©

P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
MODULE 0
PU C O
N
Dr. R.B.Ghongade,
N G
E- AD
38 E
FUZZY LOGIC & NEURAL NETWORKS
 E
AD
University Syllabus

G
N
• Teaching Scheme

PU C O
E, T GH
– Lectures/Week: 3Hrs

38
– Tutorials/Week: 1Hr
.
• Examination Scheme
E-
C F E .B

– End Semester: 60 Marks

N
R
&

– Unit Test: 20marks

D . O r.

– Attendance: 10 marks
D

– Assignments: 10 marks
BV P f.

– TW & Oral: 50 Marks

D ro

• Credits: 04
T
P
©

U
E
 E
AD
Unit I

G
Fuzzy Logic -I

N
PU C O
• Concept of Fuzzy number, fuzzy set theory
E, T GH
(continuous, discrete), Operations on fuzzy

38
sets, Fuzzy membership functions (core,
.

E-
C F E .B

boundary, support), primary and composite

N
R

linguistic terms, Concept of fuzzy relation,

&
D . O r.

composition operation (T-norm, T-conorm),

D

Fuzzy if-then rules.

BV P f.
D ro

O
T
P
©

U
E
 E
AD
Unit II

G
Fuzzy Logic -II

N
• Fuzzification, Membership Value Assignment
PU C O
E, T GH
techniques, De-fuzzification (Max membership

38
principle, Centroid method, Weighted average
.
method), Concept of fuzzy inference, Implication
E-
C F E .B

rules- Dienes-Rescher Implication, Mamdani

N
R
&

Implication, Zadeh Implication, Fuzzy Inference

D . O r.

systems -Mamdani fuzzy model, Sugeno fuzzy

D

model, Tsukamoto fuzzy model, Implementation

BV P f.

of a simple two-input single output FIS employing

D ro

Mamdani model Computing.

T
P
©

U
E
 E
AD
Unit III

G
Fuzzy Control Systems

N
PU C O
• Assumptions in a Fuzzy Control System Design,
E, T GH
Fuzzy Logic Controllers, Comparison with

38
traditional PID control, advantages of FLC,
.

E-
C F E .B

Architecture of a FLC: Mamdani Type, Example

N
R

Aircraft landing control problem, washing

&
D . O r.

machine and vacuum cleaner.

D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Unit IV

G
Artificial Neural Network -I

N
• Biological neuron, Artificial neuron model, concept of

PU C O
bias and threshold , Mc Culloch-Pits Neuron Model,
E, T GH

38
implementation of logical AND, OR, XOR functions Soft
Topologies of neural networks, learning paradigms:
.

E-
C F E .B

supervised, unsupervised, reinforcement, Linear

N
R

neuron model ,concept of error energy , gradient

&

descent algorithm and application of linear neuron for

D . O r.

linear regression, Activation functions : binary , bipolar

D

(linear, signup, log sigmoid, tan-sigmoid)

BV P f.
D ro

• Learning mechanisms: Hebbian, Delta Rule ,Perceptron

O

and its limitations

T
P
©

U
E
 E
AD
Unit V

G
Artificial Neural Network -II

N
• Multilayer perceptron (MLP) and back

PU C O
propagation algorithm, Application of MLP for
E, T GH

38
• classification and regression, Self-organizing
Feature Maps, k-means clustering, Learning
.

E-
C F E .B

vector N
R

• quantization Radial Basis Function networks:

&
D . O r.

Cover’s theorem, mapping functions (Gaussian,

D

• Multiquadrics, Inverse multi quadrics),

BV P f.

Application of RBFN for classification and

D ro

regression,
T
P

• Hopfield network, associative memories.

©

U
E
 E
AD
Unit VI
Adaptive Neuro-Fuzzy Inference Systems (ANFIS)

G
N
PU C O
• ANFIS architecture, Hybrid Learning
E, T GH
Algorithm, Advantages and Limitations of

38
ANFIS
.

E-
C F E .B

• Application of ANFIS/CANFIS for regression

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Text /ReferenceBooks

G
N
1. L-Fortuna, G. Rizzotto, M. Lavorgna, G.
PU C O
E, T GH
Nunnari, M. G. Xibilia and R.Caponetto , “Soft

38
Computing”, Springer
.

E-
C F E .B

2. James A. Freeman, David M. Skapura,

N
R

“Neural Networks Algorithms, Applications

&

and Programming Techniques”, Pearson

D . O r.D

Education.
BV P f.

3. S. N. Sivanandam, S. N. Deepa,” Principals of

D ro

soft Computing”, Wiley India

T
P
©

U
E
 E
AD
List of Tutorials/Experiments

G
N
1. Study of Fuzzy sets and operations.

PU C O
2. Study of concepts of fuzzy sets core, support, alpha cuts..
E, T GH
3. Study of fuzzy relation, Max-min composition.

38
4. Analyze t-norms and t-conorms.
.
5. Analyze Fuzzy Inference systems -Mamdani fuzzy model, Sugeno fuzzy

E-
C F E .B

model, Tsukamoto fuzzy model.

6. Analyze architecture of a FLC: Mamdani Type with Example Aircraft landing
N
R

control problem, washing machine and vacuum cleaner.

&

7. Study of learning mechanisms, approaches and activation functions in

D . O r.

ANN.
D

8. Study of Multilayer perceptron (MLP) and back propagation algorithm.

BV P f.

9.Study of Radial Basis Function networks.

D ro

10. Study of ANFIS architecture and Hybrid Learning Algorithm

O
T
P
©

Extensive use of MATLAB“!!!

U
E
 E
AD
List of Assignments

G
N
PU C O
1. Implement simple logic network using MP neuron model
2. Implement a simple linear regressor with a single neuron model.
E, T GH
3. Implement and test MLP trained with backpropagation algorithm

38
4. Implement and test RBF network.
.
5. Implement SOFM for character recognition.

E-
C F E .B

6. Perform fuzzy sets operations.

7. Implement fuzzy membership functions (triangular, trapezoidal, gbell, PI, Gamma,
N
R
&

Gaussian).
D . O r.

8. Implement defuzzification (Max-membership principle, Centroid method, Weighted

D

average method)
9. Implement FIS with Mamdani inferencing mechanism.
BV P f.

10. Implement Simulink model for Vacuum cleaner, washing machine using Fuzzy Logic
D ro

tools
T

11. Implement Fuzzy Logic Controller.

P

12. Implement perceptron learning, multilayer feed forward neural networks.

©

U
E
 E
AD
Course objectives (Theory)

G
N
• Introduction to a relatively new computing paradigm

PU C O
for creating intelligent machines useful for solving
E, T GH
complex real world problems

38
• Insight into the tools that make up the soft computing
.

E-
C F E .B

technique: fuzzy logic, artificial neural networks and

hybrid systems
N
R
&

• Awareness of the application areas of soft computing

D . O r.

technique
D

• Provide alternative solutions to the conventional

BV P f.

problem solving techniques in image/ signal

D ro

processing, pattern recognition/classification, control

T

systems
P
©

U
E
 E
AD
Course objectives (Assignments)

G
N
• Use MATLAB“ for creating artificial neural
PU C O
E, T GH
networks for solving basic problems

38
• Understand the concept of decision boundary,
.

E-
C F E .B

threshold and activation functions

N
R
&

• Implement Fuzzy Logic functions and apply

D . O r.D

them for simple problems

BV P f.

• Implement hybrid systems for basic control

D ro

applications
T
P
©

U
E
 E
AD
Pre-requisites for the course

G
N
PU C O
• Basic matrix algebra
E, T GH
• Preliminary calculus

38
Familiarity with MATLAB“
.
•
E-
C F E .B

N
R

• Preliminary knowledge of control theory,

&

digital signal processing and digital image

D . O r.D

processing
BV P f.

• Patience
D ro

O
T
P

• Resilience
©

U
E
 E
AD
What would you gain after

G
(successfully)completing the course?

N
• A completely new tool at your disposal to solve a

PU C O
wide variety of real world problems
E, T GH

38
• No matter what your main subject of interest is,
you would have an alternate solution , which may
.

E-
C F E .B

offer more adaptability, resilience and

N
R

optimization
&
D . O r.

• Soft computing domain makes a better choice for

D

students who like software and programming!

BV P f.

• Research area also opens up!

D ro

• Hopefully a lot of marks!

T
P
©

U
E
 E
AD
Reference Books

G
N
1. Introduction to the theory of neural computation, John Hertz, Anders Krogh,

PU C O
Richard Palmer, Addison –Wesley Publishing Company, 1991
E, T GH
2. Neural Networks A comprehensive foundation, Simon Haykin,Prentice Hall
International Inc- 1999.

38
3. Neural and Adaptive Systems: Fundamentals through Simulations, José C.
.
Principe Neil R. Euliano , W. Curt Lefebvre, John-Wiley & Sons, 2000

E-
C F E .B

4. Pattern Classification, Peter E. Hart, David G. Stork Richard O.Duda,Second

Edition,2000
N
R

5. Pattern Recognition, Sergios Theodoridis , Konstantinos Koutroumbas, Fourth

&

Edition, Academic Press, 2008

D . O r.

6. A First Course in Fuzzy Logic, Third Edition, Hung T. Nguyen, Elbert A. Walker,
D

Taylor & Francis Group, LLC, 2008

7. Introduction to Fuzzy Logic using MATLAB, S. N. Sivanandam ,S.Sumathi, S. N.
BV P f.

Deepa, Springer Verlag, 2007

D ro

8. Fundamentals of Neural Networks: Architectures, Algorithms And Applications,

O

Laurene Fausett, Pearson Education, Inc, 2008

T
P

9. Fuzzy Logic With Engineering Applications, Third Edition, Timothy Ross, John
Wiley & Sons,2010
©

10. Neuro-Fuzzy and Soft Computing ,J.S. Jang, C.T. Sun, E. Mizutani, PHI Learning
E

Private Limited
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
MODULE 1
PU C O
N
Dr. R.B.Ghongade,
N G
E- AD
38 E
FUZZY LOGIC & NEURAL NETWORKS
 E
AD
SOFT COMPUTING CONSTITUENTS AND CONVENTIONAL
ARTIFICIAL INTELLIGENCE

G
N
PU C O
• Soft Computing (SC): The symbiotic use of many
E, T GH
emerging problem-solving disciplines.

38
• Soft computing is an emerging approach to computing
.

E-
which parallels the remarkable ability of the human mind
C F E .B

to reason and learn in an environment of uncertainty and

N
R
&

imprecision. (Lotfi A. Zadeh, 1992 )

D . O r.

• Soft computing consists of several computing paradigms,

D

– neural networks
BV P f.
D ro

– fuzzy set theory, approximate reasoning

O
T

– derivative-free optimization methods such as genetic algorithms

P

and simulated annealing

©

U
E
 E
AD
PROBLEM SOLVING TECHNIQUES

G
N
HARD COMPUTING SOFT COMPUTING

PU C O
E, T GH

38
Precise Models Approximate Models
.

E-
C F E .B

N
R
&
D . O r.

Traditional Functional
D

Symbolic Numerical Approximate Approximation

Logic Modeling and
BV P f.

Reasoning and Randomized

Reasoning
D ro

Search Search
O
T
P
©

U
E
 E
AD
G
Methodology Strength

N
Neural network Learning and adaptation

PU C O
E, T GH Knowledge representation via fuzzy if-then
Fuzzy set theory
rules

38
Genetic algorithm and simulated annealing Systematic random search
.

E-
Conventional AI Symbolic manipulation
C F E .B

N
R
&

• The seamless integration of these methodologies forms the

D . O r.

core of soft computing

D

• The synergism allows soft computing to incorporate human

BV P f.

knowledge effectively, deal with imprecision and uncertainty,

D ro

and learn to adapt to unknown or changing environment for

T
P

better performance
©

U
E
 E
AD
An Example

G
• Neural character recognizer

N
and a knowledge base are

PU C O
used together to
determine the meaning of
E, T GH
a hand-written word

38
• The neural character
recognizer generates two
.

E-
C F E .B

possible answers "dog" and

"dag," since the middle
N
R

character could be
&

• either an "o" or "a”

D . O r.

• If the knowledge base

D

provides an extra piece of

information that the given
BV P f.

word is related to animals,

D ro

then the answer "dog" is

T
P

picked up correctly
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
 E
AD
From Conventional AI to Computational Intelligence
• Humans usually employ natural languages in

G
reasoning and drawing conclusions

N
PU C O
• Conventional AI attempts to mimic human intelligent
E, T GH
behavior by expressing it in language forms or

38
symbolic rules
.

E-
C F E .B

• AI manipulates symbols on the assumption that such

N
R

behavior can be stored in symbolically structured

&
D . O r.

knowledge bases-called physical symbol system

D

hypothesis
BV P f.

• Symbolic systems provide a good basis for modeling

D ro

human experts in some narrow problem areas if

T
P

explicit knowledge is available –Eg. an expert system

©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
A typical expert system

38 E
 E
AD
Short comings of symbolicism

G
N
• In practice, the symbolic manipulations limit the

PU C O
situations to which the conventional AI theories can be
E, T GH
applied as knowledge acquisition and representation

38
are difficult tasks
.

E-
C F E .B

• Hence more attention has been directed toward

biologically inspired methodologies such as brain
N
R
&

modeling, evolutionary algorithms, and immune

D . O r.

modeling; they simulate biological mechanisms

D

responsible for generating natural intelligence

• These methodologies are somewhat orthogonal to
BV P f.
D ro

conventional AI approaches and generally compensate

O

for the shortcomings of symbolicism

T
P
©

U
E
 E
AD
• The long-term goal of AI research is the

G
creation and understanding of machine

N
intelligence

PU C O
• Soft computing shares the same ultimate goal
E, T GH

38
with AI
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E

An intelligent system
 E
AD
NEURAL NETWORKS

G
N
DARPA Neural Network Study (1988, AFCEA International

PU C O
Press, p. 60):
E, T GH

38
... a neural network is a system composed of many simple processing
elements operating in parallel whose function is determined by network
.

E-
C F E .B

structure, connection strengths, and the processing performed at

computing elements or nodes.
N
R
&
D . O r.D
BV P f.
D ro

An artificial neuron
O
T
P
©

U
E
 E
AD
DEFINITIONS OF NEURAL NETWORKS

G
According to Haykin (1994), p. 2:

N
PU C O
A neural network is a massively parallel distributed processor that has a
E, T GH
natural propensity for storing experiential knowledge and making it

38
available for use. It resembles the brain in two respects:
.

E-
C F E .B

• Knowledge is acquired by the network through a learning process.

N
R
&

Interneuron connection strengths known as synaptic weights are

D . O r.

•
D

used to store the knowledge

BV P f.
D ro

O
T
P
©

U
E
 E
AD
According to Nigrin (1993), p. 11:

G
N
A neural network is a circuit composed of a very large number of

PU C O
simple processing elements that are neurally based. Each element
operates only on local information.
E, T GH

38
Furthermore each element operates asynchronously; thus there is no
.

E-
C F E .B

overall system clock.

N
R
&
D . O r.

According to Zurada (1992):

D

Artificial neural systems, or neural networks, are physical cellular

BV P f.

systems which can acquire, store and utilize experiential knowledge.

D ro

O
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
A multi-layered neural network
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
MULTIDISCIPLINARY VIEW OF NEURAL NETWORKS
 E
AD
FUZZY LOGIC

G
N
• Origins: Multivalued Logic for treatment of imprecision

PU C O
and vagueness
E, T GH

38
– 1930s: Post, Kleene, and Lukasiewicz attempted to
.

E-
C F E .B

represent undetermined, unknown, and other possible

intermediate truth-values.
N
R
&
D . O r.

– 1937: Max Black suggested the use of a consistency profile

D

to represent vague (ambiguous) concepts.

BV P f.
D ro

– 1965: Zadeh proposed a complete theory of fuzzy sets

O

(and its isomorphic fuzzy logic), to represent and

T
P

manipulate ill-defined concepts.

©

U
E
 E
AD
FUZZY LOGIC – LINGUISTIC VARIABLES

G
N
– Fuzzy logic gives us a language (with syntax and local

PU C O
semantics) in which we can translate our qualitative
E, T GH

38
domain knowledge.
.

E-
C F E .B

– Linguistic variables to model dynamic systems

N
R
&
D . O r.

– These variables take linguistic values that are

D

characterized by:
BV P f.

• a label - a sentence generated from the syntax

D ro

• a meaning - a membership function determined by a local

T
P

semantic procedure
©

U
E
 E
AD
FUZZY LOGIC – REASONING METHODS

G
N
– The meaning of a linguistic variable may be interpreted as an

PU C O
elastic constraint on its value.
E, T GH

38
– These constraints are propagated by fuzzy inference operations,
.
based on the generalized modus-ponens.

E-
C F E .B

N
R

– An FL Controller (FLC) applies this reasoning system to a

&

Knowledge Base (KB) containing the problem domain heuristics.

D . O r.D

– The inference is the result of interpolating among the outputs of

all relevant rules.
BV P f.
D ro

– The outcome is a membership distribution on the output space,

T
P

which is defuzzified to produce a crisp output.

©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
EVOLUTIONARY PROCESS
GENETIC ALGORITHM

E- AD
38 E
 E
AD
DEFINITION OF GENETIC ALGORITHM

G
N
PU C O
– The genetic algorithm is a probabilistic search
E, T GH
algorithm that iteratively transforms a set (called a

38
population) of mathematical objects (typically
.

E-
C F E .B

fixed-length binary character strings), each with

an associated fitness value, into a new population
N
R
&

of offspring objects using the Darwinian principle

D . O r.

of natural selection and using operations that are

D

patterned after naturally occurring genetic

BV P f.
D ro

operations, such as crossover (sexual

O
T
P

recombination) and mutation.

©

U
E
 E
AD
STEPS INVOLVED IN GENETIC ALGORITHM

G
The genetic algorithms follow the evolution process in the nature to

N
find the better solutions of some complicated problems. Foundations

PU C O
of genetic algorithms are given in Holland (1975) and Goldberg (1989)
E, T GH
books.

38
Genetic algorithms consist the following steps:
.

E-
C F E .B

¾ Initialization
N
R

¾ Selection
&

¾ Reproduction with crossover and mutation

D . O r.D

Selection and reproduction are repeated for each generation until a

BV P f.

solution is reached.
D ro

During this procedure a certain strings of symbols, known as

T
P

chromosomes, evaluate toward better solution.

©

U
E
 E
AD
HYBRID SYSTEMS

G
Hybrid systems enables one to combine various soft computing

N
paradigms and result in a best solution. The major three hybrid

PU C O
systems are as follows:
E, T GH

38
¾ Hybrid Fuzzy Logic (FL) Systems
.

E-
C F E .B

¾ Hybrid Neural Network (NN) Systems

N
R
&

¾ Hybrid Evolutionary Algorithm (EA) Systems

D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
SOFT COMPUTING: HYBRID FL SYSTEMS

G
Approximate Reasoning Functional Approximation/ Randomized

N
Search

PU C O
E, T GH
Probabilistic Multivalued & Neural Evolutionary

38
Models Fuzzy Logics Networks Algorithms
.

E-
Fuzzy Multivalued
C F E .B

Systems N Algebras
R
&

Fuzzy Logic
D . O r.

Controllers
D

HYBRID FL SYSTEMS
BV P f.

NN modified by FS FLC Tuned by NN FLC Generated

D ro

(Fuzzy Neural (Neural Fuzzy and Tuned by EA

T

Systems) Systems)
P
©

U
E

23
 E
AD
SOFT COMPUTING: HYBRID NN SYSTEMS

G
Approximate Reasoning Functional Approximation/ Randomized Search

N
PU C O
E, T GH
Probabilistic Multivalued & Neural Evolutionary

38
Models Fuzzy Logics Networks Algorithms
.

E-
C F E .B

Feedforward Recurrent
NN NN
N
R
&

Single/Multiple
D . O r.

RBF Hopfield SOM ART

Layer Perceptron
D

HYBRID NN SYSTEMS
BV P f.

NN parameters NN topology and/or

D ro

(learning rate h
O

momentum a ) weights
T
P

controlled by FLC generated by EAs

©

U
E

24
 E
AD
SOFT COMPUTING: HYBRID EA SYSTEMS

G
Approximate Reasoning Functional Approximation/ Randomized

N
Search

PU C O
E, T GH
Probabilistic Multivalued & Neural Evolutionary

38
Models Fuzzy Logics Networks Algorithms
.

E-
C F E .B

Evolution Genetic
Strategies
N Algorithms
R
&

Evolutionary Genetic
D . O r.

Programs Programs
D

HYBRID EA SYSTEMS
BV P f.

EA parameters EA-based search EA parameters

D ro

(N, P cr, Pmu) inter-twined with (Pop size, selection)

T
P

controlled by FLC hill-climbing controlled by EA

©

U
E

25
 E
AD
NEURO-FUZZY AND SOFT COMPUTING CHARACTERISTICS

G
N
• Human expertise

PU C O
– SC utilizes human expertise in the form of fuzzy if-then rules, as
E, T GH
well as in conventional knowledge representations, to solve
practical problems

38
• Biologically inspired computing models
.

E-
C F E .B

– Inspired by biological neural networks, artificial neural networks

are employed extensively in soft computing to deal with
N
R

perception, pattern recognition, and nonlinear regression and

&

classification problems
D . O r.

• New optimization techniques

D

– Soft computing applies innovative optimization methods arising

BV P f.

from various sources; they are genetic algorithms (inspired by

D ro

the evolution and selection process), simulated annealing

O

(motivated by thermodynamics), the random search method,

T
P

these optimization methods do not require the gradient vector

©

of an objective function, so they are more flexible in dealing

U
E

with complex optimization problems

 E
AD
NEURO-FUZZY AND SOFT COMPUTING CHARACTERISTICS

G
• Numerical computation

N
– Unlike symbolic AI, soft computing relies mainly on numerical

PU C O
computation. Incorporation of symbolic techniques in soft
E, T GH
computing is an active research area within this field.

38
• New application domains
.
– Because of its numerical computation, soft computing has found

E-
C F E .B

a number of new application domains besides that of AI

approaches. These application domains are mostly computation
N
R

intensive and include adaptive signal processing, adaptive

&

control, nonlinear system identification, nonlinear regression,

D . O r.

and pattern recognition.

D

• Model-free learning
BV P f.

– Neural networks and adaptive fuzzy inference systems have the

D ro

ability to construct models using only target system sample

data. Detailed insight into the target system helps set up the
T
P

initial model structure, but it is not mandatory.

©

U
E
 E
AD
NEURO-FUZZY AND SOFT COMPUTING CHARACTERISTICS

G
N
• Intensive computation

PU C O
– Without assuming too much background knowledge of the
E, T GH
problem being solved, neuro-fuzzy and soft computing rely

38
heavily on high-speed number-crunching computation to
.
find rules or regularity in data sets. This is a common
E-
C F E .B

feature of all areas of computational intelligence

N
R

• Fault tolerance
&
D . O r.

– Both neural networks and fuzzy inference systems exhibit

D

fault tolerance. The deletion of a neuron in a neural

network, or a rule in a fuzzy inference system, does not
BV P f.

necessarily destroy the system. Instead, the system

D ro

continues performing because of its parallel and

redundant architecture, although performance quality
T
P

gradually deteriorates
©

U
E
 E
AD
NEURO-FUZZY AND SOFT COMPUTING CHARACTERISTICS

G
N
• Goal driven characteristics

PU C O
– Neuro-fuzzy and soft computing are goal driven; the path leading from
E, T GH
the current state to the solution does not really matter as long as we

38
are moving toward the goal in the long run. This is particularly true
when used with derivative-free optimization schemes, such as genetic
.

E-
algorithms, simulated annealing, and the random search method.
C F E .B

Domain specific knowledge helps reduces the amount of computation

and search time, but it is not a requirement.
N
R
&

• Real-world applications
D . O r.

– Most real-world problems are large scale and inevitably incorporate

D

built-in uncertainties; this precludes using conventional approaches

that require detailed description of the problem being solved. Soft
BV P f.

computing is an integrated approach that can usually utilize specific

techniques within subtasks to construct generally satisfactory
D ro

solutions to real-world problems

T
P
©

U
E
 E
AD
APPLICATIONS OF SOFT COMPUTING

G
¾ Handwriting Recognition

N
¾ Image Processing and Data Compression

PU C O
¾ Automotive Systems and Manufacturing
E, T GH
¾ Soft Computing to Architecture

38
¾ Decision-support Systems
.
¾ Soft Computing to Power Systems

E-
C F E .B

¾ Neuro Fuzzy systems

N
R

¾ Fuzzy Logic Control

&

¾ Machine Learning Applications

D . O r.

¾ Speech and Vision Recognition Systems

D

¾ Process Control and So on

BV P f.
D ro

O
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N MODULE 7
Dr. R.B.Ghongade,
N G
FUZZY LOGIC-I
E- AD
38 E
 E
AD
Agenda

G
N
PU C O
• Brief review of conventional sets
E, T GH
• Introduction to fuzzy sets

38
.
• Membership functions
E-
C F E .B

N
R

• Operations on fuzzy sets

&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
REVIEW OF CONVENTIONAL SETS

G
• Set

N
– A collection of objects having one or more common

PU C O
characteristics
E, T GH
• Members/Elements

38
– Objects belonging to a set is represented as ‫ܣ א ݔ‬, where ‫ ܣ‬is a
set
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Subset B
– B is said to be a subset of set A, i.e., ‫ ܣ ك ܤ‬iff

G
‫ݕ׊ ܣ א ݕ ֜ ܤ א ݕ‬

N
(‫ ݕ‬belongs to ‫ ܤ‬implies ‫ ݕ‬belongs to ‫ ܣ‬for all ‫)ݕ‬

PU C O
E, T GH
• Proper subset

38
– Set B is said to be proper subset of A iff
.

E-
C F E .B

‫ ݔ׌ ݀݊ܽ ܣ ك ܤ‬, ‫ܤ ב ݔ ݐݑܾ ܣ א ݔ‬

N
R

(‫ ܤ‬is a subset of ‫ ܣ‬and there exists ‫ ݔ‬that belongs to ‫ ܣ‬but

&

does not belong to ‫) ܤ‬

D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Equal sets

G
– Two sets A and B are said to be equal iff
‫ܤ א ݕ׊ ݀݊ܽ ܣ א ݔ׊‬, ‫ݕ = ݔ‬

N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Set operations

G
N
• Intersection operation

PU C O
– For any two sets A and B if ‫ ݔ׌‬common to both A and B, then
E, T GH
‫)ܤ ת ܣ( א ݔ‬

38
where ‫ ת‬denotes the logical intersection operation
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Union Operation
– For any two sets A and B if ‫ݔ׌‬, which is a member of either

G
A or B then ‫ ܤ ׫ ܣ א ݔ‬, where ‫ ׫‬denotes the union

N
operation

PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D

• Universal Set
BV P f.
D ro

– A universal set X is a set that has all possible members of a

T
P

particular domain
©

U
E
 E
AD
Concept of Fuzziness
• When we talk about the real world , the way we describe or quantify the

G
real world , is not precise

N
• E.g., Our description of a person’s height- we use the terms short,

PU C O
medium, tall , which is imprecise
E, T GH
• Hence NOT to be PRECISE is FUZZY!

38
• Fuzzy Logic
.
– The computation that involves the logic of imprecision is much powerful than

E-
C F E .B

computation that is being carried out in a precise manner

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Fuzzy logic is all about the relative importance of precision:
How important is it to be exactly right when a rough answer

G
will do?

N
• Fuzzy logic is a convenient way to map an input space to an

PU C O
output space
E, T GH
• As complexity rises, precise statements lose meaning and

38
meaningful statements lose precision - L. A Zadeh (Father of
.

E-
C F E .B

Fuzzy Logic)
N
R

• So far as the laws of mathematics refer to reality, they are

&

not certain. And so far as they are certain, they do not refer
D . O r.

to reality -Albert Einstein

D

• Fuzzy logic is a fascinating area of research because it does a

BV P f.

good job of trading off between significance and precision -

D ro

something that humans have been managing for a very long

T
P

time
©

U
E
 E
HISTORICAL BACKGROUND

AD
• In viewing the evolution of fuzzy logic, three principal phases

G
may be discerned

N
• The first phase, from 1965 to 1973, was concerned in the

PU C O
main with fuzzification, that is, with generalization of the
E, T GH

38
concept of a set, with two-valued characteristic function
generalized to a membership function taking values in the
.

E-
C F E .B

unit interval or, more generally, in a lattice

N
R

• The basic issues and applications which were addressed

&

were, for the most part, set-theoretic in nature, and logic and
D . O r.D

reasoning were not at the center of the stage

• The second phase, 1973-1999, two key concepts were
BV P f.
D ro

introduced in this paper

O
T
P

(a) the concept of a linguistic variable;

(b) the concept of a fuzzy if-then rule
©

U
E

• Today, almost all applications of fuzzy set theory and fuzzy

logic involve the use of these concepts
 E
AD
• The term fuzzy logic was used for the first time in 1974. Today, fuzzy
logic is used in two different contexts:

G
(a) a narrow sense, in which fuzzy logic, abbreviated as FLn, is a
logical system which is a generalization of multivalued logic;

N
PU C O
(b) a wide sense, in which fuzzy logic, abbreviated as FL, is a
union of FLn, fuzzy set theory, possibility theory, calculus of fuzzy
E, T GH
if-then rules, fuzzy arithmetic and calculus of fuzzy quantifiers

38
• An important development in the evolution of fuzzy logic, marking
.

E-
C F E .B

the beginning of the third phase, 1996 - is the genesis of

N
computing with words and the computational theory of
R
&

perceptions
D . O r.

• Basically, development of computing with words and perceptions

D

brings together earlier strands of fuzzy logic and suggests that

BV P f.

scientific theories should be based on fuzzy logic rather than on

D ro

Aristotelian, bivalent logic, as they are at present

T
P

• A key component of computing with words is the concept of

©

Precisiated Natural Language (PNL)

U
E
 E
AD
CHARACTERISTICS OF FUZZY LOGIC

G
N
Some of the essential characteristics of fuzzy logic relate

PU C O
to the following:
E, T GH

38
• In fuzzy logic, exact reasoning is viewed as a limiting
case of approximate reasoning
.

E-
C F E .B

• In fuzzy logic, everything is a matter of degree

N
R

• In fuzzy logic, knowledge is interpreted a collection of

&

elastic or, equivalently, fuzzy constraint on a collection

D . O r.

of variables
D

• Inference is viewed as a process of propagation of

BV P f.

elastic constraints
D ro

• Any logical system can be fuzzified

T
P
©

U
E
 E
AD
CHARACTERISTICS OF FUZZY SYSTEMS

G
N
PU C O
There are two main characteristics of fuzzy systems
E, T GH
that give them better performance for specific

38
applications:
.

E-
C F E .B

• Fuzzy systems are suitable for uncertain or

N
R

approximate reasoning, especially for the system

&
D . O r.

where a mathematical model that is difficult to

D

derive.
BV P f.
D ro

• Fuzzy logic allows decision making with estimated

O
T
P

values under incomplete or uncertain information

©

U
E
 E
Concept of Fuzzy Number

AD
G
N
PU C O
• Precise zero
E, T GH

38
.

E-
C F E .B

N
R

• Almost zero
&

(Can tolerate a band from -

D . O r.D

1 to +1)
BV P f.
D ro

• Near zero
T

(As we go away from zero towards

P

-2 or +2 , the confidence level of

©

how near to zero , reduces)

E
 E
Fuzzy Sets

AD
• Grade of membership

G
– Every member ‫ ݔ‬of a fuzzy set ‫ ܣ‬is assigned a fuzzy index ߤ஺ (‫ )ݔ‬in the

N
interval [0 , 1], which is called as the grade of membership of ‫ ݔ‬in ‫ܣ‬

PU C O
• In a conventional /crisp set membership grade ߤ஺ (‫ )ݔ‬is either
E, T GH
0 or 1

38
• Fuzzy Sets
.

E-
– A set of ordered pairs , given by
C F E .B

N ‫ݔ = ܣ‬, ߤ஺ ‫ ݔ‬: ‫ܺ א ݔ‬
R

Where X is a universal set and ߤ஺ (‫ )ݔ‬is the grade of membership of the

&

object ‫ ݔ‬in ‫ܣ‬, usually ߤ஺ (‫ )ݔ‬lies in [0 , 1]

D . O r.

• Membership Functions
D

– A membership functionߤ஺ (‫ )ݔ‬is characterized by:

BV P f.

ߤ஺ ‫ ݔ‬: ‫ ݔ‬՜ 0,1 , ‫ܺ א ݔ‬

D ro

Where ‫ ݔ‬is a real number describing the object or its attribute , ܺ is the
T
P

universe of discourse and ‫ܺ ؿ ܣ‬

©

• Membership grade shows how likely the object belongs to the

U
E

set
 E
AD
Comparing Classical and Fuzzy Approach

G
• Classical Approach

N
– Consider a universal set T which stands for temperature

PU C O
– Cold, Normal and Hot are subsets of the universal set T
E, T GH
‫ܶ א ݁ݎݑݐܽݎ݁݌݉݁ݐ = ݈݀݋ܥ‬: ͷԨ < ‫ͳ < ݁ݎݑݐܽݎ݁݌݉݁ݐ‬ͷԨ

38
ܰ‫݈ܽ݉ݎ݋‬
.

E-
C F E .B

= ‫ܶ א ݁ݎݑݐܽݎ݁݌݉݁ݐ‬: ͳͷԨ < ‫ʹ < ݁ݎݑݐܽݎ݁݌݉݁ݐ‬ͷԨ

‫ܶ א ݁ݎݑݐܽݎ݁݌݉݁ݐ = ݐ݋ܪ‬: ʹͷԨ < ‫͵ < ݁ݎݑݐܽݎ݁݌݉݁ݐ‬ͷԨ
N
R
&

– Notice here that 24.9°C is Normal while 25.1°C is Hot!

D . O r.D

– This implies that the classical sets have rigid boundaries

BV P f.

– Because of this rigidity the expression of the data becomes

D ro

difficult and may even feel absurd to a human being!

T
P
©

U
E
 E
• Fuzzy Approach

AD
– In contrast the fuzzy sets have soft boundaries

G
– One approach to define fuzzy subsets for the universal set T would be

N
• Means that
– If temperature is around 10°C

PU C O
it is cold
E, T GH
– If temperature is around 20°C
it is normal

38
– If temperature is around 30°C
it is hot
.

E-
C F E .B

• The temperature 25°C is a

N
R

member of two fuzzy subsets

&

“Cold” and “Normal” with a

D . O r.

membership grade
D

ࣆࢀ ࡺ࢕࢘࢓ࢇ࢒ = ࣆࢀ ࡴ࢕࢚ = ૙. ૞
BV P f.
D ro

• But if temperature is 28°C then it is more likely a temperature in the

T
P

category of “Hot” , where as temperature of 22°C, belongs more to the

©

category of “Normal”
E

• This is much better way of describing the temperature!

 E
AD
Nomenclature of Fuzzy Sets

G
N
• Let the elements of set ܺ be ‫ݔ‬ଵ , ‫ݔ‬ଶ ,…‫ݔ‬௡

PU C O
• Then the fuzzy set ‫ ܺ ك ܣ‬is denoted by any of
E, T GH

38
the following nomenclatures:
.

E-
C F E .B

1. ࡭ = N
R

࢞૚ , ࣆ࡭ ࢞૚ , ࢞૛ , ࣆ࡭ ࢞૛ , … , ࢞࢔ , ࣆ࡭ ࢞࢔
&
D . O r.

࢞૚ ࢞૛ ࢞࢔
2. ࡭ = , ,…,
D

ࣆ࡭ (࢞૚ ) ࣆ࡭ (࢞૛ ) ࣆ࡭ (࢞࢔ )

BV P f.

ࣆ࡭ (࢞૚ ) ࣆ࡭ (࢞૛ ) ࣆ࡭ (࢞࢔ )

D ro

3. ࡭ =
O

, ,…,
࢞૚ ࢞૛ ࢞࢔
T
P
©

U
E
 E
AD
Typical Membership Functions

G
N
• Once we say that each member in a fuzzy set is

PU C O
associated with a membership function, we must
E, T GH
know how to characterize this membership

38
function
.

E-
C F E .B

• Various membership functions have been

N
R

employed, they are:

&

J-function
D . O r.

1.
D

2. S-function
BV P f.

3. Triangular Function
D ro

–-function
O

4.
T
P

5. Gaussian function
©

U
E
 E
The J-function

AD
• Defined as :

G
0, ‫ݑ‬൑ߙ

N
‫ݑ‬െߙ

PU C O
ߛ ‫ߙ ;ݑ‬, ߚ = ,ߙ < ‫ ݑ‬൑ ߚ
ߚെߙ
E, T GH

38
1, ‫ߚ>ݑ‬
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
The S-function
• Defined as :

G
0, ‫ݑ‬൑ߙ
ଶ

N
‫ݑ‬െߙ
2 ,ߙ < ‫ ݑ‬൑ ߚ

PU C O
ߛെߙ
ܵ ‫ߙ ;ݑ‬, ߚ, ߛ = ଶ
E, T GH ‫ݑ‬െߛ
1െ2 ,ߚ < ‫ ݑ‬൑ ߛ

38
ߛെߙ
.

E-
1, ‫ߛ>ݑ‬
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
The Triangular membership function
• Defined as :

G
0, ‫ݑ‬൑ߙ
‫ݑ‬െߙ

N
,ߙ < ‫ ݑ‬൑ ߚ
ߚെߙ

PU C O
E, T GH Ȧ ‫ߙ ;ݑ‬, ߚ, ߛ = ߛെ‫ݑ‬
,ߚ < ‫ ݑ‬൑ ߛ
ߛെߚ

38
0, ‫ߛ>ݑ‬
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
–-function
• Defined as :

G
0, ‫ݑ‬൑ߙ
‫ݑ‬െߙ

N
,ߙ < ‫ ݑ‬൑ ߚ
ߚെߙ

PU C O
ȫ ‫ߙ ;ݑ‬, ߚ, ߛ, ߜ = 1, ߚ<‫ݑ‬൑ߛ
E, T GH
ߜെ‫ݑ‬

38
,ߛ < ‫ ݑ‬൑ ߜ
ߜെߛ
.

E-
C F E .B

N 0,, ‫ߜ>ݑ‬
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Gaussian function
• Defined as :

G
(‫ ݑ‬െ ݉)ଶ
‫݉ ;ݑ ܩ‬, ߜ = ݁‫ ݌ݔ‬െ

N
2ߜ ଶ

PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Operations on Fuzzy Sets

G
N
• The main feature of operations on fuzzy sets is
PU C O
E, T GH
that unlike conventional sets, operations on fuzzy

38
sets are usually described with reference to
.
membership function
E-
C F E .B

• Thus operations manipulate the membership

N
R
&

function
D . O r.D

• Common operations defined on fuzzy sets are:

BV P f.

1. Intersection or minimum function

D ro

2. Union or maximum function

T
P

3. Fuzzy complementation
©

U
E
 E
AD
1.Fuzzy Intersection

G
ߤ஺‫ת‬஻ ‫ߤ = ݔ‬஺ ‫ߤ ר ݔ‬஻ ‫ = ݔ‬min(ߤ஺ (‫)ݔ‬, ߤ஻ (‫))ݔ‬

N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&

࡭‫࡮ת‬
D . O r.D
BV P f.
D ro

O
T
P

• Overall membership function is the minimum of

©

U
E

the two sets A and B

 E
AD
2.Fuzzy Union

G
ߤ஺‫׫‬஻ ‫ߤ = ݔ‬஺ ‫ߤ ڀ ݔ‬஻ ‫ = ݔ‬max(ߤ஺ (‫)ݔ‬, ߤ஻ (‫))ݔ‬

N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P

• Overall membership function is the maximum of

©

U
E

the two sets A and B

 ©
P
D ro
E
BV P f.
T D

ߤ஺ҧ ‫ݔ‬
D . O r.
U R
C F E .B
O & .
E, T GH ߤ஺ ‫ݔ‬

PU C O
N
ߤ஺ҧ ‫ = ݔ‬1 െ ߤ஺ ‫ݔ‬

N G
E- AD
3.Fuzzy Complementation

38 E
 E
AD
Other Fuzzy Operations

G
N
PU C O
• De Morgan’s Law:
‫ܣ = ܤ ת ܣ‬ҧ ‫ܤ ׫‬ത
E, T GH

38
For fuzzy sets:
.

E-
C F E .B

ߤ஺‫ת‬஻ = ߤ஺ҧ ‫׫‬஻ത = max(1 െ ߤ஺ , 1 െ ߤ஻ )

N
R
&

‫ܣ = ܤ ׫ ܣ‬ҧ ‫ܤ ת‬ത
D . O r.D

For fuzzy sets:

BV P f.

ߤ஺‫׫‬஻ = ߤ஺ҧ ‫ת‬஻ത = min(1 െ ߤ஺ , 1 െ ߤ஻ )

D ro

O
T
P
©

U
E
 E
AD
• Difference:

G
‫ܤځܣ = ܤ|ܣ‬ത

N
PU C O
For fuzzy sets:
E, T GH

38
ߤ஺ ‫ת‬஻ത = min(ߤ஺ , 1 െ ߤ஻ )
.

E-
C F E .B

‫ܣځܤ = ܣ|ܤ‬ҧ
N
R
&
D . O r.

For fuzzy sets:

D

ߤ஻ ‫ת‬஺ҧ = min(ߤ஻ , 1 െ ߤ஺ )
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Properties of Fuzzy Sets

G
N
1. Commutative

PU C O
E, T GH ‫ܣ׫ܤ= ܤ׫ܣ‬
‫ܣתܤ= ܤתܣ‬

38
.
2. Distributive
E-
C F E .B

‫= ܥתܤ ׫ܣ‬ ‫ܥ׫ܣ ת ܤ׫ܣ‬

N
R
&

‫= ܥ׫ܤ תܣ‬ ‫ܥתܣ ׫ ܤתܣ‬

D . O r.D

3. Associativity
‫ܥ׫ܤ ׫ܣ‬ = ‫ܥ׫ ܤ׫ܣ‬
BV P f.
D ro

‫ܥתܤ תܣ‬ = ‫ܥת ܤתܣ‬

T
P
©

U
E
 E
AD
Properties of Fuzzy Sets

G
N
4. Idempotency

PU C O
E, T GH ‫ܣ= ܣ׫ܣ‬
‫ܣ= ܣתܣ‬

38
.

E-
5. Identity
C F E .B

‫ܣ = ܺ ת ܣ ݀݊ܽ ܣ = ߶ ׫ ܣ‬
N
R
&

‫ܺ = ܺ ׫ ܣ ݀݊ܽ ߶ = ߶ ת ܣ‬
D . O r.D

Where ܺ is universal set and ߶ is a NULL set

BV P f.
D ro

O
T
P
©

U
E
 E
AD
Examples

G
• Given two discrete fuzzy sets:
0 1 0.5 0.3 0.2

N
‫ =ܣ‬, , , ,

PU C O
E, T GH 1 2 3 4 5
0 0.5 0.7 0.2 0.4

38
‫ =ܤ‬, , , ,
1 2 3 4 5
.

E-
C F E .B

• Find ‫ܣ‬ҧ , ‫ܤ‬ഥ , ‫ܤ ׫ ܣ‬, ‫ܤ ת ܣ‬, ‫ܤ|ܣ‬, ‫ܣ|ܤ‬

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Conclusion

G
N
PU C O
• Fuzzy membership v/s Probability
E, T GH
– Probability can be defined for an event that can be

38
repeated again and again
.

E-
C F E .B

– Fuzzy membership can be defined for any event,

N
R

the event need not be repeatable

&
D . O r.

• Example: Free-hand circle

D
BV P f.
D ro

O
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N MODULE 8
Dr. R.B.Ghongade,
N G
FUZZY LOGIC-II
E- AD
38 E
 E
AD
Classical and Fuzzy Relations

G
N
• Relation is of fundamental importance in all-

PU C O
engineering, science, and mathematically based fields
E, T GH

38
• Associated with graph theory, a subject of wide impact
in design and data manipulation
.

E-
C F E .B

• Intimately involved in logic, approximate reasoning,

N
classification, rule-based systems, pattern recognition,
R
&

and control
D . O r.

• Relations represent the mapping of the sets

D

• Degrees of association can be represented by

BV P f.

membership grades in a fuzzy relation by membership

D ro

grades in a fuzzy relation in the same way as degrees of

T
P

set membership are represented in the fuzzy set

©

U
E
 E
AD
Cartesian Product of Relation

G
N
• An ordered sequence of n elements is called as ordered

PU C O
n-tuple
E, T GH

38
• The ordered sequence is in the form of ܽଵ , ܽଶ , . . . , ܽ௡
.
• For the crisp sets ‫ܣ‬ଵ , ‫ܣ‬ଶ , . . . , ‫ܣ‬௡ , the set of n-tuples
E-
C F E .B

ܽଵ , ܽଶ , … , ܽ௡ , where ܽଵ ‫ܣ א‬ଵ , ܽଶ ‫ܣ א‬ଶ … , ܽ௡ ‫ܣ א‬௡ ,is

N
R

called the Cartesian product of ‫ܣ‬ଵ , ‫ܣ‬ଶ , . . . , ‫ܣ‬௡

&
D . O r.

• Cartesian product is denoted by ‫ܣ‬ଵ × ‫ܣ‬ଶ ×. . .× ‫ܣ‬௡

D

• The first element in each pair is a member of ‫ ݔ‬and the

BV P f.

second element is a member of ‫ ݕ‬formally

D ro

‫ݔ = ݕ × ݔ‬, ‫ ܻ א ݕ ݀݊ܽ ܺ א ݔ ݕ‬,

T
P

‫ݔ × ݕ ് ݕ × ݔ ݄݊݁ݐ ݕ ് ݔ ݂ܫ‬
©

U
E
 E
AD
Example: The elements in two sets A and B are given as ‫= ܣ‬

G
{0, 1} and ‫݁{ = ܤ‬, ݂, ݃} , find the Cartesian product ‫× ܣ‬
‫ܤ‬, ‫ܣ × ܤ‬, ‫ܣ × ܣ‬, ‫ܤ × ܤ‬

N
PU C O
Solution:
E, T GH
• ‫({ = ܤ × ܣ‬0, ݁), (0, ݂), (0, ݃), (1, ݁), (1, ݂), (1, ݃)}

38
• ‫݁ { = ܣ × ܤ‬, 0 , ݁, 1 , ݂, 0 , ݂, 1 , ݃, 0 , ݃1 },
.

E-
• ‫ܣ = ܣ × ܣ‬ଶ = {(0, 0), (0, 1), (1, 0), (1, 1)},
C F E .B

• ‫ܤ = ܤ × ܤ‬ଶ =
N
R
&

{(݁, ݁), (݁, ݂), (݁, ݃), (݂, ݁), (݂, ݂), (݂, ݃), (݃, ݁), (݃, ݂), (݃, ݃)}.
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Classical Relations

G
N
• A relation among classical sets

PU C O
‫ݔ‬ଵ , ‫ݔ‬ଶ , … , ‫ݔ‬௡ ܽ݊݀ ‫ݕ‬ଵ , ‫ݕ‬ଶ , … , ‫ݕ‬௡ is a subset of the Cartesian
E, T GH
product

38
• Denoted either by ܴ or by the abbreviated form
.

X × Y = (x, y)/x ‫ א‬X, y ‫ א‬Y

E-
C F E .B

• The strength of the relationship between ordered pairs of

N
R
&

elements in each universe is measured by the characteristic

function denoted by ߯, where a value of unity is associated
D . O r.D

with complete relationship and a value of zero is associated

with no relationship
BV P f.

1(‫ݔ‬, ‫ܻ × ܺ א )ݕ‬
D ro

߯௑×௒ ‫ݔ‬, ‫ = ݔ ݂ = ݕ‬ቊ

0(‫ݔ‬, ‫ܻ × ܺ ב )ݕ‬
T
P
©

U
E
 E
AD
• When the universe or the set are finite, a matrix
called as relation matrix can conveniently represent

G
the relation

N
• A two-dimensional matrix represents the binary

PU C O
E, T GH
relation

38
• Example 1: If X = {2, 4, 6} and Y = {p, q, r}, if they
.

E-
C F E .B

both are related to each other entirely, then the

N
R

relation between them can be given by:

&
D . O r.

p q r
D

2 1 1 1
BV P f.

4 1 1 1
D ro

6 1 1 1
T
P
©

U
E
 E
AD
• Example 2: Let R be a relation among the three sets
ܺ = ‫݅݀݊݅ܪ‬, ‫ ݄ݏ݈݅݃݊ܧ‬,

G
ܻ = {‫ݎ݈݈ܽ݋ܦ‬, ‫݋ݎݑܧ‬, ܲ‫݀݊ݑ݋‬, ܴ‫}ݏ݁݁݌ݑ‬

N
ܼ = {‫ܽ݅݀݊ܫ‬, ܰ݁‫݈ܽ݌‬, ܷ݊݅‫ݏ݁ݐܽݐܵ ݀݁ݐ‬, ‫}ܽ݀ܽ݊ܽܥ‬

PU C O
ܴ (‫ݔ‬, ‫ݕ‬, ‫݅݀݊݅ܪ{ = )ݖ‬, ܴ‫ݏ݁݁݌ݑ‬, ‫}ܽ݅݀݊ܫ‬
{‫݅݀݊݅ܪ‬, ܴ‫ݏ݁݁݌ݑ‬, ܰ݁‫}݈ܽ݌‬
E, T GH
{‫݄ݏ݈݅݃݊ܧ‬, ‫ݎ݈݈ܽ݋ܦ‬, ‫}ܽ݀ܽ݊ܽܥ‬

38
{‫݄ݏ݈݅݃݊ܧ‬, ‫ݎ݈݈ܽ݋ܦ‬, ܷ݊݅‫}ݏ݁ݐܽݐܵ ݀݁ݐ‬
.
The relation can be represented as follows:

E-
C F E .B

India Nepal US Canada India Nepal US Canada

N
R
&

Dollar 0 0 0 0 Dollar 0 0 1 1
D . O r.

Euro 0 0 0 0 Euro 0 0 0 0
D

Pound 0 0 0 0 Pound 0 0 0 0
Rupee 1 1 0 0 Rupee 0 0 0 0
BV P f.

Hindi English
D ro

O
T
P
©

U
E
 E
AD
Cardinality of Crisp Relation

G
N
• Suppose ݊ elements of the universe ܺ are
PU C O
related to ݉ elements of the universe ܻ
E, T GH

38
• If the cardinality of ܺ is ݊௫ and the cardinality
.

E-
C F E .B

of ܻ is ݊௬ , then the cardinality of the relation

N
R

ܴ, between these two universe

&
D . O r.

݊௫×௬ = ݊௫ × ݊௬
D

• The cardinality of the power set describing

BV P f.

this relation, ܲ(ܺ × ܻ ) is then

D ro

݊௉ ‫ = ݕ × ݔ‬2௡ೣ௡೤
T
P
©

U
E
 E
AD
Operations on Crisp Relation

G
• Union
ܴ‫߯ = ܵڂ‬ோ‫׫‬ௌ ‫ݔ‬, ‫ ݕ‬: ߯ோ‫׫‬ௌ ‫ݔ‬, ‫߯ ݔܽ݉ = ݕ‬ோ ‫ݔ‬, ‫ ݕ‬, ߯ௌ ‫ݔ‬, ‫ݕ‬

N
PU C O
• Intersection
ܴ ‫߯ = ܵ ת‬ோ‫ת‬ௌ ‫ݔ‬, ‫ ݕ‬: ߯ோ‫ת‬ௌ ‫ݔ‬, ‫߯ ݊݅݉ = ݕ‬ோ ‫ݔ‬, ‫ ݕ‬, ߯ௌ ‫ݔ‬, ‫ݕ‬
E, T GH

38
• Complement
ܴത = ߯ோത ‫ݔ‬, ‫ = ݕ‬1 െ ߯ோ ‫ݔ‬, ‫ݕ‬
.

E-
C F E .B

• Containment
N
R

ܴ ‫߯ = ܵ ؿ‬ோ ‫ݔ‬, ‫ ݕ‬൑ ߯ௌ ‫ݔ‬, ‫ݕ‬

&
D . O r.

Properties of Crisp Relations

D
BV P f.

• The properties of commutativity, associativity, distributivity,

D ro

involution(complementation) , and idempotency for the

T
P

classical sets also hold good for crisp relation

©

U
E
 E
AD
Composition

G
N
• Let ܴ be relation that relates elements from

PU C O
universe ܺ to universe ܻ
E, T GH

38
• Let ܵ be the relation that relates elements from
universe ܻ to universe ܼ
.

E-
C F E .B

• Let ܶ relate the same element in universe that ܴ

N
R

contains to the same elements in the universe ܼ

&

that ܵ contains
D . O r.D

• The two methods of the composition operations

BV P f.

are:
D ro

– Max–min composition
T
P

– Max–product composition
©

U
E
 E
AD
Max–min composition

G
• The max–min composition is defined by the set-

N
theoretic and membership function-theoretic

PU C O
expressions:
E, T GH
ܶ =ܴ‫ܵל‬

38
்߯ ‫ݔ‬, ‫ݔܽ݉ = ݖ‬
ถ min(߯ ோ ‫ݔ‬, ‫ ݕ‬, ߯ ௌ ‫ݕ‬, ‫) ݖ‬
.

E-
C F E .B

N ௬‫א‬௒
R

Max–product composition
&
D . O r.

• The max–product composition is defined by the set-

D

theoretic and membership function-theoretic

BV P f.

expressions:
D ro

ܶ =ܴ‫ܵל‬
T
P

்߯ (‫ݔ‬, ‫ݔܽ݉ = )ݖ‬

ถ (߯ ோ ‫ݔ‬, ‫ ݕ‬ȉ ߯ ௌ ‫ݕ‬, ‫) ݖ‬
©

U
E

௬‫א‬௒
 E
AD
• Example: Using max–min composition find relation between ܴ and ܵ:
‫ݕ‬ଵ ‫ݕ‬ଶ ‫ݕ‬ଷ
‫ݔ‬ଵ 1 1 0

G
ܴ = ‫ݔ‬ଶ 0 0 1
‫ݔ‬ଷ 0 1 0

N
PU C O
E, T GH ‫ݖ‬ଵ ‫ݖ‬ଶ
‫ݕ‬ଵ 0 1

38
ܵ = ‫ݕ‬ଶ 1 0
‫ݕ‬ଷ 1 1
.

E-
C F E .B

Solution:
ʅT (x1, z1) = max (min (1, 0), min (1, 1), min (0, 1))= max [0, 1, 0] = 1,
N
R
&

ʅT (x1, z2) = max (min (1, 1), min (1, 0), min (0, 1))= max [1, 0, 0] = 1,
D . O r.

ʅT (x2, z1) = max (min (0, 0), min (0, 1), min (1, 1))= max [0, 0, 1] = 1,
D

ʅT (x2, z2) = max (min (0, 1), min (0, 0), min (1, 1))= max [0, 0, 1] = 1,
ʅT (x3, z1) = max (min (0, 0), min (1, 1), min (0, 1))= max [0, 1, 0] = 1,
BV P f.

ʅT (x3, z2) = max (min (0, 1), min (1, 0), min (0, 1))= max [0, 0, 0] = 0,
D ro

O
T
P

1 1
‫ = ܵלܴ׵‬1 1
©

1 0
E
 E
AD
Extension principle of Fuzzy Sets

G
• Consider the one to one
mapping from fuzzy set A to

N
fuzzy set B, given by a

PU C O
function ‫ )ݔ(݂ = ݕ‬such that
E, T GH
‫ݕ‬ଵ = ݂ ‫ݔ‬ଵ

38
‫ݕ‬ଶ = ݂ ‫ݔ‬ଶ
.
‫ڭ‬

E-
C F E .B

‫ݕ‬௡ = ݂ ‫ݔ‬௡
N
R
&

• In terms of membership
D . O r.

grades:
D

ߤ஺ (‫ݔ‬ଵ ) ߤ஺ (‫ݔ‬ଶ ) ߤ஺ (‫ݔ‬௡ )

BV P f.

‫=ܣ‬ , ,…,
D ro

‫ݔ‬ଵ ‫ݔ‬ଶ ‫ݔ‬௡

O

ߤ஻ (‫ݕ‬ଵ ) ߤ஻ (‫ݕ‬ଶ ) ߤ஻ (‫ݕ‬௡ )

T
P

‫=ܤ‬ , ,…,
‫ݕ‬ଵ ‫ݕ‬ଶ ‫ݕ‬௡
©

where
E

ߤ஻ (‫ݕ‬௜ ) = ߤ஺ (‫ݔ‬௜ )
 E
AD
Extension principle of Fuzzy Sets

G
• Consider the many to one

N
mapping from fuzzy set A

PU C O
to fuzzy set B, given by a
E, T GH
function ‫)ݔ(݂ = ݕ‬

38
.

E-
C F E .B

N
R
&
D . O r.

ߤ஺ (‫ݔ‬ଵ ) ߤ஺ (‫ݔ‬ଶ ) ߤ஺ (‫ݔ‬௡ )

D

‫=ܣ‬ , ,…,
‫ݔ‬ଵ ‫ݔ‬ଶ ‫ݔ‬௡
BV P f.

ߤ஻ (‫ݕ‬ଵ ) ߤ஻ (‫ݕ‬ଶ ) ߤ஻ (‫ݕ‬௡ )

D ro

‫=ܤ‬ , ,…,
O

‫ݕ‬ଵ ‫ݕ‬ଶ ‫ݕ‬௡

T
P

where
©

U
E

ߤ஻ (‫ݕ‬௜ ) = ݉ܽ‫ߤ ݔ‬஺ ‫ݔ‬௝ : ‫ݔ‬௝ ‫ି ݂ א‬ଵ (‫ݕ‬௜ )

 E
AD
Example: many to one mapping

G
N
• Consider two fuzzy sets A and

PU C O
B with the mapping Aї
E, T GH
• Given

38
0.2 0.6 0.4 0.8 0.9
‫=ܣ‬ , , , ,
.

E-
C F E .B

െ1 1 െ2 2 3
N
R

determine fuzzy set B

&

Answer:
D . O r.

ߤ஻ 1 = max 0.2,0.6 = 0.6

D

ߤ஻ 4 = max 0.4,0.8 = 0.8

BV P f.

ߤ஻ 9 = max 0.9 = 0.9

D ro

0.6 0.8 0.9

T
P

‫= ܣ ݂=ܤ׵‬ , ,
1 4 9
©

U
E
 E
Fuzzy Relations

AD
• ܺ and ܻ are two universal sets. The fuzzy relation ܴ(‫ݔ‬, ‫ )ݕ‬is given as:

G
ߤ (‫ݔ‬, ‫)ݕ‬
ܴ ‫ݔ‬, ‫ = ݕ‬ோ (‫ݔ‬, ‫ܻ × ܺ א )ݕ‬
(‫ݔ‬, ‫)ݕ‬

N
PU C O
Example: Let ܺ = 1,2,3 ܽ݊݀ ܻ = 1,2 . If membership function associated
E, T GH
with each ordered pair (‫ݔ‬, ‫ )ݕ‬is given by ߤோ ‫ݔ‬, ‫ = ݕ‬exp െ ‫ ݔ‬െ ‫ ݕ‬ଶ then

38
derive the fuzzy relation ܴ ‫ݔ‬, ‫ݕ‬
.

E-
C F E .B

Solution: The fuzzy relation ܴ ‫ݔ‬, ‫ ݕ‬can be defined in two ways, one way
using the standard nomenclature is :
N
R
&
D . O r.

మ మ మ మ మ మ
݁ ି(ଵିଵ) ݁ ି(ଵିଶ) ݁ ି(ଶିଵ) ݁ ି(ଶିଶ) ݁ ି(ଷିଵ) ݁ ି(ଷିଶ)
ܴ ‫ݔ‬, ‫= ݕ‬ , , , , ,
D

(1,1) (1,2) (2,1) (2,2) (3,1) (3,2)

1.0 0.43 0.43 1.0 0.16 0.43
BV P f.

‫ݔ ܴ ׵‬, ‫= ݕ‬ , , , , ,
D ro

(1,1) (1,2) (2,1) (2,2) (3,1) (3,2)

T
P
©

U
E
 E
AD
• The second method of defining relation is

G
through the relation matrix

N
PU C O
E, T GH ܻ՜
1 2

38
1 1.0 0.43
.

E-
ܺ՝
C F E .B

N 2 0.43 1.0
R

3 0.16 0.43
&
D . O r.

• Since the membership function describes the

D

closeness between X and Y, it is obvious that a

BV P f.

higher value implies stronger relation

D ro

O
T
P
©

U
E
 E
AD
Fuzzy Relations: Formal Definition

G
N
• Definition: A fuzzy relation is a fuzzy set defined in the

PU C O
Cartesian product of crisp sets ܺଵ , ܺଶ ,…, ܺ௡ , a fuzzy relation
E, T GH
ܴ(‫ݔ‬ଵ , ‫ݔ‬ଶ ,…, ‫ݔ‬௡ ) thus is defined as

38
.

E-
ܴ(‫ݔ‬ଵ , ‫ݔ‬ଶ ,…,
C F E .B

ߤ (‫ ݔ‬, ‫ ݔ‬,…, ‫ݔ‬௡ )

‫ݔ‬௡ )= ோ ଵ ଶ
N
R

(‫ݔ‬ଵ , ‫ݔ‬ଶ ,…, ‫ݔ‬௡ ) ‫ܺ א‬ଵ × ܺଶ × ‫ܺ × ڮ‬௡

(‫ݔ‬ଵ , ‫ݔ‬ଶ ,…, ‫ݔ‬௡ )
&
D . O r.

Where
D

ߤோ : ܺଵ × ܺଶ × ‫ܺ × ڮ‬௡ ՜ [0,1]
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Operations on Fuzzy relations

G
• Fuzzy relations are very important because they can describe

N
interactions between variables

PU C O
• Let ܴ and ܵ be two binary fuzzy relations on ܺ × ܻ
E, T GH
• Intersection:

38
– The intersection of R and S is defined as
.

E-
ܴ ‫ݑ ܵ ר‬, ‫ = ݒ‬min ܴ ‫ݑ‬, ‫ ݒ‬, ܵ ‫ݑ‬, ‫ݒ‬
C F E .B

Note that ܴ: ܺ × ܻ ՜ 0,1

N
R
&

Means that the domain of ܴis the whole Cartesian product ܺ ×

D . O r.

ܻ
D

• Union:
BV P f.

– The union of R and S is defined as

D ro

ܴ ‫ݑ ܵ ש‬, ‫ = ݒ‬max ܴ ‫ݑ‬, ‫ ݒ‬, ܵ ‫ݑ‬, ‫ݒ‬

T
P
©

U
E
 E
AD
• Projection:
– A fuzzy relation ܴ is usually defined in the

G
Cartesian space ܺ × ܻ

N
PU C O
– Often a projection of this relation on any of the
E, T GH
sets ܺ or ܻ may become useful for further

38
information processing
.

E-
– The projection of ܴ(‫ݔ‬, ‫ )ݕ‬on ܺ , denoted by ܴͳ, is
C F E .B

given by:
N
R
&

ߤோଵ ‫ݔܽ݉ = ݔ‬ถ [ߤோ ‫ݔ‬, ‫] ݕ‬

D . O r.

௬‫א‬௒
D

– The projection of ܴ(‫ݔ‬, ‫ )ݕ‬on ܻ , denoted by ܴ2, is

BV P f.

given by:
D ro

ߤோଶ ‫ݔܽ݉ = ݕ‬ถ [ߤோ ‫ݔ‬, ‫] ݕ‬

T
P
©

௫‫א‬௑
U
E
 E
AD
Example of projection

G
• Consider R(x,y) from previous example:

N
PU C O
E, T GH 1 2

38
1 1.0 0.43 1.0 For X-
projection
.

E-
C F E .B

the
2 0.43 1.0 1.0
N maximum
R

value in
&

3 0.16 0.43 0.43 each row is

D . O r.

retained
D
BV P f.

X-projection
D ro

O
T
P

For Y-projection the maximum value

©

1.0 1.0
U

in each column is retained

E

Y-projection
 E
AD
Projection: Formal Definition

G
N
PU C O
• Projection of a fuzzy relation
E, T GH
ܴ(‫ݔ‬ଵ , ‫ݔ‬ଶ , … , ‫ݔ‬௡ )on to ܺ௜ × ܺ௝ × ‫ܺ × ڮ‬௞ for

38
any ݅, ݆ and ݇ in [1, ݊]is defined as a fuzzy
.

E-
C F E .B

relation ܴ௉ where
N
R
&

ߤோು (‫ݔ‬௜ , ‫ݔ‬௝ , … , ‫ݔ‬௞ )

D . O r.

ܴ௉ = ݉ܽ‫ݔ‬
ถ
D

௫ ‫א‬௑ …..௫ ‫א‬௑

‫ݔ‬௜ , ‫ݔ‬௝ , … , ‫ݔ‬௞
೔ ೔ ೖ ೖ
BV P f.
D ro

O
T
P
©

U
E
 E
AD
FUZZY IF-THEN RULES

G
N
• Linguistic Variables
PU C O
E, T GH
– Conventional techniques for system analysis are

38
intrinsically unsuited for dealing with humanistic
.

E-
systems, whose behavior is strongly influenced by
C F E .B

human judgment, perception, and emotions

N
R
&

– Principle of incompatibility:
D . O r.

• "As the complexity of a system increases, our ability to

D

make precise and yet significant statements about its

BV P f.

behavior diminishes until a threshold is reached beyond

D ro

which precision and significance become almost

T
P

mutually exclusive characteristics"

©

U
E
 E
AD
Linguistic variables and other related

G
terminology

N
• A linguistic variable is characterized by a quintuple (࢞,

PU C O
ࢀ(࢞), ࢄ, ࡳ, ࡹ) in which
E, T GH

38
x is the name of the variable
T(x) is the term set of x—that is, the set of its
.

E-
C F E .B

linguistic values or linguistic terms;

N
R

X is the universe of discourse;

&

G is a syntactic rule which generates the terms in

D . O r.

T(x); and
D

M is a semantic rule which associates with each

BV P f.

linguistic value A its meaning M(A), where M(A)

D ro

denotes a fuzzy set in X.

T
P
©

U
E
 E
AD
Example
• If age is interpreted as a linguistic variable, then its term set ܶ(ܽ݃݁)

G
could be

N
ܶ ܽ݃݁ = {‫݃݊ݑ݋ݕ‬, ݊‫݃݊ݑ݋ݕ ݐ݋‬, ‫݃݊ݑ݋ݕ ݕݎ݁ݒ‬, . . ,

PU C O
E, T GH ݈݉݅݀݀݁ ܽ݃݁݀, ݊‫݀݁݃ܽ ݈݁݀݀݅݉ ݐ݋‬, … ,
‫݈݀݋‬, ݊‫݈݀݋ ݐ݋‬, ‫݈݀݋ ݕݎ݁ݒ‬, ݉‫݈݀݋ ݏݏ݈݁ ݎ݋ ݁ݎ݋‬, ݊‫݈݀݋ ݕݎ݁ݒ ݐ݋‬,…,

38
݊‫݃݊ݑ݋ݕ ݕݎ݁ݒ ݐ݋‬, ݊‫}݈݀݋ ݕݎ݁ݒ ݐ݋‬
.

E-
• Where each term in ܶ(ܽ݃݁) is characterized by a fuzzy set of a universe
C F E .B

of discourse ܺ = [0,100]
N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Usually we use “age is young” to denote the assignment of
the linguistic value “young” to the linguistic variable age

G
• By contrast, when age is interpreted as a numerical variable,

N
we use the expression “age = 20” instead to assign the

PU C O
numerical value “20” to the numerical variable age
E, T GH

38
• The syntactic rule refers to the way the linguistic values in
.
the term set ܶ(ܽ݃݁) are generated
E-
C F E .B

• The semantic rule defines the membership function of each

N
R
&

linguistic value of the term set

D . O r.

• The term set consists of several primary terms (young,

D

middle aged, old) modified by the negation ("not") and/or

BV P f.

the hedges (very, more or less, quite, extremely, and so

D ro

forth), and then linked by connectives such as and, or, either,

T
P

and neither.
©

U
E
 E
AD
Concentration and dilation of linguistic values

G
N
• Let A be a linguistic value characterized by a fuzzy set with

PU C O
membership function ߤ஺ (ȉ)
E, T GH
• Then ‫ܣ‬௞ is interpreted as a modified version of the original

38
linguistic value expressed as
.

E-
C F E .B

௞
ߤ஺ (‫)ݔ‬
‫ܣ‬௞ = න
N
R

‫ݔ‬
&

௑
D . O r.

• The operation of concentration is defined as

‫ܣ = ܣ ܱܰܥ‬ଶ
D

• Dilation is expressed by
BV P f.

‫ܣ = ܣ ܮܫܦ‬଴.ହ
D ro

O
T
P

• Conventionally, we take CON(A) and DIL(A) to be the results of

©

applying the hedges “very” and “more or less”, respectively,

U
E

to the linguistic term A

 E
AD
• We can interpret the negation operator NOT

G
and the connectives AND and OR as

N
1 െ ߤ஺ ‫ݔ‬
ܱܰܶ ‫ܣ = ܣ‬ҧ = න
PU C O ‫ݔ‬
E, T GH
௑

38
ߤ஺ (‫ߤٿ)ݔ‬஻ ‫ݔ‬
.

‫ = ܤ ת ܣ = ܤ ܦܰܣ ܣ‬න
E-
C F E .B

N ௑ ‫ݔ‬
R
&

ߤ஺ ‫ߤ ש ݔ‬஻ ‫ݔ‬
D . O r.

‫ = ܤ ׫ ܣ = ܤ ܴܱ ܣ‬න
‫ݔ‬
D

௑
BV P f.

where A and B are two linguistic values whose

D ro

meanings are defined by ߤ஺ (ȉ)and ߤ஻ (ȉ)

T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
One more MF: Generalized bell
 E
AD
Constructing MFs for composite linguistic terms

G
N
PU C O
• Let the meanings of the linguistic terms young and old be
defined by the following membership functions:
E, T GH
1

38
ߤ௬௢௨௡௚ ‫ݔ ݈݈ܾ݁ = ݔ‬, 20,2,0 =
‫ ݔ‬ସ
.

E-
1+
C F E .B

20
1
N
R

ߤ௢௟ௗ ‫ݔ ݈݈ܾ݁ = ݔ‬, 30,3,100 =

&

‫ ݔ‬െ 100 ଺
D . O r.

1+
30
D

Where ‫ ݔ‬is the age of a given person, with the interval [0,100] as
BV P f.

the universe of discourse

D ro

O
T
P
©

U
E
 E
AD
• We can construct MFs for the following composite
linguistic terms:

G
• ݉‫ ݈݀݋ = ݈݀݋ ܮܫܦ = ݈݀݋ ݏݏ݈݁ ݎ݋ ݁ݎ݋‬଴.ହ

N
PU C O
E, T GH 1
଺
‫ ݔ‬െ 100

38
1+
30
.

E-
C F E .B

N=න
‫ݔ‬
R

௑
&
D . O r.

• ݊‫݈݀݋ ת ݃݊ݑ݋ݕ = ݈݀݋ ݐ݋݊ ݀݊ܽ ݃݊ݑ݋ݕ ݐ݋‬

D

1 1
1െ ‫ ר‬1െ
BV P f.

‫ ݔ‬ସ ‫ ݔ‬െ 100 ଺

D ro

1+ 1+
20 30
T
P

=න
©

‫ݔ‬
U
E

௑
 E
AD
• ‫)݃݊ݑ݋ݕ( ת ݃݊ݑ݋ݕ = ݃݊ݑ݋ݕ ݋݋ݐ ݐ݋݊ ݐݑܾ ݃݊ݑ݋ݕ‬ଶ
ଶ

G
1 1
‫ ר‬1െ

N
ସ ସ
‫ݔ‬ ‫ݔ‬
1+ 1+

PU C O
E, T GH 20 20
=න
‫ݔ‬

38
௑
.

E-
C F E .B

N
R
&

• ݁‫= ) ݈݀݋ ܱܰܥ ܱܰܥ(ܱܰܥ = ݈݀݋ ݕ݈݁݉݁ݎݐݔ‬

D . O r.

(( ‫ ݈݀݋‬ଶ )ଶ )ଶ
D

଼
1
BV P f.
D ro

‫ ݔ‬െ 100 ଺
1+
T
P

30
©

=න
U
E

௑ ‫ݔ‬
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
MATLAB DEMO
E- AD
38 E
 E
AD
Fuzzy If-Then Rules

G
N
• A fuzzy if-then rule (also known as fuzzy rule,

PU C O
fuzzy implication, or fuzzy conditional statement)
E, T GH
assumes the form:

38
࢏ࢌ ࢞: ࢏࢙ ࡭ ࢚ࢎࢋ࢔ ࢟ ࢏࢙ ࡮
.

E-
C F E .B

• where A and B are linguistic values defined by

N
R

fuzzy sets on universes of discourse X and Y,

&
D . O r.

respectively
D

• “‫ "ܣ ݏ݅ ݔ‬is called the antecedent or premise,

while "‫ "ܤ ݏ݅ ݕ‬is called the consequence or
BV P f.
D ro

conclusion
T
P

• Abbreviated as ࡭ ՜ ࡮
©

U
E
 E
AD
Fuzzy Intersection Re-visited
• The intersection of two fuzzy sets A and B is specified in general by a

G
function

N
ܶ ‫ ׷‬0,1 × 0,1 ՜ 0,1 ,

PU C O
which aggregates two membership grades as follows:
E, T GH
ߤ஺‫ת‬஻ ‫ߤ ܶ = ݔ‬஺ ‫ ݔ‬, ߤ஻ ‫ߤ = ݔ‬஺ ‫ߤ כ ݔ‬஻ ‫ݔ‬

38
where * is a binary operator for the function T
.
• This class of fuzzy intersection operators, which are usually referred to

E-
C F E .B

as T-norm (triangular norm) operators

N
R

• Four of the most frequently used T-norm operators are:

&

1. ‫݉ݑ݉݅݊݅ܯ‬: ܶ௠௜௡ (ܽ, ܾ) = ݉݅݊(ܽ, ܾ) = ܽ ‫ܾ ר‬

D . O r.

2. ‫ݐܿݑ݀݋ݎ݌ ܿ݅ܽݎܾ݈݁݃ܣ‬: ܶ௔௣ ܽ, ܾ = ܾܽ

D

3. ‫ݐܿݑ݀݋ݎ݌ ݀݁݀݊ݑ݋ܤ‬: ܶ௕௣ (ܽ, ܾ)| = 0 ‫ ܽ( ש‬+ ܾ െ 1)

BV P f.
D ro

ܽ, ݂݅ ܾ = 1
T

4. ‫ݐܿݑ݀݋ݎ݌ ܿ݅ݐݏܽݎܦ‬: ܶௗ௣ ܽ, ܾ = ቐ ܾ, ݂݅ ܽ = 1

P

0, ݂݅ ܽ, ܾ < 1
©

U
E
 E
Fuzzy Union Re-visited

AD
• The fuzzy union operator is specified in general by a

G
function ܵ ‫[ ׷‬0,1] × [0,1] ՜ [0,1]

N
ߤ஺‫׫‬஻ ‫ߤ ܵ = ݔ‬஺ ‫ ݔ‬, ߤ஻ ‫ߤ = ݔ‬஺ ‫ ݔ‬+ ߤ஻ ‫ݔ‬

PU C O
E, T GH
where + is a binary operator for the function S.

38
• This class of fuzzy union operators, which are often
.

E-
referred to as T-conorm (or S-norm) operators
C F E .B

• Four of the most frequently used T-conorm operators are:

N
R
&

1. ‫݉ݑ݉݅ݔܽܯ‬: ܵ௠௔௫ (ܽ, ܾ) = ݉ܽ‫ܽ(ݔ‬, ܾ) = ܽ ‫ܾڀ‬

D . O r.

2. ‫݉ݑݏ ܿ݅ܽݎܾ݈݁݃ܣ‬: ܵ௔௦ ܽ, ܾ = ܽ + ܾ െ ܾܽ

D

3. ‫݉ݑݏ ݀݁݀݊ݑ݋ܤ‬: ܵ௕௦ (ܽ, ܾ)| = 1 ‫ ܽ( ٿ‬+ ܾ )

BV P f.
D ro

ܽ, ݂݅ ܾ = 0
T
P

4. ‫݉ݑݏ ܿ݅ݐݏܽݎܦ‬: ܵௗ௦ ܽ, ܾ = ቐ ܾ, ݂݅ ܽ = 0

©

1, ݂݅ ܽ, ܾ > 1
E
 E
AD
Formalizing ࡭ ՜ ࡮ approach 1

G
N
• ࡭ ՜ ࡮ means “A is coupled with B”
PU C O
E, T GH
• Interpreted as:

38
ߤ஺ (‫ߤ כ )ݔ‬஻ (‫)ݕ‬
.

E-
C F E .B

ܴ =‫ܣ‬՜‫= ܤ×ܣ= ܤ‬න

(‫ݔ‬, ‫)ݕ‬
N
R

௑×௒
&

where * is a T-norm operator and ‫ ܣ‬՜ ‫ ܤ‬is used

D . O r.D

again to represent the fuzzy relation ܴ

BV P f.
D ro

O
T
P
©

U
E
 E
Formalizing ࡭ ՜ ࡮ approach 2

AD
G
• ࡭ ՜ ࡮ means “A entails B”

N
• Entails: “To have, impose, or require as a
PU C O
E, T GH
necessary accompaniment or consequence”

38
• Can be written as four different formulas:
.

E-
C F E .B

ҧ
1. Material implication: ܴ = ‫ ܣ‬՜ ‫ܤڂܣ = ܤ‬
N
R

ҧ
&

2. Propositional calculus: ܴ = ‫ ܣ‬՜ ‫ܣ(ڂܣ = ܤ‬ ‫)ܤ ת‬

D . O r.

3. Extended propositional calculus: ܴ = ‫ ܣ‬՜ ‫= ܤ‬

D

(‫ܣ‬ҧ ‫)ܤ ת‬
ത ‫)ܤ ׫‬
BV P f.
D ro

4. Generalization of modus ponens:

T
P

ߤோ ‫ݔ‬, ‫ߤ|ܿ ݌ݑݏ = ݕ‬஺ (‫ߤ כ )ݔ‬஻ ‫ ݀݊ܽ ݕ‬0 ൑ ܿ ൑ 1

©

U
E
 E
AD
• Based on these two interpretations and various T-

G
norm and T-conorm operators, a number of

N
qualified methods can be formulated to calculate the

PU C O
fuzzy relation ܴ = ‫ ܣ‬՜ ‫ܤ‬
E, T GH

38
• ܴ can be viewed as a fuzzy set with a two-
.

E-
C F E .B

dimensional MF
ߤோ ‫ݔ‬, ‫ߤ ݂ = ݕ‬஺ ‫ ݔ‬, ߤ஻ (‫ܽ ݂ = )ݕ‬, ܾ
N
R
&

with a = ߤ஺ ‫ ݔ‬, b = ߤ஻ ‫ ݕ‬, where the function f,

D . O r.D

called the fuzzy implication function, performs the task

of transforming the membership grades of ‫ ܣ ݊݅ ݔ‬and
BV P f.
D ro

‫ ܤ ݊݅ ݕ‬into those of (‫ݔ‬, ‫ ܣ ݊݅ )ݕ‬՜ ‫ܤ‬.

T
P
©

U
E
 E
AD
Using approach 1

G
• Four different fuzzy relations ‫ ܣ‬՜ ‫ ܤ‬result from employing four

N
of the most commonly used T-norm

PU C O
ఓಲ (௫)‫ר‬ఓಳ (௬)
1. ܴ௠ = ‫׬ = ܤ × ܣ‬௑×௒
E, T GH (௫,௬)
or ݂௠ ܽ, ܾ = ܽ ‫ܾ ר‬

38
ఓಲ (௫) ఓಳ (௬)
2. ܴ௣ = ‫׬ = ܤ × ܣ‬௑×௒ or ݂௣ ܽ, ܾ = ܾܽ
.
(௫,௬)

E-
C F E .B

ఓಲ (௫)ۨఓಳ (௬)
3. ܴ௕௣ = ‫׬ = ܤ × ܣ‬௑×௒ or ݂௕௣ ܽ, ܾ = 0 ‫ ܽ( ר‬+
N
R

(௫,௬)
&

ܾ െ 1)
D . O r.D

ఓಲ (௫)ȉఓ
Ƹ ಳ (௬)
4. ܴௗ௣ = ‫׬ = ܤ × ܣ‬௑×௒ (௫,௬)
or
BV P f.

ܽ, ݂݅ ܾ = 1
D ro

݂ௗ௣ ܽ, ܾ = ܽ ȉƸ ܾ = ቐ ܾ, ݂݅ ܽ = 1
T
P

0, ݂݅ ܽ, ܾ < 1
©

U
E
 E
AD
Using approach 2

G
• Four of the most commonly used fuzzy implication functions:

N
ଵ‫ (ר‬ଵିఓಲ (௫)ାఓಳ (௬))
1. ܴ௔ = ‫ܣ‬ҧ ‫׬ = ܤ ׫‬௑×௒ or ݂௠ ܽ, ܾ = 1 ‫( ר‬1 െ

PU C O
(௫,௬)
ܽ + ܾ)
E, T GH
(ଵିఓಲ (௫))‫(ש‬ఓಲ (௫)‫ ר‬ఓಳ (௬)

38
2. ܴ௠௠ = ‫ܣ‬ҧ ‫׬ = )ܤ ת ܣ( ׫‬௑×௒ (௫,௬)
or
.
݂௠ ܽ, ܾ = (1 െ ܽ) ‫)ܾ ר ܽ( ש‬
E-
C F E .B

(ଵିఓಲ (௫))‫ש‬ఓಳ (௬)

3. ܴ௦ = ‫ܣ‬ҧ ‫׬ = ܤ ׫‬௑×௒
N
or ݂௕௣ ܽ, ܾ = (1 െ ܽ) ‫ש‬
R
&

(௫,௬)
ܾ
D . O r.D

ఓಲ (௫)ழෝ ఓಳ (௬)
4. ܴᇞ = ‫׬‬௑×௒ (௫,௬)
where
BV P f.

1, ݂݅ ܽ ൑ ܾ
D ro

ෝ ܾ = ቐܾ
ܽ<
T
P

, , ݂݅ ܽ > ܾ
ܽ
©

U
E
 E
AD
Fuzzy Reasoning

G
• The basic rule of inference in traditional two-

N
valued logic is modus ponens, according to

PU C O
which we can infer the truth of a proposition B
E, T GH

38
from the truth of A and the implication ‫ ܣ‬՜
.

‫ܤ‬.
E-
C F E .B

N
R

• For instance, if A is identified with "the tomato

&
D . O r.

is red“ and B with "the tomato is ripe," then if

D

it is true that "the tomato is red," it is also true

BV P f.

that "the tomato is ripe."

D ro

O
T
P
©

U
E
 E
AD
premise 1 (fact): ‫ܣ ݏ݅ ݔ‬
premise 2 (rule): ݂݅ ‫ܤ ݏ݅ ݕ ݄݊݁ݐ ܣ ݏ݅ ݔ‬

G
consequence (conclusion):‫ܤ ݏ݅ ݕ‬

N
PU C O
• Much of human reasoning, modus ponens is employed in an approximate
E, T GH
manner

38
• For example, if we have the same implication rule "if the tomato is red, then it
.
is ripe" and we know that "the tomato is more or less red," then we may infer

E-
C F E .B

that "the tomato -is more or less ripe."

N
R

premise 1 (fact): ‫ܣ ݏ݅ ݔ‬Ԣ

&
D . O r.

premise 2 (rule): ݂݅ ‫ܤ ݏ݅ ݕ ݄݊݁ݐ ܣ ݏ݅ ݔ‬

D

consequence (conclusion):‫’ܤ ݏ݅ ݕ‬
BV P f.
D ro

• Where A' is close to A and B' is close to B

T
P

• When A, B, A', and B' are fuzzy sets of appropriate universes, this inference procedure is
©

called approximate reasoning or fuzzy reasoning; it is also called generalized modus

U
E

ponens(GMP)
 E
AD
Approximate reasoning (fuzzy reasoning)

G
• Let A, A', and B be fuzzy sets of X, X, and Y,

N
respectively

PU C O
• Assume that the fuzzy implication ‫ ܣ‬՜ ‫ ܤ‬is
E, T GH

38
expressed as a fuzzy relation R on ܺ × ܻ
.

E-
C F E .B

• Then the fuzzy set B induced by "x is A'" and

the fuzzy rule "if x is A then y is B” is defined
N
R
&

by
D . O r.D

ߤ஻ᇲ (‫ݔܽ݉ = )ݕ‬௫ min[ߤ஺ᇲ ‫ ݔ‬, ߤோ ‫ݔ‬, ‫] ݕ‬

BV P f.

= ‫ڀ‬௫ [ߤ஺ᇲ ‫ߤ ר ݔ‬ோ ‫ݔ‬, ‫] ݕ‬

D ro

Or
T
P

‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ܴ ל‬
©

U
E
 E
AD
Computational aspects of the fuzzy reasoning

G
N
PU C O
• Single Rule with Single Antecedent
E, T GH
Given as

38
ߤ஻ᇲ (‫ש[ = )ݕ‬௫ (ߤ஺ᇲ ‫ߤ ר ݔ‬஺ ‫ߤ ר ]) ݔ‬஻ ‫ݕ‬
.

E-
C F E .B

N = ‫ߤ ר ݓ‬஻ ‫ݕ‬
R

• ‫ ݓ‬represents a measure of degree of belief for

&
D . O r.

the antecedent part of a rule

D

• This measure gets propagated by the if-then rules

BV P f.
D ro

and the resulting degree of belief or MF for the

T
P

consequent part should not be greater than ‫ݓ‬

©

U
E
 E
AD
Graphic interpretation of GMP using
Mamdani's fuzzy implication

G
N
and the max-min composition
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Single Rule with Multiple Antecedents
• A fuzzy if-then rule with two antecedents is

G
usually written as “if x is A and y is B then z is C”

N
PU C O
• The corresponding problem for GMP is expressed
E, T GH
as

38
premise 1 (fact): ‫ܣ ݏ݅ ݔ‬Ԣ and y is B’
.

E-
C F E .B

premise 2 (rule): ݂݅ ‫ܤ ݏ݅ ݕ ݀݊ܽ ܣ ݏ݅ ݔ‬, ‫ܥ ݏ݅ ݖ ݄݊݁ݐ‬

N
R

consequence (conclusion): z ࢏࢙ ࡯ ’
&
D . O r.

• We then have
D

ߤ஼ ᇲ ‫ש = ݖ‬௫ [ߤ஺ᇲ ‫ߤ ר ݔ‬஺ ‫ש ר ] ݔ‬௬ ߤ஻ᇲ ‫ߤ ר ݕ‬஻ ‫ݕ‬ ‫ߤ ר‬஼ (‫)ݖ‬

BV P f.

௪ଵ ௪ଶ
D ro

= ‫ߤ ר ʹݓ ר ͳݓ‬஼ (‫)ݖ‬
T
P

ிூோூேீ
ௌ்ோாேீ்ு
©

where ‫ ͳݓ‬and ‫ ʹݓ‬are the maxima of the MFs of ‫ܣ ת ܣ‬Ԣ and ‫ת ܤ‬

U
E

‫ܤ‬Ԣ, respectively
 E
AD
G
N
PU C O
E, T GH

38
.

E-
C F E .B

• ࢝૚ denotes the degrees of compatibility between A and

N
R

A'; similarly for ࢝૛

&
D . O r.

• Since the antecedent part of the fuzzy rule is constructed

D

by the connective "and," ࢝૚ ‫࢝ ר‬૛ is called the firing

BV P f.

strength or degree of fulfillment of the fuzzy rule, which

D ro

represents the degree to which the antecedent part of

T
P

the rule is satisfied

©

U
E
 E
AD
• Multiple Rules with Multiple Antecedents
• The interpretation of multiple rules is usually taken as the union

G
of the fuzzy relations corresponding to the fuzzy rules

N
• Therefore, for a GMP problem written as

PU C O
E, T GH
premise 1 (fact): ‫ܣ ݏ݅ ݔ‬Ԣ and y is B’

38
premise 2 (rule 1): ݂݅ ‫ܣ ݏ݅ ݔ‬1 ܽ݊݀ ‫ͳܤ ݏ݅ ݕ‬, ‫ͳܥ ݏ݅ ݖ ݄݊݁ݐ‬
.

E-
C F E .B

premise 3 (rule 2): ݂݅ ‫ܣ ݏ݅ ݔ‬2 ܽ݊݀ ‫ܤ ݏ݅ ݕ‬2, ‫ܥ ݏ݅ ݖ ݄݊݁ݐ‬2

consequence (conclusion): z ࢏࢙ ࡯ ’
N
R
&
D . O r.

• Let ܴଵ = ‫ܣ‬ଵ × ‫ܤ‬ଵ ՜ ‫ܥ‬ଵ and ܴଶ = ‫ܣ‬ଶ × ‫ܤ‬ଶ ՜ ‫ܥ‬ଶ

D

• We simply have
BV P f.

‫ ܥ‬ᇱ = ‫ܣ‬ᇱ × ‫ܤ‬ᇱ ‫ܴ ל‬ଵ ‫ܴ ׫‬ଶ

D ro

= ‫ܣ‬ᇱ × ‫ܤ‬ᇱ ‫ܴ ל‬ଵ ‫ܣ ׫‬ᇱ × ‫ܤ‬ᇱ ‫ܴ ל‬ଶ

T
P

‫ܥ ׵‬Ԣ = ‫ܥ‬ଵ Ԣ ‫ܥ ׫‬ଶ Ԣ

©

U
E
 ©
P
D ro
E

antecedents
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
Fuzzy reasoning for multiple rules with multiple
 E
AD
The process of fuzzy reasoning or approximate
reasoning

G
• Degrees of compatibility

N
PU C O
– Compare the known facts with the antecedents of fuzzy rules to find
the degrees of compatibility with respect to each antecedent MF.
E, T GH

38
• Firing strength
– Combine degrees of compatibility with respect to antecedent MFs in a
.

E-
C F E .B

rule using fuzzy AND or OR operators to form a firing strength that

indicates the degree to which the antecedent part of the rule is
N
R
&

satisfied.
D . O r.

• Qualified (induced) consequent MFs

D

– Apply the firing strength to the consequent MF of a rule to generate a

BV P f.

qualified consequent MF. (The qualified consequent MFs represent

D ro

how the firing strength gets propagated and used in a fuzzy

implication statement.)
T
P

• Overall output MF
©

U
E

– Aggregate all the qualified consequent MFs to obtain an overall output

MF.
 E
AD
Membership Functions-Revisited

G
N
PU C O
• Features of Membership Function
E, T GH
– The feature of the membership function is defined

38
by three properties
.

E-
C F E .B

– They are:N
R
&

1. Core
D . O r.

2. Support
D

3. Boundary
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Core

G
N
PU C O
• If the region of universe is characterized by full
E, T GH
membership (1) in the set A then this gives

38
the core of the membership function of fuzzy
.

E-
C F E .B

at A N
R
&

• The elements, which have the membership

D . O r.

function as 1, are the elements of the core,

D

i.e., here ʅA(x) = 1

BV P f.
D ro

O
T
P
©

U
E
 E
AD
Support

G
N
PU C O
• If the region of universe is characterized by
E, T GH
nonzero membership in the set A, this defines

38
the support of a membership function for
.

E-
C F E .B

fuzzy set A.
N
R
&

• The support has the elements whose

D . O r.

membership is greater than 0 i.e.,μA (x) > 0.

D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Boundary

G
N
PU C O
• If the region of universe has a nonzero
E, T GH
membership but not full membership, this

38
defines the boundary of a membership; this
.

E-
C F E .B

defines the boundary of a membership

N
R

function for fuzzy set A

&
D . O r.

• The boundary has the elements whose

D

membership is between 0 and 1,

BV P f.
D ro

0 < μA (x) < 1

T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
Features of membership function
 E
AD
Important terms

G
N
PU C O
• Crossover Point: The crossover point of a
E, T GH
membership function is the elements in universe

38
whose membership value is equal to 0.5
.

E-
C F E .B

N μA(x) = 0.5
R

• Height: The height of the fuzzy set A is the maximum

&
D . O r.

value of the membership function,

D

݉ܽ‫ ݔ‬P஺ (‫)ݔ‬

BV P f.
D ro

O
T
P
©

U
E
 E
AD
Classification of Fuzzy Sets

G
• The fuzzy sets can be classified based on the membership

N
functions

PU C O
– Normal fuzzy set: If the membership function has at least one
element in the universe whose value is equal to 1, then that set
E, T GH
is called as normal fuzzy set

38
– Subnormal fuzzy set: If the membership function has the
.

E-
C F E .B

membership values less than 1, then that set is called as

subnormal fuzzy set
N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E

Normal fuzzy set Subnormal fuzzy set

 E
AD
• Convex fuzzy set: If the membership function has membership
values those are monotonically increasing, or, monotonically

G
decreasing, or they are monotonically increasing and

N
decreasing with the increasing values for elements in the

PU C O
universe ,the set is called convex fuzzy set
E, T GH
• Nonconvex fuzzy set: If the membership function has

38
membership values which are not strictly monotonically
.

E-
C F E .B

increasing or monotonically decreasing or both monotonically

increasing and decreasing with increasing values for elements
N
R
&

in the universe, then this is called as nonconvex fuzzy set

D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• When intersection is performed on two

G
convex fuzzy sets, the intersected portion is

N
also a convex fuzzy set

PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Fuzzification

G
N
• Fuzzification is the process where the crisp quantities are

PU C O
converted to fuzzy (crisp to fuzzy)
E, T GH
• By identifying some of the uncertainties present in the crisp

38
values, we form the fuzzy values
.

E-
C F E .B

• The conversion of fuzzy values is represented by the

membership functions
N
R
&

• In any practical applications, in industries, etc.,

D . O r.

measurement of voltage, current, temperature, etc., there

D

might be a negligible error

• This causes imprecision in the data which can be
BV P f.

represented by the membership functions

D ro

• Thus fuzzification process may involve assigning

T
P

membership values for the given crisp quantities

©

U
E
 E
AD
Membership Value Assignments

G
N
• Various methods to assign the membership
PU C O
E, T GH
values or the membership functions to fuzzy

38
variables are:
.

E-
C F E .B

– Intuition,
– Inference,
N
R
&

– Rank ordering,
D . O r.D

– Angular fuzzy sets,

– Neural networks,
BV P f.
D ro

– Genetic algorithms, and

T
P

– Inductive reasoning
©

U
E
 E
AD
Intuition

G
• Intuition is based on the human's own intelligence and

N
understanding to develop the membership functions

PU C O
• The thorough knowledge of the problem has to be known, the
E, T GH
knowledge regarding the linguistic variable should also be

38
known
.

E-
C F E .B

• Example: Membership function for imprecision in crisp

temperature reading
N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
Inference

AD
• This method involves the knowledge to perform deductive

G
reasoning

N
• The membership function is formed from the facts known and

PU C O
knowledge
E, T GH
• Let us use inference method for the identification of the

38
triangle
.

E-
C F E .B

• Let U be universe of triangles and A,B, and C be the inner

N
R

angles of the triangles

&

• Also шшш0 therefore the universe is given by:

D . O r.D

ܷ = {(‫ܣ‬, ‫ܤ‬, ‫)ܥ‬, ‫ ܣ‬൒ ‫ ܤ‬൒ ‫ ܥ‬൒ 0, ‫ ܣ‬+ ‫ ܤ‬+ ‫= ܥ‬

180°}
BV P f.
D ro

• There are various types of triangles, for identifying, we define

T
P

three types of triangles:

©

– I:Appropriate isosceles triangle

U
E

– R:Appropriate right triangle

– O:Other triangles
 E
AD
• The membership vales can be inferred to all of these triangle types
through the method of inference, as we know the knowledge about

G
the geometry of the triangles

N
• The membership for the approximate isosceles triangle, for the

PU C O
given conditions
E, T GH
A шшш0 and A + B + C = 180°,

38
is given as:
1
.

E-
C F E .B

ߤூ ‫ܣ‬, ‫ܤ‬, ‫ = ܥ‬1 െ min(‫ ܣ‬െ ‫ܤ‬, ‫ ܤ‬െ ‫)ܥ‬

N 60°
1
R

ߤோ ‫ܣ‬, ‫ܤ‬, ‫ = ܥ‬1 െ (‫ ܣ‬െ 90°)

&

90°
D . O r.

• The membership for the other triangles can be given as the

D

complement of the logical union of the two already defined

BV P f.

membership functions
D ro

ߤை ‫ܣ‬, ‫ܤ‬, ‫ܴ ׫ ܫ = ܥ‬
T
P

i,.e.,
©

ߤை ‫ܣ‬, ‫ܤ‬, ‫ ܫ = ܥ‬ҧ ‫ܴ ת‬ത = ݉݅݊ 1 െ ߤூ (‫ܣ‬, ‫ܤ‬, ‫)ܥ‬, 1 െ ߤோ (‫ܣ‬, ‫ܤ‬, ‫)ܥ‬
U
E
 E
AD
Example C

G
• Define the triangle for the figure shown in figure

N
with the three given angles

PU C O
• Solution: The condition is
E, T GH
‫ܣ‬൒‫ܤ‬൒‫ܥ‬൒0

38
and
.
‫ ܣ‬+ ‫ ܤ‬+ ‫ = ܥ‬180° A

E-
C F E .B

B
Here N
R

ܷ = ‫ = ܣ‬85° ൒ ‫ = ܤ‬60° ൒ ‫ = ܥ‬35° ൒ 0, ‫ ܣ‬+ ‫ ܤ‬+ ‫ = ܥ‬180°

&

The membership for the triangle are

D . O r.

ଵ
1. ߤூ ‫ = ݔ‬1 െ ݉݅݊(‫ ܣ‬െ ‫ܤ‬, ‫ ܤ‬െ ‫)ܥ‬
D

଺଴°
ଵ
=1െ ݉݅݊ 25°, 25° = 0.583
଺଴°
BV P f.

ଵ
2. ߤோ ‫ݔ‬ = 1 െ (‫ ܣ‬െ 90°)
D ro

ଽ଴°
O

ଵ
=1െ 5° = 0.944
ଽ଴°
T
P

3. ߤை ‫(݊݅݉ = ݔ‬1 െ ߤூ (‫)ݔ‬, 1 െ ߤோ (‫)ݔ‬

©

= ݉݅݊ 1 െ 0.583,1 െ 0.944 = 0.056

U
E
 E
AD
Rank Ordering

G
• The polling concept is used to assign membership values by rank

N
ordering process

PU C O
• Preferences are above the pairwise comparisons and from this the
E, T GH
ordering of the membership is done

38
• Example: Suppose 1,000 people responds to a questionnaire about
.

E-
C F E .B

the pairwise preference among five cars, ‫ ݔ‬െ

{‫ܽݎܿ݅ܯ‬, ‫݁ݎ݅ݖ݁ܦ‬, ݅ʹͲ, ‫݋݅ݎܤ‬, ‫}ݕݐ݅ܥ‬.Define a fuzzy set as A on the
N
R
&

universe of cars, “best cars”.

D . O r.D

Preferences
Micra Dezire i20 Brio City Total Percentage Rank
BV P f.

order
D ro

Micra – 515 545 523 671 2,254 22.5 2

Dezire 481 – 475 845 580 2,381 23.8 1
T
P

i20 469 624 – 141 536 1,770 17.7 4

©

Brio 457 530 470 – 649 2,114 21.1 3

U
E

City 265 425 402 389 – 1,481 14.8 5

Total 10,000
 E
AD
• Hence there is highest membership for ߤோ (‫)ݔ‬
• Thus inference method can be used to calculate the membership values

G
• From the table, it is clear that 515 preferred Micra compared to Dezire,

N
545 Astra to Dezire, etc
• The table forms an antisymmetric matrix

PU C O
• There are about ten comparisons made which gives a ground total of
E, T GH
10,000.

38
• Based on preferences, the percentage is calculated
The ordering is then performed
.
•

E-
C F E .B

• It is found that Dezire is selected as the best car

N
R

Fuzzy Set for Best Car

&
D . O r.

1
0.9
D

0.8
0.7
BV P f.

0.6
0.5
P
D ro

0.4
0.3
T
P

0.2
0.1
©

0
U
E

City i20 Brio Micra Dezire

Cars
 E
AD
Angular Fuzzy Sets

G
• The angular fuzzy sets are different from the standard fuzzy
sets in their coordinate description

N
• These sets are defined on the universe of angles, hence are

PU C O
repeating shapes every 2ߨ cycles
E, T GH
• Angular fuzzy sets are applied in quantitative description of

38
linguistic variables known truth-values
.

E-
C F E .B

• When membership of value 1 is true and that of 0 is false,

then in between ‘0’ and ‘1’ is partially true or partially false
N
R

• The linguistic values are formed to vary with T, the angle

&
D . O r.

defined on the unit circle and their membership values are

on ʅ(T)
D

• The membership of this linguistic term can be obtained

BV P f.

from where ‫ ݐ‬is the horizontal projection of the radial

D ro

vector and is given as cosT, i.e., ‫ = ݐ‬cos ߠ

T
P

• When the coordinates are in polar form, angular fuzzy sets

©

can be used
U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
Neural Networks

E- AD
38 E
 E
AD
Defuzzification

G
• Defuzzification means the fuzzy to crisp conversions

N
• The fuzzy results generated cannot be used as such to

PU C O
the applications, hence it is necessary to convert the
E, T GH
fuzzy quantities into crisp quantities for further

38
processing
.

E-
C F E .B

• This can be achieved by using defuzzification process

N
R
&

• The defuzzification has the capability to reduce a fuzzy to

D . O r.

a crisp single-valued quantity or as a set, or converting to

D

the form in which fuzzy quantity is present

BV P f.
D ro

• Defuzzification can also be called as “rounding off”

O
T
P

method
©

• Defuzzification reduces the collection of membership

E

function values in to a single scaler quantity

 E
AD
Lambda Cuts for Fuzzy Sets

G
• Consider a fuzzy set A, then the lambda cut set can
be denoted by ‫ܣ‬ఒ , where ʄranges between 0 and 1

N
PU C O
(0 чʄч1)
E, T GH
• The set ‫ܣ‬ఒ is a crisp set

38
• This crisp set is called the lambda cut set of the fuzzy
.

E-
C F E .B

set A, where
‫ݔ‬
N
R
&

‫ܣ‬ఒ = ൒ߣ
ߤ஺ (‫)ݔ‬
D . O r.D

• The value of lambda cut set is ‫ݔ‬, when the

BV P f.

membership value corresponding to ‫ ݔ‬is greater that

D ro

or equal to the specified ʄ

T
P

• This lambda cut set can also be called as alpha cut

©

set
E
 E
AD
Properties of Lambda Cut Sets

G
N
• There are four properties of the lambda cut sets,

PU C O
they are:
E, T GH

38
1. ‫ ܤ ׫ ܣ‬O = ‫ܣ‬O ‫ܤ ׫‬O
2. ‫ ܤ ת ܣ‬O = ‫ܣ‬O ‫ܤ ת‬O
.

E-
C F E .B

3. ‫ܣ‬ҧ O ് ‫ܣ‬ҧO ݁‫ ݐܽ ݐ݌݁ܿݔ‬O = 0.5

N
R
&

4. &ŽƌĂŶLJʄчɲ͕ǁŚĞƌĞɲǀĂƌŝĞƐďĞƚǁĞĞŶϬĂŶĚϭ͕ŝƚŝƐ
D . O r.

true that, ‫ܣ‬D ‫ܣ ك‬O , where the value of AϬ will be

D

the universe defined

BV P f.

• From the properties it is understood that the

D ro

standard set of operations or fuzzy sets is similar

T
P

to the standard set operations on lambda cut sets

©

U
E
 E
AD
Lambda Cuts for Fuzzy Relations

G
• The lambda cut procedure for relations is similar

N
to that for the lambda cut sets

PU C O
• Consider a fuzzy relation R , in which some of the
E, T GH

38
relational matrix represents a fuzzy set
.

E-
C F E .B

• A fuzzy relation can be converted into a crisp

N
R

relation by depending the lambda cut relation of

&
D . O r.

the fuzzy relation as:

D

‫ݔ‬, ‫ݕ‬
BV P f.

ܴఒ = ൒ߣ
D ro

ߤோ (‫ݔ‬, ‫)ݕ‬
O
T
P
©

U
E
 E
AD
Properties of Lambda Cut Relations

G
N
PU C O
• Properties of the lambda cut relations, are:
E, T GH
1. ܴ ‫ ܵ ׫‬O = ܴ O ‫ܵ ׫‬O

38
2. ܴ ‫ ܵ ת‬O = ܴ O ‫ܵ ת‬O
.

E-
C F E .B

3. ܴത O ് ܴതO
N
R
&

4. For ĂŶLJʄчɲ͕ǁŚĞƌĞɲǀĂƌŝĞƐďĞƚǁĞĞŶϬĂŶĚϭ͕
D . O r.D

then, ܴD ‫ܴ ك‬O
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Defuzzification Methods

G
• There are other various defuzzification methods employed to

N
convert the fuzzy quantities into crisp quantities

PU C O
• The output of an entire fuzzy process can be union of two or more
E, T GH
fuzzy membership functions

38
.
• Generally this can be
E-
C F E .B

given as: N
R

௡
&

‫ܥ‬௡ = ෍ ‫ܥ‬௜ = ‫ܥ‬

D . O r.

௥ ௥
D

௜
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• There are seven methods used for

G
defuzzifying the fuzzy output functions:

N
(1) Max-membership principle,
PU C O
E, T GH
(2) Centroid method,

38
(3) Weighted average method,
.

E-
C F E .B

(4) Mean-max membership,

N
R
&
D . O r.

(5) Centre of sums,

D

(6) Centre of largest area,

BV P f.
D ro

(7) First of maxima or last of maxima

T
P
©

U
E
 E
AD
Max-membership-principle

G
• This method is given by the expression
ߤ஼ ‫ כ ݖ‬൒ ߤ஼ ‫ ݖ‬, ‫ݖ׊‬

N
PU C O
• This method is also referred as height method
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Centroid method

G
• This is the most widely used method

N
• This can be called as center of gravity or center of area

PU C O
methodE, T GH
• It can be defined by the algebraic expression

38
‫כ‬
‫ߤ ׬‬஼ ‫ ݖ‬ȉ ‫ݖ݀ ݖ‬
‫= ݖ‬
.

E-
‫ߤ ׬‬஼ ‫ݖ݀ ݖ‬
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Weighted average method

G
• This method cannot be used for asymmetrical output membership

N
functions, can be used only for symmetrical output membership
functions

PU C O
• Weighting each membership function in the obtained output by its
E, T GH
largest membership value forms this method

38
• The evaluation expression for this method is
σ ߤ஼ (‫ݖ‬௜ ) ȉ ‫ݖ‬௜
.

E-
C F E .B

‫= כݖ‬
N σ ߤ஼ (‫ݖ‬௜ )
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
G
N
PU C O
E, T GH
FUZZY LOGIC-III

38
Fuzzy Inference System
.

E-
C F E .B

N MODULE 9
R
&
D . O r.

Dr. R.B.Ghongade,
D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Fuzzy Inference System

G
N
• Fuzzy inference system is a popular computing framework based on the

PU C O
concepts of fuzzy set theory, fuzzy if-then rules, and fuzzy reasoning
E, T GH
• Its applications include :

38
– automatic control
– data classification
.

E-
C F E .B

– decision analysis
– expert systems
N
R

– time series prediction

&

– Robotics
D . O r.

– pattern recognition
D

• Also called as
– fuzzy-rule-based system
BV P f.

– fuzzy expert system

D ro

– fuzzy model
T
P

– fuzzy associative memory

– fuzzy logic controller
©

– fuzzy system
E
 E
AD
Rule based Systems

G
• Worldly knowledge is very conveniently expressed in natural

N
language

PU C O
• The rule-base is one of the ways to represent knowledge
E, T GH
using natural language

38
• A generic form of rule base is :
.
ࡵࡲ ࢖࢘ࢋ࢓࢏࢙ࢋ ࢇ࢔࢚ࢋࢉࢋࢊࢋ࢔࢚ , ࢀࡴࡱࡺ ࢉ࢕࢔ࢉ࢒࢛࢙࢏࢕࢔[ࢉ࢕࢔࢙ࢋ࢛ࢗࢋ࢔ࢉࢋ]
E-
C F E .B

• This form is commonly referred to as IF-THEN rule-base form

N
R
&

• It typically expresses an inference such that if we know a fact

D . O r.

we can infer or derive another fact

D

• Given a rule, we can derive another rule OR given a rule and

BV P f.

the associated relation and another rule , we can predict what

D ro

should be the consequence

T
P
©

U
E
 E
AD
• Fuzzy information can be represented in the form of a RULE-
BASE which consists of a set of rules in conventional

G
antecedent-consequent form such as:

N
ܴ‫ ݈݁ݑ‬1: ‫ ܣ ݏ݅ ݔ ܨܫ‬, ܶ‫ܤ ݏ݅ ݕ ܰܧܪ‬

PU C O
Where A and B represent fuzzy propositions (sets)
E, T GH
• Now suppose we introduce a new antecedent A’ and consider

38
the rule:
.

E-
ܴ‫ ݈݁ݑ‬2: ‫ܣ ݏ݅ ݔ ܨܫ‬Ԣ , ܶ‫ܤ ݏ݅ ݕ ܰܧܪ‬Ԣ
C F E .B

• From this information derived from Rule 1, is it possible to

N
R
&

derive the consequent in Rule 2, i.e., B’?

D . O r.

• The answer is yes

D

• The consequent B’ is found from the composition equation:

BV P f.

‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ܴ ל‬
D ro

O
T

Where R is the relational matrix

P
©

U
E
 E
Fuzzy Implication Relations

AD
• A fuzzy implication relation for a given rule:

G
ܴ‫ ݈݁ݑ‬1: ‫ ܣ ݏ݅ ݔ ܨܫ‬, ܶ‫ܤ ݏ݅ ݕ ܰܧܪ‬

N
PU C O
is formally denoted by:
E, T GH
ߤோ೔ (‫ݔ‬, ‫)ݕ‬

38
ܴ௜ ‫ݔ‬, ‫= ݕ‬
.

E-
(‫ݔ‬, ‫)ݕ‬
C F E .B

N
R

• ߤோ೔ (‫ݔ‬, ‫ )ݕ‬is computed using various

&
D . O r.

implication rules, ݂݅ ‫ ݌( ݍ ݄݊݁ݐ ݌‬՜ ‫)ݍ‬, where

D

both ‫ ݌‬and ‫ ݍ‬are fuzzy propositions

BV P f.
D ro

‫ ܣ ݏ݅ ݔ ܨܫ‬, ܶ‫ ܤ ݏ݅ ݕ ܰܧܪ‬,
O
T
P

௣ ௤
©

U
E
 E
AD
Dienes-Rescher Implication

G
N
• If ‫ݍ ݄݊݁ݐ ݌‬states that ‫ ݌‬is true but ‫ ݍ‬is false is
PU C O
E, T GH
IMPOSSIBLE, i.e.,‫ ר ݌‬െ‫ ݍ‬is false

38
• Using DeMorgan’s law
.

E-
C F E .B

N ‫ ר ݌‬െ‫ = ݍ‬െ‫ݍ ש ݌‬
R
&

• Thus the relational matrix can be computed

D . O r.D

as:
ࣆࡾ࢏ ࢞, ࢟ = ‫[ܠ܉ܕ‬૚ െ ࣆ࡭࢏ ࢞ , ࣆ࡮࢏ ࢟ ]
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Mamdani Implications
• When fuzzy IF-THEN rules are locally true then using

G
Mamdani implication ‫ ݌‬՜ ‫ ݍ‬implies ‫ݍ ר ݌‬is TRUE

N
• Thus the relational matrix can be computed using any of the

PU C O
following expressions:
E, T GH
ࣆࡾ࢏ ࢞, ࢟ = ‫ ࢞ ࢏࡭ࣆ[ܖܑܕ‬, ࣆ࡮࢏ ࢟ ]

38
.
or

E-
C F E .B

N ࣆࡾ࢏ ࢞, ࢟ = ࣆ࡭࢏ ࢞ ȉ ࣆ࡮࢏ ࢟

R
&

• Remember that each rule is locally true

D . O r.

• Mamdani implication rule is widely used in fuzzy systems and

D

fuzzy control engineering

BV P f.

• Example: If the temperature is HOT, then the fan should run

D ro

FAST
T
P

• This rule does NOT imply , if temperature is COLD then the fan
©

U
E

should run SLOW

 E
AD
Zadeh Implication

G
N
• If ‫ ݌‬՜ ‫ ݍ‬may imply either ‫ ݌‬and ‫ ݍ‬are TRUE or ‫ ݌‬is
PU C O
FALSE
E, T GH

38
• Thus ‫ ݌‬՜ ‫ ש ݍ ר ݌ = ݍ‬െ‫݌‬
.

E-
C F E .B

• The relational matrix can be computed as:

N
R

ࣆࡾ࢏ ࢞, ࢟ = ‫ ࢞ ࢏࡭ࣆ(࢔࢏࢓[ܠ܉ܕ‬, ࣆ࡮࢏ ࢟ ), ૚ െ ࣆ࡭࢏ ࢞ ]

&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Given a set of rules we use various schemes by
which we can construct a relational matrix

G
N
between the antecedent and consequent

PU C O
• The next step would be to utilize this
E, T GH

38
relational matrix for inference
.

E-
C F E .B

• This method is commonly known as

N
R

compositional rule of inference

&
D . O r.

• Associated with each rule we have the

D

relational matrix
BV P f.
D ro

• So given a rule means given a relational

T
P

matrix, and another antecedent we can

©

U
E

compute the consequent

 E
AD
Fuzzy Compositional Rules

G
Following are the different rules for fuzzy composition operation

N
‫ܴלܣ= ܤ‬

PU C O
• max-min
E, T GH
ߤ஻ ‫ݔܽ݉ = ݕ‬௫‫א‬௑ ݉݅݊[ߤ஺ ‫ ݔ‬, ߤோ ‫ݔ‬, ‫] ݕ‬

38
• max-product
.
ߤ஻ ‫ݔܽ݉ = ݕ‬௫‫א‬௑ ߤ஺ ‫ ݔ‬ȉ ߤோ ‫ݔ‬, ‫ݕ‬
E-
C F E .B

• min-max
N
R
&

ߤ஻ ‫݊݅݉ = ݕ‬௫‫א‬௑ ݉ܽ‫ߤ[ݔ‬஺ ‫ ݔ‬, ߤோ ‫ݔ‬, ‫] ݕ‬

D . O r.

• max-max
D

ߤ஻ ‫ݔܽ݉ = ݕ‬௫‫א‬௑ ݉ܽ‫ߤ[ݔ‬஺ ‫ ݔ‬, ߤோ ‫ݔ‬, ‫] ݕ‬

BV P f.

• min-min
D ro

ߤ஻ ‫݊݅݉ = ݕ‬௫‫א‬௑ ݉݅݊[ߤ஺ ‫ ݔ‬, ߤோ ‫ݔ‬, ‫] ݕ‬

T
P

We have to choose these rules by looking at the behaviour of

©

U
E

the specific data

 E
Example

AD
• Given a rule: ‫ܣ ݏ݅ ݔ ܨܫ‬, ܶ‫ܤ ݏ݅ ݕ ܰܧܪ‬, where ‫= ܣ‬

G
଴.ଶ ଴.ହ ଴.଻ ଴.଺ ଴.଼ ଴.ସ
, , ܽ݊݀ ‫= ܤ‬ , ,

N
ଵ ଶ ଷ ହ ଻ ଽ
Infer B’ for another rule: ‫ܣ ݏ݅ ݔ ܨܫ‬Ԣ, ܶ‫ܤ ݏ݅ ݕ ܰܧܪ‬Ԣ, where ‫ܣ‬ᇱ =

PU C O
଴.ହ ଴.ହ ଴.ଷ
, ,
E, T GH
ଵ ଶ ଷ

38
• Solution: using Mamdani implication rule the relational matrix
.
R can be found as:

E-
C F E .B

ܴଵ,ଵ = ݉݅݊ ‫ܣ‬ଵ , ‫ܤ‬ଵ = min 0.2,0.6 = 0.2

N
R

ܴଵ,ଶ = ݉݅݊ ‫ܣ‬ଵ , ‫ܤ‬ଶ = min 0.2,0.8 = 0.2

&

ܴଵ,ଷ = ݉݅݊ ‫ܣ‬ଵ , ‫ܤ‬ଷ = min 0.2,0.4 = 0.2

D . O r.D

ܴଶ,ଵ = ݉݅݊ ‫ܣ‬ଶ , ‫ܤ‬ଵ = min 0.5,0.6 = 0.5

ܴଶ,ଶ = ݉݅݊ ‫ܣ‬ଶ , ‫ܤ‬ଶ = min 0.5,0.8 = 0.5
BV P f.
D ro

ܴଶ,ଷ = ݉݅݊ ‫ܣ‬ଶ , ‫ܤ‬ଷ = min 0.5,0.4 = 0.4

O

ܴଷ,ଵ = ݉݅݊ ‫ܣ‬ଷ , ‫ܤ‬ଵ = min 0.7,0.6 = 0.6

T
P
©

ܴଷ,ଶ = ݉݅݊ ‫ܣ‬ଷ , ‫ܤ‬ଶ = min 0.7,0.8 = 0.7

U
E

ܴଷ,ଷ = ݉݅݊ ‫ܣ‬ଷ , ‫ܤ‬ଷ = min 0.7,0.4 = 0.4

 E
AD
5 7 9
1 0.2 0.2 0.2
R=

G
2 0.5 0.5 0.4

N
3 0.6 0.7 0.4

PU C O
E, T GH
• Using max-min composition relation

38
‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ ܴ ל‬can be computed as:
0.2 0.2 0.2
.
ᇱ

E-
‫[ = ܣ‬0.5 0.5 0.3] ܽ݊݀ ܴ = 0.5 0.5 0.4
C F E .B

N 0.6 0.7 0.4

R
&

RC1 A’ T min RC2 A’ T min RC3 A’ T min

D . O r.

0.2 0.5 0.2 0.2 0.5 0.2 0.2 0.5 0.2

࡮Ԣ૚ = ࡮Ԣ૜ =
D

0.5 0.5 0.5 ࡮Ԣ૛ = 0.5 0.5 0.5 0.4 0.5 0.4
BV P f.

0.6 0.3 0.3 0.7 0.3 0.3 0.4 0.3 0.3

D ro

max 0.5 max 0.5 max 0.4

T
P
©

U
E

‫ܤ ׵‬ᇱ = [0.5 0.5 0.4]

 E
AD
5 7 9
1 0.2 0.2 0.2
R=

G
2 0.5 0.5 0.4

N
3 0.6 0.7 0.4

PU C O
E, T GH
• Using max-product composition relation

38
‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ ܴ ל‬can be computed as:
0.2 0.2 0.2
.
ᇱ

E-
‫[ = ܣ‬0.5 0.5 0.3] ܽ݊݀ ܴ = 0.5 0.5 0.4
C F E .B

N 0.6 0.7 0.4

R
&

RC1 A’ T prod RC2 A’ T prod RC3 A’ T prod

D . O r.

0.2 0.5 0.1 0.2 0.5 0.1 0.2 0.5 0.1

࡮Ԣ૚ = ࡮Ԣ૜ =
D

0.5 0.5 0.25 ࡮Ԣ૛ = 0.5 0.5 0.25 0.4 0.5 0.2
BV P f.

0.6 0.3 0.18 0.7 0.3 0.21 0.4 0.3 0.12

D ro

max 0.25 max 0.25 max 0.2

T
P
©

U
E

‫ܤ ׵‬ᇱ = [0.25 0.25 0.2]

 E
AD
5 7 9
1 0.2 0.2 0.2
R=

G
2 0.5 0.5 0.4

N
3 0.6 0.7 0.4

PU C O
E, T GH
• Using min-max composition relation

38
‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ ܴ ל‬can be computed as:
0.2 0.2 0.2
.
ᇱ

E-
‫[ = ܣ‬0.5 0.5 0.3] ܽ݊݀ ܴ = 0.5 0.5 0.4
C F E .B

N 0.6 0.7 0.4

R
&

RC1 A’ T max RC2 A’ T max RC3 A’ T max

D . O r.

0.2 0.5 0.5 0.2 0.5 0.5 0.2 0.5 0.5

࡮Ԣ૚ = ࡮Ԣ૜ =
D

0.5 0.5 0.5 ࡮Ԣ૛ = 0.5 0.5 0.5 0.4 0.5 0.5
BV P f.

0.6 0.3 0.6 0.7 0.3 0.7 0.4 0.3 0.4

D ro

min 0.5 min 0.5 min 0.4

T
P
©

U
E

‫ܤ ׵‬ᇱ = [0.5 0.5 0.4]

 E
AD
5 7 9
1 0.2 0.2 0.2
R=

G
2 0.5 0.5 0.4

N
3 0.6 0.7 0.4

PU C O
E, T GH
• Using max-max composition relation

38
‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ ܴ ל‬can be computed as:
0.2 0.2 0.2
.
ᇱ

E-
‫[ = ܣ‬0.5 0.5 0.3] ܽ݊݀ ܴ = 0.5 0.5 0.4
C F E .B

N 0.6 0.7 0.4

R
&

RC1 A’ T max RC2 A’ T max RC3 A’ T max

D . O r.

0.2 0.5 0.5 0.2 0.5 0.5 0.2 0.5 0.5

࡮Ԣ૚ = ࡮Ԣ૜ =
D

0.5 0.5 0.5 ࡮Ԣ૛ = 0.5 0.5 0.5 0.4 0.5 0.5
BV P f.

0.6 0.3 0.6 0.7 0.3 0.7 0.4 0.3 0.4

D ro

max 0.6 max 0.7 max 0.5

T
P
©

U
E

‫ܤ ׵‬ᇱ = [0.6 0.7 0.5]

 E
AD
5 7 9
1 0.2 0.2 0.2
R=

G
2 0.5 0.5 0.4

N
3 0.6 0.7 0.4

PU C O
E, T GH
• Using min-min composition relation

38
‫ܤ‬ᇱ = ‫ܣ‬Ԣ ‫ ܴ ל‬can be computed as:
0.2 0.2 0.2
.
ᇱ

E-
‫[ = ܣ‬0.5 0.5 0.3] ܽ݊݀ ܴ = 0.5 0.5 0.4
C F E .B

N 0.6 0.7 0.4

R
&

RC1 A’ T min RC2 A’ T min RC3 A’ T min

D . O r.

0.2 0.5 0.2 0.2 0.5 0.2 0.2 0.5 0.2

࡮Ԣ૚ = ࡮Ԣ૜ =
D

0.5 0.5 0.5 ࡮Ԣ૛ = 0.5 0.5 0.5 0.4 0.5 0.4
BV P f.

0.6 0.3 0.3 0.7 0.3 0.3 0.4 0.3 0.3

D ro

min 0.2 min 0.2 min 0.2

T
P
©

U
E

‫ܤ ׵‬ᇱ = [0.2 0.2 0.2]

 E
Approximate reasoning

AD
• It means that given any logical system , it is very difficult

G
(maybe unknown) to make exact reasoning

N
• Hence we are liberal from engineering perspective i.e., we do

PU C O
not want to be so precise as long as our system works
E, T GH

38
• We have a set of rules , so we use specific rule of inference
.
and then we infer the consequent

E-
C F E .B

• Given a rule R and given a condition A, the inference for B is

N
R

done using composition rule ‫ܴ ל ܣ = ܤ‬

&
D . O r.

• The fuzzy set associated with each rule base may be discrete
D

or continuous
BV P f.

• A rule base may contain single rule or multiple rules

D ro

• Various inference mechanisms for a single rule are

T
P

enumerated
©

U
E

• R , in case of continuous fuzzy sets is NOT defined (infinite

values) hence the computation method is different
 E
AD
Example: Single Rule with Discrete Fuzzy Set
• RULE 1: If temperature is HOT, then fan should run FAST

G
• RULE 2: If temperature is MODERATELY HOT, then fan should

N
run MODERATELY FAST

PU C O
The temperature is expressed in °F and speed in 1000 rpm
E, T GH

38
Given:
0.4 0.6 0.8 0.9
.

E-
C F E .B

‫= ܱܶܪ = ܪ‬ , , ,
N 70 80 90 100
R

0.3 0.5 0.7 0.9

&

F = FAST = , , ,
D . O r.

1 2 3 4
D
BV P f.

ᇱ
0.2 0.4 0.6 0.8
D ro

‫= ܱܶܪ ܻܮܧܶܣܴܧܦܱܯ = ܪ‬ , , ,
O

70 80 90 100
T
P

‫ ܨ ݁ݐݑ݌݉݋ܥ‬ᇱ = ‫ܶܵܣܨ ܻܮܧܶܣܴܧܦܱܯ‬

©

U
E
 E
AD
• Solution: Using Mamdani Implication Rule

G
1 2 3 4

N
70 0.3 0.4 0.4 0.4
Rule1, R =

PU C O
E, T GH 80 0.3 0.5 0.6 0.6

38
90 0.3 0.5 0.7 0.8
.
100 0.3 0.5 0.7 0.9

E-
C F E .B

N
R

• Using max-min composition relation

&

‫ ܨ ׶‬ᇱ = ‫ܪ‬Ԣ ‫ ܴ ל‬can be computed as:

D . O r.

0.3 0.4 0.4 0.4

D

‫ ܪ‬ᇱ = [0.2 0.4 0.6 0.8] ܽ݊݀ ܴ = 0.3 0.5 0.6 0.6
BV P f.

0.3 0.5 0.7 0.8

D ro

0.3 0.5 0.7 0.9

T
P
©

U
E
 E
AD
RC1 H’ T min RC2 H’ T min
0.3 0.2 0.2 0.4 0.2 0.2
ࡲԢ૚,૚ = ࡲԢ૚,૛ =

G
0.3 0.4 0.3 0.5 0.4 0.4

N
0.3 0.6 0.3 0.5 0.6 0.5

PU C O
0.3
E, T GH 0.8 0.3 0.5 0.8 0.5
max 0.3 max 0.5

38
.

E-
C F E .B

RC3 H’ T min RC4 H’ T min

N
0.4 0.2 0.2
R

0.4 0.2 0.2

ࡲԢ૚,૜ = ࡲԢ૚,૝ =
&

0.6 0.4 0.4 0.6 0.4 0.4

D . O r.

0.7 0.6 0.6 0.8 0.6 0.6

D

0.7 0.8 0.7 0.9 0.8 0.8

BV P f.
D ro

max 0.7 max 0.8

T
P

0.3 0.5 0.7 0.8

©

‫׵‬ ‫ܨ‬ᇱ= , , ,
E

1 2 3 4
 E
Example: Multiple Rules with Discrete Fuzzy Set

AD
• RULE 1: If height is TALL , then speed is HIGH

G
• RULE 2: If height is MEDIUM , then speed is MODERATE

N
The fuzzy sets for height (in feet) and speed (in m/s) are:
0.5 0.8 1 0.4 0.7 0.9

PU C O
‫ܪ‬ଵ = ܶ‫= ܮܮܣ‬ , , , ܵଵ = ‫= ܪܩܫܪ‬ , ,
5 6 7 5 7 9
E, T GH
0.6 0.7 0.6 0.6 0.8 0.7

38
‫ܪ‬ଶ = ‫= ܯܷܫܦܧܯ‬ , , , ܵଶ = ‫= ܧܶܣܴܧܦܱܯ‬ , ,
5 6 7 5 7 9
.

E-
଴.ହ ଴.ଽ ଴.଼
C F E .B

Given H = ‫= ܧܩܣܴܧܸܣ ܧܸܱܤܣ‬ , ,

ହ ଺ ଻
Compute ܵǯ = ‫ܮܣܯܴܱܰ ܧܸܱܤܣ‬
N
R
&

• Solution: Relational matrices for RULE 1 and RULE 2 are computed using Mamdani
D . O r.

Implication Rule as:

D

5 7 9 5 7 9
BV P f.

5 0.4 0.5 0.5 5 0.6 0.6 0.6

ࡾ૚ = ࡾ૛ =
D ro

6 0.4 0.7 0.8 6 0.6 0.7 0.7

T
P
©

7 0.4 0.7 0.9 7 0.6 0.6 0.6

U
E
 E
AD
ܵ ᇱ = ‫ = ܮܣܯܴܱܰ ܧܸܱܤܣ‬max ‫ܪ‬ᇱ ‫ܴ ל‬ଵ , ‫ܪ‬ᇱ ‫ܴ ל‬ଶ
= H ᇱ ‫ ל‬max[ܴଵ , ܴଶ ]

G
N
0.6 0.6 0.6
max ܴଵ , ܴଶ = 0.6 0.7 0.8

PU C O
E, T GH 0.6 0.7 0.9
0.6 0.6 0.6

38
ᇱ
‫[ = ܪ‬0.5 0.9 0.8] ܽ݊݀ ܴ = 0.6 0.7 0.8
.

E-
0.6 0.7 0.9
C F E .B

N
R
&

RC1 A’ T min RC2 A’ T min RC3 A’ T min

D . O r.

0.6 0.5 0.5 0.6 0.5 0.5 0.6 0.5 0.5

ࡿԢ૚ = ࡿԢ૜ =
D

0.6 0.9 0.6 ࡿԢ૛ = 0.7 0.9 0.7 0.8 0.9 0.8
BV P f.

0.6 0.8 0.6 0.7 0.8 0.7 0.9 0.8 0.8

D ro

max 0.6 max 0.7 max 0.8

T
P
©

0.6 0.7 0.8

U
E

‫׵‬ ܵᇱ= , ,
5 7 9
 E
AD
Multiple rules with Continuous Fuzzy Sets

G
N
• A continuous fuzzy system with two non-interactive inputs

PU C O
‫ݔ‬ଵ ܽ݊݀‫ݔ‬ଶ (antecedents) and as ingle output ‫( ݕ‬consequent) is
E, T GH
described by a collection of ‫ ݎ‬linguistic IF-THEN rules

38
௞ ௞
‫ݔܨܫ‬ଵ ݅‫ܣ ݏ‬ଵ ܽ݊݀ ‫ݔ‬ଶ is ‫ܣ‬ଶ , ‫ ݕ ݄݊݁ݐ‬௞ ݅‫ܤ ݏ‬௞
.

E-
C F E .B

for ݇ = 1,2, … , ‫ݎ‬

N
R
&

௞ ௞
where ‫ܣ‬ଵ and ‫ܣ‬ଶ are the fuzzy sets representing ݇ ௧௛
D . O r.

antecedent pairs and ‫ܤ‬௞ are the fuzzy sets representing the ݇ ௧௛
D

consequent
BV P f.
D ro

• Consider a two input system and three rule system

O
T

• The inputs to system are crisp values and we use a max-min

P
©

inference method
U
E
 E
AD
P P$ P P Bk
P$
Rule 1

G
min

N
Crisp Input, x2
Crisp Input, x1 y

PU C O

P $   P $ 
Rule 2

E, T GH Bk

38
min
.
Crisp Input, x2 y

E-
Crisp Input, x1
C F E .B


P$ P$ 
Rule 3

N
R

Bk
&
D . O r.D

Crisp Input, x1 Crisp Input, x2

y
BV P f.
D ro

Result of overlapping (max operation)՜

O

Find center of gravity to get crisp output ‫כ ݕ‬

T
P

y
y*
©

U
E
 E
AD
Structure of FIS

G
N
• The basic structure of a fuzzy inference system
PU C O
E, T GH
consists of three conceptual components:

38
– a rule base, which contains a selection of fuzzy
.

E-
C F E .B

rules; N
R

– a database(or dictionary), which defines the

&

membership functions used in the fuzzy rules;

D . O r.D

– a reasoning mechanism, which performs the

inference procedure (usually the fuzzy reasoning)
BV P f.
D ro

upon the rules and given facts to derive a

O
T

reasonable output or conclusion

P
©

U
E
 E
AD
Block Diagram of a typical FIS

G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D

• With crisp inputs and outputs, a fuzzy inference system implements a nonlinear
BV P f.

mapping from its input space to output space

D ro

• This mapping is accomplished by a number of fuzzy if-then rules, each of which

describes the local behavior of the mapping
T
P

• The antecedent of a rule defines a fuzzy region in the input space, while the
©

consequent specifies the output in the fuzzy region

E
 E
AD
MAMDANI FUZZY MODELS
• The Mamdani fuzzy inference system was proposed as

G
the first attempt to control a steam engine and boiler

N
combination by a set of linguistic control rules

PU C O
obtained from experienced human operators
E, T GH
• It is a two-rule Mamdani fuzzy inference system which

38
derives the overall output ‫ ݖ‬when subjected to two
.
crisp inputs ‫ ݔ‬and ‫ݕ‬
E-
C F E .B

• Two fuzzy inference systems were used as two

N
R

controllers to generate the heat input to the boiler and

&
D . O r.

throttle opening of the engine cylinder, respectively, to

D

regulate the steam pressure in the boiler and the speed

of the engine
BV P f.
D ro

• Since the plant takes only crisp values as inputs, we

O

have to use a defuzzifier to convert a fuzzy set to a

T
P

crisp value
©

U
E
 E
AD
The Mamdani fuzzy inference system using min and max for T-norm
and T-conorm operators, respectively

G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
The Mamdani fuzzy inference system using product and max for Tnorm
and T-conorm operators, respectively

G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Example: Single-input single-output Mamdani fuzzy model
‫݈݈ܽ݉ݏ ݏ݅ ܻ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬

G
ቐ‫݉ݑ݅݀݁݉ ݏ݅ ܻ ݄݊݁ݐ ݉ݑ݅݀݁݉ ݏ݅ ܺ ݂ܫ‬

N
‫݁݃ݎ݈ܽ ݏ݅ ܻ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬

PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Example : Two-input single-output Mamdani fuzzy model

G
‫݁݃ݎ݈ܽ ݁ݒ݅ݐܽ݃݁݊ ݏ݅ ܼ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܻ݀݊ܽ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬
‫݈݈ܽ݉ݏ ݁ݒ݅ݐܽ݃݁݊ ݏ݅ ܼ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܻ݀݊ܽ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬

N
‫݈݈ܽ݉ݏ ݁ݒ݅ݐ݅ݏ݋݌ ݏ݅ ܼ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܻ݀݊ܽ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬

PU C O
‫݁݃ݎ݈ܽ ݁ݒ݅ݐ݅ݏ݋݌ ݏ݅ ܼ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܻ݀݊ܽ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
Input-Output mapping

E- AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
MATLAB DEMO
E- AD
38 E
 E
AD
Ex 1: Fan Speed Control , single input as Temperature

G
‫ܹܱܮܵ ݏ݅ ݀݁݁݌ܵ݊ܽܨ ݄݊݁ݐ ܦܮܱܥ ݏ݅ ݁ݎݑݐܽݎ݁݌݉݁ܶ ݂ܫ‬

N
ቐ‫ܧܶܣܴܧܦܱܯ ݏ݅ ݀݁݁݌ܵ݊ܽܨ ݄݊݁ݐ ܮܣܯܴܱܰ ݏ݅ ݁ݎݑݐܽݎ݁݌݉݁ܶ ݂ܫ‬

PU C O
‫ܶܵܣܨ ݏ݅ ݀݁݁݌ܵ݊ܽܨ ݄݊݁ݐ ܱܶܪ ݏ݅ ݁ݎݑݐܽݎ݁݌݉݁ܶ ݂ܫ‬
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Ex 2: Fan Speed Control , two inputs as Temperature & Humidity
If (temperature is cold) and (humidity is low) then (fanspeed is slow)

G
If (temperature is cold) and (humidity is medium) then (fanspeed is moderate)

N
If (temperature is cold) and (humidity is high) then (fanspeed is fast)
If (temperature is normal) and (humidity is low) then (fanspeed is slow)

PU C O
If (temperature is normal) and (humidity is medium) then (fanspeed is moderate)
E, T GH
If (temperature is normal) and (humidity is high) then (fanspeed is fast)

38
If (temperature is hot) and (humidity is low) then (fanspeed is fast)
If (temperature is hot) and (humidity is medium) then (fanspeed is fast)
.

E-
If (temperature is hot) and (humidity is high) then (fanspeed is fast)
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
SUGENO FUZZY MODELS

G
N
• The Sugeno fuzzy model (also known as the TSK fuzzy model) was

PU C O
proposed by Takagi, Sugeno, and Kang in an effort to develop a
E, T GH
systematic approach to generating fuzzy rules from a given input-

38
output data set
.

E-
• A typical fuzzy rule in a Sugeno fuzzy model has the form
C F E .B

N
R

࢏ࢌ ࢞ ࢏࢙ ࡭ ࢇ࢔ࢊ ࢟ ࢏࢙ ࡮ ࢚ࢎࢋ࢔ ࢠ = ࢌ ( ࢞, ࢟),

&

• where ‫ ܣ‬and ‫ ܤ‬are fuzzy sets in the antecedent, ‫= ݖ ݈݄݁݅ݓ‬

D . O r.

݂ ( ‫ݔ‬, ‫ )ݕ‬is a crisp function in the consequent

D

• Usually ݂(‫ݔ‬, ‫ )ݕ‬is a polynomial in the input variables ‫ ݔ‬and ‫ ݕ‬but it

BV P f.

can be any function as long as it can appropriately describe the

D ro

output of the model within the fuzzy region specified by the

T
P

antecedent of the rule

©

U
E
 E
AD
Example : Fuzzy and nonfuzzy rule set-a comparison
• An example of a single-input Sugeno fuzzy model can be expressed as:

G
‫ = ܻ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬0.ͳܺ + 6.4
ቐ‫ = ܻ ݄݊݁ݐ ݉ݑ݅݀݁݉ ݏ݅ ܺ ݂ܫ‬െ0.ͷܺ + 4

N
‫ ܺ = ܻ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬െ 2

PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Example: Two-input single-output Sugeno fuzzy model

G
• An example of a two-input single-output Sugeno fuzzy model
with four rules can be expressed as:

N
‫ = ݖ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܻ݀݊ܽ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬െ‫ ݔ‬+ ‫ ݕ‬+ 1

PU C O
‫ = ݖ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܻ݀݊ܽ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬െ‫ ݕ‬+ 3
E, T GH

38
‫ = ݖ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܻ݀݊ܽ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬െ‫ ݔ‬+ 3
‫ ݔ = ݖ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܻ݀݊ܽ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬+ ‫ ݕ‬+ 2
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
TSUKAMOTO FUZZY MODELS

AD
• In the Tsukamoto fuzzy models , the consequent of each fuzzy if-then rule is represented by a

G
fuzzy set with a monotonical MF
• The inferred output of each rule is defined as a crisp value induced by the rule's firing

N
strength

PU C O
• The overall output is taken as the weighted average of each rule's output
• Since each rule infers a crisp output, the Tsukamoto fuzzy model aggregates each rule's
E, T GH
output by the method of weighted average and thus avoids the time-consuming process of

38
defuzzification
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
Example : Single-input Tsukamoto fuzzy model

AD
• An example of a single-input Tsukamoto fuzzy model can be expressed as:

G
‫ͳܥ ݏ݅ ܻ ݄݊݁ݐ ݈݈ܽ݉ݏ ݏ݅ ܺ ݂ܫ‬
ቐ‫ʹܥ ݏ݅ ܻ ݄݊݁ݐ ݉ݑ݅݀݁݉ ݏ݅ ܺ ݂ܫ‬

N
‫͵ܥ ݏ݅ ܻ ݄݊݁ݐ ݁݃ݎ݈ܽ ݏ݅ ܺ ݂ܫ‬

PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
G
N
PU C O
E, T GH
FUZZY LOGIC-IV

38
Fuzzy Logic Control
.

E-
C F E .B

MODULE 10
N
R
&
D . O r.

Dr. R.B.Ghongade,
D

VIIT, Pune-411048
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Agenda

G
N
PU C O
• Why Fuzzy Logic Control?
E, T GH
• Typical fuzzy control systems

38
.
• Classical and Fuzzy PID controller
E-
C F E .B

N
R

• Architecture of a Mamdani type Fuzzy Control

&

System
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
Why Fuzzy Logic Control?

G
N
• Fuzzy logic is a technique to embody human like

PU C O
thinking into a control system
E, T GH

38
• A fuzzy controller can be designed to emulate human
deductive thinking i.e., the process , humans use to
.

E-
C F E .B

infer conclusions from what they know

N
• Traditional control approach requires formal modeling
R
&

of the physical reality

D . O r.

• There are many systems where humans operate the

D

plant based on their experience

BV P f.

• Better the experience, better is the control mechanism

D ro

• We can utilize such principle of human way of decision

T
P

making into designing control systems

©

U
E
 E
AD
• Fuzzy control incorporates ambiguous human
logic into computer programs

G
• It suits control problems that

N
PU C O
– cannot be easily represented by mathematical models
E, T GH
– have weak model(physics of systems is poorly

38
understood)
.

E-
– have parameter variation problem
C F E .B

– unavailable or incomplete data

N
R
&

– are very complex

D . O r.

– we have good qualitative understanding of plant or

D

process operation
BV P f.
D ro

• Because of its unconventional approach, design

O
T

of such controllers leads to faster

P
©

development/implementation cycles
U
E
 E
AD
Typical Fuzzy Control Systems

G
N
• Two popular typical fuzzy control systems are

PU C O
Mamdani type and Tagaki-Sugeno (T-S) type
E, T GH

38
• Mamdani Type:
.
– Employ fuzzy sets in the consequent part of the rule
E-
C F E .B

– Incremental control action is described in the

N
R

consequent part of each rule

&
D . O r.

• T-S Type:
D

– Employ function of the input fuzzy linguistic variables

BV P f.

as the consequent of the rules

D ro

– A fuzzy dynamic model is expressed in the form of

T

local rules
P
©

U
E
 E
AD
A Classical PID Controller

G
N
PU C O
E, T GH

38
• Objective: Output should follow the command signal
.

E-
C F E .B

• PID Control Law:

1 ௧ ݀݁(‫)ݐ‬
N
R

‫ ݐ ݁ ܭ = ݐ ݑ‬+ න ݁ ‫ ݐ݀ ݐ‬+ ܶௗ
&

ܶ௜ ଴ ݀‫ݐ‬
D . O r.

• Error is multiplied by K, error is integrated over time with a time constant

D

ܶ௜ and error is differentiated

BV P f.

• In discrete domain ‫݇ = ݐ‬οܶ is the sampling time interval and ݇is the
D ro

sampling instant
૚ ࢋ ࢑ െ ࢋ(࢑ െ ૚)
T
P

࢛ ࢑ =ࡷ ࢋ ࢑ + ࢋ ࢑ + ࢋ ࢑ െ ૚ + ‫ ڮ‬οࢀ + ࢀࢊ
ࢀ࢏ οࢀ
©

U
E
 E
AD
• Consider a simpler PI controller
1 ௧

G
‫ ݐ ݁ ܭ = ݐ ݑ‬+ න ݁ ‫ݐ݀ ݐ‬

N
ܶ௜ ଴

PU C O
• Its discrete form is
E, T GH
૚

38
࢛ ࢑ =ࡷ ࢋ ࢑ + ࢋ ࢑ + ࢋ ࢑ െ ૚ + ‫ ڮ‬οࢀ
ࢀ࢏
.

E-
C F E .B

• But
N
R
&

૚
D . O r.

࢛ ࢑െ૚ =ࡷ ࢋ ࢑െ૚ + ࢋ ࢑ െ ૚ + ࢋ ࢑ െ ૛ + ‫ ڮ‬οࢀ

ࢀ࢏
D

• Giving
BV P f.

૚
D ro

࢛ ࢑ െ ࢛(࢑ െ ૚) = ࡷ ࢋ ࢑ െ ࢋ(࢑ െ ૚) + ࢋ ࢑ οࢀ
T

ࢀ࢏
P
©

U
E
 E
AD
• Thus we can write
࢛ ࢑ = ࢛ ࢑ െ ૚ + ο࢛ ࢑

G
Where ο࢛ ࢑ = ࢌ(ࢋ, οࢋ)

N
PU C O
• Hence present control action is equal to previous
E, T GH
control action +incremental control action

38
• This is the fundamental concept which led to the
.

E-
C F E .B

development of Fuzzy Logic Controller(FLC)

N
R
&

• It can be shown that:

D . O r.

‫ ݇ ݑ = ݇ ݑ‬െ 1 + ο‫݇ ݑ‬
D

Where ο‫ ݇ ݁ ݂ = ݇ ݑ‬, ο݁ ݇ , ‫ڮ‬

BV P f.
D ro

• Means incremental control action is some function

T
P

of error, change in error and higher order terms

©

U
E
 E
AD
Fuzzy PID Controller (Mamdani Type)

G
N
PU C O
E, T GH

38
.

E-
C F E .B

• Similar to classical PID controller a fuzzy PID controller can have the
N
R

following structure:
&

‫ ݇ ݑ = ݇ ݑ‬െ 1 + ο‫݇ ݑ‬
D . O r.D

• However , ο‫ ݇ ݑ‬is obtained using fuzzy rule base that provides

incremental control action , which is a function of two fuzzy
BV P f.

variables , ࢋ and οࢋ
D ro

• Classical PID controller is linear

T
P

• Fuzzy PID controller is non-linear

©

• Our goal to compute what is the incremental control action in FLC

U
E

of Mamdani Type
 E
AD
Example of a Rule-base

G
N
• Temperature control system employing a fuzzy proportional

PU C O
controller :‫)݇(݁ܭ = ݇ ݑ‬
E, T GH
• RULE 1: IF error is POSITIVE-HIGH, THEN keep the heater on

38
for a LONGER duration
.

E-
C F E .B

• RULE 2: IF error is POSITIVE-LOW, THEN keep the heater on for

N
R

a SHORTER duration
&

• Antecedent part has only ONE fuzzy variable since ‫= ݇ ݑ‬

D . O r.

‫)݇(݁ܭ‬
D

• We do not have incremental control action but a DIRECT

BV P f.
D ro

control action
O
T
P
©

U
E
 E
AD
• Temperature control system employing a fuzzy

G
PI controller :‫ ݇ ݑ = ݇ ݑ‬െ 1 + ο‫݇ ݑ‬

N
• RULE 1: IF error is POSITIVE-HIGH and change
PU C O
E, T GH
in error is POSITIVE-HIGH, THEN keep the

38
heater on for a LONGER duration
.

E-
C F E .B

• RULE 2: IF error is POSITIVE-LOW and change

N
R
&

in error is POSITIVE-LOW, THEN keep the

D . O r.

heater on for a SHORTER duration

D

‫ ݇ ݑ = ݇ ݑ‬െ 1 + ο‫݇ ݑ‬
BV P f.
D ro

௙(௘,ο௘)
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
Architecture of a FLC: Mamdani Type
 E
AD
Operation explained

G
N
• Let 40°C crisp value be fuzzified into linguistic
PU C O
values ‫ܦܮܱܥ‬, ‫ܧܶܣܴܧܦܱܯ‬, ‫ܱܶܪ‬, ‫ڮ‬
E, T GH

38
• Each linguistic value is associated with a specific
.

E-
C F E .B

membership function
N
R

• Once the data is fuzzified , it goes to the rule-

&

base and then using the inference mechanism, a

D . O r.D

fuzzy inference is derived

BV P f.

• After we get the fuzzy inference about the control

D ro

action we place a de-fuzzifier which converts the

T
P

fuzzy control action to a crisp control action

©

U
E
 E
AD
Principal Design Parameters of a FLC

G
• Fuzzification strategies and the interpretation of a

N
fuzzification operator( fuzzifier)

PU C O
• Database:
E, T GH
– Discretization/normalization of universe of discourse

38
– Fuzzy partitioning of input and output spaces
.

E-
– Completeness
C F E .B

– Choice of membership function of a primary fuzzy set

N
R

• Rule-base
&
D . O r.

– Choice of process state(input) variables and control(output)

D

variables
– Source and derivation of fuzzy control rules
BV P f.

– Types of fuzzy control rules

D ro

– Completeness of fuzzy control rules

T
P

• Fuzzy Inference mechanism

©

• De-fuzzyfication operator (defuzzifier)

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
Complete Architecture of FLC
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
Fuzzification(graphical)

38 E
 ©
P
D ro
E
BV P f.
T D

ο‫= כݑ‬
D . O r.
U R
C F E .B
O & .

‫ ߤ ׬‬ο‫(݀ ݑ‬ο‫)ݑ‬
E, T GH

‫(ߤ ׬‬ο‫)ݑ‬ο‫(݀ݑ‬ο‫)ݑ‬
PU C O
N N
E- G
AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
TRADITIONAL vs FUZZY LOGIC CONTROLLERS (FLC)
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N
FUZZY CONTROL

N G
E- AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
FUZZY LOGIC CONTROLLERS (FLC)

38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
ARCHITECTURE OF FLC

E- AD
38 E
 E
AD
DESCRIPTION OF FLC COMPONENTS

G
A Fuzzy Logic Controller usually consists of:

N
PU C O
¾ A fuzzification unit which maps measured inputs of crisp value into fuzzy
E, T GH
linguistic values to be used by a fuzzy reasoning mechanism.

38
.
¾ A knowledge base (KB) which is the collection of expert control

E-
C F E .B

knowledge required to achieve the control objective.

N
R
&

¾ A fuzzy reasoning mechanism that performs various fuzzy logic

D . O r.

operations to infer the control action for the given fuzzy inputs.
D
BV P f.

¾ A defuzzification unit which converts the inferred fuzzy control action

D ro

into the required crisp control values to be entered into the system
T
P

process.
©

U
E
 E
AD
FUZZY LOGIC CONTROLLER

G
¾ Control inputs,

N
PU C O
¾ Fuzzy sets (membership functions) of inputs,
E, T GH

38
¾ Rules,
.

E-
C F E .B

N
R

¾ Control outputs,
&
D . O r.D

¾ Fuzzy sets (membership functions) of outputs,

BV P f.
D ro

¾ Fuzzy reasoning,
T
P
©

¾ Defuzzification.
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
 E
AD
Steps in designing a simple fuzzy control system
1. Identify the variables (inputs, states, and outputs) of the plant

G
2. Partition the universe of discourse or the interval spanned by each

N
variable into a number of fuzzy subsets, assigning each a linguistic

PU C O
label (subsets include all the elements in the universe)
E, T GH
3. Assign or determine a membership function for each fuzzy subset.

38
4. Assign the fuzzy relationships between the inputs’ or states’ fuzzy
.

E-
C F E .B

subsets on the one hand and the outputs’ fuzzy subsets on the
other hand, thus forming the rule-base
N
R
&

5. Choose appropriate scaling factors for the input and output

D . O r.

variables in order to normalize the variables to the [0, 1] or the

D

΀оϭ͕ϭ΁interval
BV P f.

6. Fuzzify the inputs to the controller

D ro

7. Use fuzzy approximate reasoning to infer the output contributed

T
P

from each rule

©

8. Aggregate the fuzzy outputs recommended by each rule

U
E

9. Apply defuzzification to form a crisp output

 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
FLC AND PID CONTROLLERS

E, T GH
PU C O
N N
E- G
AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
FUZZY LOGIC CONTROLLER (MAMDANI)
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
MATLAB SCREENSHOT OF FLC - SIMULINK BLOCK
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
ADVANTAGES OF FLC

E- G
AD
38 E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
APPLICATIONS OF FLC

E- AD
38 E
 E
AD
Example of FLC

G
N
• Aircraft Landing problem

PU C O
• Necessary to simulate the final descent approach
E, T GH

38
• When aircraft lands:
– downward velocity D square of height
.

E-
C F E .B

• Thus, at higher altitudes, a large downward velocity is

desired
N
R
&

• As the height (altitude) diminishes, the desired downward

D . O r.

velocity gets smaller and smaller

D

• In the limit, as the height becomes vanishingly small, the

BV P f.

downward velocity also goes to zero

D ro

• The aircraft will descend from altitude promptly but will

touch down very gently to avoid damage.
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
velocity versus altitude

38 E
The desired profile of downward
 E
AD
Aircraft landing control problem

G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D

• Two state variables for this simulation will be the

height above ground, h, and the vertical velocity of the
BV P f.
D ro

aircraft, v
O
T
P

• The control output will be a force that, when applied to

©

the aircraft, will alter its height, h, and velocity, v

U
E
 E
AD
The differential control equations

G
N
• Mass m moving with velocity v has

PU C O
momentum:
E, T GH ‫ݒ݉ = ݌‬

38
• If no external forces are applied, the mass
will continue in the same direction at the
.

E-
C F E .B

same velocity, v
• If a force f is applied over a time interval
N
R

ο‫ݐ‬, a change in velocity will be:

&

‫ כ ݂ = ݒ‬ο‫ ݐ‬/m
D . O r.

• If we let 't = 1.0 (s) and m = 1.0 (lb s2/ ft),

D

we obtain:
‫( ݂ = ݒ‬lb), or
BV P f.

“the change in velocity is proportional to the

D ro

applied force” Simple momentum model

T
P
©

U
E
 E
AD
• In difference notation, we get:

G
‫ݒ‬௜ାଵ = ‫ݒ‬௜ +݂௜

N
PU C O
݄௜ାଵ = ݄௜ +‫ݒ‬௜ ȉ ο‫ݐ‬
E, T GH
where ‫ݒ‬௜ାଵ is the new velocity, ‫ݒ‬௜ is the old

38
velocity, ݄௜ାଵ is the new height, and ݄௜ is the old
.

E-
C F E .B

height
N
R
&

• These two “control equations” define the new

D . O r.D

value of the state variables ‫ ݒ‬and ݄ in

BV P f.

response to control input and the previous

D ro

state variable values

T
P
©

U
E
 E
AD
Construct membership functions for the height, h, the
vertical velocity, v, and the control force, f

G
• Step 1 a. Define membership functions for state variable ࢎ

N
PU C O
E, T GH Height (ft)
0 100 200 300 400 500 600 700 800 900 1000

38
Large (L) 0 0 0 0 0 0 0.2 0.4 0.6 0.8 1
Medium (M) 0 0 0 0 0.2 0.4 0.6 0.8 1 0.8 0.6
.

E-
C F E .B

Small (S) 0.4 0.6 0.8 1 0.8 0.6 0.4 0.2 0 0 0

Near zero 1 0.8 0.6 0.4 0.2 0 0 0 0 0 0
N
R
&

(NZ)
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
Step 1 b. Define membership functions for state variable ࢜

AD
•
Vertical velocity (ft/s)

G
оϯϬ оϮϱ оϮϬ оϭϱ оϭϬ оϱ 0 5 10 15 20 25 30

N
Up large 0 0 0 0 0 0 0 0 0 0.5 1 1 1
(UL)

PU C O
Up small 0
E, T GH 0 0 0 0 0 0 0.5 1 0.5 0 0 0
(US)

38
Zero (Z) 0 0 0 0 0 0.5 1 0.5 0 0 0 0 0
Down small 0 0 0 0.5 1 0.5 0 0 0 0 0 0 0
.

E-
C F E .B

(DS)
Down large 1
N 1 1 0.5 0 0 0 0 0 0 0 0 0
R

(DL)
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
Step 2. Define membership functions for control output ࢌ

AD
•
Output force (lb)

G
оϯϬ оϮϱ оϮϬ оϭϱ оϭϬ оϱ 0 5 10 15 20 25 30

N
Up large 0 0 0 0 0 0 0 0 0 0.5 1 1 1
(UL)

PU C O
Up small 0
E, T GH 0 0 0 0 0 0 0.5 1 0.5 0 0 0
(US)

38
Zero (Z) 0 0 0 0 0 0.5 1 0.5 0 0 0 0 0
Down 0 0 0 0.5 1 0.5 0 0 0 0 0 0 0
.

E-
C F E .B

small (DS)
Down large 1N 1 1 0.5 0 0 0 0 0 0 0 0 0
R

(DL)
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
• Step 3. Define the rules and summarize them in an Fuzzy Associative Memory
(FAM) table (The values in the FAM table, of course, are the control outputs)

G
N
Velocity

PU C O
E, T GH Height DL DS Zero US UL
L Z DS DL DL DL

38
M US Z DS DL DL
.

E-
C F E .B

N S UL US Z DS DL
R
&

NZ UL UL Z DS DS
D . O r.D
BV P f.

• Step 4 . Define the initial conditions, and conduct a simulation for four cycles. Since
D ro

the task at hand is to control the aircraft’s vertical descent during approach and
T
P

landing, we will start with the aircraft at an altitude of 1000 feet, with a downward
ǀĞůŽĐŝƚLJŽĨо20 ft Ɛоϭ͘
©

U
E
 E
AD
• We will use the following equations to update the state variables
for each cycle:

G
‫ݒ‬௜ାଵ = ‫ݒ‬௜ +݂௜
݄௜ାଵ = ݄௜ +‫ݒ‬௜

N
PU C O
CYCLE #0
• Initial height, ݄଴ : 1000 ft
E, T GH

38
• Initial velocity, ‫ݒ‬଴ ͗оϮϬft Ɛоϭ
• Control ݂଴ : to be computed
.

E-
C F E .B

• Height ݄ fires L at 1.0 and M at 0.6

N
R

• Velocity ‫ ݒ‬fires only DL at 1.0

&
D . O r.D

Height Velocity Output

֜
BV P f.

L (1.0) AND DL (1.0) Z (1.0)

D ro

M (0.6) AND DL (1.0) ֜ US (0.6)

T
P
©

U
E
 E
AD
G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.
D ro

O
T
P

• We defuzzify using the centroid method and get ݂଴ = 5.8 lb. This
©

is the output force computed from the initial conditions

U
E
 E
AD
CYCLE #1
• ݄ଵ =݄଴ +‫ݒ‬଴ =1000+(-20)=980 ft

G
• ‫ݒ‬ଵ =‫ݒ‬଴ +݂଴ = (-20)+(5.8)=-14.2 ft/s

N
Height ݄ଵ fires L at 0.96 and M at 0.64

PU C O
•
Velocity ‫ݒ‬ଵ fires DS at 0.58 and DL at 0.42
•
E, T GH

38
.

E-
C F E .B

Height Velocity Output

N
R

֜
&

L (0.96) AND DS (0.58) DS (0.58)

D . O r.

L (0.96) AND DL (0.42) ֜ Z (0.42)

D

M (0.64) AND DS (0.58) ֜ Z (0.58)

BV P f.

M (0.64) AND DL (0.42) ֜ US (0.42)

D ro

O
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
• We find the centroid to be ݂ଵ соϬ.5 lb
38 E
 E
AD
G
CYCLE #2

N
• ݄ଶ =݄ଵ +‫ݒ‬ଵ =980+(-14.2)=965.8 ft

PU C O
• ‫ݒ‬ଶ =‫ݒ‬ଵ +݂ଵ = (-14.2)+(-0.5)=-14.7 ft/s
E, T GH
• Height ݄ଶ fires L at 0.93 and M at 0.67

38
• Velocity ‫ݒ‬ଶ fires DS at 0.57 and DL at 0.43
.

E-
C F E .B

N
R
&

Height Velocity Output

D . O r.

֜
D

L (0.93) AND DS (0.57) DS (0.57)

L (0.93) AND DL (0.43) ֜ Z (0.43)
BV P f.

M (0.67) AND DS (0.57) ֜ Z (0.43)

D ro

M (0.67) AND DL (0.43) ֜ US (0.57)

T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
• We find the centroid to be ݂ଶ со0.4 lb
38 E
 E
AD
G
CYCLE #3

N
• ݄ଷ =݄ଶ +‫ݒ‬ଶ =965.8+(-14.7)=951.1 ft

PU C O
• ‫ݒ‬ଷ =‫ݒ‬ଶ +݂ଶ = (-14.7)+(-0.4)=-15.1 ft/s
E, T GH
• Height ݄ଷ fires L at 0.9 and M at 0.7

38
• Velocity ‫ݒ‬ଷ fires DS at 0.49 and DL at 0.51
.

E-
C F E .B

N
R
&

Height Velocity Output

D . O r.

֜
D

L (0.9) AND DS (0.49) DS (0.49)

L (0.9) AND DL (0.51) ֜ Z (0.51)
BV P f.

M (0.7) AND DS (0.49) ֜ Z (0.49)

D ro

M (0.7) AND DL (0.51) ֜ US (0.51)

T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
• We find the centroid to be ݂ଷ = 0.3 lb
38 E
 E
AD
• Final values for the state variables to finish the

G
simulation

N
PU C O
• ݄ସ =݄ଷ +‫ݒ‬ଷ =951.1+(-15.1)=936.0 ft
E, T GH

38
• ‫ݒ‬ସ =‫ݒ‬ଷ +݂ଷ = (-15.1)+(0.3)=-14.8 ft/s
.

E-
C F E .B

N Cycle 0 Cycle 1 Cycle 2 Cycle 3 Cycle 4

R
&

Height (ft) 1000.0 980.0 965.8 951.1 936.0

D . O r.D

Velocity (ft sоϭ) оϮϬ оϭϰ͘Ϯ оϭϰ͘ϳ оϭϱ͘ϭ оϭϰ͘ϴ

BV P f.
D ro

Control force 5.8 оϬ͘ϱ оϬ͘ϰ 0.3

T
P
©

U
E
 E
AD
Example of FLC

G
N
• Liquid Level Control

PU C O
• Design a controller that can be used to move the level set-
E, T GH
point from, say, 4 feet to 6 feet, the set-point-tracking

38
problem
.

E-
C F E .B

• Suppose that the tank is 10 feet tall and the tank is empty.
• We want to fill the tank to a level of 5 feet, so we make the
N
R
&

current set-point, w, equal to 5

D . O r.

• The idea is to fill the tank to the desired set-point as quickly

D

and smoothly as possible

• We want to minimize the amount of overshoot, or the time
BV P f.
D ro

that the tank has a level greater than the set-point value
O

before it finally settles down

T
P
©

U
E
 E
AD
G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&

• The current level at any time, t , is designated as h

D . O r.D

• Liquid flows out of the tank through an open valve

• This flow is designated by the letter q
BV P f.
D ro

• Liquid flows into the tank by means of a pump

T
P

• The pump flow, Q, can be regulated by the controller

©

• The tank cross-sectional area is designated by the letter A

E
 E
AD
• Equation that describes the mass balance for the liquid in the
tank as a function of time is :

G
ௗ௛

N
‫ܣ‬ =ܳെ‫ݍ‬
ௗ௧

PU C O
• Flow out of the tank, q, through the outlet pipe and the valve
E, T GH
is described as:

38
‫ = ݍ‬Ȱ‫ܣ‬௣ ʹ݄݃
.

E-
C F E .B

where N
R

Ȱ= friction coefficient for flow through both the small exit pipe
&
D . O r.

and the valve,

D

‫ܣ‬௣ =cross-sectional area of the small exit pipe

BV P f.

݃ =the gravitational constant, equal to 32.2 ft Ɛо2 (US units)

D ro

O
T
P
©

U
E
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
G
Block flow diagram

E- AD
38 E
 E
AD
Classical PID control

G
• The PID control algorithm is described as:
் ௗ௘

N
‫ܭ = ݑ‬௉ ݁ + ‫ܭ‬ூ ‫׬‬଴ ݁ ݀‫ ݐ‬+ ‫ܭ‬஽
ௗ௧

PU C O
• ‫ܭ‬௉ , ‫ܭ‬ூ , ‫ܭ‬஽ are proportional, integral, and derivative control constants, respectively
,are specific to the system in question and are usually picked to optimize the
E, T GH
controller performance and ensure that the system remains stable for all possible

38
control actions
.

E-
• If we use a PID controller, which is linear, or any other linear controller with a
C F E .B

linear plant, then the system is called a linear system

N
R

• We can use Laplace transforms to convert the linear equations in the

&

• blocks to the Laplace domain

D . O r.

• The blocks can then be combined to form a single transfer function for the entire
D

system
BV P f.

• In the real world, many systems are at least slightly nonlinear and we can linearize
D ro

them by expanding the non linear function in truncated Taylor series:

O
T
P

ஶ
݂ ௡ (ܽ) ݂ᇱ ܽ ݂ ᇱᇱ ܽ ݂௞ ܽ
©

௡ ଶ ௞
෍ (‫ ݔ‬െ ܽ) = ݂ ܽ + ‫ݔ‬െܽ + (‫ ݔ‬െ ܽ) + ‫ ڮ‬+ ‫ݔ‬െܽ
U

݊! 1! 2! ݇!
E

௡ୀ଴
+‫ڮ‬
 E
AD
• The truncated Taylor’s series for linearizing about a steady state value, in

G
this case our set-point, is given as:
ଵ

N
݄ = ݄௦ + (݄ െ ݄௦ )….(1)
ଶ ௛

PU C O
ೞ
• If we choose ݄௦ = 5 ft, we can linearize the radical term over some of the
E, T GH
control range

38
Approximate linearization.
.

E-
h (ft) ࢎ Equation
C F E .B

N (1)
R

10 3.162 3.354
&

9 3.0 3.130
D . O r.

8 2.828 2.907
D

7 2.646 2.683
Overall system block flow diagram or transfer 6 2.449 2.460
BV P f.

function for the Laplace domain 5 2.236 2.236

D ro

4 2.0 2.012
T
P

3 1.732 1.789
©

2 1.414 1.565
U
E

1 1.0 1.342
0 0 1.118
 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N
filling problem

N G
E- AD
38 E
Time-level PID response for the tank-
 E
AD
Fuzzy Control for Tank Filling Problem

G
N
PU C O
• A fuzzy control system connects input membership
E, T GH
functions, functions representing the input to the

38
controller, e, to output membership functions that
.

E-
C F E .B

represent the control action, u

N
R

• A simple fuzzy control system designed for our tank-

&

level set-point-tracking problem consists of three

D . O r.D

rules:
BV P f.

1. If the Level Error is Positive Then the Change in Control Action is

D ro

Positive
O
T

2. If the Level Error is Zero Then the Change in Control Action is Zero
P
©

3. If the Level Error is Negative Then the Change in Control Action is

U
E

Negative
 E
The input membership functions

AD
G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.

• Notice the “dead band” or “dead zone” in the membership function Zero between ±3
D ro

Inches
T

• This is optional and is a feature commonly used with on–off controllers

P

• It is easy to implement with a fuzzy controller and is useful if the control engineer
©

wishes to minimize control response to small transient-level changes

E

• This step can save wear and tear on equipment

 ©
P
D ro
E
BV P f.
T
D . O r.D
U R
C F E .B
O & .
E, T GH
PU C O
N N
E- G
AD
38 E
The output membership functions
 E
• The de-fuzzified output value from the controller is a fractional representing

AD
the required pump output for the desired level change
• It is defined by the following expression:

G
Change in Controller Action = οu = (ܳ௜ о ܳ௦௣ )/Range

N
Where,
• The term ܳ௦௣ represents the pump output (gallons per minute)

PU C O
required to maintain the set-point level
E, T GH
• The term ܳ௜ is the new pump output requested by the controller

38
• If οu>0, then the Range is defined as ܳ௠௔௫ െ ܳ௦௣ , where ܳ௠௔௫ is the
.

E-
maximum pump output
C F E .B

• If οu < 0, then the Range is defined as ܳ௦௣

N
R

• The term ܳ௦௣ must be calculated using a steady state mass balance for
&

the tank or it must be estimated in some fashion

D . O r.

• The ranges of the fuzzy output sets Positive and Negative are +2.0 to 0.0
D

ĂŶĚо2.0 to 0.0, respectively

BV P f.

• Since the Change in Controller Action is a fraction between either

D ro

• Ϭ͘ϬĂŶĚϭ͘ϬŽƌϬ͘ϬĂŶĚоϭ͘Ϭ͕ŝƚŝƐĐůĞĂƌƚŚĂƚǁĞǁŝůůŶĞǀĞƌŽďƚĂŝŶĂĐŽŶƚƌŽů
O

action outside the range ŽĨоϭ͘ϬƚŽ1.0

T
P

• Our de-fuzzification technique will require that we include numbers up to

©

2.0 in the fuzzy set or membership function Positive and numbers down to
E

оϮ͘Ϭin the fuzzy set Negative

 E
AD
• Suppose that we decide to change our set-point level from 5
feet to 8 feet

G
• The error is defined as the set-point level, 8 feet, minus the

N
current level, 5 feet, or +3 feet

PU C O
• The three rules are fired, producing the following results:
E, T GH
1. Positive error is 0.5

38
2. Zero error is 0.5
.

E-
C F E .B

3. Negative error is 0.0

N
R
&
D . O r.D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
G
N
PU C O
E, T GH

38
.

E-
C F E .B

N
R
&
D . O r.D
BV P f.

• The centroid of the “clipped” membership

D ro

functions, the shaded area is +0.5

T
P
©

U
E
 E
AD
• Since u is greater than 0.0,
ܳ௜ = (ܳ௠௔௫ оܳ௦௣ )οu +ܳ௦௣

G
or

N
ܳ௜ = 0.5(ܳ௠௔௫ +ܳ௦௣ ), since οu = 0.5

PU C O
E, T GH
• This says that the new pump output, ܳ௜ , should be

38
adjusted to be halfway between the current or set-point
.

E-
output and the maximum pump output
C F E .B

N
• After an appropriate time interval, corresponding to a
R
&

predetermined sample rate, the same procedure will be

D . O r.

repeated until the set-point level, 8 feet, is achieved!

D
BV P f.
D ro

O
T
P
©

U
E
 E
AD
SUMMARY

G
N
• Easy to set up a working fuzzy controller even for difficult-to-control

PU C O
processes (but previously controlled by operator).
E, T GH
• Easy to understand the rule base in simple cases.

38
.
• No need to use FLC in straightforward linear cases, if implementation

E-
C F E .B

reasons would not require it.

N
R
&

• Complications will arise when rule base increases.

D . O r.D
BV P f.
D ro

O
T
P
©

U
E