A data-driven acoustic emission framework for fracture mode identification in adhesive bonded joints under mode I fatigue loading

Adhesively bonded joints are extensively utilised in multi-material lightweight structures, including those with high load-bearing demands, such as wind turbine blades, and in the automotive and naval sectors. However, for certification in critical safety applications, adhesively bonded joints cannot be relied on alone and require backup measures, such as rivets and bolts, to mitigate the risk of sudden failure. A typical example is the secondary bonded parts in aircraft. To increase the reliability of adhesively bonded joints, it is crucial to have monitoring methods that accurately identify their damage evolution and different fracture modes, even under fatigue loading. This study investigates acoustic emission signatures associated with fracture behaviour in metallic adhesively bonded joints under Mode I fatigue loading. A data-driven framework combining unsupervised clustering and statistical analysis is applied to identify distinct AE signals and correlate them with cohesive and adhesive (interfacial) failures. The acquired datasets were analysed using a workflow that combines unsupervised artificial neural networks for clustering, followed by waveform inspection, and statistical analyses including distribution fitting, bootstrapping, and confidence intervals. After post-processing, the results show statistically consistent differences in AE features, particularly amplitude, duration, and energy, associated with different fracture outcomes. Importantly, interfacial (adhesive) failure is found to generate lower-amplitude signals than cohesive failure, with direct implications for AE monitoring under fatigue conditions. These findings highlight the importance of appropriate threshold selection, as high acquisition thresholds may prevent the detection of critical interfacial failure events.