Developing paradigms of data fusion for sensor - actuator networks that perform engineering tasks

ata fusion is a paradigm for integrating data from multiple sources to synthesize new information such that the whole is greater than the sum of its parts. This is a critical task in contemporary and future systems that have distributed networks of lowcost, resource-constrained sensors [1], [2]. Current techniques for data fusion are based on general principles of distributed systems and rely on cohesive data representations to integrate multiple sources of data. Such methods do not extend easily to systems in which real-time data must be gathered periodically by cooperative sensors, where some decisions become more critical than other decisions episodically. There has been extensive study in the areas of multisensor fusion and real-time sensor integration for time-critical sensor readings [3]. A distributed sensor data network is a set of spatially scattered sensors designed to derive appropriate inferences from the information gathered. The development of such networks for information gathering in unstructured environments is receiving much interest, partly because of the availability of new sensor technology that is economically feasible to implement [4]. Sensor data networks represent a class of distributed systems that are used for sensing and in-situ processing of spatially and temporally dense data from limited resources and harsh environments, by routing and cooperatively processing the information gathered. In all these systems, the critical step is the fusion of data gathered by sensors to synthesize new information. Our interest is in developing paradigms of data fusion for sensor-actuator networks that perform engineering tasks. We use automation systems as an illustrative example. Automation systems represent an important, highly engineered, trillion-dollar business in the United States. Developing paradigms of data fusion for sensor-actuator networks that perform engineering tasks.