Comparison of Statistical Models for Predicting Pipe Failures: Illustrative Example with the City of Calgary Water Main Failure

Abstract

Several statistical models have been proposed to study effects of different covariates in the failure of water pipes. In this paper, three statistical models, the Weibull proportional hazard model (WPHM), the Cox proportional hazard model (Cox-PHM), and the Poisson model (PM), were considered. Curve fitting techniques were proposed for estimation of a baseline hazard function equation for the Cox-PHM to allow for its application in break prediction. The water main failure in the City of Calgary is used for this study. From the statistical models, physical covariates (e.g., pipe diameter, length) compared with environmental covariates (e.g., temperature) were the critical factors impacting the pipe failure rate. Furthermore, WHPM and PM showed the best fit for prediction of metallic and PVC pipes, respectively. Pipe break prediction through estimation of baseline hazard by curve fitting techniques produces good results especially for a