You are on page 1of 17

ISA

TRANSACTIONS*
ELSEVIER

ISATransactions 33 (1994) 43-59

Simplified IMC-PID tuning rules


Paul S. F r u e h a u f

*'a,

I-Lung Chien b, Mark D. Lauritsen c

" Applied Control Engineering, P.O. Box 520, Hockessin, DE 19707, USA
Department of Chemical Engineering, Chang Gung College of Medicine and Technology, Kwetshan, Taoyuan, Taiwan, ROC
c DuPont Engineering, 140 Cypress Station Drit:e, Houston, TX 77090, USA

Abstract
We have significantly simplified the IMC-PID tuning rules. These new rules cover the vast majority of control
loops encountered in the chemical industry. This work is the result of a great deal of experience in successfully
applying IMC-PID tuning rules and an effort to prepare a training course on controller tuning. The simplified rules
are very similar in form to the classic open loop Ziegler-Nichols rules and use the process reaction curve method for
process testing. The two differences are that these rules are based on a less aggressive performance criteria and that
we adapt the rules for some commonly encountered special cases. This paper presents the relationship between the
simplified IMC-PID rules, the general IMC-PID rules, the Ziegler-Nichols rules and the Cohen-Coon rules. We
show that the simplified rules are less sensitive to parameter mis-estimation than other more aggressive tuning rules.
We also proposed rules for a fourth action; filtering. Filtering is available in digital controllers and smart field
transmitters. We report that filtering and derivative action cancel each other and therefore should not be used
together. We briefly outline the contents of the tuning course and finish the paper with an industrial example where
the simplified rules have been successfully applied.

Key words: Controller tuning; Control algorithm

1. Introduction
P I D c o n t r o l l e r s a r e still, by far, t h e most comm o n l y u s e d c o n t r o l l e r s in the c h e m i c a l industry.
P r o b a b l y t h e single b i g g e s t r e a s o n for this is that
a w e l l - d e s i g n e d a n d a d e q u a t e l y - t u n e d P I D cont r o l l e r m e e t s o r e x c e e d s most c o n t r o l objectives.
W e d e s i g n c o n t r o l s t r a t e g i e s for n e w plants,
m a k e c o n t r o l i m p r o v e m e n t s at existing p l a n t s a n d
troubleshoot control problems. These experiences
have shown us t h a t t h e r e is very l i m i t e d knowl-

* Corresponding author.

e d g e a b o u t c o n t r o l l e r t u n i n g at m a n y p l a n t sites.
W e have i n v e s t i g a t e d c o n t r o l p r o b l e m s for p r o cesses t h a t r e c e n t l y u n d e r w e n t c o n v e r s i o n s to
D i s t r i b u t e d C o n t r o l Systems ( D C S ) a n d f o u n d
t h a t a m a j o r i t y o f t h e c o n t r o l Mops w e r e b e i n g
run in m a n u a l simply b e c a u s e the loops w e r e
poorly tuned.
L a t e in 1992 we w e r e a s k e d to d e v e l o p a
c o n t r o l l e r t u n i n g c o u r s e to h e l p a d d r e s s this
p r o b l e m . T h e t a r g e t a u d i e n c e for t h e c o u r s e is
i n s t r u m e n t technicians, o p e r a t o r s a n d e n g i n e e r s .
O v e r the p a s t five y e a r s we have e m p l o y e d I n t e r nal M o d e l C o n t r o l - P r o p o r t i o n a l I n t e g r a l
D e r i v a t i v e ( I M C - P I D ) c o n t r o l l e r t u n i n g rules with

0019-0578/94/$07.00 1994 Elsevier Science B.V. All rights reserved


SSDI 0 0 1 9 - 0 5 7 8 ( 9 4 ) 0 0 0 0 4 - 6

44

P.S. Fruehauf et al. / ISA Transactions 33 (1994) 43-59

great success to many control loops. Early in the


course development we realized the general
I M C - P I D method was complicated and would
add significantly to the length of a training course.
Our experience with applying these rules showed
that they could be simplified and still be useful
for the vast majority of control loops. The majority of the controllers we tune are PI only. However, on occasion, we find benefit in using derivative action or filter action. We therefore propose
rules for all four actions.
The main contributions of the simplified IMCPID tuning rules to the science of controller
tuning are: the rules are based on rigorous IMCPID tuning and retain all the benefits of such, the
equations for calculating the tuning constants are
simple, the rules have been extensively tested on
real processes, for most situations only two process dynamic parameters need to be estimated
with only moderate accuracy and the rules work
well for nonlinear processes because they emphasize the initial dynamic response.
This p a p e r will show that the simplified IMCPID tuning rules are more robust than the rules
of Ziegler and Nichols [1] and Cohen and Coon
[2]. These older rules are less robust because they
are based on a more aggressive tuning criteria
(i.e., quarter wave decay). We, and others before
us, have found this tuning criteria to be a bit too
aggressive for most chemical industry applications. However, the contributions of these earlier
works are major, please do not misinterpret the
robustness analysis in this p a p e r to be a criticism
of these works. On the contrary, the fact that the
rigorous I M C - P I D tuning rules reduce to the
same form as the original Ziegler-Nichols open
loop rules further confirms just what a stroke of
empirical genius this work was. Our interest in
the work of Cohen and Coon came about because
their tuning rules correct for one deficiency of
the Ziegler-Nichols rules (i.e., sluggish closed
loop response in the rare occasion when the
process dead time is large relative to dominant
open loop time constant). Our simplified IMCPID rules also correct for this deficiency.
The first section of the paper reports the simplified tuning rules. The second section compares
the robustness of the simplified tuning rules to

other rules. A brief description of the tuning


method that we teach and how we apply the rules
is included in the third section and we finish the
p a p e r with an actual industrial example of the
application of our rules. Appendix A is attached
to document the derivation of the simplified rules.

2. Simplified IMC-PID tuning rules


The PID tuning rules based on the Internal
Model Control (IMC) design method were first
developed in 1986 by Rivera and co-workers [3].
Later extension to cover a wider range of process
models and an extension suitable for time constant dominant processes were presented by
Chien [4] and Chien and Fruehauf [5]. The IMCPID tuning method is very attractive to industrial
users because it has only one tuning parameter:
the closed loop speed of response. This parameter relates directly to the closed loop time constant for the response to setpoint changes and to
the robustness of the control loop. A smaller
closed loop speed of response decreases the robustness. Moreover the closed loop step load
response exhibits no oscillation or overshoot. Experience indicates that this minimizes controller
interactions and enhances overall process disturbance rejection.
When asked to prepare the controller tuning
course we began to investigate ways to simplify
the I M C - P I D tuning rules. Our objective was to
help the student tune the majority of control
loops that they would encounter. To simplify the
I M C - P I D tuning rules we used the following facts,
assumptions and conclusions based on our experience:
- The majority of processes are well approximated by dead time first order or dead time
integrator models.
- Tuning parameters do not need to be set to a
high degree of accuracy to achieve satisfactory
closed loop response. Accuracy to within 20%
is adequate.
- We have developed rules for setting the closed
loop speed of response (i.e, Tcl) so it can be
fixed. This eliminates the only tuning parameter that is required with the I M C - P I D rules.

P.S. Fruehauf et al./ ISA Transactions 33 (1994) 43-59

- ~

SloDe, R - -

/
f/

Definition

PI tuning rules
The first and most commonly used rules are

""

/
Controlled Verlable

/f

45

May or May Not


Level Out

Kc

(1)

2RL '

(2)

~'i = 5L,

I
I:

I
: ]

L, Apparent Dead Time

I
2"
,

,,,

Time
Fig. 1. The open loop step response of a typical process.

- The most popular type of PID controller is the


interacting form and the tuning settings are
gain (not proportional band) and reset in minu t e s / repeat (not repeats/minutes). If one is
working with tuning settings in alternative
units, conversion is simple.
- Open loop testing is used to determine the
dynamic parameters for a process.
The derivation of the simplified rules from the
rigorous rules using these principles is included in
Appendix A. Interestingly, the resulting rules reduce to a form that is very similar to the original
Ziegler-Nichols [1] open loop tuning rules. Our
rules are less aggressive and we modify the rules
for some commonly encountered special cases.
Before we present the tuning rules, we will
briefly present the open loop procedure for process testing. With the controller in manual and
the process as steady as possible, a step change in
controller output is made. The process response
will look something like Fig. 1.
If the process levels out, or would if enough
time were allowed, it is called dead time first
order. If it would not, it is called dead time
integrating. The tuning rules generally apply to
both types of processes, though modifications to
the rules are needed if the process is dead time
first order and the time constant is short relative
to the apparent dead time or when the apparent
dead time is small (i.e., L < 0.5 minutes). As a
result, we have developed three sets of rules.

where
Kc is the controller proportional gain (dimensionless),
~i is the controller reset time (minutes/repeat),
R
is the change per minute in the process
variable, (expressed as a % of the transmitter span), divided by the step change magnitude (expressed as a % of the controller
output span) (1/minutes),
L
is the apparent dead time (minutes)
These rules are derived from the IMC-PID
rules for a dead time integrating process. However, as shown in [5], these rules apply to both
dead time integrating and dead time first order
processes.
There are two types of less commonly observed responses which require modifications to
the above rules. The first is when the process is
dead time first order and the time constant is
short relative to the apparent dead time. Fig. 2
illustrates this type of response.
We have found, by performing many simulations, that the following rules give better closedloop response when the time constant divided by

Slope, R -,~,/
Final Response

63% of Final Response

_Z_~

It: :I

L, Apparent DeadTime

I
Time
Fig. 2. Open loop step response of a self regulating process.

46

P.S. Fruehauf et a l . / ISA Transactions 33 (1994) 43-59

the apparent dead time is 3 or less. The modified


tuning rules are
1
Kc = 2 R L '

(3)

Ti =

(4)

become more conservative, thus enhancing the


stability of the closed loop system. Please note
that this last rule does not apply to flow loops.
In summary, the simplified PI tuning rules are
"t

'T,

where r is the process time constant (minutes).


Notice that the only difference between this
modification and the Eqs. (1) and (2) rules is that
the reset time is, at times, significantly shorter
which speeds up the closed-loop response. This
rule is derived from the rigorous IMC-PID rules
for dead time first order processes. Often when
doing open loop testing it is difficult to observe
the open loop time constant because of noise and
other process variables that are changing that
effect the dynamic response (i.e., the process
never steadies out). If you cannot observe the
time constant just use the first of set rules. On
the occasion when you can observe the first order
time constant and the ratio is small you can take
advantage of the performance improvement that
these modifications to the rules provides.
The other special situation which requires tuning rule modification is when the apparent dead
time is very small. In this case, the tuning rules in
Eqs. (1) and (2) may result in excessive control
action thus causing closed loop stability problems
when even modest model mismatch is present. In
addition, Eq. (1) can result in a very high controller gain that will produce large changes in
controller output for small changes in setpoint. If
the controller output is tied to a valve that adjusts
the feed rate to a downstream unit operation this
can cause large flow disturbances to the downstream unit. Also, when the apparent dead time
is small the process is easy to control, so very
tight tuning is not generally necessary. In this
small dead time case, typically when L < 0.5 minutes, the tuning rules are modified to
1

Kc = ~,

(5)

r i = 4.

(6)

Notice that K c does not have L in the denominator and thus is smaller and '/'i is larger compared to Eqs. (1) and (2). The controller settings

"g

-->3
L
1

Kc
ri

2RL
5L

--<3
L
1

2RL
r

L<0.5
1

R
4

The above tuning rules work very well for the


vast majority of the control loops encountered in
the chemical industry. For specific control loops
like flow and level we provide other recommendations later in this paper. For rare occasions
when the open loop step response exhibits overshoot, oscillatory, or inverse behavior, we recommend modeling the process dynamics as a more
complicated second order response and refer to
Table 1 in [5] for controller settings.
2.1. F i l t e r a n d d e r i v a t i v e t u n i n g rules

Before we discuss the filter and derivative tuning rules it will be helpful to discuss filter action,
derivative action and the relationship between
the two.
Filter action

We use signal filtering in some loops to improve performance. One common example is in
level loops where the measurement is often noisy.
Without filter action the noise is passed by the
proportional action of the controller directly to
the control valve causing it to move unnecessarily.
Reducing this movement by decreasing the controller gain is undesirable because of the destabilizing effect this has on the level control
loop. Filter action is a superior solution. We are
not aware of any rules on how to set filter time
constant. We believe that filtering is now used
more often because it is available in many digital
controllers and smart field instruments. It has
become a fourth controller action.
The task of setting the filter time is one of
using as much as you dare without degrading the
performance of the loop. The goal is to specify a

P.S. Fruehauf et al. / ISA Transactions 33 (1994) 43-59

filter time constant that provides the benefits of


filtering, without adversely impacting closed loop
response.
We used time domain and frequency analysis
to develop a general rule. A dead time first order
process and a dead time integrating process were
simulated, with the process feedback signal being
filtered in each case. The tuning rules were used
to provide the PI controller settings. Fig. 3 illustrates one of the plots used for this analysis. It
shows the time domain responses for a dead time

47

first order process subjected to a step change in


setpoint. A filter time constant less than or equal
to 50% of the apparent dead time was selected as
the general rule because it provides the desired
filtering action without severely impacting loop
performance.
Derivative action

Derivative action can improve the performance of a control loop by shortening the closed
loop natural period. The performance improve-

Response to Step Change in Setpoint


Deadtime First-Order Lag Process

1.2

Tf

.T

2*L

e
a
s
u
r
e
d

4*L

ffi

0.8

0.6

0.5*L

= 0.2*L

0.4

0.2

10

20

30

40

50

60

70

80

90

100

~me
Fig. 3. S e t p o i n t r e s p o n s e s ~ r d i f f e r e n t v a l u e s o f f i l t e r t i m e c o n s t a n t .

48

P.S. Fruehauf et al./1SA Transactions 33 (1994) 43-59

ment is not dramatic; however, in some situations


it can help enough to warrant using it. We have
used the rigorous I M C - P I D tuning rules to derive
rules for setting derivative time which say to
make it equal to one half the apparent dead time.
In order to limit the number of rules, we elected
not to incorporate the slight gain increase and
integral decrease given by the rigorous rules.
These changes are small relative to the required
accuracy for setting these parameters.

Relationship between filter and derivative action


Filter and derivative action are essentially the
mathematical inverse of each other. They are in
series in the feedback circuit so they cancel each
other. Therefore, we propose rules that specify
one or the other but not both.
Derivative action causes the controller output
to bounce more than when it is not used, but
improves control performance. Filter action
dampens this bouncing but degrades performance. Often the output of one loop is a disturbance to another loop. Derivative action could
improve one loop's performance but could hurt
another. Filter action does the opposite. The
decision to use filter or derivative action becomes
a compromise between performance of an individual loop versus the system as a whole. Often
the best compromise is when neither is used. This
is one reason why we feel the PI only controller is
so popular.
In summary, the rules for setting filter or
derivative time are
Tf : 0.5L or "/'d= 0.5L,
where
% is the filter time constant (minutes),
~'d is the derivative time (minutes).

robustness of the simplified I M C - P I D rules to


other tuning rules in two different ways.
The first section uses time response plots to
show the effect of model mismatch for PI tuning
rules. There is a model mismatch when the process model parameters change from what they
originally were. This mimics what can happen
when the operating conditions of a chemical process are changed, because almost all processes
are non-linear to some degree. When we tune we
approximate the dynamics at a particular operating condition with a linear model.
The second section compares PID tuning rules
using stability plots. This comparison is also different from the first, in that we are looking at
mismatch between what the actual process dynamics are and what we estimate them to be.
This measure of robustness tells us how accurately we must estimate the process dynamics
when doing an open loop step test to prevent
unstable closed loop response.

PI tuning rule sensitivity to model mismatch


In this section, the simplified I M C - P I D tuning
rules are demonstrated using three numerical examples with different process dynamic characteristics. These tuning rules will be compared to the
popular tuning methods of Ziegler and Nichols
[1] and Cohen and Coon [2].
Fig. 4 shows the feedback control structure for
all the simulations in this section.
y is the controlled variable, u is the controller
output, Ys is the setpoint, d is the disturbance,
G c is the feedback controller (PI), Gp is the
process transfer function, G L is the disturbance
transfer function.

3. Robustness

Robustness is the term used to describe the


sensitivity of closed loop response to changes. A
set of tuning rules are more robust if they can
tolerate larger changes in the dynamics parameters (e.g., L and R) and still maintain adequate
response. The next two sections compare the

YS

~_*

Fig. 4. Feedback control structure used for simulations illustrated in Figs. 5-10.

P.S. Fruehauf et al./ ISA Transactions 33 (1994) 43-59

Example 1. Time-constant-dominant process:

sponse for both controlled variable and controller


output. On the contrary, both Ziegler-Nichols
and C o h e n - C o o n methods result in overshoot
and an oscillatory closed-loop response.
In order to test the robustness of the proposed
tuning rules, + 50% model mismatches in process
gain (related to R) and dead time were simulated. Fig. 6 illustrates the effect of the process
gain mismatch.
The figure clearly shows the superior closedloop behavior of the simplified IMC-PID settings
under severe model mismatch conditions.

10e-2S

Gp=G/= 5 0 0 s + l "
From the step response, R, L and T are
calculated to be R = 0 . 0 2 , L = 2 and T = 5 0 0 .
The resulting PI controller settings are

Kc
ri

Simplified
IMC-PID

Ziegler-Nichols

Cohen-Coon

12.5
10

22.5
6.7

22.5
6.6

Fig. 5 shows the disturbance closed-loop response for the simplified IMC-PID, ZieglerNichols and C o h e n - C o o n rules.
In the simulation, d is changed from 0 to 1 at
time equal to zero. Notice that the proposed
tuning rules give very smooth closed loop re-

0.08

49

Example 2. Process with short time constant and


significant dead time:
e-4S
Gp=Gt-

2s+l"

IMC-PID

0.06
0.04,

/
/

0.02

Both Z-N
and C-C

\\\ , ~"

0
-0.02

10

30

20
Time

4,0

50

(minutes)

I
.I
""0":]

_.. IMC-PID

\
'X

",

-is

/f

"

//
/

-20

"-

Bolh Z-N

"

"

i0

"

"

"

"

20
Time

30
(minutes)

Fig. 5. Load response for ExampLe 1.

"

40

5O

P.S. Fruehauf et al. / ISA Transactions 33 (1994) 43-59

50

IMC-PID

o--1 /

Both Z-N
and C-C

',,

/ ',
",,l

-0.05

-0.I

ii

'

"

'

"

IMC-PID,..,

/"

/*%,
/
/

',\

"

"

\
"

\
\
\
\

'

/
1
I
II

'

"

"

4,0

50

(minutes)

Both Z-N
and C-C.

// %

/,~,
I

0 =

-10

"

30

Time

i0--

'

20

10

/
/
/7

~'

-20"

-30

'\\/

\\ /I'
I

\\

10

20

T1hae

\\

"

\
"

30

"

40

I
50

(minutes)

Fig. 6. Load response for Example 1 for a + 50% gain mismatch.

For this example, the controller settings are

K,
~'i

Simplified
IMC-PID

Ziegler-Nichols

Cohen-Coon

0.25
2

0.45
13.3

0.45
2.9

Fig. 7 shows the disturbance rejection closedloop responses using the three tuning methods.
Notice again that the proposed tuning rules
give very smooth closed loop response for both
controlled variable and controller output, while
Ziegler-Nichols settings give very sluggish response. C o h e n - C o o n settings provides improvement over Ziegler-Nichols settings in this situation but are still a little oscillatory. Fig. 8 shows
the closed loop responses for a + 5 0 % model
mismatch in dead time. The simplified tuning
rules still provide smooth closed-loop response
while Ziegler-Nichols settings are still very slug-

gish and C o h e n - C o o n settings are more oscillatory.


Example 3. Process with small dead time:
10e-0.16s
Gp = G /

15s + 1

In this example, the process dead time is very


small (i.e., about 10 seconds). The controller settings for the three tuning methods are

K,,
ri

Simplified
IMC-PID

Ziegler-Nichols

Cohen-Coon

1.5
4

8.43
0.53

8.44
0.52

Fig. 9 shows the closed-loop responses for the


three tuning methods.
Notice that Ziegler-Nichols and C o h e n - C o o n
settings result in an almost identical response.

P.S. Fruehauf et a l . / ISA Transactions 33 (1994) 43-59

the closed loop response becomes unstable. We


estimate these parameters as part of the open
loop tuning procedure we use. A set of tuning
rules are more robust if larger errors in the
estimates can be tolerated.
These plots were developed because we wanted
a general way to assess the robustness of the
simplified IMC-PID rules before we included
them in our tuning course. The time response
plots in the previous section only test the rules
for specific values of mismatch. The plots in this
section illustrate the robustness for all values of
mismatch of interest.
Tuning rule robustness plots have been prepared for the three example processes presented
in the previous section. Fig. 11 is the plot for the
example one process. The X-axis is the ratio of
the estimated dead time (Lo) to the actual dead
time (La). The Y-axis is the ratio of the estimated

Although their responses are much faster than


the proposed tuning method, the controller output is rather oscillatory. Fig. 10 shows the response for a + 5 0 % model mismatch in dead
time. We should mention that this kind of dead
time mismatch (i.e., process having dead time of
15 seconds versus model having dead time of 10
seconds), is reasonable and can be expected to
happen from time to time. With this small model
mismatch the Ziegler-Nichols and C o h e n - C o o n
settings become unstable, while the simplified
IMC-PID rules still provides smooth and stable
closed-loop response.

Tuning rule stability plots


In this section, tuning rule stability plots are
used to compare how much we can misestimate
the process dead time and process slope before

51

IMC-PID ~ Z-N

0.5"

::1
.

"

"

"

10

'

'

30

20

Time

"

50

40

(minutes)

O r

-0.5

-1

-1.5

-2

10

"

"

"

20

Time

30

(minutes)

Fig. 7. L o a d r e s p o n s e for E x a m p l e 2.

"

4o

50

52

P.S. Fruehauf et aL / ISA Transactions 33 (1994) 43-59

process slope (Ro) to the actual process slope


(Ra).
The curves are boundaries between the regions of stable and unstable estimates. The stable
regions are above and to the right of the curves
and the unstable regions are below and to the left
of the curves. The curves are defined by the
equations for the Bode stability criteria (i.e., magnitude ratio of unity at the resonant frequency)
using the different tuning rules.
These plots can be used to compare the robustness of tuning rules in the following way.
Assume that we have just done an open loop step
test on a process and the response looks something like Fig. 1. The process dead time is misestimated, due to signal noise, to be 1.5 minutes.
Assume that the actual dead time is 2 minutes.
This defines a line perpendicular to the X-axis at
0.75. This line intersects the I M C - P I D rule curve

at 0.4 and both the Ziegler-Nichols and C o h e n Coon rule curves at 1.25.
If the true process slope is 1, the graph tells us
that if we mis-estimate the process slope to be
less than 0.4 the I M C - P I D rules will result in
unstable tuning, whereas the other rules will only
allow a mis-estimation of less than 1.25 before
unstable tuning results. For the Ziegler-Nichols
and C o h e n - C o o n rules if we estimate the process
slope to be exactly what it is we will calculate
tuning settings that produce unstable response!
We conclude that the I M C - P I D rules are more
robust because they allow for a larger mis-estimation of the slope than the other rules. More
generally, the farther the stability boundary line
is from the point (1, 1) (i.e., the point which
corresponds to exact estimates of dead time and
slope) the more robust the tuning rules are.
Fig. 12 is the robustness plot for the Example

-_

IM@PID
'

'

"

"

"

10

'

20

Time

0 ~

-0.5

r.

"

. "

"

"

30
{ m i n u t e

"

'

'

40

"

50

B }

Z-N

.1

-1.5

"

10

"

"

20
T i m e

"

30
( m i n u t

e B )

Fig. 8. Load response for Example 2 with a + 50% dead time mismatch.

40

"

"

5O

P.S. Fruehauf et a l . / ISA Transactions 33 (1994) 43-59

2 process where the process time constant is short


relative to the dead time. The Ziegler-Nichols
and C o h e n - C o o n rules are more robust for this
case but are still significantly less robust than the
simplified IMC-PID rules.
For brevity the Example 3 process plot is not
shown. The simplified IMC-PID rules are intentionally conservative for this type of process and,
therefore, are much more robust than the other
rules.

For general loops, we advocate that open loop


testing be used. This is essentially the reaction
curve testing method proposed by Ziegler and
Nichols. We teach general loop tuning as a three
step process. The steps are (1) to learn to run the
process with controller in manual and estimate L
and R, (2) to perform open loop step tests to
accurately determine L, R and r (if it is possible
to observe r), and to calculate tuning and (3) to
test new tuning with the loop in automatic by
making setpoint changes. The first step is critical
and far from trivial; however, most tuning discussions we have seen do not even address this step.
This method is used for general control loops,
e.g. temperature, pressure, composition loops, etc.
For special classes of control loops like flow and
level, we provide the following controller tuning
recommendations.

4. Tuning method taught in training course


The tuning method we teach breaks loops up
into three general categories: general loops, flow
loops and level loops where averaging level control is desired.

f
0.05

0.03

"

53

IMC-PID

0.01

-0.01

10

20

Time

30

(minutes)

O~

-0.5"

j[~V

B o t h Z-N
and
C-C

-1-

-1.5

-,2

! IMC_PID?
I

"

"

"

10

20

Time

(minut

era)

Fig. 9. Load response for Example 3.

"

"

30

54

P.S. Fruehauf et aL / ISA Transactions 33 (1994) 43-59

For flow loops we teach that the dynamics are


generally the same for all loops and therefore so
is the tuning. We recommend that the gain be set
between 0.5 to 0.7 and that the reset be set
between 0.2 and 0.3 minutes. Fine tuning can
then be done by a closed loop trial and error
method.
Level loops are generally tuned so the tank
will absorb flow disturbances so that they are not
passed on to downstream unit operations. When
this is the case, we recommend that the following
equations be used:
100% 2
Kc
ALR
'
4V
zi

Kc Q '

where A L R is the allowable level range (%), V is

IMC-PID

the measured tank volume (gallons) and Q is the


maximum flow through level control valve (gallons/minute).
The allowable level range is the range of level
that can be allowed to vary without any adverse
effects. For a surge tank, the range is often selected to be 80%. This type of level controller
tuning is referred to as averaging level control. In
the rare case when tight level control is required
the general loop method should be used.

5. Industrial

example

We have successfully applied these simplified


I M C - P I D tuning rules to many different processes. As one example, we have applied these
rules to a new 'paper' drying oven. The 'paper' is

Bolh Z-N
and C-C

0.07

i
I
I

0.05

It

0.03

t
t

0.01

-0.01

I
10

20
Time

1--

IMC-PID
I1
It
t
I
lt Jl
F /[

t|
30

(minutes)

,,f

Bo'lh Z-N

m'KI C4

II
I I I
I I I I

-1-

-2

I
I

[{
II

I ] I
I I I

t I

f I

I
lO

20

Time
(m Inut
ee )
Fig. 10. L o a d response for E x a m p | e 3 for + 5 0 % dead time mismatch.

30

P.S. Fruehauf et al. / ISA Transactions 33 (1994) 43-59

made from a polymer. Multiple or single rolls of


paper are heated using radiant heaters and then
pressed to produce the product which is used in
high performance composite materials. Many different gauge products are made with throughput
rates varying substantially. This created some
concern that the controller tuning might need to
be changed for different products. The oven is
divided into multiple heating zones. Infrared
temperature sensors are used to measure the

55

paper temperature at various locations in the


oven. PID controllers are used to control these
temperatures by manipulating the upstream
heater intensity. A complex control strategy, that
will not be discussed, sets the setpoints of these
controllers. The process is controlled with a Honeywell TDC 3000 DCS.
Shortly after startup, one of the authors was
asked to provide assistance in tuning these controllers. The process was not making quality

Example #1
Tp = 500 min, La = 2 min, Kp = 1

1.8

1.6

1.4

1,2
\\\x \

0.8

0.6

0.4

0.2

1.4

1.6

1.8

I
0.2

0.4

0.6

0.8

1.2
Lo/La

IMC Rules

-- Z-N Rules

Cohen-Coon I

Fig. 11. Stability plot for Example 1.

56

P.S. Fruehauf et al. /1SA Transactions 33 (1994) 43-59

product because there was too much variability in


the p a p e r temperatures.
One problem with the tuning was that both
filter and derivative action were being used. In
addition, they had the same tuning settings causing them to completely cancel each other. In this
particular case we judged that filtering was more
helpful because of the large amount of measurement noise. The derivative action was removed by
setting the tuning value to zero.
We applied open loop tests to each and every

loop following the procedure already outlined.


From these tests we estimated the dead time and
process slope and calculated tuning parameters.
Fig. 13 illustrates the result of one such test.
Originally all the loops were not tested because it was thought that the untested loops
would have similar responses. It turned out that
they did not. This is a common pitfall in tuning.
Some of the loops in the middle of the oven had
much larger apparent dead times than the other
loops, which resulted in significantly slower tunExample #2

T p = 2 m i n , La = 4 m i n , Kp = 1

1.8

1.6

1.4

1.2

0.8

\
0.6
N

\~

0.4

0.2

0.2

0.4

0.6

0.8

1.2

1.4

1.6

Lo/La
- - IMC Rules

- - Z-N Rules

Cohen-Coon I

Fig. 12. Stability plot for Example 2.

1.8

57

P.S. Fruehauf et al. / ISA Transactions 33 (1994) 43-59

LOOP T U N I N G

ADO PU, SP
fiND OP TRgCES
OELETE RLL
TRACES

Ip

ROD

fiRggETER.P~

I SELECT

TINEDASE]
1247TC

DELETE |
GREEN TRgCE/

I SELECT '
ADD
[DATA SOURCE] I C X A N G E
ZONE]
RT RT RT

MIN

lOg.OR

PU 175.n t
OP l g g . $ $
PU 68.5Z

J
ggagigl

171.34

OEG C

PU 145.$g

OP

$.gg$

SP 158.77
OP 4g.ggg

L247TC
g.g TO 255.g OEG C
CZ ZONE 2 SECT C TERP

g .gga

INTEGRRL iIOERIUgTIOE

lilllillil

T3 ' g i l l '

]1

KPP l i i l i l

TS e g g l g l

KNL l i i g i l

TO Igggll

6.$$
PU
OP

Q.85

liillill

171.3 ; m i n i m
4O.O , .~=i i

~ g
, ,

FIT RESPONSE WITH TWO STRAIGHT LINE


Fig. 13. Open loop step response test for the industrial application example.

ing. The original tuning for these loops, which


was extrapolated from tests on other similar loops,
produced very oscillatory response. The important lesson to remember is to test each loop. We
have seen many cases where 'identical' loops had
very different dynamic responses.
Once the testing was complete the machine
started making quality product. The tuning required about 12 hours. They had been making
bad product for two weeks costing approximately
$10 K / d a y . Interestingly, the tuning was found
be good for all types of products. We speculate
that this is partially due to the added robustness
of these rules.

especially for his historical perspective on this


subject. The Straight-Line Control Co., Inc. has
recently published a booklet on controller tuning
and control loop performance [6] that was influenced by this work. The booklet is highly recommended.

Conclusions
We have greatly simplified the IMC-PID tuning rules. These rules cover most loops encountered in the chemical industry. The tuning rules
are
For general loops:
"g

We would like to acknowledge the help of


Dave W. St. Clair, formerly of DuPont, and now
with Straight-Line Control Co., Inc., for his help
with developing the simplified tuning rules and

-->3
L

Acknowledgments
Kc
ri

,rf

2RL

2RL

5L
=

--<3
L

0.5L or ~'d = 0.5L.

L<0.5
1
4

P.S. F r u e h a u f et a l . / ISA Transactions 33 (1994) 4 3 - 5 9

58

For flow loops:


K c

= 0.5 to 0.7,

From numerous simulations and industrial application, we find setting rcj = 2 L provides very
good closed-loop performance, the resulting Kc
and
T i are

ri = 0.3 to 0.2.

For level loops:


100% 2
K c -

ALR

4 V
,

Ti

5
9RL ' which we elected

K~

K c Q "

We have shown that the simplified I M C - P I D


tuning rules are more robust then the rules of
Ziegler-Nichols and C o h e n - C o o n .
We have shown that filter and derivative action cancel each other and should not be used
together and we have proposed rules for setting
these actions.
We have developed rules for setting filter action so that maximum filtering can be obtained
without adversely affecting closed loop performance.
We have shown an example where the rules
have been successfully applied to an industrial
process.

1
to simplify to K~ - 2RL

(A.3)
(A.4)

and r i = 5 L .
Tuning rule when r / L < 3

When self regulating process behavior is observed and the dead time is relatively large compared to the process time constant, the above
tuning rules will be too conservative. A better set
of tuning rules is to tune the control loop based
on first order plus dead time response. From [5],
the PI tuning rules for process model of first
order plus dead time are
Kp

Appendix A. Derivation of the simplified tuning


rules from the rigorous IMC-PID rules

for a process model of K c

(rs +

Ls

1) '

(A.5)

K c -

Kv(rcl + L)

'

Tuning rule when r / L > 3

(A.6)

Ti :T.

From our experience, a large portion of Chemical Plant control loops are time constant dominant processes, thus the initial portion of the
dynamic step response can be approximated as a
system with integrator. The importance of the
initial response for controller tuning purposes has
been pointed out by Chien and Fruehauf [5], thus
controller tuning based on a system with integrator can provide excellent closed loop performance. The original tuning rules for a system
with integrator and dead time are
Re -

Ls

for process model of - S

Kc

2rcl + L
R(.rcl + L)2 ,

r i = 2r d +L.

(A.1)
(A.2).

It can easily be shown that R = K p / T


choosing rot = L from experience, then

1
2RL

Kc
r i = r.

and by

(A.7)
(A.8)

Tuning rule when L < 0.5

For a process with negligible dead time, typically when L < 0.5 minute, by choosing rc~ = 2
and L = 0 in Eqs. (A.1) and (A.2), we enhance
the closed-loop stability. The resulting K c and rg
are

1
K c = ~,
"J'i =

4.

(A.9)
(A.10)

P.S. Fruehauf et aL /1SA Transactions 33 (1994) 43-59

References
[1] J.G. Ziegler and N.B. Nichols, "Optimum settings for
automatic controllers", ASME Trans. 64 (1942) 759.
[2] G.H. Cohen and G.A. Coon, "Theoretical consideration
of retarded control", ASME Trans. 75 (1953) 827.
[3] D.E. Rivera, M. Morari and S. Skogestad, "Internal model
control, 4. PID Controller Design", Ind. Eng. Chem. Proc.
Des. Dev. 25 (1986) 252.

59

[4] I-L. Chien, "IMC-PID controller design - An extension",


IFAC Proc. on Adaptive Control of Chemical Processes 147,
Copenhagen, Denmark, August 1988.
[5] I-L. Chien and P.S. Fruehauf, "Consider IMC tuning to
improve controller performance", Chemical Engineering
Progress 33 (October 1990).
[6] D.W. St. Clair, Controller Tuning and Control Loop Performance, 2nd Edition, (1993) by the Straight-Line Control
Co., Inc., 3 Bridle Brook Lane, Newark, DE 19711-2003.

You might also like