Exploring the Relationship between Soil Properties and Deterioration of Metallic Pipes Using Predictive Data Mining Methods

Abstract

Soil corrosivity is considered to be a major factor for the deterioration of metallic water mains. Using a 10-point scoring method as suggested by the American Water Works Association, soil corrosivity potential can be estimated by five soil properties: (1) resistivity; (2) pH value; (3) redox potential; (4) sulfide; and (5) percentage of clay fines. However, the relationship between soil corrosivity and pipe deterioration is often ambiguous and not well-defined. In order to identify the direct relationship between soil properties and pipe deterioration, which is defined as the ratio of the maximum pit depth to pipe age, predictive data mining approaches are investigated in this study. Both single- and multipredictor based approaches are employed to model such relationship. The advantage of combining multiple predictors is also demonstrated. Among all approaches, rotation forest achieves the best result in terms of the prediction error to estimate pipe deterioration rate. Compared to the random forest method, which is next to the best, the normalized mean square error decreased 50%. With the proposed approaches, the assessment of pipe condition can be achieved by analyzing soil properties. This study also highlights the importance for collecting more reliable soil properties data.