Integrating CIRx and Machine Learning for Predictive Stormwater and Wastewater Infrastructure Management

Abstract

Municipal stormwater and wastewater utilities manage vast, largely buried asset networks whose deterioration is difficult to observe directly and expensive to inspect exhaustively, forcing asset managers to prioritize inspection and rehabilitation spending under significant uncertainty. This paper presents an integrated framework combining the Condition-Index Rx (CIRx) methodology — a composite, prescriptive condition-scoring approach that translates structural, hydraulic, consequence-of-failure, and operational data into a single actionable rehabilitation priority score — with machine learning (ML) models for predictive deterioration forecasting and anomaly detection. We describe a five-layer data engineering architecture spanning sensing and ingestion, feature and asset-twin storage, ML predictive modeling, CIRx scoring and prioritization, and decision and work-order action, and we detail how CCTV/PACP inspection data, SCADA and flow-sensor telemetry, GIS asset registries, and maintenance history are fused to compute both current condition and forecast risk. A comparative table summarizes the composition and weighting of CIRx components, and a second table compares candidate predictive model classes across precision, recall, and typical forecasting lead time. Two figures illustrate the proposed architecture and an illustrative set of ML-predicted deterioration trajectories across common asset cohorts (cast iron, vitrified clay, reinforced concrete, and PVC/HDPE pipe), showing how predicted trajectories intersect a defined CIRx intervention threshold to trigger rehabilitation prescriptions before failure occurs. We further discuss data governance, model validation against limited failure-event data, and integration with capital planning processes, concluding with open research directions including cross-utility data sharing and network-level cascading-failure modeling.