Transient morphology analysis and sparse representation for bearing fault diagnosis under variable speed condition

Abstract

Sparse representation has been extensively applied for bearing fault diagnosis under constant speed operation. However, its application to the variable speed case is confined as, unlike the constant speed case, the fault-induced transients under variable speed are more complex and the changing pattern of transient morphology along rotating speed is uncertain. As such, this paper firstly investigates the morphology of faulty bearing vibration response to reveal that the rotating speed variations have negligible effects on morphology of the fault-induced transients. Then an efficient dictionary spanned by a single atom can be constructed, where the optimal wavelet atom is selected by the correlation filtering strategy. The stage-wise orthogonal matching pursuit (StOMP) is subsequently adopted to enable the target signal to be sparsely represented and fast reconstructed. By analysing the characteristic order extracted from the reconstructed signal, the fault diagnosis can be completed. The experimental signals validate the effectiveness of the proposed method.