Reweighted Dynamic Mode Decomposition for Fault Feature Extraction of Rolling Element Bearings

Rolling element bearing is an indispensable component of rotating machinery, and many engineering cases have shown that the breakdown of equipment is related to bearing failures. Therefore, the significance of diagnosing faults in bearings cannot be overstated. However, the complex composition of bearing vibration signals makes fault information difficult to obtain directly. Conventional signal decomposition methods are effective in removing noise, but fault impulses are still mixed with harmonic components. Accurately separating fault impulses from signals under the interference of harmonics and noise has become an urgent problem. In this paper, we propose a correlated kurtosis (CK)-based reweighted dynamic mode decomposition (RDMD), which can filter out the fault-related modes by CK and reweight the fault impulses signal by these modes sub-signals. Applying the proposed RDMD and conventional signal decomposition methods to simulated and experimental signals, the results confirm that RDMD is effective in extracting fault impulses in the presence of harmonic and noise.