• Citer

A Monte Carlo algorithm for dynamic PSA based on the concept of stimulus

par Jourdain, Aymeric ;Labeau, Pierre-Etienne
Référence (March 13-17, 2011: Hilton Wilmington Riverside, Wilmington, NC), PSA 2011 international topical meeting on probabilistic safety assessment and analysis, American Nuclear Society, La Grange Park, IL
Publication Publié, 2011

Publication dans des actes

• Détails
• Contenu

Résumé :

The theory of probabilistic dynamics (TPD) was first introduced in order to overcome some of the limitations of the classical PSA methodology, by incorporating the coupling between the deterministic evolution of the process variables and discrete stochastic transitions in the delineation process of accident sequences. The Stimulus-Driven Theory of Probabilistic Dynamics (SDTPD) enriches the TPD framework by modeling in a finer fashion the competing process defining the next branching in an event tree. Each possible next event is modeled as a two-stage process: first, a so-called stimulus must be activated, i.e. conditions necessary for the event to take place must be satisfied; then a delay must elapse before the actual event occurrence. An analog Monte Carlo game can easily be implemented to solve these problems. Yet it usually turns out to be inefficient, as rare scenarios with potentially high damage are not or insufficiently sampled. To tackle this issue, an innovative algorithm properly uses the outputs of a pre-simulation of the mother branch of the event tree and the SDTPD to sample more systematically various types of branching events out of this mother branch. Compared with a classical analog simulation, this new algorithm leads to a better identification of rare sequences and a more accurate estimation of their frequency. This method is illustrated on a pressurization transient in containment. Different sampling methods of branching points along the mother branch are considered and their efficiency compared with that of the analog Monte Carlo game.