Deep Time-Series Models for Corrosion Forecasting in Urban Pipelines: A Data-Driven Review Based on Hong Kong (2007–2023)

This study investigates current limitations and future directions in the application of machine learning (ML) and computer vision (CV) for corrosion detection in pipeline systems. Challenges include architectural constraints of detection models such as YOLOv3, dataset limitations, and the need for high-resolution input processing. The role of data quality, augmentation, and transfer learning is evaluated in relation to model generalization. Interdisciplinary collaboration is identified as a critical factor in the development of predictive maintenance strategies. Integration of real-time monitoring, Internet of Things (IoT) frameworks, and advanced analytics is essential for achieving operational continuity and extending infrastructure lifespan. Sustainability and scalability are addressed through the incorporation of adaptive algorithms and robust sensor networks. Emphasis is placed on the development of predictive systems aligned with Industry 4.0, requiring improvements in data governance, cybersecurity, and cross-domain coordination.