Automatic Classification of Hydrogen-Induced Acoustic Emission Signals in High-Strength Offshore Bolts Using Time–Frequency Feature Engineering

Abstract

Hydrogen embrittlement (HE) is a critical degradation mechanism in high-strength offshore fasteners, where early-stage hydrogen-induced cracking (HIC) is difficult to detect using conventional inspection methods, due to its subsurface and time-dependent nature. This study presents an automatic acoustic emission (AE) signal classification framework for identifying hydrogen-related damage mechanisms in high-strength offshore bolts subjected to in situ electrochemical hydrogen charging under cyclic loading. Fatigue experiments were performed on modified property class 10.9 steel bolts using a bespoke axial fatigue rig integrated with localized hydrogen charging and multi-channel AE monitoring. Baseline fatigue experiments performed under uncharged conditions were additionally used to compare hydrogen-assisted and non-hydrogen-assisted AE activity. AE data was analysed using a structured framework incorporating signal filtering, feature extraction, principal component analysis (PCA), and Gaussian mixture model (GMM) clustering. To improve signal discrimination, spectral and temporal energy-distribution features, supported by continuous wavelet transform analysis, including partial-power and energy-ratio parameters, were introduced. The proposed framework enabled separation of AE signals associated with hydrogen evolution, plastic deformation, hydrogen-induced cracking, and brittle fracture. Comparison with manually classified datasets demonstrated strong agreement between automatic and physically interpreted signal clusters, while scanning electron microscopy (SEM) supported the presence of hydrogen-assisted brittle-fracture features associated with HIC-related AE activity. The introduction of spectral and temporal energy-distribution features improved cluster separability under in situ hydrogen-charged conditions. The results demonstrate that physically informed feature engineering combined with automatic clustering provides a promising proof-of-concept approach for mechanism-informed identification of hydrogen-assisted AE activity in high-strength steel fasteners.