10

CHAPTER 2

LITERATURE REVIEW

Water distribution systems consist of pipeline networks and

associated components, most of which is underground and exposed to soil

corrosion and mechanical stress from the surrounding soil, surface traffic,

and internal water pressure (Ahammed and Melchers 1997). Pipe failure in

water distribution systems disrupts the water supply to consumers and reduces

the reliability of the system. It is found that about 35% to 60% of the

supplied volume is wasted due to pipe leakages (Babovic et al 2002).

Therefore, inspection, control and planned maintenance and rehabilitation

programs are necessary to properly operate existing water distribution

systems (Saegrov et al 1999).

This chapter reviews the applications of various methods, tools and

techniques for the design and performance evaluation of water supply systems

and the optimization models in water distribution systems. Section 2.1

presents several optimization models, techniques and heuristics used for the

optimal design of water distribution systems that have been given appreciable

attention in the published literature. A literature survey on optimal operational

strategies of urban water distribution systems is given in Section 2.2. The

literature on topics such as failure analysis and prevention, reliability-based

design, and risk assessment and control in water distribution systems are

presented in Section 2.3. A review of literature on the maintenance,

rehabilitation, strengthening and replacement policies of water distribution

systems is included in Section 2.4.

11

2.1 DESIGN OF WATER DISTRIBUTION SYSTEM

A water distribution network must be designed so that it can supply

the desired quantity of water to the consumers at sufficient pressure. The

design involves specifying the sizes of different elements of the distribution

network and checking the adequacy of this network (Mays 2000). Significant

effort has been placed in developing approaches to solve for optimal designs

of water distribution systems.

2.1.1 Pipe Characteristics

A large body of literature exists on the optimization of the pipe

network design, reporting the application of classical optimization methods

(including linear programming, dynamic programming and nonlinear

programming). These methods have been used, sometimes at the cost of

considerable simplifications of the optimization models. One of the earliest

optimization approaches, the linear programming gradient method was

proposed by Alperovits and Shamir (1977). Other authors followed this

innovative course and introduced alternative derivations from the linear

programming-based gradient expressions (Quindry et al 1981, Fujiwara

et al 1987, Lansey and Mays 1989, Kessler and Shamir 1989, Fujiwara and

Khang 1990). These approaches lead to solutions in which pipes have one or

two fixed diameter segments. For practical implementation this type of

solution is unrealistic.

The state-of-the-art principles and methods of pipe network

optimization are presented by Walski (1985). Su et al (1987) presented a basic

framework for a model that can be used to determine the optimal least-cost

design of a water distribution system subject to conservation of energy and

reliability constraints. The limitation of this model is that the resulting pipe

diameters may not be commercially available pipe sizes and should be

12

rounded of to the appropriate sizes. These approximated diameters might

affect the feasibility of the resulting optimal solution.

Fujiwara and Silva (1990) proposed a heuristic method to obtain a

least-cost water distribution network design with a given reliability. The

method first determines an optimal design without the consideration of

reliability. The reliability of the network design is then assessed. An iterative

feedback procedure is then employed, which improved the reliability with a

small increase in cost.

Nonlinear programming technique has been used as an optimization

approach to solve the design optimization problem of water distribution

networks (Fujiwara and Khang 1990, Fujiwara and Khang 1991 and Varma

et al 1997) in which the diameter is taken to be a continuous variable.

A redundancy-constrained minimum-cost design of water

distribution networks is presented by Park and Liebman (1993). Redundancy

is quantified using the expected shortage due to failure of individual pipes as

a measure of reliability that permits incorporation of some considerations of

frequency, duration and severity of damage.

Developments in the field of stochastic optimization have allowed

the resolution of design optimization problems formulated as nonlinear mixed

integer problems. Genetic algorithms were used by Murphy et al (1993),

Simpson et al (1994), Dandy et al (1996), Savic and Walters (1997),

Keedwell and Khu (2005), and Keedwell and Khu (2006) and Simulated

annealing technique was applied by Cunha and Sousa (1999) and Cunha and

Sousa (2001). The applications of other evolutionary optimization algorithms

such as Ant Colony Optimization (Maeir et al 2003), Shuffled Frog Leaping

Algorithm (Eusuff and Lansey 2003), Tabu Search heuristic (Cunha and

Ribeiro 2004), Scatter Search (Lin et al 2007) and Particle-Swarm Harmony

13

Search (Geem 2009 b) to obtain the least-cost design of water distribution

network are reported in the literature. Evolutionary algorithms have also

enabled researchers to approach water distribution network design

optimization as a multiobjective problem, usually to determine the least cost

design which maximizes the benefits (Halhal et al 1997, Walters et al 1999,

Prasad and Park 2004, Kapelan et al 2005, and Lyroudia et al 2005).

Liong and Atiquzzaman (1994) proposed a powerful optimization

algorithm, Shuffled Complex Evolution (SCE) linked with EPANET, the

network simulation model to solve water distribution network design

optimization problems.

Taher and Labadie (1996) developed a prototype decision support

system WADSOP (Water Distribution System Optimization Program) to

guide water distribution system design and analysis in response to changing

water demands, timing, and use patterns; and accommodation of new

developments. WADSOP integrates a geographic information system (GIS)

for spatial database management and analysis with optimization theory to

provide a computer-aided decision support tool for water engineers. Xu and

Goulter (1999) proposed a fuzzy linear program optimization approach for the

optimal design of water distribution networks.

Wu and Simpson (2001) applied a Genetic Algorithm to the

optimal design and rehabilitation of a water distribution system. Two

benchmark problems of water pipeline design and a real water distribution

system are presented to demonstrate the application of the proposed

technique.

A Fuzzy linear programming model is formulated by Bhave and

Gupta (2004) for minimum cost design of water distribution networks. Future

water demands being difficult to predict with any uncertainty are considered

14

as fuzzy demands and modeled by trapezoidal possibility distribution

function. The proposed linear programming model avoided iterative

procedure and also provided a cheaper solution.

A self-adaptive fitness formulation for solving constrained design

optimization of water distribution networks is presented by Farmani

et al (2005). The method has been formulated to ensure that slightly infeasible

solutions with a low objective function value remain fit. The method does not

require an initial feasible solution, this being an advantage in real-world

applications having many optimization variables.

Vairavamoorthy and Ali (2005) proposed a methodology for the

optimal design of water distribution systems based on Genetic Algorithms.

The objective of the optimization is to minimize the capital cost, subject to

ensuring adequate pressures at all nodes during peak demands. The method

involves the use of a pipe index vector to control the genetic algorithm search.

The method has been tested on several networks, including networks used for

benchmark testing least cost design algorithms, and has been shown to be

efficient and robust.

A least-cost design of water distribution networks under demand

uncertainty is developed by Babayan et al (2005). A new approach to

quantifying the influence of demand uncertainty is proposed. The original

stochastic model is reformulated as a deterministic one, and it is coupled with

an efficient genetic algorithm solver to find robust and economic solutions.

Khu and Keedwell (2005) formulated a multi-objective

optimization problem for water distribution network design and results are

compared with optimization using a single-objective Genetic Algorithm and

two-objectives optimization using a non-dominated sorted Genetic

Algorithm-II (NSGAII). Atiquzzaman et al (2006) suggested a scheme to

15

assist decision makers in selecting the best alternative water distribution

network design solution which is within the available budget and tolerated

pressure deficit. A multi objective optimization algorithm, NSGA-II is

coupled with water distribution network simulation software-EPANET

(Rossman, 2000) to demonstrate the application of the same.

Geem (2006) presented a cost minimization model using Harmony

Search (HS) algorithm for the design of water distribution networks. The

model is applied to five water distribution networks and the results showed

that the Harmony Search-based model is suitable for water network design.

Zecchin et al (2006) proposed an advanced Ant Colony Optimization (ACO)

algorithm known as Max-Min Ant System (MMAS) for the water distribution

system optimization problem.

Branch and Bound integer linear programming technique is used by

Samani and Mottaghi (2006), for optimum design of municipal water

distribution networks. The constraints include pipe sizes, reservoir levels, pipe

flow velocities and nodal pressures. This procedure helped to design a water

distribution network that satisfies all required constraints with a minimum

total cost.

Lin et al (2007) have proposed an optimization procedure using

Scatter Search (SS) heuristic to solve the design optimization problem of

water distribution networks. The computational results obtained with the three

example networks indicate that SS is able to find solutions comparable to

those provided by some of the most competitive algorithms published in the

literature. Perelman and Ostfeld (2007) presented an adaptive stochastic

algorithm for optimal design of water distribution systems based on the

heuristic cross-entropy method for combinatorial optimization. Immune

Algorithm (IA) inspired by the defense process of the biological immune

16

system is proposed by Chu et al (2008) to obtain the optimal design of water

distribution networks.

Reca et al (2008) evaluated the performance of metaheuristic

techniques such as Genetic Algorithms, Simulated Annealing, Tabu Search

and iterative local search used in the design optimization of water distribution

networks. It is observed that Genetic Algorithm is more efficient when

dealing with a medium-sized network, but other methods outperformed it

when dealing with a real complex one.

A modified Harmony Search algorithm incorporating ‘particle

swarm’ concept is developed by Geem (2009 a) for optimal design of water

networks. Pierro et al (2009) have introduced two hybrid algorithms,

ParEGO and LEMMO, to solve a multi-objective design optimization

problem of water distribution networks. A real medium-size network in

Southern Italy and a real large-size network in United Kingdom under a

scenario of a severely restricted number of function evaluations are tested on

the design problem. Banos et al (2010) have analyzed the performance of a

new memetic algorithm for the optimal design of water distribution systems.

2.1.2 Network Layout

The joint problem of layout and component design of water

distribution networks is addressed by Rowel and Barnes (1982). A two-level

hierarchically integrated system of models is developed for the layout of both

single and multiple source water distribution systems. Bhave and Lam (1983)

proposed a Dynamic Programming approach to obtain a less costly

distribution layout. An integrated model for the least cost layout and design of

water distribution networks is developed by Goulter and Morgan (1985). The

model consists of two linked linear programming formulations. One linear

program determines the least cost layout of a distribution network given an

17

initial pressure distribution. The other program determines the least cost

component design given an initial pipe layout.

Goulter and Coals (1986) developed two quantitative approaches to

the incorporation of reliability measures in the least-cost design of water

distribution networks. In both approaches Linear Programming technique is

used to obtain an optimal layout design.

A model for the layout optimization of water distribution networks

under single loadings is presented by Awumah et al (1989). The zero-one

integer programming model is used to select the pipes that should form the

network, while satisfying redundancy and hydraulic requirements. A network

component optimization step, using well established design models is then

applied to this solution to refine the pipe sizes and pressure heads, thus giving

a layout and component optimal solution.

Lejano (2006) developed a method for determining an optimal

layout for a branched pipeline irrigation water distribution system given only

the spatial distribution of potential customers and their respective demands. A

mixed integer linear programming (MILP) algorithm is applied to optimize an

empirically derived objective function.

Genetic Algorithm approach is presented by Afshar (2007) for the

simultaneous layout and component size optimization of water distribution

networks. The method starts with a predefined maximum layout which

includes all possible and useful connections. An iterative design-float

procedure is then used to move from the current to a cheaper layout satisfying

a predetermined reliability set by the user.

Tanyimboh and Setiadi (2008) presented a multi-criteria maximum-

entropy approach to the joint layout, pipe size and reliability optimization of

18

water distribution systems. The capital cost of the system is taken as the

principal criterion, and so the trade-offs between cost, entropy, reliability and

redundancy are examined sequentially in a large population of optimal

solutions.

2.2 OPERATION OF WATER DISTRIBUTION SYSTEM

2.2.1 Leak Detection and Monitoring

Leakage in water supply networks can represent a large percentage

of the total water supplied, depending on the age and deterioration of the

system. As a result of water losses and increasing population, urban areas may

experience shortages of water. Coulbeck and Orr (1993) presented a

reliability perspective of the required systems and activities for control of

water distribution networks with an objective of cost control, quality control

and leakage control. The ways in which computers are being used for control

purposes are described.

Reis et al (1997) have addressed the problem of appropriate

location of control valves in a water supply pipe network and their settings via

Genetic Algorithm to obtain maximum leakage reduction for given nodal

demands and reservoir levels.

Vitkovsky et al (2000) used Genetic Algorithm technique in

conjunction with the inverse transient method to detect leakage locations and

magnitudes while simultaneously finding the friction factors in water

distribution systems.

A new method for detecting the magnitude of leaks in small

residential service zones of a drinking water distribution system is proposed

by Buchberger and Nadimpalli (2004). Several examples, based on observed

19

and simulated pipe flows are presented to demonstrate the application of the

leak detection method.

Almandoz et al (2005) presented a methodology for evaluation of

water losses based on discrimination of the two components of uncontrolled

water in a water distribution network; physical losses in mains and service

connections, and the volume of water consumed. An extended period

simulation of a water distribution network is employed in the study.

A model to support decision systems regarding quantification,

location and opening adjustment of control valves in a network system, with

the main objective to minimize pressure and consequently leakage levels is

developed by Araujo et al (2006). EPANET model (Rossman 2000) is used

for hydraulic network analysis and Genetic algorithm optimization method for

pressure control and leakage reduction. A case study is presented is used to

show the efficiency of the system by pressure control through valves

management.

Alvisi and Franchini (2009) have presented a procedure for optimal

medium-term scheduling of rehabilitation and leakage detection interventions

in a water distribution system given predetermined budget constraints. The

objectives are to minimize the lost volumes of water and break repair costs.

The optimizer used is the NSGA II multi-objective genetic algorithm.

2.2.2 Pumping Mains Operations

The problem of daily controlling a water distribution network,

including pumping devices, and storage capacities, in order to supply the

consumers at the lowest cost is addressed by Joalland and Cohen (1980).

Discrete Dynamic Programming approach is used to solve the problem.

20

Walski (1982) made an economic comparison of the cost between the lining

of main and the associated savings in the energy costs of pumping.

Chen (1988) considered a network without tanks and determined

the optimal allocation of supply between the pump sources. Dynamic

programming approach was used to select the actual pumps given the optimal

continuous outflows.

Kim and Mays (1994) described how to minimize the pumping

costs by including the rehabilitation action for each main of the hydraulic

model as a decision variable. This system cost, subject to a hydraulic

constraint formulation is minimized using a nonlinear optimization package.

Klempous et al (1997) presented a multilevel two algorithms for finding

optimal control in a static water distribution system based on the idea of

aggregation technique. The first is a simulation algorithm of the pipeline

network and the other is an algorithm for finding an optimal control at the

pumping station.

A hybrid expert system called EXPLORE has been developed by

Leon et al (2000), to manage the Seville City water supply system.

EXPLORE reduces the cost of the operation of pumping the water to the

different storage tanks. For this task EXPLORE employs the water demand

forecast to obtain an optimal daily pumping schedule. A prototype has been

tested and has provided great improvement in the electrical costs.

Sakarya and Mays (2000) used a mathematical programming

approach for determining the optimal operation of water distribution system

pumps with water quality considerations. The methodology is based upon

describing the operation as a discrete time optimal scheduling problem that

can be used to determine the optimal operation schedules of the pumps in

distribution systems.

21

Cembrano et al (2000) proposed an optimal supervisory control

system which can be used as an efficient means of scheduling water transfer

operations to achieve management goals, such as cost minimization and

quality improvement. The case study presented showed that significant

savings may be achieved by using the optimal control procedures to compute

the strategies for pumping and water transfer operation.

An optimal pump scheduling problem in water supply systems is

addressed by McCormick and Powell (2003). Medium term maximum

demand policies are assumed to be represented in daily scheduling by

constraints on power use or by penalty costs. The problem is formulated as a

Dynamic Program in which variations in daily demand for water are modeled

as a Markov process.

Zyl et al (2004) proposed the use of a hybrid Genetic Algorithm for

operational optimization of water distribution systems. Two hill-climber

strategies, the Hooke and the Jeeves and Fibonacci methods are investigated.

The objective is to find the optimal operating strategy to provide an

acceptable level of service to the customer within system constraints, while

minimizing the operational cost.

Farmani et al (2005) investigated the application of multi-objective

evolutionary algorithms to the identification of the payoff characteristics

between total cost and reliability of a water distribution system. The pipe

rehabilitation decisions, tank sizing, and pump operation schedules are the

decision variables considered.

Broad et al (2010) have proposed the application of meta-models,

which can act as a surrogate or substitute for simulation models, for optimal

operation of water distribution networks. The study considered average daily

22

pumping costs and chlorine costs and demonstrated the effectiveness of the

approach by applying it to an actual distribution network.

2.3 RELIABILITY OF WATER DISTRIBUTION SYSTEM

2.3.1 Failure Analysis and Prevention

The deterioration of pipes may be classified into structural

deterioration which diminishes their structural resiliency and their ability to

withstand various types of stress, and internal deterioration resulting in

diminishing hydraulic capacity, degradation of water quality. The

deterioration processes as well as pipe structural failure models are therefore

very complex and difficult to model. Although significant work has been done

in modeling the physical process of pipe deterioration and failure (Doleac

et al 1980, Ahammed and Melchers 1994, Rajani et al 1996), the complex

processes, lack of pertinent data and highly variable environmental conditions

posed severe challenges to these research efforts and a comprehensive model

is required.

Damelin et al (1972) considered water supply pump inter-failure

times and repair times as random variables, and assumed them to be

exponentially distributed and lognormally distributed respectively. They

studied pumps with different capacities and presented statistical data on mean

time to failure (MTTF) and mean time to repair (MTTR). The failure data

were based on inter-arrival times of working hours, not including times when

the pumps were inoperative due to scheduled outages for maintenance.

Shamir and Howard (1981) used these data for computing mean annual

number of failures, presuming that pump operates 8400 h per year with some

20 to 44 h per month for preventive maintenance and other scheduled outages.

Shamir and Howard (1979) applied regression analysis to obtain a

relationship for the breakage rate of a pipe as a function of time. Walski and

23

Pellicca (1982) developed an exponential pipe failure prediction model based

on investigation of the correlation between pipe age and break frequency.

They adopted Shamir and Howard’s (1979) model for predicting break rates.

O’Day (1982) has presented an overview of the various causes of

water-main breaks and leaks, and listed the rate of water main breaks in

15 U.S. cities. Clark et al (1982) suggested a model that combines two

equations, one to predict the time of first break and the second to predict the

number of subsequent breaks, which are assumed to grow exponentially over

time in an attempt to account for the relative impacts of various external

agents. Kettler and Goulter (1985) proposed regression equations for the

number of breaks versus diameter and time for cast iron and asbestos-concrete

mains. They observed a linear relationship between pipe breaks and age.

Andreou et al (1987) used two different approaches for modeling a pipe

failure pattern: a proportional hazard model for early states of deterioration,

and a Poisson-type model for later stages.

Quimpo and Shamsi (1991) showed that the time of failure of water

pipes follows the exponential probability density function. Many failure

probability models have also been proposed to relax the assumption of

constant failure rate. Goulter et al (1993) further examined pipe breakage

patterns with respect to both time and space. They developed a methodology

for quantifying the variation in pipe failure rates associated with temporal and

spatial clustering of water-main failures.

A practical way of assessing the impact of various pipe failure

conditions on water distribution networks is described by Jowitt and

Xu (1993). The method assesses the vulnerability of the network to the loss of

any particular pipe element, and provides a quantitative estimate of the impact

on each nodal demand, and the post-failure utilization of nodal sources and

24

pipe elements. The results of the method can be combined with pipe failure

probabilities to provide measures of network reliability.

Constantine and Darroch (1993) proposed a time-dependent

Poisson distribution, in which the cumulative number of breaks in the pipe is

a power function of the pipe age. Herz (1996) developed a new probability

distribution function to be applied in a cohort survival model to an entire

stock of pipes in a distribution system. Deb et al (1998) applied the

Herz (1996) model to several water distribution systems.

Mays (2000) reported current and future needs in the analysis of

water distribution system reliability along with a review of concepts,

techniques, and methodologies for the evaluation of water distribution

systems. The use of advanced data mining methods in order to determine the

risks of pipe bursts is proposed by Babovic et al (2002). The database of

already occurred pipe bursts has been used to establish a risk model as a

function of associated characteristics of bursting pipe (its age, diameter, pipe

material etc.), soil type in which a pipe is laid, and traffic loading.

Various studies (Rajani and Makar 2000, Katano et al 2003) have

reported different methodologies used to predict the lifespan of metallic

underground pipeline networks and all of them are dependent on data related

to corrosion attack of the pipes. Visual inspection of the metallic surface,

characterization of the pitting attack, and metallurgical analysis are required

to complete the data necessary to establish a data driven pipeline maintenance

program (Doyle and Grabinsky 2003, Srikanth et al 2005).

Misiunas et al (2005) have proposed a new continuous monitoring

approach for detecting and locating breaks in water pipelines. The continuous

monitoring technique uses a modified two-sided cumulative sum (CUSUM)

algorithm to detect abrupt break-induced changes in the pressure data.

25

Several strategies have been evaluated to obtain appropriate

predictions of pipeline condition and consequently to establish where to invest

maintenance resources to avoid failures (Doglione and Firrao 1998, Babovic

et al 2002, Sadiq et al 2004, Tesfamariam et al 2006). The lack of precise

information about failure causes and pipeline conditions are among the

primary difficulties associated with proper pipeline maintenance (Saegrov

et al 1999, Babovic et al 2002, Sadiq et al 2004).

Restrepo et al (2009) evaluated the current condition of a water

distribution system and a large amount of data for both the soil and the

corresponding pipeline section were collected. Statistical techniques have

been used to establish the sampling design and a mathematical expression for

pitting depth as a function of several experimental variables. The approaches

and techniques employed provide a useful tool for structuring maintenance

programs.

2.3.2 Reliability Analysis

Reliability analysis of a water distribution system is concerned with

measuring its ability to meet consumer’s demands in terms of quantity and

quality, under normal and emergency conditions. In the past, various concepts

of reliability and approaches for reliability assessment have been proposed

and used for analyzing the reliability of water distribution networks. The

analytical methods developed for reliability analysis of water distribution

networks considered two states: working condition and shutdown condition.

Kettler and Goulter (1985) introduced reliability constraint – the probability

of breakage should not exceed a specified acceptable level- in a least-cost

network design model. Mays (1985) reviewed many analytical techniques for

reliability estimation and concluded that cut-set method and path enumeration

method were most promising.

26

Based on the topology of the network, Goulter and Coals (1986)

introduced the probability of simultaneous failure of all links connected to a

node, which is termed as the probability of node isolation. Su et al (1987)

used minimum cut-set approach to obtain nodal and system reliabilities and

integrated these reliabilities in the least-cost design of water distribution

networks.

A review of reliability considerations and measures for water

supply systems is presented by Wagner et al (1988 a). Two probabilistic

reliability measures, reachability and connectivity, are explored for use in

water distribution systems. An algorithm is developed for the calculation of

this measure, which combines a capacitated network with a method to

efficiently search through network configurations involving multiple link

failures.

Wagner et al (1988 b) also presented simulation as a complementary

method for analyzing the reliability of water distribution networks. It is found

that simulation method enables computation of much broader class of

reliability measures than do analytical methods and also allows the detailed

modeling of the hydraulic behavior of the system.

Jacobs and Goulter (1988) made evaluation of various graph theory

approaches for their applicability to reliability analysis of water distribution

networks. Bao and Mays (1990) proposed a methodology to estimate the

nodal and hydraulic reliabilities of water distribution system that account for

uncertainties. The method is based on a Monte Carlo simulation which can be

used for the assessment of existing systems, the design of new systems or the

expansion of existing systems.

A least-cost methodology is presented by Ormsbee and

Kessler (1990) for use in upgrading existing single-source water-distribution

27

networks in order to sustain single component failure. The methodology is

developed by casting the network-reliability problem in terms of an explicit

level of system redundancy.

Deshpande (1990) presented an analytical method based on loss-of-

demand probability for reliability analysis of pumping systems. Duan and

Mays (1990) proposed a new methodology for the reliability analysis of

pumping stations for water supply systems. The methodology considers both

the mechanical failure and hydraulic failure and models the availability

capacity of a pumping station as a continuous-time Markov process.

Walski (1993) summarized the state-of-the-art in providing

reliability in water distribution and gave some practical tips for improving

reliability. The measures to improve reliability, design considerations,

operation and maintenance considerations are discussed. A review and new

concepts for incorporating reliability in optimal design of water distribution

networks is presented by Ostfeld and Shamir (1993). Methods based on

decomposition, chance constrained approach and graph theory procedures are

discussed.

The concept of hydraulic reliability has been widely used in

determining system reliability (Hobbs and Beim 1988, Duan and Mays 1990,

Fujiwara and Silva 1990, Fujiwara and Tung 1991, Fujiwara and

Ganesharajah 1993, Gupta and Bhave 1994, 1996, Prasad and Park 2004).

Taking into consideration the hydraulic requirements, the basic concept of

system reliability commonly perceived is that the water should be provided

from sources to each demand point at the desired time, at the desired pressure

and at the desired flow rate.

Kansal et al (1995) have emphasized the need of computing

mechanical reliability of the water distribution system and its various

28

components that are subjected to breaks or failures. In the connectivity

analysis presented by Yang et al (1996 a), the mechanical reliability of a water

distribution network is determined under a static condition. Yang et al (1996 b)

also applied stochastic simulation in a reliability analysis of a regional water

distribution network. The proposed simulation technique is aimed at

evaluating the impacts of component failures on meeting demand at a certain

quantity level.

Quimpo and Wu (1997) suggested a capacity-weighted reliability

surface as a tool to assess the condition of deteriorating water supply

infrastructure. The method considered the reliability to meet demands at

nodes to depend on the reliability and hydraulic capacity of all the network

elements leading from the sources.

Gargano and Pianese (2000) presented a methodology to evaluate

the reliability of water distribution systems that can be used in the design

phase and for identifying repair works to be carried out on existing systems.

The system’s overall reliability (mechanical and hydraulic) is estimated using

the overall reliability index, which is defined by the weighted mean of the

hydraulic performance indices obtained for various operating conditions.

Ostfeld et al (2002) have demonstrated an application of stochastic

simulation for reliability analysis of water distribution systems, taking into

account the quality of the water supplied, as well as hydraulic reliability

considerations. The model is limited to one random component failure at a

time, uniform time to failure and time to repair probability distribution

functions.

Kalungi and Tanyimboh (2003) presented a quantified measure for

assessing the redundancy of water distribution networks. The importance of

adopting both redundancy and reliability as the basis foe assessing system

29

performance has been demonstrated. Both performance indices are calculated

using a method based on head driven simulation, which combines the

probability that components of the network are operational at any given time

and the network’s ability to meet consumer demands.

Tolson et al (2004) demonstrated the application of Genetic

Algorithms (GA) for reliability-based optimization of water distribution

networks. The approach links genetic algorithm as the optimization tool with

the First-Order Reliability Method (FORM) for estimating network capacity

reliability. The correlations between nodal demands are shown to significantly

increase distribution network costs designed to meet a specific reliability

target.

2.3.3 Availability Assessment

The availability of the pipe network is an important criterion when

establishing maintenance policy in water distribution networks. Depending on

the network under consideration, water authorities might select either

maintenance cost or availability/reliability as their objective in the

maintenance policy. However, for primary levels of water distribution

networks, which include large pipes with very high supply capacity, pipe

breaks might cause severe supply deficiency and the use of availability

measure seems more appropriate.

Cullinane et al (1992) have presented a practical measure for

distribution system reliability, based on hydraulic availability and

incorporated in an optimal design procedure for component sizing. The

measure combines hydraulic and mechanical availability in a form that

defines the proportion of the time that the system will satisfactorily fulfill its

function. Li and Haimes (1992 a) have considered this objective in the

30

proposed model, where the deterioration behavior of a pipe is represented by

the use of a semi-Markov process.

Li and Haimes (1992 b) proposed the use of system availability,

which is defined as the weighted sum of nodal availabilities as an objective

function to find the optimal repair/replacement policy based on a set of

predefined maintenance actions at all states of a deteriorating network. The

model proposed is difficult to apply in actual water distribution networks due

to the complex relationship between nodal availability and availability of a

network’s pipes, especially when the network is rather large.

Lansey et al (1992) summarized various factors which affect the

performance of the water distribution system and also discussed about various

alternatives for the maintenance of the piping network. Reliability analysis on

two water distribution networks in Arizona by linking reliability model to a

steady state simulation model to calculate the availability of the network is

presented.

Reliability and availability analysis are performed by Shinstine

et al (2002) on two large-scale municipal water distribution networks in the

Tucson Metropolitan area in Arizona. Reliability is computed using a

minimum cut-set method and simulation model. Availability is defined as the

proportion of time the system adequately satisfies the nodal demands and

pressure heads for various possible pipe breaks in the distribution system.

A mathematical model is developed by Luong and Nagarur (2005)

that aims to support the decision to allocate funds among pipes of the water

distribution network as well as the decision to repair or replace the pipes in

the state of failure. The objective function of the model is to maximize the

total weighted long-run availability of the whole system. The concept of

hydraulic reliability is employed to determine the weight of pipes in the

31

maintenance program. The deterioration behavior of the pipe is depicted by a

semi-Markov process.

2.3.4 Risk Assessment and Control

Deb et al (1998) have proposed a framework called ‘crisis model’,

a group of prioritized models in which renovation program is determined

based on the risk of pipe breaks and the critical effects of these accidents on

the network.

Environmental assessment and Geographic Information System

(GIS)-based models are used to evaluate the impact or performance of

different phenomena on the surrounding environment. Recently, Vairavamoorthy

et al (2007) have developed a GIS-based risk analysis tool for simulation and

management of water distribution networks.

Lim et al (2010) considered an urban water infrastructure

optimization problem to reduce environmental impacts and costs. The

proposed mathematical model decreases average concentrations of the

influents supplied for drinking water, which can improve human health and

hygiene; total consumption of water resources, reducing overall

environmental impacts and the life cycle cost.

2.4 MAINTENANCE OF WATER DISTRIBUTION SYSTEM

2.4.1 Rehabilitation

Since the performance of the water distribution system depends on

the performance of every single pipe, the decision on pipe rehabilitation or

renewal should consider the individual pipe in the context of the network

performance. The literature provides a variety of models that can assist the

decision maker in scheduling rehabilitation of a water distribution system.

32

Woodburn et al (1987) combined a nonlinear programming procedure with a

hydraulic simulation program in a model designed to determine which pipes

should be replaced, rehabilitated or left alone in order to minimize cost.

The various techniques for installation, renovation and

rehabilitation of water distribution systems have been reviewed by Marshall

et al (1990). A detailed account of the development of new methods is

presented. Male et al (1990) addressed the structural deterioration of water

mains over time. A procedure is applied to New York water distribution

system that incorporates the costs of replacing water mains and those of

repairing main breaks to determine least-cost rehabilitation planning strategies. A

net present value analysis is used to investigate the best replacement policy.

A nonlinear optimization model is proposed by Lansey et al (1992)

for optimal scheduling of maintenance for water distribution systems

encompassing a wide range of alternatives. The methodology considered

major piping alternatives of replacing and cleaning, and relining. It also

considered the potential of pumping improvements while accounting for the

costs of maintenance, failure and operations for a multiple-period planning

horizon.

Kim and Mays (1994) proposed a Branch and Bound scheme to

improve the Su et al (1987) model. Arulraj and Rao (1995) introduced the

concept of Significance Index (SI) which is an optimality criterion that can be

applied heuristically to prioritize pipe rehabilitation. Halhal et al (1997)

identified the rehabilitation problem as a multi-objective one, the two

objectives being minimizing cost and maximizing benefits. The benefits to the

system included economics, flexibility, and water quality. A cost-based

approach is employed by Kleiner et al. (1998) to construct a long-term

rehabilitation strategy for water distribution networks.

33

A survey of research needs and on-going efforts in the area of

rehabilitation of water networks is presented by Saegrov et al (1999). The use

of statistical methods for estimating existing and future rehabilitation needs

and the use of software tools for prioritizing actions have been discussed. The

development on technologies for detecting leaks and for measuring pipe wall

thinning is commented.

Engelhardt et al (2000) presented a literature review on the

rehabilitation strategies for water distribution networks. Numerous

rehabilitation decision making approaches have been discussed. The review

emphasized the need to operate water networks efficiently based on a

rehabilitation strategy in order to minimize the total life-cycle cost.

UtiNets, a Decision Support System (DSS) for rehabilitation

planning and optimization of the maintenance of underground pipe networks

of water utilities, is developed by Hadzilacos et al (2000). The DSS performs

reliability based life predictions of the pipes and determines the consequences

of maintenance.

A dynamic Programming approach, combined with partial and

implicit enumeration schemes, has been used by Kleiner et al (2001), to

determine the most cost-effective plan in terms of what pipes in the water

network to rehabilitate, by which rehabilitation alternative and at what time in

the planning horizon, subject to the constraints of service requirements.

Clark et al (2002) have proposed equations that can be used to

estimate the cost of system construction, expansion, and rehabilitation and

repair for drinking water infrastructure. Equations have been developed to

estimate the cost of cement mortar lining, slip lining and corrosion control.

Selvakumar et al (2002) discussed the various types of technologies that can

be used for rehabilitation and repair of drinking water distribution

34

components. A representative costs that can be used by utility managers to

estimate order-of-magnitude budgetary costs for rehabilitation and

replacement of distribution system components are also presented.

A procedure based on a multi-objective genetic algorithm to search

for a near-optimal rehabilitation scheduling is proposed by Alvisi and

Franchini (2006). With reference to a fixed time horizon, the goal is to

minimize the overall costs of repairing and/or replacing pipes, and to

maximize the hydraulic performances of the water network.

A few methodologies for the prioritization of maintenance actions

in relation to several pipes within a water distribution network are mentioned

(Quimpo and Shamsi 1991, Arulraj and Rao 1995, Quimpo and Wu 1997). A

new prioritization approach to the problem of rehabilitation of water

distribution systems is presented by Saldarriaga et al. (2010). The proposed

algorithm selects the most important pipes to be replaced in order to achieve

two main objectives: first, to reduce loss of water due to leakage and second,

to reduce the dissipated energy in the system, making it more efficient and

reliable.

Tabesh et al (2010) presented a methodology to manage the

rehabilitation and replacement of water distribution network using hydraulic

and geospatial information systems models. The proposed indices consist of

pipe breaks and leakage analysis, hydraulic and quality performance and

mechanical reliability of the network. A novel approach is also introduced to

calculate leakage values throughout the network.

2.4.2 Strengthening and Expansion

When an existing water distribution network is deficient, several

pipes together would need strengthening. The literature related to

35

strengthening of an existing network to enhance its capability; and its

expansion to cover additional localities is presented below.

Morgan and Goulter (1985) proposed a heuristic linear

programming based procedure for the least cost layout and design of water

distribution networks. The methodology is capable of wide range of demand

pattern and pipe failure combinations. The procedure can also be extended for

use in the expansion or reinforcement of existing network systems.

Bhave (1985) used a criterion to increase the capability of the

network by optimally selecting links to be strengthened by parallel piping and

expansion of the network to cover additional localities.

Boulos and Wood (1990) proposed a method for determining

explicitly different parameters for upgrading and enhancing water distribution

networks. The method can directly determine a variety of design parameters

such as pipe diameter, pipe length, pump power, pump head, storage level and

valve characteristics; and operating parameters such as pump speed, control

valve setting and specified flow or pressure requirements.

Martin (1990) proposed a Dynamic Programming based pipeline-

capacity expansion model minimizing the cost of initial construction and

subsequent capacity expansion. The proposed design algorithm has an ability

to specify the number and size of pump stations, and the length, diameter and

pressure class of pipes to be added at the beginning of each staging interval

over the design period.

WADISCO (Water Distribution Simulation and Optimization)

computer software is developed by Walski et al (1990) which can be used for

optimal design or strengthening and expansion of a network. The algorithm

enumerates all possible pipe size combinations within the user specified size

36

ranges and tests them for feasibility against pressure requirements. The

program can handle any number of links and nodes; booster pumps and check

and pressure reducing valves.

Stochastic search method is applied to optimal strengthening and

expansion of existing networks by Savic and Walters (1997) using genetic

algorithm. Two network examples, one of a new network design and one of

parallel network expansion, illustrate the potential of the proposed computer

model GANET as a tool for water distribution network planning and

management.

Barros et al (2008) developed an optimization model for the

management and operation of a large-scale, multi-reservoir water distribution

system with preemptive priorities. When the water supply is insufficient to

meet the planned demand, appropriate rationing factors are applied to reduce

water supply. The model and its user-friendly interface form a decision

support system, which can be used to configure a water distribution system to

facilitate capacity expansion and reliability studies.

2.4.3 Repair and Replacement

A few analytical methods have been published to assist in making

pipe replacement and repair decisions. Shamir and Howard (1979) derived an

equation which computes the optimal replacement time of a water main by

equating the expected maintenance cost for repairs and the replacement cost.

O’Day (1982) presented a similar analysis to determine whether a pipe should

be relined to extend its useful life or replaced. Walski (1982) developed a

relation between the expense of increasing pump capacity and its energy

requirement, and the cost of relining a water main to determine which of the

alternatives is more cost effective.

37

The options available to utilities when trying to correct for

excessive pipe breakage are limited; they can continue to repair the breaks

as they occur or replace the offending main segment. Quimpo and

Shamsi (1991) proposed a strategy for prioritizing decisions for the

maintenance of a water distribution system. The specific components that

must be repaired or replaced are determined using a component importance

criterion that measures the overall effect of component maintenance on the

system reliability. The procedure is applied to one hypothetical and two real-

life water distribution systems.

Li and Haimes (1992 a) have developed a Semi-Markovian model

to capture the dynamic evolution of the failure mode of a deteriorating main

pipe, thus facilitating the determination of the optimal replacement/repair

decision at various deteriorating stages. An example problem is solved to

demonstrate the proposed methodology and to show the trade-off between the

system availability and the expected maintenance cost.

A comprehensive review of pipe replacement analysis is presented

by Loganathan et al (2002), from which an economically sustainable

threshold break rate for replacement of pipelines in deteriorating water

distribution systems is derived. Design charts to determine the optimal

threshold break rate as a function of pipe diameter and discount rate are also

presented.

Dandy and Engelhardt (2006) demonstrated the use of Genetic

algorithms to generate trade-off curves between cost and reliability for pipe

replacement decisions. Curves for two planning scenarios are generated. The

first identifies the trade-offs necessary for the current conditions. The second

allows the water authority to determine the required levels of future

expenditure, given funding constraints, to meet a specified level of service

over the entire planning horizon.

38

A framework for devising a short-term decision support tool for

pipe replacement is developed by Giustolisi and Berardi (2009). The approach

considered economic, technical, and management rationales as separate

objectives to produce a pipe-wise prioritization scheme which is achieved by

ranking pipes selected during a multi-objective evolutionary optimization of

replacement scenarios.

2.5 OBSERVATIONS MADE FROM THE REVIEW

The literature survey on the various Water Distribution System

(WDS) optimization models and the review of topics related to the design,

operation, reliability and maintenance of water distribution systems reveal the

following facts:

A large body of literature exists on the water distribution

system optimization, reporting the application of classical

optimization methods such as linear programming (Goulter

and Morgan 1985), non linear programming (Li and

Haimes 1992 b) and dynamic programming (Kleiner

et al 1998). On most occasions, models are deliberately

simplified to make the application possible. Some of them

have been linearized to facilitate the use of linear

programming. In others, the discrete nature of the variables is

disregarded, remaining non linear but no longer combinatorial

(Lansey et al 1992). Deterministic methods were unable to

cope up with the non linear water distribution network

problems. The stochastic optimization algorithms are quite

successful in solving such problems, though requiring large

number of evaluations (Mohan and Babu 2010).

39

Literature reveals that many researchers have addressed the

water distribution system design optimization problems.

Several studies have considered the hydraulic

availability/reliability as a performance measure for the

optimal design of water distribution systems (Bhave 2003).

All these studies assumed that a water main was structurally

sound which is not true in real-life water distribution systems.

Surveys carried out in different countries reveal that

35 – 60 % of the water is wasted in leakages in the pipelines

(Babovic et al 2002). In the literature, there appears to be only

a few decision models that address water distribution system

maintenance problems.

It is observed from the literature that the maintenance policies

in most of the proposed maintenance models are evaluated and

compared with respect to cost-effectiveness and hydraulic

reliability characteristics (Luong and Nagarur 2005, Dandy

and Engelhardt 2006). The studies on the maintenance of

water distribution systems that consider infrastructure

availability as a performance measure are not much reported

in the literature.

Several WDS maintenance models available in the literature

have investigated only pipe replacement decisions and did not

include other maintenance/rehabilitation options (Male

et al 1990, Lansey et al 1992). Also to simplify the analysis,

the replacement problems considered are limited to the case of

each pipe being replaced with the same diameter (Dandy and

Engelhardt 2006). Moreover, the studies which considered

rehabilitation problems with decision variables like tank sizing

40

and number of pumps did not consider time value of money

(Farmani et al 2005).

Actually, different failure characteristics in various pipes will

also have an effect on the optimal maintenance policy. This

aspect is not considered (Luong and Nagarur 2005). Though a

few models have incorporated failure probabilities of pipes in

the system, it is found that the maintenance models reported in

the literature have not considered the failure characteristics of

the junction joints in the pipe network and also the failure

behavior of the pumping components in the system, which

would lead to the disruption of water supply to the consumers.

The cost terms which are relevant to the WDS maintenance

problem are usually described by a set of equations, which are

non linear. The development of powerful computing systems

has enabled to carry out simulation studies for the distribution

network to evaluate the maintenance alternatives. Nevertheless,

determination of optimal maintenance strategy even for simple

networks remains difficult. To keep close to reality, this

problem must be considered in combinatorial form, whose

resolution in real-world conditions, via classical optimization

techniques, will, in general, be difficult. The use of stochastic

search algorithms to solve WDS maintenance optimization

problem has been seldom investigated in the literature.

In this study, the application of stochastic search techniques such as

Simulated Annealing, Tabu Search and Genetic Algorithm to determine the

near-optimal maintenance strategy for a real-life urban water distribution

system has been demonstrated. The deficiencies in the maintenance decision

models reported in the literature are overcome in this study.