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.

y

e

y

s + λ p = ( s + ˆ a) s + λ 1 + ˙ˆ a s + λ p

which we substitute in (7.4.19) to obtain

1

1

˙

1

s( m 2) = ( s + ˆ a) s + ˆ b

p

ˆ

1 s + ˆ

p 0)

e

y

b

u

s + λ 1 + ˙ˆ a s + λ p − s + λ p

Using (7.4.10) we have ( s + ˆ a) s + ˆ b

p 1 s + ˆ

p 0) = ( s + 1)2 and therefore

( s + 1)2

1

˙

1

s( m 2) =

e

y

ˆ b

u

s + λ

1 + ˙

ˆ a s + λ p − s + λ p

or

s( s + λ)

s + λ

1

s + λ ˙ 1

e

˙

ˆ

1 =

m 2

ˆ a

y

b

u

( s + 1)2

( s + 1)2 s + λ p + ( s + 1)2 s + λ p

(7.4.20)

˙

Because u

ˆ

p, yp, m,

∈ L∞ and ˙ˆ a, b, m ∈ L∞

L 2, it follows from Corollary 3.3.1

that e 1 ∈ L∞

L 2. Hence, if we show that ˙ e 1 ∈ L∞, then by Lemma 3.2.5 we can

conclude that e 1 0 as t → ∞. Since ˙ e 1 = ˙ yp = ayp + bup ∈ L∞, it follows that e 1 0 as t → ∞.

˙

We can continue our analysis and establish that , ˙ˆ a, ˆ b, ˙ˆ

p 0 , ˙ˆ p 1 0 as t → ∞.

There is no guarantee, however, that ˆ

p 0 , ˆ

p 1, ˆ a, ˆ b will converge to the actual values

p 0 , p 1 , a, b respectively unless the reference signal ym is sufficiently rich of order 2,

which is not the case for the example under consideration.

474

CHAPTER 7. ADAPTIVE POLE PLACEMENT CONTROL

As we indicated earlier the calculation of the controller parameters ˆ

p 0( t) , ˆ

p 1( t)

at each time is possible provided the estimated plant polynomials ( s + ˆ a( t)) , ˆ b( t) are strongly coprime, i.e., provided |ˆ b( t) | ≥ b 0 > 0 ∀t ≥ 0. This condition implies that at each time t, the estimated plant is strongly controllable. This is not surprising

because the control law is calculated at each time t to meet the control objective

for the estimated plant. As we will show in Section 7.6, the adaptive law without

projection cannot guarantee that |ˆ b( t) | ≥ b 0 > 0 ∀t ≥ 0. Projection requires the knowledge of b 0 and sgn( b) and constrains ˆ b( t) to be in the region |ˆ b( t) | ≥

b 0 where controllability is always satisfied. In the higher-order case, the problem

of controllability or stabilizability of the estimated plant is more difficult as we

demonstrate below for a general nth-order plant.

General Case

Let us now consider the nth-order plant

Z

y

p( s)

p =

u

R

p

p( s)

where Zp( s) , Rp( s) satisfy assumptions P1, P2, and P3 with the same control

objective as in Section 7.3.1, except that in this case the coefficients of Zp, Rp

are unknown. The APPC scheme that meets the control objective for the

unknown plant is formed by combining the control law (7.3.12), summarized

in Table 7.1, with an adaptive law based on the parametric model (7.4.2)

or (7.4.3). The adaptive law generates on-line estimates θa, θb of the coeffi-

cient vectors, θ∗a of Rp( s) = sn + θ∗a αn− 1( s) and θ∗b of Zp( s) = θ∗b αn− 1( s) respectively, to form the estimated plant polynomials

ˆ

Rp( s, t) = sn + θa αn− 1( s) , ˆ

Zp( s, t) = θb αn− 1( s)

The estimated plant polynomials are used to compute the estimated con-

troller polynomials ˆ

L( s, t) , ˆ

P ( s, t) by solving the Diophantine equation

ˆ

LQm · ˆ

Rp + ˆ

P · ˆ

Zp = A∗

(7.4.21)

for ˆ

L, ˆ

P pointwise in time or the algebraic equation

ˆ

S ˆ

lβl = α∗

l

(7.4.22)

7.4. INDIRECT APPC SCHEMES

475

for ˆ

βl, where ˆ

Sl is the Sylvester matrix of ˆ

RpQm, ˆ

Zp; ˆ

βl contains the coeffi-

cients of ˆ

L, ˆ

P ; and α∗l contains the coefficients of A∗( s) as shown in Table 7.1.

The control law in the unknown parameter case is then formed as

1

1

up = (Λ ˆ

LQm) u

( y

Λ p − ˆ

P Λ p − ym)

(7.4.23)

Because different adaptive laws may be picked up from Tables 4.2 to 4.5, a

wide class of APPC schemes may be developed. As an example, we present

in Table 7.4 the main equations of an APPC scheme that is based on the

gradient algorithm of Table 4.2.

The implementation of the APPC scheme of Table 7.4 requires that

the solution of the polynomial equation (7.4.21) for ˆ

L, ˆ

P or of the alge-

braic equation (7.4.22) for ˆ

βl exists at each time. The existence of this

solution is guaranteed provided that ˆ

Rp( s, t) Qm( s) , ˆ

Zp( s, t) are coprime at

each time t, i.e., the Sylvester matrix ˆ

Sl( t) is nonsingular at each time t.

In fact for the coefficient vectors l, p of the polynomials ˆ

L, ˆ

P to be uni-

formly bounded for bounded plant parameter estimates θp, the polynomials

ˆ

Rp( s, t) Qm( s) , ˆ

Zp( s, t) have to be strongly coprime which implies that their

Sylvester matrix should satisfy

| det( Sl( t)) | ≥ ν 0 > 0

for some constant ν 0 at each time t. Such a strong condition cannot be

guaranteed by the adaptive law without any additional modifications, giv-

ing rise to the so called “stabilizability” or “admissibility” problem to be

discussed in Section 7.6. As in the scalar case, the stabilizability problem

arises from the fact that the control law is chosen to stabilize the estimated

plant (characterized by ˆ

Zp( s, t) , ˆ

Rp( s, t)) at each time. For such a control law

to exist, the estimated plant has to satisfy the usual observability, controlla-

bility conditions which in this case translate into the equivalent condition of

ˆ

Rp( s, t) Qm( s) , ˆ

Rp( s, t) being coprime. The stabilizability problem is one of

the main drawbacks of indirect APPC schemes in general and it is discussed

in Section 7.6. In the meantime let us assume that the estimated plant is

stabilizable, i.e., ˆ

RpQm, ˆ

Zp are strongly coprime ∀t ≥ 0 and proceed with

the analysis of the APPC scheme presented in Table 7.4.

Theorem 7.4.1 Assume that the estimated plant polynomials ˆ

RpQm, ˆ

Zp are

strongly coprime at each time t. Then all the signals in the closed-loop

476

CHAPTER 7. ADAPTIVE POLE PLACEMENT CONTROL

Table 7.4 APPC scheme: polynomial approach.

yp = Zp( s) u

R

p

p( s)

Plant

Zp( s) = θ∗b αn− 1( s)

Rp( s) = sn + θ∗a αn− 1( s)

αn− 1( s) = [ sn− 1 , sn− 2 , . . . , s, 1] , θ∗p = [ θ∗b , θ∗a ]

Reference

Q

signal

m( s) ym = 0

Gradient algorithm from Table 4.2

˙ θp = Γ φ, Γ = Γ > 0

= ( z − θp φ) /m 2 , m 2 = 1 + φ φ

Adaptive law

α

( s)

α

( s)

φ = [ n− 1

u

n− 1

y

Λ

p, −

p]

p( s)

Λ p( s)

z = sn y

Λ

p,

θp = [ θ

p( s)

b , θa ]

ˆ

Zp( s, t) = θb αn− 1( s) , ˆ

Rp( s, t) = sn + θa αn− 1( s)

Solve for ˆ

L( s, t) = sn− 1 + l αn− 2( s),

ˆ

P ( s, t) = p αn+ q− 1( s) the polynomial equation:

ˆ

L( s, t) ·Qm( s) · ˆ

Rp( s, t) + ˆ

P ( s, t) · ˆ

Zp( s, t) = A∗( s)

or solve for ˆ

βl the algebraic equation

ˆ

S ˆ

lβl = α∗

l

Calculation

where ˆ

Sl is the Sylverster matrix of ˆ

RpQm, ˆ

Zp

ˆ

βl =[ lq , p ] ∈R 2( n+ q) , lq =[0 , . . . , 0 , 1 , l ] ∈Rn+ q q

A∗( s) = s 2 n+ q− 1 + α∗ α 2 n+ q− 2( s)

α∗l = [0 , . . . , 0 , 1 , α∗ ] ∈ R 2( n+ q)

q

Control law

up = (Λ ˆ

LQm) 1 u

( y

Λ p − ˆ

P p − ym)

A∗( s) monic Hurwitz; Λ( s) monic Hurwitz of degree

Design

n + q − 1; for simplicity, Λ( s) = Λ p( sq( s), where

variables

Λ p( s) , Λ q( s) are monic Hurwitz of degree n, q − 1,

respectively

7.4. INDIRECT APPC SCHEMES

477

APPC scheme of Table 7.4 are u.b. and the tracking error converges to zero

asymptotically with time. The same result holds if we replace the gradient

algorithm in Table 7.4 with any other adaptive law from Tables 4.2 and 4.3.

Outline of Proof: The proof is completed in the following four steps as in Example

7.4.1:

Step 1. Manipulate the estimation error and control law equations to express

the plant input up and output yp in terms of the estimation error. This step leads

to the following equations:

˙ x = A( t) x + b 1( t) m 2 + b 2 ¯ ym

up = C 1 x + d 1 m 2 + dym

(7.4.24)

yp = C 2 x + d 3 m 2 + dym

where ¯

ym ∈ L∞; A( t) , b 1( t) are u.b. because of the boundedness of the estimated

plant and controller parameters (which is guaranteed by the adaptive law and the

stabilizability assumption); b 2 is a constant vector; C 1 and C 2 are vectors whose

elements are u.b.; and d 1 to d 4 are u.b. scalars.

Step 2. Establish the e.s. of the homogeneous part of (7.4.24). The matrix A( t)

has stable eigenvalues at each frozen time t that are equal to the roots of A∗( s) = 0.

In addition ˙ θp, ˙ l, ˙ p ∈ L 2 (guaranteed by the adaptive law and the stabilizability

assumption), imply that

˙

A( t) ∈ L 2. Therefore, using Theorem 3.4.11, we conclude

that the homogeneous part of (7.4.24) is u.a.s.

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

boundedness. Let m 2 = 1 + u 2 + y 2 where · denotes the L

f

p

p

2 δ norm. Using

the results established in Steps 1 and 2 and the normalizing properties of mf , we

show that

m 2

2

f ≤ c

mmf

+ c

(7.4.25)

which implies that

t

m 2 f ≤ c

e−δ( t−τ) 2 m 2 m 2 fdτ + c

0

Because m ∈ L 2, the boundedness of mf follows by applying the B-G lemma.

Using the boundedness of mf , we can establish the boundedness of all signals in

the closed-loop plant.

Step 4. Establish that the tracking error e 1 converges to zero. The convergence

of e 1 to zero follows by using the control and estimation error equations to express

e 1 as the output of proper stable LTI systems whose inputs are in L 2 ∩ L∞.

The details of the proof of Theorem 7.4.1 are given in Section 7.7.

478

CHAPTER 7. ADAPTIVE POLE PLACEMENT CONTROL

7.4.3

APPC Schemes: State-Variable Approach

As in Section 7.4.2, let us start with a scalar example to illustrate the design

and analysis of an APPC scheme formed by combining the control law of

Section 7.3.3 developed for the case of known plant parameters with an

adaptive law.

Example 7.4.2 We consider the same plant as in Example 7.3.2, i.e.,

b

yp =

u

s + a p

(7.4.26)

where a and b are unknown constants with b = 0 and up is to be chosen so that the

poles of the closed-loop plant are placed at the roots of A∗( s) = ( s + 1)2 = 0 and

yp tracks the reference signal ym = 1. As we have shown in Example 7.3.2, if a, b

are known, the following control law can be used to meet the control objective:

˙

−a 1

1

ˆ

e =

ˆ

e +

b¯

u

0

0

1

p − Ko([1 0]ˆ

e − e 1)

s + 1

¯

up = −Kcˆ e, up =

¯

u

s

p

(7.4.27)

where Ko, Kc are calculated by solving the equations

det( sI − A + BKc) = ( s + 1)2

det( sI − A + KoC ) = ( s + 5)2

where

−a 1

1

A =

, B =

b, C = [1 , 0]

0

0