A new two-stage strategy to adaptively design and finely tune the filters for bearing fault-related mode decomposition

Abstract

Fault-excited repetitive transients are apt to be submerged in unavoidable noise. Till now, most vibration-based diagnostic methods just reject the noise outside the optimal frequency band without suppressing the noise in the candidate band, which could not extract the diagnostic information with a high signal-to-noise ratio (SNR). Motivated by this, a new two-stage strategy is proposed to reduce the out-band strong noise and further eliminate the in-band interference in different stages. In the first stage, a dichotomy-based AIC (d-AIC) picker is designed to determine the spectral segmentation points. The obtained frequency boundaries are utilized to design the filter banks so that the signal components outside the passband of each sub-band signal are completely rejected. Subsequently, in the second stage, the filter characteristics are finely tuned to further enrich the fault-related signatures and suppress the in-band interferences of these sub-band signals. Based on the distribution characteristics embedded in the whole spectral series, the designed d-AIC shows superiority in the spectrum-driven frequency division. Moreover, the proposed two-stage strategy can provide health information with a high SNR about the monitoring bearing system. The simulated and experimental scenarios give credible evidence about the advantages of the proposed strategy, which means the proposed scheme is powerful to realize a more comprehensive condition monitoring in practical application.