Supporting fault diagnosis through a multi-agent-architecture

https://doi.org/10.1016/S0378-4754(02)00016-2Get rights and content

Abstract

In the industrial environment a short down time of CNC machine-tools is increasingly important. One approach to counter the loss of production is to make the machines more reliable, thereby reducing their times of non-use. Another is to enable machine operators to solve the problem themselves, thereby avoiding the delay until maintenance staff is sent from different locations. Operators may be supported in the fault-finding process by a knowledge-based decision–support system. The questions are how to avoid the negative consequences of using such systems and how to even increase competence on the job is addressed by competence promoting multi-agent system for support (ComPASS), a system based on a multi-agent architecture, and an instrument to measure competence enhancement.

A running, functioning prototype was developed and evaluated. Based on this experience the system is currently being extended and improved, first results of which are presented here.

Introduction

In the industrial environment a short down time of CNC machine-tools is becoming more and more important. Rather than making the machines more reliable and thereby reducing their down-times, the approach taken here is to enable machine operators themselves to repair the malfunction in order to avoid the delay until qualified maintenance personnel arrives to solve the problem. Operators may be supported in the fault-finding process by a knowledge-based decision–support system. But, diagnostic tools bear the danger of people relying too much on them and their capabilities, so that after a while the formerly competent personnel loses their competence [1]. This can even occur with the use of an expert system, if it merely gives instructions to remove a fault rather than explanations on how the error was located or other possible causes for the errors, and how avoid a repetition of the same fault [2]. Another problem is the plateau-effect, which describes abrupt collapse of expert systems at the border of their domains [3].

The development of competence promoting multi-agent system for support (ComPASS), funded by DFG, German Research Foundation is aimed at overcoming the problems mentioned above in the application area of fault diagnosis of CNC machine-tools. A multi-agent structure was chosen, because it makes is possible to distribute the knowledge concerning the problem solving process over several agents [4]. Since, people memorise complicated facts better, if they are represented in a multi-perspective view [5], [6], each agent of the system supplies the user with a different view on the problem.

Section snippets

Application domain and system architecture

ComPASS was developed to assist operators in the process of locating and remedying malfunctions occurring at a CNC machine-tool. The support for the user during the diagnosis consists of the presentation of different views on the diagnostic problem and their likely solution. ComPASS consists of two deliberative, adaptive, loosely coupled agents, each furnished with its own inference strategies (Fig. 1). It lies in the user’s determination, to utilise just one agent or both. The two agents were

Diagnostic strategies

Not all the 15 diagnostic strategies defined in the literature (Fig. 2), were implemented in the running version of ComPASS, but may be added step by step. The strategies implemented so far were chosen according to the following criteria:

  • ease to understand;

  • ease to apply;

  • coverage of the whole machine;

  • distribution over the agents.

The strategies were distributed over the two agents considering the whole machine so as to give the most information about error dependencies but also to have

ComPASS support for the user

The user interface of ComPASS consists of forms containing the following:

  • radio-buttons;

  • selections in pull-down menus.

Error messages can be input by the operator or are received by the CNC module of ComPASS and analysed by the agents. There are two different starting points for the analysis of possible causes for a malfunction. The fault is described by either naming the component and having the system show possible malfunctions for it or by choosing the malfunction and letting the system add

Support for the developer

The developer of an extension or modification of ComPASS is the person responsible for the data input concerning a specific machine. For ComPASS to work properly, all data regarding a machine and the inference processed have to follow a specific format. So, to support the developer, a tool is provided that allows for the comfortable recording of the required data. Files concerning information on other machines may be appended. The forms present the developer with fields to input all the

Extended system

The prototype of ComPASS was evaluated as mentioned above, any hints for improvements were taken into consideration for an upgraded version of the system. Ultimate goal of the development is to implement the system in SMEs, which are using CNC machine-tools. For this to be feasible, several modifications are made.

Perspective

ComPASS is being enhanced to include the described features in the near future. The suitability of the approach will be investigated in two environments, a SME in the Berlin area and—considering the multi-lingual aspect of the system—in Chile. The knowledge elicitation is continuing in these two environments leading to continued improvement of the system.

The developed knowledge test for measuring competence is an appropriate tool. Long-term studies as to the competence enhancement still have to

Acknowledgements

The work described in this paper is funded by DFG under Ti 188/6-3.

References (11)

  • R. Marzi, R. John, P. John, Design and evaluation of a competence promoting decision support system, in: Proceedings of...
  • W. Coy, L. Bonsiepen, Erfahrung und Berechnung. Kritik der Expertensystemtechnik, Springer, Berlin,...
  • H. Busch, T. Hermann, K. Just, M. Rittenbruch, Systeme für Experten statt Expertensysteme, Infix, Sankt Augustin,...
  • M. von Bechtolsheim, Agentensysteme. Verteiltes Problemlösen mit Expertensystemen, Vieweg, Braunschweig,...
  • A. Paivio, Imagery and Verbal Processes, Wiley, New York,...
There are more references available in the full text version of this article.

Cited by (13)

View all citing articles on Scopus
View full text