A novel inversion method for assessing pipe corrosion condition based on upstream-downstream water quality variation in drinking water distribution systems.

Abstract

Internal corrosion in drinking water distribution systems (DWDSs) drives secondary water quality deterioration, yet scalable approaches for inferring in-service corrosion condition remain limited. In this study, 20 excavated pipe segments from a large metropolitan DWDS were investigated using paired upstream-downstream water quality measurements, inner wall image interpretation, and physicochemical characterization of corrosion scales. A residence-time-normalized variation rate was used to quantify segment-level water quality change. By comparing alternative image-weighting schemes and PCA scoring combinations, a mutually validated corrosion score was derived. The score captured the dominant contrast between Fe-rich corrosion products and Ca-enriched materials and showed strong agreement with the image-based condition index (Spearman's ρ = 0.95, p < 0.001). Among the evaluated models, a multivariable linear regression (MLR) model was selected for its strong predictive performance and interpretability. Incorporating pipe attributes and selected variation metrics, it explained 85.4% of the variance in the corrosion score (R2 = 0.854) and provided clear in-sample discrimination between severe and non-severe segments. Independent validation using 12 additional and non-excavated pipe segments achieved 83.3% classification accuracy. These results demonstrate that pipe corrosion condition can be inferred from routine water quality signals through a mechanistically interpretable and statistically grounded framework, providing new insight into how pipe wall condition shapes downstream water quality responses in drinking water distribution systems.