Acoustic emission signal clustering in CFRP laminates using a new feature set based on waveform analysis and information entropy analysis

摘要

Acoustic emission (AE) for structural health monitoring of fiber-reinforced polymers (FRP) has been under extensive study in past decades. Many of the available methods rely on using the conventional features of AE signals for clustering. These clusters are then related to a specific failure mechanism based on their behaviors. The conventional AE features are derived from a portion of the signal waveform that passes through a predetermined threshold and is heavily affected by attenuation. Therefore, recent studies on AE are more focused on waveform analysis. A key parameter when analyzing the waveform distribution is the selection of appropriate bin width. This study discusses the choice of an optimum bin width for waveform analysis. This parameter is then shown to be well correlated to the conventional threshold-dependent features of AE signals. The bin width is then used as a time-domain representation of the waveform and is used with the peak frequency for signal clustering. A series of tensile tests are performed on cross-ply and quasi-isotropic open-hole carbon FRP (CFRP) specimens. The results show that AE signal clustering using these features outperform clustering performance using conventional AE features. This approach is shown to divide signals into two clusters; a cluster with matrix dominated signals and a cluster with fiber dominated signals. The information entropy of signals in each cluster is evaluated and compared to the information entropy of noise signals.