Abstract
The detection of losses from nuclear material inventories whether by theft, leakage or other means is a problem of considerable importance. The earliest approaches to this problem were based on control chart like methods that made the assumption that the time series of material balance measurements (material balances are computed directly from analytical measurements of the amount of material present in such a manner that a zero balance means that all of the material is accounted for) was made up of observations that were statistically independent and identically normally distributed. These methods did not work well for two reasons. First, the series observations are often not independent. Second, if there are losses, they appear as outliers in the material balance series, which violates the normality assumption. The present paper reviews the literature in this area and discusses successful new chemometric methods that avoid the above two difficulties.
Keywords: nuclear material safeguards, outliers, statistical process control, robust time series analysis
2