ISA Transactions 121 (2022) 240–257

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Practice article

Simple internal model control based modified Smith predictor for

integrating time delayed processes with real-time verification
∗
Somak Karan a , , Chanchal Dey b , Surojit Mukherjee c
a
Department of Applied Electronics and Instrumentation Engineering, Haldia Institute of Technology, Haldia, West Bengal, India
b
Instrumentation Engineering, Department of Applied Physics, University of Calcutta, Kolkata, West Bengal, India
c
Yokogawa India Limited, Bengaluru, Karnataka, India

article info a b s t r a c t

Article history: Internal model control (IMC) tuned simple modified Smith predictor structure for integrating time
Received 23 February 2020 delayed processes (IPTD) is reported here. Pole position at origin implies its non-self-regulating
Received in revised form 5 April 2021 behaviour. Processes like distillation column, liquid supply to large storage tank, superheated steam
Accepted 5 April 2021
flow to turbine etc. are usually found IPTD processes. Reported modified Smith predictor (MSP) design
Available online 8 April 2021
with multiple controllers is adequate to exhibit anticipated closed loop performance for IPTD processes.
Keywords: Tuning complexity of the reported multi-controller based structure is mitigated by the sole tuning
Time delay parameter (closed loop time constant) obtained from IMC design. Proposed scheme shows considerable
Modified Smith predictor performance improvement during set point tracing by zero overshoot. Additionally, smooth as well as
IMC controller reasonably fast load recovery is ensured. Eminence of the reported scheme is established in terms of
Integrating time delayed process performance indices along with stability margins in assessment through recently reported modified
Closed loop time constant Smith predictor techniques. Real-time evaluation of the proposed design is demonstrated on an
Real time level control
indigenous set up of level control loop considered as IPTD process.
© 2021 ISA. Published by Elsevier Ltd. All rights reserved.

1. Introduction methodologies. Majhi and Atheron [7] reported modified con-

trol structure of Smith predictor [22] for unstable as well as
Pole position at origin results integrating behaviour and hence integrating processes employing gain and phase margin criteria.
to achieve desired performance from such processes is quite Subsequently, Kaya and Atherton [8] proposed MSP structure
difficult especially in presence of time delay. Due to inherent with PI-PD for delay dominated processes. Simple PID tuning rule
non-self-regulating behaviour IPTD processes provide continuous is recommended for IPTD processes by Chidambaram and Padma
fluctuations over considerable period once get disturbed from Sree [10]. Liu et al. [11] reported two degree of freedom (2-DOF)
their steady state. Moreover, IPTD processes with dominant time control structure for time delayed integrating and unstable pro-
constant exhibit undesirable sluggish behaviour. Processes like cesses. IMC regulating approach is reported by Seborg et al. [12]
fractional distillation [1], liquid supply to storage tank [2], heating for such cases also. With goal towards minimizing integral square
of batch process [3], super-heated steam supply to turbine [4] error (ISE), Ali and Majhi [13] suggested PID tuning guideline for
are the well-known integrating processes. Controlling IPTD pro- integrating processes with gain crossover frequency and Nyquist
cesses in a desired way is always a difficult task for process curve slope. Alternative tuning approach with reset compensator
engineers [5]. Moreover, selection of improper control strategy for IPTD processes are reported by Davo’ and Ba nos [14]. In IMC
and inappropriate controller settings fail to exhibit the required design by Kumar and Padma Sree [15] closed loop time constant
closed loop behaviour during transient and steady state opera- is adjusted for achieving desired process output. Combination
tions [6]. of two PD controllers in MSP design is reported by Chakraborty
To mitigate this limitation, control methodology involving et al. [16]. Medarametla and Manimozhi [17] proposed guideline
MSP technique is widely accepted. Over the last two decades, to tune PID with second order filter toward controlling pure
good numbers of publications are reported [7–21] on modifi- integrating processes. Another MSP based control scheme is re-
cation and augmentation of Smith predictor [22] based control ported by Karan and Dey [18] for achieving enhanced disturbance
rejection. Model reduction based approach for controlling inte-
∗ Corresponding author. grating processes using PID controller together with noise filter
E-mail addresses: [email protected] (S. Karan),
is reported by Goud and Rao [19]. Zhang et al. [20] proposed IMC
[email protected] (C. Dey), [email protected] based active disturbance rejection scheme for Smith predictor
(S. Mukherjee). employing available plant data. Comparable approach is recently

https://doi.org/10.1016/j.isatra.2021.04.008
0019-0578/© 2021 ISA. Published by Elsevier Ltd. All rights reserved.
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

reported for unstable delay dominated processes by Karan and 2.1. Proposed controller designing
Dey [21] where multiple controllers are involved to achieve
overall satisfactory closed loop response. Expressions for GC1 (s), GC2 (s) and GC3 (s) of the proposed MSP
Reported control schemes [7,8,10–21] for IPTD processes are (Fig. 1) are given by the following relations
targeted toward performance enhancement during closed loop
( )
1
operation. However, majority of the designs [7,10–15,17–20] are GC1 (s) = Kp 1+ , (4)
Ti s
not simple to comprehend as they involve more than one con-
GC2 (s) = β Kp , (5)
troller and incompetent to exhibit satisfactory closed loop perfor-
mance. Thus, framing of appropriate control policy for integrating GC3 (s) = γ Kp (1 + Td s). (6)
delay dominated processes is an open issue.
In Eq. (4), GC1 (s) is an IMC-PI controller [23] having tuning pa-
Here, IMC tuning guideline is reported for MSP with multiple
rameters Kp (proportional gain) and Ti (integral time). In Eq. (5),
controllers. Novelty of the proposed MSP scheme comprises sole GC2 (s) is a proportional controller having the same proportional
tuning parameter λ (i.e. closed loop time constant) which is suit- gain as with GC1 (s) i.e. Kp . GC2 (s) contains an additional tuning
able enough to tune all the controllers involved. Proposed MSP constant β which is obtained from Routh stability criterion [24]
design results zero overshoot while set point tracing and quicker for Eq. (2). GC3 (s) is PD controller having same proportional gain
load regulation. Effectiveness is demonstrated through simulation as with GC1 (s) and GC2 (s) i.e. Kp , and the derivative time Td .
results for reputed IPTD models [7,14–21]. In addition, real-time GC3 (s) has another parameter γ , achieved from Routh stability
performance verification is also substantiated by indigenously criterion [24] of Eq. (3). Notable quality of suggested MSP is
developed level control loop. Robust behaviour is established GC1 (s), GC2 (s), and GC3 (s) employ the same proportional gain
while substantial perturbations in process gain accompanied by Kp .
time delay. Superiority of suggested MSP scheme is demonstrated
through integral error measure (IAE, ISE) and integral time mul- 3. Tuning guideline
tiplied error measure (ITAE, ITSE) [9]. Moreover, smoothness in
Tuning guideline for the proposed control structure (Fig. 1)
control action for the proposed scheme is also estimated through
consisting of GC1 (s), GC2 (s), and GC3 (s) are discussed in the
computation of TV (control action variation) [9] along with other
following sections. Additional tuning parameters β and γ as-
techniques [7,14–21]. sociated with GC2 (s) and GC3 (s) are also obtained from Routh
Control structure and the design of the controllers are dis- stability criterion [24] employing the relations of Eqs. (2) and (3)
cussed in Section 2. Section 3 consists with tuning guideline for respectively.
the controllers. Stability and robustness is reported in Section 4.
Simulation study for well accepted purely integrating process 3.1. Set point controller GC1 (s)
models and real-time experimentation with indigenously devel-
oped level process is demonstrated in Section 5. Lastly conclusion Forward path controller GC1 (s) is accountable toward superior
is provided in Section 6. set point tracking. Steps toward designing of GC1 (s) for IPTD
process using conventional IMC control structure is realized by
the following steps [25,26] as shown in Fig. 2.
2. Control structure
Step 1 Integrating process model is defined by g˜P (s) without
considering its time delay [9]
K K
G (s) e−θs = Gm (s) e−θm s = e−θm s . (1) g˜P (s) = . (7)
s s
K - open loop gain and θm - dead time for IPTD model (Eq. (1)) Step 2: q (s) (IMC controller) is realized from g˜P (s)−1 (i.e. inverted
Fig. 1 exhibits the configuration of suggested MSP with con- model) derived from previous step with filter f (s) [25,26].
trollers GC1 (s) , GC2 (s), and GC3 (s). Forward path controller
s (2λs + 1)
GC1 (s) is accountable for improved servo tracking. GC2 (s) and q (s) = g˜P (s)−1 f (s) = . (8)
K (λs + 1)2
GC3 (s) are in feedback path acting towards enhanced disturbance
suppression. Relations among the process response (Y) through Step 3: IMC equivalent feedback controller is given by Eq. (9) [25,
set point variation (R) and load disturbance (D) are presented by 26]
Eqs. (2) and (3) separately. q (s)
gc (s) = . (9)
Y(s) G(s)GC1 (s) 1 − g˜P (s) q (s)
= e −θm s
, (2)
R(s) 1 + Gm (s)(GC1 (s) + GC2 (s)) Now, substituting the Eqs. (7) and (8) in Eq. (9)
−θm s
Y(s) G(s)e s(2λs+1)
= K(λs+1)2
D(s) 1 + Gm (s)(GC1 (s) + GC2 (s)) gc (s) = K s (2λs+1)
. (10)
( ) 1− ·
1 + Gm (s) GC1 (s) + GC2 (s) − GC1 (s)GC3 (s)e−θm s s K (λs+1)2
× . (3)
1 + GC3 (s)G(s)e−θm s Step 4: From Eq. (10), IMC based feedback controller is given by
Delay term is absent in denominator (Eq. (2)), GC1 (s) and 2λs + 1
gc (s) = . (11)
GC2 (s) work toward ensuring desirable set point response. How- Kλ2 s
ever, GC1 (s), GC2 (s), and GC3 (s) are present in Eq. (3) containing Step 5: For the IMC-PI controller (Eq. (11)) Kp and Ti are given by
delay term in both the numerator and denominator, account-
2
able toward desired regulatory behaviour. Here, GC1 (s) is IMC- Kp = , (12)
PI (proportional-integral) controller, GC2 (s) is P (proportional) Kλ
controller, and GC3 (s) is PD (proportional-derivative) controller. Ti = 2λ. (13)

241
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Fig. 1. Proposed MSP based closed loop structure.

Fig. 2. Conventional IMC structure.

Guideline toward selection of λ is provided in the following Subsequently, Eq. (16) can be written as
section. Relation between λ and θm is found out [9] for IPTD KKp [Ti s + 1] 1
.
−s
model gP (s) = e s i.e. from Eq. (1) where K = 1 and θm = 1. Yr = = (17)
Ti s2 + KKp [Ti s + 1] + β KKp Ti s (
s Ti s+β KKp Ti ) +1
The set point responses are shown in Fig. 3 for the proposed KKp (Ti s+1)
scheme while considering only the forward path controller GC1 (s)
To reduce the expression of Eq. (17), bracketed terms of the
with different values of λ = θm , θ2m , θ3m , θ4m . Fig. 3 clearly demon-
denominator are considered to be given by Eq. (18)
strates that λ = θm4
results quicker set point response but with
unwanted overshoot. Ti s + β KKp Ti = Ti s + 1. (18)
θm From Eq. (18), tuning parameter β is derived as
λ= . (14)
4 1
Hence, the overshoot must be restricted to avoid actuator β= . (19)
KKp Ti
saturation. Controller GC2 (s) is placed in the feedback path and
it plays significant role for stabilizing the process behaviour as Substituting the Kp and Ti values in Eq. (19)
discussed in the following section. 1 1
β= = = 0.25. (20)
2
2λ
( )
KKp Ti K Kλ
3.2. Stabilizing controller GC2 (s)
Now, substituting β from Eq. (19) in Eq. (17)
Proposed MSP design includes GC2 (s) assuring overshoot free KKp
Yr = . (21)
set point response. Proportional (P) controller GC2 (s) (Eq. (5)) s + KKp
with gain β Kp (Kp - proportional gain and β - additional parame-
Employing the Routh stability criterion [24] for Eq. (21), it is
ter) obtained from stability criterion [24] of Eq. (2)
found that the product( of open loop
) gain and the proportional
Y(s) G(s) · GC1 (s) gain must be positive i.e. KKp > 0 for IPTD process so that the
= e−θm s .
R(s) 1 + Gm (s)(GC1 (s) + GC2 (s)) desired stable behaviour can be ensured.
According to Karan and Dey [21], neglecting the time delay
3.3. Disturbance rejection controller GC3 (s)
part from Eq. (2)
Y(s) G(s) · GC1 (s) GC3 (s), PD controller (Eq. (6)) by nature with derivative time
= . (15)
R(s) 1 + Gm (s)(GC1 (s) + GC2 (s)) Td and gain γ Kp where γ is obtained from Routh stability crite-
rion [24] as follows
Substituting the expressions of GC1 (s), GC2 (s), G(s) and Gm (s)
in Eq. (15) Y(s) G(s)e−θm s
=
K
( ) D(s) 1 + Gm (s)(GC1 (s) + GC2 (s))
Y(s) · Kp 1 + T1s ( )
s i
]. 1 + Gm (s) · GC1 (s) + GC2 (s) − GC1 (s)GC3 (s)e−θm s
= Yr = [ ( ) (16) × .
R(s) 1 + s Kp 1 + T s + β Kp
K 1
1 + GC3 (s)G(s)e−θm s
i

242
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Fig. 3. Set point responses with different values of λ.

Corresponding characteristic equation is given by 4. Stability and robustness

1 + GC3 (s) G (s) e −θm s
= 0. (22)
Stability and robustness of the suggested MSP methodology
Replacing the expressions of GC3 (s) and G(s) in Eq. (22) is evaluated with parametric uncertainties of the process model
K and unwanted load fluctuations. Uncertainties are present during
1 + γ Kp (1 + Td s) · e−θm s = 0. (23) process gain estimation and time delay measurement for IPTD
s
process. Robust stability [26] of the closed loop system is ensured
Implementing first-order Pade’s approximation [21] in Eq. (23)
if
is written as
K −0.5θm s + 1 i.e. ∥∆m (jω) C (jω)∥ < 1 for ∀ωϵ (−∞, ∞). (32)
( )
1 + γ Kp (1 + Td s) · = 0. (24)
s 0.5θm s + 1 C(s = jω) is complementary sensitivity function (CSF) and
To ascertain improved load recovery [21] derivative time Td is ∆m (s = jω) is multiplicative uncertainty. CSF during set point
given by response is given by Eq. (33)
θm 4λ(jω) + 2
Td = λ = . (25) C (jω) = . (33)
4 2Kλ (jω)2 + 5λ (jω) + 2
2

Substituting Td from Eq. (25) in Eq. (24) From Eq. (32) complementary sensitivity may be given by
−0.5θm s + 1 ⏐ G (jω) e−θ s − Gm (jω)e−θm s ⏐
4s + K · γ · Kp (4 + θm s) = 0.
⏐ ⏐
0.5θm s + 1 ∆m (jω) < ⏐
⏐ ⏐. (34)
Gm (jω)e−θm s ⏐
Subsequently, Routh array is constructed from Eq. (24) where
the first column terms are Here, G (jω) e−θ s - true process model and Gm (jω)e−θm s - plant
model. With uncertainty in time delay tuning parameter must be
2θm − 2γ Kp θm Kp > 0, 2
(26)
chosen as
4 1 − γ θm Kp K > 0,
( )
(27) 1
∥C (jω)∥∞ < ⏐ −∆θ s ⏐. (35)
16γ Kp K > 0. (28) ⏐e m − 1⏐
To ascertain stability, lower and upper limit of γ is obtained Likewise, in presence of uncertainty in process gain, tuning
from Eqs. (26), (27), and (28). Hence, for absolute stability, bound- parameters need to be chosen as
ary value of γ is given by
1
1 ∥C (jω)∥∞ < |∆K|
. (36)
0<γ < . (29)
θm KKp K

In addition, if the uncertainties are present in process gain

Therefore, the intermediate value in between the upper and
accompanied by time delay simultaneously, then the tuning pa-
lower boundaries is considered to be suitable for γ as given by
the Eq. (30) rameters must be selected as
1
0.5 ∥C (jω)∥∞ < ⏐( ∆K ⏐. (37)
γ = . (30) + 1 e−∆θm s − 1⏐
)
θm KKp
⏐
K

Now, substituting the values of proportional gain Kp from However, the sensitivity and CSF accomplish the condition of
Eq. (12) and closed loop time constant λ from Eq. (14) in Eq. (30), robust performance by Eq. (38) for closed loop structure
the value of γ can be defined by Eq. (31)
∥∆m (jω) C (jω) + wm (jω) (1 − C (jω))∥ < 1. (38)
0.5 0.5
γ = θm ·K·2
= θm ·K·8
= 0.06. (31) Here, wm (jω) is the uncertainty bound and likewise stability
Kλ K·θm conditions for regulatory response can also be verified by Eq. (3).
243
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

( )
Fig. 4. Closed loop responses and related control actions for Model I Gp11 under (a) nominal and (b) perturbed conditions.

5. Results techniques suggested by Chakraborty et al. [16], Medarametla and

Manimozhi [17], Karan and Dey [18], Zhang et al. [20], and Karan
Potency of the suggested methodology is ascertained with and Dey [21] for nominal models (Eqs. (39)–(41)) are shown in
wide simulation results. Performance robustness is further ascer- Figs. 4(a), 6(a), and 8(a).
tained with well-known dead time compensation techniques [7, Similar set of graphical results for the perturbed models (+10%
14–21] reported by Majhi and Atherton [7], Davo’ and Baños [14], perturbation in open loop gain along with time delay i.e. Ĝp11 (s) ,
Kumar and Padma sree [15], Chakraborty et al. [16], Medarametla Ĝp12 (s), and Ĝp13 (s) given by Eqs. (42)–(44)) are shown in
and Manimozhi [17], Karan and Dey [18], Goud and Rao [19], Fig. 4(b), 6(b), and 8(b). Table 2 substantiates the stability of
Zhang et al. [20], and Karan and Dey [21] in presence of process Model I. Small gain theorem [26] is employed to compute the
parameters perturbation. Set value is provided by step change CSF in presence of positive perturbations in time delay (∆θm )
and subsequently after attaining the steady state, disturbance is and process gain (∆K). C(jω) for Gp11 (s) , Gp12 (s), and Gp13 (s)
provided as single pulse at the process input. Simulation study is is analysed with maximum perturbation in ∆K and ∆θm as
performed with reputed IPTD models along with real-time verifi- illustrated in Fig. 5, Fig. 7, and Fig. 9 respectively.
cation on an indigenously developed level process. Gc1 (s), Gc2 (s), 1.1
and Gc3 (s) are tuned separately for each process model with the Ĝp11 (s) = e−1.1s . (42)
s
same tuning relation for λ. Moreover, additional tuning param- 1.1
eters β along with γ are having same value for all the models. Ĝp12 (s) = e−4.4s . (43)
s
For quantitative closed loop performance assessment, rise time
0.55 −6.6s
(tr ), percentage peak overshoot (%Mp ), settling time (ts ), integral Ĝp13 (s) = e . (44)
error measures (IAE and ISE) and integral time error measures s
(ITAE and ITSE) [9] are computed. Furthermore, TV (total variation Performance indices of controllers are provided in Tables 3
in control action) [9] is also calculated for each setting. Noise and 4. Suggested MSP structure can entirely eliminate the over-
sensitivity is evaluated in existence of measurement noise [18]. shoot during servo response and quicker load recovery is also
achieved compared to Chakraborty et al. [16], Medarametla and
5.1. Model I Manimozhi [17], Karan and Dey [18], Zhang et al. [20], and Karan
and Dey [21].
This IPTD process [16,20] is considered with three different
time delays and open loop gains 5.2. Model II

1 This popular IPTD model is reported by Kumar and Padma

Gp11 (s) = e−s , (39)
s Sree [15], and Goud and Rao [19] signifying the behaviour of
1 distillation column (Fig. 10) where two outputs are considered
Gp12 (s) = e−4s , (40)
s as top and bottom products. The bottom product along with time
0.5 −6s delay is considered to be a level control system as realized by
Gp13 (s) = e . (41)
s 0.2
Gp2 (s) = e−7.4s . (45)
Process models Gp11 (s), Gp12 (s) and Gp13 (s) are having dissim- s
ilar open loop gains K = 1and0.5 along with different time delays Controller parameters for the proposed scheme are depicted
θm = 1 s, 4 s, and 6s. Here, Gp13 (s) is realized through model in Table 1(a). Closed loop performance of the proposed MSP
e −6s
reduction of 2s+1
as reported by Zhang et al. [20]. Performances of is conducted together with the settings by Kumar and Padma
the proposed MSP are verified along with the reputed dead time Sree [15], and Goud and Rao [19]. Respective process output and
compensation methodologies [16–18,20,21]. Tuning parameters control action for Model II are provided in Fig. 11(a). Improved
of controllers are provided in Table 1(a). Responses and their robustness in comparison with others’ reported schemes [15,19]
respective control action for the proposed MSP along with the is found while incorporating +20% parametric perturbations ( ∆K
244
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Fig. 5. Complementary sensitivity C (jω) and the relevant magnitude plot with +70% perturbation in ∆K = 0.7 and +20% perturbation in ∆θm = 0.2 for Model I Gp11 .
( )

( )
Fig. 6. Closed loop responses and related control actions for Model I Gp12 under (a) nominal and (b) perturbed conditions.

Table 1(a)
Controllers’ tuning parameter for IPTD process models.
Model GC1 (PI Controller) GC2 (P Controller) GC3 (PD controller)
λ Kp Ti Controller TF β Controller TF γ Td Controller TF
Gp11 (s) = 1 −s
s
e 0.25 8.0 0.5 4s+8
0.5s
2.0 0.25 0.48+0.12s
[16,17]

Gp12 (s) = 1 −4s

s
e 1.00 2.0 2.0 4s+2
2s
0.5 1.0 0.12+0.12s
Model I: Gp1 (s) = K −θ s
s
e [16]
(θ = 1,4,6)
0.5 −6s 8.01s+2.67
Gp13 (s) = s
e 1.50 2.67 3.0 3s
0.67 1.50 0.16+0.24s
[20]

Model II: Gp2 (s) = 0s.2 e−7.4s 1.85 5.4 3.7 19.98s+5.4
3.7s
1.35 1.85 0.32+0.60s
(Distillation column [15,19]) 0.25 0.06

0.05 −6s 80s+26.67

Model III: Gp3 (s) = s
e [19] 1.50 26.67 3.0 3s
6.67 1.50 1.60+2.4s
0.01 −4s
Model IV: Gp4 (s) = s
e [16] 1.0 200.0 2.0 400s+200
2s
50.0 1.0 12+12s
0.002 −3s 2000s+1333.33
Model V: Gp5 (s) = s
e 0.75 1333.33 1.5 1.5s
333.33 0.75 80+60s
(Temperature control system [16])

Model VI: Gp6 (s) = 0.s37 e−0.55s 0.14 38.61 0.28 10.81s+38.61
0.28s
9.65 0.14 2.32+0.32s
(Liquid level control system [14])

and ∆θm ) as given by Eq. (46). Small gain theorem [26] is applied is ensured as shown in Fig. 14.
toward stability analysis while combining positive perturbation in 0.24
∆K and ∆θm to obtain C(jω) (Table 2). Improved robust stability Ĝp2 (s) = e−8.88s . (46)
s
245
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Table 1(b)
Controllers’ tuning parameter for real-time level control process.
Model GC1 (PI Controller) GC2 (P Controller) GC3 (PD controller with noise filter)
λ Kp Ti Controller TF β Controller TF γ Td Controller T.F
0.09 −8s 44.44s+11.11 0.67+1.34s
Model VII : Gp7 (s) = s
e 2.0 11.11 4.0 4s
0.25 2.78 0.06 2.0 0.2s+1
(Indigenous level control system)

Fig. 7. Complementary sensitivity C (jω) and the relevant magnitude plot with +60% perturbation in ∆K = 0.6 and +20% perturbation in ∆θm = 0.8 for Model I Gp12 .
( )

20
21

( )
Fig. 8. Closed loop responses and related control actions for Model I Gp13 under (a) nominal and (b) perturbed conditions.

Process output and the related control action for Eq. (46) are 5.3. Model III
illustrated in Fig. 11(b). Enhanced robustness of proposed MSP
Model III is reported by Goud and Rao [19] obtained through
is also observed with perturbations of the controllers’ tuning
reduction [19] of the unstable process model 100s5 −1 e−6s .
parameters. All the tuning parameters associated with GC1 (s) ,
0.05
GC2 (s), and GC3 (s) along with the settings reported by [15,19] Gp3 (s) = e−6s . (47)
s
are perturbed by ±25% from their respective nominal value.
Controllers’ settings for the proposed scheme are provided
Related responses along with respective IAE values are depicted in Table 1(a) and its performance assessment against reputed
in Fig. 12(a)–(c) which visibly substantiate the advantage of sug- schemes by Goud and Rao [19], and Karan and Dey [21] are
gested structure. provided in Fig. 15(a). Robustness of the proposed MSP technique
along with others’ reported schemes [19,21] are validated with
Furthermore, responses of Model II along with control actions
+10% perturbations in K and θm (Eq. (48)), related responses and
with measurement noise (noise power = 0.001) [18] for Eqs. (45) control actions are shown in Fig. 15(b).
and (46) are illustrated in Fig. 13(a) and (b). Controller perfor- 0.055
mance indices during servo and regulatory responses are speci- Ĝp3 (s) = e−6.6s . (48)
s
fied in Tables 3 and 4 which certainly validate the effectiveness Robust stability [26] is established by complementary sensitiv-
for the proposed methodology. ity C(jω) (Table 2) and the relevant magnitude plot is illustrated
246
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

( 9.) Complementary sensitivity C (jω) and the relevant magnitude plot with +50% perturbation in ∆K = 0.25 and +10% perturbation in ∆θm = 0.6 for Model
Fig.
I Gp13 .

Fig. 10. Proposed MSP scheme for distillation column (Model II).

( )
Fig. 11. Closed loop responses and related control actions for Model II Gp2 under (a) nominal and (b) perturbed conditions.

247
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Table 2
Stability analysis using perturbed IPTD process models.
Models CSF Process gain Stability with gain Time delay Stability with time delay
C(jω) perturbation perturbation perturbation perturbation
(∆K ) (∆θm )
( ) ( )
∥C(jω)∥∞ < ∆K
1
∥C(jω)∥∞ < 1
K −jω∆θm
jω
∆θm +1
2
1.04jω+2 j0.1ω+1
Gp11 (s) = 1 −s
s
e 0.135jω2 +1.3jω+2
0.7 ∥C (jω)∥∞ < 1
0.7
0.2 ∥C (jω)∥∞ < −j0.2ω
[16,17] (+70%) (+20%)
Gp1 (s) = Ks e−θ s Gp12 (s) = 1 −4s
e
5.04jω+2
0.6 ∥C (jω)∥∞ < 1
0.8 ∥C (jω)∥∞ < j0.4ω+1
s 3.175jω2 +6.3jω+2 0.6 −j0.8ω
(θ = 1, 4, 6) [16] (+60%) (+20%)
0.5 −6s 6jω+2 j0.3ω+1
Gp13 (s) = s
e 2.25jω2 +7.5jω+2
0.25 ∥C (jω)∥∞ < 1
0.5
0.6 ∥C (jω)∥∞ < −0.6jω
[20] (+50%) (+10%)

7.44jω+2 j0.148ω+1
Gp2 (s) = 0s.2 e−7.4s 1.38jω2 +9.3jω+2
0.1 ∥C (jω)∥∞ < 1
0.5
0.296 ∥C (jω)∥∞ < −j0.296ω
(Distillation column [15,19]) (+50%) (+4%)
0.05 −6s 6.4jω+2 j0.03ω+1
Gp3 (s) = s
e [19] 0.256jω2 +8jω+2
0.035 ∥C (jω)∥∞ < 1
0.7
0.06 ∥C (jω)∥∞ < −j0.06ω
(+70%) (+1%)
0.01 −4s 4.4jω+2 j0.02ω+1
Gp4 (s) = s
e [16] 0.024jω2 +5.5jω+2
0.005 ∥C (jω)∥∞ < 1
0.5
0.04 ∥C (jω)∥∞ < −j0.04ω
(+50%) (+1%)
3.04jω+2 j0.015ω+1
Gp5 (s) = 0.002
s
e−3s (Temperature 0.002jω2 +3.8jω+2
0.001 ∥C (jω)∥∞ < 1
0.5
0.03 ∥C (jω)∥∞ < −j0.03ω
control system [16] ) (+50%) (+1%)
0.56jω+2 j0.0275ω+1
Gp6 (s) = 0.s37 e−0.55s (Liquid level 0.015jω2 +0.7jω+2
0.185 ∥C (jω)∥∞ < 1
0.5
0.055 ∥C (jω)∥∞ < −j0.055ω
control system [14]) (+50%) (+10%)
8jω+2 j0.2ω+1
Gp7 (s) = 0.s09 e−8s (Indigenously 0.72jω2 +10jω+2
0.045 ∥C (jω)∥∞ < 1
0.5
0.4 ∥C (jω)∥∞ < −j0.4ω
developed liquid level control (+50%) (+5%)
system)

Table 3
Servo performance evaluation with nominal and perturbed IPTD models.
Nominal models Scheme Nominal models Perturbed Perturbed models
models
tr (s) MP (%) ts (s) tr (s) MP (%) ts (s)

Gp11 (s) = 1 −s Proposed 1.14 0 2.39 Ĝp11 (s) = 1.20 0 2.46

se Chakraborty et al. [16] 2.07 0 10.09 2.11 0 11.14
[16,17] Medarametla and Manimozhi 2.66 0 11.63 1.1 e−1.1s (+10% 2.75 0 13.29
s
[17] perturbation)

Ĝp12 (s) = 1s.1 e−4.4s (+10%

Proposed 5.41 0 14.06 6.09 0 15.80
Gp12 (s) = 1
se
−4s
Gp1 (s) = K se
−θ s Chakraborty et al. [16] 10.3 0 49.25 10.72 0 52.62
[16,18] Karan and Dey [18] 5.94 0 17.37 perturbation) 6.37 0 19.92
(θ = 1,4,6)

Gp13 (s) = 0s.5 e−6s Ĝp13 (s) = 0.s55 e−6.6s (+10%

Proposed 7.86 0 17.28 8.21 0 17.46
Zhang et al. [20] 9.36 0 17.86 9.37 0 18.38
[20] Karan and Dey [21] 8.49 0 31.75 perturbation) 9.10 0 34.80

Gp2 (s) = 0s.2 e−7.4s Ĝp2 (s) = 0.s24 e−8.88s (+20%

Proposed 8.04 0 17.68 9.66 0 17.98
Kumar-Padma sree [15] 13.33 46 101.8 14.81 84.9 107.4
(Distillation column [15,19]) Goud and Rao [19] 11.03 64.50 64.03 perturbation) 12.30 100.80 106.00

Ĝp3 (s) = 0.055 −6.6s (+10%

Proposed 6.83 0 11.95 7.29 0 14.48
Gp3 (s) = 0.s05 e−6s [19] Karan and Dey [21] 7.26 0 19.39 s e 7.69 0 22.59
Goud and Rao [19] 9.03 60.20 56.45 perturbation) 9.35 83.10 57.51

Ĝp4 (s) = 0.011 −4.4s (+10%

Proposed 4.56 0 10.74 4.75 0 20.68
Gp4 (s) = 0.s01 e−4s [16] Chakraborty et al. [16] 8.26 0 44.62 s e 8.43 0 50.94
Karan and Dey [21] 4.78 0 12.83 perturbation) 5.21 0 23.04

Gp5 (s) = 0.002 Ĝp5 (s) = 0.0022 e−3.3s (+10%

−3s Proposed 3.32 0 8.27 3.75 0 14.12
s e Chakraborty et al. [16] 9.86 0 66.14 s 10.67 0 70.06
(Temperature control plant [16]) Karan and Dey [21] 3.56 0 10.14 perturbation) 3.80 0 20.20

Gp6 (s) = 0.s37 e−0.55s Ĝp6 (s) = 0.407 −0.605s (+10%

Proposed 0.59 0 1.89 0.67 0 2.67
Dav́o-Banǒs [14] 0.94 55.8 10.23 s e 0.96 69.2 12.52
(Liquid level control system [14]) Karan and Dey [21] 0.66 0 2.01 perturbation) 0.69 0 3.54

Gp7 (s) = 0.s09 e−8s Proposed 9.14 0 22.17 Ĝp7 (s) 0.099
s e
−8.8s (+10% 10.66 0 26.74
Majhi-Atherton [7] 9.14 52.1 44.22 10.66 44.8 49.23
(Indigenously developed liquid level perturbation)
control system)

in Fig. 16. Controller performance indices during servo and reg- niques through model reduction [19] as provided in Fig. 17(a).
ulatory responses are specified in Tables 3 and 4 which clearly Robust of the proposed MSP technique and the others’ [16,21]
validate effectiveness of suggested MSP technique. with +10% perturbations in K and θm (Eq. (50)) is illustrated in
Fig. 17(b).
5.4. Model IV 0.011
Ĝp4 (s) = e−4.4s . (50)
s
This model is well-known to the research community, re- Enhanced robust stability [26] of the proposed methodology
ported by Chakraborty et al. [16] is ensured in terms of smaller amplitude of CSF as revealed in
0.01 Table 2, Resultant magnitude plot is illustrated in Fig. 18. Closed
Gp4 (s) = e−4s . (49) loop performance measures are presented in Tables 3 and 4 which
s
clearly substantiate the benefit of suggested MSP technique.
Tuning parameters for the proposed scheme is provided in
Table 1(a). Effectiveness of the proposed methodology is validated
against others’ reported [16,21] dead time compensating tech-
248
 S. Karan, C. Dey and S. Mukherjee
Table 4
Performance analysis with nominal and perturbed IPTD models.
Nominal models Scheme Nominal models Perturbed Perturbed models
models
TV IAE ITAE ISE ITSE TV IAE ITAE ISE ITSE

Ĝp11 (s) = 1s.1 e−1.1s

Proposed 3.24 0.39 5.95 0.09 0.56 3.54 0.44 6.96 0.10 0.68
Gp11 (s) = 1
se
−s
[16,17] Chakraborty 3.81 59.41 1396 99.07 2613 (+10% perturbation) 4.98 59.46 1397 99.55 2623
et al. [16]
Medarametla 3.42 5.63 110.50 3.68 81.24 3.44 5.66 112.3 4.03 88.66
and Manimozhi
Gp1 (s) = K se
−θ s
[17]
(θ = 1,4,6)
Ĝp12 (s) = 1s.1 e−4.4s
Proposed 4.30 2.77 241 0.61 33.17 4.68 2.99 268.9 0.64 39.5
Gp12 (s) = 1
se
−4s
[16,18] Chakraborty 5.04 347 30190.25 726.2 39488.61 (+10% perturbation) 6.94 347.7 31269.74 733.2 45252.19
et al. [16]
Karan and Dey 4.39 4.11 352.90 0.93 43.03 4.78 4.26 357.80 0.94 47.47
[18]

Gp13 (s) = 0s.5 e−6s Ĝp13 (s) = 0.s55 e−6.6s

Proposed 69.42 2.64 38.97 0.89 4.63 76.42 2.74 39.51 0.95 4.84

[20,21] Zhang et al. 75.22 101.30 4452 133.30 6326.1 (+10% perturbation) 82.77 105.14 4513.6 142.29 6613.03
[20]
Karan and 84.98 5.11 87.46 1.48 11.53 93.85 5.28 87.80 1.57 12.25
Dey[21]

Gp2 (s) = 0s.2 e−7.4s Ĝp2 (s) = 0.s24 e−8.88s

Proposed 21.41 3.68 598.8 0.85 82.22 27.40 4.77 856.9 0.96 121.3

(Distillation column [15,19]) Kumar-Padma 31.12 92.46 15044 107.2 10369 (+20% perturbation) 43.03 97.05 17434 137 17310
sree [15]
249

Goud and Rao 49.49 66.91 10049 72.67 12861 64.60 93.13 14245 111.60 19504
[19]

Ĝp3 (s) = 0.055 −6.6s

Proposed 31.61 1.33 62.8 0.36 1.67 32.70 1.38 63.05 0.41 2.19
Gp3 (s) = 0.s05 e−6s [19] Karan and Dey 61.24 3.73 292.90 0.81 13.09
s e
65.06 3.74 295.10 0.84 14.32
(+10% perturbation)
[21]
Goud and Rao 88.53 27.07 1162.6 14.52 373.79 90.56 27.33 1205.1 16.99 434.58
[19]

Ĝp4 (s) = 0.011 −4.4s

Proposed 184.30 1.05 43.42 0.27 1.60 186.10 1.06 43.61 0.29 1.83
Gp4 (s) = 0.s01 e−4s [16] Chakraborty 229.60 178.9 7397.94 219.9 1303.11
s e
273.30 179.1 7420.44 220.5 1391.43
(+10% perturbation)
et al. [16]
Karan and Dey 252.00 3.35 231.50 0.61 12.63 260.40 3.36 235.60 0.63 13.82
[21]

Gp5 (s) = 0.002 Ĝp5 (s) = 0.0022 e−3.3s

−3s Proposed 594.10 0.47 7.60 0.17 0.45 628.20 0.48 7.61 0.19 0.52
s e s
(Temperature control plant [16]) Chakraborty 681.10 179.1 2896.08 219.8 598.44 (+10% perturbation) 730.10 220.8 3500.60 278.5 762.21
et al. [16]
Karan and Dey 1080 1.34 54.74 0.36 0.98 1196 1.35 56.11 0.38 1.04
[21]

Gp6 (s) = 0.s37 e−0.55s Proposed 3.86 0.1 0.95 0.03 0.01
Ĝp6 (s) = 0.407 −0.605s 3.91 0.11 1.04 0.03 0.02
s e
(Liquid level control Dav́o-Banǒs 8.20 2.96 30.72 1.55 8.55 (+10% perturbation) 9.22 3.23 33.62 1.73 9.73
system [14]) [14]

ISA Transactions 121 (2022) 240–257

Karan and Dey 4.74 0.53 14.74 0.07 0.36 4.89 0.54 14.93 0.08 0.39
[21]

Gp7 (s) = 0.s09 e−8s Proposed 19.65 1.75 165.1 0.46 5.29 Ĝp7 (s) = 0.099
s e
−8.8s 20.06 1.84 187.3 0.52 7.26
(Indigenously developed liquid level Majhi-Atherton 38.36 3.57 178.0 1.31 5.40 (+10% perturbation) 44.92 4.11 194.1 1.52 5.49
control system) [7]
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Fig. 12. Closed loop responses and equivalent IAE values (Model II) for all controllers with (a) nominal tuning parameters, (b) −25% perturbation in tuning
parameters, and (c) +25% perturbation in tuning parameters.

5.5. Model V It consists of primary and secondary inputs relating the cold
and hot water flow where goal is to keep the preferred tempera-
Model of a heat exchanger process i.e. Model V is reported by ture of output flow by controlling input flow. Control scheme of
Chakraborty et al. [16] as given by Eq. (51) the heat exchanger with the proposed methodology is shown in
0.002 Fig. 19. Corresponding responses and control actions along with
Gp5 (s) = e−3s . (51) others’ reported schemes [16,21] through model reduction [19]
s
250
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

( )
Fig. 13. Closed loop responses and related control actions with measurement noise (power = 0.001) for Model II Gp2 under (a) nominal and (b) perturbed conditions.

Fig. ) Complementary sensitivity C (jω) and the relevant magnitude plot with +50% perturbation in ∆K = 0.1 and +4% perturbation in ∆θm = 0.296 for Model
( 14.
II Gp2 .

( )
Fig. 15. Closed loop responses and related control actions for Model III Gp3 under (a) nominal and (b) perturbed conditions.

for nominal (Eq. (51)) as well as perturbed (+10% in K and θm ) Tuning parameters of the controllers involved in the pro-
models (Eq. (52)) are illustrated in Fig. 20(a) and Fig. 20(b). posed MSP methodology are provided in Table 1(a). Controller
performance measures are presented in Tables 3 and 4. Robust
0.0022 −3.3s
Ĝp5 (s) = e . (52) stability [26] is established by complementary sensitivity C(jω)
s
251
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

( 16.) Complementary sensitivity C (jω) and the relevant magnitude plot with +70% perturbation in ∆K = 0.035 and +1% perturbation in ∆θm = 0.06 for Model
Fig.
III Gp3 .

( )
Fig. 17. Closed loop responses and related control actions for Model IV Gp4 under (a) nominal and (b) perturbed conditions.

( 18.) Complementary sensitivity C (jω) and the relevant magnitude plot with +50% perturbation in ∆K = 0.005 and +1% perturbation in ∆θm = 0.04 for Model
Fig.
IV Gp4 .

252
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Fig. 19. Proposed MSP scheme for heat exchanger process (Model V).

( )
Fig. 20. Closed loop responses and related control actions for Model V Gp5 under (a) nominal and (b) perturbed conditions.

Fig. ) Complementary sensitivity C (jω) and the relevant magnitude plot with +50% perturbation in ∆K = 0.001 and +1% perturbation in ∆θm = 0.03 for Model
( 21.
V Gp5 .

(Table 2) and the relevant magnitude plot is illustrated in Fig. 21. 5.6. Model VI

Based on performance and stability assessment proposed MSP Model VI represents a liquid level process reported by Davo’
and Baños [14] as given by
scheme exhibits enhanced performance compared to Chakraborty
0.37
et al. [16], and Karan and Dey [21]. Gp6 (s) = e−0.55s . (53)
s
253
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

( )
Fig. 22. Closed loop responses and related control actions for Model VI Gp6 under (a) nominal and (b) perturbed conditions.

( 23.) Complementary sensitivity C (jω) and the relevant magnitude plot with +50% perturbation in ∆K = 0.185 and +10% perturbation in ∆θm = 0.055 for Model
Fig.
VI Gp6 .

Fig. 24. Schematic illustration of level control loop.

Tuning parameters for the proposed control scheme is pro- during closed loop operation are presented in Tables 3 and 4.
vided in Table 1(a). Responses and related control actions for 0.407
Ĝp6 (s) = e−0.605s . (54)
Eq. (53) and perturbed (+10%) model (Eq. (54)) for MSP scheme s
together with the techniques recommended by Davo’ and Small gain theorem [26] substantiates the robust stability with
considerable perturbation. Robust stability [26] is established by
Baños [14], and Karan and Dey [21] through model reduction [19] C(jω) (Table 2) and the relevant magnitude plot is illustrated in
are illustrated in Figs. 22(a) and 22(b). Performance measures Fig. 23.
254
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

Table 5
Parameter of indigenously developed level control loop [Model VII].
Parameter Description Value
Kq Constant associated with inlet flow rate of tank 1.65 cm3 / (s V)
a Cross sectional area of tank outlet hole 0.78 cm2
g Gravity constant 981 cm2 /s
A Cross sectional area of tank 81 cm2
L9 Set point or operating point for the water level in tank 9 cm
θm Time delay related to the process for long run of pipe 8 s

relation [28,29] based on the conservation of mass and rate of

change in volume of water can be written as
d (ρ V)
= ρ qi − ρ q0 . (55)
dt
Here, V - volume of tank, qi – inlet flow rate, q0 - outlet flow
rate and ρ – liquid density. Considering the upper tank to be
rectangular shaped with water level height L, its volume V = AL
(A – tank cross sectional area)
d (L)
A = qi − q0 . (56)
dt
Tank ) flow rate (qi ) is illustrated as the product of control
( inlet
signal Vp and pump discharge coefficient Kq supplied by the
manufacturer i.e.
Fig. 25(a). Experimental set up of indigenously developed level control loop.
qi = Kq Vp . (57)

Moreover, the outlet flow rate is product of cross sectional

area (a) of the discharge pipe and the rate of water flow (v) i.e.

q0 = av. (58)

The volumetric flow rate (v) at the outlet of the overhead tank
is linked with the height of water level (L) and hence Eq. (58) can
be written as
√
q0 = a 2gL. (59)

Now, substituting the expressions of qi and q0 from Eqs. (57)

Fig. 25(b). Set point tracking and load recovery during real-time experimenta-
and (59) in Eq. (56), the resulting nonlinear relation is given by
tion. Eq. (60)
d (L) Kq a√
= Vp − 2gL. (60)
dt A A
5.7. Indigenous level control
Here, water level in overhead tank can vary from 1 cm (L1 ) to
Empirical validation of suggested technique is evaluated on 9 cm (L9 ) during experimentation. Hence, the nonlinear term of
an indigenously developed level control loop along with reputed Eq. (60) can be approximated as
MSP scheme by Majhi and Atherton [7]. Schematic illustration
( )
√ √ 1
of the indigenous set up is illustrated in Fig. 24 and its snap- 2gL9 = 2gL1 + g √ (L9 − L1 ) . (61)
2g (L9 − L1 )
shot is depicted in Fig. 25(a). This level process is consisting
of a rectangular overhead tank where water is fed by a pump Here, L1 and L9 are the minimum and maximum water level
submersed into the lower reservoir. Tank level is measured by positions of overhead tank. Substituting the expression of Eq. (61)
capacitive level sensor SEN18 [27] whose output is utilized as into Eq. (60)
controlled variable and feed to Arduino Nano 3.1. Control al-
d (L9 )
( ( ) )
gorithm is developed in PC using MATLAB-Simulink compatible Kq a √ 1
( ) = Vp − 2gL1 + g √ (L9 − L1 ) .
with Arduino Nano 3.1. Control signal Vp obtained from Arduino dt A A 2g (L9 − L1 )
Nano 3.1 is fed to the submersible pump through the motor
(62)
driver L293D [27]. At the outlet port of the overhead tank a
relay operated solenoid valve is connected to provide disturbance At a liquid height of 1 cm i.e., L= L1 , control signal is considered
input for a specific duration by opening it. To incorporate large to be Vp = Vpo , hence, Eq. (62) reduces to
time delay in the control loop a long run of pipe is provided
to supply water from the lower reservoir to the overhead tank. d (L1 ) Kq a (√ )
= Vpo − 2gL1 . (63)
Arduino Nano 3.1 compatible PLX-DAQ (parallel data acquisition dt A A
tool) software is employed for real-time data acquisition of the Now, subtracting Eq. (63) from Eq. (62) and considering L9,1 =
overhead tank water level. A local LCD unit is also connected to L9 − L1
display the overhead tank water level.
For designing the control algorithm, mathematical model of dL9,1 Kq a g
= Vps − √ L9,1 . (64)
the experimental set up is required to be obtained. Mathematical dt A A 2gL9,1
255
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

( )
Fig. 26. Closed loop responses and related control actions under (a) nominal and (b) perturbed conditions of the experimental set up i.e. Model VII Gp7 .

Fig.( 27.) Complementary sensitivity C (jω) and the relevant magnitude plot with +50% perturbation in ∆K= 0.045 and +5% perturbation in ∆θm = 0.4 for Model
VII Gp7 .

where Vps is the control signal defined as Vp − Vpo . By apply- for the proposed control scheme are provided in Table 1(b).
ing Laplace Transform on Eq. (64) resulting transfer function is Considerable improvement is found during hardware experimen-
specified by tation in comparison with Majhi and Atherton [7] as illustrated in
√ Fig. 25(b). Closed loop responses and related control actions are
Kq 2gL9,1
L9,1 (s) ga Kdc also observed with identified model under nominal (Eq. (67)) and
= √ = . (65) perturbed (+10%) conditions (Eq. (68)) as illustrated in Fig. 26(a)
Vps (s) A 2gL9,1 τ1 s + 1
ga
s+1 and Fig. 26(b) respectively.
Eq. (65) is an approximated mathematical model of the ex- 0.099
Ĝp7 (s) = e−8.8s . (68)
perimental tank level process without time delay. As mentioned, s
a long run of pipe is connected to supply water to the over-
Stability and robustness is confirmed by small gain theo-
head tank which introduces time delay of nearly 8 s and hence
rem [26] with perturbations in ∆K and ∆θm , and C(jω) is cal-
substituting the values of the design parameters from Table 5
culated as presented in Table 2. Improved robust stability of MSP
experimental process can be modelled as
technique is ensured with smaller amplitude of C(jω) as shown
L9,1 (s) 1.65 0.09 in Fig. 27. Quantitative performance measures are presented in
= e−8s = e−8s . (66)
Vps (s) 18.15s + 1 s + 0.05 Tables 3 and 4 substantiate the supremacy of MSP technique.
Here, pole location of the process is quite close to the imagi-
nary axis and hence it can be approximated as IPTD in nature as 6. Conclusion
given by
A novel modified Smith predictor technique is suggested for
0.09 0.09
Model VII : Gp7 (s) = e−8s ≈ e−8s . (67) IPTD processes. Such class of processes are found to be quite
s + 0.05 s common in practice for representing the behaviour of distillation
Closed loop performance analysis is conducted with the de- column, liquid storage tank with a drain valve, and superheated
rived Model VII (Eq. (67)) as well as real-time experiment is steam flow to turbine etc. Extensive simulation study is con-
also performed on the hardware set up. Controllers’ settings ducted for substantiating the effectiveness of suggested technique
256
S. Karan, C. Dey and S. Mukherjee ISA Transactions 121 (2022) 240–257

with others’ well-known dead time compensating scheme. Nov- [12] Seborg D, Edgar TF, Mellichamp D. Process dynamics and control. New
elty of the proposed IMC scheme involves sole tuning parameter, York: John Wiley and Sons; 2006.
[13] Ali A, Majhi S. PID controller tuning for integrating processes. ISA Trans.
suitable enough to tune all the controllers involved. Real-time ex-
2010;49(1):70–8.
perimentation with a level process also justifies the supremacy of [14] Davó MA, Baños A. Reset control of integrating plus dead time processes.
the proposed technique. In future, there is scope for experimental J Process Control 2016;38:22–30.
validation with other integrating process models. [15] Kumar DBS, Padma Sree R. Tuning of IMC based PID controllers for
integrating systems with time delay. ISA Trans 2016;63:242–55.
[16] Chakraborty S, Ghosh S, Naskar AK. All-PD control of pure integrating plus
Declaration of competing interest time-delay processes with gain and phase-margin specifications. ISA Trans
2017;68:203–11.
The authors declare that they have no known competing finan- [17] Medarametla PK, Manimozhi M. Novel proportional–integral–derivative
cial interests or personal relationships that could have appeared controller with second order filter for integrating processes. Asia-Pacific
J Chem Eng 2018;13(3). https://doi.org/10.1002/apj.2195.
to influence the work reported in this paper.
[18] Karan S, Dey C. Improved disturbance rejection with modified smith
predictor for integrating FOPTD processes. SN Appl Sci 2019;1:1168. https:
References //doi.org/10.1007/s42452-019-1186-9.
[19] Goud EC, Rao AS. Design of noise filters for integrating time delay
[1] Fuentes C, Luyben WL. Control of high purity distillation columns. Ind Eng processes. Chem Prod Process Model 2019. https://doi.org/10.1515/cppm-
Chem Process Des Dev 1983;22:361–6. 2019-0056.
[2] Ogunnaike BA, Ray WH. Process dynamics modelling and control. New [20] Zhang B, Tan W, Li J. Tuning of Smith predictor based generalized ADRC
York: Oxford University Press; 1994. for time-delayed processes via IMC. ISA Trans 2020;99:159–66.
[3] Monroy NK, Ramirez M. Control systems. New Delhi: New Age [21] Karan S, Dey C. Simplified tuning of IMC based modified smith predictor
International (P) Ltd Publishers; 2003. for UFOPTD processes. Chem Prod Process Model 2020. https://doi.org/10.
[4] Howard W, Cooper I. Internal model based control for integrating 1515/cppm-2019-0132.
processes. ISA Trans 2009;519–27. [22] Smith OJ. A controller to overcome dead time. ISA J 1959;6:28–33.
[5] Wang L, Cluette WR. PID tuning controllers for integrating processes. IEEE [23] Rivera DE, Morari M, Skogested S. Internal model control for PID controller
Proc Control Theory Appl 1997;144(5):385–92. design. Ind Eng Chem Proc Des Dev 1986;25:252–65.
[6] Normey Richo JE, Camacho EF. Control of dead time processes. London: [24] Routh EJ. The advanced part of a treatise on the dynamics of a system of
Springer Verlag; 2007. rigid bodies. Part II. London: MacMillan; 1905.
[7] Majhi S, Atherton DP. A new smith predictor and controller for unstable [25] Horn IG, Arulandu JA, Gombas CJ, Van Antwerp JG, Braatz RD. Improved
and integrating processes with time delay. In: Proceedings of the 37th IEEE filter design in internal model control. Ind Eng Chem Process Des Dev
conference on decision and control. 1998, https://doi.org/10.1109/cdc.1998. 1996;35:3437–41.
758471. [26] Morari M, Zafiriou E. Robust process control. Englewood Cliffs: Prentice
[8] Kaya I, Atherton DP. A new PI-PD Smith predictor for control of process Hall; 1989.
with long dead time. In: Proc. 14th IFAC world congress; 1999. [27] Santosh KV, Joy B, Rao S. Design of an instrument for liquid level
[9] Karan S, Dey C. Modified smith predictor-based all-proportional-derivative measurement and concentration analysis using multisensor data fusion.
control for second-order delay-dominated integrating processes. Asia-Pac J Sensors 2020;1–13. https://doi.org/10.1515/2020/4259509.
J Chem Eng 2020;e2591. https://doi.org/10.1002/apj.2591. [28] Marlin T. Process control, designing processes and control systems for
[10] Chidambaram M, Padma Sree R. A simple method of tuning PID controllers dynamic performance. 2nd ed. New York: McGraw Hill; 2000.
for integrator/dead time processes. Comput Chem Eng 2003;27(2):211–5. [29] Gatzke EP, Meadows ES, Wang C, Doyle F. Model based control of a
[11] Tao Liu, Cai YZ, Gu DY, Zhang WD. New modified smith predictor scheme four-tank system. Comput Chem Eng 2000;24:1503–9.
for integrating and unstable processes with time delay. IEE Proc Control
Theory Appl 2005;152(2):238–46.

257