Data-driven prognostic for system health management represents an emerging and challenging application of data mining. The objective is to develop data-driven prognostic models to predict the likelihood of a component failure and estimate the remaining useful lifetime. Many models developed using techniques from data mining and machine learning can detect the precursors of a failure but sometimes fail to precisely predict time to failure. This paper attempts to address this problem by proposing a novel approach to find reliable patterns for prognostics. A reliable pattern can predict state transitions from current situation to upcoming failures and therefore help better estimate the time to failure. Using techniques from data mining and time-series analysis, we developed a KDD methodology for discovering reliable patterns from multi-stream time-series databases. The techniques have been applied to a real-world application: train prognostics. This paper reports the developed methodology along with preliminary results obtained on prognostics of wheel failures on train.
CITATION STYLE
Ferilli, S., Basile, T. M. A., & Di Mauro, N. (2011). Modern Approaches in Applied Intelligence. (K. G. Mehrotra, C. K. Mohan, J. C. Oh, P. K. Varshney, & M. Ali, Eds.), IEA/AIE (1) (Vol. 6703, pp. 275–284). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-21822-4
Mendeley helps you to discover research relevant for your work.