Bin Liang, Dilusha Weeraddana, Zhidong Li, Shiyang Lu, Xuhui Fan, Yang Wang, Fang Chen, Gagneet Serai, Ivan Beirne, Mitchell Hayward
Ozwater
A novel pipeline failure prediction and risk distribution model is developed using machine learning frameworks. In this application, the model is used to discover the underlying drivers of past water main breaks in a distribution network and then utilized to predict the probability of future water main breaks in the same network. Failure probability calculations are combined with existing asset criticality data to produce a risk distribution. Failure probability and risk to customers are calculated for all water main assets. Utilities can use the high resolution information to develop targeted break mitigation and asset renewal strategies. Cost effective break mitigation reduces negative customer impact and cost to serve.