Académique Documents
Professionnel Documents
Culture Documents
WM13 14:OO
1. Introduction
The design of linear feedback controllers for linear
systems has significantly matured over the past several decades. However, actual controller implementation can often lead to nonlinear system behavior. In
particular, uncertain large amplitude exogenous disturbances can drive system actuators into saturation
forcing the system to operate in a nonlinear region for
which it was not designed. Hence, t o ensure system
stability and performance actuator saturation needs to
be accounted for within the control system design process. There exists an extensive literature devoted to
the control saturation problem (see for example, [l--231
and the references therein). Specifically, [l-3,5,16] concentrate on extensions of classical optimal control theory to account for control amplitude constraints while
anti-windup compensation techniques are developed in
[4,6,8-11,221 to desensitize the effects of actuator saturation. References [7,12,14,15,17,18]provide various
Riccati and Lyapunov equation-based local and semiglobal stabilization methods for systems with actuator
constraints. Alternatively, the authors in [19-21,231 develop saturation controllers based on absolute stability
theory.
Many practical applications of feedback control sysThis research was supported in part by the
NASA/New York Space Grant Consortium under subcontract 32310-5891, the National Science Foundation
under Grant ECS-9496249, and the Air Force Office of
Scientific Research under Grant F49620-96-1-0125.
0-7803-4394-8198
$10.000 1998 IEEE
909
tectable plant with actuator amplitude and rate nonlinearities # ( u ( t ) , t )E @ and $(w(t),t) E @, t
0,
respectively,
>
(4)
4(w(s),s w s ,
(5)
ca
,n
J(Ac,Bc,Cc)
SUP
m Z ( t > &
9(.>.)E 0 0
(6)
where z(t):Elz(t)
EzUu(t) EzUv(t), z ( t ) E Rp,
ETE221= ETEz, = ,OX,
and E,T,E2u = Om,,,
is
minimized.
Note that the input rate rnodel given by (5) represents a software rate limiter that ensures that no rate
commands are sent t o the actuators that are beyond
their specified limits. To characterize the class @ of
time-varying sector bounded memoryless actuator amplitude and rate nonlinearities the following definitions
are needed. Let Mlq, Mzq E W m x m be given diagonal matrices such that MIq = diag(M1, ..., Mlq,),
Mzq = diag(Mz,,, . . . ,Adz,,,,), and Mq A'Mzq - MIq
is positive definite with diagonal entries M q , % , i=
1,.. . , m. Next, define the set of allowable nonlinearities 4(-,.) by
positive-definite matrices
n, = n m; 1 5 nc 5 na; 6 = n,
n,
ETEi, E2T,E2u, E , T , E 2 u > 0
U (control amplitude) or w (control rate)
m x m diagonal matrices
m x m diagonal matrices
diag[Ml,, MI,], diag[Mzu,MzVl
diag[Mu,M u ] diag[Hu,
,
&I
I'
(3)
Nomenclature
B,R'X",R'
S',N',P'
~ ( 0 =) GO,
@ f{4 : W"
xR+ +R"
: Mlqt9, I 4, (9,t ) q ,
>
5 MZq,9?2,
i(t) =
A q t ) - B&(G(t)),
G ( t ) = &(t),
910
(10)
4.
1.
e)
for arbitrary P, Q , P E Rnax n a , diagonal positive defi, and scalar E > 0. Furthernite matrix H E R2m~x2m
more, define ab C such that the input nonlinearity is
time-invariant, that is, $ ( q , t ) = $(q) and $(q) is contained in
for a finite range of its argument q, that
is,
Then, the function V ( 2 )= 2 T p 2 is a Lyapunov function that guarantees that the closed-loop system (1)-(5)
is globally asymptotically stable for all actuator amplitude and rate nonlinearities 4(., -) E a. Furthermore,
the performance functional (6) satisfies the bound
> 0, i
= 1,.
. . , m , are
given.
i E {m + 1,.. . ,2m},
K+ f
Note that J ( 5 0 ,A,, B,, C,) < Z T P Z O = tr P202:,
which has the same form as the H2 cost appearing
in standard LQG theory. Next, we replace 502: by
5 diag[Vl,BCV2BT],where VI E RnaXna
and
E
R r Xare
r arbitrary design weights such that VI 2 0 and
V2 > 0, and proceed by determining controller g@:s
(A,, B,,C,) that minimize J ( p ,A,, B,, Cc) f tr P V ,
where P E Rfixii,P > 0, satisfies (12).
VS
DA
where
Qi(Zi),
v,- 4* i ( i i i ) ,
min {min(K+,y-)},
-z=l,..
. ,2m
{ji. E
Ci
i E (1,. . . ,2m},
R'
: V(Z)
i = m + 1 , . . .,2m},
(15)
C,
a f A + B M l C , and
911
+ +
+
+
+
+
47(q2(t))
P,TRT:P,TL,
= A& + QA; + QCQ -
(18)
7 1 ~ ~ ~ T T ,(19)
A
QP
GTih',
GTr,
rGT = Inc,
n
71 =
n;r E IEg%
In - 7 ,
(20)
x "c
I
(21)
(23)
[ P-GP+ P
(25)
-PGT
GPG~
(28)
Remark 4.3. It is important to note that the estimate of the domain of attraction D A given by (15) for
the closed-loop system is predicated on both cl_osed and
open Lyapunov surfaces. Specifically, since C 7 B # 0,
i E { 1 , . . . ,m}, domain of attraction computations
based on open Lyapunov surfaces lead to extreme algebraic complexity [31]. Thus, for simplicity, for i E
( 1 , . . . , m } , we use closed Lyapunov surfaces for computing the domains of attraction in the presence of
control amplitude nonlinearities. Alternatively, since
C,B= 0, i E { m 1 , . . . , 2 m } , we use open Lyapunov
surfaces for computing the domain of attraction in the
presence of control rate nonlinea'rities. Hence, a combination of closed and open Lyapunov surfaces are used
to compute the estimate of the domain of attraction
V A which yields a considerable improvement over domains of attraction predicated on only closed Lyapunov
surfaces. See [23,31]for a detailed discussion on the distinction between open versus closed Lyapunov surfaces
for estimating domains of attraction.
(24)
Then
F=
>
(22)
A, = r [ A , - QaVFICYT- B,&l(HC,
-Bp R;: Pa]
GT,
B, = r Q ( I V , l ,
c, = -R;;P,G*.
I allt
-T:
sgn(q*(t)),
Iqz(t) I
{L...lflLTL),
Theorem 4.1 provides constructive sufficient conditions that yield dynamic controllers for systems
with actuator amplitude and rate nonlinearities.
When solving (16)- (19) numerically, the matrices
MI,,, Mz,, ,Ml,, Mz,, H,, , and H, appearing in the design equations can be adjusted to examine trade-offs
between performance and allowable sector bounded actuator nonlinearities +(.,.) E or f#(.) E @ b . Furthermore, as in [23], to further reduce conservatism, one can
view the scaling matrices H , and H,, as free parameters
and optimize the performance bound J with respect to
H.
Remark 4.1. A key application of Theorem 4.1 is
the case in which +(q) represents a vector of timeinvariant a c x a t o r amplitude and rate saturation nonlinearities. Specifically, let + ( q ( t ) ) = [ + I ( q l ( t ) ) ,. . . ,
912
0 ~ ~ with
1 , the step input command T = [4.2,-4.2IT.
Note that Z:l%o
= 3.0598 x lo5 so that 20 $! V A .
Figure 2 shows the response of the controlled system
output, in the presence of control amplitude and rate
saturation constraints, with the controller designed using Theorem 4.1 and the controller given by [30]. From
Figure 2 note that although the yaw rate response ~ ( t ) ,
t
0, of Theorem 4.1 design exhibits a larger overshoot as compared to the design given by [30], it provides slightly better settling time and steady state error. Specifically, the steady state value of ~ ( t t) , 0,
for Theorem 4.1 design is -4.1996 while the controller
of [30] gives a steady state value of -4.145. Furthermore, the side slip response ,tJ(t),t 2 0, for Theorem
4.1 design exhibits faster rise time and settling time.
In addition, the steady state value of P ( t ) , t 2 0, for
the design given by Theorem 4.1 and the design given
by [30] is found t o be 4.1606. In both cases, the slight
offset in the steady state behavior is attributed to the
fact that 20 $! V A .
>
>
where
yp =
[@],
6.
-2176
Conclusion
-1093
To design side slip and yaw rate step input tracking controllers we introduce the integrator states ZIas in [30].
Furthermore, as in [30] for design purposes, consider
a pseudo-equivalent feedback interconnection with I 2
subsystem interchanged with the amplitude and rate
saturation nonlinearities. This yields the augmented
dynamic system
References
[l] M. Athans and P. L. Falb, Optimal Control, A n Introduction to the Theory and Its Applications. New York:
McGraw-Hill, 1966.
[2] A. E. Bryson and Y. C. Ho, Applied Optimal Control:
Optimization, Estimation, and Control. New York:
Hemisphere Publishing, 1975.
[3] J. F. Frankena and R. Sivan, A non-linear optimal
control law for linear systems, Znt. Journal of Control,
vol. 30, pp. 457-480, 1979.
[4] N. J. Krikelis, State feedback integral control with 5ntelligent integrators, Int. Journal of Control, vol. 32,
pp. 465-473, 1980.
[5] E. P. Ryan, Optimal feedback control of saturating
systems, Int. Journal of Control, vol. 35, pp. 521-534,
1982.
[6] N. J. Krikelis and S. K . Barkas, Design of tracking
systems subject to actuator saturation and integrator
wind-up, Int. Journal of Control, vol. 39, pp. 667-682,
1984.
[7] P. 0. Gutman and P. Hagander, A new design of constrained controllers for linear systems, ZEEE Tkans.
Automat. Control, vol. 30, pp. 22-33, 1985.
[8] P. Kapasouris, M. Athans, and G. Stein, Design of
where e(t) = yp(t) - T and T is the vector of step reference inputs. Now we interpret equations (29), (30) as
equations (l),(2) and proceed with the tracking controller synthesis. However, first note that the amplitude saturation nonlinearity + ( u ( t ) ) , t 2 0, in (29) is
given by (28) with q = U, i = 2, and a,, = auz = 10
and the rate saturation nonlinearity 4 ( u ( t ) ) ,t
0, is
given by (28) with q = U , i = 2, and a,, = avz = 4.
Next, choosing RI = diag[I3,5000,50000], R2, = I 2 ,
R2, = 1 2 , VI = diag[R1,I2], V2 = I 2 , MlU = 0.99912,
Mi, = 0.99312, M2u = M2, = 1.0I2, H , = 1.2 x 10612,
and H, = 2 x lo5& a full-order (n, = n, = 7) dynamic
compensator was designed using Theorem 4.1 -with
guaranteed domain of attraction V A = { 2 : Z T P Z <
4.1211 x lo2, ICi21 5 4, i = 3,4}.
To illustrate the closed-loop behavior of the system with the controller designed using Theorem 4.1 let
[ x ~ ( 0 ) , e T ( O )=
] [ 0 1 x 3 , - ~ T ] ,4 0 ) = 02x1, and xc, =
>
913
feedback control systems for stable plants with saturating xtuators, in Proc. IEEE Conf. o n Dec. and
Control, (Austin, TX), pp. 46S479, 1988.
[9] K. J. astrom and L. Rundqwist, Integrator windup
and how to avoid it, in Proc. of American Control
Conf., (Pittsburgh, PA), pp. 1693-1698, 1989.
[lo] P. J. Campo, M. Morari, and C. N. Nett, Multivariable anti-windup and bumpless transfer: A general
theory, in Proc. of American Control Conf., (Pittus
burgh, ]?A), pp. 1706-1711, 1989.
[ll] P. Kapilsouris and M. Athans, Control systems with
rate ancl magnitude saturation for neutrally stable open
loop sy~:tems,in Proc. IEEE Conf. on Dec. and Control, (Honolulu, HI), pp. 3404-3409, 1990.
[12] Z. Lin and A. Saberi, Semi-global exponential stabilizaticm of linear systems subject to input saturation, Systems and Control Letters, vol. 21, pp. 225239, 1903.
[13] D. S. Bernstein and W. M. Haddad, Nonlinear controllers for positive real systems with arbitrary input
nonlineitrities, IEEE Trans. Automat. Control, vol. 39,
pp. 151:3-1517, 1994.
[14] Z. Lin, A. Saberi, and A. R. Teel, Simultaneous L,
stabilization and internal stabilization of linear systems
subject to input saturation: State feedback case, in
Proc. IEEE Conf. on Dec. and Control, (Orlando, FL),
pp. 380r3-3813, 1994.
(151 A. R. Teel, Semi-global stabilizability of linear null
controllable systems with input nonlinearities, in
Proc. of American Control Conf., (Baltimore, MD),
pp. 947.-951, 1994.
[16] D. S. Bernstein, Optimal nonlinear, but continuous,
feedback control of systems with saturating actuators,
Int. Journal of Control, vol. 62, pp. 1209-1216, 1995.
[17] Z. Lin and A. Saberi, A semi-global low-and-high gain
design technique for linear systems with input saturation: S3:abilization and disturbance rejection, Int. J.
Robust and Nonlinear Contr., vol. 5, pp. 381-398,1995.
[18] A. Saberi, Z. Lin, and A. R. Teel, Control of linear
systems; with saturating actuators, in Proc. of American Control Conf., (Seattle, WA), pp. 285-289, 1995.
[19] F. Tyan and D. S. Bernstein, Anti-windup compensator synthesis for systems with saturation actuators,
Int. J . Robust and Nonlinear Contr., vol. 5, pp. 521537, 1935.
[20] F. Tyan and D. S. Bernstein, Dynamic output feedback compensation for systems with input saturation,
in Proc:. of American Control Conf., (Seattle, WA),
pp. 3916-3920, 1995.
(211 W. M. Haddad and V. Kapila, Antiwindup controllers
for systems with input nonlinearities, A I A A J. Guid.,
Contr., and Dyn., vol. 19, pp. 1387-1390, 1996.
[22] K. J. h r o m and B. Wittenmark, Computer Controlled
Systems: Theory and Design. Englewood Cliffs, NJ:
Prentice-Hall Inc., 1997.
[23] W. M. 13addad and V. Kapila, Fixed-architecture controller synthesis for systems with input-output timevarying nonlinearities, Int. J. Robust and Nonlinear
Contr., vol. 7, pp. 675-710, 1997.
[24] K. McKay, Summaryof an AGARD workshop on pilot
induced oscillation, in AIAA, (Paper 94-3668), 1994.
(251 R. A. Hess and S. A. Snell, Flight control system design with rate saturating actuators, A I A A J. Guid.,
Contr., and Dyn., vol. 20, pp. 90-96, 1997.
Model
aj o y
,
10
,
12
14
16
18
Time (sec)
914