A Deep Learning Approach for Single-sensor Acoustic Emission Source Localization in Plate-like Structures

Abstract

This paper presents a deep-learning approach to localize acoustic emission (AE) sources in plate-like structures using a single sensor. This approach implicitly uses the edge reflections of Lamb waves and their dispersive characteristics to localize their sources. Specifically, the deep learning model directly learns such reflection and dispersive characteristics from the continues wavelet transform of the AE signals. Therefore, there is no need to use the traditional AE features, such as arrival time, rise time, and amplitude. In this paper, the deep learning architecture consists of a stack of two autoencoder layers and a softmax layer. To train and test the performance of the proposed deep learning approach, standard pencil lead break tests were performed next to the rivet connection of an aluminum plate with its stiffener. The result shows that this approach can accurately identify the rivet at which simulated fatigue cracks (with pencil lead break tests) generate AE waves