| description abstract | Pressure pipe failure is a common problem for water utilities worldwide and can result in high costs and disruption of customers. This study analyzed the failure rates of iron, polyvinyl chloride (PVC), asbestos cement (AC), and polyethylene (PE) mains in the water network of the Auckland, New Zealand, over a six-year period. Correlations between failure rates and a range of contributing factors were studied, and multilinear regression and machine learning (random forest, gradient boosted decision tree) models were then used to predict pipe failure rates and prioritize pipes for replacement. The study found the most important factors influencing failure rates to be diameter, age, and modeled pressure for all materials. The failure rates of all materials increased within a relatively narrow band up to a 30 year age. For pressure, failure rates were observed to increase linearly with pressure for iron and AC, while plastics (PVC and PE) displayed nonlinear trends with pressure having a greater relative impact at higher values. The pressure trends were observed for each material when considering all pipes, but also when grouping pipes by diameter or age. A focused investigation using scatterer interferometry data captured by the SENTINEL-1 satellite did not find any correlation between pipe failures and any identified ground movements from adjacent reflective surfaces. The gradient boosted decision tree model was able to include a high fraction of failing pipes in a prioritized pipe replacement list limited to 1% of the system length. | |
