ISA Transactions 53 (2014) 15831591

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

A controller based on Optimal Type-2 Fuzzy Logic: Systematic design,

optimization and real-time implementation
H.M. Fayek a, I. Elamvazuthi a,n, N. Perumal a, B. Venkatesh b
a
b

Department of Electrical and Electronic Engineering, Universiti Teknologi PETRONAS, 31750 Tronoh, Perak, Malaysia
Department of Electrical and Computer Engineering, Ryerson University, Toronto, Ontario, Canada M5B 2K3

art ic l e i nf o

a b s t r a c t

Article history:
Received 11 April 2014
Received in revised form
21 May 2014
Accepted 2 June 2014
Available online 21 June 2014
This paper was recommended for
publication by Dr. Jeff Pieper

A computationally-efcient systematic procedure to design an Optimal Type-2 Fuzzy Logic Controller

(OT2FLC) is proposed. The main scheme is to optimize the gains of the controller using Particle Swarm
Optimization (PSO), then optimize only two parameters per type-2 membership function using Genetic
Algorithm (GA). The proposed OT2FLC was implemented in real-time to control the position of a DC
servomotor, which is part of a robotic arm. The performance judgments were carried out based on the
Integral Absolute Error (IAE), as well as the computational cost. Various type-2 defuzzication methods
were investigated in real-time. A comparative analysis with an Optimal Type-1 Fuzzy Logic Controller
(OT1FLC) and a PI controller, demonstrated OT2FLC's superiority; which is evident in handling
uncertainty and imprecision induced in the system by means of noise and disturbances.
& 2014 ISA. Published by Elsevier Ltd. All rights reserved.

Keywords:
Intelligent control
Type-2 Fuzzy Logic
Fuzzy Logic Control
Optimization
Robotic arm

1. Introduction
Fuzzy Logic Control (FLC) has demonstrated superiority over
classical control, namely Proportional-Integral-Derivate (PID) controllers, especially in applications where imprecision and uncertainty are present in the system [1]. Generally, FLC has proven to be
more superior in terms of (1) noise rejection, (2) exibility, (3) the
use of human knowledge, not accurate mathematical models, and
(4) sensitivity to disturbances, which nally yields (5) overall better
system performance [2]. Type-2 Fuzzy Logic (T2FL) was introduced
to generalize Type-1 Fuzzy Logic (T1FL) [34]. T2FL is able to model
uncertainty and imprecision in a much better way [5], which makes
Type-2 Fuzzy Logic Control (T2FLC) ideal for control applications.
However, T2FL is more difcult to understand and implement than
the conventional T1FL [6]. The lack of systematic design procedures
for T2FL controllers, as well as T1FL controllers, has been a
challenge to researchers and engineers and is considered as one
of the drawbacks of T2FLC and FLC.
Various designs of T2FL controllers were reported in the
literature, most of which were designed based on heuristic
methods. For example, in [7], a T2FL controller was designed to

n
Corresponding author. Department of Electrical and Electronic Engineering,
Universiti Teknologi PETRONAS, Block 23, 31750 Tronoh, Perak, Malaysia.
Tel.: 60 53687882.
E-mail addresses: haytham.fayek@ieee.org (H.M. Fayek),
irraivan_elamvazuthi@petronas.com.my (I. Elamvazuthi).

http://dx.doi.org/10.1016/j.isatra.2014.06.001
0019-0578/& 2014 ISA. Published by Elsevier Ltd. All rights reserved.

control the speed of a DC motor and judged it against a T1FL

controller, demonstrating that the T2FL controller had better
performance. In [8], the design of an adaptive type-2 fuzzy-neuro
system for controlling the position of a servo system with an
intelligent sensor was presented. A comparative analysis between
different types and numbers of type-2 membership functions and
their impact on a T2FL controller's performance to control a
servomotor was carried out in [9].
A systemic design procedure of a stable T2FL controller was
presented in [10] based on fuzzy Lyapunov synthesis [11], where
only the rules of the T2FL controller were considered in the
systematic design process. Optimization algorithms such as
Genetic Algorithm (GA) or Particle Swarm Optimization (PSO)
are able to provide a systemic design procedure as reported in
[12]. This optimization process is conducted off-line using a
mathematical model of the desired plant, which may result in
unsatisfactory results due to the differences between the mathematical model and the actual plant. This problem is dealt with in
case of T2FL, since T2FL is able to deal with such imprecision
outstandingly. However, optimization of T2FL introduces a new
problem, which is the lengthy and computationally expensive
optimization process due to the large number of points that need
to be considered in the optimization process for each membership
function as reported in [13], where the authors used GA to design
the membership functions of a T2FL controller, and stated that the
computation time needed for the optimization process was too
lengthy. This problem was tackled in [14] by assigning xed values

1584

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

for some points of the membership functions and optimizing the

remaining points using PSO or GA without degrading the performance of the controller.
A few papers presented the implementation of a T2FL controller
in real-time. In [15], a non-optimal T2FL controller was designed and
implemented to control an autonomous mobile robot demonstrating
better results than the T1FL counterpart under similar conditions. In
[16], the design using heuristic methods and real-time implementation of a T2FL controller with different defuzzication methods
to control a DC Servomotor was presented and the computation time
was recorded as well. It was demonstrated that although T2FL was
computationally more expensive, it yielded improved handling of
noise and system disturbances than the T1FL controller. In [14], the
membership functions of a T2FL controller were optimized and the
controller was implemented on a Field-Programmable Gate Array
(FPGA) to control a motor with backlash, demonstrating better
results in case of the T1FL controller as well.
This article presents a systematic design procedure of an
Optimal Type-2 Fuzzy Logic Controller (OT2FLC), where both the
gains and membership functions of the OT2FLC are considered in
the optimization process. PSO was used to optimize the gains of
the inputs and output of the controller. While, GA was used to
optimize the membership functions by considering only two
parameters for each type-2 membership function to decrease the
computational complexity of the optimization process, while
ensuring robustness of the optimal controller. The proposed
controller is implemented in real-time to control the position of
an actual DC servomotor, investigating various defuzzication
methods. The performance and computation time of the OT2FLC
are compared against an Optimal Type-1 Fuzzy Logic Controller
(OT1FLC) and an optimal Proportional-Integral (PI) controller.
The remainder of this article is organized as follows: In Section 2,
type-2 fuzzy sets and systems theories are presented, as well as a
brief review on PSO and GA. The modeling, design methodology,
optimization and implementation of the OT2FLC to control the
position of the DC servomotor are explicated in Section 3. Section 4
depicts the real-time performance of the proposed OT2FLC and
discusses the results against the OT1FLC and the conventional PI
controller. The article is concluded in Section 5.

2. Proposed techniques
2.1. Type-2 Fuzzy Logic
Zadeh introduced Type-2 Fuzzy Sets (T2FS) as an extension to
Type-1 Fuzzy Sets (T1FS) [17,18]. T2FS are able to handle uncertainty in a much better way, which make T2FS ideal for control
applications. A T2FS, is characterized by a type-2 membership
function A~ z; z; where zA Z and A J z D 0; 1; as follows:
A~ fz; z; A~ z; zj 8 z A Z; 8 z A J z D 0; 1g

where, 0 r A~ z; z r 1, and Z is the universe of discourse.

may also be represented as follows [19]:
Z
Z
A~ z; z
2
A~
z; z
z A Z z A J z

RR
where, J z D 0; 1,
resembles union over all admissible z and
(z) [20] and Jz is called primary membership of z. Concretely, a
type-2 membership function comprises an inferior membership
function and a superior membership function; each function is
represented by a type-1 membership function. The superior
membership function is denoted as Upper Membership Function
(UMF), while the inferior membership function is denoted as
Lower Membership Function (LMF) [21] as in Fig. 1. The interval

Fig. 1. Type-2 fuzzy membership function.

between these two membership functions represents the Footprint Of Uncertainty (FOU), which distinguishes the T2FS [22].
An Interval Type-2 Fuzzy Set (IT2FS) is the most widely used
type of the T2FS [19]. It can be considered as a special case of T2FS,
whereas an IT2FS is one in which the membership grade of every
domain point is a crisp set whose domain is some interval
contained in the interval [0,1]. The membership grade of an IT2FS
is an interval set with a unity value for each secondary grade in
that set [23]. The IT2FS was used in this work.
A Type-2 Fuzzy Inference System (T2FIS) has the same IF-THEN
rules as the conventional type-1 Fuzzy Inference System (FIS) except
that the antecedent and consequent are in type-2 form as follows:
l
l
Rl : IF x1 is V~ 1 AND xn is V~ n

~ l;
THEN y is W

l 1; ; M

l
~ l is a
where, V~ n is a type-2 antecedent, yAY is the output, and W
type-2 consequent. The structure of a T2FL system, which is shown in
Fig. 2, is very similar to a T1FL system. A T2FL system comprises a
type-2 fuzzier, a type-2 rule-base, a type-2 inference engine, and
substitutes the type-1 defuzzier with an output processor which
includes a type reducer and a type-2 defuzzier [24].
The Extension Principle [25] is used to extend each type-1
defuzzication method for the corresponding type-reduced set.
A type-reducer combines the output sets in some way and then
performs a centroid calculation on this T2FS, which leads to a T1FS
that is called the type-reduced set. Type-2 defuzzication methods
used in this paper are [25,26]

 Centroid: The centroid type-reducer combines all the ruleoutput T2FS by nding their union as in Eq. (4).

B~ y [ M
8yAY
l 1 B~ l y

where, ~ l is the secondary membership function for the lth

B
rule. Finding the union of T2FS requires computing the join of
their secondary membership functions. This method involves
an enormous amount of computation, as the centroid and
membership computations have to be repeated numerous
times. Eq. (5) is used by the centroid type-reducer to calculate
~
the centroid of B:
,
Z
Z
N
i 1 yi i
Y c x

f y1 1 f yN N 
5
N

1 A J y1
N A J yN
i 1 i
where, i1,, N, and N,Jyi,fyi are associated with B~ y.
 Center of sums: The center of sums type-reducer combines the
type-2 rule output sets by adding their secondary membership

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

1585

Fig. 2. Type-2 fuzzy logic system.

functions by using Eq. (6) and then nds the centroid of the
resulting set using Eq. (5).
M

B~ y B~ l y 8 y A Y

l1

 Height: The height type-reducer replaces each type-2 output

set by a T2FS whose y-domain consists of a single point y, the
secondary membership function of which is a type-1 fuzzy set
as in Eq. (7). This single point is chosen to be the point having
the highest primary membership in the principal membership
function of the output set.
,
Z
Z
1
M
l 1 y l
Y h x

f y1 1 f yM M 
7
M
l 1 y1
1 A J y1
M A J yM
where, yl is the point having maximum membership in the lth
output set and l ; J yl ; f yl are associated with ~ l yl . This
B

method requires the least computational complexity [25].

Center of sets: The center of sets type-reducer replaces each
l
type-2 consequent set, G~ by its centroid, C l and nds a
G~

weighted average of these centroids using Eq. (8).

Z
Z
Z
Z
Y cos x

TM
l 1 C l dl
d1 A C ~ l
G

dM A C ~ M

T M
l 1 E l el

,G

e1 A E1

M
l 1 dl el
M
e
l1 l

eM A EM

G~

separate random numbers being generated for acceleration

towards pbest and gbest locations [28].
The velocity of particle i is calculated as follows:
vij vij t c1 r 1j tyij t  xij t c2 r 2j ty^ ij t  xij t

where, vij is the velocity of particle i in dimension j1, , nx at time t,

xij(t) is the position of particle i in dimension j at time t, c1 and c2 are
the cognitive and social components respectively, r1j(t) and r2j(t) are
random numbers between 0 and 1, yij(t) is the particle's personal best
position and y^ ij t is the global best position.
The position of the particle, xi(t), is updated by adding velocity
vi(t) to the current position as in Eq. (10).
xi t 1 xi t vi t 1

10

Several modications have been reported since the introduction of this algorithm such as swarm regeneration, velocity
clamping, constriction factor, etc. The addition of a linearly
decreasing inertia weight to Eq. (9), as shown in Eq. (11), to ensure
convergent behavior and to optimally tradeoff exploration and
exploitation has proven to be a signicant modication [29] to the
original algorithm and is used in this work.
vij wtvij t c1 r 1j tyij t  xij t c2 r 2j t y^ ij t  xij t

11

2.3. Genetic Algorithm

8

where, T and indicate the chosen t-norm and El is a type1 set.

2.2. Particle Swarm Optimization

PSO is a population based stochastic optimization technique
inspired by social behavior of bird ocking or sh ocking [27].
PSO optimizes the problem by iteratively trying to improve a
candidate solution with regards to a given measure of quality. The
system initially has a population of random solutions. Each
potential solution, denoted particle, is given a random velocity
and is own through the search space. The particles have memory
and each particle keeps track of previous best position and
corresponding tness. The previous best value is called pbest of
the particle. The best value of all the particles' pbest in the swarm
is called gbest. The basic concept of this technique lies in changing
the velocity of (accelerating) each particle towards its pbest and
gbest locations. Acceleration is weighted by a random term, with

GA is a heuristic search algorithm, inspired by the evolutionary

ideas of natural selection and genetics [30]. The main idea behind this
algorithm is based on the natural evolution of biological creatures,
where the ttest among a group of articial entities will likely survive
to form a new generation together with those which are produced
through gene exchange. The main GA operators are selection, crossover, and mutation. Selection is used to choose the best individuals in
a population, crossover produces new individuals by mixing couples
of selected individuals and mutation induces random changes in the
individuals. In GAs, exploitation of the search space is carried out by
the selection and the crossover operators, while exploration is carried
out by the mutation operator [3133].

3. Materials and methods

3.1. Proposed systematic design procedure and optimization
The OT2FLC has two inputs; which are the error and integral of
error and one output as depicted in Fig. 3. A PI-like structure was
selected to eliminate any steady state error which may result in
case of the Proportional Derivative (PD)-like structure, due to lack

1586

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

Fig. 3. Proposed Optimal Type-2 Fuzzy Logic Controller (OT2FLC).

Table 1
Rule-base.
E/DE
BN
MN
N
Z
P
MP
BP

BN
BN
BN
BN
BN
MN
N
Z

MN
BN
MN
MN
MN
N
Z
P

N
BN
MN
N
N
Z
P
P

Z
MN
MN
N
Z
P
MP
MP

P
N
N
Z
P
P
MP
BP

MP
N
Z
P
MP
MP
MP
BP

BP
Z
P
MP
BP
BP
BP
BP

taking into account the number of membership function associated to each input or output and the number of inputs and
outputs of the controller, which will result in an excessively large
total number of points. The authors propose to represent the
type-2 membership function using only four parameters denoted
as a, b, c, and as shown in Fig. 5b, where represents the FOU.
Eq. (12) denes the relationship between a1, a2, b1, b2, c1, c2 and
a, b, c, .
P 1 P  j P 2 P j 8 fP; P 1 ; P 2 g fa; a1 ; a2 ; b; b1 ; b2 ; c; c1 ; c2 g

12
Only two parameters are considered in the optimization
process for each membership function, which are the apex of the
membership function and the FOU, denoted as b and respectively. The bases of the membership function, denoted as a and c
are xed and distributed equally over the unity universe of
discourse to ensure robustness of the optimal controller over a
wider range of operating conditions and reduce the number of
parameters considered in this optimization problem. The apex of
the middle membership function (denoted as Z in Table 1) is
excluded from the optimization problem and is xed to zero to
minimize any steady state error. Two inequality constraints were
enforced during the optimization process as shown in Eq. (13)
Fig. 4. Seven type-2 triangular membership functions.

1 o a o bo c o 1
of integral function in PD's control nature. All inputs and output
are associated with seven triangular-shaped membership functions over a unity universe of discourse of interval [  1, 1].
Mamdani method also known as MaxMin method was used.
The rule-base is based on the MacVicar-Whelan rule-base [34] as
listed in Table 1, where, BN is Big Negative, MN is Medium
Negative, N is Negative, Z is Zero, P is Positive, MP is Medium
Positive, and BP is Big Positive. The symmetry of the rule-base is to
ensure stability and robustness of the controller after the optimization process.
The proposed design procedure incorporates two stages. The
rst stage is optimizing the gains associated with the inputs (k1
and k2) and output (k3) of the controller. During this stage the
membership functions are xed and distributed equally over the
unity universe of discourse with an overlap of 50% between
adjacent membership functions as shown in Fig. 4, which is the
most natural and unbiased choice [7]. This optimization problem is
solved using linearly decreasing inertia weight PSO due to the
relative large area of the search space and its rough surface, where
PSO was reported to have excelled in the presence of such
conditions [27,35].
The second stage of the design process is the optimization of
the membership functions. Consider a type-2 membership function shown in Fig. 5a, where it can be dened using six points,
denoted as a1, a2, b1, b2, c1, and c2. Considering all six points in the
optimization problem of each membership function is very computationally expensive and may not result in an optimum solution

13

0 o o 0:2

Considering seven membership functions associated to each of

two inputs and an output, a total of 39 points and 39 inequality
constraints are considered in this optimization process. GA was
used to solve this optimization problem, due to the relatively high
number of points considered, as well as, GA's superiority over
other optimization algorithms in dealing with constraints present
in the search space [32,33].
The objective function used in both stages of the design process
is a multiple-objective function as shown in Eq. (14). The rst
objective (O1) is the Integral Absolute Error (IAE) between the
plant input and output computed using Eq. (15) and the second
objective (O2) is the IAE as well, however, with white noise added
to the feedback signal of the closed-loop system. The multiobjective function is to optimally trade-off performance and
sensitivity of the optimal controller during the optimization
process.
FE O1 0:5O2

Oi IAE jekj;

14

8 i 1; 2

15

k1

where, O1 is the rst objective, O2 is the second objective, n is the

number of points in the run, and e(k) is the error at kth point.

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

1587

Fig. 5. Type-2 membership representation.

Table 2
Specications of the Hitec Hs-311 servomotor.
Motor type
Bearing type
Speed (4.8 V/6.0 V)
Torque, kg/cm (4.8 V/6.0 V)

3 Pole
None
0.19/0.15 s (601)
3.0/3.7

Table 3
Optimal gains of the OT2FLC, OT1FLC and PI
controller.
Controller

Optimal gains

OT2FLC
OT1FLC
PI

k1 1.124, k2 1.627, k3 1.343

k1 1.71, k2 1.257, k3 1.722
P 2.888, I 2.534

actual recorded response and the estimated model's response.

s
46:14

u s s2 11:77s 44:38

Fig. 6. Servomotor's actual measured response vs theoretical estimated response.

3.2. System modeling and implementation

Effectiveness of the proposed OT2FLC was veried through
controlling a DC servomotor. The servomotor selected for this
work is a Hitec HS-311 analog DC servomotor, which is mainly
used in robotics and Radio-Controlled (RC) applications. The
selected servomotor is part of a six Degree of Freedom (DOF)
robotic arm. The specications of the servomotor are summarized
in Table 2 [36]. The servomotor suffers from a noisy feedback
signal, which is ideal to demonstrate the superiority of the
proposed controller.
The parameters of the servomotor were estimated [37] using
the Matlab System Identication Toolbox and the identied
transfer function was deduced as in Eq. (16). Fig. 6 depicts the

16

An OT2FLC controlled closed-loop system was modeled and the

T2FIS was built using Interval Type-2 Fuzzy Logic Toolbox [38,39].
The feedback signal is the angle of the servomotor, which is
compared to the desired input angle to compute the error and
its integral. Similarly, OT1FLC and PI controlled closed-loop systems were modeled as well for comparatives analysis purposes.
Firstly, the gains of the OT2FLC, OT1FLC and PI controller were
optimized using PSO as mentioned in Section 3.1. The parameters
of the PSO algorithms were chosen as follows, the cognitive (c1)
and social (c2) components were selected to be 1 and 1.05
respectively, while the inertia weight (w(t)) was linearly decreasing from 1 to 0.3. The swarm comprised 30 particles, initialized
randomly throughout the search space. The minimum number of
iterations was set to 30, subsequently the algorithm terminates
when two termination condition are satised, which are (1) the
swarm radius becomes less than 0.1 and (2) the average change in
the tness function value becomes less than 0.0001 or at the
maximum allowed number of iterations, the 100th iteration. The
PSO converged in 32 iterations and yielded the optimal gains listed
in Table 3.
Subsequently, the membership functions of the OT2FLC and
OT1FLC were optimized using GA as described in Section 3.1. The
GA had a population size of 200, initialized randomly within the
constraints. A ranking scaling function and a stochastic uniform
selection function were used. The crossover rate was 0.8 and the
elite count was 2. The minimum number of generations was set to
50, subsequently the algorithm may terminate when the average
change in the tness function value is less than 0.0001 or at the
maximum allowed number of iterations, the 100th iteration. The
optimal membership functions are depicted in Figs. 7 and 8 for
OT2FLC and OT1FLC respectively.
To test the effectiveness of the OT2FLC against the OT1FLC and
the PI controller in real-time applications, the Hitec HS-311 analog

1588

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

Fig. 7. Optimal type-2 membership functions.

Fig. 8. Optimal type-1 membership functions.

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

1589

Fig. 9. Block diagram of the system.

Fig. 10. Simulink block diagram.

Table 4
Performance of OT2FLC with various type-2 defuzzication methods.
Defuzzication method

Centroid
Center of sums
Height
Center of sets

Fig. 11. Photograph of the hardware setup.

DC servomotor was controlled using an Arduino Uno R3 microcontroller board, which is based on the ATmega328 chip. The
microcontroller was used to interface the servomotor with Matlab/
Simulink environment. Fig. 9 illustrates the block diagram of the
entire system, while; Fig. 10 depicts the complete Simulink block
diagram of the system. A photograph of the hardware setup is
shown in Fig. 11. The sampling rate was set to 0.1 s.

4. Results and discussion

Firstly, the performance of the proposed OT2FLC to control the
actual servomotor in real-time using different defuzzication
methods is assessed. Two criteria were monitored and used to
judge the defuzzication methods against each other, which are

IAE

Execution time (s)

Run 1

Run 2

Run 3

410.3
406.8
453.1
442.6

399.2
421.9
442.0
447.4

419.9
452.1
437.9
458.4

0.00703
0.00683
0.00553
0.00829

the IAE and the execution time. The execution time is the average
total time taken by the controller to compute the output at a single
point. The computer used in this work has an Intel i3 M380,
2.53 GHz processor and 3 GB of RAM. Table 4 depicts the IAE for
three runs, as well as the execution time for each defuzzication
method. The performance of the best run of each method is
plotted in Fig. 12.
It is conspicuous from the results obtained in Table 4 and Fig. 12
that the Centroid method yields the least error and the best
results. It has a fast response and relatively lower sensitivity to
noise compared to the other methods. However, the Height
method has the fastest execution time, which can be very useful
in high-speed applications that require small sampling time.
Subsequently, the performance of the OT1FLC, as well as the
performance of the PI controller to control the servomotor under
the same system conditions were recorded. Both the OT2FLC and
the OT1FLC use the Centroid defuzzication method. Table 5
depicts the IAE for three runs, as well as the execution time for
each controller. The performance of the best run of each controller
is plotted in Fig. 13.

1590

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

Table 6
Performance of OT2FLC, OT1FLC and PI controller with a heavy free hanging load.
Defuzzication method

OT2FLC
OT1FLC
PI

IAE
Run 1

Run 2

Run 3

455.4
516.9
594.7

462.6
497.9
575.1

461.8
524.9
577.2

Fig. 12. Performance of various type-2 defuzzication methods.

Table 5
Performance of OT2FLC, OT1FLC and PI controller.
Defuzzication method

OT2FLC
OT1FLC
PI

IAE

Execution time (s)

Run 1

Run 2

Run 3

410.3
515.9
512.6

399.2
536
573.1

419.9
499.7
552.6

0.00703
0.00293
0.00029
Fig. 14. Performance of OT2FLC, OT1FLC and PI controller with a heavy free
hanging load.

Ultimately, the robotic arm was much more robust and had a
smoother and more natural movement when the OT2FLC was
employed. The controller was able to suppress chattering movements, which are caused by noise. The performance improvements
that the OT2FLC offers, comes at the expense of a slower execution
time. With the current hardware advancements and intelligent
programming techniques, this issue should not be problematic.
The slow execution time did not cause any issues when running
the system in real-time at a sampling rate of 0.1 s. Furthermore, in
high speed applications where a much smaller sampling rate is
needed, selecting a faster defuzzication method or even decreasing the number of membership functions, which would decrease
the number of rules in the rule-base signicantly, will improve the
execution time.

5. Conclusion
Fig. 13. Performance of OT2FLC, OT1FLC and PI controller.

Furthermore, a free-hanging heavy load (300 g) was connected

to the servomotor's arm by a 10 cm thread to test the performance
of all three controllers under different system parameters without
retuning the controllers. The hanging load was also a source of
unexpected disturbance and uncertainty due to its free movement.
Table 6 depicts the IAE for three runs. The performance of the best
run of each controller is plotted in Fig. 14.
From results in Tables 5 and 6, it is clear that the OT2FLC has
better overall performance compared to OT1FLC and the PI
controller. This improvement is patent in terms of noise rejection,
sensitivity to disturbances, smaller overshoot, which nally yields
a much improved overall performance.

A computationally-efcient systematic design procedure for

Type-2 Fuzzy Logic Controllers using particle swarm optimization
and genetic algorithm was proposed. The proposed Optimal Type2 Fuzzy Logic Controller (OT2FLC) was implemented in real-time
to control the position of a DC servomotor, which is part of a
robotic arm.
The success and practicality of the proposed systematic design
procedure for Type-2 Fuzzy Logic Controllers was successfully
validated. Numerous type-2 defuzzication methods were tested
in real-time. The tests revealed the superiority of the Centroid
method over other defuzzication methods. Furthermore, OT2FLC
showed much improved performance over the OT1FLC, as well as
the PI controller. This improvement is evident when handling
uncertainty and imprecision induced in the system by means of
noise and sudden disturbances.

H.M. Fayek et al. / ISA Transactions 53 (2014) 15831591

Acknowledgment
The authors would like to express their gratitude to Universiti
Teknologi PETRONAS (UTP) for supporting this work. The authors
would also like to extend their appreciation to Professor Oscar
Castillo for his authorization to use Interval Type-2 Fuzzy Logic
Control Toolbox.
References
[1] Juang J-G, Liu W-K, Lin R-W. A hybrid intelligent controller for a twin rotor
MIMO system and its hardware implementation. ISA Trans 2011;50:60919.
[2] Khongkoom N, Kanchanathep A, Nopnakeepong S, Tanuthong S, Tunyasrirut S,
Kagwa R. Control of the position DC servo motor by fuzzy logic. TENCON
2000;3:3547.
[3] Zadeh LA. The concept of a linguistic variable and its application to approximate reasoning-I. Inf Sci 1975;8:199249.
[4] Zadeh LA. The concept of a linguistic variable and its application to approximate reasoning-II. Inf Sci 1975;8:30157.
[5] Mendel JM. Computing with words: Zadeh, Turing, Popper and Occam. IEEE
Comput Intell 2007;2:107.
[6] Liang Q, Mendel JM. Interval type-2 fuzzy logic systems: theory and design.
IEEE Trans Fuzzy Syst 2000;9:53550.
[7] Tushir M, Srivastava S. Type-2 fuzzy logic controller implementation for
tracking control of DC motor. Inl J Comput Netw Secur (IJCNS) 2008;3:3441.
[8] Kayacan E, Kaynak O, Abiyev R, Trresen J, Hvin M, Glette K. Design of an
adaptive interval type-2 fuzzy logic controller for the position control of a
servo motor with an intelligent sensor. In: Proceedings of IEEE world congress
on computational intelligence, FUZZ-IEEE, Barcelona, Spain; 2010.
[9] Fayek HM, Elamvazuthi I. Type-2 fuzzy logic PI (T2FLPI) based DC servomotor
control. J Appl Sci Res 2012;8:256474.
[10] Castillo O, Aguilar L, Cazarez N, Cardenas S. Systematic design of a stable type2 fuzzy logic controller. Appl Soft Comput 2008;8:12749.
[11] Margaliot M, Langholz G. New approaches to fuzzy modeling and control:
design and analysis. Singapore: World Scientic; 2000.
[12] Wu D, Tan WW. A simplied type-2 fuzzy logic controller for real-time
control. ISA Trans 2006;45:50316.
[13] Cazarez-Castro NR, Aguilar LT, Castillo O. Hybrid genetic-fuzzy optimization of
a type-2 fuzzy logic controller. In: Proceedings of the hybrid intelligent
systems, HIS '08; 2008. p. 21621.
[14] Maldonado Y, Castillo O, Melin P. Particle swarm optimization of interval type2 fuzzy systems for FPGA applications. Appl Soft Comput 2013;13:496508.
[15] Hagras HA. A hierarchical type-2 fuzzy logic control architecture for autonomous mobile robots. IEEE Trans Fuzzy Syst 2004;12:52439.
[16] Fayek HM, Elamvazuthi I. Real-time implementation of a type-2 fuzzy logic
controller to control a DC servomotor with different defuzzication methods.
In: Proceedings of the international conference on methods and models in
automation and robotics (MMAR); 2013. p. 8691.
[17] Zadeh LA. Fuzzy logic. Computer 1988;1:8393.
[18] Karnik NN, Mendel JM. Introduction to type-2 fuzzy logic systems, Fuzzy
systems proceedings, In: 1998 IEEE International Conference on computational intelligence. vol. 2, No. 49, May 1998, p. 915,920.

1591

[19] Cazarez-Castro NR, Aguilar LT, Castillo O. Designing type-1 and type-2 fuzzy
logic controllers via fuzzy lyapunov synthesis for nonsmooth mechanical
systems. Eng Appl Artif Intell 2012;25:9719.
[20] Mendel JM, John R. Type-2 fuzzy sets made simple. IEEE Trans Fuzzy Syst
2002;10:11727.
[21] Robandi I, Kharisma B. Design of interval type-2 fuzzy logic based power
system stabilizer. Int J Electr Electron Eng 2009;3:593600.
[22] Birkin PAS, Garibaldi JM. A comparison of type-1 and type-2 fuzzy controllers
in a micro-robot context. In: Proceedings of the IEEE international conference
on fuzzy systems; 2009. p. 185762.
[23] Siahkali H, Vakilian M. Interval type-2 fuzzy modeling of wind power
generation in Genco's generation scheduling. Electric Power Syst Res
2011;81:1696708.
[24] Castillo O, Melin P, Castro JR. Computational intelligence software for interval
type-2 fuzzy logic. In: Proceedings of the 2008 workshop on building
computational intelligence and machine learning virtual organizations;
2008 p. 913.
[25] Mendel JM. Uncertain rule-based fuzzy logic systems, introduction and new
directions. Upper Saddle River, NJ: Prentice Hall; 2001; 26572.
[26] Karnik NN, Mendel JM, Liang Q. Type-2 fuzzy logic system. IEEE Trans Fuzzy
Syst 1999;7:64358.
[27] Engelbrecht AP. Fundamentals of computational swarm intelligence. England:
John Wiley; 2005.
[28] Das TK, Venayagamoorthy GK. Optimal design of power system stabilizer
using a small population based PSO. IEEE PES General Meeting; 2006.
[29] Zhang L, Yu H, Hu S. A new approach to improve particle swarm optimization.
In: Proceedings of the international conference on genetic and evolutionary
computation. Chicago, IL, USA, July 1216, 2003. p. 1349.
[30] Holland J. Adaptation in natural and articial systems: an introductory
analysis with applications to biology, control, and articial intelligence. MIT
Press; 1992.
[31] Herreros A, Baeyens E, Pern JR. Design of PID-type controllers using multiobjective genetic algorithms. ISA Trans 2002;41:45772.
[32] Goldberg DE. Genetic algorithms in search, optimization and machine learning. Boston, MA, USA: Addison-Wesley Longman Publishing Co., Inc.; 1989.
[33] Haupt RL, Haupt SE. Practical genetic algorithms. New York, USA: John Wiley &
Sons, Inc.; 1998.
[34] Macvicar-Whelan PJ. Fuzzy sets for manmachine interaction. Int J ManMach
Stud 1976;8:68797.
[35] Castillo O, Melin P. A review on the design and optimization of interval type-2
fuzzy controllers. Appl Soft Comput 2012;12:126778.
[36] HS-311 Hitec RCD, Available: http://www.hitecrcd.com/products/servos/ana
log/standard-sport/hs-311.html [1.12.12].
[37] Wada T, Ishikawa M, Ryohei, R Maruta, I, Sugie T. Practical modeling and
system identication of R/C servo motors. In: Proceedings of the 18th IEEE
conference on control applications; 2009. p. 137883.
[38] Castro JR, Castillo O, Melin P. An interval type-2 fuzzy logic toolbox for control
applications. In: Proceedings of the conference on fuzzy systems, IEEE-FUZZ;
2007. p. 16.
[39] Castro JR, Castillo O, Melin P, Rodriguez-Diaz A. Building fuzzy inference
systems with a new interval type-2 fuzzy logic. Trans Comput Sci I
2008:10414.