Background: Civil infrastructure — bridges, buildings, tunnels, dams, highways, and towers — is ageing worldwide, and deterioration from corrosion, fatigue, cracking, overloading, and environmental and seismic hazards threatens safety and serviceability. Conventional visual inspection, manual condition rating, and periodic non-destructive testing are labour-intensive, subjective, and poorly suited to continuous, large-scale monitoring.
Objective: This article reviews and critically analyses the application of artificial intelligence (AI), machine learning (ML), and deep learning (DL) to structural health monitoring (SHM) of civil infrastructure, and proposes an integrated analytical framework linking sensing, learning, and maintenance decision-making.
Methods: A critical review of peer-reviewed literature on data-driven SHM, classical ML algorithms, deep learning architectures, computer-vision-based defect detection, and intelligent infrastructure technologies (IoT, digital twins, predictive maintenance) is combined with a theoretical and comparative analysis of representative AI/ML models across vibration, strain, displacement, acoustic-emission, image, and environmental data modalities. Where actual experimental results were unavailable, hypothetical or representative comparisons are explicitly identified as such.
Results: The literature indicates that ensemble and kernel-based ML methods (random forest, support vector machine) offer interpretable, data-efficient solutions for damage classification, whereas deep learning methods (convolutional neural networks for imagery, recurrent/long short-term memory networks for time-series) provide superior automated feature extraction and predictive accuracy at the cost of larger datasets, higher computational demand, and reduced interpretability. Multimodal sensing and digital-twin integration improve robustness against environmental and operational variability, while explainable AI techniques are increasingly needed to support engineering trust and regulatory acceptance.
Conclusion: AI/ML-enabled SHM offers substantial potential to improve early damage detection, predictive maintenance, and infrastructure resilience, but practical, scalable deployment requires standardized datasets, explainable models, robust handling of environmental variability, and validated real-time edge and cloud architectures.