AI‐driven assessment of steel structures

Abstract Ageing steel infrastructure (bridges, industrial facilities, heritage buildings, offshore platforms and transmission towers) demands reliable, scalable strategies for structural safety assessment, yet conventional inspection and analysis remain slow, costly and intrusive. Recent advances in artificial intelligence (AI) are increasingly supporting every stage of the assessment process, from automated visual and sensor‐based inspection to hybrid physics–ML remaining‐life prediction and probabilistic decision support. Drawing on eleven case studies from real projects, this paper distils the conditions under which AI‐driven assessment becomes practical rather than merely possible. The cases cover anomaly detection and damage identification, virtual sensing and stress‐history reconstruction, fatigue assessment, digital‐twin‐based monitoring under real operating conditions, predictive maintenance, damage‐source‐oriented data‐driven monitoring, and condition assessment of deteriorating components. Taken together, the cases fall most naturally into categories based on the origin of their training data, and point to four recurring conditions for practical adoption: that learning is typically unsupervised by necessity, since a structure in service cannot be deliberately damaged to produce labelled examples; that hybrid physics‐plus‐AI formulations are more deployable and easier to validate than purely data‐driven ones; that sensing can be economised by instrumenting heavily for a short training campaign and monitoring cheaply thereafter; and that AI can serve as a fast surrogate for otherwise prohibitive computation. The paper concludes by outlining the steps required to move AI from research prototypes into established engineering practice, and by positioning the structural engineer as the informed decision‐maker within an AI‐augmented workflow.