Decentralized AI-driven NDT signal interpretation for cost-efficient vibration and crack monitoring in civil structures

Ensuring structural integrity in civil infrastructures demands real-time and accurate monitoring techniques, especially in aging bridges and buildings where conventional inspections fall short in frequency and coverage. This study proposes a decentralized, edge-enabled AI-based framework employing a novel 2025 algorithm, Sparse Temporal-Spatial Attention Network (STSAN), for interpreting vibration and acoustic signals from Non-Destructive Testing (NDT) sensors. The system enables on-site signal pre-processing and intelligent anomaly detection, minimizing reliance on cloud-based computation and enhancing scalability. Experimental validation was conducted on datasets obtained from bridge girders and concrete slab testbeds under controlled damage conditions. STSAN outperformed traditional CNN-RNN models and recent transformers in terms of detection accuracy (up to 96.3%) and latency reduction (by 41%) during inference. The proposed method demonstrated superior performance in early crack detection and vibration pattern differentiation, establishing it as an accurate and efficient solution for scalable, cost-effective structural health monitoring.