Threshold-triggered, fixed-window acoustic-emission (AE) acquisition remains the pragmatic choice for monitoring carbon-fibre-reinforced polymers (CFRPs), yet overlapping transients within a window can corrupt features and labels. An overlap-aware pipeline is presented that preserves the deployability of hit-based acquisition while mitigating overlap-induced errors. Hits are clustered in a low-dimensional principal-component (PCA) embedding; high-confidence cores are then obtained in the PC1–PC2 plane via the intersection of a Mahalanobis-distance ellipse and a kernel-density high-probability region. Projecting these cores onto the peak-frequency–amplitude map reveals distinct groups from which data-induced band limits are read directly (by peak-frequency bounds), avoiding continuous streaming and manual band tuning. A compact representation augments standard AE descriptors with band energies and information-theoretic entropies (wavelet, spectral, permutation, sample), and a gradient-boosted decision-tree classifier is trained on this overlap-aware feature set. Across multiple CFRP damage mechanisms and loading protocols, the approach yields consistent improvements over conventional fixed-window baselines, while sensitivity analyses over gating parameters show stable band limits and low performance variance. Computation and storage scale with the hit rate rather than mission time, enabling long-duration structural-health-monitoring deployments.

acoustic emission; carbon-fibre-reinforced polymer (CFRP); structural health monitoring (SHM); overlap-aware hit processing;