"The Design of Fault Tolerant Systems"

Fault diagnosis has been the subject of intensive research for more than thirty years, and an extensive corpus of results is now available, starting with the logical theory of diagnosis from the Artificial Intelligence community, the analytical redundancy, diagnosis observers and identification approaches from the Control community, and the decision making approaches from the field of Statistics, to mention only a few. The design of fault tolerant systems, is a more recent field, whose foundations are not yet completely stabilized, but whose understanding is getting deeper and deeper.

This talk will develop some recently developed approaches to the fault tolerance problem. It is divided into 6 parts:

1. Problem setting presents the fault tolerance problem in the frame of the more general fault handling problem,
2. System reconfiguration presents the fault tolerance strategy by which the faulty components are discarded and only the healthy components are used,
3. The functional view of fault tolerance is based on the notion of services provided by a set of components, and studies the conditions under which those services are still available under different faults,
4. Model matching is a popular approach where the behaviour of the faulty system is wished to be as close as possible to the behaviour of the healthy one. This part discusses different approaches aiming at this goal,
5. LQ control is a classical optimal control problem, whose solutions must be adapted to fault tolerance considerations, both from the conceptual and from the real-time implementation points of view,
6. Conclusion ends the talk with some considerations about future research.