Adaptive maximum second-order cyclostationarity blind deconvolution and its application for locomotive bearing fault diagnosis

Abstract

Abstract Maximum second-order cyclostationarity blind deconvolution (CYCBD) outperforms other deconvolution methods in retrieving the weak periodic impulses related to bearing incipient faults. However, the main challenge in the practical application of CYCBD is how to set several key parameters appropriately, the uppermost of which is the targeted cyclic frequency or fault period. It may attribute to the fact that the advantage of CYCBD is greatly compromised by the provided period. To overcome the above limitations, an adaptive CYCBD (ACYCBD) is presented in this article. In the proposed method, a powerful tool, envelope harmonic product spectrum (EHPS), is tailored to estimate the true cyclic frequency or period precisely. Then, the estimated result instead of a coarsely provided value is regarded as the targeted cyclic frequency. Furthermore, in the presence of heavy energy harmonics or strong external noise, EHPS still has strong robustness in the fault period identification. Compared with the original CYCBD, ACYCBD can extract the weak impulses submerged in the raw vibration signal without any prior information about the period. Finally, the effectiveness and advantages of ACYCBD are revealed by employing it on the synthesized signals and experimental data collected from a locomotive bearing test rig.