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.

sn

z =

y( n) =

y

Λ( s)

Λ( s)

α

α

φ =

n− 1( s) u, − n− 1( s) y

Λ( s)

Λ( s)

and

Λ( s) = sn + λn− 1 sn− 1 + · · · + λ 0

50

CHAPTER 2. MODELS FOR DYNAMIC SYSTEMS

is an arbitrary Hurwitz polynomial in s. It is clear that the scalar signal z and

vector signal φ can be generated, without the use of differentiators, by simply

filtering the input u and output y with stable proper filters si , i = 0 , 1 , . . . n.

Λ( s)

If we now express Λ( s) as

Λ( s) = sn + λ αn− 1( s)

where λ = [ λn− 1 , λn− 2 , . . . , λ 0] , we can write

sn

Λ( s) − λ α

α

z =

y =

n− 1( s) y = y − λ

n− 1( s) y

Λ( s)

Λ( s)

Λ( s)

Therefore,

α

y = z + λ

n− 1( s) y

Λ( s)

Because z = θ∗ φ = θ∗ 1 φ 1 + θ∗ 2 φ 2, where

θ∗ 1 = [ bn− 1 , bn− 2 , . . . , b 0] , θ∗ 2 = [ an− 1 , an− 2 , . . . , a 0]

α

α

φ

n− 1( s)

n− 1( s)

1 =

u, φ

y

Λ( s)

2 =

Λ( s)

it follows that

y = θ∗ 1 φ 1 + θ∗ 2 φ 2 − λ φ 2

Hence,

y = θ∗λ φ

(2.4.9)

where θ∗λ = [ θ∗ 1 , θ∗ 2 − λ ] . Equations (2.4.8) and (2.4.9) are represented

by the block diagram shown in Figure 2.2.

A state-space representation for generating the signals in (2.4.8) and

(2.4.9) may be obtained by using the identity

[adj( sI − Λ c)] l = αn− 1( s)

where Λ c, l are given by

−λn− 1 −λn− 2 · · · −λ 0

1

1

0

· · ·

0

 0 

Λ

c = 

.

.

.

,

l =  . 

..

. .

.. 

 .. 

0

· · ·

1

0

0

2.4. PLANT PARAMETRIC MODELS

51

φ

u αn− 1( s)

1

y

θ∗

+ ❧

Σ

Λ( s)

1

+

❆❑

φ 2 −α

θ∗

n− 1( s)

2

Λ( s)

λ

+ +

z

Σ

Figure 2.2 Plant Parameterization 1.

which implies that

α

det( sI − Λ

n− 1( s)

c) = Λ( s) ,

( sI − Λ c) 1 l = Λ( s)

Therefore, it follows from (2.4.8) and Figure 2.2 that

˙ φ 1 = Λ 1 + lu, φ 1 ∈ Rn

˙ φ 2 = Λ 2 − ly, φ 2 ∈ Rn

y = θ∗λ φ

(2.4.10)

z = y + λ φ 2 = θ∗ φ

Because Λ( s) = det( sI − Λ c) and Λ( s) is Hurwitz, it follows that Λ c is a

stable matrix.

The parametric model (2.4.10) is a nonminimal state-space representa-

tion of the plant (2.4.3). It is nonminimal because 2 n integrators are used

to represent an n th-order system. Indeed, the transfer function Y ( s) /U ( s)

computed using (2.4.10) or Figure 2.2, i.e.,

Y ( s)

Z( s) Λ( s)

Z( s)

=

=

U ( s)

R( s) Λ( s)

R( s)

involves n stable zero-pole cancellations.

The plant (2.4.10) has the same I/O response as (2.4.3) and (2.4.1)

provided that all state initial conditions are equal to zero, i.e., x 0 = 0,

φ 1(0) = φ 2(0) = 0. In an actual plant, the state x in (2.4.1) may represent

physical variables and the initial state x 0 may be different from zero. The

52

CHAPTER 2. MODELS FOR DYNAMIC SYSTEMS

effect of the initial state x 0 may be accounted for in the model (2.4.10) by

applying the same procedure to equation (2.4.2) instead of equation (2.4.3).

We can verify (see Problem 2.9) that if we consider the effect of initial con-

dition x 0, we will obtain the following representation

˙ φ 1 = Λ 1 + lu, φ 1(0) = 0

˙ φ 2 = Λ 2 − ly, φ 2(0) = 0

y = θ∗λ φ + η 0

(2.4.11)

z = y + λ φ 2 = θ∗ φ + η 0

where η 0 is the output of the system

˙ ω = Λ cω,

ω(0) = ω 0

η 0 = C 0 ω

(2.4.12)

where ω ∈ Rn, ω 0 = B 0 x 0 and C 0 ∈ Rn, B 0 ∈ Rn×n are constant matrices

that satisfy C 0 { adj( sI − Λ c) }B 0 = C { adj( sI − A) }.

Because Λ c is a stable matrix, it follows from (2.4.12) that ω, η 0 converge

to zero exponentially fast. Therefore, the effect of the nonzero initial condi-

tion x 0 is the appearance of the exponentially decaying to zero term η 0 in

the output y and z.

Parameterization 2

Let us now consider the parametric model (2.4.9)

y = θ∗λ φ

and the identity Wm( s) W − 1

m ( s) = 1, where Wm( s) = Zm( s) /Rm( s) is a

transfer function with relative degree one, and Zm( s) and Rm( s) are Hurwitz

polynomials. Because θ∗λ is a constant vector, we can express (2.4.9) as

y = Wm( s) θ∗λ W − 1

m ( s) φ

If we let

1

α

α

ψ =

φ =

n− 1( s)

u, −

n− 1( s)

y

Wm( s)

Wm( s)Λ( s)

Wm( s)Λ( s)

2.4. PLANT PARAMETRIC MODELS

53

ψ

u

αn− 1( s)

1

y

θ∗

+ ❧

Σ

Wm( s)

Λ( s) W

1

m( s)

+

❆❑

ψ 2

−αn− 1( s)

θ∗ 2

Λ( s) Wm( s)

λ

Figure 2.3 Plant Parameterization 2.

we have

y = Wm( s) θ∗λ ψ

(2.4.13)

Because all the elements of αn− 1( s) are proper transfer functions with

Λ( s) Wm( s)

stable poles, the state ψ = [ ψ 1 2 ] , where

α

α

ψ

n− 1( s)

n− 1( s)

1 =

u, ψ

y

W

2 =

m( s)Λ( s)

Wm( s)Λ( s)

can be generated without differentiating y or u. The dimension of ψ depends

on the order n of Λ( s) and the order of Zm( s). Because Zm( s) can be

arbitrary, the dimension of ψ can be also arbitrary.

Figure 2.3 shows the block diagram of the parameterization of the plant

given by (2.4.13). We refer to (2.4.13) as Parameterization 2. In [201],

Parameterization 2 is referred to as the model reference representation and

is used to design parameter estimators for estimating θ∗λ when Wm( s) is a

strictly positive real transfer function (see definition in Chapter 3).

A special case of (2.4.13) is the one shown in Figure 2.4 where

1

Wm( s) = s + λ 0

and ( s + λ 0) is a factor of Λ( s), i.e.,

Λ( s) = ( s + λ 0)Λ q( s) = sn + λn− 1 sn− 1 + · · · + λ 0

where

Λ q( s) = sn− 1 + qn− 2 sn− 2 + · · · + q 1 s + 1

54

CHAPTER 2. MODELS FOR DYNAMIC SYSTEMS

ψ

u

αn− 1( s)

1

y

θ∗

+ ❧

Σ

1

Λ

1

s+ λ 0

q ( s)

+

❆❑

ψ 2

−αn− 1( s)

θ∗ 2

Λ q( s)

λ

Figure 2.4

Plant Parameterization 2 with Λ( s) = ( s + λ 0)Λ q( s) and

Wm( s) = 1 .

s+ λ 0

The plant Parameterization 2 of Figure 2.4 was first suggested in [131],

where it was used to develop stable adaptive observers. An alternative para-

metric model of the plant of Figure 2.4 can be obtained by first separating

the biproper elements of αn− 1( s) as follows:

Λ q( s)

For any vector c = [ cn− 1 , cn− 2 , . . . , c 1 , c 0] ∈ Rn, we have

c αn− 1( s)

c

¯ c α

= n− 1 sn− 1 +

n− 2( s)

(2.4.14)

Λ q( s)

Λ q( s)

Λ q( s)

where ¯ c = [ cn− 2 , . . . , c 1 , c 0] , αn− 2 = [ sn− 2 , . . . , s, 1] . Because Λ q( s) =

sn− 1 + ¯

q αn− 2( s), where ¯

q = [ qn− 2 , . . . , q 1 , 1] , we have sn− 1 = Λ q( s)

¯

q αn− 2, which after substitution we obtain

c αn− 1( s)

c − c

= c

n− 1 ¯

q) αn− 2( s)

(2.4.15)

Λ

n− 1 +

q( s)

Λ q( s)

We use (2.4.15) to obtain the following expressions:

α

α

θ∗

n− 1( s)

n− 2( s)

1

u = b

u

Λ

n− 1 u + ¯

θ∗ 1

q( s)

Λ q( s)

α

α

( θ∗

n− 1( s)

n− 2( s)

2

− λ )

y = ( λ

y (2.4.16)

Λ

n− 1 − an− 1) y − ¯

θ∗ 2

q( s)

Λ q( s)

where ¯

θ∗ 1 = ¯ b − bn−q, ¯ θ∗ 2 = ¯ a − ¯ λ − ( an− 1 − λn− 1)¯ q and ¯ a = [ an− 2,

. . . , a 1 , a 0] , ¯ b = [ bn− 1 , . . . , b 1 , b 0] , ¯ λ = [ λn− 2 , . . . , λ 1 , λ 0] . Using (2.4.16), Figure 2.4 can be reconfigured as shown in Figure 2.5.

2.4. PLANT PARAMETRIC MODELS

55

¯

ψ

1

¯

x

u

αn− 2( s)

1

¯

+

1 = y

θ∗

✲ ❧

Σ

Λ

1

+ +

s + λ

q ( s)

0

✁✕+✻

❆❑

¯

ψ

2

−α

b

¯

n− 2( s)

n− 1

θ∗ 2

Λ q( s)

λn− 1 −an− 1

Figure 2.5 Equivalent plant Parameterization 2.

A nonminimal state space representation of the plant follows from Fig-

ure 2.5, i.e.,

˙¯ x 1 = −λx 1 + ¯ θ∗ ¯

ψ,

¯

x 1 ∈ R 1

˙¯

ψ

¯

1

= ¯

Λ 1 + ¯ lu,

¯

ψ 1 ∈ Rn− 1

˙¯

ψ

¯

2

= ¯

Λ 2 ¯ ly,

¯

ψ 2 ∈ Rn− 1

(2.4.17)

y = ¯

x 1

where ¯

θ∗ = [ bn− 1 , ¯

θ∗ 1 , λn− 1 − an− 1 , ¯ θ∗ 2 ] , ¯

ψ = [ u, ¯

ψ 1 , y, ¯