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


4.3. MONITORING

Earlier, we defined monitoring as the activity of tracking and comparing observations to expectations. Many examples of monitoring expert systems, using the MBR approach, can be found in the manufacturing and process industries. These industries already use monitoring systems based on conventional techniques. For example, this may mean tracking the value of a particular measurement over time, and warn the operator if the value exceeds a particular fixed alarm threshold. The resulting monitoring performance is often very crude, and may lead to excessive alarming (too narrow alarm limits), or under-reporting (too permissive limits). The trouble is that the monitoring limits are fixed and not related to the actual running conditions of the plant (device).

This is in contrast to the MBR approach, where monitoring is done by running a model of the physical system in parallel the system itself, producing expected values for different observations (measurements), comparing the actual observations of the system with those that the model predicts (dynamically), and reporting significant deviation. Dvorak and Kuipers (1991) presented an MBR monitoring application, called MIMIC. It exploits three different techniques:

  • Semiquantitative simulation. In Section 3 we described qualitative simulation as a reasoning technique with a number of desirable properties for inferring behavior when a purely qualitative model exists. In MIMIC, additional numerical information that is always available in real plants, is used to improve the precision of the predictions. In particular, it removes inconsistent solution and permits direct comparison of measurements with numerical predictions.
  • Tracking (measurement interpretation). MIMIC tries to maintain a set of models ("the tracking set") that is consistent with the measurements, and which represents a possible state of the target system. During diagnosis (see below), MIMIC adds interpretations to the tracking set that represent hypothesized faults. During tracking, it deletes models from the set when predictions fail to match observations.
  • Model-based diagnosis. MIMIC also performs diagnosis as part of an integrated problem-solving architecture. Its role is to update the model (tracking set). We will return to model-based diagnosis in a subsequent section.

The MIMIC architecture was applied to a water heating system on an experimental basis. Among the advantages quoted by the authors of their approach are:

  • More "intelligent" alarms limits
  • Early detection of undesired behavior
  • Availability of predicted values for non-observed variables
  • Ability to predict effect of control actions ("what-if")

Several other MBR monitoring systems have been reported in the literature, often (as for MIMIC) as part of an overall monitoring-diagnosis application. Some examples are:

  • BIOTECH (Bousson et al., 1993): Provides monitoring of key variables and diagnosis in a biological fermentation process.
  • DIAPASON (Penalva et al., 1993): A system for supervision and operator support in a continuous process in the area of nuclear fuel reprocessing.

4.4. CONTROL

Control means governing the behavior of the target system to meet stated goals. In the AI/expert system area in general, and in MBR-based systems in particular, there are not many real-world applications of closed-loop control systems, i.e., systems that directly control the target system without human intervention. On the other hand, there are numerous examples of open-loop systems, where the human is an essential link between observations and control actions. In manufacturing and process industries, one often refers to such systems as operator support systems. Their role is to provide the operator with insight into the status of the target system, and to suggest corrective or optimizing control action that the operator may implement. Such systems generally exist as an add-on to an underlying closed-loop control system that carries out low-level control in real time.

More systematically, we may distinguish between the following different ways in which expert system technology (and MBR) are being applied to control problems:

  • Expert control. This term is often used to denote systems where an expert system is used to configure and tune conventional low-level controllers. MBR is not widely used.
  • Qualitative/fuzzy control. In this case, closed-loop control is performed by AI/expert system techniques. Fuzzy logic has been widely deployed in this capacity. It can be argued that fuzzy logic is an MBR technique, but this is a matter of classification.
  • Qualitative model predictive control. In model predictive control (MPC), a model is used online to predict the consequences of various control actions. Combined with an optimizing search engine, such a model may support very sophisticated control regimes. MBR is an ideal candidate for the model part of MPC, and may be combined with numerical techniques.

In this list, the last type most directly applies MBR in practice. As explained above, most existing systems of this type are open-loop operator support systems.


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.