1.

Introduction to Nonlinear Systems

Nonlinear Control Systems

Antonio Pedro Aguiar

[email protected]

1. Introduction to Nonlinear Systems

IST-DEEC PhD Course

http://users.isr.ist.utl.pt/%7Epedro/NCS2012/

2011/2012

1
1. Introduction to Nonlinear Systems

Objective The main goal of this course is to provide to the students a solid
background in analysis and design of nonlinear control systems

Why analysis? (and not only simulation)

Every day computers are becoming more and more powerful to simulate complex
systems
Simulation combined with good intuition can provide useful insight into systems
behavior
Nevertheless
It is not feasible to rely only on simulations when trying to obtain guarantees of
stability and performance of nonlinear systems, since crucial cases may be missed
Analysis tools provide the means to obtain formal mathematical proofs
(certificates) about the systems behavior
results may be surprising, i.e, something we had not thought to simulate.

2
1. Introduction to Nonlinear Systems

Why study nonlinear systems?

Nonlinear versus linear systems
Huge body of work in analysis and control of linear systems
most models currently available are linear (but most real systems are nonlinear...)

However
dynamics of linear systems are not rich enough to describe many commonly
observed phenomena

3
1. Introduction to Nonlinear Systems

Examples of essentially nonlinear phenomena

Finite escape time, i.e, the state can go to infinity in finite time (while this is
impossible to happen for linear systems)
Multiple isolated equilibria, while linear systems can only have one isolated
equilibrium point, that is, one steady state operating point
Limit cycles (oscillation of fixed amplitude and frequency, irrespective of the
initial state)

Subharmonic, harmonic or almost-periodic oscillations;

A stable linear system under a periodic input produces an output of the same frequency.
A nonlinear system can oscillate with frequencies which are submultiples or multiples of
the input frequency. It may even generate an almost-periodic oscillation, i.e, sum of
periodic oscillations with frequencies which are not multiples of each other.
Other complex dynamic behavior, for example: chaos, biforcations, discontinuous
jump, etc...
4
1. Introduction to Nonlinear Systems

State-space model

State equation
x = f (t, x, u)
Output equation
y = h(t, x, u)

x1 u1 f1 (t, x, u)
2 3 2 3 2 3
6 x2 7 6 u2 7 6 f2 (t, x, u) 7
x = 6 . 7, u = 6 . 7, f (t, x, u) = 6 ..
6 7 6 7 6 7
4 .. 5 4 .. 5
7
4 . 5
xn um fn (t, x, u)
where x Rn is the state variable, u Rm is the input signal, and y Rq the output
signal. The symbol x = dx
dt
denotes the derivative of x with respect to time t.

5
1. Introduction to Nonlinear Systems

State-space model

State equation
x = f (t, x, u) (1)
Output equation
y = h(t, x, u) (2)

Particular cases:
Linear Systems, where the state model takes the form

x = A(t)x + B(t)u

y = C(t)x + D(t)u
Unforced state equation
x = f (t, x)
i.e., it does not depend explicitly on the input u, e.g., consider the case that there
is a state feedback u = (t, x), and therefore the closed-loop system is given by
x = f (t, x, (t, x)) = f(t, x)
Unforced autonomous (or time-invariant) state equation

x = f (x)

6
1. Introduction to Nonlinear Systems

Example - Pendulum

There is a frictional force assumed to be proportional to the (linear) speed of the mass
m. Using the Newtons second law of motion at the tangential direction

ml = mg sin () kl

where m is the mass, l is the length of the rope and k the frictional constant.

7
1. Introduction to Nonlinear Systems

Example - Pendulum

ml = mg sin () kl
State model
x1 =
x1 =
x1 = x2
x2 = gl sin (x1 ) k
x
m 2
What are the equilibrium points?

8
1. Introduction to Nonlinear Systems

Equilibrium point

A point x = x in the state space is said to be an equilibrium point of

x = f (t, x)

if
x(t0 ) = x x(t) = x , t t0
that is, if the state starts at x , it will remain at x for all future time.

For autonomous systems, the equilibrium points are the real roots of f (x) = 0.

The equilibrium points can be of two kinds:

isolated, that is, there are no other equilibrium points in its vicinity
continuum of equilibrium points.

Much of nonlinear analysis is based on studying the behavior of a system around its
equilibrium points.

9
1. Introduction to Nonlinear Systems

Example - Pendulum

State model
x1 = x2
x2 = gl sin (x1 ) k
x
m 2
Equilibrium points:

0 = x2
0 = gl sin (x1 ) k
x
m 2
which implies that x2 = 0 and sin(x1 ) = 0. Thus, the equilibrium points are

(n, 0), n Z

What is the behavior of the system near the equilibrium points?

10
1. Introduction to Nonlinear Systems

Qualitative behavior of 2nd order linear time-invariant systems

x = Ax, x R2 , A R22

Apply a similarity transformation M to A:

M 1 AM = J, M R22

where J is the real Jordan form of A, which depending on the eigenvalues of A may
take one of the three forms

1 0 k
, ,
0 2 0

with k being either 0 or 1.

11
1. Introduction to Nonlinear Systems

Case 1: Both eigenvalues are real with 1 6= 2 6= 0

1 0
J=
0 2
What is M ?
The associated eigenvectors v1 , v2 R21 must satisfy

Av1 = 1 v1
Av2 = 2 v2

Thus,

1 0 1 0
A [v1 |v2 ] = [v1 |v2 ] M = [v1 |v2 ] M 1 AM = =J
0 2 0 2

This represents a change of coordinates

z = M 1 x

and we obtain in the new referential

z1 = 1 z1
z2 = 2 z2
Why?
1 0
z = M 1 x = M 1 Ax = M 1 AM z = z.
0 2

12
1. Introduction to Nonlinear Systems

Case 1: Both eigenvalues are real with 1 6= 2 6= 0

z1 = 1 z1
z2 = 2 z2
For a given initial state (z1 , z2 )(0), the solution is given by

z1 (t) = z1 (0)e1 t
z2 (t) = z2 (0)e2 t

Eliminating time t,

z2 (0)
z2 (t) = z1 (t)2 /1
z1 (0)2 /1
Why?
z1 (t) 1 z1 (t)
= e1 t t = ln
z1 (0) 1 z1 (0)
and so,
2
2 z (t) z (t)
ln z 1 (0) ln z 1 (0) 1
z2 (t) = z2 (0)e 1 1 = z2 (0)e 1 .
At this point several combinations of the eigenvalues can arise...

13
1. Introduction to Nonlinear Systems

Case 1: Both eigenvalues are real with 1 6= 2 6= 0

(a) 1 , 2 < 0. In this case e1 t , e1 t 0 and the curves are parabolic. Consider,
without loss of generality that 2 < 1 .
Phase portrait

The equilibrium point x = 0 is called a stable node.

14
1. Introduction to Nonlinear Systems

Case 1: Both eigenvalues are real with 1 6= 2 6= 0

(b) 1 , 2 > 0. The phase portrait will retain the same character but with the
trajectories directions reversed. In this case the equilibrium point x = 0 is called
an unstable node.
(c) The eigenvalues have opposite signs. Consider for example the case 2 < 0 < 1 .
2
z (t)
ln z 1 (0) 1
z2 (t) = z2 (0)e 1 .
2
The exponent 1
is negative, thus we have hyperbolic curves.

In this case, the equilibrium point is called a saddle point.

15
1. Introduction to Nonlinear Systems

Case 2: Complex eigenvalues 1,2 = j, , R

J=

Associated eigenvectors
v1 = u + jv
v2 = u jv, u, v R2
What is M ?
A (u + jv) = ( + j) (u + jv)
A (u jv) = ( j) (u jv)
Thus,
real part Au = u v
imaginary part Av = u + v
Rearranging we have

A [u|v] = [u|v] =J

which implies that

M = [u|v] M 1 AM =

Coordinate transformation
z = M 1 x, then
z1 = z1 z2,
z2 = z1 + z2

16
1. Introduction to Nonlinear Systems

Case 2: Complex eigenvalues 1,2 = j, , R

Better insight into the solution if we work in polar coordinates,

q
r = z12 + z22 ,

= tan1 zz2 .
1

where in this case we have the following system

r = r
=

The solution is given by

r(t) = r(0)et
(t) = (0) + t

17
1. Introduction to Nonlinear Systems

Case 2: Complex eigenvalues 1,2 = j, , R

These equations represent the logarithmic spiral, and for different values of we get
i) < 0 is a stable focus
ii) > 0 is an unstable focus
iii) = 0 is a center.

18
1. Introduction to Nonlinear Systems

Other cases

a) 1 = 2 = 6= 0

z1 = z1 + kz2 , k = 0 or k = 1
z2 = z2

b) 1 = 0, 2 6= 0

z1 = 0
z2 = z2

c) 1 = 2 = 0

z1 = z2 , (k = 1)
z2 = 0

..
.

19
1. Introduction to Nonlinear Systems

Extensions to nonlinear systems (2nd order)

Consider the 2nd order nonlinear time invariant system

x1 = f1 (x1 , x2 )
x2 = f2 (x1 , x2 )

where f = (f1 , f2 )T is continuously differentiable. Moreover assume that

x1
x =
x2
is an isolated equilibrium point, i.e,
0 = f1 `x1 , x2
`

0 = f2 x 1 , x 2

Expanding the right-hand side into its Taylor series around x

x1 = f1 `x1 , x2 + a11 `x1 x1 + a12 `x2 x2 + H.O.T.

` ` `

x2 = f2 x1 , x2 + a21 x1 x1 + a22 x2 x2 + H.O.T.

where

f1 (x
1 ,x2 ) f1 (x
1 ,x2 )
a11 = x1
a12 = x2


x1 =x1 ,x2 =x2 x1 =x1 ,x2 =x2
f2 (x ,x
1 2

) f2 (x ,x
1 2

)
a21 = x1
a22 = x2
x1 =x x1 =x

1 ,x2 =x2 1 ,x2 =x2

20
1. Introduction to Nonlinear Systems

Extensions to nonlinear systems (2nd order)

Defining z = [z1 , z2 ]T , zi = xi xi , i = 1, 2,

a11 a12
A=
a21 a22

also denoted as the Jacobian Matrix and considering only the first order terms we
obtain the following linear system
z = Az.

The question now is...

What can we conclude about the behavior of the nonlinear system around an
equilibrium point from the study of the linearized system?

21
1. Introduction to Nonlinear Systems

Extensions to nonlinear systems (2nd order)

Consider the 2nd order nonlinear time invariant system

x1 = f1 (x1 , x2 )
x = f (x)
x2 = f2 (x1 , x2 )

where f1 , f2 are analytic (i.e., f1 , f2 have convergent Taylor series representation) and
f (x ) = 0. The linearization around the equilibrium point x = x provides the
following linear system
f
z = Az, A=
x
x=x
where z = x x .

If the origin z = 0 of the linearized state equation is a stable (resp. unstable)

node, or a stable (resp. unstable) focus or a saddle point, then in a small
neighborhood of the equilibrium point, the trajectories of the nonlinear system
will behave like a stable (resp. unstable) node, or a stable (resp, unstable) focus
or a saddle point, respectively.

22
1. Introduction to Nonlinear Systems

Extensions to nonlinear systems (2nd order)

However

If the Jacobian matrix A has eigenvalues on the imaginary axis, then the qualitative
behavior of the nonlinear state equation near the equilibrium point could be quite
distinct from that of the linearized state equation!

Example:
a) x = x3 , x R, is an unstable system
b) x = x3 , x R is a stable system

But they have the same linearization! (i.e., z = 0)

23
1. Introduction to Nonlinear Systems

Example - Pendulum
State model
x1 = x2
x2 = gl sin (x1 ) k
x
m 2
Equilibrium points
(n, 0), n Z

Jacobian matrix

f1 (x1 ,x2 ) f1 (x1 ,x2 )

" #
f x1 x2
0` 1
= =

f2 (x1 ,x2 ) f2 (x1 ,x2 ) gl cos x1 mk

x x=x x1 x2

x1 =x
1 ,x2 =x2

1. x = (0, 0)
s 2
f 0 1 k 1 k 4g
A1 = = 1,2 =
x x gl mk
2m 2 m l
1 =0,x2 =0

2. x = (, 0)
s 2
f 0 1 k 1 k 4g
A2 = = g k 1,2 = +
x x l
m 2m 2 m l
1 =,x2 =0

24
1. Introduction to Nonlinear Systems

Example - Pendulum

g k
Consider the case l
= 1 and m
= 0.5.

1. z1 = x 0
0 1
z1 = A1 z1 , A1 =
1 0.5
Eigenvalues: 1,2 = 0.25 j0.97. Thus the equilibrium point x = (0, 0) is a
stable focus.
2. z2 = x
0 1
z2 = A2 z2 , A2 =
1 0.5
Eigenvalues: 1 = 1.28, 2 = 0.78. Thus the equilibrium point x = (, 0) is a
saddle point.

If k = 0, 1,2 are on the imaginary axis and therefore we cannot determine the
stability of the origin through linearization!

25
1. Introduction to Nonlinear Systems

Example 3
Consider the following system
x1 = x2 ,
x2 = x1 x21 x2
The linearization around x = 0 yields

z1 0 1 z1
=
z2 1 0 z2
where the eigenvalues are j.
Is it a center?
The answer is NO, for  > 0 is stable and for  < 0 is unstable.

How do we conclude its stability?

Tip: check the evolution of the energy
d ` 2
x + x22 = 2x21 x22

dt 1 26
1. Introduction to Nonlinear Systems

Example 4

Consider the following system

x1 = x21
x2 = x22
where R is a small parameter.

a) For > 0 there exist two equilibrium points at ( , 0) and ( , 0).
Performing the linearization we get

2 0
A=
0 1

for ( , 0), which is a stable node and

2 0
A=
0 1

for ( , 0), which is a saddle point.

b) For < 0 there are no equilibrium points.

27
1. Introduction to Nonlinear Systems

Example 4

Phase portraits:

We are in the presence of a bifurcation, that is, a change in the equilibrium points
or periodic orbits or in their stability properties, as a parameter is varied.
In this example we have a saddle-node bifurcation because it results from the
collision of a saddle and a node. is the bifurcation parameter and = 0 is the
bifurcation point.
There exist other types of bifurcations, e.g. transcritical bifurcation,
super/subcritical pitchfork bifurcation and super/subcritical Hopf bifurcation.

28