Extreme value statistics for vibration spectra outlier detection

Aurélien Hazan (1) , Jérôme Lacaille (2) , Kurosh K. Madani (1)

1 SYNAPSE
2 SNECMA Villaroche [Moissy-Cramayel]

The issue of detecting abnormal vibrations from spectra is tackled in this article, when little is known about the mechanical behavior of the system, or the fault pattern. To do so, an original algorithm relying on the statistics of the maximum of log-periodograms is introduced. Receiver Operator Characteristic (ROC) curves are built and show good performance with bearing deterioration even when little wear is present.

Mots clés: fault detection; vibration; rotating machine; bearing; periodogram; maximum; excess; generalized pareto; probabilistic model

Domaines: Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]; Machine Learning [stat.ML]

Aurélien Hazan, Jérôme Lacaille, Kurosh K. Madani. Extreme value statistics for vibration spectra outlier detection. International Conference on Condition Monitoring and Machinery Failure Prevention Technologies, Jun 2012, Londres, United Kingdom. p.1. ⟨hal-00681045⟩