You are on page 1of 70

Expert System Examples 1

Expert System Examples


Expert System Examples 2
R1/XCON
Expert System Examples 3
N
one of the first commercially successful expert systems
N
application domain
N
configuration of minicomputer systems
N
selection of components
N
arrangement of components into modules and cases
N
approach
N
data-driven, forward chaining
N
consists of about 10,000 rules written in OPS5
N
results
N
uality of solutions similar to or better than human experts
N
roughly ten times faster !" vs# "5 minutes$
N
estimated savings %"5 million&year
Expert System Examples 4
System Configuration
N
complexity
N
tens or hundreds of components that can be arranged in a multitude
of ways
N
in theory, an exponential problem
N
in practice many solutions ''don(t ma)e sense((, but there is still a
substantial number of possibilities
N
components
N
important properties of individual components
N
stored in a data base
N
constraints
N
functional constraints derived from the functions a component
performs
O
e#g# *P+, memory, ,&O controller, dis)s, tapes
N
non-functional constraints
O
such as spatial arrangement, power consumption,
Expert System Examples 5
Knowledge Representation
N
configuration space
N
constructed incrementally by adding more and more
components
O
the correctness of a solution often can only be assessed after it
is fully configured
N
subtas)s are identified
O
ma)e the overall configuration space more manageable
N
component )nowledge
N
retrieved from the external data base as needed
N
control )nowledge
N
rules that govern the seuence of subtas)s
N
constraint )nowledge
N
rules that describe properties of partial configurations
Expert System Examples 6
Example Component
N
partial description of -./110 dis) controller
N
facts are retrieved from the data base and then stored in
templates
RK611*
Class: UniBus module
Type: disk drive
Supported: yes
Priority Level: buered !PR
Transer Rate: "1"
# # #
Expert System Examples 7
Example Rule
N
rules incorporate expertise from configuration experts, assembly
technicians, hardware designers, customer service, etc#
$istribute%&B%$evi'es%(
) t*e most 'urrent a'tive 'onte+t is
distributin, &assbus devi'es
- t*ere is a sin,le port disk drive t*at *as not been
assi,ned to a &assbus
- t*ere are no unassi,ned dual port disk drives
- t*e number o devi'es t*at ea'* &assbus
s*ould support is kno.n
- t*ere is a &assbus t*at *as been assi,ned
at least one disk drive and t*at s*ould support
additional disk drives
- t*e type o 'able needed to 'onne't t*e disk drive
to t*e previous devi'e is kno.n
T*en assi,n t*e disk drive to t*e &assbus
Expert System Examples 8
Configuration Task
N
chec) order1 identify and correct omissions, errors
N
configure *P+1 arrange components in the *P+ cabinet
N
configure +ni2us modules1 put modules into boxes, and
boxes into expansion cabinets
N
configure panels1 assign panels to cabinets and associate
panels with modules
N
generate floor plan1 group components and devices
N
determine cabling1 select cable types and calculate distances
between components
N
this set of subtas)s and its ordering is based on expert
N
experience with manual configurations
Expert System Examples 9
Reasoning
N
data-driven !forward chaining$
N
components are specified by the customer&sales person
N
identify a configuration that combines the selected
components into a functioning system
N
pattern matching
N
activates appropriate rules for particular situations
N
execution control
N
a substantial portion of the rules are used to determine
what to do next
N
groups of rules are arranged into subtas)s
Expert System Examples 10
Performance Evaluation
N
notoriously difficult for expert systems
N
evaluation criteria
N
usually very difficult to define
N
sometimes comparison with human experts is used
N
empirical evaluation
N
3oes the system perform the tas) satisfactorily4
N
5re the users&customers reasonably happy with it4
N
benefits
N
faster, fewer errors, better availability, preservation of
)nowledge, distribution of )nowledge, etc#
N
often based on estimates
Expert System Examples 11
Development of R1/XCON
N
-1 prototype
N
the initial prototype was developed by *arnegie 6ellon
+niversity for 37*
N
8*O9 commercial system
N
used for the configuration of various minicomputer system
families
N
first :58 11&;<0, then :58 11&;50, then other systems
N
reimplementation
O
more systematic approach to the description of control )nowledge
O
clean-up of the )nowledge base
O
performance improvements
Expert System Examples 12
Extension of R1/XCON
N
addition of new )nowledge
N
wider class of data
N
additional computer system families
N
new components
N
refined subtas)s
N
more detailed descriptions of subtas)s
N
revised descriptions for performance or systematicity reasons
N
extended tas) definition
N
configuration of ''clusters((
O
tightly interconnected multiple *P+s
N
related system 8S7=
N
tool for sales support
Expert System Examples 13
Summary R1/XCON
N
commercial success
N
after initial reservations within the company, the system was fully
accepted and integrated into the company(s operation
N
widely cited as one of the first commercial expert systems
N
domain-specific control )nowledge
N
the availability of enough )nowledge about what to do next was critical
for the performance and eventual success of the system
N
suitability of rule-based systems
N
appropriate vehicle for the encoding of expert )nowledge
N
sub>ect to a good selection of application domain and tas)
Expert System Examples 14
MYCIN
Expert System Examples 15
Overview
N
?istory
N
3793-5=
N
6@*,9
N
2ac)ground
N
.nowledge -epresentation
N
.nowledge 6anipulation
N
+ncertainty
N
Performance 7valuation
N
5dvantages and Problems
N
-eferences
Expert System Examples 16
DENDRAL
N
*ommonly considered the first expert system
N
3eveloped at Stanford in the late 1A/0s
N
7d Beigenbaum !a *S* Professor$
N
2ruce 2uchanan !a philosopher turned computer scientist$
N
Coshua =ederberg !a 9obel =aureate Deneticist$
N
5nalyEed 96- mass spectrogram FGHGIJKMNOPJK!$ data to
determine the geometric arrangement of atoms in a molecule
Expert System Examples 17
MYCIN Background
N
6edical expert system
N
3eveloped at Stanford in the 1A;0s by Beigenbaum, 2uchanan and
Qed Shortliffe !a doctor$
N
-ecommended therapy for blood&meningitis infections !RSRTUJKVRWXJK!
O
the diagnosis normally involved growing cultures of the infecting organism
!Y< hours$
O
3octors had to come up with uic) guesses about li)ely problems
O
Prescribe drugs to deal with immediate problems
N
3eveloped to explore how doctors ma)e these rough, but important,
guesses with partial information
O
5lso important in practice as there are many >unior doctors or non-
specialiEed doctors
Expert System Examples 18
MYCIN Implementation
N
Doal-directed system that uses a basic bac)ward-
chaining techniue
N
Y50 -ules written in =,SP
N
Performed as well as some experts and significantly
better than >unior doctors
N
9ever actually used in practice
N
9ot due to its performance
N
2ut rather ethical and legal issues
Expert System Examples 19
Example Rule
N
,f
N
Qhe site of the culture is blood
N
Qhe gram of the organism is neg
N
Qhe morphology of the organism is rod
N
Qhe burn of the patient is serious
Qhen
there is wea)ly suggestive evidence !0#Y$ that the identity
of the organism is pseudomonas
Expert System Examples 20
Representation
N
-ules had no variables, contexts instead
N
6@*,9 dealt with a number of implicit variables
N
Bor example there could be a patient, a culture, a few
infectious organisms#
N
6@*,9Zs )nowledge structured into [ob>ect-
parameter-value\ triples
N
[culture\ would be an ob>ect
N
[site\ would be a parameter of [culture\
N
a possible value of this parameter would be [blood\
Expert System Examples 21
Manipulation
N
6@*,9 starts out with a rule that says
N
,f there is an organism reuiring therapy, then, compute
the possible therapies and pic) the best one
N
Birst tries to see if the disease is )nown
N
if it isnZt begins reasoning process
N
2asic routine in 6@*,9
N
attempt to find the value of a parameter
Expert System Examples 22
Finding Values
N
3epending on the type of data may as) user if the value is
)nown
N
Qried to as) the most general uestion possible, so as not to
become annoying or repetitive
N
7#g#, if 6@*,9 wants to )now if morphology of organism is rod, will
as) []hat is morphology of organism4\ rather than a specific uestion
repeatedly
N
Bormat of .- is supposed to ma)e uestions reasonable
N
,f the value is not )now, 6@*,9 does bac)ward
chaining
N
Stores a list of rules that might yield a value for each
parameter
Expert System Examples 23
Uncertainty
N
6edical field must reason in the presence of
un)nown, incomplete, vague or uncertain information
N
6@*,9 used [certainty factors\
O
initially hard to defend from a sound theoretical viewpoint
O
theoretical foundations were established later !3empster-Shafer$
O
useful to see where )nowledge about uncertainty exists, and the
implications it has for the design of the system
Expert System Examples 24
Certainty Factors
N
-ange from ^1 !positive it is not the case$ to _1 !positive it is
the case$#
N
6@*,9 maintains certainty for
N
possible values of parameters !ultimately, the certainty that you have
a particular disease$
O
can maintain multiple possible values, each with its own certainty
N
validity of a rule
N
6@*,9 has rules for combining the certainty factors
Expert System Examples 25
Performance Evaluation
N
Shortliffe used
N
10 sample problems
N
< other therapy recommenders
O
5 faculty at Stanford 6ed# School, 1 senior resident, 1 senior
postdoctoral researcher, 1 senior student
N
< impartial >udges gave 1 point per problem
N
6ax score was <0
N
6@*,9` /5, Baculty` Y0-/0, Bellow` /0, -esident` Y5,
Student` a0
Expert System Examples 26
Controls
N
CudgesZ bias for&against computers
N
Cudges did not )now who recommended each therapy
N
3ifficulty of problems
N
6edical student did badly, so problems not easy
N
=evel of ,nterest
N
?ypothesis in 6@*,9 that [)nowledge is power\
N
?ave groups with different levels of )nowledge
Expert System Examples 27
Good Points
N
6@*,9 was good in that
N
,t could calculate dosages very precisely
N
3ealt well with interactions between drugs
O
5n area in which humans have trouble
N
Possesses nice explanation facilities
O
-etrieves and displays relevant rules to offer explanation of its
behavior
Expert System Examples 28
Difficulties
N
9arrow in scope
N
did not scale up well to larger problems
N
Practical concerns
N
3octors have reservations about advise from computers
N
=egal issues
Expert System Examples 29
Human Resources Expert System
N
expert systems to determine conditions and entitlements
Gbcd e for public employees in 9ew South ]ales, 5ustralia
N
main user groups
N
employees
N
managers
N
?- staff
N
accessible via ,nternet
N
http`&&www#premiers#nsw#gov#au&]or)5nd2usiness&]or)ingBorDovernment&?-7xpert#htm
N
some functionality limited to authoriEed users
N
developed by Softlaw *orporation http`&&
www#softlaw#com#au
Expert System Examples 30
Objectives
N
improve ?- advice and information
N
uality, consistency, timeliness
N
enable value-adding strategic functions
N
e#g# wor) force planning
N
extend use of technology from transaction-based 7S
to advice and information systems
Expert System Examples 31
HR Expert Principles
N
enhanced electronic decision tree
N
on-line inuiries from users determine branches
N
accessible via official ?- web sites
N
integrated with source documents
N
legislation, personnel handboo), etc#
Expert System Examples 32
HR Expert: Inquiries
Expert System Examples 33
HR Expert: Service History
Expert System Examples 34
Output
N
summary screens
N
reports
N
letters
N
applications and forms
N
audit reports
Expert System Examples 35
HR Expert: Summary Report
Expert System Examples 36
HR Expert: Full Report
Expert System Examples 37
Project
N
phase 1 - pilot pro>ect
N
a agencies, 1,"50 staff, conducted in "00"-a
N
demonstrated potential savings, user satisfaction, ualitative benefits
N
phase "
N
extension to all relevant conditions and entitlements
N
to be operational by 6ay "00Y
N
cited in the report of the 5ustralian Dovernment - ,nformation
6anagement office as an example
N
technology
N
legislative rulebase technology, SQ5Q+Q7 7xpert, by Softlaw *orp#,
*anberra, 5ustralia, http`&&www#softlaw#com#au
Expert System Examples 38
Benefits
N
employees
N
immediate and up to date
information about conditions
and entitlements
N
easy access for inuiries
N
improved data for decisions
N
increased euity
N
on-demand generation of
reports
N
standardiEed outputs and
audit reports
N
?uman -esources
N
direct access to information
about entitlements
N
less tedious wor)
O
e#g# loo)ing up information
when employees need it
O
reduced need for repetitive
wor)
N
more consistent decisions
N
on-demand generation of
reports
N
standardiEed reports
Expert System Examples 39
Issues
N
some of the input provided by the users
N
not always accurate, up to date
N
only generic conditions and entitlements
N
special cases not included
N
limited coverage
N
not all laws and regulations included
N
reuires computer and ]eb access
N
commitment and buy-in from staff and employees
Expert System Examples 40
Status
N
operational and in use
N
update to include recent changes in laws and regulations under way
N
current modules
N
6aternity =eave
N
Study Qime
N
7xtended =eave
N
-ecognition of Previous Service
N
=eaving the Service
N
:oluntary -edundancy
N
Qravel *ompensation, including 6eal and Private 6otor :ehicle
5llowances
N
?igher 3uties 5
N
llowance Salary
N
Pac)aging 5gency
N
=ist ,nuiry
Expert System Examples 41
RuleBurst Demo
N
a Blash demo of the -ule2urst enviroment is
available at http`&&
www#ruleburst#com&uploads&files&-ule2urst#html
N
a predecessor of -ule2urst was used to develop the ?-
7xpert application
Expert System Examples 42
References
N
?- 7xpert *ase Study at
http`&&www#agimo#gov#au&resources&ppt&"00a&0a0A"/sb
N
9S] governmental web site at
http`&&www#premiers#nsw#gov#au&]or)5nd2usiness&]or)ingBorDovernment&?-7xpert#htm
N
5ustralian Dovernment ,nformation 6anagement Office -eport at
http`&&www#agimo#gov#au&publications&"00Y&05&egovtfchallenges&accountability&determinations&conclusion
N
Softlaw *orporation ]eb site http`&&www#softlaw#com#au
N
Softlaw ?- 7xpert 5nnouncement
http`&&www#softlaw#com#au&content#cfm4categoryidg1"htopicidgYAhinfopageidg15"
N
-ule2urst .2 3evelopment 7nvironment http`&&www#ruleburst#com&
sites visited 03-02-05, 02-28-06
Expert System Examples 43
OSHA Hazard Awareness Advisor
N
as)s uestions about wor)place activities,
euipment, materials
N
analyEes the userZs answers
N
generates a report with common occupational
haEards, applicable OS?5 standards, and contacts
N
developed by the +#S# 3epartment of =abor,
Occupational Safety h ?ealth 5dministration !OS?5$
N
version 1#0 released in September 1AAA
N
http`&&www#osha#gov&dts&osta&oshasoft&haEexp#html
Expert System Examples 44
Objectives
N
to help identify and understand common safety and
health haEards in the wor) place
N
especially aimed at small businesses
N
designed for beginners
N
may be useful for experts as well
N
widely available through online and downloadable
versions
N
downloadable version only for 6S ]indows
Expert System Examples 45
Limitations
N
may not identify all haEards
N
will not determine compliance with OS?5 standards
N
not a substitute for safety and health professionals
Expert System Examples 46
OSHA Principles
N
expert system technology
N
accessible via official OS?5 web sites
N
integrated with source documents
N
standards, legislation, etc#
Expert System Examples 47
Hazard Awareness Advisor Inquiries
Expert System Examples 48
Hazard Awareness Advisor Report
N
highlights
N
details
Expert System Examples 49
Report Highlights
_ 7valuate the exposure to chemicals in your wor)place#
_ @our site needs a haEard communication program#
_ ,nspect ladders and ensure that wor)ers )now how to use them safely#
_ =adders should be at least a feet higher than the level they are going to reach#
_ *hec) that fire exits are unloc)ed and numerous enough for uic) escape of all wor)ers#
_ .eep passageways clear of obstructions#
_ 7valuate personal protective euipment that your wor)ers purchase for themselves#
_ Protective eye wear may be needed to protect against splashes and sprays#
_ +se laser pointers carefully# Qhey can cause eye damage#
_ Please investigate the need for head protection#
_ Please investigate the need for hard toed shoes#
_ Protective gloves may be needed because of in>uries by )nives or other hand tools#
_ Portable fire extinguishers must have maintenance service at least once a year and a written
record must be )ept to show the maintenance or recharge date#
_ 6ar) fuse boxes or brea)er boxes to identify the circuits or euipment they control#
_ 7xtension cords should not be used as a substitute for permanent wiring#
_ Qa)e care when using cleaning solvents and liuids when cleaning inside a series of deep
cabinets or similar spaces#
_ ,f you have both ammonia and bleach cleaners, ta)e care in their storage and use# Qhe mixing
of ammonia and bleach can produce dangerous chlorine gas#
Expert System Examples 50
OSHA Report Detail: Portable Ladders
@our answers indicate that your wor)ers use portable ladders# Qhe use of
any ladder is haEardous# ]or)ers may fall from them, fall with them, be
struc) by falling ladders or struc) by ob>ects dropped from wor) being
performed on the ladder#
,n>uries also result from poor ladder placement` unstable footing, wor) angle
too steep or too shallow, or placement in front of doors or passageways#
6any serious falls from ladders are the result of wor)ers standing above
the designed wor)ing height of the ladder#
Qhe haEards of ladder use can be reduced by careful selection of ladders of
appropriate height and strength, by routine inspection and maintenance,
and by training of wor)ers in safe ladder use#
,n order to safely gain access to an upper level such as a roof or platform,
the portable or extension ladder must extend at least a feet above the
point of contact# 5ny portable ladders should be tied off or held in position
during use#
Expert System Examples 51
Project
N
preceded by other S?5 advisors
N
5sbestos in (A5
N
*onfined Spaces in (A/
N
Bire Safety in (A;
N
=ead in *onstruction in (A<
N
input from
N
9ational Bederation of ,ndependent 2usiness,
N
9ational 5partment 5ssociation,
N
Synthetic and Organic *hemical 6anufacturers 5ssociation,
N
+nited 2rotherhood of *arpenters and Coiners,
N
=aborers Safety and ?ealth Bund,
N
,nternational 2rotherhood of Qeamsters,
N
other industry and labor organiEations
Expert System Examples 52
Benefits
N
owners&managers
N
easy access for inuiries
about potential haEards
N
uic) analysis of the
wor)place
N
generation of a report with
highlights, details, and
pointers for further information
N
employees
N
identification of potential
haEards
N
improved wor)ing conditions
N
possible compliance issues
O
standards and regulations
Expert System Examples 53
Issues
N
all of the input provided by the users
N
not always accurate, up to date
N
limited coverage
N
may not identify all haEards
Expert System Examples 54
Status
N
version 1#0 operational and in use since September 1AAA
N
update to include recent changes in laws and regulations 444
Expert System Examples 55
Other OSHA Expert Advisors
N
see http`&&www#osha#gov&dts&osta&oshasoft&index#html
N
5sbestos
N
*onfined Space
N
7lectronic Permit -euired *onfined Spaces !e-P-*S$
N
7lectronic ?ealth and Safety Plan !e-?5SP$
N
Bire Safety
N
?aEard 5wareness
N
=ead in *onstruction
N
=ead in Deneral ,ndustry
N
=oc)out&Qagout
O
=OQO Plus
N
Safe*are
N
%afety Pays
Expert System Examples 56
References
N
OS?5 ?aEard 5wareness 5dvisor, :ersion 1#0 September 1AAA, http`&&
www#osha#gov&dts&osta&oshasoft&haEexp#html
N
Stern, 7d !1AA<$` [OS?5 +nveils Online ?aEard 5wareness 5dvisor\,
5ccess 5merica Dovernment Services,
http`&&govinfo#library#unt#edu&accessamerica&docs&expertadvisor#html
N
Stern, 7d !1AAA$` [Qhe OS?5 ?aEard 5wareness 5dvisor\, P* 5,
6agaEine, vol 1a, no ", 6arch&5pril 1AAA
N
Shirley, -obin 7# !"00$` [9ew OS?5 ,nteractive Software 3esigned to
?elp Small 2usiness Owners\, On Qarget - 9ews for the Small 2usiness
Owner, http`&&www#reswritingservices#com&osha#html
N
:irginia ]or)ersZ *ompensation Program !"005$` [?aEard 5ssessments\,
http`&&www#covwc#com&lcarticles&archives&0000<Y#php
N
OS?5 eQools and 7lectronic Products for *ompliance 5ssistance http`&&
www#osha#gov&dts&osta&oshasoft&index#html
sites visited 07-14-05, 02-28-06
Expert System Examples 57
Gensym G2
N
real-time expert system
N
developed by Densym *orporation http`&&
www#gensym#com
N
application areas
N
chemical, oil h gas, process manufacturing, discrete
manufacturing, power utilities, water utilities,
telecommunications, government, transportation,
aerospace
N
augmented by additional modules
Expert System Examples 58
Gensym G2 Platform
Expert System Examples 59
Gensym G2 Use
http://www.gensym.com/images/pages/g2platform.jpg
Expert System Examples 60
Gensym Development Cycle
http://www.gensym.com/images/pages/xtreme-programming-lifecycle.jpg
Expert System Examples 61
Ericsson Network Management
System
N
use of D" for wireless networ) management
N
challenges in wireless networ)s
N
additional functionality
O
instant messaging, chat, ]eb access, photos, videos, i
N
increased siEe and complexity of the networ)
N
very rapid growth and change rate
N
networ) management can be very stressful
N
constant stream of alarms
O
not all are important
N
extreme pressure to identify and fix problems
Expert System Examples 62
Ericsson FMX
N
largely automated system for wireless networ) management
N
concentration on the fault management process
N
reacts to all identified events very uic)ly
O
much faster than humans
O
more reliable
O
but less flexible
N
filters out unimportant messages
N
allows networ) operators to concentrate on critical events
N
consolidates information for critical events
N
manages over 500,000 events per day
N
500 systems in 100 countries
N
50 different euipment types
N
benefits
N
increased uality of service
N
reduced operating expenses
Expert System Examples 63
FMX Screenshot
Input
Condition
Decision
Outputs
http://www.gensym.com/?p=success_stories&id=13
Expert System Examples 64
Dow Chemicals Closed Loop
Optimizer
N
energy management in a large petrochemical plant
in Seadrift, Q8
N
highly interdependent systems
N
real-time control
N
safety-critical
N
very high energy costs
N
utiliEation of waste heat from gas turbines
N
internal energy usage
N
excess energy sold to the grid
N
manual control of power generation was problematic
N
trade-off considerations must be made very fast
Expert System Examples 65
Energy Management with G2
N
modeling of the energy system
N
sensors provide input
N
important components are modeled
N
output controls actuators, informs operators
N
previous models of individual systems were not successful
for the overall energy management
N
optimiEation
N
determines the best operational plan for the current
conditions in real time
Expert System Examples 66
Seadrift Functional Diagram
juic)Qimek and a
Q,BB !+ncompressed$ decompressor
are needed to see this picture#
Expert System Examples 67
Seadrift Screenshot
juic)Qimek and a
Q,BB !+ncompressed$ decompressor
are needed to see this picture#
Expert System Examples 68
Seadrift Outcome
N
plant ran in closed loop mode A< percent of the time
N
saved 3ow %1#"5 million dollars in energy costs over
one year
N
even larger potential for savings
O
extension to other components and systems in the plant
O
more sophisticated modeling
O
usage for other plans
N
user satisfaction
N
operators were s)eptical initially, but accepted and used
the system very uic)ly
N
better view of overall plant operations
Expert System Examples 69
References
N
Densym *orporation http`&&www#gensym#com
N
Densym [Success Stories\` 3ow *hemicals http`&&
www#gensym#com&4pgsuccessfstorieshidg<
N
3ow *hemicals Seadrift Plant http`&&
www#dow#com&ucc&locations&seadrift&about&index#htm
Expert System Examples 70
Important Concepts and Terms
N
agenda
N
bac)ward chaining
N
common-sense )nowledge
N
conflict resolution
N
expert system !7S$
N
expert system shell
N
explanation
N
forward chaining
N
inference
N
inference mechanism
N
,f-Qhen rules
N
)nowledge
N
)nowledge acuisition
N
)nowledge base
N
)nowledge-based system
N
)nowledge representation
N
6ar)ov algorithm
N
matching
N
Post production system
N
problem domain
N
production rules
N
reasoning
N
-7Q7 algorithm
N
rule
N
wor)ing memory

You might also like