Collaborative Research: CISE-ANR:III: Small: Leveraging External Data for Enhanced Understanding and Causal Attribution of Anomalies in Wastewater Networks

Urban water and wastewater networks are essential to public health, environmental protection, and economic stability, yet the information used to manage these systems is often fragmented, incomplete, or difficult to retrieve and interpret. Aging infrastructure, fragmented geographic information system records, and large volumes of unstructured inspection data make it challenging for operators to detect leaks, blockages, or structural weaknesses before they cause service disruptions or environmental harm. This project addresses these challenges by developing new artificial intelligence (AI) methods to organize and interpret complex water and wastewater network data. By transforming scattered information into a coherent integrated network representation, the project aims to make water infrastructure management more efficient and reliable. The tools developed will help reduce the need for manual data review and support informed decision-making. This in turn will result in protecting community health and local environments through early detection of anomalies in water networks before they turn into costly emergencies. By making water management smarter and faster, these tools ensure more reliable service and help stabilize utility costs. Improvements to water and wastewater network data will be facilitated by a new computational framework that combines machine learning, graph-based modeling, and multimodal data integration to analyze urban water networks. The research advances methods for completing and repairing directed network representations using physics-informed flow models and attention-based neural networks, enabling the identification of missing or inconsistent connections in network data. The approach incorporates interpretable, multiresolution uncertainty quantification to assess confidence in inferred network structures and detected anomalies. In addition, the project develops multimodal learning techniques that integrate network data with video, imagery, audio, and text from inspection reports and maintenance records. The developed computational infrastructure will allow automated extraction of actionable information from traditionally unstructured sources. This enables cities to allocate resources more efficiently to ensure long-term water security and infrastructure sustainability. The methods will be trained, validated, and tested using independent wastewater network data sets from France and the United States, enabling robust evaluation and generalization across different data modalities, networks, geographic information, cities, and countries. The project also supports workforce development and transferability of skills through the training of graduate students and postdoctoral researchers in data science and AI, network modeling, and infrastructure analytics. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.