Dynamic probabilistic safety margin sensitivity analysis for model selection in nuclear power plant probabilistic safety assessment (PSA)

The safety assessment of Nuclear Power Plants (NPPs) relies on Thermal-Hydraulic (TH) codes for evaluating the safety parameters responses to postulated accidental scenarios and their safety margins with respect upper (or lower) safety thresholds. To explicitly treat the model uncertainties and account for these when assessing the safety margins, we adopt a Risk Informed Safety Margin Characterization (RISMC) definition of the safety margin that is called Dynamic Probabilistic Safety Margin (DPSM). Besides the calculation of the safety margins for the accidental scenarios considered, a NPP safety assessment also requires the quantification of the probabilities of occurrence of the accidental scenarios, based on a probabilistic model of the accident progression to be properly defined. At least two alternatives have to be considered for the probabilistic models: static traditional Event Trees (ETs) and Dynamic Event Trees (DETs). In this thesis, we propose a sensitivity analysis to be used for selecting the most proper probability assessment model, which calculates how much the TH codes inputs affect the calculated DPSM. The results of the sensitivity analysis will be of help to decide whether a static analysis is sufficient, or a dynamic assessment is, instead, necessary to give due account to the system dynamics. The methodology presented is applied to two different cases of study: a Station Black Out (SBO) accident followed by a Seal Loss Of Coolant Accident (LOCA) for a 3-loops Pressurized Water Reactor (PWR) whose dynamics is simulated by a MAAP5 model, and the accidental scenarios that can occur in a U-Tube Steam Generator (UTSG) of a NPP, whose dynamics is simulated by a SIMULINK model. Results show that, even if both systems are accurately modelled resorting to dynamic simulation models, the sensitivity analysis performed on the DPSM that are calculated for the considered accidental scenarios highlights the necessity to resort to a DET, for one case, and to an ET, on the other case, for modelling the probabilistic accident progression.