A Novel Criterion for Interpreting Acoustic Emission Damage Signals Based on Cluster Onset Distribution

Structural health monitoring (SHM) relies on non-destructive techniques such as acoustic emission (AE) that generate large amounts of data over the lifespan of systems. Clustering methods are used to interpret these data and gain insights into damage progression and mechanisms. Conventional methods for evaluating clustering results utilise clustering validity indices (CVI) that prioritise compact and separable clusters. This paper introduces a novel approach based on the temporal sequence of cluster onsets, indicating the initial appearance of potential damage and allowing for early detection of defect initiation. The proposed CVI is based on the Kullback-Leibler divergence and can incorporate prior information about damage onsets when available. Three experiments on real-world datasets validate the effectiveness of the proposed method. The first benchmark focuses on detecting the loosening of bolted plates under vibration, where the onset-based CVI outperforms the conventional approach in both cluster quality and the accuracy of bolt loosening detection. The results demonstrate not only superior cluster quality but also unmatched precision in identifying cluster onsets, whether during uniform or accelerated damage growth. The two additional applications stem from industrial contexts. The first focuses on micro-drilling of hard materials using electrical discharge machining, demonstrating, for the first time, that the proposed criterion can effectively retrieve electrode progression to the reference depth, thus validating the setting of the machine to ensure structural integrity. The final application involves damage understanding in a composite/metal hybrid joint structure, where the cluster timeline is used to establish a scenario leading to critical failure due to slippage.