Distribution flow networks are modeled by linear or nonlinear systems of balance laws. Monitoring of these networks (Faults detection and isolation) requires knowledge of certain state variables. However, in most cases it is not possible to measure all the state variables and observers based on partial differential equations modelling the network dynamic can be used. In this dissertation, the observability of the hyperbolic systems is studied first and then classical and robust PDE observers with injection of the state estimation error at boundaries or in the system dynamics are proposed. These observers provide on-line estimation of signals that are not measured. The estimation is used for the diagnosis of distribution flow systems. The performances of the observers and the diagnosis approach are validated on real flow data collected from the water distribution system (WDS) of Polytech'Lille (Cité scientifique, University of Lille 1 Sciences and Technologies), within the framework of the SUNRISE SMART CITY Project. Data are taken from the WDS in the absence and in the presence of leaks are used.
Directeur de thèse : Vincent Cocquempot, Professeur, Université Lille Sciences et Technologies. Rapporteurs : Gildas Besançon, Professeur, Université Grenoble Alpes. Valérie Dos Santos Martins, Maître de Conférences HDR, Université Lyon 1. Examinateurs : Abdel Aitouche, Professeur (co-encadrant), HEI Didier Maquin, Professeur, Université de Lorraine Fatiha Nejjari, Professeure, UPC Terrassa. Isam Shahrour, Professeur, Université Lille Sciences et Technologies.