Bayesian networks model of reliability applications to Reconfiguration and Diagnosis - Archive ouverte HAL Accéder directement au contenu
Autre Publication Scientifique Année : 2011

Bayesian networks model of reliability applications to Reconfiguration and Diagnosis

Résumé

The talk presents a new approach based on reliability to reconfigure redundant actuators when fault occurs and to diagnose the fault. The model of reliability is build with Bayesian Networks. The first application proposes a solution to control an over-actuated system that is structured as a typical consecutive-k-out-of-n: F system. In a degraded situation, we propose a reconfigured control allocation based on the on-line estimation of actuators reliability. The analytic solution proposed in the literature to compute the reliability of consecutive-k-out-of-n: F system is not appropriated to compute the reliability on-line. The graphical aspect of Bayesian Network is very interesting because it formalizes the model by coupling a generic model structure with simple parameter matrices and the inference computes the reliability of actuators according to on-line observations (evidences). It is applied on circular and linear typical consecutive-k-out-of-n: F system to estimate its reliability and provide the parameters to distribute the control efforts among the redundant set of actuators. The second application is concerned with the decision-making in fault diagnosis methods which generally relies on the analysis of fault signature vectors. The talk presents a new approach of decision-making in the case of signature vectors for various identical or similar faults. The main contributionconsists in the fusion between the reliability of components and the evaluation of the residuals in order to increase the fault isolation efficiency. The decision-making, formalised as a Bayesian network, is established with a priori knowledge on fault signatures, false alarm and missing detection probability, on line component state estimation computed by a Bayesian fusion of the component reliability and measurements. The effectiveness and performances of the method are illustrated on a heating water process.
Fichier non déposé

Dates et versions

hal-00650722 , version 1 (12-12-2011)

Identifiants

  • HAL Id : hal-00650722 , version 1

Citer

Philippe Weber, Christophe Simon, Didier Theilliol, Christophe Aubrun. Bayesian networks model of reliability applications to Reconfiguration and Diagnosis. 2011. ⟨hal-00650722⟩
74 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More