You are on page 1of 42

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281320741

Fuzzy Logic: Concepts, System Design, and Applications to Industrial


Informatics

Chapter · January 2012

CITATIONS READS

2 4,702

2 authors:

Siddhartha Bhattacharyya Paramartha Dutta


Rajnagar Mahavidyalaya Visva Bharati University
366 PUBLICATIONS   1,773 CITATIONS    296 PUBLICATIONS   2,574 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Healthcare Data Analytics View project

Call For Book: Swarm Intelligence for Cloud Computing View project

All content following this page was uploaded by Siddhartha Bhattacharyya on 29 August 2015.

The user has requested enhancement of the downloaded file.


Handbook of Research
on Industrial Informatics
and Manufacturing
Intelligence:
Innovations and Solutions
Mohammad Ayoub Khan
Ministry of Communications and Information Technology, India

Abdul Quaiyum Ansari


Jamia Millia Islamia, India
Managing Director: Lindsay Johnston
Senior Editorial Director: Heather A. Probst
Book Production Manager: Sean Woznicki
Development Manager: Joel Gamon
Development Editor: Myla Harty
Acquisitions Editor: Erika Gallagher
Typesetter: Adrienne Freeland
Cover Design: Nick Newcomer, Lisandro Gonzalez

Published in the United States of America by


Information Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: cust@igi-global.com
Web site: http://www.igi-global.com

Copyright © 2012 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or
companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.

Library of Congress Cataloging-in-Publication Data


Handbook of research on industrial informatics and manufacturing intelligence: innovations and solutions / Mohammad
Ayoub Khan and Abdul Quaiyum Ansari, editors.
p. cm.
Includes bibliographical references and index.
Summary: “This book is the best source for the most current, relevant, cutting edge research in the field of industrial
informatics focusing on different methodologies of information technologies to enhance industrial fabrication, intelligence,
and manufacturing processes”-- Provided by publisher.
ISBN 978-1-4666-0294-6 (hardcover) -- ISBN 978-1-4666-0295-3 (ebook) -- ISBN 978-1-4666-0296-0 (print & perpetual
access) 1. Industrial electronics. 2. Production engineering--Data processing. 3. Computer integrated manufacturing
systems. I. Khan, Mohammad Ayoub, 1980- II. Ansari, Abdul Quaiyum, 1962-
TK7881.H36 2012
658.4’038--dc23
2011044976

British Cataloguing in Publication Data


A Cataloguing in Publication record for this book is available from the British Library.

All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.
33

Chapter 3
Fuzzy Logic:
Concepts, System Design, and
Applications to Industrial Informatics

Siddhartha Bhattacharyya
The University of Burdwan, India

Paramartha Dutta
Visva-Bharati University, India

ABSTRACT
The field of industrial informatics has emerged as one of the key disciplines for the purpose of intelligent
management and dissemination of information in today’s world. With the advent of newer technical
know-how, the subject of informative intelligence has assumed increasing importance in the industrial
arena, thanks to the evolution of data intensive industry. Real world data exhibit varied amount of un-
quantifiable uncertainty in the information content. Conventional logic is often unable to explain the
associated uncertainty and imprecision therein due to the principles of finiteness of observations and
quantifying propositions employed. Fuzzy sets and fuzzy logic provide a logical framework for descrip-
tion of the varied amount of ambiguity, uncertainty and imprecision exhibited in real world data under
consideration. The resultant fuzzy inference engine and the fuzzy logic control theory supplement the
power of the framework in design of robust failsafe real life systems.

INTRODUCTION technology, there has been a stupendous increase


in the exchange of data based information leading
Industrial informatics (Acciani, 2011; Gomperts, to data explosion. As a result, for keeping pace
2011) has assumed importance of late thanks with this ever-increasing data and knowledge
to intelligent management and dissemination base, new subjects like informative intelligence
of information. With the rapid advancement of and intelligent informatics have come to the fore.
Information is manifested in several different
DOI: 10.4018/978-1-4666-0294-6.ch003 forms. It may be either in the form of raw data,

Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Fuzzy Logic

images, video content, speech signals or in any (Ross2003, Ross2004) through the assignment of
other electronic form. Any form of consumed real non-measurable logical qualities.
world data contains a varied amount of ambiguity This chapter is devoted to develop an under-
and imprecision, which cannot always be measured standing of the essence of fuzzy set theoretic
in practice. As such, classical computing systems concepts, the embodied logic and their appli-
seldom account for the associated uncertainty and cation to fuzzy system design as it applies to
imprecision in the principles of finiteness of ob- different fields of science, engineering, finance
servations and quantifying propositions employed. and industry. One of the important tenets of
Fuzzy sets (Zadeh, 1965; Cox, 1994; Dubois fuzzy system design is proper appraisal of fuzzy
1980; Kosko, 1997; Ross, 2003; Ross, 2004; Ber- based inference system design and fuzzy control
kan, 2000) evolved by Professor Lotfi Zadeh of the theory, which lead to the development of robust
University of California at Berkley are capable of and failsafe systems. A part of the chapter would
describing the vagueness and ambiguity inherent deal with the intricacies of fuzzy control with
in real world data. Professor Zadeh reasoned on reference to its application to industrial system
the intelligence in human reasoning to pave the design. Case studies of the applications of fuzzy
way for foundation of fuzzy sets and fuzzy logic. sets, fuzzy logic and fuzzy inference system to
He was motivated by the fact that human beings information management are also highlighted
more often communicate via natural language in this chapter. Due to the ability of handling
terms or linguistic expressions, which cannot be uncertainties in real world data, fuzzy sets and
always quantified by numeric values (Zadeh65, fuzzy logic have made a huge impact on virtually
Zadeh73, Cox94, Dubois80, Zadeh78, Zadeh94). every sphere of science, engineering and finance.
Typical human expressions include either one or The application perspectives of fuzzy sets and
several linguistic phrases, viz., very, very tall boy, fuzzy logic is also touched upon with special
very fast car, quite a few people, etc. These phrases, emphasis on the fields of industrial informatics
though aptly describe human feelings, cannot be and resource management (Chaudhari2010),
numerically quantified in the strict sense of the machine intelligence (Kyoomarsi2009), process
term. Thus, even in the absence of any precise, modeling and quality control (Berkan2000), ro-
numerical input information, human beings are botics, image processing and pattern recognition
capable of highly adaptive control. Moreover, as in industry (Simpson92), financial engineering
the complexity of a system increases, it becomes domains (Shipley2009, Meng2008), to name
more difficult and eventually impossible to make a few.
a precise statement about its behavior, culminat- The chapter is organized as follows. The
ing in a point of complexity, which cannot be limitations of the conventional crisp sets and
adjudged by means of a value. Zadeh coined the crisp logic are highlighted in the section titled
term “fuzzy” (standing for something which is FUZZY SET THEORY: CONCEPTS AND TER-
vague, obscure and imprecise) to replicate the MINOLOGIES with three real life observations.
notion of non-measurable human understanding The foundations and efficacy of fuzzy set theory
and logic. Thus, fuzzy sets form the backbone in dealing with the stated problems are discussed.
of more efficient and robust systems, which are The concept of partial membership assignment
immune to all sorts of uncertainties and impreci- in fuzzy sets is also presented in this section.
sions prevalent in the real world. Hence, fuzzy The section titled TYPES OF MEMBERSHIP
systems operate in a linguistic framework and their FUNCTIONS introduces some of the standard
strength lies in their capability to handle linguistic membership functions used to model fuzzy sets.
information and perform approximate reasoning Any fuzzy system operates on linguistic inputs.

34
Fuzzy Logic

Hence, the first and foremost phase in design- FUZZY SET THEORY: CONCEPTS
ing a fuzzy system is to obtain a set of fuzzified AND TERMINOLOGIES
inputs suitable for the system to be designed.
The section on FUZZIFICATION elucidates an This section discusses the limitations of the con-
example of application of a membership function ventional crisp set theory and crisp logic so far as
to illustrate this step of fuzzification procedure. describing real world uncertainties is concerned,
This section also justifies the effectiveness of the thereby delineating the basis for development of
assignment of fuzzy memberships in qualifying the fuzzy set theoretic paradigm. The mathemati-
uncertain propositions in real life observations. cal requisites in the buildup of fuzzy set theory
Even though fuzzy systems operate on and gen- are also elucidated in this section. In addition,
erate fuzzy data, human beings are comfortable the cardinality of a fuzzy set and different forms
with crisp data. Hence, some means of revert- of set theoretic operations on fuzzy sets are also
ing to the tangible crisp form of the fuzzified discussed in this section.
output is required for usability in the real world.
The section on DEFUZZIFICATION discusses Notion of a Fuzzy Set
the different techniques of the reverse process
of fuzzification, i.e. defuzzification. The next Before going into the details regarding the funda-
section on FUZZY OPERATORS illustrates mentals of fuzzy set theory, let us ponder on the
the collection of several operators, which are following real world observations.
applied for modifying the degree of fuzziness
of the constituent elements of a fuzzy set. The Observation 1
different measures of the degree of fuzziness
of the elements of a fuzzy set are dealt with in Consider the image pattern as presented in Figure 1.
the section titled FUZZY MEASURES. The Either of the following options would possibly
subsequent section on FUZZY LOGIC eluci- describe the exact nature of the pattern.
dates the fuzzy logical paradigm based upon the (i) Black, (ii) White, (iii) Half Black, (iv)
fuzzy set theoretic concepts. Typical examples Half White.
of the representative rule forms are presented A sizable portion of the human population would
there. A detailed analysis of the operation of the conclude the observation as being the options (iii) i.e.
fuzzy inference system based on fuzzy logic is Half Black while some would opt for (iv) i.e. Half
illustrated in this section as it applies to the man- White. Based on these options, it may be surmised
agement of uncertain and inexact information. that the pattern is either a Half Black or a Half
A comparative study of the different graphical White perspective. Assuming an intelligent and
fuzzy inference models is also made in this rational population, the mathematical conclusion is
section. The section on FUZZY CONTROL
presents a broad overview of the fundamentals Figure 1. Image pattern
of fuzzy control theory and its applications.
The general application areas of fuzzy only and
hybrid neuro-fuzzy, neuro-genetic, neuro-fuzzy-
genetic systems are discussed in the penultimate
section on APPLICATIONS OF FUZZY SETS
AND FUZZY LOGIC BASED SYSTEMS. The
DISCUSSIONS AND CONCLUSION section
concludes the chapter.

35
Fuzzy Logic

Half Black = Half White (1) Full = Empty (6)

1 1 Again we have reached to an unreachable con-


i.e. Black= White (2)
2 2 clusion in an attempt to quantify this observation.
Ok. Let’s try to adjudge another one.
1
Eliminating the common factor from both sides, Observation 3
2
Black=White (3) Here, we would consider two real life-eluding
paradoxes, which cannot always be answered
Isn’t the result in equation 3 astounding? correctly.
Making the impossible possible, i.e., equating
unequal quantities. Now, let us look into the next 1. A liar says, “I am lying”. Is what he says
observation. true or false?
2. If you have a heap of sand and remove one
Observation 2 grain at a time, when does it cease to be a
heap?
Consider a glass containing water as shown in
Figure 2. Learning from the observations 1, 2 and 3, it
If the status of the glass is to be quantified, two is evident that neither the conventional number
of the most striking conclusions would be (i) Half system nor the conventional logic (crisp logic) is
Full and (ii) Half Empty. Thinking on similar always able to interpret happenings around us and
lines as in observation 1, it can be written as that most of the real world observations exhibit
a certain amount of vagueness and imprecision.
Half Full = Half Empty (4)
Thus, it can be inferred that the conventional
crisp sets and logic are incapable of handling the
1 1
i.e. Full= Empty (5) varied amount of imprecision, uncertainty and
2 2 vagueness exhibited by real life observations.
This is because of the fact that in crisp sets/logic,
1 information is more often expressed in quantify-
Again, eliminating the common factor from
2 ing propositions, which is not always able to
both sides, it comes out to be describe the underlying uncertainty in real life
observations. To be more precise, crisp logic is
a bivalent logic, which entails only one of two
Figure 2. Glass containing water possible outcomes (TRUE or FALSE) of any
observation. Bivalence works fine at extremes
but fails everywhere else. In fact, crisp logic/
bivalence is guided by two main core principles,
viz., (i) law of excluded middle, which obviates
any possible outcome other than the extremes and
(ii) principle of non-contradiction, which stresses
only on an outcome and its negation.
On the other hand, fuzzy logic is a multi-
valued logic (Kosko97), which incorporates all

36
Fuzzy Logic

possible outcomes in an observation. Plato laid 1. The image pattern is partly Black and partly
the foundation for what would become fuzzy White, with the linguistic term partly be-
logic, indicating that there was a third region ing the membership degree of blackness/
beyond TRUE and FALSE. It was Lukasiewicz whiteness.
who first proposed a systematic alternative to 2. The image pattern contains Black with a
the bivalued logic of Aristotle. The third value membership of 0.5.
proposed by Lukasiewicz can be best translated 3. The image pattern contains White with a
as possible, and he assigned to it a numeric value membership of 0.5.
between TRUE and FALSE. Later, he explored
four-valued logic and five-valued logic, and then Thus, the fallacy regarding the blackness/
he declared that, in principle, there was nothing whiteness of the image pattern can be addressed.
to prevent the derivation of infinite-valued logic
(Kosko97). Fuzzy logic provides the opportunity Observation 2 Revisited
for modeling conditions that are inherently impre-
cisely defined. Fuzzy techniques in the form of On similar lines, if the glass of water is considered
approximate reasoning provide decision support as a fuzzy set, it can be readily stated that the
and expert systems with powerful reasoning ca- glass is partly filled with water and partly empty.
pabilities. The permissiveness of fuzziness in the As before, the linguistic term partly implies the
human thought process suggests that much of the membership of the water content in the glass.
logic behind thought processing is not traditional
two-valued logic or even multivalued logic, but Observation 3 Revisited
logic with fuzzy truths, fuzzy connectiveness
(Yager80) and fuzzy rules of inference. Hence, in The paradoxes in observation 3 are typical ex-
a fuzzy set (Zadeh65), each and every observation amples of the limitations of bivalent logic. As far
exists with a varied degree of containment in the as the first paradox is concerned, it is definitely
universe of discourse. This degree of containment difficult to surmise the attitude of the liar in terms
is referred to as the membership value of the ob- of only two outcomes, viz., not lying (truth) and
servation. A fuzzy set is a mapping from an input lying (falsity). This is because a liar always lies
universe of discourse into the interval (0, 1) that and whatever he speaks out is a lie. But there is a
describes the membership of the input variable. subtle difference between this particular statement
This mapping is referred to as fuzzification. The and any other statements he makes, since here he
reverse mechanism to revert to the bivalent crisp is testifying the fact that he is a liar. Thus, he is
logic is termed as defuzzification. both speaking a lie and a truth.
Now, let us see whether the observations 1, 2 However, if the attitude and behavior of the liar
and 3 listed above can be interpreted in terms of are considered as fuzzy sets, it can be surmised
the concept of fuzzy sets and memberships. without any dispute that he is partly true and
partly false, where the linguistic term partly, as
Observation 1 Revisited stated before, implies a membership value of the
nature of his uttered statement.
Instead of quantifying the blackness and whiteness The second paradox can be similarly dealt with
in the image pattern in Figure 1, if the pattern is by considering the heap of sand to be a fuzzy set.
considered as a fuzzy set of darkness/brightness, This analogy enables one to define a member-
then the following conclusions aptly explain the ship value for the quantity of sand in the heap to
image pattern. qualify/fail as a heap.

37
Fuzzy Logic

After having an initial impression about the the membership value of an element to unity,
power of fuzzy logic, it would be wise enough stronger is the containment of the element within
to consider the famous Russell’s barber paradox the fuzzy set. Similarly, a lower membership
mentioned below. value implies a weaker containment of the element
within the set. A fuzzy set A, characterized by a
A barber claims to shave all and only those who membership function mA (x i ) and comprising ele-
do not shave themselves. ments xi, i = 1, 2, 3, …, n, is mathematically ex-
pressed as (Zadeh65, Ross2004, Kaufmann85)
Does the barber shave himself?

= mA (x i )
From the crisp logic (Aristotelian logic) point A å xi
; i = 1, 2, 3, ..., n (7)
of view, the obvious conclusion comes out to be i

that if the barber shaves himself then he does not,


and if he does not shave himself, he does. This where, å represents a collection of elements.
again boils down to a contradiction of facts. i
 on U may also be represented
A fuzzy set A
Kosko illustrated the awesome power of multi-
as a set of ordered pairs. Each pair consists of a
valued logic by applying it to solve Russell’s barber
generic element x and its grade of membership
paradox. According to the solution proposed by
 = {(x , m (x )) | x Î U }.
function mA (x ), i.e., A
Kosko (Kosko91, Kosko92, Kosko94), the bar- 
A

ber’s claim is literally a half-truth lying exactly The support S Î [0, 1] , of such a fuzzy set A  is

A
midway between the limiting cases, midway defined as (Zadeh65, Ross2004, Kaufmann85)
between 100% truth and 100% falsity.
In precise it is an established fact that n mA (x i )
SA = {å : x i Î X "mA (x i ) > 0}
i= 1 xi
Whereas crisp sets quantify quantities, fuzzy sets (8)
qualify qualities.
 represents all
The core (C A ) of a fuzzy set A
those constituent elements, whose membership
Fuzzy Set Theoretic Concepts
values are equal to unity, i.e.
A fuzzy set is an extension of a crisp set. Where-
C A = {x i Î U | mA (x i ) = 1} (9)
as crisp sets allow only full membership or no
membership at all to the constituent elements,
fuzzy sets allow partial memberships. In a crisp  is
The bandwidth ( BWA ) of a fuzzy set A
set, membership or nonmembership of an element expressed as (Zadeh65, Ross2004, Kaufmann85)
x in a set A is described by a characteristic func-
tion χA(x), where χA(x)=1 if x∈A and χA(x)=0 if BWA = {x i Î U | mA (x i ) ³ 0.5} (10)
x∉A. Fuzzy set theory extends this concept by
defining partial memberships allowing all the
The resolution of a fuzzy set A  is determined
elements to exist in the universe of discourse U
with varying degrees of membership. A fuzzy set by the α-cut (or α-level set) of the fuzzy set. It is
A on a universe of discourse U is characterized a crisp set Aα containing all the elements of the
universal set U, that have a membership in A 
by a membership function mA (x ) Î [0, 1]. Closer
greater than or equal to α, i.e.

38
Fuzzy Logic

Aa = {x i Î U | mA (x i ) ³ a}, a Î [0, 1] (11) Figure 3. Representation of fuzzy set parameters

If Aa = {x i Î U | mA (x i ) ³ a}, then Aα is


referred to as strong α-cut. The set of all levels α
∈ (0, 1) that represents distinct α-cuts of a given
, is called a level set of A
fuzzy set A , i.e.

L A = {a | mA (x ) = a, x Î U } (12)

Figure 3 provides a graphical representation of the


core, bandwidth, support and α-level of a fuzzy set.
Since fuzzy sets are characterized by multi-
valued linguistic variables, a typical linguistic
variable x in a fuzzy set A  on the universe of
 s (x )
A
discourse U is represented by NormA s (x ) = (13)
TA (x ) = {Tx1 ,Tx2 ,Tx3 , ...,Txk } a n d hgtA s
mA (x ) = {mx1 , mx2 , mx3 , ..., mxk }, w h e r e
T ; i = 1, 2, 3, ..., k are fuzzy numbers with mem-
x
i The corresponding denormalization opera-
bership values mxi ; i = 1, 2, 3, ..., k . If x represents tion is given by (Bhattacharyya2006, Bhattacha-
the fuzzy variable speed, then TA (x ) may refer ryya2007)
to various degrees of speed like slow, medium or DeNormA s (x ) = hgtA s (x ) ´ NormA s (x ) (14)
fast speed. It may be noted that the adjectives
viz., slow, medium or fast are often referred to as
linguistic hedges in that they are used to indicate In general, for a subnormal fuzzy subset with
the fuzziness in the fuzzy set speed. These lin- support, SA s Î [L,U ], the normalization and the
guistic hedges are also fuzzy sets like the set denormalization operators are expressed as (Bhat-
speed. tacharyya2006, Bhattacharyya2007)
The maximum of all the membership values
 is referred to as the height (hgt )
in a fuzzy set A  s (x ) - L
 A
A
NormA s (x ) = ;
of the fuzzy set. If hgtA = 1, then the fuzzy set U-L (15)
is referred to as a normal fuzzy set. For a fuzzy DeNormA s (x ) = L + (U - L) ´ NormA s (x )
set where there is only one element with a mem-
bership value of 1, the element is referred to as
the prototype of the set. If however, hgtA < 1,
Fuzzy Cardinality
then it is referred to as a subnormal fuzzy set. A
normal fuzzy set is a superset of several non- The scalar cardinality of a fuzzy set A  is the
empty subnormal fuzzy subsets. summation of the membership grades of all ele-
A subnormal fuzzy subset ( A  s ) can be con-  . It is given by (Zadeh65, Ross2004)
ments of x in A
verted to its normalized equivalent by means of
the normalization operator given by (Bhattacha-
ryya2006, Bhattacharyya2007)
| A |= å
x ÎU
mA (x ) (16)

39
Fuzzy Logic

where, U is the universe of discourse. When a Inter sec tion : mA Ç B (x ) = mA (x )mB (x )
fuzzy set A has a finite support, its cardinality
(22)
can be defined as a fuzzy set. This fuzzy cardinal-
 f | and is defined by Zadeh
ity is denoted by | A In more general terms, the intersection of two
as (Zadeh65, Ross2004) fuzzy sets A and B
 is specified by a binary map-
ping T that aggregates two membership functions
 f |= a
|A å | A
(17) as

aÎL A a |

mA Ç B (x ) = T (mA (x ), mB (x )) (23)


where, α is the cut-off value, Aα is the α-level set of
the fuzzy set and L A is the corresponding level set. For example, the binary operator T may rep-
resent the multiplication of mA (x ) and mB (x ). A
Fuzzy Set Theoretic Operations collection of these fuzzy intersection operators is
referred to as T-norm (triangular norm) operators
Fuzzy set theoretic operations are analogous to (Yager80, Dubois80, Schweizer63), and they meet
crisp set operations, which lead to only the extreme the following basic requirements.
values of TRUE and FALSE. The fuzzy set op-
erations, however, yield all the possible outcomes boundary : T (0, 0) = 0;T (1, a ) = T (a, 1) = a
of an observation. The most elementary crisp set monotonicity : T (a, b) £ T (c, d ) if a £ c; b £ d
operations are union, intersection and complement
commutativity : T (a, b) = T (b, a )
operations, which essentially imply the OR, AND,
associativity : T (a,T (b, c)) = T (T (a, b), c)
and NOT operators, respectively. In the fuzzy
(24)
domain, the respective logical operators are the
max, min and complement operators. Therefore,
The first requirement ensures the correct gen-
the fuzzy logical union, intersection and comple-
eralization of crisp sets. The second requirement
ment operations (Ross2004) on two fuzzy sets
, B
 for an element x on the universe of discourse implies that a decrease in the membership values
A  and B cannot produce an increase in the
in A
X, are defined as 
membership value of the intersection of sets A

and B . The third requirement specifies that the
Union : mA È B (x ) = max[mA (x ), mB (x )] (18)
operation is insensitive to the order in which fuzzy
sets are combined, and the fourth requirement
Inter sec tion : mA Ç B (x ) = min[mA (x ), mB (x )] enables us to take the intersection of any number
(19) of fuzzy sets and any order of pair-wise groupings.
Similar to fuzzy intersection, the fuzzy union
Complement : mA (x ) = 1 - mA (x ) (20) operator is specified by the following binary map-
ping S (Yager80, Dubois80, Schweizer63).
Zadeh also defined fuzzy union and fuzzy inter-
mA È B (x ) = S (mA (x ), mB (x )) (25)
section as (Zadeh65)

Union : mA È B (x ) = mA (x ) + mB (x ) - mA (x )mB (x ) A collection of these fuzzy union operators
(21) is known as T-conorm or S-norm operators, and
they satisfy the following requirements.

40
Fuzzy Logic

boundary : S (1, 1) = 1; S (a, 0) = S (0, a ) = a ordinary membership functions since they assign
monotonicity : S (a, b) £ S (c, d ) if a £ c; b £ d a specific membership value to each element in

commutativity : S (a, b) = S (b, a ) the universe of discourse. Klir and Yuan (Giar-
associativity : S (a, S (b, c)) = S (S (a, b), c) ratano, 1993) proposed generalized membership
(26) functions for specifying the fuzziness in those situ-
ations where ordinary membership functions fail
Several T-norms and S-norms have been to precisely determine membership values due to
suggested in the literature (Yager80, Dubois80, lack of adequate information. A typical generalized
Schweizer63, Sugeno77). One example of a pair membership function is the interval-valued mem-
of S-norm and T-norm operators is the bounded bership function, which is constructed based on
sum and bounded product operators given by the upper and lower bounds of membership grades
(Yager80, Dubois80, Schweizer63, Sugeno77) for each element of the universe. The correspond-
ing fuzzy sets are referred to as interval-valued
x Å y = min(1, x + y ) (27) fuzzy sets. Thus, the membership of an element
x Ä y = max(0, x + y - 1) x in an interval-valued fuzzy set is represented by
the membership interval (α1, α2). In addition, the
interval (α1, α2) can also be treated as a fuzzy set.
TYPES OF MEMBERSHIP The resulting membership function is then referred
FUNCTIONS to as a type-2 fuzzy set. It may be noted that the
choice of a particular membership function depends
A membership function describes a fuzzy set on the system behavior under consideration. Some
(Zadeh65, Ross2004, Bezdek92, Bezdek93). If of the standard membership functions are described
the membership function for a fuzzy set is defined in the following subsections.
in such a way that the membership values of the
elements are either strictly monotonically increas- Triangular Membership Function
ing or strictly monotonically decreasing or firstly
strictly monotonically increasing followed by a Triangular curves depend on three parameters a,
strictly monotonically decreasing nature with an b and c. It is given by (Berkan, 2000)
increase in the values of constituent elements, then
the fuzzy set is referred to as a convex fuzzy set. ì
ï0 for x < a
ï
ï
Although normal and convex membership ï
ïx-a
ï
ï for a £ x < b
functions (both symmetrical and asymmetrical) b-a
f (x ; a, b, c, d ) = ï
í (28)
are commonly used in the fuzzification process, ï
ïc -x
ï for b £ x £ c
subnormal and nonconvex membership functions ï
ïc-b
ï
ï
are also used to deal with subnormal and nonconvex ï0
î for x > c
fuzzy sets. The membership functions are generally
defined as one-dimensional curves. When they are
defined in higher dimensions, these curves become Trapezoidal Membership Function
surfaces or hypersurfaces. Typical membership
functions include triangular, trapezoidal, gener- The trapezoidal membership functions depend on
alized bell shaped, Gaussian curves, polynomial four parameters, which decide the four corners of
curves, and sigmoid membership functions. These the trapezoid. The general form of a trapezoidal
membership functions are often referred to as membership function is given by (Berkan, 2000)

41
Fuzzy Logic

ì
ï0 for x < a Two-Sided Gaussian
ï
ï
ï
ïx-a Membership Function
ï
ï for a £ x < b
ïb-a
ï
ï The two-sided Gaussian membership function
f (x ; a, b, c, d ) = í1 for b £ x < c (29)
ï
ï is a combination of two Gaussian membership
ï
ïd-x
ï for c £ x < d functions characterized by means (c1, c2) and
ï
ïd-c
ï standard deviations (σ1, σ2) (Berkan, 2000). The
ï
ï
î0 for d £ x
first Gaussian function with mean c1 and standard
deviation σ1, determines the shape of the leftmost
π Membership Function curve, while the second Gaussian function with
mean c2 and standard deviation σ2 determines the
The π-shaped membership function is given by shape of the rightmost curve. The function attains
(Giarratano, 1993; Berkan, 2000) a maximum value of 1 if c1<c2.

ì
ï b Sigmoid Membership Function
ï
ïS (x ; c - b, c - , c) for x £ c
ï
ï 2
f (x ; b, c) = í
ï
ï b The sigmoid membership function is repre-
ï1 - S (x ; c, c + , c + b) for x > c sented by an asymptotic curve, which depends
ï
ï
î 2
(30) on two parameters λ andθ. It is given by (Ber-
kan, 2000)
where, S(x; a, b. c) is a membership function
defined by 1
S (x ) = (33)
1 + e - l(x - q)
ì
ï0 for x < a
ï
ï
ï
ï x-a 2 λ decides the slope of the function and θ is the
ï
ï2( ) for a £ x < b
S (x ; a, b, c) = ï b-a
function threshold.
í
ï
ï x-c 2
ï1 - 2( ) for b £ x £ c
ï
ï c-a Difference of Sigmoid
ï
ï
î1
ï or x > c
fo Membership Functions
(31)
This membership function results from the
where, a, b, and c are adjusted to fit the desired difference between two sigmoid membership
membership data. The parameter b is referred to as functions, characterized by the parameters (λ1,
the half width of the curve at the crossover point. λ2) and (θ1, θ2). Hence, the function depends on
four parameters, viz., λ1, λ2, θ1 and θ2. It is given
Gaussian Membership Function by (Berkan, 2000)

The Gaussian membership function depends on é 1 1 ù


S (x ) = ê - ú (34)
two parameters, viz., the mean c and the standard ê
ë1 + e
- l1 (x - q1 )
1+ e - l2 (x - q2 )
ú
û
deviation σ. It is given by (Berkan, 2000)

(x - c )2
-
S (x ; s, c) = e 2 s2
(32)

42
Fuzzy Logic

Product of Sigmoid S-Shaped Curve


Membership Functions Membership Function

This membership function results from the product The S-shaped membership function, as the name
of two sigmoid membership functions, character- suggests, is an S-shaped curve characterized by
ized by the parameters (λ1, λ2) and (θ1, θ2). It is two parameters a and b which determine the break
given by (Berkan, 2000) points of this membership function (Berkan, 2000).
This function is a smooth curve initiating from 0
é 1 1 ù at a to 1 at b for a<b. However, it becomes a step
S (x ) = ê ´ ú (35)
ê1 + e
ë
- l1 (x - q1 )
1+ e - l2 (x - q2 )
ú
û æa + b ö
÷
function with a jump from 0 to 1 at çç ÷when
è 2 ÷
ç ø
a≥b.
Generalized Bell Curve
Membership Function Z-Shaped Curve
Membership Function
The generalized bell curve membership func-
tion is an extension of the continuous Cau- The Z-shaped membership function is the reverse
chy–Lorentz probability distribution function of the S-shaped membership function. It is also
with the probability density function given by characterized by two parameters a and b which
(Berkan, 2000) determine its break points (Berkan, 2000). This
function is a smooth curve initiating from 1 at b
1 to 0 at a for b<a. However, it becomes a reverse
S (x ) = 2ù
(36) æa + b ö
é æx - c ö step function with a jump from 1 to 0 at ç ÷
÷
ç
pb ê
ê1 +
ç
ç ÷ú
÷ è 2 ÷
ç ø
ê è b ÷
ç øúú
ë û when b≥a.
These and score of other membership func-
tions discussed in the literature are used to model
where, c is a location parameter and β is a scale fuzzy sets depending on their characteristics.
parameter. c specifies the location of the peak of Thus, the choice of a membership function is not
the distribution while β specifies the half-width always heuristic. Several methods based on either
at half-maximum (HWHM). intuition or artificial intelligent tools like neural
The generalized bell curve membership networks or genetic algorithms (Yager, 1994;
function is characterized by three parameters Takagi, 1991; Karr, 1993) are commonly used in
a, b and c, which determine the shape and posi- designing membership functions appropriate for
tion of this membership function. It is given by describing the uncertainty in a system.
(Berkan, 2000)
Fuzzification
1
S (x ) = (37)
æ ö
2b As already stated, the process of converting a
ç x-c÷
ç1+ ÷
÷ crisp quantity to a fuzzy quantity by assigning a
ç
è a ÷
ø
suitable membership value to it is referred to as
fuzzification. The task of membership value as-
signment to a crisp variable is done by means of

43
Fuzzy Logic

Figure 4. Fuzzy membership functions for the speed of a car

a membership function. A membership function dium speed but with different degrees of member-
is essentially a curve that defines how each point ship. Similar reasoning applies for a car with a
in the input crisp space is mapped to a member- speed of 95mph. Moreover, if one observes the
ship value (or degree of membership) between membership curves in Figure 4, speeds of 54mph
0 and 1. The idea of a membership function can and 56mph also possess memberships in both the
be further clarified by considering a universe of sets slow speed and medium speed. However, μ54
discourse comprising the speed values of a car is lower in the set medium speed as compared to
ranging from 30mph to 120mph, with 30mph μ55 in the set medium speed. On the contrary μ54
and 120mph being the extreme possibilities. As is higher in the set lower speed as compared to
stated earlier, this wide range of speed values of μ55 in the set lower speed. This implies that a
the car can only be adjudged if the speed of car speed of 54mph is slower than a 55mph speed
is treated as a fuzzy set with the linguistic terms which, on the other hand, is more medium.
slow, medium and fast representing sub-ranges of Similarly, μ56 is lower in the set slow speed as
the car’s speed. Figure 4 shows three trapezoidal compared to μ55 in the same set, thereby indicat-
fuzzy membership functions for representing these ing that a 56mph speed is less slow compared to
three fuzzy sets of slow, medium and fast speed. a 55mph speed. On similar lines, as far as the set
The figure shows that a car with a speed of fast speed is concerned, μ94<μ95<μ96 means that a
35mph possesses a membership value μ=1 for 96mph speed is faster compared to a 95mph
the set slow speed, that having a speed of 75mph speed, which again is faster than a 94mph speed.
has μ=1 for the set medium speed and that with Therefore, a fuzzy description of the speed of a
a speed of 115mph exists with μ=1 for the set fast car is able to qualify the relative values of the
speed. However, a car with a speed of 55mph is different possible speeds of the car, which is
a member of both the sets slow speed and me- otherwise not clearly differentiable in terms of

44
Fuzzy Logic

meager values only. It may however be noted that A1 = {x 1 }; A0.75 = {x 1 , x 2 }; A0.5 = {x 1 , x 2 , x 3 }


A0.25 = {x 1 , x 2 , x 3 , x 4 }; A0+ = {x 1 , x 2 , x 3 , x 4 , x 6 } A0 = U
the fuzzy modeling of the speed of the car may
also be carried out by any other suitable fuzzy (39)
membership function.
The sets A1, A0.75, A0.5, A0.25 and A0+ follow di-
Defuzzification rectly from equation 11. A0 yields the universe of
discourse U since all elements in U must have at
Given the inherent power of fuzzy sets to represent least a 0 membership value in any set of the universe.
vagueness and ambiguity of real world informa-
tion, it is imperative to state that conventional Defuzzification into Scalar Values
computing relies essentially on precise crisp
values to perform a decision-making process. Before going into the details of this method, it is
Hence, a fuzzy variable must be converted to worth mentioning that a fuzzy system generates
its crisp counterpart for arriving at a precisely a fuzzy output from the fuzzy inputs, which are
quantified decision. The process of conversion essentially obtained by means of the application
of variables from the fuzzy to the crisp domain of appropriate membership functions on crisp
is referred to as defuzzification (Ross, 2004). The inputs. Hence the fuzzy output is obtained as a
input to the defuzzification process is a fuzzy logical operation of the membership functions of
set and the output of the defuzzification process the participating fuzzy inputs.
is either a crisp set or a single scalar value. The As an example, the union of the fuzzy inputs
following subsections describe the different de-   2, A
A1 , A  3 , shown in Figures 5a, 5b and 5c, char-
fuzzification processes. acterized by trapezoidal membership functions,
is given by a max operation by
Defuzzification into Crisp Sets
3
=
According to equation 11, any fuzzy set A  can A ∪ A
i= 1
i (40)

be resolved into an α-cut set Aα having elements


with a membership in A  greater than or equal to
The result of the union is also a fuzzy set
α ∈ (0, 1). An infinite number of such α-cut crisp represented in Figure 5d. Several defuzzification
equivalent sets are possible with an infinite num- techniques have been proposed in the literature
ber of possible α values. This is best illustrated (Ross2004). These include the centroid method,
by the following example. the max membership method, the weighted aver-
Consider the fuzzy set A  ={x , x , x , x , x , x } age method, the mean max membership method,
1 2 3 4 5 6
represented by Zadeh’s notation as (Zadeh65) the center of sums method, the center of largest
area method, the first (or last) of maxima method
ì1 0ü
= ï
A ï
í +
0.8 0.6 0.4 0.2
+ + + + ï
ï
ý
etc. These methods are described in the following
ï x6 ï subsections with reference to the input and output
îx 1
ï x2 x3 x4 x5 ï
þ
(38) membership functions of Figure 5.

The different α-cut sets derived using equa- Centroid Method


tion 11, for the six values of α, viz., 1, 0.75, 0.5,
0.25, 0+ (a value slightly greater than 0) and 0 This is the most commonly used defuzzification
are given by method (Takagi85, Lee90, Ross2004). In this

45
Fuzzy Logic

Figure 5. Fuzzy membership functions (a) (b) (c) input memberships (d) output fuzzy membership result-
ing out of union of (a) (b) (c)

defuzzification method, the center of gravity/ The centroid defuzzification method pro-
centroid of the union A (shown in Figure 5d) is duces a defuzzified value, which maintains a
determined and this value resembles the crisp balance in the union of fuzzy memberships due
output. For a continuous aggregated fuzzy set, to its weighted mean behavior. Figure 6a depicts
the centroid is given by  ) of Figure
the crisp output of the fuzzy union ( A
5d obtained by the centroid defuzzification
method.
òxm 
A
(x )dx
SA
xc = (41) Height/Max Membership Method
ò m  (x )dx
A
SA
This defuzzification method is generally used for
determining crisp output values of fuzzy union
 . The expression
where, SA is the support of A of participating membership functions with at
given in equation 41 reduces to most one of the membership exhibiting the
maximum membership value. Thus, this method
n is applicable in situations where a unique maxi-
å x i m  (x i ) mum membership unimodal peak exists in the
 ). In fact, the value of the fuzzy
A

xc = i= 1
n
(42) fuzzy union ( A
å m  (x i )
A
variable at this maximum membership is the
i= 1 defuzzified output. It is given by

xc = x i if mA (x i ) ³ mA (x j ) "i ¹ j (43)


for a discretized fuzzy set of n elements.

46
Fuzzy Logic

Figure 6b depicts the determination of the where, xi and xj are two extreme points of the
defuzzified output by the height/max member- maximum membership plateau shown in Figure 6c.
ship method.
Weighted Average Method
Mean Max Membership Method
It is a computationally efficient defuzzification
This method is also referred to as the middle-of- method (Ross2004), which is applicable for
maxima method (Takagi, 1985; Lee, 1990; Ross, symmetrical fuzzy membership functions. It is
2004). It is similar to the height method except given by
that it is applicable in those situations where the
fuzzy union ( A  ) does not contain a unique peak m

membership value. The fuzzy union shown in å x ci mA (x ci )


i

Figure 5d is a typical example of such a situation xc = i= 1


m
(45)
as it contains a plateau of highest membership å mA (x ci )
i
values rather than a single point membership i= 1

multimodal peak. The defuzzified value is obtained


by determining the mean of the maximum mem- where, m is the number of participating fuzzy
bership points. It is given by membership functions, x ci is the centroid of each
participating memberships.
xi + x j
xc = (44)
2

Figure 6. Defuzzification methods (a) centroid method (b) max membership method (c) mean max mem-
bership method (d) center of sums method (e) center of largest area (f) first (last) of maxima method

47
Fuzzy Logic

Center of Sums Method First (or Last) of Maxima Method

This defuzzification method is a faster method This defuzzification method (Ross2004) deter-
and is applicable to both symmetrical and asym- mines the smallest value of the domain with
metrical participating fuzzy membership functions maximized membership degree in the union of
(Ross2004). The notable feature of this method is that all the m participating fuzzy membership functions
the algebraic sum of the participating memberships A, i = 1, 2, 3, ..., m. This is achieved by first
i
is used instead of their union. Hence, the common determining the largest height in the fuzzy union
and overlapping areas are added twice. It is given by (hgtA ) by
i

n
hgtA = sup[mA (x )] (48)
ò xci å mA (x )dx i
i
x ÎU i

xc = n
i= 1
(46)
òå i= 1
mA (x )dx
i
Subsequently, the defuzzified output is the first
of maxima of the possible membership degrees,
i.e. it is determined by the greatest lower bound
where, x ci is the distance of each fuzzy variable of fuzzy variables, which possess a membership
from the centroid of its representative fuzzy degree equal to hgtA . The expression for the crisp
membership function. Figure 6d shows the deter-
i

output thus obtained is given by


mination of the defuzzified output by the center
of sums method. x c = inf{x | x Î U , mA (x ) = hgtA } (49)
x ÎU i i

Center of Largest Area Method


Alternatively, the last of maxima is defined by
The defuzzified output of this defuzzification the least upper bound of fuzzy variables, which
method (Ross2004) is the center of gravity of possess a membership degree equal to hgtA . It is
i
the participating convex fuzzy membership func- given by
tion having the largest area. It uses the centroid
method for finding the requisite center of gravity. x c = sup{x | x Î U , mA (x ) = hgtA } (50)
It is given by x ÎU i i

ò xm 
(x )dx Figure 6f shows the determination of the de-
fuzzified output value with this defuzzification
Av
SA
xc = v
(47) method.
òm 
Av
(x )dx
It is worth mentioning at this point that the
SA
defuzzification methods presented in the previ-
v

ous subsections lead to a tangible crisp output to


where, mA v is a participating convex fuzzy mem- be consumed in the real world. No matter what
bership function. The essential condition for the uncertainty prevails in real life, human beings are
application of this defuzzification method is that accustomed to playing with values and numbers.
the output fuzzy union should contain at least two Hence, a final crisp output is the obvious neces-
convex membership regions. The procedure of sity for arriving at a decision. The defuzzification
finding the crisp output value with this defuzzi- methods help in one way or other to reduce the
fication method is shown in Figure 6e. fuzziness in observations and derive a hard (crisp)

48
Fuzzy Logic

numeric value representative of the solution to the Dilator


problem at hand.
This operator dilates or stretches a fuzzy set
by increasing the membership of elements that
FUZZY OPERATORS are partly in the set. Typical examples include
somewhat, rather, quite etc. It is expressed as
As mentioned before in section on FUZZIFICA- (Ross2004)
TION, a fuzzy linguistic hedge modifies a fuzzy
set. To be precise, hedges are fuzzy operators, 1
mA (x ) = mA n (x ) for 0 £ mA (x ) £ 1 (52)
which act upon the membership function of a
fuzzy set to modify it. Considering the example
illustrated in the section, slow, medium and fast are Figure 7 also shows the operation of a rather
examples of fuzzy hedges that are applied to the dilator hedge on a fuzzy set temperature character-
fuzzy set speed to categorize the different degrees ized by a fuzzy membership function hot.
of speed. Several operators are used to form the lin-
guistic hedges in fuzzy logic. These either intensify Intensifier
the characteristics of a fuzzy membership (very,
extremely), or dilute it (somewhat, rather, quite), This operator acts as a combination of the con-
or complement it (not), or approximate a scalar to centration and dilation operators. It increases the
a fuzzy set (about, close to, approximately). Based contrast between the elements, which have more
upon the operation, these operators are classified than half-membership, and those elements, which
as (i) Concentrator (ii) Dilator, (iii) Intensifier and have less than half-membership. It is expressed
(iv) Diffuser (Ross2004, Bhattacharyya2006). The as (Ross2004)
following subsections describe these operators and
the implication of their operations on fuzzy sets. ì
) = ï
INT(A ï
í
2mA 2 (x ) for 0 £ mA (x ) £ 0.5

ï
î1 - 2[1 - mA (x )] for 0.5 < mA (x ) £ 1
2
ï
Concentrator (53)

This operator tends to concentrate the elements of


a fuzzy set by reducing the degree of membership Figure 7. Fuzzy operators concentrator and dilator
of those elements that are partly in the set. It is
generally expressed as (Ross2004)

mA (x ) = mA n (x ) for 0 £ mA (x ) £ 1 (51)

for n≥1. For example, the hedge slightly is repre-


sented by mA 1.3 (x ) . Zadeh defined the hedge very
as mA 2 (x ). Figure 7 shows the effect of application
of a very concentrator on a fuzzy set temperature
characterized by a fuzzy membership function
hot.

49
Fuzzy Logic

Figure 8. Fuzzy operators intensifier and diffuser


Typical examples include positively, definitely,
absolutely etc. Figure 8 shows the operation of
the positively intensifier operator on a fuzzy set
temperature characterized by a fuzzy membership
function hot.

Diffuser

This operator is the reverse of the intensifier


operator. It is given by (Ross, 2004)

ì
ï 1 12
ï
ï m  (x ) for 0 £ mA (x ) £ 0.5
ï 2 A
DIF(A) = ï
 í
ï
ï 1 1
ï
ï1- [1 - mA (x )]2 for 0.5 < mA (x ) £ 1
ï
î 2
(54)

One important representative example of the images exhibit a varied range of uncertainty in
diffuser operator is the generally hedge. Figure 8 the information contained therein. This is due to
also shows the effect of application of the generally the fact that there is no sharp demarcation as to
diffuser on a fuzzy set temperature characterized which level can be designated to be belonging
by a fuzzy membership function hot. to perfect black or perfect white.
Information can often be better expressed in Thus, an image can be visualized as a fuzzy
the form of images rather than discrete fuzzy set of brightness/darkness with the intensity levels
variables like temperature as in the previous of the pixels representing the fuzzy membership
examples. Images represent a map of different values of brightness/darkness in the set. Given
values intensity levels of objects of interest. this notion of an image in the background, the
Thus, there is a wide range of intensity levels in results of application of these operators on the
an image. For example, in a gray-level image, fuzzy intensity attribute of an image are shown
there are 256 different levels of gray intensity in Figure 9.
levels ranging from 0 (perfect black) to 255
(perfect white). In such a scenario, a gray-level
of 1 or 10 or 50 or 100 or 150 or 200 cannot be FUZZY MEASURES
correctly classified as perfectly black or perfectly
white. The situation is, however, trivial in case of The degree of fuzziness in a fuzzy set is deter-
a binary level image which comprises only two mined by several measures referred to as the
levels – 0 (perfect black) and 1 (perfect white). measures of a fuzzy set (Sugeno77, Ghosh93).
It is the crisp case and we are not interested in These measures include:
that in our discussion in this chapter. Similarly,
in a true-color image that we come across in our Index of Fuzziness
daily life, there are 16777216 colors in total with 0
The index of fuzziness n(A ) of a fuzzy set A
 hav-
(perfect black) and 16777215 (perfect white) and
all other possible colors in between. Therefore, ing n elements is a distance metric between the
it is evident that both gray-level and true-color set and its nearest ordinary set A, defined as

50
Fuzzy Logic

Figure 9. Application of fuzzy operators (a) Original Lena image (b) Concentrator (c) Dilator (d) Intensifier

ì
ï0 if mA (x ) £ 0.5  , characterized
The entropy E A of a fuzzy set A
mA (x ) = ï
í (55)
ï mA (x ) > 0.5
î1 if
ï by the membership function mA (x i ) , is a measure
of the degree of fuzziness in the fuzzy set. For a
 ) (Ghosh93)
The linear index of fuzziness nl (A fuzzy set comprising n elements, De Luca pro-
 is the Hamming distance version posed the fuzzy entropy function based on the
of a fuzzy set A
Shannon’s functional form as (DeLuca72, De-
of the index of fuzziness distance metric. It is
Luca74, DeLuca88a, DeLuca88b)
given by
n
1
å
n
) = 2 E A = - mA (x i ) ln(mA (x i ))
nl (A å [min{mA (x i ), 1 - mA (x i )}]
n i= 1
n ln 2 i = 1 (58)
(56) - {1 - mA (x i )} ln{1 - mA (x i )}

s )
Similarly, the linear index of fuzziness, nl (A The fuzzy entropy measure reflects the degree
for a subnormal fuzzy set A  s is defined as (Bhat- of ambiguity that corresponds to the randomness/
tacharyya2006, Bhattacharyya2007) disorder in an observation. In 1993, Bhandari and
Pal (Bhandari93) extended De Luca and Termini’s
s ) = 2
n
formula by introducing the α-order fuzzy entropy,
nl (A å [min{mAs (x i ) - L,U - mAs (x i )}]
n i= 1 which used the α-order probability entropy form. A
(57) plethora of literature is available on different forms
of evolved fuzzy entropy measures. Interested
where, L and U are the lower and upper bounds readers may refer to the literature (Fan, 2002) for
of the set. details regarding the latest forms of fuzzy entropy
measures proposed in this direction.
Fuzzy Entropy

The entropy of a fuzzy set is another measure of FUZZY LOGIC


the fuzziness of a fuzzy set. Ebanks (Ebanks83)
suggested that fuzzy entropy should satisfy five Human intelligence is manifested in ways in hu-
properties. A good survey of fuzzy entropy for man beings communicate. Human beings are more
finite universal sets can be found in the literature inclined to express and disseminate ideas using
(Pal94, Pal2000a). non-measurable phrases. These phrases are often-

51
Fuzzy Logic

times modulated by means of several adjectives/ The interpretation of an if–then fuzzy rule
adverbs, which reflect the degree of importance in a fuzzy rule base is carried out in two steps.
of the expressed ideas. As such, human beings are The first step evaluates the antecedent part of the
more amenable to linguistic expressions, which rule, where the input x is fuzzifed and the neces-
know no bounds of numeric precision. sary fuzzy operators (discussed in the section on
Fuzzy logic is a collection of conditional FUZZY OPERATORS) are applied on x. The
statements or fuzzy rules based on a linguistic second step involves the application of the result
reasoning framework, which embodies represen- of the antecedent to the consequent. This step
tation of shallow knowledge. As already stated, is also referred to as the implication, where the
the fundamental atomic terms in this linguistic or membership function of the resultant consequent
natural language-reasoning framework are often is evaluated. In fuzzy logic, a rule is fired so long
modified with adjectives or linguistic hedges. as there is a nonzero degree of similarity between
These linguistic hedges have the effect of modi- the premise and the antecedent of the rule. This
fying the membership function of a basic atom. is in sharp contrast to that in crisp logic where a
The general form of a fuzzy rule (Zadeh, 1965), rule is fired if the premise is exactly the same as
that is similar to natural language expressions, the antecedent of the rule, and the result of such
can be written as rule firing is the rule’s actual consequent.
For most applications, the fuzzy membership
IF premise (antecedent) THEN conclusion ® B
function mA ® B (x , y ) for a given relation A 
(consequent) is obtained with the minimum or product implica-
tion as
It is generally referred to as the IF-THEN rule
based form. It typically expresses an inference mA Ç B (x ) = mA (x )mB (x ) (59)
such that if a fact (premise, hypothesis or anteced-
ent) is known, then another fact (conclusion or
consequent) can be derived. It is worth noting at mA Ç B (x ) = min[mA (x ), mB (x )] (60)
this point that both the premise and the conclusion
are fuzzy entities and the collection of rules to Mamdani (Mamdani, 1977) first proposed the
describe the premise-conclusion relationship is minimum implication. Later, Larsen (Larsen80)
referred to as a fuzzy rule base. These if–then rules proposed the product implication in 1980. Inter-
specify a relationship between the input and out- ested readers are advised to refer to the literature
put fuzzy entities. Similar to crisp sets, these fuzzy (Mendel95) for details of implication methods.
relations indicate the degree of association be- Thus, the underlying logic behind the inferenc-
tween the elements of two or more sets. For two ing mechanism in human beings is aptly replicated
universes of discourse U and V, a fuzzy relation by a fuzzy inference system comprising a fuzzy
R(U ,V ) is defined as a set in the product space
rule base characterized by a shallow knowledge
(U×V) characterized by a membership function base of linguistic hedges.
mR (x , y ) , where x∈U and y∈V. Thus, a fuzzy rule This section is devoted to enable the readers to
of the form “if x is A , then y is B  ” where, x∈U gain an insight into the fuzzy logical paradigm, a
and y∈V, can be represented by a membership fuzzy inference system (FIS), its building blocks
function of the relation R = A
® B
, and the guiding fuzzy rule base. The different
mR (x , y ) = mA ® B (x , y ) . fuzzy inferencing approaches in vogue would
be touched upon with a comparative study of

52
Fuzzy Logic

the merits and demerits of each approach. The combining several linguistic connectives. Typical
concepts presented in this section would also be examples include (Ross, 2004):
supported with examples to illustrate the effective-
ness of the fuzzy logical paradigm in developing IF price is high THEN buying capacity (bc) is
information systems. very low

IF price is high And quality is good THEN bc


Rule Forms is high

As already stated, fuzzy logic comprises a col- IF price is low and quality is good THEN bc is
lection of fuzzy conditional statements/rules in medium
a linguistic framework, which form the basic
building block of a fuzzy inference system. The ELSE
knowledge base in an FIS is populated with rules
IF price is medium and quality is bad THEN bc
either by the experts or from experience of the is low
system at hand (Wang91). The rules in an FIS can
be broadly classified into three canonical forms
of rule-based forms. These include (Ross2004): Decomposition of Compound Rules
Assignment statements: These rule based
statements restrict the value of a variable by The basic properties and operations defined for
means of an assigned value. Typical examples fuzzy sets in the previous sections can be used to
are as follows (Ross2004): decompose any compound rule into a number of
simple canonical rules. These operations involve
x=large; color=yellow
the fuzzy logical AND (which is equivalent to
x=x fuzzy intersection) and OR (which is equivalent
to fuzzy union) operations between fuzzy subsets.
Conditional statements: These rule based The following four possibilities are of pedagogical
statements restrict the value of a variable based importance as far as the manifestations of com-
on a condition. Examples include (Ross2004): pound fuzzy rule forms are concerned.

IF x is large THEN color=yellow Multiple Conjunctive Antecedents

Unconditional statements: These rule based A compound rule can be formed by combining
statements restrict the value of a variable based on 1 , A
several antecedents A 2 ,A
 3 , …, A
 L such that
any condition. Some examples are (Ross2004): the conjunctive rule form takes the form
(Ross2004)
x is large

Stop  1AND A
IF x is A  2 AND A
 3 AND

¼ . AND A  L THEN y is B
S
Compound Rules (a subset of rules)

Depending on the complexity of the fuzzy infer-


Assuming a new fuzzy subset A  S and recalling
ence system, the underlying fuzzy rule base may
house both simple and compound rules formed by the equivalence of fuzzy logical AND operation

53
Fuzzy Logic

with the fuzzy intersection operation from section single antecedent based on a conditional ELSE
on FUZZY SET THEORETIC CONCEPTS, it statement in a compound rule form.
can be written as
 1 THEN (B
IF A  1 ELSE B
2)
 = A
A
S
 ÇA
 ÇA
 Ç ... Ç A
1 2
 3 L

Separating the conditional parts, this can be


Hence, the compound rule can be rewritten as decomposed into
(Ross2004)
 1THEN B
IF A  1OR IF NOT A
 1THEN B
2
 THEN B
IF A
S
 S

Nested IF-THEN Rules


Multiple Disjunctive Antecedents
These rules are also similarly dealt with as in
Similarly, for a disjunctive set of rules combined classical logic (Ross2004). The nested parts are
to form a compound rule of the form (Ross2004) interpreted separately to result in simpler forms
of the compound rules. The following example
 1OR A
IF x is A  2OR A
 3OR illustrates the decomposition of a compound nested
 L THEN y is B
S IF-THEN rule into a simpler rule.
¼ . OR A

the corresponding fuzzy superset is a fuzzy union


 1THEN æ
IF A ç
ç
è
IF A 1 ö
 2 THEN B ÷
÷
ø ( )
1 , A
of the participating fuzzy antecedents A  2 , …,
 L , i.e.
A It can be put in the form

S = A
1 È A
2 È A
 3 È ... È A
L  1AND A
IF A  2 THEN B
1
A

Thus, the compound rule is decomposed to a


simpler form as Fuzzy Inference System

 S THEN B
S A fuzzy inference system (FIS), as the name
IF A implies, is commonly used to quantify the ambi-
guity/imprecision prevalent in real world situa-
tions through qualification and modeling of the
Conditional ELSE Statements fuzziness of real world data by appropriate rule
bases. The inference system, as such, accepts
Similar to classical logic, a conditional ELSE state- and analyzes fuzzy data sets to derive a fuzzy
ment can be interpreted in fuzzy logic by treating conclusion or consequent. In general, an FIS ap-
the conditional parts separately (Ross2004). This plies a nonlinear mapping of the crisp input data
is illustrated by the following example, where two vectors into crisp outputs using a set of fuzzy
consequents are being exhibited pertaining to a rules resident in a rule base to infer on the system
behavior. For this purpose, FIS needs to convert

54
Fuzzy Logic

the crisp inputs to their fuzzy counterparts. This is aggregated and the fuzzy sets representing each
because the nonlinear mapping process is guided rule are combined into a single fuzzy set.
by the characteristic membership functions of The utility of an FIS can be further appreci-
the participating inputs and generated outputs, ated by taking recourse to a real life example,
the applied fuzzy operators and the underlying which cannot be generally solved from the con-
fuzzy if–then rules. In case of a multiple output ventional crisp logical point of view. Though the
scenario, an aggregation of the output sets is re- example to be put forward is a trivial one, yet
quired to achieve the target objective. It may be the implications reiterate the flexibility of the
noted that the fuzzy output generated by the FIS fuzzy logic on one hand and the incapability of
needs to be reverted to the crisp domain for proper the crisp logic in dealing with imprecision on
usability in the real world. Hence, a defuzzification the other.
mechanism follows all the foregoing steps of the Consider a weekend party (a regular affair)
FIS system procedure. hosted by the head of a family in a restaurant.
Figure 10 shows the block diagram of a multi- Obviously, the first consideration comes in the
input FIS comprising five inputs (x1, x2, x3, x4 x5) choice of a restaurant befitting the occasion. So,
generating a single aggregated output y. Figure what parameters guide the process of selection
10 also shows the essential components of an FIS, of a proper restaurant? This may be either its
viz., the fuzzifier, which fuzzifies the crisp inputs ambience or hygiene or publicity or efficiency
to the system by means of characteristic fuzzy of service or quality of food or affordability.
membership functions, the inference engine which It is evident that neither of these parameters is
infers on the possible system outputs based on the measurable or precise to any extent. These are
underlying rules in the fuzzy rule base, and the fuzzy or imprecise, hence uncertain. Even then
defuzzifier, which ultimately generates the crisp the head of the family is intelligent enough to
output to be used up by the real world. In addition, choose one from a list of probable restaurants in
Figure 10 also shows another component marked his locality and goes for the dinner with his fam-
as “Σ”. This component aggregates multiple rule ily members. This selection mechanism is thus
consequents resulting in a multiple output sce- guided by an inference mechanism, which elicits
nario. This happens when one or more rules in some exact responses out of inexact/imprecise
the rule base stand equally strong for the same fuzzy information. Moreover, after having dinner
antecedent at the same time. In such a situation, at the restaurant, the head of the family, being a
the consequents/ outputs for all the fired rules are sensible person, ought to offer some tips (also
fuzzy) to the bearer of the restaurant for the service

Figure 10. Block diagram of a fuzzy inference system

55
Fuzzy Logic

offered by him in terms of serving food, water, Tsukamoto Model


beverages and the like. The fuzzy tips the head
offers is also dependent on different factors which In this inference model (Passino98), the conse-
guided him in the restaurant selection process. quent part is represented by a fuzzy set with a
However, for the sake of brevity, the amount monotonic membership function of the form
of tips can be related, above all else to the two mqi (x ), where qi is a monotonic function defin-
fuzzy variables quality of food and the quality of ing the consequent. The ith rule of the underly-
service offered. This inferencing mechanism for ing rule base of the inference engine takes the
determination of the tips to the bearer stands as form
a classical example to demonstrate the power of
fuzzy inference systems in handling uncertainty IF u is ai AND v is bi THEN w
 is q
in real world problems. This example is often i

referred to as the basic tipping problem since


it deals with the problem of determination of The result of the ith rule fired with strength ti is
tips to the bearer. However, as already stated, determined as
with any FIS, since the ultimate output is a crisp
quantity, the resulting tip amount is also a scalar x i = mq-i 1 (ti ) (62)
and measurable quantity.
Different graphical models of inference are in For a system with multiple consequents, the
vogue for analyzing the fuzzy inference system aggregation of the outputs is obtained using the
(Passino98). Notable among these include the weighted sum method
(i) Mamdani, (ii) Tsukamoto, (iii) TSK (Takagi
Sugeno Kang) and (iv) Larsen models (Passino98). n
ti x i
These graphical methods are more user-friendly x' = å
i= 1 ti
(63)
than the conventional mathematical tools avail-
able in the market and even enable the beginners
to experiment and design robust fuzzy inference
systems starting from the scratch. Brief overviews Takagi-Sugeno Kang Model
of these models techniques are discussed in the
following subsections. The Takagi-Sugeno Kang model (Passino, 1998)
is similar to the Mamdani model in several respects
Mamdani Model so far as the fuzzy inference process, the input
fuzzification process and the application of fuzzy
This is the most widely accepted inference model. operators are concerned. The distinct difference
Proposed by Mamdani (Mamdani77, Passino98), between the two models lies in the fact that the
the antecedents and consequents are represented Sugeno output membership functions are either
as membership functions in the graphical view linear or constant in nature. Hence, it is compu-
of this model. As stated earlier, the inference tationally efficient and works well with linear,
mechanism is obtained using the min operator optimization and adaptive techniques but does
for a fuzzy implication and the max-min operator not fair enough with real life data sets. Moreover,
is used for the composition for the implication in this model, the consequent part is given as a
method. Since it is intuitive in nature, it has got crisp function of input variables. For an example,
widespread acceptance. Moreover, it is designed if u and v are the input variables, the consequent
to accept human inputs. part is represented as

56
Fuzzy Logic

q = f (u, v) (64) fuzzy inference engine in handling the stated


uncertainties, would aptly justify similar prob-
Thus, the ith rule may be represented as lems faced during dissemination, understanding
and analysis of vague and imprecise knowledge
IF u is ai AND v is bi THEN q is f (u,v) when it comes to industrial informatics. Thus,
this section focuses on the design perspectives
of the different facets required to solve problems
For multiple consequents, the process of ag- involving data sets entailing imprecise, vague and
gregation is carried out by means of the weighted fuzzy information content.
average method of defuzzification. The different participating input and output
fuzzy variables of the modified tipping problem
Larsen Model along with their representative membership func-
tions are shown in Figures 11 and 12 (designed
It is another graphical method of inference used using the Mamdani model of inference (Mam-
in practice (Passino98). It offers standard features dani77) and generated using MATLAB 6.5®
of a rule based inference system in that it uses (MathWorks95)).
the product operator for fuzzy implication and From Figures 11and 12 it is evident that the
the max-product operator for the composition input service (whose value ranges from 0 to 10)
operation. employs Gaussian membership functions, the
input food (whose value ranges from 0 to 10)
Modified Tipping Problem and its employs trapezoidal membership functions while
Solution Design Using Mamdani Model the input state of mind (whose value ranges from
0 to 1) employs double sided Gaussian member-
The basic tipping problem and its solution are well ship functions. The output variable tip (whose
documented in the literature (MathWorks, 1995). value ranges from 0 to 30) employs triangular
It incorporates two fuzzy input variables service memberships.
with three grades of membership (poor, good, Once the input and output variables have been
excellent) and food with two grades of membership modeled using respective membership functions,
(rancid, delicious) and the single fuzzy output vari- the fuzzy inference system applies the embedded
able tip with three grades of membership (cheap, rule base to infer the fuzzy output corresponding
average, generous). A subtle modification to the to incident fuzzy inputs to the inference system.
problem is presented here with the introduction The modified tipping problem may comprise a
of another real life uncertain fuzzy input variable collection of rules designed by experts out of
in the form of the head of the family’s state of experience. A subset of the rule base may form
mind with three grades of membership (morose, the following set of five rules.
indifferent, elated) without loss of generality. The
incorporation of the fuzzy variable state of mind 1. IF service is poor AND food is rancid AND
into the system introduces a flavor of the diverse state of mind is morose THEN tip is cheap
nature of uncertain situations often encountered in 2. IF service is excellent AND food is delicious
real world situations, be it in homely atmosphere AND state of mind is elated THEN tip is
or in a busy industrial scenario. generous
Even though the problem presented here devi- 3. IF service is good OR food is delicious OR
ates from the industrial perspective, yet a study state of mind is NOT elated THEN tip is
of the underlying uncertainty and the role of the average

57
Fuzzy Logic

Figure 11. Fuzzy membership functions for the inputs service, food and state of mind to the modified
tipping problem

58
Fuzzy Logic

Figure 12. Fuzzy membership functions for the output tip of the modified tipping problem

4. IF state of mind is indifferent THEN tip is Essentially, aggregation takes all fuzzy sets that
average represent the output for each rule and combines
5. IF service is poor OR food is delicious OR them into a single fuzzy set that is used as the
state of mind is elated THEN tip is average input to the defuzzification process. Thus, the
inputs to the aggregation process are truncated
The next part of the inferencing mechanism or modified output fuzzy sets obtained as the
is the interpretation of these rules to arrive at a output of the implication process. The output of
single or a set of conclusion/consequent. This is the aggregation process is a single fuzzy set that
generally a three-fold process. Firstly, all the fuzzy represents the output variable. Since the aggrega-
statements in the antecedent parts of the rules are tion method is commutative, the order in which
resolved to a degree of membership between 0 and the rules are executed is not important. The max
1. However, during this process, it is to be noted method happens to be the most commonly used
that if there are multiple parts to the antecedent, aggregation method.
fuzzy logic operators are invoked to resolve the Figure 13 shows the results of operation of the
antecedent to a single number between 0 and 1. designed fuzzy inference system with the embedded
This number ultimately determines the support/ rule base. The result is again generated with MAT-
strength for a particular rule to be fired. Finally, LAB 6.5® (MathWorks95). The columns indicate
the consequent part of the rule is determined by the fuzzy input/output variables. The numbers at
the implication operation (using either equation 60 the extreme left of the figure represent the five
or 61) from the antecedent part using the degree different rules of the rule base. The yellow shaded
of support for the entire rule. regions represent rules which are fired correspond-
Since different consequent outputs are obtained ing to the test input of service=5, food=9, state of
depending on the firing strengths of the different mind=0.5. From these values, it is evident that rule
rules in the rule base, the corresponding output number 3 (service=5 seems to be good; food=9
fuzzy sets need to be combined into one composite seems to be delicious even though state of mind
fuzzy set to produce the final crisp output after =0.5 does not seem to be elated), rule number 4
defuzzification. This process of combining output (state of mind=0.5 seems to be indifferent) and rule
fuzzy sets into a single set is called aggregation. number 5 (service=5 does not seem to be poor but

59
Fuzzy Logic

Figure 13. Defuzzified crisp output obtained after aggregation

food=9 seems to be delicious even though state of FUZZY CONTROL


mind=0.5 does not seem to be elated) fire simulta-
neously. The third, fourth and fifth blue shaded A fuzzy system comprises a black box guided by
triangles (from top) in the last column of Figure 13 an underlying fuzzy logic/rule base, which upon
show the corresponding multiple consequents thus accepting fuzzified input(s) delivers fuzzified
obtained. These are then aggregated using the max output(s). Hence, such a system is able to incor-
operator as stated earlier to yield the rightmost- porate ever-existing uncertainties in its functioning
bottommost blue shaded region appearing in Figure as well as its input-output relationships. The need
13. Then the centroid method of defuzzification is for design of a fuzzy control based system stems
applied to determine the crisp value of the tip (which from this ability of handling of uncertainties by
in this example comes out to be tip=15 for input fuzzy systems. In this section, the basis for the
values of service=5; food=9; state of mind=0.5) design of a fuzzy control based system would be
for the bearer at the restaurant. discussed with reference to information manage-
Thus, a fuzzy inference system is effective in ment in the industrial spectrum in particular and
handling uncertainty in information management intelligent systems in general.
and facilitating decision-making out of inexact Control systems abound in our daily life
and incomplete information. The performance of starting from the bare human physical systems to
a fuzzy inference system, however, depends on the complex industrial process control systems.
the fuzzy modeling capability of the system under A control system is defined as a collection of
consideration along with the underlying rule base hardware components designed to elicit a cer-
guiding the inference system. Thus, the more tain desired characteristic response or output
robust the rule base is, the more efficient is the through a control action (Ross2004). Control
fuzzy inference system and more is it supposed systems are typically of two types, viz., (i)
to predict the system behavior. open-loop control systems, in which the control

60
Fuzzy Logic

action does not depend on the physical system Industrial Applications of Fuzzy
output and (ii) closed-loop control systems, Logic Control
also referred to as feedback control systems,
where the control action depends on the physi- Fuzzy control have been put to use in a wide variety
cal system output. Since the proper functioning of industrial products and consumer appliances
of the closed-loop control systems depends including household appliances (Quail92), image
on the system outputs, a measurement of the processing equipment (Takagi92), water quality
outputs is an essential requirement. Once the control (Itoh87, Yagishita85), nuclear reactor
system outputs are known, the system errors are control (Kinoshita88, Bernard88), automatic train
determined and fed back to system for tuning operation systems (Yasunobu83, Yasunobu85,
the system betterment of system performance. Yasunobu87), automobile transmission control
Human physical systems are examples of open- (Kasai88), fuzzy logic controller hardware systems
loop control systems. Industrial systems are (Yamakawa86a, Yamakawa87a), fuzzy memory
closed-loop feedback control systems. Details systems (Togai86, Togai87, Watanabe88, Yamak-
regarding control system design are available awa86b, Yamakawa87b, Yamakawa88), to name
in the literature (Passino98). a few. A plethora of industrial applications using
A fuzzy inference system resides at the core fuzzy control includes blood pressure control dur-
of a fuzzy controller as shown in Figure 14. ing anesthesia (Meier92), camera auto-focusing
In addition, a knowledge base module present (Shingu89), image stabilization applications
therein guides the FIS fuzzifier with the corre- (Egusa92) etc. A good treatise on the design and
sponding membership characteristic of the input/ development of industrial fuzzy controllers is
output variables to the fuzzy rule-base system available in (Passino98). A host of different types
and the control system parameters. The system of fuzzy control domains inclusive of direct,
outputs are measured by the sensor module and nonlinear, adaptive and supervisory fuzzy control-
fed back for error estimation and further tuning lers are discussed in (Passino98) with regards to
of system parameters. The processing modules design methodologies of specific examples like
at the input end and the feedback path of the translational and rotational inverted pendulum,
controller are meant for seasoning the inputs/ two-link flexible robot system, automobile cruise
outputs either in (0, 1) or (-1, 1) amenable for control, magnetic ball suspension system, auto-
use by the controller. mated highway system, rocket velocity control,

Figure 14. Schematic of a fuzzy logic controller

61
Fuzzy Logic

temperature control, underwater vehicle control, since data exchange is always a part of the daily
tape drive servo motor control, liquid level control thoroughfare in every industry. Fuzzy logic plays
in a tank etc. Interested readers are also advised a very crucial role in extracting and analyzing
to consult (Ross2004, Phillips96, Shinskey88, relevant information from industrial transactions.
Ogunnaike94, Murrill91, Ross2003, Passino98, Another important aspect that demands the role
Wang97, Parkinson2001, Zong-Mu2004, Lee90) of fuzzy logic as far as industrial applications are
for the latest updates on the topic. concerned is the management of efficient industrial
resources for maximization of industrial profits.
The intelligence offered by the fuzzy logic can
APPLICATIONS OF FUZZY be faithfully exploited to achieve this objective
SETS AND FUZZY LOGIC as well. Chaudhury et al. (Chaudhari2010) devel-
BASED SYSTEMS oped a fuzzy model with an optimal usage of the
labor, automation, and production rate to arrive
The ability of fuzzy set and fuzzy logic in mod- at a significantly improved decision-making.
eling imprecision has been rightly explored and Niu and Dartnall (Niu2008) used a fuzzy-MRP-
put to use in a wide variety of applications in this II technique to deal with the uncertainty and
information arena. This section would illustrate the the imprecision involved in the Manufacturing
broad spectrum of different information-intensive Resource Planning (MRP-II) decision-making
applications in the frontier areas of technology. system to manage the process of production of
fast moving consumer goods. Other notable ap-
Industrial Applications plications in this direction can be found in the
literature (Das2003, Feriadi2003, Grabot2005,
These applications are generally characterized by Kahraman2007).
the interchange of a huge amount of information
in day-to-day transactions. Of late, the ITES (in- Process Modeling and Control
formation technology enabled services) industries
have emerged as the forerunner of these types of In each and every real life process, be it an indus-
data based industries. In ITES, transaction data trial one or a financial one, fuzzy sets find use in
are often under various threats, which pose dan- modeling higher and lower levels of reasoning the
ger to the integrity and validity of the data. This various aspects of control mechanisms involved
has given rise to the field of threat management. therein in the form of (i) language, (ii) problem
Researchers are constantly in search of newer solving, (iii) vision, (iv) speech recognition, for
mechanisms of shielding data from unknown yet generating the best process control settings of the
possible threats of data destruction. Threat model- embedded parameters. It is evident that the com-
ing is targeted towards achieving a secure system. plexity of a system increases with the number of
Sodiya et al. (Sodiya2007) proposed an effective parameters of a system as well as with the range
fuzzy logic-based threat modeling technique based of parameter values. It is also a known observa-
on six major categories of threats, viz., spoofing, tion that the uncertainty is manifold in a complex
tampering, repudiation, information disclosure, system. Hence, proper selection and tuning of the
denial of service and elevation of privilege using parameter values is a prerequisite in many appli-
the Mamdani model (Mamdani77) of inference. cations for achieving the desired output. Several
As the going goes, “The more the data the fuzzy based efforts have been invested in this di-
more the uncertainty”. The picture is also not rection to model system parameters either through
very much alien in case of other industries too expert opinions or through system trace so as to

62
Fuzzy Logic

attain a global set of optimum parameter settings image processing, classification and pattern rec-
appropriate for a specific task. Among the recent ognition in industry for automatic color image
advances, Majumdar and Ghosh (Majumdar2008) segmentation and edge detection (Boskovitz2002,
developed a fuzzy logic based model using the Simpson92, Zenzo98), medical image segmenta-
experience of yarn spinners for modeling the roles tion (Chang2002, Shitong2006, Shitong2007),
played by the various yarn inputs on the yarn output volume rendering (Kim2007), image enhancement
strengths. The expert system developed by them (Bhattacharyya2006, Bhattacharyya2007), video
using human knowledge base has been found to interlacing applications (Brox2006a, Brox2006b,
be robust enough to predict accurately the strength Van99, Gutierrez-Rios2004) etc.
of the output cotton yarn in the textile spinning Aside from the aforesaid applications, fuzzy
process. Cheikh et al. (Cheikh2007) devised an sets are also widely used in several other engi-
intelligent control method as a remedy to energy neering and financial applications, e.g. cogni-
crisis, pollution and global warming problems by tive science and machine learning (Joshi97,
solving the energy conversion deficiency problem Kyoomarsi2009), data clustering (Baraldi2001,
of photovoltaic cells. For this purpose they pro- Zhao2001, Yang2004, Chen2003), industrial
posed the maximum power point tracking (MPPT) portfolio management (Shipley2009, Yager81,
method of a photovoltaic system under variable Nanda2010, Wei2011), industry credit rating
temperature and insulation conditions. Another (Meng2008, Zhao2009) and the like.
interesting application which falls within this
domain and is worth mentioning as far as energy Hybrid Systems in
saving is concerned is that proposed by Lah et al. Information Management
(Lah2006). In their work, they used fuzzy logic for
taking full advantage of daylight for illuminating A synergistic symbiosis of different tenets of soft
a premise by means of controlled dynamic illumi- computing paradigm has resulted in more effi-
nance response of built environment in real-time cient and fault-tolerant hybrid systems. The soft
conditions. Osofisan and Obafaiye (Osofisan2007) computing tools like (i) neural networks operate
applied fuzzy logic for modeling the catalytic in a numeric framework and are well known for
unit of a petrochemical refinery. A host of other their learning and generalization capabilities and
applications in this domain can be found in the (ii) genetic algorithms provide powerful search
literature (Berkan, 2000). and optimization methodologies. These tools
are crucial to addressing problems in the area of
Image and Video Processing artificial intelligence and they ably supplement
and Pattern Recognition the fuzzy logical paradigm in developing robust
systems. Neuro-fuzzy techniques have been
As discussed in the section on fuzzy operators, developed to embed the intelligent structure of
images can also be regarded as fuzzy sets given the the human nervous system for solving various
varied distribution of intensity gamut. Hence, the real life problems (Taur2000, Majumdar2005,
uses of fuzzy sets and fuzzy logic in understanding Boskovitz2002, Muhammed2003, Shitong2006,
image preprocessing, processing and analysis are Shitong2007, Baraldi2001, Charalampidis2001,
self-justified. In fact, from the inception, much Bhattacharyya2006, Bhattacharyya2007, Er2008).
of the research works related to the development Neuro-fuzzy-genetic techniques have also been
of fuzzy sets and fuzzy logic have been invested evolved for the betterment of fuzzy-only systems
in the fields of pattern recognition and image (De2008, De2009, De2010), as far as management
processing (Pal2000b). Typical examples include of information is concerned.

63
Fuzzy Logic

DISCUSSIONS AND CONCLUSION ACKNOWLEDGMENT

A brief overview of the foundations of the fuzzy The authors would like to take this opportunity to
set theory and fuzzy logic is presented in this render their reverence to Dr. Parijat De, Director,
chapter. The limitations of the conventional crisp National Institute of Technical Teachers’Training and
logic in describing uncertainties encountered in Research, Chandigarh, India for his tireless efforts
real life situations are discussed with recourse to in pointing out the lexical and syntactic errors in the
some specific examples lifted from our work-a-day article and helping in improving its overall readability.
world. The power of fuzzy logic in handling such
limitations has been suitably explained by intro-
ducing the set theory-based notions and concepts. REFERENCES
This follows the introduction of the different types
of fuzzy membership functions in vogue. These Acciani, G., Fornarelli, G., & Giaquinto, A. (2011).
functions have been discussed with reference A fuzzy method for global quality index evalua-
to the appropriate mathematical forms used in tion of solder joints in surface mount technology.
practice. The application of the fuzzy membership IEEE Transactions on Industrial Informatics, 7(1),
functions for deriving at fuzzified variables has 115–124. doi:10.1109/TII.2010.2076292
been duly explained by a trivial real life example Baraldi, A., Binaghi, E., Blonda, P., Brivio, P.
of a fuzzification procedure. In addition, differ- A., & Rampini, A. (2001). Comparison of the
ent possible defuzzification procedures are also multilayer perceptron with neuro-fuzzy tech-
highlighted to emphasize the importance of the niques in the estimation of cover class mixture
usage of ultimate crisp variables in the real world. in remotely sensed data. IEEE Transactions on
As a sequel to the fuzzification and defuzzifica- Geoscience and Remote Sensing, 39(5), 994–1005.
tion procedures discussed, the chapter also empha- doi:10.1109/36.921417
sizes on the usages of different fuzzy operators in
practice. Furthermore, the manifestation of human Berkan, R. C., & Trubatch, S. L. (2000). Fuzzy
intelligence in handling uncertainties is discussed systems design principles (pp. 22–131). New
by taking recourse to fuzzy rule based systems Delhi, India: Standard Publishers Distributors.
based upon the foundations of fuzzy logic. The
Bernard, J. A. (1988). Use of rule based system for
different forms of rules that make up such systems
process control. IEEE Control Systems Magazine,
are touched upon with examples. To sum up the
8(5), 3–13. doi:10.1109/37.7735
concepts, the operation of a fuzzy inference engine,
which is a representative example of a fuzzy rule Bezdek, J. C. (1992). Computing with uncertainty.
based system, is illustrated with reference to the IEEE Communications Magazine, 30, 24–36.
design of a real life information based system doi:10.1109/35.156801
with all the necessary modalities. An overview
Bezdek, J. C. (1993). Fuzzy models—What
of fuzzy logic based control and its application
are they and why—Editorial. IEEE Transac-
to the industrial perspective is also presented for
tions on Fuzzy Systems, 1, 1–5. doi:10.1109/
the self sufficiency of the treatment. The trends
TFUZZ.1993.6027269
of research in the evolution of fuzzy logic based
systems for achieving superior performances are Bhandari, D., & Pal, N. R. (1993). Some new
also highlighted with due mention of the impor- information measure of fuzzy sets. Informa-
tant treatises and literature available as far as the tion Sciences, 67, 209–228. doi:10.1016/0020-
state-of-the-art application areas are concerned. 0255(93)90073-U

64
Fuzzy Logic

Bhattacharyya, S., Dutta, P., & Maulik, U. (2006). Chaudhari, O. K., & Khot, P. G. (2010). Fuzzy
A self supervised bi-directional neural network modeling in industry for the optimal use of avail-
(BDSONN) architecture for object extraction able resources. Advances in Fuzzy Mathematics,
guided by beta activation function and adaptive 5(2), 121–127.
fuzzy context sensitive thresholding. Interna-
Cheikh, M. S. A., Larbes, C., Kebir, G. F. T., &
tional Journal of Intelligent Technology, 1(4),
Zerguerras, A. (2007). Maximum power point
345–365.
tracking using a fuzzy logic control scheme. Re-
Bhattacharyya, S., Dutta, P., & Maulik, U. (2007). vue des Energies Renouvelables, 10(3), 387–395.
Binary object extraction using bidirectional self-
Chen, C.-T., Chen, K.-S., & Lee, J.-S. (2003).
organizing neural network (BDSONN) architec-
Fuzzy neural classification of SAR images.
ture with fuzzy context sensitive thresholding.
IEEE Transactions on Geoscience and Re-
Pattern Analysis & Applications, 345–360.
mote Sensing, 41(9), 2089–2100. doi:10.1109/
doi:10.1007/s10044-007-0072-z
TGRS.2003.813494
Boskovitz, V., & Guterman, H. (2002). An
Cox, E. (1994). The fuzzy systems handbook.
adaptive neuro-fuzzy system for automatic im-
Cambridge, MA: Academic Press.
age segmentation and edge detection. IEEE
Transactions on Fuzzy Systems, 10(2), 247–262. Das, K., Roy, T. K., & Maiti, M. (2003). Multi
doi:10.1109/91.995125 objective fuzzy inventory model for deteriorating
items with shortages over a finite time. Horizon,
Brox, P., Baturone, I., & Sánchez-Solano, S.
40(4), 252–277.
(2006). A fuzzy edge-dependent interpolation al-
gorithm. In Soft computing in image processing: De, S., Bhattacharyya, S., & Dutta, P. (2008).
Recent advances. Springer Verlag International OptiMUSIG: An optimized gray level image
Series Studies in Fuzziness and Soft Computing. segmentor. Proceedings of Sixteenth Interna-
tional Conference on Advanced Computing and
Brox, P., Baturone, I., Sánchez-Solano, S.,
Communication (ADCOM 2008) (pp. 78-87),
Gutiérrez-Ríos, J., & Fernández-Hernández, F.
Chennai, India.
(2006). Fuzzy motion adaptive algorithm for video
de-interlacing. Proceedings of Tenth International De, S., Bhattacharyya, S., & Dutta, P. (2009). A
Conference on Knowledge-Based & Intelligent differential evolution algorithm based automatic de-
Information & Engineering Systems (KES2006) termination of optimal number of clusters validated
Bournemouth. by fuzzy intercluster hostility index. Proceedings of
first IEEE International Conference on Advanced
Chang, C. L., & Ching, Y. T. (2002). Fuzzy
Computing (ICAC09) (pp. 105-111), Chennai, India.
Hopfield neural network with fixed weight for
medical image segmentation. Optical Engi- De, S., Bhattacharyya, S., & Dutta, P. (2010).
neering (Redondo Beach, Calif.), 41, 351–358. Efficient gray level image segmentation using an
doi:10.1117/1.1428298 optimized MUSIG (OptiMUSIG) activation func-
tion. International Journal of Parallel, Emergent
Charalampidis, D., Kasparis, T., & Georgiopoulos,
and Distributed Systems, 1-39.
M. (2001). Classification of noisy signals using
fuzzy ARTMAP neural networks. IEEE Trans- De Luca, A., & Termini, S. (1972). A Definition
actions on Neural Networks, 12(5), 1023–1036. of a nonprobabilistic entropy in the setting of
doi:10.1109/72.950132 fuzzy set theory. Information and Control, 20,
301–312. doi:10.1016/S0019-9958(72)90199-4

65
Fuzzy Logic

De Luca, A., & Termini, S. (1974). Entropy of L- Gomperts, A., Ukil, A., & Zurfluh, F. (2011). De-
fuzzy sets. Information and Control, 24, 55–73. velopment and implementation of parameterized
doi:10.1016/S0019-9958(74)80023-9 FPGA-based general purpose neural networks
for online applications. IEEE Transactions on
De Luca, A., & Termini, S. (1988). Vagueness in
Industrial Informatics, 7(1), 78–89. doi:10.1109/
scientific theories. In Systems and control encyclo-
TII.2010.2085006
pedia, (pp. 4993-4996).
Grabot, B., Geneste, L., Castillo, G. R., & Verot,
De Luca, A., & Termini, S. (1988). Entropy mea-
S. (2005). Integration of uncertain and imprecise
sures in fuzzy set theory. In Systems and control
orders in the MRP method. Intelligent Manufac-
encyclopedia, (pp. 1467-1473).
turing, 16, 215–234. doi:10.1007/s10845-004-
Dubois, D., & Prade, H. (1980). Fuzzy sets and 5890-x
systems: Theory and applications. New York, NY:
Gutierrez-Rios, J., Fernandez-Hernandez, F.,
Academic Press.
Crespo, J. C., & Trivino, G. (2004). Motion adap-
Ebanks, B. R. (1983). On measures of fuzziness tive fuzzy video de-interlacing method based on
and their representations. Journal of Math- convolution techniques. Proceedings of Informa-
ematical Analysis and Applications, 94, 24–37. tion Processing and Management of Uncertainty
doi:10.1016/0022-247X(83)90003-3 in Knowledge-Based Systems.
Egusa, Y., Akahori, H., Morimura, A., & Wakami, Itoh, O., Gotoh, K., Nakayama, T., & Takam-
N. (1992). An electronic video camera image stabi- izawa, S. (1987). Application of fuzzy control to
lizer operated on fuzzy theory. IEEE International activated sludge process. Proceedings of Second
Conference on Fuzzy Systems (pp. 851–858). San IFSA Congress (pp. 282-285). Tokyo, Japan.
Diego, CA: IEEE Press.
Joshi, A., Ramakrishman, N., Houstis, E. N., &
Er, M. J., & Zhou, Y. (2008). A novel framework Rice, J. R. (1997). On neurobiological, neuro-
for automatic generation of fuzzy neural networks. fuzzy, machine learning, and statistical pattern rec-
Neurocomputing, 71, 584–591. doi:10.1016/j.neu- ognition techniques. IEEE Transactions on Neural
com.2007.03.015 Networks, 8(1), 18–31. doi:10.1109/72.554188
Fan, J.-L., & Ma, Y.-L. (2002). Some new fuzzy Kahraman, C. (2007). Fuzzy set applications in
entropy formulas. Fuzzy Sets and Systems, 128, industrial engineering. Information Sciences,
277–284. doi:10.1016/S0165-0114(01)00127-0 177, 1531–1532. doi:10.1016/j.ins.2006.09.010
Feriadi, H., & Hien, W. N. (2003). Modeling thermal Karr, C. L., & Gentry, E. J. (1993). Fuzzy control
comfort for tropics using fuzzy logic. Proceedings of of pH using genetic algorithms. IEEE Transac-
Eighth IBRSA Conference, Eindhoven, Netherlands. tions on Fuzzy Systems, 1(1), 46–53. doi:10.1109/
TFUZZ.1993.390283
Ghosh, A., Pal, N. R., & Pal, S. K. (1993). Self-
organization for object extraction using a multilayer Kasai, Y., & Morimoto, Y. (1988). Electronically
neural network and fuzziness measures. IEEE controlled continuously variable transmission.
Transactions on Fuzzy Systems, 1(1), 54–68. Proceedings of International Congress on Trans-
doi:10.1109/TFUZZ.1993.390285 portation Electronics, Dearborn, MI.
Giarratano, J., & Riley, G. (1993). Expert sys- Kaufmann, A., & Gupta, M. M. (1985). Intro-
tems: Principles and programming. Boston, MA: duction to fuzzy arithmetic. New York, NY: Van
PWS-Kent. Nostrand.

66
Fuzzy Logic

Kim, J., Cai, W., Eberl, S., & Feng, D. (2007). Majumdar, A., & Ghosh, A. (2008). Yarn strength
Real-time volume rendering visualization of dual- modeling using fuzzy expert system. Journal
modality PET/CT images with interactive fuzzy of Engineered Fibers and Fabrics, 3(4), 61–68.
thresholding segmentation. IEEE Transactions
Majumdar, A., Majumdar, P. K., & Sarkar, B.
on Information Technology in Biomedicine, 11(2),
(2005). Application of an adaptive neuro-fuzzy
161–169. doi:10.1109/TITB.2006.875669
system for the prediction of cotton yarn strength
Kinoshita, M., Fukuzaki, T., Satoh, T., & Miyake, from HVI fibre properties. Journal of Textiles
M. (1988). An automatic operation method for Institute, 96, 55–60. doi:10.1533/joti.2004.0062
control rods in BWR plants. Proceedings of Spe-
Mamdani, E. H. (1977). Applications of fuzzy log-
cialists’ Meeting on In-Core Instrumentation and
ic to approximate reasoning using linguistic syn-
Reactor Core Assessment, Cadarache, France.
thesis. IEEE Transactions on Computers, 26(12),
Kosko, B. (1991). Neural networks and fuzzy 1182–1191. doi:10.1109/TC.1977.1674779
systems. Englewood Cliffs, NJ: Prentice Hall.
Meier, R., Nieuwland, J., Zbinden, A., & Hacisali-
Kosko, B. (1992). Fuzzy systems as universal hzade, S. (1992). Fuzzy logic control of blood pres-
approximators. Proceedings of the IEEE Interna- sure during anesthesia. IEEE Control Systems, 12-17.
tional Conference on Fuzzy Systems (pp. 1151-
Mendel, J. M. (1995). Fuzzy logic systems for
1162). San Diego, CA.
engineering: A tutorial. Proceedings of the IEEE,
Kosko, B. (1994). Fuzzy systems as universal 83(3), 345–377. doi:10.1109/5.364485
approximators. IEEE Transactions on Comput-
Meng, D., & Xu, Y. (2008). Lecture Notes in
ers, 43(11), 1329–1333. doi:10.1109/12.324566
Computer Science: Vol. 5009. Credit rating method
Kosko, B. (1997). Fuzzy engineering. Upper with heterogeneous information. Rough Sets and
Saddle River, NJ: Prentice Hall. Knowledge Technology (pp. 739–746).
Kyoomarsi, F., Khosravi, H., Eslami, E., & Deh- Muhammed, H. H. (2003). Unsupervised fuzzy
kordy, P. K. (2009). Optimizing machine learning clustering and image segmentation using weighted
approach based on fuzzy logic in text summariza- neural networks. Proceedings of the Twelfth In-
tion. International Journal of Hybrid Information ternational Conference on Image Analysis and
Technology, 2(2), 105–116. Processing (ICIAP’03) (pp. 308).
Lah, M. T., Zupancic, B., Peternelj, J., & Krainer, Murrill, P. W. (1991). Fundamentals of process
A. (2006). Daylight illuminance control with fuzzy control theory (2nd ed.). Research Triangle Park,
logic. Solar Energy, 80, 307–321. doi:10.1016/j. NC: Instrument Society of America.
solener.2005.02.002
Nanda, S. R., Mahanty, B., & Tiwari, M. K. (2010).
Larsen, P. M. (1980). Industrial applications of Clustering Indian stock market data for portfolio
fuzzy logic control. International Journal of management. Expert Systems with Applications,
Man-Machine Studies, 12(1), 3–10. doi:10.1016/ 37(12). doi:10.1016/j.eswa.2010.06.026
S0020-7373(80)80050-2
Niu, J., & Dartnall, J. (2008). Application of fuzzy-
Lee, C. C. (1990). Fuzzy logic in control systems: MRP-II in fast moving consumer goods manufacturing
Fuzzy logic controller – Part I. IEEE Transac- industry. In S. J. Mason, R. R. Hill, L. Mönch, O. Rose,
tions on Systems, Man, and Cybernetics, 20(2), T. Jefferson, & J. W. Fowler (Eds.), Proceedings of the
404–418. doi:10.1109/21.52551 2008 Winter Simulation Conference (pp. 1939-1945).

67
Fuzzy Logic

Ogunnaike, B. A., & Ray, W. H. (1994). Process Ross, T. J., Booker, J., & Parkinson, J. (2003).
dynamics, modeling, and control. New York, NY: Fuzzy logic and probability applications: Bridg-
Oxford University Press. ing the gap, society for industrial and applied
mathematics. Philadelphia, PA.
Osofisan, P. B., & Obafaiye, O. J. (2007).
Fuzzy logic modeling of the fluidized catalytic Schweizer, B., & Sklar, A. (1963). Associative
cracking unit of a petrochemical refinery. The functions and abstract semi-groups. Publ. Math.
Pacific Journal of Science and Technology, Debrecen, 10, 69–81.
8(1), 59–67.
Shingu, T., & Nishimori, E. (1989). Fuzzy-based
Pal, N. R., & Bezdek, J. C. (1994). Measuring automatic focusing system for compact camera.
fuzzy uncertainty. IEEE Transactions on Fuzzy Proceedings of the Third International Fuzzy
Systems, 2, 107–118. doi:10.1109/91.277960 Systems Association Congress (pp. 436–439).
Seattle, WA.
Pal, N. R., & Bezdek, J. C. (2000). Quantifying
different facets of fuzzy uncertainty. In Dubas, D., Shinskey, F. G. (1988). Process control systems
& Prade, H. (Eds.), Fundamentals of fuzzy sets – Applications, design, and tuning (3rd ed.). New
(pp. 459–480). London, UK: Kluwer Academic York, NY: McGraw-Hill.
Publishers.
Shipley, M. F. (2009). Portfolio management:
Pal, S. K., Ghosh, A., & Uma Shankar, B. (2000). A fuzzy set based approach to monitoring size
Segmentation of remote sensing images with to maximize return and minimize risk. World
fuzzy thresholding, and quantitative evaluation. Academy of Science. Engineering and Technol-
International Journal of Remote Sensing, 21(11), ogy, 52, 406–513.
2269–2300. doi:10.1080/01431160050029567
Shitong, W., Duan, F., Min, X., & Dewen, H.
Parkinson, W. J. (2001). Fuzzy and probabilistic (2007). Advanced fuzzy cellular neural network:
techniques applied to the chemical process Indus- Application to CT liver images. Artificial Intel-
tries. PhD dissertation, Department of Electrical ligence in Medicine, 39(1), 65–77. doi:10.1016/j.
and Computer Engineering, University of New artmed.2006.08.001
Mexico, Albuquerque, NM.
Shitong, W., & Min, W. (2006). A new detec-
Passino, K., & Yurkovich, S. (1998). Fuzzy control. tion algorithm (NDA) based on fuzzy cellular
Menlo Park, CA: Addison-Wesley. neural networks for white blood cell detection.
IEEE Transactions on Information Technol-
Phillips, C. L., & Harbor, R. D. (1996). Feedback
ogy in Biomedicine, 10(1), 5–10. doi:10.1109/
control systems (3rd ed.). Englewood Cliffs, NJ:
TITB.2005.855545
Prentice Hall.
Simpson, P. K. (1992). Fuzzy min–max neu-
Quail, S., & Adnan, S. (1992). State of the art in
ral networks—Part 1: Classification. IEEE
household appliances using fuzzy logic. In J. Yen,
Transactions on Neural Networks, 3, 776–786.
R. Langari, & L. Zadeh (Eds.), Proceedings of the
doi:10.1109/72.159066
Second International Workshop – Industrial Fuzzy
Control and Intelligent Systems (pp. 204–213). Sodiya, A. S., Onashoga, S. A., & Oladunjoye,
College Station, TX: IEEE Press. B. A. (2007). Threat modeling using fuzzy logic
paradigm. Issues in Informing Science and In-
Ross, T. J. (2004). Fuzzy logic with engineering
formation Technology, 4, 53–61.
applications. John Wiley.

68
Fuzzy Logic

Sugeno, M. (1977). Fuzzy measures and fuzzy Wang, L.-X., & Mendel, J. M. (1991). Gener-
integrals: A survey. In Gupta, M. M. (Eds.), Fuzzy ating fuzzy rules by learning from examples.
automata and decision processes. New York, NY: Proceedings of the IEEE International Sym-
North-Holland. posium on Intelligent Control (pp. 263–268).
Arlington, VA.
Takagi, H. (1992). Survey: Fuzzy logic applica-
tions to image processing equipment. In J. Yen, Watanabe, H., & Dettloff, W. (1988). Reconfigu-
R. Langari, & L. Zadeh (Eds.), Proceedings of rable fuzzy logic processor: A full custom digital
the Second International Workshop – industrial VLSI. International Workshop on Fuzzy System
Fuzzy control and Intelligent Systems (pp. 1–9). Applications (pp. 49-50). Iizuka, Japan.
College Station, TX: IEEE Press.
Wei, C.-C., & Chang, H.-W. (2011). A new ap-
Takagi, H., & Hayashi, I. (1991). NN-driven fuzzy rea- proach for selecting portfolio of new product
soning. International Journal of Approximate Reason- development projects. Expert Systems with Ap-
ing, 5, 191–212. doi:10.1016/0888-613X(91)90008-A plications, 38(1). doi:10.1016/j.eswa.2010.06.081
Takagi, T., & Sugeno, M. (1985). Fuzzy identifica- Yager, R. (1980). On a general class of fuzzy
tion of systems and its applications to modeling connectives. International Journal of Fuzzy
and control. IEEE Transactions on Systems, Man, Sets and Systems, 4, 235–242. doi:10.1016/0165-
and Cybernetics, 15(1), 116–132. 0114(80)90013-5
Taur, J. S., & Tao, C. W. (2000). A new neuro- Yager, R. (1981). A new methodology for
fuzzy classifier with application to on-line ordinal multiobjective decisions based on
face detection and recognition. The Journal fuzzy sets. Decision Sciences, 12, 589–600.
of VLSI Signal Processing, 26, 397–409. doi:10.1111/j.1540-5915.1981.tb00111.x
doi:10.1023/A:1026515819538
Yager, R., & Filev, D. (1994). Template-based
The MathWorks, Inc. (1995). Fuzzy logic tool- fuzzy systems modeling. Intelligent Fuzzy Sys-
box user’s guide. Retrieved from http://www. tems, 2(1), 39–54.
mathworks.com
Yagishita, O., Itoh, O., & Sugeno, M. (1985).
Togai, M., & Chiu, S. (1987). A fuzzy accelerator Application of fuzzy reasoning to the water pu-
and a programming environment for real-time rification process. In Sugeno, M. (Ed.), Industrial
fuzzy control. Proceedings of Second IFSA Con- applications of fuzzy control (pp. 19–40). Amster-
gress (pp. 147-151). Tokyo, Japan. dam, The Netherlands: North Holland.
Togai, M., & Watanabe, H. (1986). Expert system Yamakawa, T. (1986). High speed fuzzy controller
on a chip: An engine for real time expert reason- hardware system. Proceedings of Second Fuzzy
ing. IEEE Expert Systems Magazine, 1, 55–62. System Symposium (pp. 122-130). Japan.
Van de Ville, D., Rogge, B., Philips, W., & Lema- Yamakawa, T. (1987). Fuzzy controller hardware
hieu, I. (1999). De-interlacing using fuzzy based system. Proceedings of Second IFSA Congress,
motion detection. Proceedings of Third Interna- Tokyo, Japan.
tional Conference on Knowledge-Based Intelligent
Yamakawa, T. (1988). Fuzzy microprocessors –
Information Engineering Systems (pp. 263-267).
Rule chip and defuzzification chip. International
Wang, L. X. (1997). A course in fuzzy systems and Workshop on Fuzzy System Applications (pp. 51-
control. Upper Saddle River, NJ: Prentice Hall. 52). Iizuka, Japan.

69
Fuzzy Logic

Yamakawa, T., & Miki, T. (1986). The current Zadeh, L. A. (1994). Fuzzy logic, neural networks
mode fuzzy logic integrated circuits fabricated by and soft computing. Communications of the ACM,
the standard CMOS process. IEEE Transactions 37, 77–84. doi:10.1145/175247.175255
on Computers, C-35(2), 181–167. doi:10.1109/
Zenzo, S. D., Cinque, L., & Levialdi, S. (1998).
TC.1986.1676734
Image thresholding using fuzzy entropies. IEEE
Yamakawa, T., & Sasai, K. (1987). Fuzzy memory Transactions on Systems. Man and Cybernetics-
device. Proceedings of Second IFSA Congress Part B, 28, 15–23. doi:10.1109/3477.658574
(pp. 551-555). Tokyo, Japan.
Zhao, M. S., Fu, A. M. N., & Yan, H. (2001). A
Yang, Y., Zheng, C., & Lin, P. (2004). Image technique of three level thresholding based on
thresholding based on spatially weighted fuzzy probability partition and fuzzy 3-partition. IEEE
c-means clustering. Proceedings of the Fourth Transactions on Fuzzy Systems, 9, 469–479.
International Conference on Computer and In- doi:10.1109/91.928743
formation Technology (CIT’04) (pp. 184–189).
Zhao, Y., & Xu, L. (2009). Fuzzy comprehensive
Yasunobu, S., & Miyamoto, S. (1985). Automatic evaluation on the credit rating of manufactur-
train operation by predictive fuzzy control. In ing enterprises’ suppliers based on improved
Sugeno, M. (Ed.), Industrial application of fuzzy algorithm. Proceedings of Second International
control (pp. 1–18). Amsterdam, The Netherlands: Conference on Future Information Technology
North Holland. and Management Engineering (FITME ‘09) (pp.
559-563). Sanya.
Yasunobu, S., Miyamoto, S., & Ihara, H. (1983).
Fuzzy control for automatic train operation system. Zong-Mu, Y., & Kuei-Hsiang, L. (2004). A sys-
Proceedings of Fourth IFAC/IFIP/IFORS Inter- tematic approach for designing multistage fuzzy
national Congress on Control in Transportation control systems. Fuzzy Sets and Systems, 143,
Systems, Baden-Baden. 251–273. doi:10.1016/S0165-0114(03)00203-3
Yasunobu, S., Sekino, S., & Hasegawa, T. (1987).
Automatic train operation and automatic crane
operation systems based on predictive fuzzy KEY TERMS AND DEFINITIONS
control. Proceedings of Second IFSA Congress
(pp. 835-838). Tokyo, Japan. Defuzzification: The reverse mapping from
the fuzzy to the crisp domain is referred to as
Zadeh, L. A. (1965). Fuzzy sets. Information defuzzification.
and Control, 8, 338–353. doi:10.1016/S0019- Fuzzy membership function: A fuzzy
9958(65)90241-X membership function characterizes a fuzzy set.
Zadeh, L. A. (1973). Outline of a new approach It describes the behavioral characteristics of the
to analysis of complex systems and decision fuzzy set, its support, its height and its properties.
processes. IEEE Transactions on Systems, Fuzzification: It is a mapping from an input
Man, and Cybernetics, 3, 28–44. doi:10.1109/ crisp/conventional universe of discourse into the
TSMC.1973.5408575 fuzzy interval (0, 1) that describes the membership
of the fuzzy input variable.
Zadeh, L. A. (1978). PRUF-A meaning representa- Fuzzy Control: A fuzzy rule based nonlinear
tion language for natural languages. International control system which is more failsafe and sus-
Journal of Man-Machine Studies, 10, 395–460. ceptible to higher degrees of nonlinear input-out
doi:10.1016/S0020-7373(78)80003-0 relationships.

70
Fuzzy Logic

Fuzzy Inference System: A fuzzy inference Fuzzy Measures: The fuzzy measures provide
system, abbreviated as FIS, mimic human intel- a quantitative value of the degree of fuzziness in
ligence for quantifying the ambiguity/imprecision a fuzzy set.
in real world scenario through proper modeling Fuzzy Operators: These act as linguistic
of the fuzziness of real world data by appropriate hedges on fuzzy sets to modify the behavioral
rule bases. characteristics of fuzzy sets.
Fuzzy Logic: It is a multivalued logic which Fuzzy Sets: These are capable of describing
incorporates all the possible outcomes in an obser- the vagueness and ambiguity inherent in real world
vation apart from the standard bivalent/bivalued data by assigning degrees of containment to the
logic commonly dealt with in the conventional elements of the constituent sets in the universe
computing scenario. of discourses.

71

View publication stats

You might also like