Authors:
Khaoula Tidriri
;
Nizar Chatti
;
Sylvain Verron
and
Teodor Tiplica
Affiliation:
2 Avenue Notre Dame Du Lac, 49000 Angers and France
Keyword(s):
Process Monitoring, Model-based Fault Diagnosis, Analytical Redundancy, Complex Systems.
Related
Ontology
Subjects/Areas/Topics:
Engineering Applications
;
Industrial Engineering
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Robotics and Automation
;
Signal Processing, Sensors, Systems Modeling and Control
;
Surveillance
;
Systems Modeling and Simulation
Abstract:
Fault Detection and Diagnosis (FDD) approaches have become increasingly important due to the growing demand for reliability and safety for modern systems. During the last decades, many works were reported about FDD approaches, especially model-based ones. The latter relies solely on a developed model that accurately describes the system, without exploiting any additional available data. In this work, we intent to make use of the physical model as well as historical data, for both normal operating state and faulty states. Hence, the paper focuses on the validation of an experimental approach, called Fault Training Analysis, that analyzes and identifies the causal relations between residuals and faults identified and observed on the system, by dealing with real measurement data from nominal and faulty states. It results on an experimental matrix, called Fault Training Matrix, that enhances the theoretical Fault Signature Matrix. The effectiveness of the proposed approach is validated t
hrough the challenging Tennessee Eastman Process. The application results on a high fault detection rate, a high fault diagnosis rate and a small false alarm rate.
(More)