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.

MRC Schemes: Known Plant Parameters

In addition to assumptions P1 to P4 and M1, M2, let us also assume

that the plant parameters, i.e., the coefficients of Gp( s) are known exactly.

Because the plant is LTI and known, the design of the MRC scheme is

achieved using linear system theory.

The MRC objective is met if up is chosen so that the closed-loop transfer

function from r to yp has stable poles and is equal to Wm( s), the transfer

function of the reference model. Such a transfer function matching guaran-

tees that for any reference input signal r( t), the plant output yp converges

to ym exponentially fast.

A trivial choice for up is the cascade open-loop control law

k Z

R

u

m

m( s)

p( s)

p = C( s) r,

C( s) =

(6.3.6)

kp Rm( s) Zp( s)

which leads to the closed-loop transfer function

yp

k Z R k

= m m

p

pZp = W

r

k

m( s)

(6.3.7)

p Rm Zp Rp

This control law, however, is feasible only when Rp( s) is Hurwitz. Other-

wise, (6.3.7) may involve zero-pole cancellations outside C−, which will lead

to unbounded internal states associated with non-zero initial conditions [95].

In addition, (6.3.6) suffers from the usual drawbacks of open loop control

such as deterioration of performance due to small parameter changes and

inexact zero-pole cancellations.

Instead of (6.3.6), let us consider the feedback control law

α( s)

α( s)

up = θ∗ 1

u

y

Λ( s) p + θ∗ 2 Λ( s) p + θ∗ 3 yp + c∗ 0 r

(6.3.8)

shown in Figure 6.4 where

α( s) = αn− 2( s) = sn− 2 , sn− 3 , . . . , s, 1

for n ≥ 2

α( s) = 0

for n = 1

c∗ 0 , θ∗ 3 ∈ R 1; θ∗ 1 , θ∗ 2 ∈ Rn− 1 are constant parameters to be designed and Λ( s) is an arbitrary monic Hurwitz polynomial of degree n− 1 that contains Zm( s)

as a factor, i.e.,

Λ( s) = Λ0( s) Zm( s)

334

CHAPTER 6. MODEL REFERENCE ADAPTIVE CONTROL

rc∗

+ ♥

up

yp

0

Σ

Gp( s)

+ +

✂✂✍ ❆

❆❑

+

α( s)

θ∗

1

Λ( s)

α( s)

θ∗

2

Λ( s)

θ∗ 3

Figure 6.4 Structure of the MRC scheme (6.3.8).

which implies that Λ0( s) is monic, Hurwitz and of degree n 0 = n − 1 − qm.

The controller parameter vector

θ∗ = θ∗ 1 , θ∗ 2 , θ∗ 3 , c∗ 0

∈ R 2 n

is to be chosen so that the transfer function from r to yp is equal to Wm( s).

The I/O properties of the closed-loop plant shown in Figure 6.4 are

described by the transfer function equation

yp = Gc( s) r

(6.3.9)

where

c∗

G

0 kpZpΛ2

c( s) =

(6.3.10)

Λ Λ − θ∗ 1 α( s) Rp − kpZp θ∗ 2 α( s) + θ∗

We can now meet the control objective if we select the controller param-

eters θ∗ 1 , θ∗ 2 , θ∗ 3 , c∗ 0 so that the closed-loop poles are stable and the closed-loop

transfer function Gc( s) = Wm( s), i.e.,

c∗ 0 kpZpΛ2

Z

= k

m

(6.3.11)

Λ Λ − θ∗

m R

1 α Rp − kpZp θ∗

2 α + θ∗

m

is satisfied for all s ∈ C. Because the degree of the denominator of Gc( s) is

np + 2 n − 2 and that of Rm( s) is pm ≤ n, for the matching equation (6.3.11)

to hold, an additional np + 2 n − 2 − pm zero-pole cancellations must occur in

Gc( s). Now because Zp( s) is Hurwitz by assumption and Λ( s) = Λ0( s) Zm( s) 6.3. MRC FOR SISO PLANTS

335

is designed to be Hurwitz, it follows that all the zeros of Gc( s) are stable

and therefore any zero-pole cancellation can only occur in C−. Choosing

k

c∗

m

0 =

(6.3.12)

kp

and using Λ( s) = Λ0( s) Zm( s) the matching equation (6.3.11) becomes

Λ − θ∗ 1 α Rp − kpZp θ∗ 2 α + θ∗ 3Λ = ZpΛ0 Rm

(6.3.13)

or

θ∗ 1 α( s) Rp( s) + kp θ∗ 2 α( s) + θ∗ 3Λ( s) Zp( s) = Λ( s) Rp( s) − Zp( s)Λ0( s) Rm( s) (6.3.14)

Equating the coefficients of the powers of s on both sides of (6.3.14), we can

express (6.3.14) in terms of the algebraic equation

S ¯

θ∗ = p

(6.3.15)

where ¯

θ∗ = θ∗ 1 , θ∗ 2 , θ∗ 3 , S is an ( n + np − 1) × (2 n − 1) matrix that depends on the coefficients of Rp, kpZp and Λ, and p is an n + np − 1 vector

with the coefficients of Λ Rp−ZpΛ0 Rm. The existence of ¯

θ∗ to satisfy (6.3.15)

and, therefore, (6.3.14) will very much depend on the properties of the matrix

S. For example, if n > np, more than one ¯

θ∗ will satisfy (6.3.15), whereas if

n = np and S is nonsingular, (6.3.15) will have only one solution.

Remark 6.3.2 For the design of the control input (6.3.8), we assume that

n ≥ np. Because the plant is known exactly, there is no need to assume

an upper bound for the degree of the plant, i.e., because np is known

n can be set equal to np. We use n ≥ np on purpose in order to use

the result in the unknown plant parameter case treated in Sections 6.4

and 6.5, where only the upper bound n for np is known.

Remark 6.3.3 Instead of using (6.3.15), one can solve (6.3.13) for θ∗ 1 , θ∗ 2 , θ∗ 3

as follows: Dividing both sides of (6.3.13) by Rp( s), we obtain

Z

Λ − θ∗

p

1 α − kp

( θ∗

R

2 α + θ∗

3Λ) = Zp

Q + kp

p

Rp

336

CHAPTER 6. MODEL REFERENCE ADAPTIVE CONTROL

where Q( s) (of degree n − 1 − mp) is the quotient and kp(of degree

at most np − 1) is the remainder of Λ0 Rm/Rp, respectively. Then the

solution for θ∗i, i = 1 , 2 , 3 can be found by inspection, i.e.,

θ∗ 1 α( s) = Λ( s) − Zp( s) Q( s)

(6.3.16)

Q( s) R

θ∗

p( s) Λ0( s) Rm( s)

2 α( s) + θ∗

3Λ( s) =

(6.3.17)

kp

where the equality in the second equation is obtained by substituting

for ∆ ( s) using the identity Λ0 Rm = Q + kp. The parameters θ∗

Rp

Rp

i , i =

1 , 2 , 3 can now be obtained directly by equating the coefficients of the

powers of s on both sides of (6.3.16), (6.3.17).

Equations (6.3.16) and (6.3.17) indicate that in general the controller

parameters θ∗i, i = 1 , 2 , 3 are nonlinear functions of the coefficients of

the plant polynomials Zp( s) , Rp( s) due to the dependence of Q( s) on

the coefficients of Rp( s). When n = np and n∗ = 1, however, Q( s) = 1

and the θ∗i’s are linear functions of the coefficients of Zp( s) , Rp( s).

Lemma 6.3.1 Let the degrees of Rp, Zp, Λ , Λ0 and Rm be as specified in

(6.3.8). Then (i) The solution ¯

θ∗ of (6.3.14) or (6.3.15) always exists.

(ii) In addition if Rp, Zp are coprime and n = np, then the solution ¯

θ∗ is

unique.

Proof Let Rp = ¯

Rp( s) h( s) and Zp( s) = ¯

Zp( s) h( s) and ¯

Rp( s) , ¯

Zp( s) be coprime,

where h( s) is a monic polynomial of degree r 0 (with 0 ≤ r 0 ≤ mp). Because Zp( s) is Hurwitz, it follows that h( s) is also Hurwitz. If Rp, Zp are coprime, h( s) = 1, i.e., r 0 = 0. If Rp, Zp are not coprime, r 0 1 and h( s) is their common factor. We can now write (6.3.14) as

θ∗ 1 α ¯

Rp + kp( θ∗ 2 α + θ∗ 3Λ) ¯

Zp = Λ ¯

Rp − ¯

ZpΛ0 Rm

(6.3.18)

by canceling h( s) from both sides of (6.3.14). Because h( s) is Hurwitz, the cancella-

tion occurs in C−. Equation (6.3.18) leads to np + n−r 0 2 algebraic equations with

2 n− 1 unknowns. It can be shown that the degree of Λ ¯

Rp− ¯

ZpΛ0 Rm is np+ n−r 0 2

because of the cancellation of the term snp+ n−r 0 1. Because ¯

Rp, ¯

Zp are coprime,

it follows from Theorem 2.3.1 that there exists unique polynomials a 0( s) , b 0( s) of

degree n − 2 , np − r 0 1 respectively such that

a 0( s) ¯

Rp( s) + b 0( s) ¯

Zp( s) = Λ( s) ¯

Rp( s) ¯

Zp( s)Λ0( s) Rm( s)

(6.3.19)

6.3. MRC FOR SISO PLANTS

337

is satisfied for n ≥ 2. It now follows by inspection that

θ∗ 1 α( s) = f( s) ¯

Zp( s) + a 0( s)

(6.3.20)

and

kp( θ∗ 2 α( s) + θ∗ 3Λ( s)) = −f( s) ¯

Rp( s) + b 0( s)

(6.3.21)

satisfy (6.3.18), where f ( s) is any given polynomial of degree nf = n − np + r 0 1.

Hence, the solution θ∗ 1 , θ∗ 2 , θ∗ 3 of (6.3.18) can be obtained as follows: We first solve

(6.3.19) for a 0( s) , b 0( s). We then choose an arbitrary polynomial f( s) of degree nf = n − np + r 0 1 and calculate θ∗ 1 , θ∗ 2 , θ∗ 3 from (6.3.20), (6.3.21) by equating coefficients of the powers of s. Because f ( s) is arbitrary, the solution θ∗ 1 , θ∗ 2 , θ∗ 3 is not unique. If, however, n = np and r 0 = 0, i.e., Rp, Zp are coprime, then f( s) = 0

and θ∗ 1 α( s) = a 0( s), kp( θ∗ 2 α( s) + θ∗ 3Λ( s)) = b 0( s) which implies that the solution θ∗ 1 , θ∗ 2 , θ∗ 3 is unique due to the uniqueness of a 0( s) , b 0( s). If n = np = 1, then α( s) = 0, Λ( s) = 1, θ∗ 1 = θ∗ 2 = 0 and θ∗ 3 given by (6.3.18) is unique.

Remark 6.3.4 It is clear from (6.3.12), (6.3.13) that the control law (6.3.8)

places the poles of the closed-loop plant at the roots of the polynomial

Zp( s)Λ0( s) Rm( s) and changes the high frequency gain from kp to km

by using the feedforward gain c∗ 0. Therefore, the MRC scheme can be

viewed as a special case of a general pole placement scheme where the

desired closed-loop characteristic equation is given by

Zp( s)Λ0( s) Rm( s) = 0

The transfer function matching (6.3.11) is achieved by canceling the

zeros of the plant, i.e., Zp( s), and replacing them by those of the

reference model, i.e., by designing Λ = Λ0 Zm. Such a cancellation

is made possible by assuming that Zp( s) is Hurwitz and by designing

Λ0 , Zm to have stable zeros.

We have shown that the control law (6.3.8) guarantees that the closed-

loop transfer function Gc( s) of the plant from r to yp has all its poles in C−

and in addition, Gc( s) = Wm( s). In our analysis we assumed zero initial

conditions for the plant, reference model and filters. The transfer function

matching, i.e., Gc( s) = Wm( s), together with zero initial conditions guar-

antee that yp( t) = ym( t) , ∀t ≥ 0 and for any reference input r( t) that is 338

CHAPTER 6. MODEL REFERENCE ADAPTIVE CONTROL

bounded and piecewise continuous. The assumption of zero initial condi-

tions is common in most I/O control design approaches for LTI systems and

is valid provided that any zero-pole cancellation in the closed-loop plant

transfer function occurs in C−. Otherwise nonzero initial conditions may

lead to unbounded internal states that correspond to zero-pole cancellations

in C+.

In our design we make sure that all cancellations in Gc( s) occur in C−

by assuming stable zeros for the plant transfer function and by using stable

filters in the control law. Nonzero initial conditions, however, will affect the

transient response of yp( t). As a result we can no longer guarantee that

yp( t) = ym( t) ∀t ≥ 0 but instead that yp( t) → ym( t) exponentially fast with a rate that depends on the closed-loop dynamics. We analyze the effect of

initial conditions by using state space representations for the plant, reference

model, and controller as follows: We begin with the following state-space

realization of the control law (6.3.8):

˙ ω 1 = F ω 1 + gup,

ω 1(0) = 0

˙ ω 2 = F ω 2 + gyp,

ω 2(0) = 0

(6.3.22)

up = θ∗ ω

where ω 1 , ω 2 ∈ Rn− 1,

θ∗ = θ∗ 1 , θ∗ 2 , θ∗ 3 , c∗ 0

, ω = [ ω 1 , ω 2 , yp, r]

−λ

n