Time Discretisation Applied to Anomaly Detection in a Marine Engine

This paper introduces the problems associated with anomaly detection in a marine engine, and explains the benefits that the SAX representation brings to the field. Despite limitations in accuracy of the SAX representation in comparison with the normalised time series, we conclude that because of the reduction in data points that should be processed SAX should be considered further as a valid and efficient representation. Finally, a continuation of the work to make the approach more ecologically viable is briefly noted based upon Markov Chaining and Support Vector Machines.

• Type: Conference
• Year: 2007
• Where: KES
• Authors: Ian Morgan, Honghai Liu, George Turnbull, David J. Brown
• Keywords: Anomaly Detection, KES 2007, Normalised Time Series, SAX Representation