Decentralized PID controller design

for TITO processes with experimental

validation

V.D.Hajare, B.M.Patre,
A.A.Khandekar & G.M.Malwatkar

International Journal of Dynamics

and Control

ISSN 2195-268X

Int. J. Dynam. Control

DOI 10.1007/s40435-016-0252-z

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag Berlin Heidelberg. This e-offprint is
for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com.

1 23
 Author's personal copy
Int. J. Dynam. Control
DOI 10.1007/s40435-016-0252-z
Decentralized PID controller design for TITO processes
with experimental validation
V. D. Hajare1 B. M. Patre2 A. A. Khandekar3 G. M. Malwatkar3
Received: 3 October 2015 / Revised: 30 May 2016 / Accepted: 31 May 2016
Springer-Verlag Berlin Heidelberg 2016
Abstract In this paper, a design method of decentralized
proportionalintegralderivative (PID) controller for two-
input two-output processes based on pre-defined reference
transfer function is proposed. An ideal decoupler is used to
reduce the interaction among system variables. Free struc-
ture higher order diagonal controllers are computed for each
decoupled subsystem by specifying closed-loop response in
terms of dominant reference second order transfer function.
Further, to obtain controllers in PID structure, the higher
order diagonal controllers are truncated into first three terms
of Maclaurin series. The stability of resulting PID controller
is investigated. Two benchmark examples are illustrated to
show the effectiveness of the proposed controller. An exper-
imentation is performed on interacting coupled tank process
to demonstrate the applicability in real life applications.
Keywords Multivariable systems Ideal decoupler Decen-
tralized controller Parameter uncertainty Experimental
application
B B. M. Patre
bmpatre@ieee.org
V. D. Hajare
hajare.vikas@gmail.com
A. A. Khandekar
aniket.khandekar.vit@gmail.com
G. M. Malwatkar
gajananm@gmail.com
1 Cummins College of Engineering for Women, Karvenagar,
Pune, India
2 S. G. G. S. Institute of Engineering & Technology, Nanded,
India
3 Zeal College of Engineering and Research, Pune, India
1 Introduction
Many of the industrial processes posses multiple input mul-
tiple output (MIMO) dynamics. The design of controller for
MIMO systems is difficult due to process and loop inter-
actions and hence poses a challenging control task. The
controller design methods for MIMO systems reported in
the literature can be broadly classified as centralized (full
structure) and decentralized (diagonal) controllers. Different
approaches for designing centralized controllers are reported
in the literature including recent publications such as [1,2]. In
these approaches, interactions are reduced using full structure
controller and the loop controllers interact with each other.
Hence, the tuning for individual loop controller can not be
done independently which complicates the design procedure.
In process control industries, multiloop single input sin-
gle output (SISO) controllers are often used to control plants
having MIMO dynamics [3]. The most attractive advantages
of such methods are structural simplicity and the easiness to
handle loop failure. However, due to interaction the multi-
loop SISO controllers encounter more difficulties than that
of a single loop which may result in unacceptable perfor-
mance and has became an open research topic for the years
[4]. Many design methods have been reported in literature
for multiloop SISO controller design. Some of them include
detuning methods [4,5], sequential loop closing (SLC), [68]
and independent methods [9,10]. In detuning approach, the
off diagonal elements in process transfer function matrix are
ignored and the diagonal controllers are tuned based on sin-
gle loop controller design approach. The diagonal controllers
are then detuned by the detuning factor obtained from the
interaction measure like relative gain array (RGA). A well
known detuning method is the biggest log modulus tuning
(BLT) given in [4]. In this method, each individual con-
troller is first designed for corresponding diagonal element
123
 Author's personal copy
based on ZieglerNichols (ZN) tuning rules [11]. The con-
trollers are then detuned to take into account the interactions
until some prescribed limit (e.g. the biggest log modulus)
or the multivariable Nyquist stability is attained. The only
advantage of this method is its simplicity but it can lead to
sluggish or oscillatory responses. In addition to BLT method,
Chien et al. [12] have addressed similar methods to take into
accounts the interaction by detuning. Lee and Edgar [13],
and Zhang et al. [14] presented method of pole placement by
which control performance can be improved.
In SLC methods, the loops are sequentially closed one
after the other usually starting with the fastest loop. The
dynamic interaction of earlier loop is then considered in the
closing of next loop. The design of controller may depend on
the order by which the loops are closed. The iteration pro-
cedures are also essential because closing the subsequent
loops may alter the response of the previously designed
loops. Besides the detuning and SLC design methods another
well known multiloop control design is based on indepen-
dent design method reported in [10]. In this method, SISO
controllers are designed independently by using the defined
bounds for guaranteed performance. However, the resulting
performance may be poor since the methods lack in detailed
information about the controller dynamics in other loops.
All the multiloop SISO methods available in the literature
have attempted to overcome interaction among the loops.
Tuning controller parameters of one loop affects the per-
formance of the others and sometimes leads to instability
of the entire system. For stable performance, many indus-
trial decentralized controllers are tuned loosely, which causes
poor operation and higher energy costs. Another possible way
to overcome this difficulty and to make use of SISO design
methods is to construct equivalent individual loops using
decoupler. Decentralized PID controllers with decoupler are
most common choice for TITO processes. In order to auto
tune a decoupling controller for TITO processes, a method
was suggested by Ho et al. [15] using direct Nyquist array.
Gilbert et al. [16] demonstrated a method of tuning PI con-
trollers with one-way decoupling for TITO processes using a
finite number of frequency response data. Tavakoli et al. [17]
presented a decentralized PI/PID controller for TITO systems
based on non dimensional analysis. An approach of partial
decoupling based on equivalent transfer function (ETF) is
proposed in [18]. In this approach by measuring the effect of
each ETF a partial decoupler is designed. Based on decoupler,
controllers are designed for multivariable processes. An inde-
pendent PI/PID design method based on ETFs and simplified
decoupler is proposed in [19]. Nordfeldt and Hagglund [20]
presented an ideal decoupler for the controller design under
the constraints on robustness and sensitivity to measurement
noise. Jevtovic and Matausek [21] presented decentralized
controller design method with ideal decoupler. A decentral-
ized PI/PID controller design method based on gain and
123
V. D. Hajare et al.
phase margin specifications for TITO interacting processes
is given in [22]. A graphical approach of loop margins for
TITO system is presented in [23]. Maghade and Patre [24]
presented a decentralized PID controller design method in
which two different PID controllers were obtained using first
order plus dead time (FOPDT) and second order plus dead
time (SOPDT) reduced order models of decoupled subsys-
tems. In both the control strategies, decoupler given in [17]
is used to get the decoupled subsystems. Then the decoupled
subsystems were reduced into FOPDT and SOPDT models
to design the decentralized PID controllers. Both controllers
were designed using dominant pole placement approach.
The inverted decoupling internal model control (IMC) for
square stable multivariable time delay systems was proposed
in [25] wherein the main controller is of IMC structure which
directly decouples and stabilizes square stable multivariable
processes with multiple time delays. The methodology is
based on centralized inverted decoupling proposed in [26].
Application of IMC to multiloop PID controller designs
are presented by most of the researchers. In this design
methodology, the individual PID controller parameters are
expressed in terms of closed loop time constant of each loop.
It is interesting to note that, the controllers obtained based on
IMC approach may not be a standard PID structure. Hence
in order to obtain PID structure a controller reduction is
often required. Chen and Seborg [27] presented a method of
independent design combined with Nyquist stability analy-
sis. In this method, the stability region for PI controller is
identified and appropriate set for PI parameters within the
stability region was selected. Another approach of decou-
pling IMC controller based on the normalized relative gain
array (NRGA) and ETF matrix is proposed in [28]. In this
approach NRGA, and ETF are used to approximate pseudo-
inverse of the process transfer function matrix. To improve
the robustness the filters are added.
In this paper, an ideal decoupler is used reduce the inter-
action, and an equivalent multiloop SISO structure of TITO
process model is obtained. Depending on the dynamics of
each loop, a closed loop reference transfer function is spec-
ified. The nth order reference transfer function includes
dominant pole pair and remaining (n 2) non-dominant
poles. Independent higher order controller is designed for
each loop. For the implementation purpose higher order con-
troller is reduced to PID structure by using first three terms
of Maclaurin series. The stability analysis of PID controller
is carried out. Parametric uncertainty is added in the nominal
plant and performance of the proposed controller is analysed
for robust stability. Two benchmark examples are illustrated
to show the effectiveness of the proposed method. Proposed
multiloop controller is compared with prevalent controllers
in the literature. A real time experimentation is carried out
on laboratory pilot plant of interactive coupled tank to verify
the performance of proposed method. The main contribution
 Author's personal copy
Decentralized PID controller design for TITO processes with experimental validation

of this work is the design of multiloop robust PID controller where

for higher order model of decoupled subsystem without need

of model order reduction. The loop controllers are designed Q 11 (s) 0 0
such that the dominant pole placement is achieved and robust 0 Q 22 (s) 0

Q(s) = . .. .. ..
stability with respect to parametric uncertainty is attained. .. . . .
The rest of the paper is organized as follows. In Sect. 2,
0 0 Q nn (s)
decoupling and controller design methodology are intro-
duced. In Sect. 3, stability analysis is discussed. Simulation
results are included in Sect. 4. Real time experimentation and Decoupler in Eq. (3) can be represented as,
validation is given in Sect. 5. Finally conclusions are given
in Sect. 6. D(s) = Ad j[G(s)]K (s) (4)

where K (s) is a diagonal matrix. The elements kii (s) are

2 System description and controller design obtained such that common pole-zero, common dead time
and smallest gain from ith column of Ad j[G(s)] are removed
The structure of multi-variable feedback control system with and their inverse is included in kii (s). The dead time term of
decoupler is shown in Fig. 1. Q(s) in Eq. (3) can be approximated in different ways. In this
In Fig. 1, r1 , . . . , rn and y1 , . . . , yn are reference inputs work, first-order Taylor series approximation, that is ed s =
1
and outputs of the system respectively. u 1 , . . . , u n and d s+1 is used for dead time term. The decoupled subsystems
v1 , . . . , vn are controllers outputs and decoupler outputs in Eq. (3) with delay approximation can be represented in
respectively. The controller design method is based on ref- general form as
erence transfer function of typical second order systems. An
ideal decoupler is used to obtain SISO subsystems [20]. The b0ii s m ii + b1ii s (m1)ii + + bm ii
Q ii (s) = , i = 1, 2
structure of MIMO system and ideal decoupler is given by, s n ii + a1ii s (n1)ii + + an ii
(5)

G 11 (s) G 12 (s) G 1n (s)
G 21 (s) G 22 (s) G 2n (s) where m n, aii s and bii s are polynomial coefficients.

G(s) = . .. .. .. (1) Also, it is presumed that all decoupled subsystems in Q ii (s)
.. . . .
does not contain any right half plane (RHP) zero. The closed-
G n1 (s) G n2 (s) G nn (s)
loop transfer function Tii (s) between output and set-point of
ith loop is given by
and
Q ii (s)G cii
D11 (s) D12 (s) D1n (s) Tii (s) = , i = 1, 2 (6)
1 + Q ii (s)G cii
D21 (s) D22 (s) D2n (s)

D(s) = . .. .. .. (2)
.. . . . where, G cii is the controller for each diagonal subsystem in
Dn1 (s) Dn2 (s) Dnn (s) Q ii (s). According to the IMC theory [29], the closed-loop
transfer function T f ii (s) of the system is
respectively. The equivalent multiloop SISO structure is
1
T f ii (s) = , i = 1, 2 (7)
(i s + 1)qi
Q(s) = G(s)D(s), (3)
where, i indicates IMC filter constant, the term qi is the
relative order of the numerator and denominator in Q ii (s)
and chosen equal to or greater than n ii m ii . The reference
closed-loop trajectory given in Eq. (7) includes repeated real
poles. The expected response of system is critically damped
which depends on two unknown parameters, i and qi . The
selection of design parameter i in IMC is a key decision and
several guidelines have been published in the literature [29].
However, different guidelines generate different performance
Fig. 1 Block diagram of closed-loop system with decoupler for the same process. To get liberty from such limitations and

123
 Author's personal copy
V. D. Hajare et al.

to place the dominant poles, without loss of generality, the 3 Stability analysis
desired reference transfer function can be selected as
From a well known generalized Nyquist stability theorem
1 ni
2 presented in [30] assume that G system structure shown
Tdii (s) = (8) in Fig. 2 is stable where G(s), (s) are the nominal system
(i s + 1)qi 2 s 2 + 2i ni s + ni
2
and the perturbation respectively. Consider that, if is the
convex set of perturbation and  is an allowed perturbation,
where, i and ni indicate damping factor and undamped
then   is small change in  where  is any real scalar such
natural frequency respectively. The term i (0.7, 0.99)
that || 1. Then system shown in Fig. 2 is stable for all 
which is selected from peak overshoot and ni is calculated
if and only if any one of the following equivalent conditions
from expected settling time ts using relation ni = 4/(i tsi ).
is satisfied:
The closed-loop poles can be placed at desired location by
assigning value of i and ni . The term i in Eq. (8) is chosen
such that (qi 2) repeated poles are far away from dominant 1. Nyquist plot of det (I G(s)) does not encircle the
pole pair to make them non-dominant. Equating Eqs. (6) and origin, , i.e. det (I G( j)) = 0, ,
(8), the controller G cii (s) can be written as 2. i G( j) = 1, i, ,
3. G( j) < 1, , , where is spectral radius.
max
1 4. G( j) < 1,
G cii (s) = gii (s) (9)
s
Now a nominal (22) multiloop control structure from Fig. 1
where, is nominally stable if and only if

s ni
2
gii (s) = 1. G C11 G C22
Q ii (s) (1 + i s)qi 2 (s 2 + 2i ni s + ni
2 ) 2 and are stable
ni 1 + G 11 G C11 1 + G 22 G C22

Note that in Eq. (9) there is no non-minimum phase part

of Q ii (s) and hence the controller has neither causality nor G 12 G C11
0
stability issue. 2.

1 + G 11 G C11 < 1,

G 21 G C22
Multi-loop controllers obtained in Eq. (9) are of non PID 0
1 + G 22 G C22
structure. In view of the fact that most common controllers
used in process industries are of PI/PID form, to obtain a
Actuator uncertainties, and measurement uncertainties due
multi-loop PID controller structure, Eq. (9) can be approxi-
to sensor are often encountered is of lump multiple sources
mated most closely using Maclaurin series expansion as
of a multiplicative form uncertainty. In this work to identify
 
the closed-loop system robust stability in the presence of
1  gii (0) 2 such uncertainties, Eqs. (13) and (14) are used to evaluate
G cgii (s) = gii (0) + gii (0)s + s + (10)
s 2! the closed loop system robust stability.
 
The standard form of PID controller is G C (I + GG C )1 G I < 1, (13)
 
K I ii GG C (I + GG C )1 G O < 1, (14)
G cii (s) = K Pii + + K Dii s
s
1
= (K I ii + s K Pii + s 2 K Dii ) (11)
s

where, K Pii , K I ii and K Dii represent proportional, inte-

gral and derivative gain respectively. Hence, using first three
terms of Eq. (10) a standard PID controller is,


 gii (0)
K Pii = gii (0); K I ii = gii (0); K Dii = (12) Fig. 2 General G structure
2!

123
 Author's personal copy
Decentralized PID controller design for TITO processes with experimental validation

where, I and O are stable process multiplicative input 9.75

Q 22 (s) =
and output uncertainties respectively. Hence, for a specified (10.9s + 1)(21s + 1)(9s + 1)
bound of I and O , control system robust stability can be 19.4
.
evaluated by observing the magnitude plots of the left sides (14.4s + 1)(16.7s + 1)(3s + 1)
of Eqs. (13) and (14) with [0, +], which should fall
below unity. This can be conveniently performed by control Using Eq. (8), with 1 = 2 = 0.01, q1 = q2 = 3, [1 , 2 ] =
software packages, like MATLAB robust control toolbox. [0.7, 0.8] and settling time [ts1 , ts2 ] = [8, 25], the resulting
This is demonstrated in simulation Example 1. higher order free controllers G ch1 and G ch2 are obtained and
given in Eqs. (15) and (16).

352.6s 6 + 497.8s 5 + 168s 4 + 24.64s 3 + 1.805s 2 + 0.065s + 0.00092

G ch1 = (15)
s(s 5 + 101.2s 4 + 122.1s 3 + 22.98s 2 + 1.579s + 0.034)
180.5s 6 128.7s 5 33.04s 4 4.133s 3 0.2726s 2 0.0092s 0.00012
G ch2 = (16)
s(s 5 + 100.5s 4 + 52.63s 3 + 7.98s 2 + 0.48s + 0.0094)

Higher order free structure controllers G ch1 and G ch2 are

4 Simulation examples converted into PID structure using Eq. (12). The resulting
PID controller parameters are given in Table 1.The perfor-
4.1 Example 1: WoodBerry (WB) binary distillation mance of proposed method is compared with Maghade and
column process Patre [22], NDT [17] method and Shen et al. [32] methods. In
1
this example Maghade and Patre have used filters 5.5252s+1
Wood and Berry introduced the transfer function model of 1
for first and 5.1499s+1 for second PID controller.
a pilot-scale distillation column, which consists of an eight-
The desired location of closed loop poles with higher order
tray plus reboiler separating methanol and water [31]. The
controllers and PID controllers are given in Table 2. Output
WoodBerry binary distillation column process is a multi-
response of the system with higher order free controller and
variable system that has been studied extensively by many
PID controller is shown in Fig. 3. It is observed that response
researchers [17,22]. The process has the transfer function
is nearly same for both the controllers.
matrix as
For nominal model, tsi settling time in minute and I S E i
integral square error are given in Table 3. In this example,
12.8es 18.9e3s
16.7s + 1 NDT and Maghade methods are based on decoupler strategy
21s + 1 [17].
G(s) = .
6.6e7s 19.4e3s
 
10.9s + 1 14.4s + 1 1.477(16.7s+1)e2s
1
D N DT (s) = 0.34(14.4s+1)e4s
21s+1

10.9s+1 1
The decoupler determined using Eq. (4) is
and Shen et al. [32] is based on adjoint transfer matrix decou-
2.94 1.477e2s
14.4s + 1 pling strategy.
21s + 1
D(s) = . 
e4s 1 19.4 18.9e2s
10.9s + 1 16.7s + 1 D(s) Shen = 14.4s+1 12s+1
6.6 12.8
10.9s+1 16.7s+1

The Decoupled subsystems are

The unit step change is applied to the first set-point for time
t 0 keeping second set-point zero for 0 t 120 and
37.63
Q 11 (s) = changed it to unity for time t 120. The output responses
(16.7s + 1)(14.4s + 1)(s + 1)
given by various controllers are show in Fig. 4.
18.9
, From Fig. 4, it is clear that Maghade and Patres method
(21s + 1)(10.9s + 1)(7s + 1) produce oscillatory responses while Y. Shens method gives
Q 12 (s) = Q 21 (s) = 0, slow response. NDT method is comparable with the proposed
controller, however the interaction given by proposed method
and is less compared to other controllers.

123
 Author's personal copy
V. D. Hajare et al.

Table 1 Tuning Parameters for

Method [K P11 K I 11 K D11 ] [K P22 K I 22 K D22 ]
WoodBerry Example
Proposed [0.657 0.027 4.301] [0.312 0.013 2.175]
Maghade and Patre [22] [0.9733 0.0881 2.6887] [0.3134 0.0304 0.807]
Shen et al. [32] [0.06187 0.002542 0.3421] [0.02062 0.00085 0.1141]
NDT [17] [0.41 0.074 ] [0.120 0.024 ]

Table 2 Location of desired

Subsystem Desired Higher order PID controller
closed loop poles for Example 1
controller (n = 6)

Q 11 0.5 0.51i 0.5 0.51i 0.52 0.63i

Q 22 0.16 0.12i 0.16 0.12i 0.19 0.15i

1.2
Higher order Controller
PID Controller
1
1

Output (y )
1
0.8
NDT (2006)
Maghade (2012)
Shen (2012)
y1

0.6 0.5
Proposed
0.4

0
0.2 0 50 100 150 200 250
Time (min)
0
0 10 20 30 40 50 60 70 80 (a)
Time(Min)
(a)
1 NDT (2006)
1.4 Maghade (2012)
Higher order Controller
Output (y )

Shen (2012)
2

PID Controller
1.2
Proposed
0.5
1

0.8
y2

0
0.6

0.4 0 50 100 150 200 250

Time (min)
0.2
(b)
0
0 10 20 30 40 50 60 70 80
Fig. 4 Responses of Example 1. a y1 response to unit step in r1 at
Time(Min)
t = 0 and r2 at t = 120. b y2 response to unit step in r1 at t = 0
(b) and r2 at t = 120

Fig. 3 Output responses of Example 1 for higher order and PID con-
troller. a Response of y1 of Example 1. b Response of y2 of Example 1

Table 3 Performance indices

for nominal and 20 %
uncertainty in model for the WB
column
Method Nominal +20 % parametric uncertainty 20 % parametric uncertainty
[ts1 , ts2 , I S E 1 , I S E 2 ] [ts1 , ts2 , I S E 1 , I S E 2 ] [ts1 , ts2 , I S E 1 , I S E 2 ]
Proposed [13.5, 25, 1.82, 5.62] [12, 20, 1.82, 5.78] [18.7, 40, 1.71, 3.81]
Maghade [35, 22, 2.73, 4.84] [50, 60, 4.1, 7.98] [7.8, 150, 2.76, 18.02]
NDT [23, 30, 2.35, 5.32] [17, 50, 3.01, 8.38] [10.8, 30, 2.28, 4.97]
Y. Shen et al. [20, 28, 2.32, 6.92] [15, 62, 2.67, 8.16] [29.6, 34, 2.16, 6.15]

123
 Author's personal copy
Decentralized PID controller design for TITO processes with experimental validation

Nominal
1 I
O
1

Output (y1)
0.8
Magnitude (abs)

NDT (2006)
0.6
Maghade (2012)
0.5 Shen (2012)
0.4
Proposed
0.2
0
0 50 100 150 200 250
2 1 0
10 10 10 Time (min)
Frequency (rad/s)
(a)
Fig. 5 Magnitude plots of spectral radius for Example 1 1.5

Output (y )
2
To show the multi-loop control system stability of the pro-
posed method, assume that the process multiplicative input NDT (2006)
0.5
uncertainty ( I ) in loop-1 and loop-2 is (s + 0.3)/(s + 1). Maghade (2012)
It means that inputs supplied by the corresponding actua- Shen (2012)
0 Proposed
tors to the process are increased up to 100 % uncertainty at
high frequencies and 30 % uncertainty in the low frequency
range. Further, assume that the process multiplicative output 0 50 100 150 200 250
Time (min)
uncertainty ( O ) in loop-1 and loop-2 is (s + 0.2)/(2s + 1).
(b)
It means measurements provided by the corresponding sen-
sors decrease with up to 20 % uncertainty at high frequencies
and with almost 20 % uncertainty in the low-frequency range.
Figure 5 shows the magnitude plots of spectral radius in terms
1
Output (y )

of the assumed nominal, I and O for proposed control sys-

1

NDT (2006)
tem robust stability. It is observed that, I and O falls below
Maghade (2012)
the unity. 0.5 Shen (2012)
To investigate the robustness in case of process parametric Proposed
uncertainty, all four time delays, time constants and gains are
changed by 20 %. The closed loop responses for different 0
0 50 100 150 200 250
control schemes are shown in Fig. 6 and quantitative perfor- Time (min)
mance indices are tabulated in Table 3. It can be concluded
(c)
that performance of the proposed controller is robust with
less interaction among the variables.
1
Output (y2)

4.2 Example 2: Industrial-scale polymerization (ISP) NDT (2006)

reactor Maghade (2012)
0.5 Shen (2012)
Proposed
The transfer function matrix for an ISP reactor system was
introduced by Chien et al. [12] as follows: 0

22.89e0.2s 11.64e0.4s 0 50 100 150 200 250
4.572s + 1 1.807s + 1
Time (min)
G(s) =
4.689e0.2s
.
5.8e0.4s
(d)
2.174s + 1 1.801s + 1 Fig. 6 Robustness studies of Example 1 for 20 % parameter uncer-
tainty. a y1 response with +20 % model uncertainty. b y2 response with
The decoupler for this system is +20 % model uncertainty. c y1 response with 20 % model uncertainty.
d y2 response with 20 % model uncertainty

5.80e0.2s 11.64e0.2s
1.801s + 1 1.807s + 1
D(s) =
4.689
.
22.89
2.174s + 1 4.572s + 1

123
 Author's personal copy
V. D. Hajare et al.

Table 4 Tuning Parameters for

Method K P11 K I 11 K P22 K I 22 Remark
ISP reactor
Proposed 0.0248 0.0044 0.0248 0.0044 (K D11 , K D12 ) = (0.0257, 0.0257)
Xiong and Cai [5] 0.219 0.0479 0.1703 0.0946
Xiong et al. [33] 0.3137 0.0686 0.2439 0.1354 (K p12 , K i12 ) = (0.2203, 0.1013)
and (K p21 , K i21 ) = (0.0369, 0.0204)
Jin and Liu [28] 0.0098 0.0022 0.0098 0.0022

Table 5 Location of desired

Subsystem Desired Higher order controller (n = 6) PID controller
closed loop poles for Example 2
Q 11 1.33 0.65i 1.33 0.65i 1 0.54i
Q 22 1.33 0.65i 1.33 0.65i 1 0.54i

1
The Decoupled subsystems with delay approximation are 0.9 Higher order Controller
PID Controller
0.8
132.762
Q 11 (s) = 0.7

(4.572s + 1)(1.801s + 1)(0.4s + 1) 0.6

y1
54.58 0.5

+ , 0.4
(1.807s + 1)(2.174s + 1)(0.4s + 1) 0.3
Q 12 (s) = Q 21 (s) = 0, 0.2

0.1

and 0
0 5 10 15
Time(Hrs.)

132.762 (a)
Q 22 (s) =
(4.572s + 1)(1.801s + 1)(0.4s + 1)
1

0.9 Higher order Controller

PID Controller
54.58
+ .
0.8

(1.807s + 1)(2.174s + 1)(0.4s + 1) 0.7

0.6
y2

The procedure explained in previous section and using Eq. (8) 0.5

with terms 1 = 2 = 0.01, q1 = q2 = 3, 1 = 2 = 0.9

0.4

0.3
and ts1 = ts2 = 3 Hr s. The free higher order controllers 0.2

G ch1 = G ch2 are 0.1

0
0 5 10 15
Time(Hrs.)

2.925s 5 + 12.54s 4 + 16.46s 3 + 9.42s 2 + 2.431s + 0.226 (b)

G ch1,2 =
s(s 4 + 103.6s 3 + 361.7s 2 + 262.5s + 51.88)
As explained previously in Example 1, G ch1,2 is converted
into PID structure. The parameters of proposed PID con-
troller and other methods are given in Table 4. The location
of desired closed loop poles with higher order controller, and
PID controller are given in Table 5. The output response of the
system with higher order free controller and PID controller
is shown in Fig. 7 which is nearly same.
In this simulation study, the proposed method is compared
with Xiong and Cais effective transfer function (ETF) [5],
Xiong et al. [33], and Jin and Liu [28]. Xiong and Cai con-
trollers are tuned assuming that diagonal elements of transfer
function matrix are SOPDT. The Xiong et al. method is based
on the full structure controller. The tuning controller parame-
ters are given in Table 4. The controllers by Jin and Liu [28]
are based on IMC theory with Ms = 2.
Fig. 7 Output responses of Example 2 for higher order and PID con-
troller. a Response of y1 of Example 2 . b Response of y2 of Example
2
The closed-loop response of the system is shown in Fig. 8
for unit step change in first set-point from t 0 while second
set-point is changed from 0 to 1 at t = 20 and brought back
to 0 at t = 40. To investigate robustness with respect to
parameter uncertainty, all dead time elements, time constants
and steady state gains are changed by 20 %. The resulting
performance indices are given in Table 6.
From Figs. 8 and 9 it is clear that the response given
by proposed method is better than other in terms of less-
interaction. The responses given by Jin and Liu, and Xiong
and Cai methods are highly interactive while Xiong et
al.s full structure controller results comparable performance
with proposed method. Therefore, it is concluded that the

123
 Author's personal copy
Decentralized PID controller design for TITO processes with experimental validation

Response of the process (y1)

1.5 (LT) is installed in each tank to provide 420 mA corresponds
to 0 to 100 % level. The inlet flow of each tank is supplied with
the help of positive displacement pump driven by a 230 V,
1
50 Hz motor. Level control system is interfaced with personal
Jin and Liu (2014) computer using PCI 6014E and BNC 2120. These devices are
0.5 Xiong, Cai (2006) voltage compatible, therefore, the level transmitters signals
Xiong et al. (2007)
Proposed
are connected to BNC 2120 analog input channels through
current to voltage (I/V) converter circuit. The controller out-
0
0 10 20 30 40 50 60 puts which is in the range of 05 V are given to BNC 2120
Time
analog output channels and then converted into 420 mA as
(a)
input to variable frequency drives (VFDs). The control algo-
rithm is implemented in Mathworks MATLAB R2012a. To
Response of the process (y2)

Jin and Liu (2014)

1
Xiong, Cai (2006) determine the transfer function matrix model of the system
0.8 Xiong et al. (2007) step test is performed. During step test, valves (V1 , V2 and
Proposed
0.6 V3 ) shown in Fig. 11 are kept 50 % open. The accuracy of the
0.4 obtained model is verified by giving step of 5 V from com-
0.2 puter and obtaining the real time output results of the system
0
and comparing it with output of the model. The results of
0.2
the model output and real time output using pump 1 (pump
0 10 20 30 40 50 60 2 off) are shown in Fig. 12a for tank 1 and Fig. 12b for tank
Time
2 where as using pump 2 (pump 1 off) are shown in Fig. 13a
(b)
for tank 2 and Fig. 13b for tank 1.
Fig. 8 Responses of Example 2. a y1 response to unit step in r1 at Then overall system transfer function matrix for a level
t = 0 and r2 for 20 t 40. b y2 response to unit step in r1 at system is obtained as follows:
t = 0 and r2 for 20 t 40

0.43e5s 0.145e10s
29s + 1 40s + 1
controller obtained by proposed method gives comparable G(s) =
0.172e10s 0.37e5s
(17)
performance with full structure controllers.
35s + 1 27s + 1

The decoupler for this system is

5 Experimentation

2.55 1e5s
To explore the real time applicability and effectiveness of 27s + 1 40s + 1
D(s) =
1e5s
(18)
2.5
the proposed method, an experiment has been carried out on
coupled tank level system. The laboratory experimental set-
35s + 1 29s + 1
up and schematic of the coupled tank level system with data
acquisition is shown in Figs. 10 and 11 respectively. The Decoupled subsystems with delay approximation are

5.1 Experimental set-up 1.0965

Q 11 (s) =
(27s + 1)(29s + 1)(5s + 1)
The system includes coupled tank interlinked with an 0.172
,
adjustable valve V1 . At the bottom of each tank, valve V2 and (35s + 1)(40s + 1)(15s + 1)
V3 are provided for gravity flow discharge. A level transmitter Q 12 (s) = Q 21 (s) = 0,

Table 6 Performance indices of

Method
ISP reactor for nominal model
and model with 20 %
parametric uncertainty
Proposed
Xiong and Cai [5]
Xiong et al. [33]
Jin and Liu [28]
Nominal model
[ts1 , ts2 , I S E 1 , I S E 2 ]
[3, 3, 8.76, 17.52]
[10, 7, 10.74, 30.17]
[3, 5, 4.48, 12.4]
[11, 16, 24.94, 50.1]
+20 % uncertainty
[ts1 , ts2 , I S E 1 , I S E 2 ]
[3, 3, 8.76, 17.52]
[12, 5, 13.29, 31.45]
[3, 3, 4.98, 13.04]
[18, 18, 26.77, 54.49]
20 % uncertainty
[ts1 , ts2 , I S E 1 , I S E 2 ]
[3, 3, 8.76, 17.52]
[8, 10, 8.61, 30.83]
[5, 6, 4.65, 12.27]
[15, 10, 24.89, 49.31]

123
 Author's personal copy
V. D. Hajare et al.
Response of the process (y1)

1.5

Jin and Liu (2014)

0.5 Xiong, Cai (2006)
Xiong et al. (2007)
Proposed
0
0 10 20 30 40 50 60
Time
(a)
Response of the process (y2)

Jin and Liu (2014)

Xiong, Cai (2006) Fig. 10 Experimental set-up
1
Xiong et al. (2007)
Proposed

0.5

0 10 20 30 40 50 60
Time
(b)
Response of the process (y1)

1.5

Jin and Liu (2014)

Fig. 11 Schematic diagram of Coupled tank system
Xiong, Cai (2006)
0.5 Xiong et al. (2007)
50
Proposed

40
0
% Level in Tank 1

0 10 20 30 40 50 60 Model level
Actual level
Time 30
(c)
20
Response of the process (y2)

Jin and Liu (2014) 10

1
Xiong, Cai (2006)
0.8 Xiong et al. (2007) 0
Proposed 0 50 100 150 200 250 300
0.6 Time in seconds
0.4 (a)
0.2 20

0
% Level in Tank 2

15
0.2
0 10 20 30 40 50 60
Model level
Time Actual level
10
(d)
Fig. 9 Robustness studies of Example 2 for 20 % parameter uncer- 5

tainty. a y1 response with +20 % model uncertainty. b y2 response with

+20 % model uncertainty. c y1 response with 20 % model uncertainty. 0
0 50 100 150 200 250 300
d y2 response with 20 % model uncertainty Time in seconds
(b)
Fig. 12 The step test responses of model G 11 , G 21 and tank 1, 2 levels.
a For G 11 (s). i.e. y1 /u 1 . b For G 21 (s). i.e. y2 /u 1

123
 Author's personal copy
Decentralized PID controller design for TITO processes with experimental validation

40
Table 7 PID Parameters for Experimental set-up
35

30 Model level Controller For Q 11 [K p , K i , K d ] For Q 22 [K p , K i , K d ]

% Level in Tank 2

Actual level
25
Proposed [1.1508, 0.0267, 11.68] [1.3641, 0.0316,13.84]
20
NDT [17] [10.1584, 0.3865, ] [12.8357, 0.4515, ]
15
RC [19] [9.78, 0.56, ] [10.58, 0.20, ]
10

0
0 50 100 150 200 250 300
Time in seconds
of RC controller are computed using simplified IMC (SIMC).
(a) Both controllers are realized with decoupler given as
20  
10.5813s0.3919 5s
1 e
DT (s) = 11.6s0.4 5s
35s+1
40s+1 e 1
% Level in Tank 1

15 Model level
Actual level

10 In order to show the performance of the proposed con-

troller in comparison with the controller available in the
5
literature, the NDT and RC controllers are implemented in
Mathworks Matlab R2012a and the same are inter-connected
0
0 50 100 150 200 250 300 to real time coupled tank system. The performance compari-
Time in seconds son of the proposed controller and others are shown in Fig. 14.
(b) The response of the tank 1 is shown in Fig. 14a, where an
Fig. 13 The step test responses of model G 22 , G 12 and tank 1, 2 levels.
a For G 22 (s). i.e. y2 /u 2 . b For G 12 (s). i.e. y1 /u 2

726.52s 6 + 284.62s 5 + 38.15s 4 + 2.43s 3 + 0.0802s 2 + 0.0013s + 0.0009152

G che1 =
s(s 5 + 100.2s 4 + 19.58s 3 + 1.32s 2 + 0.0357s + 0.0033)
861.2s 6 + 337.4s 5 + 45.23s 4 + 2.88s 3 + 0.095s 2 + 0.0016s + 0.0001215
G che2 =
s(s 5 + 100.2s 4 + 19.58s 3 + 1.32s 2 + 0.0357s + 0.00389)

external disturbance of 10 % is added at t = 200 s, the corre-

and
0.925
Q 22 (s) =
(27s + 1)(29s + 1)(5s + 1)
0.145
(35s + 1)(40s + 1)(15s + 1)
Higher order free structure controllers G che1 , G che2 are com-
puted using Eq. (8) with terms 1 = 2 = 0.01, q1 = q2 = 3,
1 = 2 = 0.9 and ts1 = ts2 = 140 s and are mentioned on
next page.
The resulting PID controller parameters are computed
using Eq. (12) are given in Table 7. The performance of the
proposed controller is compared NDT [17] and RC [19]. In
this experiment controllers parameters for NDT are com-
puted with G M > 3 and P M > 60 . Controller parameters
sponding effect on tank 2 is shown in Fig. 14b. Similarly the
response of the tank 2 is shown in Fig. 14b, where an external
disturbance of 10 % is added at t = 400 s, the corresponding
effect on tank 1 is shown in Fig. 14a. The proposed controller
gives less overshoot as compared to NDT and RC in both
tanks as well as provides comparable performance in event
of an external disturbances. The controller action given by
the proposed controller is smooth where as controller action
by NDT and RC controllers are non-smooth. The controller
actions of the proposed controller and those of NDT and RC
controllers are shown in Fig. 15. To show the capability of the
proposed controller to eliminate the interaction, the experi-
mentation is performed, by changing the opening of V1 for
25, 50, and 75 %. The corresponding results are compared
and are shown in Figs. 16 and 17. The change in the opening
of V1 , not only changes the amount of interaction, but also,
it changes the operating model of the system. The change
in operating model causes parametric uncertainty. From, the
results, it can be seen that the controller is taking appropriate

123
 Author's personal copy
V. D. Hajare et al.

60 50
Proposed
Tavakoli (2006) 45
50 RC (2012)
40
% Level in Tank 1

35 V1 25% Open
40

% Level in Tank 1
V1 50% Open
30
V1 75% Open
30 25

20
20
15

10
10
5

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time in Seconds Time in Seconds
(a) (a)
50
50
45
45
40
40
35
% Level in Tank 2

Proposed 35 V1 25% Open

% Level in Tank 2
30 Tavakoli (2006)
30 V1 50% Open
RC (2012)
25 V1 75% Open
25
20
20
15
15
10
10
5
5
0
0 100 200 300 400 500 600 0
0 100 200 300 400 500 600
Time in Seconds Time in Seconds
(b) (b)
Fig. 14 Controller responses for level control experimental set-up. a Fig. 16 Tank level responses for level control experimental set-up for
y1 response to step in r1 at t = 0 and disturbance at t = 200. b y2 V1 opening. a y1 response to step in r1 at t = 0 and disturbance at
response to unit step in r2 at t = 0 disturbance at t = 400 t = 200. b y2 response to unit step in r2 at t = 0 disturbance at
t = 400
5
Proposed
4.5 NDT (2006) 4
RC (2012) V1 25% Open
Controller 1 Response

4 3.5 V1 50% Open

% Controller 1 Response

V1 75% Open
3.5 3
3 2.5
2.5
2
2
1.5
1.5
1
1
0.5
0.5
0
0 0 100 200 300 400 500 600
0 100 200 300 400 500 600
Time in Seconds Time in Seconds

(a) (a)
5
5
4.5
4.5
% Controller 2 Response

4
4 3.5
Controller 2 Response

3.5 3
3 2.5 V1 25% Open
V1 50% Open
2.5 2
V1 75% Open
1.5
2
Proposed 1
1.5 NDT (2006)
RC (2012) 0.5
1
0
0 100 200 300 400 500 600
0.5
Time in Seconds
0
0 100 200 300 400 500 600 (b)
Time in Seconds
(b) Fig. 17 Controller responses for level control experimental set-up for
V1 opening. a u 1 response to step in r1 at t = 0 and disturbance at
Fig. 15 Controller responses for level control experimental set-up. a t = 200. b u 2 response to unit step in r2 at t = 0 disturbance at
u 1 response to step in r1 at t = 0 and disturbance at t = 200. b u 2 t = 400
response to unit step in r2 at t = 0 disturbance at t = 400

123
 Author's personal copy
Decentralized PID controller design for TITO processes with experimental validation
action and retaining the output responses almost same for
different openings of V1 . This shows the proposed controller
is robust against parametric uncertainty. Also, it can be seen
that, the changes in the amount of interactions is handled by
the proposed controller.
6 Conclusions
This paper proposes a method to design decentralised PID
controller for TITO process based on the reference dominant
transfer function. The TITO process is decoupled through
an ideal decoupler and controllers are obtained from decou-
pled subsystems. The Maclaurin series expansion is used
to determine parameters of the PID controller. The robust
stability of the resulting PID controller is investigated. To
show the effectiveness of proposed technique, simulation
examples are included and the performance of the proposed
controller is compared with the prevalent controllers. To
show the robustness of the proposed controller, 20 % para-
metric uncertainty is added simultaneously, in all time delays,
time constants and gains. Simulation results illustrate that
this design method provides good performance with less
interaction and has robust performance under the effect of
parametric uncertainty. To show the applicability of the pro-
posed controller, an experimentation is performed on real
life interacting coupled tank liquid level system. The exper-
imental result shows that the controller outputs are smooth
which drives the pumps smoothly even under the effect of
disturbance. Such type of smooth behaviour of controller is
expected which gives long life to final control elements with-
out wear and tear.
References
1. Kumar VV, Rao VSR, Chidambaram M (2012) Centralized PI con-
trollers for interacting multivariable processes by synthesis method.
ISA Trans 51(3):400409
2. Luan X, Chen Q, Liu F (2014) Centralized PI control for high
dimensional multivariable systems based on equivalent transfer
function. ISA Trans 53(4):15541561
3. Wang QG, Huang B, Guo X (2000) Auto-tuning of TITO decou-
pling controllers from step tests. ISA Trans 39(4):407418
4. Luyben WL (1986) Simple method for tuning SISO controllers in
multivariable systems. Ind Eng Chem Process Des Dev 25(3):654
660
5. Xiong Q, Cai WJ (2006) Effective transfer function method for
decentralized control system design of multi-input multi-output
processes. J Process Control 16(8):773784
6. Chiu MS, Arkun Y (1992) A methodology for sequential design of
robust decentralized control systems. Automatica 28:9971001
7. Mayne DQ (1973) The design of linear multivariable systems.
Automatica 9:201207
8. Shiu SJ, Hwang SH (1998) Sequential design method for multi-
variable decoupling and multiloop PID controllers. Ind Eng Chem
Res 37(1):107119
9. Grosdidier P, Morari M (1986) Interaction measures for systems
under decentralized control. Automatica 22:309319
10. Hovd M, Skogestad S (1993) Improved independent design of
robust decentralized controllers. J Process Control 3(1):4351
11. Ziegler JG, Nichols NB (1942) Optimum settings for automatic
controllers. ASME Trans 64:759768
12. Chien IL, Huang HP, Yang JC (1999) A simple multiloop tuning
method for PID controllers with no proportional kick. Ind Eng
Chem Res 38:14561468
13. Lee J, Edgar TF (2006) Multiloop PI/PID control system improve-
ment via adjusting the dominant pole or peak amplitude ratio. Chem
Eng Sci 61:16581666
14. Zhang Y, Wang QG, Astrom KJ (2002) Dominant pole placement
for multi-loop control systems. Automatica 38(7):12131220
15. Ho WK, Lee TH, Xu W, Zhou JR, Tay EB (2000) The direct
Nyquist array design of PID controllers. IEEE Trans Ind Electron
47(1):175185
16. Gilbert AF, Yousef A, Natarajan K, Deighton S (2002) Tuning of PI
controllers with one-way decoupling in 22 MIMO systems based
on finite frequency response data. J Process Control 13(6):553567
17. Tavakoli S, Griffin I, Fleming PJ (2006) Tuning of decen-
tralised PI(PID) controllers for TITO processes. Control Eng Pract
14(9):10691080
18. Shen Y, Li S, Li N, Cai WJ (2011) Partial decoupling control for
multivariable processes. Ind Eng Chem Res 50(12):73807387
19. Rajapandiyan C, Chidambaram M (2012) Controller design for
MIMO processes based on simple decoupled equivalent transfer
functions and simplified decoupler. Ind Eng Chem Res 51:12398
12410
20. Nordfeldt P, Hagglund T (2006) Decoupler and PID controller
design of TITO systems. J Process Control 16:923936
21. Jevtovic BT, Matausek MR (2010) PID controller design of TITO
system based on ideal decoupler. J Process Control 20(7):869876
22. Maghade DK, Patre BM (2012) Decentralized PI/PID controllers
based on gain and phase margin specifications for TITO processes.
ISA Trans 51(4):550558
23. Le BN, Wang QG, Lee TH (2014) A graphical approach to com-
puting loop gain margins for tito systems. Trans Inst Meas Control
36(5):600603
24. Maghade DK, Patre BM (2014) Pole placement by PID controllers
to achieve time domain specifications for TITO systems. Trans Inst
Meas Control 36:506522
25. Garrido J, Vazquez F, Morilla F (2014) Inverted decoupling internal
model control for square stable multivariable time delay systems.
J Process Control 24(11):17101719
26. Garrido J, Vazquez F, Morilla F (2013) Centralized inverted decou-
pling control. Ind Eng Chem Res 52(23):78547866
27. Chen D, Seborg DE (2003) Design of decentralized PI control
system based on Nyquist stability analysis. J Process Control
13(1):2739
28. Jin QB, Liu Q (2014) Decoupling proportionalintegralderivative
controller design for multivariable processes with time delays. Ind
Eng Chem Res 53(2):765777
29. Sebor DE, Edger TF, Mellichamp DA (2004) Process dynamics
and control, 2nd edn. Wiley, New York
30. Skogestad S, Postletwaite I (1996) Multivariable feedback control:
analysis and design, 2nd edn. Wiley, New York
31. Wood RK, Berry MW (1973) Terminal composition control of a
binary distillation column. Chem Eng Sci 28(9):17071717
32. Shen Y, Sun Y, Li S (2012) Adjoint transfer matrix based decou-
pling control for multivariable processes. Ind Eng Chem Res
51(50):1641916426
33. Xiong Q, Cai WJ, He MJ (2007) Equivalent transfer function
method for PI/PID controller design of MIMO processes. J Process
Control 17:665673
123