Professional Documents
Culture Documents
6, DECEMBER
1261 1981
Short Papers
A Process-ModelControl for Linear Systems with problem with time delays. Lewis [8] and Manitius et al. [9] considered the
Delay finite spectrum assignment problem of delay systems. The resulting
control law contains certain functionals of the present state of the delay
KEIJI WATANABE AND MASAMI I T 0 element. To construct the control systems, it is necessary to continually
measure the present state of the delay element or measure the functionals.
Abstrucf-This paper is concerned with linear systems containing time Various methods of the implementation of the functionals are suggested
delay in control. A process-modelcontrol which utilizes amathematical in [4]. One of them is to use a certain model of the process. By passing the
model of the process in the minor feedback loop around the conventional control signal through the model, the functionals can be generated without
controller to overcome delay is developed.Anewprocess-modelcontrol the measurement of the state of the delay element [4]-[6], [lo]. The
system vhich canyield zerosteadystate erroranddesiredtransient resulting control systems are similar to the Smith predictor control one.
responses to step disturbances and arbitrary initial conditions is proposed. Many papers on the control of the time delay system which have been
The remarkable feature of the proposed system is thatthe states of the published embrace Smith’s principle explicitly or implicitly [IS], [19]. In
model and the integrator are unobservable in the output of the model. It is this paper, the Smith predictor method and the state space approaches
shown that this structural property is the key by which the major problems with the implementation of the functionals by a process model are said
of process-model control can be solved. generically to be process-model controls [3].
Among process-model control systems -which have been reported, there
exist significant differences of characteristics [ I 11, [13],[21]. Donoghue
I. INTRODUCTION
[ 1 I], [ 131 compared the characteristics of two control systems resulting
Time delay occurs frequently in process control problems. Compared to from application of the Smith predictor method and the optimal design
processes without delay, the presence of delay in processes greatly com- approachto a first-order lag process with delay, and showed thatthe
plicates the analytical aspects of control system design and makes satis- Smith predictor control system can yield zero steady-state error to step
factory control more difficult to achieve. disturbances, whereas the optimal system has nonzero steady-state error
Smith [l] proposed a delay compensation technique wbich utilizes a to the same disturbance even though the controller contains an integrator.
mathematical model of the process in the minor feedback loop around the Cook and Price [ 121 had objection to Donoghue’s results and pointed out
conventional controller. This technique became known as the Smith that, if the disturbances are steps of random magnitudes, then the Smith
predictor method [2]. The main advantage of the Smith predictor method predictor is optimal. However, Cook and Price did not consider the
is that time delay is eliminated from the characteristic equation of the transient responses. Considering them, one may see that the Smith
closed loop system. Thus, the design problem for the process with delay predictor suffers some shortcomings. In particular, if the process has poles
can be converted to the one without delay. near the origin in the left half plane, then the responses may be sluggish
On the other hand, an alternative design approach for systems with enough to be unacceptable. Many improvements have been tried [19], hut
delay has been discussed using state equations [4]-[9]. Fuller [4], Mee [5], the problem has not been solved to a satisfactory level within the
Kleinman [6], and Koivo er al. [7] investigated the optimal regulator framework of feedback. Consequently, disturbance attenuation has been
still one of the major problems of process-model control.
ManuscriptreceivedJuly 16, 1979; revised May 5. 1980 and December 9. 1980. Paper
recommended by W. E Schmitendorf, Chairman of the Optimal Systems Committee. Furthermore, there exist differences of characteristics among some state
K. Watanak is with the Department of Uectronic Engineering, Faculty of Engineering. space approaches. The control systems presented by Kleinman [6] and
Yamagata University. Jonan. Yonezawa-shi. Yamagata 992, Japan
M. Ito iswiththe Automatic ControlLaboratory.Faculty of Engineering. Nagoya Marshall et a/. [lo] can yield desired transient responses to arbitrary
University, Furo-cho. Chikusa-ku. Nagoya 4 6 4 , Japan. initial conditions of the system, but those of Fuller [4] and Donoghue [ 1I]
cannot. The mechanism which causes these differences has not been
clarified.
This paper is concerned with process-model control for linear systems
containing delay in control subject to step disturbances. The concept of
the error system [14], [I51 is used because of its advantages forthe
treatment of disturbances and initial conditions. A new process-model
control system, which can yield zero steady-state error and desired
transient responses to stepdisturbances and arbitrary initial conditions, in Fig. 1. Smith predictor control system
contrast to the otherexisting ones, is proposed. The remarkable feature of
the proposed system is that the states of the delay-free part of the model
and the integrator are unobservable in the output of the model. This
structural property is the key by which the major problems of process-
model control can be solved. This is illustrated by comparing the pro-
posed system with the others.
11. SYSTEM
DESCRIFTION
In this paper, we consider a single input-single output process with
delay in control described by I I'
r(t)=Cx(t)
x(0)=xo, u(T)=u~(T) -L~T<O (1)
CONTROL
111. PROCESS-MODEL
T- I+- s + l l
Id
U
s + a
-3L y ,
process
process-model
I
Fig 6. A new process-model control systemfor the regulator problem.
Fig 4. Optimal control system generalized by Cook and Price.(If M = I, then the figure
is equivalent to the Smith predictor conuoL If M = e-'L, then the figure is equivalent to The time delay is expressed as an integral-termof the output equation in
that of Marshall ef or.)
(IO), whereas it is denoted by u(r-L) in (I). Lety(r) be
x'(t)=Ax(t)+Bu(r)
v(r)=Ce-ALx(r). ( 12)
N. A NEWPROCESS-MODEL CONTROL
In this section, wewill propose a new process-model control system where d is given by
which can yield zero steady-state error and desired transient response to
step disturbances and arbitrary initial conditions.
We first suppose that both a reference input and a disturbance are zero
+
and consider a regulator problem for (1). The state at1 L for (1) is given
by (5). Let X( t ) be X( t) = x ( r + L), then (1) can be rewritten as follows: The output of the process is required to follow the step reference input r
with zero steady-state error, where r is given by
proce~s-rmodel (30)
I I
Let [ ( t ) be (b)
Fig. 7. A new process-model control system for the sewomechanism problem. (a) The
system derived from that of Fig 6 and @) the reduced system.
Then we have
i
i p t() =Axp( t ) +Bu( t )
1 i(t)=aS(t>+io(t>
g ( r > = -te-"g(r)
shown in Fig. 7 (a) remain unchanged in the system of Fig. 7 (b). That is
$(I) = t ( t + L ) to say, in Fig. 7 (b), x p and the state of the integrator are unobservable in
=eA^,[(t)+jr ei(l-T)Bo (7 (25) Y, at ra L. The direct path from u to . y L , that is, the third term on the
r- L output equation of (30). is needed to make the state of the integrator
unobservable in the output of the model. This structural property is the
and e(?)is key by which the major problems of process-model control systems can be
- 41., -
?(r)=e(t>-&" A(r-r)& 7 solved. This can be clarified via comparison of the characteristics of
( )dT. (26) process-model control systems.
Since ( A , k ) is controllable ant
(tCAL, 2) is observable, there exists a
v. COMPARISON OF CHARACTERISTICSOF %
MEPROCESS-MODEL
controlinput u ( t ) such that [ ( t ) - O as t-cc, in the sense of pole
assignment of the closed loop system. It follows from (24) and (26) that CONTROL
SYSTEMS
e( t ) -0 as r w in the same condition. The integral term on the right of
+
A. Response to a Srep Reference Inpui
(26) can be implemented by the following model in the same manner as
(13): Generally, the process-model control systems are represented by a
block diagram as shown in Fig. 8, where G ( s ) = C ( s l - A ) - ' B , G , ( s ) =
C , ( s I - A ) - ' B + D and M is the constant number. Moreover, C , is the
1 X n constant matrix and D is the constant number. It is assumed that the
controller G,(s) contains an integrator, (C,, A ) is observable, and
(34)
r- m s-0
where r ( s ) = 1,’s. Therefore, the proposition is true. = lim { G,(s) - M G ( s ) e-‘,) (39)
Remark V.1: In the proposed system, C,(s) is s-0
where d(s)= 1,’s. Equation (39) implies that the proposition is true.
G,(s)=Ce-AL(sl-A)-’B-JLCe-”rBd7.
0 Corollary V.1: The condition given by (38) is equivalent to the one that
the state of the integrator in the controlleris unobservable in y L .
If the process does not have poles at s=O, that is to say, A is nonsingular, This can be proved as follows. If (38) holds, then the process model has,
then at least, a zero at s=O. Hence, the state of the integrator is unobservable
in y,. Conversely, if the state of the integrator is unobservable iny,, then
the state does not affect yL since “unobservable” means that the output
matrix corresponding to the state is zero. Equation (38) holds.
Corollary V.2: When the process is asymptotically stable,the necessary
On the other hand, if A =0, B = 1, and C= 1, then and sufficient condition for yielding zero steady-state error to step dis-
turbances is that (38) is satisfied.
Remark V.3: In the proposed system, the states of the delay-free part
of the model and the integrator are unobservable in the output of the
model at t 2 L. These imply that poles of G , ( s ) - M G ( s ) e - ’ L are canceled
with its zeros and (38) is satisfied. Therefore, the errordecay can be made
as fast as desired.
This argument can beextended to the other processes with integral Remark V.4: In the Donoghue system shown in Fig. 3, thetransfer
properties. The proposed system satisfies (33). function of the model is given by
Remark V.2: Inthe systems of Smith, Donoghue, and Cook et d . ,
G,(s)=G(s). Then (33) is satisfied.
B. DisturbanceAttenuation
Clearly, the pole of the model is canceled with the zero. However, the
The behavior of the process-model control system subject to dis-
transfer function as s 0 becomes
+
turbances is as follows.
Proposirion V.2: In the system of Fig. 8, let the initial condition of the
system bezeroandthe reference input be zero. To yieldthe desired
transient responseto the disturbance afterL , it is necessary and sufficient
that poles of G , ( s ) - M G ( s ) e p Sare
L canceled with its zeros. Equation (38) is not satisfied.Donoghue’s system has nonzero steady-state
Proof: In the system of Fig. 8, the transferfunction from the error to step disturbance even though the controller contains an integra-
disturbance d to the output y is given by tor. The system of Fig. 4, where M=e-aL, is in the same condition.
Remark V.5: In the Smith predictor control system, the transfer func-
tion of the model isgiven by G , ( s ) - M C ( s ) e - ’ L = G ( s ) - G ( s ) e - ” L ,
where G ( s ) = C ( s l - A ) - ’ B . If the process is asymptotically stable, then
the system has 7.ero steady-state error because the impulse response of
G ( s ) - G ( s ) e - ’ L becomes zero at t-03 and (38) is satisfied. However,
poles of G(s)- G(s)ePELare not canceled with its zeros except for the
process with n= 1 and A=O. The Smith predictor control cannot yield
Poles of CJs) are zeros of I+G&s)G,(s) andpoles of G,(s)- desired transient response to disturbances. If the process has poles near
~ , together.Zeros of 1 + G,(s)G,(s) can take preassigned the origin inthe lefthalf plane,thenthe
M c ( ~ ) e - ’ taken response may be sluggish.
values in the left half plane, but poles of G , ( s ) - M G ( s ) e - s L cannot. Moreover, it should be noted that if the process is astatic with integral
Hence, it is necessary that poles of G , ( s ) - M G ( s ) e - ’ L are canceled with properties, then the Smith predictor control system has nonzero steady-
its zeros. Conversely, if poles of G,(s)- M G ( s ) e - E Lare canceled with its state error. For example, consider a process given by
1266 IEEE TRANSACTIONS ON AUTOMATIC COSTROL, VOL. AC-26.KO. 6. DECEhIBER 1981
G(s)e-IL=le-s,
S
(43)
Equation (38) is not satisfied. Hence, the Smith predictor control system
for astatic process with integral properties cannot yield zero steady-state
error to step disturbances even though the controller contains an integra-
tor. The system of Fig. 4, where M= l, is in the same condition.
The systems presented by Kleinman [6] and Marshall et al. [ 101 cannot
handle step disturbance because the step disturbances are not taken into
account in the problem formulation.
To understand better the differences of characteristics. we will present
numerical examples.
Example V.1 (Comparison of responses to a step disturbance):
Consider a process described by
1
-e-'.
s+0.2
s(s+ 1) (45) 0 20 $0 t
Fig. 10 Responses of (a) theproposed system. (b) theSmithpredictorcontrolsystem,
The response of the Smith predictor control system and the proposed one and (c) the approximate process-model control system for the astatic process with the
are calculated numerically as shown with full lines in Fig. 10, wshere zeros step disturbance.
of 1 +G,(s)G,(s) are -0.3. -0.3, -0.4, -0.4, and - 1.0. It is seen that
the proposed system can yield zero steady-state error whereas the Smith Proposition V.4: When the reference input and the dwurbance are zero
predictor control system cannot. in the process-model control system of Fig. 8. the response of the process
to the arbitraty initial condition of the system can be adjusted as desired
C. Response to .Vonzero Initial Condition
if and only if the state of the delay-free part of the model is unobservable
When the reference input and the disturbance are zero. the behatiors of in ?vL.
the process-model control systems with nonzero initial conditions are Remark V.6: The proposed system satisfies this condition as men-
shown as follows. tioned above and so the response of the process can be adjusted desirably.
First. we suppose that the initial condition of the model is zero. but that Remark V.7: In the Smith predictor control system. the transfer func-
of the process is not zero. In this case, the behavior of the system is ~ ~ ' ~of.
tion of the model is given by C ( ~ ~ - A ) - ' B - C ( S I - . ~ ) - ' BPoles
similar to the case where the process is subject to disturbances. Therefore, the model are not canceled with its zeros except for the process with n= 1
to yield desired transient response to nonzero initial condition of the and A =O. Hence. the Smith predictor control system cannot yield desired
process after L , it is necessary and sufficient that poles of G , ( s ) - response to a nonzero initial condition of the process even though the
~ b f G ( r ) e - are
' ~ canceled \pith its zeros. initial condition of the model is zero.
Second, we consider the case where the initial condition of the process Remark V.8: In the systems presented by Fuller [4] and Donoghue
and that of the model are not zero. In this case. the cancelation of poles of [ I I]. poles of the transfer function of the model are canceled with its
G l ( s ) - M C ( s ) e - " - with its zeros does not imply that the system has zeros. as seen in (40). but the state of the delay-free part of the model is
desired response. In the process-model control system of Fig.8, the not unobservable inyL. Tlus is shown as follows. In the system of Figs. 2
process and the model are connected in parallel to the control input u and and 3, the process model consists of hvo dynamical portions and so the
poles of the model are equivalent to those of the process. Hence, both response to the initial condition is not canceled with its delayed version.
response of the process and the model cannot be controlled simulta- This implies that the systems do not satisfy the unobservability of the
neously by the control input u. That is to say. there exist dynamical state of the delay-free part of the model. Therefore. if the initial condition
modes which are not controllable by u. The primary concern is the of the model is zero, then the control systems can yield desired responses
behavior of the process. If the state of the delay-free part of the model is to the initial condtion of the process, but if the initial condition of the
unobservable inyL, then contains only the state of the process. The state model is not zero, then the control system cannot have desired responses.
of the process is estimated from e and so the controller can yield the Remark V.9: In the systems proposed by Kleinman [6]. Marshall et al.
control signal u . at least, to adjust the response of the process desirably. [IO]. and Cook et 01. [ 121. the response of the model to its initial condition
However, if the state of the delay-free part of the model is not unobserva- is canceled with the delayed version, as seen in Fig. 4. where M=e-"L.
ble in .vL. then the response of the model to its nonzero initial condition This implies that the state of the delay-free part of the model is unob-
affects P. The state of the process cannot be estimated exactly from e since servable in y L after L. These systems satisfy the condition of the Proposi-
the process and the model are connected in parallel to 2 and poles of the tion V.4.
model are equivalent to those of the process. The controller cannot Remark V.10: If the integral term on the right of (7) is realized by a
generate the control signal u to adjust the response of the process as numerical computation, then poles of the closed loop system are only
desired. Consequently, the following proposition is obtained. zeros of det(sl-A-ePALBf) and the system can yield desired response
IEEE TRANSACTIONS ON AUTOMATIC COhmOL, VOL. AC-26, NO. 6, DECEhlBER 1981 1267
(47)
0 '. 5 10
'. __---
b.--- Since the impulse response maintains zero after L, we can consider a
'*op>;
0.0- t
stable system described by Cse-A~L(sI-A,)-'B,,the impulse response of
which is approximate to that of the unstable model in the interval (0, L ) .
If such a system is obtained, then the following approximate model is
s(t) l'o constructed:
-------------- _ _ _ _ _ _ _ _ _ _ _ _ _
0.0
t
1. O k
where the state of the delay-free part of the model is intended to be
unobservable in y L at r> L. Hence, the impulse response of the model is
eliminated after L and is approximate to that of (47). To yield zero
steady-state error zero, a term corresponding to the third one in (30),
which makes the state of the integratorunobservable inyL, must be added
to (48). The resulting approximate model is given by
Fig. 1 I. Response of (a) the proposed system, (b) the Smith predictor control system, (c)
Donoghue's system. (d) Cook's system where M=e-0.2 (the systems of Kleinman and Replacing (30) with (49) in the system of Fig. 7@), theapproximate
Marshall et d ) , and (e) process-model version of Manitius el nl. to nonzero initial
condition of the process. The full limes show the case where the initial condition of the process-model control system is obtained.
model is zero and the dotted lines show the m e uvhere the initial condition of the model The input-output transfer function of the approximate process-model
is not zero.
control system is given by
to arbitrary initial conditions [9]. On the other hand, if the integral term is y(s) - G,(s)G(s)e-SL
Cr(s)=--
realized by the process model, as shown in Fig. 5, then additional poles r(s) l+G,(s)(G,(s)+~(s)}
are introduced in the closed loop system as suggested in [9]. The state
corresponding to the additional poles, that is, the state of the model, is
observable in z as seen in Fig. 5. The condition of Proposition V.4 is not
satisfied. However, the transfer function of the model is given by ( s l -
A ) - ' e - A L B - ( s l - A ) - ' B e - S L from (9) and the impulse response of the
model is eliminated after L. This implies that poles of the model are
canceled with its zeros. Therefore, the behavior of the system shown in
Fig. 5 is similar to that of the system by Fuller and Donoghue.
To understand better the above differences, numerical examples are
presented.
Example V.3 (Comparison of responses to initial conditionsj:
Consider the process given by (44).When the initial condition of the
model is zero, the responses to the nonzero initial condition of the process
are calculated numerically, as shown with the full lines in Fig. 11. When
the initial condition of the model is not zero, the responses are shown with
the dotted lines in Fig. 11.
Since c(s) is the Laplace transformation of the difference in the impulse
VI. STABILIZATION
OF U N S T ~ LPROCESS
E responses of the models and zeros of 1 +G,(s)G,(s) that are in the left
half plane, the approximate process-model control system is stable if the
As mentioned above, the process-model control system has uncontrolla-
difference in the impulse responses is small enough to leave zeros of
ble modes, since the process and the model are connected in parallel to
1 +G,(s)(G,(s)+e(s)) in the left half plane. The similar argument on the
the control input u and poles of the model are equivalent to those of the
stability is seen in [9], [16]. and [17].
process. The uncontrollable modes are governed by the same poles as the
When the approximate process-model control system is stable, the
process. Therefore, if the process has poles in the right half plane. then the
steady-state and transient behaviors are as follows. Since G , ( S ) - G ( S ) ~ - ' ~
process-model control system is unstable. In this section, we ad1 consider
in (51) represents the transfer function of the proposed model and it hasa
a way to stabilize the process-model control system for unstable processes
zero at s=O, we have
by using a certain approximate model. Moreover, the results derived here
are also concerned with the discussion about the behavior of the system
lim { G , ( ~ ) - G ( s ) ~ - '=~O .} (56)
under the mismatch between the process and the model. s-0
In the proposed system, the transfer function of the model is described
by Moreover, {GSl(s)-G,(s)e-'L) is the transfer function of the approxi-
mate model which has the same structure as the proposed one. Hence,
Gl(s)-MG(s)e-SL=Ce-"L(sl-A)-'B-lLCe-ATBdT
0
lim { G , , ( s ) - G , ( s ) e - S L } = O . (57)
-C(sl-A)-'Be-'L. (46) s-0
1268 IEEE TRANSACTIONS ON AUTOMATIC COhTROL, VOL. AC-26, NO. 6, D E C E B E R 1981
Substituting (56) and (57) into (5 l), we have in the output of the model. If the process is asymptotically stable, then the
condition is also the sufficient one. The Smith predictor for astatic
lim€(S)=O. processes with integral properties and Donoghue’s system cannot satisfy
s-0
this condition. Moreover, the systems proposed by Kleinman and Marshall
Considering (58) and (33), the following equation holds: et al. cannot handle step disturbances.
3) The necessary and sufficient condition for yielding desired transient
1 limG,(s)=l
limsc,(s)-= responses to arbitrary initial conditions is that the state of the delay-free
(59)
s-0 s s-0 part of the model is unobservable in the output of the model. The systems
The output of the system can follow the step reference input aith zero proposed by Kleinman and Marshall er a/. can satisfy this condition, but
steady-state error. If the magnitude of E(S) is small, then the transient those of Fuller and Donoghue cannot.
response may be approximate to that of the system of Fig. 7(b), as seen by 4) The remarkable feature of the proposed system is that the states of
comparing (50) with (32). the delay-free part of the model and the integrator are unobservable in the
On the otherhand, the behavior of the approximate process-model output of the model at t S L . All conditions pointed out above can be
control system to disturbances is shown as follows. The disturbance satisfied. Consequently, the proposed system can yield zero steady-state
output transfer function of the approximate system is given by error and desired transient responses to step disturbances and arbitrary
initial conditions.
5 ) If the process has poles in the right half plane, then the process-model
control system is unstable. If the impulse response of the model is
eliminated after L , like the proposed one. then the system can be
stabilized by using a certain stable model. the impulse response of which
is approximate to that of the exact model.
6) In practical applications, there exists inevitable mismatch between
the process and the model. In spite of this, the unobservability of the
Substituting (57).(58). and (59) into (60). and considering that G X s ) states of the delay-free part of the model and the integrator in the output
contains an integrator, we have of the model can be satisfied because the observability of these states
depends only on the structure of the model. Therefore, the unobserva-
bhty can be satisfied without the exact knowledge of the process. If the
mismatch is small enough to maintain the stability of the closed loop
The steady-state error to step disturbances is also zero. If the magnitude system. then the system has zero steady-state error and can yield ap-
of E ( S ) is small. then the transient response to the disturbance is also proximately the desired transient responses.
approximate to that of the system of Fig. 7(b), since the impulse response This paper restricts attention to single input-single output systems for
of G S l ( s ) - G , ( s ) e - ” Lis eliminated after L. the sake of clarity of exposition. The results obtained here should be
Example VI.l (Response of the approximate process - model control sys- extended to process-model control for multivariable processes with multi-
tem): ple delays. A process-model control for multivariable discrete system is
Consider the process given by (45), which is not asymptotically stable. shown in [22].
Let the exact model be replaced with the approximate one of the form In this paper, the process-model control is developed within the frame-
work of feedback. The improvement of dlsturbance attenuation using
0.48-6.17s 0.9 e-2s feedforward is referred to in [20].
+4.18-
(S+O.l)(S+l) (s+O.l)(s+l) (62)
in the proposed system, as shown in Fig. 7@). The response of the REFERENCES
resulting approximate process-model control system is simulated digitally,
0. J. M. Smith. “A controller to overwme dead time.” IS4 J . . vol. 6, no. 2. pp.
as shown with the dotted line in Fig. IO. 28-33.1959.
Remark VI.1: An important result which emerges is that the unob- P. S. Buckley. “Automatic control of processes u.ith dead time.” Proc Frrsr Int.
Congress IFAC. 1960. pp. 33-40.
sewability of thestates of the delay-free part of the model and the S. Saa-ano, “Analog study of process-model control systems.” J . Sor. Insrrm. Conrr.
integrator in y L also plays a significant role in order to yield zero Eng.. Japun. vol. I . no. 3. pp. 198-203. 1962.
steady-state error and desired transient responses even though mismatch A T Fuller. “Optimal nonlinear control of systems with pure delay.” Inr. J . Conrr.,
vol. 8. no. 2. pp 145-168. 1968
exists between the process and the model. Moreover. it should be em- D. H. Mee. “An extension ofpredictor control for systems uith control tlme-delays.”
phasized that the states of the delay-free part of the model andthe Inr. J . Conrr.. vol. 18.no. 6.pp. 1151-116R.1973
D. L Kleinman. “Optimal control of linear systems w t h time-delay and observation
integrator are made unobservable in y L in presence of mismatch. This is nolse.” I E E E Trum Auronmr. Comr. vol AC-14. pp. 524-527. Oct. 1969.
obvious from the following fact. The observability of these states in y L H. h’. Kolvo and E. B. Lee. “Controller synthesis for linear systems uith retarded
state and control variables andquadratic cost.” Auromortcu. vol. 8. no. 2. pp.
depends only on the structure of the model. The structure of the ap- 203-208.1972.
proximate model is intended to be identical to that of the exact one. R. M. Leuis. “Control-delayed system properties via anordinary model.” Inr. J.
Hence. the unobsewability is satisfied. This implies that the unobserv- Conrr.. vol. 30. no 3. pp. 477-490. 1979.
A. Maninus and A W. Olbrot. “Finite spectrum assignment problem for systems
ability can be satisfied without the exact knowledge of the process. u,ith delays.” IEEE Tram Auronlur. Conrr.. rol. AC-24. pp 541-553, Aug. 1979
Therefore, in practical applications with inevitable mismatch between the J. E Marshall. B. Ireland. and B. Garland. “Comments on ‘an extension of predictor
control for systems with control time-delays.’“ Inr J . Conrr.. rol. 26, no. 6. pp.
process and the model, one can make the states of the delay-free part of 981-982.1977.
the model and the integrator unobservable in the output of the model. If J. F. Donoghue. “A comparison of the Smth predictor and optlmal design ap-
proaches for systems uirh delay inthe control.” IEEE Tram Ind Elerrron. Conrr.
the mismatch is small. then the process-model control system has zero lmrrurn , bel. IECI-24. pp. 109-117. Feb 1977.
steady-state error andcan yield approximately desired transient re- G Cook and X.Price. “Comments on ‘A comparison of the Smith predictor and
sponses. optimal deslgn approaches for systems with delay in the control.’“ I E E E Trans. Ind.
Elecrron Conrr. Imrun.. vol. IECI-25, pp 180-181. May 1978
J . F. Doooghue. “Further mmments on ‘A comparison of the Smith predictor and
optimal design approaches for system uith delay in the control.”’ I E E E Trans. Ind.
121. CONCLUSIONS Elecrron. Conrr. I m r m . . vol. IECI-25, pp. 379-380. Nov. 1978
S. Bhattachanya and J. B. Pearson. “On the h e a r servomechanism problem,” Inr. J .
1) The necessary and sufficient condition for yielding desired transient Conrr.. vol. 12. no. 5. pp. 795-806. 1970.
response to disturbances at r > L in the process-model control system is S. Hosoe and M.Ito. “On the error systems in linear sen0 problems.” Tram. I E E
Jupan. %,ol 92s. no. IO. pp. 369-375. 1972
that poles of the model are canceled with its zeros. The Smith predictor A. C Ioannides. G. I. Rogers. and V. Latham. “Stabilln, limits of a Smith controller
control cannot satisfy this condition. In simple systems containing a time delay.” Inr. J . Conrr.. vol. 29. no. 4. pp. 557-563,
1979.
2) To make the steady-stateerror to stepdisturbances zero, it is B. Garland and J E. Marshall. “Sensitlrit) considerations of Smith‘s method for
necessary that the state of the integrator in the controller is unobservable time-delay s>-stem:’ Elerrron Ixrr.. vol. IO. no. 15. pp 308-309.1974.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. AC-26,NO. 6, DECEMBER 1981 1269
[IS] B. Garland and J. E. Marshall, "A short bibliography on Smith's method," presented
at I E E Colloquium Time Delay Syst. Contr.-The Status of Smith's Method, London, bo[(l-A(q-1))y(t)+B(q-')~*(~-l)-~*(t)]+hu*(t-1)=0.
1979.
[I91 J. E Marshall. "Explorations of Smith's principle," IEE Colloquium TimeDelay
Syst. Contr.-The Status of Smith's Method, London, 1979. (14
[20] C. C. Prasad and P. R. Krishnaswamy. "Control of pure time delay processes,"
Chem. Eng. Sa..vol. 30, no. 2, pp. 207-215, 1975. This leads to a closed-loop system described by
[21] C. B. G . Meyer, R. K. Wood, and D. E Seborg, "Experimental exvaluation of
analytical and Smith predictors for distillation column control." Amer. Insf. Chem.
J., voL25, no. I. pp. 24-32. 1979. (boB(4-')+xA(4-'))rct)=b,B(q-')y*(t) (1.3)
[22] K Watanabe and M. Ito, "Process-model control for multivariable systems with time
delays in inputs and outputs." Trans. I E E Japan, vol. 99-c, no. 7, pp. 147-154, 1979. and
(b,B(q-')+hA(q-l))u(t)=boA(q-l)y*(t+l). (1.4)