International Journal of Systems Science

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tsys20

Maximum sensitivity (Ms)-based I-PD controller

design for the control of integrating processes
with time delay

Fuat Peker & Ibrahim Kaya

To cite this article: Fuat Peker & Ibrahim Kaya (2022): Maximum sensitivity (Ms)-based I-PD
controller design for the control of integrating processes with time delay, International Journal of
Systems Science, DOI: 10.1080/00207721.2022.2122759

To link to this article: https://doi.org/10.1080/00207721.2022.2122759

Published online: 25 Sep 2022.

Submit your article to this journal

Article views: 127

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tsys20
INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE
https://doi.org/10.1080/00207721.2022.2122759

RESEARCH ARTICLE

Maximum sensitivity (Ms)-based I-PD controller design for the control of

integrating processes with time delay
Fuat Peker and Ibrahim Kaya

Electrical and Electronics Engineering Department, Dicle University, Diyarbakır, Turkey

ABSTRACT ARTICLE HISTORY

Integrating processes, whose one or more poles are located at the origin, are common in the process Received 17 June 2022
industry. This paper focuses on maximum sensitivity (Ms)-based control of these types of processes. Accepted 5 September 2022
Integral–proportional derivative (I-PD) controllers are designed by exploiting the direct synthesis KEYWORDS
method for different forms of integrating processes. The suggested design approach is based on Time delay integrating
comparing the characteristic equation of the closed-loop system, which comprises the integrating processes; maximum
system and I-PD controller with a lead/lag filter, with the desired characteristic equation. Simple and sensitivity (Ms); direct
analytical adjusting rules are followed to determine the parameters of the I-PD controller and the synthesis (DS) method;
lead/lag filter according to desired robustness specified by maximum sensitivity (Ms). The formu- integral–proportional
las provided contain process transfer function parameters and a tuning parameter that is used for derivative (I-PD)
setting the desired Ms. The benefits of the proposed technique are demonstrated by simulation
examples and a real-time application of cart position control on an experimental set-up. Compar-
isons with some reported proportional–integral–derivative (PID) and I-PD design techniques are
presented to demonstrate the advantages of the proposed design method more evidently.

1. Introduction
can be used to classify integrating processes (Zhang
Integrating systems, in which one or more poles are et al., 2021). Typical integrating systems can be mod-
located at the origin, are one of the commonly encoun- elled as pure integrating plus dead time (PIPDT),
tered processes in the industry. These systems are integrating plus stable/unstable first-order plus dead
widespread in chemical and process industries, and time (IFOPDT/IUFOPDT) and double integrating
also in the area of mechanical engineering (Vrančić plus dead time (DIPDT) transfer functions (Kaya &
et al., 2018). Some examples of integrating processes Peker, 2020; Zhang et al., 2021).
can be given as liquid level control in tanks, batch Proportional–integral–derivative (PID) controllers
distillation columns, paper drum dryer cans, data are still frequently preferred in the process industry
communication networks, jacketed continuous stirred owing to their user-friendliness form and good per-
tank reactors, supply chain management processes, formance in many different applications. In the liter-
bioreactors and position controls of a cart (Anil & ature, quite different methods are available to design
Padma Sree, 2015a; Chakraborty et al., 2017; Visi- PID controllers (Ang et al., 2005; Åström & Häg-
oli & Zhong, 2011; Zhang et al., 2020). Integrating glund, 1995; Cetin & Iplikci, 2015; Dey & Mudi, 2009;
processes are not self-regulating systems (not asymp- O’Dwyer, 2009; Vilanova & Visioli, 2012; Wu et al.,
totically stable), namely, a bounded input causes an 2014) for controlling various processes. For the control
unbounded output (Visioli & Zhong, 2011). They are of time-delay integrating processes, different design
a subclass of unstable systems, so their control is approaches were reported, such as direct synthesis
quite difficult and requires special attention (Visioli & (DS) methods (Ajmeri & Ali, 2015; Anil & Padma Sree,
Zhong, 2011). Interested ones can find detailed infor- 2015a; Seshagiri Rao et al., 2009; Zhang et al., 2020,
mation concerning time delay integrating processes 2021), internal model control (IMC) methods (Jin &
and their control in Visioli and Zhong (2011). Num- Liu, 2014; Kaya, 2004; Kumar & Padma Sree, 2016;
ber of poles positioned at the origin or other places Rao & Padma Sree, 2010) and optimisation methods

CONTACT Fuat Peker [email protected] Electrical and Electronics Engineering Department, Dicle University, Diyarbakır, Turkey
© 2022 Informa UK Limited, trading as Taylor & Francis Group
2 F. PEKER AND I. KAYA

(Ali & Majhi, 2011; Grimholt & Skogestad, 2016; Kaya parameters of the PID with lead/lag filter were pre-
& Cengiz, 2017; Verma & Padhy, 2018; Visioli, 2001). sented using the DS method and multiple dominant
For a class of integrating processes, Ajmeri and Ali pole placement method. Tuning rules were provided
(2015) suggested a parallel control structure (PCS) to reach the desired robustness defined by Ms (Zhang
having two controllers, which are designed using the et al., 2020). In Zhang et al. (2021), by making use
DS technique. In that study, a proportional-derivative of the DS method, adjusting rules for differential for-
(PD) controller was used for improving servo response ward PID controller with lead/lag filter, were proposed
and a PID controller was used for improving reg- for time-delay integrating processes. Parameters of a
ulatory response (Ajmeri & Ali, 2015). A PID con- differential PID controller with a lead/lag filter were
troller with a lead/lag filter was designed by Anil and tuned to obtain the desired robustness specified by Ms
Padma Sree (2015a) by making use of the DS tech- (Zhang et al., 2021).
nique for different forms of integrating processes. In All studies surveyed above present adjusting rules
the aforementioned study, tuning rules were presented for classical single-loop PI/PD/PID controllers or for
in terms of parameters of the process transfer func- PI/PD/PID controllers combined in a 2DOF structure
tion and a setting parameter that is used for desired for controlling servo and regulatory responses. But,
robustness defined by maximum sensitivity (Ms) value it is a well-known fact that classical PID controllers
(Anil & Padma Sree, 2015a). In Anil and Padma Sree may not give satisfactory responses where the process
(2015b), for a class of integrating processes with dead is unstable, integrating and resonant (Kaya & Peker,
time, a PID controller with a cascaded lead/lag fil- 2020). To obtain more acceptable responses for those
ter was designed by making use of multiple domi- processes, alternative controller structures were pro-
nant pole placement techniques. Kumar and Padma posed by researchers (Aryan & Raja, 2022; Kumari
Sree (2016) proposed the design of PID controllers et al., 2021, 2022; Onat, 2019; Raja, 2021; Raja & Ali,
according to IMC principles for different forms of time 2021; D. Kumar et al., 2022). Those studies reveal that
delay integrating processes. PID parameters were pro- the use of double-loop PI-PD, IMC-PD and fractional-
vided in terms of parameters of the process transfer order IMC-PD/P (FOIMC-PD/P) controllers was gen-
function and a setting parameter that is the time con- erally proposed. Recently, in the literature, to control
stant of the IMC filter (D. B. S. Kumar & Padma Sree, time-delay integrating processes, the use of I-PD con-
2016). Grimholt and Skogestad (2016) presented opti- trollers was suggested (Chakraborty et al., 2017; Kaya,
mal tuning rules according to four different Ms val- 2020; Kaya & Peker, 2020; Peker & Kaya, 2019). The
ues (Ms = 1.40, 1.59, 1.80 and 2) for PID controller structures of I-PD and PI-PD controllers are similar,
parameters to control DIPDT processes. They used but the I-PD controller consists of one less setting
a serial (cascade) PID form for the control purpose parameter when compared to the PI-PD controller.
(Grimholt & Skogestad, 2016). For the control of inte- The I-PD controller provides similar performances to
grating processes, Srivastava and Pandit (2017) pro- the PI-PD controller, in spite of its one less parameter
vided a 2 degree of freedom (2DOF) linear quadratic to be tuned. The I-PD controller was recommended
regulator (LQR)-based PID controller and a uniquely by Chakraborty et al. (2017) for the control of PIPDT
designed set point filer. Mercader and Baños (2017) processes. In that study, formulas in terms of gain mar-
proposed a method to adjust the parameters of a PI gin, phase margin and critical gain were provided to
controller for a PIPDT process in the presence of inter- set the parameters of the I-PD controller (Chakraborty
val parametric uncertainty. Their method is based on et al., 2017). Peker and Kaya (2019) gave optimally
a parameter space approach without using approxi- adjusted formulas for the I-PD controller parameters
mations for time delay terms and ignoring parame- to control PIPDT processes. In Kaya and Peker (2020),
ter uncertainties (Mercader & Baños, 2017). Verma I-PD adjusting formulas for controlling IFOPDT and
and Padhy (2018) presented two new integral per- double IFOPDT processes were presented to achieve
formance indices to obtain PI/PID adjusting rules optimal servo response. Kaya (2020) provided optimal
according to the desired Ms to control PIPDT and analytical setting rules for adjusting I-PD controller
stable/unstable first-order plus time-delay processes. parameters to control time-delay integrating processes
In Zhang et al. (2020), analytical rules for adjusting with an inverse response. In the works (Kaya, 2020;
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 3

Kaya & Peker, 2020; Peker & Kaya, 2019), robustness 2. Maximum sensitivity (Ms)
was not considered as a priority.
Ms is the maximum value of the magnitude of the sen-
In this paper, Ms-based I-PD controller designs for
sitivity function corresponding to the inverse of the
controlling PIPDT, DIPDT and IFOPDT processes are
shortest distance between the Nyquist curve and the
proposed. The advantages of the DS technique are
critical point (−1 + j0) (Anil & Padma Sree, 2015a).
used in the controller design. The design method is
Smaller Ms values represent more stable and robust
based on comparing the characteristic equation of the
systems. The sensitivity function of a classical or con-
closed-loop system, which comprises the integrating
ventional closed-loop control system with unity feed-
system and the I-PD controller with a lead/lag filter
back is given below (Anil & Padma Sree, 2015a):
cascaded to the PD part, with the desired characteris-
tic equation. I-PD tuning rules for PIPDT and DIPDT 1
S= . (1)
processes are suggested in terms of parameters of the 1 + Gp Gc
process transfer function and a setting parameter that
where the transfer function of the plant and the con-
is used for desired robustness specified by Ms. But,
troller is represented by Gp and Gc , respectively. Gp Gc
tuning rules for IFOPDT processes are designed to
denotes the open-loop transfer function. The ampli-
give robust responses so that a Ms value is obtained
tude of the sensitivity function S can be calculated
in the interval of 1.8 ≤ Ms ≤ 2. To summarise, the
as |S| = |1/(1 + Gp (jω)Gc (jω))| (Anil & Padma Sree,
contributions of this study include the following:
2015a). The maximum amplitude value of the S gives
the measure of robustness and has the following math-
i. The I-PD controller is designed according to the
ematical expression (Verma & Padhy, 2018):
desired robustness level specified by Ms and sim-
 
ple tuning rules are derived to calculate the con-  1 
Ms = max   . (2)
troller parameters. ω∈[0,+∞) 1 + Gp (jω)Gc (jω) 
ii. The works using the DS method in the design pro-
cedure of classical PID controllers for integrating The higher the Ms value, the more sensitive the con-
processes based on desired robustness stated by troller is to parameter uncertainties. As the Ms value
Ms require a set point weighting or set point filter decreases, the sensitivity of the controller to parameter
to decrease large overshoots in the servo response. uncertainties also decreases. The relationship between
On the contrary, it is demonstrated here by sim- phase margin (Pm), gain margin (Gm) and Ms is as
ulation examples and a practical application that follows (Anil & Padma Sree, 2015a):
the suggested I-PD design approach does not need Ms
such requirements. Gm ≥ (3)
Ms − 1
Pm ≥ 2sin−1 (1/2 Ms). (4)
The benefits of the proposed design approach are
illustrated by simulation examples and a real-time As it can be understood from (3) and (4), when the
implementation carried out by controlling a cart posi- appropriate Ms value is selected, a more appropriate
tion on a cart-pendulum set-up. Comparisons with phase margin and gain margin values required for a
some available PID and I-PD controller design tech- more robust response are also selected.
niques are provided to reveal the benefits of the sug-
gested design approach more obviously.
3. I-PD controller design
The rest of the paper is organised as follows:
Section 2 provides a short explanation of Ms. Section 3 The proposed control scheme, which consists of the
includes I-PD controller designs for PIPDT, DIPDT plant, I-PD controller and lead/lag filter, is depicted in
and IFOPDT processes in detail. Simulation exam- Figure 1.
ples are provided in Section 4 and Section 5 contains In the figure, the plant is represented by G(s). Gc1 (s)
performance analysis. A practical application of the and Gc2 (s) stand for the I and PD parts of the I-PD
suggested method, which is position control of a cart controller, respectively. F(s) denotes the lead/lag fil-
on a cart-pendulum set-up, is presented in Section 6. ter. Besides, in the figure, reference input, error signal,
Section 7 includes the conclusions. input disturbance and plant output are symbolised by
4 F. PEKER AND I. KAYA

to be
  
Ke−θs Kp Td s
(s) = 1 + + Kp 1 + = 0.
s Ti s fs + 1
(10)
Using 1/1 Padé approximation for time-delay term and
rearranging (10), the following equation was obtained.

Figure 1. The proposed I-PD control structure. Ti f θ s4 + [2Ti f + Ti θ − KKp Ti (f + Td )θ] s3

+ [2Ti + 2KKp Ti (f + Td ) − KKp (f + Ti )θ] s2
R(s), E(s), d(s) and Y(s), respectively. Transfer func-
+ [2KKp (f + Ti ) − KKp θ ] s + 2KKp = 0 (11)
tions for Gc1 (s), Gc2 (s) and F(s) are given as follows:
Kp The desired characteristic equation was chosen as
Gc1 (s) = (5) follows:
Ti s
Gc2 (s) = Kp (1 + Td s) (6) (βs + 1)4 = β 4 s4 + 4β 3 s3 + 6β 2 s2 + 4βs + 1 = 0 .
as + 1 (12)
F(s) = . (7) Now, comparing the corresponding coefficients of s4 ,
cs + 1
s3 , s2 and s in (11) and (12), the following four
For modelling integrating processes, the following equations were obtained.
general transfer function model is considered:
x2 x4
v4 − =0 (13)
Ke−θ s 2x1
G(s) = . (8)  
s(Ts + b) x2 x4 x2 x2 x4 x2 x3
4v3 − + − − =0 (14)
Case (i): if T = 0 and b = 1, then G(s) is a PIPDT x1 2x1 2 2
 
process. 2 x2
6v − + x2 x4 + x2 x3 − 0.5x4 − 0.5x2 = 0
Case (ii): if T = 1 and b = 0, then G(s) is a DIPDT x1
process. (15)
Case (iii): if T > 0 and b = 1, then G(s) is an
4v − [x4 + x2 − 0.5] = 0 (16)
IFOPDT process.
For the control of time-delay integrating processes where v = β/θ , x1 = KKp θ , x2 = Ti /θ , x3 = Td /θ
given above, as mentioned before, the DS approach is and x4 = f /θ. To obtain tuning rules for calculat-
used to design the I-PD controller with a lead/lag filter. ing the I-PD controller parameters, firstly, (13)-(16)
were solved for x1 = KKp θ , x2 = Ti /θ , x3 = Td /θ and
3.1. I-PD controller design for the PIPDT model x4 = f /θ for a selected value of β/θ varying in the
interval of 0.5 ≤ β/θ ≤ 1.9. Secondly, Ms values were
For the PIPDT system, Anil and Padma Sree (2015a) computed for each corresponding β/θ using solutions
stated that using a lead/lag filter does not yield good obtained from (13) to (16). Consequently, a relation-
closed-loop responses when compared to the exist- ship between Ms and β/θ was obtained, as shown
ing approaches in the literature. Furthermore, Zhang in Figure 2. In the figure, asterisks represent actual
et al. (2020) reported that using a lead/lag filter for the β/θ values and the solid line denotes results from
PIPDT process model does not improve closed-loop the obtained formula by employing curve fitting to
performances. Hence, the filter F(s) in Figure 1 has not Figure 2. A good fitting has been achieved as shown
been used in this case. However, the I-PD controller is in Figure 2.
assumed to have a derivative filter as given below To find setting rules for parameters of the controller,
  KKp θ , Ti /θ , Td /θ and f /θ were calculated for each cor-
Td s
Gc2 (s) = Kp 1 + . (9) responding β/θ, as displayed in Figure 3 by asterisks.
fs + 1
Then, formulas for KKp θ , Ti /θ , Td /θ and f /θ were
For the closed-loop system given in Figure 1, the char- derived in terms of β/θ by making use of curve fit-
acteristic equation for PIPDT processes was found ting. The results of attained formulas are represented
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 5

Figure 2. β/θ values versus Ms for the PIPDT process model.

Figure 4. Variations of Ms and ITAE/θ 2 according to β/θ for the

servo response of PIPDT processes.

Figure 3. KKpθ , Ti/θ, Td/θ and f /θ versus β/θ for PIPDT processes.

Figure 5. Variations of Ms and ITAE/Kθ 3 according to β/θ for the

by solid, dashed, dashed-dotted and dotted lines in regulatory response of PIPDT processes.
Figure 3. Very good fittings were achieved as shown in
Figure 3. As a result, according to the desired Ms, for-
mulas for computing parameters of the controller were β values. Therefore, a trade-off exists between output
gained in terms of parameters of the process model and performance and robustness.
a setting parameter β that is used for specifying the Tuning rules for the PIPDT processes, which are
desired Ms. derived by applying curve fitting to the results denoted
The relationship between β/θ and integral of time by asterisks in Figures 2 and 3, are provided below.
absolute error (ITAE) and the relationship between
β/θ and Ms for the PIPDT process model are displayed β 45.8(Ms)2 − 91.25(Ms) + 26.68
in Figures 4 and 5 for servo and regulatory responses, =
θ (Ms)3 + 98.55(Ms)2 − 298.1(Ms) + 219.5
respectively.
∀ 1.65 ≤ Ms ≤ 6.5 (17)
From Figures 4 and 5, it is seen that the variations of
ITAE/θ 2 for servo response and ITAE/Kθ 3 for regula- 0.2853(β/θ)2 − 0.7047(β/θ) + 1.221
KKp θ =
tory response are given for the variations of β/θ . This is (β/θ ) + 0.1776
because ITAE/θ 2 for servo response and ITAE/Kθ 3 for (18)
regulatory response remain constant for the same β/θ
Ti −259.9(β/θ)2 + 611.5(β/θ) + 82.8
for the PIPDT process model. Besides, from Figures 4 =
θ (β/θ )3 − 5.577(β/θ)2 − 51.69(β/θ) + 160.3
and 5, it is realised that for a PIPDT process model, if β
(19)
increases, the Ms value decreases, but the ITAE value
⎡ ⎤
increases. This means that selecting smaller β values −0.08179(β/θ)5 + 0.3186(β/θ)4
Td ⎣
results in faster but fewer robust responses, while more = −0.08877(β/θ)3 − 0.8041(β/θ)2 ⎦ (20)
θ
robust but slower responses are achieved by higher +0.6072(β/θ) + 0.2315
6 F. PEKER AND I. KAYA

f 343.4(β/θ)2 + 473.5(β/θ) − 146.5

= .
θ (β/θ )3 − 166.1(β/θ)2 − 599.8(β/θ) + 2821
(21)

Once the transfer function of the PIPDT process

is known, the I-PD controller (with derivative filter)
parameters can easily be calculated using (17)–(21) for
a desired value of Ms.

3.2. I-PD controller design for the DIPDT model

For controlling time-delay integrating systems that
can be modelled by the DIPDT transfer function, the Figure 6. β/θ values versus Ms for the DIPDT process model.
lead/lag filter in Figure 1 is used. Hence, the charac-
teristic equation of the closed-loop system for DIPDT
processes was obtained as follows: 10v2 − [x2 x4 + x2 x3 − 0.5(x2 + x5 )] = 0 (28)
Ke−θ s 5v − [x5 + x2 − 0.5] = 0 (29)
(s) = 1 +
s2
  
Kp as + 1 where v = β/θ, x1 = KKp θ 2 , x2 = Ti /θ , x3 = Td /θ ,
× + Kp (1 + Td s) = 0.
Ti s cs + 1 x4 = a/θ and x5 = c/θ . (25)-(29) were solved for
(22) x1 = KKp θ 2 , x2 = Ti /θ , x3 = Td /θ , x4 = a/θ and
x5 = c/θ for a chosen value of β/θ varying in the range
Similar to the PIPDT case, using 1/1 Padé approxi-
of 0.9 ≤ β/θ ≤ 2.6. Then, similar to the procedure
mation for time-delay term and rearranging (22), the
explained for the PIPDT case, Ms values were com-
following equation was found.
puted for each corresponding β/θ by using solutions
Ti cθ s5 + [2Ti c + Ti θ − KKp Ti Td aθ ] s4 found from (25) to (29). Thus, a relationship between
Ms and β/θ was found, as shown in Figure 6. In the
+ [2Ti + 2KKp Ti Td a − KKp θ(Ti a + Ti Td )] s3 figure, similar to the PIPDT case, asterisks denote
+ [2KKp (Ti a + Ti Td ) − KKp θ(c + Ti )] s2 actual β/θ values and the solid line denotes results
from the obtained formula by employing curve fitting
+ [2KKp (c + Ti ) − KKp θ] s + 2KKp = 0 (23)
to Figure 6.
In this case, the desired characteristic equation was To acquire setting formulas for parameters of the
selected as follows: controller with lead/lag filter, KKp θ 2 , Ti /θ , Td /θ , a/θ
and c/θ were calculated for each corresponding β/θ
(βs + 1)5 = β 5 s5 + 5β 4 s4 + 10β 3 s3 (see asterisks in Figure 7). After that, formulas for
KKp θ 2 , Ti /θ , Td /θ , a/θ and c/θ were derived in terms of
+ 10β 2 s2 + 5βs + 1 = 0 . (24)
β/θ using the curve fitting technique. The results from
Comparing corresponding coefficients of s5 , s4 , s3 , s2 attained formulas are represented by solid, dashed,
and s in (23) and (24), the following five equations were dashed-dotted and dotted lines in Figure 7. Fig-
obtained: ures 6 and 7 demonstrate that quite good fittings were
x2 x5 achieved. Consequently, consistent with the desired
v5 − =0 (25) Ms, formulas for determining parameters of the con-
2x1
  troller with lead/lag filter were gained in terms of
4 x2 x5 x2
5v − + − 0.5(x2 x3 x4 ) = 0 (26) parameters of the process model and a tuning param-
x1 2x1 eter β that is exploited for stating the desired Ms.
 
x2 For the servo response, ITAE/θ 2 and for the regula-
10v3 − + x2 x3 x4 − 0.5(x2 x4 + x2 x3 ) = 0
x1 tory response, ITAE/Kθ 4 remain constant for the same
(27) β/θ value. Variations of ITAE/θ 2 and Ms versus β/θ
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 7

too, more robust but slower responses are achieved

by higher β values, while faster but fewer robust
responses are acquired by smaller β values.
Tuning rules for the DIPDT process transfer func-
tion model, which are achieved by applying curve fit-
ting to the results symbolised by asterisks in Figures 6
and 7, are given as follows:

10.94(Ms)3 − 55.41(Ms)2
β +107.4(Ms) − 78.4
=
Figure 7. KKpθ 2 , Ti/θ , Td/θ , a/θ and c/θ versus β/θ for DIPDT θ (Ms)4 + 2.462(Ms)3
processes. −25.49(Ms)2 + 49.67(Ms) − 32
∀ 1.9 ≤ Ms ≤ 4.5 (30)
0.08747(β/θ)2 − 0.3276(β/θ) + 0.6441
KKp θ 2 =
(β/θ ) − 0.2445
(31)
Ti −1385(β/θ)2 + 4315(β/θ ) + 475.4
=
θ (β/θ )3 − 8.251(β/θ)2 − 248.8(β/θ) + 896
(32)
⎡ ⎤
0.2242(β/θ)5 − 1.62(β/θ)4
Td ⎣
= +4.873(β/θ )3 − 7.07(β/θ )2 ⎦ (33)
θ
+4.912(β/θ ) − 0.7978
Figure 8. Variations of Ms and ITAE/θ 2 according to β/θ for the a −29.39(β/θ )2 + 116.6(β/θ) − 16
servo response of DIPDT processes. = (34)
θ (β/θ )2 − 14.4(β/θ ) + 51.34
c 313.9(β/θ)2 + 238(β/θ ) − 171.2
= . (35)
θ (β/θ )3 − 303.7(β/θ)2
+398.7(β/θ) + 2051

It should be noted that for simulation responses

under measurement noise and real-time applications,
a derivative filter may be used in the derivative por-
tion of the controller as Kp [1 + Td s/(1+(Td /N)s)] with
N = 10 to reduce the noise effects. Finally, once the
Figure 9. Variations of Ms and ITAE/Kθ 4 according to β/θ for the transfer function that belongs to the DIPDT process
regulatory response of DIPDT processes. is known, the parameters of the I-PD controller with a
lead/lag filter can simply be calculated using (30)–(35)
for a selected value of Ms.
for the servo response are depicted in Figure 8. Vari-
ations of ITAE/Kθ 4 and Ms versus β/θ for regulatory
response are illustrated in Figure 9.
3.3. I-PD controller design for the IFOPDT model
Analogous to the case of the PIPDT model, it is
observed from Figures 8 and 9 that for a DIPDT pro- Here, the I-PD controller with a lead/lag filter shown
cess model, when β increases, the Ms value decreases in Figure 1 was used for controlling the IFOPDT pro-
but the ITAE value also increases. Hence, here cesses. The characteristic equation of the closed-loop
8 F. PEKER AND I. KAYA

Figure 10. β/θ values versus θ/T for the IFOPDT process model.

 
system for IFOPDT processes was found as follows: 2 x2
10v − + x2 x4 + x2 x3 − 0.5z(x5 + x2 ) = 0
x1
Ke−θ s (41)
(s) = 1 +
s(Ts + 1)
   5v − [x5 + x2 − 0.5z] = 0 (42)
Kp as + 1
× + Kp (1 + Td s) = 0.
Ti s cs + 1 where z = θ /T, v = β/T, x1 = KKp T, x2 = Ti /T,
(36) x3 = Td /T, x4 = a/T and x5 = c/T. For different val-
ues of θ /T in the interval of 0.07 ≤ θ /T ≤ 3, (38)-(42)
Similar to previous cases, a 1/1 Padé approximation were solved for x1 = KKp T, x2 = Ti /T, x3 = Td /T,
for the time-delay term was used in (36) to find the x4 = a/T and x5 = c/T according to β/T which was
following equation: adjusted to achieve Ms in the range of 1.8 ≤ Ms ≤ 2.
Namely, for each θ /T value, β/T was adjusted to obtain
Ti Tcθ s5 + [Ti T(θ + 2c) + Ti cθ − KKp Ti Td aθ ] s4
suitable KKp T, Ti /T, Td /T, a/T and c/T values so that
+ [2Ti T + Ti (θ + 2c) + 2KKp Ti Td a 1.8 ≤ Ms ≤ 2 was reached for the closed-loop con-
− KKp θ(Ti a + Ti Td )] s3 trol system. Then, β/T was defined in terms of θ /T
using the curve fitting technique. Finally, formulas for
+ [2Ti + 2KKp (Ti a + Ti Td ) − KKp θ(c + Ti )] s2 KKp T, Ti /T, Td /T, a/T and c/T were derived in terms
+ [2KKp (c + Ti ) − KKp θ] s + 2KKp = 0 (37) of β/T by making use of curve fitting. Eventually, for
the IFOPDT process model, tuning formulas give the
As in the case of the DIPDT process model, the desired controller parameters that result in a robust response
characteristic equation was chosen as (24) and then for which the Ms value in the range of 1.8 ≤ Ms ≤ 2
corresponding coefficients of s5 , s4 , s3 , s2 and s in (24) is achieved. Three intervals were defined for θ /T for a
and (37) were compared to obtain the following five better curve fitting, and formulas for β/T, KKp T, Ti /T,
equations: Td /T, a/T and c/T were obtained for each interval.
For all intervals, solutions of (38)–(42) and formulas
x2 x5 z
v5 − =0 (38) gained from curve fittings are illustrated in Figures 10
2x1
  and 11.
4 x2 (z + 2x5 ) x2 x5 z Once more, the obtained results from solutions of
5v − + − 0.5x2 x3 x4 z = 0
2x1 2x1 (38)–(42) are represented by asterisks and derived for-
(39) mulas are denoted by solid, dashed, dashed-dotted and

x2 x2 (z + 2x5 ) dotted lines in Figures 10 and 11.
10v3 − + + x2 x3 x4 For the first interval of 0.07 ≤ θ /T ≤ 0.9, by apply-
x1 2x1
ing curve fitting to the results depicted by asterisks
− 0.5zx2 (x4 + x3 )] = 0 (40)
in Figures 10(a) and 11(a), the following formulas to
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 9

Figure 11. KKpT, Ti/T, Td/T, a/T and c/T versus β/T for the IFOPDT process model.

evaluate the parameters of the I-PD controller with a I-PD controller with a lead/lag filter were obtained as
lead/lag filter were found. follows:
⎡ ⎤  
0.4541(θ/T)3 + 0.1906(θ/T)2 0.03208(θ/T)4 + 0.7188(θ/T)3
⎣+1.485(θ/T) ⎦ β −1.053(θ/T)2 − 0.8711(θ/T) + 1.406
=
β +0.02199 T (θ/T)2 − 2.455(θ/T) + 1.586
= (43)
T (θ/T) + 0.6758
  (49)
2.229(β/T)4 − 8.867(β/T)3
 
+13.95(β/T)2 − 10.23(β/T) + 4.863 0.4299(β/T)4 + 0.5291(β/T)3
KKp T =
(β/T)2 + 1.355(β/T) − 0.1334 −4.776(β/T)2 + 0.2428(β/T) + 7.901
KKp T =  
(44) (β/T)5 − 2.092(β/T)4 − 1.767(β/T)3
  +2.366(β/T)2 + 2.846(β/T) + 1.424
Ti −0.2772(β/T)4 + 0.4989(β/T)3
= (50)
T −0.308(β/T)2 + 5.104(β/T) − 0.127
(45)  
  7.375(β/T)5 − 26.03(β/T)4 + 14.12(β/T)3
Td 0.6253(β/T)4 − 1.528(β/T)3 Ti +20.01(β/T) 2
+ 3.007(β/T) − 17.61
= =  
T +1.541(β/T)2 − 0.2514(β/T) + 0.1577 T (β/T) − 5.644(β/T)4 + 11.81(β/T)3
5

(46) −7.474(β/T)2 − 10.19(β/T) + 14.21

  (51)
a −0.6562(β/T)4 + 2.149(β/T)3
=  
T −3.082(β/T)2 + 2.447(β/T) − 0.0901
0.2977(β/T)3 − 0.8144(β/T)2
(47) −0.1784(θ/T) + 1.246
Td
⎡ ⎤ = (52)
−0.7822(β/T)5 + 2.951(β/T)4 T (β/T)2 − 3.86(β/T) + 3.734
c
= ⎣−4.03(β/T)3 ⎦ ⎡ ⎤
T 2.219 exp{−(((β/T) − 1.663)/0.1625)2 }
+2.638(β/T) − 0.4683(β/T) + 0.1735
2
⎢+5.642 exp{−(((β/T) − 4.033)/1.959)2 }⎥
(48) ⎢ ⎥
a ⎢+0.7788 exp{−(((β/T) − 1.859) ⎥
=⎢
⎢
⎥
⎥
For the second interval of 0.9 < θ/T ≤ 2, similar T ⎢/0.1033) }2
⎥
to the first interval, by applying curve fitting to the ⎣+0.1435 exp{−(((β/T) − 2.089) ⎦
results represented by asterisks in Figures 10(b) and /0.08907)2 }
11(b), formulas to calculate the parameters of the (53)
10 F. PEKER AND I. KAYA

⎡ ⎤
3.588 exp{−(((β/T) − 1.551) Similar to the DIPDT case, for simulation responses
⎢/0.7163)2 } ⎥ under measurement noise and real-time applications,
⎢ ⎥
⎢+2.837 exp{−(((β/T) − 2.679) ⎥ a derivative filter may be used in the derivative por-
⎢ ⎥
⎢ ⎥ tion of the controller as Kp [1 + Td s/(1+(Td /N)s)] with
c ⎢ /0.2964)2}
⎥
=⎢ ⎥. (54) N = 10 to suppress the noise. So, once the transfer
T ⎢−0.4349 exp{−(((β/T) − 1.978)⎥
⎢ ⎥
⎢/0.0886)2 } ⎥ function that belongs to the IFOPDT process is known,
⎢ ⎥
⎣+0.8416 exp{−(((β/T) − 2.331)⎦ the parameters of the I-PD controller with a lead/lag
/0.2213)2 } filter can readily be calculated using (43)–(60).

Finally, for the third interval of 2 < θ/T ≤ 3, similar to 4. Simulation examples
the previous two intervals, by applying curve fitting to
the results signified by asterisks in Figures 10(c) and Three simulation examples are provided in this section
11(c), formulas for computing the parameters of the to demonstrate the utility and benefits of the suggested
I-PD controller with a lead/lag filter were derived as design method. Comparisons with some reported PID
follows: and I-PD design approaches are presented. For a fair
comparison, neither set point weight nor set point fil-
⎡ ⎤ ter was used in all design methods. In the simulation
2.952 exp{−(((θ/T) − 3.139)/4.089)2 }
⎢+0.01361 exp{−(((θ/T) − 2.242) ⎥ examples, the legends in the control signals figures are
β ⎢ ⎥
⎢
= ⎢/0.0975) }
2 ⎥ omitted because they are the same as the legends in the
T ⎣ ⎥
+0.0782 exp{−(((θ/T) − 2.206) ⎦ output response figures of the related example.
/0.3972) }
2

(55) 4.1. Example 1

⎡ ⎤ A PIPDT process studied by Chakraborty et al. (2017)
0.3236 exp{−(((β/T) − 2.786)
⎢/0.3036)2 } ⎥ is taken into consideration:
⎢ ⎥
⎢+0.0096 exp{−(((β/T) − 2.915)⎥ e−s
KKp T = ⎢
⎢/0.0084)2 }
⎥ (56)
⎥ G(s) = .
⎢ ⎥ s
⎣+0.0211 exp{−(((β/T) − 2.897)⎦
/0.03283)2 } I-PD controller parameters recommended by Chak-
⎡ ⎤ raborty et al. (2017) were given as Kp = 0.844,
3.938 exp{−(((β/T) − 2.953) Ti = 3.55 and Td = 0.482. For the proposed I-PD
⎢/0.0367)2 } ⎥
⎢ ⎥ design approach for the above PIPDT model, tun-
⎢
Ti ⎢−0.443 exp{−(((β/T) − 2.916)⎥⎥ ing formulas provided in (17)-(21) were used to
=⎢ ⎥ (57)
⎢/0.0029) }
2
T ⎥ determine the controller parameters as Kp = 0.6340,
⎣+9.702 exp{−(((β/T) − 2.916)⎦
Ti = 4.4147, Td = 0.1533 and f = 0.3814 for Ms = 2.
/0.3485)2 } Anil and Padma Sree (2015a) suggested a PID con-
⎡ ⎤
6.675 exp{−(((β/T) − 2.795) troller to control the PIPDT processes. According
Td ⎢ /0.1569)2 } ⎥ to formulas provided in their work, PID parame-
=⎢
⎣+1.415 exp{−(((β/T) − 2.9) ⎦
⎥ (58) ters for Ms = 2 (for a fair comparison with the pro-
T
/0.02957)2 } posed design method) were calculated as Kp = 0.8352,
⎡ ⎤ Ti = 4.0976 and Td = 0.3556. Zhang et al. (2020) also
0.2795 exp{−(((β/T) − 2.959)
⎢/0.03181)2 } ⎥ recommended a PID controller for controlling PIPDT
a
=⎢
⎣
⎥ (59) processes. Parameters of the PID for Ms = 2 (for a fair
T +1.171 exp{−(((β/T) − 3.024)⎦
comparison with the proposed design method) were
/0.8031)2 }
found using tuning rules provided in their study to
⎡ ⎤
6.403 exp{−(((β/T) − 2.796) be Kp = 0.8916, Ti = 4.3673 and Td = 0.4267. A unit
c ⎢/0.1613)2 } ⎥ step reference input change at t = 0 s, and a step input
=⎢ ⎥ . (60)
T ⎣+1.348 exp{−(((β/T) − 2.897)⎦ disturbance, the magnitude of which is 0.1, at t = 40 s
/0.02995)2 } were inserted into all closed-loop systems. The results
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 11

Figure 12. Closed-loop responses for the nominal case of Figure 15. Control signals for +25% changes in parameters of
example 1. the process model of example 1.

Figure 13. Control signals for the nominal case of example 1.

Figure 16. Closed-loop responses for −25% changes in parame-
ters of the process model of example 1.

Figure 14. Closed-loop responses for +25% changes in parame-

ters of the process model of example 1. Figure 17. Control signals for −25% changes in parameters of
the process model of example 1.

for the nominal case are demonstrated in Figures 12

and 13 for closed-loop responses and related control For closed-loop responses in the nominal case
signals, respectively. Output responses and associated depicted in Figure 12, it can be seen that although the
control signals of closed-loop systems are illustrated in design method recommended by Chakraborty et al.
Figure 14 and 15, respectively, for +25% changes, and (2017) and the proposed one seem to have similar
in Figure 16 and 17 for −25% changes in all process performances for the servo response, the proposed
model parameters, namely the gain and time delay. design method has no overshoot when the figure is
12 F. PEKER AND I. KAYA

well examined. The design methods of Anil and Padma

Sree (Anil & Padma Sree, 2015a) and Zhang et al.
(2020) have excessive overshoot for set point track-
ing. Inspecting control signals in the nominal case
depicted in Figure 13, it is realised that for the servo
responses, the proposed design method has a lesser
initial control effort than others and design methods
recommended by Anil and Padma Sree (2015a) and
Zhang et al. (2020) have very large and aggressive ini-
tial control efforts. Also, it can be inferred from control
signals that the proposed design method and design
method recommended by Chakraborty et al. (2017) Figure 18. Closed-loop responses for the nominal case of
consist of plausible and admissible control signals. example 2.
Taking up the responses for +25% changes in param-
eters of the process model demonstrated in Figure 14,
it can be said that the proposed design method and the
design method suggested by Chakraborty et al. (2017)
were affected approximately in the same way. It is also
clear that other design methods are more influenced by
+25% changes in the parameters of the process model.
For responses in the case of −25% changes in param-
eters of the process model, analogous to the case of
+25% changes, it can be deduced that the proposed
design method and the design method recommended
by Chakraborty et al. (2017) were almost similarly
affected.
Figure 19. Control signals for the nominal case of example 2.

4.2. Example 2
model were found to be Kp = 0.2370, Ti = 14.4608,
A process model, which has fourth-order transfer Td = 1.5596, α = 0.9996 and β = 1.2888. Similar to
function with one integrator, is considered here: Anil and Padma Sree (2015a), Zhang et al. (2020) sug-
gested a PID with a lead/lag filter for the IFOPDT
e−2s
G(s) = . process model. Parameters of the PID with a lead/lag
s(s + 1)(0.5s + 1)(0.25s + 1) filter for the IFOPDT process model were evaluated as
Taylor series expansion was used to obtain PIPDT Kp = 0.3647, Ti = 10.4462, Td = 0.9345, α = 1.7398
and IFOPDT process models as G(s) ≈ e−3.75s /s and and β = 0.2351 based on the formulas presented by
G(s) ≈ e−2.75s /s(s + 1), respectively. For the proposed Zhang et al. (2020). A unit step reference input change
I-PD design technique for the IFOPDT model, since at t = 0 s, and a step input disturbance, the magni-
θ /T = 2.75, formulas (55)-(60) were exploited to com- tude of which is 0.1, at t = 100 s were injected into
pute parameters of the suggested I-PD controller all closed-loop systems. The results for the nominal
with a lead/lag filter as Kp = 0.2570, Ti = 12.9525, case are depicted in Figures 18 and 19 for closed-loop
Td = 3.2148, a = 1.3337 and c = 3.1877. I-PD con- responses and corresponding control signals, respec-
troller parameters for the PIPDT model suggested tively. Output responses and related control signals of
by Chakraborty et al. (2017) were found to be closed-loop systems are shown in Figures 20 and 21
Kp = 0.2249, Ti = 13.3246 and Td = 1.8108. Anil and for +25% changes and in Figures 22 and 23 for −25%
Padma Sree (2015a) recommended a PID with a changes in all parameters, i.e. the gain, time constants,
lead/lag filter for the IFOPDT process model. Param- and time delay of the process model.
eters of the PID with a lead/lag filter suggested by It is apparent from Figure 18 that the proposed
Anil and Padma Sree (2015a) for the IFOPDT process design method shows the best performance for the
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 13

Figure 20. Closed-loop responses for +25% changes in parame-

Figure 23. Control signals for −25% changes in parameters of
ters of the process model of example 2.
the process model of example 2.

suggested by Zhang et al. (2020) gives a very large ini-

tial control effort for the servo response. The proposed
design method and the design method recommended
by Chakraborty et al. (2017) have similar and rea-
sonable initial control efforts for set point tracking.
Analysing responses for +25% changes in parameters
of the process model demonstrated in Figure 20, it can
be observed that the proposed design method and the
design method recommended by Chakraborty et al.
(2017) seem to be the least and approximately equally
influenced. It is also realised that the design method
Figure 21. Control signals for +25% changes in parameters of suggested by Zhang et al. (2020) is the most influenced
the process model of example 2. one for +25% of changes in process model parame-
ters. For responses in the case of −25% changes in
parameters of the process model, similar to the case
of +25% changes, it can be inferred that the proposed
design method and the design method recommended
by Chakraborty et al. (2017) were affected closely.

4.3. Example 3
Consider a process transfer function, which has double
integrators, given below:
e−s
G(s) = .
s2
Figure 22. Closed-loop responses for −25% changes in parame-
ters of the process model of example 2.
For the proposed I-PD controller for controlling
DIPDT processes, tuning rules given in (30)–(35) were
used to compute tuning parameters of Kp = 0.1319,
servo response. For the nominal case of regulatory Ti = 9.9631, Td = 1.1468, a = 3.9148 and c = 1.1303
responses, although the design method of Zhang et al. for Ms = 2. Parameters of serial PID controller for
(2020) has a less overshoot and settling time than the DIPDT model recommended by Grimholt and
other design methods, it gives excessive overshoot for Skogestad (2016) were calculated as Kp = 0.0625,
set point tracking. From the control signals shown in Ti = 8.64 and Td = 8.76 for Ms = 2 (for a fair com-
Figure 19, it can be observed that the design method parison with the proposed design method). Anil and
14 F. PEKER AND I. KAYA

Figure 24. Closed-loop responses for the nominal case of Figure 25. Control signals for the nominal case of example 3.
example 3.

Padma Sree (2015a) recommended a PID with a

lead/lag filter for DIPDT process models. Param-
eters of the PID with lead/lag filter suggested by
Anil and Padma Sree (2015a) for Ms = 2 (for a
fair comparison with the proposed design method)
were found according to the formulas presented
by them as Kp = 0.1341, Ti = 9.8887, Td = 3.8752,
α = 1.1610 and β = 1.1321. Zhang et al. (2020), sim-
ilar to Anil and Padma Sree (2015a), recommended
a PID with a lead/lag filter having parameters of
Kp = 0.1768, Ti = 9.5912, Td = 3.5403, α = 0.5039 Figure 26. Closed-loop responses for +25% changes in parame-
and β = 0.2638, which were found for Ms = 2 (for ters of the process model of example 3.
a fair comparison with the proposed design method)
based on equations given in their work. A unit step
reference input change at t = 0 s, and a step input dis-
turbance, the magnitude of which is 0.1, at t = 100 s
was inserted into all closed-loop systems. The results
for the nominal case are shown in Figures 24 and 25 for
closed-loop responses and associated control signals,
respectively. Figures 26 and 27, respectively, depict
output responses and corresponding control signals of
closed-loop systems for +25% changes in all param-
eters (the gain and time delay) of the process model.
Figures 28 and 29 give similar results for the case of
−25% changes in all parameters (the gain and time Figure 27. Control signals for +25% changes in parameters of
delay) of the process model. the process model of example 3.
It is clear from Figure 24 that the proposed design
method has better performance for the servo response.
For the nominal case of regulatory responses, although control effort for the servo response. Also, it can easily
the design method of Zhang et al. (2020) has a be realised that the proposed design method has the
less overshoot than other design methods, its servo lowest initial control effort for the servo response. For
response has a very large overshoot. From the nom- closed-loop responses in the case of parameter changes
inal case of control signals illustrated in Figure 25, it shown in Figures 26 and 28, it can be said that the
is obviously observed that the design method recom- proposed design method seems to be the least affected
mended by Zhang et al. (2020) has the largest initial among all.
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 15

Table 1. Performance analyses of example 1.

Servo Regulatory

Method ITAE TV OS ITAE TV OS

Proposed PIPDT for 11.86 0.4666 0 3.075 0.1864 0.1625
Ms = 2
Chakraborty et al. 8.868 0.5828 0.0256 1.762 0.2159 0.1198
Anil and Padma 8.863 1.6310 0.4128 2.01 0.1942 0.1278
Sree PIPDT for
Ms = 2
Zhang et al. PIPDT 9.157 1.9934 0.3586 2.139 0.2138 0.1205
for Ms = 2

Figure 28. Closed-loop responses for −25% changes in parame- Table 2. Performance analyses of example 2.
ters of the process model of example 3. Servo Regulatory

Method ITAE TV OS ITAE TV OS

Proposed IFOPDT 99.88 0.1729 0 65.34 0.1865 0.42
Chakraborty et al. 125.5 0.1504 0.0268 93.7 0.1575 0.4224
Anil and Padma 119 0.4182 0.4568 86.57 0.1766 0.4466
Sree IFOPDT
Zhang et al. IFOPDT 53.24 3.5975 0.5436 34.24 0.2417 0.2823

Table 3. Performance analyses of example 3.

Servo Regulatory

Method ITAE TV OS ITAE TV OS

Proposed DIPDT for 60.1 0.0704 0 75.26 0.2370 0.7340
Ms = 2
Grimholt and 76.74 0.1288 0 240.5 0.2094 0.7445
Skogestad for
Figure 29. Control signals for −25% changes in parameters of Ms = 2
the process model of example 3. Anil and Padma 53.52 0.3128 0.4123 73.13 0.2375 0.7216
Sree DIPDT for
Ms = 2
Zhang et al. DIPDT 45.57 0.5047 0.3350 53.33 0.2143 0.5150
5. Performance analysis for Ms = 2

Performance analyses for all design methods given in

examples 1, 2 and 3 are presented in this section. In design method has somewhat larger ITAE values, it has
performance analyses, nominal cases were taken into smaller TV values than other design methods. Also,
consideration and performance summaries for servo from Table 1, though the proposed design method
and regulatory responses were presented separately. In results in a somewhat larger overshoot for the reg-
the performance analysis of all design methods, ITAE ulatory response, it has no overshoot for the servo
of error signal e(t), total variation (TV) of input signal response. Looking at the ITAE values for servo and
u(t) and maximum amount of overshoot (OS) of out- regulatory responses given in Table 2, it is seen that the
put y(t) were used. TV is a good criterion to evaluate design method recommended by Zhang et al. (2020)
the smoothness of the control (or input) signal (Anil & has smaller ITAE values than others, but it results in a
Padma Sree, 2015a; Skogestad, 2003). Lesser TV values very large overshoot for the servo response. From the
represent smoother input signals that are looked for TV values of servo responses provided in Table 2, it
in control systems, especially in practical applications. is realised that the proposed design method and the
Calculation of the TV belongs to an input signal given design method suggested by Chakraborty et al. (2017)

by TV = ∞ n=0 |un+1 − un |, where un is the input sig- have similar and reasonable values, but the design
nal at an nth instant (Anil & Padma Sree, 2015a). method recommended by Zhang et al. (2020) has a
Tables 1–3 give performance analyses of examples 1, very high value which means it can cause problems
2 and 3, respectively. in practice. Considering TV values belong to regula-
Examining Table 1 both for servo and regula- tory responses in Table 2, it is once again seen that
tory responses, it is seen that although the proposed the design method suggested by Zhang et al. (2020)
16 F. PEKER AND I. KAYA

Figure 30. Cart-pendulum set-up.

has a higher value than others and other methods 6. Practical application on an experimental
have values close to each other. For the amounts of set-up
overshoots presented in Table 2, it takes the attention
Cart-pendulum set-up manufactured by Feedback
that proposed design method has no overshoot for
Instruments (Feedback Instruments Ltd., 2010) was
the servo response. For the overshoots of regulatory
used to demonstrate real-time application of the sug-
responses given in Table 2, the design method recom-
gested I-PD design method. The set-up (see Figure 30)
mended by Zhang et al. (2020) has a lesser overshoot
comprises a cart and two pendulum arms that are con-
since it results in a very fast and large overshoot for
nected to the cart. The cart moves on a 1-meter track
the servo response, and the remaining design meth-
by pulling a belt attached to it. At the end of the rail, a
ods have close values of overshoots. Analysing Table 3,
dc motor is available. By pulling the belt in two direc-
it is observed that the design method recommended
tions via the dc motor, the movement of the cart is
by Zhang et al. (2020) has a smaller ITAE value than
assured. The value of the force applied to the cart is
others for both servo and regulatory responses. How-
determined by the value of the voltage applied to the dc
ever, similar to the case of example 2 given in Table 2, it
motor. That’s why the input variable or control signal
has a very large overshoot for set point tracking. Han-
of the set-up is the voltage. On the set-up, for mea-
dling TV values of servo responses given in Table 3,
suring cart position and pendulum angle, there exist
it is clearly realised that the proposed design method
two optical encoders. Some control implementations,
has the smallest TV value. When the TV values that
for example, cart position control, crane control or sta-
belong to the regulatory responses in Table 3 are exam-
bilisation of the pendulum at the upright position are
ined, it can be observed that all design methods have
performed by applying an appropriate voltage to the
values that are not far from each other. Looking at
dc motor (Feedback Instruments Ltd., 2010). The con-
the amount of overshoots provided in Table 3, it once
trol scheme that belongs to the cart-pendulum set-up
again draws attention that there is no overshoot in the
is displayed in Figure 31.
servo response of the proposed design method. The
In this study, only the cart position control was car-
design method recommended by Grimholt and Sko-
ried out. Since pendulum arms swing when the cart
gestad (2016) has also no overshoot for servo response,
moves, there exists a disruptive effect on the position
but it has a longer settling time than the proposed
control of the cart due to swings of pendulum arms.
one (see Figure 24). For the overshoots of regulatory
So, swings of pendulum arms were considered as dis-
responses provided in Table 3, similar comments made
tortion. The cart position has an integrating behaviour
for the case of example 2 deduced from Table 2 can be
and was identified from many input-output data that
derived.
belong to the cart as in the following (Peker & Kaya,
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 17

Figure 31. The control scheme belonging to the cart-pendulum

set-up.

2017):
Figure 32. Closed-loop responses for real-time application on
0.64394 the cart-pendulum set-up.
G(s) = .
s(0.13605s + 1)
It was assumed that there is 0.1 s time delay in the
system. So, the system can be handled as an IFOPDT
system as G(s) ≈ 0.64394e−0.1s /s(0.13605s + 1) and
the ratio of the θ /T was found as 0.7350. Formulas pro-
vided in (43)-(48) were used to compute parameters of
the proposed I-PD controller with a lead/lag filter as
Kp = 10.1475, Ti = 0.6587, Td = 0.0731, a = 0.1043
and c = 0.0647. Parameters of the PID with a lead/lag
filter suggested by Anil and Padma Sree (2015a) were
found as Kp = 10.9602, Ti = 0.4705, Td = 0.0487,
α = 0.1359 and β = 0.0373. PID with a lead/lag fil-
ter recommended by Zhang et al. (2020) was adjusted
according to the formulas provided in their work to be Figure 33. The control signals for real-time application on the
cart-pendulum set-up.
Kp = 14.0453, Ti = 0.5719, Td = 0.1012, α = 0.0532
and β = 0.0130. For the design method suggested
by Chakraborty et al. (2017), a PIPDT model is with a magnitude of 0.3 metre and a period of 100 s,
required. Hence, by taking advantage of the Tay- was applied to the cart. Also, a unity step input distur-
lor series expansion, the PIPDT model was iden- bance was inserted into the system at t = 80 s. Cart
tified as G(s) ≈ 0.64394e−0.23605s /s. For this identi- position responses and associated control signals are
fied PIPDT model, I-PD controller parameters sug- illustrated in Figures 32 and 33, respectively. It should
gested by Chakraborty et al. (2017) were found be pointed out that legends in the figure of control sig-
to be Kp = 5.5492, Ti = 0.8395 and Td = 0.1140. nals, namely Figure 33, are the same as the legends in
Also, as mentioned before, a derivative filter was Figure 32.
used in derivative part of the proposed controller as From real-time responses shown in Figure 32, it can
Kp [1 + Td s/(1+(Td /N)s)] with N = 10 to suppress the be seen that although the design method suggested by
noise. The same derivative filter was also used in Chakraborty et al. (2017) has a close response to the
the design methods of Anil and Padma Sree (2015a) proposed one for set point tracking, it has a large over-
and Zhang et al. (2020) since they do not contain shoot for disturbance rejection. Also, from Figure 32
a derivative filer. The design method recommended it can be observed that though the design methods of
by Chakraborty et al. (2017) has already its deriva- Anil and Padma Sree (2015a) and Zhang et al. (2020)
tive filter. Similar to simulation examples, for a fair have a somewhat lesser overshoot than the proposed
comparison, in all design methods neither set point design method for disturbance rejection, they give a
weighting nor set point filter was used. For all design large overshoot for set point tracking. When looking at
methods, a reference trajectory, which is a pulse input the control signals displayed in Figure 33, it takes the
18 F. PEKER AND I. KAYA

attention that design methods recommended by Anil future study, a 2DOF control structure including two
and Padma Sree (2015a) and Zhang et al. (2020) result I-PD controllers, one for the set point tracking and
in very large initial control efforts for set point track- the other for disturbance rejection, can separately be
ing. Owing to this situation, separate windows had to designed based on desired Ms value so that more satis-
be opened to see other control signals. Furthermore, factory closed-loop responses can be obtained for both
it can be deduced from Figure 33 that the proposed the set point tracking and disturbance rejection.
design method has reasonable and acceptable control
signal values for both set point tracking and distur-
bance rejection. Besides, as it is mentioned before, the Acknowledgements
input variable or control signal of the set-up is the Authors would like to thank Dicle University for providing
voltage and it is seen from Figure 33 that very large cart-pendulum set-up to perform real-time applications.
control signal values occur for the design methods of
Anil and Padma Sree (2015a) and Zhang et al. (2020).
It may rightly be asked whether this does not harm the Disclosure statement
system. The answer is yes, but here the system is pro- No potential conflict of interest was reported by the author(s).
tected by a saturation block in the software that does
not allow voltages to be applied to the system outside
the range of 0–5 volts. When the control signals are Data availability statement
analysed, it becomes clear that some vibrations may be
present in the control signals. This is due to the swing- Data sharing is not applicable to this article as no
ing of the pendulum arms, which occurs when the cart datasets were generated or analysed during the current
moves as explained earlier. Therefore, these vibrations study.
appear in the control signals to immobilise the cart in
the set position.
Notes on contributors

7. Conclusion Fuat Peker was born in Konya, Turkey. He received a

B.Sc. degree in electrical-electronics engineering from Dokuz
Ms-based simple I-PD controller tuning rules for con- Eylül University, İzmir, Turkey in 2012 and an M.Sc. degree
trolling PIPDT, DIPDT and IFOPDT processes have in electrical-electronics engineering from Dicle University,
Diyarbakır, Turkey in 2017. He is now a Ph.D. student at the
been derived. The DS technique was used in the con-
department of electrical-electronics engineering at Dicle Uni-
troller designs. The advantage of the proposed design versity Institute of Science. He currently works as a research
method was presented by three simulation examples. assistant at Dicle University. His research interests include
Simulation results reveal that the proposed design proportional–integral–derivative controller, optimal control,
method has important advantages over other design robust control, direct synthesis approach, system modelling
methods, especially methods that use the DS tech- and identification.
nique. In the proposed design approach, there is no Ibrahim Kaya was born in Diyarbakır, Turkey, in 1971. He
need to use a set point weighting or set point filter, received a B.Sc. degree in electrical and electronics engineer-
which are utilised by other design methods that use the ing from Gaziantep University, Gaziantep, Turkey in 1994 and
DS technique, to decrease large overshoots at the servo a D.Phil. degree in control engineering from the University of
Sussex, Brighton, UK in 1999. He is currently working as a
response. The reason for this is that the proposed I-PD
professor in the department of electrical and electronics engi-
design method yields no/almost no overshoot for the neering at Dicle University, Diyarbakır, Turkey. His primary
servo response. A practical application of the proposed research interest lies in system modelling, identification tech-
design approach was demonstrated by controlling a niques and model-based controller design methods in different
cart position on an experimental set-up. Although the control structures.
set point tracking responses of the proposed controller
design method are very satisfactory, its disturbance
rejection performance is slightly poor since a single ORCID
controller is used for both the set point tracking and Fuat Peker http://orcid.org/0000-0003-0611-225X
disturbance rejection. To overcome this situation, as a Ibrahim Kaya http://orcid.org/0000-0002-8393-1358
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 19

References of Measurement and Control, 42(16), 3123–3134. https://doi.

org/10.1177/0142331220941657
Ajmeri, M., & Ali, A. (2015). Direct synthesis based tuning
Kaya, I., & Cengiz, H. (2017). Optimal tuning of PI/PID con-
of the parallel control structure for integrating processes.
trollers for integrating processes with inverse response. 2017
International Journal of Systems Science, 46(13), 2461–2473.
21st International Conference on System Theory, Control and
https://doi.org/10.1080/00207721.2013.871369
Computing (ICSTCC), 717–722. https://doi.org/10.1109/IC
Ali, A., & Majhi, S. (2011). Integral criteria for optimal tuning
STCC.2017.8107121.
of PI/PID controllers for integrating processes. Asian Journal
Kaya, I., & Peker, F. (2020). Optimal I-PD controller design
of Control, 13(2), 328–337. https://doi.org/10.1002/asjc.278
for setpoint tracking of integrating processes with time
Ang, K. H., Chong, G., & Li, Y. (2005). PID control
delay. IET Control Theory & Applications, 14(18), 2814–2824.
system analysis, design, and technology. IEEE Trans-
https://doi.org/10.1049/iet-cta.2019.1378
actions on Control Systems Technology, 13(4), 559–576.
Kumar, D., Aryan, P., & Raja, G. L. (2022). Decoupled double-
https://doi.org/10.1109/TCST.2005.847331
loop FOIMC-PD control architecture for double integral
Anil, C., & Padma Sree, R. (2015a). Tuning of PID controllers
with dead time processes. Canadian Journal of Chemical
for integrating systems using direct synthesis method. ISA
Engineering, 1–13. https://doi.org/10.1002/cjce.24355
Transactions, 57, 211–219. https://doi.org/10.1016/j.isatra.20
Kumar, D. B. S., & Padma Sree, R. (2016). Tuning of IMC based
15.03.002
PID controllers for integrating systems with time delay. ISA
Anil, C., & Padma Sree, R. (2015b). PID control of integrating
Transactions, 63, 242–255. https://doi.org/10.1016/j.isatra.
systems using multiple dominant poleplacement method.
2016.03.020
Asia-Pacific Journal of Chemical Engineering, 10(5), 734–742.
Kumari, S., Aryan, P., Kumar, D., & Raja, G. L. (2022).
https://doi.org/10.1002/apj.1911
Hybrid dual-loop control method for dead-time second-
Aryan, P., & Raja, G. L. (2022). A novel equilibrium opti-
order unstable inverse response plants with a case study on
mized double-loop control scheme for unstable and inte-
CSTR. International Journal of Chemical Reactor Engineer-
grating chemical processes involving dead time. Interna-
ing, June, 1–11. https://doi.org/10.1515/ijcre-2022-0035
tional Journal of Chemical Reactor Engineering, June, 1–20.
Kumari, S., Aryan, P., & Raja, G. L. (2021). Design and
https://doi.org/10.1515/ijcre-2022-0007.
simulation of a novel FOIMC-PD/P double-loop control
Åström, K. J., & Hägglund, T. (1995). PID controllers - theory,
structure for CSTRs and bioreactors. International Jour-
design, and tuning (2nd ed). ISA.
nal of Chemical Reactor Engineering, 19(12), 1287–1303.
Cetin, M., & Iplikci, S. (2015). A novel auto-tuning PID con-
https://doi.org/10.1515/ijcre-2021-0140
trol mechanism for nonlinear systems. ISA Transactions, 58,
Mercader, P., & Baños, A. (2017). A PI tuning rule for integrat-
292–308. https://doi.org/10.1016/j.isatra.2015.05.017
ing plus dead time processes with parametric uncertainty.
Chakraborty, S., Ghosh, S., & Kumar Naskar, A. (2017).
ISA Transactions, 67, 246–255. https://doi.org/10.1016/j.isat
I–PD controller for integrating plus time-delay processes.
ra.2017.01.025
IET Control Theory & Applications, 11(17), 3137–3145.
O’Dwyer, A. (2009). Handbook of PI and PID controller tuning
https://doi.org/10.1049/iet-cta.2017.0112
rules (3rd ed). Imperial College Press.
Dey, C., & Mudi, R. K. (2009). An improved auto-tuning
Onat, C. (2019). A new design method for PI–PD control of
scheme for PID controllers. ISA Transactions, 48(4), 396–
unstable processes with dead time. ISA Transactions, 84,
409. https://doi.org/10.1016/j.isatra.2009.07.002
69–81. https://doi.org/10.1016/j.isatra.2018.08.029
Feedback Instruments Ltd. (2010). Digital pendulum control
Peker, F., & Kaya, I. (2017). Identification and real time con-
experiments manual. In User Manual (No. 1160-33936S; 33-
trol of an inverted pendulum using PI-PD controller. 2017
936S). www.feedback-instruments.com.
21st International Conference on System Theory, Control and
Grimholt, C., & Skogestad, S. (2016). Optimal PID control
Computing (ICSTCC), 771–776. https://doi.org/10.1109/IC
of double integrating processes. IFAC-PapersOnLine, 49(7),
STCC.2017.8107130.
127–132. https://doi.org/10.1016/j.ifacol.2016.07.228
Peker, F., & Kaya, I. (2019). Integral-proportional deriva-
Jin, Q. B., & Liu, Q. (2014). Analytical IMC-PID design in
tive (I-PD) controller tuning for pure integrating processes
terms of performance/robustness tradeoff for integrating
with time delay. 2019 International Conference on Applied
processes: From 2-Dof to 1-Dof. Journal of Process Con-
Automation and Industrial Diagnostics (ICAAID), Septem-
trol, 24(3), 22–32. https://doi.org/10.1016/j.jprocont.2013.
ber, 1–6. https://doi.org/10.1109/ICAAID.2019.8934961.
12.011
Raja, G. L. (2021). Enhanced design of a PI-PD based smith
Kaya, I. (2004). Two-degree-of-freedom IMC structure and
predictor for industrial plants. IFAC-PapersOnLine, 54(21),
controller design for integrating processes based on gain and
79–84. https://doi.org/10.1016/j.ifacol.2021.12.014
phase-margin specifications. IEE Proceedings - Control The-
Raja, G. L., & Ali, A. (2021). New PI-PD controller design
ory and Applications, 151(4), 481–487. https://doi.org/10.10
strategy for industrial unstable and integrating processes
49/ip-cta:20040658
with dead time and inverse response. Journal of Con-
Kaya, I. (2020). Integral-proportional derivative tuning for
trol, Automation and Electrical Systems, 32(2), 266–280.
optimal closed loop responses to control integrating
https://doi.org/10.1007/s40313-020-00679-5
processes with inverse response. Transactions of the Institute
20 F. PEKER AND I. KAYA

Rao, C. V. N., & Padma Sree, R. (2010). IMC based controller and Applications, 148(2), 180–184. https://doi.org/10.1049/
design for integrating systems with time delay. Indian Chemi- ip-cta:20010197
cal Engineer, 52(3), 194–218. https://doi.org/10.1080/001945 Visioli, A., & Zhong, Q. (2011). Control of integral processes with
06.2010.547972 dead time. Springer. https://doi.org/10.1007/978-0-85729-
Seshagiri Rao, A., Rao, V. S. R., & Chidambaram, M. (2009). 070-0.
Direct synthesis-based controller design for integrating Vrančić, D., Huba, M., & Oliveira, P. M. (2018). PID controller
processes with time delay. Journal of the Franklin Insti- tuning for integrating processes. IFAC-PapersOnLine, 51(4),
tute, 346(1), 38–56. https://doi.org/10.1016/j.jfranklin.2008. 586–591. https://doi.org/10.1016/j.ifacol.2018.06.159
06.004 Wu, H., Su, W., & Liu, Z. (2014). PID controllers: Design and
Skogestad, S. (2003). Simple analytic rules for model reduc- tuning methods. 2014 9th IEEE Conference on Industrial
tion and PID controller tuning. Journal of Process Control, Electronics and Applications, 808–813. https://doi.org/10.11
13(4), 291–309. https://doi.org/10.1016/S0959-1524(02)000 09/ICIEA.2014.6931273.
62-8 Zhang, W., Cui, Y., & Ding, X. (2020). An improved ana-
Srivastava, S., & Pandit, V. S. (2017). A 2-Dof LQR based lytical tuning rule of a robust PID controller for inte-
PID controller for integrating processes considering robust- grating systems with time delay based on the multiple
ness/performance tradeoff. ISA Transactions, 71, 426–439. dominant pole-placement method. Symmetry, 12(9), 1449.
https://doi.org/10.1016/j.isatra.2017.09.010 https://doi.org/10.3390/sym12091449
Verma, B., & Padhy, P. K. (2018). Optimal PID controller design Zhang, W., Cui, Y., Ding, X., Yin, Z., & Wang, Y. (2021). A
with adjustable maximum sensitivity. IET Control Theory & novel tuning method of differential forward robust PID con-
Applications, 12(8), 1156–1165. https://doi.org/10.1049/iet- troller for integrating systems plus time delay based on direct
cta.2017.1078 synthesis method. International Journal of Systems Science,
Vilanova, R., & Visioli, A. (Eds.). (2012). PID control in the third 52(2), 238–262. https://doi.org/10.1080/00207721.2020.
millennium. Springer. https://doi.org/10.1007/978-1-4471- 1825871
2425-2.
Visioli, A. (2001). Optimal tuning of PID controllers for inte-
gral and unstable processes. IEE Proceedings - Control Theory