Retour au site du LAAS-CNRS

Abstract

Every model-based diagnostic approach relies on a representation of a real-world system, in this paper called believed system. The believed system is used along with the observations about the real-world system to generate a diagnostic problem to be solved. In this paper it is firstly argued that believed systems can differ from real-world systems in many different manners. As so, properties of believed systems, diagnostic problems and diagnostic results are introduced. Then, a series of relations between these properties are proved. The importance of such relations, sometimes seen as intuitive, is that they are necessary to formally prove the accordance between the real-world system and the believed system; to formally prove that a believed system and a diagnostic problem will produce high-quality diagnostic results; or even to ease diagnostic algorithms, since for systems and problems with certain properties, different model-based diagnostic approaches produce the same diagnostic results. In order to introduce the referred properties and reasoning about them a framework of diagnosis based on the difference between the believed and the real systems is proposed.

Mots-Clés / Keywords

Mots-Clés / Keywords

Abstract

This paper addresses the problem of the maintenance of a complex system of heterogeneous components. With the ecreasing costs of sensors, it now becomes possible to really benefit of on-line adaptive prognostic methods in order to support the optimization of the maintenance actions and scheduling. In this paper, we propose a generic health monitoring architecture that encompasses the available prognostic methods and provides a common support for the maintenance decision on a complex system. The output of this architecture takes into account the composed nature of the system by not only providing component prognoses but also function prognosis. In order to provide a prognosis for the whole system, component prognoses are combined by taking into account functional dependencies in the system.

Mots-Clés / Keywords

Résumé

En général, les méthodes de diagnostic utilisent un seul type de connaissances, par exemple un modèle analytique ou qualitatif, une base d'observations numériques, une base de cas préalablement résolus, etc. Si ces méthodes produisent de bons résultats sur des systèmes assez simples, elles deviennent très difficiles à implémenter sur des systèmes complexes comme des véhicules. En effet, il devient alors impossible ou trop coûteux d'acquérir la connaissance nécessaire à leur mise en oeuvre. L'approche suivie par le laboratoire commun Autodiag est d'utiliser différentes formes de connaissances hétérogènes dans une méthode de diagnostic intégrée. Pour cela, nous proposons d'utiliser différentes méthodes de diagnostic, comme le diagnostic à base de modèles, le diagnostic par reconnaissance de formes ou la recherche sémantique d'informations, dans un cadre commun appelé MODE. En définissant un langage commun, MODE permet à toutes ces méthodes de collaborer et de séquencer l'application de chacune en fonction des données disponibles et des contraintes de l'utilisateur. Cette approche est illustrée sur des données réelles issues du domaine de l'automobile.