<![CDATA[Structural health monitoring (SHM) has shifted from periodic visual inspection and threshold-based signal processing toward data-driven diagnosis powered by machine learning (ML). This article synthesizes recent developments in ML-based SHM across civil, aerospace, and composite structures, focusing on sensing modalities, learning paradigms, and damage-related tasks (detection, localization, quantification, and prognosis). Drawing on a thematically organized body of twenty-five peer-reviewed sources published mainly between 2021 and 2026, the review traces the field's progression from supervised classifiers trained on labeled vibration or strain data toward unsupervised and physics-informed models capable of operating under scarce or unlabeled damage data and fluctuating environmental and operational conditions. The discussion highlights persistent challenges, including the scarcity of real damage-state data, sensitivity to environmental and operational variability, limited interpretability of deep architectures, and the difficulty of transferring models across structures. The article's novelty lies in proposing an integrated conceptual framework that links sensing, feature representation, learning strategy, and decision support into a single life-cycle pipeline, rather than treating these as isolated research strands. The synthesis further identifies digital twins, physics-informed learning, and explainable artificial intelligence as convergent directions for closing the gap between laboratory-validated ML-SHM models and trustworthy, field-deployable infrastructure management systems]]>
Copyrights © 2026
