Detecting Fault Conditions in Distributed Sensor Networks Using Dynamic Bayesian Networks

George Chin Pacific Northwest National Laboratory
Sutanay Choudhury Pacific Northwest National Laboratory
Lars Kangas Pacific Northwest National Laboratory
Sally McFarlane U.S. Department of Energy
Kevin Widener Pacific Northwest National Laboratory

Category: Instruments

A Bayesian network (BN) is a probabilistic graphical model, where nodes represent random variables and directed edges represent conditional dependencies. A dynamic Bayesian network (DBN) models the stochastic evolution of a set of variables over time. We are developing DBN models for identifying and analyzing fault conditions occurring in atmospheric radiation and measurement sensor networks. In these models, the different variables represent the various measurements that may be taken across a distributed sensor network. The variables are linked in the DBN to convey both causal and temporal relationships. Using the DBN, we may then compute the probabilities that particular measurements are unexpected or anomalous and detect whether parts of the sensor network are behaving unusually or erratically. A full set of DBNs of varying complexities is currently under construction to cover different collections and ranges of measurements and time frames.

This poster will be displayed at ASR Science Team Meeting.