Robust Adaptive Control by Petros A. Ioannou, Jing Sun - HTML preview

PLEASE NOTE: This is an HTML preview only and some elements such as links or page numbers may be incorrect.
Download the book in PDF, ePub, Kindle for a complete version.

the tracking error equation

Z

e

pQ 1

1 =

¯

u

R

p

pQm

in the following state-space form

In+ q− 1

˙ e =  −θ∗

1

− − −−

e + θ∗up, e ∈ Rn+ q

(7.7.22)

0

e 1 = C e

where C

= [1 , 0 , . . . , 0] ∈ Rn+ q and θ∗ 1 , θ∗ 2 are the coefficient vectors of RpQm −

sn+ q, ZpQ 1, respectively.

Let eo = e − ˆ e be the state observation error. Then from the equation for ˆ e in

Table 7.5 and (7.7.22), we have

˙ˆ e = Ac( te + ˆ

KoC eo

˙ eo = Aoeo + ˜

θ 1 e 1 ˜

θup

(7.7.23)

where

In+ q− 1

A

o =

−a∗

− − −−

0

520

CHAPTER 7. ADAPTIVE POLE PLACEMENT CONTROL

is a stable matrix; Ac( t) = ˆ

A − ˆ

B ˆ

Kc and ˜

θ 1 = θ 1 − θ∗ 1 , ˜ θ 2 = θ 2 − θ∗ 2. The plant input

and output satisfy

yp = C eo + C ˆ e + ym

R

Q

u

pX Q 1

1 Y Rp

p

=

¯

u

y

A∗

p +

A∗

p

(7.7.24)

where X( s) , Y ( s) are polynomials of degree n + q − 1 that satisfy (7.3.25) and A∗( s) is a Hurwitz polynomial of degree 2( n+ q) 1. Equation (7.7.24) for up is established

in the proof of Theorem 7.3.1 and is used here without proof.

Step 2. Establish e.s. for the homogeneous part of (7.7.23). Let us first

examine the stability properties of Ac( t) in (7.7.23). For each frozen time t, we

have

det( sI − Ac) = det( sI − ˆ

A + ˆ

B ˆ

Kc) = A∗c( s)

(7.7.25)

i.e., Ac( t) is stable at each frozen time t. If ˆ

Zp( s, t) Q 1( s) , ˆ

Rp( s, t) Qm( s) are strongly

coprime, i.e., ( ˆ

A, ˆ

B) is strongly controllable at each time t, then the controller gains

ˆ

Kc may be calculated at each time t using Ackermann’s formula [95], i.e.,

ˆ

Kc = [0 , 0 , . . . , 0 , 1] G− 1

c

A∗c( ˆ

A)

where

Gc = [ ˆ

B, ˆ

A ˆ

B, . . . , ˆ

An+ q− 1 ˆ

B]

is the controllability matrix of the pair ( ˆ

A, ˆ

B). Because ( ˆ

A, ˆ

B) is assumed to be

strongly controllable and ˆ

A, ˆ

B ∈ L∞ due to θp ∈ L∞, we have ˆ

Kc ∈ L∞. Now,

˙ˆ

d

K

˙

c = [0 , 0 , . . . , 0 , 1]

−G− 1

c

GcG− 1

c

A∗c( ˆ

A) + G− 1

c

A∗

dt c( ˆ

A)

Because θp ∈ L∞ and ˙ θp ∈ L 2, it follows that ˙ˆ

Kc( t) ∈ L 2, which, in turn, implies

that

˙

Ac( t) ∈ L 2. From Ac being pointwise stable and

˙

Ac( t) ∈ L 2, we have that

Ac( t) is a u.a.s matrix by applying Theorem 3.4.11. If ˆ

Zp( s, t) Q 1( s) , ˆ

Rp( s, t) Qm( s)

are not strongly coprime but ˆ

Zp( s, t) , ˆ

Rp( s, t) Qm( s) are, the boundedness of ˆ

Kc

and

˙ˆ

Kc( t) ∈ L 2 can still be established by decomposing ( ˆ

A, ˆ

B) into the strongly

controllable and the stable uncontrollable or weakly controllable parts and using

the results in [95] to obtain an expression for ˆ

Kc. Because Ao is a stable matrix the

homogeneous part of (7.7.23) is e.s.

Step 3. Use the properties of the L 2 δ norm and the B-G Lemma to establish

boundedness. As in Example 7.4.2., we apply Lemmas 3.3.3, 3.3.2 to (7.7.23) and

(7.7.24), respectively, to obtain

ˆ

e

≤ c C eo

yp

≤ c C eo + c ˆ e + c ≤ c C eo + c

(7.7.26)

up

≤ c ˆ

e + c yp ≤ c C eo + c

7.7. STABILITY PROOFS

521

where

·

denotes the L 2 δ norm for some δ > 0.

We relate the term C eo with the estimation error by using (7.7.23) to express

C eo as

C eo = C ( sI − Ao) 1(˜

θ 1 e 1 ˜

θ 2 ¯ up)

(7.7.27)

Noting that ( C, Ao) is in the observer canonical form, i.e., C ( sI − Ao) 1 =

αn+ q− 1( s) , we have

A∗( s)

o

¯

n

s¯ n−i

C eo =

θ

A∗

1 ie 1 ˜

θ 2 i ¯ up) ,

¯

n = n + q − 1

i=0

o

where ˜

θi = [˜

θi 1 , ˜

θi 2 , . . . , ˜

θi¯ n] , i = 1 , 2. Applying Swapping Lemma A.1 to each

term under the summation, we have

s¯ n−i ˜

Λ

s¯ n−i

θ

p( s) Q 1( s)

˜

θ

e

A∗

1 ie 1 =

1 i

1 + Wci( s) ( Wbi( s) e 1) ˙˜

θ 1 i

o( s)

A∗o( s)

Λ p( s) Q 1( s)

and

s¯ n−i ˜

Λ

s¯ n−i

θ

p( s) Q 1( s)

˜

θ

¯

u

A∗

2 i ¯

up =

2 i

p + Wci( s) ( Wbi( s

up) ˙˜

θ 2 i

o( s)

A∗o( s)

Λ p( s) Q 1( s)

where Wci, Wbi, i = 0 , . . . , n + q − 1 are transfer matrices defined in Lemma A.1

with W ( s) =

s¯

n−i

. Therefore, C e

Λ

o can be expressed as

p( s) Q 1( s)

Λ

¯

n

s¯ n−i

s¯ n−i

C e

p( s) Q 1( s)

˜

o

=

θ

e

¯

u

A∗

1 i

1 ˜

θ 2 i

p

+ r 1

o( s)

Λ

Λ

i=0

p( s) Q 1( s)

p( s) Q 1( s)

Λ

α

α

=

p( s) Q 1( s)

˜

θ

n+ q− 1( s) e

n+ q− 1( s) ¯ u

A∗

1

1 ˜

θ 2

p

+ r 1 (7.7.28)

o( s)

Λ p( s) Q 1( s)

Λ p( s) Q 1( s)

where

Λ

¯

n

r

p( s) Q 1( s)

1 =

W

A∗

ci( s)[( Wbi( s) e 1) ˙˜

θ 1 i − ( Wbi( sup) ˙˜

θ 2 i]

o( s)

i=0

From the definition of ˜

θ 1, we have

˜

θ 1 αn+ q− 1( s) = θ 1 αn+ q− 1( s) − θ∗ 1 αn+ q− 1( s)

=

ˆ

Rp( s, t) Qm( s) − sn+ q − Rp( s) Qm( s) + sn+ q

= ( ˆ

Rp( s, t) − Rp( s)) Qm( s) = ˜

θa αn− 1( s) Qm( s)

(7.7.29)

where ˜

θa = θa − θ∗a is the parameter error. Similarly,

˜

θ 2 αn+ q− 1( s) = ˜ θb αn− 1( s) Q 1( s)

(7.7.30)

522

CHAPTER 7. ADAPTIVE POLE PLACEMENT CONTROL

where ˜

θb = θb − θ∗. Using (7.7.29) and (7.7.30) in (7.7.28), we obtain

b

Λ

Q

1

1

C e

p( s) Q 1( s)

˜

m( s)

o

=

θ

e

¯

u

A∗

a αn− 1( s)

1 ˜

θb αn− 1( s)

p

+ r 1

o( s)

Q 1( s) Λ p( s)

Λ p( s)

Λ

Q

Q

=

p( s) Q 1( s)

˜

θ

m( s)

y

m( s)

u

A∗

a αn− 1( s)

p − ˜

θb αn− 1( s)

p

o( s)

Q 1( sp( s)

Λ p( s) Q 1( s)

+ r 1

(7.7.31)

where the second equality is obtained using

Qm( s) e 1 = Qm( s) yp,

Q 1( sup = Qm( s) up

Noting that

1

1

αn− 1( s)

u

y

Λ

p = φ 1 ,

αn− 1( s)

p = −φ 2

p( s)

Λ p( s)

we use Swapping Lemma A.1 to obtain the following equalities:

Qm( s) ˜

Q

θ

m( s)

y

˙˜ θ

Q

a φ 2 = ˜

θa αn− 1( s)

p + Wcq Wbq ( s) φ 2

a

1( s)

Q 1( sp( s)

Qm( s) ˜

Q

θ

m( s)

u

˙˜ θ

Q

b φ 1 = ˜

θb αn− 1( s)

p + Wcq Wbq ( s) φ 1

b

1( s)

Q 1( sp( s)

where Wcq, Wbq are as defined in Swapping Lemma A.1 with W ( s) = Qm( s) . Using

Q 1( s)

the above equalities in (7.7.31) we obtain

Λ

C e

p( s) Qm( s) ˜

o =

θ

A∗

p φ + r 2

(7.7.32)

o( s)

where

Λ

r

p( s) Q 1( s)

˙˜

˙˜

2 = r 1 +

W

θ

θ

A∗

cq ( s) Wbq ( s) φ 1

b + Wcq ( s) Wbq ( s) φ 2

a

o( s)

From Table 7.5, the normalized estimation error satisfies the equation

m 2 = ˜

θp φ

which can be used in (7.7.32) to yield

Λ

C e

p( s) Qm( s)