Trabalho Condicao Setor

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Domain of Attraction Estimates Obtained through a

Sector Condition Method


Leonardo Broering Groff and Esmael de Oliveira Flores

AbstractThis paper addresses the problem of stability analysis of a polynomial system by verifying a sector condition. To
this end, we first introduce some preliminary concepts regarding
sector nonlinearities and the estimation of domains of attraction,
as well as the LMI conditions required for the system stability.
Then, we qualitatively analyze the system and apply the method
in order to obtain an initial region of asymptotic stability.
With this, we use the trajectory-reversing method to expand the
obtained region. Finally, we compare the estimates to the real
region of attraction of the system and make some concluding
remarks.

I. I NTRODUCTION
The subject of nonlinear control is an important area of
automatic control, which has been getting a lot of attention
from researchers and designers of several areas of knowledge,
such as aircraft control, robotics and biomedical engineering
[1]. One of the main reasons why the nonlinear systems theory
is applied instead of the mature linear one is its ability to
describe phenomena observed in real physical systems, such
as multiple equilibrium points, robust limit cycles and other
complex dynamical behavior [2].
However, the control of nonlinear systems also poses some
challenges. First of all, the superposition principle does not
hold, forcing the use of more complex mathematical tools.
Also, nonlinear system present a broad spectrum of characteristics and behaviors, so that solutions for nonlinear control
problems are generally restricted to specific classes of systems,
and generic control methods can be quite conservative [3].
An important class of nonlinear systems is that of polynomial systems, that is, systems where the time derivatives
of the state variables depends on a polynomial function of
the systems state variables. Several methods for dealing with
this class of systems were proposed, such as the Sum of
Squares (SOS) [4], the Linear Fractional Representation (LFR)
[5], the Linear Parameter-Varying (LPV) systems [6], [7],
the Differential Algebraic Representation (DAR) [8], [9], and
methods based on a sector condition [3], [10], [11].
Most of the cited works rely on the Lyapunov theory [3],
[1] to prove the systems stability in terms of its equilibrium
points and on Linear Matrix Inequalities (LMIs) [12] to eficiently compute numerical solutions to the problems presented.
Also, since in general the stability of nonlinear systems doesnt
hold globally, convex optimization problems [13] are used

to find estimates of the domain of attraction of the systems


equilibrium points.
In this paper, we will follow the ideas presented in [10] to
analyze the stability of a polynomial system. However, unlike
in that paper, the considered nonlinearity only satisfy a sector
condition locally, so aditional LMI constraints are used to
ensure that the trajectories belong to a set that satisfies the
sector condition.
This paper is organized as follows: Section II presents some
preliminary concepts needed for the execution of this work and
the problem statement, Section III presents the methodology
followed, Section IV shows the results obtained, alogn with
some discussion about them and in Section V some concluding
remarks are made.
II. P RELIMINARIES AND P ROBLEM S TATEMENT
Consider the following nonlinear system:
x(t)

= f (x(t)),

where x Rn is the system state vector and f (x(t)) is a


polynomial function of x. In this case, the system can also be
represented as:
x(t)

= Ax(t) + H(z(t))
z(t) = Lx(t),
where A, H
dimensions, z
system with a
(z(t)) Rp
Fig. 1.

(2)

and L are constant matrices of appropriate


Rp can be seen as the output of the linear
transfer function G(s) = L(sI A)1 H and
the nonlinear feedback element, as shown in

A. Sector Condition
A nonlinearity (z(t)) verifies a sector condition if
z(t)2 z(t)(z(t)) z(t)2 , t 0, z Bz ,

*This work was supported by CAPES (PhD Scholarship) and CNPQ, Brazil.
L. B. Groff and E. de O. Flores. are with the UFRGS - Department of Electrical Engineering, Av. Osvaldo Aranha 103, 90035-190
Porto Alegre-RS, Brazil. E-mails: leonardo.groff@ufrgs.br,

esmaelflores@gmail.com

(1)

Fig. 1. Block diagram of system (2)

(3)

and in that case, the inequality


[(z(t)) z(t)]0 [(z(t)) z(t)] 0, t 0, z Bz
(4)
holds [3], and (z(t) is said to belong to the sector [, ]. An
example of a sector nonlinearity can be seen in Fig 2.
In this paper, we consider that (z(t)) is a decentralized
(also called decoupled) nonlinearity, that is:

1 (z1 (t))
2 (z2 (t))

(z(t)) =
(5)
.
..

.
p (zp (t))
We also consider that it belongs to a sector [0, ], with

11 . . . 1p

..
..
= ...
.
.
1p . . . pp

(6)

so that the following inequality holds for any diagonal positive


definite matrix = diag(1 , ..., p ):
[(z(t))]0 [(z(t)) z(t)] 0, t 0, z Bz .

(7)

In the general case where (z(t)) belongs to a sector [1 , 2 ],


the property (7) can be recovered by a loop transformation [3].
B. Stability Condition

result, we need to find an estimate of the region of attraction


(RA) of the origin of the system.
C. Region of Attraction
Since there is a nonlinear element in the loop, the global
stability of the origin may not be guaranteed. In this sense, let
(t, x0 ) be the trajectory of system (2) initialized in x0 . Then,
the RA for (2) is defined by [14]:
Definition 1: The RA of the origin of system (2) is the set
of all initial states x0 Rn for which (t, x0 ) 0 as t .
This means that the RA is the set of initial conditions whose
trajectories converge to the origin. However, the exact characterization of the RA is, generally, a hard task [3]. Hence, the
following definition of a region of asympotic stability (RAS)
is useful:
Definition 2: A region R0 is a RAS with respect to the
origin of system (2) if 0 R0 and (t, x0 ) 0 as
t , x0 R0 .
By this definition, it is clear that a RAS is a subset of the RA
of the origin. The idea then is to find a RAS for the origin of
the system and use it as an estimate of the RA. One method
of finding such a RAS is to consider a level surface of the
Lyapunov function associated to the system that belongs to
the set B. Representing the Bz as a polytope defined by the
inequalities:
Bz = {z : a0k z 1, k = i, ..., p}.

In order to analyze the stability of system (2), consider a


Lyapunov candidate function V (x) = x0 P x and the following
Lemma, adapted from [10]:
Lemma 1: If there exist a symmetric positive definite matrix
P and a diagonal positive definite matrix satisfying:
 0

A P + P A P H + L0
<0
(8)

2
then V (x) 0 provided z Bz , with V (x) = x0 P x.
The proof of Lemma 1 is similar to the proof of Proposition
5 from [10] when saturation is not taken into account, except
in our case (z) only belongs to the sector [0, ] locally. In
this sense, Lemma 1 only proves that, given 0 Bz , the origin
of the system is stable. In order to accomplish a more practical

(9)

From the definition of z, we have that z Bz x Bx ,


with:
Bx = {x : a0k Lx 1, k = 1, ..., p}.

(10)

From Lemma 1, a region E = x : x0 P x 1 will be an


invariant set if E Bx , which is guaranteed if the following
LMI is satisfied [12], [8]:


1 a0k L
0.
(11)
P
With this, we can state that if (8) and (11) are satisfied, than
E is a RAS for the origin of the system (2).
D. Optimization Problems
The LMIs (8) and (11) provide conditions for guaranteeing
that a set of initial conditions belong to the RA of the origin.
Now, we wish to find parameters that lead to a maximization
of that set. Since the stability conditions are cast in the form
of LMIs, we can formulate convex optimization problems in
the form:

0.5

minimize j()
P,

subject to:

-0.5
(z)
Sector [a, b]

-1
-4

-3

-2

-1

Fig. 2. A nonlinearity that satisfies a sector condition locally.

,
(8), (11)

(12)

where j() is an optimization criterion, such as:


OP1: Maximize the minor axis: j = and P I;
OP2: Maximize all axes with the same weight: j =
trace(P )

E. Problem Statement

B. Obtaining an Ellipsoidal RAS

Consider the nonlinear system described by the following


state equations:

To obtain an estimate of the RA of the origin using Lemma 1


and (11), we first need to rewrite the system as (2). To do so,
we consider the following representation of (13):




0 0
0
1
x+
(z)
x =
1 1
1 2
| {z }
| {z }
H
 A
1 0
(18)
z=
x
0 2
| {z }
 3L
z
(z) = 13 .
z2

x 1 = x2
x 2 = (1 + x21 )x1 + (2 + 8x22 )x2 + u.

(13)

u = 2x1 4x2
The problems we wish to address in the present work are:
Problem 1: Determine all the equilibrium points of the
system.
Problem 2: Obtain an ellipsoidal estimate of the RA of the
origin.
Problem 3: Apply the reversed-trajectory method to enlarge
the obtained estimate.
Problem 4: Analyze the conservatism of the RAS obtained.

Next, we analyze the sector condition (7). Substituting (z),


we find:

III. M ETHODOLOGY AND R ESULTS

1 (z16 11 z14 12 z13 z2 ) + 2 (z26 22 z24 21 z23 z1 ) 0


(19)

A. Qualitative Analysis
Initially, we determine all the equilibrium points of the
system. This can be done by solving (13) for x = 0, leading
to:
x2 = 0

x1 = 0 .
(x21 1)x1 = 0 x1 = 1

x1 = 1

(14)

Next, we proceed to analyze the behavior of each equilibrium


point. To do this, we compute the Jacobian of the state
equations of the system finding:


f
0
=
3x21 1
x

1
24x22 2

(15)

Evaluating the Jacobian at the equilibrium point p1 = [0, 0]


and calculating the eigenvalues:


 
f
0
1
1 = 1
=
,
,
(16)

1
2
2 = 1
x x=p1
so this equilibrium point is a stable node. On the other hand,
evaluating the Jacobian at p2 = [1, 0] or p3 = [1, 0] we find:


f
0
=

2
x x=p2 or p3


1
,
2

so p2 and p3 are saddle points.

Choosing 12 = 21 = 0, (7) is verified if:


1 z14 (z12 11 ) + 2 z24 (z22 22 ) 0,

(20)

which holds for all z such that |z1 | 11 and |z2 | 22 .


Note that the polytope Bx can be described in terms of , as
follows:
p



0
(11 )1
p
a1 =
, a2 =
(21)
0
(22 )1
Performing a linear search in the parameters 11 and 22 ,
we find the following that maximizes j, for both criteria
(OP1 and OP2):


0.28
0
=
(22)

0.92
Fig. 3 presents the guaranteed RAS obtained with OP1 and
OP2.

0.6

E with OP1
E with OP2
Bx

0.4
0.2

x2

In this section, we will present the methodology used to


address the proposed problems and the results obtained by
this methodology.

0
-0.2
-0.4
-0.6

1 = 0.73
,
2 = 2.73

(17)

-0.6

-0.4

-0.2

0.2

x1

Fig. 3. Block diagram of system (2)

0.4

0.6

C. Enlarging the Estimated RA

x = f (x).

0.6
E with OP1
Enlarged RAS t = 0.5s
Enlarged RAS t = 1s
Enlarged RAS t = 1.5s

0.4

x2

0
-0.2
-0.4

-0.6

-0.4

-0.2

x1

0.2

0.4

0.6

0.8

Fig. 4. Enlarged estimates with different simulation times using trajectoryreversing method
1

-1

-2
-2

-1

-0.5

0.5

1.5

x1

Fig. 6. RA boundaries (red) and some trajectories starting near it (black)

Another way of obtaining an estimate of the RA is by


finding the trajectories that bound the RA. In general, this is a
very complex task, but in this particular second-order system it
can be done with relative ease. Reversing the stable trajectories
of the saddle points we can find two separatrices which are
the boundaries of the RA. Fig. 6 shows the results of applying
this method.
In Fig 7, all the results are presented for comparison.
We can see that the use of a quadratic Lyapunov function
and the sector-condition add some degree of conservatism
to the estimated ellipsoid. However, it provides a decent
approximation for some applications and a useful starting point
in case more precise estimations are required.

In this paper we applied a sector condition to find an


estimate of the region of attraction of the system using LMI
restrictions. Then, we used the trajectory-reversing method
to obtain enlarged estimates at the cost of increased computational burden. Comparing the results with the real region of attraction, we found that the methods worked very
well in finding increasingly larger estimations, and that the
sector condition provides a useful initial estimate. In higher
dimension systems, however, the method of trajectory-reversal
might become impractical and innacurate, strengthening the

t = 1.5s, 100 points


t = 2s, 200 points
t = 2.5s, 500 points
t = 3s, 2500 points
E with OP1

0.5

-1.5

IV. C ONCLUDING R EMARKS

0.2

-0.8

(23)

With this, we can find better approximations of the system


RA. Starting with the initial estimation E, we discretize the
surface of the ellipsoid using 100 points, and then simulate
the system with intervals of t = 0.5s, t = 1s and t = 1.5s.
The estimates obtained are shown in Fig 4. The increase in the
size of the estimated RA is clear. However, the method stops
working well for simulation times greater than 1.5 seconds,
since the trajectories of the system have different speeds
when approaching infinity, wich happens because the RA of
the origin is unbounded. Using more points to discretize the
surface of E allows larger simulation times. For example, Fig.
5 shows the result of simulations for t = 2s using 100 points,
t = 2.5s using 250 points and t = 3s using 1000 points.
Note that the computational complexity increases sharply.
Another problem is that the trajectories have finite escape
times, so even if a very large number of points are used in
the discretization, some trajectories cease to exist at around
t = 6s.

-0.6
-1

x2

Using the trajectory-reversing method [3], we will now


enlarge the obtained estimated RA. This method consists in
simulating the reversed trajectories of the system of interest,
by the simulation of

0.6
0.4

x2

x2

0.2
0

0
-0.2

-0.5

-0.4
-0.6

-1
-2

-1.5

-1

-0.5

0.5

1.5

x1

-0.8
-1.5

-1

-0.5

0.5

x1

Fig. 5. Enlarged estimates using increased number of point to discretize the


surface of E

Fig. 7. Comparison of all methods

1.5

importance of an estimate that doesnt rely on the simulation


of the system.
R EFERENCES
[1] J.-J. E. Slotine, W. Li et al., Applied nonlinear control. Englewood
Cliffs, NJ: Prentice-Hall, 1991.
[2] S. Sastry, Nonlinear Systems, Analysis, Stability, and Control, 1st ed.
London: Springer, 1999.
[3] H. K. Khalil, Nonlinear Systems, 2nd ed. Upper Saddle River: PrenticeHall, 1996.
[4] P. A. Parrilo, Structured semidefinite programs and semialgebraic
geometry methods in robustness and optimization, Ph.D. dissertation,
California Institute of Technology, 2000.
[5] L. El Ghaoui and G. Scorletti, Control of rational systems using linearfractional representations and lmis, Automatica, vol. 32, no. 9, pp.
12731284, 1996.
[6] A. L. Do, J. M. Gomes da Silva Jr., O. Sename, and L. Dugard, Control
design for lpv systems with input saturation and state constraints:
an application to a semi-active suspension, in Proc. Conference on
Decision and Control and European Control Conference, 2011.
[7] M. Q. Nguyen, J. M. Gomes da Silva Jr., O. Sename, and L. Dugard,
Semi-active suspension control problem: some new results using an
LPV/H state feedback input constrained control, in Proc. IFAC
Conference on Advances in PID Control, 2012.
[8] D. F. Coutinho, A. S. Bazanella, A. Trofino, and A. S. e Silva, Stability
analysis and control of a class of differential algebraic nonlinear systems, International Journal of Robust and Nonlinear Control, vol. 14,
no. 16, pp. 13011326, 2004.
[9] D. F. Coutinho and J. M. Gomes da Silva Jr., Computing estimates
of the region of attraction for rational control systems with saturating
actuators, IET Control Theory and Applications, vol. 4, no. 3, pp. 315
325, 2010.
[10] E. B. Castelan, S. Tarbouriech, and I. Queinnec, Stability and stabilization of a class of nonlinear systems with saturating actuators, in Proc.
of the 9th IFAC Symposium on Nonlinear Control Systems (NOLCOS
2013), 2013.
[11] , Control design for a class of nonlinear continuous-time systems,
Automatica, vol. 44, pp. 20342039, 2008.
[12] S. P. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear matrix
inequalities in system and control theory. Philadelphia: SIAM, 1994,
vol. 15.
[13] S. Boyd and L. Vandenberghe, Convex optimization.
Cambridge:
Cambridge University Press, 2004.
[14] S. Tarbouriech, G. Garcia, J. M. Gomes da Silva Jr, and I. Queinnec,
Stability and stabilization of linear systems with saturating actuators.
London: Springer-Verlag, 2011.

You might also like