LEARNING AIR-TO-GROUND INTEGRATED FUZZY GUIDANCE

SYSTEM USING TABU SEARCH

Mohamed Rizk, SM IEEE Ahmed ElSayed
Faculty of Engineering, Department of Computer Science and Engineering,
Alexandria University, University of Bridgeport,
Alexandria, Bridgeport, CT,
Egypt. USA.
mrmrizk@ieee.org eng_ahmedabdelsalam@yahoo.co.uk

ABSTRACT
In this paper we consider the problem of air to ground missile guidance system
using fuzzy controller which uses Proportional Navigation and Pure Pursuit
methods. The fuzzy controller’s rules are defined using Tabu Search (TS). The
missile model is assumed as three degree of freedom (3DOF) (assuming the
analysis in the vertical plane only). This paper presents numerical results for
applying the learned fuzzy controller to the missile in different situations of control
parameters.

Keywords: Fuzzy Control, Missile Guidance, Tabu Search.

1 INTRODUCTION of energy, which enables the missile to detect the

presence of the target, and a control system, which
In the last few In the last few years there has computes and analyzes the received data to get a
been an increasing interest in the applications of the control command suitable for the position of target.
fuzzy set theory in practical control problems. Fuzzy Missiles which use an active homing guidance is
control is applied to processes that are too complex completely independent, the missile does not require
to be analyzed by conventional techniques. The any signal from any external source or any guidance
missile guidance system is one of these systems intelligence [1].
which are complex system to analyze. There are The homing guidance systems have two major
many ways to design the guidance system such as guidance methods:
homing guidance, command guidance and beam-
rider guidance. In this paper we present the problem Pure pursuit method: it's a method in which the
of ground to air integrated missile guidance system, missile velocity vector is always directed toward the
which means that the controller works direct to the instantaneous target position.
missile dynamics without the autopilot and the
actuator, using fuzzy controller. First, the used Proportional navigation method: it’s a method in
guidance method, homing guidance, will be which the rate of change of missile heading is
explained; then a scenario of the missile mission will directly proportional to the rate of rotation of the
be explained. LOS (Line of Site) from the missile to target.
The active homing guidance method does not need
any external guidance equipment which makes the
1.1. Homing Guidance Systems: missile works in any place without the need of
building any fixed structure, except launcher. But the
A homing guidance system is defined as a distortion of the tracking and guidance equipment
guidance system by which a missile steer itself when the missile hits the target and destroys itself
toward a target by an internal mechanism without the makes some problems of this method.
need of external source for tracking the target or
itself. The homing guidance systems are classified 1.2. Missile mission description :
into three general types:
The missile mission presented in this paper is an
Active homing air-to-ground missile mission, which means that an
Semi active homing aircraft will launch a missile on a stationary target
Passive homing (such as a Tank, Artillery position…). Typically the
launching occurs far from the target and the
The active homing guidance system, in its consequence of aligning the missile's flight path with
simplest form consists of a transmitter and receiver the target early in flight causes the missile to fly in at

1
a shallow angle. The objective of this work is to hit where
the targets at the top where the fronts and sides are
more strongly protected. In this paper we design a
θ is attitude in radian
guidance controller and learn the controller rules q is budy rotation rate in rad/sec
with Tabu Search algorithm to get a small final miss
distance and the final attitude of the missile in the m is missile mass in kg.
order of 900. 2
g is the accelerati on of gravity in m/s
2 MISSILE MODEL 2
I yy is the moment of inertia about y axis in Kg.m
From the analysis of the forces of aerodynamics 2
W& is the accelerati on in the Z body axis in m/s
around the missile shown in Fig. 1, we can get the
following equations which describe the motion of the 2
q& is the change in body rotation rate in rad/s
missile [2] [3] [4]

T is the thrust in the X body axis in N

3
ρ is the air density in Kg/m
2
S ref is the reference area in m

C X is the coefficien t of aerodynami c force in

t he X axis
C Z is the coefficien t of aerodynami c force in
t he Z axis
C M is the coefficien t of aerodynami c moment
Fig. 1: Forces and variables around the missile
airframe about the Y axis
T + Fx d ref is the reference length in m
U& = − qW − g sin θ .......... ...( 1)
m
Fz δ is the fin angle in radians
W& = + qU + g cos θ .......... .........( 2 )
m FX is the aerodynami c force in the X body
M
q& = .......... .......... .......... .......... ..( 3 ) axis in N
I yy
FZ is the aerodynami c force in the Z body
θ& = q .......... .......... .......... .......... .........( 4 )
U e = U cos (θ ) + W sin (θ ).......... .......( 5 ) axis in N
W e = −U sin (θ ) + W cos (θ ).......... .....( 6 ) M is the aerodynami c moment along the Y
X& me = U e .......... .......... .......... .......... ( 7 ) body axis
Z& me = W e .......... .......... .......... .......... .( 8 ) q is the dynamic pressure in Pa
V is the airspeed in m/s
where
α is the incidence in radians
Fx = qS ref C x (Mach, α )
U is the velocity in the X body axis in m/s
Fz = q S ref C z (Mach, α, δ )
W is the velocity in the Z body axis in m/s
M = q S ref d ref C M (Mach, α, δ, q )
U e is the velocity in the X earth axis in m/s
1 2 We is the velocity in the Z earth axis in m/s
q= ρV
2
X me is the X position of the missile in the X
2 2
V = U +W earth axis in m
−1 ⎛ W ⎞
α = tan ⎜ ⎟
⎝U ⎠

2
 Z me is the Z position of the missile in the Z coordinate of the space of the missile motion, and the
value of maximum and minimum of the output is the
earth axis in m safe limit of the elevation angle which the missile
can have.
We will define also two terms as shown in Fig. 2,
which are 3.2 Using Pure pursuit method:
λ : missile to t arg et LOS (Line of S ite) angle in rad In pure pursuit method the controller will get the
λ& : rate of change of the missile to t arg et LOS two velocity components with respect to the earth
(Line of S ite) in rad / sec coordinate Ue, We as its two inputs and will produce
the elevation angle δ (fin deflection angle) as an
These two terms are used in the first controller output.
design which is based on the proportional navigation The reaction of the elevation angle on the missile
method. acceleration is the same as before. First consider the
membership function of the two inputs and the
output as shown in Fig. 5 [6] [7]. Second, the fuzzy
-Ze rules of the controller are learned using Tabu Search.
The value of the maxima and the minima of the
universe of discourse for each input and output are
normalized to the initial velocity of the missile at
launching time, and the value of maximum and
minimum of the output is the safe limit of the
elevation angle which the missile can have.

4 LEARNING FUZZY RULES USING TABU

λ
SEARCH
X Tabu Search is one of the algorithms that can be
Fig. 2: LOS angle from missile to target used in the optimization problems Fig. 3. In this
section we will use the Tabu Search to learn the
fuzzy rules of the missile guidance controller by
3 FUZZY CONTROLLER DESIGN consider it as an optimization problem [8].
In the two methods used the relation between the
This section presents a design of an integrated input variables and the output variable are not
fuzzy guidance system using two different types of obvious then the incremental learning algorithm
homing guidance. In the two methods the fuzzy rules based on Tabu Search is used. The final miss
table is determined by the tabu learning of the fuzzy distance between the missile and the target is the
rules. energy function for this optimization problem. This
energy function is a function of the controller rules
3.1 Using proportional navigation method: only and all the other dynamical variables are
assumed to be constant during the optimization
In proportional navigation method the fuzzy process. The initial position can vary only in the
controller will get the LOS angle from missile to starting of every learning loop but constant in the
& optimization process.
target ( λ ) and the rate of change of this angle ( λ ) By applying the algorithm for the two previous
as two inputs, and will produce the elevator angle δ controllers, the fuzzy rules shown in Table (1) (2)
(fin deflection angle) as an output. could be found
From the missile dynamics, the fin deflection
angle will change the acceleration normal to the
missile body and the moment about the missile Y
axis, which will change the position of the missile in
space [2].
First let us consider the membership function of
the two inputs [5] and the output as shown in Fig. 4.
Second, the fuzzy rules of the controller are learned
using Tabu Search.
The value of the maxima and the minima of the
universe of discourse for each input and output are
obtained from the maximum and the minimum of the
Fig. 2: Tabu Search Algorithm

3
 μUe
Table 1: PN method controller fuzzy rule table
NB NS Z PS PB

λ& VS S MS M MB B
λ
VS NM PB PB PB PB PB -1 1
Ue the first input of the
S NS NP NM PM NP NP controller

MS Z NS Z PS NP NP
M PS PS PS NM NP NP μWe
NB NS Z PS PB
MB NM PB NB NP Z NM
B NS NS Z PM NP NP

Table 2: Pure Pursuit method controller fuzzy -1

We the second input of the
1

rule table controller

Ue
NB NS Z PS PB μδ
We NB NM PM PB
NS PS
Z
NB PS PB PB PM Z
NS PM PB PS NM PB
Z NS PS PB PM NS
PS PB Z PM PB PM -300 300
δ The output of the
PB PB Z PS Z PS controller

Fig. 5: Pure Pursuit method inputs and output

membership functions

μVS
λ
S MS M MB B
5 NUMERICAL RESULTS

The controller is simulated [9] with numerical

example for the missile with the previous model and
λ
λ, The first input to the following numerical configurations:
the controller
U0 = 900 m/s (The initial velocity component in
the direction of the missile body X axis)
W0 = 0 m/s (The initial velocity component in
μ λ& VS S MS M MB B
the direction of the missile body Z axis)
q0 =0 rad/s (The initial angular velocity of the
missile about the missile body Y axis)
θ0 = 0 rad (The initial missile attitude angle)
λ& , The second λ&
The value of Zme0 and the target position are
chosen to get the best performance of every
μδ controller.
NB NM PM PB
In this section the numerical results of the
NS PS
Z simulation process will be seen for the two types of
the controllers.

5.1 Using proportional navigation (PN) method

0 0
and learned by Incremental Learning based
-30 30
δ The output of the on Tabu Search:
controller
In this method the used value of the Zme0 for
Fig. 4: PN method inputs and output learning the controller is
membership functions Zme0 = -2000 m (The initial missile altitude)

4
And the target position is: 7 REFERENCES
Xt0 = 2000 m
Zt0 = 0 [1] M. Abdel Rahim: Design of a Robust Controller
The resultant miss distance in order = 6.1 m for a Command Guidance System, (PhD. Thesis,
θ = -106 Faculty of Engineering, Alexandria University),
This means that the effective head of missile will (1994).
be approximately perpendicular to the thin area of [2] A.G. Biggs, B.E.: A Mathematical Model of the
target, which means that probability of missile to hit Missile System Suitable for Analogue Computation,
the target is 100 %. Australian Defense Scientific Service, Weapon
Fig. 6 shows the trajectory of the missile for this Research Establishment, Report SAD 20, no. 8
case. J.S.T.U. D3, (1954).
[3] Jan Roskam: Airplane Flight Dynamics and
5.2 Using Pure pursuit (PP) method and Automatic Flight Control, (Roskam Aviation and
learned by Incremental Learning based on Engineering Co.),( 1979).
Tabu Search: [4] "Aerospace Toolbox," Matlab, Mathworks Inc.
[5] Gerard Leng : Missile guidance algorithm design
In this method the used value of the Zme0 for using inverse kinematics and fuzzy logic, Fuzzy Sets
learning the controller is and Systems, Science Direct, 79, 287-295,(1996).
Zme0 = -4000 m (The initial missile altitude) [6]. L. A. Zadeh, Fuzzy Set, Information and Control,
And the target position is: vol. 8, pp. 338-353, (1965).
Xt0 = 4000 m [7] J. M. Mendel: Fuzzy Logic Systems for
Zt0 = 0 Engineering: A Tutorial, proc. IEEE, vol. 83, no. 3,
The resultant miss distance in order = 0.93 m pp. 345-377, (1995).
θ = -870 [8] Maurizio Denna: Giancarlo Mauri, Anna Maria
This means that the effective head of missile will Zanaboni. Learning Fuzzy Rules with Tabu Search
be approximately perpendicular to the thin area of An Application to Control, Fuzzy Systems, IEEE,
target, which means that probability of missile to hit pp.1063-6706, (1999).
the target is 100%. [9] "Fuzzy Logic Toolbox," Matlab, Mathworks Inc.
The head of the launching position of the missile
can be changed in a range of 400 m, and the
maximum value of miss distance will be in order of 6
m.
Fig. 7 shows the trajectory of the missile in this
case.

6 CONCLUSIONS

This paper presented the design of an air-to-

ground integrated fuzzy guidance system using fuzzy
controller. The missile model used was the nonlinear
exact 3DOF model which shows the fact that fuzzy
controller can be used for any complex system with
acceptable error.
Two methods of guidance were used in this
paper. In these two methods the relation between the
input variables and the controlled variable were not Fig. 6: PN method Missile trajectory
clear then the incremental learning algorithm based
on Tabu Search was used with the final miss distance
between the missile and the target as the energy
function. This energy function is a function of the
controller rules only and all the other variables were
assumed to be constant during the optimization
process. The initial position can vary only in the
starting of every learning loop but constant in the
optimization process. From the simulation results we
can see that the Tabu Search algorithm produced the
same rules even though the starting rules are
different, which means that the algorithm always
converge to the global minimum from any starting
rule.
Fig. 7: PP method Missile trajectory