Fast Spectral Correlation Detector for Periodic Impulse Extraction of Rotating Machinery

Abstract

Periodic impulse components are a typical symptom of rotating machinery failure, which are often masked by heavy background noise. It is of great practical significance to research how to obtain periodic impulse components to achieve fault diagnosis of rotating machinery. Fast spectrum correlation (Fast-SC), as a typical non-stationary and nonlinear signal processing method, has been studied in feature extraction by suppressing background noise to enhance periodic impulse components. However, effectively determining the bandwidth of Fast-SC is still a challenging issue. To address this issue, a novel method called Fast-SC detector is developed, which decomposes the measured signal of rotating machinery into multiple Fast-SC slices with different frequency bands. The whale optimization algorithm (WOA) is utilized to optimize the number of Fast-SC slices with minimum mean entropy as the optimal Fast-SC bandwidth. In addition, an adaptive combination morphological filter (ACMF) is applied to suppress the residual noise and narrowband impulses in the WOA-based Fast-SC slices to enhance the fault features. Finally, the Fast-SC detector is constructed using the average denoising Fast-SC slices to infer the type of rotating machinery. The proposed Fast-SC detector is compared with existing methods by applying simulation signal model and experimental data. The results prove that Fast-SC detector is superior to some excellent periodic impulse extraction algorithms in extracting the symptoms of rotating machinery failure.