Computer Vision-Based Deep Learning Approach for Automated Delamination Detection and Classification in Carbon Fiber-Reinforced Polymer Composites

To ensure the structural reliability and safety of carbon fiber-reinforced polymer (CFRP) materials, the detection and classification of delamination are crucial. Therefore, this study proposes a novel approach for classifying delamination areas in CFRP samples based on computer vision-based deep learning algorithms. The proposed approach utilized signals acquired by piezoelectric transducers fixed in the inspected structure using the Lamb wave transmission-reception method. The methodology involved detecting delamination induced by fatigue loading in CFRP samples and classifying it through segmented regions. The dataset was divided into two classes based on damage levels: D1 (up to 10 cm²) and D2 (above 10 cm²). Time-frequency representations were generated the class imbalance was addressed with random oversampling. The EfficientNet architecture was employed for classifier development, achieving an accuracy of 83.33% and an area under curve (AUC) of 0.94, demonstrating feasibility in CFRP damage classification.