Robust Localization and Classification of Barely Visible Indentations in Composite Structures by Fusion of Ultrasonic Damage Indices

Abstract

This study aims to detect, localize, and assess the severity of barely visible indentation damage in a composite sandwich structure using ultrasonic guided waves. A quasistatic loading was gradually applied on a specimen of carbon fiber reinforced epoxy resulting in dents on the surface. Lamb-wave measurements, from a sensor network mounted on the panel's surface, were taken for the intact condition and three damage cases (0.2, 0.5, and 2.7-mm dents). Three approaches were adopted to define the damage indices (DIs) toward anomaly detection, namely, amplitude variation, symbolic dynamics, and root mean square deviation. Data fusion was performed between measurements from multiple excitation frequencies for single and multiple DIs, where the anomaly combination between all the frequencies and the DIs was called a total anomaly. An imaging algorithm was implemented for damage localization in conjunction with single and combined DIs. It was shown that combining the effects of different frequencies and/or different DIs increases the robustness and consistency of the damage detection and localization process. Moreover, a distance-based classification technique was applied using features from single DIs and the combined anomaly measure. Accuracies higher than 91% were attained for the majority of the cases tested.