Prediction of Likelihood of Failure of Underground Linear Assets Using Survival Analysis

Abstract

Water and sewer pipe assets are challenging to manage because they are buried in the ground. The scope of this paper was to determine the likelihood of failure (LoF) of water and sewer pipes, which was mainly their physical failure. The pipes that possibly deteriorated similarly were grouped either by judgmental or statistical methods. Survival analysis was used for this study due to its ability to include censored data, which was common for underground assets that do not get inspected very often and are difficult to keep track of. Parametric and non-parametric models were developed. The parametric model is better in predicting the LoF of underground assets. For all the data sets tested, the gamma distribution fitted the best. It is essential to manage assets in groups. By grouping similarly behaved pipes together, survival curves can be developed to predict their LoF effectively. Smaller asset groups would lead to many survival curves and it could be difficult to manage. However, accurate survival curves may be difficult to generate for large groups of assets. A compromise has to be made for the right group sizes so that an asset manager can effectively manage his assets. While studying the sewer pipes, it was found that for every 100 ft. increments in pipe lengths, there was an 18.2% drop in survival probability. This problem could be solved if pipe conditions were assessed in similar length groups. The survival curve is different for every group of assets.