Supporting fault diagnosis through a multi-agent-architecture
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,...
Cited by (13)
A measurement approach for overcoming unbalanced overwork in multi-agent systems
2013, International Journal on Artificial Intelligence ToolsManaging the information systems in the industrial domain
2016, Cogent Business and ManagementResearch on intelligent monitoring and fault diagnosis system based on Multi-Agent model
2013, Applied Mechanics and MaterialsFault diagnosis and optimization for agent based on the D-S evidence theory
2012, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)Diagnosis system of manufacturing equipment based on structure-fault tree
2011, Advanced Materials Research