A survey of generative models for image-based structural health monitoring in civil infrastructure

Accurately assessing and monitoring the condition of structures is essential for ensuring the safety and integrity of civil infrastructure. Over the past decade, image-based structural health monitoring technologies have emerged as powerful tools to enhance efficiency and improve the objectivity of structural evaluations. The integration of deep learning technologies with these monitoring systems has significantly improved the efficiency and reliability of structural condition diagnostics. Of particular interest are specifically Variational Autoencoders, Generative Adversarial Networks, and Diffusion Models, which have gained increasing attention due to their versatility in data generation and ability to address fundamental challenges in structural monitoring. While image-based structural health monitoring encompasses both damage detection and structural response measurements, this review primarily focuses on local-level monitoring applications such as damage detection, where generative models have demonstrated particular effectiveness in addressing challenges like limited data availability and environmental variations. This paper provides a comprehensive analysis of these generative models, examining their underlying concepts, mechanisms, and applications in image-based structural health monitoring. Key applications are reviewed, including structural damage detection, data augmentation for training, and emerging areas such as image quality enhancement and domain generalization. Our analysis presents the current state of generative models in structural monitoring, identifying critical challenges and promising future research directions. This systematic review serves as a foundational resource for researchers and practitioners in the field, offering insights into current achievements and potential advancements.