The deterioration of pipes in urban water distribution systems is of concern to waterutilities throughout the world. This deterioration generally leads to pipe breaks andleaks, which may result in reduction in the water-carrying capacity of the pipes fromtuberculation of interior walls of the pipe. Deterioration can also lead to contaminationof water in the distribution systems. Water utilities which are already facing tightfunding constraints incur large expenses in replacement and rehabilitation of watermains, and hence it becomes critical to evaluate the current and future condition of thesystem for making maintenance decisions. Quantitative estimates of the likelihood ofpipe breaks on individual pipe segments can facilitate inspection and maintenancedecisions. A number of statistical methods have been proposed for this estimationproblem. This thesis focuses on comparing these statistical models on the basis of shorttime histories. The goals of this research are to estimate the likelihood of pipe breaks inthe future and to determine the parameters that most affect the likelihood of pipe breaks.The various statistical models reviewed in this thesis are time linear and timeexponential ordinary least squares regression models, proportional hazards models(PHM), and generalized linear models (GLM). The data set used for the analysis comes from a major U.S. city, and the data includes approximately 85,000 pipe segments withnearly 2,500 breaks from 2000 through 2005. The covariates used in the analysis arepipe diameter, length, material, year of installation, operating pressure, rainfall, land use,soil type, soil corrosivity, soil moisture, and temperature. The Logistic GeneralizedLinear Model fits can be used by water utilities to choose inspection regimes based on arigorous estimation of pipe breakage risk in their pipe network.
展开▼
