This paper describes research on the DAME Project into the use of Correlation Matrix Memories for imprecise pattern matching on large volumes of time series data using the Grid. This technology, named AURA has been developed over a period of 15 years at the University of York. In the DAME project, we have begun to apply AURA technology to the problem of searching aero-engine vibration data with the aim of finding partial matches to vibration anomalies and thus providing information that can be used to form an early prognosis / diagnosis of potential faults. Within the DAME project, the underlying AURA technology has been improved in terms of its speed and memory usage and adapted for use as a Grid service under Globus Toolkit 3. At the mid-point in the project (July 2003), research is ongoing into encoding techniques suited to fast processing of time series data. In the second half of the project, we intend to investigate the use of advanced data mining methods in combination with a distributed grid enabled implementation of AURA to support flexible and very high performance pattern matching over large amounts of data.
|Published - Sept 2003