Damage Identification for Guided Wave Testing of Composite Structures Using Statistical Features

Abstract

Guided wave structural health monitoring (GW-SHM) is essential for detecting damage in composite materials. Conventional damage identification approaches require knowledge of material properties to calculated deviation of monitoring signal from baseline and are limited to specific damage types or materials. Deep learning has emerged as a more automated method, but it requires high computational power. To address this, we propose using two features: correlation coefficient deviation (CCD) and root mean squared deviation (RMSD). CCD captures the changes in phase of monitoring signal due to presence of the damage. RMSD on the other hand is sensitive to changes in the amplitude. When combined with a binary random forest classifier, these features achieve performance comparable to deep learning. We tested our algorithm on two datasets with different damage types, recording accuracy of 94.4% for Open Guided Waves (OGW) dataset and 99.2% for NASA Prognostic Center of Excellence-Guided Waves (PCoE) dataset. These lightweight models are suitable for in-situ monitoring, offering practical application for damage identification.