Applications of Fuzzy Logic: Evolution

and Case Study

December 9, 2021 1
 Overview

• Commercial Applications of Fuzzy Logic

• Evolution in Growth of Fuzzy Applications
• Fuzzy Logic and the World
• Example: Sendai Subway Control
• Case Study: Fuzzy-based Path Ordering
• Limitations
• Probability vs. Possibility
• Conclusions
Pg.2, Dec 9, 2021
 Commercial Fuzzy Applications

• Two broad categories of commercial

applications:
– Industrial process control: Used for modeling and
controlling complex industrial systems
– Modeling human intelligence: Development of
fuzzy expert systems using the representation and
inference techniques of fuzzy logic able to model
imprecise processes of human experts.

Pg.3, Dec 9, 2021

Fuzzy Statistics

Pg.4, Dec 9, 2021

 Fuzzy Logic and the World
• Most important players in Asia are China and
Japan with India and Singapore playing minor
roles
• All major Japanese companies involved in
integrating fuzzy technology ($2-3 B)
• Fuzzy logic technology largely ignored in
USA (NASA, Otis)
• France and Germany among European nation
conducting fuzzy logic research (Peugeot,
VW)

Pg.5, Dec 9, 2021

 Fuzzy Logic in Japan
Hitachi Automatic control for subway trains in Sendai yielding smoother,
more efficient ride with higher stopping precision
Nippon Electric Controlling temperatures in glass fusion at the Notogawa and
Takatsuki factories
Nissan Anti-skid brake systems and automatic transmissions for cars

Canon, Minolta, Ricoh Auto-focusing in cameras by choosing the right subject in the picture
frame to focus
Panasonic Jitter removal in video camcoders by distinguishing between jitters
and actual movement of subjects
Matshushita, Toshiba, Vacuum cleaners that use sensors to gather information about floor
Sanyo, Hitachi and dirt conditions and then use a fuzzy expert system to choose the
right program
Yamaichi Computerized trading programs which mimic the approximate
reasoning processes of experienced fund managers
Matshushita A/C that make judgments based on factors such as number of persons
in room and optimum degree of comfort

Pg.6, Dec 9, 2021

 Sendai Subway Control
• Research on automation of train operations in
Japan begins in 1960
• Conventional automatic train operation (ATO)
based on PID control
• Preset target speed which the ATO tries to
achieve by powering and braking
• In practice, many changes in running
conditions such as gradient of track and the
braking force of rolling stock

Pg.7, Dec 9, 2021

 Sendai Subway Control (2)
• PID works best with linear systems
• Train is a complex non-linear system and all
ATO use simplified linear representations of
the train’s operation to apply to PID
• Fuzzy rules attempts to emulate the reasoning
and control decisions of an experience subway
driver
• Minimize difference between desired speed
and actual speed

Pg.8, Dec 9, 2021

 Sendai Subway Control (3)

ATO by conventional control ATO with fuzzy control

Pg.9, Dec 9, 2021

 Sendai Subway Control (4)
• Satisfy goals like good riding comfort and
accurate stopping
• Sendai subway operational since July 1987.
• Hitachi engineers compared fuzzy and
conventional controllers in 300,000
simulations and 3,000 rider less test runs.
• Reduced stop-gap distance 2.5 times, doubled
the comfort index, reduced power
consumption by 10%

Pg.10, Dec 9, 2021

 Case Study
• Fuzzy-based method for path selection under
additive quality of service constraints (delay),
where the information available for making
routing decisions is inaccurate.
• Identify a feasible path while minimizing the
overall setup time.
• Performance evaluation compare between the
fuzzy approach, the optimal solution, and
greedy approach.

Pg.11, Dec 9, 2021

 Basics of Networks
router
• millions of connected workstation
computing devices: hosts, end- server
systems mobile
– PCs workstations, servers local ISP
– PDAs phones, toasters
running network apps
• communication links regional ISP
– fiber, copper, radio, satellite
– Links have different
bandwidth
• routers: forward packets
• Packet: a piece of message company
network

Pg.12, Dec 9, 2021

 ATM Networks
• ATM is a connection-oriented low-layer networking
concept
• An end-to-end path called a virtual channel must be
set up in advance, using an ATM signaling (control)
protocol, before any data cells can be sent
• All cells of a virtual channel travel on the same path
• Cells arrive in the order that they were sent
• Switches must maintain state about the virtual
channels passing through them

Pg.13, Dec 9, 2021

 QoS and ATM
• Quality of Service (QOS)
– a specification of the desired (or acceptable)
grade of service required for a traffic flow
– some traffic is delay-sensitive (e.g., voice)
– some traffic is loss-sensitive (e.g., data)
– some traffic is both (e.g., compressed video)
– QoS requested at time of call setup
– ATM network tries to provide requested QoS

Pg.14, Dec 9, 2021

 Introduction
• Modern networks define parameters that
represent QoS they expect to receive from the
network.
• QoS requirements for a connection set-up
include:
– Bandwidth
– Loss ratio
– Average/Maximum delay
– Delay variation

Pg.15, Dec 9, 2021

 Introduction (2)
• Finding path that satisfies all QoS
requirements is NP-complete where # of
additive QoS parameters ¸ 2

• Present algorithms depend on deterministic

knowledge of availability of resources

• Network resources should be used efficiently

Pg.16, Dec 9, 2021

 Introduction (3)
• Fuzzy approach used for the following reasons
– Multiple parameters
– Intuitive
– Advantage over non-fuzzy approach
• Development of a mechanism for solving
complex routing problems in communications
networks
• Complex routing problems in uncertain
environment analyzed in simple rules and
decision process modeled by sets of these rules

Pg.17, Dec 9, 2021

 Inaccurate Information in Networks

• Unavailable parameters
– Information about topology and availability of
resources collected through routing protocols
– Existing protocols do not advertise all relevant
parameters necessary for making routing decisions
– Network management mechanisms only provide
partial values
– Lack of practical tools that provide QoS
information

Pg.18, Dec 9, 2021

 Inaccurate Information
in Networks (2)
• Inaccurate calculations

– Computation of link state parameters based on

traffic measurements and forecasts

– Values usually only averages or bounds, which

causes inaccuracies

Pg.19, Dec 9, 2021

 Inaccurate Information
in Networks (3)
• Obsolete information
– Dynamic parameters are significantly affected by
temporal network conditions
– Difficult to have the most current view of the
parameters available on the other links and nodes
– Requires frequent updates
– Trade-off between overhead costs and accuracy of
the network state information that the path
selection algorithm depends on

Pg.20, Dec 9, 2021

 Inaccurate Information
in Networks (4)
• Aggregation in large networks
– Sheer growth in information makes it practically
impossible to maintain accurate knowledge about
all nodes and links.
– Concept of hierarchical routing used to overcome
scalability problem
– Nodes and links of each hierarchy recursively
aggregated into higher levels representing
collection of lower nodes
– Increase in aggregation, decrease in accuracies

Pg.21, Dec 9, 2021

 Inaccurate Information
in Networks (5)
• Partial information
– Routing information exchanged between nodes
when under same network administration

– Difficult to do so in interconnected networks

which include multiple operators and private
networks

– Networks only publish partial information

Pg.22, Dec 9, 2021

 Impact of Uncertain Parameters
• Connections with only bandwidth demands,
impact of inaccuracies relatively minimal
• Connections with end-to-end delay demands
inaccuracies cause significant impact on:
– Routing complexity
– Path selection process becomes intractable

Pg.23, Dec 9, 2021

 Simple Example
• Not sufficient to consider only probability of
failure of selected path, but also connection
setup time to successfully establish connection

A Pf = 0.2

Pf = 0.24
S T

S: Source
B C D E T: Destination
Pf = 0.18

Pg.24, Dec 9, 2021

 Proposed Approach
• Algorithm whose target function is to find a
feasible path that meets the end-to-end delay
requirement, while minimizing the average
connection setup time
• Assumptions
– Routing environment: ATM or MPLS
– Information available to source node is inaccurate
– Crankback allowed

Pg.25, Dec 9, 2021

 Assumptions
• Connection setup time taken to cross each other node in
network is known
• With each link (u,v) E2is a toll t(u,v) representing
time spent for propagation along (u,v) and the queuing
and processing times at v
• Propagation delay constant, queuing and processing
times almost constant at each node
• WLOG t(u,v)=1 for each link (u,v) E, thus, the setup
time of a given path is directly proportional2 to the
number of hops along this path.

Pg.26, Dec 9, 2021

 Assumptions (2)
• Source node can approximate probability of
connection setup failure for each possible
given path as a function of its estimated delay
to the destination
• Density function of the accumulative delay of
N N
P(n0,nk): pdf(e1… ek)=pdf(e1) pdf(e2)… pdf(ek)
• Probability ofR failure P(n0,nk)
1
D

Pf(P(n0,nk))= pdf(e1… ek) (x) dx

Pg.27, Dec 9, 2021

 Problem Definition
• Adopting connection setup time as cost
function
• Objective: Minimizing the average connection
setup time
• Given:
– directed graph G=(N,E)
– candidate paths L={P(1), P(2)…P(L)}
– source (S) and destination (T) node
• Problem: Order L for routing algorithm such
that avg connection setup time is minimal
Pg.28, Dec 9, 2021
 Routing Algorithm
• Input: An ordered list of L candidate S-T paths
• Output: Feasible path P(i) or a rejection
message
1. i=1
2. Fwd connection setup request from S to T along
P(i)
3. If P(i) satisfies all constrains; path feasible
4. Else if i<L then i=i+1 return to step 2
5. Rejection: No feasible path

Pg.29, Dec 9, 2021

 Fuzzy-based Ordering
• Input:
– H(i): number of hops in P(i)
– Pf(i): failure probability of P(i)
• Output:
– Cost(i): Cost of P(i) in terms of connection setup
time
• Inputs used to define fuzzy rules and fuzzy
logic membership functions
• Output used to determine preference order of
connection paths
Pg.30, Dec 9, 2021
 Rule Matrix
• IF H(i) is LOW AND Pf(i) is LOW THEN Cost(i) is
VERY LOW
• IF H(i) is LOW AND Pf(i) is MEDIUM THEN
Cost(i) is LOW
H(i)
Low Medium High
Very Low Medium
Low Low
Pf(i) Medium Low Medium High

Medium High Very

High
High
Pg.31, Dec 9, 2021
 Membership Function
1.0
Degree of
Membership Low Medium High

5.5 10
Number of Hops

1.0
Degree of
Membership Low Medium High

40%
Probability of Failure

Pg.32, Dec 9, 2021

 Input Degree of Memberhip
Low Medium High

7 hops => 0.667

Medium, 0.33 High

Number of Hops

Low Medium High

10% Pf => 0.25

Medium, 0.75 Low

Probability of Failure

Pg.33, Dec 9, 2021

 Defuzzification
Rule H(i) Pf(i) Fuzzy Output
Consequent response
value
1 L: 0 L: 0.75 VL 0
H(i)=7
2 L: 0 M: 0.25 L 0
Pf(i)=10%
3 M: 0.667 L: 0.75 L 0.667
4 L: 0 H: 0 M 0
5 H: 0.33 L: 0.75 M 0.33
6 M: 0.667 M: 0.25 M 0.25
7 H: 0.33 M: 0.25 H 0.25
8 M: 0.667 H: 0 H 0
9 H: 0.33 H: 0 VH 0

Pg.34, Dec 9, 2021

 Defuzzification (2)
• Using the fuzzy output matrix and the output
response value:
– Very Low = VL = max{rule1}=0
– Low = L = max{rule2, rule3}=0.667
– Medium = M = max{rule4, rule5, rule6}=0.33
– High= H = max{rule7, rule8}=0.25
– Very High = VH = max{rule9}=0

Pg.35, Dec 9, 2021

 Defuzzification (3)
• The output center points of the defuzzification
method are:
– Very Low Center = vlc = 0
– Low Center = lc = 25
– Medium Center = mc = 50
– High Center = hc = 75
– Very High Center = vhc = 100
• The output center points of the defuzzification
method as explained in Fuzzy sets and Fuzzy Logic:
Theory and Applications by Klir and Yuan
Pg.36, Dec 9, 2021
 Defuzzification (4)
• Path cost calculated using the center of gravity
method
• Cost(pi)=
v lc ¤ V L + lc ¤ L + m c ¤ M + h c ¤ H + vh c ¤ V H
VL+ L+ M + H + VH = 41:64
• Calculate the cost of the rest of the paths and
sort them in an increasing order
Cost(P (1)) · C ost(P (2)) · : : : · C ost(P (L))
• Requests are forwarded according to routing
algorithm
Pg.37, Dec 9, 2021
 Example Simulation Result
Path Failure Success # Hops Optimal Greedy Fuzzy
Probability Probability

Upper 0.20 0.80 2 2nd 2nd 3rd

Middle 0.24 0.76 1 1st 3rd 1st

Lower 0.18 0.82 5 3rd 1st 2nd

T(Fuzzy) = 0.76 * (1 * 2) + 0.24 * 0.82 * (1 * 2 + 5 * 2) + 0.24 * 0.18 * 0.8 * (1 * 2

+ 5 * 2 + 2 * 2) = 4.4

T(Greedy)= 0.82 * (5 * 2) + 0.18 * 0.8 * (5 * 2 + 2 * 2) + 0.8 * 0.2 * 0.76 (1 * 2 + 2

* 2 + 5 * 2) = 10.65

T(Optimal)= 0.76 * (1 * 2) + 0.24 * 0.8 * (1 * 2 + 2 * 2) + 0.24 * 0.2 * 0.82 * (1 * 2

+ 2 * 2 + 5 * 2) = 3.3

Pg.38, Dec 9, 2021

 Simulation Results
• Fuzzy algorithm higher than optimal by 10%
while Greedy algorithm higher than optimal
by 90%

Average
connection
setup time

Probability of path failure

Pg.39, Dec 9, 2021

 Simulation Results (2)
• Fuzzy algorithm higher than optimal by 11%
while Greedy algorithm higher optimal by
100%

Average
connection
setup time

Average number of hops per candidate path

Pg.40, Dec 9, 2021

 Simulation Results (3)
• No significant gap between the three
algorithms

Average
number of
crankbacks per
connection

Probability of path failure

Pg.41, Dec 9, 2021

 Case Study Summary
• Average deviation of fuzzy algorithm from optimal
is 10%; more than acceptable for practical networks
• Significant advantage over optimal since it does not
require the knowledge of the exact pdf for each
additive QoS constraint of every link
• Easily implemented in routing protocols of
communication networks with minimal
computational overhead while significantly
improving overall setup time

Pg.42, Dec 9, 2021

 Possibility vs. Probability
Sum of possibility distribution Sum of probability distribution
does not need to sum to 1 should equal 1

Directed at the uncertainty in Concerned with uncertainty in

the description of an event the outcome of clearly defined
and randomly occurring events

Pg.43, Dec 9, 2021

 Problems and Limitations
• Stability: No theoretical guarantee that fuzzy
system does not go chaotic
• Learning capability: Capability of machine
learning, memory, or pattern recognition
through hybrid fuzzy-neural systems
• Determining or tuning good membership
functions not always easy
• Verification and validation requires extensive
testing with hardware in the loop.
Pg.44, Dec 9, 2021
 Conclusions
• Long term implications for industry and
business due to commercialization of fuzzy
logic
• Need to market innovative products at
cheaper price
• Can not afford to ignore fuzzy technology

Pg.45, Dec 9, 2021

 References
1. S. Dutta, Fuzzy Logic Applications: Technological and Strategic Issues, IEEE
Transactions on Engineering Management, Vol 40, No 3, pg 237 – 254, 1993
2. A. Cohen, E. Korach, M. Last, and R. Ohayon, A Fuzzy-based Path Ordering
Algorithm for QoS Routing in Non-deterministic Communication Networks, Fuzzy
Sets and Systems, Vol 150, pg 401-417, 2005
3. T. Munakata and Y. Jani, Fuzzy Systems: An Overview, Communications of the
ACM, Vol 37, No 3, pg 69-76, 1994
4. Carey Williamson, CPSC 641: Performance Issues in High Speed Networks
Homepage, Retrieved on 8 February 2005 from
http://pages.cpsc.ucalgary.ca/~carey/CPSC641/
5. Anirban Mahanti, CPSC 441: Computer Communications Homepage, Retrieved
on 8 February 2005 from http://pages.cpsc.ucalgary.ca/~mahanti/CPSC441/
6. S. Kaehler, Fuzzy Logic Tutorial, Retrieved on 4 February 2005 from
http://www.seattlerobotics.org/encoder/mar98/fuz/flindex.html
7. C. Elkan, The Paradoxical Success of Fuzzy Logic, Proceedings of the Eleventh
National Conference on Artificial Intelligence, pg 698—703, AAAI Press, 1993.

Pg.46, Dec 9, 2021