Application of 'matched' wavelets to identification of metallic transients

Abstract

A class of signals that usually precede failure of machinery are metallic transients due to momentary contact of supposedly noncontacting moving parts. In many cases these transients can be modeled as sums of spectral components. It is possible to construct models of metallic transients by summing together shifted and dilated copies of a single damped sinusoid. Wavelets are inherently suited to the analysis of metallic transient signals because they are based on shifts and dilations of a single function. It is possible to tailor the wavelet used in the wavelet transform to give an optimal response to the damped sinusoids that compose the metallic transient. If these metallic transients are generated by cyclostationary processes, it is shown that moments of the wavelet transforms of these signals can be used to study correlation between the individual components of the signal in both time and frequency.<>