PID controller tuning methods: a novel teaching approach

A.O’Dwyer, School of Control Systems and Electrical Engineering,

Dublin Institute of Technology, Kevin St., Dublin 8.
e-mail: aidan.odwyer@dit.ie

Abstract: This paper discusses an approach to the teaching of PID controller tuning methods to final year
bachelors degree and taught masters students in control engineering at Dublin Institute of Technology,
Ireland. The method involves analytically calculating the gain margin, phase margin and maximum
sensitivity for PI and PID controlled systems whose process is modeled in first order lag plus time delay
(FOLPD) form. Students can examine the performance of many tuning rules from graphical results,
allowing insight to be developed as to the most rational choice of the tuning rule for the application. Some
preliminary work done to develop an expert system to allow a greater automation of the procedure for
recommending a tuning rule, for user defined requirements, is also described.

1. Introduction

The ability of proportional integral (PI) and proportional integral derivative (PID) controllers to
compensate most practical industrial processes has led to their wide acceptance in industrial applications. It
has been stated, for example, that 98% of control loops in the pulp and paper industries are controlled by PI
controllers (Bialkowski, 1996) and that, in more general process control applications, more than 95% of the
controllers are of PID type (Åström and Hägglund, 1995).
The requirement to choose either two or three controller parameters has meant that the use of tuning
rules to determine these parameters is popular. The author has tabulated 219 such tuning rules in the
literature to specify the PI controller terms, with 381 tuning rules defined to specify the PID controller
parameters (O’Dwyer, 2003). Though the use of tuning rules is practically important, it is a challenge to
allow students to gain perspective on the advantages and limitations inherent in the method. Traditionally,
the topic has been taught by briefly considering the major classes of tuning rule methods, followed by a
more detailed study, often in the laboratory, of one or more tuning rules (typically the process reaction
curve tuning method of Ziegler and Nichols (1943) and the ultimate cycle method of Ziegler and Nichols
(1942)). A number of textbooks offer a more comprehensive approach (e.g. Ogunnaike and Ray (1994),
Marlin (2000), Wilkie et al. (2002), Seborg et al. (2004)), though the tuning rule methods discussed are
presented in a stand-alone manner. To the authors’ knowledge, no textbook gives a comprehensive
approach to choosing a tuning rule based on a performance specification and knowledge of the process to
be controlled.
This paper will describe such an approach to tuning rule selection taught at the Dublin Institute of
Technology to students taking control engineering as an option subject in both the final year of the four-
year honours Batchelor of Engineering (B.E.) degree in Electrical/Electronic Engineering and the one-year
taught Masters of Engineering (M.E.) degree in Advanced Engineering. The process is modeled as a first
order lag plus time delay (FOLPD) model, and compensated by PI or PID controllers whose parameters are
specified using appropriate tuning rules. The gain margin, phase margin and maximum sensitivity of the
compensated system, as the ratio of time delay to time constant of the process varies, are analytically
determined. The results of the analytical calculations are plotted and are used by the students to judge the
performance and robustness of the compensated system, the appropriateness of the tuning rule for a given
process and the range of time delay to time constant ratios over which it is sensible to apply various tuning
rules. The teaching approach is based on research work previously published by Ho et al. (1995), (1996)
and O’Dwyer (1998), (2001).
The paper is organised as follows. The formulae taught to the students for analytically calculating the
gain margin, phase margin and maximum sensitivity, for the PI compensated process, are developed in
Section 2. Graphical results showing the performance and robustness of FOLPD processes, compensated
with sample PI and PID tuning rules, are shown in Section 3. In Section 4, a tuning rule that achieves
constant gain and phase margins for all values of delay is outlined. Section 5 outlines some work done on
an expert system implementation of the method, for teaching purposes. Conclusions of the work are drawn
in Section 6.
2. The analytical calculation of gain margin, phase margin and maximum sensitivity – PI controller

The controller and process model are respectively given by

1
G c ( s)=K c 1+
( Tis ) (1)
−sτ m
Km e
Gm (s )=
1+ sT m (2)
The controller is a so-called ideal PI controller; 98 tuning rules have been defined for the control of
processes modeled in FOLPD form using this controller (O’Dwyer, 2003). Then
− j ωτ m
Km e K c ( jωT i +1)
Gm ( jω)Gc ( jω)=
1+ jωT m jωT i (3)
From the definition of gain and phase margin, the following sets of equations are obtained:
φm =arg [ Gc ( jωg )Gm ( jωg ) ]+ π (4)
1
A m=
|Gc ( jω p )G m( jω p )| (5)
where
ωg and ω p are given by
|Gc ( jw g )Gm ( jw g )|=1 (6)
arg [ Gc ( jw p )Gm( jw p )] =- p (7)
From Equation (3),
K m K c 1+ ω2 T
√ 2
i
Gm ( jω) Gc ( jω)= ∠−0 . 5 π + tan−1 ωT i −tan −1 ωT m−ωτ m
2
ωT i 1+ω T
√ m2 (8)
Therefore, from Equation (4)
φm =π−0 .5 π+tan−1 ω g T i−tan−1 ω g T m−ωg τ m (9)
with
ωg given by the solution of Equation (6) i.e.
K m K c 1+ω 2 T 2
√ g i =1
ω g T i √1+ω 2 T 2
g m (10)
Also, from Equations (5) and (8),
ω T 1+ω p 2 T m2

ωp
A m=
1
= p i
|G m ( jω p )Gc ( jω p )| K m K c 1+ ω 2 T 2 √ p i (11)
with given by the solution of Equation (7) i.e.
−0.5 π+tan−1 ω p T i −tan−1 ω p T m−ω p τ m=−π (12)
From Equation (10),
w g may be determined analytically to be
2

ωg=
√ T i( K
c
2 K 2 −1
m

An analytical solution of Equation (12) (to determine

)+ √( K c 2 K m2−1 ) T i2+ 4 K c2 K m2T m2
2 T iT

ω
m2

p ) is not possible. An approximate analytical

(13)

solution may be obtained if the following approximation for the arctan function is made:
π π π
tan−1 x≈ x , ¿|x|<1 ¿ tan−1 x≈ − ,¿|x|>1 ¿
4 and 2 4x (14)
This is quite an accurate approximation, as is shown in Figures 1a and 1b.

Figure 1a: Arctan x and its approximation (Equation 14)

arctan x
approximation

Figure 1b: % error in taking approximation (Equation 14) to arctan(x)

Considering Equation (12), four possibilities present themselves if the approximation in Equation (14) is to
be used. These possibilities, together with the formula for
w p that may be determined analytically for
each of these cases, are
1 1

ω p T i >1,¿ω p T m>1 ¿ :
ω p=
√
π± π 2 −4 πτ m

4 τm
( −
Ti Tm )
(i) (15)
π

w p T i >1,¿w p T m<1 ¿ :
ω p=
√
π± π 2 −
Ti
( 0 .25 πT m+ τ m )
2 ( 0. 25 πT m +τ m)
(ii)
(16)
π

(iii)
w p T i <1,¿w p T m>1 ¿ :
(17)
ω p=
√ Tm
4 τ m−πT i

2π
ω p=
(iv)
w p T i <1,¿w p T m<1 ¿ : 4 τ m + π ( T m−T i )
(18)
The gain and phase margin of the compensated system, for each of the tuning rules, as a function of
t m /T m , may be calculated by applying Equations (9), (11), (13) and the relevant approximation for
w p from Equations (15) to (18).
The maximum sensitivity is the reciprocal of the shortest distance from the Nyquist curve to the (-1,0)
point on the Rl-Im axis. It is defined as follows:
 M max = Max ¿
1
all ¿ ω| |
1+G ( jω)G ( jω)
m c (19)
For a FOLPD process model controlled by a PI controller,
2

|G m ( jω )G c ( jω)|=
1+ ω T 2 K K
i c m √
1+ω 2 T 2 ωT i √ m (20)
and
−1 −1
arg [ Gm ( jω)G c ( jω) ]=−0 . 5 π −tan ωT m+tan ωT i −ωτ m (21)
The maximum sensitivity may be calculated over an appropriate range of frequencies corresponding to
0 0
phase lags of 100 to 260 .
The method can be extended to the determination of the gain margin, phase margin and maximum
sensitivity of FOLPD process models, compensated by the classical PID controller structure, and the ideal
PID controller structure (with some approximation), in a straightforward manner. These controller
structures are given by equations (19) and (20), respectively.
1 1+sT d
G c ( s )=K c 1+
( Tis )( 1+ sαT d ) (19)
1
Gc ( s)=K c 1+
( +T s
Tis d )(20)
88 tuning rules have been defined for the control of processes, modeled in FOLPD form, using these
controllers (O’Dwyer, 2003).

3. Simulation results

Space considerations dictate that only representative simulation results may be provided; an extensive
set of simulation results covering many of the PI controller and PID controller tuning rules are available
(O’Dwyer, 2000). The MATLAB package has been used in the simulations. Figures 2 to 7 show how gain
margin, phase margin and maximum sensitivity vary as the ratio of time delay to time constant varies, if
some PI tuning rules are used (Figures 2 to 4) and corresponding PID tuning rules for the classical
controller structure (with α = 0.1) are used (Figures 5 to 7). In these results, Z-N refers to the process
reaction curve method of Ziegler and Nichols (1942); W-W refers to the process reaction curve method of
Witt and Waggoner (1990); IAE reg, ISE reg and ITAE reg refer to the tuning rules for regulator
applications that minimise the integral of absolute error criterion, the integral of squared error criterion and
the integral of time multiplied by absolute error criterion, respectively, as defined by Murrill (1967) for PI
tuning rules and Kaya and Scheib (1988) for PID tuning rules based on the classical controller structure.
Figures 8 to 15 show gain and phase margin comparisons between corresponding PI and PID controller
tuning rules. These results would allow the following general points to be outlined in the classroom:
 It is clear that the gain margin is generally less when the PID rather than the PI tuning rules are
considered, over the ratios of time delay to time constant taken; the difference between the phase
margins is less clear cut. This suggests that these PID tuning rules should provide a greater degree of
performance than the corresponding PI tuning rules, but may be less robust.
 Comparing the individual tuning rules, it is striking that the ISE based tuning rules have generally the
smallest gain margin and have also a small phase margin, suggesting that this is a less robust tuning
strategy. The results in Figures 4 and 7 confirm these comments.
 No general conclusion can be reached as to the best tuning rule (as expected); it is interesting, though,
that many tuning rules may be applied at ratios of time delay to time constant greater than that normally
recommended. One example may be seen in Figures 5 to 7, where the gain margin, phase margin and
maximum sensitivity (associated with the use of the PID tuning rule for obtaining minimum IAE in the
regulator mode) tends to level out when the ratio of time delay to time constant is greater than 1;
normally, the tuning rule is used when the ratio is less than 1 (Murrill, 1967). On the other hand, it is
clear from Figures 8 and 9 that there is a significant degradation of performance when using the PID
 tuning rule of Witt and Waggoner (1990) and the PI tuning rule of Ziegler and Nichols (1942) for large
ratios of time delay to time constant, which is compatible with application experience.
 The decision between the use of a PI and PID controller to compensate the process, depends on the ratio
of time delay to time constant in the FOLPD model, together with the desired trade-off between
performance and robustness, as expected. It turns out, however, that the analytical method explored
allows the calculation of a far wider range of gain and phase margins for PI controllers; it is also true
that stability tends to be assured when a PI controller is used (O’Dwyer, 2000). Thus, a cautious design
approach is to use a PI controller, particularly at larger ratios of time delay to time constant.

Figure 2: Gain margin Figure 3: Phase margin Figure 4: Maximum sensitivity

4 120 8

- = Z-N
3.5
2.5
* = IAE reg 100
100
7
6

+ = ITAE reg 90 6
5.5
3 80
o = ISE reg 5 Figure 10: Gain
80
5
2.5 4 60 120
70 4.5
2
3 100 4
60 4
2 40
2.5
2.8 3.5 100
90
100 3.5
3
50

2.6 3
1.5 2040 90
80 3 - = IAE reg PID
2
1.5
3 30 80 2.5 o = IAE reg PI
2.4 80
70 100
2.81 0 1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0
20 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 20 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
70 90
2.2
2.6 60
2
2.5 Tm 10
T 60
1.5
Tm
2.4
21 0
m
60
50
80 1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
2.2 Figure 5: Gain margin
0 0.2 0.4 0.6 0.8
Figure 1
6: Phase 1.2 1.4 1.6 1.8
50 margin
70
2
Figure 7: Maximum sensitivity
1.8 2 4040
2 40 60
1.6 - = W-W 30
1.5 50
1.8
* = IAE reg
1.4 1.5
3020
20
1.6+ = ITAE reg 20
40
1.2
o = ISE reg 10
30
1.4 1 10 0
1 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
1.2 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
1 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 10
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 T
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

4. Design of tuning rules to achieve constant gain Figure 12: Gain

and phase margins for all values of delay

Normally, the gain and phase margins of the compensated systems tend to increase as the time delay - = ISE reg PID
increases, T
supporting
m the common view that PI T and PID T
controllers are less suitable for m
the control of
m o = ISE reg PI
dominant time delay processes. However, it is possible to design a PI tuning rule that facilitates the
Figure
achievement of 8: Gain margin
a constant gain comparison
and phase margin as the time delay of the
Figure process
9: Phase modelcomparison
margin varies; this tuning
rule has the structure: c m m m , i K =aT /K τ
m . This new tuning rule design, outlined below, is T =T
taught to the students; a similar method, for the design of a similar tuning rule for the classical PID
controller of a FOLPD process model, is also taught to the students.
For the -FOLPD
= W-Wprocess
PID model and the PI controller, Equations (1) and (2) apply i.e,
−sτ m
o = Z-N PI K e
Gm (s)= m
1+ sT m
1
with
Gc ( s)=K c 1+
( ) Tis
T
From Section 2,
φm =π−0 .5 π+tan−1 ω g T i−tan−1 ω g T m−ωg τ m (Equation 9) Figure 14: Gain

- = ITAE reg PI
o = ITAE reg PI
Tm Tm
 T
with
ωg given by the solution of
K m K c 1+ω 2 T 2
√ g i =1
ω g T i √1+ω 2 T 2
g m (Equation 10)
and
ω p T i 1+ω p 2 T m2

ω p given by the solution of

A m=
√
K m K c 1+ω 2 T 2
p i (Equation 11)
with
−0.5 π+tan−1 ω p T i −tan−1 ω p T m−ω p τ m=−π (Equation 12)
If
Kc T
and i are designed as follows:
aT m
Kc=
K mτ m (21)
and
T i =T m (22)
Then Equation (12) becomes
−0.5 π−ω p τ m=−π (23)
i.e.
ω p=π /2 τ m
(24)
Substituting Equation (24) into Equation (11) gives
A m=πT m /2 K m K c τ m (25)
Substituting Equation (22) into Equation (10) gives
K m K c /ω g T m=1 (26)
i.e.
ω g =K m K c /T m (27)
Substituting Equations (22) and (27) into Equation (9) gives
φm =0 .5 π−K m K c τ m /T m (28)
Finally, substituting Equation (21) into Equation (25) gives
A m=π /2 a (29)
and substituting Equation (21) into Equation (28) gives
φm =0 .5 π−a (30)
Some typical tuning rules are shown in Table 1.

Table 1: Typical PI controller tuning rules – FOLPD process model

a Kc Ti Am φm
π /3 1.047 T m /K m τ m Tm 1.5 π /6
π /4 0.785 T m / K m τ m Tm 2.0 π /4
π /6 0 .524 T m / K m τ m Tm 3.0 π /3
5. Expert system implementation

In Section 3, data has been defined as MATLAB variables representing gain margin and phase margin
values, as the ratio of time delay to time constant varies, for most of the 98 PI controller tuning rules for
FOLPD process models (O’Dwyer, 2000). Though the graphical results based on this data gives students a
good insight into the usefulness of the tuning rules (at larger time delays, for example), the sheer volume of
results generated means that a more automatic means of comparing the tuning rules would be useful. Thus,
some preliminary work on the development of an expert system was carried out. The expert system asks the
user to input the desired range of gain and phase margin of the compensated system. The expert system will
then indicate the PI tuning rules that fit the criteria, outputing the gain margin and phase margin associated
with each rule, together with a recommendation.
The gain margin and phase margin data, associated with each tuning rule, was first exported to a
Microsoft Access database file. A Microsoft Visual Basic (VB) front end was developed using intrinsic VB
controls, to provide the user with a friendly and intuitive interface. On correct completion of a logon
process, the main user screen, shown in Figure 16, is loaded and a connection to the Microsoft Access
database is made using a VB data control object. The database is local to the VB programme (i.e. it is on
the same PC as the VB application), though it could be put on a Local Area Network (LAN) or a Wide
Area Network (WAN). The upper portion of this screen is used as an input interface. It consists of a set of
text boxes and labels into which the user may enter the ratio of delay to time constant (TD/TC on Figure
16), the required gain and phase margins and an acceptable variation (called tolerance on Figure 16) on the
gain and phase margins.
The process of retrieval starts when the Execute button is clicked. A Structured Query Language (SQL)
query, using parameters from the input text boxes, was used to return a recordset of matching tuning rules.
A VB DBGrid object, bound to the data control object, was used to display the recordset. If the system
finds fewer tuning rules than a low threshold value, or more tuning rules than a high threshold value, then
the user is prompted to widen or narrow the default gain and phase margin variation of 10%, respectively.
A secondary SQL query using the tuning rule number returned by the first query, was used to access
another database containing the tuning rule sources and the formulae associated with each tuning rule. The
right hand panels in the lower half of the user screen (Figure 16) were used to display this information.

6. Conclusions

The paper has reported on the teaching of PI and PID controller tuning methods by determining the
performance and robustness of a PI and PID controlled FOLPD process, with the parameters of the
controllers determined by a variety of tuning rules. The method allows an analytical approach to be taken to
the evaluation for a large number of tuning rules. The development of the analytical procedure increases
student understanding of the concepts of gain margin, phase margin and maximum sensitivity; in addition,
the plotted results allow the students to understand immediately, for each tuning rule, the performance and
robustness of the compensated system, the appropriateness of the tuning rule for a given process and the
range of time delay to time constant ratios over which it is sensible to apply the tuning rule. Limited
comparisons of the tuning rules are possible, as is shown in Section 3. The expert system reported
facilitates student motivation as well as providing a platform for further project work. In short, the
approach outlined clarifies the topic of tuning rule selection and informs subsequent laboratory work.
Finally, the method complements iterative methods of controller design, using Bode plots, based on gain
and phase margin specifications that are also covered in the courses.

Figure 16: Main User Screen

References

Åström, K.J. and Hägglund, T. (1995). PID Controllers: Theory, Design and Tuning (Instrument Society of
America, Research Triangle Park, North Carolina, 2nd Edition).
Bialkowski, W.L. (1996). in The Control Handbook, ed. W.S. Levine, (CRC/IEEE Press, Boca Raton,
Florida, 1996), 1219-1242.
Ho, W.K., Hang, C.C. and Zhou, J.H. (1995). IEEE Transactions on Control Systems Technology, 3, 245-
248.
Ho, W.K., Gan, O.P., Tay, E.B. and Ang, E.L. (1996). IEEE Transactions on Control Systems Technology,
4, 473-477.
Kaya, A. and Scheib, T.J., Control Engineering (July 1988), pp. 62-65.
Marlin, T.E. (2000). Process Control: designing processes and control systems for dynamic performance,
Mc-Graw-Hill, 2nd edition.
Murrill, P.W., Automatic control of processes (International Textbook Co., 1967)
O'Dwyer, A., Proc. Irish Signals and Systems Conference (Dublin Institute of Technology, Ireland, 1998),
227-234.
O’Dwyer, A., Technical Report AOD-00-12 (Dublin Institute of Technology, Ireland, 2000)
O’Dwyer, A., Proc. Irish Signals and Systems Conference (National University of Ireland, Maynooth,
Ireland, 2001), 96-100.
O’Dwyer, A. (2003). Handbook of PI and PID controller tuning rules, Imperial College Press.
Ogunnaike, B. and Ray, W. (1994). Process dynamics, modeling and control, Oxford University Press.
Seborg, D.E., Edgar, T.F. and Mellichamp, D.A. (2004). Process dynamics and Control, 2nd edition, Wiley.
Wilkie, J., Johnson, M. and Katebi, R. (2002). Control Engineering: an introductory course, Palgrave Ltd.
Witt, S.D. and Waggoner, R.C., Hydrocarbon Processing (June 1990), 74-78.
Ziegler, J.G. and Nichols, N.B. (1942). Transactions of the ASME, 64, 759-768.
Ziegler, J.G. and Nichols, N.B. (1943). Transactions of the ASME, July, 433-444.

Appendix 1: Glossary of symbols and abbreviations used in the paper

 Am = gain margin
FOLPD model = First Order Lag Plus time Delay model
Gc ( s) = PID controller transfer function
Gm (s) = Process model transfer function
∞

IAE = integral of absolute error =

∫0 |e(t)|dt
∞ 2

ISE = integral of squared error =

∫0 e (t)dt
∞

ITAE = integral of time multiplied by absolute error =

∫0 t|e(t)|dt
K c = Proportional gain of the controller
K m = Gain of the process model
M max = maximum value of closed loop sensitivity
PI controller = proportional integral controller
PID controller = proportional integral derivative controller
s = Laplace variable
T d = Derivative time of the controller
T i = Integral time of the controller
T m = Time constant of the FOLPD process model
f m = phase margin
tm = time delay of the process model
w = angular frequency