Expert Systems

Applied to Engineering

Class 2

Jonny Carlos da Silva, Dr.Eng.

UFSC- Mechanical Engineering Department

BRAZIL
Key points - last class

 Expert Systems- Definition

 Advantages
 Knowledge Engineering Process
 Interaction between Expert and Knowledge Eng.
 Incremental approach
 Course Evaluation Scheme
 Technical Analysis
 Prototype development
 Final test

–
Second class- outline

 Expert System- History

 Power of Heuristics
 Advantages Related to Expert Systems:
 Elements of an Expert System
 Typical differences between ES and
conventional programs
 Growth in the Application of Expert Systems
 Issues on Feasibility Study

–
Expert System- History

 In the early 1970s, AI researchers concluded: What

was needed, in relation to problem solving, was
specific knowledge about a particular and limited
application domain, instead of generic knowledge
applied to several domains (the previous GPS
paradigm).
 This conclusion led to the development of knowledge-
based systems or expert systems (ES).
 Expert Systems (ES) - becomes one of the main topics of AI.
 Evolution:
 simple concepts generated in research -> practical applications

 Power of Heuristics

ES Knowledge
=
Cause - effect

Practical Rules
Knowledge / experience of
(rules of thumb)
domain experts
Heuristics

HEURISTIC = informal knowledge, shortcuts, quick solution search, no detailed

analysis or comparative analysis.

A heuristic often provides a correct solution to a problem, but it can occasionally provide incorrect
answers. Application of an acceptable choice in return for a completely right one.

WHY DOES THIS OCCUR?

Distinction between heuristic and algorithmic
approach
Expert: has skills that
allows to complete from
experiences and quickly
focus on the core of a
given problem.
Non-Expert: Systematic, methodical and
procedural approach.

Example: A prospective home buyer with a defined architectural plan wants a cost
estimate before committing to closing the contract with a builder.

Case 1: A certain builder determines the price of the work after a detailed analysis
of the costs. (Algorithmic Approach)

Solution characteristics: Careful calculation, the advantage of practically

guaranteeing the precise result, very little risk for the manufacturer, involves
considerable effort and time.

WHAT HAPPENS IF THE BUYER NEEDS THE PRICE QUOTATION FOR

TODAY?
Distinction between heuristic and algorithmic
approach

Case 2: An expert builder. (Heuristic approach)

Solution characteristics: compare the size of the house with

others and analyze the differences, increase or decrease the
estimate, perform the estimate in about 30 minutes, quick
quotation, possibility of a poor consideration making the
estimate invalid.

HOWEVER, IF THE INDIVIDUAL IS A TRUE EXPERT,

IT IS UNLIKELY THAT WILL HAPPEN.
Some distinctions between
ES and conventional programs:
 The separation of knowledge from a specific
domain from how it is applied;
 Use of highly specific knowledge;
 Application of a heuristic approach, instead of an
algorithmic method;
 Ability to explain how the solutions were
obtained;
 Knowledge is presented explicitly.

–
 Different Perspectives on ES

Function Relationship
Manager What can I use it for?
Programmer How can I best implement it?
Researcher How can I expand it?
User How can this system help me?
Is it worth the changes and investment for acquisition?
How reliable is the system?
Advantages Related to Expert Systems:
 Increased availability of knowledge - more available experience;
 Cost reduction - cost of access to knowledge;
 Risk reduction - use in hazardous environments;
 Permanence - stored knowledge;
 Multiple expertises - combining the knowledge of several experts;
 Quick answer;
 Complete, consistent and impartial response;
 Intelligent tutor;
 Intelligent database;
 High performance - level of competence equal to or greater than a expert in
the field of knowledge.
 Explanation - an ES should explain in detail the
reasons that led to a conclusion;
Elements of an Expert System
 Key definitions
 User interface - communication mechanism between the user
and the ES.
 Explanation module - clarifies reasons found by the ES.
 Working memory - a "database" of facts used by the rules.
 Inference Engine - mechanism that decides which rules are
satisfied, prioritizes the satisfied rules and executes the highest
priority rule.
 Agenda - a list of the rules prioritized by Inference Engine,
whose conditions are satisfied by the facts or objects in the
working memory.
 Acquisition module - provides the user with an automatic way
to insert knowledge into the system, without requiring coding by
the Knowledge Engineer.

Key definitions
 RULE - corresponds to a small and modular
part of knowledge (chunk).
 Some comparisons:
 Models proposed by Newel and Simon's cognitive
theory (basis of expert systems)
 Operational Memory (facts) - acts as short-term
memory, since it is used for temporary storage of
knowledge during the solution of a problem.
 Knowledge Base (rules) - represents a long-term
memory, where permanent knowledge is stored.

–
 Short-term and long-term memories
... Although long-term memory can store hundreds of thousands of parts, the
capacity of short-term memory is very small.

TRY TO MENTALLY VIEW THE MAXIMUM NUMBERS POSSIBLE!
 HOW MANY CAN YOU VIEW?

In the search for a solution, a part of the knowledge is activated (or stimulated) and triggers
the activation of other parts.
–
For such a process to occur, it is necessary to have a cognitive processor that tries to
determine which RULES will be activated by a certain set of stimuli (FACTS).

Inference Engine - corresponds to the cognitive processor.

Analogy with a neuron

 Other Definitions

 Knowledge Engineering - ES development

process
 Knowledge Engineer - responsible for developing
an ES
 SHELL - computational tool to implement an ES.
 User - can be the end user (without domains
knowledge), the knowledge engineer or the
domain expert.


–
Typical differences between ES and
conventional programs
Feature Conventional Prog. Expert System

Control Explicit code Inference Engine

Control and data Implicit Integration Explicit separation

Solution Algoritmic Rules and Inference

Data input Assumed correct It can be incomplete, incorrect

Unforeseen Input Difficulty handling Very adequate

Output Always correct It varies with the problem

Explanation None Present

Applications Numeric, files and text Symbolic manipulation

Execution Usually sequential Most appropriate rules

System design Structured Little or none

Ease of change Difficult Reasonable

Expansion Via jumps Incremental

 Growth in the Application of Expert Systems
 Since the early 1970s (when the ES paradigm was established) several
systems have been developed, here are some examples:
– MYCIN (76) Diagnosis of disease based on blood sample and specification of
treatment for infection. Developed by STANFORD.

XCON/R1 (82) Computer systems configuration:

Note: Developed by Digital Equipament Corporation (DEC) in
partnership with Carnegie Mellon University, with its application, DEC
has saved millions of dollars per year.
PROSPECTOR (78) Geological analysis of data for mineral prospecting.
Obs: Through the PROSPECTOR a mine worth $ 100 million was
discovered.
ISIS (84) Generation of labor allocation plans for the factory floor. Developed by
Carnegie Mellon University and Westinghouse Electric Corporation.

Monitoring and diagnosis of operational conditions of large generators in

GENAID (86)
real time. Developed by Westinghouse Electric Corporation with support
from Texas Utilities and Carnegie Mellon University.

Monitoring and diagnosis of the liquid oxygen supply process in a

LES (87) spaceship tank. Developed by MITRE Corporation and NASA-KSC
(Kennedy Space Center)
Other ES Projects
PTRANS (83) Support for manufacturing control and distribution of DEC. Joint
development between DEC and Carnegie-Mellon University.

DENDRAL Interpretation of mass spectrograms to identify chemical elements.

DELTA (83) Diagnosis / Repair of GE locomotives. Developed by General Electric .

INTERNIST (82) Support to the physician in the process of diagnosing hospital infections.
Developed by the University of Pittsburgh.

REACTOR Reactor accident diagnosis / repair.

STEAMER (84) Operating instructions for steam engines. Developed by the American
Navy in collaboration with BBN Corporation.

CADHELP CAD Instruction

SOPHIE Instruction for fault diagnosis in electronic circuits.

DIPMETER Geological analysis of data for oil prospecting.

Feasibility of Expert Systems for certain
types of problems
Correct application + correct reason = positive influence on the final success of the project

Right application + inadequate reason = end of the project not so satisfactory.

The criteria for analyzing the viability of ES consist of several factors

that can be grouped into two categories:
1. Suitability of the application area
2. Availability of resources
Adequacy of the application area

Many project failures involving SE can be attributed to an inappropriate definition of the application.
Some questions to consider: ·

Does the problem really exist?

This is the first question to be considered. The risk of developing a project for a problem that doesn't
really exist is that the solution will not have a real impact on the organization's operation.

The project must represent considerable value for the organization. The higher the value,
the more the organization will be willing to commit to the development of the system.
Some key questions
 Are the problem entries always correct and complete?
 Can the problem be solved by other means?
 Does the problem go through the phone test?
 Is the Expert System approach justified?
 Resource Availability
 Even though a specific problem may satisfy all of the
above criteria, it may still not become a successful
implementation. Some factors can compromise the
project:
–
Resources
 The knowledge engineer must be highly trained.
 As such projects do not exist in isolation, they are
subject to some restrictions commonly encountered in
government, industry, and academic organizations
that affect project development.
 Therefore, the next questions must be considered very
carefully.
 Is there managerial support for the project?
 Time availability?
 Tools and training?
–
Questions related to the EXPERT
 Expert´s availability?
 Is there support from Expert?
 Is the expert competent?
 Is the expert well articulated?
 The highly qualified expert in the problem who says:
"the solution is so obvious, that anyone can see" or
 "I cannot explain why this is the solution only that it is",
not of great assistance.
 Is the expert physically close?