@

초록

Structural Health Monitoring (SHM) of laminated composites using Lamb Wave (LW) signals has proven to be an effective method for early damage detection and assessing structural integrity. However, its effectiveness is often compromised by complex multimode interactions and noisy environments. To address these challenges, this study introduces a comprehensive deep learning framework for damage classification and spatial localization using attention-driven architectures. Initially, a baseline Multi-head Attention Transformer (MAT) was developed to classify and localize multi-class damage from LW signals. To enhance robustness, a hybrid model combining the Multi-head Attention Transformer with a Convolutional Neural Network (C–MAT) was introduced for damage quantification in laminated composites. This hybrid model exhibited exceptional performance, achieving 97.66% classification accuracy and an R2 value of 97.08%, indicating improved localization precision. A noise-robustness study was conducted using Additive White Gaussian Noise (AWGN) and Laplace noise across various Signal-to-Noise Ratio (SNR) levels (10–25 dB) to simulate realistic operational environments and assess the stability of the model under degraded conditions. Under noisy environments, the model consistently maintained over 92% accuracy for both damage detection and localization. Even under severe noise conditions at 10 dB SNR, the proposed model maintained strong robustness, achieving classification accuracies of 92.19% and 95.31% under AWGN and Laplace noise, respectively, with corresponding localization performance of R2 = 92.15% and 92.58%. © 2026 Elsevier Ltd.

키워드