Abstract FR:

System diagnosis is a very challenging task due to the variety of symptoms, probable causes, impact analysis and the selection of the most suitable solution. Most system diagnosis activities today are human-based, for this reason, error-prone. The automation of the diagnosis decisions faces several challenges, such as: high complexity, lack of automatic knowledge transfer, and disconnection between context-oriented positive decisions taken in dissimilar domains. To address some of these issues, the contribution of this thesis is tow-fold. Firstly, we propose an adaptive framework for diagnosis validation and transfer of information from successful cases for future use in similar situations. We show that this mechanism allows a post-validation of successful diagnosis actions optimizing the diagnosis process and increasing its accuracy. Secondly, we introduce an event ontology and concepts related to semantic tag clouds; we show how to manage the related activities to build an ontology-driven diagnosis. We formalize these concepts in order to derive diagnosis actions and validate the successful ones. We also propose augmented event and augmented action models to validate the diagnosis. Lastly, we embed a timestamps approach and we consider a series of temporal operators defining the events relative temporal positions; this approach allows a more fine grain interpretation of the system behavior. A combination of the proposed mechanisms is used to complete the main functions of the diagnosis engine. We highlight the complexity of the approach via a case study and we illustrate partial solutions for different concepts introduced.