Expert systems: AI decision support tools for fire safety assessment - IRSN - Institut de radioprotection et de sûreté nucléaire Accéder directement au contenu
Poster De Conférence Année : 2023

Expert systems: AI decision support tools for fire safety assessment

Résumé

Fire safety studies conducted at IRSN aim to assess the adequacy and effectiveness of risk control measures taken by the operators of nuclear facilities. In support of these studies, IRSN has been developing for several years expert systems based on the Bayesian Belief Network methodology. These tools, taking advantage of the two-zone model SYLVIA, make it possible to better understand the behaviour of a nuclear facility and its systems in a fire situation. Specifically, this is achieved by dynamically crossing the knowledge of the physical phenomena involved, with the knowledge specific to the facility. As a result, expert systems allow for a better targeting of safety issues when carrying out risk assessments. An illustrative example is provided by the SyPRE expert system, developed for the study of pressure effects on fire-barriers resulting from a fire in nuclear facilities. A limiting factor for creating complex expert systems is the size of the simulation database. Thus, the challenge is to increase the complexity of expert systems by taking into account specific features of a nuclear facility, without having to exponentially increase the size of the knowledge base and the time to build it. An initial area of development was to adapt the expert system to the expert's use in order to make the best use of the knowledge base. This has been achieved through the development of the SE-Toolbox which allows an expert to set up his own expert system. A second area of development currently being investigated at IRSN consists of studying the extent to which machine learning techniques (neural networks) could be used to create approximation models of the SYLVIA software used to create the expert system knowledge base. The advantage of these techniques is that the computational time required to build large databases is considerably reduced, allowing a greater number of causal links to be characterized numerically. These areas of development are illustrated on the poster.
Fichier principal
Vignette du fichier
Poster_Plumecocq et al.pdf (339.9 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Licence : Copyright (Tous droits réservés)

Dates et versions

irsn-04176829 , version 1 (03-08-2023)

Identifiants

  • HAL Id : irsn-04176829 , version 1

Citer

William Plumecocq, Jean Paul Joret, Eric Chojnacki. Expert systems: AI decision support tools for fire safety assessment. 14th International Symposium on Fire Safety Science, Oct 2023, Tsukuba (JAPAN), Japan. ⟨irsn-04176829⟩
41 Consultations
14 Téléchargements

Partager

Gmail Facebook X LinkedIn More