Sequential test procedures for detecting protracted materials losses

Sequential tests are required for detecting protracted (trickle) losses of strategic special nuclear materials from a single materials control unit (MCU). We compared applicable tests including modified versions of Page's test and power-one procedures. We used simulated data from a MCU in a conversion/fabrication process that took into account process variations, materials holdup, and measurement uncertainties. Comparisons were made over a 60-day accounting period under different loss scenarios. Some important findings include: (1) no single test procedure is best for all diversion scenarios; (2) power-one procedures are best for protracted losses that occur early in the accounting period and Page's test is best for late loss occurrence; (3) if holdup and process variations are not included in the Inventory Difference model but are present in the process, then assuming steady-state conditions, false-alarm probabilities can double. 19 refs., 16 figs., 3 tabs