Graph Neural Networks in Structural Health Monitoring: A Scoping Review

Interest in graph neural networks (GNNs) has surged in recent years due to their effectiveness across diverse domains. GNNs encode information about a system’s underlying mechanics as nodes and edges, providing insights that traditional artificial neural networks, which tend to behave as black boxes, cannot provide. Although GNNs excel with graph-structured data, such as molecules and social networks, their applicability extends to various data types, including images. Yet, a thorough and up-to-date review of their application and potential in the field of structural health monitoring (SHM) in civil infrastructure is currently lacking. This scoping review fills that gap by mapping existing research, highlighting advantages, open challenges, and future directions. Following the Preferred Reporting Items for Systematic reviews and Meta-Analysis extension for Scoping Review (PRISMA-ScR) protocol, 46 journal articles and two conference papers published up through October 2024 were retrieved from Web of Science and systematically analyzed. Discussing core concepts regarding various GNN types and categorizing their applications based on machine learning (ML) and SHM criteria highlights each approach’s benefits and limitations. A systematic review of methodologies and strengths from outside SHM identifies transferable strategies. Consequently, this study suggests future research directions to enhance damage detection and life-cycle assessment in SHM.