Name	Self-adaptive monitoring systems
Theme	Diagnosis of large scale discrete event systems
Begin	2002
State	Ongoing
Description	Monitoring systems, like Calicot, are often faced with signals of varying quality. Taking advantage of context knowledge may generally facilitate signal processing and improve the results which are next exploited by diagnosis. We have proposed a system architecture to adapt such monitoring systems to the current context [73]. This means adapting signal processing algorithms to, e.g. the noise type and level of the raw signal or the shape of interesting waves, as well as to current failure or disease hypotheses inferred by diagnosis. Furthermore, diagnosis can also be adapted to the knowledge level allowed by signal processing or current possible diagnosis hypotheses. Our adaptive system has a pilot module that operates at three stages. It chooses and tunes the signal processing algorithms that analyze the input signal by using decision rules that are learned [17] automatically and then used to choose the most relevant algorithm according to the current context. It activates or deactivates processing tasks devoted to the extraction of specific events from the input signal. It adapts the diagnosis to the current resolution of the signal analysis. In our approach, diagnosis is achieved by chronicle recognition. Chronicles are organized in a hierarchy of chronicle bases: more abstracted chronicles use less objects (event descriptions) and fewer attributes; more specific chronicles are more precise and more detailed. The pilot chooses the current chronicle base according to the level of abstraction imposed by the recognition task or the representation of events. For example, a more abstracted chronicle base needs less low-level computation than a more refined one but the abstraction level may be to high to discriminate between diseases. Using this kind of chronicle hierarchy leads to a smarter use of computational resources.