Evaluating Safeguards Statistical Assumptions Via Stochastic Simulation

Year
2021
Author(s)
Chris Gazze - International Atomic Energy Agency
Sukesh K. Aghara - University of Massachusetts Lowell
Lohith Annadevula - University of Massachusetts Lowell
Logan J Joyce - University of Massachusetts Lowell
Kenneth D Jarman - Pacific Northwest National Laboratory
Jose Gomera - Brookhaven National Laboratory
Katherine M. Bachner - Brookhaven National Laboratory
Claude F. Norman - International Atomic Energy Agency
Robert Binner - International Atomic Energy Agency
File Attachment
a540.pdf1.64 MB
Abstract
A stochastic simulation was built and tested to estimate achieved detection probabilities (DPs) on a stratum basis, over a tailorable range of diverted amounts from 0 to 2 SQ, using typical IAEA inspection data: i.e., SQ in stratum, number of items, number of gross/partial/bias defect measurements conducted, and realistic relative standard deviation (RSD) values for typical IAEA verification measurements. For bulk strata, the model calculates achieved DP at 0.01 SQ diversion increments; for item strata, the model calculates DP using the smallest realistic diversion increment (e.g., a plate, pin, or coupon). After successfully benchmarking against IAEA deterministic models, the simulation was used to test the sensitivity of DP to certain standard assumptions and selected input parameters. First, the equal defect assumption was tested; the results suggest significant complexity in the effectiveness of partial defect measurements. Next, the authors explored the sensitivity of DP to the assumed RSD of attribute tests. Then, the authors compared non-normal models for instrument performance (e.g., logistic, step, or arbitrary functions) to the typical results from a normal distribution (characterized by RSD). This last comparison was supplemented with experimentally derived performance data for an HM-5. The HM-5 was used to make enrichment measurements on both LEU and HEU MTR fuel elements as plates were removed, and the results fit with logistic curves and plugged into the simulation. These stochastic DP results were compared to DP estimates from a deterministic model assuming a normal curve and typical RSD, yielding insights that could improve effectiveness in the field. These early results illustrate the potential of stochastic models to better understand achieved DP and to improve safeguards effectiveness.