Intelligent identification and interpretability study of damage in composite laminates based on elastic guided wave and deep learning

This study proposes an integrated intelligent framework that combines the dispersion characteristics of elastic guided waves with deep learning for the automatic identification of crack relative location and severity in composite laminates. The study first establishes a database of dispersion curve images covering various damage scenarios through parametric finite element modeling. It fully uses the inherent physical feature extraction capability of dispersion curves. Subsequently, a dedicated convolutional neural network (CNN) is designed to automatically learn damage-sensitive features from the images. The Grad-CAM method is introduced to visualize key pixel regions, enhancing the interpretability of the model's decisions. The key innovation lies in the physics-guided intelligent diagnosis. On one hand, multiscale mechanical modeling clarifies the physical relationship between damage and wave propagation. On the other hand, the hierarchical feature extraction capability of the CNN overcomes the limitations of traditional empirical criteria. Results show that the model’s macro-average F 1 -scores exceed 97.46%. The confusion matrix heatmap reveals that the model's recognition highest accuracy exceeds 99.4%. In conclusion, this study not only verifies the feasibility of integrating elastic guided waves with deep learning but also elucidates the physical response mechanisms of damage-sensitive modes. It provides a novel, high-precision, and trustworthy approach for the structural health monitoring of composite laminates.