Real-time monitoring and predictive maintenance of urban water networks using digital twins

The maintenance of urban water systems is often reactive. Maintenance crews rely on incident reports and institutional knowledge to ensure that the network remains operational. Unfortunately, this reactive approach often strains maintenance crew resources while posing the risk of the cascading failure of urban water infrastructure; the increasing frequency of extreme weather events further compounds this risk. In this work, we introduce a machine learning-driven digital twin approach for the near real-time monitoring and proactive identification of maintenance issues to ensure the effective operation of urban water systems. This approach uses machine learning-based anomaly detection methodologies to identify changes in the flow regimes, detect blockages, and flag anomalous behavior in the wastewater network. In this talk, through a real-world application, we demonstrate the efficacy of the proposed approach and the various steps (e.g., data filtering, accounting for model drift) involved in deploying a digital twin for real-world application. Additionally, this talk also focuses on the human-computer interaction aspect of the digital twins in urban water systems; specifically, our experiences co-designing a platform for communicating actionable information to the maintenance teams and decision-makers in the city. Through this talk, we aspire to provide the attendees with an insight into the application and the socio-technical challenges involved in using digital twins in the real world.