Damage mode identification in carbon/epoxy composite via machine learning and acoustic emission

Abstract Combining acoustic emission (AE) and machine learning algorithms to understand damage and failure of carbon fiber reinforced polymer (CFRP) under bending loads is a challenging task in their practical applications. This work aims to identify different characteristics of the signal induced by damage patterns of carbon/epoxy composites using machine learning model. By carrying out wavelet packet analysis of the AE signals allowing the identification of four macroscopic damage modes. Four types of signals corresponding to matrix cracking, delamination, fiber/matrix debonding and fiber breakage are characterized by the frequency bands of the main energy distribution in the original waveform. The frequency bands are (62.5–125 kHz) or (125–187.5 kHz), (187.5–250 kHz), (250–312.5 kHz) and (312.5–375 kHz) and above, respectively. To test the actual performance of the established model, called support vector machine (SVM) classifier, several precracked and untreated specimens have been fabricated and subjected to three‐point bending test. The classification result of constructed classifier was compared with the k ‐means algorithm, which is widely accepted for classifying AE signals, and the similarity of the two results is analyzed. The results indicated that the similarity of different clusters exceeded 92%, 84%, 83% and 77%, respectively. It can be seen that the SVM classifier was considered promising to provide new ideas for the health monitoring of composite structures.