You are on page 1of 26

Journal Pre-Proof

Optimal Tuning of Decentralized Fractional Order PID Controllers for TITO


Process using Equivalent Transfer Function

S.K. Lakshmanaprabu, Mohamed Elhoseny, K. Shankar

PII: S1389-0417(18)30683-1
DOI: https://doi.org/10.1016/j.cogsys.2019.07.005
Reference: COGSYS 868

To appear in: Cognitive Systems Research

Received Date: 25 September 2018


Revised Date: 9 April 2019
Accepted Date: 21 July 2019

Please cite this article as: Lakshmanaprabu, S.K., Elhoseny, M., Shankar, K., Optimal Tuning of Decentralized
Fractional Order PID Controllers for TITO Process using Equivalent Transfer Function, Cognitive Systems
Research (2019), doi: https://doi.org/10.1016/j.cogsys.2019.07.005

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will
undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing
this version to give early visibility of the article. Please note that, during the production process, errors may be
discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier B.V.


JOURNAL PRE-PROOF

Optimal Tuning of Decentralized Fractional Order PID


Controllers for TITO Process using Equivalent Transfer
Function
Lakshmanaprabu SK1,* Mohamed Elhoseny2 , Shankar K3

1Department of Electronics and Instrumentation Engineering, B. S.Abdur Rahman Crescent Institute of

F
Science and Technology, Chennai, India. Email: prabusk.leo@gmail.com

O
2Faculty of Computers and Information, Mansoura University,
Egypt, mohamed_elhoseny@mans.edu.eg

O
3School of Computing, Kalasalingam Academy of Research and Education, Krishnankoil, India.
Email: shankarcrypto@gmail.com

PR
Abstract: E-
This paper presents a method of designing independent fractional order PI/PID controller for two
PR
interacting conical tank level (TICTL) process based on the equivalent transfer function (ETF) model and
simplified decoupler. The TICTL process is decomposed into independent single input single output
(SISO) model using ETF. A bat optimization algorithm is utilized to independently design a diagonal
fractional order PI/PID controller based on ETF model. The effectiveness of the proposed method is
AL

illustrated with simulation examples and also the experimental TICTL process utilized to validate the
proposed method.
N

Keywords: Fractional order PI/PID control, TITO process, Optimal Control, Decoupler, Equivalent
transfer function, Bat algorithm.
R
U

1. INTRODUCTION:
JO

Controlling the multi input multi output interconnected process is the most challenging problems for
control system engineer. Several authors have been proposed control designing procedures for
multivariable systems but still, the researchers working on this problem to enhance the control
performance for MIMO process. There are three major control schemes such as decentralized
(multiloop), decoupled control, centralized control has been presented for MIMO process.

1
JOURNAL PRE-PROOF

In general, decentralized (multiloop) control scheme has been widely used in the industrial process
because of its advantages in failure tolerance and easy implementation. However, Model predictive
control (MPC) is capable of handling MIIMO control problem effectively; But MPC is used mostly used on
higher level to provide setpoints to the lower level PID controllers [1]. The MPC is employed in
supervisory level, where the sampling time of MPC is higher than the lower level PID control loops
.Managing the coupling problem using MPC can be troublesome due to the limited bandwidth of MPC
[2]. Hence, the PI/PID based controllers are used to control the process at the lower level. The

F
improvement in the lower level control system increases the overall performance of multivariable

O
control. The multiloop PI/PID controllers generally utilized technique in the multivariable process. The
purpose of utilizing the PI/PID in multi loop because of its simple structure, failure tolerance and

O
capacity of meeting user specification. But the tuning of PI/PID is difficult due to the interaction effect

PR
between input and output, because the design of one loop depends on another loop. There has been
many designing procedure proposed in the literature such as detuning [3], sequential loop closing,
independent loop [4] and relay auto tuning method. In the detuning method, the controller parameter is
E-
found for most important loop transfer function without considering interaction effect and then the
controller gains are detuned by considering interaction effect to meet some user control specification.
PR

But the performance and stability of closed loop system has not been discussed properly in the detuning
procedures. In the sequential tuning methods, controllers are tuned while closing the loop one after
other, but the final controller design completely depends on the order of other controller.
AL

The multiloop PI/PID Controller provides better control performance for the process with modest
interaction. It fails to provide reasonable control performance when the interaction effects between
N

loops are significant. In such a case decoupler based control scheme is preferred for MIMO process
three types of decouplers are available in the literature such as ideal, inverted, and simplified decoupler
R

[5]. The ideal decoupler formed using the inversion of process transfer function which may result in
U

complex dynamics. The simplified decoupler is widely used to develop an ETF model and decentralized
controller is designed for corresponding ETF model.
JO

VU TNL and Lee [6] have designed independent IMC based PI/PID controller using EOTF model, where
the higher order EOTF model is approximated into reduced order model using maclaurin series. The
EOTF resulted in a higher order model, it requires model reduction techniques to form reduced order
model which make the controller design easier. Hajare VD and BM Patre [7] have approximated the
higher order EOTF model to reduced order model using frequency response fitting. The formulation of

2
JOURNAL PRE-PROOF

EOTF for higher order process is complex and decoupler design also makes controller complicated
control structure. Xiong et.al [8] has demonstrated a method of designing decentralized PI/PID
controller using effective transfer function. The ETF model is developed using the information of
effective relative gain array, relative gain array (RGA), relative frequency of open loop transfer function
model. Vijay kumar V et. al [9] has designed ETF model using normalized relative gain array (RNGA) and
relative average time array (RARTA) to form ETF model and then controller is tuned using maclaurin
series for corresponding EFT model. Wang et. al [10] has developed systematic design for full

F
dimensional PID control for higher order process using approximated decoupler but the robustness of

O
closed loop control system cannot be guaranteed.

O
Yuling shen et al [11] expanded the normalized decoupling method for higher order process.
Rajapandiyan C and Chidambaram M [12] have proposed new method of decoupled process

PR
approximation for higher order process, where the simplified decoupler with ETF approximation method
is combined to form new approximation procedure for decoupled process, and then the diagonal
E-
controller is designed using simplified internal model control (SIMC) method for approximated
decoupled process.
PR

The integer order PI/PID controller still dominated in the industries. Recently, the fractional order
controller has received extensive consideration in industries and academia. The three hundred years old
fractional calculus has some interesting history, however last few decades the fractional calculus has
AL

been gain popularity in control system engineering and other engineering application. The researcher
have demonstrated that the FOPID controller outer performs than the integer order PID controller for
many application. However there is lack of tuning methods available compared to PI/PID controller. The
N

analytical and numerical based tuning method for FOPI/FOPID is reported in the literature. The
R

complete review of FOPID tuning methods are discussed in [13].Generally, the FOPID controller tuned to
meet user defined specification (time, frequency domain) by analytical method and optimization
U

method. Mostly, the optimization based tuning methods are used to obtain the FOPI/FOPID controller
JO

parameters. In [14], the particle swarm optimization, differential evolutionary optimization, bat
optimization used to tune the fractional order PI/PID controller for SISO process. The design of multiloop
FOPID controller is difficult for MIMO process because of its coupling effect between input and output.

The main objective of this paper is to make the student to understand about the real time industrial
control problem. Also, this papers helps control system student community to understand about the
fractional order control system in simple manner. The design of coupled system is difficult due to its

3
JOURNAL PRE-PROOF

coupling effect; hence, the coupled interconnected system is separated into equivalent single input
single output system to make design of control system easier. The design of one loop controller depends
on the other loop, so it is always complicated to design a FOPID controller for multivariable system. By
using independent loop method, the MIMO process is decomposed into independent single input single
output (SISO) for tuning of controller easily. In this paper, the FOPID controller is used as a diagonal
controller with simplified decoupler. The diagonal controller parameters are obtained using bat
optimization algorithm for minimum values of time weighted integral absolute error.

F
O
2. Literature Survey

Chen, K., (2018) proposed a synthesis tuning method for PIλ to satisfying the frequency domain

O
specification such as gain crossover frequency and phase margin. In general, the synthesis methods for

PR
tuning controllers are proposed for reduced order model. The optimal PIλ controllers are tuned for
higher order process according to the specified phase margin and crossover frequency [15].
E-
Bingul z and Oguzhan Karahan (2018) compared the PID and FOPID controller performance where the
controller parameters are tuned using artificial bee colony and particle swarm optimization. The
controller tuned with three different cost function such as settling time, rise time, overshoot, and
PR

steady-state error [16].

Pritesh Shah and Sudhir Agashe (2017) demonstrated the effect of order of differentiation (µ) and
AL

integration (λ) on time domain specification. The PID and FOPID controller implemented for the
quadruple tank level system [17].
N

Haji, V.H. and Monje, C.A., 2018 presented multiloop FOPID controller for distillation column process
using improved bat algorithm. The multivariable FOPID controllers are designed with and without
R

decoupler, the unknown controller parameters are obtained using different optimization method with
U

different time domain specifications as a objective function. The distillation column process parameters
such methanol in the distillate and bottom compositions is controlled using two feedback loops, where
JO

the two loops contains ten FOPID controller parameters. The unknown ten parameters are optimally
tuned for minimizing the integral error criteria [18].

Zhang, F (2018) presented FOPID tuning method using state transition algorithm. The effect of sample
time on the objective function of optimization for FOPID is discussed [19]. Kasireddy Idamakanti et al

4
JOURNAL PRE-PROOF

(2018) developed a IMC based integer and non integer controller for automatic generation control of
two area power system using the method of simplified decoupling techniques [20].

Jin, Q et al. 2016 developed a IMC based controller for TITO system, the complex model is decomposed
into independent SISO model and then the higher order SISO model is reduced using Maclaurin series
expansion. Then, the robust IMC based controller is tuned for reduced order model [21].

Lakshmanaprabu, S.K., el al (2017) developed IMC based Fractional Order PID controller for TITO process

F
where the complex model is decomposed into SISO model and Controller parameters tuned

O
independently using novel bat algorithm [22].

O
San-Millan, A et al (2017) demonstrated an effectiveness of PI and PID controller for the real time
system, and the TITO real model is identified as the first order model by closing the other loops. The

PR
effective open loop transfer function is developed from the real time laboratory prototype of a hydraulic
canal. The reduced effective model was utilized for design of PI and PID controller using frequency
domain specifications [23].
E-
Moradi, M., 2014 proposed centralized FOPID controller for TITO process and the controller parameters
PR

are tuned using the genetic algorithm. The feedback control for TITO system is designed with centralized
fractional order PID controllers which contains 20 tuning parameters. The tuning of controller with multi
objective function is requires lot of computation power for finding the optimal centralized FOPID
AL

controller parameters [24].

Lakhdar Chaib etal., (2015) demonstrated Bat Algorithm (BA) based FOPID controller for power system
N

stabilizer. The BA is applied to find the optimal controller values for the four different fitness function
such as IAE, ISE, Integral Absolute Time Error (ITAE) and Integral Square Time Error (ITSE). The
R

comparison of different fitness function for FOPID tuning are recorded and concluded that the
U

overshoot with Integral Absolute Error (IAE), Integral Time weighted Absolute Error (ITAE) is bigger than
Integral Squared Error (ISE), Integral Time weighted Squared Error (ITSE). However, IAE and ITAE have
JO

better response and settling time than ISE, ITSE [25].

Nasirpour, Naeimeh, and Saeed Balochian (2017) developed a FOPID controller tuning method for TITO
air-conditioning system. The air-conditioning system is decomposed into SISO system using decoupling
techniques and the FOPID controller parameters are optimally tuned using PSO algorithm through

5
JOURNAL PRE-PROOF

minimization of multi objective function which includes settling time, rise time, overshoot and integral
time squared error [26].

Katal, N. and Narayan, S., 2017 proposed a optimal fractional order PID controller for two tank liquild
level system. The controller is tuned using multi-objective variant of particle swarm optimization. The
multi objective minimization problem is solved using Pareto optimal set which is a simplified illustration
of data. It is very difficult to use Pareto optimal set for m dimensional optimization problems. The level

F
diagrams are utilized for helping the selection of better solution from Pareto optimal set [27].

O
3. Basic fractional calculus

O
Fractional calculus is an emerging technique in engineering and sciences, it has some unique features
and it represents system completely. Non integer order differentiator and Integrator are represented by

PR
q
differ-integral operators a D t . The fractional derivative and fractional integral combined and expressed

in generalized form,
E-
 dq
 q q0
PR

 dt

a D t  1 q0
q
(1)
t
  d  q

a
 q0
AL

Where q is a fractional order and ‘a’ is an initial conditions. Many definitions are proposed for fractional
differ- integral. The Riemann and Liouville(R-L) definition for fractional derivative is the most popular,
N

this is defined as follows,


R
U

t
d q f (t ) 1 dn
n 
a Dt f (t ) 
q
= (t   n-q-1 f ( d (2)
JO

d (t  a) (n  q) dt 0
q

Where n is the integer which is not less than q i.e. n-1 ≤ q < n and Γ is the Gamma function.

inf

t
z-1  t
(z) = e dt (3)
0

6
JOURNAL PRE-PROOF

The R-L definition for fractional integral is given by

t
1 dn
n 
a Dt
q
f (t )  (t   q-1 f ( d (4)
(q ) dt 0

The fractional order transfer function can be obtained easily for the system with non integer order
differential equation.

F
a Dqt y (t )    an 1 Dt n1 y (t )  an Dt n y (t )  b1 Dt1 u (t )    bm Dtm u (t ) (5)

O
The fractional order transfer function of a single variable system with zero initial condition can be

O
defined by

PR
L  0 Dt f (t )  s F ( s)

(6)

Therefore,
E-
b1s1  b2 s2    bm sm
G (s)  (7)
PR

a1s1  a2 s 2    am s n

ηi , γi are order of numerator and ai , bi are real numbers.


AL

In this paper, the outloup approximation for the fractional order is used for simulation. The fractional

derivative or integral s is approximated into integer order transfer function using a recursive
N

distribution of poles and zeros .


R

N 1  ( s z ,n )
s  k  (8)
1  ( s  p ,n )
U

n 1
JO

The approximated transfer function assigned in the pre specified frequency range. K is a regulated gain,
it is adjusted in both sides of equation for a unit gain at 1 rad/s. The frequencies of poles and zeros are
given by,

 z ,1  l  ;  p , n  z , n ε , n=1 N ; z , n 1   p , n  , n=1 N-1 (9)

7
JOURNAL PRE-PROOF

  h l  ;   h l 
 /N (1 )/ N

4. Equivalent Transfer function

The TITO model of this process is defined by the G(s),

 g (s) g12 ( s ) 
G ( s )   11
g 22 ( s ) 
(10)
 g 21 ( s )

F
The multiloop controller transfer function matrix is,

O
 g (s) 0 
Gc ( s)   c1
gc 2 ( s) 
(11)
 0

O
The effective open loop transfer function is developed by incorporating decouplers in the MIMO

PR
process. The open loop transfer function between input and output is developed while other loop is
closed. It is not easy to develop an EOTF model by closing other loops, because closing other loops
requires controller, so the EOTF model is completely depends on the other loop controller.
E-
The EOTF model between u1 and y1 is developed while loop 2 is closed with the feedback controller gc2.
It is assumed that the gc2 gives perfect control, where the output of y2 attains setpoint with no transient.
PR
g12 ( s ) g 21 ( s )
g11eotf ( s )  g11 ( s ) 
g 22 ( s ) (12)
g 21 ( s ) g12 ( s )
g 22eotf ( s )  g 22 ( s ) 
g11 ( s ) (13)
AL

The EOTF model resulted in higher order dimension, which is difficult to use directly in the controller
design, hence the model has to be reduced using any approximation method. To avoid this
N

computational complexity the equivalent transfer function (ETF) model obtained using relative gain
array (RGA), relative normalized gain array (RNGA), relative average resitance time array (RARTA)
R

methods.
U

The steady state information and transient response information of MIMO process are incorporated to
JO

form ETF model, which is used to develop a controller separately. The normalized gain (KN,ij), RGA( Λ ),
RNGA(  ), Relative average resistance time( γ ) and RARTA (  ) are found to develop a ETF model.

Consider the FOPDT transfer function matrix (G(s)) for TITO process is given below,

8
JOURNAL PRE-PROOF

 k11 11 s k12 


  1 e e 12 s 
 12  1
G(s)=  11 
(14)
 k21 k22  22 s 
 e 21s e 
  21  1  22  1 

Where Kij is the process gain between ith output and jth input; τij is the time constant of FOPDT model
between ith output and jth input; θij is the dead time of FOPDT model between ith output and jth input (i, j

F
= 1,2).

O
The process gain at steady state,

O
k k12 
K=G(s=0)=  11
 k 21 k 22  (15)

PR
Relative gain array RGA,

1  k11k 22 -k12 k 21 
= 
E-
K  -k12 k 21 k 22  (16)
PR

The normalized gain array for TITO process G(s) is,

 K N,11 K N,12 
KN = 
 K N,21 K N,22  (17)
AL

k ij k ij
where K N,ij = = , the  ij is the average residence time of ith output to jth input.
σij τij +θij
N

RNGA,
R

11 12 
  K N  K NT , where  = 
21 22  (18)
U

RARTA,
JO

  12 
 =  11
 21  22 

(19)

ˆ ij ij
where  ij   is a relative average residence time,  =   , where  is RNGA
 ij  ij
and  is the RGA.

9
JOURNAL PRE-PROOF

The ETF model is developed using RGA, RNGA and RARTA as,

k ij 1 -γ θ s
ĝ ij (s)= e ij ij (20)
Λij γijτijs+1

The ETF model for TITO process is,

 gˆ ( s ) gˆ12 ( s ) 
Ĝ(s)=  11  (21)
 gˆ 21 ( s ) gˆ 22 ( s ) 

F
O
In the presence of decoupler, the TITO system behaves like two independent loops. There are many
decoupling methods are available such as ideal decoupler, inverted decoupler, normalized decoupler.
The wang et.al (2000) decoupler is a recommended realizable decoupler where the extra time delay is

O
incorporated.

PR
 g12 ( s )  v (12 11 ) s 
 e  v (22 21 ) s  e 
g11 ( s )
D(s)=  
 g 21 ( s )  v (21 22 ) s  (22)
 e
E- e  v (11 12 ) s 
g
 22 ( s ) 
PR

where,

1 if   0
v ( )  
0 if   0
AL

5. Fractional order PI/PID control

The FOPID is a special case of PID controller where it is less sensitive to parameter variation of controller
N

The fractional order PID controller provides more flexibility in tuning of controller with additional two
tuning parameters.
R

The transfer function of FOPID controller Gfopid(s) is given below,


U
JO

 1 
G fopid (s)=K p  1   t s  (23)
 ti s 
d 
 

where Kp is the proportional gain, ti is the integral time constant, λ is the order of integrator, td is the
derivative time and µ order of derivative. The performance FOPID has been demonstrated by many
researchers that the FOPID controller outer performs than conventional PID controller due to the

10
JOURNAL PRE-PROOF

additional tuning parameter such as integrator order () and differentiator order() [36] The addition two
parameters provide better flexibility in adjustment of gain and phase characteristics in closed loop
control system.

The FOPID controller tuned to meet the time domain performance criteria such as integral square error
(ISE), Integral absolute error (IAE), integral time absolute error (ITAE). The ISE based tuning provides the
lower overshoot in the closed control response than IAE based controller setting. But the ISE based

F
controller may tends to settle in large settling time. The ITAE based controller provides overshoot

O
response with faster settling time. ITSE is less sensitive and it is not comfortable computationally [37].
In this paper, fractional order PI/PID controller tuned using bat optimization algorithm. The controllers

O
are independently designed using each ETF model.

PR
6. Bat Optimization Algorithm

In evolutionary computation techniques, Bat optimization has gained popularity in all engineering
E-
nonlinear multimodal problems [28]. The natural forging strategies of bat is inspired and mimicked into
bat optimization algorithm .The bat emits sounds pulses and receive back to identify the prey. The pulse
PR

emission rate varies depending on the location of species. Initially bats flies randomly with velocity Vi at
position xi with fixed frequency fmin to search prey. It updates its frequency and pulse emission rate
depending on the position of prey and then the bat position is updated to move near to prey. The
loudness varies from maximum to minimum when the bat moves towards the prey. The bats starts with
AL

position xit and velocity vit at frequency fi to find the prey. The bat founds new solution (prey) by
adjusting frequencies and loudness, then the quality of the solution is ranked by loudness and pulse
N

emission rate which is developed based on the global solution (prey). The bat algorithm flow chart is
shown in the figure 1.
R
U
JO

11
JOURNAL PRE-PROOF

F
Start

O
Initialize the bat population, pulse frequency rates (ri) and loudness A0

O
Calculate fitness value for each bat location (JRMSE)

PR
Generate new locations by adjusting fi and updating vi

Consider First bat


E-
rand> ri
PR

Generate local solution around the best location

Replace new temporary location with the solution of local search


AL

Evaluate fitness of the new temporary location of this bat

Consider next
N

Next If the new


bat
iteration A0> rand location
R

f(xi) > f(x*)

Accept temporary location as


U

new Location increase ri and Keep old location as new location


decrease A0
JO

All the bats are


considered?

Rank the location, select best location and save the ri and A0

Is termination
criteria satisfied?

12

End
JOURNAL PRE-PROOF

Figure 1. Flow Chart of Bat Optimization Algorithm

The position xit and velocity vit at frequency fi are given by,

F
O
The frequency of ith bat,

f i =f min +(f max -f min )

O
Here  is a random vector drawn from uniform distribution. (24)

PR
The velocity of ith bat ,

t
v it =v it-1 +(x it -x best ) fi
E- (25)

Where xbestt is the global best location.


PR

The bat new position,

x it =x it-1 +v it (26)
AL

The best position is found by random walk,

x new = x old + wA t (27)


N

Where ‘w’ is a random number from -1 to 1. AT is an average loudness of all bats. The loudness variation directly
R

proportional to the closeness of prey position, when the bat moves towards to the prey then it reduces its
U

loudness and increases the pulse emission rate. That is given as,
JO

A it+1 = μ a A it , rit+1 = ri0 1- exp (-γt)  (28)

Where μa is a constant in the range of [0, 1] and γ is a positive constant. When time reaches infinite the rit equal to
rio.

For any case, 0< μa < 1 and γ > 0

A it  0, rit  ri0 , as t  (29)

13
JOURNAL PRE-PROOF

For the easy implementation of BOA, the standard recommended μa, γ values are chosen as 0.9.

6.1 Tuning of -FOPID Controller using BOA

The minimization of integrated error performance indices are the general specific control requirement. The
Integral time absolute error which yields faster settling time with less overshoots. The optimization for controller
design problem is formulated and the constraints bounds are given below,

J   ysp (t)  y1 (t) dt (30)

F
0

O
Where ysp is the set points of controller and y1 are the output of ETF model, t is the simulation time (in sec). The ‘t’
is the time weighting factor which give less weight to initial error and large weightage to steady state error. The

O
minimum value of ‘J’ produce controller response with less overshoot and faster settling time.

PR
The tuning of FOPID controller parameters using optimization techniques are depends on the selection of objective
function for controller tuning. The different kind of objective function gives different values of FOPID controller
parameters. The ITAE based indices provide better controller with faster servo tracking and disturbance rejection
E-
without offset. The main reason for selection of ITAE for the FOPID controller design is that the order of integrator
produce offset for a long time. Hence, the ITAE is utilized for providing more penalties to the offset error. The ITAE
PR
is a better performance indicator of the closed loop control scheme where overshoot, setting time, rise time,
offset are considered. Therefore, the ITAE is used as the objective function of multiloop FOPID controller design.
The controller with small values of ITAE considered as best controller for system and also it indicates that the
controller response is fast.
AL

7. Simulation Study
N

7.1. Case 1: VL Column System


R

The bench mark example of VL column given by luyben (1986) is a TITO process
U

with higher interaction between input and output. The Transfer function model
of VL column is given below,
JO

  2.2e  s 1.3e 0.3 s 


 
G p ( s )   7 s 11.8 s 7s  1 
 0.35 s (31)
  2.8 e 4.3 e 
 9.5 s  1 9.2s  1 

14
JOURNAL PRE-PROOF

The normalized gain matrix, RGA, RNGA, RARTA are found using equation
16,17,18,19 and then the ETF model is obtained using equation 18.

 1.353 0.959 s 2.078 0.2656 s 


 6.91s  1 e 6.91s  1
e 
Ĝ(s)=   (32)
 4.4769 e 1.593 s 2.6455 0.334 s 
e
 8.41s  1 8.794 s  1 

F
The EOTF/ETF models are developed by assuming the perfect control assumption, but this

O
assumption can be validated only by incorporating decoupler in the open loop model.

O
The simplified decoupler D(s) is developed using equation 22.

PR
 1 0.5909 

D(s)= (5.9907 s  0.6512) 1.45 s 
 0.7 s  (33)
e e
 9.5s  1 
E-
Additional time delay included in the decoupler D(s),
PR

 g (s) g12 ( s )   1 d12 ( s )e0.7 s   g11


*
(s) 0 
G ( s ).D ( s )   11     (34)
 g 21 ( s ) g 22 ( s )   d 21 ( s ) e 0.7 s
  0 *
g 22 ( s ) 

AL

The decoupled processes are,

* 1.3535 0.9559 s
g11 (s)  e (35)
6.691s  1
N
R

* 2.6455 1.034 s
g 22 (s)  e (36)
8.794s  1
U

Rajapandiyan C et.al (2012) obtained the multiloop PI controller parameter using


JO

simplified internal model control (SIMC) with EOTF model.

The SIMC based PI controller gc(s) for FOPDT process gp(s) is given by,

k  s  1 
g p (s)  e ; g c  kc  1   (37)
 s 1  ti s  1 

15
JOURNAL PRE-PROOF

  
kc    ; ti  min( ,8 ) (38)
 2 K 

The reported (Rajapandiyan et.al) PI controller setting is,

 2.5858 1 1  0 
Gc ( s )    (39)
 6.691s 
 1 
 0 1.6066  1
8.276 s  

 

F
The optimal FOPI/FOPID controller settings are obtained via minimization of ITAE for ETF model

O
( g11*(s), g22*(s)) using BOA. The optimal controller values are tabulated in table.1

O
Table 1 The optimal FOPID values of ETF model based on IAE

PR
Kc ti λ td µ ITAE
Loop1 -2.588 6.691 1 - - 2.661
PI E-
Loop2 1.6066 8.276 1 - - 3.473
Loop1 -2.772 6.7 1 - - 2.618
FOPI
PR

Loop2 1.955 8.96 1.002 - - 3.326


Loop1 -3.91 7.1689 1.005 0.32 0.96 1.521
FOPID
Loop2 2.7 9.20 0.98 0.59 0.979 3.041
AL

The closed loop response for V-L column is shown in figure 2. The unit step
N

changes are applied at both references. At t=50, the setpoint of loop2 is changed
R

from 0.25 to 1. The proposed controller is compared with FOPI controller and
U

SIMC based controller proposed by Rajapandiyan and Chidambaram. From the


JO

figure 2, it is observed that the proposed FOPID controller produce better


performance than the FOPI, PI controller. The interaction effect reduced
effectively with faster regulatory response by the proposed FOPID controller.

16
JOURNAL PRE-PROOF

1.4

setpoint(yr1)
1.2 PI(SIMC)
FOPI
FOPID
1

0.8
y1

0.6

F
0.4

O
0.2

O
0 10 20 30 40 50 60 70 80 90 100
time(sec)

Fig. 2 Servo and regulatory response for the VL column (loop1)

PR
1.4
E-
setpoint(yr2)
PR
1.2 PI(SIMC)
FOPI
FOPID
1

0.8
AL
y2

0.6

0.4
N

0.2
R

0
0 10 20 30 40 50 60 70 80 90 100
U

time(sec)
JO

Fig.3 Servo and regulatory response for the VL column (loop 2)

7.2. Case 2: The TICTL process Experimental setup

To demonstrate the practical applicability, the two interacting conical tank level process is
considered and model parameters are identified from the experimental data. This obtained

17
JOURNAL PRE-PROOF

model is further used for simulation study and resulted controller is implemented on the
laboratory two interacting conical tank level process (TICTLP) setup. The TICTLP is shown in
figure. 3 which consist of two tanks in series with variable speed pump. The laboratory setup of
TICTLP is interfaced to personnel computer using ADuC841 micro controller based data
acquisition (DAQ) card. The DAQ has 8 channel analog inputs and 2 channel analog outputs with
12 bit resolution. The input and output voltage range of DAQ is 0V to 5V. The water levels in the

F
tanks 1, 2 are measured using ABB 2600 series differential pressure transmitters. The output of

O
transmitter is 4 - 20 mA current signal, which is converted into 0 to 5V range for interfacing
with ADuC841 DAQ card. The variable speed pumps P1, P2 with thyristor driver circuit are fitted

O
to provide liquid inflow to tanks 1 and 2 respectively for the input ranges from 0 to 5V. The

PR
input voltage is given to tyrister driver circuit (TE10 A) to produce varying voltage (0 - 230V) for
running a pump at different speed. PWM signal is generated and given to tyristor driver circuits
(TE10 A ) to produce the varying voltage (0 - 230V) for running a pump at different speed. The
E-
change in applied voltages is directly proportional to the speed and inflow rate of liquid.
PR
AL
N
R
U

Fig.4 Real time experimental setup of TICTP


JO

Table 2 Technical Specification

Part name Description


DPT ABB 2600T SERIES
pressure transmitter
Pump Maximum height of Tank 1 and 2

18
JOURNAL PRE-PROOF

Rotameter Input Voltage to pump 1 and 2


TE10A Thyristor power control

7.2.1. Black box modeling

Obtaining the transfer function model of the plant is not a simple task for MIMO process. The

F
accurate modeling can improve the controller performance and it can help the controller to
reduce the effect of uncertainties. The aim is to make the water level in conical tank remained

O
stable, and then first order model is to be captured by applying step changes in each input. The

O
system is allowed to take new steady state at fixed input flow rate.

PR
The TICTLP is highly nonlinear and coupling process. The nonlinearity is due to the variation in
the area of conical tank, hence, the linearized transfer function matrix developed around single
operating points. The operating points are selected based on the domain knowledge and input
E-
output characteristics. In this study, the nominal operating points are fixed as [V1=50% (2.5V),
V2=50% (2.5V), h1=49% (24.5cm), h2=51.2% (25cm)]. The nominal input voltages v1= 50%,
PR

v2=50% are applied to pumps and then tank level allowed to take steady state output. The both
tanks took 1000 secs to reach steady state, and then the step change is introduced at input
voltage v1 from 50% (2.5 V) to 60% (3 V) while the input 2 (V2) voltage remained at constant
AL

50% (2.5V). The output level h1, h2 are recorded with the sampling time of 1 sec, from this
recorded data the first order plus dead time (FOPDT) transfer function (g11(s)) between input V1
N

and h1 and transfer function (g22(s) between input v1 and h2 is identified using process reaction
R

curve method. Similarly, the same steps are followed to obtain the transfer function g21 (s)
between V2 and h1 and the g22 (s) between V2 and h2.
U

The identified transfer function model of TICTLP at operating condition [V1=50% (2.5V), V2=50%
JO

(2.5V), h1=49% (24.5cm), h2=51.2% (25cm)] is,

 1.8361 11.5 s 0.723 


 340.7   1 e e 19.2 s 
415.4   1
G P (s)=  
 0.74 19.1s 1.89 12.4 s 
 407.3   1 e 365.6  1
e 
 

19
JOURNAL PRE-PROOF

The RGA, RNGA, average residence time (Tar) and RARTA are found to develop ETF model.

The gain array (K), relative gain array (  ) of TICTLP is,

1.8361 0.7234   1.182 -0.182 


K=   =
 0.74 1.89  ;  -0.182
 1.182 

The normalized gain matrix found using equation (17),

F
O
 0.0052 0.0017 
KN = 
 0.0017 0.0050 

O
PR
The RNGA, RARTA found using equation (18,19),

 1.1246 -0.1246   0.9511 0.6833


=   ; =  
-0.1246 1.1246   0.6833 0.9511
E-
The ETF model transfer function parameter is found using equation (20)
PR

 1.553 10.9 s 3.966 13.1s 


 324.1 s  1 e 283.9 s  1
e 
Ĝ P (s)=  
 4.057 12.6 s 
AL

1.598
 278.3 s  1 e e 11.8 s 
 347.7 s  1 
N

The wang et.al (2000) decoupler is used in the system to incorporate the extra time delay in the
system.
R
U

 134.167 s  0.3938 7.7 s 


 1 e 
415.4 s  1
D(s)=  
JO

 143.145 s  0.3915 6.7 s 


 e 1 
 407.3 s  1 

The independent decomposed process is considered for controller design. The controller is
designed independently using,

20
JOURNAL PRE-PROOF

1.553 1.598
*
g11 (s)  e 10.9 s ; g 22
*
(s)  e 11.8 s
324.1 s  1 347.7 s  1

The controller setting obtained using the BOA as discussed in the section 4.1. The controller
settings are tabulated in table.3. The BOA optimization is adopted with population size 20,
iteration size 100, etc. The initial guess of lower and upper bounds of loop1 controller Kp1 is
fixed between the range of [0-20], Ki values between the range of [0-10], Kd values between

F
the range of [0-10]. Similarly, loop2 controller parameters kp2,ki2,kd2 are fixed using IMC

O
based tuning. The BOA is implemented in MATLAB 2014a, the simulation were carried out using
the personnel computer with 3.2 GHz with Intel V process with 4 GB of random access memory.

O
The optimal values of FOPID controller parameters obtained by BOA with IAE, ITAE are

PR
tabulated in table 3.

E-
Table 3. The optimal FOPID values of ETF model based on ITAE
Kc ti λ Kd µ IAE ITAE ISE
PR

Loop1 0.623 87.2 - - - 11.41 846.6 6.18


PI
Loop2 0.605 94.4 - - - 12.53 941.4 6.69
AL

Loop1 6.82 116.7 0.991 - - 10.92 821.1 5.92


FOPI
Loop2 7.05 121.5 0.995 - - 12.04 902.82 6.20
N

FOPID Loop1 10.1 105.3 0.998 10.32 0.752 10.14 798.3 5.47

Loop2 9.74 112.8 1.00 9.81 0.673 11.89 891.05 5.84


R
U
JO

21
JOURNAL PRE-PROOF

50

Setpoint (h1 )
45
PI
40 FOPI
FOPID
35
% Level in Tank 1 (h1)

30

25

20

15

10

F
0
0 500 1000 1500 2000 2500 3000 3500 4000
time(sec)

O
Fig 5 Output response h1 for TICTLP

O
100

90

PR
80

PI
70
FOPI
% Controller output (V1)

FOPID
60

50

40 E-
30

20

10

0
PR
0 500 1000 1500 2000 2500 3000 3500 4000
time(sec)

Fig.6 Controller response for TICTLP

60
AL

50

Setpoint (h2)
40
PI
% Level in Tank 2(h2)

FOPI
FOPID
30

20
N

10
R

0
0 500 1000 1500 2000 2500 3000 3500 4000
time(sec)

Fig. 7 Output response for TICTLP


U
JO

100

90
PI
80
FOPI
FOPID
70
% Controller output (V2)

60

50

40

30

20

10

0
0 500 1000 1500 2000 2500 3000 3500 4000
time(sec)

22
JOURNAL PRE-PROOF

Fig. 8 controller response for TICTLP

The proposed controller is implemented for TICTLP and experimental results of closed loop
response are shown in figure 5,6,7,8. The step variation is introduced in closed loop system,
where the setpoint of liquid level in tank 1 (h1) is applied at 500 sec and liquid level in tank 2
(h2) is applied at 1000 sec. At t=1000 sec, the setpoint of h1 is changed from 20 cm to 30 cm

F
while the liquid level in tank 2 is unchanged. The setpoint variation in tank 1 level will affects

O
the liquid level in tank 2 due to the interaction flow. But, the proposed controller compensates
the interaction effect faster than the FOPI, PI controller. The servo tracking of proposed

O
controller is faster than the FOPI, PI Controller with minimum settling time and overshoot. It

PR
can be clearly observed that the proposed controller track and regulate the desired liquid level
faster than the PI, FOPID controller.

8. Conclusion
E-
Generally, it is quite difficult to design a fractional order PI/PID controller for multivariable
PR

process. In this paper, the independent design of FOPI/FOPID controller is designed with
decoupler for TITO process. The ETF model is developed for TITO system using the information
of RGA, RNGA and then the FOPID controller parameters are independently designed for ETF
AL

model. The FOPID controller parameters are tuned using bat optimization algorithm to achieve
minimum value of ITAE. The proposed controller validated for simulation example. In addition,
N

the controller is implemented for the real time laboratory TIFCTL process. It is concluded that
proposed FOPID controller has enhance performance than the centralized PI and centralized
R

FOPI controller. The FOPID controller provides optimal performance when the order of
U

integrator is close to 1. This paper provides insightful understanding of multivariable controller


JO

design for complex industrial control problem. For a dynamical and nonlinear system, a new
dynamical decoupling method can be designed with improved robustness and adaptability.

23
JOURNAL PRE-PROOF

Reference

[1]. Q. Xiong, W.J. Cai, M.J. He, “ Equivalent transfer function method for PI/PID controller design of MIMO
process”, Journal of process control, 17 (2007), pp.665-673.
[2]. Juan Garrido, Francisco Vazquez, Fernando Morilla, “Centralized multivariable control by simplified
decoupling”, journal of process control.
[3]. Chandra shekar besta, Manickam Chidambaram, “Tuning of multivariable PI controllers by BLT method for

F
TITO systems”, Chemical Engineering Communication, Vol. 203, issue 4, pp 527-538, 2016.
[4]. Vu TNL, Lee M, Independent design of multiloop PI/PID controller for interacting multivariable process”,

O
journal of process control, 20(8), 2010, pp.922-933.

O
[5]. Cai, W. J.; Ni, W.; He, M. J.; Ni, C. Y. Normalized Decoupling, “ A New Approach for MIMO Process Control
System Design. Ind. Eng.Chem. Res. 2008, 47 (19), 7347−7356.

PR
[6]. TNL Vu , Lee M, “ Multiloop PI controller design based on direct synthesis for interacting multi-time delay
process”, ISA Transaction, Vol.49 pp.79-86, 2010.
[7]. Hajare VD, Patre BM, “Decentralized PID controller for TITO process using characteristics ratio assignment
E-
with an experimental application”, ISA Transaction(2015), http://dx.doi.org/10.1016/j.isatra.2015.10.008
[8]. Xiong Q, Cai WJ, “Effective transfer function method for decentralized control system of multi-input multi-
PR
output process, Journal of process control, Vol.16, pp.773-784, 2006.
[9]. Vijaykumar V, Rao VSR, Chidambaram M, “Centralized PI controller for interacting multivariable processes by
synthesis method”, ISA Transaction, Vol.51, pp.400-409, 2012.

[10]. Wang, Qing-Guo, Bin Huang, and XinGuo. "Auto-tuning of TITO decoupling controllers from step
AL

tests." ISA transactions 39, no. 4, pp. 407-418, 2000.

[11]. Yuling Shen, Youxian Sun, Wei Xu, “Centralized PI/PID controller design for multivariable processes”,
N

Industrial and Engineering Chemistry research, 53, pp.10439-10447, 2014.


[12]. Rajapandiyan C, Chidambaram M, “Controller design for MIMO process based on simple decoupled
R

equivalent transfer function and simplified decoupler”, Industrial & Engineering Chemistry Research, Vol.51,
pp.12398-12410, 2012.
U

[13]. Pritesh Shah, Sudhir Agashe, “Review of fractional order PID controller”, Mechatronics, Vol. 38, pp. 29-41,
JO

2016.
[14]. Agababa MP, “ Optimal design of fractional order PID controller for five bar linkage robot using a new
particle swarm optimization algorithm, Soft computing, 1-13,2015.

[15]. Chen, K., Tang, R., Li, C. and Lu, J., 2018. Fractional order PIλ controller synthesis for steam turbine speed
governing systems. ISA transactions, 77, pp.49-57.

24
JOURNAL PRE-PROOF

[16]. Bingul, Z. and Karahan, O. Comparison of PID and FOPID controllers tuned by PSO and ABC algorithms for
unstable and integrating systems with time delay. Optimal Control Applications and Methods, 39(4), pp.1431-
1450. 2018
[17]. Pritesh Shah and Sudhir Agashe 2017. Experimental analysis of fractional PID controller parameters on
time domain specifications. Progress in Fractional Differentiation and Applications, Vo.3 (2),pp.141-154, 2017
[18]. Haji, V.H. and Monje, C.A., 2018. Fractional-order PID control of a MIMO distillation column process using
improved bat algorithm. Soft Computing, pp.1-20.https://doi.org/10.1007/s00500-018-3488-z

F
[19]. Zhang, F., Yang, C., Zhou, X. and Gui, W., 2018. Fractional-order PID controller tuning using continuous
state transition algorithm. Neural Computing and Applications, 29(10), pp.795-804

O
[20]. Kasireddy Idamakanti, Abdul Wahid Nasir, and Arun Kumar Singh. "IMC based controller design for

O
automatic generation control of multi area power system via simplified decoupling." International Journal of
Control, Automation and Systems 16, no. 3 (2018): 994-1010.

PR
[21]. Jin, Q., Wang, Q. and Liu, L., 2016. Design of decentralized proportional–integral–derivative controller
based on decoupler matrix for two-input/two-output process with active disturbance rejection
structure. Advances in Mechanical Engineering, 8(6), doi 10.1177/1687814016652563
E-
[22]. Lakshmanaprabu, S.K., Banu, U.S. and Hemavathy, P.R., 2017. Fractional order IMC based PID controller
design using Novel Bat optimization algorithm for TITO Process. Energy Procedia, 117, pp.1125-1133.
[23]. San-Millan, A., Feliu-Talegón, D., Feliu-Batlle, V. and Rivas-Perez, R., 2017. On the Modelling and Control
PR

of a Laboratory Prototype of a Hydraulic Canal Based on a TITO Fractional-Order Model. Entropy, 19(8), p.401.
[24]. Moradi, M., 2014. A genetic-multivariable fractional order PID control to multi-input multi-output
processes. Journal of Process Control, 24(4), pp.336-343.
AL

[25]. Chaib, Lakhdar, Abdelghani Choucha, and Salem Arif. "Optimal design and tuning of novel fractional order
PID power system stabilizer using a new metaheuristic Bat algorithm." Ain Shams Engineering Journal 8, no. 2
(2017): 113-125.
N

[26]. Nasirpour, Naeimeh, and Saeed Balochian. "Optimal design of fractional-order PID controllers for multi-
input multi-output (variable air volume) air-conditioning system using particle swarm optimization." Intelligent
R

Buildings International 9, no. 2 (2017): 107-119.


U

[27]. Katal, N. and Narayan, S., 2017. Design of robust fractional order PID controllers for coupled tank systems
using multi-objective particle swarm optimisation. International Journal of Systems, Control and
JO

Communications, 8(3), pp.250-267


[28]. Xin-She Yang, Amir Hossein Gandomi, (2012) "Bat algorithm: a novel approach for global engineering
optimization", Engineering Computations, Vol. 29 Issue: 5, pp.464-483.

25

You might also like