Multi-source signal analysis in thin-walled spatial structural health monitoring: an overview of deep learning-based approaches

Spatial structures exhibit complex load-transfer mechanisms, strong spatial coupling, and high sensitivity to environmental actions, which impose increasing demands on spatial structural health monitoring (SSHM). Advances in sensing technologies have enabled SSHM systems to acquire substantial data. Based on structural characteristics and monitoring requirements, this review defines multi-source data as complementary information differing in source, location, or temporal scale, and summarizes data foundations from field monitoring, numerical simulation, and experiments. Following the information flow from acquisition and quality assurance to condition estimation, deep learning (DL) methods are reviewed for compressive sensing, missing data recovery, response prediction, anomaly and damage detection. Analysis indicates that DL has potential for nonlinear feature extraction and spatiotemporal correlation modeling. However, existing studies remain dominated by specific applications and homogeneous data. Future directions involving heterogeneous data fusion, graph neural network (GNN), physics-informed neural network (PINN), and performance assessment are discussed. This review establishes a unified framework of SSHM, supporting the transition of DL from algorithmic validation toward practical applications.