Deterministic Sensitivity and Uncertainty Methodology for Best Estimate System Codes applied in Nuclear Technology

Nuclear Power Plant (NPP) technology has been developed based on the traditional defense in depth philosophy supported by deterministic and overly conservative methods for safety analysis. In the 1970s [1], conservative hypotheses were introduced for safety analyses to address existing uncertainties. Since then, intensive thermal-hydraulic experimental research has resulted in a considerable increase in knowledge and consequently in the development of best-estimate codes able to provide more realistic information about the physical behaviour and to identify the most relevant safety issues allowing the evaluation of the existing actual margins between the results of the calculations and the acceptance criteria. However, the best-estimate calculation results from complex thermal-hydraulic system codes (like Relap5, Cathare, Athlet, Trace, etc..) are affected by unavoidable approximations that are un-predictable without the use of computational tools that account for the various sources of uncertainty. Therefore the use of best-estimate codes (BE) within the reactor technology, either for design or safety purposes, implies understanding and accepting the limitations and the deficiencies of those codes. Taking into consideration the above framework, a comprehensive approach for utilizing quantified uncertainties arising from Integral Test Facilities (ITFs, [2]) and Separate Effect Test Facilities (SETFs, [3]) in the process of calibrating complex computer models for the application to NPP transient scenarios has been developed. The methodology proposed is capable of accommodating multiple SETFs and ITFs to learn as much as possible about uncertain parameters, allowing for the improvement of the computer model predictions based on the available experimental evidences. The proposed methodology constitutes a major step forward with respect to the generally used expert judgment and statistical methods as it permits a) to establish the uncertainties of any parameter characterizing the system, based on a fully mathematical approach where the experimental evidences play the major role and b) to calculate an improved estimate of the computed response and relative improved (i.e. reduced) uncertainty