A novel method of improving the readability of the spectral correlation function using the reassignment technology for cyclostationary signals

Recently, cyclostationary analysis has played a prominent role in mechanical fault diagnosis due to the ability to extract the weak feature. It has been proved that plenty of vibration signals from faulty rotating machinery exhibit second-order cyclostationarity and the spectral correlation can achieve the detection of the cyclic frequency associated with the fault. However, the spectral correlation still has issues with the inaccuracy of fault detection because of its poor localization of the signal components caused by the spectral leakage. To address this issue, a modified cyclostationary analysis method named Reassigned Spectral Correlation (RSC) is proposed to improve the readability of the bi-frequency (spectral frequency and cyclic frequency) representation, i.e., spectral correlation. Due to the fact that the integration of the spectral correlation function over the spectral frequency is equivalent to the squared envelope spectrum of the signal, it implies that the cyclic frequency of the spectral correlation essentially corresponds to the frequency of the squared envelope spectrum. Therefore, the RSC method first calculates the reassigned coordinates of the squared envelope spectrum and then assigns the values of spectral correlation to these reassigned coordinates over the cyclic frequency, so as to produce a better localization of the signal components. The localization performance of the proposed RSC method is validated using three typical cyclostationary signal models, and the effect of spectral resolution on the proposed method is investigated. Experimental results from cavitation-induced vibration data of a centrifugal pump and bearing vibration data with a local fault at the inner race demonstrate the effectiveness of the proposed RSC method for diagnosing faults in rotating machinery.