A Robust-to-parameterization Clustering Fusion Approach for Monitoring Damage Onsets and Growth using Acoustic Emission

Abstract

A methodology is presented for acoustic emission (AE) data processing and interpretation, well suited for material characterization of laboratory specimens and in-service structural health monitoring. The methodology relies on the AE streaming in which relevant transients are detected using wavelet-based wave picking, feature vector cleansing with a Mahalanobis-based procedure and a new approach for pattern recognition relying on clustering fusion. The proposed clustering fusion method emphasizes damage kinetics related to the continuous-time definition of AE signals, quantifies uncertainty on clusters (AE sources) evolution, and evaluates the robustness of the results with respect to the change in the parameterization using mutual information. Illustrations concern the early detection and monitoring damage onsets and evolution in a thermoplastic thermostable composite tubular bandage for high-speed rotating engine.