An unsupervised pattern recognition approach for AE data originating from fatigue tests on polymer-composite materials

This work investigates acoustic emission generated during tension fatigue tests carried out on a carbon fiber reinforced polymer (CFRP) composite specimen. Since massive fatigue data processing, especially noise reduction, remains an important challenge in AE data analysis, a Mahalanobis distance-based noise modeling has been proposed in the present work to tackle this problem. A sequential feature selection based on Davies-Bouldin index has been implemented for fast dimensionality reduction. An unsupervised classifier offline-learned from quasi-static data is then used to classify the data to different AE sources with the possibility to dynamically accommodate with unseen ones. With an efficient proposed noise removal and automatic separation of AE events, this pattern discovery procedure provides an insight into fatigue damage development in composites in the presence of millions of AE events.