Nonlinear Systems and Control

Lecture # 38

Observers

Exact Observers

p. 1/1
 Observer with Linear Error Dynamics

Observer Form:

x = Ax + (y, u), y = Cx

where (A, C) is observable, x Rn , u Rm , y Rp

From Lecture # 24: An n-dimensional SO system

x = f (x) + g(x)u, y = h(x)

is transformable into the observer form if and only if

h iT  
n1
= h, Lf h, Lf h , rank (x) = n
x
h iT
b= 0, 0, 1 , =b
x
p. 2/1
 [adif , adjf ] = 0, 0 i, j n 1

[g, adjf ] = 0, 0j n2
Change of variables:

i = (1)i1 adi1
f , 1in

T h i
1 , 2 , n =I
x
z = T (x)

p. 3/1
 x = Ax + (y, u), y = Cx

x = Ax + (y, u) + H(y C x)
x = x x
x = (A HC)x
Design H such that (A HC) is Hurwitz
What about feedback control?
Let u = (x) be a globally stabilizing state feedback
control
u = (x)

x = Ax + (y, u) + H(y C x)

p. 4/1
How would you analyze the closed-loop system?

x = Ax + (Cx, (x x))
x = (A HC)x

We know that
the origin of x = Ax + (Cx, (x)) is globally
asymptotically stable
the origin of x = (A HC)x is globally exponentially
stable
What additional assumptions do we need to show that the
origin of the closed-loop system is globally asymptotically
stable?

p. 5/1
 Circle Criterion Design

x = Ax + (y, u) L(M x), y = Cx

where (A, C) is observable, x Rn , u Rm , y Rp ,

M x R , () = [ 1 (1 ), . . . , ( ) ]T

x = Ax + (y, u) L(M x N (y C x)) + H(y C x)

x = x x
x = (A HC)x L[(M x) (M x N (y C x))]

x = (A HC)x L[(M x) (M x (M + N C)x)]

Define
z = (M + N C)x
(t, z) = (M x(t)) (M x(t) z)
p. 6/1
 x = (A HC)x L(t, z)
z = (M + N C)x

def
G(s) = (M + N C)[sI (A HC)]1 L

0 +
- n -
z -
G(s)
6

() 
h iT
(t, z) = 1 (t, z1 ), . . . , (t, z )

p. 7/1
Main Assumption: i () is a nondecreasing function

(a b)[i (a) i (b)] 0, a, b R

If i (i ) is continuously differentiable
di
0, i R
di

zi i (t, zi ) = zi [i ((M x)i ) i ((M x)i zi )] 0

z T (t, z) 0

p. 8/1
By the circle criterion (Theorem 7.1) the origin of

x = (A HC)x L(t, z)
z = (M + N C)x

is globally exponentially stable if

def
G(s) = (M + N C)[sI (A HC)]1 L

is strictly positive real

Design Problem: Design H and N such that G(s) is

strictly positive real
Feasibility can be investigated using LMI (Arcak &
Kokotovic, Automatica, 2001)
p. 9/1
Example:

x1 = x2 , x2 = x31 x32 + u, y = x1
" # " #
0 1 h i 0
A= , C= 1 0 , = 3
,
0 0 y + u
" #
0 h i
3 d
L= , M = 0 1 , () = , = 3 2 0
1 d
" #
h1
h= , N
h2

1 s + N + h1
G(s) = (M + N C)[sI (A HC)] L=
s2 + h1 s + h2

p. 10/1
From Exercise 6.7, G(s) is SPR if and only if

h1 > 0, h2 > 0, 0 < N + h1 < h1

1
h1 = 2, h2 = 1, N = 2
3
s+ 2
G(s) =
(s + 1)2

x 1 = x2 + 2(y x1 )

3
x 2 = y + u x2 +
3 1
2 (y x1 ) + (y x1 )

p. 11/1
What about feedback control?
Let u = (x) be a globally stabilizing state feedback control
Closed-loop system under output feedback:

x = Ax + (y, (x x)) L(M x)

x = (A HC)x L(t, z)
z = (M + N C)x

How would you analyze the closed-loop system?

(t, z) depends on x(t). How would you show that is

well defined?

What about the effect of uncertainty?

p. 12/1