Detecting Proliferation Activities via System-centric Integration and Interpretation of Multi-Modal Data Collected from a System of Sensors

This paper illustrates the potential benefit that real-time detection of anomalies via system-centric process monitoring (PM) may have as a proliferation deterrent. This benefit is illustrated by demonstrating the robust performance that the proposed system-centric PM approach delivers in alerting the potential existence of proliferation activities at monitored facilities. Under system-centric PM, sensors are deployed to observe unit operations comprising the monitored facility. Multi-modal sensor data are collected and/or analyzed by unattended on-line measurement and monitoring systems (MMS), which are then processed at the system-centric level by data integration and interpretation (DII) module to detect and keep track of the occurrences of abnormal patterns of interest. Anomalies of interest for detection may represent undeclared plant operations characterized by specific signatures, such as patterns of events occurring at different locations of the monitored facility and at different stages of operations. The interest here is to directly look for operational anomalies rather than indirectly inferring them by measuring mass balance inconsistencies resulting from their occurrences. The underlying assumption is that defined abnormal patterns correspond to signatures that manifest themselves whenever proliferation activities are conducted. In order to illustrate the benefits of assessing the nonproliferation condition of monitored facilities based on multi-modal data collected from a system of sensors, we consider a multi unit facility consisting of five unit operations observed using twenty two sensors distributed around the monitored facility. We assume that the three operational anomalies of interest for detection correspond to the protracted diversion of special nuclear material (SNM) into three different streams that do not normally receive this material. The three diversion scenarios considered are achieved by abnormally operating the unit operations. For detecting them, we consider a DII approach based on PM, which does not directly compute mass balance calculations, but rather monitors for anomaly patterns related to potential diversion of SNM caused by the execution of proliferation activities. By effectively integrating collected observations in time and space, the proposed system-centric PM approach tries to infer whether these specific abnormal patterns of events have occurred and how many times within a given time period. Simulation results are discussed.