Surveillance de systèmes continus à l'aide de méthodes n'utilisant pas de modèles formels : Application aux systèmes médicaux
Sylvie Charbonnier · HAL (Le Centre pour la Communication Scientifique Directe) · 2006
In this document, our studies on monitoring of dynamic systems without using formal models are presented. The applications concern medical systems. Information available is a set of measurements recorded on the system by sensors and some expert knowledge, which cannot be modelled by differential equations. Relevant features are extracted from the signals and merge using decision models that copy the expert reasoning. Two paths were explored : - trend analysis which enables the relevant transients of a signal to be enhanced, so as to use them in a decision making process - pattern recognition where the static links between variables are exploited. Knowledge about diagnosis processes can be extracted using data mining techniques when a large data base is available. Applications are about patient monitoring in Intensive Care Units and physiological states diagnosis (sleep/wake stages, hemorrhagic choc).