Progressive Failure Analysis of Quasi-isotropic Self-reinforced Polyethylene Composites by Comparing Unsupervised and Supervised Classifications of Acoustic Emission Data

Unsupervised and supervised pattern recognition（ PR）techniques are used to classify the acoustic emission（ AE） data originating from the quasi-isotropic self-reinforced polyethylene composites,in order to identify the various mechanisms in the multiangle-ply thermoplastic composites. Ultra-high molecular weight polyethylene / low density polyethylene（ UHMWPE / LDPE）composites were made and tested under quasi-static tensile load. The failure process was monitored by the AE technique. The collected AE signals were classified by unsupervised and supervised PR techniques, respectively. AE signals were clustered with unsupervised PR scheme automatically and mathematically. While in the supervised PR scheme,the labeled AE data from simple lay-up UHMWPE / LDPE laminates were utilized as the reference data.Comparison was drawn according to the analytical results. Fracture surfaces of the UHMWPE / LDPE specimens were observed by a scanning electron microscope（ SEM） for some physical support. By combining both classification results with the observation results,correlations were established between the AE signal classes and their originating damage modes. The comparison between the two classifying schemes showed a good agreement in the main damage modes and their failure process. It indicates both PR techniques are powerful for the complicated thermoplastic composites. Supervised PR scheme can lead to a more precise classification in that a suitable reference data set is input.