Skip to Main content Skip to Navigation

Méthodologie pour le placement des capteurs à base de méthodes de classification en vue du diagnostic

Abstract : The presented study is in the field of the decision-making aid for the monitoring and the diagnosis of complex systems such as chemical processes. Our work permitted to design a methodology allowing placing the most relevant sensors on a process for its diagnosis from the analysis of historical data. This methodology is based on the association of methods used for measurement of information quantity (Shannon's entropy) delivered by signals coming from a system and for the classification of data. From time evolution data of sensors (constituting the set of all possible sensors) following scenarios of failure (for example simulated on a dynamic simulator), it is possible to identify the most relevant sensors in the set of the possible sensors and to obtain a model of the process with a level of abstraction such as it is usable for the diagnosis. A procedure for the model adaptation in the case of recognition of unknown failures has also been proposed Tests of feasibility on concrete industrial cases have been carried out in simulation by using first of all a dynamic process simulator universally distributed in industry: HYSYS then on a new chemical reactor developed by the Laboratoire de Génie Chimique of Toulouse. These works are part of a project supported by the "Institut pour une Culture de Sécurité Industrielle" (ICSI).
Document type :
Complete list of metadatas
Contributor : Emilie Marchand <>
Submitted on : Tuesday, November 8, 2005 - 1:50:02 PM
Last modification on : Monday, October 19, 2020 - 11:11:24 AM


  • HAL Id : tel-00010906, version 1


Antonio Orantes Molina. Méthodologie pour le placement des capteurs à base de méthodes de classification en vue du diagnostic. Automatique / Robotique. INSA de Toulouse, 2005. Français. ⟨tel-00010906⟩



Record views


Files downloads