Damage Assessment and Classification of Glass Fiber-Reinforced Composite Materials During Tensile Tests Based on Acoustic Emission and Unsupervised Learning Approaches

Abstract

Advancements in composite materials design have rendered glass fiber-reinforced polymer composites (GFRPC) an effective candidate for various engineering and industrial applications. Some attractive characteristics of GFRPC are low specific mass and high specific mechanical stiffness and strength. Identifying and classifying damage mechanisms is difficult for these types of materials. Several scientific works investigated the failure mechanisms of GFRPC considering different approaches, such as non-destructive techniques and various learning algorithms. However, studies to reliably identify mechanical failures in GFRPC are ongoing. Therefore, the present work investigates non-destructive testing (NDT) combined with unsupervised learning algorithms to identify damage mechanisms in GFRPC during tensile tests. The NDT used for this study is the acoustic emission (AE) technique that detects the waveforms of the different failure mechanisms during testing. Therefore, these waveforms are classified based on unsupervised learning algorithms such as principal component analysis (PCA) and self-organizing maps. The PCA method selects the most suitable AE descriptors to identify the different damage mechanisms. Thus, the self-organizing maps algorithm performs the clustering analysis and classification of the failure mechanisms. The obtained information from the AE descriptors is compared to AE data from the technical literature associated with specific damage mechanisms, matrix cracking, matrix/fiber debonding, delamination, and fiber breaking. Electron microscope images of the failures are presented to validate the results.