Statistical Analysis Of Convergence And Error Propagation In Stochastic Model For Safeguards Inspection

IAEA safeguards experts use detection probability (DP) as the primary effectiveness metric for nuclear material inventory and flow verification activities. The DP is calculated over a spectrum of diversion scenarios (from a few items with gross defects to a large number of items with bias defects), and the worst-case (lowest) DP is reported. Deterministic models using statistical distributions are generally used to compute achieved DP for individual nuclear material strata to assess the effectiveness of the IAEA verification inspections. The models get involved as the total number of defect types in the multi-defect sample space increases. The model must first calculate the item selection probability as well as the identification probability at each step in each stratum. Once done, the model then aggregates the DP from each stratum to calculate an aggregate detection probability (ADP) for the entire facility. The model must consider the broad range of ways in which material could be diverted within the facility to add up to that total and determine the minimum ADP over all possible diversion strategies. The stochastic approach to simulating inspection of items in a nuclear facility involves random selection of items containing nuclear material and predicting the results of measurements for these samples. Stochastic methods offer greater flexibility to model scenario development to simulate different diversion strategies and measurement characteristics. Stochastic methods rely on multiple simulations for each scenario to generate a distribution of DP values to compute the average and the uncertainty. A critical attribute of stochastic methods for such an application is a rigorous analysis of convergence (in terms of number of stochastic “trials”) and error, or precision in the estimated DP. This paper describes the method and its application to international safeguards inspection performance evaluation.