Conception d'un Système Expert pour l'étude de la Sécurité Incendie de Bâtiments dans un Contexte Pluridisciplinaire

In this thesis we study the improvements that artificial intelligence techniques can bear in the field of fire safety management of buildings. Fire safety is a component of the technical function of a building and therefore can be examined at the design stages. Nowadays it is mainly taken into account by application of regulation rules. We suggest a method that differs in two ways: first it concentrates on the search of risk factors and the best measures for their elimination. In the other part, it is a global method which integrates knowledge from different sources and which considers all the functions of the building (Architecture, Economy, Technique). The expert systems technique seemed to offer the best frame to experiment the new approach. The works have concerned mainly: – The elaboration of a global method for fire safety of buildings by formalising the current knowledge in the field. – The elaboration of a methodology for developing an operational and pedagogic multi-expert system. Two kinds of results are presented: – A set of tools like "TRIAGE", "ARBRE", "NASS", "ICONOGRAPH." o TRIAGE is the expert system dedicated to multi-disciplinary studies of fire safety. Two modes are provided: the focalised one in which the user consults a specialist on an isolated subject, and the global mode where the user is guided by a tasks scheduler to reach predefined goals. o ARBRE is an expert-oriented tool for the setting of causal trees with subjective probabilities. o NASS is an environment for developing expert systems capable of managing many knowledge sources in natural language. o ICONOGRAPH is a graphic interface for fast description of objects and relations in a structured universe. It uses an auto-organizational model (said with mutation rules). – The set of methodological elements used to produce these tools. We propose a technique that we call semography. It is supposed to guide the designer of an expert system on the kind of knowledge to search for and on how to proceed. The semography technique is based on system theory and cognitive ergonomy, and proposes models of experts' discourse universe and techniques to stimulate and analyse the discourse itself. The techniques and tools proposed can be translated to other fields than fire safety.