Sewer and drainage systems are critical components of urban infrastructure, particularly in densely populated cities like Hong Kong. With approximately 4,565 kilometers of underground pipelines, comprising 39% sewers and 51% stormwater drains, Hong Kong faces increasing infrastructure challenges due to aging assets. Notably, 45% of the pipeline network exceeds 30 years in age, while 15% surpasses 50 years, elevating the risk of structural failure, blockages, and sewer overflows. These failures compromise environmental safety, public health, and urban resilience.

Despite routine inspections, most condition data are embedded in PDF-based CCTV reports, which are not readily machine-readable or actionable. This lack of structured, real-time condition data hinders proactive maintenance and timely rehabilitation. Moreover, there is limited integration between drainage performance data and other critical geospatial and environmental factors such as soil characteristics, traffic load, land use, and rainfall variability. Without a unified platform for data fusion and predictive modelling, decision-makers struggle to identify high-risk assets or prioritize interventions effectively.

To address this pressing need, this project proposes a smart and sustainable modelling framework for proactive sewer and drainage infrastructure management. By developing a fully integrated GIS-based platform, automated condition extraction tool, and data-driven deterioration models, the project aims to revolutionize how drainage asset management is conducted in Hong Kong. This is the first large-scale attempt in the region to create a unified, dynamic, and operational tool for predicting network deterioration and supporting condition-based maintenance planning. The proposed approach will be co-developed in collaboration with key government departments and stakeholders to ensure practical relevance and broad scalability.

Project objectives:
1. To develop an automated pipeline for extracting condition data from historical CCTV reports, converting them from unstructured PDF documents into structured datasets suitable for analysis and visualization.
2. To enrich the CCTV-based dataset with external geospatial and environmental variables, including soil properties, weather data, land use, and traffic conditions—and integrate them into a unified GIS database that supports spatially contextualized decision-making.
3. To build a deterioration prediction model for sewer and stormwater pipelines, accounting for age, environmental stressors, operational history, and physical attributes, using advanced survival analysis and machine learning techniques.
4. To design and implement a web-based decision support platform for municipal and infrastructure stakeholders. This platform will enable real-time condition tracking, failure risk mapping, sewer overflow monitoring, and interactive dashboards for planning and policy support.

Key Findings and Deliverables
1. Development of Smart Monitoring Systems for Blockage and Overflow

• A novel monitoring system using contactless flow sensors was successfully developed to detect and analyze overflow/blockage events in small-sized sewer networks.
• The study demonstrated that existing problems include high sedimentation, insufficient inspection, and backflow-induced overflows.
• A smart sensor-based system integrating flow depth and velocity readings significantly improved detection accuracy and real-time response capabilities.

2. Application of Deep Learning and IoT for Sewer Condition Assessment

• Deep learning models, specifically CNN-based architectures, were applied for automated condition rating using image datasets from sewer inspections.
• The models achieved high classification accuracy across defect severity levels and demonstrated scalability across regions.
• The proposed pipeline includes data collection from CCTV, preprocessing, model training, and output visualization via web dashboard tools for municipal decision-makers.

3. GIS and Data-Driven Modeling for Risk Prioritization

• A hybrid GIS–ML framework was developed to predict failure risk in aging sewer pipes.
• The integration of geospatial pipe attributes, hydraulic parameters, and inspection data into ML models like Random Forest and Gradient Boosting Machines yielded over 85% accuracy in predicting structural condition.
• The model allowed visualization of risk hotspots and prioritization for proactive maintenance.