ON SETTING NRC ALARM THRESHOLDS FOR INVENTORY DIFFERENCES AND PROCESS UNIT LOSS ESTIMATORS: CLARIFYING THEIR STATISTICAL BASIS WITH HYPOTHESIS TESTING METHODS AND ERROR PROPAGATION MODELS FROM JAECH, BOWEN AND BENNETT AND IAEA

Major fuel cycle facilities in the U.S. private sector are required to respond-at predetermined alarm levels-to various special nuclear material loss estimators in the material control and accounting (MC&A) area. This paper presents U.S. Nuclear Regulatory Commission (NRC) policy, along with the underlying statistical rationale, for establishing and inspecting the application of thresholds to detect excessive inventory differences (ID). Accordingly, escalating responsive action must be taken to satisfy NRC's MC&A regulations for low -enriched uranium (LEU) fuel conversion/fabrication plants and LEU enrichment facilities. The establishment of appropriate ID detection thresholds depends on a site-specific goal quantity, a specified probability of detection and the standard error of the ID. Regulatory guidelines for ID significance tests and process control tests conducted by licensees with highly enriched uranium are similarly rationalized in definitive hypothesis testingincluding null and alternative hypotheses; statistical errors of the first, second, third, and fourth kinds; and suitable test statistics, uncertainty estimates, prevailing assumptions, and critical values for comparisons. Conceptual approaches are described in the context of significance test considerations and measurement error modelsincluding the treatment of so called \"systematic error variance\" effects as observations of random variables in the statistical sense, using as complementary references Jaech (USAEC, 1973), Bowen and Bennett (USNRC, 1988) and the IAEA statistical manual (IAEA, 1989).