Acoustic emission cluster analysis of composite laminates with different interfacial fiber orientations based on gaussian mixture model

Abstract

The present study aimed to find a method with higher efficiency to cluster acoustic emission events. The acoustic emission signals of composite laminates with different interfacial fiber orientations were captured during the Double cantilever beam (DCB) experiments. Through feature selection of the preprocessed data using the Relief F algorithm, it was found that most of the information in the acoustic emission signals can be represented by the amplitude, inverse frequency, center frequency, peak frequency and so on. Subsequently, principal component analysis and expectation-maximization and Gaussian mixture models were adopted to cluster the data with reduced dimensions. This method can not only distinguish among different damage mechanisms but also show the level of damage concentration according to their types. Furthermore, once the clusters were assigned to different damage mechanisms, they could be identified precisely using both the amplitude and peak frequency. Finally, based on the loading curves of the four types of specimens, the cumulative events and energies of the acoustic emission were compared and analyzed, and different damage mechanisms were identified.