A diagnostic protocol for the monitoring of bearing fault evolution based on blind deconvolution algorithms

The detection and identification of bearing faults at their initial stage is pivotal in order to avoid catastrophic failures. However, the vibration contribution related to early stage bearing faults are frequently weak and masked by strong background noise and mechanical interferences. In this scenario, blind deconvolution algorithms can be exploited for extracting impulsive patterns related to incipient bearing faults. Maximum Correlated Kurtosis Deconvolution (MCKD) and Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA) proved to be effective for fault diagnosis in rotating machines. However, their effectiveness on monitoring the progressive degradation of rolling element bearings has not yet been exhaustively studied. In this paper, the experimental data from an endurance test are investigated by means of MCKD and MOMEDA. The results in terms of incipient fault detection and fault identification accuracy are discussed from different perspectives, highlighting advantages and limits of such blind deconvolution approaches. In particular, an original diagnostic protocol is proposed, based on a condition indicator computed from the cumulative of the blind deconvolution maximized criterion combined with a non-parametric statistical threshold. The proposed indicator is sensitive to the fault degradation as well as the fault type.