Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


2. BACKGROUND AND HISTORY

A majority of early expert systems were diagnostic (or classification) systems, especially in the medical, electronic, and manufacturing domains (Buchanan and Shortliffe, 1984). We can better understand the reasons why model-based reasoning techniques were developed by briefly recapitulating the history of diagnostic systems. Early systems like MYCIN and INTERNIST relied on a knowledge base containing diagnostic rules:

IF symptoms1 THEN fault1
IF symptoms2 THEN fault2
Etc.

A rule is a direct link between an observed set of symptoms and the fault (or disease in the medical domain) that these symptoms indicate. Many extensions to this simple pattern were needed in practice, such as certainty factors, screening rules, etc. Each rule expresses a fragment of a domain expert's knowledge in a direct, compiled form. The advantages are easy comprehension and very efficient application (by a rule interpreter) of knowledge in this form. Indeed, many successful experts systems were, and still are, developed along those lines.

However, it soon became clear that the compiled diagnostic rule format for expressing diagnostic knowledge had distinct drawbacks:

  • The system could not provide any conclusion if the expert had not entered a rule covering the exact symptoms that were observed.
  • The expert reasoning behind the compiled form was lost, thereby impeding understanding and verification by other professionals.
  • Unless carefully planned, the collection of diagnostic rules soon became unwieldy, and hard to maintain and improve upon.
  • Building a new system often meant starting from scratch with a new expert, not being able to capitalize on knowledge already available in other rule systems.

In short, diagnostic systems developed according to the compiled rule paradigm tend to be lacking in coverage, clarity, flexibility, and cost-effective extendability. The model-based reasoning paradigm was developed, first in the diagnostic domain, in order to answer these deficiencies. In an MBR-based diagnostic system, the set of diagnostic rules is replaced by a model of the target system (patient, device), defining its structure and behavior (Davis, 1984):

  • Structural model: What are the constituent parts of the target system, and how are the parts related (in a topological, spatial, and/or temporal sense)?
  • Behavioral model: Given a set of "input" stimuli, what are the "output" responses of the target system? The causality of the target is often modeled.

Later, we will categorize and see specific examples of structural/behavioral models. Using a model of this kind, an MBR diagnostic engine is able to:

  • Use the behavioral model to detect discrepancies between the faulty and normal behavior of the target system, thereby indicating symptoms.
  • Use the behavioral and structural models to trace back from observed discrepancies to the underlying root causes, thereby indicating a diagnosis.

This approach has a number of advantages over the compiled rule approach, including:

  • The diagnostic system is able to diagnose "new" faults, faults that could not be acquired by just formalizing empirical knowledge.
  • Engineering or scientific knowledge, available in textbooks and other sources, can be applied in building the models, thereby avoiding personal biases.
  • The models are compositional, thereby allowing the gradual accumulation of model component libraries and easing model construction.
  • The MBR approach lends itself to better explanation of conclusions and improved verification of diagnostic capabilities.

The MBR approach is not without problems, including:

  • It often involves reasoning algorithms with combinatorial complexity, leading to excessive or unacceptable computation time.
  • Creating the required model may include insight and effort that is not available, or the model may in fact be unknown (e.g., certain physiological processes).

The distinction between expert systems built using the compiled rule paradigm and systems using models were recognized during the 1980s. The two types of systems were given different labels, such as "shallow" (compiled rules) and "deep" (model-based) systems. Model-based systems were also called "second-generation expert systems." Many researchers and practitioners saw the advantages of both approaches, and proposed hybrid solutions. In the diagnostic domain, these hybrids generally work along the following principles:

  • The knowledge base contains both a set of diagnostic rules and a model, and the system has appropriate inference engines for both types of knowledge.
  • Confronted with a diagnostic case, the rules are first applied. If they yield a solution, the search is over.
  • If the rules fail, the MBR machinery is invoked. It will hopefully produce a solution, possibly after spending considerable time on this.
  • If successful, the MBR solution is formulated as a rule and added to the rule base for efficient solution of similar cases in the future.

The MBR paradigm is not restricted to the diagnostic domain, even if most results have been reported for diagnostic applications. Other types of problems that have been attacked with this approach include design/configuration problems, and monitoring and control problems (the latter two often in combination with diagnostic systems). A recent survey of methods and applications of MBR can be found in Dague (1995). In a later section, some examples of application in these various domains will be described briefly.


Previous Table of Contents Next

footer nav
Use of this site is subject certain Terms & Conditions.
Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details.