An empirical mode decomposition-independent component analysis based approach for detecting localized and distributed faults in rolling bearing diagnostics

Rolling bearing diagnostics still represents an open research field, especially when distributed faults are looked for rather than localized faults. In fact, distributed faults are typically due to a progressive growth of surface wear. A low-quality manufacturing, in terms of material or process, can even constitute another cause of distributed fault or representing an accelerating factor for the fault development. Classical strategies adopted for diagnosing localized faults can barely recognize this type of faults. However, certain approaches based on the extraction of the spectral components building the vibrational signature of the bearing can be exploited to diagnose both localized and distributed faults. This paper aims at presenting an approach that can be exploited for this purpose. The algorithm is based on a combined use of Empirical Mode Decomposition (EMD) and Independent Component Analysis (ICA). EMD is exploited as a pre-processing step to decompose the original signal into multiple time-series, the so-called intrinsic mode functions. These time series are then processed by ICA in order to extract those components that can be related to the fault. The non-stationary content of the distributed fault is taken into account by both methods. The effectiveness of the whole procedure in tackling the distributed faults diagnostic issue is presented on simulated data. A sensitivity analysis is presented as well.