Ageing big-diameter underground steel pipelines pose significant sustainability and operational challenges for bulk water distribution utilities. These challenges include but are not limited to an increased risk of underground big-diameter pipelines failure and rising costs for asset condition assessments. We developed and tested a unsupervised machine-learning framework to improve pipeline condition assessment, predictive maintenance, and inspection prioritisation using real-world secondary data. We combined mixed-data clustering, non-linear dimensionality reduction, anomaly detection, and association rule mining to identify complex patterns in the condition of underground steel pipelines without excavation. Our results indicate that mixed-type clustering methods produce stable, well-separated condition groups and outperform numeric-only methods. Non-linear embeddings show clear separability, and anomaly detection reliably pinpoints high-risk pipeline segments. Association rules reveal hidden connections between pipeline attributes, enhancing clarity and engineering relevance. This framework enables data-driven decision-making, reduces unplanned maintenance, and supports efficient resource use. By extending underground big-diameter pipeline asset lifespans and boosting operational reliability. This study supports sustainable asset management and improved operations in bulk water distribution pipelines systems through practical, scalable unsupervised analytics.

Keywords

Unsupervised Machine Learning, Big-diameter steel pipeline, Water Utilities