Corrosion Prediction on Sewer Networks with Sparse Monitoring Sites: A Case Study

Abstract

Sewer corrosion is a widespread and costly issue for water utilities. Knowing the corrosion status of a sewer network could help the water utility to improve efficiency and save costs in sewer pipe maintenance and rehabilitation. However, inspecting the corrosion status of all sewer pipes is impractical. To prioritize sewer pipes in terms of corrosion risk, the water utility requires a corrosion prediction model built on influential factors that cause sewer corrosion, such as hydrogen sulphide (H$_2$S) and temperature. Unfortunately, monitoring sites of influential factors are very sparse on the sewer network such that a reliable prediction has often been hampered by insufficient observations – It is a challenge to predict H$_2$S distribution and sewer corrosion levels on the entire sewer network with a limited number of monitoring sites. This work leverages a Bayesian nonparametric method, Gaussian Process, to integrate the physical model developed by domain experts, the sparse H$_2$S and temperature monitored records, and the sewer geometry to predict corrosion risk levels on the entire sewer network. A case study has been conducted on a real data set of a water utility in Australia. The evaluation results well demonstrate the effectiveness of the model and admit promising applications for water utilities, including prioritizing high corrosion areas and recommending chemical dosing profiles.