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


4.5. DIAGNOSIS

Model-based diagnosis (MBD) has received more widespread practical application than any other application of MBR techniques. It appears that this is a particularly fertile ground for using these techniques. Diagnosis is a task of high practical and direct economic significance. Improving availability of, e.g., a manufacturing plant by as little as a few tenths of a percent can yield enormous benefits. A survey of techniques and applications of diagnostic advisory systems, including systems using MBR techniques, can be found in Kramer and Fjellheim (1995).

One can generalize the flow of activities in the diagnostic process to the following steps:

  1. Fault detection: The fault manifests itself by some observable symptom. In the MBD approach, detection will be done by comparing actual with model-predicted values of one or several variables.
  2. Fault isolation: Based on the detected symptom(s), one will try to trace back to the location of the underlying cause, e.g., which component in the circuit causes the faulty output. Again, a model may support this tracing.


    FIGURE 2 Generic MBD system architecture.

  3. Fault identification: It may be desirable to go one step further after isolation, namely to identify the precise nature of the localized fault (e.g., short circuit). In MBD, this ability assumes fault models in addition to the nominal model.
  4. Fault correction: Finally, a complete troubleshooting session should conclude with a recommended action, such as to repair or replace a component. This falls outside the scope of MBD, but may be an essential part of a deployed system.

A generic system architecture that supports this diagnostic process, using the model-based approach, is shown on Figure 2 (Leitch et al., 1992). The architecture is representative of a large number of MBD applications, even though details may differ significantly from system to system.

In this architecture, a system model is run in parallel with the system itself, producing predictions of what the observations "ought to be." A discrepancy detector module tracks observations and predictions, and triggers on significant deviations (note that this may in itself be a nontrivial operation, perhaps involving data validation and statistical processing). The discrepancies are fed into a candidate generator module, which proposes one or more fault candidates. These are used to modify the behavior of the system model according to the hypothesized fault. If the predictions now match the observations, we know that the fault has been identified. The whole diagnostic process is steered by a diagnostic supervisor module, which may range in complexity from a simple invocation interface to a sophisticated strategy reasoner.

A recent and very substantial MBD application is the TIGER system (Travé-Massuyés and Milne, 1996). The application domain of TIGER is monitoring and diagnosis of gas turbines, an area of large industrial interest. The TIGER system has been deployed at a large petroleum company. Some technical features characterizing this application are:

  • It uses a qualitative model of the turbine and a qualitative simulation algorithm to predict normal turbine behavior.
  • Fault detection is done by matching the actual measurements with the simulated normal behavior (just as in the reference architecture).
  • The diagnosis algorithm is patterned after the framework of Reiter (1987): it collects conflict sets, i.e., sets of components that cannot behave normally according to the observations.
  • Augmenting the MBD algorithm, TIGER also has a real-time expert system capable of monitoring alarms quickly and with guaranteed response time.
  • Finally, the system has a temporal reasoning module for expressing and monitoring temporal dependencies in the turbine.

Other successful expert systems that use an MBD approach are (many other systems could have been listed):

  • DCP (Fjalestad et al., 1994). Uses a mixed topological/numerical model to support pollution monitoring and diagnosis at a fertilizer plant.
  • GIOTTO (Cermignani and Tornielli, 1993). A generic tool for building MBD systems for continuous and static processes.
  • KARDIO (Bratko et al., 1989). Uses a qualitative model of the human heart to generate a diagnostic system for heart diseases.


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.