Studies in Informatics and Control, 32(3) 89-98 September 2023 ISSN: 1220-1766 eISSN: 1841-429X 89

Design of an Improved Interval Type-2 Controller Using

FCM and Supervised Clustering Algorithms
Anup Kumar MALLICK*, Achintya DAS
Department of Electronics & Communication Engineering, Kalyani Government Engineering College,
Kalyani, Nadia, 741235, India
[email protected] (*Corresponding author), [email protected]
Abstract: In the last few decades, the interval type-2 fuzzy controller has gained popularity in comparison with the type-
1 fuzzy controller. This is due to the capability of the interval type-2 fuzzy controller to better handle uncertainty and
imprecision. However, modeling an intervaltype-2 fuzzy controller brings about several hurdles. One of the challenges of the
design of an interval type-2 fuzzy controller is the generation of the primary membership functions with the purpose of formulating
the upper and lower membership functions. This paper proposes a technique by which two Gaussian primary membership functions
are generated for an interval type-2 fuzzy set. The means for the Gaussian membership functions are generated by employing two
clustering techniques, namely the standard fuzzy c-means clustering algorithm and a recently developed supervised clustering
algorithm. The standard deviations of the Gaussian membership functions are optimally selected by using the differential evolution
algorithm. Besides, the number of rules required for the proposed model is much smaller. The proposed controller is applied to an
armature-controlled DC motor and the obtained simulation results are compared with those obtained for the conventional interval type-2
fuzzy controller. The robustness of the proposed controller is also checked by adding noise and an impulse disturbance during the simulation.
Keywords: Interval Type-2 controller, Fuzzy C-means Clustering, Differential Evolution, Armature-Controlled DC Motor,
Robust Control, Optimal Control.

1. Introduction
The type-1 (T1) fuzzy set (FS) can model a system these two methods. IT2FSMC is used in many
based on a single expert’s knowledge. With the control applications, for example, to control a
type-2 (T2) fuzzy set, introduced by Zadeh 3-DOF helicopter (Zeghlache et al., 2017), to
(1975) and pioneered by Mendel & John (2002), control a robot manipulator (Nafia et al., 2018)
it is possible to accommodate the knowledge and so on. Using a genetic algorithm, an optimal
of several experts. In type-2 fuzzy systems, the type-2 fuzzy controller is implemented for the
membership functions (MFs) are themselves velocity regulation of a DC motor in (Maldonado
fuzzy instead of crisp as in type-1 fuzzy sets. & Castillo, 2012). Behera et al. (2017) have
Although T2 fuzzy sets possess better handling applied an interval type-2 controller in the
capabilities for uncertainties and imprecision, automatic generation control of a restructured
their real-world applications require complex power system. Control of a one-wheel vehicle
practical considerations (Wu & Mendel, 2019). As in the real world has been developed using an
an intermediate solution, the more practical model adaptive interval type-2 fuzzy controller in (Chiu
for handling uncertainties is the interval type-2 & Hung, 2020). Zakovorotniy & Kharchenko
(IT2) fuzzy set. In interval type-2 fuzzy sets, the (2021) designed an optimal controller associated
membership is an interval instead of T1 fuzzy sets with speed control using the interval type-2 fuzzy
as in traditional generalized type-2 fuzzy sets. In set. The interval type-2 controller has also been
the recent past, the modeling of control systems designed based on the variations of the footprint
using IT2 FS has attracted increased interest from of uncertainty. Zhou et al. (2021) have analyzed
the research community. the variation trend of variable gains, in relation to
the increase of footprint of uncertainty. Şahin &
Liang & Mendel (2000) described the theory and Ulu (2023) have proposed an IT2 controller with
the design steps for an interval type-2 fuzzy logic a dynamic footprint of uncertainty. The dynamic
system. Several researches and applications for footprint of uncertainty has been defined as a
the IT2 controller were followed thereafter. Wu function of system error. In (Şahin & Ulu, 2023),
& Tan (2004) proposed a singleton interval type-2 the IT2 controller with the dynamic footprint of
fuzzy controller to design the liquid-level control. uncertainty has been applied for controlling the
An interval type-2 fuzzy sliding mode controller altitude of a quadcopter.
(IT2FSMC) is proposed in (Hsiao et al., 2008).
IT2FSMC is a combination of the interval type- One of the important tasks of an interval
2 fuzzy logic control (IT2FLC) and the sliding- type-2 fuzzy set is to generate multiple type-
mode control (SMC) which shares the benefits of 1 fuzzy sets and then generate the upper and

https://doi.org/10.24846/v32i3y202308 ICI Bucharest © Copyright 2012-2023. All rights reserved

90 Anup Kumar Mallick, Achintya Das

lower membership functions from these type- (FOU). The FOU enables IT2 FS to represent the
1 membership functions by using some fuzzy uncertainty in decision making. An example of
operations. Similarly, an interval type-2 fuzzy an IT2 FS is shown in Figure 1 to indicate the
logic controller should incorporate multiple type- different components of an IT2 FS.
1 fuzzy control strategies for the construction
of the upper and lower membership functions.
However, most of the IT2 FLC modeling found
in literature either uses standard symmetrical
equally spaced membership functions or the
parameters of upper and lower membership
functions are obtained by applying some
optimization techniques. Not much attention
is given to generating the primary membership
functions to formulate the upper and lower Figure 1. A schematic of an interval type-2 fuzzy set
membership functions. This paper addresses
The following definitions related to interval type-2
the above issue, and proposes a systematic
FS are used throughout the paper.
method for generating two sets of membership
functions for the interval type-2 controller. Definition 1[(IT2 FS)]
First, two primary membership functions are
generated by using the clustering techniques A T2 FS à is characterized by µÃ(x,m) and
and the differential evolution (Storn, 2008; Das expressed by the following equation:
& Suganthan, 2011) algorithm. Then, the upper
and lower membership functions are generated A = ∫ ∫ µ A ( x, m) / ( x, m) (1)
x∈ X m∈J x
by taking the fuzzy union and the fuzzy
intersection of these two primary membership Here, X is the universe of discourse of the variable
functions. This work addresses another challenge x, m ∈ J x ⊆ (0,1) is the primary membership
for interval type-2 controller, i.e., the number of value of x, and µÃ(x,m) denotes the secondary
rules. The proposed controller vividly reduces membership grade that is 1 for IT2 FS.
the required number of rules, thereby easing
the computation. An armature-controlled DC Definition 2 [(FOU)]
motor is used to check the performance of the The FOU of à is defined as the union of all of its
proposed controller. Noise and disturbance are primary MFs, and is expressed by equation (2):
added during simulation to check the robustness
of the proposed controller. The remainder of this FOU ( A ) = ( x, m);
∪∀x∈X J x =
paper is structured as follows. Section 2 outlines (2)
the interval type-2 fuzzy sets and interval type-2 where m ∈ J x ⊆ [0,1].
fuzzy logic controllers. Section 3 illustrates the Definition 3 [(Upper and lower MF)]
proposed fuzzy controller. Section 4 provides
the description of the applied process. The The upper bound of FOU(Ã) is called the UMF
simulation results are presented and discussed and is denoted by µ A ( x) . Similarly, the lower
in section 5. Section 6 includes the conclusion of bound of FOU(Ã) is called LMF and is expressed
this paper and refers to its possible future scope. by µ A ( x) . That is,

2. Preliminaries =UMF FOU ( A ) for ∀x ∈ X , (3)

and
2.1 Interval Type -2 Fuzzy Set
=LMF FOU ( A ) for ∀x ∈ X . (4)
Unlike type-1 fuzzy sets, in IT2 FS (Mendel &
Wu, 2010), each membership function consists 2.2 Interval Type-2 Fuzzy
of two MFs, i.e., upper membership function Logic Controller
(UMF) and lower membership function (LMF).
The area in the interval between the upper MF An interval type-2 fuzzy controller comprises
and the lower MF is called footprint of uncertainty four major building blocks, i.e. a fuzzifier, an

https://www.sic.ici.ro
 Design of an Improved Interval Type-2 Controller Using FCM and Supervised Clustering Algorithms 91

inference engine, a type reducer, and a defuzzifier

(Chakraborty et al., 2015). A block diagram of
an IT2 fuzzy logic controller (FLC) is shown in
Figure 2.

Figure 3. Membership functions of e, ∆e and u

Figure 2. Block diagram of an interval type-2 FLC with NB: negative big, NM: negative medium, NS:
negative small, ZE: zero, PS: positive small, PM:
positive medium, and PB: positive big
3. The Proposed Interval
Type-2 Controller Then the respective control actions for the
sampled input space are calculated by the modified
The proposed interval type-2 fuzzy controller is version of the original MacVicar-Whelan rule base
developed in the following three stages, namely (MacVicar-Whelan, 1976) with the 7 linguistic
generation of membership functions, formation variables listed in Table 1.
of a fuzzy inference engine, and defuzzification.
Table 1. Rule base for generating the control action
3.1 Generation of Membership
∆e/e NB NM NS ZE PS PM PB
Functions
NB NB NB NB NM NS NS ZE
The common membership functions used in FLCs NM NB NM NM NM NS ZE PS
are the Gaussian, triangular, and trapezoidal one. NS NB NM NS NS ZE PS PM
For the proposed IT2 controller, the Gaussian ZE NB NM NS ZE PS PM PB
membership functions are considered for PS NM NS ZE PS PS PM PB
two reasons. First, the Gaussian membership PM NS ZE PS PM PM PM PB
function requires a lower number of parameters PB ZE PS PS PM PB PB PB
in comparison with other membership functions.
Secondly, the proposed controller generates By combining, [e, ∆e] with the control action u
membership functions based on traditional and obtained in Table 1, a control surface is generated
an improved fuzzy c-means clustering, and the as it is illustrated in Figure 4.
shape of the Gaussian function features a higher
suitability or proximity to the FCM-like clustering
algorithm. Two MFs are generated for each subset
of antecedents and the consequence. A detailed
step-by-step description of the membership
functions` generation is provided below.

3.1.1 Generation of Control Surface

The control surface consists of error (e), change
of error (∆e), and control action (u). The MFs of
e, ∆e and u are defined in the common interval of
[-1, 1] as it is shown in Figure 3. This is one of
the unbiased & natural ways of MF distribution. Figure 4. The generated control surface
The input space (e, ∆e) is uniformly sampled in
1,000 points in the interval of [-1, 1].

ICI Bucharest © Copyright 2012-2023. All rights reserved

92 Anup Kumar Mallick, Achintya Das

3.1.2 Clustering of the Control Surface data points coming from each cluster. After the
segregation of core and boundary data points,
Next, the control surface is clustered into c number the class labels of the boundary data points are
of clusters. As each process input e and ∆e can be predicted by a k-nn classifier based on the class
divided into 3 major groups, i.e. negative, zero, labels of the core data points. In this paper, the
& positive, c is selected as 3*3=9 to cover the value of k in the k-nn classifier is set to 5. The final
control surface significantly. The control surface cluster centers are found by averaging the data
is first clustered into 9 groups by traditional fuzzy points belonging to that cluster. The cluster centers
c-means clustering. It provides one set of means obtained by traditional FCM and by Clusterwise
for the Gaussian membership functions. A second mean provide two sets of means for the primary
set of cluster centers is obtained by using a newly Gaussian membership functions.
developed supervised clustering algorithm. The
details on supervised clustering can be found in 3.1.3 Selection of the Standard Deviations
(Bandyopadhyay et al., 2007; Mukhopadhyay & of the Primary MFs
Maulik, 2009). The two major steps of supervised
clustering are to first segregate the core and The standard deviations of the Gaussian
boundary data points, and then to apply supervised membership functions are optimally selected
learning in clustering. From the membership by differential evolution (DE). Here, each
matrix of FCM, the membership values of the chromosome consists of 27 genes (9 genes for the
data points for all the clusters are found. Among standard deviation of the e component, the next 9
the membership values of a data point for all the genes for the ∆e component, and the last 9 genes
clusters, let us assume that the highest membership for the u component). In each run, DE optimizes
of a data point (xi) is represented by µhi and the the value of standard deviations by minimizing the
second highest membership is denoted by µnhi, sum of squared errors (SSE) of the data points of
and the difference between these two membership the control surface.
values is represented by ∆µi. Data points with a k1
higher ∆µi are assumed to be near a cluster center
and are designated as core data points. On the
= SSE ∑
i =1
(act (ui ) − est (ui )) 2 (6)
other hand, the data points with a lower ∆µi are
considered to be at the overlap of at least two Here, k1 is the number of data points, est(u) is the
clusters and are defined as boundary data points. estimated control action for a population with a
The membership-related threshold for deciding if certain set of standard deviations, and act(u)is the
a data point belongs either to the core data set or to output found from the control surface. The best
the boundary data set plays a significant role in the solution found at the last iteration of DE is translated
final cluster solution. Mallick & Mukhopadhyay as the standard deviations of the respective Gaussian
(2018) proposed a scheme, namely Clusterwise MFs. The pseudo-code of the differential evolution
mean to separate the core data points and boundary algorithm used here is given in Algorithm 1.
data points. In Clusterwise mean, the core data
3.1.4 Formation of UMF and LMF
set is formed by taking core data points from each
cluster. Hence, the membership-related threshold is As the control surface is clustered into 9 groups,
selected cluster-wise. In this method, the threshold each antecedent (e, ∆e), and consequence (u)
of a cluster is the mean value of membership consist of 9 membership functions. Two sets
differences ∆µi for all the data points belonging to of primary Gaussian MFs for the antecedents
that cluster only, and is determined by equation (5). and the consequence are formed by considering
1 cluster centers as the means, and the solution of
∆µthreshold ( j ) = ∑ ∆µ p (5) differential evolution as the standard deviations.
n j p∈C j
These two Gaussian membership functions shall
Here, Cj denotes the jth cluster, nj is the number be defined as GMF1 and GMF2. Then the upper
of data points in jth cluster, p is the index of data membership functions (UMFs) are generated by
points belonging to jth cluster, i is a set given using the fuzzy MAX operation over GMF1 and
by {1,2,3,…,n}, and n is the number of all data GMF2 as it is given in equation (7).
points. The data points with ∆µi ≥ ∆µthreshold of UMF = max ∀i µGMF 1 ( x(i )), µGMF 2 ( x(i ))
that cluster are regarded as core data. The overall (7)
core data set is formed by amalgamating the core for ∀x ∈ X

https://www.sic.ici.ro
 Design of an Improved Interval Type-2 Controller Using FCM and Supervised Clustering Algorithms 93

Algorithm 1: Pseudocode to generate standard

deviations of the Gaussian MFs
Require: Control surface data, popsize, Maxiter, means
of the Gaussian MFs
Ensure: Standard deviations of the Gaussian MFs
Randomly generate initial population
1
xi,G; i =1, 2, ..., popsize
2 Evaluate cost f (xi,G) using procedure SSE
3 Find the best population pbest of the population xi,G
4 G←0
5 while G < Maxiter do
6 G←G+1
7 for i = 1 to popsize do
Select randomly r1 and r2 such that
8
i ≠ r1 ≠ r2
9 Generate F from normal distribution
vi ,G +1 = xi ,G
10
+ F ∗ ( pbest − xi ,G + xr1,G − xr 2,G )
11 Calculate cost f(vi,G+1)
12 if f (vi,G+1) < f (xi,G) then
13 xi,G+1 = vi,G+1
14 else xi,G+1 = xi,G
15 end
16 end
17 Find the pbest of the population xi,G+1
18 end
19 Procedure SSE(f)
Formulate UMF&LMF using Means and
20 translating genes of the chromosome as the
standard deviations
Evaluate the estimated outputs (u) for all the
21
input data points (e, ∆e) of the control surface
k1

∑ ( act (u ) − est ( ui ) ) ;
2
Calculate f
= i
22
i =1
k1: no. of data points on the control surface
23 Return f
24 End

Figure 5. Membership functions (MFs) of the proposed IT2 FLC: UMFs of (a) error (e); (b) change of error
(∆e) and (c) control action (u); LMFs of (d) error (e); (e) change of error (∆e) and (f) control action (u)

ICI Bucharest © Copyright 2012-2023. All rights reserved

94 Anup Kumar Mallick, Achintya Das

Similarly, lower membership functions (LMFs) Then,

are obtained by taking the minimum of GMF1
r n = max( LMF 1 ,..., LMF k , ..., LMF N ), (13)
and GMF2 as in equation (8).
and
LMF = min ∀i µGMF 1 ( x(i )), µGMF 2 ( x(i ))
(8)
for ∀x ∈ X r n = max(UMF 1 ,..., UMF k ,..., UMF N ), (14)
The upper and lower MFs obtained for the where N is the total number of rules. Next, the
antecedents and the consequence of the proposed fuzzy output is defuzzified. In this paper, the
IT2 FLC are shown in Figure 5. Nie-Tan method of defuzzification (Nie & Tan,
2008; Wu, 2012) was used. In this method, the
3.2. Fuzzy Inference System crisp output y of an IT2 FLC is estimated by the
following in equation (15).
Let us assume that the process input at an instant
P
be denoted by ein and ∆ein. The membership
intervals of e in are computed by means of ∑ y (r n
n
+ r n)
projection of ein to the e component of all the y= n =1
P
, (15)
9 rules. Similarly, the membership intervals of ∑ (r n
+r ) n

∆ein are also calculated. Then the firing interval n =1

[ f k , f k ] at kth rule is estimated by minimum where P is the number of output discrete points.
t-norm. In Example 1, the procedure of the fuzzy
inference system is illustrated. 4. The Applied Process
Example 1. Let the projections of ein on the kth The armature-controlled DC motor is widely
LMF and kth UMF be expressed by fL1 and fU 1 , used in many industrial applications. Hence,
respectively, and the positions of ∆ein on the kth researchers still use DC motors as a process for
LMF and kth UMF be expressed by fL 2 and fU 2 , validating their proposed controller (Sankeshwari
respectively. Then, the firing strength of LMF of & Chillé, 2019; Li et al., 2021). The block diagram
kth rule denoted by f k is calculated as it is given of the armature-controlled DC motor is shown in
in equation (9). Figure 6 and the transfer function of the armature-
f k = min( fL1, fL 2) (9) controlled DC motor is given by equation (16).
θ m (s) KT
= (16)
Similarly, the firing strength of UMF of k rule th Ea ( s ) JLa s + ( Ra J + La B ) s 2 + ( Ra B + K b KT ) s
3

expressed by f k is estimated by equation (10).

f k = min( fU 1, fU 2) (10)

Next, the inferences due to firing of kth rule

are obtained from the intersection of f k with
LMFk (u ) , and from the intersection of f k with
UMFk (u ) . Let the consequence fuzzy output of
LMF and UMF of kth rule be denoted by LMFk and
UMFk, respectively. Thus,
LMF k = min( f k , LMFk (u )), (11) Figure 6. Block diagram of an armature- controlled
DC motor
and
UMF k = min( f k , UMFk (u )). (12) Here, θm: angular position of the motor shaft, ea:
applied voltage, KT: torque constant of motor,
3.3. Defuzzification J: the moment of inertia, Kb: voltage constant
of the motor, B: viscous friction coefficient, La:
The total output of LMF (and UMF) for a discrete armature inductance, and Ra: armature resistance.
output point is obtained by MAX aggression of all In an armature-controlled DC motor, the armature
the rules. Let the output LMF and UMF at nth data inductance is negligibly small compared to
point be denoted by r n and r n , respectively. armature resistance (Mandal, 2014). Thus,

https://www.sic.ici.ro
 Design of an Improved Interval Type-2 Controller Using FCM and Supervised Clustering Algorithms 95

the transfer function in equation (16) can be

approximated as equation (17).
θ m (s) KT / Ra J
= 2
Ea ( s ) s + ( B / J + K b KT / Ra J ) s (17)
5. Results and Discussion Figure 8. Matlab simulation block diagram

For uniformity purposes, all the simulations The value of Ra is set at 0.4 ohm, that of J at 0.5
were carried out by a fourth-order Runge-Kutta kg-m2, that of Kb at 1.25 V-s/rad, the value of KT at
solver with a fixed interval of 0.01 seconds. In 1 N-m/A, and that of B at 0.01N-m/rad (Rahman,
order to illustrate the performance of the proposed 2017). The simulation results for the step response
model, a conventional IT2 FLC is considered for with the conventional IT2 controller and with the
comparison purposes. The membership functions proposed controller are shown in Figure 9(a) and
of the conventional IT2 FLC are considered to in Figure 9(b), respectively. From Figure 9, it can
be symmetrical triangles (except for the two MFs be seen that the proposed IT2 controller has a
at the extreme ends) as it is shown in Figure 7. faster response with a lower overshoot, and also
The simulation is done in Matlab R2016a for the
settles earlier in comparison with the conventional
block diagram shown in Figure 8. For an unbiased
IT2 controller. The robustness of the proposed
comparison, the values of all the gain parameters,
controller is checked in two ways; by applying
namely Ge, G∆e, and GPD in Figure 8, were
optimized using differential evolution (DE) for a random noise, and by adding a sudden impulse
both the proposed model and for the conventional disturbance. The mean and the variance of the
fuzzy logic controller. In DE, Integral square error random noise are 0 and 0.1, respectively, and the
(ISE) is considered as the objective function given applied disturbance is of 30-unit impulse.
in equation (18).
nL
ISE = ∑ e (i),
i =1
2
(18)

where e(i) is the error, i.e. the difference between

the desired step response and the actual step
response at the ith discrete point for a set of [Ge,
G∆e, GPD]; n is the number of discrete data points,
and L is the sampling time, i.e. 0.01 seconds in this
paper. Here, n is considered to be 3000, as after
3 seconds the fluctuation in the response reduces
sufficiently. The algorithm of differential evolution
used here for the optimized selections of the gain
parameters is similar to Algorithm 1, except that
the cost function is calculated using equation (18).

Figure 9. Step responses of the DC motor in the

no noise and no disturbance conditions for (a) the
conventional IT2 controller; (b) the proposed
IT2 controller

Figure 7. MFs of conventional IT2 controller; solid

lines: UMFs, dashed lines: LMF of e/ ∆e/ u

ICI Bucharest © Copyright 2012-2023. All rights reserved

96 Anup Kumar Mallick, Achintya Das

The simulation results for the step response in the From Figure 10, it appears that the fluctuation in
presence of the noise and disturbance are shown response due to the random noise for the proposed
in Figure 10 and in Figure 11, respectively. controller is much lower than for the conventional
controller. Figure 11 indicates that the proposed
controller is more aggressive than the conventional
controller at the event of disturbance. During the
simulation, it was found that the conventional
controller required 0.54 seconds, whereas the
proposed controller takes only 0.06 seconds
to settle after applying the disturbance. It was
also seen that the proposed controller features a
lower overshoot (2.81%) in comparison with an
overshoot of 9.89% for the conventional controller
due to the disturbance.

The values of different time domain performance

indices, namely rise time (tr), peak overshoot (Mp),
settling time (tss), integral absolute Error (IAE),
integral time absolute error (ITAE), integral square
error (ISE), and integral time squared error (ITSE)
are included in Table 2.
Table 2. Comparative performance of conventional
and proposed IT2 controller
IT2 Controller used
Disturbance

Figure 10. Step responses of the DC motor in the

Noise/

Time domain
presence of random noise with a variance of 0.1 for performance index Conventional Proposed
(a) the conventional IT2 controller; (b) the proposed
IT2 controller
tr (sec.) 1.22 0.49
Mp% 4.59624 4.29711
disturbance applied
No noise and no

tss (sec.) 1.44 0.63

IAE 0.68807 0.28447
ITAE 0.29912 0.08630
ISE 0.49016 0.19074
ITSE 0.15704 0.02434
tr (sec.) 1.21 0.49
Mp% 4.80191 4.28470
Random noise with a
Variance of 0.1

tss (sec.) 1.42 0.63

IAE 0.68119 0.28314
ITAE 0.30557 0.08655
ISE 0.48081 0.18938
ITSE 0.15218 0.02408
tr (sec.) 1.22 0.49
Unit impulse disturbance
applied at 2.5 seconds

Mp% 4.59624 4.29711

tss (sec.) 1.44 0.63
IAE 0.72002 0.28640
ITAE 0.38856 0.09130
ISE 0.49220 0.19079

Figure 11. Step responses of the DC motor with 30- ITSE 0.16255 0.02445
unit impulse disturbance applied at 2.5 seconds for
(a) the conventional IT2 controller; (b) the proposed From Table 2, it appears that the proposed
IT2 controller IT2 controller outperforms the conventional

https://www.sic.ici.ro
 Design of an Improved Interval Type-2 Controller Using FCM and Supervised Clustering Algorithms 97

IT2 controller for all the performance indices the conventional IT2 controller to only 9 for
considered here. the proposed IT2 controller. From the obtained
simulation results, it can be seen that the
6. Conclusion and Future Scope proposed controller outperforms the conventional
IT2 controller, and that it also shows robustness
This paper prescribes a systematic approach to the noise and the applied disturbance.
to modelling an interval type-2 fuzzy logic
controller. It proposes a way to generate The present work considers the shape of primary
primary type-1 membership functions in order membership functions as being Gaussian. Further
to formulate the upper and the lower membership research may also be conducted with the purpose
functions for the interval type-2 controller. The of generating an IT2 controller with triangular or
number of rules is also reduced from 49 for trapezoidal membership functions.

REFERENCES
Bandyopadhyay, S., Mukhopadhyay, A. & Maulik, Machine Studies. 8(6), 687-697. doi: 10.1016/S0020-
U. (2007) An improved algorithm for clustering gene 7373(76)80030-2.
expression data. Bioinformatics. 23(21), 2859–2865.
doi: 10.1093/bioinformatics/btm418. Maldonado, Y. & Castillo, O. (2012) Genetic design
of an interval type-2 fuzzy controller for velocity
Behera, A., Adhya, D. G. & Parida, S. K. (2017) regulation in a DC motor. International Journal of
Analysis of robust interval type-2 fuzzy PID Advanced Robotic Systems. 9, 1-8. doi: 10.5772/51188.
controller for AGC in a restructured power system.
In: 7th International Conference on Power Systems, Mallick, A. K. & Mukhopadhyay, A. (2018) Different
December 21-23, Pune, India. pp. 168-174. Schemes for Improving Fuzzy Clustering Through
Supervised Learning. In: 2nd International Conference
Chakraborty, S., Konar, A., Ralescu, A. & Pal, N. on Computational Intelligence, Communications, and
R. (2015) A fast algorithm to compute precise type- Business Analytics, July 27-28, 2018, Kalyani, India.
2 centroids for real-time control applications. IEEE pp. 155-164.
Transactions on Cybernetics. 45(2), 340-353. doi:
10.1109/TCYB.2014.2308631. Mandal, A. K. (2014) Introduction to Control
Engineering: Modeling, Analysis and Design. India,
Chiu, C. H. & Hung, Y. T. (2020) One wheel vehicle New Age International Publishers.
real world control based on interval type 2 fuzzy
controller. Mechatronics. 70, 102387. doi: 10.1016/j. Mendel J. M. & John R. I. (2002) Type-2 fuzzy sets
mechatronics.2020.102387. made simple. IEEE Transactions on Fuzzy Systems.
10(2), 117-127.
Das, S. & Suganthan, P. N. (2011) Differential
evolution: A survey of the state-of-the-art. IEEE Mendel, J. M. & Wu, D. (2010). Perceptual
Transactions on Evolutionary Computation. 15(1), Computing: Aiding People in Making Subjective
4-31. doi: 10.1109/TEVC.2010.2059031. Judgments. Wiley-IEEE Press, New Jercy, USA.
ISBN: 0470478764
Hsiao, M.-Y., Li, T.-H. S., Lee, J.-Z., Chao, C.-H. &
Tsai, S.-H. (2008) Design of interval type-2 fuzzy Mukhopadhyay, A. & Maulik, U. (2009) Towards
sliding-mode controller. Information Sciences. 178(6), improving fuzzy clustering using support vector
1696-1716. doi: 10.1016/j.ins.2007.10.019. machine: Application to gene expression data. Pattern
Recognition. 42(11), 2744-2763. doi: 10.1016/j.
Li, J., Ma, K. & Wu, Z. (2021) Prescribed performance patcog.2009.04.018.
control for uncertain flexible-joint robotic manipulators
driven by DC motors. International Journal of Nafia, N., Kari, A. E., Hassan A. & Mjahed M. (2018)
Control, Automation and Systems. 19, 1640-1650. doi: Robust interval type-2 fuzzy sliding mode control
10.1007/s12555-020-0311-2. design for robot manipulators. Robotics. 7(3), 40. doi:
10.3390/robotics7030040.
Liang, Q. & Mendel, J. M. (2000) Interval type-
2 fuzzy logic systems: Theory and design. IEEE Nie, M. & Tan, W. W. (2008) Towards an efficient
Transactions on Fuzzy Systems. 8(5), 535-550. doi: type-reduction method for interval type-2 fuzzy logic
10.1109/91.873577. systems. In: 2008 IEEE International Conference on
Fuzzy Systems, June 1-6, 2008, Hong Kong, China.
MacVicar-Whelan, P. J. (1976) Fuzzy sets for man- IEEE. pp. 1425-1432.
machine interaction. International Journal of Man-

ICI Bucharest © Copyright 2012-2023. All rights reserved

98 Anup Kumar Mallick, Achintya Das

Rahman, Z-A. S.A. (2017) Design of a fuzzy logic Wu, D. & Tan, W. (2004) A type-2 fuzzy logic
controller for controlling position of D.C. motor. controller for the liquid-level process. In: 2004 IEEE
International Journal of Computer Engineering In International Conference on Fuzzy Systems, July 25-
Research Trends. 4(7), 285-289. 29, 2004, Budapest, Hungary. IEEE. pp. 953-958.

Sankeshwari, S. S. & Chillé, R. (2019) Performance Zadeh, L. (1975) The concept of a linguistic variable
analysis of disturbance estimation techniques for and its application to approximate reasoning-I.
robust position control of DC motor. International Information Sciences. 8(3), 199-249. doi:
Journal of Control, Automation and Systems. 18, 486- 10.1016/0020-0255(75)90036-5.
494. doi: 10.1007/s12555-018-0838-7.
Zakovorotniy, A. & Kharchenko, A. (2021) Optimal
Storn, R. (2008) Differential Evolution Research – speed controller design with interval type-2 fuzzy sets.
Trends and Open Questions. In: Chakraborty, U.K. In: IEEE 2nd KhPI Week on Advanced Technology
(ed.) Advances in Differential Evolution. Studies (KhPIWeek), September 13-17, 2021, Kharkiv,
in Computational Intelligence. Springer, Berlin, Ukraine. IEEE. pp. 363-366.
Heidelberg. vol. 143, pp. 1-31.
Zeghlache, S., Benslimane T., Amardjia, N. &
Şahin, İ. & Ulu, C. (2023) Altitude control of a Bouguerra, A. (2017) Interval type-2 fuzzy sliding
quadcopter using interval type-2 fuzzy controller with mode controller based on nonlinear observer for a
dynamic footprint of uncertainty. ISA Transactions. 3-DOF helicopter with uncertainties. International
134, 86-94. doi: 10.1016/j.isatra.2022.08.020. Journal of Fuzzy Systems. 19, 1444-1463. doi:
10.1007/s40815-016-0226-5.
Wu, D. (2012) An overview of alternative type-
reduction approaches for reducing the computational Zhou, H., Zhang, C., Tan, S., Dai, Y. & Duan, J.
cost of interval type-2 fuzzy logic controllers. In: 2012 (2021) Design of the footprints of uncertainty for
IEEE International Conference on Fuzzy Systems, a class of typical interval type-2 fuzzy PI and PD
June 10-15, 2012, Brisbane, QLD, Australia. IEEE. controllers. ISA Transactions. 108, 1-9. doi: 10.1016/j.
pp. 1-8. isatra.2020.08.009.

Wu, D. & Mendel, J. M. (2019) Recommendations

on designing practical interval type-2 fuzzy systems.
Engineering Applications of Artificial Intelligence. 85,
182-193. doi: 10.1016/j.engappai.2019.06.012.

https://www.sic.ici.ro