23 Some General Insights Regarding Development of NPP Dukovany PSA Input Parameters Set Made on the Base of Broad Analysis of Operational Experience (PSAM-0066)

The development of quantitative input parameters of NPP Dukovany PSA study has been intentionally based, to the maximum possible extent, on available plant specific experience. To achieve this goal, periodic update of all kinds of PSA model input parameters has become an integral part of plant Living PSA project. In 2003, a broad analysis of plant specific experience with regard to common cause failures was carried out. In 2004, a complete review and update of the other PSA quantitative inputs (except for human error probabilities) was done.

The main output of detailed elaboration of plant specific experience was a new set of PSA model quantitative parameters - initiating event frequencies, component failure rates and on demand failure probabilities, common cause failure models parameters and unavailabilities due to maintenance. Some additional useful conclusions were reached during the analysis, which can be summarized into a set of general statements.

The main purpose of the contribution is to comment this general experience and to discuss and confront it with some stereotypes appearing sometimes in the approaches to estimation of PSA model parameters. For example, in derivation of initiating event frequencies, generic data sets are used sometimes with a premise that plant specific experience is missing completely or is too rare, without more detailed consideration of real plant data. Similarly, CCF probabilities are based, from time to time, just on generic information transfer. The validity of general “automatic” assumptions has been made a subject of discussion in the first part of the presentation.

In the second part of the presentation, a general overview of quantitative results of NPP Dukovany data analysis is made. Besides that, some methodological aspects (the role, Bayesian approach should play in the process of PSA quantification) and potential data collection problems (common cause failures caused by human) are discussed.