![]() |
|||
![]()
|
![]() |
![]() |
![]() |
4.1. NEURAL NETWORKSNeural networks are being used in the field of complex machine diagnostics as they have the ability to learn a machine's operating characteristics. Bruel and Kjaer (1996) believe that neural technology must be used as it has the ability to "learn by input of data about its surrounding environment". One European EUREKA project, NEURAL-MAINE (EUREKA project 1250), (Harris et al., 1995) aims to advance the technology available for complex machine diagnostics by the use of multiple sensor technology, data fusion and neural networks.
The NEURAL-MAINE system is currently under development by the project consortium. It will compromise a number of neural network based local fusion systems and an overseer system. The local fusion systems will represent one part of a complex machine, for example a bearing. Each part of a machine will have its own local fusion system. Thus a turbine with twelve bearings will have twelve local fusion systems. The local fusion system will be responsible for taking all of the sensor data (of varying types) and fusing it into one model, which represents the piece of machinery for that specific time slice, as shown in Figure 5. The fused data model will be then passed to the neural network which will evaluate it as being normal or unique (novelty). If the data are different then the local fusion system will flag that a novelty situation has occurred, and allow a more complex diagnosis to continue at the local level. The overseer system will sit above the local fusion systems, and take as input the output from the local fusion systems and also other machinery operating parameters, as shown in Figure 6. The NEURAL-MAINE consortium identify the following objectives for the overseer system:
To accomplish this, the overseer will work in two modes - firstly as a novelty detection system and secondly as a diagnostic aid. 4.2. DISTRIBUTED ARTIFICIAL INTELLIGENCEAnother new field that is starting to take shape is that of Distributed Artificial Intelligence systems (DAI). DAI uses the approach that a large problem can be separated up into specific areas where "agents" can perform specific intelligence-related tasks. The agents have the ability to control their own problem solving and communicate with each other. Interactions between agents usually involve cooperation and communication with each other with the aim of "enhancing their individual problem solving and to better solve the overall application problem" (Jennings, 1994). ARCHON (ARchitecture for Cooperative Heterogeneous ON-line systems) was Europe's largest project in the area of Distributed Artificial Intelligence (DAI) (Huhns, 1994). The project resulted in the development of a general-purpose architecture, software framework and methodology which has been used to support the development of a number of DAI systems in a number of real world industrial environments including the power generation industry, control of a cement kiln, control of a particle accelerator and control of robotic equipment.
ARCHON provides a decentralised software framework for creating Distributed AI (DAI) systems for industrial applications and provides a methodology which can offer guidance on how to decompose an application to best fit with the ARCHON approach. One application that was created using the ARCHON architecture is CIDIM (Huhns, 1994). CIDIM provides an intelligent aid to Control Engineers (CE) whose task is to ensure that there is a constant electric supply to customers (Varga et al., 1994). The control engineer's main tasks are to carry out maintenance work safely with the cooperation of field engineers, to identify faults on the power network and to carry out restoration to provide continuous electricity supply. The control engineer uses an electric network control system which allows remote operation of circuit breakers across the network and reports any automatic switch operation that occurs in response to a fault. The system also displays alarms and provides the engineer with various other pieces of information such as load readings. The system covers the High-Voltage Network (HVN) and has partial coverage of the Low-Voltage Network (LVN). The rest of the switching operations for the LVN are done manually by the field engineer who is in contact with the control engineer. CIDIM provides fault diagnosis, lightning detection, user-driven restoration planning and automatic re-checking of restoration plans. The main advantage of CIDIM is that it automatically collates the information which previously the control engineer had to manually acquire and integrate from several separate stand-alone systems. The CIDIM systems consists of seven agents, with the ARCHON architecture helping to share information between agents which results in a greater consistency across the application (see Figure 7).
|
![]() |
|
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. |