Effect of Breakage Level One in Design of Water Distribution Networks

Water Resour Manage
DOI 10.1007/s11269-010-9701-1
Effect of Breakage Level One in Design of Water

Distribution Networks
Mohammadjafar Soltanjalili · Omid Bozorg-Haddad ·

Migual A. Mariño
Received: 7 January 2010 / Accepted: 11 July 2010

© Springer Science+Business Media B.V. 2010
Abstract Design of water distribution networks (WDNs) that do not consider per-
formance criteria would possibly lead to less cost but it could also decrease water
pressure reliability in abnormal conditions such as a breakage of pipes of the net-
work. Thus, awareness of the situation of consumption nodes, by considering water
pressures and the amount of water that is being supplied, could be an effective source
of information for designing high performance WDNs. In this paper, Two-loop and
Hanoi networks are selected for least-cost design, considering water pressures and
the amount of water supplied on each consumption node under breakage level
one, using the honey-bee mating optimization (HBMO) algorithm. In each state
of design, a specific pressure is defined as the minimum expected pressure under
M. Soltanjalili (B) · O. Bozorg-Haddad

Department of Irrigation and Reclamation, Faculty of Agricultural Engineering
and Technology, College of Agriculture and Natural Resources, University of Tehran,
Karaj, Tehran, Iran
e-mail: jalili@ut.ac.ir
O. Bozorg-Haddad
e-mail: obhaddad@ut.ac.ir
M. A. Mariño
Department of Land, Air and Water Resources, University of California,
139 Veihmeyer Hall, Davis, CA 95616-8628, USA
e-mail: MAMarino@ucdavis.edu
M. A. Mariño
Department of Civil and Environmental Engineering, University of California,
M. A. Mariño
Department of Biological and Agricultural Engineering, University of California,
M. Soltanjalili et al.
breakage level one which holds the pressure reliability in the considered range. Also,
variations of some criteria such as reliabilities of pressure and demand, vulnerability
of the network, and flexibility of the design are analyzed as a tool for choosing the
appropriate state of design. Results show that a minor increase in the cost of design
could lead to a considerable improvement in reliabilities of pressure and demand
under breakage level one.
Keywords Water distribution networks · Design · Performance criteria ·

Honey-bee mating optimization (HBMO) algorithm · Breakage level one
1 Introduction
Traditionally, the optimal design of WDNs considered minimization of economic

cost. Nowadays, however, minimization of design cost is considered as the objective
function of a WDN design problem and diameters of the network’s pipes as its
decision variables. In a discrete optimization problem with 10 locations for the pipes
and 10 commercial diameters available for each pipe, the decision space for such a
∧
problem would be 10 10 different states.
It is clear that searching in such an expanded discrete decision space, without using
an optimization tool is not possible. In addition, calculating a solution in a reasonable
time is an important factor. Linear programming (LP) and nonlinear programming
(NLP) are optimization tools capable of determining optimal solutions. Alperovits
and Shamir (1977), Quindry et al. (1981), Goulter et al. (1986), Kessler and Shamir
(1989), and Fujiwara and Khang (1990) used LP and NLP to design WDNs.
The time-consuming process of direct search methods and the inefficiency of
gradient methods to solve discrete problems decrease the efficiency of traditional
algorithms such as LP and NLP. Thus, evolutionary algorithms, especially genetic
algorithm (GA), have been used to solve optimization problems. Simpson et al.
(1994) used the GA to solve discrete and nonlinear problems. Pipe diameters are
discrete variables which cause more application of the GA. Dandy et al. (1996), Savic
and Walters (1997), and Montesinos et al. (1999) used different methods based on
the GA to solve WDN problems. Cunha and Sousa (1999) used simulated annealing
(SA) algorithm for optimal design of WDNs (Hanoi and New York networks).
Lippai et al. (1999) applied Opt-Quest software in the optimal design of the New
York WDN. Geem (2005) proposed harmonic search (HS) algorithm in the design
of two benchmark problems (Two-loop and Hanoi networks) in WDN design.
Walski (2001) indicated that concentration on cost minimization without using
performance criteria, such as reliability of supplying demand or nodal pressure and
network reliability are important reasons for the inefficiency of proposed models in
the design of WDNs. Thus, design of WDNs should include minimization of project
cost subject to supplying acceptable pressure. Farmani et al. (2005) considered the
effects of different types of failure, such as pump being out of service and pipe
breakage, on the performance of WDN and also the influence of those considerations
on the reliability and the cost of design. They stated that reliability in a WDN is the
probability of meeting a desirable operation in the system, as it had been expected in
the designing period. Fujiwara and De Silva (1990) proposed a method based on
cost minimization considering a specific reliability. Todini (2000) considered cost
Effect of Breakage Level One in Design of Water Distribution Networks
minimization and system flexibility maximization as objective functions and the

optimal trade-off between those two objectives was found. Prasad and Park (2004)
modified Todini’s (2000) criteria and reported that an increase in flexibility criteria
is not enough reason for increasing reliability criteria.
In recent years, simulation of social behavior of insects such as ants and bees has
been developed as algorithms to solve optimization problems. Afshar et al. (2007)
used the honey-bee mating optimization (HBMO) algorithm to solve nonlinear
constrained and unconstrained problems. They showed the capability of HBMO in
determining near-optimal solutions compared to GA and considerable proximity
to optimal solutions compared to Lingo 8. Bozorg Haddad et al. (2008) used the
HBMO algorithm to extract an optimal policy for rehabilitation of WDN for periods
of 30 and 100 years. The HBMO algorithm was also used by Ghajarnia et al. (2009) to
determine the least-cost design of WDN with respect to different levels of reliability
during an operational period. They also identified critical nodes of the network by
considering different nodal pressure reliabilities.
In this paper, the HBMO algorithm is used as the optimization tool for optimal
design of Two-loop and Hanoi networks considering pressure reliability conditions
when breakage level one occurs. In most existing investigations, the reliability
criterion is calculated after least-cost design. In this paper, however, the least-cost
design of WDNs subject to pressure reliability is considered as the optimization
problem. Results show that an acceptable increase in the amount of cost could lead
to a considerable improvement in the reliability of WDNs.
2 Breakage, Its Levels and Their Probabilities in WDNs
Breakage in WDNs means failure of pipes in a period of time which is specified

to their repairs. Different levels of breakage show how many pipes are breaking in
one period simultaneously. Breakage level zero means that no breakage occurs in a
period irrespective of the useful life operational period of the WDN. Breakage level
one means that in a period of time just one breakage happens, irrespective of total
number of breakages in the useful life operational period of WDN. Thus, breakage
level two means that two breakages happen simultaneously in one period. Urban
WDNs are mostly designed in looped shape. Usually in looped WDNs, including
case studies considered in this paper, there are some nodes which are terminated
with only two pipes. Breakage level two may cause them to be out of any supply
at all, because no pipes are terminating and supplying them. As the result of this
situation, the hydraulic simulator would not be able to analyze the network anymore,
unless the solitary nodes are eliminated from the process of hydraulic simulation. As
stated previously, the aim of this paper is to strengthen the network so as to be able
to satisfy the demands of all nodes in a breakage situation. To be able to analyze
the network even under breakage level two and have an economic approximation
of the cost which it imposes to the design of the network, this paper considers the
analyzable states of breakage level two in a Two-loop network to design the network
to be reliable with respect to simultaneity of every analyzable breakage level two.
The statement “analyzable breakage level two” means the states of simultaneity of
two breakages which do not eliminate the connection of any consumption node with
the pipes of the network. These analyzable states are shown in Table 1.
Table 1 Analyzability and non-analyzability of simultaneous breakage of any couple of the net-
work’s pipes
Number Number of the network’s pipes
of the 2 3 4 5 6 7 8
network’s
pipes
2 NA** A* A* A* NA** A*
3 A* A* A* NA** A*
4 A* A* A* A*
5 NA** A* NA**
6 A* NA**
7 A*
8
A* Analyzable, NA** Non analyzable
It has been determined by a statistical calculation that the occurrence of breakage

level two is unlikely to happen in comparison with breakage level one. Equation 1
achieves the probability of occurrence of breakage with different levels:
L P−L
P B B
Pr = 1− (1)
L Y×P Y×P
in which: Pr = probability of occurrence of breakage level L; L = number of

simultaneous breakages (level of breakage); P = number of the network’s pipes;
B = total probable number of breakages in the network during an operational
period; and Y = number of periods for occurrence of breakage in the network during
operational period, which is achieved dividing the length of the operational period
by the time required to repair each breakage. Figure 1 shows changes in probability
of occurrence of breakage with different levels considering different total number
of breakages in the network during an operational period with Y = 25 and P = 7.
It is seen in the figure that in a few number of breakages, the probability of facing
periods which do not experience any breakages (level zero) is high while there is
a low probability of facing periods in which breakage level one happens. Also, by
increasing the total number of breakages in an operational period of the network, as
the level of breakage increases, the difference between its occurrence probability
and the probability of breakage level one also increases. For a total number of
breakages less than 15, the occurrence probability of breakages level 3 and higher
are achieved nearly zero and also there is a considerable difference between the
occurrence probability of breakage levels 1 and 2. Although moving ahead along
the horizontal axis shows that the probabilities of breakage levels 1 and 2 are getting
closer to each other, when one considers the useful life operational period of the
WDNs, the parameter Y could have values much greater than 25. Thus, regarding
Eq. 1, an increase in the value of parameter Y would cause the difference between
the probabilities of breakage level one and higher levels to remain stable for a greater
number of total breakages.
1
Breakage Level 0 Breakage Level 1 Breakage Level 2 Breakage Level 3
0.9 Breakage Level 4 Breakage Level 5 Breakage Level 6 Breakage Level 7
0.8
0.7
0.6
Probability
0.5
0.4
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45 50
Number of Breakages
Fig. 1 Changes in probability of occurrence of breakage with various levels, considering total
number of breakages in the network during operational period with Y = 25 and P = 7
3 HBMO Algorithm
The HBMO algorithm is an evolutionary algorithm based on bee behavior that works
as a hybrid tool including GA, SA, and local search (LS) algorithms which improve
the capability of the preceding individual algorithms by combining them into one
procedure. The HBMO algorithm includes three repetitive stages: (1) selection; (2)
reproduction; and (3) improvement. Naturally, in each hive there is a queen which is
in fact the best bee of the colony. In the mating season, the queen flies to a proper
place and attracts the attention of drones with a special dance. Afterwards with a
fast sudden move, it flies towards the sky, which is known as mating flight, and in
this manner only some of the best drones are able to mate with the queen. After
the mating flight, the queen returns to the hive and starts laying eggs. The workers
then breed the broods and the best broods are selected for becoming a new queen.
Selected broods are fed by a special nutrient called royal jelly. After growing up,
the best nurtured brood will replace the present queen provided that it is better than
the queen. Hereby successive generations of honey-bees in the hive will move toward
evolution. In this algorithm, firstly a random set of decision variables is generated and
is assumed to be the queen. Other random answers are then generated and the best
ones are combined with the queen (queen generation). Afterwards, the new answers
(broods) are improved by some predefined functions, and the best answer is then
replaced by the queen if it is better. One of the advantages of the HBMO algorithm
is its ability to weighing the operators and functions based on their performance
and desirability observed in previous generations. That is to say, an evolutionary
function which has done well in the previous generation will be allowed to generate
a bigger portion of the initial population of the next generation. This rule has the
Start
Define algorithm and model input parameters
Generating a set of initial random solutions, keeping the best, based on the objective function, as the queen and the other
ones as the drones
Simulated
Simulated annealing (generating
No annealing Yes
Utilizing previous trail solutions new trail solutions and replacing
(The limits of
the previous ones with them)
mating flight)
Generating new solutions (breeding process) using cross over operators and heuristic
functions (workers) between the best solution (queen) and trail solutions according to
their fitness value
Improving the newly generated set of solutions (feeding selected broods and queen with
the royal jelly) utilizing heuristic functions and mutation operators
No
Sorting new solutions in accordance with their fitness value,
selecting the best new solutions and the best new trail solutions
Is the new best

Yes solution better No
Substituting the best solution
than the
previous one?
Examination
Yes
of Finish
termination
criterion
Fig. 2 Computational flowchart of the algorithm
same justification about the values of the algorithm’s parameters. That is, if the value
of the probability for the crossover function in a generation improves the solution
and objective function, then the algorithm must change their values for the next
generation using previous nearby values. Thus, initial values of parameters do not
have a great influence on the progress of the algorithm. The computational flowchart
of the algorithm is presented in Fig. 2. More information about HBMO algorithm is
included in Bozorg Haddad et al. (2008).
4 Optimal Design Model of WDNs in Breakage Level One Condition
In this model, pipe diameters are considered as decision variables to be selected

from commercially available diameters. The objective function is the total cost of
network pipes. Constraints include the pressure provided in each consumption node
after hydraulic analysis of the set of solutions in which every node has to be at least
more than the minimum pressure requirement under normal conditions and also
more than the minimum expected pressure when breakage level one occurs. There
are two penalty functions in the optimization model: (1) one related to providing less
pressure of P Min under normal conditions and (2) another related to less pressure of
Pexp when breakage level one happens. Penalty functions are defined so as to prevent
the best infeasible solution to be better than any feasible one. Thus, the optimization
model can be expressed as:

NI
Min. OF = Costi + PF1 + PF2 (2)
i=1
Costi = Ci (Di , Li ) (3)

NJ
PF1 = K × (Pmin − P j + 1)2 , if P j < Pmin (4)
j=1

NJ
PF2 = K × (Pexp − P j + 1)2 , if P j < Pexp (5)
j=1
in which: Costi = cost of ith pipe; PF(1 and 2) = penalty functions for provid-
ing respectively minimum pressures in normal and breakage level one conditions;
C(Di , Li ) = cost function of ith pipe with diameter of D and length of L; P j =
pressure of jth node; Pexp = minimum expected pressure in breakage level one
conditions which varies under each state of design; NJ = number of consumption
nodes; and NI = number of pipes. Hydraulic calculations are done using EPANET2
software (Rossman 2000), which was linked to the HBMO algorithm. The algorithm
was run with 110 number of drones in the hive (sample solutions), and allowed to run
for 5000 mating flights.
5 Performance Criteria in WDNs
Five criteria are defined that present performance of the system under different costs
and conditions.
5.1 Reliability (Re)
This criterion contains the concept of providing minimum required pressure (Pmin )
under different hydraulic simulation states and is calculated as:
NS × N J − N P≺Pmin
Re = (6)
NS × N J
where NS = number of different states for design and consequently hydraulic
simulations of the network, which would lead to a set of pressures achieved for each
consumption node of the network. The aforementioned states involve simulations
of looped designed network without considering pressure reliability under breakage
condition and reliable looped designed network under normal condition. Again, the
preceding reliable network under breakage of just one pipe could be any pipe of
the network except pipe number 1. N P≺P min = total number of nodes in which their
pressures are less than the minimum required under all aforementioned designs and
simulations states.
The minimum pressure required is a fixed value for all the states of design. On
the other hand, the penalty functions are added to the objective function in: (1)
least-cost design when the pressure provided in each node is less than P Min and (2)
reliable design of WDN when the pressure is less than Pexp under breakage level one
condition.
5.2 Vulnerability (Vu)
In breakage level one condition, some nodal pressure possibly could be provided less
than P Min . Thus, vulnerability presents the maximum difference between provided
pressures in all nodes and P Min among different reliable designs under breakage
level one:
P Min − Min(P)
Vu = (7)
P Min
where Min(P) = minimum pressure among all consumption nodes considering a
specific value for minimum expected pressure in each state of simulation (NS).
5.3 Flexibility (Fl)
This criterion involves both reliability and vulnerability of the system and is calcu-
lated as:
Fl = R e × (1 − Vu) (8)
5.4 Hydraulic Benefit of Pressure (HBP )
This criterion was proposed by Carrijo and Reis (2004). To calculate HB P , two
positive values are considered as the minimum and maximum permissible pressures
for each node. The pressure of each node and the pressure hydraulic benefit of it (ψ j)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130
Fig. 3 Changes in with respect to changes in pressure
are then calculated with Eqs. 9 and 10. Finally, HB P would be the sum of ψ j in all
consumption nodes:

NJ
H BP = ψj (9)
j=1
0.5
P j −Pmin
i f Pmin < P j < Pmax
ψj = Pmax −Pmin (10)
0 i f P j < Pmin or P j > Pmax
in which Pmax is the maximum allowable nodal pressure in the network. The greater
the amount of ψ j, the greater the reliability on node j. Figure 3 shows an increasing
relation between P j and ψ j. If the existing pressure of a node falls outside of the
allowable boundaries, its ψ j will be equal to zero. It seems not to be a right option
for ψ j to increase at the same rate as the amount of P j along the interval between
minimum and maximum defined allowable pressures. This is because both very small
and excessive values for the pressure provided at nodes would affect reversely the
performance of WDNs. Thus, it is preferable to define HB P so that its maximum
amount would be achieved between minimum and maximum allowable pressures.
5.5 Hydraulic Benefit of Demand (HBD )
Also proposed by Carrijo and Reis (2004), this criterion is similar to HB P and is
weighed considering the amount of demand in each specific node. So, the more
demand in a node, the higher the priority of supplying the demand for the same
node.
⎛ ⎞
NJ
P j − Pmin 0.5 ⎜
⎜ QD j ⎟
⎟
H BD = ×⎜ ⎟ (11)
Pmax − Pmin ⎝NJ ⎠
j=1 QD j
j=1
where Q D j = demand of jth node.
6 Case Study
The Two-loop network is a benchmark network that is used as the first case study
in this paper. Alperovits and Shamir (1977) presented this simple network which
does not have any pump and storage tank. Figure 4 shows a schematic of the Two-
loop network with eight pipes and seven nodes. Length of all pipes is 1,000 m and
the Hazen–Williams coefficient is assumed to be 130 for all pipes. The minimum
required pressure in all nodes (P Min ) is equal to 30 m-H2 O and Pexp has discrete
values between 0 and 30 with 3 m-H2 O as intervals.
100 100
2 1 1
3 2
160 150
7 3
270 120
4
5 4
Reservoir
150 155 Node
Pipe
8 5 Elevation
Demand
200 330
6
7 6
160 165
Fig. 4 Schematic of two-loop network

940 560
Reservoir Demand
13 12
12
Node Length 1200
11
3500
Pipe
500
11
950
10
105 805 170 900 370 310 280 615 525 525
31 32 25 26 27 16 15 14 10 9
33 34 26 27 28 15 14 13 9 850
16 8
32 25 750
150 860 950 850 300 550 500 800 800
865
17 2730 550
1300 8
360 820 17
1345 850
7
30 24 18
1750
31 1350
24 1230 18
1600 60 7
800 4
19 400
6
290 1045 1275 19 130 725
360 850 1005
29 28 23 20 3 4 5 6
30 29 23 20 3 4 5
21 1150
2000 1500 2650 2200 2 900 1450
890 1350
930
1500 2
21
22 1
1
485 100
500
22
Fig. 5 Schematic of Hanoi network
The Hanoi network in Vietnam is used as the second case study. This simple
network involves 34 pipes and 32 nodes without any pump and storage tank.
Elevation of all nodes is zero and here the Hazen–Williams coefficient of all pipes
is 130. Figure 5 shows a schematic of the Hanoi network.
7 Results and Discussion
In this paper, optimal design of WDNs (Two-loop and Hanoi) with the aim of
minimizing their total cost are designed so as to provide minimum required pressures
in normal conditions and minimum expected ones in breakage level one conditions.
7.1 Two-Loop Network
Under each reliable state of design, which guarantees minimum expected pressures
of 0, 3, ..., 30 m-H2 O under breakage level one condition, the diameter of each pipe
varies in a specific range (Fig. 6). For instance, it is expected to achieve relatively
less values of diameters for different pipes of the network under unreliable looped
state of design in comparison with the reliable state of design which guarantees 30 m-
25
pipe 1 pipe 2 pipe 3 pipe 4 pipe 5 pipe 6 pipe 7 pipe 8
20
Diameter (Inches)
15
10
0
Tree shaped WDN
Unreliable looped
12
15
18
21
24
27
30
WDN
Minimum Expected Pressure (m-H2o)
Fig. 6 Changes of each pipe’s diameter for various states of design of two-loop network
H2 O in breakage level one condition. Table 2 presents statistical measures relevant

to the mentioned variability for each pipe. For example, the value located at the
junction of row 2 (numbered “1”) and column 2 (minimum, in inches) illustrates
that under different states of design which are previously mentioned in detail, the
least achieved value for pipe number 1 is 18 in. This and other statistical parameters
calculated herein for each pipe, considering different design states, could give the
designers a viewpoint that will be helpful to identify the more important pipes of
the network considering the influence of the size of their diameter on the hydraulic
performance of the network. Because the aim of this paper is to achieve the least-cost
design subject to different levels of reliability (different design states), the resulting
information could differentiate the effect of enlarging and reducing the diameter
of different pipe sizes. Figure 6 shows changes of each pipe’s diameter for various
states of design. In the first state (tree-shaped network on the horizontal axis),
Table 2 Statistical measures for diameters achieved for each pipe of two-loop network under various
reliable design states
Pipe Minimum Average Maximum ST. DEV Variance C.V. Median Mode
numbers (inches) (inches) (inches) (inches) (inches)
1 18 20 20 0.93 0.87 0.05 18 18
2 16 20 20 1.62 2.62 0.08 16 16
3 14 18 18 1.35 1.82 0.07 16 14
4 10 16 16 1.96 3.85 0.12 12 12
5 12 14 14 0.93 0.87 0.07 14 14
6 10 14 14 1.75 3.05 0.12 12 10
7 14 18 18 1.40 1.96 0.08 16 16
8 12 16 16 0.89 0.80 0.06 14 14
the model is allowed to select a zero value for diameters. Thus, a least-cost design
leads to a tree-shaped network. In the second state (unreliable looped network),
the model is not allowed to select zero value for diameters and also the network
is not expected to provide discrete values of minimum expected pressures under
breakage level one condition. This has led to a looped network but unreliable to
provide specific amounts of pressure under breakage level one conditions. In both
mentioned states of design, providing minimum required pressure (30 m-H2 O) has
been considered necessary. Other states guarantee providing minimum required
pressure under normal condition and also minimum expected pressures 0 to 30
m-H2 O (each discrete value on the horizontal axis) under breakage level one con-
dition. As it is seen in Fig. 6, diameters of most of the network’s pipes have been in-
creased with respect to the rise in the amount of minimum expected pressure. This is
more obvious looking at the curves related to pipe numbers 4 and 8. Also, under two
tree shaped and unreliable looped states, diameters of these pipes are zero and nearly
zero, respectively. Considering the next values for diameter of these pipes along the
horizontal axis, it could be inferred that increasing their diameter would lead to a
rise in the amount of reliability under breakage level one. On the other hand, when
the minimum expected pressure in breakage condition is 21 m-H2 O, diameters of
different pipes are closer to each other in comparison with other states. This would
practically ease preparation and installation of the network’s elements. Figure 7
shows the relation between performance criteria and total cost under different
design states. It shows the ascending trend of reliability and flexibility criteria and
descending trend of the vulnerability criterion with rise in the cost of network. The
reliability of the system in the state of 21 m-H2 O as the minimum expected pressure
100
Reliablity Vulnerability Flexibility
90
80
Performance Criteria (%)
70
60
50
40
30
20
10
0
419 (unreliable
514 (Pexp=0)
524 (Pexp=3)
542 (Pexp=6)
554 (Pexp=9)
572 (Pexp=12)
610 (Pexp=15)
650 (Pexp=18)
680 (Pexp=21)
740 (Pexp=24)
790 (Pexp=27)
870 (Pexp=30)
looped WDN)
Cost (×10^3 $)
Fig. 7 Relation between performance criteria and total cost under various design states of two-loop
network
900
800
Cost (×10^3 $)
700
600
500
0 3 6 9 12 15 18 21 24 27 30
Fig. 8 Relation between total costs with respect to rise in minimum pressure expected in two-loop
network
is equal to 0.85, which can be presented as an alternative design state. The rea-
son is closeness of the diameters of pipes connecting to a consumption node. In this
situation, if one of the pipes were broken, the other could considerably cover its
portion of the node’s demand. Otherwise, if the larger one breaks, the smaller one is
not able to supply any critical node. Figure 8 presents the ascending relation among
the total costs with respect to a rise in minimum pressure expected. It is seen here
again that after point 21 on the horizontal axis, the change in cost curve is occurring
with steeper slope. Also, Figs. 7 and 8 show that an increase in the minimum expected
pressure from 0 to 21 under breakage condition has increased the cost of design from
$514 to $680 × 10∧3 (33% rise in cost) and also has increased the reliability of the
system by 15%. A cost increase of $ 190 × 10∧3 occurs between points 21 and 30
on the horizontal axis of Fig. 8. This again confirms the desirability of this point
for selecting diameters of the network’s pipes. Figures 9 and 10 respectively present
pressure and demand hydraulic benefit in all nodes under breakage of each pipe
solely. It is shown in the figures that as minimum expected pressure increases, the
value of these criteria increases as well. Also, it could be inferred from the figures
that in each state of hydraulic simulation of the network, breakage of pipes 2 and
3 have the most influence on the sum of these criteria in the network. This could
be rationalized by considering the location of these pipes at the beginning of the
network. It is shown in Fig. 6, which introduces suggested diameters for different
pipes under each state of design, that in most of the reliable design states, suggested
diameters for pipes 2 and 3 are close to each other. Because these two pipes play the
role of distributors of the water delivered from the reservoir, and also considering
their nearly equal diameters, it would be expected that breakage of each of them
would make a serious hardship for the other one and will cause a shortage of pressure
provided on consumption nodes. Also, it is seen that in the most of reliable design
6
Pmin=0 Pmin=3 Pmin=6 Pmin=9
Pmin=24 Pmin=27 Pmin=30 Unreliable looped WDN
Hydraulic Benefit at All Nodes (Pressure)
0
2 3 4 5 6 7 8
Network's Pipes Number
Fig. 9 Pressure hydraulic benefit in all nodes under breakage of each pipe solely in two-loop network
states, breakage of pipe 2 has a larger influence on reliability criteria of the network.
This is because of the bigger diameter for this pipe suggested by the model. When
pipe 2 does not exist, it will force pipe 3 to supply the network, while its diameter
is smaller than that of pipe 2. Naturally, this would negatively affect demand and
1
0.9 Pmin=12 Pmin=15 Pmin=18 Pmin=21
Hydraulic Benefit at All Nodes (Demand)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2 3 4 5 6 7 8
Fig. 10 Demand hydraulic benefit in all nodes under breakage of each pipe solely in two-loop
network
10
9
Hydraulic Benefit at each Node (Pressure)
0
2 3 4 5 6 7
Number of Consumption Nodes
Fig. 11 Sum of pressure hydraulic benefits in each consumption node with respect to breakage of
various pipes in two-loop network
pressure reliability criteria. Figures 11 and 12 show the sum of pressure and demand
hydraulic benefits criteria in each consumption node with respect to breakage of
different pipes. The values of pressure hydraulic benefit criterion in nodes 6 and
7 and also in node 2 have been achieved more than in other nodes. The reason
which causes node 5 to be located at the top of the demand hydraulic benefit curve
2
1.8 Pmin=24 Pmin=27 Pmin=30 Unreliable looped WDN
Hydraulic Benefit at each Node (Demand)
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0
2 3 4 5 6 7
Fig. 12 Sum of demand hydraulic benefits in each consumption node with respect to breakage of
various pipes in two-loop network
is the high value of pressure and supply in comparison with other nodes. Node 6 has
the greatest portion of total demand of the network. But because of low pressures
provided in this node in the most of design states (especially in the states that their
minimum expected pressures are less than 24 m-H2 O), node 6 has been located at the
lowest level in Fig. 12. Figure 13 shows the cumulative number of nodes in which the
pressure provided in them is less than 30 m-H2 O under breakage of different pipes
are delineated for different design states. Thus, each curve is showing the results
of its own design state with a specific value for minimum expected pressure or the
minimum pressure which is guaranteed by applying that state of design. The curves
are cumulative for more clarity. Therefore, the vertical distance between each two
consecutive curves shows the number of nodes with pressure provided less than 30 m-
H2 O, under the design state related to the upper curve. For example, the vertical
distance between two lowermost curves at point 2 on the horizontal axis equals 5 (10
− 5). It means that at the state of design which guarantees 3 m-H2 O in breakage level
one condition, in the condition of breakage of pipe 2, the number of nodes in which
their pressure provided would be less than 30 m-H2 O is 5. The curve which is located
before the uppermost one guarantees 30 m-H2 O in breakage level one condition. As
it is clear from the figure, as we go down from this curve, the distance between curves
has an ascending trend. That means the less minimum pressure expected, the more
nodes with the minimum pressure provided less than the minimum required one.
Reliable design regarding breakage level two in the Two-loop network has been
done for only one minimum expected pressure, which is the state of design that
guarantees the pressure equal to 30 m-H2 O in breakage level two conditions. As
mentioned previously in Section 2, by neglecting the non analyzable occurrence
conditions of breakage level two in the two-loop network, the network is reliable to
supply the demand of all consumption nodes with 30 m-H2 O pressure in analyzable
Cumulative Number of Nodes with Less Pressure Provided than
50
45
40
35
Minimum Required
30
25
20
15
10
0
2 3 4 5 6 7 8
Fig. 13 Cumulative number of nodes with pressure provided less than 30 m of water under breakage
of various pipes, for various design states in two-loop network
Table 3 Statistical measures for diameters achieved for each location’s pipes of Hanoi network
under various reliable design states
1 0 20 40 23 533 1 20 0
0 28 40 19 358 1 35 40
40 40 40 0 0 0 40 40
2 40 40 40 0 0 0 40 40
0 10 40 20 400 2 0 0
0 30 40 20 400 1 40 40
3 0 10 40 20 400 2 0 0
0 10 40 20 400 2 0 0
0 20 40 23 533 1 20 40
4 0 25 40 17 300 1 30 30
0 14 30 16 249 1 12 0
0 3 12 6 36 2 0 0
5 0 20 40 23 533 1 20 0
0 10 40 20 400 2 0 0
0 10 40 20 400 2 0 0
6 0 18 40 21 425 1 15 0
0 13 40 19 356 1 6 0
0 8 30 15 225 2 0 0
7 0 0 0 0 0 – 0 0
0 10 40 20 400 2 0 0
0 25 40 17 300 1 30 30
8 0 11 24 13 164 1 10 0
0 19 30 13 169 1 22
0 11 24 13 164 1 10 0
9 0 23 30 15 225 1 30 30
0 10 40 20 400 2 0 0
0 0 0 0 0 – 0 0
10 20 26 30 5 24 0 27 30
0 0 0 0 0 – 0 0
0 3 12 6 36 2 0 0
11 0 3 12 6 36 2 0 0
0 9 20 11 111 1 8 0
16 20 24 3 11 0 20 20
12 0 16 24 11 117 1 20 20
0 5 20 10 100 2 0 0
0 3 12 6 36 2 0 0
13 0 10 40 20 400 2 0 0
0 8 30 15 225 2 0 0
0 15 30 17 300 1 15 30
14 0 23 30 15 225 1 30 30
0 0 0 0 0 – 0 0
0 10 40 20 400 2 0 0
15 0 20 40 23 533 1 20 40
0 10 40 20 400 2 0 0
0 10 40 20 400 2 0 0
breakages level two. Ignoring pipe number 1 which is the only connection between
the reservoir and the network, there are 21 states of breakage level two which is
possible to happen in the two-loop network. As it is stated in Section 2, some states
Table 3 (continued)
16 0 10 40 20 400 2 0 0
0 30 40 20 400 1 40 40
0 0 0 0 0 – 0 0
17 0 10 40 20 400 2 0 0
0 10 40 20 400 2 0 0
0 20 40 23 533 1 20 40
18 0 10 40 20 400 2 0 0
0 30 40 20 400 1 40 40
0 0 0 0 0 – 0 0
19 0 20 40 23 533 1 20 0
0 20 40 23 533 1 20 40
0 3 12 6 36 2 0 0
20 0 15 40 19 367 1 10 0
0 18 40 21 425 1 15 0
0 8 30 15 225 2 0 0
21 0 5 20 10 100 2 0 0
0 12 24 14 192 1 12 24
0 5 20 10 100 2 0 0
22 0 9 20 11 111 1 8 0
0 0 0 0 0 – 0 0
0 7 16 8 68 1 6 0
23 0 18 40 21 425 1 15 0
0 3 12 6 36 2 0 0
0 18 40 21 425 1 15 0
24 30 38 40 5 25 0 40 40
0 0 0 0 0 – 0 0
0 0 0 0 0 – 0 0
25 0 20 40 23 533 1 20 0
0 8 30 15 225 2 0 0
0 8 30 15 225 2 0 0
26 0 10 40 20 400 2 0 0
0 10 40 20 400 2 0 0
0 20 40 23 533 1 20 0
27 0 20 40 23 533 1 20 0
0 0 0 0 0 – 0 0
0 20 40 23 533 1 20 40
28 0 8 30 15 225 2 0 0
0 10 40 20 400 2 0 0
0 20 40 23 533 1 20 0
29 0 14 20 10 91 1 18 20
0 0 0 0 0 – 0 0
0 8 20 10 96 1 6 0
30 0 0 0 0 0 – 0 0
0 14 20 10 91 1 18 20
0 8 20 10 96 1 6 0
31 0 9 20 11 111 1 8 0
0 9 20 11 111 1 8 0
0 3 12 6 36 2 0 0
Table 3 (continued)
32 0 6 12 7 48 1 6 12
0 7 16 8 68 1 6 0
0 3 12 6 36 2 0 0
33 0 8 16 9 85 1 8 0
0 3 12 6 36 2 0 0
0 8 20 10 96 1 6 0
34 0 12 16 8 64 1 16 16
0 0 0 0 0 – 0 0
0 4 16 8 64 2 0 0
would lead to disconnection of some nodes of the network. Ignoring these states
of breakage level two, there would remain 15 states of breakage level two which
are analyzable. Thus, the two-loop network is designed in a reliable manner which
guarantees 30 m-H2 O under analyzable breakage level two states and the cost of
design is $1,160.
7.2 Hanoi Network
Reliable design of the Hanoi network is impossible using available commercial

diameters. Thus, the model is allowed to select more than one pipe for each location,
up to three pipes. If it is necessary to install more than one pipe between two
nodes, it is assumed that they are installed in parallel. Table 3 presents statistical
measures relevant to diameters of pipes selected for different locations of the
network. Figure 14 shows changes of diameter of equivalent pipe for each location
120
Location 1 Location 2 Location 3 Location 4 Location 5 Location 6
100 Location 13 Location 14 Location 15 Location 16 Location 17 Location 18
80 Location 31 Location 32 Location 33 Location 34
Diameter (Inches)
60
40
20
0
Tree shaped WDN Unreliable looped 0 10 20 30
WDN
Fig. 14 Changes of each location’s equivalent pipe diameter for various states of design of Hanoi
network
of the network for various states of design. As it is seen in Fig. 14, diameters of
most of the network’s locations have been increased with respect to the rise in
the amount of minimum expected pressure. This is more obvious looking at curves
related to the diameter of equivalent pipes for locations 1, 2, 14, and 31, that
shows a prominent increase in reliability of the network with respect to rise in their
diameters, when breakage level one happens. On the other hand, when the minimum
expected pressure in breakage condition is 20 m-H2 O, diameters of equivalent pipes
are closer to each other in comparison with other states. This would practically
ease preparation and installation of the network’s elements. Figure 15 shows the
relation between performance criteria and total cost for various design states.
It is seen that the difference between reliability criteria under different states is at
its uppermost level between unreliable state and reliable ones in comparison with
the difference between each two reliable states. Because in an unreliable state the
network has not been designed for breakage condition, facing this condition would
shock the network and will cause lots of nodes to fail in supplying the demands and
providing required pressures. As it is expected with an increase in cost, the trend
of flexibility and vulnerability are opposite. Considering the equations for flexibility
and vulnerability, low (nearly zero) pressures would cause the vulnerability criteria
to reach an undesirable value of 1 and so the value of 0 (or nearly zero) for flexibility
of network under the two states of design which are unreliable looped network and
minimum expected pressure of zero under breakage condition. Figure 16 presents
the ascending relation between total costs with respect to rise in minimum expected
pressure. It is seen that the slope of the curve decreases after point 20 on the
horizontal axis. By considering Fig. 14, it is clear that the diameter of pipes has not
Reliablity Vulnerability Flexibility

100
90
80
Performance Criteria (%)
70
60
50
40
30
20
10
0
6.1 (unreliable looped 7.9 (Pexp=0) 8.2 (Pexp=10) 8.6 (Pexp=20) 8.7 (Pexp=30)
WDN)
Cost (×10^6 $)
Fig. 15 Relation between performance criteria and total cost under various design states of Hanoi
network
8.7
Cost (×10^6 $)
8.4
8.1
7.8
0 10 20 30
Fig. 16 Relation between the total costs with respect to rise in minimum pressure expected in Hanoi
network
been increased considerably between points 20 and 30. Figures 17 and 18 respectively
present pressure and demand hydraulic benefit in all nodes under breakage of each
pipe solely. As it is seen in the figures, these criteria have the least value under
30
Pmin=0 Pmin=10 Pmin=20 Pmin=30 Unreliable looped WDN
Hydraulic Benefit at All Nodes (Pressure)
25
20
15
10
0
3 4 5 6 7 8 9 13 14 15 16 17 18 19 20 23 24 25 26 27 28 29 30 31 32 33 34
Network's Location Number
Fig. 17 Pressure hydraulic benefit in all nodes under failure of each location’s pipes in Hanoi
network
1
0.9
Hydraulic Benefit at All Nodes (Demand)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
3 4 5 6 7 8 9 13 14 15 16 17 18 19 20 23 24 25 26 27 28 29 30 31 32 33 34
Fig. 18 Demand hydraulic benefit in all nodes under breakage of each location’s pipes in Hanoi
network
failure in locations 3, 4, 18, 19, and 20, which shows the importance and sensitivity of
these pipes in supplying demands and providing required pressures of consumption
nodes. In Fig. 19, the sum of pressure hydraulic benefits for each consumption node
30
Hydraulic Benefit at each Node (Pressure)
25
20
15
10
0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Fig. 19 Sum of pressure hydraulic benefits in each consumption node with respect to breakage of
various location’s pipes in Hanoi network
2
1.8
Hydraulic Benefit at each Node (Demand)
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Fig. 20 Sum of demand hydraulic benefits in each consumption node with respect to breakage of
various location’s pipes in Hanoi network
has been delineated with respect to breakage of different pipes. Clearly, nodes 8–
14 and 28–32 are more sensitive than others. This is because the sum of the values
of hydraulic benefit of pressure under breakage of different pipes in each state of
design in aforementioned nodes is relatively less than other consumption nodes of
the network. In Fig. 20, the sum of demand hydraulic benefits for each consumption
node has been delineated with respect to breakage in different pipes. This criterion
gives specific weights to pressure hydraulic benefit in the nodes regarding the amount
of their demand. It means that the importance of providing the required pressure in
each node depends on its value of required demand. For example, by considering
nodes 4 and 5, at the point of 20 m-H2 O, as minimum expected pressure, the pressure
provided in node 4 is six units more than its value in node 5. But because of the
demand of node 5 in comparison with node 4, the value of demand hydraulic benefit
of node 5 is almost five times more than its value in node 4. As going ahead along
Fig. 20, it is seen that the peaks of the curves are related to the nodes in which their
demand are more than others. By comparing nodes 3 and 7, which are almost the
same in their value of demand, it is seen that providing more pressure in node 3
has caused more value for demand hydraulic benefit for this node. In Fig. 21, the
cumulative number of nodes in which the pressure provided in them is less than
30 m-H2 O under breakage of the pipes of different locations have been delineated
for different design states (as in the two-loop network). It is seen here again, as we
come down along the figure, the vertical distance between the curves is increasing,
especially under the failure of locations 3, 4, 18, and 19. Also, it is seen that the
curve related to minimum expected pressure of 30, conforms to the curve related
to minimum expected pressure of 20, meaning that under these states no node has
provided the minimum pressure less than 30.
Cumulative Number of Nodes with Less Pressure Provided than
80
70
60
Minimum Required
50
40
30
20
10
0
3 4 5 6 7 8 9 13 14 15 16 17 18 19 20 23 24 25 26 27 28 29 30 31 32 33 34
Fig. 21 Cumulative number of nodes with pressure provided less than 30 m of water under breakage
of various location’s pipes, for various design states in Hanoi network
8 Concluding Remarks
This paper presented the optimization of reliable design of two-loop and Hanoi
WDNs, using the HBMO algorithm, under breakage level one condition. The
objective function consisted of minimization of the total cost of design. Constraints
considered providing defined minimum expected pressures under breakage of each
pipe in the two-loop network and failure of each pipe location in the Hanoi network,
pressure and demand hydraulic benefits, and other performance criteria. Results
of the two-loop network showed that the composition of pipes that guarantees
21 m-H2 O with breakage level one condition is acceptable, considering performance
criteria and the cost of design. The interesting point about the states of design which
guarantee 21 m-H2 O in the two-loop and 20 m-H2 O in the Hanoi networks was the
closeness of diameters achieved for the networks that considerably ease preparation
and installation of the network’s elements. It could be inferred that the number of
nodes, with provided pressure less than 30 m-H2 O, are maximum under breakage of
pipes, which will cause the reliability of the network to reach to its minimum value.
The reason for pressure and demand hydraulic benefits under unreliable states of
design to be more than their values under some states of reliable design, is that
breakage of some pipes would lead to incalculable pressures as the result of hydraulic
simulation of the network. Thus, it seems to be necessary to do further research to
develop methods which could help to make these situations known.
The two-loop network was designed to be reliable to supply all demands of the
network with a pressure of 30 m-H2 O in analyzable breakage level two states as iden-
tified in Table 1. The cost of design was considerably more than the reliable state of
design, guaranteeing 30 m-H2 O in breakage level one condition. The aforementioned
cost was achieved for designing the network under analyzable states of breakage level
two. However, there are some states of non-analyzable states of breakage level two
which have the same chance of occurrence as analyzable ones. Thus, the achieved
cost of design for breakage level two was not the one which strengthened the network
to be able to supply all the nodes under all possible breakage level two conditions.
The scope of this research, however, was to strengthen the network in order to supply
all the consumption nodes in breakage conditions.
By a statistical computation it was determined that the higher breakage levels
are much less probable than breakage level one. It has been understood that by
increasing the total number of breakages in the operational period of the network
as the level of breakage increases, the difference between its occurrence probability
and the probability of breakage level one increases. Although moving ahead along
the horizontal axis showed that the probability of breakage levels one and two are
getting closer to each other, but by considering the useful life operational period of
the WDNs, the parameter Y could have values much greater than 25. Thus, from the
Eq. 1, it could be inferred that an increase in the value of parameter Y would cause
a considerable difference between the probabilities of breakage level one and higher
levels remain stable for a greater number of total breakages.
References
Afshar A, Bozorg Haddad O, Mariño MA, Adams BJ (2007) Honey-bee mating optimization
(HBMO) algorithm for optimal reservoir operation. J Franklin Inst 344(5):452–462
Alperovits E, Shamir U (1977) Design of optimal water distribution systems. Water Resour Res
13(6):885–900
Bozorg Haddad O, Adams BJ, Mariño MA (2008) Optimum rehabilitation strategy of water distrib-
ution systems using the HBMO algorithm. J Water Supply Res Technol 57(5):337–350
Carrijo IB, Reis LFR (2004) Operational optimization of WDS using a genetic algorithm with
multiobjective function and operating rules extracted through data mining. In: Simposio Inter-
nazionale di Ingegneria Sanitaria Ambientale, 2004, Taormina. SIDISA 2004–Proceedings, vol 1.
Catania-Italia: CSISA-Centro Studi Di Ingegneria Sanitaria Ambientale Onlus, pp 01–325
Cunha MC, Sousa J (1999) Water distribution network desing optimization simulated annealing
approach. J Water Resour Plan Manage 125(4):215–221
Dandy GC, Simpson AR, Murphy LJ (1996) An improved genetic algorithm for pipe network
optimization. J Water Resource Research 32(2):449–458
Farmani R, Walters GA, Savic D (2005) Trade-off between total cost and reliability for anytown
water distribution network. J Water Resour Plan Manage 131(3):161–171
Fujiwara O, De Silva U (1990) Algorithm for reliability-based optimal design of water networks. J
Environ Eng 116(3):220–230
Fujiwara O, Khang DB (1990) A two-phase decomposition method for optimal design of looped
water distribution networks. Water Resour Res 26(4):539–549
Geem ZW (2005) Optimal cost design of water distribution networks using harmony search. Journal
of Engineering Optimization 38(3):259–280
Ghajarnia N, Bozorg Haddad O, Mariño MA (2009) Reliability based design of water distribution
network (WDN) Considering the reliability of nodal pressures. World Environmental and Water
Resources Congress (ASCE). doi:10.1061/41036(342)579
Goulter IC, Lussier BM, Morgan DR (1986) Implications of head loss path choice in the optimization
of water distribution networks. Water Resour Res 22(5):819–822
Kessler A, Shamir U (1989) Analysis of the linear programming gradient method for optimal design
of water supply networks. Water Resour Res 25(7):1469–1480
Lippai I, Heaney JP, Laguna M (1999) Robust water system design with commercial intelligent
search optimizers. J Comput Civil Engineering 13(3):135–143
Montesinos P, Garcia-Guzman A, Ayuso JL (1999) Water distribution network optimization using a
modified genetic algorithm. Water Resour Res 35(11):3467–3473
Prasad TD, Park NS (2004) Multiobjective genetic algorithms for design of water distribution net-
works. J Water Resour Plan Manage 130(1):73–82
Quindry GE, Brill ED, Liebman JC (1981) Optimization for looped water distribution systems. J
Environ Eng 107(4):665–679
Rossman LA (2000) EPANET 2 users manual. United States Environmental Protection Agency,
Cincinnati. http://www.epa.gov/nrmrl/wswrd/dw/epanet.html
Savic DA, Walters GA (1997) Genetic algorithms for least-cost design of water distribution net-
works. J Water Resour Plan Manage 123(2):67–77
Simpson AR, Dandy GC, Murphy LJ (1994) Genetic algorithms compared to other techniques for
pipe optimization. J Water Resour Plan Manage 120(4):423–443
Todini E (2000) Looped water distribution networks design using a resilience index based heuristic
Approach. J Urban Water 2(3):115–122
Walski (2001) The wrong paradigm-Why water distribution optimization doesn’t work. J Water
Resour Plan Manage 127(4):203–205

Effect of Breakage Level One in Design of Water Distribution Networks

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Effect of Breakage Level One in Design of Water Distribution Networks

Transféré par

Droits d'auteur :

Formats disponibles

Water Resour Manage

Effect of Breakage Level One in Design of Water

Mohammadjafar Soltanjalili · Omid Bozorg-Haddad ·

Received: 7 January 2010 / Accepted: 11 July 2010

M. Soltanjalili (B) · O. Bozorg-Haddad

Keywords Water distribution networks · Design · Performance criteria ·

Traditionally, the optimal design of WDNs considered minimization of economic

minimization and system flexibility maximization as objective functions and the

2 Breakage, Its Levels and Their Probabilities in WDNs

Breakage in WDNs means failure of pipes in a period of time which is specified

It has been determined by a statistical calculation that the occurrence of breakage

in which: Pr = probability of occurrence of breakage level L; L = number of

Define algorithm and model input parameters

Is the new best

Fig. 2 Computational flowchart of the algorithm

4 Optimal Design Model of WDNs in Breakage Level One Condition

In this model, pipe diameters are considered as decision variables to be selected

Costi = Ci (Di , Li ) (3)

5 Performance Criteria in WDNs

5.1 Reliability (Re)

5.2 Vulnerability (Vu)

5.3 Flexibility (Fl)

5.4 Hydraulic Benefit of Pressure (HBP )

Fig. 3 Changes in with respect to changes in pressure

5.5 Hydraulic Benefit of Demand (HBD )

where Q D j = demand of jth node.

Fig. 4 Schematic of two-loop network

Fig. 5 Schematic of Hanoi network

7 Results and Discussion

7.1 Two-Loop Network

Minimum Expected Pressure (m-H2o)

H2 O in breakage level one condition. Table 2 presents statistical measures relevant

7.2 Hanoi Network

Reliable design of the Hanoi network is impossible using available commercial

Reliablity Vulnerability Flexibility

Vous aimerez peut-être aussi