Diagnostic à base de modèles non linéaires. : Application au circuit carburant d'une turbomachine

Abstract : The current gas turbine regulation systems are based on complex architectures that manufacturers tend to make more modular with more cost effective technologies while ensuring a greater or equal level of reliability. In this context, the fuel system health monitoring, which aims to identify critical hydraulic components dysfunction, allows to reduce maintenance costs, to improve maintainability level and to ensure gas turbine availability. The present study focuses on the development of performant and robust diagnosis methods for the detection and isolation of faults affecting primary fuel system hydraulic functions. Existing nonlinear model based residual generation methods are presented and applied to the fuel system. The analytical approach for decoupling, combined with extended Kalman filters, helps fault isolation by generating residual structures. A new approach based on differential flatness theory is proposed for nonlinear systems fault diagnosis with an application to the fuel system. Sliding mode differentiators are used to estimate derived signals that are necessary for the application of some residual generation methods. Numerical simulations illustrate the efficiency of obtained results. An experimental application is presented using a real data set from a partial test bench provided by Turbomeca company of the SAFRAN group.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-01249601
Contributor : Abes Star <>
Submitted on : Monday, January 4, 2016 - 2:55:39 PM
Last modification on : Friday, May 17, 2019 - 11:38:57 AM

File

SIFI_MOHCINE_2015.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01249601, version 1

Citation

Mohcine Sifi. Diagnostic à base de modèles non linéaires. : Application au circuit carburant d'une turbomachine. Automatique / Robotique. Université de Bordeaux, 2015. Français. ⟨NNT : 2015BORD0054⟩. ⟨tel-01249601⟩

Share

Metrics

Record views

597

Files downloads

1394