Modelling Pipe Failure using Statistical models
The degradation of pipe factor of safety upon pipe installation has increased interest in buried infrastructure asset management. Several modelling approaches using statistical models have been proposed to explain the effects of covariates in the failure of water pipes. In this thesis, Cox Proportional Hazard Model (Cox-PHM) was used in the prediction of number of break. Curve fitting techniques were proposed for estimation of baseline hazard function and the resulting equation applied in break prediction. The results from the model were compared to the results from Weibull Proportional Hazard Model (WPHM) and Poisson Model. Further, Cox-PHM was used to determine the time to failure of metallic pipes. Results indicated that physical factors e.g. diameter, were the critical factors impacting pipe failure and the occurrence of a particular break-type. Further, results indicated that the effects of covariates differ according to material type with PVC and DI pipe showing low and high vulnerability to breaks, respectively. Additionally, when mean time to failure (MTF) of a pipe was analyzed, it revealed that after the occurrence of a failure, time to failure for DI decreases significantly compared to CI pipe. Results indicated that when a pipe has had four breaks, it should be considered for replacement. The prediction results from the models revealed that different models depending on pipe material can be used to model the evolution of breaks for the City of Calgary Water Network (CCWN). Therefore, Poisson Model and WPHM performed best in the prediction of the PVC, and both DI and CI pipes, respectively. Results from Cox-PHM indicate that the estimation of the baseline hazard function using curve fitting techniques captures the trend of metallic pipes especially for the young water networks. It is therefore recommended that a combination of models should be used based on the rate of deterioration and material type of the system rather than a single model.