Feedback Stabilization of Perturbed Nonlinear Discrete-Time Systems

Feedback Stabilization of Perturbed Nonlinear Discrete-Time Systems

Copyright © 1996 IFAC (3th Triennial World Congress, San Francisco, USA 2b-14 5 FEEDBACK STABILIZATION OF PERTURBED NONLINEAR DISCRETE-TIME SYSTEMS ...

328KB Sizes 0 Downloads 42 Views

Copyright © 1996 IFAC (3th Triennial World Congress, San Francisco, USA

2b-14 5

FEEDBACK STABILIZATION OF PERTURBED NONLINEAR DISCRETE-TIME SYSTEMS I Cesar Cruz' and Henk Nijrneijer" • CICESE, Mexico 2 ,.. University of l'wente, The Netherlands 3

Abstract. A robu:-it feedback IStabilization problem of nonlincar discrde-time syst.f!m~ i~

studied, A sufficient condition for feooback ~tabilization of a class of perturbed nonlinear systermi (which are not necessarily feedback linearizable) i~ obtained_ It is l:ihown that. if the Jacobian linearization of the llominalsYl:itcm io stabilizable l then the pert.urbed non linear system ilS asymptotically stabilizablc. A linear feedback that stabilizes the nominal system also locally stabilizes the perturbed system. A simulation example illustrate the obtained results.

Keywords. :\"onlinear discrete-time systems 1 feedback stabilization pert.urbed j

systems,

robllstnes~_

I. INTRODVCTlO:\

The stabilization problem plays a central role in dynamical control systems, A matter of practical importance is t.o determine the extent of aHymptotic stability of an asymptotically stable equilibrium point, Le., how large a perturbation from the equilibrium can be allowed and we can still be sure t hat the traject.ory will return to that equilibrium. In the majority of the applications 1 we encounter perturbed system:;; which are the product of modf:~lillg errors) uncertaintie~ and disturbances. Hence a property desirable of cOlltrollers is t.heir robustness, i.e. 1 their capacity to work suitably even when t.hif'! controlled. system differ::; from lhe system designed. .For continuom; time nonlinear systems several papers have been published where the stability of the closcdloop system i8 analyzed when uncertainties are present (e.g .. Khalil, 1992; Liao, et al., 1992: Marino and Tomei, H193). It. ha.., been proved that. a controller may tolerate uncertainties if the clo.'ieu-loop nominal system has an RSymptoti(:ally stable equilibrium, and t.he perturbation sati'ifiili some olher conditiollf:i. In the discrete-time case, in (Byrnes, et aI., 199:3) it is shown t.hat. any nOIlI Research performed during a visit in Alltumn'94 of Cesar Cruz at the University of 'l'wente, supported by the Faculty of Applied Mathematics. 2 Department of Electronics and Telecommunications, p.a. Box 434944, San Dif'gu, CA 92143, USA. PhUIlp.: +52.(617).4450lx2,~30; fax: +rI2.(617).4.'iISa; e-mail: ccruzOciCCSC.UlX 3 Department of Applied Mathematic!'<, p.a. Box 217, 7500 AE Enschede, The Netherland!'J. Phone: +31.53.893442; fax: 31.53.340733; e-mail: [email protected]

linear system with Lyapunov-stable unforced dynamics can always be globally stabilized by smooth state feedback if suitable cont.rollability-like rank conditions are satisfied. In particular, in (Cruz and Alvarez, 1994) the authors have di-.;cm:;soo the stability of a class of nonlinear fef!dback linearizablf' :
2424

2. PROBLEM STATEMENT

ConlSider the class of discret.e-time lIonlinear systems d epending on a real-valued parameter vector c E n q

1:(k I

I) ~.

(1 )

f(x (k) . lI(k ),£) ,

where x :..:; (X t , ___ ,x n ) and Tt -- (u! , ... , u m ) are smooth 10(:a1 coordinat.es for the state spa(X~ Ai and input i:ipace U respe(~t ive ly, \Ve 3s;;urne that f i:-; a analytic vcct.orfield ill J: , 'It and 1.;:, anrl that f(ll, 0, E)

(2)

0, '1£.

c

Problem 1. (Robust feedback stabilization of nonlinear control systems): Con.id,,· the .• y.• tem (1 ) .• atisfying (2). Under which mnditions does there exist" smoolll. static staJ.e feedba ck

/(x ,u) c' f(x , u,a). the system (1) corresponding to the value e ::: 0 is called the unpertur-bed or nominal sy&tern, while it will be called lite pC7'turbed syst em whe n e .f O. T he verturbed sys t.em could result from modcling errors , l:lg iug of (Jh'y~i(:81 cumponents, uncertainties and d~t.urUaIl(: es which exit-I. ill any TCH listic problem .

Let us int. rodu<.:c the notation B(O , R) to indicate the set of v"d oTS 1: satisfying 11 x - (I 11 < R ; that. is, it. is the sct. of all point.~ who...e Ellclidt'..an dist ance fro m 0 ie,: less t han R .

u(k) -- (» (x(k)) ,

where () , M loop system

+ 1)

(3)

J(J:(k)),

(3) is stable if there ., an Ro >

a for which the following

is true: For e very R < Ro. there is an r, 0 < r < R, such t.ha t if x(O ) i, in,ide R(O, r ), t.hen x (k ) i, inside R(O, R) for a ll k 2: a, (a) is aBymptotically .table if it stable aud B(O . r ) (;811 he chOljcn such that IIx (O) 1I


=>

lim x (k) =

k -- .x,

a.

Defin;tion f. (Luenuerger, 1979) A function v defined un Ii region n of the stat e space of (3) and containing t he equilibrium point. ;1: . - 0 i~ a Lyapunov function for the disc rete· tirnp. r.;ystem (8) if it sati.<;fies t.h p. foll owing three requiremenb;: (1) 1.1 is (:o ntinuous. (2) 1! (3:') has R unique minimum at O. (3) The function L'!.v(x(k)) . 11 (f(x(k))) - v(x (k)) sa!.isne. L'!.1'{x(k)) SO, for all J: E n. Theo rem 1. (Asymptotic stability) (Luenberger,1979) H t here ex',ts 8 L.Yapunov function v(x) in B(a , Ro) with Ce nt.er the o rigill , s uch that the function .6.v{x) is strict ly negative at every point (except O) ~ t hen the origin or t.he non linear syste m (:i) is asymptotically stable.

Rm, w;U, 0'(0) - 0 , .•".oh that the closed-

x(k

+ I)

-, f(r.(k) ,cr(x (k)),e:)

(r.)

has 0 a.s a locally asympto tically stable equilibrium for all c in a neighborhood of l: ;;;;;; 0 ? Note that 0' is selected ind~pende nt. of c. Let us write the ~y8tcm (5) in the form

x(k

+ I) =

/( x (k) , n (x(k)))

+ g(x(k) , o(x (k)) , E) , (6)

where g(x , n(x) ,E) = f( x , <>(x) , €) - /(x , n(x ») is the perturbation function . \Ve make the following a....sllmption :

(A) : The linearizat;on of Ihe nominal system x(k + 1) -- j(x(k ), tI(k))

Dejinition 1. (Luenberger, 1979) Let x = 0 be an equilibrium point of tlw nonlinear syst.em x(k

~

around the origin} is

(7)

stabili ~ able.

If We define Ao ~ ~(a,O) and Ro = ~(a, a), thell the Assumption (A) guarantees t he existe nce of a matrix FE R.mxn such t.hat all eigt mvalue~ of (.1\0 + BoF) have modulu:o> st.ric:tiy less than one .

Proposition £. Consider t he nomina) syst e m (7). A..." umc Assumption (A) holds and 1<1. F he such that (An I RoF) is asymptotically st.a ble, t hen the c1ose
+ I)

= j(x( k ), Fx(k) ).

has 0 as a locally asymptot ically

Proof.

stabl~

(8)

e;quilibriurn.

Suppo,e Assumption (A) hoIck It follows that

for any positive df!finitc ITwtrix Q = qT , t,he solution p .;.;; pT of the discrct.c-timt, ~ Lyapunov matrix equation

(Ao + BoF)Tp(Ao 1 RoF) - P ~ -Q , is a posit.ive definite matrix see, e.g. : (Vidyli.'Sagar, 1993) . \Ve will ~how that v(~: ) =c x T Px is also M Lyapunov funet.ion for the c1",cd-Ioop system (8) in a lIeighborhood of /(a, a) ~ o. To thl'i end, we evaluat.e t.1I(x(k)) = IAox(k)

Our goal in this not.c is to study the following problem.

+ Ro,,(k) + j

(x(k ), u(k))

-A"x(k) - Bou(k)rr .

2425

.PIA"x(k) I R"vik:1 = I(A" I B"F) x( k) I

,PI(Ao

-

I,

+ j(x(k), v(k)) (x( k) , Fx(klll

T

Suppose v(x) is we ll-defined in B(O, ~). Let R be arbitrary with 0 < R < ~. Let R, < R be selected so that if x E B(O , R,) the n i(x , Fx) E H(O . ~) (Le. , if the state lies illtiide 8(0, RI) , it will not jump o ut of B(O , R,,) in one step). The continuity of t.he funct.ion I gllarant fCi'j the existence of RI'

.

+ BoF)x(k) + I, (x(k) , Fx(k»)J

-3:I'(kjPx(k) where

It ...

Fx. I,- (x. Fx)

6

fix . F.r) - (Ao

.cc

+ BoF ) x,

Let m ,.,. minRI <; lI r U:=; nu V(3:), the exist.ence oft-he minimum m is guaranteed from t.he continuity of v(x) aud the compactnpJ;.'; of the region defined uy R, :Sl\ x I\ :S Ra . Moreover m > v(O) :;ince ll has a unique minimum at O. From t.he conti nuity tJ[ v(x), it. is po.':iSible to select an r, 0 < r < R, such that for x E B(O, r) it holds v(x) < m . This is be(:allSf! near the origin v mu...,'t take value.s cIOtiC to viOl = o. If x(O) E H(O ,r), '.he n v(x(O» < m . Sillce t. v(x) < 0 t he value of" decrea,es wit h time. Therefore, the corresponding trajectory can never go out~idf! 8(0 , RI), and consequently it can ne ver go outside R(O, R), i.c ..

t.v( 3:(k» .o -xT(k)Qx(k)

t2x T (k) (Ao I

with

+ BuF}7' pi,

(x(k) , Fx(k»)

iT (x(!:) . F.r(k») pi, (x(k). F.r(k)) ,

i, (x , Fx) satisfies

Ili, (x(!:), Fx(k))11 <", Ilx(k}11 + c~ IIFx(k)11 :S c,II:"(k)11 + c,c3 1Ix(k)ll, so

\I I(x(!:) , Fx(k»

\I

- Aox(k) - BoFx(k) II x(k) II

Ilx(O)1\ < r ""

I\x(/')I\ < R, < R ,

'I k ~ O.

a.

< c, + C2 03,

Finally, it i< suffic ient to show that v(x (k» ~ 0 k~ Since t.v(x) < 0 for all:c except at origin, t hen vex) must act.ually deCrCA!;f! rnoU4)tonousiy. Thus v(x(k» COIIverges t.o some limiting value m. The only que.c;t.ion i~ whether it is possible fur .ft > viOl = O. This i, nut poosible since v(x(k» eOnVi!rges to tnl it must be t rue that t.v(x(k) converges to zero. But t.v(x) < 0 except at origin. Thus, since L\v(x(k» is continuous x(k) m1L~·t converge to origin and v(x(k)) must (:onverge to v(O) . By Theorem 1, we (,;onclude that the origin is (locally) asymptot.ically stable equilibrium for the nomiutl) sy:-;tem (8). 00.

for all x satisfying I\x(k)1\ < r

t.v(x(k)):S

- IIQllllx(k)II' t 2I1xT(k) IIII(Ao + BoFfll· ·IIPIIIII,(x(k). FT.(k»11 + lifT (x(k). FX(kll lIIIPl\lli, (x(k) , Fx(k))ll ·

Now Jet

- xT(k)Qx(k) :S -).,»>,,,(Q)

II(Au

I

RoFl"'I l ~ cs. I\PI\ -

Ilx(k)1I

2

).,m •• W)

=

,

-c,llx(k)11 ,

= C6,

C, =

2cscn.

Hel1l.:e

t.v(x(k))

< -c, l\x(k) I\' + c, (c, + C,C3) I\x(k) 1\' +C6 (c, + c,c3)' l\x(k) I\' =

I- c, + e7 (c , +c. (c, t

t C,C3:1

c,c3)'lllx(k)II' .

choosing

The following theorem shows how t he origin of (1) will remain asymptotically sta ble fOT !;mall valuC2:0 of parameter g 'I 0; that is the robust stabilizatioll Problem 1 is solvahle for small e. Theorem 3. Consider t he class of discrete-time nonlinear perturbed systems ( 1) . Assume that Assumption (A) hold,. Then the perturhed system (5) has 00,<; a locall y a.~ymptotic:Rll'y stable ('lCluilibrium for c sufficie:ntly small.

Proof. Assume t hat Assumpt.ion (A) holds. We will show that v(x) ;;... x T Px is also a Lyapunov function for the closoo-loop perturbed s~'stem

x(k ensures that t.v (x(k))

< 0, for all IIx(k)1\ < r.

From standard Lyapunov Theorems see, e.g. } (Luenberger, 1979) , we have t hat all traj ector ies starting inside 8(0, r) will ('on verge to the origi n.

+ 1) =

I(x(k) , FX( k),e:)

(with u - Fx , where F such that (Ao + BuF) is asymptotically siable) in a neighborhood of (O,O,c), i.e .,

x(k

2426

+ 1) '- (A. + B,F)x(k) + ft(x(k) , PX(k},E),

IIp(x(kllll S (C8 f.(3:,1-'X.<:} ~ fix , FT, c) - (A.

+ B. F)x,

(9)

+ C9) IIx(k)II 2 + (c6 1 CIO ) IIx(k)II Ill,

(x(k) , Fx(k) ,E)11

+c. IIf, (3(k) , Fx(k)")112 < C" IIx(k)II 2 +C'2 II,,(k) 11 [Cl

with

8f

A, - ,,(O,O,e), vX

+ R.F 0

A,

(Ao

{)f B, - Ou (0, 0, e),

+ A (e»

1 (80

+ B(e»F.

+ B,Fj'"P(A + B.F) -

1>v(x(k» - xT(k)l(A,

Plx (k)

+2x'" (k) (A, +- H,F)" P j,(x(k) , Fx(k), c)

+Jl" (x(k), Px(k),c) Pf, (x(k), Fx(k),e) , i(x,t,,) +- g(x,Px,e) where Let f(x , F x,e ) g(x, FX ,e) ~ f(x , Fx ,e) - I(x,Fx), thu8, we write (9) in the form I, (x , Fx . c) ~ J (x, Fx) - (110 I BoP) x - (.4 (e) + H (c) F) x.

~. C"

for all IIx(k)II

+ C3 (C2 + 1'2"")IIlx(k)II , 2 +C6 [Cl -1- (" + C3 (C2 + c21)1 II x(k)II 2 2 IIx(k)1I +- c" (k, + k21) IIx(k)II 2 2 ~c, (k, + kn) II x(k) " ,

< c"

I C,

where k, = c, '1 '" I C,C3, k, = C2e",

IIp(x(kllll < (k3 +- k41

+ k,1') IIx(k)II' ,

\I IIx(k) 11

< T"

wit.h

1>1'(x(k)) S -x" (k)Qx(k)

+ (k3 + k,"") + k.1') IIx(k) 11' ,

I g (x, Fx ,.) Choose

Now IIJ(x(k)' Fx(k» - (Ao we

choo~e 1:\11

+ BoF)x(k) 11 < (c, + C2C3) IIx(k)II ,

arbitrary

,0

> U and assume that

IIg(x(k), FJ:(k),£) - (A(e) I B(e)FJx(k)11 < c, IIx(k)11 I cno IIFx(k)II (10)

S (c, holds for some "'Y

:5

')'0-

I

c2c:n'<>!lIx(k)11

So

11 f, (x(k). Fx(k) , e} I I H ' , + C3(C, + C2~O) , 11 x (k) 11 < c, , for all x satisfying

IIx(k) II < 1\OW W~

,

'2 = mm

(

'"

(

c. ~ C2

.,) + C2"YO

).

have t hat

1>1'(x(k»

= -x"l"(k)Qx(k) + p(x(k»,

then 1>v(x(k)) S -x T (k)Q3:(k), for all IIx(k)1I < T3, if g(x, Fx, e) - (A(e) + B(e) F),""sti:;fics (10) with"") S "")1. In this point, we repeat the :-isme tlrguments of the pre-

vious proof, i.e., we show t,hat all trajectories starting inside B(O, r3) converge to the origin provided (10) iH sati"ified with 1 ~ ')'1 . Remark 1. If one consiuer/') I he closed-loop syst.em

x(k

+ 1) ~ I

(3'(k) , Fx(k),e)

and linearize6 it at. x ~ 0 for



+ 1)

=

(11)

0, we obtain the

- (A" 'f BoF) x(k). Clearly th" linf".arization of (11) about x .,-:- 0 for small g will have it~<;

linearization x(k

eigenvalues clooe t.o the eigfmvalues of (Ao + HoP). Since all the eigenvalues of (AD + lioF) have modulus strictly

less than one the same hold,~ true for the linearizat.ioll of (11) for sufficiently small ,.

wh ere

+ BaF)'" prAte) + B(e)F) x +xT (A(E) + B(e)P)'" l' (A(e) + B(e) F) x

3, EXA'vIPLE

p(x) = 2xT (Aa

12xT (Aol BoF)T Ph(x , Fx , e)

+2XT (A( e)

Consider the following Ilonlinear system wliich is not feedback linearizahlc

+ B(e)F)'" PJ.{x , Fx ,.)

x,(H I) =x3(k)(a, 1 x,(k» ,

I fnx , Fx , e:)P/, (x. FX, E),

x2(k

+ 1) = x, (k) + (a, + xz(k)) u(k) ,

x3(k + 1) ~x2(k) (a3 ;. x,(k» - x3(k)u(k),

and 1'H:ttisfiet;

2427

t.he or igin is an equilibrium point of the nominall'iYl'it.em, i.e., j(O , O) __ O. T he linearization a round U is z(k

+

E,

..

E,

,.

I) - Aoz(k) f- Bou(k) ,

.

• •

with



.

.,

~

.

,

":, (k

+ I)

+ x , (k))·-

t l) =

A,z(k)

••

~.

"-.- '

• ( • .bl

."f; E,

[ (0.006

+ l.00(je,)(1

+ e,) [< 1 - (0.07 + 0,07"2), (12)

+ 1.00602 )(1

I E,))2 ,( 13)

]n Fig. 1.a we t;how the guarant.eed stabilit.Y region given by the conditions (12) and (1 3).

B ,u(k) ,

Remark 2. If onc would change t.h e original F , then aL':'O thL,", region in t he (El,e2)-plane may change; other F's could lead to more or less c' )J1servat ivp. h oun<.lli, i.e., the size or the region iu l ite (e"1~ E2)-space can be change, sce Fig.1.b, we choose F - I- I,O ,O[ such that the p oh have b een fixed gt t he origin, in this C&ie, t he values of E l , c2 sat,is fy t he conditions :

with

t.h ~ closerl~loop

. ..

,

[-(0.07+ 0.07£, )[< 1- «0.006

x3(k)u(k) ,

The linearization around 0 is

*



11 . _1

has all it.~ root~ wit h modulus strictly less t han one if and only if El and £2 :-;atisf.\· t he inequali t ie8

+ x,(k)) + E, xo(k), + (az + x.(k)) u(k ) + E,u(k) ,

x3(k I 1) ~.. x,( k) (a3

" ."

_:'I n

Fig. I. Guaranteed stability regions given by: (La) t he ineq ualities (12) ann (13), and (1. b) the inequalities (14)

- xo(k) (a,

x.(k I I) =, x ,(k)

and

.

.

.,.

" where [R", A"R", A~B"J ha., full ran k (n = 3) for lhe nominal parameter values a] :-: ; u" '''' a:'1 = 1, so (AO l Bo) is .tabilizahle, If we ch.,.,.., u ~ F z, wilh F z ~ -1.006z, + O.07Z3i then t.he closed-loop poJ('f) have been fixed a t 0.1, 0.2 a nd - 0.3. Let us ~llPp~ C that a·1 and a2 are perturbed by £ 1 = D.a] ano £2 = Aa:,? r{'~.,pectively. Then t.he pert,llrherl :'iy~tem is givt'n by

•• '" '" ,

perturbed :;ystem is given by

z(k + I) - (A , + R, F) z(k) = (Ao t BoF):(k) t (A (E) + B(e)F)z(k), where

I (1 + e,Je, 1< 1,

J - [(I

,

+ "de, I > O.

(14)

The following numerical re<.; ults are obtained with t he first choice for F .

In Fig .2, we show t he response of the closed-loop IiAo

A(E)

+

1 ] BoF - 0,006 0 0.07 , [ o I 0

+ B(E)F =

0

0

e, ]

0 0 - 1.006e, OO.07e, [ 000

'

By T heorem 3 the perturbed ~yste m has a asymptotic.ally stable closed-loop linf'..ari7.at ion for El. £2 sufficiently s mall. 111 the present setting (Af:'" R~F) will be a stahle matrix, if the following hold... : from the SchurCohu Criterion (LaSalle, 1986) ~ the characteristic polynomial

.\', - (0.07

+ 0.07C2)' + (O.OO(j

I L006o,)(1

nearizctl I)y~t.em under t.he values of Cl = 10 and € :2 ,,:: 0.05, wlwn the states are ~tarting at Zl = Z2 ...." 0.4 a nd Z3 = 0.5. Fig.3.a exhibit s t:he behavior of t,he nonlin ear syste m controlled by the linear feedback under the :;aUle perturbat.ionl';, but. with initial condit ions: XI X2 = 0.01 and X3 = 0.1. From Figs.2 and 3.a, we can see t.he obvioUb fact that. t,h(' stability region of the nOHlinear syste m constitutel'l f\ !-imall region in t.he stability region or t he linearized syst.em. Finally~ F ig.3.b shows the corm:-;ponding respons~ of lhe closed-loop linearized :;ystem exactly wit. h the same r:olldit ioIlS of the experiment shown in Fig.:l.a to illustrat.r. t hat in this case, both r ~ ponses are very ~ im i l a r .

+ e,) ,

2428

4,

-::~

"

o

5

'0

'5

-: :: ~

"

o

!i

10

15

20

25

38

- ~--

25

---- =1

3D

35

.0

Time (k:I

L

,3 D,5

-0,: 0f ~ 10 5

15

25

20

I

o

5

15

'0

'"

25

]n this note we have presented all study of the robU/'j l. feedback stabilization proble m for a class o f perturbed non linear dlli<:rete·time. \V.: have I)hown that. the COIltroller ils ruLm;l if the li nearir.at.ion oft·he nominal system is st.abilizaLle. We haVf~ remarked that t.hi."i robustnes.o;; property depends of the choice of the feedbac k applied. that is, the si:u~ of the ... tability reg ion may be cha nge. \Ve have generali:r.ed the rel'ults given in (Crnz and Alvarez, 1994) for the cla....."i 01' lIlultivaria ble systemH that are not necesHarily feedback lin carizablc.

.0 35 nm'(I<1

3D

u,:~ -'0

I

•0

Timeltl

20

35

30

3S

I

.0

REFERENCES Byrncs, C,I. , W, Lin and H K. Ghosh (1993), Stabilizat.ion of di.-;cret.e-time nnnlinr.ar system:; by smoot.h state feedback , System.' & Control',elters, 21 , pp,-

Ome (ij

Fig. 2 . .H.csponse of the linearized ~i.)'stem under t he perturbs!,ions s, ~ 10 and 02 ~ 0,05; wilh . , (0) = "2(0) -- 0 ,4 and "3 (0) - 0,5_

CO~CLCSJONS

25,,)-263,

Cruz, C. and J . Alvarez (1994) _ Stability Analysis uf Li ncarizing Controllers f(,r a C lass of Perturbed Nonlinear Discrete-time SY:"itcms, Proe. 1994 American Control Conference , Balt.imore, .YlD , CSA, pp,5.58,562.

Xl_:fF~_3_ o

211

.110 Time

x2

~

0

11

20

40

fkl

o,;ftF'_____ -0.1

":F------~

liD

,2

20

60

Time IkI

o:~r-~----11 -<1.1

!;=o---=,,;:---'."o--'.. ,;! Time IkI

zJ

';F....--o-11 -G.I 0

20

40

'0 Time M

_:Ew------1 _:F-----11 u

u

0204060 13.a)

Time (kJ

02040&0 Time fleJ

!l_bl

Khalil, 1l.K. (1992), No nlin. ar systems, Macmillan I'll bli~hing Company. LaSalle, J,P, (19S6),The Stability and Control of Discrete Processes, Applied 1\1athematical Sciences, Vol. 62, Springf:r- Verlag. Liao, T., L, Fll and Ch, H
Luenberger, I),G. (1979), Introduction to Dynamic SystentS, Theory, Models, & Applications, Johll Wiley & Sons, Marillo ~ R . and P . Tomei (lH93) . Robust st.abi li:r.ation of feedback linearizable t. un e· var ying uncertain 11011linear systems, Automniica, Vo1.29~ No. i , pp.181189. Vidyasagar, M, (199:]). Non/inea,' systems analy,';... Se c.ond F...d., Prentice Hall.

Fig. 3. (3.a) RespOIl8e of the nonlineRr system under the perr.urbations c, - 10 and "2 " 0.05; with x,(O) ~ xAO) ,~ om and X3(0). (3,b) RCBponse of the lincarized ~y.stf!m ~xact. ly under the same conditions.

Remark ,'1. T1H.~ approximal,ioIl givf'n in (Cruz ann AIvarez 1 1994) is for nonlinear feedback linearizable systeJTIN only. it is clear that such a strategy is not possible to apply to the dA...';S of non linear s'ystem ao;; in this ex· ampl~ which it; not feedback lineari:t.ablc tiystem.

2429