Wood damage identification and crack-mode recognition based on acoustic emission multi-parameter fusion and hierarchical clustering

Accurate acoustic emission (AE)-based identification of damage scales and crack modes in wood can be hindered by overlap between microscopic damage events and macroscopic crack-related signals. This study developed and preliminarily evaluated a cascaded hierarchical framework that separates damage-scale identification from crack-mode recognition. AE signals from three-point bending tests of Zelkova schneideriana were characterized using energy, mean value, centroid frequency, and peak frequency. After principal component analysis, K-means++ was used to distinguish microscopic damage events from macroscopic crack-propagation events, and RA–AF analysis was then applied only to the macroscopic-event subset to distinguish tensile-dominated, shear-dominated, and mixed-mode characteristics. The identified characteristics were generally consistent with synchronously observed surface crack morphologies. Validation using ZS2 tested at a different loading rate and ZS3 containing a pre-existing crack showed similar damage and crack-mode characteristics. These findings indicate that separating damage scales before RA–AF-based crack-mode recognition provides a more interpretable approach for characterizing fracture-related AE responses in wood and suggests preliminary cross-condition transferability, although broader validation is still required.