m tnp slfts

Awurka Coo -e
Sam Fruim, CaIIsrma.Jin WAI2 - 9:35
Sliding Mode Control of
HAVE DASH II Missile Systems *
J. Huang, C.F. Lin
American GNC Corporation
9131 Mason Ave.
Chatsworth, CA 91311

Abstract: Sliding mode control is applied to design au- based on the more complicated missile model and the perfor-
topilot for bank-to-turn (BTT) missile systems. The design mance is evaluated using the full-scale six degree-of-freedom
is compared with the feedback linearization approach and HAVE DASH II missile model.
shows a significant improvement over the latter.
2. Formulation of Sliding Mode Controller
1. Introduction It is known that sliding mode control can be used to achieve
The nonlinearity and uncertainty inherent in the HAVE trajectory tracking in presence of certain model uncertainty.
DASH II missile dynamics have been a great challenge to the In this section, we will formulate the sliding mode controller
high performance autopilot design. In recent years, much for a class of input-output feedback linearizable nonlinear
attention has been focused on usmg linear optimal control systems. Details are referred to [SL]. We consider the system
methods to design missile autopilot [LL], [LJGENV], [XX]. described by
Although these advanced multivaxiable approaches are com-
pared favorably to the classical one-loop-at-a-time method, i(t) = f(x()) + g(z(i))u(t)
they still manifest themselves incapabilities when perform- y(t) = h(x(t)) (2.1)
ing large and rapid maneuvers. A natural step towards im-
proving the autopilot's performance is to incorporate the where 4(t) is the n-dimensional plant state, u is m-
more accurate nonlinear missile model into the design pro- dimensional plant input, y is n-dimensional plant output,
cess. A typical nonlinear approach is the well-known feed- f: Rn R'and g: R- Rn xR,a and, h: R Rm + are
back linearization approach [HSM] which uses the feedback smooth functions. Sliding mode controller for system (2.1)
and/or coordinates transform to linearize the nonlinear sys- can be designed by the following steps.
tem, then addresses the design issues on the linearized sys- 1. Performing input-output feedback lineania-
tem thus obtained. However, success of the feedback lin- tion. System (2.1) is said to be input-output feedback
earization approach is hinged on the availability of the ac- linearizable if there exist constants P ,** Pm and input-
curate description of the model. Indeed, severe model un- output mapping of the form
certainty mainly due to the aerocoefficients may degrade the
effectiveness of the feedback linearization approach. A more Ft u i(i)
realistic approach should tolerate the model uncertainty to
some degree. With this regard, some robust schemes such Y(2) | = B(z(t)) A(z(t))
+
| (2.2)
as sliding mode control have been proposed [SS]. ymlp-)(t)
In this paper, we will apply sliding mode method to de-
%. ,~[

Y(R^)(g)~~~~~~~
Uin (t)
sign autopilot for the bank-to-turn HAVE DASH II missile where Ui i== , m are components of u and y, B :
-

system. The design wiU be carried out in two steps. First, R"-- R' and A: Rn- Rm x Rm are smooth with A(z)
a nonlinear inversion law is synthesized to decouple and lin- invertible. For if this is the case, the following feedback
earize the input-output dynamrics of a nominal missile sys- control
tem. Then based on this decoupled and linearized input-
output relation, the slidimg mode method is further applied u(t) = A- (x(t))(-B(1(t)) + v(t)) (2.3)
to design a switching mechanism to account for the model
uncertainty. The performance of the resulting controller is )m (t)]T E Rm, yields
where v(t) = [vi (t), e . ,v
compared to that of the feedback linearization approach. It
should be noted that similar approach ha. been applied to Y(PI(t) = Vi(t), i = 1,*..,nm (2.4)
the control of aircraft in [SI]. However, our development is which clearly exhibits a decoupled linear input-output struc-
'Research supported under USAF Contract No. F0o63-91- C-0055, ture. The integers Pl,i**, pm are called relative degree of
Air Force Armment Directorate, WL/MNAG (2.1).

183
 2. Specifying sliding surface. On the basis of the Remark 2-3: It is interesting to note that the above
input-otput linearized system (2.4), we can define m sliding formulation include the feedback linearization or dynamic
surfaces si, i = 1, * , m, as follows: inversion as a special case. To see this, let Ai = 0 in (2.9),
then (2.9) is such that j = 0 for t > 0. However, such a goal
si(t) = e(P'-')+ki(Pi le(P' 2) can be achieved by feedback linearization controllers only if
there are no uncertainties in the plant model. By contrast,
+ +ki2e4') +kilei + kio eidt (2.5) even in the presence of certain types of uncertainties, the
where- ei = - ri with ri the reference trajectories; and sliding condition (2.7) can still be maintained by suitably
choosing the values of Ai [SL] It should be also pointed out
kj(P_j- * kio are such that
I
that unless JT pi = n, there is no guarantee for closed-
A" + kd(p,_.)AP-1 + * + ki1A + kio (2.6) loop stability since certain unobsolvable modes called zero-
dynamics cannot be stabilized by the above controller. (I].
is Hulwitz polynomial.
3. Achieving sliding condition. The closed-loop sys- 3. HAVE DASH II Missile Model
tem is said to satisfy the sliding condition if the following
applies. The rigid body motion equations of the HAVE DASH II
misile in a body fixed frame are described by
l < -7i7SiI, ('ii > 0)
2 dt- (2.7)
where in,i = M,rn, are positive numbers. Note that P =-4QR+ L/Its
4:
sliding condition will mak-e si(t) = 0 and si(t) = 0 in a
finite time. Since 4(t) = 0 is a stable differential equation, QI= 12- "PR+M/Iy
satisfaction of si(t) = 0 by ei(t) in turn leads to aymptotic
tracking. A = IV PQ + N/IIz
hz
Let 'ynr
v sin$seceq + cos$seceR
cos$Q - sin$R
S *],Y(i *(n)
(2.8) t P + sin4tanEQ + cossitanER
[ [
y(p)
1
L Jr Vrm = -{cosacoso3(F
m
+ g,)

sgn(s) = [sgn(si),.. ,sgn(sm)]

sin,8(Fy + gy)
+t sinacos3(F2 +gz)}
where sgn(-) is the sign fimetion. Then it can be verified Q - (Pcosa + Rsina)tanl3
that a control law that achieves the sliding condition (2.7) 1
is given by + {-sinar(F, + g) + cosc(F, + gz)}
u = A-'((i - Y(P)) - B - Asgn(s)) (2.9) p = Psina - Rcosa
1
where A = diag [AlaI,Am] with Ai a positive number + mV/rn {-cosaksin3(F, + o£) + cos,3(F, +9gJ)
greater than given positive number *i. Note that
- sinasin (F, +gz)} (3.1)
(s4-y?') = .-P' +kije(,) where
+ +ki2e42) + k1e(1) + kioei
g. = -mgsine
so (As- Y(P) does not depend on u.
9s, = mgcosesin'
Remark 2-1: The integral term in (2.5) can be omit- 9* = mgcosecos4
ted by setting kio = 0. Since the sliding condition also
implies sE(t) = 0, the asymptotical tracking can still be
achieved by the control law (2.9) as long as, for i-= * *m In (3.1), P, Q, and R are the angular rates in the body
fixed frame; V1, ca, and P are the missile velocity, angle
ki(p_ 1) *... kil are such that APi-l+ kj(,p,.Il)A"p2 +'v+kjl of attack, and sideslip angle with respect to the body fixed
are Hulwitz.
frame; and 0, $, P are pitch angle, roll angle, and yaw an-
Remark 2-2: The discontinuity of the sign function will gle. The other parameters are three moments of the inertial
cause chattering in the closed-loop system. In practice, the about the body frame I, I'', and I2,,missile mass m, and
sign finction sgn(si) is often replaced by the saturation gravity constant g. (FX, F,, FZ) and (L, M, N) are aerody-
function sat( .L) where namic forces and rolling, pitching, and yawing moments,
sat(Z) = r, iffrj I respectively, all about the body frame. They are given by
sat(x) = sgn(z), if 1l1 2 1 (2.10) F: = kFPVT2CZ Fs ,=kppV,2Cy
1844
 F, = kFPV2C, L = kmpIQC, The entries of the matrices CFo, CF., CMO, CMU are gener-
M = kMpVI CmI N kMpvmCn (3.2) ally functions of a, /3 and Mm. Using a simple curve fitting
scheme based on the least square error criterion gives
where p is the atmospheric density; kF and km are
constants determined by the missile's geometry; and -0.57 + 0.0083at 0.004 0 0
C:, C,, C,, C,CCm, and C. are aerodynamic coefficients CF O-= -0.21# , CFU1 =
0 0 0.08
given in terms of an aerodynamic table. The functional Czo(a,M6m) J = -0.09 0 0
form of the aerodynamic coefficients Ci, i = r,y, z,l,m,n is 0.116/3 1 - 0 -0.127 0
given as follows CMO = [ Cmo(0, Mm) , Cmu = -0.675 0 0
0.08p3 Mu L 0 0 -0.584
Ci Cio(ctr3,Mm) + Ci 6(o, 66,Mm)
+ Ci66(t ba, Mm) + Ciar& bx, Mm) (3-3) where C,o(a, Mm), and CmO(a, Mm) are given as follows
where- Mm = Vm/c is Mach number with c beimg the speed CzO(aC Mm) = Czl(Ca)+Cz2(C)Mm
of sound. b6, b., and 6, are the effective pitch, roll and Cmo(ca,Mm) = Cml(C()+Cm2(0.)Mm (3.6)
yaw fin control deflections, respectively. They are converted
from the corresponding commanded deflections 6ce, 6ac, 6,rc with
through actuator and fin mixer. The HAVE DASH II nissile
system has four tail fins arranged asymmetrically. The com- Czl(C) = -0.0015a3 + 0.0125ar2 - 0.5052a + 0.0429
manded deflections produce four fin deflection commands bi6 Cz2(a) = 0.0006a3 - 0.0138a2 + 0.1230a - 0.0191
according to Cmi(a) = -0.0055a3 + 0.2131a2- 2.7419t - 0.0381
ble = 6e -6ac + 6r, Cm2(0() = 00014A3 - 0.0623a2 + 0.8715a - 0.4041
62c = -bec 6c +6rc
b3c= -6Cc-6a006rc The other missile parameters are given by G =
32.174 ft/sec2, m = 9.89 slug, c = 968 ft/sec,
64C = CC- 6ac -6r KF = 0.1534 ft2, KM = 0.0959 ft31 p = 5.124 x
These fin deflection commands 6i,,i = 1, *,4 are the in- 10- slugs/ft3, I., - 1.1913 slug ft2, I,, =
puts to the fin actuators which give the actual fin deflections 100.51 slug f12, and 1h, = 100.57 slug ft2, where c and p
6,, i = 1,.* ,4, via a second order system are computed at the altitude of 40, 000 feet.
d bi
dt b- ]
~_r -2 w ]
bi The performance output is chosen to be the following

Y1 = Pcosa +Rsinad! P. -

where (a=0.7, wa = 35Hz. It is assumed that the deflec- Y2 = a

tion limits of the fins are ± 25deg. The deflection rates are (3.7)
limited by ±400 deg/sec. FinaUly, the actual fin deflections Y/3 =
are mixed to produce the effective pitch, roll and yaw fin where P, represents the roll rate with respect to the stability
deflections according to axs. The basic requiremnet for the closed-loop system is
rapid and precise command following ability of P. and a.
Cc= -(6b-2-63+64) Also, the sideslip angle 8 is required to be less than one
4
degree.
6a = -_i(Cl
+ 62 J 3 64)
+
4. Sliding Mode Controller
6,r 41(6i
= + 62-63-64)
To put the missile model in the form described by (2.1), we To make the design easier, we will assume the following
need to approximate the aerodynamic coefficients by well
defined affine functions as follows * The actuator -dynamics are ignored, that is, we assume

[ [
b6e=6, e b=a6bac, and 6, = 670,.
Cc Csc CZa Cr 6e
Cy
CZ I Cyo 1+1 Cie CY,a C,,r
CZo C,c C,a, C,.z J br
6a
J
* The variations of V1m is ignored, that is, we assume
Vm =O.
dy CFO + CF.U (3.4) * The coupling between the force and control deflections
is ignored, that is, we assume CFU = 0.
C,

Cn : I -

L9 CU+ m
Cmo + Cm.
CnO+CnCc Cn-a Cnr,
CM.u
C14 C,r I 6,e
Cma Cmr E 64 Note that the first two assumptions are mainly made for
convenience,
br J topilot design.which has been customarily practiced in au-
The third assumption is based on the ob-
(3.5) servation that the coupling between the force and control
185
 deflections are relatively small in comparison with the cou- for the reason of space) show that the induced sideslip angle
pling -between the moment and control deflections since the is well below 0.2 deg.
aileron, rudder, and elevator are principally moment pro-
ducing devices. [RSJ. It is interesting to make a comparison between the sliding
mode controller and the feedback linearization controller.
Under the above assumptions, the missile system can The feedback linearization controller can be obtained from
be put into the standard form of (2.1) with x - (2.9) by letting Ai = 0 and assigning appropriate values to
[P Q R Vm at p 'P $ e]T, iU = [ie ba r4]T, and kij. Figures 3 and 4 show the performance of the feedbak
y = [P. a /3]T. It can be easily verified that this system linearization controller with klo = 55, k2l = 21, k20 = 225,
has a vector relative degree (1,2,2). In fact, some calcula- k3l = 24, k32 = 350. For a 10 deg angle of attack change,
tions yield the following input-output relation this controller performs almost as good as the sliding mode
controller, however, For a 20 deg angle of attack change, a
PS ] [Bi All A1 A13 -E significant performance degradation in terms of- about 50%
= B2 + A21 A22 A23 ba overshoot in roll rate response is observed.
/3 J L B3 J L A31 As
A32 A
A33 L
A
Or J Acknowledgement
= B(z) + A(z)u (4.1) The authors wish to thank Johnny Evers of USAF Arma-
ment Directorate of Wrigh}t Laboratory for his suggestions
with A(z) givei by and comments.
r cosa ina References
A(r) = kMVPV2 _ cosatantl I sinatanj5
C M8(4.2)
[HLCED] J. Hunag, C.F. Lin, J.R. Cloutier, J.H. Evers, and
C.D. D'Souza, " Robust Feedback Linearization Approach
to Autopilot Design" Proceedings of the First Control Ap-
It can be checked that A(z) is invertible for x in a neighbor- plications Conference September, 1992.
hood of x = 0. Therefore, the sliding mode controller can [HSM] HUNT, L.R., SU R. AND G. MEYER, "Design
be readily synthesized using our general formulation (2.9). for Multi-input Nonlinear Systems," Differential Geomet-
Specifically, the sliding surface is given, according to (2.5) ric Control Theory, R.W. Brocket, R.S. Millman and H.
with k,o = 0, by Sussman edc., Birkhauser, pp. 268-298, 1983.
[I] ISIDORI, A, Nonliner Control Systems, Springer Verlag,
s = S2 ]= 2+ k2le2 J (4.3) 1989.
83 j 3 + kA,le3 [LCEJW] LIN, C.F., J.R. CLOUTIER, J. EVERS, J.
where el = P5-P.0, e2 = a - aC, and e3 = /; and Ps, JUANG AND Q. WANG, "High Performance, Adaptive,
and a, are commanded stability ais roll rate and angle of Robust Bank to Turn Missile Autopilot Design," Proceed-
attack. Then ings of AIAA Guidance, Navigation and Control Confer-
ence, Vol. 3, pp 123 - 137, 1991.
Fs1 -Pa0 [L] LIN, C. F., "Modern Navigation, Guidance, and Control
52- = k21Qi-0)-a
(S-Y(P) = [ 2- a& ] = il(- ) ac ] Processing," Prentice Hall, 1991.
L 3- k31 i
[LL] LIN, C.F. AND S.P. LEE, "Robust Missile Autopilot
The design parameters are given by kA2 = 20, A31 = 8, Design Using a Generalized Singular Optimal Control Tech-
A1 = 100, A2 = 30, A3 = 16. Finally, to alleviate the chat- nique," J. of Guidance, Control and Dynamics, Vol. 8, No.
tering due to the switching function, smoothing technique as 4, pp 498-507, 1985.
described in Remark 2-2 is adopted. That iS, replace 8gn(sj) [RS] ROMANO, J. J. AND S.N. SINGH, 'l-0 Map Inver-
by sat(si/ci) with ei =1.0 E2 = 0.5, E3 = 0.5. Note that all sion, Zero Dynamics and Flight Control" IEEE Trans. on
these parameters are selected largely on the trial-and-error Aerospace and Electronnic Systems, VOL. 26, NO. 6, pp
basis. 1022-1028, 1990.
The performance of this- autopilot for a typical maneu- [SI] SINGH, S.N. AND A. 1YER, "Nonlinear Decoupling
ver is shown in Figures 1 and 2 where all initial conditions Sliding Mode Control and Attitude Control of Spacecraft,"
are zero except that Vm(O) = 2662ft/sec. The miMile is IEEE lrans. von aerospace and Electronic Systems, VOL.
required to rol at 90 degree per second in both moderate 25, No. 5, pp 621-633, 1989.
(10 degree) and large (20 degree) angles of attack. It is
seen that the autopilot is quite capable of performing this [SL] SLOTINE, J.-J. E. AND W. Li, Applied Nonlinear
maneuver. In both cases, the autopilot shows an excellent Control, Prentice Hall, 1991.
tracking ability to angle of attack command. The roll rate [SS] SLOTINE, J.-J E. AND S.S. SASTRY, "TrackingCon-
command following is also satisfactory though a nearly 10% trol of Nonlinear Systems Using Sliding Surfaces with Am
overshoot in the roll rate response is observed when a 20 deg plications to Robot Manipulators," Int. J. Control, 39, 2,
angle of attack change occurs. Other figures (not included 1983.
188
 X 10
"0
w50-
S-I

co
00O 1 2 3 4 5 0 1 2 3 4 5
Time (sec) Time (sec)
Figure Ia Angle of attack response (solid) (10 degree com- Figure lb Roll rate response (solid) of sliding mode con-
mand) of sliding mode controller troller with a 10 degree angle of attack maneuver
-q 30 0-11
100 ,

t20- 50
"0
_S-

'S 10
1X
Cu

V-

1 2 3 4 5
rvN.!--
Time (sec) tme (5c)
Figure 2a Angle of attack response (solid) (20 degree com- Figure 2b Roll rate response (solid) of sliding mode con-
mand) of sliding mode controller troller with a 20 degree angle of attack maneuver
100 1
~50
'~10

5
'S -

o ~~~~~~~~~0
_-5
cuc - 0 1 2 3 4 5
1 2 3 4 5
Time (sec)
Time (sec)
Figure 3a Angle of attack response (solid) (10 degree com- Figure 3b Roll rate response (solid) of feedback lineariza-
nand) of feedback linearization controller tion controller with a 10 degree angle of attack maneuver
-q 30 1501 -l - I
| n
U
a0 100
3, (
(20
"0
-

50
'IO
'S 10 S-I0

5
Time (sec) Time (sec)
Figure 4a Angle of attack response (solid) (20 degree corn- Figure 4b Roll rate response (solid) of feedback lineariza-
mand) of feedback linearization controller tion controller with a 20 degree angle of attack maneuver
187