Yannick Pencolé Diagnostic décentralisé de systèmes à événements discrets : application aux réseaux de télécommunications , jun , 2002 , Document Abstract This thesis deals with the monitoring and the diagnosis of complex dynamic systems such as telecommunication networks. These systems are composed of a set of interconnected equipments. With the help of sensors, a supervisor is able to receive alarms from the system. The objective of the supervisor is to deal with these alarms in order to guess the possible states and the possible failures of the system. The purpose of this work is to provide a help for interpreting alarms and giving the possible failures that could have occurred on the system. The developped approach uses model-based diagnosis techniques. Given a behavioural model of the set of supervised components, this approach consists in efficently using this model in order to analyse the alarm stream on-line. We propose to represent diagnoses as compacted transition systems based on failure events (reduced transducers). Because of the complexity and the distributed nature of supervised systems, our approach is based on the ``divide and conquer'' paradigm. Firstly, we compute a set of local diagnoses based on local behaviours. Secondly, the computation of the global diagnosis is based on a local diagnoses merging operation. A strategy for applying this merging operation was developped in order to be efficient. This work was done in the context of a RNRT scientific project called MAGDA. Finally, a platform for the decentralised diagnosis of dynamic systems was developped. This platform validates our approach on telecommunication network examples: Transpac network and a SDH network. This web site is maintained by René Quiniou using the Weave system from the Caravel project: http://www-caravel.inria.fr. Last modification: 10-07-2011 09:54:25 Version française