Artificial Intelligence for Structural Condition Assessment and Rehabilitation: Recent Advances and Future Directions

The growing need to ensure the safety, resilience, and sustainability of existing building structures has accelerated the adoption of artificial intelligence (AI) for structural condition assessment and rehabilitation. This critical narrative review synthesizes 82 retained sources, including 33 application-oriented sources, through a transparent, structured literature search and study-selection process; it is not a formal systematic review or meta-analysis. To organize this fragmented evidence base, the review introduces the Data-to-Decision (D2D) Continuum, a unifying conceptual framework that traces eight engineering stages from data acquisition through damage detection, localization, quantification, condition and performance assessment, prognosis, reliability and risk assessment, to rehabilitation decision support. Classical machine learning, deep and temporal models, physics-guided and probabilistic approaches, and emerging foundation models are examined according to the engineering output required at each stage. The strongest evidence concerns bounded defect detection and localization, whereas uncertainty-aware prognosis, risk-informed rehabilitation selection, multi-site validation, and governed deployment remain markedly less mature. By integrating existing monitoring, digital-twin, life-cycle risk, and maintenance-decision concepts into an interface-centered evidence chain, the D2D framework clarifies what must be validated before an AI output can responsibly influence an intervention.