Ultrasonic stochastic localization of hidden defects in composite materials

The paper presents a non-deterministic technique to localize hidden defects in carbon fiber reinforced plastics, using multimodal probability beliefs developed from ultrasonic signals. This technique estimates the location of hidden defects, utilizing repetitive amplitudes in multiple Ascans collected for a common target. The multimodal beliefs represent probability density estimates of amplitudes in an A-scan, depicting multiple defects. Location of defects is magnified using Bayesian theorem, while reducing uncertainty. The proposed technique proves to be significantly useful for localizing hidden defects in one dimension, compared to conventional techniques. Multiple carbon fiber composite specimens varying in thickness, numbers of plies and lamina layout are inspected. Ultrasonic A-scans, collected from specimens, are processed using the proposed technique, validating its robustness and accuracy. The results of detection and localization of delamination and low profile defects, such as porosity, are presented in this paper.