Fuzzy Systems

* Fuzzy Systems Toolbox, M. Beale and H Demuth

How can fuzzy systems be used in a world where
measurements and actions are expressed as crisp
values?

Fuzzy Systems (Cont.)
* Fuzzy Systems Toolbox, M. Beale and H Demuth
* Fuzzify crisp inputs to get the fuzzy inputs
* Defuzzify the fuzzy outputs to get crisp outputs
Fuzzy Systems (Cont.)
90 Degree F.
It is too hot!
Turn the fan on high
Set the fan at
90% speed
Input Fuzzifier Fuzzy System Defuzzifier output
Fuzzy Set: Vector Representation
Two vectors can represent fuzzy discrete
sets or fuzzy continuous sets,
Support Vector (universe vector)
Grade Vector (membership vector]

Example: Vector Representation
Define the concept of tall over heights
from 5 to 7 feet, using MATLAB
S = [5.00 5.50 6.00 6.50 7.00];
G = [0.00 0.25 0.50 0.75 1.00];
0
0.25
0.5
0.75
1
5 5.5 6 6.5 7
Grade
Fuzzification
Process of making a crisp quantity fuzzy
Vector representation can be viewed as either
a discrete or an approximation of a continuous
set ( use linear interpolation]
Crisp input
Fuzzy Grade
Example: Fuzzification
Define fuzzy set near 5
S = [ 0:10];
G = [0.0 0.1 0.3 0.5 0.8 1 0.8 0.5 0.3 0.1 0];
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9 10
Near 5
0.5
0.9
Fuzzy Systems
How do you make a machine smart?
Put some FAT in it!
A FAT enough machine can model any
process
A FAT system can always turn inputs to
outputs and turn causes to effects and turn
questions to answer
FAT stands for Fuzzy Approximation
Theorem
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Fuzzy Systems (cont.)
FAT idea has a simple geometry
Cover a curve with patches!
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Fuzzy Systems (cont.)
Knowledge as rules
Each piece of human knowledge, each rule of
the form
IF this then that defines a patch.
All the rules define patches
Try to cover some wiggly curve
The better the patches cover the curve, the
smarter the system
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Knowledge as Rules
How do you reason?
You want to play golf on Saturday or Sunday
and you dont want to get wet when you play.
Reach it with rules!
If it rains, you get wet!
If you get wet, you cant play golf
If it rains on Saturday and wont rain on Sunday
You play golf on Sunday!
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Artificial Intelligence: AI
Knowledge is rules
Rules are in black-and-white language
Bivalent rules
AI has so far, afer over 30 years of research,
not produced smart machines!
Because they cant yet put enough rules in the
computer (use 100-1000 rules, need >100k}
Throwing more rules at the problem

*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Knowledge as Rules
Fuzzy researchers have built hundreds of
smart machines that work!
Yes, we need rules
No, we dont need a lot of rules for many
tasks
We need Fuzzy rules
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Knowledge as Rules
Every term in one of our rules is Fuzzy
Every term is vague, hazy, inexact, sloppy
One human rule covers all these cases
AI rule covers one precise case
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Fuzzy Systems (cont.)
Fuzzy rule relates fuzzy sets
If X is A, then Y is B
A and B are fuzzy sets and subset of X and Y
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Build a Fuzzy System
3 Steps
Pick the nouns or variables
Example: X be input and Y be output
Let x be temperature and Y be change in motor speed
Cause, effect. Stimulus, response!
Pick the fuzzy sets
Define fuzzy subsets of the nouns X and Y
Pick the fuzzy rules
Associate output to the input
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Design motor speed controller for air
conditioner
Step 1: assign input and output
variables
Let X be temperature in Fahrenheit
Let Y be the change in motor speed of the
air conditioner
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Design motor speed controller for air
conditioner
Step 2: Pick fuzzy sets
Define subsets of the noun X and Y
Say 5 fuzzy sets on X
Cold, Cool, Just Right, Warm, and Hot
Say 5 fuzzy sets on Y
Stop, Slow, Medium, Fast, and Blast
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Input Fuzzy set

*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Output Fuzzy set

*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Design motor speed controller for air
conditioner
Step 3: Assign a motor speed set to
each temperature set
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Rules
If temperature is cold then motor speed stop
If temperature is cool then motor speed slows
If temperature is just right then motor speed is medium
If temperature is warm then motor speed is fast
If temperature is hot then motor speed blasts
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Fuzzy Relation
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Example: Build a Fuzzy System
Fuzzy system with 5 patches
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
FAT Theorem
You can always cover a curve with a finite
number of fuzzy patches
Let the cuts overlap
Sloppy rules give big patches
Fine rules give small patches
You pay for precision!
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
FAT Theorem (Cont.)
Rough cover of the non linear System
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
FAT Theorem (Cont.)
finer cover of the non linear System
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Fuzzy Associative Memory
Which rule fires or activates at which
time?
They all fire all the time
They fire in parallel
All rules fire to some degree
Most fire to zero degree
The result is a fuzzy weighted average
*Fuzzy Thinking:The new Science of Fuzzy Logic, Bart Kosko
Additive Fuzzy System
Stores m fuzzy rules of the form
If X = A
j
then Y = B
j,
then computes the output
by defuzzifiy the summed (MAXed) of the
partially fired then-part fuzzy sets B
j

*Fuzzy Engineering, Bart Kosko
Graphical Technique of
Mamdani (Max-Min] Inference
If x
1
k
is A
1
k
and x
2
k
is A
2
k
Then Y
k
is B
k
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Graphical Technique of
Mamdani (Max-Min] Inference
If x is A Then Y is B
MATLAB
XS = support vector of X;
XG = grade vector of X;
YS = support vector of y;
YG = grade vector of Y;
A = Crisp input;
B = ifthen_min(XS,XG,A,YG);
Where function B= ifthen_min (); calculate
[YG_row,YG_column] = size(YG);
B = Min(fuzzify(XS,XG,A)*ones(1,YG_column),YG);

Graphical Technique of
Max-Product Inference
If x
1
k
is A
1
k
and x
2
k
is A
2
k
Then Y
k
is B
k
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Graphical Technique of
Max-Product Inference
If x is A Then Y is B
MATLAB
XS = support vector of X;
XG = grade vector of X;
YS = support vector of y;
YG = grade vector of X;
A = Crisp input;
B = ifthen_prod(XS,XG,A,YG);
Where function B= ifthen_prod (); calculate
B = fuzzify(XS,XG,A)
*
YG;

Max: Fuzzy Union & Fuzzy Union
using algebraic sum
Variation on Fuzzy Implication
When simulating human reasoning with fuzzy rules in
decision making and expert systems, the algebraic
sum and product often will give a more intuitively
pleasing result
* Fuzzy Systems Toolbox, M. Beale and H Demuth, * *Fuzzy Engineering, Bart Kosko
Example: temp. = 65 degree F.
If temperature is just right then motor speed is medium
*Fuzzy Thinking, Bart Kosko
Example: temp. = 63 degree F.

*Fuzzy Thinking, Bart Kosko
If temperature is cool then motor speed slows
If temperature is just right then motor speed is medium
Example: t = 63 degree F. (Cont.)
*Fuzzy Thinkring, Bart Kosko
Example: t = 63 degree F. (Cont.)
Summed (MAXed) of the partially fired then-
part fuzzy sets
*Fuzzy Thinkring, Bart Kosko
OR OUTPUT
Example: t = 63 degree F. (Cont.)
Defuzzify to find the output motor speed
*Fuzzy Thinkring, Bart Kosko
Defuzzification
Convert fuzzy grade to Crisp output
*Fuzzy Engineering, Bart Kosko
Defuzzification (Cont.)
Centroid Method: the most prevalent and
physically appealing of all the defuzzification
methods [Sugeno, 1985; Lee, 1990]
Often called
Center of area
Center of gravity
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Defuzzification (Cont.)
Max-membership principal
Also known as height method
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Defuzzification (Cont.)
Weighted average method
Valid for symmetrical output membership functions

*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Formed by weighting
each functions in the
output by its respective
maximum membership
value
Defuzzification (Cont.)
Mean-max membership (middle of maxima)
Maximum membership is a plateau

*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Z* = a + b
2
Defuzzification (Cont.)
Center of Largest area
If the output fuzzy set has at least two convex
subregion, defuzzify the largest area using centroid

*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Defuzzification (Cont.)
First (or last) of maxima
Determine the smallest value of the domain with
maximized membership degree
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Example: Defuzzification
Find an estimate crisp output from the following
3 membership functions
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Example: Defuzzification
CENTROID
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Example: Defuzzification
Weighted Average
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Example: Defuzzification
Mean-Max
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Z* = (6+7)/2 = 6.5
Example: Defuzzification
Center of largest area
Same as the centroid method because the complete
output fuzzy set is convex
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Example: Defuzzification
First and Last of maxima
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Defuzzification
Of the seven defuzzification methods presented,
which is the best?
It is context or problem-dependent
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Defuzzification: Criteria
Hellendoorn and Thomas specified 5 criteria
against which to measure the methods
#1 Continuity
Small change in the input should not produce the large
change in the output
#2 Disambiguity
Defuzzification method should always result in a unique
value, I.e. no ambiguity
Not satisfied by the center of largest area!
*Fuzzy Logic with Engineering Applications, Timothy J. Ross
Defuzzification: Criteria (Cpnt.)
Hellendoorn and Thomas specified 5 criteria
against which to measure the methods
#3 Plausibility
Z* should lie approximatly in the middle of the support region
and have high degree of membership
#4 Computational simplicity
Centroid and center of sum required complex computation!
#5 Constitutes the difference between centroid,
weighted average and center of sum
Problem-dependent, keep computation simplicity
*Fuzzy Logic with Engineering Applications, Timothy J. Ross