Cluster analysis on damage pattern recognition in carbon/epoxy composites using acoustic emission wavelet packet

Abstract

The purpose of this paper is to investigate the damage pattern recognition in carbon/epoxy composite laminates by Acoustic Emission technique. Three specimens with layup of [90/0/90/0]s were fabricated and subjected to three-point bending test. The dataset was clustered and analyzed by three unsupervised algorithms, namely, k-means, Self-Organizing Map, and Fuzzy C-Means. The results show that the signals can be divided into four clusters, which correspond to matrix cracking, delamination, fiber/matrix debonding, and fiber breakage, respectively. A method of energy feature extraction based on wavelet packet decomposition was used to analyze the clustering center to obtain the frequency band with relatively high energy corresponding each damage mode further determined the actual damage pattern of these signals with disputed classification results. The results of k-means illustrated that little difference in the determination of matrix cracking, preferring to assign more signal to inter-laminar damage and higher peak frequency to fiber breakage compared to the other two algorithms. It can verify the clustering results well and k-means algorithm performs better in damage pattern recognition of composites.