The design and development of multi-agent based RFID middleware system for data and devices management

Abstract

Radio frequency identification technology (RFID) has emerged as a key technology for automatic identification and promises to revolutionize business processes. While RFID technology adoption is improving rapidly, reliable and widespread deployment of this technology still faces many significant challenges. The key deployment challenges include how to use the simple, unreliable raw data generated by RFID deployments to make business decisions; and how to manage a large number of deployed RFID devices.

In this thesis, a multi-agent based RFID middleware which addresses some of the RFID data and device management challenges was developed. The middleware developed abstracts the auto-identification applications from physical RFID device specific details and provides necessary services such as device management, data cleaning, event generation, query capabilities and event persistence. The use of software agent technology offers a more scalable and distributed system architecture for the proposed middleware. As part of a multi-agent system, application-independent domain ontology for RFID devices was developed. This ontology can be used or extended in any application interested with RFID domain ontology.

In order to address the event processing tasks within the proposed middleware system, a temporal-based RFID data model which considers both applications’ temporal and spatial granules in the data model itself for efficient event processing was developed. The developed data model extends the conventional Entity-Relationship constructs by adding a time attribute to the model. By maintaining the history of events and state changes, the data model captures the fundamental RFID application logic within the data model. Hence, this new data model supports efficient generation of application level events, updating, querying and analysis of both recent and historical events.

As part of the RFID middleware, an adaptive sliding-window based data cleaning scheme for reducing missed readings from RFID data streams (called WSTD) was also developed. The WSTD scheme models the unreliability of the RFID readings by viewing RFID streams as a statistical sample of tags in the physical world, and exploits techniques grounded in sampling theory to drive its cleaning processes. The WSTD scheme is capable of efficiently coping with both environmental variations and tag dynamics by automatically and continuously adapting its cleaning window size, based on observed readings.