10/6/2015

(3) Delta Method

Convenient when the second order system is likely
to be oscillatory with low damping.
Method consists in approximating the trajectory
with short arcs of circles. ..
.
The differential equation: y + d ( y , y ) + y = 0
is arranged in form the state equations:
.

x1 = x 2
x 2 = (x 1 + d ( x 1 , x 2 ) )
.

10/6/2015

Slope equation:

dx 2 (x1 + d ( x1 , x2 ) )
=
dx1
x2

Cross multiplying: x 2 dx2 + (x1 + d ( x1 , x 2 ) )dx1 = 0

If the function d ( x1 , x2 ) is sufficiently
,it can be
taken as constant for small changes in operating points
Then integrate the equation:
We get:

x 2 dx2 + (x1 + d (.) )dx1 = 0

x 22 + (x1 + d (.) ) = R 2
2

This is equation to a circle with center at ( d (.), 0 )

and radius = R

10/6/2015

Part of the trajectory can be drawn as circle with

the centre at ( d , 0 ) and radius R.
Construction requires evaluation of d at different
points in state plane.
Drawing arcs of circles from the centers at ( d , 0 )
and appropriate radius.
Following procedure can be used for construction of
the trajectory:
1) Identify the term to be include in function d ( x1 , x2 )
2) Plot d ( x1 , x 2 ) vs. x1 or x 2 as case may be.
0
0
3) From a given initial state P1 , ( x1 , x 2 )
oject lines
to determine d1 and set this length against the
negative real axis locating ( d 1 , 0 ) .

10/6/2015

4) With centre at ( d1 , 0 ) and radius determined by the

position of the initial state, drew an arc of a circle
forming a part of trajectory.
5) Repeat Steps 3 and 4 at the new point on the
trajectory and complete the trajectory repeating these
steps.

Figure for Delta method of Phase plane trajectory

construction

10/6/2015

Example of a hydraulic speed control system :

..

e + 0 . 33 e + (1 + 0 . 58 e 2 ) e = 0
Defining the error, e(t) and its derivative as the state
variables:
.

x1 = x 2
.

x 2 = x 1 0 . 33 x 2 0 . 58 x 13

3
d can be identified as 0 .58 x1 + 0 .33 x 2 which
can be split into two terms:

d 1 ( x1 ) = 0 .58 x13

and

d 2 ( x 2 ) = 0 .33 x 2

10/6/2015

d 1 ( x1 ) is plotted against x1 .
d 2 ( x 2 ) is plotted against x 2 .
(1 )
From an initial condition point, values of d 2 ( x 2 )
and d 1(1) ( x1 ) are measured.
These are laid out towards the negative x1 axis as

d (1) = d 1(1) + d 2(1)

With point ( d (1) , 0 ) as centre and radius equal to
distance between the centre and initial condition
point, an arc of a circle is drawn .
The procedure is repeated at other points as well to
completer the trajectory.
.

10/6/2015

Limit Cycles
A limit cycle is an isolated closed trajectory.
Discovery by H. Poincare

Examples:
? Heart beat, pacemaker, neurons, daily rhythms in
body temperature, chemical reactions, ...

10/6/2015

A limit cycle is an isolated closed trajectory;

this means that its neighboring trajectories are not
closed.
they spiral either towards or away from the limit cycle.
Thus, limit cycles can only occur in nonlinear systems.
In a linear system exhibiting oscillations closed
trajectories are neighbored by other closed trajectories
A stable limit cycle is one which attracts all
neighboring trajectories.
A system with a stable limit cycle can exhibit selfsustained oscillations most of the biological processes
of interest are of this kind.

10/6/2015

Neighbouring trajectories are repelled from

unstable limit cycles.
half-stable limit cycles are ones which attract
trajectories from one side and repel those on the
other.
Trajectory of Limit Cycle has to be both closed,
indicating the periodic nature of motion, and
Isolated , indicating the limiting nature of the cycle
with neighboring trajectories converging to or
diverging from it.

10/6/2015

We have seen examples of linear and non-linear

systems which have a family of periodic solutions.
These occur near any centre of the system.
Any such periodic trajectory is called a cycle. There
is a type of cycle which is isolated i.e. it is not part
of a family of cycles.
We are interested in the question of stability of such
cycles.
That is, do trajectories which are near to a cycle,
approach it or leave it as t? 8 ?

10

10/6/2015

If all trajectories which are sufficiently close to a

cycle, converge to that cycle as t? 8 ,
we call the cycle an asymptotically stable limit cycle
of the system.
If no trajectories which are sufficiently close to a
cycle, diverge from that cycle as t? 8 ,
we call the cycle an unstable limit cycle of the
system.
If some trajectories which are sufficiently close to a
cycle, converge to that cycle as t? 8 and others
diverge from that cycle,
we call the cycle a semi-stable limit cycle of the
system.

11

10/6/2015

Are inherently a non-linear phenomenon

? What about .closed orbits in linear systems?

X = AX

They can have solutions that

are closed orbits!
? Linearity implies: if x(t) is a solution so is c x(t)
? Amplitude is determined by Initial conditions
? In contrast: in limit cycles in nonlinear systems, the
structure of the system determines the amplitude
and shape of the limit cycle.

12

10/6/2015

Oscillation: A system oscillates when it has a

nontrivial periodic solution.

x ( t ) = x ( t + T );

for all t 0

The linear oscillation is not practical because :

It is not structurally stable.
Infinitesimally small perturbations may change the
type of the equilibrium point to a stable focus
(decaying oscillation) or unstable focus (growing
oscillation)
The same problems exist with oscillation of
nonlinear systems due to a center equilibrium point
(e.g., pendulum without friction)

13

10/6/2015

Van der Pol Oscillator

..

y+ m 1 y

)y +
.

y = 0

Equation arose in connection with

nonlinear electrical circuits used.
Damping term
Causes large amplitude oscillates
to decay, but pumps them back if
they become too small
? Van der Pol Oscillator has a unique
stable Limit cycle for each m > 0

14

10/6/2015

Poincar-Bendixson Theorem
As closed orbits are of much interest there
should be a method to establish that there
exists such an orbit, and there is: PoincarBendixson Theorem:
Suppose that:
1. R is a closed bounded subset of the plane;
2. dx/dt = f(x) (state equations) is a
continuously differentiable vector field on
an open set containing R;
3. R does not contain any Fixed Points (Singular
points);
4. There exists a trajectory C that is confined in
R, in the sense that it starts in R and stays
in R for all future time.
Then either C is a closed orbit, or it spirals
towards closed orbit as time tends to
infinity .
So, R contains a closed orbit.

15

10/6/2015

Bendixsons Criteria for the existence of the limit

cycle:
Bendixson has shown that a sufficient condition for
the absence of the limit cycle or any sort of closed
trajectory within a region is that ,
Given the state equation of a non-linear system as:
.

x1 = f1 ( x1 , x 2 )
.

x 2 = f 2 ( x1 , x 2 )
Then on the boundary of the region:

f 1 ( x1 , x 2 ) +
f 2 ( x1 , x 2 ) = fixed sign
x1
x 2

16

10/6/2015

Proof: Slope equation of trajectory is given as:

dx 2
f 2 ( x1 , x 2 )
=
dx 1
f 1 ( x1 , x 2 )
on
cross multplying
f 1 dx 2 f 2 dx 1 = 0

This equation gives relation that must exit along any

solution trajectory in the state plane.
Gauss Theorem states that a surface integrals over a
a complete domain R is related to a line integral
around the closed path C which is boundary of
region:

17

10/6/2015

f
+
f
R x1 1 x 2 2 dx 1 dx 2 =

( f dx
1

+ f 2 dx 1 )

If a limit cycle C is assumed to be bounding the

region R, Then,
The RHS of the Gauss Theorem is zero since the
intergral is zeros everywhere on C.
Thus , the LHS must be zero
Consequently,

f1 +
f2
x1
x2

must change sign with R.

18

10/6/2015

Example:
Consider system represented by the following
differential equation:
2
d 2 y
10 dy dy
2

0
.
1

+
y
+
y
=0

dt
3 dt dt

State space representation cab be given as:

dx 1
= x2
dt
dx 2
10
= 0 .1
(x 2 )2 x 2 x1 x12 = 0
dt
3

so ,

f 1 ( x1 , x 2 ) = x 2 ;

19

10/6/2015

10

f 2 ( x1 , x 2 ) = 0 . 1
(x 2 )2 x 2 x1 x12
3

Thus ,
f 1 ( x1 , x 2 ) = 0
x1
and

and
f 2 ( x1 , x 2 ) = 0 . 1 10 x 22
x2

f 1 ( x1 , x 2 ) +
f 2 ( x1 , x 2 )
It is a apparent that
x1
x2
does not have fixed sign.
By Bendixsons criteria, limit cycle may exit.

20

10/6/2015

Concept of Feedback Linearization

Consider a class of non-linear system in the form of:
d
x = f ( x ) + G ( x )u
dt
y = h( x )
If there exit a state feedback control :
u = a ( x ) + b ( x )v
and a change of state variables, z = T ( x ) ,
That transform the non-linear system into an
equivalent linear system.
Then the classical linear control techniques can be
applied to resulting closed loop system.

21

10/6/2015

Feedback Linearization can possible in two forms:

(1) Input-state linearization
In this the full state equation is linearized
(2) Input-Output Linearization
In this case emphasis is given on linearizing the
input-output map from u to y, even if the state
equation is only partially linearized.

Input-state linearization
Consider example of Pendulum system, having
following differential equation:
d 2q
dq
=

a
sin
q

b
+ cT
2
dt
dt

22

10/6/2015

Where T is applied torque and q is angle subtended by

rod and the vertical axis.
Objective is to stabilize the pendulum at an angle

q =d

Choosing the state variables as x1 = q d ; x 2 = q

So that desired equilibrium point is located at the origin
of the state space
The state equation is obtain as:

dx 1
dx 2
= x2 ;
= a (sin( x1 + d ) ) bx 2 + cT
dt
dt

For system to maintain equilibrium at the origin ,

torque must have steady state component Tss, that
satisfies

0 = a (sin d ) + cT ss

23

10/6/2015

Define a control variable u as: u = T T ss

Rewrite state equation:
dx 1
= x2;

dt
dx 2
= a (sin( x 1 + d ) sin d ) bx 2 + cu
dt

From inspection of the state equation shows that if

we choose u as below to cancel nonlinear term:

u=

a
(sin( x1 + d ) sin d ) + v
c
c

results in:
dx 1
dt

= x2;

dx 2
= bx 2 + v
dt

24

10/6/2015

Thus stabilization problem for non linear system has

been converted to stabilization problem for a
controllable linear system
We can now proceed to design a stabilizing state
feedback control law as:
v = k 1 x1 + k 2 x 2
To locate the eigen values of the closed loop
system:
dx 1
dx 2
= x2;
= k 1 x1 + ( k 2 b ) x 2
dt
dt
In the left half of s-plane. The overall state feedback
control law is given by:

25

10/6/2015

u =

a
(sin( x1 + d ) sin d ) + 1 (k 1 x1 + k 2 x 2 )
c
c

Generalization of non-linearity cancellation

Not to except that we can cancel nonlinearities in
every nonlinear system.
Certain structural property of the system that allow
to perform such cancellation.
To cancel a nonlinear term a ( x ) by subtraction,
the control u and the nonlinearity a ( x ) must
appear together as a sum u + a ( x )
To cancel a nonlinear term g ( x ) by division,
control u and the nonlinearity g ( x ) must appear
together as a product g ( x ) u .

26

10/6/2015

If the matrix g ( x ) is non singular in the domain of

interest , then it can be cancelled by u = b ( x ) v
1
where b ( x ) = l ( x ) .
Ability to use feedback to convert a nonlinear state
equation into a controllable linear state equation by
cancelling nonlinearities requires the nonlinear
state equation to have the structure:
d
x = Ax + B b 1 ( x ) [u a ( x ) ]
dt
Where A = n X n , B = n X p , pair (A,B) is controllable
Functions
a (x) : R n
R p ; b ( x ) : R n
R pXp

27

10/6/2015

are defined in the domain D x R n that contains

the origin.
If the state equation takes the form as above, then
one can linearize it via the state feedback law:
u = a ( x ) + b ( x )v
To obtain the linear state equation:
d
x = Ax + Bv
dt
For stabilization, one design v = Kx such that A + BK
is Hurwitz . The overall nonlinear stabilizing state
feed back control law is:
u = a ( x ) + b ( x ) Kx

28

10/6/2015

For Pendulum stabilization problem having the state

equation as:

dx 1
= x2;
dt
dx 2
= a (sin( x1 + d ) sin d ) bx 2 + cu .
dt
System state equation can be rearranged in matrix
form as:

x. 0 1 x 0 a

1
.1 =
+
c
u

(
sin(
x
+
d
)

sin
d
)

x2 0 b x2 1

29

10/6/2015

Comparing with structure,

d
x = Ax + B b 1 ( x ) [u a ( x ) ]
dt
we get
0
A =
0
a (x) =

1
;

b
a
c

(sin(

0
B = ;
1
x 1 + d ) sin d

b (x) =

1
;
c

For Linearization, control law used :

u = a ( x ) + b ( x )v

30

10/6/2015

If a non-linear state equation does not have the

structure as above for one choice of state variables,
it might do so for
another choice. .
.
x 2 = x12 + u
For example: x1 = a sin x 2 ;
One can cannot cancel nonlinear term a sin x2
But if change variable by the. transformation:

z 1 = x1 and z 2 = x1 = a sin x 2

So new
. state equation:
.
z 1 = z 2 ; z 2 = ax 12 cos x 2 + au cos x 2
Nonlinearities can be cancelled by the control:

1
u= x +
v
a cos x 2
2
1

31

10/6/2015

State equation in new coordinates (z1, z2) can be found

by transforming to express (x1,x2) in terms of (z1,z2).
Transformed state equation:
.

z1 = z 2 ;

z 2 = a cos sin

z2
2

( z 1 + u )
a

Transformation from one x-coordinates to zcoordinates is done through z = T ( x )

The map T must be invertible. An inverse map T 1 (.)
1
x
=
T
( z ) for all z T ( D x )
must exit such that
where Dx is the domain of T.
Since the derivatives of z and x should be continuously ,
both T(.) and inverse of T be continuously
differentiable.
A continuously differentiable map with a continuously
differentiable inverse is known diffeomorphism.

32