Double autocorrelation-based cyclicity evaluation for repetitive transients feature extraction

Abstract

Abstract The vibration response caused by bearing local defects has impact and periodicity in waveform, which provides a standard for the frequency band selection in envelope analysis. However, most periodicity measurements without prior knowledge belong to sparsity evaluation, while the defects of sparsity index in nature are inevitable. Inspired by the periodic component extraction of autocorrelation function, a novel cyclicity measurement based on double autocorrelation calculation is proposed. With the help of normalization, this approach can distinguish periodic impulses from random impulses by using the periodic sub-maxima of the envelope autocorrelation. Considering the influence of the noise component on the autocorrelation of the periodic signal, the sub-maximums are maintained by threshold processing. On this basis, the re-autocorrelation is calculated to identify the periodic sub-maximum. Finally, as a non-prior index, a demodulation band selection is also proposed in combination with an impulsivity evaluation. The results of the proposed method are analyzed and verified by comparison with typical methods.