Single Sensor Localization and Characterization of Acoustic Emission Sources in Metallic Panels: A Deep Learning Approach

Abstract

This paper presents data-driven localizing and characterizing of acoustic emission (AE) sources in metallic panels. Deep learning is used to leverage the multimodal and dispersive reverberations of such sources in real-world panels with geometric features, such as rivet connections, doublers, and stringers. The deep networks used in this study are stacked autoencoders trained on the wavelet transform and frequency spectrum of AE waveforms. To evaluate the performance of this deep learning-based approach, Hsu-Nielsen pencil lead break sources were generated on a real-world Boeing-777 fuselage panel. The results show that the proposed approach can localize and characterize AE sources with only one sensor placed in the far field.