Summary
           Drinking water supply systems in relatively dense urban and rural areas consist mainly of pipes usually installed underground, whose structures are changing over time. This greatly hinders the work of maintaining and carrying an accurate record of some key actions complementary (inventories, replacement and/or repair of pipelines, failures and damages management, demand management service, etc.) necessary not only for proper infrastructure maintenance, but also for a comprehensive and effective management of the service.
In reference to the above mentioned, and in order to contribute to perform efficiently the work of planning, design, construction, operation and maintenance of water supply networks, it is necessary to count with an updated information system of the hydraulic infrastructure, its technical state and its deterioration. It may well address not only the management and planning of the network, but also recording and precise quantification of the real state of structural deterioration and reliability of supply system in general, allowing statistical modeling. 
In this thesis, a review of failure models in individual sections of pipe water supply networks is conducted. Additionally, it assesses the adequacy of the data quality and quantity for feedback from these models. It also describes the most appropriate statistical methods to model these pipes failures and the most suitable are applied to the case of a network in a city of the Spanish Mediterranean coast. This database is characterized by its short history of breakages and high censorship.
The overall objective of this study (review, evaluation, adaptation and application of survival models in engineering), primarily addressed the problem of identifying the pattern of reliability and survival of water pipes and its networks as a first approach with free distribution techniques such as Kaplan Meier and life-tables, obtaining highly coincidental results that reinforce confidence in the coherence and consistency of these analyses. 
Moreover, Cox regression models are applied (after verifying assumptions, estimating parameters for maximum likelihood and revise the adjustment to the data) to identify most influential risk factors (physical and environmental). The aim is to model the relationship between failure rate and environment of the distribution system. In another phase of this work, the best fitting parametric family is identified, among more than ten models of possible distributions of pipe’s life. Model parameters are estimated by least squares and maximum likelihood, with very similar results, and finally selected the best fit to the data (using Anderson-Darling indicator and Pearson correlation).
The study notes a slight tendency to overestimate pipes reliability in the Cox method, compared with the Kaplan-Meier estimates and the parametric model estimates (lognormal model), from 35-40 years of the network operation. The study also revealed that survival and the failure risk influencing factors goes beyond age, including such as material component of the pipeline and working conditions.