DELAMINATION CLASSIFICATION IN COMPOSITE PLATES USING GUIDED WAVES AND 1D-CNN

This study assesses the effectiveness of Ultrasonic Guided Waves (UGWs) combined with a 1D Convolutional Neural Network (1D-CNN) for detecting and classifying delamination in carbon fiber reinforced polymer (CFRP) laminates. A dataset of 912000 waveforms from undamaged and damaged CFRP plates was analyzed using a shallow 1D-CNN with two convolutional blocks and a fully connected layer, designed for computational efficiency and robust generalization. The model processed data with a 70-30-10 split for training, validation, and testing, achieving an average accuracy of 90.27%. The analysis showed high recall and precision rates - over 90% in about 75% of classes and over 95% in about 40% of the classes. These metrics were particularly significant when the waveforms propagated parallel to the fiber direction: averaging Precision, Recall, and F1-Score at 96.2%. They highlighted the effectiveness of the model over a range of delamination depths and orientations. However, performance differences were observed in cases with deeper delamination and perpendicular wave propagation due to suboptimal wave excitation. These results highlight the potential of 1D-CNNs to effectively classify complex UGW signals for real-time health monitoring of structural composite components.