Extraction of Damage-Sensitive Signal Features in Damaged Carbon/Epoxy Composite Structures Using Statistical Pattern Recognition

Abstract

A statistical pattern recognition-based methodology for determining damagesensitive signal features and corresponding damage extent from nondestructive ultrasonic guided wave (UGW) testing of carbon/epoxy composite laminates has been established. Changes in signal features can be a function of several parameters, which can include operational and environmental variability in the experiment. Similarly, damage-sensitive features are functions of several random variables, and selecting optimal features are often subject to human bias. Multivariate statistical analysis (MSA) was utilized to determine a high variance in jointly distributed random variables, experiment parameters, and signal features to systematically distinguish damage-sensitive signal features. In Data Sets 1 and 2, the variances were examined with respect to experiment parameters in composite laminates with a range of damage levels. In Data Set 3, the variances of several signal features were examined with respect to changing notch sizes over a sweep of frequencies. The relative damage extent was distinguished in Data Sets 1 and 2 using multivariate analysis and damage-sensitive signal features were distinguished in Data Set 3.