ISA Transactions 47 (2008) 94–100

www.elsevier.com/locate/isatrans

Comparative study of decay ratios of disturbance-rejection magnitude

optimum method for PI controllers
Satja Lumbar ∗ , Damir Vrančić, Stanko Strmčnik
Jožef Stefan Institute, Ljubljana, Slovenia

Received 6 January 2007; accepted 7 May 2007

Available online 15 August 2007

Abstract

One of the key time-domain closed-loop performance requirements is the closed-loop response decay ratio. In this paper, the decay ratios of
the disturbance-rejection magnitude optimum (DRMO) tuning method [Vrančić D, Strmčnik S, Kocijan J. Improving disturbance rejection of PI
controllers by means of the magnitude optimum method. ISA Trans 2004; 43: 73–84; Vrančić D, Strmčnik S. Achieving optimal disturbance
rejection by using the magnitude optimum method. In: Pre-prints of the CSCC’99 conference. 1999. p. 3401–6] are analyzed and compared
to decay ratios of two other modern tuning methods, i.e. the Kappa–Tau tuning method (based on time-domain step-response characteristics)
[Åström KJ, Högglund T. PID controllers: Theory, design, and tuning. 2nd ed. Instrument Society of America; 1995] and the non-convex
optimization tuning method (based on frequency response) [Panagopoulos H, Åström KJ, Hägglund T. Design of PI controllers based on non-
convex optimization. Automatica 1998; 34: 585–601; Panagopoulos H, Åström KJ, Hägglund T. Design of PID controllers based on constrained
optimisation. IEE Proc Control Theory Appl 2002; 149 (1): 32–40]. It is shown that the DRMO method results in such a closed-loop response
that the decay ratio is within a relatively narrow interval when compared to the other two methods.
c 2007, ISA. Published by Elsevier Ltd. All rights reserved.

Keywords: Disturbance rejection; Decay ratio; Controller tuning; PI controller

1. Introduction which is a relatively demanding procedure. Some popular

PID controllers are the most widely used controllers
in the chemical and process industry due to their very
good balance between robustness and performance. However,
efficient closed-loop control depends on the quality of PID
controller tuning. The most often applied tuning rules for PID
control are those based on the process open-loop step response
or detection of one or several points of process frequency
response [3,5–10]. It has been estimated [1] that more than 95%
of the control loops used in process control are of the PI and
PID type.
When dealing with closed-loop regulation, a frequent
time-domain requirement is, amongst others, the closed-loop
response decay ratio. In order to satisfy a prescribed decay ratio,
the closed-loop response has to be optimized in time domain,
∗ Corresponding address: Department of Systems and Control, Jožef Stefan
Institute, Jamova 39, 1000 Ljubljana, Slovenia. Tel.: +386 1 477 3732; fax:
+386 1 477 3994.
E-mail address: satja.lumbar@ijs.si (S. Lumbar).
methods [4,5] tend to make use of an indirect approach that is
supposed to guarantee acceptable decay ratios by optimizing
the minimal distance of the open-loop transfer function to
the critical point in the polar plot. Such approaches result in
stable closed-loop responses for different process models but
the values of the decay ratios sometimes vary considerably.
In contrast, while using the disturbance-rejection magnitude
optimum (DRMO) tuning method [1,2,11–13], we have
observed that the decay ratios have rather similar values for
a variety of process models, thereby offering a significant
advantage. Furthermore, the DRMO method is relatively simple
to apply since, in its basic form, it does not require any form
of optimization (i.e. retuning). To confirm our observations,
we tested the DRMO method on various process models often
encountered in the process and chemical industry and the
associated decay ratios were analyzed and compared with two
other modern tuning methods.
This paper presents our results and is set out as follows.
Section 2 provides basic definitions. A study on uniformity
of the decay ratios of the DRMO method is given in

c 2007, ISA. Published by Elsevier Ltd. All rights reserved.

0019-0578/$ - see front matter
doi:10.1016/j.isatra.2007.05.006
 S. Lumbar et al. / ISA Transactions 47 (2008) 94–100 95

Fig. 2. Typical closed-loop configuration using a 1DOF controller.

If we assume that there is no input disturbance (d = 0), the

transfer function between the reference and the process output
is:
Y (s) G P (s)G C (s)
G C L (s) = = . (3)
R(s) 1 + G P (s)G C (s)
The controller is determined in such a way that
Fig. 1. Definition of decay ratio.
G C L (0) = 1, (4)
Section 3. Section 4 provides a comparison of the decay
d |G C L ( jω)|
 2k 
ratios obtained with the DRMO method and two other modern lim = 0; k = 1, 2, . . . , kmax , (5)
ω→∞ dω2k
tuning methods [3–5]. Finally, the conclusions are provided in
Section 5. for as many values of k as possible [1,11,12]. This technique
is variously called as magnitude optimum (MO) [15], modulus
2. Basic definitions optimum [3], or Betragsoptimum [11] and results in a fast and
non-oscillatory closed-loop time-response for a large class of
2.1. Decay ratio process models [15–17].
For a PI controller:
Decay ratio is considered as one of the most basic closed- Ki
loop parameter. It is defined by the following relation: G C (s) = K P + , (6)
s
B where K P and K i are the proportional gain and the integral
dr = , (1) gain respectively, the MO method results in the following
A
expressions for these controller parameters [1]:
where B is the difference between the second peak and
the second valley of the closed-loop response and A is the A3
KP = , (7)
difference between the first peak and the first valley, as depicted 2(A1 A2 − K PR A3 )
in Fig. 1 [14]. A2
Ki = , (8)
2 (A1 A2 − K PR A3 )
2.2. Maximum sensitivity
where A1 –A3 are the so-called “characteristic areas” [1,18,19],
and K PR is the static gain of the process.
Maximum sensitivity Ms is defined as the inverse of the
If the process is described by the following transfer function:
minimum distance of the open-loop transfer function to the
critical point (−1) in the polar plot: 1 + b1 s + b2 s 2 + · · · + bm s m
G P = K PR , (9)

1
 1 + a1 s + a2 s 2 + · · · + an s n
,
 
Ms = max  (2)
ω 1 + G p (iω)G c (iω)  then the characteristic areas of the process can be defined as [1,
18,19]:
where G p (iω) and G c (iω) are the process and the controller
transfer function respectively. A1 = K PR (a1 − b1 + Tdel )
" #
T2
2.3. Magnitude optimum method A2 = K PR b2 − a2 − Tdel b1 + del + A 1 a1
2!
Fig. 2 shows the process in a closed-loop configuration ..
.
with the controller, where signals r , u, d, y and e represent a " # (10)
k i b
reference, controller output, input disturbance, process output X Tdel k−i
Ak = K PR (−1) k+1
(ak − bk ) + (−1) k+i
and control error respectively. One possible controller design i!
i=1
objective is to maintain the closed-loop magnitude (amplitude) k−1
as flat and as close to unity over as wide frequency range as
X
+ (−1)k+i−1 Ai ak−1 .
possible [12]. i=1
96 S. Lumbar et al. / ISA Transactions 47 (2008) 94–100

However, by using the original MO method, disturbance

rejection is degraded when dealing with lower-order processes,
since slow process poles might become almost entirely
cancelled by controller zeros. If the cancelled poles are
excited by disturbances and are slow compared to dominant
closed-loop poles, poor attenuation of load disturbances might
occur [1]. Consequently, an exponential approach to the
reference (so-called “long-tails”) is observed in these cases.
Such closed-loop responses are usually not desired in practice.
Degradation of disturbance rejection is expected, since the
MO method aims at achieving optimal reference tracking
instead of disturbance rejection. Recently, a modification of the
MO tuning method, referred to as the “disturbance-rejection
magnitude optimum” (DRMO) method, has been proposed [1].
It considerably improves closed-loop disturbance-rejection
performance, especially for lower-order processes. When Fig. 3. Decay ratios for closed-loop systems with a PI controller and processes
applying the DRMO tuning method, the following expressions G P1 to G P10 .
were obtained for the PI controller structure [1]:
1
q G P5 = ;
ξ2 − sgn(ξ2 )A1 A22 − A1 A3 (1 + sT1 ) (1 + sT2 )2
2
KP = , (11) 12 − 2T1
ξ1 T1 = {5.5, 5, 4.5, 4, 3.5, 3}; T2 = , (18)
2
(1 + K P K PR )2 1 12
Ki = , (12) G P6 = ; n = {3, 4, 5, 6, 7, 8}; T = , (19)
2A1 (1 + sT )n n
where ξ1 and ξ2 are 1
G P7 =
;
(1 + sT ) (1 + skT ) 1 + sk 2 T 1 + sk 3 T


ξ1 = K PR
2
A3 − 2K PR A1 A2 + A31
(13) k = {0.9, 0.7, 0.5, 0.4, 0.3, 0.2};
ξ2 = A1 A2 − K PR A3 . 11
T = , (20)
k + k2 + k3
1 − sTz 12 − Tz
3. Study of decay ratios for DRMO tuning method G P8 =
3 ; Tz = {1, 2, 3, 4, 5, 6}; T p = ,
1 + sT p 3
As mentioned, the decay ratio is one of the most (21)
basic closed-loop parameters obtained from the process time 1 + sTz
response. In this section, decay ratios measured for a variety G P9 =
3 ; Tz = {0.5, 1, 2.5, 4.5, 5.5, 7};
1 + sT p
of process models in a closed-loop configuration with a 12 + Tz
PI controller tuned by the DRMO method are presented. Tp = , (22)
3
The following process models, covering processes of lower
and higher orders, processes with delay, non-minimum phase 1
G P10 = ;
processes and processes with zeros on the left half-plane, were (1 + 4s) (1 + 4s (1 − iα)) (1 + 4s (1 + iα))
selected: α = {0.2, 0.3, 0.4, 0.5, 0.7, 1} . (23)

e−sTdel These process models were used in the closed-loop

G P1 = ; Tdel = {12, 11, 10, 6, 2, 1}; configuration (Fig. 2) with input step disturbance of magnitude
1 + sT
T = 12 − Tdel , (14) 1 (d = 1(t)). The calculated decay ratios (Eq. (1)) for all the
process models are depicted in Fig. 3.
e−sTdel Similar decay ratios (values between 0.017 and 0.024) can
G P2 = ; Tdel = {11, 10, 8, 6, 4, 1};
(1 + sT )2 be observed for all the process models except for some of the
12 − Tdel non-minimum phase process models (G P8 in Eq. (20)), which
T = , (15)
2 give noticeably lower decay ratios. The histogram of decay
e−sTdel ratios is shown in Fig. 4.
G P3 = ; Tdel = 4;
(1 + sT1 ) (1 + sT2 )
4. Comparison of decay ratios for some modern tuning
T1 = {3, 2.5, 2, 1.5, 1, 0.5}; T2 = 12 − Tdel − T1 , (16) methods
1
G P4 = ; T1 = {11, 10, 9, 8, 7, 6};
(1 + sT1 ) (1 + sT2 ) The DRMO tuning method was compared with two modern
T2 = 12 − T1 , (17) tuning methods: the Kappa–Tau (denoted as KT) tuning
 S. Lumbar et al. / ISA Transactions 47 (2008) 94–100 97

Table 1
Controller parameters for processes (24)–(28)

Tuning method Parameters G P1 G P2 G P3 G P4 G P5

DRMO KP 0.651 2.176 0.276 0.328 0.268
Ki 0.455 4.041 0.045 0.176 0.804

KT Ms = 1.4 KP 0.535 1.32 0.077 0.141 0.005

Ki 0.334 2.289 0.018 0.11 0.020

KT Ms = 2 KP 0.633 1.93 0.164 0.179 0.158

Ki 0.325 2.591 0.027 0.101 0.472

NC Ms = 1.4 KP 1.145 3.036 0.280 0.340 0.023

Ki 0.715 5.266 0.064 0.266 0.097

NC Ms = 2 KP 1.22 4.13 0.266 0.294 0.255

Ki 0.685 6.988 0.048 0.184 0.854

Fig. 4. The histogram of the decay ratios for closed-loop systems with a PI
controller and processes G P1 to G P10 .

Fig. 5. Determination of process static gain, lag time and apparent dead time Fig. 6. Response on input disturbance (r = 0, d = 1) of process G P1 (24).
from process response on unity step input.
4.3. Simulation results
method [3] and non-convex tuning in frequency domain
The sets of tuning rules for PI control described in
(denoted as NC) [4,5]. A short description of these tuning
Sections 3, 4.1 and 4.2 were applied to the following process
methods follows in the next two sub-sections.
models:
4.1. Kappa–Tau (KT) tuning method 1
G P1 = , (24)
(s + 1)3
This method basically derives from the original Ziegler– 1
Nichols rules [3]. The desired maximum sensitivity (Eq. (2)) is G P2 = , (25)
(s + 1) (1 + 0.2s) (1 + 0.04s) (1 + 0.008s)
used as a tuning parameter. If the process is stable, its dynamics
is characterized by three parameters: the static gain K PR , the e−15s
G P3 = , (26)
apparent dead time L, and the apparent lag T (Fig. 5), which (s + 1)3
can all be obtained from a simple open-loop experiment. 1 − 2s
The procedure of calculating these parameters is given in [3]. G P4 = , (27)
(s + 1)3
4.2. Non-convex (NC) based optimization tuning method G P5 = e−s . (28)
The PI controller parameters, calculated from all three
This method is based on non-convex optimization in tuning methods, are given in Table 1. For KT these parameters
frequency domain [4]. The controller parameters are adjusted are calculated for Ms = 1.4 and Ms = 2, and for NC method
until a certain value of sensitivity Ms (Eq. (2)) is achieved Ms = 1.4, 1.6, 1.8 and 2.
at highest integral gain (K i ). This method requires either the Figs. 6–10 show the closed-loop responses on input distur-
process model transfer function or the frequency characteristics bance for processes G P1 (24) to G P5 (28). In Fig. 11 the corre-
of the process. sponding decay ratios are depicted for all three tuning methods.
98 S. Lumbar et al. / ISA Transactions 47 (2008) 94–100
Fig. 7. Response on input disturbance (r = 0, d = 1) of process G P2 (25).
Fig. 8. Response on input disturbance (r = 0, d = 1) of process G P3 (26).
Fig. 9. Response on input disturbance (r = 0, d = 1) of process G P4 (27).
The decay ratios of the DRMO tuning method range
between 0.0083 and 0.0239, while the other two methods have
Fig. 10. Response on input disturbance (r = 0, d = 1) of process G P5 (28).
Fig. 11. Decay ratios of processes (24)–(28). Top: DRMO tuning method (+),
the KT tuning method with Ms = 1.4 (4) and the KT tuning method with
Ms = 2 (◦). Bottom: The NC tuning method with Ms = 1.4 (O), NC tuning
method with Ms = 1.6 (4), NC tuning method with Ms = 1.8 (◦), NC tuning
method with Ms = 2 (×).
significantly wider intervals of decay ratios for Ms ≥ 1.6
(Table 2). On the other hand, most decay ratios obtained from
the KT and NC methods for Ms = 1.4, become zero and
the interval of decay ratios becomes narrower than that of
the DRMO method. Zero decay ratio results in overdamped
response and inefficient closed-loop performance, as mentioned
in Section 2.3. This can be clearly observed in Figs. 6–10 in all
cases where the decay ratio is equal to zero.
Maximum sensitivity (Eq. (2)) values are also calculated for
all tested processes and tuning methods. Fig. 12 clearly shows
that the NC method guarantees constant maximum sensitivity
(by optimization) since it sets Ms to a fixed value. On the other
hand, the variance of maximum sensitivity of the KT method
varies significantly, even more than the DRMO method.
Expectation of interdependency between the decay ratio and
maximum sensitivity seems to be natural. For the same process
model an increase in maximum sensitivity yields an increase in
 S. Lumbar et al. / ISA Transactions 47 (2008) 94–100
Fig. 12. Maximum sensitivity function for processes (24)–(28). Top: DRMO
tuning method (+), the KT tuning method with Ms = 1.4 (4) and the
KT tuning method with Ms = 2 (◦). Bottom: The NC tuning method with
Ms = 1.4 (O), NC tuning method with Ms = 1.6 (4), NC tuning method with
Ms = 1.8 (◦), NC tuning method with Ms = 2 (×).
Table 2
Decay ratio intervals for DRMO, KT and NC tuning methods
Minimum value Maximum value
DRMO 0.0083 0.0239
KT Ms = 1.4 0 0.0073
KT Ms = 2 0 0.1018
NC Ms = 1.4 0 0
NC Ms = 1.6 0 0.02506
NC Ms = 1.8 0.0025 0.0549
NC Ms = 2 0.0132 0.0896
decay ratio and vice versa, which is clearly shown in Fig. 11.
However, Fig. 11 also shows that some responses have higher
value of decay ratio for Ms = 1.6 than other responses for
Ms = 2.0. Therefore it may be concluded that, setting the
maximum sensitivity to a certain value does not necessarily
yield a relatively constant decay ratio.
5. Conclusions
The disturbance-rejection magnitude optimum (DRMO)
tuning method applied to a PI controller results in efficient
input disturbance rejection for many process models frequently
encountered in process and chemical industries. The aim of this
paper was to determine whether the closed-loop time-responses
are uniform in terms of the decay ratios. Our experiments on
a wide range of process models established that decay ratios,
except for non-minimal phase process models, are between
0.0167 and 0.0239, whereas decay ratios of the non-minimum
phase process models are significantly smaller.
In addition, we observed that the DRMO method, when
compared with the Kappa–Tau method and the non-convex
optimization method, gives less variable closed-loop responses
in terms of decay ratios in most cases. Narrower decay ratio
99
intervals for KT and NC methods are observed only for
the lowest values of maximum sensitivity (Ms = 1.4) due
to over-conservative tuning, which actually results in zero
decay ratio and degraded disturbance-rejection performance
(so-called “long tails”).
References
[1] Vrančić D, Strmčnik S, Kocijan J. Improving disturbance rejection of PI
controllers by means of the magnitude optimum method. ISA Trans 2004;
43:73–84.
[2] Vrančić D, Strmčnik S. Achieving optimal disturbance rejection by
using the magnitude optimum method. In: Pre-prints of the CSCC’99
conference. 1999. p. 3401–6.
[3] Åström KJ, Hägglund T. PID controllers: Theory, design, and tuning. 2nd
ed. Instrument Society of America; 1995.
[4] Panagopoulos H, Åström KJ, Hägglund T. Design of PI controllers based
on non-convex optimization. Automatica 1998;34:585–601.
[5] Panagopoulos H, Åström KJ, Hägglund T. Design of PID controllers
based on constrained optimisation. IEE Proc Control Theory Appl 2002;
149(1):32–40.
[6] Besançon-Voda A. Iterative auto-calibration of digital controllers:
Methodology and applications. Control Eng Pract 1998;6:345–58.
[7] Gorez R. A survey of PID auto-tuning methods. Journal A 1997;38:3–10.
[8] Ho WK, Hang CC, Cao LS. Tuning of PID controllers based on gain and
phase margin specifications. In: Proceedings of the 12th world congress
IFAC, vol. 5. 1993. p. 267–70.
[9] Skogestad S. Simple analytic rules for model reduction and PID controller
tuning. J Process Control 2003;13:291–309.
Interval [10] Wang QG, Hang CC, Bi Q. Process frequency response estimation from
relay feedback. Control Eng Pract 1997;5:1293–302.
0.0156
[11] Kessler C. Über die Vorausberechnung optimal abgestimmter Regelkreise
0.0073
Teil III. Die optimale Einstellung des Reglers nach dem Betragsoptimum,
0.1018
Regelungstechnik 1955;3:40–9.
0
[12] Whiteley AL. Theory of servo systems, with particular reference to
0.02506
stabilization. J IEE, Part II 1946;93(34):353–72.
0.0524
[13] Lumbar S, Vrančić D. Study on disturbance rejection magnitude optimum
0.0764
method decay ratios. 2006. Report DP-9421. Available on: http://
www-e2.ijs.si/People/Satja.Lumbar/Bibliography/DP9412_decays.pdf.
[14] Wade H. Trial and error: An organized procedure, InTech, 2005. Available
on: http://www.isa.org/InTechTemplate.cfm?Section=Article_Index1&
template=/ContentManagement/ContentDisplay.cfm&ContentID=44007.
[15] Umland JW, Safiuddin M. Magnitude and symmetric optimum criterion
for the design of linear control systems: What is it and how does it
compare with others? IEEE Trans Ind Appl 1990;26:489–97.
[16] Vrančić D, Peng Y, Danz C. A comparison between different PI controller
tuning methods. Report DP-7286. J. Stefan Institute, Ljubljana. 1995.
Available on: http://www-e2.ijs.si/Damir.Vrancic/bibliography.html.
[17] Vrančić D, Lumbar S. Improving PID controller disturbance rejection
by means of magnitude optimum. Report DP-8955. Ljubljana: J.
Stefan Institute; 2004. Available on: http://www-e2.ijs.si/Damir.Vrancic/
bibliography.html.
[18] Vrančić D, Peng Y, Strmčnik S. A new PID controller tuning method
based on multiple integrations. Control Eng Pract 1999;7:623–33.
[19] Vrančić D, Strmčnik S, Jurčić Ð. A magnitude optimum multiple
integration tuning method for filtered PID controller. Automatica (Oxf.)
2001;37:1473–9 [Print ed.].
Satja Lumbar, born in Ljubljana in 1980, received
his B.Sc. in electrical engineering at the Faculty
of Electrical Engineering, University of Ljubljana in
2005. Since 2005, he has been a researcher in the
Department of Systems and Control. His research
interests include PID control, predictive control, flight
control and related fields (modeling, simulation).
100 S. Lumbar et al. / ISA Transactions 47 (2008) 94–100

Damir Vrančić received the B.S., M.S. and Ph.D.
degrees in electrical engineering at the Faculty
of Electrical and Computer Engineering, University
of Ljubljana, Slovenia, in 1991, 1995 and 1997,
respectively. Since 1992, he has been with the
Department of Systems and Control at J. Stefan
Institute. His research interests include PID control,
tuning, modeling, simulation and design of anti-windup
protection.
Engineering, University of Ljubljana and Polytechnics, Nova Gorica.
Stanko Strmčnik was born in Vitanje, Slovenia in
1949. He received his B.Sc., M.Sc., and Ph.D. from the
Faculty of Electrical Engineering, University of Ljubl-
jana in 1972, 1975 and 1979 respectively. In 1973 he
joined J. Stefan Institute in Ljubljana. Since 1986, he
has been the head of the Department of Systems and
Control at the same institute. His research interests con-
cern mathematical modelling, identification, process
control and non-technical aspects of automation. He is
also an associated professor at the Faculty of Electrical