Modelling Pipe Failure using Statistical models
Abstract
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.