MCAGN: Multilayer Coupled Attention Graph Networks for Composites’ Damage Detection

Damage detection in carbon fiber reinforced polymer (CFRP) composites is essential for maintaining structural integrity and preventing catastrophic accidents. Recently, machine learning-based damage detection methods have garnered increasing attention. However, existing approaches often face challenges in extracting relational information about damage, both within guided wave paths and between damage samples. These limitations result in insufficient detection stability under varying training sample conditions and increase the risk of model overfitting when relying solely on high-dimensional information extracted from guided wave signals. To address these challenges, this paper proposes a multi-layer graph network with a coupled attention mechanism based on feature fusion. In the proposed method, by encoding sensor coordinates and fusing guided wave signal features, positional information can be integrated into the guided wave signals. A multi-layer graph network is further employed to extract signal features, while the inter-layer coupled attention mechanism and multi-layer graph construction fully explore the relational information associated with the damage. Experimental datasets collected from real CFRP composites demonstrate the effectiveness of the proposed method. Comparative results show that the multi-layer graph neural network achieves superior stability and accuracy, even with limited training samples and noisy training samples. These findings highlight the robustness and practicality of the proposed approach for damage detection in CFRP composites.