Functional evaluation of an event detection ensemble to detect anomalous system behavior

Surveillance systems are established to manage the complexity of ensuring processes of interest behave as expected. They are formed in response to public demand that systems—natural and human-made—be predictable, managed, and under control. This paper examines the functionality of an event detection ensemble used to detect anomalous conditions a system of interest may exhibit. The event detection ensemble consists of an agent and its associated sensors, models, and detectors. In addition, the subsystem is presented in the broader context of a generalized surveillance design framework. A logical organization of components is presented as well as a demonstration implementation scheme. The event detection ensemble is evaluated for functionality using two hypothetical test cases representing real-world applications. Event signatures based upon prediction, extrapolation, and domain discrepancies are characterized and simulated. In addition, a radial basis neural network is employed to create a model capable of distinguishing these types of discrepancies. Finally, application and benefits of this design approach is discussed with respect to designing surveillance systems for human-made and natural systems. Discussion includes how this design framework can be applied to detect events in a wide array of industrial engineering applications.