, a course correction is then made to place the missile on course to the target.
Terminal guidance in the target area is provided by the optical Digital Scene Matching
Area Correlation (DSMAC) system, which compares a stored image of target with the
actual target image.
Fig (1.7) TERCOM, DSMAC missiles guidance technique
The Tomahawk missile {Fig. (1.8), Table (1.1)} provides a long-range, highly survivable,
unmanned land attack weapon system capable of pinpoint accuracy. The Surface Navy’s
deep strike capability resides in the Tomahawk missile system – the proven weapon of
choice for contingency missions.
12
Table (1.1) Tomahawk specifications
Specifications
Primary Function Long-range subsonic cruise missile for attacking land
targets.
Contractor Hughes Missile Systems Co., Tucson, Ariz.
Power Plant Williams International F107-WR-402 cruise turbo-fan
engine; solid-fuel booster
Length 18 feet 3 inches (5.56 meters); with booster: 20 feet 6
inches (6.25 meters)
Weight 2,650 pounds (1192.5 kg); 3,200 pounds (1440 kg) with
booster
Diameter 20.4 inches (51.81 cm)
Wing Span 8 feet 9 inches (2.67 meters)
Range Land attack, conventional warhead: 600 nautical miles
(690 statute miles, 1104 km)
Speed Subsonic - about 550 mph (880 km/h)
Guidance System Inertial and TERCOM
Warheads
Conventional: 400 kg Bullpup, or
Conventional submunitions dispenser with combined
effect bomblets, or
WDU-36 warhead w/ PBXN-107 explosive & FMU-148
fuze, or
200 kt. W-80 nuclear device
Date Deployed 1983
13
Fig (1.8) Tomahawk missile
1.2.1.4 Chinese nuclear missiles
Figure (1.9) shows different types of Chinese nuclear missiles.
Fig. (1.9) Chinese nuclear missiles DONG FENG/JULANG SERIES MISSILES
Chinese nuclear missiles are classified according the range an the warhead weight
most of them are guided either by inertial/fiber optic, strap down inertial system or
GPS.
1.3 AIM Missiles
1.3.1 Missile components
Guided missiles are made up of a series of subassemblies as shown in Fig. (1.10). The
various subassemblies form a major section of the overall missile to operate a missile
14
system, such as guidance, control, armament (warhead and fusing), and propulsion. The
major sections are carefully joined and connected to each other. They form the complete
missile assembly. The arrangement of major sections in the missile assembly varies,
depending on the missile type.
The guidance section is the brain of the missile. It directs its maneuvers and causes the
maneuvers to be executed by the control section. The armament section carries the
explosive charge of the missile, and the fusing and firing system by which the charge is
exploded. The propulsion section provides the force that propels the missile.
Fig. (1.10) Missile components
1.3.1.1 Guidance and control section
The complete missile guidance system includes the electronic sensing systems that
initiate the guidance orders and the control system that carries them out. The elements
for missile guidance and missile control can be housed in the same section of the missile,
or they can be in separate sections.
15
There are a number of basic guidance systems used in guided missiles. Homing-type,
air-launched, guided missiles are currently used. They use radar or infrared homing
systems. A homing guidance system is one in which the missile seeks out the target,
guided by some physical indication from the target itself. Radar reflections or thermal
characteristics of targets are possible physical influences on which homing systems are
based. Homing systems are classified as active, semiactive, and passive.
(a)-Active homing system
In the active homing system as shown in Fig. (1.11), target illumination is supplied by a
component carried in the missile, such as a radar transmitter. The radar signals
transmitted from the missile are reflected off the target back to the receiver in the
missile. These reflected signals give the missile information such as the target's distance
and speed. This information lets the guidance section compute the correct angle of
attack to intercept the target. The control section that receives electronic commands
from the guidance section controls the missile’s angle of attack. Mechanically
manipulated wings, fins, or canard control surfaces are mounted externally on the body
of the weapon. They are actuated by hydraulic, electric, or gas generator power, or
combinations of these to alter the missile's course.
Fig. (1.11) Active homing system
(b)-Semiactive homing system
In the semiactive homing system as shown in Fig (1.12), the missile gets its target
illumination from an external source, such as a transmitter carried in the launching
16
aircraft. The receiver in the missile receives the signals reflected off the target, computes
the information, and sends electronic commands to the control section. The control
section functions in the same manner as previously discussed.
Fig. (1.12) Semiactive homing system
(c)-Passive homing system
In the passive homing system, the directing intelligence is received from the target.
Examples of passive homing include homing on a source of infrared rays (such as the
hot exhaust of jet aircraft) or radar signals (such as those transmitted by ground radar
installations). Like active homing, passive homing is completely independent of the
launching aircraft. The missile receiver receives signals generated by the target and then
the missile control section functions in the same manner as previously discussed.
1.3.1.2 Armament Section
The armament system contains the payload (explosives), fusing, Safety and Arming
(S&A) devices, and target-detecting devices (TDDs).
(a)-Payload
The payload is the element or part of the missile that does what a particular missile is
launched to do. The payload is usually considered the explosive charge, and is carried in
the warhead of the missile. High-explosive warheads used in air-to-air guided missiles
17
contain a rather small explosive charge, generally 10 to 18 pounds of H-6, HBX, or PBX
high explosives. The payload contained in high-explosive warheads used in air-to-
surface guided missiles varies widely, even within specific missile types, depending on
the specific mission. Large payloads, ranging up to 450 pounds, are common. Comp B
and H-6 are typical explosives used in a payload. Most exercise warheads used with
guided missiles are pyrotechnic signaling devices. They signal fuse functioning by a
brilliant flash, by smoke, or both. Exercise warheads frequently contain high explosives,
which vary from live fuses and boosters to self-destruct charges that can contain as
much as 5 pounds of high explosive.
(b)-Fusing
The fusing and firing system is normally located in or next to the missile's warhead
section. It includes those devices and arrangements that cause the missile's payload to
function in proper relation to the target. The system consists of a fuse, a Safety and
Arming (S&A) device, a Target-Detecting Device (TDD), or a combination of these
devices. There are two general types of fuses used in guided missiles—proximity fuses
and contact fuses. Acceleration forces upon missile launching arm both fuses. Arming is
usually delayed until the fuse is subjected to a given level of accelerating force for a
specified amount of time. In the contact fuse, the force of impact closes a firing switch
within the fuse to complete the firing circuit, detonating the warhead. Where proximity
fusing is used, the firing action is very similar to the action of proximity fuses used with
bombs and rockets.
(c)-Safety and arming devices
S&A devices are electromechanical, explosive control devices. They maintain the
explosive train of a fusing system in a safe (unaligned) condition until certain
requirements of acceleration are met after the missile is fired.
(d)-Target detecting devices
TDDs are electronic detecting devices similar to the detecting systems in VT fuses. They
detect the presence of a target and determine the moment of firing. When subjected to
18
the proper target influence, both as to magnitude and change rate, the device sends an
electrical pulse to trigger the firing systems. The firing systems then act to fire an
associated S&A device to initiate detonation of the warhead. Air-to-air guided missiles
are normally fused for a proximity burst by using a TDDwith an S&A device. In some
cases, a contact fuse may be used as a backup. Air-to-surface guided missile fusing
consists of influence (proximity) and/or contact fuses. Multifusing is common in these
missiles.
1.3.1.3 Propulsion section
Guided missiles use some form of jet power for propulsion. There are two basic types of
jet propulsion power plants used in missile propulsion systems—the atmospheric (air-
breathing) jet and the thermal jet propulsion systems. The basic difference between the
two systems is that the atmospheric jet engine depends on the atmosphere to supply the
oxygen necessary to start and sustain burning of the fuel. The thermal jet engine
operates independently of the atmosphere by starting and sustaining combustion with its
own supply of oxygen contained within the missile.
There are three types of atmospheric jet propulsion systems—the turbojet, pulsejet, and
ramjet engines. Of these three systems, only the turbojet engine is currently being used
in Navy air-launched missiles. A typical turbojet engine includes an air intake, a
mechanical compressor driven by a turbine, a combustion chamber, and an exhaust
nozzle. The engine does not require boosting and can begin operation at zero
acceleration.
Thermal jets include solid propellant, liquid propellant, and combined propellant
systems. As an AO, you come in contact with all three systems. The solid propellant and
combined propellant systems are currently being used in some air-launched guided
missiles. The majority of air-launched guided missiles used by the Navy use the solid
propellant rocket motor. They include the double base and multibase smokeless powder
propellants as well as the composite mixtures. Grain configurations vary with the
different missiles. Power characteristics and temperature limitations of the individual
rocket motors also vary. In some guided missiles, different thrust requirements exist
19
during the boost phase as compared to those of the sustaining phase. The dual thrust
rocket motor (DTRM) is a combined system that contains both of these elements in one
motor. The DTRM contains a single propellant grain made of two types of solid
propellant—boost and sustaining. The grain is configured so the propellant meeting the
requirements for the boost phase burns at a faster rate than the propellant for the
sustaining phase. After the boost phase propellant burns itself out, the sustaining
propellant sustains the motor in flight over the designed burning time (range of the
missile).
1.3.2 Sidewinder AIM 9 missiles
Tables (1.2) and (1.3) show the characteristics of major types of AIM 9 missiles.
20
Table (1.2) Characteristics of major types of AIM 9 missiles.
Subtype AIM-9J AIM-9L AIM9-M AIM-9P AIM-9R
Service USAF Joint Joint USAF USN
Origin AIM-9E AIM-9H AIM-9L AIM-9J/N AIM-9M
Detector PbS InSb InSb InSb Focal
PlaneArray
Cooling Peltier Argon Argon Argon
Dome MgF2 MgF2 MgF2 MgF2 Glass
Reticle speed
(Hz)
100 125 125 100 Focal Plane
Array
Modulation AM FM FM FM Focal
PlaneArray
Tracking rate
(deg/s)
16.5 classified Classified >16.5 Classified
Electronics Hybrid Solid state Solid state Solid state Solid state
Warhead Blast
/fragmentation
Annular BF Annular BF Annular
BF
Annular BF
Fuse Passive-IR IR/Laser IR/Laser IR/Laser IR/Laser
Manufacturer Hercules
/Aerojet
Hercules
/Bermite
MTI/Hercules Hercules
/Aerojet
MTI/Hercules
Type MK.17 MK36 MOD.7,8 MK.36 MOD.9 SR.116 MK.36MOD.9
launcher AERO-111 Common Common Common Common
Length (m) 3 2.9 3 3 2.9
Span (m) 0.55 0.6 0.6 0.55 0.6
Mass (kg) 70 90 90 90 90
Velocity
(Mach)
<3 <3 <3 <3 <3
21
Table (1.3) Characteristics of AIM-9B, 9D, 9E , 9G, 9H
Subtype AIM-9B AIM-9D AIM-9E AIM-9G AIM-9H
Service Joint USN USAF USN USN
Origin NWC AIM-9B AIM-9B AIM-9D AIM-9G
Detector PbS PbS PbS PbS PbS
Cooling Uncooled Nitrogen Peltier Nitrogen Nitrogen
Dome
window
Glass MgF2 MgF2 MgF2 MgF2
Reticle speed
(Hz)
70 125 100 125 125
Modulation AM AM AM AM AM
Track rate
(deg/s)
11.0 12.0 16.5 12.0 >12.0
Electronics Thermionic Thermionic hybrid Thermionic Solid state
Warhead Blast/frag. Continuous
rod
Blast/frag. Continuous
rod
Continuous
rod
Fuse Passive-IR Passive-
IR/HF
Passive-IR Passive-
IR/HF
Passive-
IR/HF
Manufacturer Thiokol Hercules Thiokol Hercules Hercules
Bermite
Type MK.17 MK.36 MK.17 MK.36 MK.36
launcher Aero-111 LAU-7A Aero-111 LAU-7A LAU-7
Length(m) 3 2.8 2.9 2.8 2.8
Span(m) 0.5 0.6 0.5 0.6 0.6
Mass (kg) 70 90 75 90 89
Velocity
(Mach)
<3 <3 <3 <3 <3
22
1.3.2.1 The imaging Sidewinder-AIM-9R
The AIR-9R is the latest production sidewinder, using an imaging seeker which is a
fundamental departure from the established design. Developed by the naval weapons
center, the AIM-9R uses a modified AIM-9R control actuator, while retaining the fuse,
warhead, motor, wings and canards of its predecessor.
The imaging seeker is built around a focal plane array imaging device, analogous to
the CCDs employed in modern television cameras. A focal plane array has a much
greater instantaneous field of view than a reticle seeker (conventional technique), and
stares at the target and its immediate background, tracking the target by means of
contrast lock similar to that employed by TV guided weapons such as Maverick or
GBU-15. In this fashion, the seeker can account for the background contrast and reject
it, while also providing the potential to discriminate between multiple targets and
countermeasures such as flares.
The WGU-19 seeker uses a three gimbal stabilized platform mounting a visible band
focal plane array device, most likely a 256X256 element InSb array, or a higher
resolution PtSi device on a peltier cooled substrate, to provide coverage down to 4
micron band. The video signal produced by the array is then digested and processed by
a software programmable digital image processor, which tracks the targets and feeds
the autopilot with data so that it can send steering commands to the control actuators.
The AIM-9R is the latest naval sidewinder, using the airframe, fuse and motor of the
AIM-9M with a new digital imaging seeker. The new seeker employs a focal plane array
imaging device effective to visual wavelengths, mounted on a gimbaled stabilized
platform. The use of the imaging seeker has provided a vast improvement in target
detection range, off bore-sight angle, rejection of background and ability to selectively
aim for vulnerable areas of the target. Imaging seekers are immune to jamming
techniques effective against reticle seekers.
23
The AIM-9 provides a major increase in target acquisition range over established
subtypes, with much better tracking performance, and the ability to reject both
background terrain and clouds. The total field of view of the seeker is much greater,
allowing acquisition of off-bore-sight and maneuvering targets, while the software
provides for intelligent selection of an aim point when impacting a target, also the
manufacturer claim effective counter-countermeasures capability against known and
postulated jamming or seduction techniques.
1.3.3 AIM-120 AMRAAM Slammer
The AIM-120 advanced medium-range air-to-air missile (AMRAAM) is a new
generation air-to-air missile. It has an all-weather, beyond-visual-range capability and
is scheduled to be operational beyond 2000. AMRAAM is a supersonic, air launched,
aerial intercept, guided missile employing active radar target tracking, proportional
navigation guidance, and active Radio Frequency (RF) target detection. It employs
active, semi-active, and inertial navigational methods of guidance to provide an
autonomous launch and leave capability against single and multiple targets in all
environments as shown in Fig. (1.13).
Fig. (1.13) Comparison between AIM 7 and AIM 120 guidance technique
24
The AMRAAM weighs 340 pounds and uses an advanced solid-fuel rocket motor to
achieve a speed of Mach 4 and a range in excess of 30 miles. In long-range engagements
AMRAAM heads for the target using inertial guidance and receives updated target
information via data link from the launch aircraft. It switches to a self-guiding terminal
mode when the target is within range of its own monopulse radar set. The AIM-120
also has a "home-on-jam" guidance mode to counter electronic jamming. With its
sophisticated avionics, high closing speed, and excellent end-game maneuverability,
chances of escape from AMRAAM are minimal. Upon intercept an active-radar
proximity fuse detonates the 40-pound high-explosive warhead to destroy the target. At
closer ranges AMRAAM guides itself all the way using its own radar, freeing the launch
aircraft to engage other targets.
AMRAAM is a follow-on to the AIM-7 Sparrow missile series {Fig (1.14)} and
Table (1.4). The missile is faster, smaller and lighter, and has improved capabilities
against low-altitude targets. It incorporates an active-radar with an inertial reference
unit and micro-computer system, which makes the missile less dependent upon the fire-
control system of the aircraft. Once the missile closes on a target, its active radar guides
it to intercept. This enables the pilot to aim and fire several missiles simultaneously at
multiple targets. The pilot may then perform evasive maneuvers while the missiles
guide themselves to their targets.
25
Table (1.4) Specification data of AIM 120 (AMRAAM)
Primary Function Air-to-air tactical missile
Contractor Hughes Aircraft Co. and Raytheon Co.
Power Plant High performance (solid fuel)
Length 143.9 inches (3.66 m)
Launch Weight 335 pounds (150.75 kg)
Diameter 7 inches (177.8 mm)
Wingspan 20.7 inches (525.8 mm)
Range <20 miles (<32 Km)
Speed Supersonic < 4 Mach
Guidance System Active radar terminal/inertial midcourse
Warhead Blast fragmentation
Unit Cost $386,000
Relative costs of
AMRAAM
components
Guidance 68%
Control 9%
Fuse 9%
Warhead 2%
Propulsion 6%
Airframe 6%
Date Deployed September 1991
Aircraft platforms
Navy: F-14D and F/A-18
Air Force: F-15 and F-16
NATO: German F-4, British Tornado and Sea Harrier
26
Fig. (1.14) AIM 7 air to air missile
The AIM-120 grew out of a joint agreement, no longer in effect, among the United
States and several NATO nations to develop air-to-air missiles and to share the
production technology. The AMRAAM program was established as a result of Joint
Service Operational Requirement for an Advanced Air-to-Air Tactical Missile needed
in the post-1985 time frame. The AMRAAM program began with a 1975 study which
recommended that future aerial threats be engaged at 3-40 miles of range.
The AIM-120A is a non-reprogrammable missile (requires a hardware change to
upgrade the missile software). The AIM-120B/C is reprogrammable through the missile
umbilical using Common Field-level Memory Reprogramming Equipment (CFMRE).
The AIM-120C has smaller aero surfaces to enable internal carriage on the Air Force
F-22 aircraft. The USAF All-Up-Round (AUR) container houses an internal cable
which enables up to four missiles to be reprogrammed while in the container. USN
containers are not equipped with the cable and must be opened to reprogram the
27
missile. All three AMRAAM variants are currently approved for use on the F-15C/D/E,
F-16C/D, and F/A-18C/D aircraft.
Four wings, four fins (control surfaces), and the wiring harness cover are mounted
externally, providing additional distinguishing features from other similar missiles,
such as AIM-7 Sparrow. The AIM-120C utilizes ‘’clipped’’ wings and fins in order to
meet the internal carriage requirements of the F-22. AMRAAM consists of the
following major sections: Guidance, Armament, Propulsion, and Control. Other
components include a wiring harness, harness cover, Thermally Initiated Venting
System (TIVS), and wing and fin assemblies.
1.3.3.1 Weapons guidance section
Weapons Guidance Unit. The Weapons Guidance Unit (WGU) consists of the radome,
seeker, servo, transmitter-receiver, electronics unit, Inertial Reference Unit, Target
Detection Device (TDD), the harnesses, and frame structure. All units except the TDD
are contained within a sealed structure composed of the pyroceramicradome, titanium
skin sections, and aluminum aft bulkhead. The TDD, RF and video processor, and the
antennas are attached to the aft skin section as a complete testable assembly.
Electronics group functions include radar signal processing, seeker servo control, and
all of the computations performed in the central data processor. The WGU-16B is used
on AIM-120A missiles, the WGU-41/B is used on AIM-120B missiles, and the WGU-
44/B is used on AIM-120C missiles. Guidance sections on AIM-120B and AIM-120C
missiles contain Electronic Erasable Programmable Read Only Memory which allow
reprogramming of the missile software. Missile software versions are denoted by Tape
and Revision Numbers.
1.3.3.2 Weapons propulsion section
Weapons Propulsion Unit. The Weapons Propulsion Unit (WPU)-6/B consists of an
airframe, integral rocket motor, a blast tube and exit cone, and an Arm/Fire Device
(AFD) with a visible safe-arm indicator. The high performance rocket motor utilizes a
reduced smoke, hydroxyl terminated, polybutadiene propellant in a boost sustain
configuration, an asbestos-free insulated case (an integral part of the airframe), and an
28
integral aft closure, blast tube, and nozzle assembly with a removable exit cone to
facilitate control section installation/removal. Wings are attached in wing sockets at the
forward end of the propulsion section. Provisions are included within this section for
mounting the filter rectifier assembly.
1.3.3.3 Weapons control section
Weapons Control Unit. The Weapons Control Unit (WCU)-11/B consists of four
independently controlled electro-mechanical servo actuators, four lithium-aluminum
batteries connected in parallel, and a steel fuselage section that is bolted to the
propulsion section aft skirt. Each actuator consists of a brushless DC motor ballscrew,
an infinite resolution potentiometer directly coupled to the output shaft, and pulse
width modulated control electronics. The output shaft is engaged directly to a squib
actuated lock so that it does not interfere with the fin (control surface) installation and
removal. The wiring harness cover extends from the aft end of the guidance section to
the forward end of the control section. Its primary purpose is to provide protection for
the wiring harness. The main wiring harness electrically connects the umbilical
connector, guidance section, and control section. The wiring harness cover also houses
the TIVS. The TIVS is designed to vent rocket motor pressure in the event the missile is
exposed to a fuel fire. The TIVS consists of an external thermal cord which, when
ignited, triggers an Out-Of-Line Device (OOLD) that ignites a Linear Shape Charge
that weakens the rocket motor, allowing the rocket motor to vent without exploding.
The OOLD prevents the shaped charge from detonating should the booster in the
OOLD inadvertently detonate due to causes such as high impact. The unit has an
additional safety feature that causes it to “reset” within nine to thirteen units of gravity,
such as the acceleration experienced during missile launching. This feature prevents the
system from functioning during missile free flight so that the associated aerodynamic
pressures do not inadvertently enable the TIVS and thereby degrade the missile
performance. An indicator is on the wiring harness cover showing the condition of the
TIVS, either ‘’ENABLE’’ or ‘’DISABLE’’. Only TIVS equipped missiles are deployed
aboard Aircraft Carriers. The WPU-6/B Propulsion Section (with TIVS) meets the fast
cook-off and sympathetic detonation requirements of the IM program and the policy
delineated in OPNAV Instruction (OPNAVINST) 8010.13B. The other requirements
29
(bullet impact, fragment impact, and slow cook-off) have not been met with the current
configuration. However, the WPU-6/B has been granted the appropriate waivers for
shipboard use.
1.3.3.4 Wing and fin assemblies
Wing and fin assemblies provide flight control of the missile. The four wings are
detachable, stationary flight surfaces with ball fasteners to facilitate quick installation
and removal. The four fins provide the movable control surfaces. The AIM 7 is
interchangeable with AIM-120A and AIM-120B missiles. The AIM-120C utilizes
‘’clipped’’ wings and fins in order to meet the internal carriage requirements on ATD
starting in FY98 that will advance the AMRAAM development to provide a full-up
integration. AMCOM will deliver 2 AMRAAM fire units in 2Q FY99 for Marine Corps
operational suitability and effectiveness testing.
1.3.4 Air to air missiles features and techniques
To understand the benefits of applying the control strategy, many topics in that field
are needed to be discussed. These topics are very important in evaluating any type of
missile. The following subsections present these topics.
1.3.4.1 Target acquisition technique
It is classified into many ways to detect and track a target, these types are classified
according to the range of the missiles and target type, such as TV guided missiles, laser,
radar, and infrared ones (IR). All these different technique follow some guidance laws in
each guidance phases to verify the best trajectory of the missile that it should follow to
keep the target in the required dome origin (shifted) to shorten the duration time until
hitting the target.
1.3.4.2 Aim point when impacting the target
It is the best point in the target at which the missile can cause the greatest damage
after hitting it which is usually the center of the target image on the focal plane array on
the missile dome.
30
1.3.4.3 Proximity fuse distance and locking time
Proximity fuse distance is the maximum distance between the target and the seeker
within which the missile can cause great damage to that target. This is determined by
the type of the missile, the type and quantity of the ammunition, the target geometrical
feature and the control strategy of the missile.
The proximity fuse in AIM-9R functions according to the intensity of the IR rays
coming from the target indicating the target-seeker distance.
Locking time is the time required since the missile enters the proximity fuse envelope
until activating the missile ammunition which could be constant or variable time.
Locking time is very beneficial to secure direct and close impact of the target.
1.3.4.4 Jamming and seduction techniques
All countermeasures that have been established to protect aircraft from missiles such
as Electronic Countermeasures are usually attached to the aircraft at the front part. The
theory of this countermeasures is simply the capability of the instrument to guide the
missile wrongly away from the real position of the target which is done electronically,
but simple jamming and seduction techniques are activated either by dropping flares for
heat seeking missile or chaff for the radar missiles.
1.3.4.5 Reticle seeker (conventional technique)
It is used in most air-to-air missiles as a way to determine the location of the target on
the dome window of the missile in such a way using the wave shift and frequency of the
detector rotating element.
The reticle seeker is the most common optical system design employed in conventional
heat seeking missiles. It was invented by the Germans during the latter phase of World
War 2. It provides a means of using a single detector element to produce an error signal
in rectangular coordinates, for a point target somewhere within the cone which
represents the view of the seeker.
The technique is based on the idea of mechanically chopping the light flux which
impinges on a detector, in such a fashion that the characteristics of the chopped light
pulses vary with position of the light source in the field of view. Because the detector
31
produces an electrical signal directly proportional to the impinging light flux, electronic
hardware can be built to extract a positional error signal in X, Y coordinates, suitable
for driving a missile autopilot with traditional linear controllers.
1.3.4.6 Target seeker distance calculation
Target seeker distance can be easily evaluated from the targets area on the focal plane
array through the missile dome window using the similarity and the target pre-
calculated geometry using the following simple relation.
Target-seeker distance (x) is inversely proportional to the square root of the area of
the target on the focal plane array of the missile detecting screen. Eq. (1.1) is used to
determine target-seeker distance.
Tc
arTarget-seeker distance=
(1.1)
Where (ar) represents the target area on the frontal screen of the missile (m2), Tc
represents a constant depending on the target geometry and the projection angle of the
target on the frontal screen of the missile (target orientations).
This thesis consists of six chapters and two appendices:
- Chapter 1 includes appropriate background about different types of missiles.
- Chapter 2 includes literature review, problem formulation, objectives and thesis
layout.
- Chapter 3 describes modeling procedure of air-to-air missiles.
- Chapter 4 discusses different types of neural network models used for control.
- Chapter 5 explains the fuzzy logic technique and how to apply it for control.
- Chapter 6 presents the results of the computer simulations of the proposed
system.
- Chapter 7 discusses the final conclusions and future work.
- Appendix A presents the Simulink models for the air-to-air missile, neural
network controller, fuzzy logic controller and the whole integrated control
system.
- Appendix B presents an example of a gust effect as a system disturbance.
32
Chapter 2
LITERATURE REVIEW
2.1 Introduction
The development and application of most present day systems and control theory
were spurred on by the need to resolve aerospace problems. This is roughly the problem
of analyzing and designing guidance law and flight control systems (autopilot) for
tactical missiles or aircraft. Therefore, it is beneficial to review the development of
systems and control theory. The guidance and control laws used in current tactical
missiles are mainly based on classical control design techniques. These control laws were
developed in the 1950s and have evolved into fairly standard design procedures (Locke,
1955). Earlier guidance techniques worked well for targets that were large and traveled
at lower speeds. However, these techniques are no longer effective against the new
generation targets that are small, fast, and highly maneuverable. For example, when a
ballistic missile reenters the atmosphere after having traveled a long distance, its radar
cross section is relatively small, its speed is high and the remaining time to ground
impact is relatively short. Intercepting targets with these characteristics is a challenge
for present-day guidance and control designs. In addition, the missile-target dynamics
are highly nonlinear partly because the equations of motion are best described in an
inertial system while the aerodynamic forces and moments are best represented in a
missile and target body axis system.
Moreover, unmodeled dynamics or parametric perturbations usually exist in the plant
modeling. Because of the complexity of the nonlinear guidance design problem, prior
approximations or simplifications have generally been required before the analytical
guidance gains can be derived in the traditional approaches (Lin, 1991) [2]. Therefore,
one does not know exactly what the true missile model is, and the missile behavior may
change in unpredictable ways. Consequently, one cannot ensure
optimality of the resulting design. In the last three decades, optimality-based guidance
designs have been considered to be the most effective way for a guided missile engaging
the target (Bryson and Ho 1969; Lin, 1991; Zarchan,1994) [3], [2] and [4]. However, it is
also known from the optimal control theory that a straightforward solution to the
33
optimal trajectory shaping problem leads to a two-point boundary value problem
(Bryson and Ho,1969) [3], which is too complex for real-time onboard implementation.
Based on the reasons given above, advanced control theory must be applied to a
missile guidance and control system to improve its performance. The use of intelligent
control systems has infiltrated the modern world. Specific features of intelligent control
include decision making, adaptation to uncertain media, self organization, planning and
scheduling operations. Very often, no preferred mathematical model is presumed in the
problem formulation, and information is presented in a descriptive manner. Therefore,
it may be the most effective way to solve the above problems. Intelligent control is a
control technology that replaces the human mind in making decisions, planning control
strategies, and learning new functions whenever the environment does not allow or does
not justify the presence of a human operator. Artificial neural networks and fuzzy logic
are two potential tools for use in applications in intelligent control engineering. Artificial
neural networks offer the advantage of performance improvement through learning by
means of parallel and distributed processing. Many neural control schemes with back-
propagation training algorithms, which have been proposed to solve the problems of
identification and control of complex nonlinear systems, exploit the nonlinear mapping
abilities of neural networks (Miller et al., 1991 [5]; Narendra and Parthasarthy, 1990
[6]). Recently, adaptive neural network algorithms have also been used to solve highly
nonlinear flight control problems. A fuzzy logic-based design that can resolve the
weaknesses of conventional approaches has been cited above. The use of fuzzy logic
control is motivated by the need to deal with highly nonlinear flight control and
performance robustness problems. It is well known that fuzzy logic is much closer to
human decision making than traditional logical systems. Fuzzy control based on fuzzy
logic provides a new design paradigm such that a controller can be designed for
complex, ill defined processes without knowledge of quantitative data regarding the
input-output relations, which are otherwise required by conventional approaches
(Mamdani and Assilian, 1975 [7]; Lee, 1990 [8] ;Driankovet al., 1993 [9]). An overview
of neural and fuzzy control designs for dynamic systems was presented
by Dash et al. (1997) [10]. Very few papers have addressed the issue of neural or fuzzy-
based neural guidance and control design. The published literature in this field will be
introduced in this paper. The following sections are intended to provide the reader with
34
a basic, and unified view of the concepts of intelligent control. Many potentially
applicable topologies are well studied. It is hoped that the material presented here will
serve as a useful source of information by providing for solutions for current problems
and future designs in the field of guidance and control engineering.
2.2 Conventional Guidance and Control Design
Tactical missiles are normally guided from shortly after launch until target
interception. The guidance and control system supplies steering commands to
aerodynamic control surfaces or to correct elements of the thrust vector subsystem so as
to point the missile towards its target and make it possible for the weapon to intercept a
maneuvering target.
2.2.1 Guidance
From the viewpoint of a control configuration, guidance is a special type of
compensation network (in fact, a computational algorithm) that is placed in series with a
flight control system (also called autopilot) to accomplish an intercept. Its purpose is to
determine appropriate pursuer flight path dynamics such that some pursuer objective
can be achieved efficiently. For most effective counterattack strategies, different
guidance laws may need to be used to accomplish the mission for the entire trajectory.
First, midcourse guidance refers to the process of guiding a missile that cannot detect its
target when launched; it is primarily an energy management and inertial
instrumentation problem. When a radar seeker is locked onto a target and is providing
reliable tracking data, such as the missile-target relative range, line-of-sight (LOS)
angle, LOS angle rate and bore sight error angle, the guidance strategy in this phase is
called terminal guidance. Steering of
the missile during this period of flight has the most direct effect on the final miss
distance. The steering law should be capable of achieving successful intercept in the
presence of target maneuvers and external and internal disturbances.
2.2.2 Flight control system
The flight control system executes commands issued based on the guidance law with
fidelity during flight. Its function is three-fold: it provides the required missile lateral
35
acceleration characteristics, it stabilizes or damps the bare airframe, and it reduces the
missile performance sensitivity to disturbance inputs over the required flight envelope.
2.2.3 Conventional design methods
The principles behind controlling guided missiles are well known to control engineers.
Since the basic principles were extensively covered by Locke (1955) [11], a large number
of control technologies have been developed to improve missile performance and to
accommodate environmental disturbances. These techniques are mainly based on
classical control theory. Many different guidance laws have been exploited based on
various design concepts over the years. Currently, the most popular terminal guidance
laws defined by Locke (1955) [11] involve LOS guidance, LOS rate guidance, command-
to-line-of-sight(CLOS) guidance (Ha and Chong, 1992) [12] and other advanced
guidance strategies, such as proportional navigation guidance (PNG) (Locke, 1955) [11],
augmented proportional navigation guidance (APNG) (Zarchan, 1994) [13] and optimal
guidance law based on linear quadratic regulator theory (Bryson and Ho, 1969 [3];
Nazaroff, 1976 [14]), linear quadratic Gaussian theory (Potter, 1964 [15]; Price and
Warren, 1973 [16]) or linear exponential Gaussian theory (Speyer et al., 1982). Classical
guidance laws different from these guidance laws were discussed by Lin (1991) [2], and
the performance of various guidance laws was extensively compared. Among the current
techniques, guidance commands proportional to the LOS angle rate are generally used
by most high speed missiles today to correct the missile course in the guidance loop. This
approach is referred to as PNG and is quite successful against nonmaneuvering targets.
While PNG exhibits optimal performance with a constant velocity target, it is not
effective in the presence of target maneuvers and often leads to unacceptable miss
distances. Classical and modern guidance designs were compared by Nesline and
Zarchan (1981) [17]. The midcourse guidance law is usually a form of PNG with
appropriate trajectory-shaping modifications for minimizing energy loss. Among the
mid-course guidance laws, the most effective and simplest one is the explicit guidance
law (Cherry, 1964) [18].The guidance algorithm has the ability to guide the missile to a
desired point in space while controlling the approach angle and minimizing a certain
appropriate cost function. The guidance gains of the explicit guidance law are usually
selected so as to shape the trajectory for the desired attributes (Wang, 1988 [19]; Wang
36
et al., 1993 [20]). Other midcourse guidance laws are theoretically optimal control-based
approaches (Glasson and Mealy, 1983 [21]; Cheng and Gupta, 1986 [22]). These
research efforts have produced many numerical algorithms for open-loop solutions to
problems using digital computers. However, the main disadvantage of these algorithms
is that they generally converges slowly and are not suitable for real-time applications.
Unfortunately, only rarely is it feasible to determine the feedback law for nonlinear
systems which are of any practical significance. The flight control system used in almost
all operational homing missiles today is a three loop autopilot, composed of a rate loop,
an accelerometer, and a synthetic stability loop. Generally, the controller is in a form of
proportional-integral-derivative (PID) parameters, and the control gains are deter-
mined by using classical control theory, such as the root locus method, Bode method or
Nyquist stability criterion (Price and Warren, 1973 [16]; Neslineet al.,1981 [17]; Nesline
and Nesline, 1984 [23]). Modern control theory has been used extensively to design the
flight control system, such as in the linear quadratic techniques (Stallard, 1991 [24]; Lin
et al., 1993 [25]), generalized singular linear quadratic technique (Lin and Lee, 1985)
[26], design technique (Lin, 1994) [27] and feedback linearization (Lin, 1994) [27]. Over
the past three decades, a large number of guidance and control designs have been
extensively reported in the literature. For a survey of modern air-to-air missile guidance
and control technology, the reader is referred to Cloutieret al. (1989) [28]. Owing to
space limitations, only representative ones were cited above. For further studies on
various design approaches that have not been introduced in this section, the reader is
referred to Lin (1991) [2] and Zarchan (1994) [13]. Current highly maneuverable
fighters pose a challenge to contemporary missiles employing classical guidance
techniques to intercept these targets. Guidance laws currently in use on existing and
fielded missiles may be inadequate in battlefield environments. Performance criteria will
probably require application of newly developed theories, which in turn will necessitate
a large computation capability compared to the classical guidance strategy. However,
advances in microprocessors and digital signal processors allow increased use of
onboard computers to perform more sophisticated computation using guidance and
control algorithms.
37
2.3 Neural Net-based Guidance and Control Design
The application of neural networks has attracted significant attention in several
disciplines, such as signal processing, identification and control. The success of neural
networks is mainly attributed to their unique features:
(1) Parallel structures with distributed storage and processing of massive amounts of
information.
(2) Learning ability made possible by adjusting the network interconnection weights and
biases based on certain learning algorithms.
The first feature enables neural networks to process large amounts of dimensional
information in real-time (e.g. matrix computations), hundreds of times faster than the
numerically serial computation performed by a computer. The implication of the second
feature is that the nonlinear dynamics of a system can be learned and identified directly
by an artificial neural network. The network can also adapt to changes in the
environment and make decisions despite uncertainty in operating conditions.
2.3.1 Supervisory control
The neural controller in the system is utilized as an inverse system model as shown in
Fig. (2.1). The inverse model is simply cascaded with the controlled system such that the
system produces an identity mapping between the desired response (i.e., the net-work
input r) and controlled system output y. This control scheme is very common in robotics
applications and is appropriate for guidance law and autopilot designs. Success with this
model clearly depends on the fidelity of the inverse model used as the controller
(Napolitano and Kincheloe, 1995 [29]; Guezet al., 1998 [30]). In the terminal guidance
scheme proposed by Lin and Chen (1999) [31], a neural network constructs a specialized
on-line control architecture, which offers a means of synthesizing closed-loop guidance
laws for correcting the guidance command provided by the PNG. The neural network
acts as an inverse controller for the missile airframe. The results show that it can not
only perform very well in terms of tracking performance, but also extend the effective
defensive region. Moreover, based on its feature of adaptively, the neural net-based
guidance scheme has been shown to provide excellent performance robustness. It was
also demonstrated by Cottrell et al. (1996) [32] that using a neuro control scheme of this
type for terminal guidance law synthesis can improve the tracking performance of a
38
kinetic kill vehicle. Hsiao (1998) [33] applied the control scheme to treat the disturbance
rejection problem for the missile seeker. In addition, a fuzzy neural network control
architecture, called the fuzzy cerebellar model articulation controller (fuzzy CMAC),
similar to this scheme, was proposed by Geng and MaCullough (1997) [34] for designing
a missile flight control system. The fuzzy CMAC is able to perform arbitrary function
approximation with high speed learning and excellent approximation accuracy. A
control architecture based on the combination of a neural network and a linear
compensator was presented by Stecket al. (1996) [35] to perform flight control
decoupling. In Zhu and Mickle (1997) [36], a neural network was combined with a linear
time-varying controller to design the missile autopilot.
Fig. (2.1) Supervisory control scheme
2.3.2 Hybrid control
Psaltiset al. (1987) [37] discussed the problems associated with this control structure by
introducing the concepts of generalized and specialized learning of a neural control law.
It was thought that off-line learning of a rough approximation to the desired control law
should be performed first, which is called generalized learning. Then, the neural control
will be capable of driving the plant over the operating range and without instability. A
period of on-line specialized learning can then be used to improve the control provided
by the neural network controller. An alternative is shown in Fig. (2.2), it is possible to
utilize a linear, fixed gain controller in parallel with the neural control law. This fixed
39
gain control law is first chosen to stabilize the plant. The plant is then driven over the
operating range with the neural tracking performance of a kinetic kill vehicle. Hsiao
(1998) [33] applied the control scheme to treat the disturbance rejection problem for the
missile seeker. In network tuned online to improve the control. The guidance law (Lin
and Chen, 1999) [31] and flight control system (Stecket al., 1996) possess a similar
control scheme of this type.
Fig. (2.2) Hybrid control scheme
2.3.3 Model reference control
The two control schemes presented above do not consider the tracking performance. In
this scheme, the desired performance of the closed-loop system is specified through a
stable reference model, which is defined by its input-output pair {r(t), yR (t)}. As shown
in Fig. (2.3), the control system attempts to make the plant output y(t) match the
reference model output asymptotically. In this scheme, the error between the plant and
the reference model outputs is used to adjust the weights of the neural controller. In
papers by Light body and Irwin (1994) [38], Psaltiset al. (1987) [37] discussed the
problems the neural net-based direct model reference adaptive control scheme was
applied to design an autopilot for a bank-to-turn missile. A training structure was
suggested in these papers to remove the need for a generalized learning phase.
40
Techniques were discussed for the back propagation of errors through the plant to the
controller. In particular, dynamic plant Jacobian modeling was proposed for use as a
parallel neural forward model to emulate the plant.
Fig. (2.3) Model reference control scheme
2.3.4 Internal model control (IMC)
In this scheme, the role of the system forward and inverse models is emphasized. As
shown in Fig. (2.4), the system forward and inverse models are used directly as elements
within the feedback loop. The network NN1 is first trained off-line to emulate the
controlled plant dynamics directly. During on-line operation, the error between the
model and the measured plant output is used as a feedback signal and passed to the
neuro-controller NN2 . The effect of NN1 is to subtract the effect of the control signal
from the plant output; i.e., the feedback signal is only the influence due to disturbances.
The IMC plays a role as a feedforward controller. However, it can cancel the influence
due to unmeasured disturbances, which cannot be done by a traditional feed-forward
controller. The IMC has been thoroughly examined and shown to yield stability
robustness (Hunt and Sbarbaro-Hofer, 1991) [39]. This approach can be extended
readily to autopilot designs for nonlinear airframes under external disturbances.
41
Fig. (2.4) Internal model control scheme
2.3.5 Adaptive linear or nonlinear control
The connectionist approach can be used not only in nonlinear control, but also as a part
of a controller for linear plants. The tracking error cost is evaluated according to some
performance index. The result is then used as a basis for adjusting the connection
weights of the neural network. It should be noted that the weights are adjusted on-line
using basic back propagation rather than off-line. The control scheme is shown in Fig.
(2.5)
Fig. (2.5) Adaptive control scheme
2.3.6 Predictive control
Within the realm of optimal and predictive control methods, the receding horizon
technique has been introduced as a natural and computationally feasible feedback law.
In this approach, a neural network provides prediction of future plant response over a
specified horizon. The predictions supplied by the network are then passed on to a
42
numerical optimization routine, which attempts to minimize a specified performance
criteria in the calculation of a suitable control signal (Montague et al., 1991 [40]; Saint-
Donat et al., 1994 [41]). Figure (2.6) shows the complete scheme of a predictive
controller.
Fig. (2.6) Predictive control scheme
2.3.9 Example
A hybrid model reference adaptive control scheme is described here, where a neural
network is placed in parallel with a linear fixed-gain independently regulated autopilot
as shown in Fig. (2.7) (McDowell et al., 1997 [42]). The linear autopilot is chosen so as to
stabilize the plant over the operating range and provide approximate control. The
neural controller is used to enhance the performance of the linear autopilot when
tracking is poor by adjusting its weights. A suitable reference model is chosen to define
the desired closed-loop autopilot responses ¨ Zrefand ¨ Yrefacross the flight envelop.
These outputs are then compared with the actual outputs of the lateral autopilot ¨ Z and
¨ Y to produce an error measurement vector [ezey]T , which is then used in conjunction
with an adaptive rule to adjust the weights of the neural network so that the tracking
error will be minimized. A direct effect of this approach is to suppress the influence
resulting from roll rate coupling.
43
Fig. (2.7) Model reference control of coupled lateral dynamics
2.4 Fuzzy Logic Based Guidance and Control Design
The existing applications of fuzzy control range from micro-controller based systems
in home applications to advanced flight control systems. The main advantages of using
fuzzy are as follows:
(1) It is implemented based on human operator’s expertise which does not lend itself to
being easily expressed in conventional proportional-integral- derivative parameters of
differential equations, but rather in situation/action rules.
(2) For an ill-conditioned or complex plant model, fuzzy control offers ways to
implement simple but robust solutions that cover a wide range of system parameters
and, to some extent, can cope with major disturbances.
The sequence of operations in a fuzzy system can be described in three phases called
fuzzification, inference, and defuzzification shown as in Fig. 11. A fuzzification interface
converts input data into suitable linguistic values that may be viewed as labels of fuzzy
sets. An inference mechanism can infer fuzzy control actions employing fuzzy
implication and the rules of the interface in fuzzy logic. A defuzzification interface yields
a nonfuzzy control action from an inferred fuzzy control action. The knowledge base
involves the control policy for the human expertise and necessary information for the
proper functioning of the fuzzification and defuzzification modules. Fuzzy control was
first introduced and applied in the 1970’s in an attempt to design controllers for systems
that were structurally difficult to model. It is now being used in a large number of
44
domains. Fuzzy algorithms can be found in various fields, such as estimation, decision
making and, especially, automatic control.
2.4.1 Fuzzy proportional integral derivative (PID) control
In this case, fuzzy rules and reasoning are utilized on-line to determine the control
action based on the error signal and its first derivative or difference. The conventional
fuzzy two-term control has two different types: one is fuzzy-proportional-derivative
(fuzzy-PD) control, which generates a control output from the error and change rate of
error, and is a position type control; the other is the fuzzy-proportional integral (fuzzy-
PI) control, which generates an incremental control output from the error and change
rate of error, and is a velocity type control (Driankovet al., 1993) [9]. Figure (2.8) shows
a fuzzy-PD controller with normalization and denormalization processes. In Mizumoto
(1992) [43] and Qiao and Mizumoto (1996) [44], a complete fuzzy-PID controller was
realized using a simplified fuzzy reasoning method. Control schemes of these types can
be easily designed and directly applied to guidance and control system design. In fuzzy
logic terminal guidance design, the LOS angle rate and change of LOS angle rate can be
used as input linguistic variables, and the lateral acceleration command can be used as
the output linguistic variable for the fuzzy guidance scheme (Mishra et al., 1994) [45].
The LOS angle rate and target acceleration can also be used as input linguistic variables
to obtain an alternative fuzzy guidance scheme (Mishra et al., 1994 [45]; Lin et al., 1999
[31]). It has been shown that these fuzzy guidance schemes perform better than
traditional proportional navigation or augmented proportional navigation schemes, i.e.,
smaller miss distance and less acceleration command. A terminal guidance law was
proposed by Leng (1996) [46] using inverse kinematics and fuzzy logic with the LOS
angle and LOS angle rate constituting the input linguistic variables. A complete PID
guidance scheme employing heading and flight path angle errors was proposed by
Gonslaves and Caglayan (1995) [47] to form the basis for fuzzy terminal guidance. The
fuzzy-PD control scheme has also been applied to various missile autopilot designs
(Schroeder and Liu, 1994 [48]; Lin et al., 1998 [49]). Input-output stability analysis of a
fuzzy logic-based missile autopilot was presented by Farinewataet al. (1994) [50]. A
fuzzy logic control for general lateral vehicle guidance designs was investigated by
Hessburg (1993) [51]. In the papers by Zhao et al. (1993) [52] and Ling and Edgar (1992)
45
[53], fuzzy rule-based schemes for gain-scheduling of PID controllers were proposed.
These schemes utilize fuzzy rules and reasoning to determine the PID controller’s
parameters. Based on fuzzy rules, human expertise is easily utilized for PID gain
scheduling.
Fig. (2.8) Fuzzy PD controller
2.4.2 Hybrid fuzzy controller
Fuzzy controllers can have inputs generated by a conventional controller. Typically,
the error is first input to a conventional controller. The conventional controller filters
this signal. The filtered error is then input to the fuzzy system. Since the error signal is
purified, one needs fewer fuzzy sets describing the domain of the error signal. Based on
this specific feature, these types of controllers are robust and need a less complicated
rule base.
2.4.3 Fuzzy adaptive controller
The structure is similar to that of fuzzy PID controllers. However, the shapes of the
input/output membership functions are adjustable and can adapt to instantaneous
error. A typical fuzzy adaptive control scheme is shown as in Fig. (2.9) Since the
membership functions are adaptable, the controller is more robust and more insensitive
to plant parameter variations (Dash and Panda, 1996) [54]. In a paper by Lin and Wang
(1998) [49], an adaptive fuzzy autopilot was developed for bank-to-turn missiles. A self-
organizing fuzzy basis function was proposed as a tuning factor for adaptive control. In
Huang et al. (1994) [55], an adaptive fuzzy system was applied to autopilot design of the
X-29 fighter.
46
Fig. (2.9) Typical adaptive fuzzy control scheme
2.4.4 Fuzzy Sliding Mode Controller (SMC)
Although fuzzy control is very successful, specially for control of non-linear systems,
there is a drawback in the designs of such controllers with respect to performance and
stability. The success of fuzzy controlled plants stems from the fact that they are similar
to the SMC, which is an appropriate robust control method for a specific class of non-
linear systems. The fuzzy SMC as shown in Fig. (2.10) can be applied in the presence of
model uncertainties, parameter fluctuations and disturbances, provided that the upper
bounds of their absolute values are known (Driankovet al., 1993 [9]; Ting et al., 1996
[56]; Palm and Driankov, 1997 [57]).
Fig. (2.10) Fuzzy sliding mode control scheme
2.4.5 Fuzzy model following controller
To have the advantages of a fuzzy logic controller with a desired level of performance, a
fuzzy adaptive controller can be used in a model-following control system as shown in
47
Fig. (2.11) In this scheme, the error between the plant output and the reference model
output is used to adjust the membership functions of the fuzzy controller (Kwong and
Passino,1996) [58].
Fig. (2.11) Fuzzy model-following control scheme
2.4.6 Hierarchical fuzzy controller
In a hierarchical fuzzy controller as shown in Fig. (2.12), the structure is divided into
different levels. The hierarchical controller gives an approximate output at the first
level, which is then modified by the second level rule set. This process is repeated in
succeeding hierarchical levels (Kandel and Langholz, 1994 [59]).
Fig. (2.12) Hierarchical fuzzy control system
48
2.4.7 Optimal control
A fuzzy logic system can be utilized to realize an optimal fuzzy guidance law. In this
approach, exact open-loop optimal control data from the computed optimal time
histories of state and control variables are used to generate fuzzy rules for fuzzy logic
guidance. First, data related to the state and control variables of optimal guidance are
generated using several scenarios of interest. The fuzzy logic guidance law possesses a
neurofuzzy structure. Critical parameters of the membership functions of linguistic
variables are presented in the connecting weights of a neural network. The collected
data are then used to train the network’s weights by using the gradient algorithm or
other numerical optimization algorithms. After training has been performed
successfully, missile trajectories and acceleration commands for the optimal solution
and fuzzy logic guidance solution will be close during actual flight using these scenarios.
This approach can effectively resolve the computational difficulty involved in solving the
two-point boundary-value problem. The problem considered by Bouletet al. (1993) [60]
was that of estimating the trajectory of a maneuvering object using fuzzy rules. The
proposed method uses fuzzy logic algorithms to analyze data obtained from different
sources, such as optimal control and kinematical equations, using values sent by sensors.
2.4.8 Example
Figure (2.13) shows a fuzzy logic oriented architecture employed in a fuzzy terminal
guidance system (Gonsalvs and Caglayan, 1995) [47]. The architecture is duplicated for
both the heading and flight path angle channels. Guidance path errors drive in parallel
with a PD and a PI controller. The results produced by the fuzzy PD/PI controllers
(uPDand uPI,respec-tively) are combined via a fuzzy weighting rule-base. The combined
control utotalis then processed via a gain scheduler to account for variations over the
flight envelope. A fuzzy terminal guidance system can readily achieve satisfactory
performance that equals or exceeds that of conventional guidance approaches with
additional advantages, such as intuitive specification of guidance and control logic, the
capability of rapid prototyping via modification of fuzzy rule-bases and robustness to
sensor noise and failure accommodation.
49
Fig. (2.13) A fuzzy terminal guidance system
It should be noted that fuzzy control systems are essentially nonlinear systems.
Therefore, it is difficult to obtain general results from the analysis and design of
guidance and control systems. Furthermore, knowledge of the aerodynamics of missiles
is normally poor. Therefore, the robustness of the resulting
designs must be evaluated to guarantee stability in spite of variations in aerodynamic
coefficients.
2.5 Gain-Scheduling Guidance and Control Design
Gain scheduling is an old control engineering technique which uses process variables
related to dynamics to compensate for the effect caused by working in different
operating regions. It is an effective way to control systems whose dynamics change with
the operating conditions. It is normally used in the control of nonlinear plants in which
the relation-ship between the plant dynamics and operating conditions is known, and for
which a single linear time-invariant model is insufficient (Rugh, 1991 [61]; Hualin and
Rugh, 1997 [62]; Tan et al., 1997 [63]). This specific feature makes it especially suitable
for guidance and control design problems. Gain-scheduling design involves three main
tasks: partitioning of the operating region into several approximately linear regions,
designing a local controller for each linear region, and interpolation of controller
parameters between the linear regions. The main advantage of gain scheduling is that
controller parameters can be adjusted very quickly in response to changes in the plant
dynamics. It is also simpler to implement than automatic tuning or adaptation.
50
2.5.1 Conventional Gain-Scheduling (CGS)
A schematic diagram of a CGS control system is shown in Fig. (2.14). As can be seen,
the controller parameters are changed in an open-loop fashion based on measurements
of the operating conditions of the plant. A gain-scheduled control system can, thus, be
viewed as a feedback control system in which the feedback gains are adjusted using
feedforward compensation (Tan et al., 1997) [63]. Gain scheduled autopilot designs for
tactical missiles have been proposed by Balas and Packard (1992) [64] and Eberhardt
and Wise (1992) [65]. An approach to gain scheduling of linear dynamic controllers has
been considered for a pitch-axis autopilot design problem. In this application, the linear
controllers are designed for distinct operating conditions using H- methods (Nichols et
al., 1993 [66]). A gain scheduling eigenstructure assignment technique has also been
used in autopilot design (Piou and Sobel, 1996) [67].
Fig. (2.14) Conventional gain-scheduling control scheme
2.5.2 Fuzzy Gain-Scheduling (FGS)
The main drawback of CGS is that the parameter change may be rather abrupt across
the boundaries of the region, which may result in unacceptable or even unstable
performance. Another problem is that accurate linear time-invariant models at various
operating points may be difficult, if not impossible, to obtain. As a solution to these
problems, FGS has been proposed, which utilizes a fuzzy reasoning technique to
determine the controller parameters (Takagi and Sugeno, 1985) [68]. For this approach,
human expertise in the linear control design and CGS are represented by means of fuzzy
51
rules, and a fuzzy inference mechanism is used to interpolate the controller parameters
in the transition regions (Ling and Edgar, 1992 [53]; Tan et al., 1997 [63]). Figure (2.15)
shows the fuzzy gain-scheduled control scheme. The Takagi-Sugeno fuzzy models
provide an effective representation of complex nonlinear systems in terms of fuzzy sets
and fuzzy reasoning applied to a set of linear input-output submodels. Based on each
models, fuzzy gain-scheduling controllers can be obtained by means of linear matrix
inequality methods (Zhao et al., 1996) [69]. An H_ gain-scheduling technique using
fuzzy rules was also proposed by Yang et al. (1996) [70] to ensure stability and
performance robustness. The FGS technique has been used in missile guidance design
(Hessburg, 1993 [51]; Lin et al., 1999 [31]) and aircraft flight control design (Gonsalves
and Zacharias, 1994 [71]; Wang and Zhang, 1997 [72]; Adams et al., 1992 [73]). A
robust fuzzy gain scheduler has also been designed for autopilot control of an aircraft
(Tanaka and Aizawa, 1992) [74]. In a paper by Pedrycz and Peters (1997) [75] a
controller of this type was applied for attitude control of a satellite.
Fig. (2.15) Fuzzy gain-scheduling control scheme
2.5.3 Neural Network Gain-Scheduling (NNGS)
NNGS can incorporate the learning ability into gain-scheduling control (Tan et al.,
1997) [63]. The training example consists of operating variables and control gains
obtained at various operating points and their corresponding desired outputs. The main
advantage of NNGS is that it avoids the need to manually design a scheduling program
52
or determine a suitable inferencing system. A representative neural gain scheduling PID
control scheme is shown in Fig. (2.16). In Chai et al. (1996) [76], an on-line approach to
gain-scheduling control of a nonlinear plant was proposed. The method consists of a
partitioning algorithm used to partition the plant’s operating space into several regions,
a mechanism that designs a linear controller for each region, and a radial basis function
neural network for on-line interpolation of the controller parameters of the different
regions. A neural controller design technique for multiple-input multiple-output
nonlinear plants was presented by Maia and Resende (1997) [77]. This technique is
based on linearization of a nonlinear plant model at different operating points. Then a
global nonlinear controller is obtained by interpolating or scheduling the gains of the
local operating designs. The neural gain-scheduling technique has been used in various
fields, such as hydroelectric generation (Liang and Hsu, 1994) [78], process control
(Cavalieri and Mirabella, 1996) [79], robotic manipulators (Wang et al., 1994) [80] and
aircraft flight control systems (Chu et al., 1996; Jonckheereet al., 1997).
Fig. (2.16) Neural network gain-scheduling PID control scheme
2.5.4 Neural-Fuzzy Gain-Scheduling (NFGS)
NFGS is implemented using a neural-fuzzy net-work that seeks to integrate the
representational power of a fuzzy inferencing system and the learning and function
approximation abilities of a neural net-work to produce a gain-scheduling system (Tan
et al., 1997 [63]; Tomescu and VanLandingham, 1997 [81]). As in NNGS, interpolation
53
of the controller parameters is adaptively learned by a neural-fuzzy network. Unlike to
FGS, the fuzzy rules and membership functions can be refined using learning and
training data. In contrast to NNGS, NFGS provides a more meaningful interpretation of
the network; in addition, expert knowledge can be incorporated into the fuzzy rules and
membership functions. The control scheme is shown in Fig. (2.17).
Fig. (2.17) Neural-fuzzy gain-scheduling control scheme
2.6 Guidance Laws
Song and Zhang [82] applied a passive homing missile’s variable structure
proportional navigation with terminal angular constraint. The objective was to design a
variable structure proportional navigation law for the passive homing missile so that the
missile can hit the target with a minimum miss-distance and desired impact attitude
angle within the required overload when it flies at a lower trajectory.
Choi and Chaw [83] applied a nonlinear adaptive guidance considering target
uncertainties and control loop dynamics. The fourth order state equation for integrated
guidance and control loop is formulated considering target uncertainties in control loop
dynamics. The state equation is further changed into the normal form by nonlinear
54
coordinate transformation. A nonlinear adaptive guidance law was proposed to
compensate the uncertainties in both target acceleration and control loop dynamics. The
proposed law adopted the sliding mode control approach with adaptation for unknown
bound of uncertainties.
Innocenti [84] applied a nonlinear guidance technique for agile missiles. This technique
presents new approaches to the guidance of agile missiles. They are based on nonlinear
discontinuous control techniques applied to the generation of guidance laws capable of
taking advantage of the vehicle’s post-stall capabilities. Agility and maneuverability
requirements imply a higher bandwidth and robustness for the guidance loop, which are
addressed by a variable structure controller format. Formal stability considerations are
presented, and the guidance structures are validated using nonlinear simulation.
Anderson and Burkhalter [85] applied a design of a guided missile interceptor using a
genetic algorithm. An apportioned genetic algorithm was used to manipulate a solid
rocket design code, an aerodynamic design code and a three-loop autopilot to produce
guided missile interceptor designs capable of accurately engaging a high-speed /high-
altitude target. Definition of the optimization problem required 29 design variables and
four primary goals were established to assess the performance of the interceptor
designs. Design goals included the following: minimize miss distance, minimize intercept
time, minimize takeoff weight, and minimize loading.
Zhou and Chundi [86] applied motion tracking sliding-mode guidance of a homing
missile with bearings-only measurements. Motion tracking sliding-mode control theory
was utilized to design a homing-missile guidance law, which induces missile target line of
sight angular-velocity-rate to track an arbitrary given signal. An adaptive two-step filter
was used to estimate missile-target relative distance, relative velocity, and target
acceleration involved in the guidance law during the whole guidance period.
Recently, neurofuzzy modeling techniques have been successfully applied to modeling
complex systems, where traditional approaches hardly can reach satisfactory results due
to the lack of sufficient domain knowledge. Much research has been done on
55
applications of neural networks (NNs) for identification and control of dynamic systems
[87]. According to their structures, the NNs can be mainly classified as feedforward
neural networks and recurrent neural networks [87]. It is well known that a
feedforward neural network is capable of approximating any continuous functions
closely. However, the feedforward neural network is a static mapping. Without the aid
of tapped delays the feedforward neural network is unable to represent a dynamic
mapping. Although much research has used the feedfoward neural network with tapped
delays to deal with dynamical problem. The feedforward neural network requires a
large number of neurons to represent a dynamic response in the time domain [88].
Moreover, the weight updates of the feedforward neural network do not utilize the
internal information of the neural network and the function approximation is sensitive
to the training data.
2.7 Advantages over conventional designs
(1) Fuzzy guidance and control provides a new design paradigm such that a control
mechanism based on expertise can be designed for complex, ill-defined flight dynamics
without knowledge of quantitative data regarding the input-output relations, which are
required by conventional approaches. A fuzzy logic control scheme can produce a
higher degree of automation and offers ways to implement simple but robust solutions
that cover a wide range of aerodynamic parameters and can cope with major external
disturbances.
(2) Artificial Neural networks constitute a promising new generation of information
processing systems that demonstrate the ability to learn, recall, and generalize from
training patterns or data. This specific feature offers the advantage of performance
improvement for ill-defined flight dynamics through learning by means of parallel and
distributed processing. Rapid adaptation to environment change makes them
appropriate for guidance and control systems because they can cope with aerodynamic
changes during flight.
2.8 Problem Formulation
Missile guidance control system is a complicated work due to the missile nonlinear
model that contains a cross coupling effect between its lateral and longitudinal motions.
56
The missile also is subjected to a continuous disturbance (gust effect) while its speed
reaches more than 5 Mach in some types. In spite of all these circumstances the miss
distance between the missile and the target should not be more than 5 meters to secure a
great damage to the target. In order to achieve such requirements a neuro-fuzzy control
system will be used in the present work.
2.9 Main Objective of the Thesis
This technique will be applied to AIM 9R as a passive homing missile using the
simplest guidance law that is CLOS (command to line of sight) to eliminate any errors
developed from different guidance laws to easily evaluate the applied control strategy.
The miss distance and minimum locking time should be calculated according to this
strategy that should satisfy the proximity fuse distance and acceptable minimum locking
time.
2.10 Thesis Layout
In this thesis, the concept of incorporating fuzzy logic into a neural network has
been discussed. In contrast to the pure neural network or fuzzy system, the fuzzy neural
network (FNN) possesses both their advantages. It combines the capability of fuzzy
reasoning in handling uncertain information and the capability of artificial neural
networks in learning from processes Fig. (2.18). In missiles controlling branch it is
needed to minimize the output performance errors until fulfilling the required
specifications such as the minimum locking time and the maximum fuse distance.
Nowadays there are different types of missiles using different types of control strategy
some of them uses reference frame unit as they are controlled by a remote base or an
aircraft. Consequently controlling such missiles by reference angles ,, with respect
to the fixed frame of reference. Another type of control strategies doesn’t use reference
frame unit such as all heatseeking and active radar missiles (passive and active homing
types). In this thesis, both types of control strategies are discussed. It presented a
technique for the preliminary design of a control system using neurofuzzy systems for a
diverse data range of a highly nonlinear MIMO 5_DOF model (AIM 9R air to air
missile) which is composed of 3_body axes velocities and rotating in both pitch and yaw
direction only as the rolling motion is prevented due to gyroscopic stabilizers (anti roll
57
system) attached at the rear fins. The model has cross coupling effects between the
longitudinal and lateral motions due to the coupled equations of motion. MIMO model
was divided into two separated SISO models taking into consideration the nonlinear
cross coupling effects, then generating the neural network controller for each separated
SISO model, blending the output performance using a 25 rule a Mamdani fuzzy logic
controller for each SISO model to be able to deal with the system uncertainty due to
cross coupling effect and affecting gust in the whole system when instantaneously
applying both inputs to the model. The main goal is to fulfill certain specifications within
its limited ranges such as the proximity fuse distance and the min target locking time
during the whole mission of the missile.
Fig. (2.18) Integrated control model for the guidance of AIM 9R missile
58
Chapter (3)
MATHEMATICAL MODELING OF AN AIR TO AIR MISSILE
3.1 Introduction
The first step in this research work is the development of a mathematical model of
AIM 9R air-to-air missile. There are a number of simple models commonly used for
dynamic analysis depending on the type of analysis to be performed. In this chapter it is
started by developing a comprehensive model sophisticated enough to be used in general
dynamic analysis. The model provides the ability of implementing various modern
control systems such as neural network and fuzzy logic control. This chapter is
organized as following:
3.2 The Rigid –Body Equations
The rigid-body equations of motion are derived from the first simple principles.
Let m in Fig. (3.1) be an element of mass of the missile, having velocity v relative to
fixed axes, and let F be the resultant force that acts on it.
Fig. (3.1) Missile body axes
59
Newton’s second law then gives the equation of motion of m as:
F= mvd
dt (3.1)
Summing Eq. (2.1) for all elements m in the missile to give:
F= mvd
dt=
dt
d v m (3.2)
The quantity F is a summation of all the forces that act upon all the elements.
The internal forces are those exerted by one element upon another occurring in equal
and opposite pairs, by Newton’s third law, and hence contribute nothing to the
summation. Thus F=F is the resultant external force acting on the missile.
The absolute velocity of m is
v = vc+ rd
dt (3.3)
wherevcis the velocity of the mass center. Thus,
v m= (vc + rd
dt) m (3.4)
v m=mvc +dt
d r m (3.5)
Since C is the mass center, then r m=0,so that
v m=mvc (2.6)
Where m is the total mass of the missile.
Thus Eq. (3.6) becomes:
F=mvcd
dt (3.7)
This equation relates the external force on the missile to the motion of its mass center.
The relationship between the external moment and the rotation of the missile is needed.
60
It is obtained from a consideration of the moment of momentum. The moment of
momentum of m is by definition h=rv m.
dt
d( h)=
dt
d(rv) m (3.8)
dt
d( h) =
rd
dtv m + r
vd
dt m (3.9)
Now from Eq. (3.3):
rd
dt=v-vc (3.10)
Also,
rvd
dt m=r F= G (2.11)
Eq. (3.11) follows from Eq. (3.1)and introduces G, the moment of F about C. Eq.
(3.9) then becomes
G=dt
d( h)-(v- vc) v m (3.12)
Since vv=0, Eq. (3.12) becomes
G=dt
d( h)+ vcv m (3.13)
Eq. (3.13) is now summed for all elements, as was Eq. (3.1)and becomes
G=dt
d ( h) + vc v m (3.14)
By an argument similar to that for F, G is shown to be the resultant
external moment about C, denoted G. h is called the moment of momentum, or
angular momentum of the missile and is denoted by h. Using Eq. (3.3) and noting that
vcvc=0 then Eq. (3.14) is reduced to as follows:
61
G= hddt (3.15)
3.3 Evaluation of the Angular Momentum (h)
From the definition of (h)
h= ( h) = (rv) m (3.16)
Let the angular velocity of the missile be
=Pi + Qj + Rk (3.17)
Where P, Q, R are the scalar components of respectively, and i, j, k are unit vectors
in the directions of x, y, z.
Now the velocity of a point in a rotating rigid body is given by
v= vc + r (3.18)
thus,
h= r ( vc + r) m (3.19)
h= rvc m+ r ( r) m (3.20)
Since vc is constant with respect to the summation, and since r m=0, then the first
sum in Eq. (2.20) is ( r m)vc=0.
The second sum is conveniently expanded by the rule for a vector triple product.
r ( r) = r2 –r( r) (3.21)
62
Since the vector r has the components (x, y, z) then Eq. (3.20) finally becomes
22 2( ) m (Px Qy Rz) myx z h r (3.22)
The scalar components of Eq. (3.22) are:
hx=P ( y2+z2) m-Q xy m - R xz m (3.23)
hy=-P xy m + Q (x2+z2) m - R yz m (3.24)
hz=-P xz m-Q yz m+R (x2+y2) m (3.25)
The summations that occur in these equations are the moments and products of inertia
of the missile (A, B, C, D, E , J), then the equations can be written in matrix form as:
x
y
z
h A J E P
h J B D Q
E D C Rh
(3.26)
where
A=2 2(y z ) m
B=2 2(x z ) m
C= 2 2(x y ) m
D= yz m
E= xz m
J= xy m
3.4 Euler’s Equation of Motion
It is noted from Eq. (3.15) that it is required the derivatives with respect to time of Eq.
(3.26). If axes of reference are non-rotating, then it is clear that, as the missile rotates,
63
the moments and products of inertia will vary, and derivatives of A, B,C, D, E, J will
appear in the equations, as well as derivatives of P, Q, R. This is the most undesirable
situation. It can be avoided, however, if the frame of reference Cxyz is fixed to the
missile and move with it, where A, B, C, D, E, J will be constants.
The derivative of a vector A, referred to a frame of reference rotating with angular
velocity is
Ad
dt=
A
t + A (3.27)
where
A Ax
tt i +
Ay
tj +
Az
tk (3.28)
The vector equations of motion Eq. (3.7), (3.15) become when referred to the frame of
reference C x y z, fixed to the missile (body axes of the missile).
Fx=m(U. + QO- RV) (3.29)
Fy=m(V. + RU - PO) (3.30)
Fz=m(O.+ PV - QU) (3.31)
dxdt
dydt
ddt z
L R Q h
M R P h
N Q P h
(3.32)
64
3.5 Orientation and Position of the Missile
Since the frame of reference adopted for the equations of motion is fixed to the
missile, and moves with it the position and orientation of the missile can’t be described
relative to it.
For this purpose an earth fixed frame of reference Ox’y’z’ is introduced as shown in
Fig. (3.2). Let oz’ be taken vertically downward, and Ox’ horizontal in the vertical plane
containing the initial velocity vector of the mass center. The origin O is assumed to
coincide with C at t=0.
The orientation of the missile is then given by a series of three consecutive rotations,
whose order is important. The missile is imagined first to be oriented to that its axes are
parallel to x’y’z’. It is then in the position Cx1y1z1. The following rotations are then
applied.
1. A rotation about cz1, carrying the axes to Cx2y2z2 (bringing Cx to its final
azimuth).
2. A rotation about cy2, carrying the axes to Cx3y3z3 (bringing Cx to its final
elevation).
3. A rotation about cx3, carrying the axes to their final position Cxyz (giving the final
angle of bank to the wings).
65
Fig. (3.2) Missile orientation around fixed frame of reference
3.6 The Flight Path Determination
It is recommended to have the coordinates of the flight path relative to the fixed
frame x’y’z’. To do this the velocity components in the direction x1,y1,z1 are required.
Let these be denoted U1,V1,O1, and similarly let subscripts 2, 3 denote components in the
directions of (x2,y2,z2) and (x3,y3,z3) respectively. Therefore:
dt
dx'=U1 (3.33)
dt
dy'=V1 (3.34)
dt
dz'=O1 (3.35)
here
U1=U2cos -V2sin (3.36)
V1=U2sin +V2cos (3.37)
O1=O2 (3.38)
66
U2=U3cos +O3sin (3.39)
V2=V3 (3.40)
O2=-U3sin +O3cos (3.41)
and
U3=U (3.42)
V3=Vcos -Osin (3.43)
O3=Vsin +Ocos (3.44)
Combining Eqs. (3.36) through (3.44) gives:
dt
dx'=Ucos cos +V(sin sin cos -cos sin )+O(cos sin cos + sin sin )
(3.45)
dt
dy'=Ucos sin +V(sin sin sin +cos cos )+O(cos sin sin -sin cos )
(3.46)
dt
dz'=-Usin +Vsin cos +Ocos cos (3.47)
The coordinates (x’,y’,z’) are then given by the integral of Eqs. (3.45), (3.46) and (3.70).
Integrating these equations in other than special simple cases is obviously a formidable
problem, but can be easily solved using MATLAB (simulink editor).
67
3.7 The Orientation of the Missile
The orientation of the missile would be presented in terms of angular velocity
components (P, Q, R). Let (i, j, k) be unit vectors, with subscripts 1,2,3 denoting
directions (x1,y1,z1),etc.
Let the missile experience, in time t, an infinitesimal rotation from the position
defined by , , to that corresponding to ( + ), ( + ), ( + ). The
vector. representing this rotation is given approximately by:
i1+ j3+ k2 (3.48)
Then the angular velocity is exactly:
=
t 0lim
t
= ϕ.
i1 + ϴ . j3 + ψ.
K2 (3.49)
If the components of given in Eqn. (3.48) be projected onto Cxyz, and remembering
that =Pi + Q j + R k, obtaining the following relations:
P= ϕ.- ψ.
sin (3.50)
Q= ϴ .cos + ψ.
cos sin (3.51)
R= ψ.cos cos - ϴ
. sin (3.52)
68
In order to find the angles ( , , ) in terms of (P, Q, R), the above equations must be
solved. If they are regarded as algebraic equations in (ϕ. ,ϴ
., ψ.
) then:
ϕ. = P + Qsin tan +Rcos tan ϴ (3.53)
ϴ .
= Qcos -Rsin (3.54)
ψ. = (Qsin +Rcos )sec ϴ (3.55)
Using Eqn’s. (2.56), (2.57), (2.58), ( , , ) are given by:
= t
0
(P + Qsin tan + Rcos tan )dt (3.56)
= t
0
(Qcos -Rsin )dt (3.57)
= t
0
(Qsin +Rcos )sec dt (3.58)
However it can be solved analytically only in simplified forms, or using MATLAB.
3.8 The Applied Forces and Moments
The equations of motion derived before contain the following forces and moments:
1-The external forces acting on the missile (Fx, Fy, Fz).
2-The external couples acting on the missile (L, M ,N).
3-The control surface hinge moments (He, Hr, Ha).
4-The generalized control forces (Fe,Fr,Fa).
The forces and moments in 1, 2, 3 above are the external actions on the missile .They
are of two kinds gravitational and aerodynamics ,and are in general determined by
69
orientation, configuration, and motion of the missile : i.e, by the variables ( , , ),
( e, r, a), (U,V,O), (P,Q,R).The generalized control forces, on the other hand are
internal forces.
The external forces are as following:
From Fig. (3.3) the components of the weights in the directions of the axes are found to
be.
Xg=-mg sin (3.59)
Yg=mg cos sin (3.60)
Zg=mg cos cos (3.61)
Fig. (3.3) Weight forces acting on the missile during flight
With the aerodynamic forces (including propulsive forces) denoted by (X, Y, Z), the
resultant external forces are:-
Fx=X- mg sin (3.62)
Fy=Y+ mg cos sin (3.63)
Fz=Z+ mg cos cos (3.64)
70
Since gravity does not contribute to the moments about the mass center, then (L, M, N)
are entirely aerodynamic. The effect of mass unbalance of the control surface in
producing a gravitational hinge moment will be assumed small and neglected, so that the
hinge moments (He, Hr, Ha) are also purely aerodynamic.
Finally, the equations of motion of the missile are as follows:
1-Force equations
Using Eqn’s (3.29), (3.30), (3.31), (3.62), (3.63) and (3.64), force equations are given by:
X- mg sin = m(U. + QO – RV) (3.65)
Y+ mg cos sin = m(V. + RU – PO) (3.66)
Z+ mg cos cos = m(W. + PV- QU) (3.67)
2-Moment equations
Using Eqn’s (3.23-3.25) and (3.26), moment equations are given by:
L=AP.- ER
.+ QR (C-B) - E PQ (3.68)
M=BQ.
+ RP (A-C) + E( 2P - 2
R ) (3.69)
N=-EP.
+ CR. + PQ (B-A)+E QR (3.70)
3-Missile orientation equations
ϕ. = P + Qsin tan +Rcos tan ϴ (3.71)
ϴ .
= Qcos -Rsin (3.72)
ψ. = (Qsin +Rcos )sec ϴ (3.73)
71
4-Flight path equations
dt
dx'=Ucos cos +V(sin sin cos -cos sin )+W(cos sin cos +sin sin )
(3.74)
dt
dy'=Ucos sin +V(sin sin sin +cos cos )+W(cos sin sin -sin cos )
(3.75)
dt
dz'= - U sin + Vsin cos + Wcos cos
(3.76)
For AIM canard configuration, further simplification may be made since the xy (as well
as xz) is also a plane of symmetry (i.e., E=0) and moment of inertia about the y axis is
generally equal to that about the z axis (i.e., B=C), and so due to antiroll gyroscopic
stabilizers rolling about x axis will be mostly prevented so angular velocity about x axis
P will almost equal to zero, hence the resultant simplified equations are:
M=BQ. (3.77)
Where Q. is angular acceleration about y axes and M is given by:
M= M + M
+ M '' + M '
' + M '' (3.78)
N=CR. (3.79)
Where R. is angular acceleration about z axes and N is given by:
N= N + N
+'N
' +'N
' +'N
' (3.80)
72
The mathematical model derived herein is built in Matlab / Simulink environment to
facilitate various computer simulations. Figure (3.4) shows the computation algorithm of
the missile model while Fig. (3.5) shows the missile detailed model in simulink editor.
(3.4). Appendix A includes some detailed models describing the building process.
Fig. (3.4) Missile model algorithm
73
Fig. (3.5) AIM 9 Mathematical Modeling
74
3.9 AIM 9R Model Characteristics
3.9.1 Model verification
The model was established and examined using (Simulink-MATLAB) simulations.
The output performance was then verified with another simplified model (3 DOF
Model) expressing only the longitudinal motion [89].
Figure (A1) shows 3 DOF model in Simulink model appendix A. This model will be used
in AIM 9R missile model verification. Figures (3.6) and (3.7) show the verification
process using Step and frequency response respectively.
Fig. (3.6) Pitch angle step response for 5 DOF and 3 DOF models
Fig.
(3.7) Pitch angle dynamic response for 5 DOF and 3 DOF models, linearization was
performed at frequency =.3 (rad/s) as an example
75
Figure (3.6) and (3.7) indicate that AIM 9R missile model that was represented by 5
DOF model is verifying 3 DOF missile model with some errors due to the cross coupling
effect in 5 DOF model rather than 3 DOF model. Also the 5 DOF is highly nonlinear
more than the simplified 3 DOF model.
3.9.2 Cross effect representation
As mentioned before the model has a considerable cross coupling effect between its
longitudinal and lateral axes. This effect can be expressed through Fig. (3.8), (3.9) and
(3.10) using some quantitative measure such as mean value and standard deviation of
the cross coupling effect.
Fig. (3.8) Cross coupling effect (yaw on pitch)
Fig. (3.9) Mean value of the cross coupling effect versus input frequencies at a
constant amplitude of .1(rad) rudder deflection as an example
76
Fig. (3.10) Standard deviation of cross coupling effect (yaw on pitch)
Figures (3.9) and (3.10) illustrate the cross coupling effect on longitudinal motion that
reaches the maximum value at rudder deflection input frequency =0.1 rad/s then decays
sharply until reaching rudder deflection input frequency =0.5 rad/s then continuing to
decay gradually according to the frequency increase.
3.9.3 Wind gust effect (model disturbance)
To determine the gust effect it is needed to examine the whole system without gust
effect and then with gust effect as shown in Fig. (3.11). This is done by applying a
Dryden gust through the pitch angular velocity Q and angle of attack .
Fig. (3.11) Gust effect (pitch motion)
77
3.10 Summary
(a) Derivation of AIM 9R missile model including the aerodynamic and
gravitational forces and moments was implemented.
(b) Derivation of missile orientation and flight path equations was implemented.
(c) AIM 9 R model was verified using 3 DOF model for step and sine wave inputs.
(d) Cross coupling effect between longitudinal and lateral motions was examined
and illustrated using mean value and standard deviation methods. Resulting
that the maximum cross coupling effect on longitudinal motion is at rudder
deflection frequency =0.1 rad/s.
(e) Wind gust effect was illustrated and examined according to Dryden gust model.
78
Chapter (4)
NEURAL NETWORK FOR MISSILE GUIDANCE
4.1 Introduction
As it was discussed in the previous chapter the AIM 9R missile model was built and
verified with another simplified one. Now the main objective is to use such model with
the most suitable controllers to drive this missile in the right trajectory that allow the
missile to hit the target and cause a great damage to its structure.
This chapter discusses the capabilities of the neural network, different types of neural
network controllers and how to apply these types of controllers to the AIM 9R missile
model in spite of the existing cross coupling effect between the longitudinal and lateral
motions and wind gust (system disturbance). The development of the neural network
passes through many phases during the last century and will be illustrated briefly as
follows.
Scientists have just begun to understand how biological neural networks operate. It
is generally understood that all biological neural functions, including memory, are
stored in the neurons and in the connections between them. Learning is viewed as the
establishment of new connections between neurons or the modification of existing
connections.
The neurons considered here are not biological. They are extremely simple
abstractions of biological neurons, realized as elements in program or perhaps as
circuits made of silicon. Networks of these artificial neurons do not have a fraction of the
power of the human brain, but they can be trained to perform useful functions.
Concepts and their accompanying mathematics are not sufficient for a technology to
mature unless there is some way to implement the system. For instant, the mathematics
necessary for the reconstruction of images from computer-aided tomography (CAT)
79
scans was known many years before the availability of the high-speed computers and
efficient algorithms finally made it practical to implement a useful CAT system.
The history of neural networks has progressed through both conceptual innovations
and implementation developments. These advancements, however, seem to have
occurred in fits and starts rather than by steady evaluation.
Some of the background work for the field of neural networks occurred in the late
19th and early 20th centuries. This consisted primarily of interdisciplinary work in
physics, psychology and neurophysiology by such scientists as Hermann von Helmholtz,
Ernst Mach and Ivan Pavlov. This early work emphasized general theories of learning,
vision, conditioning, etc., and did not include specific mathematical models of neuron
operation.
The modern view of neural networks began in the 1940s with the work of Warren
McCulloch and Walter Pitts, who showed that the networks of artificial neurons could,
in principle, compute any arithmetic or logical function. Their work is often
acknowledged as the origin of the neural network field [90].
McCulloch and Pitts were followed by Donald Hebb, who proposed that classical
conditioning (as discovered by Pavlov) is present because of the properties of individual
neurons. Donald Hebb proposed a mechanism for learning in biological neurons [90].
The first practical application of artificial neural networks came in the late 1950s,
with the invention of the perceptron network and associated learning rule by Frank
Rosenblatt. Rosenblatt and his colleagues built a perceptron network and demonstrated
its ability to perform pattern recognition. This early success generated a great deal of
interest in neural network research. Unfortunately, it was later shown that the basic
perceptron network only could solve a limited class of problems [91].
At about the same time, Bernard Widrow and Ted Hoff introduced a new learning
algorithm and used it to train adaptive linear neural networks, which were similar in
80
structure and capability to Rosenblatt’s perceptron. The Widrow –Hoff learning rule is
still in use today [92].
Unfortunately, both Rosenblatt’s and Widrow’s networks suffered from the same
inherent limitations. Rosenblatt and Widrow were aware of these limitations and
proposed new networks that would overcome them. However, they were not able to
successfully modify their learning algorithms to train the more complex networks [93].
Many people, influenced by Minsky and Papert, believed that further research was a
dead end. This, combined with the fact that there were no powerful digital computers on
which to experiment, caused many researchers to leave the field. For a decade neural
network research was suspended [94].
Some important work, however, continued during the 1970s. In 1972 Teuvo
Kohonenan James Anderson independently and separately developed new neural
networks that could act as memories. Stephen Grossberg was also very active during
this period in the investigation of self-organizing networks [95].
Interest in neural networks had faltered during 1960s because of the lack of new
ideas and powerful computers with which to experiment. During the 1980s both of these
impediments were overcome, and research in neural networks increased dramatically.
New personal computers and workstations, which rapidly grew in capability, become
available. In addition, important new concepts were introduced as follows.
Two new concepts were most responsible for the rebirth of neural networks. The first
was the use of statistical mechanics to explain the operation of a certain class of
recurrent network, which could be used as an associative memory. This was described in
a seminal paper by physicist John Hopfield [96].
The second key development of the 1980s was the backpropagation algorithm for
training multilayer perceptron networks, which was independently discovered by
81
several different researchers. The most influential publication of the backpropagation
algorithm was by David Rumelhart and James McClelland [97], [98].
Well, so much for the history of neural networks to this date. The real question is,
“what will happen in the next twenty years?” Will neural networks take a permanent
place as a mathematical /engineering tool, or will they fade away as have so many
promising technologies? At present, the answer seems to be that neural network will not
only have their day but will have a permanent place, not as a solution to every problem,
but as a tool to be used in appropriate situations.
4.2 Theory and Applications
4.2.1 Single-input neuron
A single-input neuron is shown in Fig. (4.1). The scalar input P is multiplied by the
scalar weight W to form WP, one of the terms that are sent to the summer. The other
input, 1, is multiplied by a bias b and then passed to the summer. The summer output n,
often referred to as the net input, goes into a transfer function f, which produces the
scalar neuron output a.
Fig. (4.1) Single Input neuron
The neuron output is calculated as:
a= (wp+b) (4.1)
82
4.2.2 Multiple-input neuron
Typically, a neuron has more than one input. A neuron with R inputs is shown in
Fig. (4.2). The individual inputs P1, P2, …PRare each weighted by correspondingelements
W1,1,W1,2,…..W1,Rof the weighting matrix W.
Fig. (4.2) Multiple Input Neuron
The neuron has a bias b, which is summed with the weighted input to form the net input
n:
n= W1,1 P1 +W1,2 P2+ …..W1,R PR+b (4.2)
This expression can be written in matrix form :
n= WP+b, (4.3)
where the matrix W for the single neuron case has only one row.
Now the neuron output can be written as:
a= (WP+b) (4.4)
83
4.2.3 A layer of neurons
A single-layer networks of S neorons is shown in Fig. (4.3). Note that each of the R
inputs is connected to each of the neurons and that the weight matrix now has S rows.
Fig. (4.3) Layer of S Neurons
The layer includes the weight matrix, the summers, the bias vector b, the transfer
function boxes and the output vector a.
Each element of the input vector p is connected to each neuron through the weight
matrix W. Each neuron has a bias b, a summer, a transfer function f and an output a.
The outputs from the output vector aare given by.
a= (WP+b) (4.5)
4.2.4 Multiple layers of neurons
Now consider a network with several layers. Each layer has its own weight matrix W,
its own bias vector b, a net input n and an output vector a. It is needed to introduce
84
some additional notation to distinguish between these layers. Superscripts will be used to
identify the layers. Specifically, number of the layer will be appended as a superscript to
the names for each of these variables.
Now some different types of neural network kits are prepared to be applied to
different control systems. These neural networks depend mainly on recurrent neural
networks such as inverse models, feedforward, feedback linearization…etc [17].
4.3 Neural network controllers
4.3.1 Control with inverse models
Conceptually, the most fundamental neural network based controllers are probably
those using the ‘inverse’ of the process as the controller as shown in Fig. (4.4). The
simplest concept is called direct inverse control.
The principle of which is that if the process can be described by:
y(t+1)=g(y(t),….,y(t-n+1),u(t),…,u(t-m)) (4.6)
where
y(t-1)……y(t-n+1) are the past values of output signal
u(t-1)……u(t-m) are the past values of control signal
a network is trained as the inverse of the process:
u(t)=g-1(y(t+1),y(t),….,y(t-n+1),u(t-1),u(t-m)) (4.7)
The inverse model is subsequently applied as the controller for the process by inserting
the desired output, the reference r(t+1), instead of the output y(t+1). Figures (4.5)
through Fig. (4.7) represent the dynamic response of this model without any disturbance
effect.
85
Fig. (4.4) Direct inverse control
Fig. (4.5) Control by Direct Inverse Control at steady state flight
86
Fig. (4.6) Control by Direct Inverse Control at climb angle =90 deg
Fig. (3.7) Control by Direct Inverse Control at initial velocity =700 m/s
It is clear from Fig. (4.5), (4.6) and (4.7) that the control signal is varying according to
the initial conditions of the launched missile. So in case of training off line all these
initial conditions should be taken into consideration otherwise training should be
performed on line. Training on line requires small sample time and very fast training
algorithm with a very fast microprocessor. In mean while if the contribution of this
87
neural network controller is for few seconds in the missile mission it will be prefered if
the training was done off line as it will be discussed later.
4.3.2 Control by feedforward model
Using inverse models for feedback control leads to a dead-beat type of control, which is
unsuitable in many cases. If a PID controller has already been tuned for stabilizing the
process, an inverse model can be used for providing a feedforward signal directly from
the reference model as shown in Fig. (4.8). The dynamic response of this model is shown
in Fig. (4.9)
Fig. (4.8) Control by Feedforward
Fig. (4.9) Control by Feedforward model at steady state flight
88
4.3.3 Control by feedback linearization
Feedback linearization is a common method for controlling certain classes of
nonlinear processes. This method provides a simple example of a discrete input-output
linearizing controller that is based on a neural network model of the process. The
discrete feedback linearization structure is given as follows.
y’ (t)=f(y(t-1),….,y(t-n),u(t-2),…u(t-m))+g(y(t-1),…,y(t-n),u(t-2),…,u(t-m))u(t-1) (4.8)
fand g are two separate networks. A feedback linearization controller {Fig. (4.10)} is
obtained by calculating the controls according to:
u(t)=w(t)-f(y(t),…,y(t-n+1),u(t-1),…,u(t-m+1))/g(y(t),…,y(t-n+1),u(t-1),…,u(t-m+1)) (4.9)
Selecting the virtual control, w(t), as an appropriate linear combination of past outputs
plus the references enables an arbitrary assignment of the closed-loop poles. Figure
(4.11) represent the dynamic response of this model without any disturbance effect.
Fig. (4.10) Control by feedback linearization
89
Fig. (4.11) Control by feedback Linearization at steady state flight
Finally, it is obvious from Fig. (4.9), (4.10) and (4.11) that the inverse model has a great
advantage in controlling the AIM 9R missile model with the best performance rather than
feedback linearization or feedforward models. The feedforward model has bad performance
compared with the inverse one beside it has a very high frequency control signal that can not
match a practical missile actuator band width.
Finally, when inverse control technique is applied to the AIM 9R missile model in
chapter 2 the missile response can be examined with a sine wave input as follows:
-Figure (4.12) shows the missile pitch motion using NN controller (desired and actual ptch
angle).
-Figure (4.13) shows the integral square error using NN controller that is relatively high when
used for the whole mission.
-Figure (4.14) shows the missile absolute velocity of the missile using NN controller during the
whole mission varying sharply with the elevator deflection.
90
-Figure (4.15) shows the missile angle of attack using NN controller indicating the maximum
angle of attack = 9 that maintain the low angle of attack aerodynamic missile model used in
the present work.
-Figure (4.16) shows the missile elevator deflection angle using NN controller according to the
reference input (reference pitch angle).
Fig. (4.12) Pitch motion using NN controller with sine wave input
Fig. (4.13) Integral square error using NN controller with sine wave input
91
Fig. (4.14) Missile absolute velocity using NN controller with sine wave input
Fig. (4.15) Missile angle of attack using NN controller with sine wave input
92
Fig. (4.16) Elevator deflection angle using NN controller with sine wave input
4.4 Summary
(a) Explaining the basic structures of the common neural network.
(b) Using different neural network model for control.
(c) Selecting the best performance neural network controller .
(d) Applying the neural network controller after training to the AIM 9R missile
model.
(e) Evaluating the performance of neural network controller and the complete
system performance.
93
Chapter (5)
MISSILE GUIDANCE USING FUZZY LOGIC CONTROL
5.1 Introduction
The term fuzzy logic has been used in two different senses. It is thus important to clarify
the distinctions between these two different usages of the term. In a narrow sense, fuzzy
logic refers to a logical system that generalizes classical two-valued logic for reasoning
under uncertainty. In a broad sense, fuzzy logic refers to all of the theories and
technologies that employ fuzzy sets, which are classes with unsharp boundaries.
For instance, the concept of “warm room temperature” may be expressed as an
interval (e.g., [70 C or78 C] in classical set theory. However, the concept does not have
a well-defined natural boundary. A representation of the concept closer to human
interpretation is to allow a gradual transition from “not warm” to “warm”. In order to
achieve this, the notion of membership in a set needs to become a matter of degree. This
is the essence of fuzzy sets.
The power of the fuzzy logic appears when there is a need to control a nonlinear
model. Also the fuzzy logic enables control engineers to easily implement control
strategies used by human operators.
In the present work a real application of missile guidance used a fuzzy logic controller
to deal with the nonlinearity in the system and any other uncertainty such as cross
coupling effect between lateral and longitudinal motions and the gust effect.
5.2 Applying Fuzzy Inference System for Control
The fuzzy inference technique can be very useful in control engineering. A standard
control system would utilize a numerical input and produce numerical output and so
should a fuzzy controller. The knowledge base contains the set of inference rules chosen
to achieve the control objectives and the parameters of the fuzzy systems used to define
the data manipulation in the fuzzification, inference engine, and defuzzification
processes as shown in Fig. (5.1). The input to the fuzzification process is the measured or
estimated variable that appears in the antecedent part of the if-then rule. This input
94
variable has associated linguistic values to describe it. Each linguistic value is defined by
a membership function, parameterized by data from the knowledge base .In the
inference engine the decision –making logic is conducted, inferring control laws from the
input variables through fuzzy implication. The final step is the defuzzification process
where a crisp control command is determined based on the inferred fuzzy control law
[99, 100].
Fig. (5.1) The structure of a Fuzzy Logic Control System
5.2.1 Designing membership functions (Fuzzification process)
Designing membership functions for a fuzzy set is a very important step in the
fuzzification process. Type number of membership function is affected by many factors
such as the type of the dynamic model, the nonlinearity of the model and the degree of
the affecting disturbance (uncertainty of the system). A membership function can be
designed in three ways:
(a) Interview those who are familiar with the underlying concept and later adjust it
based on a tuning strategy.
(b) Construct it automatically from data.
(c) Learn it based on feedback from the system performance.
The first approach was the main approach used by fuzzy logic researchers until 1980.
Due to lack of systematic tuning strategies in those days, most fuzzy systems were tuned
through a trial and error process. Figure (5.2) shows one of the simplest type of
membership functions which is a triangular type membership function. There are many
other types such as Guassian and trapezoid.
95
Fig. (5.2) Triangular membership function
In this work the triangular membership function was chosen for all the used fuzzy sets
(error, change in error, control signal). Figures (5.3), (5.4) and (5.5) show the
membership functions for the error signal, the change in error signal, and the control
signal respectively.
Fig. (5.3) Membership function for error input (e)
Fig. (5.4) Membership function for error rate input (ce)
96
Fig. (5.5) Membership function for control signal output (u)
5.2.2 Rule Derivation
The knowledge of the relationship between inputs and outputs of fuzzy controllers
are expressed as a collection of if-then rules to form what so called the rule-base of the
fuzzy controller. Four methods, possibly used in combination, are mainly used to
generate the rule-base of a fuzzy controller.
1-Expert experience and control engineering judgment [99, 100, 101].
In this approach the knowledge and experience of the designer are used to manually
construct the rule-base of the controller according to observation of the system.
2-Observation of an operator’s control action [102, 103, 104].
Many practical control tasks are difficult to describe by an explicit mathematical model.
However, they are successfully performed by skilled human operators. The input/output
data can be used to construct fuzzy control rules to mimic the behavior of human
operator.
3-Fuzzy model of the plant [105, 106].
A fuzzy model of the plant can be thought of as a linguistic description of the
dynamic characteristics of plant using fuzzy logic and inference. A set of subsequent
fuzzy control rules can be designed based on the fuzzy model of the plant to be
controlled.
4-Learning approaches [107, 108].
Motivated by the need of a systematic method to generate and modify fuzzy rule-bases,
much research is being conducted on developing learning approaches. This technique
begins with the self-organizing controllers which consist of two levels of fuzzy rule bases.
97
The first rule base is the standard control fuzzy rule base. The second level contains a
fuzzy rule base consisting of meta-rules, which attempt to assess the performance of the
closed loop control system and subsequently used to modify the standard rule base.
Learning approaches based on evaluation theory, such as genetic algorithms, may have
a promising potential towards the derivation of fuzzy rule bases. Also, neural network
can be used to form a hybrid neurofuzzy control system that is capable of self-tuning.
4.2.2.1 Structure of fuzzy rules
A fuzzy rule is the basic unit for capturing knowledge in many fuzzy systems. A fuzzy
rule has two components: an if-part (also referred to as the antecedent) and a then-rule
(also referred to as the consequent):
IF <antecedent> THEN <consequent>
The antecedent describes a condition, and the consequent describes a conclusion that
can be drawn when the condition holds.
The structure of a fuzzy rule is identical to that of a conventional rule in artificial
intelligence. The main difference lies in the content of the rule antecedent-the antecedent
of the fuzzy describes an elastic condition (a condition that can be satisfied to a degree)
while the antecedent of a conventional rule describes a rigid condition (a condition that
is either satisfied or dissatisfied). For instance, consider the two rules below:
R1: IF the annual income of a person is greater than 120K, THEN the person is rich.
R2: IF the annual income of a person is high, THEN the person is rich.
Where high is a fuzzy set defined be membership function. The rule R1 is a conventional
one, because its condition is rigid. In contrast R2 is a fuzzy rule because its condition can
be satisfied to a degree for those people whose income lies in the boundary of the fuzzy
set High representing high annual income. The consequent of fuzzy rules can be
classified into two categories:
(a) Crisp consequent: IF….THEN y=a
Where a isnonfuzzy numeric value or symbolic value.
(b) Fuzzy consequent: IF….THEN y is A
Where A is a fuzzy set.
98
4.2.3 Fuzzy rule-based inference
The algorithm of fuzzy rule-based inference consists of three basic steps and an
additional optional step.
(1) Fuzzy matching: Calculate the degree to which the input data match the
condition of the fuzzy rules. Figure (5.6) shows the degree of matching for two
rules using two different inputs through conjunctive conditions.
Fig. (5.6) Fuzzy matching process
(2) Inference: Calculate the rule’s conclusion based on its matching degree. Figure
(5.7) shows the clipping method for fuzzy inference for the first rule at matching
degree =0.3.
Fig. (5.7) Fuzzy inference process
99
(3) Combination: Combine the conclusion inferred by all fuzzy rules into a final
conclusion. Figure (5.8) shows the combining fuzzy conclusions inferred by the
clipping method.
Fig. (5.8) Fuzzy combination process
(4) (Optional) Defuzzification: for applications that need a crisp output (e.g., in
control system), an additional step is used to convert a fuzzy conclusion into a
crisp one. Many ways can be used for such a purpose such as MOA (Maximum of
area) and COA (Center of area). The centroid (or COA) defuzzification method
calculates the weighted average of a fuzzy set as shown in Fig. (5.9). The result of
applying COA defuzzification to a fuzzy conclusion can be expressed by the
formula:
A i ii
A ii
( )y yy
( )y
(5.1)
Where A is the membership of area A.
100
Fig. (5.9) Defuzzification process
In the present work these steps are applied to the inputs of the fuzzy logic
controller (system error, change in error) the control action (control signal) is then
generated at the defuzzification process. Figure (5.10) represent 25 rule-based
inference of Mamdani type fuzzy logic controller and final defuzzification step.
Fig. (5.10) Rule base calculations
101
5.2.4 Response pattern of a simple generic fuzzy control system
The simple control strategy of the fuzzy control system can be developed as shown in
Fig. (5.11), maps the error, e=yd–y into the control action u. At the heart of this scheme
is a fuzzy logic control algorithm that operates in discrete time steps of period T and
maps the normalized values of error, en (t), and the change in error, cen(t), defined
respectively as
en(t)=noe e(t) (5.2)
cen(t)=noce (e(t)-e(t-T)) (5.3)
wherenoeand noceare the corresponding normalization factors, into changes in the
control action un (t) via rules of the form
If en (t) is P and cen(t) is N then un(t) is Z.
Where P, N and Z are short for positive, negative and zero, defined as fuzzy sets over the
normalized domains and definition of the relevant variables as shown in Fig.(5.12) and
Table (5.1).
Fig. (5.11) Architecture of the generic fuzzy control system
In reference to Fig. (5.11) the computed value of ( u) is subsequently denormalized and
integrated (added to the past value of (u) to form the present value of (u)) as follows:
102
u nu(t) u(t T) (t)de u
(5.4)
wherede u is the corresponding denormalization factor.
Fig. (5.12) Step response of a simple generic fuzzy controller
Table (5.1) Control rules for a simple generic fuzzy controller
Region 1 2 3 4
IF error is P N N P
AND error rate is N N P P
THEN control signal (δu) is Z N Z P
In the same way the 25 rule-based Mamdani type fuzzy logic controller was established
and simulated through MATLAB then generating the final control action as shown in
Fig. (5.13). The main effort was done in tuning the fuzzy logic controller normalized and
denormalized factors to get the best output performance of the whole system.
103
Fig. (5.13) Fuzzy inference system control action
5.3 Fuzzy Logic Control Application
Figure (5.7) shows the response of the missile in terms of its pitch angle when
controlling using a neural network and fuzzy logic controllers. It is significant from Fig.
(5.14) that the output performance of the Fuzzy Logic Controller is better than the
Neural Network Controller output performance.
Fig. (5.14) Comparison between Fuzzy logic controller and NN controller
104
5.4 Comparison between Different Fuzzy Logic Controllers and PD
Controller
Figure (5.15) and (5.16) show the effect of using different normalized and denormalized
factors on the performance of fuzzy logic controller for pitch motion. Figure (5.17)
shows the difference of using a simple control law (PD controller)and the fuzzy logic
controller on pitch motion.
Fig. (5.15) Effect of normalized and denormalized factors
Fig. (5.16) Integral square error using PD and FL controllers for pitch motion
105
Fig. (5.17) Comparison between PD and FL controllers with combined sine wave input
for pitch motion
5.5 Low pass Filter
It is required to apply a low pass filter to blend the fuzzy logic controller input
signals (e, e.). The dynamic performance of that filter is shown in Fig. (5.18) and (5.19).
The low pass filter is Butterworth type of cutoff frequency = 11 (rad/s).
Fig. (5.18) Dynamic response of a low pass filter
106
Fig. (5.19) Integral square error for different Fuzzy logic controllers using low pass
filter
Figure (5.20) shows the error and error rate for the pitch motion using a low pass filter
that gives a smooth curvature expressing the real rate of increase and decrease of the
actual error, while Fig. (5.21) shows the error and error rate using the regular delay that
gives a bad display of error rate.
Fig. (5.20) Pitch error rate using filter
107
Fig. (5.21) Pitch error rate using delay only
5.6 Effect of Using Neural Network in the First Time Period
It is clear now that using fuzzy logic controller is better than PD controller or NN
controller in this application. But for the first time period (first 5-seconds) it is obvious
from Fig. (5.22) that using a neural network controller with fuzzy logic controller is
better than using fuzzy logic controller only, as the integrated square error when using
NN controller is less than that when using fuzzy logic controller alone.
Fig. (5.22) Effect of using NNC for the first 5 seconds
108
5.7 Fuzzy Logic Controller Output Signal
The output signal from the fuzzy logic controller is shown in Fig. (5.23) according
to the error (between the desired and the actual) and rate of error (e ,e ) with sine wave
input.
Fig. (5.23) Error, error rate and control signal as functions of time due to a sine wave
input to a fuzzy logic controller
5.8 Neurofuzzy Structure
Figure (5.24) shows the internal structure of the NeuroFuzzy controller
implemented through the Simulink editor.
Fig. (5.24) Neurofuzzy structure
109
5.9 The Effect of Flight Regimes on Neurofuzzy Controller Performance
Figure (5.25) shows the effect of flight regime (velocity) on the output performance of
the Fuzzy logic and Neural Network controllers.
Fig. (5.25) Two types of fuzzy logic controllers output performance
5.10 Composite Fuzzy Guidance Law
The justification for selecting any one of the guidance laws from this set is based on the
following heuristic analysis. When the target is far away, proportional navigation
guidance law (low level controller) yields moderate acceleration commands. However, as
the range becomes small, the acceleration commands will become very large. Thus,
under small range conditions, alternate guidance laws that use bounded acceleration
commands must be found. Since bang-bang guidance law (high level controller)
intercepts targets using bounded acceleration at every time instant, this guidance law is
ideal for use under small range conditions. The augmented proportional navigation
(medium level controller) can serve as the “bridge” guidance scheme between
proportional navigation and bang-bang law. Figure (5.26) shows the composite guidance
law with the selected parameters K1, K2, K3 calculated from Fuzzy inference system.
110
Fig. (5.26) Composite guidance law
Eighteen fuzzy inference rules are used in the composite fuzzy inference system to
calculate K1, K2, K3 as shown in Fig. (5.27) using the ruled-base inference principle as
discussed in the beginning of this chapter.
Fig. (5.27) Defuzzification process
111
Figures (5.28), (5.29) and (5.30) show the values of K1, K2, K3 according the Fuzzy
inference system inputs (Range, Range rate). Thus calculating the contribution of each
controller in the final control action.
Fig. (5.28) Fusion Fuzzy system output K1
Fig. (5.29) Fusion Fuzzy system output K2
Fig. (5.30) Fusion Fuzzy system output K3
112
Finally, the missile response to various inputs, as simulated using MATLAB and
missile model of chapter 2 when applying fuzzy logic controller of chapter 4 is as
follows:
(a) Sine wave input for pitch motion
- Figure (5.31) shows the missile pitch motion (desired and actual) reflecting the
effect of using fuzzy logic controller rather than PD controller.
- Figure (5.32) shows the integral square error for PD and FL controller indicating
that FL controller is obviously better than PD controller as the integral square
error for FL controller is less than the integral square error of the PD controller.
- Figure (5.33) shows the missile angle of attack using NN controller indicating the
maximum angle of attack = 9 that maintain the low angle of attack aerodynamic
missile model used in the present work.
- Figure (5.34) shows the missile elevator deflection angle using FLC according to
sine wave input as a reference input indicating that the maximum deflection of
the elevator is 21 and the nominal deflection is 9 .
- Figure (5.35) shows the missile absolute velocity increasing from 250 up to 550
m/s according to sine wave input for 40 s mission interval.
Fig. (5.31) Comparison between PD and FL controllers with sine wave input for pitch
motion
113
Fig. (5.32) Integrated square error for PD and FL controllers with wave input for pitch
motion
Fig. (5.33) Angle of attack using FLC with sine wave input
114
Fig. (5.34) Elevator deflection angle using FLC with sine wave input
Fig. (5.35) Absolute velocity of the missile using FLC with sine wave input
(b) Step input for pitch motion
- Figure (5.36) shows the missile pitch motion (desired and actual) reflecting the
effect of using fuzzy logic controller rather than PD controller.
115
- Figure (5.37) shows the integral square error for PD and FL controller indicating
that FL controller is obviously better than PD controller as the integral square
error for FL controller is less than the integral square error of the PD controller.
- Figure (5.38) shows the missile angle of attack using NN controller indicating the
maximum angle of attack = 7 that maintain the low angle of attack aerodynamic
missile model used in the present work.
- Figure (5.39) shows the missile elevator deflection angle using FLC according to
sine wave input as a reference input indicating that the maximum deflection of
the elevator is 13 and the nominal deflection is .45 .
- Figure (5.40) shows the missile absolute velocity increasing from 250 up to
830m/s according to sine wave input for 40 s mission interval.
Fig. (5.36) Comparison between the step responses of the missile due to fuzzy logic and
PD controllers for 0.3 rad step input for the pitch motion
116
Fig. (5.37) Comparison between FLC and PD controller errors for 0.3 rad step input for
the pitch motion
Fig. (5.38) Angle of attack response for 0.3 rad step input using FLC
117
Fig. (5.39) Elevator deflection angle response for 0.3 step rad step input using FLC
Fig. (5.40) Absolute missile velocity response for 0.3 rad step input using FLC
(c) Ramp input for pitch motion
- Figure (5.41) shows the missile pitch motion (desired and actual) reflecting the
effect of using fuzzy logic controller rather than PD controller.
118
- Figure (5.42) shows the integral square error for PD and FL controller indicating
that FL controller is obviously better than PD controller as the integral square
error for FL controller is less than the integral square error of the PD controller.
- Figure (5.43) shows the missile angle of attack using NN controller indicating the
maximum angle of attack = 5 that maintain the low angle of attack aerodynamic
missile model used in the present work.
- Figure (5.44) shows the missile elevator deflection angle using FLC according to
sine wave input as a reference input indicating that the maximum deflection of
the elevator is 7 and the nominal deflection is 1.2 .
- Figure (5.45) shows the missile absolute velocity increasing from 250 up to 750
m/s according to sine wave input for 40 s mission interval.
Fig. (5.41) Path angle response using PD, FL controllers with ramp input
119
Fig. (5.42) Comparison between the integral square error for PD, FL controllers with
ramp input
Fig. (5.43) Angle of attack response using FLC with ramp input
120
Fig. (5.44) Elevator deflection angle response using PD, FL controllers with ramp input
Fig. (5.45) Missile velocity response using FLC with ramp input
121
5.11 Summary (a) Applying fuzzy logic controller to AIM 9 R missile model and rule-base
derivation.
(b) Using low pass filter in order to blend the pitch error signal express the rate of
error properly.
(c) Evaluation of fuzzy logic controller performance compared with PD and neural
network controllers indicating that the FLC is better than using PD or NNC for
the whole mission.
(d) In order to decrease the error in the first time period, a superposition between
FLC and NNC should be used specially in the first time period.
(e) Development of a neurofuzzy structure using both types of controllers (FLC and
NNC).
(f) Multi-controllers and multi-guidance laws should be used to maintain the highest
performance of the controllers for all regimes of flight.
122
Chapter (6)
A HYBRID NEUROFUZZY CONTROL SYSTEM FOR MISSILE
GUIDANCE
6.1 Introduction
In this chapter an integrated control system for a multi-input multi-output air-to-air
missile based on both neural network and fuzzy logic controllers is presented. The
proposed control system is shown in Fig. (6.1) using MATLAB where the reference
inputs that represent the angles between the target position on focal plane array and the
missile C.G or (the angles between the tracking camera axes and missile body axes). A
superposition between the NN and FL controller’s outputs is shown in Fig. (6.1) where
the final control signals represent the elevator and rudder deflections. The main goal is
to control an air-to-air missile to follow the desired trajectory according to the actual
missile track rate with minimum error between desired and actual trajectories to secure
a minimum proximity fuse distance between the missile and the target (aircraft) to
achieve the required impact against the target. This control system is exposed to a
significant uncertainty due to cross coupling effect between inputs and due to the gust
affecting the missile actual performance.
Fig. (6.1) AIM 9 Model with 2 separate controllers
123
6.2 Methodology
Eventually, this project can be briefly explained in the following seven steps:
1.Deal with the (multi-input multi-output) model as two (single-input single-output)
models.
2. Train neural network controllers for the AIM model separately using applied kits.
3. Apply (25 rule MAMDANI fuzzy logic controllers) to both NN controllers models
after connecting them to the actual (multi-input multi-output) model as FLC will
enhance the output performance in case of existing gust or cross coupling effect as it is
the best way to deal with large uncertainty without having instability in the dynamic
system as shown in Fig (6.2).
Fig. (6.2) NeuroFuzzy structure
124
4.Simulation and Results
The proposed control system is established and simulated using MATLAB/SIMULINK
software. Appendix A includes block diagrams for MATLAB/SIMULINK dynamic
models of the AIM 9R missile. The reference input comes from the angle between the
moving camera and missile axis. The results of simulation are examined for step, ramp,
sine wave before applying the actual input data from the camera location. The results
are accepted in accordance with the data sheet of this missile referred to Tables (1.2)
and (1.3).
5- In order to track the target according to its location of the target on missile focal
plane array (missile dome window) it is required to transfer target location according to
a fixed frame of reference to its location according to missile body axes. This can be
done by transferring the target location to the missile body axes at fixed frame of
reference origin then transferring this location to the actual missile body axes as shown
in Fig. (6.3).
Fig. (6.3) Target location transformation from fixed frame of reference coordinate
to missile body axes coordinate
6-To plot the target and seeker flight actual paths it is required to calculate the aircraft
orientation angles then integrate Eq.’s (3.50), (3.51) and (3.52) using
MATLAB/SIMULINK editor to get the aircraft trajectory. Consequently the missile
125
trajectory from the intensity and the position of the target on the plane focal array at the
dome of the missile is calculated. Both aircraft and missile trajectories are shown in Figs
(6.2), (6.4) and (6.6).
Fig (6.4) Actual trajectories for target and missile
Fig. (6.5) 3D Plot for missile trajectory in both X,Y axes with respect to time
126
Fig. (6.6) 3D Plot for missile trajectory in both X, Z axes with respect to time
6- The maximum fuse distance can be calculated from the target-seeker difference
distance using Fig. (6.7) then verified according to the amount and type of missile
fragmentation. As the amount of kinetic energy absorbed by the target through the
fragments impulsion is inversely proportional to the miss distance between the aircraft
ad the missile. The ammunition type used in the missile affect also the amount of energy
absorbed by the target in the moment of collapse. Eq. (6.1) expresses the effect of the
ammunition type used in the missile and target-seeker miss distance.
tc.s
c.rc.r
LK.E. .E .
r (6.1)
K.E. = kinetic energy absorbed by the target in the moment of collapse (joule)
= fragmentation efficiency
127
c.rE = total energy of the fragments (joule)
tc.sL = average target cross section length (m)
c.rr = target-seeker miss distance (m)
Locking time is the shown in Fig. (6.7) and equals to half of the maximum locking time.
Fig. (6.7) Calculation of maximum fuse distance according to simulation data
7-Evaluating the integrated control system output error between the desired and actual
flight paths. It is required to calculate the angle of attack error, side slipping angle error
that results from the false desired trajectory coming from the inclination of the missile
axial axes to the missile actual velocity vector. Thus, subtracting these errors from the
total error gives the integrated control system errors as shown in Figs. (6.8) and (6.9).
128
Fig. (6.8) Total error in Z direction due to AOA and control system errors
Fig. (6.9) Total error in Y direction due to SSA and control system errors
129
Chapter (7)
DISCUSSIONS AND CONCLUSIONS
7.1 Discussions
In the present research work an initial research effort on nonlinear control design
methods for air-to-air missiles autopilots was conducted. Two different autopilots logics
were discussed: nonlinear fuzzy logic controller and neural network controller (inverse
model technique). The performance of these autopilots was demonstrated in nonlinear
simulations. Due to the wide variation in the flight conditions encountered by the
changing center of gravity location during flight, altitude, variable missile velocity and
the target-seeker distance that judge the type of applied guidance law, old linear
autopilots require gain scheduling with respect to Mach number, altitude, center of
gravity location of the missile. In the present thesis to cope with the system nonlinearity
fuzzy logic controller is used supported with neural network controller in the first time
period. But for the wide variation in the flight conditions fuzzy fusion system is used to
apply the adequate autopilot and the corresponding guidance law in accordance with the
flight conditions.
In the present aerodynamic model of AIM 9 missile has been established. Then PD,
neural network, fuzzy logic and combined controllers were applied separately and
respectively. The performance has been measured for different inputs.
The combination of fuzzy logic and neural network controllers is examined also to get
the best performance of the whole system especially with the affecting gust and the
existing cross coupling effect.
The simulation results in chapter 4, 5 show the performance of the dynamic system
for the three different types of controllers (PD, fuzzy logic and neural network) for
different types of inputs (step input, ramp input, combined sine wave). The final results
130
were statically evaluated by using the ISE (integrated square error) as shown in Table
(7.1).
Table (7.1) ISE for PD, NN, FL controllers
Controller type Step input Ramp input Combined sine wave input
PD controller .00388 .00633 .112
NNC --- --- 1.75
FLC .00318 .00580 .066
It is clear form Table (7.1) that the fuzzy logic performance controller has the best
performance compared with the PD and NN controllers as it has the smallest values of
ISE for different types of inputs.
The combination of fuzzy logic and neural network controllers has an advantage of
reducing the net errors in the first 5_seconds of the mission but for the rest of the
mission the performance decays rapidly. Table (7.2) explains the feature of the
combined and fuzzy logic controllers.
Table (7.2) Comparison between combined (NNC and FLC) and FLC only
Type of controller ISE
after 1
second
ISE
after 2
second
ISE
after 3
second
ISE
after 4
second
ISE
after 5
second
ISE
at the end of
the mission
Combined controller .009 .011 .020 .027 .035 .112
FLC .017 .020 .032 .034 .035 .066
It is noticed from Table (7.2) that the performance using both FLC and NNC in the
first 5 seconds time interval of the mission is better than the performance using FLC
only. As the fuzzy logic controller command is initially increasing slowly to reach the
desired value according to the incremental control signal value, but for the rest of the
mission it is recommended to use FLC only as using NNC may loose some stability
against the uncertainty even when used with FLC on the contrary when using FLC
131
alone. As it is robust enough when used alone to stabilize the whole system against the
existing uncertainty (cross coupling effect, gust). Consequently it is recommended to use
both types of controllers NNC, FLC in the first 5_seconds and using the FLC alone for
the rest of the mission.
Due to different regimes and wide input data range the Mamdani Fusion Fuzzy
System Classifier will enhance the output performance using multi controllers for
different regimes. Finally it is recommended that the locking time for triggering should
not exceed 0.065 seconds, the maximum proximity fuse distance should not be less than 5
meter. These specifications are according to the dynamic model data used in the
simulation.
7.2 Conclusion
(a) In order to fulfill the required specifications of maximum proximity fuse
distance and target locking time, a hybrid neurofuzzy control system should be
used in the first 5 seconds of the mission while fuzzy logic controller should be
used alone the rest of the mission.
(b) To enhance the performance of the missile during its whole mission it is
required to use multi guidance laws and multi neurofuzzy control systems to
cope with the uncertainty of the missile model according to flight conditions and
the target-seeker distance.
(c) To apply different guidance laws along the whole mission it is required to use
variable missile body axes, that means the origin of the missile body axes will be
shifted according to the target-seeker distance and applicable guidance law.
(d) To verify the maximum damage to the target during collapse it is required to
use variable locking time that activate the proximity fuse at the closest distance
to the target. This can be done by activating the proximity fuse at the moment
that (miss _ dis tan ce)t
= the smallest value= 0
132
7.3 Recommendations for Future Work
1-Exprimental application of the control techniques in the present work to verify the
results that has been reached from the computer simulations.
2-Optimization of the guidance system of the air-to-air missile according to the target
maneuver data (predicted path, velocity and acceleration). This will require multiple
types of guidance techniques verifying the best performance at any target-missile
regime.
133
REFERENCES
[1] http://vectorsite.tripod.com/avbomb2.html
[2] Lin, C. F. Modern Navigation, Guidance, and Control Processing. Prentice-Hall,
Englewood Cliffs, NJ, U.S.A., 1991.
[3] Bryson, A. E., Jr. and Y. C. Ho Applied Optimal Control. Blaisdell, Waltham, MA,
U.S.A., 1969.
[4] Zarchan, P. Tactical and Strategic Missile Guidance, 2nd Ed. AIAA, Inc., Washington,
D.C., U.S.A., 1994.
[5] Miller, W. T., R. S. Sutton, and P. J. Werbos Neural Networks for Control. MIT Press,
Cambridge, MA, U.S.A., 1991.
[6] Narendra, K. S. and K. Parthasarthy Identification and control of dynamical systems
using neural networks. IEEE Trans. Neural Networks, 1990, pp.4-27.
[7] Mamdani, E. H. and S. Assilian An experiment in linguistic synthesis with a fuzzy logic
controller. Int. J. Man Machine Studies, 1975, 7(1), pp.1-13.
[8] Lee, C. C. Fuzzy logic in control systems: fuzzy logic controller part I. IEEE Trans. Syst.
Man and Cyb., 1990, 20(2), pp. 404-418.
[9] Driankov, D., H. Hellendoorn, and M. ReinfrankAn Introduction to Fuzzy Control.
Springer, Berlin, Germany, 1993.
[10] Dash, P. K., S. K. Panda, T. H. Lee and J. X. Xu Fuzzy and neural controllers for
dynamic systems: an overview. Proc. Int. Conf. Power Electronics, Drives and Energy
Systems, Singapore, 1997.
[11] Locke, A. S. Guidance. D. Van Nostrand Co., Princeton, NJ, U.S.A. 1955
[12] Ha, I. and S. Chong Design of a CLOS guidance law via feedback linerization. IEEE
Trans. Aero.Electr. Syst., 1992, 28(1), 51-63.
[13] Zarchan, P. Tactical and Strategic Missile Guidance, 2nd Ed. AIAA, Inc., Washington,
D.C., U.S.A. 1994.
[14] Nazaroff, G. J. An optimal terminal guidance law. IEEE Trans. Automat. Contr., 1976,
21(6), 407-408.
[15] Potter, J. E. A Guidance Navigation Separation Theorem. AIAA, Washington, D.C.,
U.S.A. 1964, AIAA Paper 64-653
134
[16] Price, C. F. and R. S. Warren Performance Evaluation of Homing Guidance Laws for
Tactical Missiles. TASC Tech. Rept. TR-170-4, The Analytic Sciences Co., Reading,
MA,U.S.A. 1973.
[17] Nesline, F. W. and P. Zarchan A new look at classical vs. modern homing missile
guidance. AIAA J. Guid. Contr., 1981, 4(1), 78-85.
[18] Cherry, G. W. A General Explicit, Optimizing Guidance Law for Rocket-Propellant
Spacecraft., AIAA, Washington, D.C., U.S.A. 1964, AIAA Paper 64-638
[19] Wang, K. Optimal control and estimation for grazing angle problem. Proc. American
Control Conf., Atlanta, GA, U.S.A. 1988.
[20] Wang, Q., C. F. Lin, and C. N. D’Souza Optimality-Based Midcourse Guidance.,
Washington, D.C., U.S.A., 1993, AIAA Paper 93-3893
[21] Glasson, D. P. and G. L. Mealy Optimal Guidance for Beyond Visual Range Missiles.
AFATL-TR-83-89, USAF, Eglin AFB, FL, U.S.A., 1983.
[22] Cheng, V. H. L. and N. K. Gupta Advanced midcourse guidance for air-to-air missiles.
J. Guid. and Contr., 1986, 9(2), 135-142.
[23] Nesline, F. W. and M. L. Nesline How autopilot requirements constrain the
aerodynamic design of homing missiles. Proc. American Contr. Conf., San Diego, CA,
U.S.A., 1984.
[24] Stallard, D. V. An Approach to Autopilot Design for Homing Interceptor Missiles.,
AIAA, Washington, D.C., U.S.A., 1991, AIAA Paper 91-2612.
[25] Lin, C. F., J. Cloutier, and J. Evers Missile autopilot design using a generalized
Hamiltonian formulation. Proc. IEEE 1st Conf. Aero. Contr. Syst., Westlake Village,
CA, U.S.A., 1993.
[26] Lin, C. F. and S. P. Lee Robust missile autopilot design using a generalized singular
optimal control technique. J. Guid., Contr., and Dyna., 1985, 8(4), 498-507.
[27] Lin, C. F. Advanced Control System Design. Prentice-Hall, Englewood Cliffs, NJ,
U.S.A., 1994.
[28] Cloutier, J. R., J. H. Evers, and J. J. Feeley Assessment of air-to-air missile guidance
and control technology. IEEE Contr. Syst. Mag., 1989, 9(6), 27-34.
[29] Napolitano, M. R. and M. Kincheloe On-line learning neural-network controllers for
autopilot systems. J. Guid., Contr., and Dyna, 1995, 33(6), 1008-1015.
135
[30] Guez, A., J. L. Eilbert, and M. Kam Neural network architecture for control. IEEE
Contr. Syst. Mag., 1998, 8(2), 22-25.
[31] Lin, C. L. and Y. Y. Chen. Design of advanced guidance law against high speed
attacking target. Proc. Natl. Sci. Counc. ROC(A), 1999, 23(1), 60-74.
[32] Cottrell, R. G., T. L. Vincent, and S. H. Sadati Minimizing interceptor size using neural
networks for terminal guidance law synthesis. J. Guid., Contr., and Dyna., 1996, 19(3),
557-562.
[33] Hsiao, Y. H. Adaptive Feedforward Control for Disturbance Torque Rejection in
Seeker Stabilizing Loop. M.S. Thesis. Feng Chia University, Taichung, Taiwan,
R.O.C., 1998.
[34] Geng, Z. J. and C. L. MaCullough Missile control using fuzzy cerebellar model
arithmetic computer neural networks. J. Guid., Contr. and Dyna, 1997, 20(3), 557-565.
[35] Steck, J. E., K. Rokhsaz, and S. P. Shue Linear and neural network feedback for flight
control decoupling. IEEE Contr. Syst. Mag., 1996, 16(4), 22-30.
[36] Zhu, J. J. and M. C. Mickle Missile autopilot design using a new linear time-varying
control technique. J. Guid., Contr., and Dyna, 1997, 20(1), 150-157.
[37] Psaltis, D., A. Sideris, and A. Yamamura Neural con-trollers. Proc. 1st Int. Conf.
Neural Networks, San Diego, CA, U.S.A., 1987.
[38] Lightbody, G. and G. W. Irwin Neural model reference adaptive control and
application to a BTT-CLOS guidance system. Proc. IEEE Int. Conf. Neural Networks,
Orlando, FL, U.S.A., 1994.
[39] Hunt, K. J. and D. Sbarbaro Hofer Neural networks for nonlinear internal model
control. IEE Proc. Pt. D, 1991, 138(5), 431-438.
[40] Montague, G. A., M. J. Willis, M. T. Tham, and A. J. Morris Artificial neural networks
based multivariable predictive control. Proc. IEE 2nd Int. Conf. Artificial Neural
Networks, Bournemouth, U.K., 1991.
[41] Saint-Donat, J. N. Bhat, and T. J. McAvoy Neural net based model predictive control.
In: Advances in Intelligent Control, Chap. 8. C.J. Harris Ed. Taylor and Francis, 1994.
[42] McDowell, D. M., G. W. Irwin, and G. McConnell Hybrid neural adaptive control for
bank-to-turn missiles. IEEE Trans. Contr. Syst. Tech., 1997, 5(3), 297-308.
[43] Mizumoto, M. Realization of PID controllers by fuzzy control methods. IEEE Int. Conf.
Fuzzy Syst., Piscataway, NJ, U.S.A., 1992.
136
[44] Qiao, Q. Z. and M. Mizumoto PID type fuzzy controller and parameters adaptive
method. Fuzzy Sets and Systems, 1996, 78, 23-25.
[45] Mishra, S. K., I. G. Sarma, and K. N. Swamy Performance evaluation of two Fuzzy-
logic-based homing guidance schemes. J. Guid., Contr., and Dyna., 1994, 17(6), 1389-
1391.
[46] Leng, G. Missile guidance algorithm design using inverse kinematics and fuzzy logic.
Fuzzy Sets and Systems, 1996, 79, 287-295.
[47] Gonsalves, P. G. and A. K. Caglayan Fuzzy logic PID controller for missile terminal
guidance. Proc. 1995 IEEE Int. Symp.Intell.Contr., Monterey, CA, U.S.A., 1995.
[48] Schroeder, W. K. and K. Liu An Appropriate Application of Fuzzy Logic: A Missile
autopilot for dual control implementation. 1994 IEEE Int. Symp.Intell.Contr.,
Columbus, OH, U.S.A., 1994.
[49] Lin, C. K. and S. D. Wang A self-organizing fuzzy control approach for bank-to-turn
missiles. Fuzzy Sets and Systems, 1998, 96, 281-306.
[50] Farinewata, S. S., D. Pirovolou, and G. J. Vachtsevanos An input-output stability
analysis of a fuzzy controller for a missile autopilot’s yaw axis. Proc. 3rd IEEE Conf.
Fuzzy Syst., Orlando, FL, U.S.A., 1994.
[51] Hessburg, T. Fuzzy logic control for lateral vehicle guidance. Proc. 2nd IEEE Conf.
Contr. Appl., Vancouver, BC, Canada, 1993.
[52] Zhao, Z. Y., M. Tomizuka, and S. Isaka Fuzzy gain scheduling of PID controllers.
IEEE Trans. Syst., Man, Cyb., 1993, 23(5), 1392-1398.
[53] Ling, C. and T. F. Edgar A new fuzzy gain scheduling algorithm for process control.
Proc. American Contr. Conf., Chicago, IL, U.S.A., 1992.
[54] Dash, P. K. and S. K. Panda Gain-scheduling adaptive control strategies for HVDC
systems using fuzzy logic. Proc. Int. Conf. Power Electronics, Drives and Energy
Systems, New Delhi, India, 1996.
[55] Huang, C., J. Tylock, S. Engel, and J. Whitson Comparison of Neural-Network-Based,
Fuzzy-Logic-Based, and Numerical Nonlinear Inverse Flight Controls, AIAA,
Washington, D.C., U.S.A., 1994, AIAA Paper 94-3645.
[56] Ting, C. S., T. H. S. Li, and F. C. Kung An approach to systematic design of the fuzzy
control system. Fuzzy Sets and Systems, 1996, 77, 151-166.
137
[57] Palm, R. and D. Driankov Stability of fuzzy gain schedulers: sliding-mode based
analysis. Proc. 6th IEEE Int. Conf. Fuzzy Systems, Barcelona, Catalonia, Spain, 1997.
[58] Kwong, W. A. and K. M. PassinoDynamically focused fuzzy learning control. IEEE
Trans. Syst., Man, Cyb., 1996, 26(1), 53-74.
[59] Kandel, A. and G. Langholz Fuzzy Control Systems. CRC Press, Boca Raton, FL,
U.S.A, 1994.
[60] Boulet, V., E. Druon, D. Willaeys, and P. Vanheeghe Target estimation using fuzzy
logic. Proc. 1993 IEEE Int. Conf. Syst., Man and Cyb., Piscataway, NJ, U.S.A, 1993.
[61] Rugh, W. J. Analytical framework for gain scheduling. IEEE Contr. Syst. Mag., 1991,
11(1), 79-84.
[62] Hualin, T. and W. J. Rugh Overtaking optimal control and gain scheduling. Proc.
American Contr., Conf., Albuquerque, NM, U.S.A, 1997.
[63] Tan, S., C. C. Hang, and J. S. Chai Gain scheduling: from conventional to neurofuzzy.
Automatica, 1997, 33(3), 411-419.
[64] Balas, G. J. and A. K. Packard Design of robust time varying controllers for missile
autopilot. First IEEE Conf. Contr. Appl., Dayton, OH, U.S.A, 1992.
[65] Eberhardt, R. and K. A. Wise Automated gain schedules for missile autopilots using
robustness theory. First IEEE Conf. Contr. Appl., Dayton, OH, U.S.A, 1992.
[66] Nichols, R. A., R. T. Reichert, and W. J. Rugh Gain scheduling for controllers: a
flight control example. IEEE Tans. Contr. Syst. Tech., 1993, 1(2), 69-79.
[67] Piou, J. E. and K. M. Sobel Application of gain scheduling eigenstructure assignment
to flight control design. Proc. 1996 IEEE Int. Conf. Contr. Appl., Dearborn, MI,
U.S.A, 1996.
[68] Takagi, T. and M. Sugeno Fuzzy identification of systems and its applications to
modeling and control. IEEE Trans. Syst., Man, Cyb., 1985, 15(1), 116-132.
[69] Zhao, J., V. Wertz, and R. Gorez Fuzzy gain scheduling controllers based on fuzzy
models. Proc. 5th IEEE Int. Conf. Fuzzy Syst., New Orleans, LA, U.S.A, 1996.
[70] Yang, C. D., T. M. Kuo, and H. C. Tai gain scheduling using fuzzy rules. Proc. 35th
Conf. Dec. Contr., Kobe, Japan, 1996.
[71] Gonsalves, P. G. and G. L. Zacharias Fuzzy logic gain scheduling for flight control.
Proc. 3rd IEEE Conf. Fuzzy Syst., Orlando, FL, U.S.A, 1994.
138
[72] Wang, J. and W. Zhang A dynamic back propagation algorithm with application to
gain scheduled aircraft flight control system design. Proc. Intell. Infor.Syst., Los
Alamitos, CA, U.S.A, 1997.
[73] Adams, R. J., A. G. Sparks, and S. S. Banda A Gain scheduled multivariable design for
a manual flight control 1system. First IEEE Conf. Contr. Appl., Dayton, OH, U.S.A,
1992.
[74] Tanaka, T. and Y. Aizawa A Robust Gain Scheduler Interpolated into Multiple Models
by Membership Functions., Washington, D.C., U.S.A, 1992, AIAA Paper 92-4553.
[75] Pedrycz, W. and J. F. Peters Hierachical fuzzy controllers: Fuzzy gain scheduling. 1997
IEEE Int. Conf. Syst. Man, Cyb., Orlando, FL, U.S.A, 1997.
[76] Chai, J. S., S. Tan, and C. C. Hang Gain scheduling control of nonlinear plant Using
RBF neural network. Proc. IEEE Int. Symp.Intell.Contr., Dearborn, MI, U.S.A, 1996.
[77] Maia, C. A. and P. Resende Neural control of MIMO nonlinear plants: a gain-
scheduling approach. Proc. 12th IEEE Int. Symp.Intell.Contr., Istanbul, Turkey, 1997.
[78] Liang, R. H. and Y. Y. Hsu Scheduling of hydroelectric generations using artificial
neural networks. IEE Proc. Gener.Transm.Distrib., 1994, 141(5), 452-458.
[79] Cavalieri, S. and O. Mirabella Neural networks for process scheduling in real-time
communication systems. IEEE Trans. Neural Networks, 1996, 7(5), 1272-1285.
[80] Wang, Q., D. R. Broome, and A. R. Greig Intelligent gain-scheduling using neural
networks for robotic manipulators. Workshop on Neural Network Applications and
Tools, Liverpool, U.K., (1994).
[81] Tomescu, B. and H. F. VanLandinghamNeurofuzzymultimodel control using Sugeno
inference and Kohonen tuning in parameter space. 1997 IEEE Int. Conf. Syst., Man,
Cyb., Orlando, FL, U.S.A., (1997).
[82] Song, J.M. and Zhang, T.Q. “Passive homing missile’s variable structure proportional
navigation with terminal angular constraint” Aeronaut, 2001, vol.14, pp.83-87.
[83] Chio, J.Y. andChaw, D.K. “Nonlinear adaptive guidance considering target
uncertainties and control loop dynamics” IEEE Trans.,vol. 1 pp.506-511, 2001
[84] Innocenti, M. “nonlinear guidance technique for agile missiles” Control Eng., 2001, vol
9, pp.1131-1144.
139
[85] Anderson, M.B. and Burkhalter, J.E. “Design of a guided missile interceptor using a
genetic algorithm” Spacecr Rockets , 2001, vol 38, pp.28-35.
[86] Zhou, Di and Mu. Chundi “Motion tracking sliding mode guidance of a homing missile
with bearings-only measurements” Trans JpnSoc Aeronaut Space Sci, 2000, vol. 43,pp.
130-136.
[87] Narendra K. S. and Parthasarathy K., ”Identification and control of dynamical systems
using neural networks,”IEEE Trans. Neural Networks, 1990, vol. 1,pp.4-27.
[88] Park Y. M., Choi M. S., and Lee K. Y., ”An optimal tracking neuro-controller for
nonlinear dynamic systems,” IEEE Trans. Neural Networks, 1996, vol. 7, pp.1099-1110.
[89] Hamdy A., and El Desoky S., ” Development of the preliminary design of the missile
command guidance and testing “ M. Sc.Thesis, Cairo university, 1997
[90] Ku C. C. and Lee K. Y.,”Diagonal recurrent neural networks for dynamic systems
control, ” IEEE Trans. Neural Networks, 1995, vol. 6,pp.144-156.
[91] Brdys M. A. and Kulawski G. J., ”Dynamic neural controllers for induction motor,”
IEEE Trans. Neural Networks, 1999,vol. 10,pp. 340-355.
[92] Widrow B., Hoff M.E., ”Adaptive switching circuits,”1960 IRE WESCON convention
Record, New York: IRE Part 4, 1960, pp. 96-104.
[93] Rosenblatt F., ”The perceptron: A probabilistic model for information storage and
organization in the brain,” Psychological Review, 1958, vol. 65, pp. 386-408.
[94] Minsky M. and Papert S., Perceptrons, Cambridge, MA: MIT Press, 1969.
[95] Kohonen T., ”Correlation matrix memories,” IEEE Transactionson Computers, 1972,
vol. 21,pp. 353-359.
[96] Hopfield J. J., ”Neural networks and physical systems with emergent collective
computational properties, ” Proceedings of the National Academy of Science, 1982, vol.
79, pp. 2554-2558.
[97] Rumelhart D. E., Hinton G. E. and Williams R. J., ”Learning representations by back-
propagating errors,” Nature, 1986, vol. 323, pp. 533-536.
[98] Chow T. W. S. and Fang Y.,”A recurrent neural-network-based real-time learning
control strategy applying to nonlinear systems with unknown dynamics,” IEEE Trans.
Indust.Electron., 1998, vol. 45,pp. 151-161.
[99] Takagi T. and Sugeno M., “Fuzzy identification of systems and its applications to
modeling and control,” IEEETans, Syst., Man, Cybern., 1985, vol. 15, pp. 116-132.
140
[100] Sugeno M. and Yasukawa T.,”A fuzzy-logic-based approach to qualitative
modeling,”IEEE Trans. Fuzzy Syst., 1993, vol. 1, pp. 7-31.
[101] Wang L. X., Adaptive Fuzzy System and control: Design and Stability Analysis.
Englewood Cliffs, NJ: Prentice-Hall, 1994.
[102] Dickerson J. A. and Kosko B., ”Fuzzy function approximation with ellipsoidal
rules,”IEEE Trans. Syst., Man, Cybern., 1996, vol. 24, pp. 542-560.
[103] Cherkassky V., Gehring D., and Mulier F., “Comparison of adaptive methods for
function estimation from samples,” IEEE Trans. Neural Networks, 1996, vol. 7, pp.
969-984.
[104] Chen S. and billings S. A., “Neural networks for nonlinear dynamic system modeling
and identification,” Int. J. Control, 1992, vol.56, pp. 319-346.
[105] Choi J., Van Landingham H. F., and Bingulac S., “ A constructive approach for
nonlinear system identification using multilayer perceptrons,” IEEE Trans. Syst.,
Man, Cybern. B, 1996, vol. 26, pp. 307-312.
[106] Mendel J. M. and Mouzouris G. C.,”Designing fuzzy logic systems,” IEEE Trans.
Circuits Syst., 1997vol. 44, pp. 885-895.
[107] Sugeno M., “Theory of Fuzzy Integrals and Its Applications,” Doctoral Thesis, Tokyo
Institute of Technology, 1974.
[108] Adlassnig K. P., “Fuzzy set theory in medical diagnosis, “ IEEE Trans. Syst., Man,
Cybern., 1986vol. 16, pp, pp. 260-265.
[109] Menon P. K. and Iragavarapu V. R., “Blended Homing Guidance Law Using Fuzzy
Logic,” Optimal Synthesis Inc., 1998
[110] Lu-Ping Tsao and Ching-Shaow Lin, “A new optimal guidance law for short-range
homing missiles,” Department of system engineering, Taiwan, R.O.C., 2000
[111] Narendra K. S. and Parthasarathy K., ”Identification and control of dynamical
systems using neural networks, ”IEEE Trans. Neural Networks, 1990, vol. 1, pp.4-
27.
[112] Ku C. C. and Lee K. Y.,”Diagonal recurrent neural networks for dynamic systems
control, ” IEEE Trans. Neural Networks, 1995, vol. 6, pp.44-156.
[113] Takagi T. and Sugeno M., “Fuzzy identification of systems and its applications to
modeling and control,” IEEETans, Syst., Man, Cybern., 1985, vol. 15, pp. 116-132.
141
[114] Menon P. K. and Iragavarapu V. R., “Blended Homing Guidance Law Using Fuzzy
Logic,” Optimal Synthesis Inc., 1998
[115] Lu-Ping Tsao and Ching-Shaow Lin, “A new optimal guidance law for short range
homing missiles, ” Department of system engineering, Taiwan, R.O.C., 2000
[116] Hosny, A.M., Zeyada, Y.F., Hassaan, G.A., “Development of a neurofuzzy control
system for the guidance of air to air missile” MDP 8, Cairo, 2004, vol 1, pp. 465-475.
142
Appendix A
Integrated Control System
143
M 9R Aerodynamic Model
144