Abstract

Water Distribution Network (WDN) is the most important element in water supply systems. According to the Canadian Water and Wastewater Association (CWWA), there are more than 112,000 kilometers of water mains in Canada and their replacement cost is estimated to be $34 billion. Since majority of pipelines are frequently above 100 years old, they are prone to failure and outbreaks of disease derivable to drinking water are inevitable. Breakage in water infrastructure can result in disruptions and damage to other surrounding infrastructure such as road networks or structures. Moreover, unscheduled emergency rehabilitation works can cause interruption to traffic, households and businesses. Therefore, it is important to assess the unknown condition of WDNs to find their respective rate of deterioration in order to prevent disastrous failures or sudden shutdowns.
Determining pipe condition through cost-effective assessments will grant very poor condition pipes to be considered first in order to avoid related risk and devastating failures. The problem here is that in most cases, there are limited data about condition of water mains due to the underground location of the pipelines and their restricted access. Several pipes were installed 100 years ago and they have not been examined until a problem occurred. An extensive literature review shows the absence of comprehensive and generalized maintenance model for scheduling the rehabilitation and replacement of individual pipelines in the whole network based on their remaining useful life. Previous research efforts concentrated mostly on developing models, which utilize long-term data and consider solely the pipe segments not the whole network. Since pipe segments are connected together, the performance of one pipe affect the performance of other pipes in the neighborhood. This is the reason that pipes should be considered as a network rather than individual pipeline. This shows the need for a model which could forecast the behavior of each pipeline and the whole network based on available data simultaneously.
This study aims to develop a model that can predict remaining useful life to optimize the needed intervention plans based on the available budget. For this purpose, a statistical condition model is developed which utilizes characteristics of a pipeline to predict its condition. In this model, Delphi study identifies the most important factors affecting deterioration of water pipelines at first, through three rounds of questionnaires sent to selected experts. The findings show that important factors are mainly physical factors such as pipe age, pipe material, etc. After that, Fuzzy Analytical Hierarchy Process (FAHP) and Entropy Shannon are employed to prioritize the selected factors in previous step and calculate their weights based on their relative importance. Results reveal that pipe installation, age and material are the most effective parameters in deterioration. These weights are used to find the condition index of the pipeline from pipe characteristics, soil and water properties. Upon determining the condition index, the remaining useful life is estimated using the developed artificial neural network (ANN). Ultimately, the budget is allocated efficiently and different repair and replacement strategies are scheduled based on the remaining useful life and breakage rate of the pipelines utilizing the developed near optimum Genetic Algorithm (GA)-based model. Data of the water distribution network of the city of Montréal is used to develop, train and validate the developed models. Results indicate that 30.7 km of the pipelines of Montreal should be replaced in the next 20 years and 2610 km are in need of both major and minor rehabilitations. This research proposes a framework for optimized replacement and maintenance plans based on the remaining useful life and condition of the pipelines which will help operators for efficient budget allocation and better management of needed intervention plans.