Wavelet Based Feature Extraction for Acoustic Emission

Abstract

A new feature extraction method for acoustic emission signals is presented. The method works in the time?frequency domain, as compared to the time domain from which features have previously been taken. The method employs a wavelet transform followed by a simple optimisation of a series of 2D Gaussian functions via simulated annealing. The solution to the optimisation is taken as a feature set to replace or embellish the usual data set ? rise time, energy, duration etc. The aim is to produce a feature set which heeds the modal nature of AE signals. The method is applied to a set of data from a box girder fatigue test, detailed by Manson et al (2001). The time domain feature extraction is compared with the time?frequency feature extraction, and the results are presented for comparison