Analysis and comparison of multiple features for fault detection and prognostic in ball bearings

Thi Phuong Khanh Nguyen (1) , Amor Khlaief (1) , Kamal Medjaher (1) , Antoine Picot (2) , Pascal Maussion (2) , Diego Tobon (3) , Bertrand Chauchat (3) , Regis Cheron (3)

1 LGP - Laboratoire Génie de Production
2 LAPLACE-CODIASE - COmmande et DIAgnostic des Systèmes Electriques
3 Alstom Transport - Alstom Transport

Feature extraction is one of the most important elements in Prognostics and Health Management (PHM) systems. Numerous techniques have been proposed for fault detection, diagnostics and prognostics in ball bearings which are key components of rotating machineries, widely used in the industry. Considering the strengths and weaknesses of these techniques, this paper aims at evaluating and analyzing different features in all three signal processing domains: time, frequency and time-frequency. The crucial indicators related to normal and abnormal cases are extracted from both vibration signals and stator current signals. Then, a new metric is proposed to measure the evolution of these indicators with respect to degradation levels of bearings. The performance of every indicator is analyzed to study which feature(s) is(are) better than other(s) and which feature(s) is(are) the best appropriate for vibration and current signals. These results could be effectively used in future for fault detection, diagnostics and prognostics applications.

Mots clés: Feature extraction; Fault detection; Fault prognostics

Domaines: Autre

Thi Phuong Khanh Nguyen, Amor Khlaief, Kamal Medjaher, Antoine Picot, Pascal Maussion, et al.. Analysis and comparison of multiple features for fault detection and prognostic in ball bearings. Fourth european conference of the prognostics and health management society 2018, Jul 2018, Utrecht, Netherlands. pp.1-9. ⟨hal-02111843⟩