Random Inspection Planning For Misuse Detection In Safeguards

The IAEA uses random inspections (RIs) to, inter alia, provide credible assurance that declared nuclear facilities are not used for undeclared purposes. These inspections are random in the sense that they are scheduled randomly in date and time, with short notice given to the inspected site. The IAEA has interest in employing statistical models for RI planning that take advantage of any potential efficiency gains while maintaining a high level of effectiveness. This paper first introduces the model parameters that are necessary for a quantitative analysis of RI models for misuse inspections (in the remainder the short expression is: RI models), and discuss their importance. Then, using the model parameters, the set of all RI models is introduced and four examples of RI models are presented. Next, for any RI model the probability that any facility is selected at least once per year for a RI, and - regarding the objective of a RI - the probability that a misuse is detected within T days after its start, where the parameter T is the duration of misuse signatures at the facility, is derived. Next, the question is addressed which RI model should be chosen for RI planning: If no further constraints from the Inspectorate are imposed on the RI models (e.g., maximum unpredictability, resource constraints leading to an upper number of RIs, etc.), then the RI model that maximizes the achieved detection probability for a given set of input parameters should be selected. This maximization problem, however, is by no means trivial, because the maximization is performed over a set of RI models and not over a subset of real numbers. Finally, the functionality and features of the software prototype TRIPS (Tool for Random Inspection Planning in Safeguards) are demonstrated, and future work topics are highlighted.