Attribution Non-Commercial (BY-NC)

28 tayangan

Periodic Compensation of Continuous-Time Plants

Attribution Non-Commercial (BY-NC)

- Sample Exam 1 EE 210
- Matlab Paper
- Process Automation
- Fuzzy Co-Ordination of FACTS Controllers for Damping Power S
- Robust Decoupling Control Design for Twin Rotor System Using Hadamard Weights
- First Chapters
- Control System Analysis & Design by Frequency Response
- Transfer Functions of Solar Heating Systems for Dynamic Analysis
- Planar Running Robot
- International Journal of Soft Computing, Mathematics and Control (IJSCMC)
- system
- Track 7
- Cover Page
- dp
- PID Controller
- 2 Feedback Principles
- 07041979
- control objective bits
- S1706-PolesAndZeros
- lec1.pdf

Anda di halaman 1dari 7

5, MAY 2007

Sarit K. Das and Jayati Dey

bust stability for single-input–single-output (SISO), linear time invariant

(LTI) plants having both right-half plane (RHP) poles and zeros, a job LTI

controllers fail to do. In addition, for strictly proper plants this controller

achieves model matching ensuring at the same time that the periodic os-

cillations present in the plant output are insignificant in magnitude. The

design steps are straightforward and linear algebraic in nature.

troller, robust stability.

I. INTRODUCTION

Fig. 1. Periodic controller.

For a linear time invariant (LTI) system, a suitable measure of loop

robustness is the minimum distance of the nyquist plot of loop transfer

function P (s) from the (01; 0) point. It is well known that the in-

of the approximation increasing with the frequency of the periodic vi-

verse of the peak of the frequency response of the transfer function

brations. In this note, the controller parameters have been suitably de-

1=(1+ P (s)) equals this minimum distance. In literature, 1=(1+ P (s))

signed so as to achieve pole as well as zero-placement leading to a

is often referred to as sensitivity transfer function S (s) and therefore

the robustness measure becomes kS k0

robust averaged system that matches a given model. One then expects

11 [1], [2]. It may be noted that

kS k1 2 signifies simultaneous0 satisfaction of gain margin (GM) the original periodic system to be robust as well. (In fact, [10, Th. 5.1]

2 and phase margin (PM) 30 [1]. claims via the averaging approach that arbitrary GM compensation can

be achieved using a periodic controller.) Simulation results justify this

For a single-input–single-output (SISO), LTI plant having RHP

expectation. The simulation also shows that the controller considered

poles pj ; j = 1; 2; . . . ; k , and zeros zi ; i = 1; 2; . . . ; l, closed-loop

results in insignificant output ripples.

LTI compensation can only achieve [1]

Section II of this note presents the controller and obtains the aver-

aged equivalent of the closed-loop system corresponding to it. Sec-

k

kS k1 max jz i + p j j : tion III presents the steps for controller design and Section IV some

i

j =1 jz i 0 p j j (1)

examples.

Clearly, if the plant has RHP poles and zeros in close vicinity, the loop

robustness will be poor.

Several attempts have been made in literature to improve matters II. THE AVERAGED SYSTEM

using periodic controllers. Some of these results pertain to discrete-

time systems [3]–[5]. Some use the generalized sampled-data hold ap- The following notations are used. For any function, [:]( ) =

[:]( + T ), [:] := T 01 0 [:]( )d . Further if [:] = 0, then

proach [6], [7]. These, however, have significant intersampling ripples T

and poor robustness properties [8], [9]. The high frequency (! ) contin- [:](0k) := 1 1 1 [:]( )(d )k with [:](0i) = 0 for i = 1; . . . ; k . Also,

uous-time periodic controllers in [10] and [11] are seen to yield robust define

designs even for plants having RHP poles and zeros. Unfortunately,

O(1) oscillations (of frequency ! ) crop up in the response significantly

to make these results unacceptable. A variation of these controllers as

( ) := e0k (2)

considered in [12] is, however, seen to yield zero input responses cor-

where km and ( ) are as in Fig. 1. Note that has the property =

01 .

responding to a type of nonlinear systems with very small oscillations.

In this note, a controller based essentially on [12] has been considered.

The analysis and synthesis of the periodically compensated system has Consider a plant x_ = Ax + Bu, y = Cx + d, with

bn0r sn0r + 1 1 1 + b1 s + b0

been carried out using the averaging principle as in [10]–[12]. Ac-

cording to this principle, the stability and the response of a periodic C (sI 0 A)01 B = B0 (s) = n

system (satisfying certain constraints [11]) can be adequately approxi- A 0 (s ) s + an01 sn01 + 1 1 1 + a1 s + a0

(3)

mated by an equivalent averaged (time-invariant) system, the accuracy

where d is the output disturbance. Further let be the measurement

noise. Now consider the two degree-of-freedom observer canonical

form controller of Fig. 1 which may be described as

Manuscript received September 14, 2005; revised March 17, 2006 and Jan-

uary 11, 2007. Recommended by Associate Editor U. Jonsson.

The authors are with the Department of Electrical Engineering, Indian Insti-

+ [G + ! r (!t)L] (y 0 ) + Qv

tute of Technology, Kharagpur-721302, India (e-mail: skdas@ee.iitkgp.ernet.

in).

Digital Object Identifier 10.1109/TAC.2007.895911 u = Hz + gm+1 (y 0 ) + qm+1 v z 2 Rm (4)

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 5, MAY 2007 899

T

F =

0 m0 2 I m0 2 m0

_ m + !m km (!t) = 0

( 1) 1 ( 1) ( 1)

FT

_ i + !(!t)qm+1 [ki + km i ] = 0; i = 1; . . . ; m 0 1

K = 0m2(m01) K H = 012(m01) 1 _ m + !qm+1 (!t)km m = 0 (8)

G = [ g 1 1 1 1 g m ]T L = [ l1 1 1 1 lm ]T _i(1) + ! (!t)li + !ki (!t)gm+1 + !i lm (!t)

F = [ f 1 1 1 1 f m ]T K = [ k 1 1 1 1 k m ]T +!km (!t)i gm+1 = 0; i = 1; . . . ; m 0 1

Q = [q1 1 1 1 qm ]T Gm+1 = 012(m01) gm+1 T _m(1) + !(!t)km gm+1 m + ! (!t)lm m = 0

Qm+1 = 012(m01) qm+1 T : which yields

Then, the overall system equation becomes (5), as shown at the bottom i = 0 [(ki =km ) + ci ] ; i = 1; . . . ; m 0 1 m = cm

of the page. i = 0ci qm+1 [ + ci ] m = qm+1 [m + cm ]

Note that some of the elements in (5) are of O(!r ), some of O(! ) i(1) = 0li (01) + ci(1) 0 ci gm+1 + (ki lm =km ) (01)

and the rest of O(1). For averaging to be applicable, it is first necessary

[10] to transform all the coefficients to O(1). First assume r = 1. To +ci lm d; i = 1; . . . ; m 0 1

normalize the O(! ) terms present in the expression for z_ , set m(1) = gm+1 m 0 cm lm d + cm(1)

(9)

where is as given by (2) and ci , ci , cm , ci(1) , cm are constants of

(1) = 8z + 2(1) x + 9v + 30(1) (d 0 )

integration. Choose = cos(!t). Then, i and i , i = 1; . . . ; m, are

(6)

to have = 0 (else the integration will produce a t term). Therefore

8 = I m0 21m1 10 01 1m1 0 2m

T

( 1) ( 1) ( 1) 1

must be cos(!t) for r = 1. Without any loss of generality, one may

1

set ci = 0, i = 1; . . . ; m 01, and cm = 0. Also set cm = 1 to have the

9 = [ 1 1 1 m ]T 1 convenient property that m = m 01 . For r = 1 then, this procedure

3 =

0(1) 1 1 1 m T 1(1) (1)

reduces the system (5) to the amplitude normalized form shown in (10)

2 = 3 C:

(1) 0(1)

at the bottom of the page.

For r = 2, however, even after the transformation (6) is carried out,

Then the following O(! ) terms will remain in the system equation:

+ 3_ 0(1)(d 0 ) + 30(1)(d_ 0 _ ) i(1) = ! 0li (01) 0 li ci(1) + (ki lm =km ) (01)

:

= 8_ z + 8 [G + ! (!t)L + gm+1 (F + !(!t)K )] i =1; . . . ; m 0 1; m(1) = gm+1 0 !lm d + cm(1)

2 fCx + (d 0 )g + 8 F + !(!t)K z Since for further normalization of such O(! ) terms they need to

(11)

+ 8 [Q + qm+1 fF + !(!t)K g] v + 2_ (1) x be integrated as well, they must be of zero-mean. To ensure this set

+ 2(1) [Ax + gm+1 B [Cx + (d 0 )] ci(1) = 0, i = 1; . . . ; m 01, and lm = 0. Now, use the transformation

+BHz + qm+1 Bv] + 9_ v + 9_v

+ 30(1)(d 0 ) + 30(1)(d_ 0 _ ):

_ (7) (2) = (1) + 2(2) x + 30(2) (d 0 ) + 31(2) (d_ 0 _ ) (12)

x_

z_

= [G + !r (!t)L]AC++ggmm BC BH +1 x

[F + !(!t)K ] C F + !(!t)K z

+1

+1

+1 +1

+1

x

y = [C 0] +d (5)

z

x_ gm BC 0 BH 80 2 BH 80

= 8GC 0 8F 80 2 0 2A +BH

1 1

+1 (1)

_ (1) 8 2 + gm 8F C + gm 2 BC + 2 A 2 BH 80 + 8F 80

0 1

(1) (1)

1

(1) +1 +1 (1) (1) (1)

1 1

2 BH 80 9 + qm 2 B v + 9 v_

+1

1 1

(1) +1 (1) +1 (1)

1

0 2 BH 80 3 + gm 2 B (d 0 ) + 3 (d 0 _ )

+1 0(1)

1 1

0(1) +1 (1) 0(1) +1 (1) 0(1)

x

y = [C 0] +d (10)

(1)

900 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 5, MAY 2007

and equate the O(!) terms in the expression for _ (2) to zero to obtain 3 = 111 0

0(1) 1(1) 0 111 0 g

(m r +1)(1) m+1

T

(k 1 j )(k)

( k k

0 j )! F

1 j j

T

(2) 0 1(2) (m 2)(2) (m 1)(2) 1(k) (m r +1)(k)

T

T k = 2 ; . . . ; r 0 1; j = 0 ; . . . ; k 0 1

+ [k k 1 1 1 k 0 0] 0 l k 0 C

1 2 m 1

T ( 2) m 1

(13) 3 0 0 = (01) 0 0 j !(r(r0011)!

(r 1 j )(r )

0 j )! F

r 1 j j

T T

0(2) 0 1(2) (m 2)(2) (m 1)(2) 1(r ) (m r +1)(r )

l j = 0; . . . ; r 0 2

+ [k k 1 1 1 k 0 0] 0 k 0 T ( 2) m 1

3 = 0 l 0 [k 1 1 1 k 0 0] =k

1 2 m 1 (14) ( r) T

m

0(r ) m r +1 1 m 1 m

3 =0 111 0 0 T

+ F 0 1 1 1 0 0

(15)

1(2) 1(2) (m 1)(2)

0 111 0

T

r 1

1(r ) (m r )(r ) (m r +1)(r )

(19)

where where

( = 0 ! 0 0 l k = 1; . . . ; r 0 1

r k ( k)

0 li 02) + ci(2) ; i = 1; . . . ; m 0 1

i k)

(2) =

i

(

i

(

i r) = 0 l 0 + c ; i = 1; . . . ; m 0 r + 1

i

( r)

i(r )

( 01)(2) =

m 0 lm01 (01) d + cm(2) : and

( 0 = 0l 0 (0 r +1)

d + c( 0 :

Note that for (m01)(2) to be bounded, (for r = 2) must be chosen

m r +1)(r ) m r +1 m r +2)(r )

One may generalize the above to normalize the system (5) having normalized form:

O(!r ) terms using the sequence of transformations (1) as given by

x_

(6) and (k) , k = 2; 3; . . . ; r , given by

_ ( )

= A x + B v + B v_ + C (d 0 )

T T ;0 T ;1 T ;0

r (r )

01 + C (d_ 0 _ ) + 1 1 1 + C dtd (d 0 )

r

( = 0 +2 x+

k

3 00 d 010 (d 0 )

k j

T ;1 T ;r

r

dt 010

k) (k 1) (k) (k 1 j )(k) (16)

k j

x

j =0

y = [C 0] + d: (20)

( ) r

and to ensure that these substitutions be well-posed, it is required that (Explicit forms of AT , BT ;i and CT ;i are presented in [13].) Note that

all the O(! r0k+1 ) elements of 2(k01) and 3i(k01) be of zero mean the coefficient matrices AT , BT ;i and CT ;i are bounded and smooth.

and (m0r+1)(r) be bounded, for which set lm0k+2 = 0 and ci(k01) = Now, let v be a step function. Since v_ (t)jt=0 tends to 1, for well-

0, k = 2; 3; . . . ; r, and i = 1; . . . ; m 0 k +1, and to have (0r+1) = posedness, one must have BT ;1 = 0 which, it can be verified, requires

0 choose and as qm+1 = 0. Further, for the moment, assume d = = 0. Then, (20)

may be averaged to yield

sin(!t); if r is even

cos(!t); if r is odd

(17)

x_

_

= A x + B v y = [C 0] x

T T ;0 (21)

(r ) (r ) (r )

Then, one obtains

which, following [11, Th. 4.4], is globally -equivalent to (5) in the

2 = 111 0 0 111 0 g C sense that, if AT has all its eigenvalues in the LHP, then for any > 0,

T

(1) 1(1) (m r +1)(1) m+1

2 = (01) 0 0 j !(k(k0011)!

1

k

where y (t) is the moving average (Definition 3.3, [11]) of the output

0 j )! F

k 1 j j

(k)

j =0 y(t) of (20) and hence (5). Note that since the output matrix in (20)

2 111 0 0 1 1 1 0 CA 0 0 T k 1 j is time-invariant, here y (t) = y (t), which signifies that the ripples

1(k) (m r +1)(k)

present in the periodically compensated system output will also tend to

k = 2; . . . ; r 0 1 zero for sufficiently large ! . The same conclusion can be obtained from

a physical point of view as well. Since a signal sin !t passing through

and an integrator gets attenuated by a factor ! , it follows from Fig. 1 that

for the choices of li = 0, i = m 0 r + 2; . . . ; m, r 2, as made ear-

0 lier, the controller output contains oscillations of O(1) amplitude only.

2 = (01) 0 0 j !(r(r0011)!

2

r

0 j )! F

r 1 j j

(r ) Obviously then the plant output would contain the periodic oscillations

j =0 of O(1=! r ) only.

2 111 0 0 1 1 1 0 CA 0 0

1(r ) (m r +1)(r )

T r 1 j

Referring again to (21), in order to obtain an averaged character-

+ F 0 1 1 1 0 0

istic equation linear in the controller parameters, it is further neces-

01110 C

T

r 1

1(r ) (m r )(r ) (m r +1)(r )

+ 0 l k0 [k 1 1 1 k 0 0] C

( r) m r +1

m

1 m 1

T

(18) tion so that BT ;0 v B T ;0 v and BT ;1 v_ B T ;1 v_ . Also, let d and

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 5, MAY 2007 901

( )( ) (

CT ;i di =dti d 0 C T ;i di =dti d 0 , i )(

; ; r . Then, ) = 1 ... F (s ) = 0 f 1 0 f s0 111

2 2

2

following [11, Rem. 4.7], the averaged equivalent of (20) as given by

(22) at the bottom of the page, where 0f 0 s 0 0f s 0 +s

m 1

2 m 2

m

m 1 m

Q ( s ) = [q + q s + 1 1 1 + q s 0 + q s ]

2 m 1 m

1 2 m m+1

8 = I 00K=k 2 0 0 0 21 G (s ) = 0 g 1 1 1 0 g 0 s 0 (26)

T

(m 1) (m 1) (m 1) 2 2 m 2

1 m 1

0 g +l 0 0 s 0 0g s

m

K = [k 1 1 1 k 0 0]

( r) m 1 m

m m r m+1

1 m 1

9= 0 2 0 q

1 (m 1) m+1

T

L(s) = (01) 0 [l + l s + 1 1 1 + l 0 s 0 0 ]

r ( r)

1 2 m r

m r 1

and

The output corresponding to the averaged system (22) is shown in K (s) = k1 + k2 s + 1 1 1 + km01 sm02

(23) at the bottom of the page, where the characteristic polynomial is

given by with

ki = (

2

0 1) fm ki

+ k kk i m 01 0 kk0 i 1

; i = 1; . . . ; m 0 1

s) := A (s) F (s) + K (s) + Z(s)

1( 0 (24) km 2

m m

in forming the closed loop characteristic equation. Without any

with the effective loop zero polynomial being loss of generality, then, one may remove them by setting ki , =0

i = 1 ... 1

; ; m 0 . However, in order to ensure that (0r) 6 , km =0

Z(s) = B0 (s)G(s) must be nonzero. (However, the value of km is arbitrary.) Then, s )

1(

+ (1 0 2 )[sB0 (s) 0 bn01 A0 (s)] gm+1 sm01 becomes

2 (k1 + 1 1 1 + km01 sm02 )=km (25) s) := A (s)F (s) + Z (s)

1( 0 (27)

x_ BH 8

_ (1)

= 8GC 0 8F 80 2 + Ag 8F C + G 1

(1) m+1 m+1 CA bn01 Gm+1 H + bn01 (01) LH + 8F 801

x

(1)

+ 8Q 0 8F 8009 + q 8F v + Q0 1

m+1 m+1

v_ +

0

8G 0 8F80 3 + g 1

0(1) m+1 8 F (d 0 )

+ 0 (d_ 0 _ ) for r = 1

Gm+1

x_ BH 8

_ (r)

= 8GC 0 8F 80 2 0 8F80 GA C + g 8F C + G

1

(r )

1

m+1 m+1 m+1 CA 0(01) b 0 0 r

n r

(r)

LH + 8F 801

2 x + 8Q 0 8F 8009 + q 8F v + Q0 v_ 1

m+1 m+1

(r )

+ 8G 0 8F80 3 0 8F0 80 G + g 8F (d 0 ) +

1

0(r )

1

m+1 m+1 Gm+1

0 (d_ 0 _ ) for r 2

x

y = [C 0] + d; r 1 (22)

(r )

2

Y (s ) =

1 (s) V (s) + s)

1( D (s )

2

+ 1 (s) N (s ) +

1 (s) D(s)

B0 (s) (0r) l k (k + 1 1 1 + k 01 sm02 )

0 N (s )

1 m

1 (s) (23)

902 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 5, MAY 2007

2(1 0 2 )gm+1 sm01 + bn0r A0 (s)L(s): (28)

as

Y (s ) (s)Q(s) :

= B01(

V (s ) s) (29)

Further, (29) alongwith (27) and (28) yields the averaged equivalent

error response transfer function of the closed-loop system, where the

error E s( )= ( ) ( )

V s 0 Y s , as

A0 (s) F (s) 0 b 01 (1 0 2 )g +1 s 01 m Fig. 2. (a) Unit step response. (b) Its deviations from the averaged value for

E (s ) = ( )

n m

Example 1.

V s

1 (s)

B0 (s) G(s) + (1 0 2 )g +1 s 0 Q(s) m

+ V (s ) which shows that B (s) must contain the RHP zeros of B0 (s) and

m

0 (s)L(s) Next, it is evident from (30) that for e [=

+ b 0 A1(

n

s ) V (s ):

r

(30) lim sE (s)] to be robustly zero, one must have

ss

!0

!0 s 2 [ each of the three terms in RHS of (30)] = 0. For a step

lim s

s

Next, from (23) the averaged disturbance-to-output and the noise-to- input and for a desired transfer function having unity dc gain, for

output transfer functions are, respectively, obtained as example, it follows from (27), (28), (26), and (34) that these terms

2 m01

Y (s) = A0 (s) F (s)+ bn0rL(s) 0 (1 0 )bn01 gm+1 s

will be zero: i) for a type-1 plant automatically and ii) for a type-0

plant when F (0) = = 0

f1 and L (0) = = 0

l1

1 (s)

.

D (s ) The above observations may now be concretized in the form of an al-

(31) gorithm. First, however, it would be convenient to define the following:

2

Y (s) = B0 (s) G(s) + (1 0 )gm+1 s : F (s) := sm + fm sm01 + 1 1 1 + f1

m

N (s ) 1 (s) (32)

Q(s) := qm+1 sm + 1 1 1 + q2 s + q1

G(s) := gm+1 sm + 1 1 1 + g2 s + g1 (35)

III. CONTROLLER DESIGN

L(s) := lm0r sm0r01 + 1 1 1 + l2 s + l1

The basic mechanism how a periodic controller does what LTI con-

Z (s) := zm+n0r sm+n0r + 1 1 1 + z1 s + z0

A0 (s)F (s) := sm+n + m+n01 sm+n01 + 1 1 1 + 1 s + 0

trollers can not is as follows. Compensating (3) using a LTI compen-

() ()

sator having a transfer function G0 s =F0 s , one obtains a closed- :

loop characteristic equation 1(s)1(s) := sm+n + m+n01 sm+n01 + 1 1 1 + 1 s + 0

(36)

Design Step 1: Following Section II, in order that the system be

where 1(s) may be assumed to consist of the n desired closed-loop ( )

well posed and linear (in the controller parameters), set !t

poles and 1(s) the remaining ones. Now, if one chooses 1(s), then ( )

and !t as in (17) and k_i=0, i = 1 ...

; ; m 0 , li 1 =0 ,

G0 (s) and F0 (s) get fixed up automatically. However, (27) shows that =

i m0r + 1 ...

; ; m, and for step input qm+1 =0. Also

for the periodically compensated system the loop transfer function is choose km to be an arbitrary number >0 (say 1). Further, in order

P (s) = Z (s)=A0 (s)F (s) where Z (s), as given by (28), does not, to decide about the controller order note from (28) and (35) that

unlike the LTI case, generically contain the plant zeros. (This, in fact, 2 +1

m 0 r number of the controller parameters, gi and li , assign

amounts to loop zero-placement.) Now, the advantage of using a peri- ( + )

the coefficients of m n 0 r th order loop zero polynomial Z s . ()

odic controller arises from the fact that here one may choose both s 1( ) Then one must have m 0 r2 +1 + +1

m n 0 r ) m n.

()

and F s a priori (arbitrarily for r =1

, but subject to some constraints For type-1 or higher plants, choose m = n. However, as noted

1 ()

for r > ) so that the resulting Z s leads to a loop transfer function above, for type-0 plants one must preset l1 =0 which requires

()

P s having satisfactory robustness margin. The controller parameters = +1

m n .

()

gi and li corresponding to such a Z s can be evaluated solving (28) ( ) 1( )

Design Step 2: Choose the polynomials F s and s satisfying

and (26). the following conditions.

Now suppose a plant (3) with RHP poles and zeros is to be com- i)

pensated such that in addition to the loop becoming robustly stable:

( ) 1( )

i) the input-output transfer function becomes B s = s , and ii) ro- m+n0i = m+n0i ; i = 1; . . . ; (r 0 1); for r >1 (37)

bust zero-error tracking is effected. To achieve i), note from (29) that

[which follows from (36), (35), and (27)].

B (s ) = B0 (s)Q(s) ) B0 (s)Q(s) = B(s)1(s) ii) 1( ) ()

s contains the LHP zeros of B0 s if they are not

1(s) 1(s)1(s)

(34)

()

present in B s .

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 5, MAY 2007 903

()

iii) For a type-0 plant F s contains one root at origin (to Stage 2: Next following (26) evaluate gi , li as

ensure robust zero-error tracking), such that the corre-

()

sponding Z s as obtained following (27) leads to an

acceptable loop characteristics. Root locus and/or loop gk = 0 gk 1 k m 0 1

2

shaping considerations should prove to be useful while

l =

lk = (01)r k 1k m0r

making these choices which, expectedly, are not unique.

It may be emphasized in this context that here the search (0r) : (41)

space made available to the designer is much larger

(compared to LTI designs) so that satisfactory designs gm = 0 m

(g =) + (0r)lm0r

are much more likely to be obtained. Examples presented

in Section IV illustrate one way to make these choices. gm+1 = 0 gm+1

Design Step 3: Now find the controller parameters as follows.

i) Following (26) evaluate the fi ’s as

IV. EXAMPLES

fi = 0 fi = ; i = 1 ; 2; . . . ; m 0 1 ( + 1)( s 0 =s s2) ( + 3)( 1)

s 0 , is to be

2

Example 1: A plant, s

fm = 0 fm : (38) compensated such that it: i) resembles 0 s 0 = s2( 2) ( + 2)( + 1 s 6

)

j , ii) robustly tracks a step input, and iii) has good loop robustness

ii) Obtain the qi ’s, i ; ; ;m = 1 2 ...

, by matching the +1 properties. Consider a periodic controller of the form shown in Fig. 1

coefficients of the like powers of s in (34) after carrying with order m = =3

n . The first step is to make suitable choices for

out the due cancellations. Then, following (26), evaluate ( ) 1( )

the polynomials F s and s . In order to do so in a tractable fashion,

the qi ’s as one may try to reduce the number of design variables by engineering as

much LHP pole-zero cancellations as possible. The remaining design

variable may then be chosen using trial and error. Here the choices have

qi = qi = ; i = 1 ; 2; . . . ; m + 1 :

2 been made as follows.

(39)

1( ) ( +1)

i) Design Step 2) requires that s has a factor s . This may

iii) The remaining controller parameters gi , li can be valu-

() ()

be ensured by letting both F s and Z s have the same factor.

2 () ()

ii) As one of the desired poles is at 0 , let both F s and Z s have

ated using a two-stage linear procedure in which first the

values of some intermediate variables gi , li are obtained,

( + 2)

the factor s .

in terms of which, next, gi , li are evaluated.

() (+)

iii) Let the remaining factor of F s be s p , p a design parameter.

Stage 1: Equating the coefficients of sk in the expressions for Z s () 1( ) ()

Also let s and Z s have the same factor. Note that although

given by (28) and (35), k = 0 ...

; ; m n 0 r , one obtains the + ()

the choice of p does not affect the S s expression for the nom-

inal plant, it must be sufficiently far in the LHP so as to ensure

matrix equation

reduced sensitivity to plant pole-zero variations.

iv) The remaining factor of 1( ) ( + )

s may be chosen at s p1 , p1

G a design parameter, such that (a) the two plant poles at 0 and 1

[B

..

. Ab ] 111 = Z (40) 1

arrive at 0 6 j , (b) the pole at 03 becomes 0p1 , and (c) the

L remaining two loop zeros occur in the LHP.

The choice of p = 15 and p1 = 10 is found to lead to an acceptable

where result. For p1 = 10 ( )

, Z s , following (27), becomes

b0 0 111 0 0

b1 b0111 0 0 Z (s) = (s + 1)(s + 2)(s + 15)(10s2 + 25s + 20) (42)

.. .. .. .. ..

. . . . . The corresponding S s ( ) = ( 1)( + 3) ( + 1 )( + 10)

s s0 s =s 6j s and

111 bn01 a0 (1 0 ) its frequency response shows that kS k1 .1

2

bm01 bm02 b0

B=

111 b0 + bn01 a1 (1 0 ) =1

2 2

bm bm01 b1 The corresponding controller parameters for km may now be

..

.

..

.

..

.

..

.

..

.

obtained as follows. Since F s( ) = ( +2)( +1)( +15)

s s s , (38) yields

2

+ bn0 an0 (1 0 )

1 1

2

2R (m+n 0r+1)2(m+1)

a0 0 0 111 0 From (34) one obtains, Q s ( ) = 02(s + 15)(s + 10) which in turn,

a1 a0 0 111 0 following (39), yields

q1 = 0187:1581; q2 = 031:1930; q3 = 01:2477 and q4 = 0:

.. .. .. .. .

. ..

. .

0 0 0 111 0

.

()

For the Z s given by (42), (40) becomes

G = [ g1 1 1 1 gm ] T 2 R m 2 ( +1) 1 02 0 0 0 0 0 g1 600

+1

01 02 0 0 03 0 g2 1690

L = [ l1 1 1 1 lm0r ]T 2 R m0r 2 1 01 02 0 2 03 = 1835

( ) 1

g3

Z = [ z0 1 1 1 zm n0r ]T 2 R m n0r

+

( + +1) 21 0 1 01 01:3971 1 2 g4 940 :

0 0 1 02:8088 0 1 l1 205

Now, evaluate the gi , li solving (40). 0 0 0 1 0 0 l2 10

904 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 5, MAY 2007

( 5 2)( +5 2) 2 4 ( +5 5)(

Example 2: Consider the plant s0 : s : = : s2 s : s0

5 5)

: which pertains to a cart-pendulum system [2]. It is to be robustly

2 5385(

compensated such that the transfer function becomes 0 : s0

5 2) ( + 0 5)( + 1 )( + 5 5)

: =s : s 6 j s : . (Here, a LTI compensator can

35 6667

only obtain kS k1 : .) Consider a periodic controller of order

m n= =4 =2

, and choose, satisfying the condition (37) for r ,

( ) = ( +5 2)( + )( + ) 1( ) = ( +5 2)( + )( +

F s s : s p s p0 2 , s s : s p s p0 0

4 ) 4 = 15

6 jp1 with p0 > and p1 arbitrary. For p0 = 20and p1 one

( ) = ( + )( + 5 5)( + 5 2)(519 +

obtains, following (4), Z s s p s : s : s3

2607 + 1585 + 521)

s2 s ( ) = ( 5 5)( +

and consequently S s s2 s 0 : s

15) ( + 0 5)( + 11 20)( + 1 )

2

=s : s = 1 555

6 j s 6 j for which kS k1 : .

=1

The corresponding controller parameters for p =1 and km have

= 20000

been obtained and response found satisfactory for ! .

V. CONCLUSION

Fig. 3. Control input to the plant for Example 1. (a) Averaged system. (b) Ac-

tual system. Examples presented show that the proposed high frequency ! peri- ( )

odic controller achieves model matching (with robust tracking) as well

as satisfactory robust compensation for plants having both RHP poles

and zeros. The control input has small oscillations of O (1)

only and,

consequently, the oscillations present in the plant output are insignifi-

(1 )

cant, being of O =!r , where r is the relative order of the plant. The

improvement in the robustness, however, does not entail improvement

in the disturbance/noise rejection characteristics.

REFERENCES

[1] W. A. Wolovich, Automatic Control Systems. Rochester, NY: Saun-

ders College Publishing, 1994.

[2] J. C. Doyle, B. A. Francis, and A. R. Tannenbaum, Feedback Control

Theory. New York: Macmillan, 1993.

[3] P. P. Khargonekar, K. Polla, and A. Tannenbaum, “Robust control of

linear time-invariant plants using periodic compensation,” IEEE Trans.

Autom. Control, vol. 30, no. 11, pp. 1088–1096, Nov. 1985.

[4] S. K. Das and P. K. Rajagopalan, “Techniques of analysis and robust

Fig. 4. (a) Disturbance rejection. (b) Noise attenuation for Example 1. control via zero placement of periodically compensated discrete time

plants,” in Control and Dynamic Systems, Advances in Theory and Ap-

plications. New York: Academic, 1996, vol. 74.

= 187 1581

[5] S. K. Das and P. K. Rajagopalan, “Periodic discrete-time systems: Sta-

Solving this and using (41), one obtains g1 : , bility analysis and robust control using zero placement,” IEEE Trans.

g2 = 1 1448 10 = 795 4037

: 2 3 , g33 : = 10 = 1 0622 10

, g 4 0 , l1 : 2 3, Autom. Control, vol. 37, no. 3, pp. 374–378, Mar. 1992.

and l2 = 1 0538 10

: 2 . [6] P. T. Kabamba, “Control of linear systems using generalized sampled-

data hold functions,” IEEE Trans. Autom. Control, vol. AC-32, no. 9,

A simulation carried out using MATLAB-SIMULINK shows that pp. 772–783, Sep. 1987.

the unit step response of the compensated system, shown in Fig. 2(a) [7] W.-Y. Yan, B. D. O. Anderson, and R. R. Bitmead, “On the gain margin

for ! = 10000 , is as desired, the deviation as well as the ripples being improvement using dynamic compensation based on generalized sam-

of the order of 04

10 (= 1 ) =! as shown in Fig. 2(b). The corresponding pled-data hold functions,” IEEE Trans. Autom. Control, vol. 39, no. 11,

pp. 2347–2354, Nov. 1994.

averaged and actual plant inputs are presented in Fig. 3(a) and (b), re- [8] J. S. Frudenberg, R. H. Middleton, and J. H. Braslavsky, “Robustness

spectively. of zero shifting via generalized sampled-data hold functions,” IEEE

It would be interesting to compare this design with the LTI com- Trans. Autom. Control, vol. 42, no. 12, pp. 1681–1692, Dec. 1997.

pensation for the same plant presented in [1]. Although the response [9] A. Feuer and G. C. Goodwin, “Generalized sample hold functions-fre-

obtained is identical, kS k1 for the latter is 6.9 (the achievable limit quency domain analysis of robustness, sensitivity, and intersample dif-

ficulties,” IEEE Trans. Autom. Control, vol. 39, no. 5, pp. 1042–1047,

as per (1) being 3), and simulations show that it has a GM of 1.5

( = 1 2 0 8) GM

May 1994.

Kmax =Kmin : = : only while the present design has a [10] S. Lee, S. M. Meerkov, and T. Runolfson, “Vibrational feedback con-

375 ( 60 0 16)

Kmax =Kmin = : for ! = 10000 . Further, the robust- trol: Zeros placement capabilities,” IEEE Trans. Autom. Control, vol.

ness with respect to location of the plant zero at 2 has been examined. AC-32, no. 7, pp. 604–611, Jul. 1987.

[11] K. Shujaee and B. Lehman, “Vibrational feedback control of time delay

It is seen that the present design maintains stability till the zero is made systems,” IEEE Trans. Autom. Control, vol. 42, no. 11, pp. 1529–1545,

1.6 (for which the LTI design is unstable), but fails when it is 1.5. It has

been verified that kS k1 01 for the perturbed plant corresponding to the Nov. 1997.

[12] S. Lee and S. M. Meerkov, “Vibrational feedback control in the

zero being 1.6 is 0.146. If the zero is at 1.5, the nyquist plot crosses the problem of absolute stability,” IEEE Trans. Autom. Control, vol. 36,

(01,0) point. no. 4, pp. 482–485, Apr. 1991.

[13] J. Dey, “Periodic comepnsation of continuous-time plants,” Ph.D. dis-

Further, Fig. 4(a) and (b) shows the frequency responses of the dis- sertation, Indian Inst. Technol., Kharagpur, India, 2006.

turbance-to-output and noise-to-output transfer functions as obtained [14] J. S. Shamma and M. A. Dahleh, “Time-varying versus time-invariant

following (31) and (32), respectively. It is seen that unlike the robust- compensation for rejection of persistent bounded disturbances and ro-

ness properties the periodic controller does not seem to improve the bust stabilization,” IEEE Trans. Autom. Control, vol. 36, no. 7, pp.

838–847, Jul. 1991.

disturbance/noise rejection properties. In fact, it has been shown in [15] H. Chapellat and M. A. Dahleh, “Analysis of time-varying control

[14] and [15] that so far as disturbance rejection is concerned, LTI con- strategies for optimal disturbance rejection and robustness,” IEEE

trollers can always achieve what a periodic controller can. Trans. Autom. Control, vol. 37, no. 11, pp. 1734–1745, Nov. 1992.

- Sample Exam 1 EE 210Diunggah olehdoomachaley
- Matlab PaperDiunggah olehsusmarohi
- Process AutomationDiunggah olehfocusvin
- Fuzzy Co-Ordination of FACTS Controllers for Damping Power SDiunggah olehranga247
- Robust Decoupling Control Design for Twin Rotor System Using Hadamard WeightsDiunggah olehqadeer62
- First ChaptersDiunggah olehMusa Yazar
- Control System Analysis & Design by Frequency ResponseDiunggah olehDozdi
- Transfer Functions of Solar Heating Systems for Dynamic AnalysisDiunggah olehSaif Evony
- Planar Running RobotDiunggah olehnarumugai29
- International Journal of Soft Computing, Mathematics and Control (IJSCMC)Diunggah olehIJRAP
- systemDiunggah olehsonti11
- Track 7Diunggah olehMohamed Shamseldein
- Cover PageDiunggah olehJake Lobrigas
- dpDiunggah olehhmthant
- PID ControllerDiunggah olehVikash Tiwari
- 2 Feedback PrinciplesDiunggah olehSadiq
- 07041979Diunggah olehShaheer Durrani
- control objective bitsDiunggah olehRamesh Dongara
- S1706-PolesAndZerosDiunggah olehArmando Malone
- lec1.pdfDiunggah olehAlif Izani Shahri
- CN10_DCgainSSE.pdfDiunggah olehHamzeh
- Aspen_HYSYS_DYNAMICS_Training_Course.pdfDiunggah olehendospora
- ENGD3038 - TosionalDiunggah olehLegendaryN
- 104UBICCV2no4_104_2Diunggah olehUbiquitous Computing and Communication Journal
- InformeDiunggah olehAndres Lozano
- nicholsIDiunggah olehbbmathi
- 4_380proj1_rprtDiunggah olehakita
- lec7Diunggah olehdanilodsp
- CICSDiunggah olehCS & IT
- Control System 2Diunggah olehRajeevSangam

- Selection of Gas Engine for a Power Plant using the Even Swaps Method.Diunggah olehAcimSultan
- aculyn_38Diunggah olehMiguel Vera Muñoz
- Wireless Communication and Rf System Design Using Matlab and SimulinkDiunggah olehThương HD
- Coordinates and Celestial SphereDiunggah olehncl12142
- DC PrinciplesDiunggah olehKaveewat
- Sol Page One QqqDiunggah olehbree789
- Experiment 9Diunggah olehMay Lacdao
- Regenerative Breaking SystemDiunggah olehDishant Chauhan
- molding defectsDiunggah olehAnurag Srivastava
- 4 Group 17 Elements UpdatedDiunggah olehNorzilah Mazahar
- Iscar CatalogoDiunggah olehJorge O. Ferreyra L.
- A Taxonomic Revision of Desmodesmus Serie DesmodesDiunggah olehDiana Mabel
- Energy_Balance_Notes_2008.pdfDiunggah olehDyo Mande
- Lab 3_Types of Chemical ReactionsDiunggah olehalextzhao1996
- assignment-2.docxDiunggah olehBadulla M. Twaaha
- huitransmissionlines 1111.pdfDiunggah olehAhmed Salem
- Ion Chromatography Column Product Application (Diapositivas).pdfDiunggah olehJosé Ignacio Aparicio Méndez
- 2014 DeepAssurance Cementing_OverviewDiunggah olehAQUILES CARRERA
- FLP Motors Litt BBLDiunggah olehshaikkareemullah70
- Synthesis, Characterization and Catalytic Performance of Mo-TUD-1Diunggah olehAnoop
- Lead Paint (1)Diunggah olehkulihat_hijau
- X-Ray Computed Tomography (CT)Diunggah olehrvpilot
- Chromatography Chemistry SbaDiunggah olehJasonKoylass
- Bioenergetics Energy Flow, Secondary Production, Ecol Efic Mad Cockroaches 18-ShanholtzerDiunggah olehCHRISTEROP
- Fully Convolutional Neural NetworkDiunggah olehIlma Arifiany
- BP(Public)1Diunggah olehKai Esperon
- Energy and the EnvironmentDiunggah olehkwakseok812
- TN27 Australian Standard for Precast BC and HeadwallDiunggah olehgemotorres
- Abstract - Earthquake Effects on Single Layer Lattice Domes With Supporting FramesDiunggah olehARNoshnagh
- Journal Bearings Practice.pdfDiunggah olehPrabhath Kumara