Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Copyright © IFAC Nonlinear Control Systems, ELSEVIER

Stuttgart, Germany, 2004


IFAC
PUBLICATIONS
www.elsevier.com/locate/ifac

STABILIZATION OF A CASCADE-CONNECTED
NONLINEAR SYSTEM WITH AN INPUT
CONSTRAINT

Hisakazu Nakamura' Nami Kidane' Yuh Yamashita··


Hirokazu Nishitani •

• Graduate School of Information Science


Nara Instititue of Science and Technology
Email: hisaka-n@is.aist-nara.ac.jp.nami-ki@is.aist-nara.ac.jp.
nisitani@is.aist-nara.ac.jp
•• Graduate School of Information Science and Technology
Hokkaido University
Email: yuhyama@ist.hokudai.ac.jp

Abstract: This paper forcuses on the problem of stabilizing a cascade-connected


nonlinear system with an input constraint. We provide a construction method using
a Lyapunov function for a system that has an input-to-state stable subsystem and an
unstable subsystem and then propose a controller based on the Lyapunov function.
Finally, we confirm the effectiveness of the proposed method by computer simulation.
Copyright © 2004 IFAC
Keywords: constraints, cascade control, nonlinear control systems, stabilization
method, Lyapunov function, state feedback

1. INTRODUCTION 2. PRELIMINARIES

In this section, we summarize some important


definitions and properties.
Actual systems have input constraints in order
to restrict their actuators or protect the systems. 2.1 Class K function
A few controllers have been proposed for non-
linear systems with input constraints (Lin and Definition 1. (Class K function).
Sontag, 1995) (Lin and Sontag, 1991) (Malisoff A continuous function 0 : [0, a) -> lR+ is said to
and Sontag, 2000). belong to class K if it is strictly increasing and
However, there have been no research on stabiliza- 0(0) = O.
tion of a cascade-connected nonlinear system with If a = 00 and lim r _ oo o(r) = 00, the function is
an input constraint. In this paper, we consider said to belong to class K 00'
a nonlinear cascade-connected system that has
an input-to-state stable (ISS) subsystem and an Definition 2. (Class K L function).
unstable subsystem. We provide a construction A continuous function (3 : [0, a) x [0,00) -> [0,00)
method using a Lyapunov function for the system is said to belong to class K L if, for each fixed s,
with an input constraint, and propose a controller the function
for the system based on the Lyapunov function.
0: [0,1) -> [0,00)
Moreover, we confirm the performance of the pro- (1)
posed method by computer simulation. rr--+(3(r,s)

1207
belongs to class K and, for fixed r, the function Definition 7. (ISS-Lyapunov function). A smooth
function V : Rn -> R+ is called an ISS-Lyapunov
4J: [0,00) -> [0,00)
(2) function for system (6) if there exist Koo-functions
S f--> {J(r, s)
Ql' 01, and K functions a2 and ~, such that
is decreasing and lim 5 -+ oo 4J(s) = o. (8)
for any x E Rn and
2.2 Control Lyapunov juntion and stability of the 8V
8x f(x, u) ~ -a2(1!xlJ) (9)
system
for any x E Rn and any u E R m so that IIxll >
In this subsection, we consider the following input- ~(llull)·
affine system.
Lemma 1. A smooth function V is an ISS-Lyapunov
:i; = f(x) + g(x) . u, (3)
function for system (6) if and only if there exist
where x E Rn is state, u E D ~ Rm is input, K oo functions Ql' 01, a2, a3 such that eq. (8) holds
f : Rn -> Rn, 9 : Rm -> Rn are continuous and
mapping, and f(O) = o. 8V
8x f(x, u) ~ -a2(lIxlJ) + a3(lIulJ)· (10)
Definition 3. (Control Lyapunov funtion).
A smooth radially unbounded positive-definite Theorem 1. The system is ISS if and only if it
function V : Rn -> R+ is a control Lyapunov admits an ISS-Lyapunov function.
function if the following inequality is satisfied:
in( {LfV + LgV· u} < 0, "Ix =I O. (4) 3. PROBLEM STATEMENT
uEU
In this paper, we consider the following cascade-
Definition 4. (Local control Lyapunov function).
connected input-affine system:
A smooth radially-unbounded positive-definite
function V : Rn -> R+ is said to be a Local control El : :i; = f(x) + 'lj;(x, z) + g(x, z)u
(11)
Lyapunov function for a domain of attraction D E 2 : i = a(z) + h(z)u,
if the following inequality is satisfied: where x ERn!, Z E Rn, is state, u E U 2 ~ R ffi
in( {LfV + LgV· u} < 0, "Ix E D\{O}, (5) is an input, f : Rn! -> Rn!, 'lj; : R n \ X !Rn\ -> !Rn \,
uEU and h : !Rn, -> !Rn, are continuous mapping that
where D is an open subset of Rn that satisfies satisfies f(O) = 0, 'lj;(0, 0) = 0 and a(O) = 0, and
OE D. 9 : !Rn, X !Rn, -> !Rn! X !Rm , h : !Rn, -> !Rn, X !Rm
is a continuous function matrix.
Definition 5. (Small control property (scp)).
The objective of this paper is to stabilize the
For each E, the Lyapunov function V has a small
system (11) with the following input constraint:
control property if there exists a 8 > 0 such that
if 0 =lllxll < 8, then there is some u with lIull < E U2={uE!Rm!lIull<I}. (12)
such that LfV + LgV· u < O.
We assume the subsystem
El : :i; = f(x) + 'lj;(x, z) + g(x, z)u (13)
2.3 Input-to-state-stability
is ISS, 'lj;(x,O) = 0, and the following ISS-
Lyapunov function W for system (13) is known:
We consider the following system in this subsec-
tion: ~~h (11x1D ~ W(x) ~ 01 (1Ixll), (14)
:i; = f(x, u), (6) W(x) ~ -a2(lIxID + a(lI(z, u)II), (15)

where f : Rn X Rm -> Rn satisfies f(O, 0) = O.


where the functions Ql,01,a2,a are in K oo .
Moreover, we assume for the other subsystem
Definition 6. (Input-to-state-stability (ISS)).
System (6) is input-to-state stable if there exist
E2 : i = a(z) + h(z)u, (16)
a class KL function {J(.,,) and a class K function we have a control Lyapunov function that satis-
')'('), called a gain function, such that for any input fies the small control property and the following
u(·) E L~ and any Xo E !Rn, the response of x(t) inequality:
of system (6) in the initial state x(O) = Xo satisfies
IIx(t)1I ~ {J(lI x oll, t) + ')'(lIuOll oo ) (7)
Since El is ISS, the following proposition is held
for all t 2: O. obviously.

1208
Proposition 1. If we can stabilize subsystem ~2 Proof. Byeq. (22) and U(z), we can obtain the
by small input in the neighborhood of the origin, following equation:
we can stabilize system (11).
U = LaU _ LaU + V(L aU)2 + IIL h UI1 4 (23)
1 + Vi + IILhUII2
In the domain of attraction, LaU < IILhUII.
4. STABILIZATIO OF SUBSYSTEM ~2 Therefore,

In this section, we consider the problem of sta-


(24)
bilizing subsystem ~2 and construct a controller. in D. Because inputs Ui are continuous and U < 0,
We use the controller proposed in (Lin and Son- subsystem (16) is asymptotically stable.
tag, 1991) to stabilize subsystem ~2. For subsys-
tem ~2, it is clear that the following propositions
are held. Therefore, we can obtain the following lemma.

Proposition 2. Consider system (16) with input Lemma 2. If we can stabilize system (16) by input
constraint (12). The following input minimizes the (22), there exists a positive-definite continuous
derivative of the given CLF U: function wO that satisfies the following inequal-
ity:
(18) U ~ -w(U(x)), (25)
and that satisfies the following equation:
Furthermore the minimum of U is obtained by the
following equation: lim w(r) = O. (26)
r->O

(19) Moreover, there exists a class K oo function 1) such


that
Proposition 3. Consider system (16) with input (27)
constraint lIulI < 1. Assume there exists a such
that for all U(z) < a,
Consequently, we can obtain the following theo-
LaU rem, which is one of the main results of this paper.
P2 ()
Z = IILhUII < 1. (20)

Then, the set Theorem 3. Consider system (16) with input con-
straint (11). Input (22) stabilizes the system
D = {zIU(z) < a} (21) asymptotically in the domain of attraction Rnl x
D.
is a domain of attraction.

The proof is clear by theorem 2 and the property


By using these propositions, we can obtain the of ISS.
modified version of Theorem 1 in (Lin and Sontag,
1991).
5. MODIFICATION OF U(Z)
Theorem 2. Consider system (16) with input con-
straint (12). Let U(z) be a local clf for the system
In the previous section, we have constructed input
and D be the domain of attraction defined by eq.
U to asymptotically stabilize the subsystem ~2.
(21).
However, the function w(·) is not a class K func-
Then, the input tion. Hence, in this section, we modify function U
to a function such that its derivative is class K.
LaU + j(L aU)2 + IIL hUI1 4
We choose a class K function 4J0 such that
Ui = - 1+ Vi+ II L h U II 2
. 4J(r)
(LhU '" 0) hm - () < 00. (28)
r-O W r
Ui = 0
(LhU = 0) Consider a function 71'0 such that
(i=l, ... ,m) [X 4J(r)
(22) 71'(x) = io w(r) dr (0 ~ r < a). (29)

asymptotically stabilizes the origin of the system Based on eq. (28) and positive definition of func-
in the domain of attraction D. Moreover, the tion w('), we can always say 0 ~ ~ < 00.
input is continuous. Therefore, the function 71'(.) is well-defined.

1209
Consider the following function U: Consider the following function W:
• lJ(z) W(x) = O(W(x», (40)
lJ(z) = 1r(lJ(z)) + a -lJ(z)· (30)

This U satisfies the following properties: where the derivative of W is obtained as follows:
• U(z) = 0 (z = 0) W = 17(W(X» . [-Q2(llxID + Ci(llzID]. (41)
• U(z) ---+ 00 (lJ(z) ---+ a).
Therefore, we can obtain the following theorem.
Then,
~ cP(lJ(z». a . Theorem 4. The following function V is a Lya-
lJ = w(lJ(z» lJ + (a _ lJ(Z»2 lJ punov function for system (32):
a .
~ -cP(lJ(z)) + (a -lJ(z))2lJ. (31) V(x, z) = W(x) + U(z) (42)

Proof·
6. CONSTRUCTION OF A LYAPUNOV V(x, z) ~ 17(W(X» . [-Q2(lIxll) + Ci(lIzll)]
FUNCTION FOR A a . (43)
CASCADED-CONNECTED SYSTEM - cP(lJ(z» + (a _lJ(z»2 lJ
In the previous section, we have obtained a Lya- In the case Ilxll > x(llzll), it is clear that V(x, z) <
punov function such that its derivative is class K. o. All we have to do is to prove V < 0 in the case
However, the sum of the Lyapunov functions for Ilxll < x(lIzll)·
subsystems W + U is not a Lyapunov function for (i) Case: Ci(x) < cP 0 ~(x) when x ---+ 0
the entire cascade-connected system (11) as usual.
Accordingly, we construct the Lyapunov function
for system (11) in this section. V(x, z) ~ -17(W(X» . Q2(llxll)
When we use eq. (22) for inputs, the system can + inf cP o. ~3)(r) . Ci 0 gJ"1 0 lJ(lIzll)
be rewriten by the following equations: OSr<g;l(a) a r
a .
El : x = f(x) + T/J(x, z) + g(x, z) . u(z) (32)
- cP(lJ(z» + (a _lJ(Z»2 lJ
E 2 : i = a(z) + h(z) . u(z).
<0.
Since subsystem El in eq. (32) is ISS for (z, u) and
u(z) ~ a4(1lzll), El is also ISS for z. Therefore, we (ii) Case: Ci(x) ~ cP0~(x) when x ---+ 0
can easily obtain the class K oo functions Q2, Ci and
class K function Xsatisfy the following conditions:
W~ -Q2(lIxll) + Ci(lIzll), (33) V(x, z) ~ -17(W(x» . Q2(llxll)
l
and if Ilxll > x(lIzID, + cP 0 Q3 0 Q3 (lJ(lIzll)) Ci(a -1 (lJ(llzID)
CioQ3l(lJ(lIzll) !:3
Q2(llxll) > Ci(lIzll)· (34) a .
Then, consider a function 0(·) such that
- cP(lJ(z» + (a _lJ(z))2 lJ

O(x) = l x
17(r)dr, (35)
< o.
Therefore, V is a Lyapunov function.
where the function 17(·) : IR ---+ IR is defined as
follows:
If 7. CONSTRUCTION OF INPUTS
. Ci(x)
Ilm < 1, (36) In the previous section, we have constructed a
x-O cP 0 Q3(x) Lyapunov function for system (32), and controller
. f cP o_Q3(r) (22) makes the entire system asymptotically sta-
17X=
( ) m (const), (37) ble. However, input (22) uses only the information
o:5 r <g;l(a) a(r)
of subsystem E 2 . Therefore, when the residual
and if of subsystem El is large, the convergence of the
I· Ci(x) >1 (38) system becomes slow.
lm cP0~ (
x-O x )
- '
Then, we construct a controller with theinforma-
x) = cPo~oX-loQl(X). tion of both El and E2 for system (11) in this
( (39) section. Because V is a Lyapunov function when
17 CioX-loQl(x)
we use input (18), we can obtain the following
It is clear that function 0(·) is well-defined. proposition.

1210
Proposition 4. A function V as eq. (42) is a CLF Then,
for the system (11). 2
(; = 2z2 _ + vis) 2z (1
1 + y'1
+ (2z)2
Moreover, we can use the following theorem.
2
2z (vis - y'1 + (2z)2)
Theorem 5. When we use a function Vasa CLF, 1 + y'1 + (2z)2
the state z of subsystem I;2 is kept in the domain
~ _z2 + Izl 3
of attraction D.
~ -U(z) + U(z)~ = -w(U(z)). (53)
Therefore, when we decrease the value of V, z We choose </> as
must not escape from the domain of attraction
D. Now, we can use a CLF with good properties.
</>(U(z)) = U(z) = z2. (54)
We rewrite system (11) into the following system: Then, the function 7l' is obtained as follows:
i = ](5;) + g(5;) . u.
ior ~dr
(44)
7l'(x)=
Then, r - T2
1 1
= - 2r 2 - 210g( 1 - r'i ). (55)
(45)
Therefore, we can obtain U:
Now, we can construct a controller that stabilizes • Z2
system (11) by theorem 2. U = -21zl - 210g(1 - Izl) + 1 _ Z2' (56)

On the other hand, W(x) = x 2 is an ISS-


8. EXAMPLE Lyapunov function. Then,
Consider the following system with input con- . 2 4xz
W=-2x +--
straint lul ~ 1: 1 + Izl

:t [~~] = [~ ~] [~~] + [n u. (46)


~ -2x + 41xllzl
2

~ _x 2 + 4z 2 . (57)

We choose the following new coordinates: Therefore,


x(x) = 4x. (58)
Hence,
Then, system (46) can be transformed to the
following diagonal system: Ql(X)=X, Qs(x) = x,
(59)
I;l ::i; = -x + U </>(x) = x, u(x) = 4x, x(x) = 4x.
(48)
I;2 : i = z + u. Then,
Subsystem El is cleary ISS, so we consider the </> 0 Qs(x) = x < u(x). (60)
stabilization of z. We choose a Lyapunov function
candidate U = z2 for subsystem I;2. Then, the Hence,
input that minimizes the derivative of U is given 1](x) = 1. (61)
in the following equation:
Therefore
u = - sgnz. (49)
Then, we obtain the following equation: (62)
(; = 2z (z - sgn z) Consequently, we obtain the following LyapunoY
= 2z 2 - 21z1 function:
1
= 2U(z) - 2U'i (z). (50) 1 z2
V = _x 2 - 21zl- 210g(I-lzl) + - - . (63)
4 1 - z2
Therefore the domain of attraction D = {Izl < I},
a = 1 and Therefore,
(51) 1 2z 2 2z 2
L·V = __ x 2 + - - + - - (64)
f 4 1 - Izl 1 - z2
Then, the input can be obtained as the following 1 2z 2z
equation: L·V= -x+ - - + - - (65)
9 4 1 - Izl 1 - Z2
z(1 + VS) 2 2
P = -x (1 - z2) + 8z (1 + Izl) + 8z
2
(66)
u(z) = - 1 + y'1 + (2z)2 (52) 2
Ix(1 - z2) + 8z(1 + Izl) + 8z 2j

1211
Finally, we can obtain the following input that method of a Lyapunov function for the system.
stabilize the system in the domain of attraction Finally, we have confirmed the effectiveness of the
Izl < 1: proposed method by computer simulation.

LjV + J(L j V)2 + ILgVI 4 1


u =- --':-1-+'-.jr.l:=+=;=;IL:=g::;:::;V~12'--· IL gVI
(67) 10. REFERENCES
(L g # 0)
u=o Ishidori, A. (1999). Nonlinear Control Systems If.
(L g = 0). Springer. London.
Lin, Y. and E. D. Sontag (1991). A universal for-
We show the simulation results in Fig. 1 and Fig. mula for stabilization with bounded controlls.
2. Systems & Control Letters 16, 393-397.
Lin, Y. and E. D. Sontag (1995). Control-
-x
- - - '1
Lyapunov universal formulas for restricted in-
puts. Control- Theory and Advanced Technol-
ogy 10, 1981-2004.
Malisoff, M. and E. D. Sontag (2000). Universal
formulas for feedback stabilization with re-
~ . spect to Minkowski balls. Systems & Control
Letters 40, 247-260.
, Sepulchre, R., M. Jankovic and P. Kokcr
,, tovic (1997). Constructive Nonlinear Control.
Springer. London.
Sontag, E. D. and Y. Wang (1995). On character-
_I.L--------'------'-------' izations of the input-tcrstate stability prop--
Timo erty. Systems & Control Letters 24, 351-359.
Fig. 1. Simulation Result: State

••L- --'-_ _ -===- ----,J

Time

Fig. 2. Simulation Result: Input

From the results, we can find that the state


converges with the origin and the input does not
exceed 1.

9. CONCLUSION

In this paper, we have considered a cascade-


connected nonlinear system that includes an
input-tcrstate stable subsystem and an unstable
subsystem. The objective in this paper is to con-
struct the input under an input restriction. We
have proposed a controller for the system with
an input constraint based on the local CLF and
ISS. Moreover, we have provided a construction

1212

You might also like