Infrastructure

The most advantageous way to formulate the design of a water distribution system

under transient conditions is as a multi-objective optimization problem. The first

objective in such a design is a least-cost optimization problem using pipe diameters

as decision variables. The second objective is to minimize the likelihood of a damaging

transient event. In this article, a new surge damage measure called surge damage

potential factor is introduced. Evolutionary algorithms are applied to produce a set of

Pareto-optimal solutions in the search space of pipe cost and surge damage potential.

The model was tested on the New York City tunnel system. Comparison of the

proposed method with that of a conventional approach shows that the modifications

of pipe size in the design process can result in an effective and inexpensive surge

control strategy.

Optimal transient network design:

A multi-objective approach

A
water distribution system (WDS) is designed and operated to
consistently deliver water from source to consumer in suffi-
cient quantity, of acceptable quality, at appropriate pressure,
and as economically as possible. After choosing the critical
loading conditions for a distribution system design (i.e., the
greater of peak-hour demand or peak-day demand and a chosen fire flow),
optimization methods are applied to select the most economical set of pipe
BONG SEOG JUNG, sizes that will produce the desired range of pressures in the network. The
PAUL F. BOULOS, rationale behind an economical design is that by selecting the smallest pos-
sible diameter pipe set to minimize overall cost, pressures are marginally
AND TOM ALTMAN
higher than an acceptable level for the specified design loading conditions.
However, because the design problem is posed as a static one (i.e., the
design loads are not treated as dynamic variables), a conventional design
could well be suboptimal or even seriously inadequate for handling hydrau-
lic transient events.
Transient events in a WDS are inevitable and usually occur because of
actions at pump stations and control valves. Boulos et al (2006) described
typical events that require transient considerations: pump startup or shut-
down, valve opening or closing (i.e., any variation in a cross-sectional flow
area), changes in boundary pressures (e.g., adjustments in the water level
at reservoirs, pressure changes in tanks), rapid changes in demand condi-
2011 © American Water Works Association
118 APRIL 2011 | JOURNAL AWWA • 103:4 | PEER-REVIEWED | JUNG ET AL
tions (e.g., hydrant flushing), changes in transmission (1996) compared a genetic algorithm (GA) approach
conditions (e.g., pipe break, line freezing), or pipe filling with both complete enumeration and nonlinear pro-
or draining. When transient events occur too quickly, gramming in the context of pipeline optimization.
they induce a rapid change in flow rate within the system Although much of the pipeline optimization research
and cause potentially objectionable pressure surges—also has been concerned with systems under steady or near-
called water hammers—that could lead to unacceptable steady flow conditions, few optimization approaches
operating conditions. These conditions can cause breaches have dealt with the operating conditions pertaining to
in the hydraulic and physical integrity of the distribution system integrity, safety, and performance. Laine and
system. High-pressure transients (upsurges) can lead to Karney (1997) applied optimization to a simple pipeline
system failure and excess leakage, whereas low-pressure connecting a pump and a storage reservoir. A complete
transients (downsurges) can create vacuum conditions enumeration scheme and a probabilistic selection pro-
and pipeline collapse as well as opportunities for con- cedure were incorporated with both transient and
taminated groundwater to intrude into the distribution steady-state analysis. Lingireddy et al (2000) described
system at a leaky joint or break, which can affect public a surge tank design model based on a bilevel genetic
health. The volume of the intrusion can range from a few optimization framework that produces optimal tank
gallons to hundreds of gallons (Boyd et al, 2004). Because sizes while satisfying a specified set of pressure con-
all pipeline systems eventu- straints. Jung and Karney
ally leak and hydraulic (2004) considered the effect
transients occur continu- When transient events occur too quickly, of transients on the choice
ously in WDSs, it is not sur- of optimal diameter in a
prising that low-pressure they induce a rapid change in flow rate network considering both
transient events offer con- within the system and cause potentially steady and transient crite-
siderable potential to draw ria. More recently, Jung
untreated and possibly haz- objectionable pressure surges. and Karney (2006) pre-
ardous water into the pip- sented an optimum selec-
ing system. Water quality tion approach of hydraulic
studies have emphasized the need for transient analysis devices for water-hammer control in a WDS. GA and
of large pipe networks to properly assess the potential particle swarm optimization were used to optimize the
level of intrusion associated with negative pressure events preliminary selection, sizing, and placement of surge
and the resulting consequences on disinfectant residual protection devices. Previous approaches, however, have
effectiveness (Boulos et al, 2006; Friedman et al, 2004; focused separately on the optimal component size in the
LeChevallier et al, 2003). steady and transient levels; few approaches have considered
Pressure surges can create serious consequences for the interaction of both analysis and the effect of selecting
pipeline systems if not properly recognized and addressed. component size on surge control. In addition, these
Many hydraulic transient approaches have been devel- approaches have been limited to single-objectives, so their
oped to identify system weak points, to predict the poten- results make it difficult for the decision-maker to fully
tially damaging effects of hydraulic transients under appreciate the significant amount of interaction taking
various worst-case scenarios, and to evaluate how they place between steady and transient hydraulic analyses.
may be eliminated or controlled (Jung et al, 2007; Boulos In this article, the optimal design of a WDS under
et al, 2005; Wood et al, 2005; Wylie & Streeter, 1993). transient conditions is formulated as a multi-objective
In particular, Boulos et al (2005) provided a detailed optimization problem. Unlike most optimization models
transient analysis flow chart for the selection of compo- in which demands are set to their end-of-life levels, this
nents for surge control and suppression in WDSs and approach assumes that the demand loadings vary
concluded that a transient analysis should always be throughout the design life of the system. The first objec-
performed to determine the effect of each proposed strat- tive is formulated as a least-cost optimization problem
egy on the resulting system performance. Jung et al (2007) with pipe diameters as the decision variable. The second
further argued that only a systematic transient analysis objective is to minimize the likelihood of damaging tran-
can be expected to resolve complex transient character- sient events. In this study, the authors use a new surge
izations and adequately protect WDSs. damage measure called the surge damage potential factor
Optimization methods have been widely applied to (SDPF). For any transient event, the SDPF is defined as
many problems associated with WDS design, manage- the integration of the transient pressures that are lower
ment, and operation. Alperovits and Shamir (1977) than the minimum required level (e.g., datum) or higher
applied linear programming, and Lansey and Mays than the maximum allowable transient pressure level
(1989) suggested using nonlinear programming to opti- (e.g., pipe ratings). Evolutionary algorithms are applied
mize component sizing and the operational decisions to produce a set of Pareto-optimal solutions in the search
arising in WDSs. Simpson et al (1994) and Dandy et al space of pipe cost and SDPF. The model is tested (by
2011 © American Water Works Association
JUNG ET AL | PEER-REVIEWED | 103:4 • JOURNAL AWWA | APRIL 2011 119
simulation) on the New York City tunnel system, and pipeline with a 1-m3/s flow and a 1,000-m/s wave speed.
relevant conclusions are presented. When the rate of flow is changed rapidly, the kinetic
energy associated with the flowing water can be con-
PRESSURE SURGE CONTROL IN WDS verted rapidly into strain energy in the fluid and the pipe
Because pressure surges can introduce contaminants wall, thereby causing either abnormally high surge pres-
into the WDS, break pipes, or damage hydraulic equip- sure or stresses. This relationship can be represented by
ment such as pumps or valves, it is not surprising that a Joukowski equation showing the change in pressure,
many protective strategies have been proposed. These called a potential surge or Joukowski pressure change,
include solutions ranging from system modification and H, and is directly proportional to the change in the
operational considerations to the addition of one or flow velocity, V
more dedicated surge protection devices (Boulos et al, c c
2006, 2005; Walski et al, 2003). The most common H =  V =  Q (1)
g gA
surge protection strategies use various protection
devices to store water or otherwise delay the change in in which c is the wave speed, g is the gravitational accel-
flow rate or to discharge water from a given pipe. For eration, and A is the cross-sectional area of pipe. Figure
example, air-release/vacuum-breaking valves are 1 shows the effect of pipe size on a Joukowski surge
installed at high points in a pipeline to prevent negative head when the 1-m3/s flow rapidly decreases to zero. As
pressure and cavitation by emitting air into the pipe pipe size increases, the resulting reduction in velocity
when the line pressure drops below atmospheric condi- decreases the magnitude of the surge pressure. For exam-
tions. A pressure-relief valve prevents excessive high ple, the increase of pipe size from 1 to 2 m results in a
pressure subsequent to an upsurge by ejecting water 75% decrease of surge pressure (from 130 to 32 m). If
out of a side orifice. No two systems are hydraulically pipe size modifications are implemented after a system
identical; therefore, the ultimate choice and combina- is in place, they are often expensive, but if they are
tion of surge protection devices will usually differ. examined early in the design process, they may form part
Other surge protection strategies are to influence the of an effective and inexpensive surge approach. As a
root transient causes of flow changes, such as adjusting result, this approach incorporates the effects (of varying
valve or pump operations. Too-rapid valve closure/ pipe sizes) on surge control to minimize the likelihood
opening or pump shutdown/startup may lead to water- of a damaging transient event. This is in conjunction
column separation or excessively high surge pressures. with the traditional least-cost optimization problem with
The transient effect might be eased or avoided through the selection of pipe diameters.
comprehensive operator training or locking out a quick
operation mechanism in the system. MATHEMATICAL FORMULATION
System modifications can also be considered—such as Whether designing a WDS using trial-and-error enu-
pipe reinforcement (i.e., increasing a pipe’s pressure rat- meration methods or with formal optimization tools, a
ing), rerouting conduits, using larger diameter pipes, broad range of concerns must be considered. Overall cost
changing the pipe material, or making strategic changes is likely to be the primary factor and includes costs for
in system topology. Consider the example of a single system construction, operation, and maintenance. The

FIGURE 1 Effect of pipe size on Joukowski surge head

Joukowski Surge Head—m

3,246

812
361
203
130
90
66 51
40 32
27

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2

Pipe Size—m

2011 © American Water Works Association

120 APRIL 2011 | JOURNAL AWWA • 103:4 | PEER-REVIEWED | JUNG ET AL
initial capital investment for the system includes pipes, Minimize Pipe Cost = 冱 Ck (Dk, Lk) (2)
pumps, tanks, and valves. Energy consumption occurs k僆Npipe

over time as the system is operated. The main constraints

are that the nodal demands are supplied at a minimum Minimize SDPF = 冱 冮 ⏐Hi (t)⏐dt,
(3)
pressure and that network flows and pressure heads must i僆Nnode

satisfy the governing equilibrium laws of conservation of where H < H*minimum or H > H*maximum
energy and mass.
In this article, the optimal design of a WDS under Subject to the governing transient equations
transient conditions is formulated as a two-objective
optimization problem. The first objective is formulated 1 ∂Q ∂H R n–1
in Eq 2 as a least-cost optimization problem with the   +  +  Q⏐Q⏐ =0 (4)
gAp ∂t ∂x x
selection of pipe diameters as the decision variables. The
second objective is to minimize the likelihood of a damag-
ing transient event, which is measured by the parameter ∂H a2 ∂Q
SDPF, defined as the integration of the transient pressures     =0 (5)
∂t gAp ∂x
that are lower than the minimum required level (e.g.,
datum) or higher than the maximum allowable transient and a set of algebraic constraints
pressure level (e.g., pipe ratings). As shown in Figure 2,
the SDPF is the area of transient pressure below the Hi (t) = C1, Qi (t) = C2, where t = 0, ∀i苸 Nnode (6)
minimum required pressure, H*minimum and above the
maximum allowable pressure, H*maximum. An SDPF value
of zero represents a condition of no damage with a given f [H9 (t), Qi (t)] = C3, where t > 0, i = boundary nodes (7)
transient event; an SDPF value with a higher value has a
chance of greater surge damage. Therefore, the second
objective, given in Eq 3, is formulated to minimize the Hi (t)  Hminimum i, where t = 0, ∀i苸 Nnode (8)
SDPF. The SDPF approach may be applied to the whole
WDS, or to particular subsystems, including single pipes.
In contrast to traditional optimization models in which Dk 苸 {D}, ∀k苸 Npipe (9)
demands are set to their end-of-life levels, this approach
assumes that the demand loadings vary throughout the in which Dk = discrete pipe diameters selected from the
design life of the system. The pipe network layout, nodal set of commercially available pipe sizes {D} (Eq 9); Ck(Dk
demands, and minimum head requirements are assumed Lk) = cost of pipe k with diameter Dk and length Lk; dt
to be known. The optimal design of water distribution is the integration of transient pressures along the tran-
networks can be stated mathematically as: sient period (t); and H = piezometric head; H*maximum and

FIGURE 2 Measuring pipe surge damage potential factor

| H |dt where H > H *maximum

Maximum allowable pressure

H *max
Pressure Head

H *min
Minimum required pressure

| H |dt where H > H *maximum

Time

H—piezometric head, H*maximum—maximum permissible head, H*minimum—minimum permissible head

2011 © American Water Works Association

JUNG ET AL | PEER-REVIEWED | 103:4 • JOURNAL AWWA | APRIL 2011 121
 H*minimum are the maximum and minimum permissible
FIGURE 3 Flowchart of pipeline optimization heads (e.g., representing pipe ratings, health concerns for
negative pressures), respectively. Equations 4 and 5 rep-
Decision variable: pipe size
resent the momentum equation and mass conservation
for transient flow in closed conduits (Wylie & Streeter,
1993). Here x is distance along the centerline of the
Hydraulic model: conduit; t is time; Q = volumetric flow rate; a = celerity
Optimization model: steady and
of the shock wave; Ap = cross-sectional area of the pipe;
NSGA transient
analyses and g = acceleration resulting from gravity. The friction
term R in Eq 4 can be represented by
Objective functions: pipe cost and
surge damage potential factor R = fp x/2g Dp A2p, n = 2, (10)

NSGA—nondominated sorting genetic algorithm or

R = x/(0.278 CDp2.63)1/0.54, n = 1/0.54, (11)

FIGURE 4 New York City tunnel system schematic
in which fp = Darcy-Weisbach friction factor and C =
Hazen-Williams roughness coefficient. The two hyperbolic
partial differential calculations in Eqs 4 and 5 are subject
R1 to the initial conditions of Eq 6 and boundary conditions
of Eq 7, in which C1, C2, and C3 are constants. Initial
conditions are typically taken as steady. Simple boundary
P1 P15
conditions of constant reservoir level and fixed demand are
N2 assumed, but combined relationships between H and Q are
N15 typical for most boundaries. Equation 8 requires that the
P2 nodal pressure H for any node i (where total number of
P14 nodes is Nnode) is equal to or greater than a specified mini-
N3
mum pressure Hminimum for a steady-state condition.
N14
P3

N4
MULTI-OBJECTIVE OPTIMIZATION
P13
The remaining but challenging question is how to apply
P4 an optimization method to the suggested problem of
WDS optimization. Single-objective optimization algo-
N5 N13
rithms have been applied to solve WDS problems. Gra-
P5
P12 dient-based mathematical optimization methods (Boulos
N18 P19
N6
et al, 2006; Walski et al, 2003) have provided efficient
N12
P17 P18 computational procedures for achieving a lower-cost
P6
P11
solution, but the methods suffered from some disadvan-
N7
tages: (1) being ineffective at reaching the global optimal
P7
N11 solution because of zero-gradient optimality criteria,
P19 which easily trapped a search process at a local optimal
N8 solution; (2) the lack of flexibility in handling discrete
P8 P10
N20 design variables and optimizing a partial network that is
N9 often required for many practical engineering designs; (3)
P9
the complexity of implementing and using the techniques;
N10 P21 P20
and (4) the requirement to compute either first- or sec-
ond-derivative information to generate improvements in
the objective function. Several researchers have used GA
P16
optimization for solving such complex WDS optimiza-
N16
tions (Dandy et al, 1996; Simpson et al, 1994). The ant
colony optimization (Maier et al, 2003) and the shuffled
N17 frog leaping algorithm (Eusuff & Lansey, 2003) have also
been applied for obtaining specific optimal designs of
WDSs. These methods offer significant advantages over
N—node, P—pipe size in millimetres; R—reservoir
gradient-based optimization approaches in that they do
2011 © American Water Works Association
122 APRIL 2011 | JOURNAL AWWA • 103:4 | PEER-REVIEWED | JUNG ET AL
not require any gradient information and search for the then evaluates the system and creates a new set of system
optimal solution by continuing to evaluate multiple solu- alternatives for the next iteration. The iterations continue
tion vectors simultaneously. until an optimal or an acceptable solution is reached.
Multi-objective optimization algorithms have been intro-
duced to solve WDS problems with multiple conflicting CASE STUDY
criteria or design objectives (Jeong & Abraham, 2006; The proposed method is illustrated using the New York
Farmani et al, 2005; Prasas et al, 2004; Prasas & Park, City tunnel system (Schaake & Lai, 1969). The network
2004; Kapelan et al, 2003). As opposed to the single-objec- (Figure 4) has been extensively studied for steady-state
tive optimization method for finding the best solution, conditions. It comprises 22 nodes (20 demand nodes), 21
which corresponds to the pipes, and one source node.
minimum or maximum value The system is gravity-driven
of the objective function, Water quality studies have emphasized and draws water from the
multi-objective optimization Hillview reservoir to the
cannot produce a single opti- the need for transient analysis of large downstream network. The
mal solution with conflicting objective of the optimization
objectives. The interaction
pipe networks to properly assess the problem is to add new pipes
among different objectives potential level of intrusion associated parallel to the existing ones.
instead gives rise to a set of The new pipe diameters need
compromised solutions, with negative pressure events. to be selected from 15 avail-
known as Pareto-optimal able sizes. A single demand
solutions. Kapelan et al pattern (57,130 L/s) was
(2003) applied the multi-objective optimization approach considered and a minimum allowable hydraulic grade
to the sampling design for WDS with the objectives of was specified for each node. The network and cost data
maximizing the calibrated model accuracy by minimizing are shown in Dandy et al (1996).
the relevant uncertainties and the total sampling design cost. Since the system was first examined in 1969 by Schaake
Prasas et al (2004) investigated the booster disinfection and Lai, many researchers have used it to test the numer-
facility location and injection scheduling problem in a ical effectiveness, efficacy, and performance of their respec-
WDS. They formulated the problem as a multi-objective tive techniques (Eusuff & Lansey, 2003; Maier et al, 2003;
optimization model to minimize the total disinfectant Wu et al, 2001; Savic & Walters, 1997; Dandy et al, 1996).
dose and to maximize the volumetric demand within However, all of these approaches were based only on
specified residual limits. Farmani et al (2005) and Prasas steady-state optimization. In this study, the optimization
and Park (2004) presented a multi-objective approach to process includes the effect of a surge event as well as the
a WDS design that minimized network cost and maxi- existing steady-state considerations. By doing so, different
mized network reliability by providing excess head
greater than the minimum allowable head. Jeong and
Abraham (2006) considered a physical attack scenario in FIGURE 5 Pareto-optimal solutions of pipe cost
a water infrastructure system and offered a model to and surge damage potential factor
generate a set of optimal operational strategies to mini-
mize consequences of intentional physical attacks. They
80
used a multiobjective GA to minimize the degree of the
disruption of critical infrastructure services, economic 70
loss, and the number of customers affected.
Surge Damage Potential Factor

In this article, for the given dual-objective (pipe cost and 60

SDPF) problem, nondominated sorting GA (NSGA), devel-
50
oped by Srinivas and Deb (1994), is used to circumvent
subjective decision-making and to generate Pareto-optimal 40
solutions for the multi-objective optimization problem.
Figure 3 shows a flow chart of the framework for optimiz- 30

ing the pipeline system considering the dual-objective

20
problem. First, an optimization program initializes the
pipe sizes as decision variables and the pipe cost is calcu- 10
lated. The hydraulic model then analyzes the given system
0
and uses the optimization program to check whether the 42 43 44 45 46 47 48 49 50
solution satisfies the required constraints in Eqs 4–9 and
then computes the second objective function in Eq 3. With Pipe Cost—$ millions

the dual-objective function values, the optimization model

2011 © American Water Works Association
JUNG ET AL | PEER-REVIEWED | 103:4 • JOURNAL AWWA | APRIL 2011 123
design decisions would be required, and the restrictive the wave characteristic method (Ramalingam et al, 2009;
search based on limited operating conditions is likely Boulos et al, 2006).
suboptimal for a broader range of demand loadings. The multi-objective method, NSGA, was applied to
To introduce transient conditions into this case study, a satisfy Eqs 2–9. For this problem, the probability of
variety of possible causes could be selected. For conve- mutation was set to 0.025, the probability of (single-
nience, a valve opening that point) crossover was set
increases the demand at node to 0.9, the population size
10 from 28 L/s to 4,814 L/s Because pressure surges can introduce was set to 400, the length
for 1 s was chosen to charac- of each chromosome was
terize the transient perfor- contaminants, break pipes, or damage set to 84, and the simula-
mance of the system. This tions run for 100 genera-
increased demand may be
hydraulic equipment, it is not surprising tions. For this problem,
the result of a fire flow, a that many protective strategies have 16 decision variables
burst pipe, an operator error, including the “do-noth-
or a temporary increase in been proposed. ing” option made up a
water consumption. The solution space of 1621 or
maximum permissible heads 1.93 × 1025 possible pipe
* *
H maximum and H minimum in Eq 3 are assumed to be 304.8 combinations. The NSGA initialized the population of
m and 54.9 m, respectively, for the whole system. Because pipe diameters, calculated the cost of pipelines and the
of the rapid demand increase at node 10, a reduced pres- SDPF of transient network design model to satisfy the
sure wave moves through the system. This wave is reflected given constraints and then created a new population
from the upstream reservoir and then propagates back and for the next generation. The SDPF approach was
forth in the system while being tracked numerically using applied to the whole WDS into which a single transient
the method of characteristics (Wylie & Streeter, 1993) or event was introduced.

TABLE 1 The pipe size of Pareto-optimal solutions

Pipe Size—mm
Multi-
Objective Pipe Cost 1–3, 5–8,
Solutions $ millions SDPF* 10–14, 20 4 9 15 16 17 18 19 21

Dandy† 38.8 99.86 ‡ 3,000 2,100 2,400 2,100 1,800 1,800

M1 42.5 67.5 3,000 2,100 3,000 1,800 3,000 1,500
M2 43.2 67.4 900 3,000 2,100 3,000 1,800 3,000 ,1500
M3 43.3 65.6 3,300 2,100 3,000 1,800 3,000 1,500
M4 43.4 60.8 900 3,000 2,100 3,000 1,800 3,000 1,500
M5 43.6 55.4 3,000 2,100 3,000 2,100 3,000 1500
M6 44.2 50.6 1,500 3,000 2,100 3,000 1,800 3,000 1,500
M7 44.5 49.7 900 3,000 2,100 3,000 2,100 3,000 1,500
M8 44.6 44.5 1,800 3,000 2,100 3,000 1,800 3,000 1,500
M9 45.0 38.2 2,100 3,000 2,100 3,000 1,800 3,000 1,500
M10 45.5 31.7 2,400 3,000 2,100 3,000 1,800 3,000 1,500
M11 46.0 25.2 2,700 3,000 2,100 3,000 1,800 3,000 15,00
M12 46.5 19.1 3,000 3,000 2,100 3,000 1,800 3,000 1,500
M13 47.0 13.8 3,300 3,000 ,2100 3,000 1,800 3,000 1,500
M14 47.5 9.7 3,600 3,000 2,100 3,000 1,800 3,000 1,500
M15 48.0 8.7 3,900 3,000 2,100 3,000 1,800 3,000 1,500
M16 48.1 5.0 3,300 3,000 2,100 3,000 2,100 3,000 1,500
M17 48.6 0.9 3,600 3,000 2,100 3,000 2,100 3,000 1,500
M18 49.1 0.0 3,900 3,000 2,100 3,000 2,100 3,000 1,500

M—multi-objective solution, SDPF—surge damage potential factor

*SDPF shown in Eq 2
†Optimal results from Dandy et al (1996)
‡Blank spaces indicate no pipe required.

2011 © American Water Works Association

124 APRIL 2011 | JOURNAL AWWA • 103:4 | PEER-REVIEWED | JUNG ET AL
 The resulting solutions, obtained after 100 genera- parison of the transient results obtained using Dandy’s
tions, are shown in Figure 5. The x axis plots pipe cost result and three results of the multi-objective solutions
in Eq 2 and the y axis plots the SDPF in Eq 3. The inter- (M1, M9, and M18) at node 19. The minimum pressures
action among different objectives gives rise to a set of of Dandy, M1, M9, and M18 are 50.1, 53.1, 52.9, and
compromised Pareto-optimal solutions. Each solution 55.4 m, respectively. The results indicate that the previ-
on the Pareto-optimal curve of Figure 5 is not dominated ous approaches considering the steady-state design alone
by any other solution. In going from one solution to are, not surprisingly, inadequate for coping with a water-
another, it is not possible to improve on one objective hammer event. The results also suggest that the proper
without making the other objective worse; for example, sizing of pipe diameters is crucial to prevent water ham-
improving the first objective of minimizing the pipe cost mer. As a result, if the modifications of pipe size are
worsens the second objective of minimizing the SDPF. considered in the design process, they can form a reliable
This leads to a tradeoff relationship between pipe cost and cost-effective surge control strategy.
and SDPF in which a decision-maker can choose a pre-
ferred solution. CONCLUSION
Table 1 shows the pipe cost, SDPF, and pipe diame- Transient analysis, despite its significant concern for
ters chosen in the Pareto-optimal solutions that cor- WDS design, is a complicated problem, and so is the
respond to the 18 distinct results. The tradeoff relation- optimization of a transient control strategy for WDSs.
ship provides useful data on the cost-effectiveness of The purpose of this article was to obtain an optimal
adding pipe capacity to reduce surge damage. The pipe network design by considering simultaneously
results indicate that when the pipe cost is increased steady and transient conditions. The objectives were to
from $42.5 million to $45 million (multiobjective solu- minimize both pipe cost and the likelihood of a damag-
tions [M] M1–M9), the 6% additional investment in ing transient event. A parameter called SDPF is defined
pipe can achieve a 43% reduction in surge damage as the integration of the transient pressures that are
(67.5–38.2%). Similarly, when the pipe cost is increased lower than the minimum required level or higher than
from $45 million to $49.1 million (M9–M18), the 9%
additional investment in pipe can achieve the condition
of no damage with the given transient event. A deci-
sion-maker can use the information in Figure 5 and TABLE 2 Violating maximum and minimum pressure
Table 1 to evaluate the marginal rate of tradeoff heads and their locations
between pipe capacity and SDPF. The Dandy et al
(1996) results are shown in Table 1 to compare the Multi- Maximum Minimum
optimal result of steady-state analysis alone with the Objective
Solutions Node Head—m Node Head—m
authors’ multiobjective approaches, including both
steady-state analysis and transient analysis for surge Dandy* † 17 41.2
protection. Dandy’s pipe cost is lower than those of the M1 17 41.2
Pareto-optimal solutions, but its SDPF is much higher, M2 17 41.2
because it did not include any consideration of surge M3 19 41.3
protection. Table 1 also shows the size of pipe P9 M4 19 42.4
increased from 0 to 3,900 mm as the pipe cost increases. M5 17 41.2
This occurs because pipe P9 is next to the surge-creat- M6 19 44.3
ing node 10 and the proper sizing of pipe P9 is most M7 17 42.4
crucial for controlling the surge pressure effectively. M8 19 45.6
Table 2 shows the violating maximum and minimum M9 19 46.9
pressure heads and their locations. In this study, the M10 19 48.4
minimum required head, 54.9 m, is the dominant factor M11 19 49.9
for controlling surge pressure. Therefore, no solutions M12 19 51.4
in the Pareto-optimal curve or in the Dandy et al (1996) M13 19 52.8
results violate the maximum allowable pressure. Figure M14 19 52.8
6 shows the transient head profiles obtained using Dan- M15 19 52.8
dy’s result and three multiobjective results (M1, M9, and M16 19 52.9
M18) at node 17. Because pipe cost is increased with a M17 19 54.3
larger diameter, the resulting reduction in velocity M18
decreases the magnitude of the pressure wave, increasing M—multi-objective solution
the minimum pressure. The minimum pressures of *Optimal results from Dandy et al (1996)
Dandy, M1, M9, and M18 are 41.2, 41.2, 46.9, and †Blank spaces indicate no violation of maximum permissible heads.

55.8 m, respectively. Similarly, Figure 7 shows the com-

2011 © American Water Works Association
JUNG ET AL | PEER-REVIEWED | 103:4 • JOURNAL AWWA | APRIL 2011 125
FIGURE 6 Transient head profile at N17

Dandy M9
M1 M18
120

100
Pressure Head—m

80

60

40

20
0 20 100 150 200 250 300

Time—s

Dandy—optimal results from Dandy et al (1996), M—multi-objective solution, N—node

FIGURE 7 Transient head profile at N19

Dandy M9
M1 M18
120

100
Pressure Head—m

80

60

40

20
0 20 100 150 200 250 300

Time—s

Dandy—optimal results from Dandy et al (1996), M—multi-objective solution, N—node

the maximum allowable transient pressure level. Evo- be introduced into the system to collect useful data.
lutionary algorithms were applied to produce a set of The cumulative results would then be incorporated
Pareto-optimal solutions in the search space of pipe cost into determining Pareto-optimal solutions. For clarity
and SDPF. The case study using the New York City tun- of presentation in this study, only one transient event
nel system indicated that the previous approaches, which was shown.
considered steady-state design alone, are inadequate Moreover, the authors considered the optimal selec-
for coping with a water-hammer event. In addition, the tion of pipe diameters for a surge protection strategy. A
study showed that pipe size is significant to controlling more global approach will ultimately also be consid-
transient response; as a result, proper pipe selection and ered, in addition to pipe size, transient properties (e.g.,
transient consideration can minimize the damage of operation speed), system characteristics (e.g., system
water-hammer events and form an effective and inexpen- topography, pipe material and thickness), and transient
sive surge control strategy. protection devices. This comprehensive design frame-
To provide a comprehensive analysis of a WDS work will offer a more complete range of systematic
design (or to fully evaluate the sensitivity to transients surge protection strategies and result in more reliable
of an existing system), several transient events should cost-optimization solutions.
2011 © American Water Works Association
126 APRIL 2011 | JOURNAL AWWA • 103:4 | PEER-REVIEWED | JUNG ET AL
ABOUT THE AUTHORS Division in 2008. Paul F. Boulos is president and chief
Bong Seog Jung is a senior engineer, operating officer of Innovyze, Broomfield, Colo. Tom
Innovyze, 618 Michillinda Ave., Ste. Altman is a professor in the Department of Computer
200, Arcadia, CA 91007; bongseog. Science and Engineering, University of Colorado, Denver.
[email protected]. He has focused
on hydraulic transients and optimiza- Date of submission: 06/30/2010
tion of water distribution systems for Date of acceptance: 08/13/2010
the past 10 years. His article “The
Need for Comprehensive Transient JOURNAL AWWA welcomes
Analysis of Water Distribution Systems” won AWWA’s comments and feedback
Best Paper Award in the Engineering & Construction at [email protected].

REFERENCES Lansey, K.E. & Mays, L.W., 1989. Optimization Model for Water Distri-
bution System Design. Jour. Hydraulic Engrg., 115:10:1401.
Alperovits, E. & Shamir, U., 1977. Design of Optimal Water Distribution
Systems. Water Resources Res., 13:6:885. LeChevallier, M.W.; Gullick, R.W.; Karim, M.R.; Friedman, M.; & Funk,
J.E., 2003. The Potential for Health Risks From Intrusion of Con-
Boulos, P.F.; Lansey, K.E.; & Karney, B.W., 2006. Comprehensive Water
taminants Into the Distribution System From Pressure Tran-
Distribution Systems Analysis Handbook for Engineers and Plan-
sients. Jour. Water & Health, 1:1:3.
ners (2nd ed.). MWH Soft Publ., Broomfield, Colo.
Lingireddy, S.; Funk, J.E.; & Wang, H., 2000. Genetic Algorithms in
Boulos, P.F.; Karney, B.W.; Wood, D.J.; & Lingireddy, S., 2005. Hydrau-
Optimizing Transient Suppression Devices. Proc. ASCE 2000
lic Transient Guidelines for Protecting Water Distribution Sys-
Joint Conference on Water Resources Engineering and Water
tems. Jour. AWWA, 97:5:111.
Resources Planning and Management, Minneapolis, Minn.
Boyd, G.R.; Wang, H.; Britton, M.D.; Howie, D.C.; Wood, D.J.; Funk,
Maier, H.R.; Simpson, A.R.; Zecchin, A.C.; Foong, W.K.; Phang, K.Y.;
J.E.; & Friedman, M.J., 2004. Intrusion Within a Simulated
Seah, H.Y.; & Tan, C.L., 2003. Ant Colony Optimization for Design
Water Distribution System Due to Hydraulic Transients. II: Vol-
of Water Distribution Systems. Jour. Water Resources Planning
umetric Method and Comparison of Results. Jour. Envir. Engrg.,
& Mngmnt., 129:3:200.
130:7:778.
Prasas, T.D. & Park, N., 2004. Multiobjective Genetic Algorithms for
Dandy, G.C.; Simpson, A.R.; & Murphy, L.J., 1996. An Improved
Design of Water Distribution Networks. Jour. Water Resources
Genetic Algorithm for Pipe Network Optimization. Water
Planning & Mngmnt., 130:1:73.
Resources Res., 32:2:449.
Prasas, T.D.; Walters, G.A.; & Savic, D.A., 2004. Booster Disinfection
Eusuff, M.M. & Lansey, K.E., 2003. Optimization of Water Distribution
of Water Supply Networks: Multiobjective Approach. Jour.
Network Design Using the Shuffled Frog Leaping Algorithm.
Water Resources Planning & Mngmnt., 130:5:367.
Jour. Water Resources Planning & Mngmnt., 129:3:210.
Ramalingam; D.; Lingireddy, S.; & Wood, D.J., 2009. Using the WCM
Farmani, R.; Walters, G.A.; & Savic, D.A., 2005. Trade-off Between
for Transient Modeling of Water Distribution Networks. Jour.
Total Cost and Reliability for Anytown Water Distribution Net-
AWWA, 101:2:75.
work. Jour. Water Resources Planning & Mngmnt., 131:3:161.
Savic, D.A. & Walters, G.A., 1997. Genetic Algorithms for Least-cost
Friedman, M.; Radder, L.; Harrison, S.; Howie, D.; Britton, M.; Boyd, G.;
Design of Water Distribution Network. Jour. Water Resources
Wang, H.; Gullick, R.; LeChevallier, M.; Wood, D.; & Funk, J., 2004.
Planning & Mngmnt., 123:2:67.
Verification and Control of Pressure Transients and Intrusion in
Distribution Systems. AwwaRF, Denver. Schaake, J. & Lai, D., 1969. Linear Programming and Dynamic Pro-
Jeong, H.S. & Abraham, D.M., 2006. Operational Response Model for gramming Applications to Water Distribution Network Design.
Physically Attacked Water Networks Using NSGA-II. Jour. Com- Rep. 116, Dept. of Civil Engrg., Mass. Inst. of Technol., Cam-
puting in Civil Engrg., 20:5:328. bridge.

Jung, B.S.; Karney, B.W.; Boulos, P.F.; & Wood, D.J., 2007. The Need Simpson, A.R.; Dandy, G.C.; & Murphy, L.J., 1994. Genetic Algorithms
for Comprehensive Transient Analysis of Distribution Systems. Compared to Other Techniques for Pipe Optimization. Jour.
Jour. AWWA, 99:1:112. Water Resources Planning & Mngmnt., 120:4:423.

Jung, B.S. & Karney, B.W., 2006. Optimization of Transient Protection Srinivas, N. & Deb, K., 1994. Multi-Objective Function Optimization
Devices Using GA and PSO Approaches. Jour. Water Resources Using Non-Dominated Sorting Genetic Algorithms. Evolutionary
Planning & Mngmnt., 132:1:44. Computing, 2:3:221.

Jung, B.S. & Karney, B.W., 2004. Fluid Transients and Pipeline Optimi- Walski, T.M.; Chase, D.V.; Savic, D.A.; Grayman, W.M.; Beckwith, S.; &
zation Using GA and PSO: The Diameter Connection. Urban Koelle, S., 2003. Advanced Water Distribution Modeling and
Water Jour., 1:2:167. Management. Haestad Press, Waterbury, Conn.

Kapelan, Z.S.; Savic, D.A.; & Walters, G.A., 2003. Multiobjective Sam- Wood, D.J.; Lingireddy, S.; Boulos, P.F.; Karney, B.W.; & McPherson,
pling Design for Water Distribution Model Calibration. Jour. D.L., 2005. Numerical Methods for Modeling Transient Flow in
Water Resources Planning & Mngmnt., 129:6:466. Distribution Systems. Jour. AWWA, 97:7:104.

Laine, D.A.; & Karney, B.W., 1997. Transient Analysis and Optimization Wu, Z.Y.; Boulos, P.F.; Orr, C.H.; & Ro, J.J., 2001. Using Genetic Algorithms
in Pipeline—A Numerical Exploration. 3rd International to Rehabilitate Water Distribution Systems. Jour. AWWA, 93:11:74.
Conference on Water Pipeline Systems, The Hague, the Wylie, E.B. & Streeter, V.L., 1993. Fluid Transients in Systems. Prentice
Netherlands, 281. Hall, Englewood Cliffs, N.J.
2011 © American Water Works Association
JUNG ET AL | PEER-REVIEWED | 103:4 • JOURNAL AWWA | APRIL 2011 127