Challenges and Prospects in Anomaly Detection of Sewer Monitoring Data: Annotating Synthetic Sewer Data with Known Sensor Failures

Abstract

Managing sewer monitoring data poses challenges, often revealing quality issues. This study explores the feasibility of employing crowdsourcing methods, akin to artificial artificial intelligence, on sewer flow monitoring data. We devised a data annotation project, relying on the expertise of 12 sewer researchers to retrieve artificial anomalies introduced to the data. While recognizing limitations in annotation skills, we compiled time series from 7 locations and implemented visualizations for improved interpretation. Evaluation using the F1 criterion yielded mediocre scores (0.625±0.226), highlighting challenges in interpreting noisy raw data and varying analyst mental models. Despite the potential for data-driven modeling in urban drainage research, our results suggest challenges in obtaining annotations through crowdsourcing. Further work should focus on standards for data annotations, community efforts in data labelling, and identifying role model utilities providing open access to routine wastewater datasets.