Single-sensor acoustic emission source localization in plate-like structures: a deep learning approach

Abstract

Acoustic emission (AE) source localization in plate-like structures with geometric features, such as stiffeners, usually requires a large number of sensors. Even without any geometric feature, such approaches are usually accurate only within the convex area surrounded by sensors. This paper proposes a deep learning approach that only requires one sensor and can localize acoustic emission sources anywhere within a metallic plate with geometric features. The idea is to leverage the edge reflections of acoustic waves as well as their multimodal and dispersive characteristics. This deep learning approach consists of three autoencoder layers and a regression layer. The input to the first autoencoder layer is the continuous wavelet transform of AE signals and the output of the regression layer is the estimated coordinates of AE sources. To validate the performance of the proposed approach, Hsu-Nielsen pencil lead break tests were performed on an aluminum plate with a stiffener. The results show that the proposed approach has no blind zone and can localize AE sources anywhere on the plate.