Mining satellite telemetry data: Comparison of rule-induction and association mining techniques

An artificial Satellite's health and functioning in space is monitored by analysing the stream of Telemetry Data received at the Satellite Ground Control Centre. This data is received continually during the entire mission life of the Satellite. Hence a huge volume of Data is collected for any given satellite. Data Mining techniques are now regularly applied to analyse such large quantity of data for outlier analysis, fault detection in past data and fault prediction. In the present work, hidden relationships between apparently unrelated Telemetry parameters are mined by applying Rule Induction and Association Rule mining algorithms separately on a data set with randomly selected parameters. From the results obtained, a comparison is made to assess the suitability of either of these techniques to the application domain. This paper presents a finding that Rule Mining Algorithm is most appropriate for the domain application since Association rule mining is not amenable to data sets of Satellite Telemetry Parameters which are highly unrelated but whose functional inter-relationships during specified on-orbit stages are to be mined.