Building intelligent sensor networks with multiagent graphical models

Abstract

Modern society relies heavily on various equipments. To ensure productive operation, avoid downtime and reduce maintenance cost, engineers must constantly determine whether equipment is operating normally and what is the small set of devices that is highly likely the culprit of abnormality. A sensor network is often deployed to gather and process the key information in this decision process. With the traditional centralized approach for sensor network monitoring, the data transmission can introduce delay, the centralized processing can create a bottleneck, and the central unit must have access to all the knowledge needed. The intelligent sensor network is a promising alternative, where a set of distributed agents embody local sensors, local computing resources, local knowledge, and inference procedures, and cooperate through limited communication. This chapter introduces, at the application development level, the approach based on multiply sectioned Bayesian networks, a probabilistic framework for agent inference in intelligent sensor networks. Through a case study, key technological steps involved in applying the framework are linked together and practitioners are facilitated in mapping theoretical intricacies to practical reality. © 2008 Springer-Verlag Berlin Heidelberg.