Journal of Process Control 12 (2002) 887–895

www.elsevier.com/locate/jprocont

Performance comparison between PID and

dead-time compensating controllers
Ari Ingimundarson*, Tore Hägglund
Department of Automatic Control, Lund Institute of Technology, Box 118, S-221 00 Lund, Sweden

Received 9 May 2001; received in revised form 4 February 2002; accepted 5 April 2002

Abstract
This paper is intended to answer the question: ‘‘When can a simple dead-time compensator be expected to perform better than a
PID?’’. The performance criterion used is the integrated absolute error (IAE). It is compared for PI and PID controllers and a
simple dead-time compensator (DTC) when a step load disturbance is applied at the plant input. Both stable and integrating pro-
cesses are considered. For a fair comparison the controllers should provide equal robustness in some sense. Here, as a measure of
robustness, the H1 norm of the sum of the absolute values of the sensitivity function and the complementary sensitivity function is
used. Performance of the DTC’s is given also as a function of dead-time margin (DM). # 2002 Published by Elsevier Science Ltd.
Keywords: Performance comparison; PID control; Dead-time compensators

1. Introduction transfer function and compared to best achievable

performance of a DTC, that is when it has infinite gain.
In [16] a control structure was presented which The performance criterion used was the integrated
became one of the main solutions to deal with processes squared error (ISE). DTC’s with very high gains can have
with long dead time. Recently a renewed interest in arbitrary small robustness toward dead-time errors, see
dead-time compensators has been noted in the control [12], even though their robustness measured with tradi-
literature. Extensions to integrating and unstable pro- tional measures like amplitude margin, phase margin or
cesses have been presented (see for example [7] and [5]). maximum sensitivity, can be good. Therefore, in the cur-
Despite this, little has been written about when DTC’s rent paper, performance of DTC’s is given as a function of
should be used. Commonly in textbooks on process dead-time error sensitivity, measured with the dead-time
control a few pages are devoted to explain the control margin (DM) or the smallest error in the dead time
structures of common DTC’s but recommendations and which causes instability. Another reference where the
guidelines about when to use DTC’s are very rare. subject is treated is [9]. There the robustness of the two
The most common control structure in the process structures is not treated specially and the range of process
industry is the PI. The derivative gain is often turned dynamics for which the comparison is made is small.
off. This applies specially when long dead times are The layout of this paper is the following. The com-
present. Therefore, given that the current control struc- parison criteria is treated in Section 2. The DTC’s and
ture is a PI two options should be compared. One is to their tuning is presented in Section 3. In Section 4 the
add the derivative part to the PI making it a PID. The results are presented. This is followed by a discussion in
other is to implement a DTC. Section 5. Finally conclusions are drawn in Section 6.
One reference where a similar comparison was done is
[15]. There, within the IMC framework, a PID controller
was designed for a first order plus dead time (FOPDT) 2. Comparison criteria

The purpose of this paper is to give insight into the

* Corresponding author. Fax:+46-46-138118. problem of choosing between a DTC and a PI(D) within a
E-mail address: ari@control.lth.se (A. Ingimundarson). typical process control environment. While providing
0959-1524/02/$ - see front matter # 2002 Published by Elsevier Science Ltd.
PII: S0959-1524(02)00017-3
888 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895

results general enough to be useful in that environment, a step load disturbance is applied at the plant input.
certain assumptions are necessary to limit the size of the This comparison was performed for the FOPDT model
paper. for a griding of the interval
One question that arises is what information should be
assumed to be available to select between control struc- T 2 ½0:01; 10
ð4Þ
tures. Detailed knowledge can not be assumed to be
available since this is usually not the case within the pro- while L and Kp were kept equal to 1. This range includes
cess industry. In connection with the usage of DTC’s, ‘‘almost’’ pure dead-time processes to ‘‘almost’’ dead-time
plants are often referred to as being dead-time dominated. free processes.
What is meant by this is that the ratio between dead-time It should be pointed out that while the comparison
and other dynamics is large and accordingly a decision to made in this paper is for load disturbances only, the
implement a DTC is taken. In this article it is assumed main strength of DTC’s has on the other hand been its
that process information is available in terms of a first- set point response. Furthermore it should be pointed
order plus dead-time (FOPDT) model, denoted with out that it has been a subject of many papers to improve
the load disturbance response of DTC’s. As was said
Kp Ls before, the DTC’s presented here are simple and there is
PðsÞ ¼ e ð1Þ
Ts þ 1 much room for improvement.
The block diagram of the loop is shown in Fig. 1. C(s)
This is the simplest way to represent the division of is the controller while P(s) is the process to be con-
process dynamics into pure dead-time and other trolled. l is the load disturbance affecting the system
dynamics. In the case of integrating processes it is while n is the measurement noise. For a FOPDT model
assumed that the FOPDT is in serial with an integrator on interval (4), the PI(D) with the lowest IAE and with
equal or better robustness was compared to the IAE of
Kp the DTC with the tuning obtained from the FOPDT
PðsÞ ¼ eLs ð2Þ
sðTs þ 1Þ model. The robustness condition will be introduced in
the next section.
In the process industry, processes are commonly mod- Since the PI(D) controller parameters are only subject
eled with these transfer functions so most control engi- to a robustness constraint, the comparison presented is
neers are familiar with their parameters. The comparison not dependent of a specific tuning rule of the PI(D).
is made for these processes only. By making sure con-
trollers fulfill a robustness constraint, the results should be 2.1. Robustness constraints
valid for plants sufficiently close to these models.
Another question that arises is the complexity of the Caution has to be shown when comparing perfor-
controller structures. The most commonly used struc- mance of control structures because of the ever present
ture in the process industry is the PI which has two trade off between robustness and performance. The
parameters. A more complex structure with more para- comparison should be made under the assumption that
meters might show better performance but still it might robustness of the control structures is similar. A small
never be implemented since this requires more expertize deviation in the process should not result in a great
and advanced maintenance than the PI. In this paper, difference in performance between the structures.
an effort has been made to keep the DTC’s as simple as The robustness measure used here is the H1 norm of
possible. The DTC structures considered all contain a the sum of the absolute values of the sensitivity function
model of the process. Most parameters are related to this and the complementary sensitivity function
model. When the plant is equal to the model, the set-





point response is given by a FOPDT transfer function  ¼ sup
Sði!Þ
þ
Tði!Þ
ð5Þ
with unit gain. !

1
eLs ð3Þ
Tr s þ 1

Tr is selected with dead-time sensitivity in mind. This

applies to the stable and integrating plant case. The
models and the tuning of the DTC’s will be introduced
in Section 3.
Within the process industry regulatory performance is
usually more important that set point response. There-
fore, the performance criterion used was the IAE when Fig. 1. Block diagram of disturbance signals.
 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895 889

where capture dead-time sensitivity of DTC’s. The reason is

that it is caused by large loop gain in the right-half
1 CP plane of the Nyquist diagram, see [1,13], while the
S¼ T¼ ð6Þ
1 þ CP 1 þ CP robustness measure is related to regions on the left side
of the Nyquist diagram. Dead-time sensitivity can be
This robustness measure or similar measures have taken into account by proper tuning of the DTC’s. The
appeared in various places in the control literature. In general principle in this article was to select a tuning so
[14] it was shown how the Nyquist curve of a closed- that the amplitude of the loop gain was strictly smaller
loop system with a given  is guaranteed to stay out of than 1 for frequencies larger than
the crossover
fre-
the region given by the contours of quency, !c (smallest frequency where
Cði!ÞPði!Þ
¼ 1).

1 þ jL j
¼ ð7Þ
j1 þ L j 3. Controllers

Furthermore it was shown how this parameter, with a 3.1. Stable case
suitable selection of weights for l and n is equivalent to
the generalized robustness margin, see [17]. The block diagram for the DTC can be seen in Fig. 2.
In [10] and [2] the following condition for robust per- P(s) is the real process to be controlled. P0(s) is the model
formance appears: of the process with dead time, G0(s) is the model without.




This way the controller for the DTC as indicated by
W1
Sði!Þ
þ W2
Tði!Þ
<18! ð8Þ Fig. 1 can be written as

For a typical process control problem the weights W1 Cc ðsÞ

and W2 should have a special shape. Weight W1 should be CðsÞ ¼ ð9Þ
1 þ Cc ðsÞG0 ðsÞð1  eLs Þ
large at low frequencies to assure good load–disturbance
rejection. Weight W2 usually increases with frequency to
guarantee robustness toward model perturbations at high The model P0(s) was the FOPDT model given by Eq. (1)
frequencies. The larger W1 can be the better disturbance and Cc(s) was a PI controller set to
rejection. Larger W2 means better robustness toward
multiplicative uncertainties is assured. 1 1 Ts þ 1
The control structures presented in this article have C c ð sÞ ¼ G ð sÞ ¼ ð10Þ
Tr s 0 Tr Kp s
certain inherited qualities. All of them have infinite gain
at low frequencies resulting in asymptotic rejection of a
step load disturbance. In the nominal
case

it can be where Tr is a design parameter. This parametrization is
shown with simple analysis that
Sði!Þ
! 0 as ! related to H2 optimal control. The actual criteria it mini-
when ! ! 0. At high frequencies


the controllers have mizes is the integrated squared error (ISE) but because of
constant gain which means
Tði!Þ
! 0 as 1=! when its simplicity it is used even here. For references about this
! ! 1. Given two controllers with these qualities and parametrization, see [4].
equal  there is a pair of weights, W1 and W2, such that In the nominal case, when there is no model error,
condition (8) is fulfilled for both controllers. These P0(s)=P(s), then Tr is the time constant from set point
weights can be constructed by choosing them to be 1= ysp to output y.
for the frequency for which the supremum is achieved in
Eq. (5). For other frequencies


1 would be put equal to
W YðsÞ 1
the smaller value of
1=Sði!Þ
for the two controllers. ¼ eLs ð11Þ
Ysp ðsÞ Tr s þ 1

W2 would
be chosen similarly to be the smaller value of

1=Tði!Þ
. The implication of this is that there is a set of
plants, around the nominal one, for which both con-
trollers satisfy the robust performance condition given
by Eq. (8).

2.2. Dead-time sensitivity

A useful concept when dealing with DTC’s is the

dead-time margin (DM) or the smallest change in the
dead-time which will cause instability. The robustness
measure presented in the previous section does not Fig. 2. Block diagram of the simple DTC.
890 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895

Furthermore, the loop gain L(s)=C(s)P(s) is then given by

eLs
LðsÞ ¼ ð12Þ
Tr s þ 1  eLs

The robustness properties of this structure were ana-

lyzed in [13]. There it was shown that L(i!) never has a
real part smaller than -1/2. This means the phase margin
and amplitude margin are minimum 60 and 2, respec-
tively. It was also shown how the dead-time margin Fig. 3. The improved dead-time compensating controller for integrat-
depends on Tr/L. Some values are given in Table 1. For ing processes.
Tr =L 5 0.34 the dead-time margin is larger than 100%.
This was the value used for the comparison. Again, for
this value the amplitude of the loop gain is always less Tuning of this structure has been presented in [3]. The
than 1 for frequencies larger than the crossover frequency. controller in [7] will be referred to as the MSP (modified
In the nominal case the transfer function from l to y is Smith predictor) and the one in [8] as the IMSP
given by (improved MSP).
  As with most DTC’s, these controllers contain a
eLs model of the plant to be controlled. This is the two-
yðsÞ ¼ P0 ðsÞ 1  l ð sÞ ð13Þ
Tr s þ 1 parameter model

This is easily seen to have an over damped response if l Kp s

P0 ðsÞ ¼ G0 ðsÞes ¼ e ð15Þ
is a step. This means that the IAE is equal to the inte- s
grated error (IE) and an analytical expression can be
obtained by a Taylor series expansion of the exponent As was explained before, the comparison is made for
in Eq. (13) and applying the final-value theorem. This the collection of FOPDT models in serial with an inte-
gives grator as given by Eqs. (2) and (4). A two-parameter
ð1 model approximation of the model given by Eq. (2) was
SðsÞ obtained by putting =T+L.
IAE ¼ lim eðtÞdt ¼ lim s 2 ¼ Kp ðTr þ LÞ ð14Þ
t !1 0 s !0 s The equation for the controller according to Fig. 1 is

For example, a DTC with DM equal to 35% will have Cc þ F þ FCc G0

IAE from a step load disturbance equal to 1.24KpL. C¼ ð16Þ
1 þ Cc ðG0  P0 Þ
With a very high gain DTC, obtained by Tr 0, one
could get close to IAE=KpL. That is the lower limit
with the presented tuning. The transfer function F(s) is given by
Notice also that even if the IAE is independent of T, 8
the response is not. With a large T settling time will also > 1
>
< : MSP
be very large. 2Kp
FðsÞ ¼ 0:72 0:4s þ 1 ð17Þ
>
>
: : IMSP
3.2. The modified Smith predictor, integrating case Kp 0:04s þ 1

DTC’s for integrating plants have been presented in Finally Cc is a constant given by
among others [8,11]. The one chosen for the comparison
in this article is the one presented in [8], see Fig. 3. The 1
Kr ¼ ð18Þ
reason for this is that this structure is very simple and Kn Tr
replacement from a PI should be straight forward. This
structure is an improvement of a structure presented in where Tr is the time constant of the transfer function
[7] which will be considered for comparison as well. from set point in the nominal case when P=P0,

YðsÞ 1
¼ es ð19Þ
Ysp ðsÞ Tr s þ 1
Table 1
Dead time margin in percentage with the corresponding minimum Tr/L
Then the loop gain can be expressed as a function of
DM (%) 5 15 25 35 >100
Tr/L 0.03 0.10 0.16 0.24 0.34
only dead time  and the time constant from reference
value to the output, Tr
 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895 891
 
es s þ FðsÞTr s þ FðsÞKp To simplify discussions the range of T/L is split into
LðsÞ ¼ ð20Þ three regions.
sðTr s þ 1  es Þ

It can be seen that the loop gain depends only on the 4.1.1. Top region. T/L 2[5,10].
ratio Tr = and therefore the DM as well. In accordance Here the performance of the PI is better than the
with the tuning of the DTC for stable processes, this DTC. This is not unexpected since when L=0 the poles
ratio was chosen so that the amplitude of L(i!) for fre- of the closed loop system can be placed arbitrarily with
quencies larger than the cross-over frequency would be a PI controller. The PID performs best of the controller
strictly smaller than 1. This gave Tr/=0.534. structures.
The tuning recommended in [7] was to set Tr equal to
T in the model (2). This parameter setting is included in 4.1.2. Middle region. T/L 2[0.1,5].
the simulation study that follows. Notice though that In this region the performance of the DTC is better
when T is small this tuning can give arbitrary small DM. than the PI. The largest difference is when T/L 1.
There the reduction in IAE following a switch to a DTC
3.3. The PID parametrization is at least 25%.
For T/L < 1 the PID performance is similar to the
The PID was parametrized on parallel form DTC. To obtain a DTC with better performance than a
  PID, it would have to be tuned more aggressively with
1 Td s lower DM.
CðsÞ ¼ K 1 þ þ ð21Þ
Ti s Td =Ns þ 1
4.1.3. Bottom region. T/L=[0.01,0.1].
with N=10. N is sometimes considered a tuning para- The PI has higher IAE over the region but the per-
meter in the study of PID controllers. To limit the formance is quite similar. Notice that the performance
complexity of the PID controller it was fixed to a typical curve for the PI is leveling out and approaching a con-
value as recommended in textbooks on process control. stant value as the FOPDT process approaches a pure
For PI, Td was set to 0. dead-time process. Only a DTC with smaller DM would
have superior performance in this interval.
The performance of the PID approaches the perfor-
4. Results mance of the PI when the ratio is decreasing. Investiga-
tion of the derivative gain showed it approaches zero for
4.1. Stable case smaller T/L.

The main result of the paper for the stable case is 4.2. Integrating case
shown in Fig. 4. IAE is displayed for PI, PID and the
DTC for different values of T/L. The performance of The results in the integrating case are shown in Fig. 5.
the DTC is only dependent on the Tr/L ratio determined As the IAE varies considerably over the T/L range the
by the dead-time margin and it is therefore constant
over the entire range of the T/L ratio.

Fig. 5. Results for the integrating case. What is shown is the IAE nor-
Fig. 4. IAE results for the different control structures in the stable malized with the IAE of IMSP with Tr/L=0.534 (—), PI (- - -), PID (. - .),
case. DTC(—),PI (- - -), PID (. - .) . IMSP with Tr=T (    upper), MSP with Tr=T (    lower).
892 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895

IAE of all control structures was normalized with the seems to touch the contour in more than one place. In
IAE of the IMSP with Tr/L=0.534. Figs. 7–9 the sensitivity stars for the different para-
The performance of the DTC’s is better than the PI meters are shown. Notice that the scale for IAE change
for all values of T/L. Largest difference is for T/L 3. is different on each figure. Generally speaking one can
There the IAE of the PI is 3.3 times larger than the IAE say that none of the structures seemed significantly
for the DTC. For small T, when the process approaches more sensitive to parametric changes than the others.
a pure dead-time process with an integrator, the ratio of For varying Kp the surprising thing is the increase in
the IAE between PI and DTC approaches a constant IAE for the PI on the right side. Comparing the load
value. The PI has around 2.5 times higher IAE than the disturbance responses, one can see that the damping in
DTC in this case. the PI response decreases more as a function of DKp ,
If the IMSP and MSP are tuned as recommended than for the PID and DTC. For varying T the behavior
with Tr=T, the performance depends strongly on T/L is rather good for all structures when the change is lar-
but is always superior to the PI. ger than -0.5. For smaller values, the performance of the
The PID performs much better than the other struc- DTC and the PID increases very quickly before the loop
tures for large T/L. This it not surprising since if L=0 becomes unstable (DT < -0.65). The PI on the other
in Eq. (2) the poles of the closed-loop system can be hand remains stable with a small increase in IAE on
placed arbitrarily with a PID. For small T/L the PID that portion of the interval. The PID is most sensitive
has a performance very similar to the IMSP.

4.3. Sensitivity analysis

The PI(D) controllers were compared to the DTC’s

under the assumption that they fulfilled the robustness
constraint in Eq. (5) for equal or less . Yet it is inter-
esting to see how IAE changes when the true process is
not the same as the nominal model. As there are only
three parameters, a sensitivity star was determined for
each parameter, for the stable and integrating case. This
was done for T/L ratio equal to 1. The sensitivity stars
show how IAE changes, relative to the IAE for the cor-
rect model, as a function of parameter variation.

4.3.1. Stable case

In Fig. 6 the Nyquist diagram is shown. Also shown,
on the left half plane is the  contour (encircling -1). It
can be seen that all controllers touch the contour
meaning they all have equal . Notice that the PID Fig. 7. Sensitivity star for K, stable case. DTC (—), PI (- - -), PID (. - .).

Fig. 6. Nyquist diagram for the stable case, T/L=1. DTC (—), PI (- -
-), PID (. - .). Fig. 8. Sensitivity star for T, stable case. DTC (—), PI (- - -), PID (. - .).
 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895 893

Fig. 9. Sensitivity star for L, stable case. DTC (—), PI (- - -), PID (. - .). Fig. 11. Sensitivity star for K, integrating case. IMSP (—), PI (- - -),
PID (. - .).

toward change in dead-time, L. The IAE increases for

all structures for positive change in dead-time.

4.3.2. Integrating case

The dead time compensator for the sensitivity study
was the IMSP with Tr/L=0.543. The Nyquist diagram
for the integrating case is shown in Fig. 10. Again all
controllers lie tight up to the robustness constraint. The
sensitivity stars are shown in Figs. 11–13. Again one can
say that no structure is significantly more sensitive than
the others. For a change in Kp it is obvious that the
change in IAE for the DTC has a dependence that is very
close to linear in DKp and the dependence is opposite to
the other structures.

5. Discussion
Fig. 12. Sensitivity star for T, integrating case. DTC (—), PI (- - -),
There are many aspects that have not been considered PID (. - .).
in this comparison. The PID and DTC control structures

Fig. 10. Nyquist diagram for the integrating case, T=1. IMSP (—), Fig. 13. Sensitivity star for L, integrating case. IMSP (—), PI (- - -),
PI (- - -), PID (. - .). PID (. - .).
894 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895

see [1]. To reach this optimal point by manual tuning

might be difficult

6. Conclusions

The paper has focused on comparing the performance

of the PID control structure with simple DTC’s. The
purpose has been to gain insight about when each con-
trol structure should be used. IAE performance was
determined as a function of the ratio between time con-
stant and dead time, T/L, for a collection of FOPDT
models.
In the stable case, the main result is that for T/L in
Fig. 14. Contour lines for fixed . Optimal IAE solution is marked the interval [0.1,5] a substantial decrease in IAE can be
with  . obtained by switching from a PI to a DTC. The perfor-
mance of a PID is better than a DTC for T/L on [1,10],
each have advantages that should be taken into otherwise it is similar to the DTC. A superior perfor-
account. The PID is a well known structure frequently mance of the DTC on the interval [0.1,1] could be
found in distributed control systems as standard mod- obtained by accepting a smaller DM.
ules. One advantage of the DTC is that the dead time In the integrating case, the performance difference is
enters the control law directly, opening the possibility for much greater between PI and the DTC. A switch to a
gain scheduling if the dead time is measurable. Further- DTC might reduce IAE to about one third of the IAE
more, DTC’s can be designed with certain dead-time for PI. The PID performs best over a large portion of
margin if variations in dead-time are known, see [3]. the area but it has been shown it might be difficult to
Another aspect ignored here is the control signal obtain this optimal performance with manual tuning.
amplitude or energy. This could be checked by looking
at the transfer function from measurement noise, n, to
the control signal, C(s)S(s). The performance of the PID References
for large T/L was often followed by large values of this
transfer function. [1] K.J. Astrom, T.H. Hägglund, The future of PID control, Control
It was said in the introduction that recommendations Engineering Practice 9 (2001) 1163–1175.
for when to use DTC’s are scares in the literature. An [2] J. Doyle, B. Francis, A. Tannenbaum, Feedback Control Theory,
Macmillan, New York, 1992.
exception is [6]. There, a DTC is recommended for [3] A. Ingimundarson, T. Hagglund, Robust tuning procedures for
stable processes if the feedback fraction dead time, L/ dead-time compensating controllers, Control Engineering Prac-
(T+L) is larger than 0.7. This translates to a T/L ratio tice 9 (2001) 1195–1208.
smaller than 0.4. As indicated in Fig. 4 the benefits of [4] D. Laughlin, D. Rivera, M. Morari, Smith predictor design for
replacing a PI start for larger values of T/L. robustperformance. Int. J. Control 46 (1987) 477–504.
[5] S. Majhi, D. Atherton, Obtaining controller parameters for a new
Smith predictor using autotuning, Automatica 36 (2000) 1651–
5.1. PID considerations 1658.
[6] T. Marlin, Designing Processes and Control Systems for
The DTC is a model based control strategy. Imple- Dynamic Performance, McGraw-Hill International Editions,
1995.
menting it necessitates an identification of a process
[7] M. Matausek, A. Micic, A modified smith predictor for control-
model. Having said that a PID performs similar to a ling a process with an integrator and long dead-time, IEEE
DTC it should be noted that to obtain this performance Transaction on Automatic Control 41 (1996) 1199–1203.
with a PID some kind of identification of the process [8] M. Matausek, A. Micic, On the modified smith predictor for
would be necessary. Achieving similar performance with controlling a process with an integrator and long dead-time,
a manual tuning would be difficult. And this is not only IEEE Transaction on Automatic Control 44 (1999) 1603–1606.
[9] C. Meyer, D.E. Seborg, R.K. Wood, A comparison of the Smith
due to the increase from two to three parameters. To predictor and conventional feedback control, Chemical Engi-
support this claim, some contour lines in the PID para- neering Science (1976).
metric space is shown in Fig. 14 for the integrating case [10] M. Morari, E. Zafiriou, Robust Process Control, Prentice Hall,
when T/L=1. For several values of Td the contour lines Englewood Cliffs, NJ, 1989.
in K and Ti are shown for the same  as the IMSP has [11] J. Normey-Rico, E. Camacho, Robust tuning of dead-time com-
pensators for processes with an integrator and long dead-time,
when tuned with fixed Tr/ ratio. It can be seen that the IEEE Transaction on Automatic Control 44 (1999) 1597–1603.
surface has sharp edges on it. Sharp edges of the para- [12] Z. Palmor, Stability properties of Smith dead-time compensa-
metric surfaces of PID’s has been reported elsewhere, torcontrollers. Int. J. Control 32 (1980) 937–949.
 A. Ingimundarson, T. Hägglund / Journal of Process Control 12 (2002) 887–895 895

[13] Z. Palmor, M. Blau, An auto-tuner for Smith dead time- 4, PID controller design. Ind. Eng. Chem. Proc. Des. Dev 25
compensator. Int. J. Control 60 (1994) 117–135. (1986) 252–265.
[14] Panagopoulos, H., Astrom, K.J., 2000. PID control design and [16] O.J.M. Smith, Closed control of loops with dead time, Chemical
H1 loop shaping. Int. J. Robust Nonlinear Control, 10, 1249– Engineering Progress 53 (1957) 217–219.
1261. [17] Zhou, K., 1998. Essentials of Robust Control. Prentice Hall:
[15] D.E. Rivera, M. Morari, S. Skogestad, Internal model control— Englewood Cliffs, NJ, 1998.