Fault-related impulsive component detection for vibration-based diagnostics in the presence of random impulsive noise

Abstract

Abstract In this paper, the authors propose the methodology for vibration-based diagnostics towards local damage detection in rolling element bearings in the presence of non-Gaussian noise. In real-life cases, the main problem making the analysis difficult is the non-Gaussianity of the high-energy noise present in the operational environment. Because of this fact, popular impulsiveness-related detection techniques cannot be used. In the presented article, a real-life data measured in an industrial scenario will be presented and a proposition of an approach to cyclic component extraction will be discussed. The proposed approach takes advantage of the Cyclic Spectral Coherence (CSC) map as multidimensional data representation. It can be very useful for indicating cyclic modulated components in the otherwise non-cyclic signal content. However, due to the limitations of statistics used in CSC map calculation impacting the quality of CSC map in the presence of non-cyclic impulsive behavior in the signal, Nonnegative Matrix Factorization idea is used as a method for component separation. The presented method allows for obtaining carrier-related and modulation-related features of the component of interest. The main advantage of the presented method is pairing very useful multidimensional data representation (CSC) with very potent decomposition technique (NMF), that allows to reconstruct separated components in the final stage of the analysis. As a consequence, the time series of the component can be reconstructed based on the carrier feature.