A METHOD TO COMPARE PSA MODELS IN A MODULAR PSA

摘要

Most Probabilistic Safety Assessment (PSA) models are typically based on the fault tree and event tree approach. Since decades, these PSA models serve as important tool for safety demonstration and as support for regulatory issues. In some industries, for example in nuclear power plants, PSA models increased in size and complexity over the recent years as a consequence of the scope extension to external hazards, new applications and new requirements (due post-Fukushima insights and recommendations). Therefore, for the purpose of quality assurance, it is worth to have a precise log of any model evolutions to guaranty compliance with standards and to ensure that models reflect the reality of plants. However, in the database architecture of currently used PSA tools, only meta data information can be obtained concerning model modifications. Analysts (users), developers and reviewers may need to have deep insights on different model transitions (set of modifications), and then go through details in order to verify and justify (for example to safety authorities) the set of modifications applied to a PSA model. Currently, those activities are performed manually and can be time-consuming and error-prone since PSA models may contain dozens of thousands of model objects. In this article, a method is presented to automatically compare PSA models. The method provides important feedback to model engineers for example to verify model modifications applied since an ancient model state, for checking or diagnosis purposes (for instance to understand the impact and importance of individual model modifications). Further, the result (the differences) can be used to automatically generate modification reports to trace and justify model modifications. Finally, it can serve as preliminary step for the purpose of model fusion (the "combination" of models) which is an ultimate step to concurrent modeling. The comparison method has been implemented in Andromeda, a research software developed at EDF R&D to develop and test new modeling approaches in a so-called "modular PSA". A modular PSA treats models by smaller pieces (the "modules") what constitutes a crucial paradigm for the comparison method of this article.