Alpha-stable distribution based methods in the analysis of the crusher vibration signals for fault detection

Abstract

In the field of the industrial signal processing it is important to provide appropriate models of analyzed data. Signals acquired on the faulty components often possess impulsive behavior resulting in nonexistence of moments, heavy tails and as such they cannot be analyzed with application of the standard methods related to Gaussian models. In past years varying methods allowing for the analysis of the impulsive data have evolved. We recall few of them such as α-stable distribution modeling, dependency measures connected with this distribution, namely codifference, covariation, fractional lower order covariance (FLOC). These measures when applied can detect frequency bands in which the information about the fault is present (informative frequency band) and also identify/localize the fault. Application of such measures, especially FLOC towards spectrogram will result in creation of the lag-frequency dependency map from which one can extract mentioned information. Authors provide in this paper a novel procedure for local damage detection combining local maxima enhancement with dependency analysis in the frequency bands using FLOC.