Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Pmmdings aftbe 41d lEEE

Conferenceon Decisiou and Contml


Las Vegas, Nevada USA, December 2002 WeA05-2
Robust Control Design Using the PID Controller'
Guillermo J. Silva Aniruddha Datta3 S.P. Bhattacharyya

Abstract function

In this paper we present useful tools and criteria for designing a


PI or a PID controller for a first-order system with time delay. where k is the steady-stste gain, L is the apparent time delay, and
These tools show the importance of knowing apriori the set of T is the apparent time constant. W e will consider the feedback
controller parameter values that stabilize the closed-loopsystem. control system shown in Fig. 1 where r is the command signal;
The characterization of these stabilizing sets was derived in ear- y is the output of the plant, C(s) given by (1) is the plant t o be
lier works [I, 21 and semes as the basis of the synthesis tools controlled, and C(s) is the controller. In this paper; we consider
presented in this paper.

1 Introduction

In general, the task ofa controller is to maintain B desired system


performance while coping with any system disturbances. In the
case of the PID controller, to achieve a desired performance, the
user needs to select carefully the amount of each contmi action: Fig. 1: Feedback control system.
proportional, integral and derimtive. The main problem that can
occur if the parameters &re chosen incorr&ly is -illation and
Instability. In an industrial situation this can lead to material the following controllers: (a) Constant gain controller: C(s) =
lass and destruction ofequipment. This is the reasan why setting kp; (b) PI controller: C(s) = k,++; (e)PID controller: C(8) =
the parameters of a PID controller is an important step in the kp+ $ +kds where kp is the proportional gain, ki is the integral
design of PID control systems. g a n and kd is the derivative gain.

In this paper we discuss different approaches for designing PI and The following theorems [I, 21 provide an analytical characteriza-
PID controllers for firstader plants with time delay. As will he tion of the e t of controller parameten for which the closed-loop
e n later, i t is important far the designer to know spriori the set system in Fig. 1 is stable. Throughout the paper, w e a ~ ~ u m
that
s
of stabilizing controller parameter Miues. This characterization the plant is open-loop stable, i.e, T > 0 and that k > 0, L > 0.
wm derived in [I, 21 and forms the basis of the synthesis tools We start by considering the case of a comtant gain controller.
presented in this paper. First, the c- of 8. system with a~
unknow time delay is considered. It ia assumed that an upper
bound on the time delay is known and using the Walton-Marshall T h e o r e m 2.1 Under the above ossumplions on k and L, the
procedure 131 and earlier stabilization results 121, a robust PID get of oll stabilizing gains kp for D given open-loop stable plant
controller design is derived. Next, we incorporate into the PID with honsJer function C(s) M in ( 1 ) is given by
controller design, time performance specifications such BS settling
time and maximum overshoot.

The paper is o r g a n i d as follows. Section 2 contains some prs-


1
-i<k,<-
'J
kL
a;+??
L2

where a, is the solution of the quotion tan(a) = - $ a in the


liminary stabilization results that form the basis of the design
took preented In later sections. Section 3 considers the case internal (f ,x).
of a system with unknown delay which requires the design of a
robust PID controller. Section 4 introduces time domain per- Moreover, by explicitly evaluating the derivative with respect t o
formance specifications into the PID design. Finally, Section 5 L , it can he shown [4] that the upper bound for k, given in
presents some concluding remark. Theorem 2.1 is a monotonically decreasing function of the time
delay L of the system. This is formalized in the following lemma.

2 Stabilization of First-order Systems with Time delay


L e m m a 2.1 The upper bound fov kp given in (2) b a mono-
The design techniques presented in this paper are bssed on char- tonically d e e d n g function of the time delay L oJ the system.

'This work -
acterizing the plant to be controlled by the following transfer

supported in part by the National Science


Foundation under Grant No. ECS-9903488, in part by the Texas
Advanced Technology Program under Grant No. 000512-0099-
For the case of PI controllers, we are interested in determining
the regipion in the kp-k(parameter space for which the dosed-loop
system is stable. The following result provides the range of k,
d u e s for which B solution exists to the PI stabilization proh
1999, and in part by the National Cancer Institute under Grant lem. Then, the complete stabilizing region can be determined
No. CA 90301. following the procedure described in [I].
'IBM Server Group, 11400 Burnet Road,Austin, TX 78758,
U.S.A., *mail: guilsilvOuus. i b m . cam
3Department of Electrical Engineering, Texas A & M T h e o r e m 2.2 Undw the e&ve os9umpfi0nr on k and L. the
University, College Station, TX 77843-3128, U.S.A., email: mnge of k, values fw which (1 solution -1s to the PI dabi-
datta@ee.tamu.edn, bhattOee.tmu.edu lizatirm pmblern o/ D givm open-Imp *table plant u i t h hawfw

0-7803-7516-5/02/$17.00 02002 lEEE 1313


function G(s) (U in (1) is given b y In Fig. 2, the parameters mj. b j , and w j , for j = 1,2, are defined
as follows:
-61 < k, < &-
v h m ai is the solution of the equation
T
tan(a) =--a
L
in the internal ( ,; r)
where 11, 1 2 , IT? > 21 are the solutions of kkp + COS(T) -
Moreover, for eoch k , inside the mnge in (3)$ the stabilizing k; fzsin(r) = 0 in the interval (0,~).
values ore given by
To conclude this section, we present the procedure introduced by
0 < k; < min
j=1,3.5. ...
{ 2[sin(y)+ f -zj eos(q)
I>+ Walton and Marshall in 1987 131. This procedure is a systematic
way t o study the stability of systems with a single time delay,
i.e., s y s t e m with the following characteristic equation:
wh- zj, j = 1 , 2 , 3 ,..., am the sdtltions of kkp cos(z)
+ n(s)eC"
~

$isin(=) = 0 armnged in ine-ing order of magnitude. 6(s] = d ( s ) (8)


where d ( e ) and .(a) are polynomials with real coefficients,
deg[d(s)]= q, de&(.)] = p , and L > 0 is the time delay of
Finally, for the case of a PID controller, the following theorem the system. The condition for stability is t h a t all the roots of
provides an analytical characterization of the set of controller the characteristic equation (8) lie in the open left-half of the com-
parameten ( k p ,ki, kd) for which t h e closed-loob system is stable. plex s-plane. Thus, in our c-, the basic problem of stability
is t h a t of determining the range (or ranges) of values of L for
which this occurs.
Theorem 2.3 The mnge of k, valuer fov which a given open-
loop stable plant, with tmnsferJ"nctionG(s] (U in (l), continues The procedure developed by Walton and Marshall allows us t o
to h o w closed-loop stability with a PID eontmller in the loop w analyze the behaviour of the roots of (8) as L inme- from 0 t o
@"en b y m and consists of three basic steps. The first step is t o examine
the stability of (8) for L = 0 and determine the number of roots,
-61 < k , < [$I sin(a1) - C O S ( U I ) ] (5)
if any, of a(,) = 0 not lying in the open left-plane. The second
step considers the case bf an infinitesimally small positive L. For
this d u e there will be an infinite number of new roots and it
where a1 is the solatton of the equation
is necessary t o find out where in the complex plane these roots
T have appeared. T h e third and final step is t o find positive values
t a n ( a ) = --a (6) of L,if any, a t which there are roots of 6(s) = 0 lying on the
T+L
imaginary axis and then t o determine whether these roots merely
in the interval (0,~). F o r k , values ouMde thw rnnge, there ape touch the axis or whether they cross from one half-plane t o the
no stabilizing PID contmllers. The eomplete stabilizing region other with increasing L. Roots crossing from left t o right rue
*.given by: ($e Fig. 2) considered destabilizing and those crossing from right t o left are
considered otabilizmg.
1. For each k,, E (-i,i), the cmss-section of the stabilizing
We can summarize this procedure as follows 151.
-on in the ( k i , kd) space CS the hnpemid T.
2. For Iop = 1, the cross-section of the stabilizing -on in Step 1. Examine the stability at L = 0.
the ( k ; ,k d f space is the hiangle A.
Step 2. Csnsider an infinitesimally small positive L. If q > p , all
3. For eoch k p E ( i . k . , , : = [f.l.i.(.l)-~~~(al)]), the new rOGtS will lie in the open left-half plane and this step c m
the cross-section of the stabilizing W o n in the ( k ; ,kd) be omitted. If q = p, the location of the new roots is determined
space i s the q u a d d a t e d Q. by the sign of W ( u * ]for large w , where
w(w') = d(jw)d(-jw) - n(.jw)n(-jw). (91

Step 3. Determine the positive roots of W ( w 2 )= 0,the come-


sponding positive values of L and the nature of these roots. If
there are no repeated roots,then the stabilizing and destabilizing
mots alternate. Moreover, in the majority ofs y s t e m the highest
root is always destabilizing. Thus, we can label the roots in de-
scending order as destabilizing, stabilizing, and so on. The same
procedure c m be used for the corresponding d u e s of L in order
t o determine for what values of L all the roots of 6(s,L) = 0 lie
in t h e opes left-half plane.

3 Robust C o n t r o l l e r Design

In this Section we consider the problem of stabilizing B first-order


Fig. 2: The stabilizing region of (k;:kd) for: (a) -t < system with time delay, where the time delay of the system is
k, < (b) kp = t; (c) t
i; < kp < k,,,. unknown but lies inside a known interval. This generates a one-
dimensional family of plants. Similar problem such as stabilizing

1314
plant families with uncertain delays have generated interesting Proof. T h e idea of the proof is to follow the threestep procedure
research results (see [SI and the references therein). kr we will introduced in Senion 2. This procedure allows us t o analyze the
show shortly, it is sufficient t o design a P, PI or PID controller behaviour of the mots of 6 ( s , L ) = U when the time delay L of
that stabilizes the system with the upper bound on the time the system increaSeS from 0 to +m.
delay. This guarantee that the plant family is stable for any
value of the time delay inside the given interval. Let us denote by k; and k: some controller parameter values
that stabilize the delay-free system and the system described by
We start by considering the plant family G(s): (IO) with t h e time delay set to L'. Also, we rewrite 6(s, L ] as
follows
+
6(s, L ) = d ( s ) n(s)e-"
where
where L is unknown and
d(s) = Tsz+s
L E lL1, L21 . n(8) = kk;s+ kk:.
For convenience we will focus on the c ~ s eof an open-loop stable
system, i.e., B system where T > U. Step 1: stability at L = 0 . This follors from our a s s u m p
tion concerning k; and k:. Note that in Theorem 2.2, we have
3.1 Robust Stabilization U s i n g a C o n s t a n t G a i n imposed this (reasonable) requirement on any stabilizing PI con-
In this c a e , the controller C ( s ) is a constant gain, i.e., troller.
C ( s ) = k, . Step 2 . increment L from 0 to an infinitesimally small and
The problem of characterizing all constant gain controllers for the positive number. Since the degree of d ( s ) is bigger than the
plant family P(s) is to determine all the values of the controller degree of 4 s ) we conclude that all the new roots lie in the open
parameter kp for which the family of first-order systems with le&-half plane (see Section 2).
time delay L E [L, L z ] is stable. Now, from Theorem 2.1, for a
fixed value L' E [ L , , L z ] ,the range of stabilizing constant gain Step 3: potential crossing points on the imaginary axis. First,
values is given by we determine W ( w z )fmm (9):
+ (1- k2k;')w2
1
-i<kp<-
T/7
k L'
where a1 is the solution of the equation
a:+- Then,
W ( w 2 )= T2wr

W ' ( w 2 ) = 2 T Z w 2+ 1- k'k;'
- k'k;'

.
. (12)

Now, the roots of W ( w z ) are given by the following expression

in the interval (4,~). Moreover, Lemma 2.1 states that the


upper bound in the previous expression is a monotonically de- which can be rewritten as
creasing function of the time delay L of the system. Thus,

Clearly, the expression inside the square root is always bigger


than 1. Thus. we have
for any L' E [ L , , L z ] . Notice that in the previous inequality, the
two parameters 0 1 are not the same as can be seen from equation
(11) where L appeam explicitly.

Thus, it is not difficult t o see that if we design for L z , the set of


resulting stabilizing constant gain controllers will stabilize any
element of the plant family G(8).

3.2 Robust Stabilization U s i n g a PI C o n t r o l l e r


In this subsection, we consider the problem of determining the
set of PI controllen that stabilize the plant family P(a) in (10). + sgn[wg] = 8gn [I - k*k;'] .
Since the controller C ( s ) is now given br
Now, we are only intereted in the positive roots of W ( w 2 ) .From
the previous expresliom, it is clear that for a fixed d u e of the
C ( 8 )= kp +-
ki
controller parameter kp # i,there is only one positive root of
the family of closed loop characteristic polynomials 6 ( s , L ) be- W ( w z )since w: and w: h a w opposite signs.
comes
6(s. L ) = Ts' + +
s +
(kkp,s kki)eCL' Consider t h a t k; is such that is the only positive root of
W ( w z ) . We now determine the value of S given by
where L E [ L I La].
, As in the previous s u k c t i o n , our objective
is to determine the values of k, and ki for which the entire family
of elm& loop characteristic polynomials 6(s, L ) is stable.
s = - .
sonlW'fw?N
, .,.
We Start by presenting the following result which is based an the
material introduced in Section 2.
= --sgn [ l - k'k;']
and from (13)we conclude that S = 1. Thus, this root is desta-
Lemma 3.1 Consider the system wtth tmnsferfunetion (10). bilking. For the case when k; is such that wz is the only positive
If (I given PI cDnholler stabilizes the delay-free system and the root of W ( w 2 ) ,a similar analysis yields
system with L = L* > 0 , then the same PI mntmller stabilizes
the system v L E [U, L']. s = sgn[W'(w;)l= 1

1315
Moreover, if k; = i.
then (12) reduces t o the form by (IO). Since the controller C(8) in question is now a PID
controller, i.e.,
W ( w z ) = Tzu4 - k Z k f Z.
C ( s )= kp + k + kds
In this case there is oniy one positive mot a t w: = and it is the family of closed-loop characteristic polynomials 6(s, L ) be-
not difficult t o see that comes
S = sgn(W'(w:)] = sgn[ZTkkf]. 6(s,L) = Tsz + 8 + ( k k d + kk,a + kk,)eCL"
Since k;, k: stabilize the delay-free system, it follows that k f > where L E [ L I , L z ] . We want t o determine the values of the
0. Hence, we conclude that S = 1 for k; = f . controller parameters kp, k,, kd, for which the entire family of
closed-loop characteristic polynomials a(s, L) is stable.
Thus, in any case, there is only one positive root of W ( w z ) ,and
this root is always destabilizing. Now, the corresponding values As in the last subsection, we start by presenting an interesting
of L are given by 151: lemma based on the results introduced in Section 2. This lemma
states that if a given PID controller stabilizes the delay-free sys-
tem and the system in (10) with time delay L', then it also
stabilizes the same system for all time delay L E 10, L'].
Solving for L we obtain L = b l . b,. . ., where 0 < bl < bz < . . .
are real numbers. This means t h a t a t L = bi, two roots of
6 ( e , L ) = 0 cross from left t o Eight of the imaginary axis. Then,
two more cross a t L = bz and 50 on. We conclude that the only
region of stability is 0 5 L < L,,, where Lmm== bi.

Now, since the closed-loop system is stable for L', then it fol-
lows from the previous discussion that L' is inside the intern1 Proof. The proof of this lemma is similar t o that of Lemma 3.1
(0,Lmar). Hence, the closed loop system is stable for L E 10, L - ] and follows the three-step procedure introduced in Section 2.
and this concludes the proof of t h e lemma.
Let us denote by k;, k f and k; some controller parameter values
In view of Lemma 3.1, any controller taken from the set of sta- that stabilize the delay-free System and the system described by
bilizing PI controllers obtained for the given first-order system
(IO) with L = L'. Also. we rewrite 6 ( s , L ) as follows
with the time delay set t o L z (known upper bound of the time
delay) will stabilize the piant family G(s). This set can be ob- 6(s, L ) = d ( s ) + n(s)eCLa
tained using Theorem 2.2 and following the algorithm presented
where
in ( 1 ) . The following example illustrates this property.
d(s) = Tsz+s

E x a m p l e 3.1 Consider the plant family


n(s) = kk;s2 + kk;s + kkf .

Step 1: stability a t C = 0.' This follows from our s u m p t i o n


concerning k;, k f and k;. Note that in Theorem 2.3 (see also
where L E [ l , 3 ] sec. By Vszng Theorem 2.2 and the pmcedure Fig. 2), we have imposed this (reasonable) requirement on any
developed i n [ I ] we um f i n d the set of stnbilving PI ulntmllers stabilizing PID controller.
fw different values of the time delay L E ( 1 , 3 ] . Fig. 3 shows
thesesetsforL=1,1.5,2,3. Asean b e s e e n f m m t h i s f i g u r e t h e Step 2 : increment L from 0 t o an infinitesimally small and
positive number. Since the degree of d ( s ) is equal t o the degree
of n(s) we need t o consider the behaviour of W ( w 2 )for large w z
I.iI ,/- (see Section 2). From (9) we have:
W ( w z ) = (Tz - k2k;')w4 + (1 + 2k'k;kf - k*k;')w* - kzkf2 .
Now, for large w z we have
lim W ( w Z )= (T' - k'k'' lim w4 .
wa-+m

It follows from Fig. 2 that -E


< k; < (see also 121). Thus,
- kZkZz > 0 so W ( w 2 )> 0 for large w? We conclude that
the infinite number of new roots occurs in the left half-plane.

Step 3: potential crossing points on the imaginary axis. From


the expression for W ( w z ) derived in the previous Step w e h a w

W ' ( w z )= 2(Tz - kZk;')wZ + (1+2kzk;kf - k*k;') .


Now. the roots of W ( w 2 ) me given by
Fig. 3: Sets of stabilizing PI controllers for Example 3.1
- ( 1 + 2 k 2 k ; k f - k'k;')
U:,"; =
intersection of all these sets U the set corresponding to L = 3 2(TZ - k2kz2)
(dashed am). T h w , any PI conhollerfrom thU set Unll stabilize
the whole family ofplants described by G(8). a +
(1 2kzk;k: - kzk;z)z + 4 ( T z - kzkZ2)kzkfZ
Z(Tz - kZkiz)
which can be rewritten as
a PID C o n t r o l l e r
3.3 R o b u s t S t a b i l i z a t i o n U s i n g
In this subsection, we consider the problem of finding the set ~ (1 +Zk*k;k; - k2k;l)
of PID controllers that stabilize t h e plant family P(s) described WllWZ --
2(TZ- k2k;2)

1316
where (1 function of the time delay L . This upper bound is given by the
4(Tz - kzk;z)kzk;' Jollouir~gezpression:
7=
( 1 + 2 k z k j k f - k2k1;z)2
Clearly, the expr-ion inside the square root is always greater
than 1 . Thus, we have where 01 is the solution of the equation
T
tan(u) = --a
T+L
in the interval (0,a). By sweeping me? the time delay L , we
+ sgnlw;] = -sgn [I + 2 k 2 k ; k r - k'k;'] (14) obtained the plot s h o w in Fig. 4. This plot shows the behoMour
of the upper bound kuPP ~9 D function OJ the time delay of the
since TZ- kzkZz >0 BS pointed out in Step 2. Moreover, system.

.; = - (1 + 2kzk;kf - kZ$')
2 ( P - kZk;2) k-45
sgnlw;] = sgn [ l + 2 k z k ; k t - k'k;'] .
Since w; and w; have opposite signs, we conclude t,hat for kp =
k;, ki = k,?, kd = k;, there is only one positive root of W ( w z ) .

Consider now that the controller parameter values are such that
U: is the only positive root of W(wz). Then, the value of S is
given by

s = sgn[w'(w:)]

= sgn [-(I + 2k'k;kl - k2k;')(1 +6 ) Fig. 4 Plot of the upper bound kupp a
s a function of t h e
+I t 2k2k;k: - k'k;'] time delay for Example 3.2.

= sgn [-(1+2kZk;k; -kZk;')&] As we can see fmm this plot, the uppep bound k,,, is a mono-
t o n i d l y dec-ing function OJ t h e time delay of the system.
= -sgn [l + 2 k z k ; k l - k'k;'] This implies that if we design, JOT ezomple, for L = 1, we will
obtain PID contmllers that pmduce an unstable behaviour for
and from (14) we conclude that S = 1. Thus, this root is desta- some of the members OJ the plant family G(8). Thus, we toke
bilizing. For the c a ~ ewhen the controller parameters are such k,,, = 1.3045 (value eomsponding to L = 3 ) os D safe upper
that uf is the only positive mot of W(w2), a similar analysis bound for the contmller pawmetet- k p .
yields
s = SSn[W'(w;)l = 1 . Nezt, 'we f.z the controller pommeter kp at 1.2. Using Fig. 2
we now determine the stabilizing mgion oJ (ki, k d ) values JOT
Thm, in any case; there is only one positive mot of W ( w z ) ,and
dlffemnt values of L E [ 1 , 3 ] . Fig. 5 shows these set8 for L =
this root is always destabilizing. Now, the corresponding =lues
1 , 1 . 5 , 2 , 3 . A s urn be seen fmm this figure the intersection of
of L are given by 151:

Solving for L we obtain L = b l , bz, .. ., where 0 < b i < bz < . . .


are real numbers. This means that a t L = b l , two roots of
6 ( s , L ) = 0 cross from left to right of the imaginary axis. Then,
two more cross at L = bz and so on. We conclude that the only
region of stability is 0 6 L < L,,,, where Lma= = 61.

Now, since the closed-lwp system is stable for L * , then it fol-


lows from the previous discussion that L' is inside the interval
(0,Lmor).Hence the closed loop system is stable for L E 10, L']
and this concludes the proof of the lemma. w
In view of Lemma 3.2, we conclude that when solving the PlD
stabilization problem for the plant family G ( s ) in (IO), it i s suf-
ficient to solve the PID stabilization problem for the case of
L = L z (known upper hound on the time delay). This will gen- Fig. 5: Sets of stabilizing ( k i , kd) for Example 3.2 (kn =
erate a set of PID controllem that stabilizes the family of plants 1.2).
with time delay in [O,Lz],which is a superset of [ L i , L z ] . We
now present a simple example to illustrate these observations. all these ret8 is the set corresponding to L = 3 (shaded a m ) .
Thus, when kp is set equal to 1.2, any (k., kd) fmm this set will
stabilize the whole Jomilg ofplants described by G ( s ) . A
Example 3.2 Cowider the same plant family ~ 1 sin Ezomple
s.1:
5(s)= Le& 4 Time Domain Performance SpeciAeationr
S+l
where L E 11.31 see. By using Theorem 2.9 we urn study the Before designing a controller, it i s important to understand the
behaviour OJ the upper bound OJ the contmller pammeter k, a3 primary objmtive of control. A common type of control objective

1317
is to follow a setpoint or a Step input. Specifications on setpoint Although the design presented above are essentially brute force
following may include requirements on rim time, Settling time, optimization searches, nevertheless the fact that the results of
decay ratio, maximum overshoot and steady-state offset for step Section 2 can be used to confine the search t o the stabilizing set
changes in the setpoint. These are time domain performance makes the design problems orders of magnitude easier.
specifications that need to be incorporated into the design of the
controller.
6 Concluding R e m a r k s
The purpose of the integral term in a PI controller is to achieve
zero steady-state offset when tracking step inputs. Thus, we can In this paper we have presented some tools for designing PI or
employ the time domain specifications 171 mentioned above to PID controllers for first-order systems with time delay. These
quantify the performance of a PI controlled closed-loop system. tools require apriori knowledge of the set of controller p a r a m e
The characterization of all stabilizing (kp,k,)values provided in ter values that stabiiiee the closed-loop system. First, the ease
[I] enables us to graphically display the variation of these perfor- of a system with unknown time delay is considered. This gen-
mance indices over the entire stabilizing region in the parameter erates a family of plants that needs to be stabilized. It is only
space. Using such a tool, we can select the (kp.ki) vdalues that known an upper bound of the time delay. Using the character-
meet the performance specifications. Moreover, using the results isation derived in earlier works [I,21, a robust PID controller
of Section 3,we can obtain a robust PI controller when the time is proposed for the plant family. Next, some time domain spec-
delay of the system is not known. The following example illus- ifications are introduced into the design and again, the set of
trate the procedure involved. stabilizing controller parameters plays B crucial role. Using the
computational capabilities of modern computers, it is possible to
graphically display the variation of several performance indices
E x a m p l e 4.1 Consider the pmblem of choosing stabilizing PI over the entire stabilizing region in the parameter space. Using
gains for the following plant: such a tool, we can select the controller parameter values that
meet the performance specifications
qs)= Le-..
4.+1
~I ..
where L E 10.8.11. The . .
Derfomance . .
soecifieotions that we are
required to meet when deszgning the PI contmller are the follow-
ing: (1) Settling time 5 30 sees; (2) Overshoot 5 20%.

A s in ezample 3.1: we know that for L = 1 we obtain a set of


mbwt wntmllers that stabilize the whole family of plants P(8).
The stabilizing region in the kp-k. plane for L = 1 can be ob-
tained using the olgodthm in 111. Nezt, we obtain by simulation
the tmnsient responss of the closed-loop system fo?'the (k,,ki)
values imide the stabilizing &on for L = 1. In order to ob-
tain a redient OT non-fmgde contmller (a concept intmdueed in
[8]), we only consider the (k,,k;) values imide the follolving boz
defined in the pammeter space:
2 5 kp 5 4 and 0.5 5 ki 5 1.5 .
In this way we can alleviate the controller fmsilitv pmblem to
some d e n t . Fiq. 6 disployr the vanation of the manmum m e _
Fig. 7: T i m e response for t h e closed-loop system of Ex-
shwt mkibited by the closed-loop system od D fanetion of t h e ample 4.1. . '

pmpTtiona1 and integml gains. This kind of plot ollows us to

References
.111. .. S. P. "PI Sta-
Silva, G. J., Datta, A., and Bhattachsryya,
bilization of Firstarder Systems with Timedelay," Automatica,
Vol. 37,pp. 2025-2031,2001.
(21 Silva, G. J., Datta, A., and Bhattacharyya, S. P. "New
Results on the Synthesis of PID Controllers," IEEE nasaetions
on Automatic Control, Vol. 47,pp. 241-252, 2002.
131 Walton, K., and Marshall, 3. E. "Direct Method for TDS
stability analysis," P r o d i n g a of IEE, Part D, Vol. 134, pp.
101-107, 1987.
(41 Silva, G. J., Datta, A., and Bhattacharyya, S. P. "Stabi-
lization of Time Delay Systems," Proceedings of the American
Control Conference, pp. 963-970,2000.
[5] hlarshall, J. E., Gorecki, H., Korytowski, A., and Walton,
K. Timedelay Systems: Stability and Performance Criteria with
Applications, Ellis Horwood, New York, 1992.
Fig. 6: Plot of maximum overshoot versus (k,,k<) for Ex- 161 Kharitonov, V. L., and Zhabko, A. P. "Robust Stability of
ample 4.1. Timedelay Systems," IEEE ltansactions on Automatic Control,
Vol. 39,pp. 2388-2397,1994.
pefom D s m z h for those (kp,ki) values that meet the desired
[7] Astrom, K. 3, and Hagglund, T. PID Controllers: TheoT,
performance specifications. Among all these values, we set the Design and lhlirning, Instrument Society of America, Research
contmllerpommeters to: kp = 2.1053,k; = 0.7105. Fig. 'Ishows Triangle Park, NC, 1995.
the tine m p n s e of the corresponding closed-loop system. As we
181 Keel, L. H, and Bhattacharyya, S. P. "Robust, Fragile or
can see fmm this figurn, the closed-lwp system with the designed Optimal?" IEEE ltansactions on Automatic Control, Vol. 42,
PI controller meets the performance specifications and by design pp. 1098-1105.1997.
we ore also guaranteed to have a non-fmgile PI controller. A

1310

You might also like