ARTICLE IN PRESS

Control Engineering Practice 13 (2005) 373–382

Adaptive flight control design for nonlinear missile

Antonios Tsourdos*, Brian A. White
Department of Aerospace, Power & Sensors, Cranfield University, RMCS Shrivenham, Swindon SN6 8LA, UK
Received 31 January 2003; accepted 24 April 2004
Available online 20 June 2004

Abstract

The focus of this work is to investigate the possible benefits of modern nonlinear control design methods for missile autopilot
design. The basic requirement for an autopilot is firstly fast responses because of the short amount of time involved in the end-game.
A slow response could easily cause a miss if the target has the capacity to perform high-g evasive manoeuvers. Secondly, minimum
error is an obvious requirement if the missile is to achieve a kill. Finally, robustness to model uncertainties is important in order for
the missile to achieve its objective in the physical environment.
In the first part of this paper input–output approximate linearisation of a nonlinear missile has been studied. A method for
controlling the nonlinear system that is input–output linearisable is examined that retains the order of the system in the linearisation
process, hence producing a linearised system with no internal or zero dynamics. Desired tracking performance for lateral
acceleration of the missile is achieved by using a nonlinear control law that has been derived by selecting the lateral velocity as the
linearisation output. Simulation results are shown to exercise the final design and show that the linearisation and controller design
are satisfactory. Then an adaptive nonlinear controller is designed that guarantees tracking performance when the uncertain
parameters vary within a stability bound. An autopilot combining an indirect adaptive controller with approximate feedback
linearisation is proposed in order to achieve asymptotic tracking. Adaptation is introduced to enhance closed-loop robustness, while
approximate feedback linearisation is used to overcome the problem of unstable zero dynamics. Computer simulations show that
this approach offers a possible autopilot design for nonlinear missiles with uncertain parameters.
r 2004 Elsevier Ltd. All rights reserved.

Keywords: Nonlinear control; Feedback linearisation; Flight control; Missile system; Adaptive control

1. Introduction dynamic pressure. Several approaches, including adap-

tive control (Lin & Cloutier, 1991), nonlinear control
The performance of aerospace systems such as (White, Tsourdos, & Blumel, 1998) and gain scheduling
aircraft, spacecraft and missiles is highly dependent on (Shamma & Cloutier, 1993) have been used to alleviate
the capabilities of the guidance, navigation and control these tracking problems. While gain scheduling is
systems. To achieve improved performance in such conceptually simple and has been proven successful, it
aerospace systems, it is important that more sophisti- has virtually no guarantee of stability in the transitional
cated control systems be developed and implemented. In periods between operating points and relies on the fact
particular, as the performance envelope is expanded, the that the scheduling variables should only change slowly.
control schemes must become adaptive and nonlinear, Furthermore, there is a heavy design overhead due to
to provide performance over a greater range, in the face the large number of linear controllers which must be
of uncertain or changing operating conditions. derived and, as the performance demands of modern-
The tracking performance of a missile is also day missile systems become more stringent, alternatives
dependent on the location within the flight envelope to linear control are of increasing practical significance.
and varies with factors such as Mach number and Feedback linearisation is a popular method used in
nonlinear control applications, and there have been
*Corresponding author.
several flight control demonstrations (Snell, 1992).
E-mail addresses: a.tsourdos@cranfield.ac.uk (A. Tsourdos), Dynamic model inversion is the feedback linearisation
b.a.white@cranfield.ac.uk (B.A. White). method employed to design the missile autopilot. This

0967-0661/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conengprac.2004.04.023
 ARTICLE IN PRESS
374 A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382

method is very effective in applications to aircrafts and estimates the plant parameters from the control signal
missiles. The main drawback of dynamic model inver- and the output measurement. The estimated parameters
sion is the need for high-fidelity nonlinear force and are then used to update the controller gains according to
moment models that must be invertible in real time, one of the several control methodologies.
which implies a detailed knowledge of the plant The parameter identifier is used in the outer loop
dynamics, and the approach tends to be computation- design and continuously adjusts the parameter estimates
ally intensive. In general, dynamic model inversion is based on observation error. The certainty equivalence
sensitive to modelling errors. The application of robust principle suggests that these parameter estimates that
and/or adaptive control can alleviate this sensitivity and, are converging to their true values may be employed to
therefore, the need for detailed knowledge of nonlinea- asymptotically achieve the desired objective as para-
rities. meter estimates converge to their true value. The
In this paper an adaptive nonlinear control design adaptive scheme developed for the lateral missile flight
technique is applied to the autopilot for the missile control system is presented in the following sections.
model which is aerodynamically controlled. Missile Other adaptive schemes such as direct adaptive
motion is modelled to be nonlinear with unknown control schemes are discussed in details in Sastry
parameters. Based on the model, we adopt a design and Bodson (1989). In the schemes of that form,
procedure similar to Sastry and Bodson (1989), basically parameters do not need to converge to their true value
an adaptive feedback linearisation method. In this but they are required to stay bounded and converge to
scheme, unknown parameters are estimated and based some constant. Typically, if the system is persistently
on these estimates, control parameters are updated. exciting, then all the parameters will converge to their
Computer simulations show that this approach is very true values.
promising to apply the autopilot design for the missiles
which are highly nonlinear in aerodynamics with
unknown parameters. 2. Missile model
The missile model can be represented in the general
nonlinear state space The missile model used in this study derives from a
’ ¼ f ðx; yÞ þ gðx; yÞu;
xðtÞ nonlinear model produced by Horton of Matra-British
yðtÞ ¼ hðx; yÞ: ð1Þ Aerospace (Horton, 1992). This study will look at the
reduced problem of a 2 DOF controller for the pitch and
Typically the control law is based on a vector y# which is yaw planes without roll coupling. The angular and
an online estimate of the true parameter vector y: The translational equations of motion of the missile airframe
update laws for these adjusted parameters are deter- are given by
mined as part of the design and shall be such that the

closed loop system stability is preserved. The conver- 1 1 1
r’ ¼ Iyz rV0 Sd dCnr r þ Cnv v þ V0 Cnz z ;
gence of these parameters estimate to their true value y 2 2
is a necessary condition in the indirect schemes of 1
adaptive control. v’ ¼ rV0 SðCyv v þ V0 Cyz zÞ  Ur; ð2Þ
2m
An indirect adaptive controller consists of a para-
meter identification scheme and a controller whose gains where the variables are defined in Fig. 2 and Tables 1
are calculated on-line based on estimates of the plant and 2. Eqs. (2) describe the dynamics of the body rates
model parameters. The structure of the plant is assumed and velocities under the influence of external forces (e.g.
a priori, but the coefficients or parameters involved are Cyv ) and moments (e.g. Cnr ), acting on the frame. These
estimated based on the available input/output informa- forces and moments are derived from wind tunnel
tion. Fig. 1 shows the schematic diagram for indirect measurements and by using polynomial approximation
adaptive control schemes. The identification block algorithms, Cyv ; Cyz ; Cnr ; Cnv and Cnz (Horton, 1992) can
be represented by polynomials which can be fitted to the
set of curves taken from look-up tables for different
Parameter Estimates flight conditions. A detailed description of the model
Identifier
can be found in Horton (1992).
The aerodynamic forces and moments acting on the
airframe are nonlinear functions of Mach number,
U(t) X(t)
longitudinal and lateral velocities, control surface
Ym U(t) Y(t) deflection, aerodynamic roll angle and body rates.
Controller Plant Control of the missile will be accomplished in this
paper by controlling an augmented version of lateral
Fig. 1. Schematic diagram for indirect adaptive control schemes. acceleration.
 ARTICLE IN PRESS
A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382 375

p
u

v
w
q
r
y
z
Fig. 2. Airframe axes.

Table 1 where the Mach number M; and the total velocity V0 are
Nomenclature of missile model slowly varying.
Nomenclature From the viewpoint of autopilot design, the usual
output for control purposes is the lateral acceleration of
f Lateral acceleration
s Incidence angle the centre of gravity f : This is not always possible to
r Yaw rate measure directly as the accelerometer is usually placed
M Mach number ahead of the c.g. and will thus pick up some angular
U Velocity along the roll axis acceleration as well as c.g. acceleration. It will, however,
x Force along the roll axis
be assumed that it is directly measurable for this study
y Force along the pitch axis
d Missile diameter and no significant difficulty is seen in the design process
V0 Total velocity as a consequence of this choice. Indeed, the use of the
z Rudder angle accelerometer moment arm is beneficial in removing the
nonminimum phase zero that is inherent in the c.g.
Table 2 acceleration measurement. Another output that can be
Coefficients of missile model considered is the sideslip velocity v: This can be obtained
Cyv 0:5½ð25 þ M  60jsjÞð1 þ cos 4lÞ by measurement of incidence and knowledge of forward
þ ð26 þ 1:5M  30jsjÞð1  cos 4lÞ speed. This has some advantages over lateral accelera-
tion in that the resulting system is minimum phase, but
Cyz 10 þ 0:5½ð1:6M þ 2jsjÞð1 þ cos 4lÞ
þ ð1:4M þ 1:5jsjÞð1  cos 4lÞ
the relationship between sideslip velocity and accelera-
tion is nonlinear and thus acceleration is not as
Cnr 500  30M þ 200jsj accurately controlled as a direct measurement. Finally,
an augmented acceleration signal f% is considered. This
Cnv sm Cyv ; where produces a synthetic signal that removes the nonmini-
sm ¼ d 1 ½1:3 þ 0:1M þ 0:2ð1 þ cos 4lÞjsjþ
0:3ð1  cos 4lÞjsj  ð1:3 þ m=500Þ
mum phase attributes of other lateral acceleration
measurements.
Cnz sf Cyz ; where
sf ¼ d 1 ½2:6  ð1:3 þ m=500Þ
3. Augmented lateral acceleration
The dynamic equation for lateral acceleration can be
derived (White et al., 1998) and is given by The conditions for feedback linearisation of nonlinear
systems are restrictive and it is of practical interest to
f ¼ v’ þ Ur; investigate situations where these conditions are met
f ¼ V 0 ðCyv v þ V0 Cyz zÞ approximately. Such a case is the design of autopilot for
nonlinear missiles. Continuing the work of Krener
¼ V 0 ½ðCyv0 þ CyvM M þ Cyvs jsjÞv (1984), who gave conditions for approximate full state
þ V0 ðCyz0 þ CyzM M þ Cyzs jsjÞz linearisation of nonlinear multi-input systems, Hauser,
¼ V 0 ½ðC% yv0 v þ C% yvs jvjÞv þ V0 C% yz0 z þ V0 C% yzs jvjz Satry, and Kokotovic (1992) discuss approximate input–
output linearisation of single-input single-output sys-
¼ yv v þ yz z
tems which fail to have relative degree. In contrast to the
¼ fðvÞ þ cðv; zÞ; ð3Þ extended linearisation (Baumann & Rugh, 1986) and
 ARTICLE IN PRESS
376 A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382

pseudo-linearisation (Reboulet & Champetier, 1984), tion can be carried out to give
the technique presented in Hauser et al. (1992) does not
d2 f% % % %
approximate the system by a linear system or a family of 2
¼ afv v’ þ afr r’ þ bfz z’
dt
linear systems but rather by a single nonlinear system %
that is input–output linearisable. ¼ afv ½yv v þ ðyr  UÞr þ yz z
% %
This approach to the tracking problem differs from þ afr ½nr r þ nv v þ nz z þ bfz z’
the work of Isidori and Byrnes (1990) who provide % % % %
(fragile) conditions under which one can exactly track ¼ ½afv yv þ afr nv v þ ½afv ðyr  UÞ þ afr nv r
the output of a finite-dimensional exosystem. In % % %
þ ½afv yz þ afr nz z þ bfz z’
contrast, Hauser et al. (1992) provide approximate
% % % %
tracking of a large class of output signals that is valid ’
¼ a2v f v þ a2r f r þ b2z f z þ bfz z; ð9Þ
under a wide class of nonlinear perturbations in the
system model. where
% % %
a2v f ¼ ½afv yv þ afr nv 
3.1. Modelling
¼ ½ðy2v þ yr nv Þyv þ ðyv ðyr  UÞ þ yr nr Þnv ;
To reduce the zero problem, consider using an % % %
a2r f ¼ ½afv ðyr  UÞ þ afr nv 
augmented acceleration output f;% where
¼ ½ðy2v þ yr nv Þðyr  UÞ þ ðyv ðyr  UÞ þ yr nr Þnr ;
f% ¼ f  yz z: ð4Þ
%
The use of f% can be justified on the grounds that the %
b2z f ¼ ½afv yz þ afr nz 
%

acceleration provided directly by the rear control surface

¼ ½ðy2v þ yr nv Þyz þ ðyv ðyr  UÞ þ yr nr Þnz : ð10Þ
yz z is usually small in comparison to the main lift
acceleration f : It also causes the minimum phase zero to Substituting for v; as before, gives
occur as the lift force developed by the fin appears !
% %
directly on the c.g. acceleration variable. By removing d2 f% a2v f 2f% a2v f
¼ f þ ar  yr r
this term, the effect of the fin is cancelled from the dt2 yv yv
accelerometer signal. Using f;% we now have !
%
2f% a2v f %
f% ¼ f  yz z þ bz  yz z þ bfz z’ ð11Þ
yv
¼ yv v þ yr r: ð5Þ
or
This is not a direct function of the control surface z;
d2 f% % % % %
’
hence the relative degree is increased to 1. Differentiat- ¼ a% f2f f þ a% 2r f r þ b%2z f z þ bfz z; ð12Þ
ing the output f% gives dt2
where
df%
¼ yv v’ þ yr r’ % a2v
dt a% 2f f ¼ ;
y
¼ yv ½yv v þ ðyr  UÞr þ yz z þ yr ½nr r þ nv v þ nz z
v 
2f % a2v
¼ ½y2v þ yr nv v þ ½yv ðyr  UÞ þ yr nr r þ ½yv yz þ yr nz z a% r ¼ a2r  yr ;
yv
% % %

¼ afv v þ afr r þ bfz z; ð6Þ 2f% a2v
%
bz ¼ b2z  yz : ð13Þ
where yv
% These equations now give the dynamics of the augmen-
afv ¼ ½y2v þ yr nv ;
ted output f:%
%
afr ¼ ½yv ðyr  UÞ þ yr nr ;
% 3.2. Augmented acceleration feedback control
bfz ¼ ½yv yz þ yr nz : ð7Þ
This is now a direct function of the input z and hence These equations can be used to control the dynamics
the relative degree is 1. Substituting for v; as before, of f% by again constructing an error function
gives
! ! f%e ¼ f%d  f;% ð14Þ
% % %
df% afv f% afv f% afv where f%d is the desired lateral acceleration. With this
¼ f þ ar  yr r þ bz  yz z: ð8Þ
dt yv yv yv definition, the following feedback can be used
In order to achieve an order of 2 for the dynamics and % % % % df%e
b%2z f z þ bfz z’ ¼ a% 2f f f  a% 2r f r þ ke f%e þ ke’ ð15Þ
hence eliminate the zero dynamics, another differentia- dt
 ARTICLE IN PRESS
A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382 377

or In most cases yr E0; and if yr is assumed negligible, then



% %
b%fz
’z ¼ 1 a% 2f%f  a% 2f%r  b%2f%z þ ke f%e þ ke’ dfe : ð16Þ nz nv
% %E  U y  y : ð23Þ
f r z
bfz dt
bfz z v

This gives the closed loop equation The ratios of moment to force can also be approximated
d2 f% df%e by
2
¼ ke f%e þ ke’ : ð17Þ nz Jyy
dt dt E  lf ;
If the acceleration component in f%d is small, then yz m
nv Jyy
d2 f% d2 f%e E  ls ; ð24Þ
2
E 2 ð18Þ yv m
dt dt
where lf is the fin moment arm, and ls is the static
and hence
margin. With these approximations, the stability of the
d2 f%e df%e zero can be established as
2
þ ke’ þ ke f%e ¼ 0; ð19Þ
dt dt Jyy
sEUðlf  ls Þ : ð25Þ
where the constants ke and ke’ make (19) Hurwitz. Note m
that the relative degree of 1 for the dynamics has forced Hence a stable zero results if the fin moment arm is
the use of the bz z’ term. significantly greater than the static margin. This is
The configuration for the controller is shown in Fig. 3. usually the case in most agile missiles as the static
It can be seen from the figure that an extra pole has margin is made as close to zero as possible for a stable
been inserted in the forward path. This effectively missile and negative for an unstable missile. If the zero is
cancels the zero that must be in the quasi-linear large, then the z’ term can be neglected, and hence (15)
equations which are order two, relative degree one. becomes
The zero must be at the solution of
% % % df%e
% % b%2z f z ¼ a% 2f f f  a% 2r f r þ ke f%e þ ke’ ð26Þ
b%2z f z þ bfz z’ ¼ 0 ð20Þ dt
which gives a zero at or


% 1 % % df%e
b%2z f z ¼ % a% 2f f f  a% 2r f r þ ke f%e þ ke’ : ð27Þ
s ¼  %: ð21Þ b%2z f dt
bfz
% % The configuration for the simplified controller is shown
From the definition of b%2z f and bfz ; we have in Fig. 4.
%
b%2z f ½yv ðyr  UÞ þ yr nr ½nz  nv =yv yz 
% ¼ : ð22Þ 3.3. Simulation results
bfz ½yv yz þ yr nz 
Fig. 4 shows the nonlinear controller structure. A fast
linear actuator with natural frequency of 250 rad=s has

f f
fd 1_ _1 ζ fd 1_ ζ
G _ G
bζ r bζ r
s
_ _ _ _ _
+ + + +
_
bζ

_ _
ar a2r

_ _
af a2f
o_ _o
fe o
fe
keo ke
ke _ ke _
fe fe

Fig. 3. Augmented acceleration controller configuration. Fig. 4. Simplified augmented acceleration controller configuration.
 ARTICLE IN PRESS
378 A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382

Lateral acceleration
150

100

(m/sec2)
50

-50
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Time(sec)

150

100
(m/sec2)

50

-50
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Augmented Lateral acceleration

0
(m/sec)

-5

-10

-15
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Sideslip velosity
Fig. 5. Acceleration and lateral velocity for ad ¼ 100:

been included in the nonlinear system. The error 4. Adaptive nonlinear control
dynamics are constructed using the ad signal and the
feedback of the actual states—velocity, rate and accel- The design presented in the previous section was for
eration. the nominal missile model. However, neither the Mach
The error coefficients in Eq. (27) are chosen to satisfy number nor the mass of the missile remain constant. As
a Hurwitz polynomial. For the second order error the flight conditions vary, both Mach number and mass
equation in each channel, ke ¼ 2zwn and ke’ ¼ w2n ; where vary. From Eq. (1) and Tables 1–3 it can easily shown
wn ¼ 60ðrad=sÞ and z ¼ 0:65: The speed of response is that the different aerodynamic coefficients are multi-
significantly faster than the open loop missile response linear functions of Mach number and mass. Hence, as
and so should exercise the dynamics of the nonlinear these two variables vary, the aerodynamics coefficients
missile sufficiently for meaningful conclusions to be vary. Since the feedback control law was design only for
drawn. the nominal values of the aerodynamic coefficients, any
The results of a 100 m=s2 demand in acceleration is variations of these will cause inexact cancellation of the
shown in Figs. 5 and 6 for the 100 m=s2 case. The figures system’s nonlinearities, i.e. inexact decoupling.
show almost identical step responses with some varia- Consider a SISO nonlinear system of form (2) under
tion in peaks and steady state values for the body rate, parametric uncertainty
the actuator movement and the lateral velocity. The
’ ¼ F ðx; yÞ þ Gðx; yÞu;
xðtÞ
difference between the lateral acceleration and the
augmented acceleration shows that there is a good yðtÞ ¼ hðx; yÞ; ð28Þ
match between the two and that steady state values are where x1 ¼ r; x2 ¼ v; u ¼ z and
very close. This illustrates the small effect that the fin " #
1 1 1 1 1
force has on the missile acceleration and justifies the use 2 Iyz rV0 Sd 2 dCnr r 2 Iyz rV0 SdCnv v
F ðx; yÞ ¼ 1 ; ð29Þ
of the augmented acceleration. The results also show 2m rV0 SCyv v Ur
that the actuator does not significantly affect the design.
" #
The nonlinear approach is also shown to be reasonably 1 1 2
2 Iyz rðV0 Þ SdCnz zÞ
accurate, as the predicted and actual performance are Gðx; yÞ ¼ 2
: ð30Þ
1
very close. 2m rðV0 Þ SCyz zÞ
 ARTICLE IN PRESS
A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382 379

0.5

0.4

(rad/sec2)
0.3

0.2

0.1

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Yaw rate

0.03

0.02

0.01
(rad)

-0.01

-0.02

-0.03
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Fin angle
Fig. 6. Rate and fin angle for ad ¼ 100:

Further, assume F ðxÞ and GðxÞ have the form estimates Ld

i
Fh

x’# 1 ¼ Ld
X
n
#
F ðx; yÞ ¼ yFi Fi ðxÞ; F h6L
h F# h

x’# 2 ¼ Ld
2hþLd i1 # #
i¼1
F G LF h6L
h F#
h þ LG# LF# h; ð35Þ
X
m
Gðx; yÞ ¼ yG
i Gi ðxÞ ð31Þ
where from Eqs. (6) and (9)
i¼1
% %
with yF and yG vectors of unknown parameters and LF h ¼ afv v þ afr r;
Gi ðxÞ and Gi ðxÞ known functions. The estimates of these % %
L2F h ¼ a2v f v þ a2r f r;
functions are given by %
Xn LG LF h ¼ b2z f z: ð36Þ
# yÞ ¼
Fðx; y# Fi Fi ðxÞ;
i¼1
As in Sastry and Bodson (1989), since these estimates
Xm are not linear in the unknown parameters yi ; we define
# yÞ ¼
Gðx; y# G
i Gi ðxÞ; ð32Þ each of the parameters products to be a new parameter.
i¼1 For example
where y# j are the estimates of the unknown aerodynamic X
n X
m

parameters yj and are multi-linear functions of Mach LGd

LF h ¼ yFi yG
j LG LF h ð37Þ
i¼1 i¼1
number and mass. Fi ðxÞ and Gi ðxÞ are obtained by F ðxÞ
and GðxÞ by parameterising them according to y# Fi and and we let YARp be the large p-dimensional vector of
y# G
i : Now let us replace the control law (27) by all multi-linear parameter products yFi ; yG F G
j ; yi yj ; y .
1 h d i The vector containing all the estimates is denoted by
uad ¼ L2F h þ nad ð33Þ #
YAR p
with F6Y  Y # representing the parameter
LGdLF h
error.
with Due to the indirect nature of our approach, this over-

 parametrisation does not increase the complexity of the
d2 f%d df%d # #
nad ¼ þ k ’
e  x 2 þ ke ðf%d  x1 Þ; ð34Þ closed loop system since a parameter identifier is to be
dt2 dt
used to estimate the unknown parameters yj : The
where ke ; ke’ are chosen as before in Eq. (27), x1 ¼ f% and parameter vector Y is, however, constructed here in
%
x2 ¼ df=dt and x# i1 ¼ LiF h are replaced by their order to show the stability of the resulting adaptive
 ARTICLE IN PRESS
380 A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382

Table 3 system. The control law (33) now becomes

Characteristics of the Horton model
x’* 2 ¼ L2F h þ ½LG LF h  Lgd h ad  Ld
LF hu 2hþn
F ad
r0 Sea level air density 1:23 kg=m3
c2  þ ½L L h  L d
¼ ½L2F  L h ad þ nad : ð38Þ
F G F G LF hu
r Air density r0  0:094h
Subtracting n in error equation from both sides gives
h Altitude in km
e2 þ ke’ e þ ke e ¼ ½LG L1F h  LGdL1F hu
h ad
a0 Sea level velocity of sound 340 m=s 
c2  þ k L  L
þ ke’ ½L2F  L cF
F e’ F
a Velocity of sound a0  4h m=s
¼ FT wðx; uad ðxÞÞ; ð39Þ
d Reference diameter (calibre) 0:2 m
where wT 6½LG LF huad ðxÞjL2F hjLF h: Therefore, in the
S Reference area pd 2 =4 ¼ 0:0314 m2 closed loop, for the approximate system, we have in
compact form
l Length of the missile 2:7 m
e’ ¼ Ae þ W T ðx; uad ðxÞÞF; ð40Þ
sp Wing span 0:6 m where A is a Hurwitz matrix and note that if f6y 
#
y-0 as t-N; then F-0 as t-N: To estimate the
m Mass 125 kg nominal
150 kg full unknown parameters, we consider an observer-based
100 kg all burnt identifier proposed in Taylor, Kokotovic, Marino, and
Kanellakopoulos (1989), Kudva and Narendra (1973).
Iz ; Iy Lateral inertia 67:5 kg m2 nominal First we rewrite Eq. (28) as
75 kg m2 full !
60 kg m2 all burnt yF
x’ ¼ ðF1 ?Fn jG1 u?Gm uÞ G
xcg Centre of gravity 1:6 m full y
1:5 m all burnt
m
6 Z T ðx; uad ðxÞÞy: ð41Þ
1:3 þ 500
Consider the following identifier system:
xcp Centre of pressure (body+wings) x0cp ¼ 1:3 þ 0:1M þ 0:2jsj
x45 x’ ¼ Að #
# x#  xÞ þ ZT ðx; uad ðxÞÞy;
cp ¼ 1:3 þ 0:1M þ 0:3jsj

xf Centre of pressure (fins only) 2:6 m y’# ¼  Zðx; uÞPðx#  xÞ; ð42Þ

xr Fin moment arm 1:5 d2 m where A# is a Hurwitz matrix, x# is the observer state, x is
the plant state in Eq. (28), and P > 0 is a solution to the
sm Static margin (body+wings) xcp  xcg Lyapunov equation AP # þ PA# ¼ lI with l > 0: We
assume that all the states x in Eq. (28) are available and
sf Fin moment arm for lateral motion xf  xcg
hence x# and y# are given by Eq. (42). We also assume that
y is a vector of constant but unknown parameters. Then
# e þ Z T ðx; uÞf;
e’# ¼ A#
’ ¼  Zðx; uÞPe#
f ð43Þ

Outer Loop

Controller Parameter
Adaptive Controller Observer
Parmeters Estimates

External Controller Internal Controller Nonlinear System

vd e u x v
K(e) Ψ(x,v) f(x)+g(x)u h(x)
+
−

Linearized System

Inner Loop

Fig. 7. Adaptive tracking control design.

 ARTICLE IN PRESS
A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382 381

100 0

Lateral acceleration
80 -2

60 -4

(m/sec)
40 -6

20 -8

0 -10

-20 -12
0 0.2 0.4 0.6 0 0.2 0.4 0.6
Time(sec) Velocity v

0.02

0.4 0.01

Fin Angle(rad)
Yaw Rate

0
0.2
-0.01

0 -0.02

-0.03
0 0.2 0.4 0.6 0 0.2 0.4 0.6
Time(sec) Time(sec)
Fig. 8. Nonadaptive with uncertainty 35% in Mach number and mass for acceleration demand ad ¼ 100:

150 0
Lateral acceleration

100
-5
(m/sec)

50

-10
0

-50 -15
0 0.2 0.4 0.6 0 0.2 0.4 0.6
Time(sec) Velocity v

0.4 0.01
Fin Angle(rad)

0.3 0
Yaw Rate

0.2 -0.01

0.1 -0.02

0 -0.03
0 0.2 0.4 0.6 0 0.2 0.4 0.6
Time(sec) Time(sec)
Fig. 9. Adaptive with uncertainty 35% in Mach number and mass for acceleration demand for ad ¼ 100:

is the observer error system where e#6x#  x is the (5) e# and f converge exponentially to zero if Zðx; uÞ is
observer state error and f ¼ y#  y is the parameter R tþs rich (i.e. (d1 ; d2 ; s > 0 such that 8t :
sufficiently
error. d1 Ip t ZZ T dtpda I:
Properties of the observer-based identifier in Eq. (43)
are given in Sastry and Bodson (1989) and are
However, since Zðx; uÞ is a function of state x; condition
(1) fALN ; 5 cannot be verified ahead of time.
(2) with e#ð0Þ ¼ 0; fðtÞpfð0Þ 8tX0; The block diagram of our adaptive lateral flight
(3) e#ALN -L2 ; control design for the nonlinear missile is shown in Fig. 7
(4) if Z T ðx; uad Þ is bounded then e#ALN and e#-0 as while results of the nonadaptive and adaptive schemes
t-N; are shown in Figs. 8 and 9, respectively. Good tracking
 ARTICLE IN PRESS
382 A. Tsourdos, B.A. White / Control Engineering Practice 13 (2005) 373–382

performance for variation up to 35% in Mach number Hauser, J., Satry, S. S., & Kokotovic, P. (1992). Nonlinear control via
and mass is achieved. approximate input–output linearization: The ball and beam
example. IEEE Transactions on Automatic Control, 37(3),
392–398.
Horton, M. (1992). A study of autopilots for the adaptive control of
5. Conclusions tactical guided missiles. Master’s thesis, University of Bath.
Isidori, A., & Byrnes, C. (1990). Output regulation of nonlinear
Dynamic model inversion is the feedback linearisation systems. IEEE Transactions on Automatic Control, 35,
method employed to design the missile autopilot. The 131–140.
Krener, A. (1984). Approximate linearization by state feedback and
main drawback of dynamic model inversion is the need coordinate change. Systems and Control Letters, 5, 181–185.
for high-fidelity nonlinear force and moments models Kudva, P., & Narendra, K. (1973). Synthesis of an adaptive observer
that must be invertible in real time, which implies a using lyapunov’s direct method. International Journal of Control,
detailed knowledge of the plant dynamics, and the 18, 1201–1210.
Lin, C., & Cloutier, J. (1991). High performance, adaptive, robust
approach tends to be computationally intensive. In
bank-to-turn missile autopilot. In AIAA guidance, navigation,
general, dynamic model inversion is sensitive to model- control conference (pp. 123–137).
ling errors. In this paper an adaptive nonlinear control Reboulet, C., & Champetier, C. (1984). A new method for linearizing
design technique is applied to the autopilot for the nonlinear systems: The pseudolinearization. International Journal
missile model which is aerodynamically controlled. of Control, 40, 631–638.
Missile motion is modelled to be nonlinear with Sastry, S., & Bodson, M. (1989). Adaptive control, stability, covergence
and robustness. Englewood Cliffs, NJ: Prentice-Hall Inc.
unknown parameters. In the adaptive scheme used in Shamma, J., & Cloutier, J. (1993). Gain-scheduled missile autopilot
this paper, unknown parameters are estimated and design using lpv transformations. Journal of Guidance, Control and
based on these estimates, control parameters are Dynamics, 16(2), 256–263.
updated. Computer simulations show that this approach Snell, 1992. Nonlinear inversion flight control for a supermaneuverable
is very promising to apply the autopilot design for the aircraft. Journal of Guidance, Control and Dynamics, 15(4),
976–984.
missiles which are highly nonlinear in aerodynamics Taylor, D., Kokotovic, P., Marino, R., & Kanellakopoulos, I. (1989).
with unknown parameters. Adaptive regulation of nonlinear systems with unmodeled
dynamics. IEEE Transactions on Automatic Control, 34,
405–412.
References White, B.A., Tsourdos, A., & Blumel, A. (1998). Lateral acceleration
control design of a nonlinear homing missile. In Fourth IFAC
nonlinear control systems design symposium (pp. 708–713).
Baumann, W., & Rugh, W. (1986). Feedback control of nonlinear
systems by extended linearization. IEEE Transactions on Automatic
Control, 31, 40–46.