Uncertainty importance is used to rank sources of uncertainty in the input variables for their degree of contribution to the uncertainty of the output variable(s). Such ranking is used to plan in reducing epistemic uncertainty in the output variable(s). In application to Thermal-Hydraulic calculations using RELAP5, TRAC and TRACE codes involving uncertainty, uncertainty ranking can be used to confirm the results of Phenomena Identification and Ranking Table (PIRT) utilized in available uncertainty quantification methodologies. Several methodologies have been developed to address uncertainty importance assessment. Existing methodologies for uncertainty importance are not practical for some applications such as some Thermal-Hydraulic calculations due to required computational time and resources. A new efficient uncertainty importance ranking method is proposed as part of a broader research conducted by the authors for comprehensive TH code uncertainty assessment. Given the computational complexities of the TH codes, the proposed uncertainty importance measure is defined in multiples of standard deviation (xσ) changes in a given input parameter or variable over the resulting changes in the standard deviation of output variable (such as a Figure of Merit). The total uncertainty range resulted from propagation of uncertainties is obtained from several available methodologies e.g., CSAU, GRS, UMAE and the recently proposed integrated methodology IMTHUA, proposed by the author. There are some difficulties in assessment of non-linearity of some input changes vs. variations in the output variables, which require special treatment. Different levels of input change (multiples of standard variation) are devised for accurate ranking of uncertainty contributors. Comparing the output change as a fraction of the overall uncertainty range will result in a ranking index to show the contribution of each uncertainty source. In this paper a brief overview of the importance analysis as well as the difference between uncertainty-importance vs. importance uncertainty will be first given. Current methodologies for uncertainty importance will be discussed and their applicability to TH analysis will be discussed. A description of the proposed methodology, along with an example of its application to LOFT LBLOCA uncertain parameters will be discussed.
A kernel estimate method for characteristic function-based uncertainty importance measure
