Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Computers and Electrical Engineering 45 (2015) 10–21

Contents lists available at ScienceDirect

Computers and Electrical Engineering


journal homepage: www.elsevier.com/locate/compeleceng

Fractional-Order Sliding Mode Controller Design for a Modified


Quadruple Tank Process via Multi-Level Switching q
S. Sutha a,⇑, P. Lakshmi b, S. Sankaranarayanan c
a
Department of Electronics and Instrumentation Engineering, Jayaram College of Engineering and Technology, Tiruchirappalli 621014, Tamil Nadu, India
b
Department of Electrical and Electronics Engineering, College of Engineering, Anna University, Guindy, Chennai 600025, Tamil Nadu, India
c
Department of Instrumentation and Control Engineering, National Institute of Technology, Tiruchirappalli, Tamil Nadu, India

a r t i c l e i n f o a b s t r a c t

Article history: The paper deals with the level control of a Modified Quadruple Tank Process (MQTP). In
Received 13 August 2014 that, the interaction is introduced between the bottom two tanks which causes an addi-
Received in revised form 17 April 2015 tional non-linear dynamical component of the conventional Quadruple Tank Process
Accepted 20 April 2015
(QTP). Since the process has an inbuilt and imposed uncertainties, a robust Sliding Mode
Controller (SMC) is initially designed to drive the system to the desired operating point
via the sliding surface. Further, the conventional sliding surface is altered with fractional
Keywords:
order dynamics. It makes a Fractional-Order SMC (FrSMC) and is proposed for both finite
Modified Quadruple Tank Process
Fractional-Order Sliding Mode Controller
time convergence and counteract to the uncertainties present in the system. The undesir-
Multi-Level Switching able chattering effect is reduced by introducing a novel exponential Multi-Level Switching
Performance indices (MLS) technique in order to protect control valve. Simulation results show the efficacy of
Robustness the proposed Multi-Level Switching FrSMC (MFrSMC) in terms of fast convergence with
high robustness when compared to conventional SMC.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Many industrial processes that have currently been implemented are multivariable processes, that is, multiple outcomes
of a single process. Thus, multivariable control is always a complex platform to implement. The controlling methodology of
such a complex process becomes a tedious task. As a result, a complex, nonlinear and multivariable MQTP, which is an
advanced version of its prequel benchmark process known as a Quadruple Tank Process (QTP), with parametrical outages
or variations, is used for analysis and control. Under the above-mentioned conditions in a complex process, a robust con-
troller can be implemented; for example, a robust and nonlinear Sliding Mode Controller (SMC) can be used. Recent studies
that have focused on overcoming the defects of the SMC in terms of improved robustness, a faster convergence rate and a
reduced chattering effect in the control output have been carried out based on researchers’ interests. Regarding the finite
time convergence or fast response of a controller, forcing the process to the desired track has resulted in the advancement
of the composite controller of Fractional-Order SMC (FrSMC), which combines the properties of fractional dynamics and
SMC.
The performance of each controller is compared with that of a standard and classic Proportional and Integral (PI) con-
troller because of its ease of implementation in an industrial environment and the flexibility provided by the PI controller

q
Reviews processed and recommended for publication to the Editor-in-Chief by Guest Editor Dr. H.A. Shabeer.
⇑ Corresponding author.
E-mail addresses: suthaa_s@yahoo.co.in (S. Sutha), p_lakshmi@annauniv.edu (P. Lakshmi), kokilamsankar@gmail.com (S. Sankaranarayanan).

http://dx.doi.org/10.1016/j.compeleceng.2015.04.012
0045-7906/Ó 2015 Elsevier Ltd. All rights reserved.
S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21 11

in coordinating the process. Controller tuning strategies for the PI controller have been proposed and explained in detail in
various studies [1]. Process modelling and the analysis of the controller’s dynamical conditions, along with theoretical and
practical explanations, were presented in [2]. A complex and challenging multivariable QTP was proposed and enumerated
with a PI controller for various working configurations, such as Minimum Phase (MP) and Non Minimum Phase (NMP) con-
ditions [3]. To understand the basics of SMC in a simple manner, the author discussed the application and design of SMC for
various mechanical or electromechanical systems by providing numerical examples [4]. Initially, SMC was designed for a
basic entity known as a coupled tank system with a Single Input and Single Output (SISO) [5]. After the design and
implementation of the conventional controllers for the benchmark QTP, which is operated in NMP mode, an advanced robust
control strategy named SMC was applied [6]. Recent efforts to reduce the complexity and improve the standards of research
on control have achieved slight improvements. Through such efforts, the QTP was modified during the interactions between
the bottom two tanks; as a result, the size and complexity of the system increased due to the presence of an additional zero,
thereby causing the system to become more oscillatory. This proposed concept is named MQTP, for which various configu-
rations and corresponding transfer functions have been derived [7]. An extension of the modelling of MQTP, involving an
inverted decoupling along with the implementation of a PI controller, has been designed and simulated in NMP mode [8].
Furthermore, the fundamentals of the corresponding fractional calculus have been identified and studied with respect to
a large number of applications [9]. Fractional-order dynamics combined with SMC in two different networks of sliding sur-
faces, such as PDa and PIl Da , have been compared with conventional linear sliding surface networks, such as PD and PID, for a
nonlinear coupled tank process [10]. To improve the convergence rate with a robust property, a Fractional-Order
Proportional Derivative (FOPD) sliding surface SMC has been designed and applied to an Antilock Braking System (ABS)
under multiple benchmark road conditions [11]. To enable a faster rate of convergence towards the desired trajectory, a
novel fractional terminal sliding mode approach was obtained from the Lyapunov stability criterion; in addition, to compen-
sate for the singularity conditions, a nonsingular fractional fast terminal SMC was obtained and applied to a numerical exam-
ple [12,13]. The design of adaptive and robust fractional fuzzy SMC was applied to a multivariable nonlinear system (in
which an adaptive law updated the controller with respect to changes in the parameters of the system in the
fractional-order condition), which was compared with the conventional form of the adapting law to highlight the effective-
ness of the proposed methodology [14]. A fractional SISO system for which a stable SMC was designed using two different
stability criteria and two fractional chaotic systems were synchronized; this approach is known as pseudo-state SMC
synchronization and was highlighted using an example [15]. For a fractional chaotic system, FrSMC is used for the control
system. However, to highlight the importance of the proposed method, a nominal system that is used to control both termi-
nologies was elucidated using numerical examples [16]. SMC applied with fractional-order dynamics led to a PDa sliding sur-
face that was then applied to a mechanical system. To reduce the undesirable chattering phenomena, an intelligent fuzzy
function was used instead of a discontinuous signum function. To further improve the fuzzy performance, an optimization
tool (genetic algorithm) has been implemented [17]. Novel fractional-order dynamics combined with a conventional SMC to
produce a novel three-dimensional autonomous Liu system was implemented on commensurate and incommensurate sys-
tems. Finally, fuzzy methodology was used to reduce the chattering phenomena in the proposed controller [18]. Along with
the study of different types of parametric models using different numerical examples, fractional-order parametric models
have been used to achieve higher accuracy in all cases of operation, and estimations of the parameters have been used to
implement certain control strategies [19]. The application of FrSMC for an incommensurable plant represents a novel imple-
mentation, as explained via numerical example [20]. A detailed analogy involving the development of a FrSMC-based extre-
mum seeking controller for a nonlinear system with optimization has been discussed; this controller was mainly developed
for its fast tracking performance and the improvement achieved in control accuracy [21]. To utilize both fractional-order sys-
tems and the Lyapunov stability theorem for a class of nonlinear systems, a conventional Lyapunov function was replaced
with a fractional-order function to improve the stability criteria in a previous study; this improvement was verified by
numerical simulations [22]. A fractional-order sliding surface was used to compensate for the mismatch, based on the back
stepping concept. This proposed concept was supported by abundant numerical examples [23]. In another study, a sliding
surface was formed using the fractional integral formula, and a special feedback law was used to form the control algorithm
for a nonlinear differential inclusion system such that all states were forced to converge towards the surface and reach the
desired equilibrium point [24]. The main objectives of the present work are: (i) design of FrSMC and SMC to control the level
of bottom two tanks of MQTP, (ii) a novel MLS technique is introduced in order to reduce the chattering in control signal, and
(iii) comparison of above said controllers in terms of fast convergence, better tracking and robustness.
This paper is structured as follows. In Section 2, a description of the dynamic model of the MQTP and an analysis of the
system are presented. The basics of fractional calculus, the design of the SMC and FrSMC, and the concept of a novel MLS are
developed in Section 3. In Section 4, the simulation results are presented, and the performances of the proposed controller
are analysed to demonstrate the controller’s efficacy. Finally, conclusions are presented in Section 5.

2. Modelling of the MQTP

The laboratory MQTP is shown schematically in Fig. 1. The purpose of this plant is to control the level of the two lower
tanks. The plant’s inputs are the voltage signals applied to the pumps. Fig. 2 depicts the experimental setup of the MQTP
consisting of four interconnected tanks with a common water source. Pump 1 discharges water to tank 1 and tank 4
12 S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21

Fig. 1. Schematic diagram of the MQTP.

simultaneously; the flows are indicated by rotameters 1 and 4, respectively. Similarly, pump 2 discharges water to tank 2 and
tank 3, and the flows are indicated by rotameters 2 and 3, respectively. The split of the flow from pump 1 and pump 2 can be
varied by manually adjusting the valves in tank 1 and tank 2. Tanks 1 and 2 also receive water by gravity flow from tank 3
and tank 4, respectively. The split valves in the rotameters can be manually adjusted to substantially alter the characteristics
of the system. The interacting tank system is so named due to the direct coupling between the lower tanks. The strength of
the coupling is controlled by means of a manual valve, which results in considerable variations in the process characteristics.

2.1. Dynamic model of the MQTP

A nonlinear process model is developed based on mass balance and Bernoulli’s law,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dh1 ðtÞ b a1 b a3 b a12 c Kp
¼ 1 2gh1 ðtÞ þ 3 2gh3 ðtÞ  12 sgnðh1 ðtÞ  h2 ðtÞÞ 2gjh1 ðtÞ  h2 ðtÞj þ 1 1 V 1 ðtÞ
dt A1 A1 A1 A1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dh2 ðtÞ b2 a2 b4 a4 b12 a12 c2 Kp2
¼ 2gh2 ðtÞ þ 2gh4 ðtÞ þ sgnðh1 ðtÞ  h2 ðtÞÞ 2gjh1 ðtÞ  h2 ðtÞj þ V 2 ðtÞ ð1Þ
dt A2 A2 A2 A2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dh3 ðtÞ b a3 ð1  c2 ÞKp2 dh4 ðtÞ b a4 ð1  c1 ÞKp1
¼ 3 2gh3 ðtÞ þ V 2 ðtÞ ¼ 4 2gh4 ðtÞ þ V 1 ðtÞ
dt A3 A3 dt A4 A4

Fig. 2. Experimental setup of the MQTP.


S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21 13

where

Ai – Cross-section of the tank (cm2).


ai – Cross-section of the outlet hole (cm2).
hi – Water level in the tank (cm).
bi – Outlet valve ratio; i = 1 to 4.
cj – Fraction of water flow from pump j.
v j – Voltage input of pumps 1 and 2 (volts); j = 1, 2.
g – Acceleration due to gravity (cm/s2).
b12 – Connected valve ratio.

When the sum of the flow ratios lies in the range of 1< (c1 þ c2 Þ 6 2, the system operates in MP mode (see Table 1).

3. Formulation of the control law

3.1. Basics of Fractional-Order Calculus

Fractional calculus has a 300-year mathematical history. Over the past three centuries, this subject was the exclusive
domain of mathematicians; only in the past few years has fractional calculus been implemented in several (applied) fields
of engineering science and economics. Perhaps fractional calculus will become the calculus of the 21st century. The
fractional-order differentiator can be denoted by a general fundamental operator as a generalization of the differential
and integral operators and is defined as follows:
8 da
>
> dta
RðaÞ > 0
>
<
a
Dt ¼ 1 RðaÞ ¼ 0 ð2Þ
>
>
>Rt
: a
a
ðdtÞ RðaÞ < 0
where a is the fractional order, which can be a complex number; the constant a is related to the initial conditions. The three
definitions used for the general fractional differential and integral are the Grunwald–Letnikov (GL) definition presented in
(3) and the Riemann–Liouville (RL) and Caputo definitions. Two commonly used definitions exist for the general fractional
differentiation and integration, i.e., the Grunwald–Letnikov (GL) and the Riemann–Liouville (RL) definitions, respectively.
 
1X n
j a
Dat f ðt Þ ¼ limh!0 a ð1 Þ f ðt  jhÞ ð3Þ
h j¼0 j

For n  1 < a < n, the initial conditions for the fractional-order differential equations with Caputo’s derivatives are of the
same form as those for the integer-order differential equations.

3.2. Conversion to standard form

To obtain the control law in SMC, the actual systems must be converted to the standard or canonical form of the
equations. Thus, for conversion purposes, the error variables are taken as phase variables, which are defined as follows, with
certain assumptions made to facilitate attaining the proper control law.

Assumptions

1. The two bottom tanks are operated under standard or constant voltages such that the derivative of the control action is
not needed. With this assumption, alterations are made to the actual dynamics of the system, and the relative degree of
the system becomes 2.

Table 1
Parameters of the process.

Process parameters MP Operating point


Ai (cm2)/ai (cm2) 176.7146/2.01062
(c1 ; c2 Þ/ðK P1; K P2 Þ (0.6, 0.6)/(70, 64)
0 0 0 0
ðh1 ; h2 ; h3 ; h4 Þ in cm (16.76, 16.11, 6.326, 8.16)
ðV 01 ; V 02 Þ
in (volts) (3.5, 3.5)
(b1 ; b2 ; b3 ; b4 ; b12 Þ (0.6, 0.7, 0.4, 0.385, 0.25)
14 S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21

2. The level of liquid in tank 1 is always greater than that in tank 2 (h1 > h2 Þ. This assumption is made to avoid the signum or
the switching function in the equations so that the derivative of the equations can be easily obtained.

Based on the above-mentioned assumptions, the rewritten equations become


pffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi c kp1 pffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi c kp2
h_ 1 ¼ K 11 h1 þ K 31 h3  K 12 h1  h2 þ 1 V 10 ; h_ 2 ¼ K 22 h2 þ K 42 h4 þ K 12 h1  h2 þ 2 V 20
A A ð4Þ
pffiffiffiffiffi ð1  c Þkp2 pffiffiffiffiffi ð1  c Þkp1
h_ 3 ¼ K 33 h3 þ 2
V 2 ; h_ 4 ¼ K 44 h4 þ 1
V1
A A
b a pffiffiffiffiffiffi
where K ij ¼ Aij 2g ; V i0 – Constant input voltage; j – Index of the tank, for which the dynamics are explained; and i – Index
of the tank written with respect to the jth tank.
The following equations are derived based on the assumptions and equations described above, which leads to the stan-
dard form of the equation.
x1 ¼ h1  href1 ; x3 ¼ h2  href2
where x1 ; x3 – Error generated in the system or the assumed state variable; h1 ; h2 – Actual height of tanks 1, 2; and href1 ; href2
– Reference heights of tanks 1, 2.
Thus, to obtain the phase variable format or simply a standard form of the equations, the following procedure is used. The
phase variable format is given by Eqs. (5) and (6).

x_ 1 ¼ h_ 1 ¼ x2
! !   !
K 11 K 12 K 12 K 31 K 33 K 31 kp2 ð1  c2 Þ ð5Þ
€1 ¼ 
x_ 2 ¼ h pffiffiffiffiffi þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4  þ pffiffiffiffiffi V2
2 h1 2 h1  h2 2 h1  h2 2 2 h3 A

x_ 3 ¼ h_ 2 ¼ x4
! !   !
K 22 K 12 K 12 K 42 K 44 K 42 kp1 ð1  c1 Þ ð6Þ
€2 ¼ 
x_ 4 ¼ h pffiffiffiffiffi þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2  þ pffiffiffiffiffi V1
2 h2 2 h1  h2 2 h1  h2 2 2 h4 A

3.3. Sliding Mode Control

Generally, sliding mode control is used to monitor the desired trajectory or any functional or input parameter changes,
within known bounds. According to the Hurtzwiz stability criterion, a sliding surface is constructed as follows:
S1 ¼ x2 þ m1 x1 ð7Þ
where m1 > 0 is a positive constant. The constant determines the slope of the sliding surface. Upon checking the stability of
the sliding surface itself, the positive definite Lyapunov can be obtained as follows:
1 2
V n2 ðS1 Þ ¼ S >0 ð8Þ
2 1
Differentiating equation Eq. (8) leads to a derivative Lyapunov function:

V_ n2 ðS1 Þ ¼ S1 S_ 1 < 0 ð9Þ

where S_ 1 ¼ x_ 2 þ m1 x2 substituting Eq. (5) into the above equation results in the following:
! !   !
K 11 K 12 K 12 K 31 K 33 K 31 kp2 ð1  c2 Þ
S_ 1 ¼ m1 x2  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4  þ pffiffiffiffiffi V2 ð10Þ
2 h1 2 h1  h2 2 h1  h2 2 2 h3 A

As a result, by substituting Eq. (10) into Eq. (9), Eq. (11) becomes
" ! !   ! #
K 11 K 12 K 12 K 31 K 33 K 31 kp2 ð1  c2 Þ
V_ n2 ðS1 Þ ¼ S1 m1  p ffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi x4  þ p ffiffiffiffiffi V2 < 0 ð11Þ
2 h1 2 h1  h2 2 h1  h2 2 2 h3 A

To ensure that V_ n2 ðS1 Þ < 0, the control law must be


!1 " ! !   #
K 31 kp2 ð1  c2 Þ K 11 K 12 K 12 K 31 K 33
V2 ¼  pffiffiffiffiffi m1  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4  þ a2  sgnðS2 Þ ð12Þ
2 h3 A 2 h1 2 h1  h2 2 h1  h2 2
 
where a P ðF þ gÞ ; F > ^f  f ^f – Estimated function; f – Actual function.
S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21 15

Control law V 1 is derived in a similar manner. S2 ¼ x4 þ m2 x3 is derived as follows


!1 " ! !   #
K 42 kp1 ð1  c1 Þ K 22 K 12 K 12 K 42 K 44
V1 ¼  pffiffiffiffiffi m2  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2  þ a1  sgnðS1 Þ ð13Þ
2 h4 A 2 h2 2 h1  h2 2 h1  h2 2

Thus, by using the above-described control law, the Lyapunov function becomes V_ n2 ðS1 Þ < 0.

3.4. Fractional-Order Sliding Mode Controller

The FrSMC is obtained from the Fractional-Order Sliding Surface and is given by Eq. (14):
S ¼ m1 x1 þ Da x1 ð14Þ
Rearranging Eq. (14) yields

S1 ¼ m1 x1 þ Da D1 x_ 1 ; S1 ¼ m1 x1 þ Dða1Þ x2 ð15Þ


Upon differentiating Eq. (15), we have

S_ 1 ¼ m1 x2 þ Dða1Þ x_ 2 ð16Þ
For simplicity, Eq. (16) is rearranged into the following form

S_ 1 ¼ 0 ¼ Dð1aÞ m1 x2 þ x_ 2 ð17Þ
By considering the following Lyapunov function
1 2
V ð S1 Þ ¼ S >0 ð18Þ
2 1
differentiating Eq. (18) yields
_ 1 Þ ¼ S1  S_ 1 ¼ S1 ðDð1aÞ m1 x2 þ x_ 2 Þ
VðS
! !   ! !
K 11 K 12 K 12 K 31 K 33 K 31 kp2 ð1  c2 Þ
¼ S1 Dð1aÞ m1 x2 þ pffiffiffiffiffi þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4  þ pffiffiffiffiffi V2 6 0 ð19Þ
2 h1 2 h1  h2 2 h1  h2 2 2 h3 A

Using fractional-order dynamics to ensure that Eq. (19) is negative definite, The sliding control law is as follows:
!1 " ! !   #
K 13 kp2 ð1  c2 Þ ð1aÞ K 11 K 12 K 12 K 13 K 33
V2 ¼  pffiffiffiffiffi D m1  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4  þ a2  sgnðS1 Þ ð20Þ
2 h3 A 2 h1 2 h1  h2 2 h1  h2 2

Control law V 1 is derived in a similar way from S2 ¼ m2 x3 þ Da D1 x_ 3 :


!1 " ! !   #
K 24 kp1 ð1  c1 Þ ð1aÞ K 22 K 12 K 12 K 24 K 44
V1 ¼  pffiffiffiffiffi D m2  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2  þ a1  sgnðS2 Þ ð21Þ
2 h4 A 2 h2 2 h1  h2 2 h1  h2 2

The finite response or convergence can be defined using FrSMC, which is not possible for a conventional SMC, by setting
Eq. (14) equal to zero and then rearranging and integrating it to produce the following equation:
Z tf   Z x1 ðtÞ
a 1 a
dt ¼  x1 d x1 ð22Þ
t0 m1 x1 ð0Þ 1

Because x1 ðtÞ ¼ 0, the final steady-state error must be zero, and upon integrating Eq. (22), the finite time of convergence is
obtained as follows:
 ð1=aÞ
a þða1Þ
t¼ jx1 ð0Þj a sec ð23Þ
ða  1Þm1

3.5. Multi-Level Switching (MLS) control

A novel MLS algorithm is presented to address the serious issue of chattering in SMC. For MLS, different operating
conditions of the system are studied and classified into three different operating regimes with different error of occurrence
conditions. The variables used are the following: different operating regions (LS – Lower level operating region, MS – Middle
Level Operating Region, and HS – Higher Level Operating Region), different error occurring regions (ZR – Around Zero Error
Condition, MR – Medium Error Occurrence Condition, and HR – High Error Occurrence Condition), and different switching
levels (L – Low level switching, M – Medium level switching, and H – High level switching).
16 S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21

The Exponential MLS variable gain is

g 0 ¼ ðVariable lev el switching lev elðupper  lowerÞlimitsÞ=ðfull lev el switching ðupper  lowerÞlimitsÞ

The logical combination of switching is presented in Table 2, and the various ranges for the operating regions with an MLS
gain is presented in Table 3. The chattering is reduced and smoothed by the MLS component. The smooth transition of
 
1
switching is given by GF ðsÞ ¼ g sþ1 . Then, the sgnðSÞ term is replaced by the ML variable gain combined with the smooth-
0
 
     t
1 1 1 jSj g
ness term, which is given by L ðGF ðsÞÞ j S j¼ L g 0 sþ1
 j S j¼ g0 e 0 . The control law containing the MLS gain compo-
nent for Tank 2 is given by Eq. (24).
!1 " ! !      t #
K 13 kp2 ð1  c2 Þ ð1aÞ K 11 K 12 K 12 K 13 K 33 j S1 j  g 0
V2 ¼  pffiffiffiffiffi D m1  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4  þ a2  e
2 h3 A 2 h1 2 h1  h2 2 h1  h2 2 g0
ð24Þ

Similarly, the control law containing the MLS gain component for Tank 1 is given by Eq. (25).
!1 " ! !      t #
K 24 kp1 ð1  c1 Þ ð1aÞ K 22 K 12 K 12 K 24 K 44 j S2 j  g 0
V1 ¼  pffiffiffiffiffi D m2  pffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x4 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2  þ a1  e
2 h4 A 2 h2 2 h1  h2 2 h1  h2 2 g0
ð25Þ

Thus, the combination of a smoothing filter and the MLS gain collectively reduces the excess control effort and sudden
switching transients in the final control element, respectively. The modulus of the sliding surface j S j is considered to satisfy
the Lyapunov criterion such that the obtained control law with the proposed methodology does not affect the stability of the
system. Consider a Lyapunov function and its first derivative, as in Eqs. (18) and (19). Upon proper substitution, the first
derivative of the Lyapunov function is obtained as indicated in Eq. (19), and then, the control law obtained in Eq. (24) for
Tank 1 is substituted into Eq. (19); finally, Eq. (26) is obtained, which is clearly a negative definite function. This result is
adequate for solving the stability problem. Thus, MLS does not alter the system stability region of the operation. Because
the condition g 0 > 0 is always valid, the exponential function also becomes positive. Thus, the function is always negative
as long as the conditions outlined above are not violated.
   t !
_VðS1 Þ ¼ S1  S_ 1 ¼ S1 a2  jS1 j e g0 < 0 ð26Þ
g0

4. Simulation results and discussion

The simulation results of MSMC and MFrSMC are discussed and compared for servo, regulatory and robustness responses
in the following sections.

4.1. Servo response

To test the ability of the proposed MFrSMC, multiple step changes in the process variables are given (initially, 26.32% of
the positive step change at 600 s and 52.64% of the negative step change at 1200 s). The MFrSMC is observed to display better
tracking performance with a faster settling time (albeit with some overshoot) and to reduce the undesirable chattering effect
in the control signal. However, MSMC shows some steady-state error. The corresponding variations in the process variables
and manipulated variables with SMC with and without MLS and MFrSMC are illustrated in Figs. 3 and 4. The theoretical
response times or convergence times, according to Eq. 23, are 64.23 s and 58.97 s for Tanks 1 and 2, respectively.
The simulation showed response times of 98 s and 97 s for Tanks 1 and 2, respectively. The results indicate that the prac-
tical and the mathematical finite time convergence times are in good agreement.

Table 2
Combinational logic table.

Set point change Disturbance (Error range) Resultant switching


LS ZR L
MS ZR L
HS ZR M
LS MR M
LS HR H
S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21 17

Table 3
Lookup table for the process variables and the corresponding operational bounds.

Setpoint change (Servo) Disturbance (Error Level) Switching level g0


[Level 1-(17–22), Level 2-(12–15)]- MS Error < absolute (0.25)-(ZR) Low level = (L) = (0.75 to 2.15) L = (0.28)
[Level 1-(22–26), Level 2-(15–19)]-HS Absolute (0.5)<Error < absolute (2.5)-(MR) Medium level = (M) = (1.5–2.6) M = (0.22)
[Level 1-(13–15), Level 2-(8–12)]-LS Error > absolute (2.5)-(HR) High level = (H) = (2.1 to 3.4) H = (0.27)

25
Level in Tank 1 (cm) MFrSMC
20 Set Point
SMC
MSMC
15

10

0
0 200 400 600 800 1000 1200 1400 1600 1800
Time (sec)

25
MFrSMC
Level in Tank 2 (cm)

20 SMC
Set Point
15 MSMC

10

0
0 200 400 600 800 1000 1200 1400 1600 1800
Time (sec)

Fig. 3. Servo response for change in the level of Tanks 1 and 2.

6
Control Signal 1 (volts)

SMC MFrSMC MSMC

0 200 400 600 800 1000 1200 1400 1600 1800


Time (sec)
6
Control Signal 2 (volts)

SMC MFrSMC MSMC

0 200 400 600 800 1000 1200 1400 1600 1800


Time (sec)

Fig. 4. Change in the manipulating variable for changes in levels of Tanks 1 and 2.

4.2. Regulatory response

Figs. 5 and 6 depict the performance of the two controllers with respect to disturbance rejection. After reaching a steady
state, a sudden disturbance d1 (drain 11.11% of water in tank 1) is introduced at 250 s, and d2 (22.22% increase in the flow
rate of Tank 1) is introduced at 500 s.
18 S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21

25

Level in Tank 1 (cm)


20

15
SMC
10 MFrSMC
Set Point
5 MSMC

0
0 100 200 300 400 500 600 700 800
Time (sec)
25
Level in Tank 2 (cm)

20

15
SMC
10 MFrSMC
Set Point
5 MSMC

0
0 100 200 300 400 500 600 700 800
Time (sec)

Fig. 5. Regulatory response for changes in the levels of Tanks 1 and 2.

6 SMC MSMC MFrSMC


Control Signal 1 (volts)

0 100 200 300 400 500 600 700 800


Time (sec)

6
Control Signal 2 (volts)

MSMC SMC MFrSMC

0 100 200 300 400 500 600 700 800


Time (sec)

Fig. 6. Change in manipulating variable for changes in the levels of Tanks 1 and 2.

Based on the response, the proposed controller is inferred to reject the disturbance quickly and with less chattering of the
control signals. The same amount of disturbance is also observed in Tank 2.

4.3. Robustness response

Simulation studies are performed to demonstrate the robust performance of the controllers presented. The robust control
action of the controller is tested by varying any of the dynamic parameters of the system. In the case of the MQTP, the inter-
action valve of the bottom tanks (1, 2) is physically changed from its nominal condition. To test the operational quality
afforded by the proposed method, the interaction valve is varied by adjusting the valve at various instances in both the pos-
itive and negative direction with both a gradual and an instantaneous pattern, as pictorially represented in Fig. 7(c).
Based on the simulated results, MFrSMC effectively handles the change in the system parameter in the MQTP interaction
valve and maintains a constant robust performance compared to the conventional MSMC, as illustrated in Figs. 7(a) and (b).
S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21 19

Level in Tank 1 (cm)


20

15 SMC
MFrSMC
Set Point
10 MSMC

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Time (sec)
20
Level in Tank 2 (cm)

15

10
MFrSMC
SMC
5 Set Point
MSMC
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Time (sec)
(a)

6 SMC MSMC MFrSMC


Control Signal 1 (volts)

0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Time (sec)
6
Control Signal 2 (volts)

SMC MSMC MFrSMC

0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Time (sec)
(b)
Interaction Valve Opening (%)

80

60

40

20

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Time (sec)
(c)
Fig. 7. Robustness of levels of Tanks 1 and 2 by changing the b12 positions gradually and instantaneously: (a) variations in the process variable, (b)
variations in the control input, (c) gradual and instantaneous change in the interaction valve b12 from its nominal position.

4.4. Comparison of the performance indices

Table 4 summarizes the performance of different controllers. The effectiveness of the proposed MFrSMC is compared with
that of the MSMC in terms of quality indices, such as the settling time and the peak overshoot, and error indices, such as ISE,
20 S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21

Table 4
Comparison of the performance of various controllers.

Controllers Performance indices


Quality indices Error indices
Loop OS (%) Ts (sec) Servo Regulatory
4
ITAE (10 ) IAE ISE ITAE (104) IAE ISE
MSMC h1 – 183 52.3 1101 9321 10.01 809.2 7450
h2 – 213 58.73 1113 8591 8.819 750.1 6604
MFrSMC h1 20.89 98 30.76 794.6 8052 4.267 583 6538
h2 22.61 99 31.59 777.1 7419 4.195 558.4 5845

IAE and ITAE, for the h1 loop and h2 loop for set point tracking and disturbance rejection. The proposed method shows lower
error indices than the other methods considered, as well as a considerable overshoot. However, the settling time is reduced
for the MFrSMC.

5. Conclusion

In this study, MFrSMC and MSMC controllers were used to maintain the same level of performance in a complex nonlinear
multivariable MQTP. Although SMC and MSMC controllers provide robust performance in this respect against uncertainties,
they have the drawback of a non-finite time convergence. In theory, the FrSMC concept is proven to provide a finite conver-
gence, which in turn helps to converge to a required equilibrium point much faster compared to other controllers. To support
the proposed controller theory, the simulation results indicate that a FrSMC affords better performance in level control with
a shorter convergence time. In addition, the proposed controller scheme rejects the uncertainties at a faster rate than the
other controllers mentioned above. An undesirable chattering effect, which reduces the life of the final control elements,
may sometimes lead to actuator failure (e.g., in the case of a pneumatic valve). Chattering phenomena commonly occur
in a SMC because the SMC is a switching controller. Thus, the novel approach of MLS was implemented, which involves
studying the behaviour of a system and tuning the MLS gain according to the current status of the operating point. The
proposed methodology reduces the chattering effect to a great extent, and an exponential term is used to ensure a smooth
transition of gain from one level to another level. Based on the control signal responses, which were described in the results
and discussion section, the chattering effect is more reduced in the MFrSMC than in the conventional SMC and MSMC. In
addition, the performance index table highlights the performance of the MFrSMC.

References

[1] O’Dwyer A. Handbook of PI and PID controller tuning rules. 2nd ed. Ireland: Dublin Institute of Technology, Imperial College Press; 2006. chap. 3.
[2] Bequette BW. Process control: modeling, design, and simulation. Prentice Hall Professional; 2003.
[3] Johansson KH. The quadruple-tank process: a multivariable laboratory process with an adjustable zero. IEEE Trans Control Syst Technol
2000;8(3):456–65.
[4] Slotine JJE, Li W. Applied nonlinear control, Vol. 199. Englewood Cliffs (NJ): Prentice-Hall; 1991 (No. 1) chapter 7.
[5] Almutairi NB, Zribi M. Sliding mode control of coupled tanks. Mechatronics 2006;16(7):427–41.
[6] Biswas PP, Srivastava R, Ray S, Samanta AN. Sliding mode control of quadruple tank process. Mechatronics 2009;19(4):548–61.
[7] Numsomran A, Tipsuwanporn V, Tirasesth K. Modeling of the modified quadruple-tank process. In: SICE annual conference. IEEE; 2008. p. 818–23.
[8] Numsomran A, Tipsuwanporn V, Trisuwannawat T, Tirasesth K. Design of PID controller for the modified quadruple-tank process using inverted
decoupling technique. In: 11th International Conference on Control, Automation and Systems (ICCAS). IEEE; 2011. p. 1364–8.
[9] Monje CA, Chen Y, Vinagre BM, Xue D, Feliu-Batlle V. Fractional-order systems and controls: fundamentals and applications. Springer; 2010.
[10] Delavari H, Ranjbar AN, Ghaderi R, Momani S. Fractional order control of a coupled tank. Nonlinear Dynam 2010;61(3):383–97.
[11] Tang Y, Zhang X, Zhang D, Zhao G, Guan X. Fractional order sliding mode controller design for anti lock braking systems. Neurocomputing
2013;111:122–30.
[12] Zhao G. Fractional-order fast terminal sliding mode control for a class of dynamical systems. Math Problems Eng 2013.
[13] Dadras S, Momeni HR. Fractional terminal sliding mode control design for a class of dynamical systems with uncertainty. Commun Nonlinear Sci
Numer Simulat 2012;17(1):367–77.
[14] Luo J, Liu H. Adaptive fractional fuzzy sliding mode control for multivariable nonlinear systems. Discrete Dynam Nat Soc 2014.
[15] Shi B, Yuan J, Dong C. Pseudo-state sliding mode control of fractional SISO nonlinear systems. Adv Math Phys 2013.
[16] Yuan J, Shi B, Zeng X, Ji W, Pan T. Sliding mode control of the fractional-order unified chaotic system. Abstract and applied analysis. Hindawi Publishing
Corporation; 2013.
[17] Delavari H, Ghaderi R, Ranjbar A, Momani S. Fuzzy fractional order sliding mode controller for nonlinear systems. Commun Nonlinear Sci Numer
Simulat 2010;15(4):963–78.
[18] Faieghi MR, Delavari H, Baleanu D. Control of an uncertain fractional-order Liu system via fuzzy fractional-order sliding mode control. J Vib Control
2011. 1077546311422243.
[19] Tavakoli-Kakhki M, Haeri M, SalehTavazoei M. Simple fractional order model structures and their applications in control system design. Eur J Control
2010;16(6):680–94.
[20] Valerio D, da Costa JS. Fractional sliding mode control of MIMO nonlinear noncommensurable plants. J Vib Control 2013. 1077546313480543.
[21] Yin C, Chen Y, Zhong SM. Fractional-order sliding mode based extremum seeking control of a class of nonlinear systems. Automatica
2014;50(12):3173–81.
[22] Chen D, Zhang R, Liu X, Ma X. Fractional order Lyapunov stability theorem and its applications in synchronization of complex dynamical networks.
Commun Nonlinear Sci Numer Simulat 2014;19(12):4105–21.
S. Sutha et al. / Computers and Electrical Engineering 45 (2015) 10–21 21

[23] Wang Z. Synchronization of an uncertain fractional-order chaotic system via backstepping sliding mode control. Discrete Dynam Nat Soc 2013. Article
ID 732503.
[24] Balochian S. Sliding mode control of fractional order nonlinear differential inclusion systems. Evol Syst 2013;4(3):145–52.

S. Sutha is currently a research scholar (P.T) at the College of Engineering, Guindy campus, Chennai, and is working as an associate professor in the
Department of Electronics and Instrumentation Engineering, Jayaram College of Engineering and Technology, Trichy, Tamil Nadu, India. She received her
B.E. degree in electronics and instrumentation engineering from Shanmugha College of Engineering, Thanjavur, and her M.E. degree in control and
instrumentation from the College of Engineering, Guindy, Chennai. Her present research interests are nonlinear control, optimization techniques and
advanced process control. Contact: suthaa_s@yahoo.co.in

P. Lakshmi received her B.E. degree from the Government College of Technology, Coimbatore, Tamil Nadu, India. She obtained her master’s and Ph.D.
degrees from the College of Engineering, Guindy, Anna University, Chennai. Presently, she is working as a professor in the Department of EEE, College of
Engineering, Guindy, Chennai, Tamil Nadu. Her current research interests lie in the areas of intelligent controllers, optimization techniques, process control
and power system stability. Contact: p_lakshmi@annauniv.edu

S. Sankaranarayanan received his B.E. degree in electrical and electronics engineering from Velammal Engineering College, Chennai, Tamil Nadu, India, and
received his M.E. degree in control and instrumentation from the College of Engineering, Guindy, Anna University, Chennai. He is currently pursuing his
Ph.D. in smart metering techniques for gas and water at the National Institute of Technology, Trichy, Tamil Nadu, India. His research interests are nonlinear
control and sliding mode control. Contact: kokilamsankar@gmail.com

You might also like