Automated Damage Detection in Carbon-Fiber Reinforced Polymers Using Air-Coupled Ultrasound and Neural Networks
e-Journal of Nondestructive Testing · 2026-07 · 31 (7)
DOI: 10.58286/33402 ↗At a glance
Journal article in e-Journal of Nondestructive Testing, 2026-07. Written by Olga Popovych and 3 co-authors.
Abstract (original language)
This study investigates the performance of a ResNet50-based U-Net and a YOLOv5s model for the detection of delamination in carbon fiber reinforced polymers using air-coupled ultrasound data acquired with piezoelectric and laser excitation. Four datasets were employed to analyze the influence of excitation type on convolutional neural network performance with limited training data. The U-Net architecture was optimized using a two-stage training strategy, while YOLOv5s followed a single-stage approach. Model generalization was evaluated across different CFRP types. Computed tomography and lock-in thermography were used for the ground truth. The U-Net model trained on combined piezoelectric and laser-excited air-coupled ultrasound data showed a tendency to overestimate defect areas and exhibited inconsistent generalization behavior, likely caused by extensive data augmentation of laser excitation data. In contrast, the YOLOv5s model trained on piezoelectric excitation data demonstrated stable and consistent overestimation across different CFRP materials, indicating robust generalization capabilities. Increasing the dataset size improved training performance but did not significantly enhance validation results. The findings underline the critical role of dataset composition, excitation strategy, and model optimization in achieving reliable deep-learning-based non-destructive evaluations. The results further support the future integration of defect area prediction into composite repair assessment workflows based on certified repair schemes
Cite
APA
Popovych, O., Lai, K. W., Orellana, A. E. M., & Grosse, C. U. (2026). Automated Damage Detection in Carbon-Fiber Reinforced Polymers Using Air-Coupled Ultrasound and Neural Networks. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/33402
Vancouver
Popovych O, Lai KW, Orellana AEM, Grosse CU. Automated Damage Detection in Carbon-Fiber Reinforced Polymers Using Air-Coupled Ultrasound and Neural Networks. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/33402
BibTeX
@article{Popovych2026,
author = {Popovych, Olga and Lai, Kher Wei and Orellana, Ana E. Menendez and Grosse, Christian U.},
title = {Automated Damage Detection in Carbon-Fiber Reinforced Polymers Using Air-Coupled Ultrasound and Neural Networks},
journal = {e-Journal of Nondestructive Testing},
year = {2026},
volume = {31},
number = {7},
doi = {10.58286/33402}
}
EUVIMED