See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222297633

Simple Analytic Rules for Model Reduction and PID Controller Tuning

Article in Journal of Process Control · April 2004

DOI: 10.1016/S0959-1524(02)00062-8 · Source: DOAJ

CITATIONS READS

2,017 6,708

1 author:

Sigurd Skogestad
Norwegian University of Science and Technology
628 PUBLICATIONS 28,404 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sigurd Skogestad on 14 May 2018.

The user has requested enhancement of the downloaded file.

 Journal of Process Control 13 (2003) 291–309
www.elsevier.com/locate/jprocont

Simple analytic rules for model reduction and PID

controller tuning§
Sigurd Skogestad*
Department of Chemical Engineering, Norwegian University of Science and Technology, N-7491 Trondheim, Norway

Received 18 December 2001; received in revised form 25 June 2002; accepted 11 July 2002

Abstract
The aim of this paper is to present analytic rules for PID controller tuning that are simple and still result in good closed-loop behavior.
The starting point has been the IMC-PID tuning rules that have achieved widespread industrial acceptance. The rule for the integral
term has been modified to improve disturbance rejection for integrating processes. Furthermore, rather than deriving separate rules for
each transfer function model, there is a just a single tuning rule for a first-order or second-order time delay model. Simple analytic rules
for model reduction are presented to obtain a model in this form, including the ‘‘half rule’’ for obtaining the effective time delay.
# 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Process control; Feedback control; IMC; PI-control; Integrating process; Time delay

1. Introduction Step 2. Derive model-based controller settings. PI-set-

tings result if we start from a first-order model, whereas
Although the proportional-integral-derivative (PID) PID-settings result from a second-order model.
controller has only three parameters, it is not easy,
without a systematic procedure, to find good values There has been previous work along these lines,
(settings) for them. In fact, a visit to a process plant will including the classical paper by Ziegler amd Nichols [1],
usually show that a large number of the PID controllers the IMC PID-tuning paper by Rivera et al. [2], and the
are poorly tuned. The tuning rules presented in this closely related direct synthesis tuning rules in the book
paper have developed mainly as a result of teaching this by Smith and Corripio [3]. The Ziegler–Nichols settings
material, where there are several objectives: result in a very good disturbance response for integrat-
ing processes, but are otherwise known to result in
1. The tuning rules should be well motivated, and rather aggressive settings [4,5], and also give poor per-
preferably model-based and analytically derived. formance for processes with a dominant delay. On the
2. They should be simple and easy to memorize. other hand, the analytically derived IMC-settings in [2]
3. They should work well on a wide range of are known to result in a poor disturbance response for
processes. integrating processes (e.g., [6,7]), but are robust and
generally give very good responses for setpoint changes.
In this paper a simple two-step procedure that satisfies The single tuning rule presented in this paper works well
these objectives is presented: for both integrating and pure time delay processes, and
for both setpoints and load disturbances.
Step 1. Obtain a first- or second-order plus delay
model. The effective delay in this model may be 1.1. Notation
obtained using the proposed half-rule.
The notation is summarized in Fig. 1. where u is the
manipulated input (controller output), d the dis-
§
turbance, y the controlled output, and ys the setpoint
Originally presented at the AIChE Annual meeting, Reno, NV,
USA, Nov. 2001. (reference) for the controlled output. gðsÞ ¼
y

u denotes
* Tel.: +47-7359-4154; fax: +47-7359-4080. the process transfer function and c(s) is the feedback
E-mail address: [email protected] part of the controller. The
used to indicate deviation
0959-1524/03/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved.
PII: S0959-1524(02)00062-8
292 S. Skogestad / Journal of Process Control 13 (2003) 291–309

where we have used b=1 for the proportional setpoint

weight.

2. Model approximation (Step 1)

The first step in the proposed design procedure is to

obtain from the original model go(s) an approximate
first- or second-order time delay model g(s) in the form

k
Fig. 1. Block diagram of feedback control system. In this paper we gð s Þ ¼ es
consider an input (‘‘load’’) disturbance (gd=g). ð
1 s þ 1Þð
2 s þ 1Þ
k0
¼ es ð4Þ
ðs þ 1=
1 Þð
2 s þ 1Þ

variables is deleted in the following. The Laplace vari-

able s is often omitted to simplify notation. The settings Thus, we need to estimate the following model infor-
given in this paper are for the series (cascade, ‘‘interact- mation (see Fig. 2):
ing’’) form PID controller:

  Plant gain, k

I s þ 1  Dominant lag time constant,
1
Series PID : cðsÞ ¼ Kc   ð
D s þ 1Þ

I s  (Effective) time delay (dead time), 
 Optional: Second-order lag time constant,
2 (for
Kc   dominant second-order process for which
2 > ,
¼
I
D s2 þ ð
I þ
D Þs þ 1 ð1Þ

I s approximately)

where Kc is the controller gain, tI the integral time, and tD If the response is lag-dominant, i.e. if
1 > 8y
the derivative time. The reason for using the series form is approximately, then the individual values of the time
that the PID rules with derivative action are then much constant
1 and the gain k may be difficult to obtain, but
simpler. The corresponding settings for the ideal (parallel at the same time are not very important for controller
form) PID controller are easily obtained using (36). design. Lag-dominant processes may instead be
approximated by an integrating process using
1.2. Simulations.
k k k0
¼ ð5Þ
The following series form PID controller is used in all
1 s þ 1
1 s s
simulations and evaluations of performance:




I s þ 1
D s þ 1
uðsÞ ¼ Kc ys ðsÞ  yð s Þ ð2Þ

I s
F s þ 1

with
F=
D and =0.01 (the robustness margins have

been computed with =0). Note that we, in order to
avoid ‘‘derivative kick’’, do not differentiate the setpoint
in (2). The value =0.01 was chosen in order to not bias
the results, but in practice (and especially for noisy
processes) a larger value of a in the range 0.1–0.2 is
normally used. In most cases we use PI-control, i.e.

D=0, and the above implementation issues and differ-
ences between series and ideal form do not apply. In the
time domain the PI-controller becomes

uðtÞ ¼ u0

ðt 
1
þ Kc ðbys ðtÞ  yðtÞÞ þ ðys ð
Þ  yð
ÞÞd
ð3Þ Fig. 2. Step response of first-order plus time delay process,

I 0 gðsÞ ¼ kes =ð
1 s þ 1Þ.
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 293

which is exact when t1!1 or 1/t1!0. In this case we the various approximated terms. In addition, for digital
need to obtain the value for the implementation with sampling period h, the contribu-
tion to the effective delay is approximately h/2 (which is
def
 Slope, k0 ¼ k=
1 the average time it takes for the controller to respond to
a change).
The problem of obtaining the effective delay  (as well In terms of control, the lag-approximation (8) is con-
as the other model parameters) can be set up as a para- servative, since the effect of a delay on control perfor-
meter estimation problem, for example, by making a mance is worse than that of a lag of equal magnitude
least squares approximation of the open-loop step (e.g. [8]). In particular, this applies when approximating
response. However, our goal is to use the resulting the largest of the neglected lags. Thus, to be less con-
effective delay to obtain controller settings, so a better servative it is recommended to use the simple half rule:
approach would be to find the approximation which for
a given tuning method results in the best closed-loop  Half rule: the largest neglected (denominator)
response [here ‘‘best’’ could, for example, bye to mini- time constant (lag) is distributed evenly to the
mize the integrated absolute error (IAE) with a specified effective delay and the smallest retained time
value for the sensitivity peak, Ms]. However, our main constant.
objective is not ‘‘optimality’’ but ‘‘simplicity’’, so we
propose a much simpler approach as outlined next. In summary, let the original model be in the form
Q 
2.1. Approximation of effective delay using the half rule Tj0inv þ 1
j
Q e0 s ð9Þ
We first consider the control-relevant approximation
i0 s þ 1
i
of the fast dynamic modes (high-frequency plant
dynamics) by use of an effective delay. To derive these
approximations, consider the following two first- where the lags
i0 are ordered according to their magni-
order Taylor approximations of a time delay transfer tude, and Tj0inv > 0 denote the inverse response (negative
function: numerator) time constants. Then, according to the half-
rule, to obtain a first-order model es =ð
1 s þ 1Þ, we use
1 1
es 1  s and es ¼ ð6Þ
es 1 þ s
20
20 X X h

1 ¼
10 þ ;  ¼ 0 þ þ
i0 þ Tj0inv þ
2 2 i53 j
2
From (6) we see that an ‘‘inverse response time con- ð10Þ
stant’’ T0inv (negative numerator time constant) may be
approximated as a time delay:
and, to obtain a second-order model (4), we use
  T0inv s
T0inv s þ 1 e ð7Þ
30

1 ¼
10 ;
2 ¼
20 þ ;
2

30 X X h ð11Þ
This is reasonable since an inverse response has a  ¼ 0 þ þ
i0 þ Tj0inv þ
deteriorating effect on control similar to that of a time 2 i54 j
2
delay (e.g. [8]). Similarly, from (6) a (small) lag time
constant t0 may be approximated as a time delay:
where h is the sampling period (for cases with digital
1 implementation).
e
0 s ð8Þ The main basis for the empirical half-rule is to main-

0 s þ 1
tain the robustness of the proposed PI- and PID-tuning
rules, as is justified by the examples later.
Furthermore, since
Example E1. The process
T0inv s þ 1 s 0 s T0inv s 
0 s
e e e e 1

0 s þ 1 g0 ð s Þ ¼
0 þT0inv þ
0 ðs þ 1Þð0:2s þ 1Þ
¼ e ð Þs ¼ es

is approximated as a first-order time delay process,

it follows that the effective delay  can be taken as the g(s)=kes+1/(
1s+1), with k=1, =0.2/2=0.1 and
sum of the original delay 0, and the contribution from
1=1+0.2/2=1.1.
294 S. Skogestad / Journal of Process Control 13 (2003) 291–309

2.2. Approximation of positive numerator time constants 0:1

k ¼ 20:75 ¼ 1:5; ¼ þ 0:1 ¼ 0:15;
2
We next consider how to get a model in the form (9), 0:1
if we have positive numerator time constants T0 in the
1 ¼ 1 þ ¼ 1:05
2
original model g0(s). It is proposed to cancel the
numerator term (T0s+1) against a ‘‘neighbouring’’ or as a second-order time delay model with
denominator term (
0s+1) (where both T0 and
0 are
positive and real) using the following approximations: 0:1 0:1
k ¼ 1:5;  ¼ ¼ 0:05;
1 ¼ 1;
2 ¼ 0:1 þ ¼ 0:15
2 2
3. Derivation of PID tuning rules (step 2)
8
>
> T0 =
0 for T 0 5
0 5  ðRule T1Þ
>
> 3.1. Direct synthesis (IMC tuning) for setpoints
>
> T0 = for T 0 5  5
0 ðRule T1aÞ
>
>
>
>
T0 s þ 1 <1 for  5 T0 5
0 ðRule T1bÞ Next, we derive for the model in (4) PI-settings or

0 s þ 1 >
> T0 =
0 for
0 5 T0 5 5 ðRule T2Þ PID-settings using the method of direct synthesis for
>
>
>
> setpoints [3], or equivalently the Internal Model Control
>
>
>
> ð
~ 0 =
0 Þ def approach for setpoints [2]. For the system in Fig. 1, the
: for
~ 0 ¼ minð
0 ; 5Þ T0 ðRule T3Þ
ð
~ 0 
0 Þs þ 1 closed-loop setpoint response is
ð12Þ
y gðsÞcðsÞ
¼ ð14Þ
Here  is the (final) effective delay, which exact value ys gðsÞcðsÞ þ 1
depends on the subsequent approximation of the time
constants (half rule), so one may need to guess  and where we have assumed that the measurement of the
iterate. If there is more than one positive numerator output y is perfect. The idea of direct synthesis is to
time constant, then one should approximate one T0 at a specify the desired closed-loop response and solve for
time, starting with the largest T0. the corresponding controller. From (14) we get
We normally select
0 as the closest larger denomi-
nator time constant (
0 > T0) and use Rules T2 or T3. 1 1
cð s Þ ¼ ð15Þ
The exception is if there exists no larger
0, or if there is gðsÞ 1
smaller denominator time constant ‘‘close to’’ T0, in ðy=ys Þdesired 1
which case we select
0 as the closest smaller denominator
time constant (
0 < T0) and use rules T1, T1a or T1b. To We here consider the second-order time delay model
define ‘‘close to’’ more precisely, let
0a (large) and
0b g(s) in (4), and specify that we, following the delay,
(small) denote the two neighboring denominator con- desire a simple first-order response with time constant
stants to t0. Then, we select
0=
0b (small) if T0/
0b <
0a/
c [2,3]:
T0 and T0/
0b < 1.6 (both conditions must be satisfied).

Derivations of the above rules and additional exam- y 1
¼ es ð16Þ
ples are given in the Appendix. ys desired
c s þ 1

Example E3. For the process (Example 4 in [9]) We have kept the delay  in the ‘‘desired’’ response
because it is unavoidable. Substituting (16) and (4) into
2ð15s þ 1Þ (15) gives a ‘‘Smith Predictor’’ controller [10]:
g0 ð s Þ ¼ ð13Þ
ð20s þ 1Þðs þ 1Þð0:1s þ 1Þ2
ð
1 s þ 1Þð
2 s þ 1Þ 1
cð s Þ ¼ ð17Þ
k ð
c s þ 1  es Þ
we first introduce from Rule T2 the approximation

c is the desired closed-loop time constant, and is the
sole tuning parameter for the controller. Our objective
15s þ 1 15s is to derive PID settings, and to this effect we introduce
¼ 0:75 in (17) a first-order Taylor series approximation of the
20s þ 1 20s
delay, es 1  s. This gives

(Rule T2 applies since T0=15 is larger than 5, where ð
1 s þ 1Þð
2 s þ 1Þ 1
cð s Þ ¼ ð18Þ
 is computed below). Using the half rule, the process k ð
c þ Þs
may then be approximated as a first-order time delay
model with which is a series form PID-controller (1) with [2,3]
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 295

1
1 1 1
Kc ¼ ¼ ;
I ¼
1 ;
k ð
c þ Þ k0 ð
c þ Þ
ð19Þ

D ¼
2

3.2. Modifying the integral time for improved

disturbance rejection

The PID-settings in (19) were derived by considering

the setpoint response, and the result was that we should
effectively cancel the first order dynamics of the process
by selecting the integral time
I=
1. This is a robust
setting which results in very good responses to setpoints
and to disturbances entering directly at the process out-
put. However, it is well known that for lag dominant Fig. 3. Effect of changing the integral time
I for PI-control of
‘‘almost integrating’’ process gðsÞ ¼ es =ð30s þ 1Þ with Kc ¼ 15. Unit
processes with
1  (e.g. an integrating processes), the setpoint change at t=0; load disturbance of magnitude 10 at t=20.
choice
I=
1 results in a long settling time for input
(‘‘load’’) disturbances [6]. To improve the load dis-
turbance response we need to reduce the integral time, is much larger than 1.1 With a PI controller c=Kc
but not by too much, because otherwise we get slow (1+
11 s) the closed-loop characetristic polynomial
oscillations caused by having almost have two inte- 1+gc then becomes
grators in series (one from the controller and almost one

I 2
from the slow lag dynamics in the process). This is illu- s þ
I s þ 1
strated in Fig. 3, where we, for the process, k0 KC

es =ð
1 s þ 1Þ with
1 ¼ 30;  ¼ 1 which is in standard second-order form,
02 s2 þ 2
0 s þ 1;
with
consider PI-control with Kc=15 and four different rffiffiffiffiffiffiffiffiffi
values of the integral time:
I 1 pffiffiffiffiffiffiffiffiffiffiffiffiffi

0 ¼ 0
; ¼ k0 Kc
I ð20Þ
k Kc 2

I=
1=30 [‘‘IMC-rule’’, see (19)]: excellent set-
point response, but slow settling for a load dis- Oscillations occur for < 1. Of course, some oscilla-
turbance. tions may be tolerated, but a robust choice is to have

I=8=8 (SIMC-rule, see below): faster settling =1 (see also [11] p. 588), or equivalently
for a load disturbance.

I=4: even faster settling, but the setpoint Kc
I ¼ 4=k0 ð21Þ
response (and robustness) is poorer.

I=2: poor response with ‘‘slow’’ oscillations. Inserting the recommended value for Kc from (19)
then gives the following modified integral time for pro-
A good trade-off between disturbance response and cesses where the choice
I=
1 is too large:
robustness is obtained by selecting the integral time
such that we just avoid the slow oscillations, which
I ¼ 4ð
c þ Þ ð22Þ
corresponds to
I=8 in the above example. Let us
analyze this in more detail. First, note that these ‘‘slow’’
oscillations are not caused by the delay  (and occur at a 3.3. SIMC-PID tuning rules
lower frequency than the ‘‘usual fast’’ oscillations which
occur at about frequency 1/). Because of this, we To summarize, the recommended SIMC PID settings2
neglect the delay in the model when we analyze the slow for the second-order time delay process in (4) are3
oscillations. The process model then becomes

es 1 k k0
gðsÞ ¼ k k ¼ 1
From (20) and (22) we get
0=
I/2, so !0
1=
10
1 ¼ 2

1I . Here

1 s þ 1
1 s þ 1
1 s s

15
I, and it follows that !0
1 1.
2
Here SIMC means ‘‘Simple control’’ or ‘‘Skogestad IMC’’.
where the second approximation applies since the 3
The derivative time in (25) is for the series form PID-controller in
resulting frequency of oscillations !0 is such that (
I!0)2 (1).
296 S. Skogestad / Journal of Process Control 13 (2003) 291–309

1
1 1 1 offset for load disturbances occuring at the input. To

Kc ¼ ¼ ð23Þ
k
c þ  k0
c þ  remove this offset, we need to reintroduce integral
  action, and as before propose to use

I ¼ min
1 ; 4ð
c þ Þ ð24Þ

I ¼ 4ð
c þ Þ ð27Þ

D ¼
2 ð25Þ

Here the desired first-order closed-loop response time It should be noted that derivative action is required to

c is the only tuning parameter. Note that the same rules stabilize a double integrating process if we have integral
are used both for PI- and PID-settings, but the actual action in the controller.
settings will differ. To get a PI-controller we start from a
first-order model (with
2=0), and to get a PID-con- 3.4. Recommended choice for tuning parameter
c
troller we start from a second-order model. PID-control
(with derivative action) is primarily recommended for The value of the desired closed-loop time constant
c
processes with dominant second order dynamics (with can be chosen freely, but from (23) we must have   <

2 > , approximately), and we note that the derivative
c < 1 to get a positive and nonzero controller gain.
time is then selected so as to cancel the second-largest The optimal value of
c is determined by a trade-off
process time constant. between:
In Table 1 we summarize the resulting settings for a
few special cases, including the pure time delay process, 1. Fast speed of response and good disturbance
integrating process, and double integrating process. For rejection (favored by a small value of
c)
the double integrating process, we let let
2!1 and 2. Stability, robustness and small input variation
introduce k00 =k0 /
2 and find (after some algebra) that (favored by a large value of
c).
the PID-controller for the integrating process with lag
approaches a PD-controller with A good trade-off is obtained by choosing
c equal to
the time delay:
1 1
Kc ¼  ;
D ¼ 4ð
c þ Þ ð26Þ SIMC-rule for fast response with good robustness :
k 4ð
c þ Þ2
00

c ¼  ð28Þ

This controller gives good setpoint responses for the This gives a reasonably fast response with moderate
double integrating process, but results in steady-state input usage and good robustness margins, and for the

Table 1
SIMC PID-settings (23)–(25) for some special cases of (4) (with tc as a tuning parameter)

Process g(s) Kc
I
Dd

es 1
1   –
First-order k min
1 ; 4ð
c þ Þ
ð
1 s þ 1Þ k
c þ 

es 1
1  
Second-order, Eq. (4) k min
1 ; 4ð
c þ Þ
2
ð
1 s þ 1Þð
2 s þ 1Þ k
c þ 

Pure time delaya kes 0 0e –

es 1 1
Integratingb k0  4ð
c þ Þ –
s k0 ð
c þ Þ

es 1 1
Integrating with lag k0  4ð
c þ Þ
2
sð
2 s þ 1Þ k0 ð
c þ Þ

es 1 1
Double intergratingc k00  4ð
c þ Þ 4ð
c þ Þ
s2 k0 4ð
c þ Þ2

a
The pure time delay process is a special case of a first-order process with
1=0.
b
The integrating process is a special case of a first-order process with
1!1.
c
For the double integrating process, integral action has been added according to Eq. (27).
d
The derivative time is for the series form PID controller in Eq. (1).
def
e
Pure integral controller cðsÞ ¼ KsI with KI ¼ K
Ic ¼ kð
c1þÞ.
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 297

second-order time delay process in (4) results in the fol- upper bound of 2). The maximum allowed time delay
lowing SIMC-PID settings which may be easily mem- error is
/=PM [rad]/(!c.), which in this case gives
orized (
c=):
/=2.14 (i.e. the system goes unstable if the time
delay is increased from  to (1+2.14)=3.14).
As expected, the robustness margins are somewhat
0:5
1 0:5 1 poorer for lag-dominant processes with
1 > 8, where
Kc ¼ ¼ 0 ð29Þ
k  k  we in order to improve the disturbance response use

I ¼ minf
1 ; 8g ð30Þ
I=8. Specifically, for the extreme case of an integrat-
ing process (right column) the suggested settings give

D ¼
2 ð31Þ GM=2.96, PM=46.9 , Ms=1.70 and Mt=1.30, and
the maximum allowed time delay error is
=1.59.
Of the robustness measures listed above, we will in the
The corresponding settings for the ideal PID-controller following concentrate on Ms, which is the peak value as
are given in (37) and (38). a function of frequency of the sensitivity function S=1/
(1 +gc). Notice that Ms < 1.7 guarantees GM > 2.43
4. Evaluation of the proposed tuning rules and PM > 34.2 [2].

In this section we evaluate the proposed SIMC PID 4.1.2. Performance

tuning rules in (23)–(31) with the choice
c=. We first To evaluate the closed-loop performance, we consider
consider processes that already are in the second-order a unit step setpoint change (ys=1) and a unit step input
plus delay form in (4). In Section 4.2 we consider more (load) disturbance (gd=g and d=1), and for each of the
complicated processes which must first be approximated two consider the input and output performance:
as second-order plus delay processes (step 1), before
applying the tuning rules (step 2). 4.1.2.1. Output performance. To evaluate the output
control performance we compute the integrated abso-
4.1. First- or second-order time delay processes lute error (IAE) of the control error e=yys.
ð1
4.1.1. Robustness  
IAE ¼ eðtÞdt
The robustness margins with the SIMC PID-settings 0
in (29)–(31), when applied to first- or second-order time
delay processes, are always between the values given by which should be as small as possible.
the two columns in Table 2.
For processes with
148, for which we use
I=
1 4.1.2.2. Input performance. To evaluate the manipulated
(left column), the system always has a gain margin input usage we compute the total variation (TV) of the
GM=3.14 and phase margin PM=61.4 , which is input u(t), which is sum of all its moves up and down. TV
much better than than the typical minimum require- is difficult to define compactly for a continuous signal,
ments GM > 1.7 and PM > 30 [12]. The sensitivity and but if we discretize the input signal as a sequence, [u1,
complementary sensitivity peaks are Ms=1.59 and u2, . . ., ui . . . ], then
Mt=1.00 (here small values are desired with a typical
X
1  
TV ¼ uiþ1  ui 
i¼1
Table 2
Robustness margins for first-order and integrating time delay process
using the SIMC-settings in (29) and (30) (tc=y) which should be as small as possible. The total variation
k s k0
is a good measure of the ‘‘smoothness’’ of a signal.
Process g(s)
1 sþ1 e es
s In Table 3 we summarize the results with the choice
c ¼
0:5
1 0:5 1
Controller gain, Kc k  k0   for the following five first-order time delay processes:
Integral time,
I
1 8
Gain margin (GM) 3.14 2.96
Case 1. Pure time delay process
Phase margin (PM) 61.4 46.9
Sensitivity peak, Ms 1.59 1.70 Case 2. Integrating process
Complementary sensitivity peak, Mt 1.00 1.30 Case 3. Integrating process with lag
2=4
Phase crossover frequency, !180. 1.57 1.49 Case 4. Double integrating process
Gain crossover frequency, !c. 0.50 0.51 Case 5. First-order process with
1=4
Allowed time delay error,
/ 2.14 1.59

The same margins apply to a second-order process (4) if we choose Note that the robustness margins fall within the limits

D=
2, see (31). given in Table 2, except for the double integrating
298 S. Skogestad / Journal of Process Control 13 (2003) 291–309

Table 3
SIMC settings and performance summary for five different time delay processes (tc=y)

Case g(s) Kc tI tDc Ms Setpointa Load disturbance

IAE
IAE(y) TV(u) IAE(y) TV(u) IAEmin

1
1 kes 0 0d – 1.59 2.17 1.08 k 2.17 k 1.08 1.59
0 es 0:5 1 1 0 2
2 k s k0   8y – 1.70 3.92 1.22 k0  16 k  1.55 3.27
0 es 0:5 1 1 0 2
3 k sð4sþ1Þ k0   8y
2=4y 1.70 5.28 1.23 k0  16 k  1.59 5.41
s
0:0625 1 1
4 k00 es:2 k00  2 8y 8y 1.96 7.92 0.205 k00 2
128 k00 3 2.34 5.49
es 0:5
1 2 1
5 k 4sþ1 k  ¼ k
1=4y – 1.59 2.17 4.11 k 2 k 1.08 2.41
a
The IAE and TV-values for PID control are without derivative action on the setpoint.
b
IAEmin is for the IAE-optimal PI/PID-controller of the same kind
c
The derivative time is for the series form PID controller in Eq. (1).
d
Pure integral controller cðsÞ ¼ KsI with KI ¼ K
Ic ¼ 0:5
k .

process in case 4 where we, from (27), have added inte- for the double integrating process in case 4. Note also
gral action, and robustness is somewhat poorer. that the setpoint response can always be modified by
introducing a ‘‘feedforward’’ filter on the setpoint or
4.1.2.3. Setpoint change. The simulated time responses using b 6¼ 1 in (3).
for the five cases are shown in Fig. 4. The setpoint
responses are nice and smooth. For a unit setpoint 4.1.2.4. Load disturbance. The load disturbance
change, the minimum achievable IAE-value for these responses in Fig. 4 are also nice and smooth, although a
time delay processes is IAE= [e.g. using a Smith Pre- bit sluggish for the integrating and double integrating
dictor controller (17) with
c=0]. From Table 3 we see processes. In the last column in Table 3 we compare the
that with the proposed settings the actual IAE-setpoint- achieved IAE-value with that for the IAE-optimal con-
value varies between 2.17 (for the first-order process) to troller of the same kind (PI or series-PID). The ratio
7.92 (for the more difficult double integrating process). varies from 1.59 for the pure time delay process to 5.49
To avoid ‘‘derivative kick’’ on the input, we have for the more difficult double integrating process.
chosen to follow industry practice and not differentiate However, lower IAE-values generally come at the
the setpoint, see (2). This is the reason for the difference expense of poorer robustness (larger value of Ms), more
in the setpoint responses between cases 2 and 3, and also excessive input usage (larger value of TV), or a more com-
the reason for the somewhat sluggish setpoint response plicated controller. For example, for the integrating pro-
cess, the IAE-optimal PI-controller (Kc ¼ 0:91 1
k0   ;

I ¼ 4:1) reduces IAE(load) by a factor 3.27, but the
input variation increases from TV=1.55 to TV=3.79, and
the sensitivity peak increases from Ms=1.70 to Ms=3.71.
The IAE-optimal PID-controller (Kc ¼ 0:80 1
k0   ;

I ¼ 1:26;
D ¼ 0:76) reduces IAE(load) by a factor 8.2
(to IAE=1.95k0 2), but this controller has Ms=4.1 and
TV(load)=5.34. The lowest achievable IAE-value for the
integrating process is for an ideal Smith Predictor con-
troller (17) with
c=0, which reduces IAE(load) by a factor
32 (to IAE=0.5k0 2). However, this controller is unrealiz-
able with infinite input usage and requires a perfect model.

4.1.2.5. Input usage. As seen from the simulations in the

lower part of Fig. 4 the input usage with the proposed
settings is very smooth in all cases. To have no steady-
state offset for a load disturbance, the minimum
achievable value is TV(load)=1 (smooth input change
with no overshoot), and we find that the achieved value
Fig. 4. Responses using SIMC settings for the five time delay pro-
cesses in Table 3 (
c =y). Unit setpoint change at t=0; Unit load dis-
ranges from 1.08 (first-order process), through 1.55
turbance at t=20. Simulations are without derivative action on the (integrating process) and up to 2.34 (double integrating
setpoint. Parameter values:  ¼ 1; k ¼ 1; k0 ¼ 1; k00 ¼ 1. process).
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 299

4.2. More complex processes: obtaining the effective tings are given. For some processes (El, E12, E13, E14,
delay E15) only first-order approximations are derived, and
only PI-settings are given. The model approximations
We here consider some cases where we must first (step for cases E2, E3, E6 and E13 are studied separately; see
1) approximate the model as a first- or second-order (41), (13), (42) and (43). Processes El and E3–E8 have
plus delay process, before (step 2) applying the pro- been studied by Astrom and coworkers [9,13], and in all
posed tuning rules. cases the SIMC PI-settings and IAE-load-values in
In Table 4 we summarize for 15 different processes Table 4 are very similar to those obtained by Astrom
(E1–E15), the model approximation (step 1), the SIMC- and coworkers for similar values of Ms. Process E11 has
settings with
c= (step 2) and the resulting Ms-value, been studied by [14].
setpoint and load disturbance performance (IAE and The peak sensitivity (Ms) for the 25 cases ranges from
TV). For most of the processes, both PI- and PID-set- 1.23 to 2, with an average value of 1.64. This confirms

Table 4
s
Approximation gðsÞ ¼ k ð
1 sþ1e Þð
2 sþ1Þ, SIMC PI/PID-settings (tc=y) and performance summary for 15 processes

Case Process model, g0(s) Approximation, g(s) SIMC settings Performance

c
k 
1
2 Kc
I tD Ms Setpointa Load disturbanceb
IAE
IAE(y) TV(u) IAE(y) TV(u) IAEmin

1
E1 (PI) ðsþ1Þð0:2sþ1Þ 1 0.1 1.1 – 5.5 0.8 – 1.56 0.36 12.7 0.15 1.55 1
ð0:3sþ1Þð0:08sþ1Þ
E2 (PI) ð2sþ1Þð1sþ1Þð0:4sþ1Þð0:2sþ1Þð0:05sþ1Þ3
1 1.47 2.5 – 0.85 2.5 – 1.66 3.56 1.90 2.97 1.26 1.39
E2 (PID) 1 0.77 2 1.2 1.30 2 1.2 1.73 2.73 2.84 1.54 1.33 1.99
2ð15sþ1Þ
E3 (PI) ð20sþ1Þðsþ1Þð0:1sþ1Þ2
1.5 0.15 1.05 – 2.33 1.05 – 1.55 0.46 4.97 0.45 1.30 3.82

E3 (PID) 1.5 0.05 1 0.15 6.67 0.4 0.15 1.47 0.25 15.0 0.068 1.45 64
1
E4 (PI) ðsþ1Þ4
1 2.5 1.5 – 0.3 1.5 – 1.46 5.59 1.15 5.40 1.10 1.93
E4 (PID) 1 1.5 1.5 1 0.5 1.5 1 1.43 4.31 1.27 3.13 1.12 3.49
1
E5 (PI) ðsþ1Þð0:2sþ1Þð0:04sþ1Þð0:0008sþ1Þ 1 0.148 1.1 – 3.72 1.1 – 1.59 0.45 8.17 0.30 1.41 4.1
E5 (PID) 1 0.028 1.0 0.22 17.9 0.224 0.22 1.58 0.27 43.3 0.056 1.49 27
ð0:17sþ1Þ2 d
E6 (PI) sðsþ1Þ2 ð0:028sþ1Þ
1 1.69 – 0.296 13.5 – 1.48 6.50 0.67 45.7 1.55 10.1
d
E6 (PID) 1 0.358 1.33 1.40 2.86 1.33 1.23 1.95 3.19 2.04 1.55 1
2sþ1
E7 (PI) ðsþ1Þ3
1 3.5 1.5 – 0.214 1.5 – 1.66 7.28 1.06 8.34 1.28 1.23
E7 (PID) 1 2.5 1.5 1 0.3 1.5 1 1.85 5.99 1.02 6.23 1.57 1.22
1 d
E8 (PI) sðsþ1Þ2
1 1.5 – 0.33 12 – 1.76 6.47 0.84 36.4 1.78 3.2
d
E8 (PID) 1 0.5 1.5 1.5 4 1.5 1.79 2.02 4.21 2.67 1.99 40
es
E9 (PI) ðsþ1Þ2
1 1.5 1.5 – 0.5 1.5 – 1.61 3.38 1.31 3.14 1.15 1.34
E9 (PID) 1 1 1 1 0.5 1 1 1.59 3.03 1.29 2 1.10 1.60
es
E10 (PI) ð20sþ1Þð2sþ1Þ 1 2 21 – 5.25 16 – 1.72 6.34 12.3 3.05 1.49 2.9
E10 (PID) 1 1 20 2 10 8 2 1.65 4.32 22.8 0.80 1.37 4.9
ðsþ1Þ
E11 (PI) ð6sþ1Þð2sþ1Þ2
es 1 5 7 – 0.7 7 – 1.63 11.5 1.59 10.1 1.20 1.37
E11 (PID) 1 3 6 3 1 6 3 1.66 9.09 2.11 6.03 1.24 1.86
ð6sþ1Þð3sþ1Þe0:3s
E12 (PI) ð10sþ1Þð8sþ1Þðsþ1Þ 0.225 0.3 1 – 7.41 1 – 1.66 1.07 18.3 0.15 1.39 2.1
2sþ1 s
E13 (PI) ð10sþ1Þð0:5sþ1Þ e 0.625 1.25 4.5 – 2.88 4.50 – 1.74 2.86 6.56 1.61 1.20 3.39
sþ1 d
E14 (PI) s 1 1 – 0.5 8 – 2 3.59 2.04 17.3 3.40 3.75
sþ1
E15 (PI) sþ1 1 1 1 – 0.5 1 – 2 2 1.02 2.85 3.00 1.23
a
The IAE- and TV-values for PID control are without derivative action on the setpoint.
b
IAEmin is for the IAE-optimal PI- or PID-controller.
c
The derivative time is for the series form PID controller in Eq. (1).
s
d
Integrating process, gðsÞ ¼ k0 sð
e2 sþ1Þ.
300 S. Skogestad / Journal of Process Control 13 (2003) 291–309

that the simple approximation rules (including the half results with PI-control and PID-control. We note from
rule for the effective delay) are able to maintain the ori- Table 4 that there is a close correlation between the
ginal robustness where Ms ranges from 1.59 to 1.70 (see value of
2 = and the improvement in IAE for load
Table 2). The poorest robustness with Ms=2 is changes. For example,
2 = is infinite for case E1, and
obtained for the two inverse response processes in E14 indeed the (theoretical) improvement with PID control
and E15. For these two processes, we also find that the over PI control is infinite. In cases E5, E6, E8, E3, E10
input usage is large, with TV for a load disturbance and E2 the ratio
2 = is larger than 1 (ranges from 7.9 to
larger than 3, whereas it for all other cases is less than 2 1.6), and there is a significant improvement in IAE with
(the minimum value is 1). The inverse responses pro- PID control (by a factor 22–1.9). In cases E11, E9, E4
cesses E14 and E15 are rather unusual in that the pro- and E7 the ratio
2 = is less than 1 (ranges from 1 to 0.4)
cess gain remains finite (at 1) at high frequencies, and and the improvement with PID control is rather small
we also have that they give instability with PID control. (by a factor 1.6 to 1.3). This improvement is too small in
The input variation (TV) for a setpoint change is large most cases to justify the additional complexity and noise
in some cases, especially for cases where the controller sensitivity of using derivative action.
gain Kc is large. In such cases the setpoint response may In summary, these 15 examples illustrate that the
be slowed down by, for example, prefiltering the setpoint simple SIMC tuning rules used in combination with the
change or using b smaller than 1 in (3). (Alternatively, if simple half-rule for estimating the effective delay, result
input usage is not a concern, then prefiltering or use of b in good and robust settings.
> 1 may be used to speed up the setpoint response.)
The last column in Table 4 gives for a load dis-
turbance the ratio between the achieved IAE and the
minimum IAE with the same kind of controller (PI or 5. Comparison with other tuning methods
series-PID) with no robustness limitations imposed. In
many cases this ratio is surprisingly small (e.g. less than Above we have evaluated the proposed SIMC tuning
1.4 for the PI-settings for cases E2, E7, E9, E11 and approach on its own merit. A detailed and fair com-
E15). However, in most cases the ratio is larger, and parison with other tuning methods is virtually impos-
even infinity (cases E1 and E6-PID). The largest values sible—because there are many tuning methods, many
are for processes with little or no inherent control limi- possible performance criteria and many possible mod-
tations (e.g. no time delay), such that theoretically very els. Nevertheless, we here perform a comparison for
large controller gains may be used. In practice, this three typical processes; the integrating process with
performance can not be achieved due to unmodeled delay (Case 2), the pure time delay process (Case 1), and
dynamics and limitations on the input usage. the fourth-order process E5 with distributed time con-
For example, for the second-order process gðsÞ ¼ stants. The following four tuning methods are used for
1
ðsþ1Þð0:2sþ1Þ (case E1) one may in theory achieve perfect comparison:
control (IAE=0) by using a sufficiently high controller
gain. This is also why no SIMC PID- settings are given 5.1. Original IMC PID tuning rules
in Table 4 for this process, because the choice
c==0
gives infinite controller gain. More precisely, going back In [2] PI and PID settings for various processes are
to (23) and (24), the SIMC-PID settings for process E1 are derived. For a first-order time delay process the
‘‘improved IMC PI-settings’’ for fast response ("=1.7)
1
1 1 are:
Kc ¼ ¼ ;
I ¼ 4
c ;
D ¼
2 ¼ 0:1 ð32Þ
k
c
c



1 þ
These settings give for any value of tc excellent 0:588 2 
IMC PI : Kc ¼ ;
I ¼
1 þ ð33Þ
robustness margins. In particular, for tc!0 we get k  2
GM=1, PM=76.3 , Ms=1, and Mt=1.15. However,
in this case the good margins are misleading since the and the PID-settings for fast response (e=0.8) are
gain crossover frequency, !c 1=
c , approaches infinity
as
c goes to zero. Thus, the time delay error
 ¼ 0:769
1
IMC series-PID : Kc ¼ ;
I ¼
1 ;
PM=!c that yields instability approaches zero (more k 
ð34Þ
precisely, 1.29
c) as
c goes to zero. 

D ¼
The recommendation given earlier was that a second- 2
order model (and thus use of PID control with SIMC
settings) should only be used for dominant second-order Note that these rules give
I5
1, so the response to
process with t2 > , approximately. This recommenda- input load disturbances will be poor for lag dominant
tion is justified by comparing for cases E1-E11 the processes with t1 .
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 301

5.2. Astrom/Schei PID tuning (maximize KI) Remark. We have here assumed that the PID-settings
given by Ziegler and Nichols (K0c ¼ 0:6Ku ,
0I ¼ Pu =2,
0
Schei [14] argued that in process control applications
D ¼ Pu =8) were originally derived for the ideal form
we usually want a robust design with the highest possi- PID controller (see [15] for justification), and have
ble attenuation of low-frequency disturbances, and translated these into the corresponding series settings
proposed to maximize the low-frequency controller gain using (36). This gives somewhat less agressive settings
def Kc and better IAE-values than if we assume that the ZN-
KI subject to given robustness constraints on the

I settings were originally derived for the series form. Note
sensitivity peaks Ms and Mt. Both for PI- and PID- that Kc/
I and Kc
D are not affected, so the difference is
control, maximizing KI is equivalent to minimizing the only at intermediate frequencies.
integrated error (IE) for load disturbances, which for
robust designs with no overshoot is the same as mini- 5.4. Tyreus–Luyben modified ZN PI tuning rules
mizing the integral absolute error (IAE) [5]. Note that
the use of derivative action (
D) does not affect the IE The ZN settings are too aggressive for most process
(and also not the IAE for robust designs), but it may control applications, where oscillations and overshoot
improve robustness (lower Ms) and reduce the input are usually not desired. This led Tyreus and Luyben [4]
variation (lower TV—at least with no noise). Astrom [9] to recommend the following PI-rules for more con-
showed how to formulate the minimization of KI as an servative tuning:
efficient optimization problem for the case with PI con-
trol and a constraint on Ms. The value of the tuning Kc ¼ 0:313Ku ;
I ¼ 2:2Pu
parameter Ms is typically between 1.4 (robust tuning)
and 2 (more aggressive tuning). We will here select it to
be the same as for the corresponding SIMC design, that 5.5. Integrating process
is, typically around 1.7.
s
The results for the integrating process, gðsÞ ¼ k0 e s ,
5.3. Ziegler–Nichols (ZN) PID tuning rules are shown in Table 5 and Fig. 5. The SIMC-PI con-
troller with
c= yields Ms=1.7 and IAE(load)=16.
In [1] it was proposed as the first step to generate The Astrom/Schei PI-settings for Ms=1.7 are very
sustained oscillations with a P-controller, and from this similar to the SIMC settings, but with somewhat better
obtain the ‘‘ultimate’’ gain Ku and corresponding ‘‘ulti- load rejection (IAE reduced from 16 to 13). The ZN PI-
mate’’ period Pu (alternatively, this information can be controller has a shorter integral time and larger gain
obtained using relay feedback [5]). Based on simulations, than the SIMC-controller, which results in much better
the following ‘‘closed-loop’’ settings were recommended: load rejection with IAE reduced from 16 to 5.6. How-
ever, the robustness is worse, with Ms increased from
P-control : Kc ¼ 0:5Ku 1.70 to 2.83 and the gain margin reduced from 2.96 to
1.86. The IMC settings of Rivera et al. [2] result in a
PI-control : Kc ¼ 0:45Ku ;
I ¼ Pu =1:2 pure P-controller with very good setpoint responses, but
there is steady-state offset for load disturbances. The
PID-controlðseriesÞ : Kc ¼ 0:3Ku ;
I ¼ Pu =4; modified ZN PI-settings of Tyreus–Luyben are almost
identical to the SIMC-settings. This is encouraging since

D ¼ Pu =4: it is exactly for this type of process that these settings
were developed [4].

Table 5
Tunings and performance for integrating process, g(s)=k0 eys/s

Setpointb Load disturbance

Method Kc.k y
0

I/
D/ a
Ms IAE(y) TV(u) IAE(y) TV(u)

SIMC (
c=) 0.5 8 – 1.70 3.92 1.22 16.0 1.55

IMC (e=1.7y) 0.59 1 – 1.75 2.14 1.32 1 1.24
Astrom/Schei (Ms=1.7) 0.404 7.0 – 1.70 4.56 1.16 13.0 1.88
ZN-PI 0.71 3.33 – 2.83 3.92 2.83 5.61 2.87
Tyreus–Luyben 0.49 7.32 – 1.70 3.95 1.21 14.9 1.59
ZN-PID 0.471 1 1 2.29 2.88 2.45 3.32 3.00
a
The derivative time is for the series form PID controller in Eq. (1).
b
The IAE- and TV-values for PID control are withput derivative action on the setpoint.
302 S. Skogestad / Journal of Process Control 13 (2003) 291–309

Table 6
Tunings and performance for pure time delay process, g(s)=kes

Setpointb Load disturbance

Method Kc.k0 KI.
I/c
D/a Ms IAE(y) TV(u) IAE(y) TV(u)

SIMC (
c=) 0 0.5 – 1.59 2.17 1.08 2.17 1.08

IMC-PI (E=1.7) 0.294 0.588 – 1.62 1.71 1.22 1.71 1.22
Astrom/Schei (Ms=1.6) 0.200 0.629 – 1.60 1.59 1.08 1.59 1.08
Pessen 0.25 0.751 – 1.80 1.45 1.30 1.45 1.30
ZN-PI 0.45 0.27 – 1.85 3.70 1.53 3.70 1.53
Tyreus–Luyben 0.313 0.071 – 1.46 14.1 1.22 14.1 1.22
IMC-PID (E=0.8) 0 0.769 0.5 2.01 1.90 1.06 1.38 1.67
ZN-PID 0.3 0.6 0.5 Unstable
a
KI=Kc/
I is the integral controller gain.
b
The derivative time is for the series form PID controller in Eq. (1).
c
The IAE- and TV-values for PID control are without derivative action on the setpoint.

5.6. Pure time delay process IAE=3.70, and the Tyreus–Luyben controller is extre-
mely sluggish with IAE=14.1. This is due to a low value
The results for the pure time delay process, of the integral gain KI.
g(s)=kes, are given in Table 6 and Fig. 6. Note that Because the process gain remains constant at high fre-
the setpoint and load disturbances responses are iden- quency, any ‘‘real’’ PID controller (with both propor-
tical for this process, and also that the input and output tional and derivative action), yields instability for this
signals are identical, except for the time delay. process, including the ZN PID-controller [2]. (However,
Recall that the SIMC-controller for this process is a the IMC PID-controller is actually an ID-controller, and
pure integrating controller with Ms=1.59 and it yields a stable response with IAE=1.38.)
IAE=2.17. The minimum achievable IAE-value for any The poor response with the ZN PI-controller and the
controller for this process is IAE=1 [using a Smith instability with PID control, may partly explain the
Predictor (17) with tc=0]. We find that the PI-settings myth in the process industry that time delay processes
using SIMC (IAE=2.17), IMC (IAE=1.71) and cannot be adequately controlled using PID controllers.
Astrom/Schei (IAE=1.59) all yield very good perfor- However, as seen from Table 6 and Fig. 6, excellent
mance. In particular, note that the excellent Astrom/ performance can be achieved even with PI-control.
Schei performance is achieved with good robustness
(Ms=1.60) and very smooth input usage (TV=1.08).
Pessen [16] recommends PI-settings for the time delay 5.7. Fourth-order process (E5)
process that give even better performance (IAE=1.44),
but with somewhat worse robustness (Ms=1.80). The The results for the fourth-order process E5 [9] are
ZN PI-controller is significantly more sluggish with shown in Table 7 and Fig. 7. The SIMC PI-settings

Fig. 5. Responses for PI-control of integrating process, gðsÞ ¼ es =s,

with settings from Table 5. Setpoint change at t=0; load disturbance Fig. 6. Setpoint responses for PI-control of pure time delay process,
of magnitude 0.5 at t=20. gðsÞ ¼ es , with settings from Table 6.
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 303

Table 7
1
Tuning and peformance for process gðsÞ ¼ ðsþ1Þð0:2sþ1Þð0:04sþ1Þð0:008sþ1Þ ðE5Þ

Setpointb Load disturbance

Method Kc
I
D a Ms IAE(y) TV(u) IAE(y) TV(u)

SIMC-PI ð
c ¼ Þ 3.72 1.1 – 1.59 0.45 8.2 0.296 1.41

Astrom/Schei (Ms=1.6) 2.74 0.67 – 1.60 0.58 6.2 0.246 1.52
ZN-PI 13.6 0.47 – 11.3 1.87 207 0.137 13.9
Tyreus–Luyben 9.46 1.24 – 2.72 0.50 35.8 0.131 2.91
SIMC-PID ð
c ¼ Þ 17.9 1.0 0.22 1.58 0.27 43.3 0.056 1.49
ZN-PID 9.1 0.14 0.14 2.39 0.24 39.2 0.025 3.09
a
The derivative time is for the series form PID controller in Eq. (1)
b
The IAE- and TV-vaules for PID control are without dervative action on the setpoint.

again give a smooth response [TV(load)=1.41] with 6.2. Measurement noise

good robustness (Ms=1.59) and acceptable disturbance
rejection (IAE=0.296). The Astrom/Schei PI-settings Measurement noise has not been considered in this
with Ms=1.6 give very similar reponses. IMC-settings paper, but it is an important consideration in many
are not given since no tuning rules are provided cases, especially if the proportional gain Kc is large, or,
for models in this particular form [2]. The Ziegler– for cases with derivative action, if the derivative gain
Nichols PI-settings give better disturbance rejection Kc
D is large. However, since the magnitude of the
(IAE=0.137), but as seen in Fig. 7 the system is close to measurement noise varies a lot in applications, it is dif-
instability. This is confirmed by the large sensitivity ficult to give general rules about when measurement
peak (Ms=11.3) and excessive input variation (TV=13.9) noise may be a problem. In general, robust designs (with
caused by the oscillations. The Tyreus–Luyben PI-set- small Ms) with moderate input usage (small TV) are
tings give IAE=0.131 and a much smoother response insensitive to measurement noise. Therefore, the SIMC
with TV=2.91, but the robustness is still somewhat rules with the recommended choice
c=, are less sen-
poor (Ms=2.72). As expected, since this is a dominant sitive to measurement noise than most other published
second-order process, a significant improvement can be settings method, including the ZN-settings. If actual
obtained with PID-control. As seen from Table 7 the implementation shows that the sensitivity to measure-
performance of the SIMC PID-controller is not quite as ment noise is too large, then the following modifications
good as the ZN PID-controller, but the robustness and may be attempted:
input smoothness is much better.
1. Filter the measurement signal, for example, by
sending it through a first-order filter 1/(tFs+1);
6. Discussion see also (2). With the proposed SIMC-settings
one can typically increase the filter time constant
6.1. Detuning the controller

The above recommended SIMC settings with
c=,

as well as almost all other PID tuning rules given in the
literature, are derived to give a ‘‘fast’’ closed-loop
response subject to achieving reasonable robustness.
However, in many practical cases we do need fast con-
trol, and to reduce the manipulated input usage, reduce
measurement noise sensitivity and generally make oper-
ation smoother, we may want detune the controller.
One main advantage of the SIMC tuning method is that
detuning is easily done by selecting a larger value for
c.
From the SIMC tuning rules (23) and (24) a larger value
of
c decreases the controller gain and, for lag-dominant
processes with
1 > 4(
c+), increases the integral time.
Fruehauf et al. [17] state that in process control appli-
cations one typically chooses
c > 0.5 min, except for Fig. 7. Responses for process 1=ðs þ 1Þð0:2s þ 1Þð0:04s þ 1Þð0:008s þ 1Þ
flow control loops where one may have
c about 0.05 ðE5Þ with settings from Table 7. Setpoint change at t=0; load dis-
min. turbance of magnitude 3 at t=10.
304 S. Skogestad / Journal of Process Control 13 (2003) 291–309

F up to about 0.5y, without a large affect on
2=1. The series-form SIMC settings are Kc=0.5,
1=1
performance and robustness. and tD=1. The corresponding settings for the ideal PID
0 0
2. If derivative action is used, one may try to controller in (35) are K0c =1,
I =2 and
D =0.5. The
remove it, and obtain a first-order model before robustness margins with these settings are given by the
deriving the SIMC PI-settings. first column in Table 2.
3. If derivative action has been removed and filter-
ing the measurement signal is not sufficient, then Remarks:
the controller needs to be detuned by going back
to (23)–(24) and selecting a larger value for
c. 1. Use of the above formulas make the series and
ideal controllers identical when considering the
feedback controller, but they may differ when it
6.3. Ideal form PID controller comes to setpoint changes, because one usually
does not differentiate the setpoint and the values
The settings given in this paper (Kc,
1,
D) are for the for Kc differ.
series (cascade, ‘‘interacting’’) form PID controller in 2. The tuning parameters for the series and ideal
(1). To derive the corresponding settings for the ideal forms are equal when the ratio between the deri-
(parallel, ‘‘non-interacting’’) form PID controller vative and integral time,
D =
I approaches zero,
that is, for a PI-controller (
D=0) or a PD-con-

 troller (
I=1).
1 3. Note that it is not always possible to do the
c0 ðsÞ ¼ K0c 1 þ 0 þ
D 0
s

I s reverse and obtain series settings from the ideal
ð35Þ
K0c  0 0 2 0  settings. Specifically, this can only be done when
¼ 0
I
D s þ
I s þ 1
I0 5 4
D
0
. This is because the ideal form is more

I s
general as it also allows for complex zeros in the
controller. Two implications of this are:
we use the following translation formulas
(a) We should start directly with the ideal PID


 controller if we want to derive SIMC-settings
0
D 0
D for a second-order oscillatory process (with
Kc ¼ Kc 1 þ ;
I ¼
I 1þ ;

I
I complex poles).
0
D ð36Þ (b) Even for non-oscillatory processes, the ideal

D ¼
D
1þ PID may give better performance due to its

I less restrictive form. For example, for the
process gðsÞ ¼ 1=ð
s þ 1Þ4 (E4), the minimum
achievable IAE for a load disturbance is
The SIMC-PID series settings in (29)–(31) then corre- IAE=0.89 with a series-PID, and 40% lower
spond to the following SIMC ideal-PID settings (
c=): (IAE=0.52) with an ideal PID. The optimal
settings for the ideal PID-controller
0:5 ð
1 þ
2 Þ (K0c =4.96,
I0 =1.25,
D
0
=1.84) can not be

1 4 8 : K0c ¼ ;
I0 ¼
1 þ
2 ; represented by the series controller because
k 

2 ð37Þ
I0 < 4
D
0
.
0

D ¼
2
1þ

1
6.4. Retuning for integrating processes
0 0:5
1 
2  0

1 5 8 : Kc ¼ 1þ ;
I ¼ 8 þ
2 ; Integrating processes are common in industry, but
k  8 control performance is often poor because of incorrect

2 ð38Þ

D ¼
0
settings. When encountering oscillations, the intuition

2
1þ of the operators is to reduce the controller gain. This is
8 the exactly opposite of what one should do for an inte-
grating process, since the product of the controller gain
We see that the rules are much more complicated when Kc and the integral time
I must be larger than the value
we use the ideal form. in (22) in order to avoid slow oscillations. One solution
is to simply use proportional control (with tI=1), but
Example. Consider the second-order process this is often not desirable. Here we show how to easily
gðsÞ ¼ es =ðs þ 1Þ2 (E9) with the k=1, =1,
1=1 and retune the controller to just avoid the oscillations with-
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 305

out actually having to derive a model. This approach

has been applied with success to industrial examples.
Consider a PI controller with (initial) settings Kc0 and

I0 which results in ‘‘slow’’ oscillations with period P0
(larger than 3
I0, approximately). Then we likely have
es
a close-to integrating process gðsÞ ¼ k0 for which the
s
product of the controller gain and integral time (Kc0tI0)
is too low. From (20) we can estimate the damping
coefficient and time constant t0 associated with these
oscillations of period PØ, and a standard analysis of
second-order systems (e.g. [12] p. 118) gives that the
corresponding period is

rffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi
2 2
I0
I0
P0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi
0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi 0K
2 ð39Þ
1 2 1 2 k c0 k0 Kc0
where we have assumed 2 < < 1 (significant oscilla-
tions). Thus, from (39) the product of the original con- Fig. 8. Industrial case study of retuning reboiler level control system.
troller gain and integral time is approximately

2
1
I0 eventually accepted to increase Kc by a factor 7.7 and
I
Kc0 
I0 ¼ ð2 Þ2 0
k P0 by a factor 24, that is, Kc
I was increased by
7.7.24=185. The much improved response is shown in
To avoid oscillations ð 5 1Þ with the new settings we the ‘‘after’’ plot in Fig. 8. There is still some minor
must from (21) require Kc
I54/k0 , that is, we must oscillations, but these may be caused by disturbances
require that outside the loop. In any case the control of the down-

 stream distillation column was much improved.
Kc
I 1 P0 2
5 2 ð40Þ
Kc0
I0
i0 6.5. Derivative action to counteract time delay?

Here 1= 2 0:10, so we have the rule: Introduction of derivative action, e.g.
D=/2, is
 To avoid ‘‘slow’’ oscillations of period P0 the pro- commonly proposed to improve the response when we
duct of the controller gain and integral time should be have time delay [2,3]. To derive this value we may in
increased by a factor f 0:1ðP0 =
I0 Þ2 . (17) use the more  exact 1st
 order Pade approximation,
es  2 s þ 1 = 2 s þ 1 . With the choice
c= this
Example. This actual industrial case originated as a results in the same series-form PID-controller (18)
project to improve the purity control of a distillation found
  above,
  but in addition we get a term
column. It soon become clear that the main problem 2 s þ 1 = 0:5 2 s þ 1 . This is as an additional derivative
was large variations (disturbances) in its feed flow. The term with
D=/2, effective over only a small range,
feed flow was again the bottoms flow from an upstream which increases the controller gain by a factor of two at
column, which was again set by its reboiler level con- high frequencies. However, with the robust SIMC set-
troller. The control of the reboiler level itself was tings used in this paper (
c=), the addition of deriva-
acceptable, but the bottoms flowrate showed large var- tive action (without changing Kc or
I) has in most cases
iations. This is shown in Fig. 8, where y is the reboiler no effect on IAE for load disturbances, since the integral
level and u is the bottoms flow valve position. The PI gain KI ¼ Kc =
I is unchanged and there are no oscilla-
settings had been kept at their default setting (Kc=0.5 tions [5]. Although the robustness margins are some-
and
I=1 min) since start-up several years ago, and what improved (for example, for an integrating with
resulted in an oscillatory response as shown in the top delay process, k0 es =s, the value of Ms is reduced from
part of Fig. 8. 1.70 (PI) to 1.50 (PID) by adding derivative action with
From a closer analysis of the ‘‘before’’ response we
D=/2), this probably does not justify the increased
find that the period of the slow oscillations is P0=0.85 complexity of the controller and the increased sensitivity
h=51 min. Since
I=1 min, we get from the above rule to measurement noise. This conclusion is further con-
we should increase Kc.
I by a factor f 0.1.(51)2=260 to firmed by Table 6 and Fig. 6, where we found that a PI-
avoid the oscillations. The plant personnel were some- controller (and even a pure I-controller) gave very good
what sceptical to authorize such large changes, but performance for a pure time delay process. In conclu-
306 S. Skogestad / Journal of Process Control 13 (2003) 291–309

sion, it is not recommended to use derivative action to 1. The half rule is used to approximate the process
counteract time delay, at least not with the robust set- as a first or second order model with effective
tings recommended in this paper. delay , see (10) and (11),
2. For a first-order model (with parameters k,
1
6.6. Concluding remarks and ) the following SIMC PI-settings are sug-
gested:
 As illustrated by the many examples, the very
simple analytic tuning procedure presented in 1
1  
Kc ¼ ;
I ¼ min
1 ; 4ð
c þ Þ
this paper yields surprisingly good results. Addi- k
c þ 
tional examples and simulations are available in
reports that are available over the Internet where the closed-loop response time
c is the tuning
[18,19]. The proposed analytic SIMC-settings are parameter. For a dominant second-order process (for
quite similar to the ‘‘simplified IMC-PID tuning which
2 > , approximately), it is recommended to add
rules’’ of Fruehauf et al. [17], which are based derivative action with
on extensive simulations and have been verified
industrially. Importantly, the proposed Series-form PID :
D ¼
2
approach is analytic, which makes it very well
suited for teaching and for gaining insight. Spe-
cifically, it gives invaluable insight into how the Note that although the same formulas are used to
controller should be retuned in response to pro- obtain Kc and
I for both PI- and PID-control, the
cess changes, like changes in the time delay or actual values will differ since the effective delay y is
gain. smaller for a second-order model (PID). The tuning
 The approach has been developed for typical parameter
c should be chosen to get the desired trade-
process control applications. Unstable processes off between fast response (small IAE) on the one side,
have not been considered, with the exception of and smooth input usage (small TV) and robustness
integrating processes. Oscillating processes (with (small Ms) on the other side. The recommended choice
complex poles or zeros) have also not been con- of
c ¼  gives robust (Ms about 1.6–1.7) and somewhat
sidered. conservative settings when compared with most other
 The effective delay  is easily obtained using the tuning rules.
proposed half rule. Since the effective delay is the
main limiting factor in terms of control perfor-
mance, its value gives invaluable insight about Acknowledgements
the inherent controllability of the process.
 From the settings in (23)–(25), a PI-controller Discussions with Professors David E. Clough, Dale
results from a first-order model, and a PIDcon- Seborg and Karl J. Astrom are gratefully acknowl-
troller from a second-order model. With the edged.
effective delay computed using the half rule in
(10) and (11), it then follows that PI-control
performance is limited by (half of) the magnitude Appendix. approximation of positive numerator time
of the second-largest time constant
2, whereas constants
PID-control performance is limited by (half of)
the magnitude of the third-largest time constant,
3. In Fig. 9 we consider four approximations of a real
 The tuning method presented in this paper starts numerator term (Ts + 1) where T > 0. In terms of the
with a transfer function model of the process. If notation used in the rules presented earlier in the paper,
such a model is not known, then it is recom- these approximations correspond to
mended to use plant data, together with a
regression package, to obtain a detailed transfer
function model, which is then subsequently ðT0 s þ 1Þ
Approximation 1 : T0 = 0 5 1
approximated as a model with effective delay ð  0 s þ 1Þ
using the proposed half-rule.
ðT0 s þ 1Þ
Approximation 2 : T0 = 0 4 1
ð  0 s þ 1Þ
7. Conclusion
ðT0 s þ 1Þ 1
A two-step procedure is proposed for deriving PID- Approximation 3 :
ð  0 s þ 1Þ ð 0  T0 Þs þ 1
settings for typical process control applications.
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 307

imation 3 is assymptotically correct (and best) at low

frequency. Approximation 4 is is asymptotically correct
at both high and low frequencies.
Furthermore, for control purposes it is most critical
to have a good approximation of the plant behavior at
about the bandwidth frequency. For our model this is
approximately at ! ¼ 1= where  is the effective delay.
From this we derive:
1. If T0 is larger than all denominator time
constant (
0 ) use Approximation 1 (this is the
only approximation that applies in this case
and it is always safe).
2. If
0 5 T0 5 5 use Approximation 2.
(Approximation 2 is ‘‘unsafe’’, but with
T0 5 5 the resulting increase in Ms with the
suggested SIMC-settings is less than about
ðTsþ1Þ
0.3).
Fig. 9. Comparison of g0 ðsÞ ¼ ð
a sþ1Þð
b sþ1Þ with
a 5 T 5
b (solid 3. If the resulting
3 ¼
0  T0 is smaller than 
line), with four approximations (dashed and dotted lines):
use Approximation 3.
g1 ðsÞ ¼ ðð
T=
bÞ
a sþ1Þ
, g2 ðsÞ= ð
T=
a
b sþ1Þ
, g3 ðsÞ ¼ ð
3 sþ1Þ1ð
b sþ1Þ with
3 ¼
a  T, and 4. If the resulting
4 is larger than  use
1
a
b
g4 ðsÞ ¼ ð
4 sþ1Þ with
4 ¼ T . Approximation 4.
The first three approximations have been the basis for
deriving the correspodning rules T1–T3 given in the
ð T 0 s þ 1Þ paper. The rules have been verified by evaluating the
Approximation 4 :
ð 0a s þ 1Þð 0b s þ 1Þ resulting control performance when using the approxi-
1 mated model to derive SIMC PID settings. Some spe-

 cific comments on the rules:
 0a  0b
sþ1  Since the loss in accuracy when using
T0
Approximation 3 instead of Approximation 4
is minor, even for cases where Approxima-
For control purposes we have that tion 4 applies, it was decided to not include
 Approximations that give a too high gain are Approximation 4 in the final rules.
‘‘safe’’ (as they will increase the resulting gain  Approximation 1,
margin)
 Approximations that give too much negative ðT0 s þ 1Þ
k
phase are ‘‘safe’’ (as they will increase the ð
0 s þ 1Þ
resulting phase margin)
where k ¼ T
00 5 1 is good for
0 5 . It may be
and by considering Fig. 9 and we have that safely applied also when
0 < , but then gives
1. Aprroximation 1 (with T05
0 ) is always safe conservative controller settings because the gain
(both in gain and phase). It is good for fre- k ¼ T0 =
0 is too high at the important frequency
quencies ! > 1=
0 : 1/. This is the reason for the two modifications
2. Approximation 2 (with T0 4
0 ) is never safe T1a and T1b to Approximation 1. For example,
(neither in gain or phase). It is good for 2sþ1
for the process g0 ðsÞ ¼ ð0:2sþ1 es , Approxima-
Þ2
! > 5=T. k
tion 1 gives 0:2sþ1 es with k ¼ T0 =
0 =10. With
3. Approximation 3 is good (and safe) for

c ¼  ¼ 1 the SIMC-rules then yield Kc=0.01
! < 1=ð
0  T0 Þ. At high frequencies it is
and
I =0.2 which gives a very sluggish reponse
unsafe in gain.
with IAE(load)=20 and Ms=1.10. With the
4. Approximation 4 is good (and safe) for
modification k ¼ T0 = ¼ 2 (Rule T1a), we get
! > 1=
4 ¼T0 =ð
0a
0b Þ. At low frequencies it
Kc=0.05 which gives IAE(load)=4.99 and
is somewhat unsafe in phase.
Ms=1.84 (which is close to the IAE-optimal PI-
settings for this process).
‘‘Good’’ here means that the resulting controller set-
tings yield acceptable performance and robustness.  The introduction of e
0 instead of
0 in Rule
Note that approximations 1 and 2 are asymptotically T3, gives a smooth transition between Rules
correct (and best) at high frequency, whereas approx- T2 and T3, and also improves the accuracy of
308 S. Skogestad / Journal of Process Control 13 (2003) 291–309

Approximation 3 for the case when
0 is we first introduce from Rule T3 the approximation
large.
 We normally select
0 ¼
0a (large), except ð0:17s þ 1Þ2 1 1
¼
when
0b is ‘‘close to T0’’. Specifically, we select ð s þ 1Þ ð1  0:17  0:17Þs þ 1 0:66s þ 1

0 ¼
0b (small) if T0 =
0b <
0a =T0 and
T0 =
0b < 1.6. The factor 1.6 is partly justified Using the half rule we may then approximate (42) as
because 8=5=1.6, and we then in some an integrating process, gðsÞ ¼ k0es =s; with
important cases get a smooth transition when
there are parameter changes in the model k0 ¼ 1;  ¼ 1 þ 0:66 þ 0:028 ¼ 1:69
g0 ðsÞ.
or as an integrating process with lag, gðsÞ ¼ kes =
sð
2 s þ 1Þ, with
Example E2. For the process
k0 ¼ 1;  ¼ 0:66=2 þ 0:028 ¼ 0:358;
ð0:3s þ 1Þð0:08s þ 1Þ
g0 ð s Þ ¼ k
2 ¼ 1 þ 0:66=2 ¼ 1:33
ð2s þ 1Þð1s þ 1Þð0:4s þ 1Þð0:2s þ 1Þð0:05s þ 1Þ3
ð41Þ
Example E13. For the process
we first introduce from Rule T3 the approximation
2s þ 1
g0 ð s Þ ¼ es ð43Þ
ð10s þ 1Þð0:5s þ 1Þ
0:08s þ 1 1
0:2s þ 1 0:12s þ 1 the effective delay is (as we will show) =1.25. We then
get e
0 =min(
0 ; 5)=min(10, 6.25)=6.25, and from
Using the half rule the process may then be approxi- Rule T3 we have
mated as a first-order delay process with
2s þ 1 ð6:25=10Þ 0:625
¼
10s þ 1 ð6:25  2Þs þ 1 4:25s þ 1
 ¼ 1=2 þ 0:4 þ 0:12 þ 30:05 þ 0:3 ¼ 1:47;
Using the half rule we then get a first-order time delay

1 ¼ 2 þ 1=2 ¼ 2:5 approximation with

or as a second-order delay process with k ¼ 0:625;  ¼ 1 þ 0:5=2 ¼ 1:25;

 ¼ 0:4=2 þ 0:12 þ 30:05 þ 0:3 ¼ 0:77;
1 ¼ 2;
1 ¼ 4:25 þ 0:5=2 ¼ 4:5

2 ¼ 1 þ 0:4=2 ¼ 1:2

Remark. We here used
0 ¼
0a ¼ 0:2 (the closest larger References

time constant) for the approximation of the zero at
T0=0.08. Actually, this is a borderline case with
[1] J.G. Ziegler, N.B. Nichols, Optimum settings for automatic con-
T0 =
0b ¼ 1:6, and we could instead have used
0 ¼
0b ¼
trollers, Trans. A.S.M.E. 64 (1942) 759–768.
0:05 (the closest smaller time constant). Approximation [2] D.E. Rivera, M. Morari, S. Skogestad, Internal model control. 4.
using Rule T1b would then give 0:08sþ1
0:05sþ1 1, but the PID controller design, Ind. Eng. Chem. Res. 25 (1) (1986) 252–265.
effect on the resulting models would be marginal: the [3] C.A. Smith, A.B. Corripio, Principles and Practice of Automatic
resulting effective time delay  would change from 1.47 Process Control, John Wiley & Sons, New York, 1985.
to 1.50 (first-order process) and from 0.77 to 0.80 (sec- [4] B.D. Tyreus, W.L. Luyben, Tuning PI controllers for integrator/
dead time processes, Ind. Eng. Chem. Res. (1992) 2628–2631.
ond-order process), whereas the time constants (
1 and [5] K.J. Astrom, T. Hagglund, PID Controllers: Theory, Design and

2 ) and gain (k) would be unchanged. Tuning, 2nd Edition, Instrument Society of America, Research
Triangle Park, 1995.
Example E6. For the process (Example 6 in [9]), [6] I.L. Chien, P.S. Fruehauf, Consider IMC tuning to improve
controller performance, Chemical Engineering Progress (1990)
33–41.
ð0:17s þ 1Þ [7] I.G. Horn, J.R. Arulandu, J. Gombas, J.G. VanAntwerp,
g0 ð s Þ ¼ ð42Þ R.D. Braatz, Improved filter design in internal model control,
sðs þ 1Þ2 ð0:028s þ 1Þ Ind. Eng. Chem. Res. 35 (10) (1996) 3437–3441.
 S. Skogestad / Journal of Process Control 13 (2003) 291–309 309

[8] S. Skogestad, I. Postlethwaite, Multivariable Feedback Control, Chemical Engineering, Norwegian University of Science and
John Wiley & Sons, Chichester, 1996. Technology, Trondheim, 2001. http://www.chemeng.ntnu.no/
[9] K.J. Astrom, H. Panagopoulos, T. Hagglund, Design of PI con- users/skoge/diplom/prosjekt01/hellem/.
trollers based on non-convex optimization, Automatica 34 (5) [16] D.W. Pessen, A new look at PID-controller tuning, Trans. ASME
(1998) 585–601. (J. of Dyn. Systems, Meas. and Control) 116 (1994) 553–557.
[10] O.J. Smith, Closer control of loops with dead time, Chem. Eng. [17] P.S. Fruehauf, I.L. Chien, M.D. Lauritsen, Simplified IMC-PID
Prog. 53 (1957) 217. tuning rules, ISA Transactions 33 (1994) 43–59.
[11] T.E. Marlin, Process Control, McGraw-Hill, New York, 1995. [18] O. Holm, A. Butler, Robustness and performance analysis of PI
[12] D.E. Seborg, T.F. Edgar, D.A. Mellichamp, Process Dynamics and PID controller tunings, Technical report, 4th year project.
and Control, John Wiley & Sons, New York, 1989. Department of Chemical Engineering, Norwegian University of
[13] T. Hagglund, K.J. Astrom. Revisiting the Ziegler-Nichols tuning Science and Technology, Trondheim, 1998. http://www.chem-
rules for PI control. Asian Journal of Control (in press). eng.ntnu.no/users/skoge/diplom/prosjekt98/holm-butler/.
[14] T.S. Schei, Automatic tuning of PID controllers based on trans- [19] S. Skogestad, Probably the best simple PID tuning rules in the
fer function estimation, Automatica 30 (12) (1994) 1983–1989. world. AIChE Annual Meeting, Reno, Nevada, November 2001
[15] S. M. Hellem, Evaluation of simple methods for tuning of PID- http://www.chemeng.ntnu.no/users/skoge/publications/2001/
controllers. Technical report, 4th year project. Department of tuningpaper-reno/.

View publication stats