In recent years, the prediction of the deterioration of water supply pipes within the framework of asset management has become an important part of the maintenance and management of urban infrastructure. In particular, the need to anticipate the risk of deterioration in advance is increasing year by year in the pursuit of longer service life and more efficient maintenance of sewer pipes.

Our research team is using machine learning methods to build a model for predicting deterioration to address this problem. Specifically, the model is trained by combining information on the diameter, material and year of laying of the pipeline with the number of years until failure, and analysing the key features related to the degree of deterioration. The model has the ability to predict future deterioration risks with high accuracy based on historical data.

Furthermore, the accuracy of our model is confirmed by using ROC curves and comparing them with predictions of deterioration based on age alone. We will also apply supervised machine learning to pipelines where deterioration has already been investigated, and link this to GIS information to visualise areas where deterioration is progressing.

We hope that this initiative will pave the way for ensuring sustainable water resource supply and economical and effective asset management at the same time.