Practical Guidelines for Tuning PID Controllers by Using MOMI Method

Damir Vrančić and Stanko Strmčnik II. THE MOMI PID CONTROLLER TUNING METHOD
J. Stefan Institute,
Jamova 39, SI-1001 Ljubljana, Slovenia A magnitude optimum multiple integration (MOMI)
[email protected], [email protected] tuning method is based on a magnitude optimum (MO)
frequency criterion which makes the frequency response
from set-point to plant output as close to one as possible
Abstract – The magnitude optimum multiple integration for low frequencies.
(MOMI) tuning method for the PID controllers provides non- If GCL(s) is the closed-loop transfer function from the
oscillatory closed-loop response for a large class of process set-point (w) to the process output (y), the controller is
models. However, one must account for certain additional determined in such a way that
obstacles that have to be overcome to enable application of the
method in practice. A few practical guidelines for performing GCL (0) = 1
multiple integrations (MI) from the process step response in
practice, for re-tuning controller parameters, and for ∂ n GCL ( jω ) (1)
=0
calculating the parameters of the two-degrees-of-freedom ∂ω n
controller in order to improve disturbance rejection ω =0

performance, are given. for as many n as possible (Åström and Hägglund, 1995).
Such criterion results in a fast and non-oscillatory
closed-loop time response for a large class of process
I. INTRODUCTION models.
In order for the MO method to be applied by using the
The Ziegler-Nichols tuning rules (Ziegler and Nichols, following PID controller transfer function:
1942) were the very first tuning rules for PID controllers,
and it is surprising that they are still widely used today.  
U (s ) 1 sTd
Their popularity lies in their simplicity and efficiency. This GC (s ) = = K 1 + + , (2)
is why so many different tuning rules which are based on E (s )  sTi 1 + sT f 
 
the same tuning procedures have subsequently been
developed (Gorez, 1997). where u is the controller output and e is the controller error
Following the work of Ziegler and Nichols, a variety of (e=w-y), and the following process transfer function:
PID tuning methods have been developed. In general, these
Y (s ) 1 + b1 s + b2 s 2 +  + bm s m
methods can be divided into two main groups: the direct G P (s ) = = K PR e − sTdel , (3)
and the indirect tuning methods (Åström et al., 1993; U (s ) 1 + a1 s + a2 s 2 +  + an s n
Gorez, 1997).
The direct tuning methods do not require a process an explicit identification of the parameters KPR, a1, a2, a3,
model, while the indirect methods calculate controller a4, a5, b1, b2, b3, b4, b5, and Tdel is required (Vrančić, 1997;
parameters from an identified model of the process. Vrančić et al., 1997, Vrančić et al., 1998b). However, it is
Recently, a new indirect tuning method which is based well known that accurately estimating such a number of
on an implicit process model was developed (Vrančić et parameters from real measurements could prove to be very
al., 1996). The multiple integrations (MI) method (Rake, problematic.
1987; Strejc, 1960) is used for the implicit process However, this problem can be avoided by using the
identification. However, the areas, calculated by using the concept of multiple integrations (Rake, 1987; Strejc,
multiple integrations from the open-loop process response, 1960). Following Rake, (1987), the following areas can be
are directly used for the calculation of the controller expressed by integrating the process open-loop step
parameters rather than for the process identification in response (y(t)), after applying the step-change ∆U at the
order to meet the so-called magnitude optimum (MO) process input at t=0:
criterion (Åström and Hägglund, 1995; Hanus, 1975;
Kessler, 1955). It was found out that in this way, by using
the so-called magnitude optimum multiple integration A1 = y1 (∞ )
(MOMI) method, the magnitude optimum criterion can be  , (4)
met for a very large set of process models (low-order, high- Ak = y k (∞)
order, highly non-minimum phase and/or processes with
larger time delays) merely by measuring the process open- where
loop step response without the need for additional “fine”
y (t ) − y (0)
tuning. The excellent tuning results were also achieved on y 0 (t ) =
several laboratory set-ups (Vrančić, 1997; Vrančić et al., ∆U
1997; Vrančić et al., 1998b). t

However, one must account for certain additional y1 (t ) = ∫ [ y 0 (∞ ) − y 0 (τ )]dτ

. (5)
obstacles that have to be overcome to enable application of 0

the method in practice or to improve disturbance rejection 

t
for some certain processes. Such problems and y k (t ) = ∫ [Ak −1 − y k −1 (τ )]dτ
corresponding solutions are closely studied in this paper. 0
 In order to meet quite demanding MO frequency However, relatively small errors in calculating the
criterion (1), the PID controller parameters can be process steady-state gain (KPR) could lead to relatively
calculated in the following way (Vrančić, 1997; Vrančić et large errors in the calculated areas. Such errors are
al., 1997, Vrančić et al., 1998b): especially noticeable when dealing with a process
corrupted by noise. To improve the accuracy of the
A3 A4 − A2 A5 calculated KPR, the process step response should be
Td = 2
(6)
A3 − A1 A5 averaged in time intervals t=[t0, t1] (before making step
change) and t=[tint, tfin] (after the new steady-state was
A3 already been achieved) in the following way (see Fig. 1):
K= (7)
(
2 A1 A2 − A3 K PR − Td A1
2
) y a 0 = y (t ); t = [t 0 , t1 ]
. (9)
Ti =
A3
(8)
[
y a1 = y (t ); t = tint , t fin ]
A2 − Td A1 The process steady-state gain is then simply calculated
as:
Note that the PI controller parameters can be expressed
from (7) and (8) simply by applying Td=0. y a1 − y a 0
Also note that Equations (6) to (8) hold when the filter K PR = . (10)
time constant is fixed to Tf=0. However, choosing e.g. ∆U
Tf=Td/10 still does not seriously affect the result of the Note that y(0) in (5) should be replaced by ya0.
calculation of the PID controller parameters (see Vrančić, How are the time instants t0 and tfin chosen? Numerous
1997, and Vrančić at al, 1998a, Vrančić et al., 1998b). experiments on several process models and laboratory
plants showed that good practical results are usually
III. GUIDELINES FOR PRACTICAL WORK obtained when choosing:

The previous section showed that the implementation of t1 − t 0 = 0.1l0.3 ⋅ (tint − t1 )

. (11)
the magnitude optimum multiple integrations (MOMI) t fin − t int = 0.1l0.3 ⋅ (t int − t1 )
method is very simple and straightforward. Only the
process step response has to be measured, and some
integrations (summations) to be performed in order to The proposed integration procedure will now be
calculate areas A1 to A5 (A1 to A3 for PI controller). illustrate using an example.
However, there are always some additional obstacles that The following process model was chosen:
have to be overcome in order to be able application of the
method in practice. In this section, a few practical 1
GP ( s ) = . (12)
guidelines for deriving areas from the process step (1 + 4s )3
response will be given, as well as some modifications of
the tuning procedure where the calculated controller gain is A random noise, generated by the MATLAB function
too high, or even negative, or where using a two-degrees- RANDN, and amplified by factor 0.05, was added to the
of-freedom controller. process step response. The process output and input signals
are shown in Fig. 2. The following time intervals were
A. Performing multiple integrations in practice chosen: t0=0s, t1=10s, tint=50s, and tfin=60s. Values ya0 and
ya1 were calculated by averaging process output signal
Areas A1 to A5 can be calculated from the final values during intervals t=[t0, t1] and t=[tint, tfin] (9) which resulted
(t=∞) of signals y1(t) to y5(t) (4). Of course, in practice it is in ya0=-6.97⋅10-4, and ya1=0.996. Using (10), the calculated
sufficient to wait until process step response settles. Fig. 1 process gain was KPR=0.997. Functions y1(t) to y5(t) were
shows a typical process step response. At t=t1, a step- calculated from (5), where integrations were performed in
change is applied to the process input. The process the time interval t=[t1, tint]. Areas A1 to A5 were calculated
practically reaches the steady-state value at t=tint, so all the from y1(tint) to y5(tint). The following values of the areas and
integrations in (5) can be made in the time interval t=[t1, controller parameters were obtained:
tint]. process : K PR = 0.997, A1 = 11.87, A2 = 93.47,
A3 = 604.1, A4 = 3433, A5 = 1.762 ⋅ 10 4
. (13)
PI : K = 0.595, Ti = 6.46
PID : K = 2.50, Ti = 9.92, Td = 2.74

The ideal values, obtained on the process without noise

present, were the following:
process : K PR = 1, A1 = 12, A2 = 96,
A3 = 640, A4 = 3840, A5 = 2.15 ⋅ 10 4
. (14)
PI : K = 0.625, Ti = 6.67
PID : K = 2.31, Ti = 9.87, Td = 2.59
Fig. 1. Process input and output during step-change experiment.
 It is clear that the obtained controller parameters (13) are
close to the ideal ones (14). 1
K≤ . (19)
2 A1 A2
− 2 K PR
Process output
A3
1
When limiting the controller gain of the PI controller, of
course, only (15) is used. Note that the proposed re-tuning
0.5
of controller parameters can also be used in cases when
slower and more robust controller should be designed (by
0
decreasing gain K), or if a faster, but more oscillatory,
0 10 20 30 40 50 60 70 80
response is required (by increasing gain K).
Time [s] The proposed modified tuning procedure will now be
Process input
illustrated.
1 The following process model was chosen:
0.8
2
0.6 GP ( s ) = . (20)
0.4 (1 + 5s )(1 + s )
0.2
0 The multiple integrations were performed on the process
−0.2 step response (y), and the following values of the process
0 10 20 30 40 50 60 70 80
Time [s] steady-state gain and areas were obtained from (4) and (5):

Fig. 2. Process output (y) and controller output (u) during the open-loop K PR = 2, A1 = 12, A2 = 62, A3 = 312,
experiment on the process with present noise. . (21)
A4 = 1562, A5 = 7812
B. Re-tuning the controller parameters In the next step, the PI and PID controller parameters
were calculated from (6) to (8):
In some cases, the controller parameters, obtained by
using the MOMI method, have to be re-tuned due to some PI : K = 1.3, Ti = 5.03s
. (22)
practical reasons, namely that when tuning the PID PID : K = ∞, Ti = 6s, Td = 0.833s
controllers for a first-order or second-order process the
controller gain is, in accordance with the MO tuning By fixing the controller gain at K=10, and by applying
criterion, theoretically infinite. In practice (when there is (15) and (16), the following modified PID controller
process noise), the calculated controller gain can have a parameters were obtained:
very high positive or negative value. In this case, the
controller gain should be limited to some acceptable value, K = 10, Ti = 5.85s, Td = 0.725s . (23)
which depends on the controller and the process Fig. 3 shows the closed-loop process responses when
limitations. using the original PI controller and the modified PID
The remaining two controller parameters can now be controller parameters. It is clear that the closed-loop
calculated according to the limited (fixed) controller gain process response when using such modified PID controller
from (7) and (8): is very good.
A1
Ti = (15)
1
K PR + 1.2
2K
and 1

A3  A1 A2 1 
Td = 2
 − − K PR  (16) 0.8
A1  A3 2K 

if 0.6

1
K> (17)
2 A1 A2 0.4
− 2 K PR
A3
0.2
and
Td = 0 (18) 0
0 2 4 6 8 10 12 14 16 18 20
Time [s]
if
Fig. 3. Process output (y) and controller output (u) during the closed-loop
experiment with: __ modified PID controller, -- PI controller.
C. Modified tuning procedure for 2-degrees-of-freedom performed on several process models, the following
PI controllers expression was derived (Vrančić, 1997)1:

It is frequently claimed that a drawback of the MO  AA 

tuning approach is that the process poles are cancelled by β = 0.7 + 0.5 1 2 − 1 , (28)
K A 
the controller zeros. This may lead to poor attenuation of  PR 3 
load disturbances if the cancelled poles are excited by The proposed tuning procedure will now be illustrated.
disturbances, and if they are slow compared to the The following process model was chosen:
dominant closed-loop poles (Åström and Hägglund, 1995).
Poorer disturbance rejection can be observed when 1
GP (s ) = . (29)
controlling low-order processes. In such cases, disturbance (1 + 40s )(1 + 4s )(1 + s )
rejection can be significantly improved by using a two-
degrees-of-freedom PI (PID) controller. However, the The following values of the process steady-state gain
controller parameters have to be recalculated according to and areas were obtained: KPR=1, A1=45, A2=1821, and
the changed controller structure. A3=7.29⋅104. Parameters of the classical PI controller
The controller parameters will be calculated for the (β=1), and the modified PI controller are calculated from
simple two-degrees-of-freedom PI controller, shown in Fig. (7), (8), (25), (26a), and (28):
4 (see e.g. Åström and Hägglund, 1995).
PI : β = 1, K = 4.04, Ti = 40.05s
(30)
modified PI : β = 0.762, K = 4.11s, Ti = 22.7 s

Fig. 5 shows the closed-loop process responses when

using the classical PI controller and the modified PI
controller parameters. It is clear that disturbance rejection,
Fig. 4. A two-degrees-of-freedom PI controller. when using the modified PI controller, is significantly
improved.
By following the same tuning objective as given in (1),
the following PI controller parameters are derived as a Process output; __ PI, −− modified PI
function of parameter β: (see Vrančić (1997)):
1

( )( 2
K 2 1 − β 2 K PR A3 + A1 − 2 K PR A1 A2 +
3
) (24)
+ 2 K (K PR A3 − A1 A2 ) + A3 = 0 0.5

A1
Ti = 2
(25) 0

(1 − β )
0 50 100 150 200 250 300 350 400
1 KK PR 2
K PR + +
Time [s]
Process input; __ PI, −− modified PI
2K 2 5

From (24), the controller proportional gain K can be 4

expressed in the following way: 3

( A1 A2 − K PR A3 ) − S1
2

K= , (26a)
(1 − β )(K
2
PR
2 3
A3 + A1 − 2 K PR A1 A2 ) 1

0
0 50 100 150 200 250 300 350 400
if KPRA3-A1A2 < 0, and Time [s]

( A1 A2 − K PR A3 ) + S1 Fig. 5. Process output (above) and controller output (below) during the
K= , (26b) closed-loop experiment with:
(1 − β )(K
2
PR
2 3
A3 + A − 2 K PR A1 A2
1 ) __ classical PI controller, -- modified PI controller.

if KPRA3-A1A2 > 0, where

IV. CONCLUSIONS
S1 = (K PR A3 − A1 A2 ) −
2

. (27a)
2
(
− A3 (1 − β ) K PR A3 + A1 − 2 K PR A1 A2
2 3
) The purpose of this paper was to present how to
overcome some obstacles in order to be able to apply the
In case when β=1, or KPR2A3+A13-2KPRA1A2=0, the MOMI tuning method in practice. Namely, tuning result
proportional gain is calculated from (7) by applying Td=0. can become quite sensitive to the process noise, the
The remaining question is how to choose the new calculated controller gain can become quite high (positive
parameter β? By using optimisation of the closed-loop or negative), and the MOMI tuning method could result in
responses on the reference and disturbance step changes, poor disturbance rejection. All of the counted problems
were treated in the present paper.

1
Criteria used for optimisation will be given in the final
version of the paper.
 It was shown that the tuning result can become relatively
insensitive to the process corrupted by noise by properly
choosing integration interval of the process step response.
The calculated gain of the PID controller, when applying
the MO criterion, could be too high for successful
implementation in practice. It was shown that in this
particular case the PID controller parameters can be simply
re-tuned, according to arbitrary chosen controller gain,
without the need for additional process identification stage.
The MO technique may lead to poor attenuation of load
disturbances. It was shown that disturbance rejection can
be significantly improved by using a two-degrees-of-
freedom controller structure.

REFERENCES

[1] Åström, K. J., T. Hägglund, C. C. Hang, and W. K. Ho (1993).

Automatic Tuning and Adaptation for PID Controllers - A Survey.
Control Engineering Practice, 1 (4), pp. 699-714.
[2] Åström, K. J., and T. Hägglund (1995). PID Controllers: Theory,
Design, and Tuning. Instrument Society of America, 2nd edition.
[3] Gorez, R. (1997). A survey of PID Auto-Tuning Methods. Journal
A, 38 (1), pp. 3-10.
[4] Hanus, R. (1975). Determination of controllers parameters in the
frequency domain. Journal A, XVI (3).
[5] Kessler, C. (1955). Über die Vorausberechnung optimal
abgestimmter Regelkreise Teil III. Die optimale Einstellung des
Reglers nach dem Betragsoptimum. Regelungstechnik, Jahrg. 3,
pp. 40-49.
[6] Rake, H. (1987). Identification: Transient- and Frequency-
Response Methods. Systems & Control Encyclopedia; Theory,
Technology, Applications, Madan G Singh, ed., Pergamon Press.
[7] Strejc, V. (1960). Auswertung der dynamischen Eigenschaften von
Regelstrecken bei gemessenen Ein- und Ausgangssignalen
allgemeiner Art. Z. Messen, Steuern, Regeln, 3 (1), pp. 7-10.
[8] Vrančić, D. (1997). Design of Anti-Windup and Bumpless Transfer
Protection. Doctoral thesis.
[9] Vrančić, D., Y. Peng, S. Strmčnik, and R. Hanus (1996). A new
tuning method for PI controllers based on a process step response.
Pre-prints of the CESA'96 IMACS Multiconference, Symposium
on Control, Optimization and Supervision, Lille, 2, pp. 790-794.
[10] Vrančić, D., Y. Peng, and J. Petrovčič (1997). A new simple auto-
tuning method for PID controllers. Pre-prints of the 2nd IFAC
Workshop on New Trends in Design of Control Systems,
Smolenice, pp. 457-462.
[11] Vrančić, D., Y. Peng, S. Strmčnik, and Đ. Juričić (1999a). A
multiple integration tuning method for filtered PID controller.
Accepted by the IFAC World Congress, Beijing, 1999.
[12] Vrančić, D., Y. Peng, and S. Strmčnik (1999b), A new PID
controller tuning method based on multiple integrations, To appear
in Control Engineering Practice.
[13] Ziegler, J. G., and N. B. Nichols (1942). Optimum settings for
automatic controllers. Trans. ASME, 64, pp. 759-768.