Vibration-based identification of lubrication starved bearing using spectral centroid indicator combined with minimum entropy deconvolution

For bearings under starved lubrication conditions, direct contacts occur between metallic surfaces, leading to random impulses in vibration response. This paper proposes a vibration-based scheme for starved lubrication identification. Minimum entropy deconvolution (MED) is utilized to enhance the weak random impulses caused by lubricant starvation. Moreover, a starved lubrication feature (SLF) based on the spectral centroid indicator is constructed to diagnose starved lubrication. The efficacy is verified by a comparative study conducted on a bearing test bench. Bearing vibration signals under two lubrication conditions are collected at different rotating speeds. The kurtosis of raw signals under starved lubrication is less than 4, but increases to above 7 after MED. For evaluation of lubrication conditions classification, the Davies-Bouldin index (DBI) of SLF is 0.042, significantly better than that of traditional characteristics (RMS DBI = 0.484, kurtosis DBI = 0.559). The results indicate that the SLF can identify starved lubrication conditions under different speeds.