Abstract
In this academic investigation, we employed Acoustic Emission (AE) monitoring and a Convolutional Neural Network (CNN) to scrutinize hydrogen-induced crack behavior in hydrogen-precharged 304 austenitic stainless steel during tensile stress. This study’s pivotal findings reveal that AE monitoring adeptly captures sound wave signal alterations induced by material stress, especially during the critical phase of crack initiation, from the late stage of strengthening to the necking stage. Additionally, industrial-grade Computed Tomography (CT) scans corroborated the presence of a singular principal crack during these phases, aligning with the crack type identified through unsupervised clustering analysis of Short-Time Fourier Transform (STFT)-processed AE signals using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm. The developed CNN model, demonstrating a 98.32% accuracy rate in validation, effectively discriminated between the signals corresponding to distinct stages of material damage. This research underlines the efficacy and practicality of integrating AE monitoring with deep learning for hydrogen induced damage detection in materials.
