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


3.8. NEURAL AND FUZZY RULE-BASED EXPERT SYSTEMS FOR AUTOMATED WELDING

Neural networks are also being used to acoustically determine if a weld is good or bad for Navy ships. The welding arc sound emits a unique frequency spectrum according to the welding conditions such as electrode extension, and unstable arc largely changes the pattern of the spectrum, providing the useful information about the phenomena of the molten pool and the transfer of droplets. Similar pattern recognition techniques using neural networks are adopted in grinding such that the change of grinding state can be detected before the occurrence of the grinding burn or the chatter.

A fuzzy rule-based expert system was developed for modification of welding conditions by applying fuzzy rules (Fukuda and Kamio, 1990). IF parts define the fuzziness of unacceptable defects of the weld that is produced under the previously determined conditions, and THEN parts define the degree of modification of an input variable. The expert system is not a fully fuzzy system, because knowledge is not enough to develop the membership functions in the THEN parts. However, it provides a fundamental basis for constructing a truly adaptive and automated control loop using appropriate sensors for robotic welding.

3.9. EXPERT SYSTEM DEVELOPMENT TOOLS

Because of a shell's ease of use and availability for commercial personal computers, it is quickly becoming the realistic means of creating an expert system. When creating an expert system with a shell, only the knowledge base needs to be created. However, there is a limit to what existing shells can do in manufacturing and engineering applications; expert systems made with a shell might run slower and have limited I/O capabilities.

The knowledge base and inference engine can be created with code in AI programming languages, such as Prolog and Lisp, according to the user's needs. However, it takes much time and needs extensive programming skills, especially for manufacturing engineers. Furthermore, there are some problems for utilizing these AI programming languages in manufacturing and engineering applications, for example, welding:

  1. In the determination of optimum welding conditions, numerical computations are largely required, such as FEM simulation of welding heat conduction and parameter estimation of empirical formulas using statistical analysis.
  2. Data and control flows for welding condition setting and real-time control imply some concurrency, as shown in Figure 5.

By using a programming tool for knowledge representation, such as OPS83, the user can also construct his own inference mechanism and link it with programs written in C. Thus, welding engineers' experience could be also reflected in meta-knowledge rather than the knowledge itself.

For conventional process control applications, the expert system is regarded as a separate knowledge module and combined with the control part in a distributed control system. To fully bring out latent potentialities of the expert system, hierarchical integration between the expert system part and the control part is a possible solution, since many real systems are represented as an inherently hierarchical structure. An object-oriented programming methodology integrates heuristic knowledge and model-based knowledge for complex, ill-structured problems into the class hierarchy formality.

The frame language integrates static attributes and dynamic operations as a whole. Each frame is viewed as a basic knowledge block and can be used in the different parts of the knowledge base. The static knowledge of objects is expressed by "attribute" slots of the frame. Numerical calculating procedures and heuristic logic are expressed by "method" slots and "rule" slots, respectively. The expert system part and the process control part are tightly coupled at all levels of class hierarchy. Some object-oriented expert system development shells on workstations or personal computers are available, for example, Ext Kernel with ESP (Extended Self-contained Prolog) on MELCOM PSI workstation in Japan.

3.10. KNOWLEDGE REPRESENTATION USING HIGH-LEVEL PETRI NETS

To realize expert control systems that require real-time processing, it is necessary to model the necessary knowledge clearly and simply, and implement it on a parallel and distributed processing architecture. High-level colored Petri nets have been used as a knowledge representation tool for image understanding, where output tokens represent extracted features at various decision levels, and directly applied to vision-based real-time mobile robot navigation. The similar approaches were applied to the determination and adjustment of welding conditions (Yasuda and Tachibana, 1996). Use of Petri nets insures some important properties for expert systems, such as boundedness or safeness, liveness or freedom from deadlock, and reversibility.

4. RESEARCH ISSUES

4.1. HIERARCHICAL INTEGRATION OF INTELLIGENT STRATEGIES

An intelligent manufacturing system can be seen as the basic multilayered structure of intelligent machines, in which the layers are composed according to the principle of increasing precision with decreasing intelligence. This principle can be generalized for the development of expert systems with a low-level intelligent controller that consists of several parts, each with its own type of intelligence.

At the lower levels of automation, normal signal processing plays an important role, as used in neural networks and conventional mathematical methods. The knowledge available may be structured or not, and symbolic or numeric. Symbolic knowledge represents global features, while numeric knowledge represents specific or individual features. At the lower levels, before the pieces of symbolic knowledge are applied, they should resort to conventional numerical procedures. At the lowest level, user-written routines and simple logic are very well suited.

At the higher levels of automation, explicitly described knowledge with advanced reasoning strategies plays an important role. The corresponding knowledge representation can be found in high-level object-oriented descriptions, such as in sophisticated expert system shells. An integration of various AI-based strategies in an intelligent manufacturing scheme requires an integration of various methods of information processing, involving both typical symbolic processing and connectionist methods such as neural networks.


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.