Modeling and Monitoring of Hybrid Dynamic Systems with Feed-Forward Neural Networks: application to two tanks hydraulic system

Nadhir Messai (1) , Philippe Thomas (2) , Dimitri Lefebvre (3) , Bernard Riera (1) , Abdellah Elmoudni (4)

1 CRESTIC - Centre de Recherche en Sciences et Technologies de l'Information et de la Communication - EA 3804
2 CRAN - Centre de Recherche en Automatique de Nancy
3 GREAH - Groupe de Recherche en Electrotechnique et Automatique du Havre
4 SET - Laboratoire Systèmes et Transports

Within the model based diagnosis community, Fault Detection and Isolation (FDI) techniques for hybrid systems require the ability to discern between all the modes and to identify at each time the current mode. Unfortunately, these necessary conditions are very restrictive because, on one hand, only few partial results have been reported on the notion of discernability between the modes and, on the other hand, all the switching sequences must be systematically investigated. To overcome these drawbacks, this paper proposes to use Feed-forward Neural Networks in order to build average models of Hybrid System. This alternative can be particularly interesting either when we can not describe all the system's modes by Ordinary Differential Equations (ODEs) or when we can not investigate all the switching sequences. Once the Neural Networks models are obtained they are used to generate residuals and to achieve FDI without any need to discern or to estimate the current mode.

Mots clés: Fault detection; Neural Network; Identification; Hybrid Dynamic Systems

Domaines: Automatique / Robotique

Nadhir Messai, Philippe Thomas, Dimitri Lefebvre, Bernard Riera, Abdellah Elmoudni. Modeling and Monitoring of Hybrid Dynamic Systems with Feed-Forward Neural Networks: application to two tanks hydraulic system. Workshop on Advanced Control and Diagnosis, Nov 2005, Mulhouse, France. pp.103-109. ⟨hal-00103137⟩