Local damage detection based on vibration data analysis in the presence of Gaussian and heavy-tailed impulsive noise

Abstract

Local damage detection in bearings focuses on the identification of periodically impulsive components. Popular methods assume presence of either non-Gaussian noise or different frequency band for informative and non-informative impulses, and use statistics to select appropriate band. Here two impulsive sources occupy the same frequency range: a fault-related signal of interest, and non-cyclic noise describing random events during particular technological process (crushing, sieving etc.). The task is formulated as damage detection in presence of non-Gaussian impulsive noise. We propose to use Nonnegative Matrix Factorization of spectrogram for separation of cyclic and non-cyclic impulsive components. Partial information is fused into a single data set for each component. Finally, post-processing is implemented to allow to recover the time series of each component. New method allows to detect and extract impulsive signal (damage in bearing) in presence high amplitude non-cyclic impulsive signal. Moreover, the algorithm allows to properly indicate lack of any damage.