Interpretable AI for Guided-Wave SHM of Anisotropic Composites: Toward Trustworthy Damage Diagnosis

The use of Artificial Intelligence (AI) in Structural Health Monitoring (SHM) offers powerful capabilities for damage detection, localization, and quantification in complex composite structures. However, many data-driven models act as black boxes, providing limited interpretability and weak connection to the underlying wave physics. This work investigates a physics-guided AI framework based on dispersion-aware sparse decomposition and tensor analysis for guidedwave–based SHM. The approach combines compact, physically meaningful features—linked to propagation distance, attenuation, and dispersion—with machine learning to achieve efficient and interpretable decision-making. A comparative study is conducted against conventional data-driven approaches on a representative composite structure under various damage scenarios. Results show that while purely data driven models can achieve high detection accuracy, the proposed hybrid framework reaches acceptable performances with enhanced robustness and explainability. The physically grounded representation potentially allows direct insight into the damage mechanisms, enabling reliable monitoring and facilitating the integration of AI within maintenance and certification processes. This interpretable hybrid approach thus provides a promising route toward trustworthy, physics-informed SHM of next-generation aerostructures.