Delivering water in sufficient quantity and acceptable quality is the main objective of water distribution networks (WDN) and at the same time is the main challenge. Many factors affect the delivery of water through distribution networks. Some of these factors are relevant to water quality, quantity and the condition of the infrastructure itself. The deterioration of water quality in the WDN leads to failure at the water quality level, which can be critical because it is closest to the point of delivery and there are virtually no safety barriers before consumption. Accordingly, developing a powerful monitoring system that takes into consideration water demand distribution, the vulnerability of the distribution system and the sensitivity of the population to the deterioration of water quality can be very beneficial and, more importantly, could save lives if there was any deterioration of water quality due to operational failure or cross-contamination events. In this paper, a framework for a water quality monitoring system that considers water demand distribution, the vulnerability of the system and the sensitivity of the population using fuzzy synthetic evaluation and optimization algorithms is developed. The proposed approach has been applied to develop a monitoring system for a real WDN in Saudi Arabia.

INTRODUCTION

Water transported by a distribution network usually passes through a complex treatment process before its distribution. Once the treated water is pumped into the network, there are usually no defensive lines that could protect consumers from any deterioration of water quality due to any unexpected consequence that happened during the transportation. These consequences might be gradual, such as a variation in chlorine levels, or rapid, such as contamination caused by intentional criminal or terrorist attacks. Many factors could affect water quality within water distribution networks (WDN) directly or indirectly. Some of these factors are relevant to the infrastructure of the system, such as pipe materials, age and breaks. Some are related to the operational practices of the network, such as the water age in the network. To prevent, control and/or reduce any potential deterioration of water quality and to ensure that the WDN is functioning the way it should, constant and comprehensive assessment and monitoring of the WDN should be established.

In 1990 and 1992, Lee and Deininger were the first to develop a scientific criterion to determine representative water quality monitoring locations by developing a coverage method concept. Since then, several improvements have been made to the original approach developed by Lee & Deininger (1992) to reduce the complexity by considering steady and extended period simulations for water demand, adding key parameters, including water age and pipe diameters, and considering rapid water quality deterioration rather than considering only gradual changes in water quality. Although most of the studies that dealt with identifying the optimal locations of monitoring stations (MSs) in WDN have used water demand as the key parameter for locating the MSs (Lee & Deininger 1992; Kumar et al. 1997; Kessler et al. 1998; Liu et al. 2011, 2012), other approaches have used other parameters such as water age rather than considering the demand (Tryby & Uber 2001). In addition, there were several objectives used to determine the optimal locations of MSs, such as minimizing the percentage of population at risk due to sudden contamination attack, minimizing the exposed population, minimizing the detection time of the contaminant, minimizing the contaminated water consumed, minimizing the number of times the system fails to detect a contaminant, and minimizing the distance that the contaminant would reach in the pipelines (Berry et al. 2004, 2005). In general, integer programming was applied to locate the MSs (Lee & Deininger 1992; Kumar et al. 1997; Berry et al. 2005; Liu et al. 2011); however, other optimization algorithms were also used, such as genetic algorithms, robust optimization and Bayesian belief networks (Al-Zahrani & Moied 2003; Grayman et al. 2006; Watson et al. 2009).

In most of the studies that dealt with identifying the optimal locations of MSs in WDN, water demand was the only or the key parameter. Moreover, factors such as the vulnerability of the system were not considered, despite its importance and relation to the contamination events. Furthermore, it was assumed that the population would be exposed equally to any deterioration consequences in the WDN. However, the sensitivity of the population to contamination events varies based on several factors such as age, standard of living and population density. In this study, an approach for determining the optimal locations of a monitoring system was developed considering water demand, the vulnerability of the system, and population sensitivity to water quality deterioration.

METHODOLOGY

Developing a comprehensive and representative water monitoring system consists of two main steps. The first step is prioritizing the risk for the different regions in the city based on the system vulnerability and population sensitivity to any deterioration of water quality. The second step is determining the optimal locations of MSs to maximize the covered demand (monitored demand), and takes into consideration the regional vulnerability and sensitivity risks. Detailed descriptions of the two phases are explained in the following paragraphs.

Step I: Risk prioritization

Risk prioritization requires an aggregation of measurable and non-measurable factors to estimate vulnerability, sensitivity and overall risk at specific points in the WDN using routinely collected data. Figure 1 provides the prioritization framework for determining vulnerability, sensitivity and overall risk indices. The framework consists of four levels or generations of factors, aggregated in a hierarchical fashion to develop different risk indices. The top of the pyramid represents the overall risk, which depends on two factors, i.e. vulnerability and sensitivity in the first level. These two factors are determined through aggregation of various factors in the previous levels. The vulnerability index is determined by aggregating factors related to hydraulics, structural integrity and the quality of the water pumped through the WDN. The population sensitivity index is linked to the presence of certain groups of consumers served by the WDN, who may suffer from various consequences if the water quality in the WDN deteriorates. These indices were developed for different regions in the WDN based on the local data for each region (Francisque et al. 2009).
Figure 1

Framework for the risk index prioritization of water mains.

Figure 1

Framework for the risk index prioritization of water mains.

Due to the complexity of the system, variation of the collected data (measurable and non-measurable) and difficulty in obtaining precise data for some of the factors, an index-based approach using fuzzy synthetic evaluation (FSE) was used to describe vulnerability, sensitivity and overall risk. An estimate of risk at a given location in the WDN will provide a representative value for a predefined geographical region. A region represents an influence zone in which the values of all contributing factors are assumed to be fixed. Once the risk values are determined in various sectors of the WDN, they can be ordered, ranked or prioritized based on the overall risk index.

Fuzzy rule-based modelling

The fuzzy set theory was first developed by Zadeh (1978) to methodically incorporate human reasoning in decision making. In FSE, variables are used as inputs for decision making, in which numeric or non-numeric data can be evaluated using natural linguistics such as ‘low’, ‘med’ and ‘high’ (Ross 2009). In this study, FSE is used to quantify and estimate indices for the hydraulics of the system, structural integrity, water quality, vulnerability, sensitivity and, consequently, risk index. The FSE follows the following steps (Sadiq & Rodriguez 2004; Sadiq et al. 2004a; Khan & Sadiq 2005; Francisque et al. 2009).

Fuzzification

Fuzzification can be defined as a process by which measurable and non-measurable input data are transformed into a homogeneous scale (0 to 1), or the process of changing crisp values into fuzzy values (Francisque et al. 2009; Ross 2009). When using different types of input data, which could be a mix of measurable (i.e. pH, total dissolved solids (TDS), etc.) and non-measurable (i.e. standard of living, activity, etc.) parameters, fuzzy logic is preferred over crisp logic, since crisp logic is limited to two classes only; True or False, 0 or 1, High or Low, etc. However, input data can be classified using a wide range of classes (for example: High, Med and Low) when using fuzzy logic. Therefore, for crisp logic a single input data can be a member of only one class, while it can belong to more than one class in fuzzy logic. This advantage gives more freedom in classifying different parameters and considering the fuzziness or uncertainity involved. Accordingly, due to the nature of the input parameters used in this study, fuzzy logic was used.

Triangular, Gaussian and trapezoidal fuzzy subsets are usually used to fuzzify input data (Wang 1992; Lee 1996; Sadiq et al. 2004a, 2004b; Francisque et al. 2009) as shown in Figure 2. Using these fuzzy subsets, input data can be mapped to fuzzy sets. In this study, triangular and trapezoidal membership functions were used. For example, suppose Figure 3 shows a fuzzy set for the TDS level in water. There are three fuzzy subsets, low (μlow), med (μmed) and high (μhigh). If the TDS in water is 200 ppm, then it is said that the TDS level has a membership of 1 to the ‘low’ fuzzy subset, and if the level is 800 ppm, then it is said that the TDS level has a membership of 1 to the ‘high’ fuzzy subset. If the TDS level is 450 ppm, then it is said that it has a membership of 0.25 to the ‘low’ fuzzy subset and 0.75 to the ‘med’ fuzzy subset.
Figure 2

Membership functions: (a) Triangular and (b) Trapezoidal.

Figure 2

Membership functions: (a) Triangular and (b) Trapezoidal.

Figure 3

TDS membership function.

Figure 3

TDS membership function.

To fuzzify the attributes used in this study, four types of membership functions were used, as shown in Figure 4, three of which are triangular and one trapezoidal. The thresholds (A, B, C, D, E and F) for these membership functions (as shown in Table 1) were developed from water quality standards published in literature and local operational standards (AWWA 2002; Sarbatly & Krishnaiah 2007; Gupta 2008; WHO 2008;Francisque et al. 2009; USEPA 2009).
Table 1

Fuzzy set thresholds and types

Thresholds
 
ParameterNature of fuzzy setABCDEFType
TDS Data (ppm) 300 500 700 ∞  Triangular 1 
Temperature Data (C°) 20 25 30 100  Triangular 1 
pH Data 5.5 6.5 8.5 9.5 14 Trapezoidal 
Free chlorine Data (ppm) 0.2 0.3 1.2 1.3 ∞ Trapezoidal 
Turbidity Data (NTU) 0.5 0.8 ∞  Triangular 1 
Pipes type Percentage of badness 25 50 75 100  Triangular 3 
Potential industrial intrusion Percentage by area 25 50 75 100  Triangular 3 
Pipe age (Option 2) Average age 20 30 40 60  Triangular 2 
Pipe break (Option 3) Breakage ratio 0.25 0.5 0.75  Triangular 3 
Schools No. of elementary students 709 1,418 2,835 ∞  Triangular 1 
No. of intermediate students 317 633 1,266 ∞  Triangular 1 
No. of secondary students 273 546 1,092 ∞  Triangular 1 
Hospitals No. of beds 40 80 120 160 ∞ Triangular 2 
Pressure Nodes with low and high pressure (%) 25 50 100   Triangular 2 
Optimal diversions (%) 25 50 100   Triangular 2 
Velocity Pipes with low and high velocity (%) 25 50 100   Triangular 2 
Water age Nodes with high water age 25 50 100   Triangular 2 
Population Population density 9,420 18,840 300,000   Triangular 2 
Sewer system coverage Percentage of area not covered by sewer system 25 50 75 100  Triangular 3 
Water table Dry-wet pipes (%) 25 50 100   Triangular 2 
Thresholds
 
ParameterNature of fuzzy setABCDEFType
TDS Data (ppm) 300 500 700 ∞  Triangular 1 
Temperature Data (C°) 20 25 30 100  Triangular 1 
pH Data 5.5 6.5 8.5 9.5 14 Trapezoidal 
Free chlorine Data (ppm) 0.2 0.3 1.2 1.3 ∞ Trapezoidal 
Turbidity Data (NTU) 0.5 0.8 ∞  Triangular 1 
Pipes type Percentage of badness 25 50 75 100  Triangular 3 
Potential industrial intrusion Percentage by area 25 50 75 100  Triangular 3 
Pipe age (Option 2) Average age 20 30 40 60  Triangular 2 
Pipe break (Option 3) Breakage ratio 0.25 0.5 0.75  Triangular 3 
Schools No. of elementary students 709 1,418 2,835 ∞  Triangular 1 
No. of intermediate students 317 633 1,266 ∞  Triangular 1 
No. of secondary students 273 546 1,092 ∞  Triangular 1 
Hospitals No. of beds 40 80 120 160 ∞ Triangular 2 
Pressure Nodes with low and high pressure (%) 25 50 100   Triangular 2 
Optimal diversions (%) 25 50 100   Triangular 2 
Velocity Pipes with low and high velocity (%) 25 50 100   Triangular 2 
Water age Nodes with high water age 25 50 100   Triangular 2 
Population Population density 9,420 18,840 300,000   Triangular 2 
Sewer system coverage Percentage of area not covered by sewer system 25 50 75 100  Triangular 3 
Water table Dry-wet pipes (%) 25 50 100   Triangular 2 
Figure 4

TDS membership function: (a) Triangular 1, (b) Triangular 2, (c) Triangular 3 and (d) Trapezoidal.

Figure 4

TDS membership function: (a) Triangular 1, (b) Triangular 2, (c) Triangular 3 and (d) Trapezoidal.

Aggregation

Aggregation is the process by which fuzzy sets representing the outputs for each parameter or element (child element) [μlow, μmed, μhigh] are combined or aggregated to produce a single output for the group of elements (parent fuzzy set output) (MathWorks 2012). Fuzzy sets produced from fuzzification for all elements and weights calculated for each element were used to determine the aggregated fuzzy set for the parent group using matrix multiplication (Sadiq & Rodriguez 2004; Francisque et al. 2009).

Defuzzification

The process by which fuzzy sets [μlow, μmed, μhigh] are converted to a representative crisp value is called defuzzification (Wang 1992; Francisque et al. 2009). There are several methods to defuzzify fuzzy sets; however, the weighted average method or scoring is preferred by many researchers, especially in environmental applications (Silvert 2000; Sadiq & Rodriguez 2004; Francisque et al. 2009). According to the weighted average method, to convert the fuzzy sets into a crisp value, each fuzzy set will be multiplied by a constant weight (a, b and c) and the product summation is the crisp value as follows: 
formula
1

Step II: Locating MSs

In this phase, the water demand as well as the risk indices produced in step I were used to determine the optimal locations of MSs using the demand coverage method (DCM). Basically, DCM assumes that if a certain percentage of water at node X is originally coming from node Y, then setting an MS at X will cover demand at Y if this portion of the water is equal to or greater than the coverage threshold (CT). In this study, CT was defined as 60%. In other words, if 60% or more of the water in node X is originating from node Y, then placing an MS at X will cover nodes X and Y. Defining these water pathways is a complex operation, and requires intensive modelling. Further details about DCM can be found in the literature (Lee 1990; Lee & Deininger 1992; Liu et al. 2011). The MSs with the highest demand and risk index are the optimal MSs.

The main objective of identifying the optimal locations of MSs in WDN is to increase the representativeness of the monitoring system and, consequently, maximize the monitored (covered) demand, taking into consideration the existing risks associated with delivering water to each region. Mathematically, the objective function for maximizing the coverage can be expressed as follows: 
formula
2
 
formula
3
subject to: 
formula
4
 
formula
5
where m is the number of total nodes covered by node i, di,j is the demand at node j covered by node i, Ij is the risk index for node j, xi is an integer value that determines if there is or is not an MS at node i, n is the number of total nodes in the network, MS is the maximum allowable number of MSs to be used for the network, and w is the number of potential monitoring stations (PMS) covered by node i.
Furthermore, a regional constraint was added to ensure that every region in the network will have at least one MS as shown in Equation (6). 
formula
6
where z is the total number of MSs in region A and xA,i is an integer variable xi for the nodes in region A. Figure 5 shows the flowchart summarizing the process of locating MSs.
Figure 5

Flow chart summarizing the procedure for locating optimal MSs.

Figure 5

Flow chart summarizing the procedure for locating optimal MSs.

APPLICATION

The presented approach was applied to a real WDN to enhance the existing monitoring system and to illustrate the execution of the proposed approach.

Study area

The developed approach was applied to Al-Khobar WDN, which is located on the eastern coast of Saudi Arabia. The city has an area of approximately 64 km2 with a population of about 580,000, which is expected to rise to approximately 814,000 by 2025 (MEP 2010). The total length of the network is approximately 472,652 m. Desalinated water from the Al-Aziziah plant and groundwater wells are the two main sources of water supply for the Al-Khobar WDN. Detailed information about the Al-Khobar WDN and its hydraulics has been reported by Al-Zahrani & Al-Ghamdi (2008). Figure 6 shows the skeleton of the Al-Khobar WDN.
Figure 6

Skeleton of Al-Khobar WDN.

Figure 6

Skeleton of Al-Khobar WDN.

RESULTS AND DISCUSSION

Risk prioritization

First, the Al-Khobar network was divided into several regions based on the locations of the existing water quality MSs. The Thiessen method was used to divide the WDN into different regions, as shown in Figure 7(a). Next, based on the collected data, the fuzzy set thresholds and types summarized in Table 1 are used to develop the risk indices.
Figure 7

Al-Khobar WDN: (a) sub-regions, (b) vulnerability risk index for water quality, (c) sensitivity of the regions and (d) total risk index for each region.

Figure 7

Al-Khobar WDN: (a) sub-regions, (b) vulnerability risk index for water quality, (c) sensitivity of the regions and (d) total risk index for each region.

For the hydraulic index, which includes pressure, velocity and water age, the analysis of these components showed that the hydraulic properties in general are within the acceptable limits. The average pressure for all demand scenarios for all the regions varies between 7 and 41 m. The highest pressure in the city usually occurred in the city center, especially at region 94 and its surroundings, since the main pumping station is located in this region. The average velocity for all demand scenarios ranges between 0 and 4.58 m/s. Although very few pipes have a velocity of 2 m/s or higher, the most frequent violations were velocities lower than 0.4 m/s. Unlike pressure and velocity, the water age is within the recommended range in all regions for all demand scenarios, which implies that water age will have no effect in the prioritization of risk and developing the hydraulic index. A summary of the highest and lowest risk indices for all regions is shown in Table 2.

Table 2

Indices for regions with the highest and lowest total risk

 Regions
Index747576777981829498102103104105119120121
Water quality 0.21 0.14 0.18 0.19 0.32 0.26 0.27 0.28 0.20 0.19 0.15 0.20 0.23 0.25 0.24 0.45 
Structural integrity 0.59 0.59 0.61 0.59 0.60 0.61 0.63 0.59 0.18 0.18 0.30 0.18 0.18 0.18 0.37 0.43 
Hydraulics 0.21 0.80 0.29 0.21 0.21 0.00 0.21 0.75 0.29 0.21 0.21 0.00 0.21 0.22 0.31 0.30 
Vulnerability 0.34 0.51 0.36 0.33 0.38 0.29 0.37 0.54 0.22 0.19 0.22 0.12 0.20 0.21 0.31 0.40 
Sensitivity 0.75 0.68 0.35 0.26 0.17 0.03 0.06 0.86 0.46 0.21 0.39 0.10 0.10 0.14 0.60 0.59 
Total risk 0.54 0.60 0.35 0.30 0.28 0.16 0.21 0.70 0.34 0.20 0.30 0.11 0.15 0.18 0.45 0.49 
 Regions
Index747576777981829498102103104105119120121
Water quality 0.21 0.14 0.18 0.19 0.32 0.26 0.27 0.28 0.20 0.19 0.15 0.20 0.23 0.25 0.24 0.45 
Structural integrity 0.59 0.59 0.61 0.59 0.60 0.61 0.63 0.59 0.18 0.18 0.30 0.18 0.18 0.18 0.37 0.43 
Hydraulics 0.21 0.80 0.29 0.21 0.21 0.00 0.21 0.75 0.29 0.21 0.21 0.00 0.21 0.22 0.31 0.30 
Vulnerability 0.34 0.51 0.36 0.33 0.38 0.29 0.37 0.54 0.22 0.19 0.22 0.12 0.20 0.21 0.31 0.40 
Sensitivity 0.75 0.68 0.35 0.26 0.17 0.03 0.06 0.86 0.46 0.21 0.39 0.10 0.10 0.14 0.60 0.59 
Total risk 0.54 0.60 0.35 0.30 0.28 0.16 0.21 0.70 0.34 0.20 0.30 0.11 0.15 0.18 0.45 0.49 

Infrastructure and the structural integrity of the WDN were judged based on pipe breaks, age and material, in addition to potential intrusions of wastewater or industrial waste should pipe breaks occur. The historical records of Al-Khobar municipality indicate that most pipe breaks occurred in the central and southern regions of the city, which are the oldest in the city (44 years). The main pipe material in the central and southern regions is asbestos, and PVC in the northern regions. Cracks and breakages have been widely reported in asbestos pipes. According to the Al-Khobar water authority, approximately 65% of the total breaks occur in asbestos pipes. On the other hand, asbestos is considered as a carcinogenic material, although its risk relevant to drinking water is not substantial. Accordingly, the infrastructures in the central and southern regions are not in good condition and might be a potential source of intrusions.

Pipe material, age, breaks, sanitary coverage, industrial and wastewater intrusions, water table levels and the soil surrounding the pipes give a clear view of the infrastructural condition of the WDN. It is obvious that the southern regions of the city have multiple issues. The structural integrity risk index ranges between 0.18 and 0.63 (Table 2). As expected, regions in the southern part of the city have a higher risk index compared to the city center and the northern part. Regions in the north, such as 98 and 119, have PVC pipes, relatively newer pipes, a low pipe breakage ratio and no industrial activity or potential intrusions (either industrial or wastewater), which explains the relatively low risk index in these regions. Regions in the south, such as 79 and 82, have asbestos pipes, older pipes, a high pipe breakage ratio and potential industrial intrusions, which explains the higher risk index compared to the northern regions. For other parameters (TDS, pH, free chlorine, temperature and turbidity), the analysis indicates that they are within the acceptable standards. Accordingly, a water quality risk index for each region was developed by aggregating the physico-chemical and microbial factors as shown in Table 2. It is obvious that the water quality risk is low in the WDN, since the water quality parameters are generally within the recommended standards. Water quality indices range between 0.14 and 0.43, which is a relatively low risk (Table 2).

The vulnerability index of the WDN is developed by aggregating the hydraulic properties, water quality and structural integrity. The major factors affecting vulnerability are hydraulic properties and structural integrity, since the water quality risk index is low all over the city. Regions which have a high risk due to hydraulic properties and structural integrity, i.e. regions 75 and 94, are the regions in the city center which have high vulnerability. Regions in the north, such as 98 and 102, have a low risk in terms of hydraulic properties and structural integrity, which is reflected in the vulnerability index. Regions in the extreme south of the city, such as regions 77 and 82, have a low risk index due to hydraulic properties, but they have a high risk index due to structural integrity, which explains why the vulnerability indices in that zone are higher than the northern regions, but at the same time less than the regions in the city center, as shown in Figure 7(b).

The sensitivity of the region gives a sense of the possible health risk consequences in case of any possible contaminant intrusion into the network. The standard of living, population density, activity, the number of beds in hospitals and the number of students are measures for regional sensitivity in case of any possible deterioration of water quality within the WDN. There are groups in the community who are more sensitive to any deterioration of water quality than other groups. Patients and school students, as an example, are more sensitive to waterborne illnesses, as their immune system is either weak or not fully developed (Francisque et al. 2009). Aggregation of all these factors determines the sensitivity of each region. It is obvious that regions 74, 75, 94, 120 and 121 are more sensitive than the other regions. These factors were aggregated to develop the sensitivity risk index for the different regions in the city as shown in Figure 7(c). The sensitivity index ranges between 0.03 and 0.86. Regions in the center, such as 74, 75, 94 and 121, are more sensitive to any deterioration of water quality in the WDN. Regions in the north and in the south, such as 81, 82, 102 and 119, have the least sensitivity, mainly because they have low population density, a low number of students and fewer residential areas.

Total regional risk is based on the aggregations of vulnerability and sensitivity of each region. Total risk indices, as shown in Figure 7(d), range between 0.11 and 0.70. In general, the results indicate that the northern and southern regions have the least risk index, which is a reflection of the generally low vulnerability and sensitivity of these regions. Regions in the city center are the most sensitive regions in the city, as shown in Figure 7(d). This explains the relatively high risk in the city center compared to the other regions in the city. Table 2 shows the detailed risk indices for all the regions.

Locating MSs

Currently, there are 15 MSs distributed over the Al-Khobar WDN as shown in Figure 8. In general, most of the MSs are located close to major pumping locations such as the Makkah tank, central pumping stations and pumping wells, to monitor water quality directly after pumping to ensure that chlorine levels and TDS are within the acceptable limits. However, this distribution of MSs ignores the fact that water quality can deteriorate during delivery through contaminant intrusions, cracks or due to intentional criminal and/or terrorist acts.
Figure 8

Locations of MSs at Al-Khobar WDN.

Figure 8

Locations of MSs at Al-Khobar WDN.

In the Al-Khobar WDN, there are 871 non-zero nodes, water demand nodes, which can be considered as PMSs and have an equal chance of being chosen during the optimization process based on the objective functions and constraints defined previously.

Risk prioritization using FSE showed that each region has different characteristics to the other regions, which makes the expected consequences in case of water quality deterioration vary for the different regions. Severe consequences are expected in some regions more than others, since these regions have higher risk (vulnerability and sensitivity). Therefore, demand and regional risk were used to optimally locate the MSs.

It was noticed that the demand coverage of the monitoring system increases as the number of proposed MSs increases, as shown in Table 3. Based on the current practices with 15 MSs, only 19.56% of the demand can be monitored; however, the coverage increased to 59.18% when 50 MSs were used. Figure 9 shows the optimal distribution of MSs when 15, 20, 30 and 50 MSs were proposed. Each region is monitored by at least one MS, as shown in Figure 9, with the exception of region 81, which is not populated and, accordingly, is not served by the distribution network. The figure indicates that more MSs are located in the central area of the city compared to the extreme northern and southern parts. The central part of the city is the most populated zone of the city, with higher demand and higher sensitivity. In addition, the risk assessment showed that the central regions are also vulnerable. Accordingly, the higher number of MSs is justified by the high demand and high risk in these regions.
Table 3

Percentage of demand coverage for the proposed MSs

Number of MSs 15 20 30 50 
Demand coverage (%) 19.56 31.94 44.97 59.18 
Number of MSs 15 20 30 50 
Demand coverage (%) 19.56 31.94 44.97 59.18 
Figure 9

Locations of proposed MSs for: (a) 15 MSs, (b) 20 MSs, (c) 30 MSs and (d) 50 MSs.

Figure 9

Locations of proposed MSs for: (a) 15 MSs, (b) 20 MSs, (c) 30 MSs and (d) 50 MSs.

CONCLUSIONS

Many guidelines and best practices have been developed in the past for an effective water quality monitoring in WDN. In most cases, MSs are located based on the water demand. Nodes with higher water demand are more likely to be considered as an MS, since it will monitor a significant amount of water in the WDN. However, these considerations ignore the fact that there are other factors that might have a significant effect on locating the MSs, such as the vulnerability of the system itself and the sensitivity of the population to the consequences of water quality deterioration. Vulnerable and sensitive regions deserve more monitoring compared to other regions. This study proposed an approach for locating MSs based on water demand, the vulnerability of the system and population sensitivity. For this purpose, framework modelling based on FSE was constructed to develop risk indices for the regions in the WDN. These indices, as well as water demand, were used as key optimization parameters to determine the optimal locations of the MSs. The developed model was applied to the Al-Khobar WDN, considering several sets of MSs that were examined. The study found that: (1) most of the MSs are located in the central regions of the WDN, since they have the highest water demand, vulnerability and sensitivity, and (2) the proposed approach ensures that every region in the city will be monitored by at least one MS regardless of the demand.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the support provided by the Deanship of Scientific Research (DSR) at King Fahd University of Petroleum and Minerals (KFUPM) for funding this work through project No. RG1004-1&2. Thanks are also extended to the Ministry of Water (Al-Khobar branch), Ministry of Education and Ministry of Economy & Planning for providing the necessary data to complete the study.

REFERENCES

REFERENCES
Al-Zahrani
M.
Al-Ghamdi
A.
2008
Hydraulic Analysis of Al-Khobar Water Distribution System
.
Revised Final Report, KACST Research No. APR-22-003
.
Al-Zahrani
M.
Moied
K.
2003
Optimizing water quality monitoring stations using genetic algorithms
.
The Arabian Journal for Science and Engineering
28
(
1
),
57
75
.
AWWA
2002
Effects of Water Age on Distribution System Water Quality.
US Environmental Protection Agency
,
Washington, DC
.
Berry
J.
Hart
W.
Phillips
C.
Uber
J.
2004
A general integer-programming-based framework for sensor placement in municipal water networks
. In:
Proceedings of the World Water and Environmental Resources Congress
.
Berry
J.
Fleischer
L.
Hart
W.
Phillips
C.
Watson
J.
2005
Sensor placement in municipal water networks
.
Journal of Water Resources Planning and Management
131
(
3
),
237
243
.
Francisque
A.
Rodriguez
M. J.
Sadiq
R.
Miranda
L. F.
Proulx
F.
2009
Prioritizing monitoring locations in a water distribution network: a fuzzy risk approach
.
Journal of Water Supply: Research and Technology – AQUA
58
(
7
),
488
509
.
Grayman
W.
Ostfeld
A.
Salomons
E.
2006
Locating monitors in water distribution systems: red team-blue team exercise
.
Journal of Water Resources Planning and Management
132
(
4
),
300
304
.
Gupta
S.
2008
Hydrology and Hydraulic Systems
.
Waveland Press, Inc
,
IL, USA
.
Kessler
A.
Ostfeld
A.
Sinai
G.
1998
Detecting accidental contaminations in municipal water networks
.
Journal of Water Resources Planning and Management
124
(
4
),
192
198
.
Kumar
A.
Kansal
M.
Arora
G.
1997
Identification of monitoring stations in water distribution system
.
Journal of Environmental Engineering
123
(
8
),
746
752
.
Lee
B. H.
1990
Locating Monitoring Stations in Water Distribution Networks. PhD Dissertation
,
University of Michigan
.
Lee
B. H.
Deininger
R. A.
1992
Optimal locations of monitoring stations in water distribution system
.
Journal of Environmental Engineering
118
(
1
),
4
16
.
Liu
H.
Yuan
Y.
Zhao
M.
Lu
J.
Zheng
X.
Zhao
H.
2011
Optimization of monitoring point placement of water quality in the water distribution system based on TEVA-SPOT
. In
ICPTT 2011: Sustainable Solutions for Water, Sewer, Gas, and Oil Pipelines
.
ASCE
, pp.
1419
1428
.
MathWorks
2012
Fuzzy Logic Toolbox: for Use with MATLAB: User's Guide
.
MathWorks, Incorporated
.
Ministry of Economy and Planning (MEP)
2010
Population and Housing Census 2010
.
Central Department of Statistics and Information
,
Saudi Arabia
.
Ross
T.
2009
Fuzzy Logic with Engineering Applications
.
Wiley
,
New York, NY
.
Sadiq
R.
Rodriguez
M. J.
2004
Fuzzy synthetic evaluation of disinfection by-products – a risk-based indexing system
.
Journal of Environmental Management
73
(
1
),
1
13
.
Sadiq
R.
Kleiner
Y.
Rajani
B.
2004b
Aggregative risk analysis for water quality failure in distribution networks
.
Journal of Water Supply: Research and Technology — AQUA
53
(
4
),
241
261
.
Sarbatly
H.
Krishnaiah
D.
2007
Free chlorine residual content within the drinking water distribution system
.
International Journal of Physical Sciences
2
(
8
),
196
201
.
Silvert
W.
2000
Fuzzy indices of environmental conditions
.
Ecological Modelling
130
(
1/3
),
111
119
.
Tryby
M.
Uber
J.
2001
Representative water quality sampling in water distribution systems
. In:
Bridging the Gap: Meeting the World's Water and Environmental Resources Challenges
.
ASCE
, pp.
1
10
.
USEPA
2009
EPA Drinking Water Advice Note No. 5: Turbidity in Drinking Water
.
The Environmental Protection Agency
.
US Environmental Protection Agency
.
Wang
L.
1992
Fuzzy systems are universal approximators
. In
IEEE International Conference on Fuzzy Systems 1992
.
IEEE
, pp.
1163
1170
.
WHO
2008
Guidelines for Drinking-water Quality
, 3rd edn,
Vol 1
.
World Health Organization
,
Geneva
.
Zadeh
L. A.
1978
Fuzzy sets as a basis for a theory of possibility
.
Fuzzy Sets and Systems
1
,
3
28
.