multi agent diagnosis and repair
STW project DIT 5780

There is an increasing interest in developing automated systems that are able to operate in a dynamic and error-prone environment. For such autonomous systems, powerful diagnostic tools have to be used to detect possible faulty behaviour of the system without human intervention. Currently, the dominant and most successful approach to diagnosis for stand-alone systems is the model-based diagnosis approach.
Automated systems often operate in a distributive environment where other, related, systems are operating, sharing (and sometimes competing for) resources, plans and goals. Multi-agent systems technology tries to design and implement such autonomous systems (agents) operating in a common environment.
It turns out that the dominant model-based diagnosis approach is not simply applicable in a distributed multi-agent context.

research aims
The current project aims to adapt, extend and apply diagnosis methods and tools to such a distributed multi-agent system. More in particular, it aims
  • to extend the classical model-based diagnosis approach for single systems to distributed multi-agent based systems. The result will be a detailed description of a multi-agent architecture for diagnosis;
  • to develop distributed, multi-agent based diagnostic techniques incorporating existing single-agent based techniques. As a result, a description and prototypical implementation of a multi-agent based diagnosis system will be delivered;
  • to show the feasibility of the approach by applying it to a distributed air traffic control problem. The result will be a demonstrator for diagnosis and repair in a distributed agent-based system for arrival-departure scheduling.

project organization
To extend the classical model-based diagnosis approach, the research will be carried out in three PhD-research tracks (diagnosis and communication/negotiation, diagnosis and conflict detection and diagnosis and coordinated repair). All three tracks will be integrated in one case application: a distributed system for arrival and departure scheduling.
This collaboration makes it possible to develop a multi-agent system for diagnosis and repair in which the three aspects are fully integrated. Considering the complexity of the application domain, this surely would be beyond the reach of an single Ph.D. student. Moreover, for this larger project, the utilization partner (NLR) is able to provide a domain expert who will aid in the acquisition of domain knowledge and in the realization of a prototype.

The case application will be developed in close cooperation with the National Aerospace Laboratory (NLR). Currently, NLR is involved in several research projects aiming at the operational integration of existing tools for arrival and departure planning management, together with those derived from the planning and routing function of the ground movement concept.

This airport planning problem is an ideal test case for the proposed multi-agent diagnosis research as it involves multiple individual parties (such as ATC, ground control, aircraft, air transportation companies) that have cooperate in order to satisfy their, possibly conflicting, goals. Furthermore, the tools developed in the project can be used to generate the models of the (distributed) subsystems that have to be diagnosed. Such diagnoses have to be performed frequently and efficiently as the system has to be robust for abnormalities (e.g., delays) that occur quite frequently in the domain of air transport.
Finally, the proposed research application is a promising one: successful solutions to the arrival and departure planning problem may have huge financial impacts, as air transport is likely to increase in the future.

NLR will apply the results of the research in a demonstrator for diagnosis and repair in a distributed agent-based system for arrival and departure scheduling.
NLR has also the intention to apply the research results in future diagnostics projects at NLR. Possibilities are the JSF (Joint Strike Fighter) project and other projects in aircraft health management and prognostics.

research partners
Delft University of Technology    
Maastricht University
    Utrecht University
Parallel and Distributed
Systems Group
Institute for Knowledge
and Agent Technology
Data and Knowledge
Intelligent Systems