Fiber-Optic Sensor-Based Structural Health Monitoring with Machine Learning: A Task-Oriented and Cross-Domain Review.

Structural health monitoring (SHM) plays an increasingly important role in managing aging, safety-critical infrastructure under growing environmental and operational demands. In recent years, fiber-optic sensors (FOSs) have attracted significant attention for SHM applications due to their immunity to electromagnetic interference, durability in harsh environments, multiplexing capability, and suitability for both localized and fully distributed measurements. In parallel, advances in machine learning (ML) have enabled new approaches for extracting actionable insights from large, high-dimensional sensing datasets. This paper presents a systematic review of FOS-based SHM systems integrated with ML across civil, transportation, energy, marine, and aerospace infrastructures. Following PRISMA 2020 guidelines, peer-reviewed studies were identified and synthesized to examine sensing principles, deployment configurations, data characteristics, and learning-based analytical strategies. Fiber optic technologies are categorized into point-based, quasi-distributed, and fully distributed systems, and their capabilities for capturing strain, temperature, and spatiotemporal structural responses are critically evaluated. ML approaches are examined from a task-oriented perspective, including damage detection, localization, severity assessment, environmental compensation, and prognosis, with emphasis on the alignment between sensing configurations and appropriate learning paradigms. Key challenges remain, particularly regarding large data volumes, environmental variability, limited labeled damage datasets, model generalization, and system-level integration. Emerging directions such as physics-informed and hybrid learning, transfer learning, uncertainty-aware modeling, and integration with digital twins are discussed as pathways toward more robust and scalable SHM systems. By jointly addressing sensing physics and data-driven intelligence, this review provides a structured reference and practical roadmap for advancing intelligent FOS-based SHM in next-generation infrastructure.