Issue 78

D. S. Lobanov et alii, Fracture and Structural Integrity, 78 (2026) 207-219; DOI: 10.3221/IGF-ESIS.78.12

the fiber orientation. In contrast, for samples with a ‘scratch’ defect, the dominant clusters differ. It can be assumed that this is due to the fibers being cut by the blade when the defect was applied to samples with fibers laid in the warp direction. Certain characteristics can be identified in the clusters obtained based on the characteristics of each parameter. Cluster 1 is characterized by signals with low amplitude, short duration and short rise time. The signals in Cluster 2 have average values for each parameter. Cluster 3 comprises signals with high amplitude, high duration values and average rise times (but, unlike Cluster 2, the variation in values is less significant). In Cluster 4, high rise times and average values for the other two parameters can be identified. For clarity, Tab. 3 presents the characteristics of the signal parameters. Presumably, the signals in cluster 1 indicate failure in the matrix, those in clusters 2 and 3 indicate various stages of delamination, and those in cluster 4 indicate friction between the fibers and their slippage out of the matrix.

№ Сluster

Amplitude

Duration

Rise time

Prevalence

1 2 3 4

Low (145– 300 μV)

Short (10–115 µs) Medium (63–191 µs) Long (128–176 µs) Medium (54–78 µs)

Low (2–34 µs)

Dominant (47–67%) Moderate (7–14%)

Medium (400– 1100 μV) High (800– 2240 μV) Medium (250– 360 μV)

Medium (7–24 µs) Medium (12–18 µs)

Low (7–10%)

High (20–28 µs)

Moderate (8–32%)

Table 3: Characteristics of the AE signal parameters obtained during the clustering process. Subsequently, a cluster analysis was performed on the peak frequencies of the signals to validate the results obtained. The k-means++ method was also used to determine the initial centroids. Points as far apart as possible were selected to ensure faster convergence and a higher-quality initial partition. Since the peak frequency partition was used to validate previously obtained data, the array of points was divided into the same number of clusters, K=4. Fig. 13 presents the clustering results in relative units. Tab. 4 also provides a summary of the clustering results. As the parameters were similar for a clearer comparison, the clustering of signals differs for samples cut along the warp and weft. However, the analysis allowed for the identification of more precise peak frequency ranges. Cluster 1 corresponds to the low-frequency range (57–84 kHz), indicating matrix degradation within the fiber. An intermediate range (101–276 kHz) has been identified, the signals within which presumably indicate the onset of delamination, and a mid-frequency range (277–432 kHz), corresponding to a disruption in fiber adhesion and friction between fibers. In samples cut along the warp, cluster 4 is also identified – signals with high peak frequencies (over 450 kHz); however, their percentage does not exceed 2%.

Figure 13: Signals grouped by cluster for different sample types under the ‘peak frequency’ parameter.

216