• Based on intuition and judgment • No need for a mathematical model • Provides a smooth transition between members and nonmembers

• Relatively simple, fast and adaptive • Less sensitive to system fluctuations • Can implement design objectives, difficult to express mathematically, in linguistic or descriptive rules.

Applications Domain
• Fuzzy Logic • Fuzzy Control
– Neuro-Fuzzy System – Intelligent Control – Hybrid Control

Some Interesting Applications
• • • • • • • Ride smoothness control Camcorder auto-focus and jiggle control Braking systems Copier quality control Rice cooker temperature control High performance drives Air-conditioning systems

• Fuzzy Pattern Recognition • Fuzzy Modeling

Conventional or crisp sets are binary. An element either belongs to the set or doesn't. {True, false} {1, 0}

µ B =1 µC =1

Crisp Set/Subset µA =0 µ A =1
Subset A Subset B

Universe (X)
Subset C

1

Fuzzy Indicators
• Examples of fuzzy measures include close, heavy, light, big, small, smart, fast, slow, hot, cold, tall and short.
• Can you distinguish between American and French person? • Some Rules:
– – – – – – If If If If If If speaks English then American speaks French then French loves perfume then French loves outdoors then American good cook then French plays baseball then American

e.g. On a scale of one to 10, how good was the dive?

10

9 9.5

9

Fuzzy Indicators
• Rules may give contradictory indicators {good cook, loves outdoors, speaks French} • The right answer is a question of a degree of association • Fuzzy logic resolves these conflicting indicators
– Membership of the person in the French set is 0.9 – Membership of the person in the American set is 0.1

• Fuzzy ≠ Probability • Probability deals with uncertainty and likelihood • Fuzzy logic deals with ambiguity and vagueness

• Fuzzy ≠ Probability • Example #1
– Billy has ten toes. The probability Billy has nine toes is zero. The fuzzy membership of Billy in the set of people with nine toes, however, is nonzero.
(cite: Professor Marks)

Example #2
– A bottle of liquid has a probability of ½ of being rat poison and ½ of being pure water. – A second bottle’s contents, in the fuzzy set of liquids containing lots of rat poison, is ½. – The meaning of ½ for the two bottles clearly differs significantly and would impact your choice should you be dying of thirst. – 50% probability means 50% chance that the water is clean. – 50% fuzzy membership means that the water has poison.
(cite: Bezdek)

#1

#2

2

• Crisp membership functions (µ) are either one or zero. • e.g. Numbers greater than 10.
A ={x | x>10} 1 x 10

• The set, B, of numbers near to 2 can be represented by a membership function
µB(x)

µA(x)

µ B ( x) = e −| x − 2|
0 1 2 3

x

Fuzzy Sets
• A fuzzy set, A, is said to be a subset of B if

µ A ( x) ≤ µ B ( x)
• e.g. B = far and A=very far. • For example...
2 µ A ( x) = µ B ( x)
Tall Tall or Short? Short

Very Tall

Fuzzy Measures

7’ Tall 6’ Medium 5’ 4’ Very Short Short

µ
Very Short 1.0

Membership Function
Short Medium Tall Very Tall

4’

5’

6’

7’

µ = [µvs , µ s , µ m , µt , µvt ]

3

µ
Very Short 1.0

Membership Function
Short Medium Tall Very Tall

Fuzzy Logic Operations
Fuzzy union operation or fuzzy OR

4’ Short

5’

6’ Medium Tall

7’

µ = [0, 1, 0, 0, 0] µ = [0, 0, 0.5, 0.5, 0]

µ A+ B ( x) = max [ µ A ( x), µ B ( x)]

Fuzzy Logic Operations
Fuzzy intersection operation or fuzzy AND

Fuzzy Logic Operations
Complement operation

µ A⋅B ( x) = min [ µ A ( x ), µ B ( x)]

µ A ( x) = 1 − µ A ( x)

µ A⋅B ( x) = min [ µ A ( x ), µ B ( x)]

Fuzzy Logic Operations
Fuzzy union operation or fuzzy OR
1

µA(x)

µB(x)

µ A+ B ( x) = max [ µ A ( x), µ B ( x)]
0 1 2 3

Fuzzy intersection operation or fuzzy AND

x

0

1

2

3

x

µ A⋅B ( x) = min [ µ A ( x ), µ B ( x)]
Complement operation

µA⋅B(x)

µ A ( x) = 1 − µ A ( x)
0 1 2 3

x

4

µ A+ B ( x) = max [ µ A ( x), µ B ( x)]
µA(x)
1

µB(x)

0

1

2

3

x

0

1

2

3

x

µA+B (x)

0

1

2

3

x

• Fuzzifier converts a crisp input into a fuzzy variable. • Definition of the membership functions must
– reflects the designer's knowledge – provides smooth transition between member and nonmembers of a fuzzy set – simple to calculate

• Typical shapes of the membership function are Gaussian, trapezoidal and triangular.

• Assume we want to evaluate the health of a person based on his height and weight. • The input variables are the crisp numbers of the person’s height and weight. • Fuzzification is a process by which the numbers are changes into linguistic words

µ
Very Short 1.0

Fuzzification of Height
Short Medium Tall Very Tall

µ
Very slim 1.0

Fuzzification of Weight
Slim Medium Heavy Very heavy

4’ VS = very short S = Short M = Medium etc.

5’

6’

7’

150lb VS = very slim S = Slim M = Medium etc.

200lb

250lb

300lb

5

• • • • •

Rules reflect experts decisions. Rules are tabulated as fuzzy words Rules can be grouped in subsets Rules can be redundant Rules can be adjusted to match desired results

• Rules are tabulated as fuzzy words
– – – – Healthy (H) Somewhat healthy (SH) Less Healthy (LH) Unhealthy (U)

• Rule function f

f = {U , LH , SH , H }

f
Height
Very Short Short Medium Tall

Weight
Very Slim Slim Medium Heavy Very Heavy

H SH LH U U

SH H H SH LH

LH SH H H H

U LH LH SH SH

U U U U LH

f = {U , LH , SH , H }

Very Tall

Membership of Height
• For a given person, compute the membership of his/her weight and height • Example:
– Assume that a person height is 6’ 1” – Assume that the person’s weight is 140 lb

µ
Very Short 0.7 0.3 4’ 5’ 6’ 7’ Short Medium Tall Very Tall

µheight = {µVS µS µM µT µVT} µheight = {0 0 0.7 0.3 0}

6

Membership of Weight

µ
Very Slim 0.8 Slim Medium Heavy Very Heavy

Weight
Very Slim Slim Medium Heavy Very Heavy Very Short Short Medium Tall Very Tall

Height

H SH LH U U

SH H H SH LH

LH SH H H H

U LH LH SH SH

U U U U LH

0.2 150 lb 200 lb 250 lb 300 lb

µWeight = {µVS µS µM µH µVH} µWeight = {0.8 0.2 0 0 0}

Weight 0.8 0.2
SH H H SH LH
Medium Heavy Very Heavy

Weight 0.8 0.2
0 0
Medium (0) Heavy (0) V.Heavy (0) V. Short (0) Short (0)

Height

Short

SH LH U U

SH H H H

LH LH SH SH

U U U LH

Height

Very Short

H

LH

U

U

0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

0.7 0.3
Very Tall

0.7 0.3
V. Tall (0)

0.7 0.3
0

0.2 0.2
0

f
Weight 0.8
V. Short (0) Short (0)

Weight
0.2
0 0
Very Short Short

0.7
0.2
SH H H SH LH H

0.8
SH LH U U

0 0

0.7 0.3
V. Tall (0)

0.7 0.3
0

0.2 0.2
0

0.7 0.3
Very Tall

0.3 0.2

f = {U , LH , SH , H } f = { 0.3, 0.7, 0.2, 0.2}

f = {U , LH , SH , H } f = { 0.3, 0.7, 0.2, 0.2}

7

• Use the fuzzified rules to compute the final decision. • Two methods are often used. – Maximum Method (not often used) – Centroid

• Fuzzy set with the largest membership value is selected. • Fuzzy decision:

f = {U , LH , SH , H }

f = { 0.3, 0.7, 0.2, 0.2}

• Final Decision (FD) = Less Healthy • If two decisions have same membership max, use the average of the two.

f
0.7 0.3 0.2

FD =

∑µ D = µ ∑µ

U

DU + µ LH DLH + ... ... µU + µ LH + ... ...

FD =

0.3 × 0.2 + 0.7 × 0.4 + 0.2 × 0.6 + 0.2 × 0.8 = 0.4429 0.3 + 0.7 + 0.2 + 0.2

Crisp Decision Index (D) = 0.4429

f
0.7 0.3 0.2

f
0.75 0.25

Crisp Decision Index (D) is the centroid D = 0.4429

Fuzzy Decision Index (D) 75% in Less Healthy group 25% in Somewhat Healthy group

8

• Assume that we need to evaluate student applicants based on their GPA and GRE scores. • Let us assume that the decision should be Excellent (E), Very Good (VG), Good (G), Fair (F) or Poor (P) • An expert will associate the decisions to the GPA and GRE score. They are then Tabulated.

• Assume that we need to evaluate student applicants based on their GPA and GRE scores. • For simplicity, let us have three categories for each score [High (H), Medium (M), and Low(L)] • Let us assume that the decision should be Excellent (E), Very Good (VG), Good (G), Fair (F) or Poor (P) • An expert will associate the decisions to the GPA and GRE score. They are then Tabulated.

Excellent Very Good Good Fair Poor

= 95-100% = 90 - 94% = 80 - 89% = 70 - 79% = 0 - 69%

• Fuzzifier converts a crisp input into a fuzzy variable. • Definition of the membership functions must
– reflects the designer's knowledge – provides smooth transition between member and nonmembers of a fuzzy set – simple to calculate

µ

GRE

• Typical shapes of the membership function are Gaussian, trapezoidal and triangular.

µ

GRE

= {µL , µM , µH }

9

µ

GPA

Fn

µ

GPA

= {µL , µM , µH }

GRE=900
• Assume a student with GRE=900 and GPA=3.6 • A decision on the classification of the applicant is needed.
– Excellent – Very good – Etc.

µ

GRE

µ

GRE

= {µL = 0.8 , µM = 0.2 , µH = 0}

GPA=3.6

µ

GRE

= {µL = 0.8 , µM = 0.2 , µH = 0}

µ

GPA

= {µL = 0 , µM = 0.6 , µH = 0.4}

µ

GPA

= {µL = 0 , µM = 0.6 , µH = 0.4}

10

Fn
0.6 0.4 0.2

• Converting the output fuzzy variable into a unique number • Two defuzzifier methods are often used. – Maximum Method (not often used) – Centroid

• Fuzzy set with the largest membership value is selected. • Fuzzy decision: Fn = {P, F, G,VG, E} Fn = {0.6, 0.4, 0.2, 0.2, 0} • Final Decision (FD) = Poor Student • If two decisions have same membership max, use the average of the two.

Fn

FD = FD =

∑ µ fn µ = ∑µ
f i f i

E

× E + µVG ×VG + ..... µ E + µVG + .....

0 ×100 + 0.2 × 90 + 0.2 × 80 + 0.4 × 70 + 0.6 × 60 = 70 0.2 + 0.2 + 0.4 + 0.6

Final Decision (FD) = Fair Student

11

F
Reference

Input

Fuzzy Controller

u

Output

System

Feedback (error, change in error)

Speed Regulation
LN MN SN E ZE SP MP LP LN LN LN LN LN SN ZE SP MN LN LN LN MN ZE ZE SP SN LN LN MN SN ZE SP MP CE ZE LN MN SN ZE SP MP LP
3.0

SP MN SN ZE SP MP LP LP

MP SN ZE ZE MP LP LP LP

LP SN ZE SP LP LP LP LP

2.5 Position 2.0 1.5 1.0 0.5 0.0 0 1 2 3 4 5 Time 6 7 8 9 10

Tracking (HPD)
3

Conventional non-adaptive Tracking Control
• Requires mathematical models • Nonlinear processes are linearized • Poor Performance when model deviates • Difficult to tune for unknown dynamics • Poor performance for widely varying operation

Position

2

1

0 0 1 2 3

Time

12

Conventional non-adaptive Tracking Control
• Poor performance in noisy environments • Inclusion of some design objectives can be challenging (e.g. comfort of ride and safety)

Merits of Conventional Control

• Engineers are comfortable with the classical control designs.
– Well-established technologies – Verifiable overall system stability – System’s reliability can be evaluated.

Irony
• Stability and reliability studies
• Neural and Fuzzy Control • Based on intuitions and judgments. • Relatively simple, fast and adaptable • Can implement design objectives, difficult to express mathematically, in linguistic or descriptive rules.

are based on linearized models linearly

• Most systems do not behave •

Most systems do drift

Why Fuzzy Control?
• No need for mathematical model • Less sensitive to system fluctuations. • Design objectives difficult to express mathematically can be incorporated in a fuzzy controller by linguistic rules. • Implementation is simple and straight forward.

Position Tracking of Induction Motor
25 Position (rev) Position Track Motor Position

0

Time(sec)

20

13

MULTI-LAYER FUZZY CONTROL
System Inputs Execution Layer ELC ELC ELC
µCE
1

µE
1

LN

MN

SN ZE

SP

MP

LP

0 -6 LN MN

-3 SN

-1

0 ZE

1 SP

3 MP

6 LP

Supervisor Layer MLFC outputs

0 -3

-2

-1

0

1

2

3

14