Source reconstruction for acoustic emission signals clustering and events nature identification. Application to a composite pipe bending test

Abstract

This article deals with the application of clustering to acoustic emission data. Although many results are available in the literature for small scale applications showing clustering linking the clusters to the physical nature of the events, the performances are in general limited in large structures due to the effect of propagation on the characteristics of acquired signals. We propose here a methodology coupling source reconstruction, compensating the propagation, and clustering to recover the physical nature of the events. The methodology is exemplified on data from a bending test of a composite pipe, enabling the identification of matrix cracking, fiber debonding and fiber breakage where clustering directly on the data leads to data clustered based on the source–sensor distance.