Journal of Process Control 24 (2014) 475484

Contents lists available at ScienceDirect

Journal of Process Control

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

Control of a distillation column by type-2 and type-1 fuzzy logic

PID controllers
Michele Miccio , Bartolomeo Cosenza 1
Dipartimento di Ingegneria Industriale (Dept. of Industrial Engineering), Universit degli Studi di Salerno, Via Giovanni Paolo II, 132, 84084 Fisciano, SA,
Italy

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

Article history: The aim of this paper is to develop a type-1 and a type-2 fuzzy logic PID controller (type-1 FLC and type-2
Received 11 July 2013 FLC, respectively) for the control of a binary distillation column, the mathematical model of which is char-
Received in revised form acterized by both high nonlinearities and parameter uncertainties. Attention was focused on the tuning
13 December 2013
procedure proposed by the authors and representing a development of the original Jantzen [1] method
Accepted 15 December 2013
for type-1 and type-2 fuzzy controllers, in particular including input type-2 Gaussian membership func-
Available online 15 January 2014
tions. A theoretical explanation of the differences in fuzzy controller performance was in fact provided
in the light of simulation results. The performance of a type-1 FLC was then compared in simulation with
Keywords:
Type-2 fuzzy logic PID controller
the one of type-2 FLC. All the simulation results conrmed the robustness and the effective control action
Feedback control of each fuzzy controller, with evident advantages for the type-2 FLC.
Non-linear dynamical system 2013 Elsevier Ltd. All rights reserved.
Mathematical modeling
Uncertainty
Distillation column

1. Introduction
It is well known that the control of many industrial chemical
processes characterized by high nonlinear dynamics and parameter
uncertainties does not give satisfactory responses using conven-
tional controllers. It is true that type-1 fuzzy logic controllers
(type-1 FLCs) [2] have been reported to be successfully used for
a number of complex and nonlinear processes, often coupled with
an optimal control: Karr et al. [3] developed a fuzzy control of an
exothermic chemical reaction using genetic algorithms; Cetinkaya
et al. [4] used a fuzzy-relational models-dynamics matrix control
for an optimal temperature control in a batch polymerization reac-
tor; Shahraz and Bozorgmehry Boozarjomehry [5] proposed a fuzzy
sliding mode control approach for nonlinear chemical processes;
while Lima et al. [6,7] designed a fuzzy model-based predictive
controller to control polymerization processes and proposed the
application of predictive control using fuzzy logic for a polymer-
ization system.
Despite their popularity, scientic research has recently shown
that in many cases type-1 FLCs can have difculties in minimiz-
ing the effect of uncertainties in the plant model, which are often
Corresponding author. Tel.: +39 089 964148.
E-mail addresses: mmiccio@unisa.it (M. Miccio), bartolomeocosenza@hotmail.it
(B. Cosenza).
1
Tel.: +39 328 7062159.
responsible for the degradation of the same process control. There-
fore, the choice of type-1 FLCs and then of type-1 fuzzy sets [8] may
not always represent the optimal solution to a control problem,
whereas a possible alternative is using type-2 fuzzy logic con-
trollers (type-2 FLCs) that make use of type-2 fuzzy sets [34,47]
characterized by a larger number of parameters and, as a conse-
quence, of freedom degrees. The type-2 fuzzy logic control has been
applied in many elds like anesthesia control [43], level control [9],
marine diesel engine control [10] and more recently vehicle non-
linear active suspension systems [11], autonomous mobile robot
control [46], cable-driven parallel mechanism [12], Kundur test sys-
tem [13], biochemical reactor control [1417], continuous stirred
tank reactor [18] and inverted pendulum model [19].
Fuzzy logic has been applied to control of a distillation col-
umn, too. Luo et al. [20] developed a fuzzy-neural net-based
inferential control for a high purity distillation column, while a
neuro-fuzzy modeling and control of a batch process involving
simultaneous reaction and distillation was developed by Wilson
and Martinez [21]. More recently Fabro et al. [22] proposed a
startup of a distillation column using intelligent control techniques
(fuzzy sets optimized by genetic algorithms), Worapradya and
Pratishthananda [23] developed a real-time control of a binary
distillation column using HGA fuzzy supervisory PI controllers; Fer-
nandez de Canete et al. [24] applied an adaptive neurofuzzy control
using soft sensors to continuous distillation; the control of binary
distillation column using PI controllers was proposed by Javadi and
Hosseini [25]; while Barcel-Rico et al. [26] realized the modeling

0959-1524/$ see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jprocont.2013.12.007
476 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484
Fig. 1. Distillation column.
and the control of a continuous distillation tower through fuzzy
technique. Other control strategies were applied to the distillation
column as an inferential control system of distillation composi-
tions using dynamic partial least squares regression by Kano et al.
[27,28]; a globally linearized control system design of a constrained
multivariable distillation column was instead developed by Jana
et al. [29]; Bezzo et al. [30] used a model predictive control for
middle-vessel continuous distillation columns; Kawathekar and
Riggs [45] realized a nonlinear model predictive control of a reac-
tive distillation column; a nonlinear adaptive control algorithm
for a multicomponent batch distillation column was developed by
Murlidhar and Jana [48]; and Sivakumar et al. [31] used a fuzzy
model predictive control in multivariable control of distillation col-
umn.
The main aim of the distillation control method developed in
this paper is to reach the set point value for the head concentra-
tion x1 starting from a given initial condition and, above all, to
have x1 tracking the set point, despite the variation of the system
parameters, in particular of the relative volatility .
2. Materials and methods
2.1. Distillation column model
The system model, here considered, is a conventional two-
component distillation column (Fig. 1) [32,33]. It is well known that
distillation is based on the separation of two or more components
of a liquid mixture by virtue of the differences in boiling points of
the components. Fractional distillation is the most common form of
separation technology used in petroleum reneries, petrochemical
and chemical plants. Design and operation of a distillation column
depends on the feed and desired products. The feed, i.e., the mix-
ture to be separated, enters the central section of the column as a
liquid, a liquidvapor mixture or a vapor. The stripping section is
the name given to the part of the column below the feed section.
In this section the concentration of the more volatile component
decreases. The enriching section is instead the name given to the
part of the column above the feed section. In this section the con-
centration of the more volatile component increases. On the top of
the column there is a condenser, the overhead vapor containing the
most volatile components from the feed moves from the top of the
column to the condenser and the condensate is collected in a tank
from which the distillate and the reux are taken. The reux is fed
back to the head of the column to provide liquid ow above the
feed point (as it is necessary to keep the liquid ow rate constant)
and it moves down the column in countercurrent ow with the
Fig. 2. Generic equilibrium stage.
vapor owing up the column. The distillate (i.e., overhead product)
contains liquid with the same composition specied in the design
of the distillation column. On the bottom there is a reboiler. The
residue is taken from the bottom as a liquid stream, then it is partly
vaporized in the reboiler and re-introduced in the column where
ows back up the column in countercurrent with the liquid moving
down the column.
The trays of a distillation column represent distillation stages,
where some of the vapor moving up the column is condensed and
some of the liquid moving down the column vaporizes. The trays
are characterized by many holes or bubble caps to allow the vapor
to pass through. Their utility is that to increase the contact time
between the vapor and the liquid in the column. If there are two
components in a liquid feed (as in this case study), a greater amount
of the more volatile component will vaporize and a larger amount of
the less volatile component will remain in the liquid condensate.
In the simplest limiting case, the substance with the lower boil-
ing point will condense in the top condenser and the species with
higher boiling point will leave the bottom of the column as a con-
densed liquid. The application of the column distillation considered
in the paper regards the separation of liquid streams containing
two components (binary mixture). As in a mixture of benzene and
toluene, the rst is the light component (the component that boils
at lower temperature) and the second one is the heavy component
(the component that boils at a higher temperature).
The dynamical mass balance on the component in the generic
stage (Fig. 2), excluding the feed stage, the condenser and the
reboiler, is as follows:
dMi xi
= Li1 xi1 + Vi+1 yi+1 Li xi Vi yi (1)
dt
The steady state assumption holds for the overall mass balance;
therefore, it is:
Vi = Vi+1 (2)
Li = Li+1 (3)
The overall mass balances for the feed tray (Fig. 3) are:
Vnf = Vnf +1 + F(1 q) (4)
Lnf = Lnf +1 + Fq (5)
where q is the rate (percentage) of liquid present in the feed and is
expressed as:
Lnf Lnf 1
q= (6)
F
whereas its complement to 1 (1 q) represents the rate of vapor
present in the feed.
The overall mass balance in the condenser is:
LD + D = V2 (7)
 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484 477

characteristic is that of incorporating uncertainty about the mem-

bership function into fuzzy set theory.
A fuzzy set A is so dened [34]:
 
A = A (x, u)/(x, u) (15)
xX u Jx

with x X, u Jx [0.1]. A (x, u) [0.1] is a type-2 membership

function and is a secondary grade while the primary membership
of x is the domain of the secondary membership function (Fig. 4a).
This kind of type-2 fuzzy set is also called general type-2 fuzzy set,
to distinguish it from a particular subcase (i.e., the interval type-2
fuzzy sets). An interval type-2 fuzzy set AI (Fig. 4b) can be instead
Fig. 3. Feed stage.
dened [3436] as a particular case of type-2 fuzzy logic and is
expressed as:
 
1
AI = (16)
where LD and D are the molar ow rates for the reux and distillate xX u Jx [0,1]
(x, u)
respectively.
The overall mass balance in the reboiler is: The membership function of a general type-2 fuzzy set A is
three-dimensional. In this case the third dimension represents the
B = LNS1 Vreboiler (8) value of the membership function at each point on its domain
where LNS1 is the total amount of the liquid entering the reboiler, (the footprint of uncertainty) characterized by two dimensions.
B is the bottom product ow rate and V is the ow of vapor from the The main difference with a normal type fuzzy set is that the sec-
reboiler that is re-injected into the column. The component mass ondary grade of an interval set A (x, u) can assume only two
I
balance is values: 0 or 1 and not values between 0 and 1 as a normal
set.
dx1 VR (y2 x1 )
in the condenser : = (9) As a preliminary study and in order to limit the compu-
dt MD
tational load, only interval type-2 fuzzy sets are treated in
this work. The mathematics that is needed for such sets is
dxi LR xi1 + VR yi+1 LR xi VR yi
in the rectifying section : = in fact simpler than that required for general type-2 fuzzy
dt MT sets.
(from i = 2 to i = nf 1) (10)
2.2.2. Type-2 fuzzy system structure
In Fig. 5 a type-2 fuzzy logic system [34,35,37] is shown,
dxnf LR xnf 1 + Vs ynf +1 + Fzf Ls xnf VR ynf characterized by four components: a Rules-Base, a Fuzzier, an
in the feed tray : =
dt MT Inference-Engine and an Output-Processor. The main difference
(11) between a type-1 (Fig. 6) and a type-2 fuzzy logic system (Fig. 5) is
represented just by the type-2 output processor. This last compo-
nent in fact maps a type-2 fuzzy set into a type-1 fuzzy set with the
dxi Ls xi1 + Vs yi+1 Ls xi Vs yi type-reducer block and then transforms the fuzzy output in a crisp
in the stripping section : =
dt MT output with a defuzzier block.
The structure of a type-2 fuzzy rule coincides with that of a type-
(from i = nf + 1 to i = ns 1) (12)
1 fuzzy rule. Considering in fact a type-2 FLS with p inputs x1 X1 ,
and . . ., xp Xp and one output y Y, and assuming that there are only
dxns Ls xns1 Bxns Vs yns M rules, the lth rule of the rule base has the following form [34]:
in the reboiler : = (13)
dt MB R1 : IF x1 is F1l and . . . and xp is Fpl , THEN y is Gl 1 = 1, . . ., M (17)
The vaporliquid thermodynamic equilibrium relationship for
each component is: 2.2.3. Type reduction and defuzzication
xi The output of the fuzzy inference engine block (Fig. 5) is a type-2
yi = (14) output fuzzy set, which must be type-reduced. The center of sets
1 + ( 1)xi
(the most commonly used type-reduction method) is expressed as:
The feed conditions are taken as follows, i.e., F = 1 mol/min,
zf = 0.5, q = 1, and the model parameters are given the following   
base values: = 1.5, ns = 41, nf = 21, reboiler vapor ow rate equal to Ycos (x) = [yl , yr ] =      
3.2 mol/min. The reux ow rate is the system manipulation vari- y1 y1 ,yr1 yM yM ,yrM f 1 f 1 ,f 1
l l -
able as usual in distillation control. It is well known that the reux 
is taking a part of the liquid from the condenser and feeding it back 1
  M  (18)
to the column, where it moves down in countercurrent ow with f M f M ,f M  
M

the vapor owing up the column. - f i yi / fi

i=1 i=1
2.2. Type-2 fuzzy logic
In Eq. (18) Ycos (x) is an interval set, whereas yl and yr are its
i
2.2.1. Interval type-2 fuzzy sets end-points, [f , f i ] and [yli , yri ] are the interval ring level of the ith
-
Type-2 fuzzy sets generalize traditional fuzzy sets (type-1 fuzzy rule and the centroid of the consequent interval type-2 fuzzy set,
sets) and were proposed for the rst time by Zadeh [8]. Their main respectively.
478 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484

Fig. 4. (a) General type-2 fuzzy set. The secondary membership function in the third dimension can assume values [0,1]. (b) Interval type-2 fuzzy set. The secondary
membership function in the third dimension can assume only two values: [0,1].

Fig. 5. Type-2 fuzzy logic system.

It is well known that Eq. (18) may be computed using the 

A (x) is the upper membership function associated with the upper
KarnikMendel iterative method [34,44]. Finally the crisp output bound of FOU () and denoted as:
can be obtained from yl and yr as a simple average:
A (x) FOU(A)
 xX (20)
yl + yr
y(x) = (19) A (x) is instead the lower membership function associated with
2 -
the lower bound of FOU (A) and denoted as (Table 1):
2.2.4. Management of uncertainty A (x) FOU(A) xX (21)
Type-2 fuzzy logic shows all its potential only in environments -
full of uncertainties. It is well known that uncertainty is an inherent
part in the real control systems. The measurement noise, the coarse 3. Theory and calculations
estimation of process parameters, the system parameter changes,
even the ambiguity in the meaning of words used in the rules (FLC) 3.1. Fuzzy control strategy
are all source of uncertainties [34,38,39]. The FOU (Footprint of
Uncertainty) is just the main characteristic of the type-2 fuzzy sets In this section the control of distillation control is presented
and represents the entire interval type-2 fuzzy set. focusing the attention on the presence of uncertainty and its nega-
Its shading (Fig. 7) denotes interval sets (the domain) for the tive effects on the system performance.
secondary membership functions (Figs. 4 and 7). With the FOU it is The main aim of the distillation control developed in this paper
possible to take into account all uncertainties present in the system is to reach the set point value for x1 (head concentration) starting
and minimize consecutively their negative effects. The limits of the from an initial condition and, above all, to have the x1 concen-
FOU (Fig. 4) are a lower and an upper membership function (Fig. 7). tration tracking the set point, despite the variation of the system

Fig. 6. Type-1 fuzzy logic system.

 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484 479

Fig. 7. Upper and Lower membership functions for a triangular type-2 membership
function.

parameters, particularly the relative volatility . This latter is a

measure of the difference between the vapor pressure of the more
volatile components and that of the less volatile components in
a liquid mixture. It is very sensitive to pressure, temperature and
feed composition variations. Quite often, the actual numerical value
used for the relative volatility in industrial distillation design and
operation is based on approximations. turns out a signicant
source of uncertainty in distillation and the hypothesis of a time
variation of this parameter is realistic. Therefore, the variation of
is taken into consideration in this paper; conversely, if not taken
properly into account, it might degrade system performance (see
Figs. 8 and 9).
Fig. 8. (a) Solution diagram x1 vs Reux for three different values. (b) Zoom of
A conventional feedback control architecture was considered in
Fig. 8a.
this work, making use of type-1 and type-2 fuzzy PID controllers,
which are described in detail in the next section.
In Fig. 8a the solution diagram is shown for x1 in the system con-
sidered, when changing the value of as a parameter. It is easy to
note by zooming (Fig. 8b) that decreasing the value of from 1.58
to 1.5 and from 1.5 to 1.4 (with the same value of reux) the value
of x1 decreases. This change, while keeping stable the steady-state
operating conditions, is not desirable because it leads to lower x1
concentrations. It is important to note that if the variation (small
as is) of a system parameter (the main source of uncertainty) is
not taken into consideration and well handled, it may degrade the
performance of the system. A sudden and unexpected increase in
temperature, for example, can lead to a consistent decrease of the
value of relative volatility, with drastic effects on the product qual-
ity. In Fig. 9 the approach to steady state in the knee region of
Fig. 8 at Reux = 2.706 is shown as a time series chart in the time Fig. 9. Time trajectory of the system x1 concentration for different values at xed
domain. Reux = 2.706: = 1.58; = 1.5; = 1.4.

Table 1
A similar result can be seen in Fig. 10, but this time increas-
Fuzzy set nomenclature.
ing the value of the reboiler vapor ow rate (from 3.1 to 3.2 and
Type-2 fuzzy set from 3.2 to 3.5). Also in this case, the variation of this variable
I Interval type-2 fuzzy set
may have negative effects on the controlled variable x1 . A high
Fi Fuzzy set in X (antecedent set)
FLC Fuzzy logic controller
performance controller must be able to control the system, driving
FOU Footprint of Uncertainty it toward the desired set-point value and preventing the system
G Fuzzy set in Y (consequent set) to work in undesirable operating conditions. Type-1 and above all
Jx Interval [0.1] type-2 fuzzy logic PID controllers (that make use of type-2 fuzzy
R Rule
sets) may represent an effective choice to handle such uncertainty
y Output variable (crisp output)
Y y domain that is often not taken into account in the design of the controller,
Ycos Interval set minimizing its negative effects on the system. This can be achieved
i
i
[f , f ] Interval ring level of the ith rule by changing the value of the reux ow rate properly, increasing or
[yli , yri ] Centroid of the consequent interval type-2 set decreasing it according to the case. The manipulation of the reux
x Input variable (crisp input) ow rate is therefore an efcient and fast way to control the col-
X Universe of discourse ( domain)
umn head concentration. The regulation of the head composition
Greek symbol using the reux ow rate as manipulated variable is actually used
 Type-1 membership function in industrial processes and represents the fastest way to control
A Type-2 membership function
this composition. The alternative is that of acting on the vapor ow
480 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484

Table 2
Type-1 fuzzy sets.

Error fuzzy sets

Negative = exp(0.5 ((error + 0.5)/0.3)2 );
Zero = exp(0.5 (error/0.35)2 );
Positive = exp(0.5 ((error 0.5)/0.3)2 );

Derivative error fuzzy sets

Negative = exp(0.5 ((delta error+0.5))/0.3)2 );
Zero = exp(0.5 (delta error/0.35)2 );
Positive = exp(0.5 (delta error 0.5/0.3)2 );

Output = [2;1;0;1;2]

Table 3
Type-2 fuzzy sets.
Fig. 10. Time trajectory of the system x1 concentration at xed Reux = 2.706 for
Error fuzzy sets
different reboiler vapor ow rate values: 3.1 mol/min; 3.2 mol/min; 3.5 mol/min.
NegativeLower = exp(0.5 ((error + 0.5)/0.34)2 );
NegativeUpper = exp(0.5 ((error + 0.5)/0.38)2 );
rate from the reboiler, which is located in the stripping section, but ZeroLower = exp(0.5 (error/0.30)2 );
the consequence is further delaying the system response. ZeroUpper = exp(0.5 (error/0.345)2 );
In the next section the fuzzy PID controllers designed are
PositiveLower = exp(0.5 ((error 0.5)/0.30)2 );
described in detail. PositiveUpper = exp(0.5 ((error 0.5)/0.345)2 );

3.2. Fuzzy controllers Derivative error fuzzy sets

Fuzzy PID controllers are adopted in this work to effectively
determine the manipulated variable and, hence, to guarantee high
control performances.
NegativeLower = exp(0.5 ((delta error + 0.5))/0.18)2 );
NegativeUpper = exp(0.5 ((delta error + 0.5))/0.30)2 );
ZeroLower = exp(0.5 (delta error/0.18)2 );
ZeroUpper = exp(0.5 (delta error/0.30)2 );

It is well known that FLC (fuzzy logic control) and PID (pro- PositiveLower = exp(0.5 (delta error 0.5)/0.18)2 );
portional integral derivative) control algorithms have been and PositiveUpper = exp(0.5 (delta error 0.5)/0.39)2 );
continue to be an important, effective and useful research eld Output = [2;1;0;1;2]
[1,40]. Even if the conventional PID have been successfully used
in the process industries for long time, remaining the most com-
Table 4
monly used algorithm today for their ease and functionality, it is Rule base matrix.
true that fuzzy logic control has increasingly covered a wide range
of practical areas. Experimental results in fact conrm that a fuzzy Negative Zero Positive

PID controller is able to perform better than the conventional PID Negative 2 1 0
controller [41]. The synergic fusion of these two control algorithms Zero 1 0 1
Positive 0 1 2
is therefore an useful and effective way for process systems char-
acterized by complexity and nonlinearity.
The tuning method for type-1 and type-2 fuzzy controllers is Therefore each fuzzy controller with Sugeno inference method
that of Jantzen [1] and each fuzzy controller is a function of three [42] makes use of Gaussian membership functions and is charac-
inputs: error, derivative error and integral error terized by two inputs: error and derivative error; and one output
(manipulation variable). It is easily understood from Fig. 11 that the
U = [f (GE error, GCE derivative error)
structure of this controller is the same for both type-1 and type-2
+ GIE integral error] GU (22) fuzzy logic PID controllers, that is same inputs, same output and
same rule base. The main differences are just in the fuzzy sets: the
way by which the inputs are processed has been explained above
The advantages to use derivative error and integral error are
(see Section 2.2).
those to reduce overshoot and to remove steady state error, respec-
In Tables 2 and 3 all type-1 and type-2 fuzzy sets are shown.
tively. It is common to separate the integral action in the fuzzy PD + I
In Table 4 the rule base matrix, which is equal for type-1 and
(FDP + I) controller [40] (Fig. 11), to avoid a rule base with three
type-2 FLCs and is constituted by 9 rules, is shown.
inputs, which would be rather big and complicate.
The center of type-2 Gaussian membership func-
tions is the same of that of type-1 Gaussian
membership functions. The amplitude values of type-2 fuzzy
sets are, on the contrary, chosen by minimizing an objective
function (i.e., IAE) [17,18]. Even if type-2 fuzzy PID controller
implementation is more complex than the type-1 counterpart, as
type-2 fuzzy sets are more demanding than the type-1 ones, the
tuning procedure developed by the authors and based on Jantzen
method allows to speed up type-2 fuzzy controller design.

4. Results

A code has been written in Matlab language with Simulink

support, for the solution of the dynamic mathematical model char-
Fig. 11. Fuzzy proportional integral derivative controller structure. acterized by a non-linear behavior, due to the system complexity.
 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484 481

Fig. 15. Time trajectory of the system manipulation variable (reux ow rate) for set
Fig. 12. Time trajectory of the system x1 concentration for set point tracking
point tracking (x1,SP = 0.99) under a disturbance in the parameter (linear variation
(x1,SP = 0.99) under a disturbance in the parameter (linear variation from 1.5 to
from 1.5 to 2 throughout the simulation time).
1.3 throughout the simulation time) and another disturbance in the vapor ow rate
(step from 3.2 to 3.5 at t = 30 min).

Typically, running the code takes a time of the order of 10 s on a

Intel Core 2 Duo processor T5500 CPU (1.6 GHz, 667 MHz FSB, 2 MB
L2 cache).
In the simulation results reported in this work (Figs. 1220),
the controlled variable x1 was always considered to have an ini-
tial condition equal to 0.98. In Fig. 12, with an imposed set-point
x1,SP = 0.99, the performance of the system controlled by type-1 and
type-2 FLCs is shown by disturbing the system with a linear varia-
tion of the parameter from 1.5 to 1.3 during the entire simulation,
i.e., in a time as long as 50 min; further, a step in the vapor ow from
3.2 to 3.5 is superimposed at time t = 30 min. It is evident that the
type-2 FLC is able to reach the set point value earlier than type-1
Fig. 16. Time trajectory of the system x1 concentration for set point tracking (step
from 0.99 to 0.995 at t = 28 min) and disturbance rejection (step in from 1.56 to
1.8 at t = 35 min).

Fig. 13. Time trajectory of the system manipulation variable (reux ow rate) for set
point tracking (x1,SP = 0.99) under a disturbance in the parameter (linear variation
from 1.5 to 1.3 throughout the simulation time) and another disturbance in the Fig. 17. Zoom of Fig. 16 following the step from 1.56 to 1.8 at t = 35 min.
vapor ow rate (step from 3.2 to 3.5 at t = 30 min).

Fig. 18. Time trajectory of the system manipulation variable (reux rate) in set point
Fig. 14. Time trajectory of the system x1 concentration for set point tracking tracking (step in x1,SP from 0.99 to 0.995 at t = 28 min) and disturbance rejection (step
(x1,SP = 0.99) under a disturbance in the parameter (linear variation from 1.5 to in from 1.56 to 1.8 at t = 35 min).
2 throughout the simulation time).
482 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484

Table 5
Performance indexes of fuzzy PID controllers.
  2
  2
IAE = |e(t)|dt ISE = e(t) dt ITAE = t|e(t)|dt ITSE = te(t) dt
0 0 0 0

Type-2 FLC 0.02978 0.0001095 0.2194 0.0001652

Type-1 FLC 0.04083 0.0001724 0.3259 0.0003262

Table 6
Performance indexes of fuzzy PID controllers.
  2
  2
IAE = |e(t)|dt ISE = e(t) dt ITAE = t|e(t)|dt ITSE = te(t) dt
0 0 0 0

Type-2 FLC 0.03081 0.0001095 0.247 0.0001762

Type-1 FLC 0.04175 0.0001714 0.3614 0.0003355

Table 7
Performance indexes of fuzzy PID controllers.
  2
  2
IAE = |e(t)|dt ISE = e(t) dt ITAE = t|e(t)|dt ITSE = te(t) dt
0 0 0 0

Type-2 FLC 0.0455 0.0001368 0.814 0.0009795

Type-1 FLC 0.05027 0.0001673 0.8871 0.001014

Fig. 19. Type-1 and type-2 fuzzy set for the error, the rst input of the fuzzy con-
troller, at the beginning of each simulation.
FLC and with a lower overshoot. The calculated values of the IAE,
ISE, ITAE and ITSE shown in Table 5 conrm the simulation results.
In each case, in fact, the performance index of the type-2 FLC is
lower than the type-1 counterpart. In Fig. 13 the behavior of the
system manipulation variable (reux ow rate) is further shown.
The effects of the vapor step at t = 30 min are barely visible in Fig. 12
(the system controlled variable seems not to notice this change),
Fig. 20. Zoom of Fig. 19 in the saturation region.
but are more evident just in the behavior of the system reux ow
rate shown in Fig. 13.
A similar analysis was further pursued by varying the parameter
in the opposite direction. In Fig. 14 the time trajectory of the
system x1 concentration is shown when is allowed to linearly
grow as a disturbance from 1.5 to 2 during the entire simulation,
i.e., 50 min. The result of the simulation conrms what had been
stated previously, with a type-2 FLC showing a better performance
in terms of response speed and lower overshoot.
Similarly, Fig. 15 shows the behavior of the manipulation vari-
able (reux ow rate) and Table 6 the performance indexes of fuzzy
PID controllers.
Analogous results are shown in Figs. 1618 for set point tracking
and disturbance rejection. In this case, the set-point was changed
stepwise from 0.99 to 0.995 at t = 28 min and, later on, was
stepped from 1.56 to 1.8 at t = 35 min. Although the differences in
the performance results between type-1 and type-2 FLC are not so
evident as in the previous simulation results, the type-2 FLC still
has the best performance. The zoom of Fig. 16 shown in Fig. 17
in fact conrms the previous statement. As in the preceding cases,
Fig. 18 conrms the same behavior as concerns the system manip-
ulation variable and Table 7 endorses the benets of type-2 fuzzy
PID controllers as far as the performance indexes.
In order to give a more sound explanation of the advantages
provided by a type-2 FLC in the above results, the attention was
focused on the control loop error, i.e., the rst input to both type-
1 and type-2 FLCs. By tracking the error during each simulation
run, it was observed that the earliest value of the error, i.e., the
rst input to the type-1 FLC at the beginning of the simulation, is
located at one extreme of its range revealing a condition of satu-
ration (see the black curve in the leftmost part of Figs. 19 and 20).
Later on, the error decreases because of the loop controlling action,
the rst input moves toward the center of its range and, after some
time, the system reaches the set point and the error approaches
zero.
For the type-2 FLC, on the contrary, the presence of a lower
and an upper membership function as the extremes of the rst
input range allows for more error handling. It was observed that
the type-2 FLC does not reach the saturation condition of its type-1
counterpart and the error moves toward the center of its range (see
the shaded band in the central part of Fig. 19) in a shorter time, thus
ensuring faster and a more effective control action.
 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484
5. Conclusions
The application of type-2 and type-1 FLCs to a complex and
nonlinear dynamical system, like a distillation column presenting
parameter uncertainty, has been analyzed by simulations. The
uncertainty in the main process parameters, i.e., the relative
volatility and the reboiler vapor ow rate, has been handled by
introducing a time variation law for them. All the simulation results
demonstrate the effectiveness of type-2 FLC in achieving a very
high control performance and allowing a faster and more precise
control of the process, both for set point tracking and disturbance
rejection, with less amount of overshoot as compared to the type-1
counterpart controller. The favorable time trajectories of both the
controlled variable and the manipulation variable, as obtained by
the type-2 FLC, are also conrmed by values of the conventional
performance indexes: as an example, according to the simulations
carried out, the type-2 fuzzy control ensures an average improve-
ment of 27% in the IAE value with respect to the type-1 controller
counterpart. Moreover, the type-2 FLC, because of the built-in pres-
ence of a lower and an upper membership function as the extremes
of its rst input range, allows for a smoother error handling and
prevents the occurrence of saturation conditions. All in all, the sim-
ulation results produced in this work are in line and conrm what
was shown in previous papers on fuzzy control [1418]. It can be
concluded that the robustness of the type-2 FLC does not require a
tuning optimization or a frequent retuning; moreover, the tuning
method of type-2 fuzzy controller developed by the authors and
based on Jantzen method allows to speed up type-2 fuzzy controller
design.
In the light of simulation results, a theoretical discussion on
differences in fuzzy controller performance provides a sound expla-
nation about the reasons for the type-2 fuzzy controller best
performance. The results in the paper are very encouraging because
from the error pattern and from a suitable modication of type-2
fuzzy set, it will be possible to further boost the performance of
type-2 fuzzy control compared with its type-1 counterpart, without
using complex and discouraging optimization algorithms.
References
[1] J. Jantzen, Nonlinear Fuzzy PID Control, in Foundations of Fuzzy Control, John
Wiley & Sons Ltd., Chichester, UK, 2007.
[2] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338353.
[3] C.L. Karr, S.K. Sharma, W.J. Hatcher, T.R. Harper, Fuzzy control of an exother-
mic chemical reaction using genetic algorithms, Engineering Applications of
Articial Intelligence 6 (1993) 575582.
[4] S. Cetinkaya, Z. Zeybek, H. Hapoglu, M. Alpbaz, Optimal temperature control in
a batch polymerization reactor using fuzzy-relational models-dynamics matrix
control, Computers and Chemical Engineering 30 (2006) 13151323.
[5] A. Shahraz, R. Bozorgmehry Boozarjomehry, A fuzzy sliding mode control
approach for nonlinear chemical processes, Control Engineering Practice 17
(2009) 541550.
[6] N.M.N. Lima, F. Manenti, R. Maciel Filho, M. Embirucu, M.R. Wolf Maciel, Fuzzy
model-based predictive hybrid control of polymerization processes, Industrial
and Engineering Chemistry Research 48 (2009) 85428550.
[7] N.M.N. Lima, L.Z. Linan, R.M. Filho, M.R. Wolf Maciel, M. Embirucu, F. Grcio,
Modeling and predictive control using fuzzy logic: application for a polymer-
ization system, AIChE Journal 56 (2010) 965978.
[8] L.A. Zadeh, The concept of a linguistic variable and its applications to approxi-
mate reasoning-1, Information Sciences 8 (1975) 199249.
[9] D. Wu, W.W. Tan, Genetic learning and performance evaluation of interval type-
2 fuzzy logic controllers, Engineering Applications of Articial Intelligence 19
(2006) 829841.
[10] C. Lynch, H. Hagras, V. Callaghan, Using uncertainty bounds in the design of
an embedded real-time type-2 neuro-fuzzy speed controller for marine diesel
engines, in: Proceedings of FUZZ-IEEE Conference, Vancouver, Canada., 2006,
pp. 72177224.
[11] J. Cao, H. Liu, P. Li, D. Brown, Adaptive fuzzy logic controller for vehicle active
suspension with interval type-2 fuzzy membership functions, in: Proceedings
of the IEEE-FUZZ Conference, Hong Kong, China, 2008, pp. 13611373.
[12] C.D. Li, J.Q. Yi, Y. Yu, D.B. Zhao, Inverse control of cable-driven parallel mech-
anism using type-2 fuzzy neural network, Acta Automatica Sinica 36 (2010)
459464.
483
[13] I. Robandi, B. Kharisma, Design of interval type-2 fuzzy logic based power sys-
tem stabilizer, International Journal of Electrical Power and Energy Systems
Engineering 2 (2009) 7380.
[14] M. Galluzzo, B. Cosenza, A. Matharu, Control of a nonlinear continuous bioreac-
tor with bifurcation by a type-2 fuzzy logic controller, Computers and Chemical
Engineering 32 (2008) 29862993.
[15] M. Galluzzo, B. Cosenza, Control of the biodegradation of mixed wastes in a con-
tinuous bioreactor by a type-2 fuzzy logic controller, Computers and Chemical
Engineering 33 (2009) 14751483.
[16] M. Galluzzo, B. Cosenza, Type-2 fuzzy control of a bioreactor, in: Proceedings
of the IEEE International Conference on Intelligent Computing and Intelligent
Systems, Shangai, China, 2009, pp. 700704.
[17] M. Galluzzo, B. Cosenza, Nonlinear fuzzy control of fed-batch reactor for
the penicillin production, Computers and Chemical Engineering 36 (2012)
273281.
[18] M. Galluzzo, B. Cosenza, Control of a non-isothermal continuous stirred tank
reactor by a feedback-feed forward structure using type-2 fuzzy logic con-
trollers, Information Sciences 181 (2011) 35353550.
[19] Y.M. Li, Y.-Y. Sun, Type-2 TS fuzzy impulsive control of non-
linear systems, Applied Mathematical Modeling 36 (2012)
27102723.
[20] R.F. Luo, H.H. Shao, Z.J. Zhang, Fuzzy-neural-net-based inferential control
for a high-purity distillation column, Control Engineering Practice 3 (1995)
3140.
[21] J.A. Wilson, E.C. Martinez, Neuro-fuzzy modeling and control of a batch pro-
cess involving simultaneous reaction and distillation, Computers and Chemical
Engineering 21 (1997) S1233S1238.
[22] J.A. Fabro, L. V.R., F.N. Arruda Jr., Startup of a distillation column using intel-
ligent control techniques, Computers and Chemical Engineering 30 (2005)
309320.
[23] K. Worapradya, S. Pratishthananda, Real-time control of a binary distillation
column using HGA fuzzy supervisory PI controllers, in: Tencon 2005 2005 IEEE
Region 10, 2005, pp. 814818.
[24] J. Fernandez de Canete, P. Saz-Orozco, S. Gonzalez-Perez, Application of adap-
tive neurofuzzy control using soft sensors to continuous distillation, Computer
Aided Chemical Engineering 25 (2008) 465470.
[25] S. Javadi, J. Hosseini, Control of binary distillation column using PI controllers,
in: Proceeding of the 9th WSEAS International Conference on Signal Processing,
Computational Geometry and Articial Vision, 2009, pp. 145152.
[26] F. Barcel-Rico, J.M. Gozlvez-Zafrilla, J.L. Dez, A. Santaf-Moros, Modelling and
control of a continuous distillation tower through fuzzy techniques, Chemical
Engineering Research and Design 89 (2011) 107115.
[27] M. Kano, K. Miyazaki, S. Hasebe, I. Hashimoto, Inferential control system of dis-
tillation compositions using dynamic partial least squares regression, Journal
of Process Control 10 (2000) 157166.
[28] M. Kano, N. Showchaiya, S. Hasebe, I. Hashimoto, Inferential control of dis-
tillation compositions: selection of model and control conguration, Control
Engineering Practice 11 (2003) 927933.
[29] A.K. Jana, A.N. Samanta, S. Ganguly, Globally linearized control system design
of a constrained multivariable distillation column, Journal of Process Control
15 (2005) 169181.
[30] F. Bezzo, F. Micheletti, R. Muradore, M. Barolo, Using MPC to control middle-
vessel continuous distillation columns, Journal of Process Control 15 (2005)
925930.
[31] R. Sivakumar, S. Manic, V. Nerthiga, R. Akila, K. Balu, Application of fuzzy model
predictive control in multivariable control of distillation column, International
Journal of Chemical Engineering and Applications 1 (2010) 3842.
[32] J.D. Seader, E.J. Henley, Equilibrium-Stage Separation Operations in Chemical
Engineering, John Wiley and Sons, Ingls, 1981.
[33] J.D. Seader, E.J. Henley, D.K. Roper, Separation Process Principles, John Wiley &
Sons, Inc., Hoboken, NJ, 2010.
[34] J.M. Mendel, Uncertainty Rule-Based Fuzzy Logic Systems: Introduction and
New Directions, Prentice-Hall, Englewood Cliffs, NJ, 2000.
[35] Q. Liang, J.M. Mendel, Interval type-2 fuzzy logic systems, Proceedings of the
IEEE Transactions on Fuzzy Systems 8 (2000) 535550.
[36] J.R. Castro, O. Castillo, L.G. Martnez, Interval type-2 fuzzy logic toolbox, Engi-
neering Letters 15 (2007) 8998.
[37] N.N. Karnik, J.M. Mendel, Introduction to type-2 fuzzy logic systems,
in: Proceedings of the IEEE-FUZZ Conference, Anchorage, AK, 1998, pp.
915920.
[38] G.J. Klir, M.J. Wierman, Uncertainty-Based Information, Physica-Verlag, Heidel-
berg, Germany, 1998.
[39] L.A. Zadeh, Toward a generalized theory of uncertainty (GTU) an outline,
Information Sciences 172 (2005) 140.
[40] J. Jantzen, Tuning Of Fuzzy PID Controllers, Technical University of Denmark,
1998, Report 98-H 871.
[41] E. Natsheh, K.A. Buragga, Comparison between conventional and fuzzy logic
PID controllers for controlling dc motors, International Journal of Computer
Science Issues 7 (2010) 128134.
[42] H. Takagi, M. Sugeno, Fuzzy identication of systems and its applications to
modelling and control, IEEE Transactions on Systems, Man, and Cybernetics 15
(1985) 116132.
[43] O. Castillo, G. Huesca, F. Valdez, Evolutionary computing for optimizing
type-2 fuzzy systems in intelligent control of non-linear dynamic plants, in:
Proceedings of the North American Fuzzy Information Processing Society,
NAFIPS, Ann Arbor, MI, 2005, pp. 247251.
484 M. Miccio, B. Cosenza / Journal of Process Control 24 (2014) 475484

[44] N.N. Karnik, J.M. Mendel, Operations on type-2 fuzzy sets, Fuzzy Sets and Sys-
tems 122 (2000) 327348.
[45] R. Kawathekar, J.B. Riggs, Nonlinear model predictive control of a reactive dis-
tillation column, Control Engineering Practice 15 (2007) 231239.
[46] R. Martnez, O. Castillo, L.T. Aguilar, Optimization of interval type-2 fuzzy logic
controllers for a perturbed autonomous wheeled mobile robot using genetic
algorithms, Information Sciences 179 (2009) 21582174.
[47] J.M. Mendel, Q. Liang, Pictorial comparisons of type-1 and type-2 fuzzy logic
systems, in: Proceedings of the IASTED International Conference on Intelligent
Systems and Control, Santa Barbara, CA, 1999, pp. 280285.
[48] G.M. Murlidhar, A.K. Jana, Nonlinear adaptive control algorithm for a multi-
component batch distillation column, Chemical Engineering Science 62 (2007)
11111124.