Documente Academic
Documente Profesional
Documente Cultură
I
n many engineering applications, partial stability (stabil- cifically, consider the equation of motion for the slider-
ity with respect to part of the systems states) is often crank mechanism shown in Fig. 1 given by [10], [11]:
necessary. In particular, partial stability arises in the
study of electromagnetics [1], inertial navigation systems m( ( t ))&&( t ) + c( ( t ))& 2( t ) = u ( t ),
[2], spacecraft stabilization via gimballed gyroscopes (0 ) = , &(0 ) = & t 0 ,
0 0, (1)
and/or flywheels [3], combustion systems [4], vibrations in
rotating machinery [5], and biocenology [6], to cite but a
where
few examples. For example, in the field of biocenology in-
volving Lotka-Volterra predator-prey models of population 2
r cos sin
dynamics with age structure, if the birth rate of some of the m( ) = mBr 2 + mAr 2 sin + ,
species preyed upon is left alone, then the corresponding l 2 r 2 sin 2 (2)
population increases without bound while a subset of the
prey species remains stable [6, pp. 260-269]. The need to
consider partial stability in the aformentioned systems r cos sin
c( ) = mAr 2 sin +
arises from the fact that stability notions involve equilib- l 2 r 2 sin 2
rium coordinates as well as a hyperplane of coordinates that l 2(1 2sin 2 ) + r 2 sin 4
is closed but not compact. Hence, partial stability involves cos + r ,
(l 2 r 2 sin 2 )3 / 2 (3)
motion lying in a subspace instead of an equilibrium point.
Additionally, partial stabilization, that is, closed-loop stabil-
ity with respect to part of the closed-loop systems state, and mA and mB are point masses, r andl are the lengths of the
also arises in many engineering applications [3], [5]. Spe- rods, and u() is the control torque applied by the motor.
Now suppose we choose the feedback control law u = ( , & )
.
cifically, in spacecraft stabilization via gimballed gyro-
scopes, asymptotic stability of an equilibrium position of so that the angular velocity of the crank is constant; that is,
the spacecraft is sought while requiring Lyapunov stability &( t ) a s t , w h e re > 0. T h i s i m p l i e s t h a t
of the axis of the gyroscope relative to the spacecraft [3]. Al- ( t ) = t as t . Furthermore, since m( ) and c( )
ternatively, in the control of rotating machinery with mass are functions of , we cannot ignore the angular position .
imbalance, spin stabilization about a nonprincipal axis of in- Hence, since does not converge, it is clear that (1) is unsta-
ertia requires motion stabilization with respect to a ble in the standard sense but partially asymptotically stable
subspace instead of the origin [5]. Perhaps the most com- with respect to & (see Definition 1 below).
mon application where partial stabilization is necessary is Our next example involves a nonlinear system originally
adaptive control, wherein asymptotic stability of the studied as a simplified model of a dual-spin spacecraft to in-
closed-loop plant states is guaranteed without necessarily vestigate the resonance capture phenomenon [12] and more
achieving parameter error convergence [7]-[9]. recently studied to investigate the utility of a rota-
To further demonstrate the utility and need for partial tional/translational proof-mass actuator for stabilizing
stability theory, we consider two simple examples. Spe- translational motion [13]. The system (see Fig. 2) involves an
eccentric rotational inertia on a translational oscillator giving
rise to nonlinear coupling between the undamped oscillator
B and the rotational rigid-body mode. The oscillator cart of
mass M is connected to a fixed support via a linear spring of
l r stiffness k. The cart is constrained to one-dimensional mo-
tion, and the rotational proof-mass actuator consists of a
A mass m and mass moment of inertia I located at a distance e
Motor
from the carts center of mass. Lettingq, q, &
& , and denote the
translational position and velocity of the cart and the angular
position and velocity of the rotational proof mass, respec-
Figure 1. Slider-crank mechanism. tively, the dynamic equations of motion are given by
Chellaboina (ChellaboinaV@missouri.edu) is with the Deaprtment of Mechanical and Aerospace Engineering, University of Missouri, Colum-
bia, MO 65211, U.S.A. Haddad is with the School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0150, U.S.A.
0272-1708/02/$17.002002IEEE
66 IEEE Control Systems Magazine December 2002
[ ]
( M + m)q&&( t ) + me &&( t )cos ( t ) & 2( t )sin ( t ) + kq ( t ) = 0, addressing stability theory for autonomous and nonauto-
nomous systems within a unified framework.
(4)
Yet another application of partial stability theory is the ex-
&&( t ) + meq&&( t )cos ( t ) = 0, tra flexibility it provides in constructing Lyapunov functions
( I + me 2 ) (5)
for nonlinear dynamical systems. Specifically, generalizing
Lyapunovs stability theorem to include partial stability
where t 0, q (0 ) = q 0 , q&(0 ) = q&0 , (0 ) = 0 , and &(0 ) = & 0 . Note
weakens the hypotheses on the Lyapunov function (see The-
that since the motion is constrained to the horizontal plane,
orem 1), thus enlarging the class of allowable functions that
the gravitational forces are not considered in the dynamic
can be used in analyzing system stability. Perhaps the clear-
analysis. Analyzing (4), (5) (see Example 1 for details), it fol-
est example of this is the Lagrange-Dirichlet stability problem
lows that the zero solution (q ( t ),q&( t ), ( t ), &( t )) (0 ,0 ,0 ,0 ) to
[6] involving the conservative Euler-Lagrange system with a
(4), (5) is unstable in the standard sense but partially
& nonnegative-definite kinetic energy functionT and a positive-
Lyapunov stable with respect toq, q, & and (see Definition 1).
definite potential function U. In this case, the Lagrange-
Once again, standard Lyapunov stability theory cannot be
Dirichlet energy function V = T + U is only nonnegative-defi-
used to arrive at this result since the angular position of
nite and hence cannot be used as a Lyapunov function candi-
the rotational proof mass cannot be ignored from (4), (5)
date to analyze the stability of the system using standard
and ( t ) as t .
Lyapunov theory. However, the Lagrange-Dirichlet energy
Another important application of partial stability theory
function can be used as a valid Lyapunov function within par-
is the unification it provides between time-invariant stabil-
tial stability theory to guarantee Lyapunov stability of the
ity theory and stability theory for time-varying systems.
Lagrange-Dirichlet problem (see Example 2).
Specifically, in most nonlinear control textbooks, the pre-
In this article, we present partial stability theorems for
sentations on stability theory for nonlinear time-invariant
nonlinear dynamical systems and present a unification be-
systems and nonlinear time-varying systems are often sepa-
tween partial stability theory for autonomous systems and
rated, with the latter being labeled as an advanced topic on
stability theory for nonlinear time-varying systems. This
the subject. Furthermore, some textbooks (see, for exam-
unification allows for time-varying stability theory to be pre-
ple, [14, p. 82]) remark that time-varying systems can be eas-
sented as a special case of autonomous partial stability the-
ily handled as part of time-invariant systems by appending
ory so that time-varying and time-invariant stability theory
another state to represent time. Specifically, consider the
can be discussed in juxtaposition in a first course on nonlin-
time-varying nonlinear dynamical system given by
ear systems. Furthermore, partial stability theory is a funda-
mental topic of nonlinear system theory and should be part
x&( t ) = f ( t , x ( t )), x( t0 ) = x 0 , t t0 , (6)
of a nonlinear control course in its own right. Note that our
partial stability definitions are different from those given in
where x ( t ) R n , t t 0 , and f :[t 0 , t 1 ) R n R n . Now define
[3] and [15] in that we do not require the whole system ini-
x 1( ) = x ( t ) and x 2( ) = t , where = t t 0 , and note that the so-
tial condition to lie in a neighborhood of the origin. The defi-
lution x ( t ), t t 0 , to the nonlinear time-varying dynamical
nition used in this article is more closely related to the
system (6) can be equivalently characterized by the solution
notion of stability of sets [16], [17]. This alternative defini-
x 1( ), 0, to the nonlinear autonomous dynamical system
tion is key to providing the above unification. Finally, we
x& 2( ) = 1, x 2 (0 ) = t 0 , (8) M
where x& 1() and x& 2() in (7) and (8), respectively, denote dif- k
ferentiation with respect to . However, in this case, stability I
results for time-invariant systems do not apply to the aug-
mented system (7), (8) since one of the states, namely, the
m
state x 2 representing time, is unbounded. As a consequence,
the presentation on autonomous and nonautonomous sta-
bility theory is often separated, and students are often puz-
zled as to whether the two theories are fundamentally
distinct or somehow connected. However, writing the x
time-varying nonlinear system (6) as (7), (8), it is clear that
partial stability theory provides a natural formulation for Figure 2. Rotational/translational proof-mass actuator.
z z
||x01|| = ||x01|| =
||x01|| = ||x01|| =
x(t) x(t)
y1 y1
y2 y2
(a) (b)
Figure 3. (a) Partial Lyapunov stability with respect to x 1 . (b) Partial asymptotic stability with respect to x 1 : x 1 = [ y1 y2 ]T , x 2 = z ,
and x = [ x 1T x 2 ]T .
p p
differentiable function V :D R n 2 R. Furthermore, we as- x1 V ( x1 ,x2 ) x1 , ( x 1 ,x 2 ) D R n2 , (16)
sume that the solution ( x 1( t ), x 2( t )) to (9), (10) exists and is
unique for all t 0. It is important to note that unlike standard
theory (see, for example, [19, Theorem 2.4]), the existence of V& ( x 1 , x 2 ) x 1 ,
p
( x 1 ,x 2 ) D R n2 , (17)
a Lyapunov function V ( x 1 , x 2 ) satisfying the conditions in
Theorem 1 is not sufficient to ensure that all solutions of (9),
then the nonlinear dynamical system given by (9),
(10) starting in D R n 2 can be extended to infinity since nei-
(10) is exponentially stable with respect to x 1 uni-
ther of the states of (9), (10) serve as an independent vari-
formly in x 20 .
able. We do note, however, that continuous differentiability of
vi) If D = R n1 and there exists a continuously differentia-
f1(,) and f2(,) provides a sufficient condition for the exis-
ble function V :R n1 R n 2 R and positive constants
tence and uniqueness of solutions to (9), (10) for all t 0.
, , , p 1satisfying (16) and (17), then the nonlinear
Theorem 1. Consider the nonlinear dynamical system
dynamical system given by (9), (10) is globally expo-
(9), (10). Then the following statements hold:
nentially stable with respect to x 1 uniformly in x 20 .
i) If there exists a continuously differentiable function
Proof:
V :D R n 2 R and a class K function () such that
i) Let > 0 be such that B ={x 1 R n1 : x 1 < } D, define
= ( ), and define D = {x 1 B : there exists x 2 R n 2
V (0 , x 2 ) = 0 , x 2 R n2 , (11) such that V ( x 1 , x 2 ) < }. Since V& ( x 1 , x 2 ) 0, it follows
that V ( x 1( t ), x 2( t )) is a nonincreasing function of time
and hence D R n 2 is a positive invariant set with re-
( x 1 ) V (x 1 , x 2 ), ( x 1 ,x 2 ) D R n2 , (12) spect to (9), (10). Next, since V (,) is continuous and
V (0 , x 20 ) = 0, x 20 R n 2 , there exists = ( , x 20 ) > 0
such that V ( x 1 , x 20 ) < , x 1 B , and hence B D .
V& ( x 1 , x 2 ) 0 , ( x 1 ,x 2 ) D R n2 , (13) Hence, for all ( x 10 , x 20 ) B R n 2 , it follows that
x 10 B \ DL and t 0, x 1( t ) x 10 e p
,
t
establishing exponential stability with respect to x 1
V ( x 1( t ), x 2( t )) = V ( x 10 , x 20 ) + V& ( x 1( s ), x 2( s ))ds
uniformly in x 20 .
0
t vi) The proof follows as in iv) and v).
V ( x 10 , x 20 ) ( x 1( s ) )ds Remark 1. By setting n1 = n and n2 = 0, Theorem 1 special-
0 izes to the case of nonlinear autonomous systems of the form
()
V ( x 10 , x 20 ) $ t . x& 1( t ) = f1( x 1( t )). In this case, Lyapunov (respectively, asymp-
totic) stability with respect to x 1 and Lyapunov (respectively,
asymptotic) stability with respect to x 1 uniformly in x 20 are
Letting t (V ( x 10 , x 20 ) L ) / ( $ ), it follows that equivalent to the classical Lyapunov (respectively, asymp-
V ( x 1( t ), x 2( t )) L, which is a contradiction. Hence, totic) stability of nonlinear autonomous systems. Further-
L = 0, and, since x 10 B was chosen arbitrarily, it fol- more, note that in this case there exists a continuously
lows that V ( x 1( t ), x 2( t )) 0 as t for all x 10 B . differentiable function V :D R such that (12), (14), (15)
Now, since V ( x 1( t ), x 2( t )) ( x 1( t ) ) 0, it follows hold if and only if V () is such that V (0 ) = 0, V ( x 1 ) > 0, x 1 0,
that ( x 1( t ) ) 0 or, equivalently, x 1( t ) 0 t , V ( x 1 )f1( x 1 ) < 0, x 1 0 [19]. In addition, if D = R n1 and there
establishing asymptotic stability with respect to x 1 . exist class K functions (),() and a continuously differen-
iv) Let > 0 be such that x 10 < . It now follows from (15) tiable functionV () such that (12), (14), (15) hold if and only if
that V ( x 1( t ), x 2( t )) is a nonincreasing function of time V () is such that V (0 ) = 0, V ( x 1 ) > 0, x 1 0, V ( x 1 )f1( x 1 ) < 0,
and hence x 1( t ) D =
{x 1 R n1 : there exists x 2 R n 2 x 1 0, and V ( x 1 ) as x 1 .
such that V ( x 1 , x 2 ) ( )}, t 0. Next, since () is a It is important to note that there is a key difference be-
class K function, it follows that there exists > 0 such tween the partial stability definitions given in Definition 1
that ( ) < ( ), and it follows from (14) that and the definitions of partial stability given in [3] and [15]. In
( x 1( t ) ) V ( x 1( t ), x 2( t )) ( ) < ( ), t 0. Hence, particular, the partial stability definitions given in [3] and
x 1( t ) B , t 0. Now, with = , the proof follows as in [15] require that both the initial conditions x 10 and x 20 lie in a
the proof of iii). neighborhood of the origin, whereas in Definition 1, x 20 can
v) Let > 0 and = ( ) be given as in the proof of i) . Now be arbitrary. As will be seen in the next section, this differ-
(17) implies that V& ( x 1 , x 2 ) 0, and hence it follows ence allows us to unify autonomous partial stability theory
that V ( x 1( t ), x 2( t )) is a nonincreasing function of time with time-varying stability theory. Lyapunov (respectively,
and D R n 2 D R n 2 is a positive invariant set with asymptotic) stability with respect to x 1 given in Definition 1
respect to (9), (10). Thus, it follows from (16) and (17) is referred to in [3] as x 1 -stability (respectively, x 1 -asymp-
that for all t 0 and ( x 10 , x 20 ) D R n 2 , totic stability) for large x 2 , whereas Lyapunov (respectively,
asymptotic) stability with respect to x 1 uniformly in x 2 given
in Definition 1 is referred to in [3] as x 1 -stability (respec-
V& ( x 1( t ), x 2( t )) x 1( t ) V ( x 1( t ), x 2( t )) ,
p
tively, x 1 -asymptotic stability) with respect to the whole of
x 2 . Note that if a nonlinear dynamical system is Lyapunov
(respectively, asymptotically) stable with respect to x 1 in
which implies that the sense of Definition 1, then the system is x 1 -stable (re-
spectively, x 1 -asymptotically stable) in the sense of the defi-
nition given in [3] and [15]. Furthermore, if there exists a
t
continuously differentiable function V :D R n 2 R and a
V ( x 1( t ), x 2( t )) V ( x 10 , x 20 )e
.
class K function () (respectively, and class K function ())
such that V (0 ,0 ) = 0 and (12), (14) (respectively, (12), (14))
It now follows from (16) that hold, then the nonlinear dynamical system (9), (10) is x 1 -sta-
V ( x 1 )f1( x 1 , x 2 ) ( x 1 ), ( x 1 ,x 2 ) D R n2 , (19)
1
x 12 + max x 22 + x 42 ,
2 2
( )
then the nonlinear dynamical system given by (9), (10) is as-
ymptotically stable with respect to x 1 uniformly in x 20 . More- which implies that V () satisfies (12) and (14). Now, since
over, if V :R n1 R is radially unbounded, then the system V& ( x 1 , x 2 , x 3 , x 4 ) = 0, it follows from ii) of Theorem 1 that (4),
(9), (10) is globally asymptotically stable with respect to x 1 (5) is Lyapunov stable with respect to x 1 , x 2 , and x 4 uni-
uniformly in x 20 . formly in x 3 . Furthermore, since the system involves a non-
Proof. The proof is a direct consequence of Theorem 1 linear coupling of an undamped oscillator with a rotaional
with V ( x 1 , x 2 ) replaced by V ( x 1 ). rigid-body mode, it follows that x 3( t ) as t .
Example 1. In this example, we use Theorem 1 to show Example 2. In this example, we apply Theorem 1 to a
that the rotational/translational proof-mass model (4), (5) is Lagrange-Dirichlet problem involving a conservative Eu-
partially Lyapunov stable with respect toq, q, & To show
& and . ler-Lagrange system with a nonnegative-definite kinetic en-
& &
this, let x 1 = q , x 2 = q , x 3 = , x 4 = and consider the ergy function T and a positive-definite potential function U.
Lyapunov function candidate Specifically, we consider the motion of the spherical pendu-
lum shown in Fig. 4, where denotes the angular position of
the pendulum with respect to the vertical z-axis and de-
notes the angular position of the pendulum in the x-y plane, m
V ( x1 ,x2 , x3 ,x4 ) =
denotes the mass of the pendulum, L denotes the length of
1
2
[ kx 12 + ( M + m)x 22 + ( I + me 2 )x 42 + 2mex 2 x 4 cos x 3 . ] the pendulum, k denotes the torsional spring stiffness, and g
denotes the gravitational acceleration. Definingq = [ ]T to be
& & T
(20) the generalized system positions and q& =[ ] to be the gen-
eralized system velocities, it follows that governing equa-
Note that V ( x 1 , x 2 , x 3 , x 4 ) = (1 / 2)x 12 + (1 / 2)x~ T P( x 3 )x~, where tions of motion are given by the Euler-Lagrange equation
x~ = [x 2 x 4]T and
d L L
(q ( t ),q&( t )) (q ( t ),q&( t )) = 0 ,
M+m me cos x 3 dt q& q
P( x 3 ) = .
me cos x 3 I + me 2 q (0 ) = q 0 , q&(0 ) = q&0, t 0 ,
(21) (26)
function andV ( x 1 , x 2 ) = V (q ,q& ) ( x 1 ). Furthermore, note ( x 10 , x 20 ) R n1 R n 2 , x 1( t ) R ={x 1 R n1 :W ( x 1 ) = 0} a s
that V (0 , x 2 ) = 0, x 2 R, and V& ( x 1 , x 2 ) = 0. Now, it follows t .
from i) of Theorem 1 that the Euler-Lagrange system given Proof. Assume (29) and (30) hold. Then it follows from
by (27), (28) is partially Lyapunov stable with respect to x 1 . Theorem 1 that the nonlinear dynamical system given by
Finally, it can be easily shown via simulations that the Eu- (9), (10) is Lyapunov stable with respect to x 1 uniformly in
ler-Lagrange system given by (27), (28) is not Lyapunov sta- x 20 . Let > 0 be such that B D and let = ( ) > 0 be such
bility in the standard sense. t h a t i f x 10 B , t h e n x 1( t ) B , t 0. N o w, s i n c e
In the case of time-invariant systems, the Barbashin- V ( x 1( t ), x 2( t )) is monotonically nonincreasing and bounded
Krasovskii-LaSalle invariance theorem [18], [19] shows from below by zero, it follows from the Bolzano-Weierstrass
theorem [20] that lim t V ( x 1( t ), x 2( t )) exists and is finite.
Hence, since for every t 0,
z
k y
W ( x ( ))d V&( x ( ), x ( ))d
t t
1 1 2
0 0
= V ( x 10 , x 20 ) V ( x 1( t ), x 2( t )),
L
it follows that lim t W ( x 1( ))d exists and is finite. Next,
t
0
x L since x 1( t ) is uniformly continuous and W() is continuous
on a compact set B , it follows that W ( x 1( t )) is uniformly
Lsin continuous at every t 0. It now follows from the Barbalats
mg
lemma [19] thatW ( x 1( t )) 0 as t . Finally, if in addition
D = R n1 andW1() is radially unbounded, then, as in the proof
of iv) of Theorem 1, for every x 10 R n1 there exists , > 0
such that x 10 B and x 1( t ) B , t 0. The proof now follows
Figure 4. Spherical pendulum. by repeating the above arguments.