Multi-Case EPR Strategy for the Development of Sewer Failure Performance Indicators

A number of different pipe failure prediction models have been developed in the past to actively manage the sewer systems. Here, a novel methodology for developing mathematical models for the prediction of sewer system failures (i.e. number of collapses and blockages over some fixed time period) as a function of different potential explanatory factors (e.g. pipe material, diameter, age, soli type, traffic load, etc.) is presented. The methodology is based on the application of recently developed Evolutionary Polynomial Regression (EPR) technique to a group of (rather than several individual) sewer systems. The methodology is tested and verified on a real-life case study involving two UK sewer systems. The results obtained clearly demonstrate that, when compared to the two individual models, in the case of a grouped pipe failure prediction model local effects are neglected in favour of a more general description of the underlying deterioration phenomenon. As a consequence, even though the grouped pipe failure prediction model has reduced accuracy (i.e. reduced model fit) it is less prone to data over-fitting phenomenon.