AN ANOMALY DETECTOR APPLIED TO A MATERIALS CONTROL AND ACCOUNTING SYSTEM*

Large amounts of safeguards data are automatically gathered and stored by monitoring instruments used in nuclear chemical processing plants, nuclear material storage facilities, and nuclear fuel fabrication facilities. An integrated safeguards approach requires the ability to identify anomalous activities or states in these data. Anomalies in the data could be indications of error, theft, or diversion of material. The large volume of the data makes analysis and evaluation by human experts very tedious, and the complex and diverse nature of the data makes these tasks difficult to automate. This paper describes our early work in the development of analysis tools to automate the anomaly detection process. Using data from accounting databases, we are modeling the normal behavior of processes. From these models we hope to be able to identify activities or data that deviate from that norm. Such tools would be used to reveal trends, identify errors, and recognize unusual data. Thus the expert's attention can be focused directly on significant phenomena