Sewerage Report Analysis for Report-Based Failure Prediction in Pipes

Ofek Aloni, Lukas Guericke, Nicolas Caradot, Barak Fishbain

Abstract: Maintenance of sewerage networks is costly, and the funds required will only increase with population size and pipe aging. Smart tools for future planning can help reduce costs and focus repairs on the pipes that need them most, preventing failures. Early research shows promise of Machine Learning (ML) models such as Random Forests (RF) in predicting the evolution in time of the structural state of pipes. We take a different approach, using ML models to directly predict pipe failures. Training data utilizes past reports of pipe failures, combined with GIS data with pipe parameters. This makes it immediately applicable to many municipalities, as this data is often available. This is especially relevant when CCTV inspections assessing the structural condition of pipes are scarce.