Skip to Main content Skip to Navigation
Theses

Surveillance de procédés à base de méthodes de classification : conception d'un outil d'aide pour la détection et le diagnostic des défaillances

Abstract : The present work belongs to the field of decision support systems for complex process monitoring, such as chemical and petrochemical plants. Since it is not always possible to obtain a mathematical model for these processes, it is necessary to consider other approaches such as learning and classification methods, in order to identify their different operating modes (normal or faulty). We propose a strategy based on Data Mining techniques, which allows the construction of a discrete event model of the process behavior using historical and online data. This strategy consists on an offline learning stage for the elaboration of a first reference model. This model, in the form of a finite state automaton, must be validated and completed by the process expert. A second online stage consists in tracking the identified process states. A deviation is detected when a given number of observations are not recognized into any expected functional state. At this stage a new learning procedure is proposed in order to identify the nature of the deviation. The new resulting classes and information concerning the descriptors involved are presented to the expert as support for his diagnosis. A decision support software tool for monitoring processes using LAMDA classification algorithm has been developed based on the proposed strategy. LAMDA method uses Fuzzy logic theory and introduces the adequacy concept for the assignment of an element to a class. Within the context of CHEM European project the principal aspects of our work were tested on different industrial and pilot plants.
Document type :
Theses
Complete list of metadata

Cited literature [3 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00010247
Contributor : Emilie Marchand <>
Submitted on : Thursday, September 22, 2005 - 2:08:58 PM
Last modification on : Monday, October 19, 2020 - 11:09:49 AM
Long-term archiving on: : Friday, April 2, 2010 - 10:41:30 PM

Identifiers

  • HAL Id : tel-00010247, version 1

Citation

Tatiana Kempowsky-Hamon. Surveillance de procédés à base de méthodes de classification : conception d'un outil d'aide pour la détection et le diagnostic des défaillances. Automatique / Robotique. INSA de Toulouse, 2004. Français. ⟨tel-00010247⟩

Share

Metrics

Record views

1561

Files downloads

5635