Fuzzy Inference

System
Bridging Human Reasoning and Machine Intelligence

Overview Membership Functions Calculations Graphs Applications

OVERVIEW What is Fuzzy Logic?

CRISP (Classical) LOGIC FUZZY LOGIC

● TRUE or FALSE only ● Degrees of truth (0 to 1)

● Value is 0 or 1 ● Partial membership allowed

● Sharp boundaries ● Smooth, gradual transitions

VS
● Temp > 30°C = "Hot" (yes/no) ● Temp = 28°C → "Somewhat Hot"

● Rigid — no in-between ● Mirrors human reasoning

Key Insight

Fuzzy Logic was introduced by Lotfi Zadeh in 1965. It models imprecise, vague, and uncertain information — the kind humans deal
with every day. Instead of asking 'Is it hot?', fuzzy logic asks 'How hot is it?' and returns a degree from 0.0 to 1.0.
 FIS ARCHITECTURE

CRISP FUZZIFI- RULE DEFUZZI- CRISP

INPUTS CATION BASE FICATION OUTPUT

Temperature Maps crisp → fuzzy IF-THEN rules Aggregation Fan speed

Pressure membership degrees from expert + Centroid / Valve position
Speed knowledge Max method Control signal

Fuzzification Inference Engine Defuzzification

Converts precise numerical inputs into fuzzy Applies fuzzy rules (IF-THEN) to the fuzzified Converts fuzzy output back into a crisp value.
set membership values using Membership inputs and evaluates each rule using AND/OR Common methods: Centroid of Area (COA),
Functions (MF). operators. Mean of Maxima (MOM).
 MEMBERSHIP FUNCTIONS

Triangular MF Trapezoidal MF Gaussian MF

μ(x) = max(0, min((x-a)/(b-a), (c-x)/(c- μ(x) = max(0, min((x-a)/(b-a), 1, (d-

μ(x) = exp(-(x-c)² / (2σ²))
b))) x)/(d-c)))

Parameters: Parameters: Parameters:

a = left foot, b = peak, c = right foot a,b = rising; c,d = falling edge c = center, σ = standard deviation

Best for: Best for: Best for:

Simple, widely used in real-time systems For 'plateau' regions of certainty Smooth, natural ideal for sensor data

Example — Temperature Universe of Dodoma [0°C → 50°C]

COLD COOL WARM HOT

[0, 0, 10, 20] [10, 20, 25] [20, 30, 35] [30, 40, 50, 50]
 MEMBERSHIP FUNCTION GRAPH — Temperature

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0° 4° 8° 12° 16° 20° 24° 28° 32° 36° 40° 44° 48°
Cold Cool Warm Hot
 FUZZY RULES & INFERENCE ENGINE

Rule Base — IF-THEN Structure Fuzzy Operators

Rules encode expert knowledge in natural language. Each rule maps fuzzy input
conditions to fuzzy output conclusions. AND (T-norm)
Minimum: min(μA, μB)

IF Temp is Cold AND Humidity is Low Product: μA × μB

R1
THEN Fan Speed is Slow

OR (S-norm)
IF Temp is Warm AND Humidity is Medium
R2 Minimum: max(μA, μB)
THEN Fan Speed is Medium
Product: μA + μB − μA·μB

IF Temp is Hot OR Humidity is High

R3
THEN Fan Speed is Fast
NOT

IF Temp is Cool AND Humidity is Low Minimum: 1 − μA

R4 Product: 1 − μA
THEN Fan Speed is Slow

Mamdani Inference: Uses min() for AND, clips the output MF at the rule strength. Each active rule contributes a clipped fuzzy output that is then
aggregated.
 STEP-BY-STEP CALCULATION EXAMPLE

Problem: Temperature = 28°C, Humidity = 60%. Find Fan Speed.

STEP 1: Fuzzification STEP 2: Rule Firing

μ_Cold(28) = trapmf(28, 0,0,10,20) = 0.00 R1: min(μ_Cool=0.20, μ_Hum_Low=0.40) = 0.20

μ_Cool(28) = trimf(28, 10,20,30) = 0.20 → Fan_Slow fires at α₁ = 0.20

μ_Warm(28) = trimf(28, 22,30,38) = 0.75 R2: min(μ_Warm=0.75, μ_Hum_Med=0.60) = 0.60

μ_Hot(28) = trapmf(28, 32,42,50,50) = 0.00 → Fan_Medium fires at α₂ = 0.60

R3: max(μ_Hot=0.00, μ_Hum_High=0.00) = 0.00

→ Fan_Fast fires at α₃ = 0.00

STEP 3: Defuzzification — Centroid of Area (COA)

Fan_Slow MF center: 20% | Fan_Medium MF center: 50% | Fan_Fast MF center: 80%

COA = Σ(αᵢ × centroid_i) / Σαᵢ = (0.20 × 20 + 0.60 × 50 + 0.00 × 80) / (0.20 + 0.60 + 0.00) = 44% Fan Speed
 DEFUZZIFICATION METHODS

Centroid of Area (COA) Mean of Maxima (MOM) Bisector Method

Most Popular Simple Balanced

z* = ∫ z·μ(z) dz / ∫ μ(z) dz z* = (z_min_max + z_max_max) / 2 ∫_{a}^{z*} μ(z)dz = ∫_{z*}^{b} μ(z)dz

Finds the geometric center of the aggregated

Averages the x-values where the output MF Finds the vertical line dividing the area under the
fuzzy output. Smooth and well-behaved. Used in
reaches its maximum. Simple to compute. aggregated output into two equal halves.
most applications.

Smooth output, considers whole distribution Fast computation, easy to implement More robust than COA for asymmetric MFs

Computationally expensive for continuous MFs Ignores shape outside the maximum region Can differ from COA significantly
 FIS TYPES — COMPARISON

Key Differences
Interpretability
100 Mamdani FIS
80 ▸ Fuzzy output (MF-based)
60 ▸ Requires defuzzification
Real-time Control Accuracy
40 ▸ Best: Interpretability
20
0 Sugeno FIS
▸ Crisp/linear output
Nonlinear Computational ▸ No defuzzification needed
Modeling Speed ▸ Best: Speed & Optimization

Tsukamoto
Expert Rule
Support ▸ Monotonic output MFs
▸ Weighted avg. output

Mamdani Sugeno Tsukamoto ▸ Best: Smooth response

 REAL-WORLD APPLICATIONS

Industrial Control Automotive Systems Medical Diagnosis

Cement kilns, chemical plant process control, Anti-lock brakes (ABS), automatic transmission, Disease risk assessment, drug dosage
HVAC automation, robotic arm positioning engine control, cruise control systems controllers, ECG signal classification

Consumer Electronics Finance & Trading Robotics & AI

Camera autofocus, washing machine load Credit risk scoring, stock market prediction, Path planning under uncertainty, human-robot
sensing, rice cooker temperature control fraud detection systems interaction, neural-fuzzy hybrids
 ADVANTAGES & LIMITATIONS

80
Advantages
Scalability 92
65 Handles imprecision naturally

95 Easy to implement expert knowledge

Computational Cost 55
80 No need for precise mathematical model

88 Works well with nonlinear systems

No Training Data 10
95
Robust to noisy inputs

20
Handles Uncertainty 85
90 Limitations

90 Rule base can get very large

Interpretability 20
95 MF design is subjective

0 20 40 60 80 100 No automatic learning (pure FIS)

Crisp Logic Neural Nets Fuzzy Logic

Hard to validate formally
KEY TAKEAWAYS

Fuzzy Inference System

in a Nutshell
01 Fuzzification converts crisp inputs into membership degrees using MFs (triangular, trapezoidal, Gaussian)

02 Rule Base encodes expert knowledge as IF-THEN rules evaluated using AND (min) / OR (max) operators

03 Inference Engine fires applicable rules, clips or scales output MFs based on rule activation strength

04 Defuzzification (COA, MOM, Bisector) converts aggregated fuzzy output back to a crisp control value