RELIABILITY-OPTIMIZATION BASED MODELS

F. Bouchart, N. Duan, I. Goulter

K.E. Lansey, L.W. Mays,
Y.C. Su and Y.K. Tung

1989
Book Chapter WWRC-89-22

In

Reliability Analysis of
Water Distribution Systems

Chapter 14

F. Bouchart N. Duan
Department of Civil Engineering Center for Research in Water Resources
University of Manitoba University of Texas
Winnipeg, Manitoba, Canada Austin, Texas

I. Goulter K.E. Lansey

Department of Civil Engineering School of Civil Engineering
University of Manitoba Oklahoma State University
Winnipeg, Manitoba, Canada Stillwater, Oklahoma

L.W. Mays Y.C. su

Center for Research in Water Resources Center for Research in Water Resources
University of Texas University of Texas at Austin
Austin, Texas Austin, Texas

Y.K. Tung
Wyoming Water Research Center
University of Wyoming
Laramie, Wyoming
 CHAPTER 14

RELIABILITY-OPTIMIZATION BASED MODELS

Francois Bouchart, Ning Duan, Ian Goulter, Kevin E. Lansey,

Larry W. Mays, Yu-Chun Su and Y.K. Tung

14.1 RELIABILITY-BASED OPTIMIZATION MODEL FOR WATER

DISTRIBUTION SYSTEMS

This section presents the basic framework for a model that can be
used to determine the optimal (least cost) design of a water distribution
system subject to continuity, conservation of energy, nodal head bounds,
and reliability constraints. Reliability is defied as the probability of satis-
fying nodal demands and pressure heads for various possible pipe failures
(breaks) in the water distribution system. The overall model includes
three models that are linked: a steady-state simulation model, a reliability
model, and an optimization model. The simulation model is used to
implicitly solve the continuity and energy constraints and is used in the
reliability model to define minimum cut-sets. The reliability model
which is based on a minimum cut-set method determines the values of
system and nodal reliability. The optimization model is based on a
generalized reduced gradient method. Examples are used to illustrate the
model.

14.1.1 Previous Models

Numerous optimization models have been developed and reported

in the literature. However, very few optimization models have included
the reliability aspects in the design of water distribution systems. Damelin,
Shamir and Arad (1972) developed a model to evaluate the reliability of
supplying a known demand pattern in a given water supply system in

472
which shortfalls were caused by random failures of the pumping equip-
ment. However, there is no optimization performed in their model and
only p u p failures are considered. The major components considered in
the Shamir and Howard (1979) model are the storage volume and the
standby pumps.

Tung (1985) discussed six techniques for water distribution system

reliability evaluation and concluded that the cut-set method is the most
efficient technique in evaluating the system reliability of water distri-
bution systems. Goulter and Coals (1986) developed a model in which
they incorporated reliability aspects into a linear programming model.
One disadvantage of the Goulter and Coals (1986) model is the assumption
that all the pipes connecting a node have similar diameters and hence
have similar values of the failure probability which is not applicable in
real pipe networks. A second disadvantage is a theoretical weakness in the
node isolation approach. The model considers a node to be adequately
supplied as long as there is at least one link connecting it to the rest of the
network. In addition, the model is not able to determine the design of
pumps and storage tanks or consider multiple loading conditions.

Mays and Cullinane (1986) introduced methods for evaluating the

reliability of individual water distribution system components which
included the concepts of mean time to failure analysis and stress-strength
or load-resistance analysis. They also concluded that the most promising
methods for determining the system reliability and availability for simple
series-parallel combination systems were the cut-set method and the path-
enumeration method.

Wagner, Shamir and Marks (1986) developed a model for the

calculation of probabilistic reliability measures for water distribution
systems. They defined the term "reachability" to denote the situation
when one specified demand node in the network is connected to at least
one source. Since there are an exponential number of configurations
which are used in searching and computing the reachability of demand
nodes, methods for calculating these probability measures must be effi-
cient which may not be applicable to large or even moderately sized,
complex systems. Also, Wagner, Shamir and Marks (1986) could not
include the multiple loading conditions into the model. Furthermore,
this model did not incorporate the reachability aspects into an optimi-
zation model for the design of water distribution systems.

14.1.2 Problem Description

The overall optimization problem for the design of water distri-

bution networks considering the reliability aspect can be stated as follows:

473
 Objective:

Minimize Cost f (D,L) (14.1.1)

subject to

Conservation of flow and conservation of energy constraints

(14. I .2)

Pressure head bound

- (14.1.3)
H2H2H

Reliabilities constraints - for both system and demand nodes

R2R (14.1.4)

Non-negativity constraints

D20 (14.1.5)

where D is the pipe diameter; L is the pipe length; H is the nodal

pressure head; and R is the reliability.

For a water distribution network design problem, the objective is to

minimize the total cost of the network. The variables in the objective
function are the nodal pressure head, H, the flowrate, Q, and the pipe dia-
meter D. Also the pipe length, L, is a parameter in the objective function.
Since the flowrate, Q, can be written in terms of H via the flow equation it
does not appear in the objective function, equation (14.1.1).

The constraints of this optimization model are:

1. Conservation of Mass and Enerm - The total amount of water

flow into a junction node should be equal to the total amount of water
flow out from this junction node. For each junction node the continuity
equation can be written as

CQh- Ca,, = Qe (j equations) (14.1.6)

4 74
where Qin is the flowrate of the water into the junction; Qout is the flow-
rate of the water from the junction; Qe is the external demand at the
junction node; and j is the number of junction nodes.

The conservation of energy constraint for each primary loop states

that the sum of the energy loss in each pipe should equal to the energy put
into the liquid. For each primary loop which is an independent closed
path, the conservation of energy equation can be written for the pipe in
the loop as follows

ChL = C E P (m equations) (14.1.7)

where hL is energy loss in each pipe, Ep is energy produced by a pump, and

m is number of primary loops.

For f fixed grade nodes, f-1 independent conservation of energy

equations can be written for paths of pipes between any two fixed grade
nodes as follows

(f - 1 equations) (14. I .8)

where AE is difference in total grade between the two fixed grade nodes
and f is the number of fixed grade nodes.

2. Head Bounds - The pressure head H at each demand node

should be within a specified range of a minimum required head H and a
maximum R:
-
H 2 H 2 H for each demand node (14.1.9)

3. Reliabilitv Constraints - Both the system and nodal reliabilities

should be greater than or equal to a lower bound which may be different
for the different reliabilities.

Rs(H,D)2 -sR for the network (14.1.10)

R(H,D) 2 R for each demand node (14.1.II)

4. Non-negativitv Constraints on pipe diameter, equation (14.1.5).

4 75
 By using the formulation proposed by Lansey and Mays (1985) f i e
above model can be used to determine the design of pumps and storage
tanks while considering multiple loading conditions. The model here
only considers the reliability aspect of the pipe network.

14.1.3 Model AIgorithm

Three models have been combined to develop the framework for a

methodology to define the overall optimal (least cost) reliability based
design of water distribution systems. A simple flowchart is presented in
Fig. 14.1.1 illustrating the basic concepts of this methodology. The optimi-
zation model used is the generalized reduced gradient model, GRG2, by
Lasdon and Waren (1979, 1984) which solves an optimization problem
with a nonlinear objective function and nonlinear constraints. A simu-
lation model is used to implicitly solve the continuity and conservation of
energy constraints, equation (14.1.21, at each iteration of the optimization
procedure. The simulation model adopted is the University of Kentucky
Model known as KYPIPE by Wood (1980) which simulates steady state flow
in a water distribution system based upon the continuity and conservation
of energy equations. The model determines the pressure head at each
node satisfying continuity. A reliability model is used to determine the
nodal and system reliabilities at each iteration of the optimization proce-
dure using the minimum cut-set method. A minimum cut-set is a set of
system components which, when failed, causes failure of the system but
when any one component of the set has not failed, does not cause system
failure (Billinton and Allan, 1983).

For each iteration of the optimization model (GRGZ), the reliability

model computes the values of the system and nodal reliabilities based
upon the minimum cut-set method. By assuming that a pipe break can be
isolated from the rest of the system, minimum cut-sets are determined by
closing a pipe or combination of pipes in the water distribution system and
using the simulation model (KYPFE) to obtain the values of pressure
head at each demand node of the system. By comparing these pressure
heads with minimum pressure head requirements, the reliability model
can determine whether or not this pipe or combination of pipes is a mini-
mum cut-set of the system as well as of the demand nodes whose pressure
head does not meet the minimum requirement. This procedure is
repeated until all the combinations of pipes have been considered and
hence all the minimum cut-sets of the demand nodes and the total system
have been determined. The reliability model then computes the values of
system and nodal reliability and returns to the optimization model.

The optimization model determines whether the solution is opti-

mal or not. If not, the optimization model based on the reduced gradient

476
 Subroutine

GCOMP

System Simulation Model .) System Reliability Model

(Univ. of Kentucky Model) 4 (Minimum Cut Set Method)

F i g u r e 14.1.1 Flowchart of Reliability Based Optimization

Model of Water Distribution Systems

477
method determines a new solution point in the direction of improving
the value of the objective function (minimizing cost) and a new iteration
begins. The model stops either with an optimal solution or with a mes-
sage that the solution is infeasible which means that at least one of the
constraints cannot be satisfied.

14.1.4 The Reliabilitv Model

Water main failures, as pointed out by O'Day (1982), are principally

due to excessive load, temperature or corrosion. Break rates are dependent
upon pipe size, geographic location, method of pipe manufacture, soil
type, etc. Kettler and Goulter (1983) have shown that the failure rate for
cast iron pipes in Winnipeg, Canada, has a strong relationship with pipe
diameter.

The method used herein to determine the probability of failure of

individual pipes is similar to the procedure developed by Goulter and
Coals (1986). The probability of failure of pipe j, Pj, can be determined
using the Poisson probability distribution

P. = b e-Pi (14.1.12)
J
and
pj = rj Lj (14.I.13)

where pj is the expected number of failures per year for pipe j; rj is the
expected number of failures per year per unit length of pipe j; and Lj is the
length of pipe j.

Assuming there are n components (pipes) in the i-th minimum

cut-set of a water distribution system then the failure probability of the j-th
component in the i-th minimum cut-set is Pi,j which can be obtained by
equation (14.1.12). The failure probability of the i-th minimum cut-set is

P(MCi) = n P i , j = P
j= 1 i, 1
Pi,* ... p.
cn
(14.1.14)

The basic assumption is that the occurrence of the failure of the compo-
nents within a minimum cut-set are statistically independent.

478
 For example, if a water distribution network has four minimum
cut-sets, MCI, MC2, MC3, and MC4 for the system reliability, the failure
probability of the system, Ps,is then defined as (Billinton and Allan, 1983)

S
MC,UMC~UMC,UMC,

= &Ci)
i =1

- $ $ P( MC, n MC.
i = l j=i+l J

+ xP(MCinMC.nMCk
i<jck I

MC, nMC, nMC, n MC4 (14.1.15)

By applying the principle of inclusion and exclusion (Ross, 1985) which

gives the upper bound of PSIequation (14.1.15) can be reduced to

MC, u MC, uMC, u MC,

(
= P MC1) (
+ P MC,) + P (MC3)
+P(MC4)

= P(MCi) (14.1.16)
i =1

Expressing equation (14.1.16) in a more general form as

Ps = 2
i=l
P(MCi) (14.1.17)

then the minimum system reliability, Rs,is expressed as

M
RS = I - P s = l-&(MC.) 1 (14.1.18)
i =1

479
where M is the number of minimum cut-sets in the system. The same
approach can be applied to compute the values for the nodal reliabilities.

The procedure of determining the reliability of a water distribution

system based on the minimum cut-set method is as follows. For a given
initial solution (pipe diameters), one of the pipes in the system is closed (a
simulated break) which means that no flow can pass through this pipe.
Then a simulation is performed using KYPIPE to determine the pipe flows
and nodal pressures throughout the network. This simulation model
implicitly satisfies nodal continuity so that the resulting pressure heads at
demand nodes of the system are compared to see if they satisfy the mini-
mum pressure head. If the pressure head at any one of the nodes in the
system is below the minimum required level due to the closure (simula-
ted failure) of this pipe, this closed pipe is a minimum cut-set of the sys-
tem and the nodes whose pressure head does not meet the required level.
This pipe is then opened and another pipe is closed and the simulation
model is run again. This procedure is repeated until all pipes in the
system have been considered.

The model then closes combinations of pipes simulating simul-

taneous failures of two pipes and KYPIPE is run for each combination. If
the resulting pressure head at any one of the demand nodes of the system
goes below the required level, the closed pipes are a cut-set of the system
and the demand nodes whose pressure are not satisfied. If any one of the
pipes in this cut-set is already a minimum cut-set of the system, by defi-
nition, this cut-set is not a minimum cut-set of the system; otherwise, it is
a minimum cut-set of the system. A similar comparison is made for the
demand nodes.

After all combinations of pipes have been considered by the above

procedure, all the minimum cut-sets of both the system and the demand
nodes are obtained. The values of the failure probability of each pipe is
obtained by using the Poisson distribution as previously described. The
probability of failure of a minimum cut-set is then equal to the product of
the failure probabilities of the pipes of this minimum cut-set, assuming
that the failures of pipes are independent. The system reliability is then
obtained by equation (14.1.18). Similarly, the reliability of each demand
node in the water distribution system can be obtained. Later in this sec-
tion, examples are presented to illustrate that the simultaneous failure of
combination of pipes may not have a major impact on the reliability of the
system because of the small probability of the failures of combination of
pipes.

The generalized reduced gradient code, GRG2, by Lasdon and

Waren (1984) has been used to solve the optimization model. GRG2

480
requires a user-supplied subroutine GCOMP for the purposes of com-
puting the constraint and objective function values of the decision
variables, GCOMP can also be used to read in initial values of any user-
required constants. GRG2 is a modular program written to provide
dynamic memory allocation with all arrays set up as portions of one large
main array so that redimensioning of arrays is never required.

Generalized reduced gradient methods such as GRG2 require an

initial solution to start the optimization search. GRG2 does have the
option of using an initial solution provided by the user or to start from an
arbitrary solution, as determined by the lower bounds of the decision
variables. If the initial solution is an infeasible solution, a phase I optimi-
zation is initiated which minimizes an objective function consisting of the
sum of infeasibilities until a feasible point is found. Once this is achieved,
the actual objective function replaces the sum of infeasibilities and the
actual optimization phase is initiated. Using an initial point provided by
the user allows the inclusion of engineering judgment in selecting a good
initial solution which may or may not be feasible. In either case, experi-
ence has shown that a good user-provided initial point results in less
computer time than initializing the algorithms from the lower bounds.
Although the model is able to give an optimal solution to the problem, a
global optimal solution can not be guaranteed. Multiple starting points
could be used to obtain different local optimal solutions and the one with
lowest cost would be chosen as the final design.

GRG2 like most nonlinear programming codes require the gradi-

ents of the objective function and constraints with respect to the decision
variables to determine search directions. Due to the way the reliabilities
and the definition of a minimum cut-set must be determined, these gradi-
ents cannot be computed analytically. A numerical scheme must be used
which perturbs each decision variable individually, computes the objec-
tive function and constraints for the perturbed values and then calculates
the gradients, e.g., af/& = &/Ax.

14.1.5 Example Application

A hypothetical one-loop 14pipe network as shown in Fig. 14.1.2 is

used to illustrate the model. This example has a requirement to design a
least cost water supply network to satisfy the demands at thirteen demand
nodes and meet the reliability constraints. The length of pipes and asso-
ciated initial failure probabilities for an initial solution, with all pipe
diameters being 36 inches, are listed in Table 14.1.1. The values of eleva-
tion, demand and initial pressure of the nodes resulting from the initial
flow and geometry condition are listed in Table 14.1.2. The pump in pipe 1

481
Figure 14.1.2 Water Distribution Network of Example

482
x
w
0 0 0 0 0 0 0 0 0 0 0 0 0 0
o o o o o o o o o o o o I n o
0 ~ 0 0 0 0 ~ 0 ~ 0 0 0 m P
rl r l d r l r l rl d r l r (
 Table 14.1.2 C h a r a c t e r i s t i c s of Demand Nodes

Node Elevat i o n I n i t i a l Pressure

Number (feet) (psi)
~

1 0.75 2518.0 62.68

2 0.75 2517.0 63.11
3 0.75 2498.0 71.41
4 1.50 2492.0 74.00
5 0.75 2508.0 67.01
6 3.00 2523.0 60.50
7 0.00 2513.0 64.85
8 1.50 2532.0 56.62
9 1.50 2512.0 65.34
10 0.75 2520 .O 61.89
11 0.75 2540.0 53.16
12 0.75 2530.0 57.49
13 1.50 2552.0 47.96
FGN -14.25 2323.0 0.00
has a useful pumping power of 550 horsepower. The objective function
for pipe cost (Quindry, Brill and Liebman, 1981) used in this example is

Minimize f (D,L) = 1.1 Df:24 L.. (14.122).

'J '1
i,j

where Dij is the Pipe diameter (inches) and Lijis the pipe length (feet). The
constraints are equations (14.1.6), (14.1.7), (14.1-8) solved implicitly using
KYPIPE and

Hi 2 92.3 feet (40 psi) for all demand nodes (14.1.23)

Rs 2 0.80 for system (14.1.24)

R.1 2 0.95 for all demand nodes (14.1.25)

D.. 2 0 for all pipes (14.1.26)

'I

The Hazen-Williams formula is used in KYPIPE to determine the pipe

flows. A Hazen-Williams roughness coefficient of 130 is used for all pipes.
English units are used for all computations. Note that the continuity and
conservation of mass constraints are implicitly satisfied in the simulation
model, and hence, the explicit constraints considered by GRG2 are equa-
tions (14.1.23) to (14.1.26). A convergence tolerance of 0.001 is used for all
constraints.

Using failure data obtained from the City of St. Louis (1985), a
regression equation was developed to compute the average break rate, PR
(breaks/mile/year), as

-0.1363D
= 0.819e (14.1.27)

where D is the pipe diameter in inches.

The initial and optimal minimum cut-sets and values of reliability

for the example system and for each demand node of the system are listed
in Tables 14.1.3. The bptimal value of the objective function is $652,000.
The final values of the pipe diameters and associated failure probabilities
of the network are listed in Table 14.1.4. This example required 1,157
seconds for execution on the CDC Dual Cyber 170/750 mainframe system
at The University of Texas at Austin.

485
 Table 14.1.3 Minimum Cut-Sets and Reliabilities of
the System and Demand Nodes

Initial Optimal
Minimum C u t S e t s of System
System System
(Pipe Number)
R e l i a b i lity Re1 i a b i l i t y

1, 2 , 3 , 4 , 5 , 6, 8, 9, 11, 13,

7 & 10, 7 & 12, 7 & 14, 1 0 t 12, 0.88963 0.e39as

10 & 14, 12 & 14.

486
 Table 1 4 . 1 . 4 O p t i m a l S o l u t i o n for Example

Multiple P i p e Single Pipe

Minimum C u t S e t s Minimum C u t S e t s
Pipe
Diameter Failure Diameter Failure
Number
( i n c h e s) Probility (inches) Probility

1 36.00 0.012530 36.00 0.012530

2 22.69 0.011481 14.40 0.016322

3 15.95 0.021315 13.15 0.025172

4 17.77 0.019547 12.80 0.0257 93

5 27.07 0.014618 31.73 0.013360

6 21.07 0.017221 25.30 0.015239

7 35.79 0.008813 21.30 0.011994

8 20.29 0.017692 36.00 0.012530

9 29.40 0.009773 36.00 0.008788

10 13.98 0.023841 12.93 0.025558

11 11.68 0.028074 11.76 0.027894

12 35.13 0.012681 11.13 0.029391

13 20.39 0.006206 36.00 0.004404

14 9.77 0.023500 12.10 0.019091

487
 The previous solution for the example computed minimum cut-
sets considering the failure of single pipes and the simultaneous failure of
two pipes. Considering simultaneous failure of two pipes has a small
probability of occurrence and requires considerably more computational
effort to compute the minimum cut-sets at each iteration of the optimiza-
tion procedure. The example has been resolved considering only
minimum cut-sets consisting of single pipe failures.

The nodal and system minimum cut-sets and reliabilities for the
initial and final solution for the example are listed in Table 14.1.5. The
optimal pipe sizes and associated failure probabilities for the examples are
listed in Table 14.1.4. The optimal objective function values and compu-
tation time for Example 1 are $651,690 and 1,157 seconds considering both
single and simultaneous two pipe failures minimum cut-sets as opposed
to the $622,470 and 515.1 seconds considering only single pipe minimum
cu t-sets.

Pipes 7/ 10,12 and 14 are in a loop and are in minimum cut-sets for
the simultaneous failure of two pipes but are not in minimum cut-sets
when considering single pipe failures. The fact that these pipes are not in
the minimum cut-sets for the single pipe failure case excludes them from
the reliability constraints and causes the model to use larger size pipes in
other locations such as pipes 5,7,9 and 13 in order to satisfy the reliability
constraints.

The system reliability for the optimal design using both single pipe
and siqultaneous failure of two pipes is 0.83985 as given in Table 14.1.3
and the system reliability using the single pipe failure is 0.83528 in Table
14.1.5. The last column in Table 14.1.5 lists the system and nodal
reliabilities of the optimal design when considering both single and sim-
ultaneous pipe failure. These values are not significantly different than
those considering the single pipe failures. The difference in cost for these
two design is $29,220 for a very small change in system reliability, 0.83985
as compared to 0.83797 for the design which considered single pipe mini-
mum cut-sets. Comparing the nodal reliabilities in Tables 14.1.3 and
14.1.5, there are rather insignificant differences in the nodal reliabilities.
The last column in Table 14.1.5 lists the system and nodal reliabilities
using the optimal design (Table 14.1.4) based upon single pipe minimum
cut-sets, but including the failure of both single pipes and two pipes failing
simultaneously. This illustrates that when the minimum cut-sets of two
pipes failing simultaneously are considered for this example that the
system and nodal reliabilities are reduced only slightly. The small differ-
ences in system reliability from considering only one pipe minimum cut-
sets as compared to one pipe and then two pipes failing simultaneously;

488
 T a b l e 1 4 . 1 . 5 Minimum Cut-Sets a n d R e l i a b i l i t i e s of t h e System
a n d Demand Nodes ( S i n g l e Pipe Minimum C u t - S e t s )

i
Initial Optimal
Minimum Cut Sets of System System
System System
( P i p e Number) Reliability*
Reliability Reliability

1, 2, 3, 4, 5 , 6,
8, 9, 11, 1 3 . 0.89031 0.83797 0 . e3528

Initial Optimal
Node Minimum Cut S e t s Nodal
Nodal Nodal
Number ( P i p e Number) Reliability*
Reliability Reliability

1 1, 2, 5, 9. 0.9574
0.9500 0.9500
2 1, 5, 9.
0.9662 0.9653 0.9638
3 1, 4, 8.
0.9624 0.9500 0.9500
4 1, 6, 8. 0.9624
0.9597 0.9597
5 1, 9, 13. 0.9743 0.9743 0.9725

6 1, 3, 9, 13. 0.9617
0.9500 0.9500
8 I, 9 . 0.9787
0.9787 0.9771
9 1, 8 . 0.9749 0.9749 0.9749
10 1. 0.9875 0.9875 0.9875
11 1, 9 . 0.9787 0.9787 0.9787
12 1, 9 . 0.9787 0.9787 0.9774
13 1, 9, 11. 0,9662 0.9508 0.9508

*Based upon t h e optimal d e s i g n (Table 1 4 . 1 . 4 ) for s i n g l e pipe

minimum c u t - s e t s but considering t h e reliabilities
for b o t h s i n g l e p i p e a n d t w o pipe f a i l u r e s .

489
however, this can result in rather significant differences in the cost of the
optimal solution.

The suggested design procedure is to determine the optimal design

of a network considering only single pipe failures. The reliabilities of
optimal design are then computed with both the single and simultaneous
pipe failures. If the decision maker finds these reliabilities unacceptable,
the system can be redesigned considering both single and simultaneous
failures. When making this decision two points should be kept in mind.
First, the optimal solution consists of continuous pipe diameters and
rounding to a discrete solution will slightly affect the cost and reliabilities
of the system. Second, the pipe failure data is not perfect knowledge.
When reviewing the reliability values, the number of significant figures
which can be considered as accurate must reflect this knowledge.

14.1.6 Conclusion
An optimal (least cost) reliability based design and evaluation
model of a water distribution network has been presented. The objective
of the model is to minimize the cost of the water distribution system
subject to continuity, conservation of energy, nodal pressures, and relia-
bility constraints. The model has been used on a few simple networks to
test its validity in designing water distribution systems. This model also
can be used in optimizing the extension of an existing water distribution
network by setting the diameter of existing pipes to constant values and
optimizing the extension. The optimal design of the extension is a func-
tion of the reliability of the existing system since the pipe breaks in the
existing system may cause nodal head constraints not to be satisfied in the
extension. The advantages of this model compared with other models in
the literature are: 1) it has successfully included the reliability aspects into
an optimization model; and 2) it can give an optimal (least cost) design of
a water distribution system while simultaneously satisfying the continu-
ity, conservation of energy, nodal pressure head bound, and reliability
constraints. Additional details of this basic framework of the methodology
presented herein can be found in Su (1986).

There are two disadvantages of the model. First, the resulting pipe
diameters may not be commercially available pipe sizes so that these
resulting pipe diameters must be rounded to the appropriate sizes. These
rounding diameters might affect the feasibility of the resulting optimal
solution and further research is being done to solve this problem. The
second disadvantage is that the model can require considerable computa-
tional effort to determine the optimal design of large looped networks.
The computation time required for this model depends on the number of
pipes and number of loops and to a lesser extent on the initial solution

490
used in the optimization. More research is in progress on reducing the
computational time required to execute the model.

More research should also be done to incorporate other system

components such as pumps, valves, storage tanks, etc. The major problem
is how to define the reliability of a system which includes all these com-
ponents. Also, a system under multiple loading conditions with normal
demand conditions and emergency demands such as fire demands, should
be considered in the least cost design model to ensure that the nodal pres-
sures are met for these loads. Furthermore, research may be performed to
extend this model to incorporate the repair and replacement type analysis.

14.2 WATER DISTRIBUTION SYSTEM DESIGN UNDER

UNCERTAINTIES

14.2.1 In trod uc ti on

A chance constrained optimization methodology is presented for

the minimum cost design of water distribution networks. This method-
ology attempts to account for the uncertainties in required demands,
required pressure heads, and roughness coefficients. The optimization
problem is formulated as a nonlinear programminggroblem which is
solved using a generalized reduced gradient method. Details of the
mathematical model formulation are presented along with example
applications. Results illustrate that uncertainties in demands, pressure
head requirements and roughness can have significant effects on the
network design and cost.

There is currently no universally accepted definition or measure of

the reliability of water distribution systems. In general, reliability is
defined as the probability that a system performs its mission within
specified limits for a given period of time in a specified environment.
Over the past two decades, there have been many models developed for
the analysis and the minimum cost design of water distribution networks
(e.g., Alperovits and Shamir, 1977; Quindry et al., 1981; Morgan and
Goulter, 1986; Lansey and Mays, 1987). Only a very few models have been
reported that attempt to consider the reliability of the network and the
various components. Coals and Goulter, (1985) presented three
approaches by which the probability of failure of individual pipes can be
related to a measure of the overall system reliability in a linear program-
ming minimum cost design procedure. No models explicitly consider the
uncertainties in demands, pressure heads, and pipe roughness.

49 1
 The real issue of water distribution system reliability concerns the
ability of the system to supply the demands at the nodes or demand points
within the system at required minimum pressures. The conventional
design process for water distribution systems is a trial and error procedure
that attempts to find a design that represents a least-cost solution that can
satisfy demands. These trial and error methods make no attempt to ana-
lyze or define any reliability aspects of the designed system and have no
guarantee that the resulting system is a minimum cost system. The resul-
ting system design is not based upon consideration of the various design
uncertainties.

Mays and Cullinane (1986) presented a review of methods that can

be used to define the component reliabilities for the various components
of a water distribution system. These methods are based upon using time-
to-failure and time-to-repair data for the various components of the water
distribution system to 'define reliability and availability. Su et al. (1987)
presented a procedure for modeling reliability in water distribution net-
work design that more realistically considers reliability. That is, the
reliability is defined in terms of the ability of the system to supply the
demands at the nodes or demand points within the system at or above
minimum pressure heads. This model uses component failure rates to
compute component failure probabilities which are then used to define
nodal and system reliabilities. A minimum cut-set method is used in the
nodal and system reliability detennination. The key issue in this approach
was to relate failure probability of the pipes to meeting specified demands
(flow rates) at minimum pressure heads at the demand nodes. The proce-
dure was linked to a nonlinear programming optimization model to
determine a minimum cost water distribution system considering nodal
and system reliabilities as constraints.

Water distribution systems are designed to service consumers over

a long period of time. Because the number and types of future consumers
are impossible to define with any accuracy, the projected future required
demands and required pressure heads for design are very uncertain. Ano-
ther uncertain parameter in the design of a system is the system capacity.
The capacity is affected by corrosion of pipes, deposition in pipes, even the
physical layout and installation of the system which has a marked effect
on the carrying capacity. The change in system capacity can be reflected in
the roughness coefficient of the pipes (Hudson, 1966). Since the impact of
the different mechanisms that decrease system capacity is not known,
there is uncertainty in the projections of the coefficients of roughness.
The variation of roughness is illustrated in the work by Hudson who
compared the Hazen-Williams roughness coefficient for seven U.S.cities
as a function of age of pipe.

492
 No models have been developed for the minimum cost design of
water distribution networks that directly consider the uncertainties in
demand requirements, pressure head requirements, and roughness coeffi-
cients. Previous models considered uncertainties in delivering flows and
pressure heads during pipe failures. There have been many works
reported in the literature that deal with uncertainties in water supply
forecasting and modeling; however, very little work has been performed
in developing a model that directly considers the uncertainties of required
demand and other system parameters in the design of water distribution
sys tems.

The objective of this section is to present a methodology which

incorporates the uncertainties in required demands, required pressure
heads and roughness coefficients in the design of water distribution
systems. In the technique, each of the parameters is considered as inde-
pendent. The required demand, Q, and the required pressure head are
dependent upon consumer needs whereas the roughness coefficient
depends upon other factors, such as the types of pipes in the system, age,
amount of deposition, etc. The methodology is presented through the
formulation of an optimization model for the design of water distribution
systems. This optimization model is based upon a nonlinear Chance-
constrained fofiulation and can be solved d i n g generalized reduced
gradient methods, such as GRG2 by Lasdon and Waren (1984)# Details of
the mathematical model are given in addition to examples to illustrate the
methodology.

14.2.2 Model DeveloDment

The basic optimization model for water distribution system design

can be stated in general form as

Min. Cost = f(Di,j) (14.2.1)

i,jEM

subject to the following constraints

j = 1,. ..,J (nodes) (14.2.2)

x h k = O k = 1,...,K(loops) (14.2.3)
i,jE k
H. 2 H. j = 1,...,J (I4.2.4)
I 7

493
 D..2 0 (14.2.5)
111

The objective function is to minimize cost as a function of the dia-

meter Di,j/ for the set of possible links, M, connecting nodes i and j in the
network. Constraint equation (14.2.2) is the continuity equation used to
satisfy demand at each node in which q,j is the flow rate in the Pipe
connecting nodes i and j, and Qj is the external demand at node j. This
constraint is written for each node j in the network. Constraint equation
(14.2.3)states that the sum of the head losses, hk, around each loop k = l,...,
K is equal to zero. Equation (14.2.4) defines the lower bound, €3j ,on the
pressure head, Hp at each node.

The discharge q,j in each pipe connecting nodes i and j can be

expressed using the Hazen-Williams, so that equation (14.2.2) can be
expressed as

(14.2.6)

where K, is a constant to account for flow units; Ci,j is the Hazen-

Williams roughness coefficient; Hi and Hj are the pressure heads at nodes
i and j, respectively; Li,j is the length of the pipe connecting nodes i and j;
and Di,jis the pipe diameter of the pipe connecting nodes i and j. Substi-
tuting the Hazen-Williams equation into equation (14.2.2) makes equation
(14.2.3) unnecessary since it is satisfied. The constraint set for the deter-
ministic model now consists of equations (14.2.4, 14.2.5 and 14.2.6).

Considering the demands, Qk the minimum pressure head

requirements, E p and the pipe roughness coefficients,
variables, the chance constrained formulation of the
expressed as

Min.Cost = [f(Di,j)] (14.2.7)

i,jtsM
subject to

0.54
H.- H.
(I4.2.8)
 j = 1,...,J (14.2.9)

D.. 2 0 (I4.2.1 0)
'PI

The objective function (14.2.7) is expressed in terms of minimizing

the costs. Constraint equation (14.2.8) is expressed as the probability, P( 1, of
satisfying demands, i.e. that demands are equalled or exceeded with
probability level a..Similarly, constraint equation (14.2.9) expresses the
1
mobability of the minimum pressure head being satisfied, i.e. the pressure
heads equal or exceed the minimum pressure head with probability level
bj-

14.2.3 Deterministic Equivalent Form of Model

-
The above model (14.2.7) (14.2.10) can be transformed to a deter-
ministic form using the concept of the cumulative probability distribution.
Although the theory is general and the variables may follow any distri-
bution, the demands, Qk pressure minimum heads, + H and roughness
coefficients, Ci,p are assumed to be normal random variables, with means,
p, and standard deviations, 0, expressed as

(14.2.11)

(14.2.12)

and
(14.2.13)

Constraint equation (14.2.9)can be written in terms of the standard

normal random variable as
r 1

(14.2.14)

Using the definition of the cumulative probability distribution, this can be

simplified to

495
 (14.2.15)

in which F is the cumulative probability.

The deterministic form of constraint equation (14.2.9) can then be

expressed as

(14.2.16)
Hi - %j2 (Ja
F

Because PH., SH. and pj are all specified, this constraint can be written as a
J- -J
simple bound constraint

(14.2.17)

Similarly, constraint equation (14.2.8)can also be expressed in a

deterministic form. The first step is to rewrite the constraint in the
following form

'(cqi,j-4
1 0) 2 aj (14.2.18)

and then express the pipe flow using equation (14.2.6)

0.54
H. - H.

[y (14.2.19)

Because both Qj and Ci!j are considered normal random variables

and are assumed statistically independent, the term on the left side of the
inequality

(14.2.20)

496
is also a normal random variable with mean

(14.2.21)

and standard deviation

rr

Constraint equation (14.2.19) can now be written as

(14.2.23)

which can be simplified to

s La.
J (14.2.24)

The deterministic form of constraint equation (14.2.19) is then

(14.2.25)

in which pwj and owj are defined by equations (14.2.21) and (14.2.22),
- and oc are equal to zero,
respectively. If the standard deviations OQ,W
the parameters are known with certainty. -
The deterministic formulation of the chance constrained model is
expressed by the objective function (14.2.7) and constraint equations
(14.2.25), (14.2.17), and (14.2.10) as

497
 Min. Cost = f ( Di,j) (14.2.7)
i,jEm

subject to

(14.2.25)

(14.2.17)

D.. 2 0 (I4.2.1 0)
14

This is a nonlinear programming problem in which equations

(14.2.17) and (14.2.10) are treated as simple bound constraints. Equation
(14.2.17) is a simple bound because the right-hand side

is known as previously discussed. Constraint equation (14.2.25) expresses

the relationship for the decision variables Hj and Di,j. The nonlinear
problem consists of one nonlinear constraint (14.2.25) for each node and a
simple bound for each decision variable, i.e., one for each pipe link and
one for each node. The total number of decision variables is the sum of
the number of nodes (number of €39 and the number of pipe links
(number of diameters).

The model formulation assumes that the decision variables, H and

D, are independent of the random variables, Qj, Hj, and C. This can be
done through the zero order decision rule for chance-constrained pro-
gramming (Charnes and Cooper, 1963; Charnes and Sterdy, 1966).

14.2.4 Solution Techniaue

The above deterministic model formulation of the chance con-

strained model is nonlinear because of the nonlinear objective function
and nonlinear constraints (14.2.25). The generalized reduced gradient
code, GRG2, by Lasdon and Waren (1984)has been used to solve the deter-
ministic form of the chance constrained model. This model is briefly
described in section 14.1.

498
14.2.5 Example Applications

To illustrate the use of the model, two examples are included. The
first is a simple hypotheticd network, shown in Fig. 14.2.1, which has two
loops and eight pipes, each 3,280 feet in length, and mean demands at each
node as shown in the figure. All the nodes are assumed to be at the same
elevation and the pressure head at the source, node 1, is 196.8 feet. The
mean nodal pressure head requirement at each node is 100 feet, and the
mean Hazen-Williams roughness coefficient is 100 for each link. The
formulation, however, does not assume that all nodes have the same
pressure head requirement or that they are on a level plane. Also, a gen-
eral network with one or more source nodes can be incorporated, if
desired. The pipe cost equation used in both examples is

COST = 0.331LD’” (14.2.26)

where D is the pipe diameter in inches and L is the length of the link in
feet. This equation is representative for thickness Class 23 cast iron pipe
and includes the cost for purchasing, hauling, and laying (U.S. Army -
Corps of Engineers, 1980). Any cost function, linear or nonlinear, could be
used as equation (14.2.26) is used only for the purpose of illustrating the
model application.

Several computer runs were made using various values of the

standard deviation of the demand, pressure head, and roughness coeffi-
cient, in order to illustrate the impact of different levels of uncertainty on
the design cost. The standard deviation selected for the nodal demands
were 0.0, 0.10, and 0.25 MGD. Selected standard deviations for the pressure
heads were 0, 5, and 10 f t and for the Hazen-Williams roughness were 0,5,
and 10. A standard deviation equal to zero refers to the case of no uncer-
tainty, and the larger the standard deviation, the greater the uncertainty.
Computer runs were made for various values of a and P ranging from
0.5 to 0.99. Using a = 0.5 (p = 0.5) is equivalent to using mean values of the
nodal demands and the pressure heads. Higher levels of a and p refer to
more stringent performance requirements so that the likelihood of system
failure is reduced.

Figure 14.2.2 illustrates the impact of increasing the standard

deviation of the roughness coefficient and the nodal demands, indepen-
dently, while assuming the nodal pressure head requirement is known
- = 0). With only Hazen-Williams roughness being
with certainty (OH
uncertain (OH = 0 and OQ = o), as expected, the higher the reliability
requirementTthe greater the cost of the design. The same is true for

499
1.0MGD 1.0MGD

2 1

7 3

* 1.0MGD

8 5

Figure 14.2.1 Exampie I: Eight Pipe Network

500
 420000 ~i i

410000

400000

A 390000
UP
Y

4J
U
J
380000
0
u 370000

360000

350000

340000

330000
0.5 0.6 0.7 0.8 0.9 1.0

Figure 1 4 . 2 . 2 C o s t vs R e l i a b i l i t y Requirement (%=o)

365000

355000

-
a?
Y
350000 1
JJ
(D 345000
0
0

340000

335000

330000
0.5 0.6 0.7 0.8 0.9 1.0

P
Figure 14.2.3 Cost vs Pressure Head R e l i a b i l i t y Requirement
(0= 0 =o>
Q C

501
different standard deviations of nodal demand and known Hazen-
Williams roughness coefficient (OH= 0 and oc = 0). Figure 14.2.3 shows
the change in cost with increasing-b for two standard deviations of nodal
pressure and no uncertainty in the nodal demands or roughness coeffi-
cients (OC = 0, OQ = 0 >.The same trend is apparent in this figure as seen in
Fig. 14.2.2.

The optimal continuous diameters could be converted to discrete

diameters by considering them as equivalent pipe diameters and deter-
mining the lengths of two pipes which make up the link and have the
same hydraulic characteristics. All of the optimal solutions were branched
networks as expected for the optimal design of systems under a single
demand pattern. The nonlinear programming problem for this example
consisted of 16 decision variables, 6 nonlinear constraints, and 14 simple
bounds. The computation time required to determine a design was
approximately 2 seconds of CPU time on the University of Texas Dual
Cyber 170/750.
The second application considered a more realistic size network
consisting of 33 pipes and 16 nodes (Fig. 14.2.4) with the pipe lengths listed
in Table 14.2.1. This system was also assumed to be on a level plane
although this restriction is not necessary since the pressure head require-
ment may be different for each node. The demand is assumed to be
uniform throughout the system with a 1 mgd mean demand at each node.
The pipe roughness has a mean of 130 for all pipes and the mean mini-
mum pressure head requirement is 92.3 ft. at every node. The pressure
head at the source node was fixed at 135.0 ft. The model was executed for
values of a and p equal to 0.5,0.75, and 0.90, with the pipe diameters and
total costs listed in Table 14.2.2. As in the previous example, the cost of the
system increases with the reliability requirement. The networks for a$ =
0.5, 0.75 and 0.90 are presented in Figs. 14.2.5,14.2.6,and 14.2.7, respectively.

The nonlinear programming problem for this example consisted of

49 decision variables, 16 nonlinear constraints, and 49 simple bounds. The
computation time required for a typical problem was 100 CPU seconds on
the University of Texas Dual Cyber System. However, when computing
the gradients in GRG2 the numerical finite difference scheme was
employed. Analytically calculating the gradients of the constraints which
can be written in a closed form would result in significant savings with
the final computation time being about 20-25%of that required when
using the numerical gradients.

502
 SaURCE
( n o t to s c z l e )

Figure 14.2.4 Example 2 Network

503
 TABLE 14.2.1 Pipe Lengths for Example 2

Pipe No. Length (ft.) Pipe No. Length (ft.)

1 12000 17 9000
2 12000 18 6000
3 12000 19 6000
4 9000 20 6000
5 6000 21 6000
6 6000 22 6000
7 6000 23 6000
8 6000 24 6000
9 6000 25 6000
10 6000 26 6000
11 6000 27 9000
12 6000 28 6000
13 6000 29 6000
14 6000 30 6000
15 6000 31 6000
16 6000 32 12000
33 12000

504
TABLE 14.2.2 Optimal Solution for Different a and b for Example 2

Pipe a = b = 0.5 a = b = 0.75 a = b = 0.90

1 22.8 10.4 24.2

2 18.8 21.7 15.9
3 20.9 30.0 31.4
4 0.0 0.0 0.0
5 0.0 0.0 11.4
6 21.5 0.0 11.4
7 0.0 0.0 16.4
8 0.0 0.0 0.0
9 0.0 0.0 12.9
10 15.3 0.0 0.0
11 10.5 0.0 0.0
12 10.5 11.7 0.0
13 0.0 11.7 0.0
14 0.0 0.0 0.0
15 10.0 10.5 0.0
16 10.0 10.5 12.7
17 11.5 14.5 0.0
18 0.0 0.0 0.0
19 0.0 12.9 1.3
20 12.5 0.0 0.0
21 0.0 0.0 14.1
22 0.0 0.0 0.0
23 0.0 0.0 0.0
24 9.6 21.5 28.9
25 0.0 18.2 0.0
26 17.5 10.2 0.0
27 0.0 0.0 12.8
28 15.7 0.0 0.0
29 0.0 16.6 17.6
30 12.8 13.4 14.1
31 0.0 11.2 17.6
32 0.0 0.0 0.0
33 0.0 0.0 0.0

cost $2,144,300 $2,258,400 $2,613,533

505
N

SOURCE

( n o t to scale)

Figure 14.2.5 Optimal Network f o r (a=P=0.50)

506
N

SOURCE

( n o t to s c a l e )

Figure 14.2.6 I Optimal Network f o r (a=P=0.75)

507
N

SOURCE
( n o t to s c a l e )

Figure 14.2.7 Optimal N e t w o r k for ( a=p=o.gO )

508
Summary

In most of the previous sections of this report, the mechanical

reliability of the system, i.e. the likelihood of component failure, and its
effect on the hydraulic performance was the primary measure of a sys-
tem's reliability. However, the hydraulic performance reliability is also
affected by the uncertain nature of the demands placed on the system, as
discussed in Chapter 8, and the uncertain rate of decay in the network, as
reflected in the pipe roughness coefficients. Using a nonlinear chance
constraint formulation, a methodology to consider these uncertainties has
been developed. The approach is similar to the load-resistance analysis
presented in Chapter 8 but considering the ability of the system to meet the
user's demands on flow and pressure head rather than an individual
element.

The application of the methodology showed that as the reliability

measure increased, the cost of the system increased as did its ability to
perform satisfactorily under a wider range of conditions. For a specific
network, a relationship between cost and the reliability measure can be
developed which will allow a decision-maker to better understand this
trade-off. Thus, more informed decisions can be made which better reflect
the goals of the decision-maker and the community.

14.3 RELIABILITY-CONSTRAINED PIPE NETWORK MODEL

14.3.1 Introduction

The explicit consideration of reliability is one of the most difficult

tasks facing researchers working on the development of least-cost opti-
mization design models for water distribution networks. Reliability has
traditionally been addressed in a deterministic sense by providing redun-
dancy in the form of loops within the distribution network. The existence
of loops, however, provided a major difficulty for least-cost design models,
which by their very nature would try to remove redundant or unnecessary
loops in the search for minimum cost (Templeman, 1982). In many cases,
the problem was overcome by specifying minimum allowable pipe sizes in
all links (e.g., Alperovits and Shamir, 1977; Quindry et al., 1981). While
connectivity is ensked by these minimum pipes, redundancy in terms of
having adequate and independent flow paths to each node is not guar-
anteed, and the resulting networks can operate as implicitly branched,
non-redundant systems (Goulter et al., 1986). This redundancy problem
has been partially overcome in the design models of Rowel1 and Barnes
(1982) and Morgan and Goulter (1985) by consideration of a wide range of

509
 loading conditions and pipe failure combinations such that every Iink was
on a primary suppIy path for at least one design condition. The improve-
ment in redundancy consideration was obtained, however, at the cost of
mathematical exactitude within the technique.

There is, however, a critical factor missing in the way in which

redundancy /reliability is considered in these models. Reliability of a water
distribution network is also defined by the probability of the network fail-
ing to perform, either through failure of one or more of its components,
e.g., pipes or pumps, or through the demand on the system being greater
than that assumed for design purposes. Very little research activity has
been reported on the inclusion of these probabilistic issues in optimization
design models for water distribution networks. This section reports on a
means of incorporating these probabilistic issues in iterative procedures
based upon network design models.

14.3.2 Problem Statement

Most of the approaches to the probabilistic aspects of reliability in

water distribution systems can be separated into two main groups. Inclu-
ded in the first group are those approaches and models which address the
reliability of the system as a whole. These approaches examine the relia-
bility of the major components, namely, the supply, treatment, and
distribution stages, in terms of the performance of the overall system.
These problems are addressed by Hobbs (19851, Shamir and Howard (1985),
Germanopoulis et al. (1986), and Hobbs and Beim (1986).

The approaches and models in the second group are directed at the
reliability of specific components of the overall system, e.g., the distribu-
tion network. This section discusses issues related to the reliability of the
distribution network only and as such falls into this second area. The
concepts examined relate to pipe size choice and its role in determining
the network reliability. Some of the earlier work in the field of distribu-
tion network reliability examined the reliability question in terms of the
probability distribution of the demand on the network and the implica-
tions of the flow required within a pipe exceeding its design capacity, e.g.,
Cullinane (1986) and Tung (1986). Wagner et al. (1986) examined the
reliability of the network by looking at the actual shape or layout of the
network. Studies by Goulter and Coals (1986), Cullinane (19861, and Mays
et al. (1986) addressed the reliability question through use of the probability
of individual pipe failure. Both Goulter and Coals (1986) and Mays et al.
(1986) also showed how the concept of pipe breakage probability could be
incorporated into a least-cost model for water distribution network design.

510
 In this paper, the least-cost design models of Alperovits and Shamir
(1977) and Goulter and Coals (1986) are combined and extended to recog-
nize the probabilistic nature of the reliability issue. The extension is
achieved through the introduction of reliability constraints, and by the
inclusion of probabilistic factors in the coefficients in the formulation.
The reliability constraints restrict the probability that the number of pipe
breaks in a particular link in a given time period is greater than an accep-
table value. Restrictions on the probability that the required flow in a pipe
is greater than its design capacity is achieved by modifications to the flow
values used in the formulation.

A network design provided by an optimization model incorpora-

ting these reliability constraints is, however, only the least-cost "solution"
for the probabilistic parameters applied through reliability constraints and
the model coefficients. Furthermore, the probabilistic values do not in
themselves give true or exact measures of the r e d network reliability. A
series of computational steps are required to convert the probability values
into reasonable measures of overall system performance. Various
numerical expressions which can be used to guide changes to the model
parameters such that the solutions obtained achieve improved overall
reliabilities at least cost, are examined and evaluated in the following
sections.

14.3.3 Model Formulation

The procedure is primarily based upon the least-cost linear program

design model of Alperovits and Shamir (1977),which can be summarized
as follows:

Objective Function:

(I4.3.1)

subject to

Length Constraint:
Ngi)
A L..'I = LNi for all links i (14.3.2)
j= 1

511
 Minimum Head Requirements at Nodes:

m
for all nodes m (14.3.3)

Head Losses Around Loop:

c Ng;ijLij
i E p(m)
= 0 for all loops (14.3.4)

where

C.. cost per unit length of j-th diameter in link i ($/m)

1'
MHm maximum permissible head loss between source
and node m

HS - Hm
HS pressure head at source (m)
L.. length of pipe of diameter j in link i
'I
Hm minimum allowable pressure head at node m (m)

Jij
hydraulic gradient of pipe of j-th diameter in link i
(derived from the Hazen-Williams formula) (m/m)
LNi total length of link i (m)

ND(i) number of candidate diameters in link i

NL total number of links in network

p(m> = set of links on path from source to node m

The pipe breakage considerations of Goulter and Coals (1986) can be

included in this formulation by the addition of the following constraint

'
set which restricts the average number of breaks per time period permitted
in each link.
ND i )
r..L.. 2 Ri . for all links i (14.3.5)
'I 1J
i

512
 where

r.. = average number of breaks per year per unit length for j-th
'I diameter pipe in link i (breaks/year/m>
Ri = maximum acceptable average number of pipe breaks per
year in link i

Probabilitv Distribution of Pipe Breakages

The formulation described by equations (14.3.1)- (14.3.5)does not

have any recognition of the probabilistic features of either the rij factors or
the Q k demand characteristics. In order to incorporate these probabilistic
issues, the formulation must be modified as follows. The rij in equation
(14.3.5)is the number of failures per unit length for a particular pipe dia-
meter, which may vary significantly from year to year. In order to restrict
the probability that the number of breaks in a link in any given year is
greater than some acceptable value, equation (143.5) can be rewritten as
follows

Nf)rijLij I mi for all links i (14.3.6)

j= 1

where

RHS, -

inverse of the cumulative distribution function

of r..
1J
probability of the number of breaks per year in
link i
VAR(rij) = variance of r..
'I

Probabilitv Distribution of Demand at a Node

In designing a water distribution network, the flow demands that

must be provided at minimum specified pressures are generally taken as
deterministic values. Little or no consideration is given to the possibility
of these demands being exceeded. A more realistic view of the design
flows, however, would recognize that the demands themselves are

513
probabilistic and that there is a finite probability that the demands on the
system will at some time be greater than the design capacity. For example,
one criteria for design flows may be the 10-year, 1-day maximum demand.
The 20-year, I-day maximum demand will be larger than this value and is
likely to occur at least once during the life of the distribution system.

Once the actual demand exceeds the design capacity, at least ont? link
in the network will be required to carry a flow greater than its design
capacity. Two situations can arise from this condition. The pipes can
deliver the flow but at pressures below the minimum specified or the
pipes will not be able ;t provide the flow at all. Either kituation represents
a "failure" of the network.

Tung (1986) approached the problem by assuming that the required

demand and head at a demand node are random variables, with known
probability distributions. A series of chance constraints for the demands
and pressure heads at each node could than be formulated for inclusion in
a least-cost design model. The approach used in this model does not allow
the use of explicit chance constraints for these flow issues. Rather, the
probabilistic component is addressed through inclusion in the coefficients
of the model that are directly related to flow, e.g., Jij.

The actual incorporation of flow is achieved through the use of the

following equation to define the demands at each node.

NQ, = E Demandk) + F-'( pk) SID( Demandk) (14.3.7)

(
where

NQk
- value of the demand at node k to be used
as the design flow (volume/unit time)
$Demandk) - avera e (ex cted) value of demand at
node % (vocme/unit time)
ST Demand = standard deviation of demand at node k
k) (volume/unit time)
- inverse of the cumulative distribution
I

function of Q,
- probability that demand at node k will
P, exceed the design demand

Solution of the optimization formulation described by equations

(14.3.11,(14.3.3),(14.3.4),(14.3.6),and using the hydraulic gradient Jij values

5 14
derived from (14.3.7) provides a least-cost network with explicit consider-
ation of both breakage or mechanical based reliability issues and flow
exceedance probabilistic reliability measures contribute to the assessment
of reliability. However, as mentioned earlier, the probabilistic measures
contained with equations (14.3.6) and (14.3.7) are, however, only surrogates
for the true reliability of the system, and do not give a real indication of
the reliability of the network. The means to combine the two measures
into a single comprehensive measure of pipe network reliability is yet to
be available. Once a network solution has been obtained from the
optimization model, it must therefore be examined to see if it contains
satisfactory levels of reliability as assessed by the design engineer. If the
network is judged to have satisfactory reliability, the problem is solved. If,
on the other hand, it is felt that the network is unsatisfactory, then adjust-
ments must be made to the parameters of the least-cost optimization
model and the network solved again.

The question now arises as how to change the parameters of the

model such that the requested improvements in reliability are obtained at
the smallest increase in cost. In other words, a method which provides a
measure of the sensitivity of the least-cost solution to changes in "relia-
bility" oriented model parameters is needed. Before proceeding with an
examination of these methods, it is important to note that there are a
number of similarities in the way in which improvement to one of the
measures of reliability causes improvements to other measures.

A number of researchers, e.g., Clark et al. (1982), Ciottoni (1983),and

Kettler and Goulter (1985) have shown that pipe failure rates are strongly
related to diameter with the larger diameter pipes having lower rates of
failure than pipes with smaller diameters. Improvements to the pipe
breakage or mechanical reliability can therefore be achieved by selecting
larger diameter pipes. However, these larger diameter pipes also have
greater hydraulic capacities for the same pressure conditions and therefore
also provide increased flow capabilities. These increases in the flow capa-
bilities reduce the probability that the flow required in a link (derived
from demands at nearby nodes) will be larger than its design capacity.
Similarly, improvements (reductions) to the flow exceedance probabilities
at nodes will also improve mechanical reliability by causing the selection
of larger diameter, high capacity, pipes which, in turn,have reduced rates
of breakage.

Under such conditions, it might be expected that unsatisfactory

system reliability can be identified and improvements subsequently
achieved through modifying either the mechanically based Ri and F-l(y)
parameters in equation (14.3.6) or the flow exceedance reliability based F-
1(pk) parameter in equation (14.3.7). Unfortunately, this is not generally

5 15
true. Areas with good mechanical reliability may in fact have low flow-
based reliability probability. Consider a pipe near a source. The pipe %
likely to be of large diameter as it carries flow for a large downstrem
portion of the network and will therefore have a low failure rate and
correspondingly high mechanical reliability. The same pipe, however,
may have a low flow exceedance-based reliability if the probability of 5 e
flow exceeding the design capacity is relatively large.

It can be seen, therefore, that while there is some complementarv

interaction between the two reliability conditions, care must be taken to
ensure that areas of low reliability are always identified and the correct
steps to overall system improvement undertaken. The following sections
examine a number of ways of identifying system parameters which should
be changed in order to achieve improvements to system reliability at least
increase to system cost.

14.3.4 Means of Improving- System

- Reliabilitv

There are two major issues which must be examined in determin-

ing how to improve system reliability. The first is which measure is to be
used to define system reliability. The second is which parameters should
be modified in order to achieve the desired improvement in reliability.
Unfortunately, relatively little is known about either problem. Most
optimization design models simply provide a network solution that may
or may not address the probabilistic reliability issues of this paper. Gen-
erally, no assessment of reliability is provided, and furthermore, no
attempt is made to modify model parameters if reliability should be
unsatisfactory.

Cullinane (1986) suggests a measure for system performance which

considers both the mechanical reliability of the network components and
the implications of failures of those components on the hydraulic perfor-
mance of the network relative to its design criteria. The measure was not
incorporated into an optimization design model. No indication was given
on how to identify or change model parameters such that the desired
improvements to network reliability were achieved with the lowest cost
implications. Goulter and Coals (1986) developed optimization based
procedures which use a mechanical reliability measure to define system
reliability in t e r n of the probability of node isolation, i.e., the probability
of all links connected to a node failing simultaneously. A means of deter-
mining which link in the vicinity of the node has the worst probability of
node isolation to change in order to obtain improved system reliability
(reduced probability of node isolation) at least-cost was also provided. The
measure used to identify the link for which the optimization model para-
meters were to be changed was derived from the dual variables in the

516
linear programming solutions and the expected number of breaks in the
links connected to the critical node.

It is important to note, however, that while the work of Goulter and

Coals (1986) was one of the few to address the criteria for improving distri-
bution network reliability, it was very simplistic and took no account of
either the flow exceedance issues in network reliability or the probability
distribution of the rates of pipe breakage. One of the primary features
examined in this section is how to combine the mechanical and flow based
reliability factors to provide a balanced measure for identifying how to
modify the parameters in equations (14.3.6),and (14.3.7). There are, how-
ever, similarities between the method of Goulter and Coals (1986) and the
problems of the two reliability measures which can be exploited. Goulter
and Coals (1986) lower the deterministic Ri values in their model in order
to achieve the desired improvements. A reduction in y in equation
(14.3.7) has the same general effect, although it achieves it through a
probabilistic measure. The dual variable based method of Goulter and
Coals (1986) therefore appears to have some applications to this extended
problem and is used as the basis for three of the following measures.

At present, there is no explicit means of combining the two proba-

bilistic failure measures in either a general statement of system reliability
or in a single explicit measure which can be used to guide changes to the
optimization to achieve more reliable networks. For this reason, a series
of overall system reliability and improvement measures are formulated
and critically compared.

The reliability measure chosen to assess system performance is a

heuristic measure termed "the probability of 'node failure'.'' This mea-
sure incorporates both pipe and demand failure through the following
expression

P(no "node failure") = P(no "node isolation")

.
P (no "demand failure") (14.3.8)

where P( ) = probability of Occurrence of event.

It should be noted that the term P(no "node isolation") is the relia-
bility measure developed by Goulter and Coals (1986). However, in this
case, the simplistic consideration of node isoIation is improved by com-
bining it with the flow based probability measure P(no "demand failure").

The next problem is that if any part of the network is considered to

be unreliable, according to equation (14.3.8),how should the parameters of
the optimization model be modified in order to achieve the desired

517
improvement? One obvious consideration is that any improvement in
reliability should increase the overall network cost. However, before
proceeding with examination of the methods used to identify the appro-
priate model changes, it is necessary to define a new parameter. Let WBi
represent the "relative importance" or weighting of link i in the network.
In the context of this paper, the relative importance of a particular link is
determined by how much flow that link carries relative to the flows in
other links of the network. The actual value of the weight for a particular
link is calculated by

Flow in link i
WB, = (14.3.9)
Total demand in network

The range of measures which were examined as to their suitability

for determining how to improve the network reliability is described in
Table 14.3.1. It should be noted that application of each measure requires
that the portion of the network which has unsatisfactory reliability be
identified. Within the table, the measures are described in relation to the
reliability at the single node, k, having the worst "reliability" (as defined by
equation 14.3.8)and those links connected directly to that node are to be
addressed by the measures. The measures are, however, generally appli-
cable to more than one node at a time, with the appropriate connections
being examined in each case. The description relative to a single node is
for simplicity in explanation only.

14.3.5 Application of Methodolodes

The measures were assessed by application to the network shown in

Fig. 14.3.1. The pertinent data for the network is given in Tables 14.3.2-
14.3.4. All links in the network have total length of 1,000 m.

As equation (14.3.6)is nonlinear, the model cannot be solved

directly by linear programming. It can obviously be solved by nonlinear
optimization approaches. However, since the value of the dual variable is
a crucial element in most of the improvement measures examined (see
Table 14.3.1)' and the nonlinearity in equation (14.3.6)is not "severe,"
linear programming with successive approximations was used. Experi-
ence with the formulations and this solution approach showed the
combination to be very efficient.

In applying the model to the sample problem, the following "step

sizes" were used to modify the model parameters when reliability was
deemed to be unsatisfactory. For modifications (reductions to a particular
value Ri, a step size corresponding to 10% of the initid Ri was used. For

5 18
 Table 14.3.1 Description of Measures to Determine How to Improve
Network Reliability

Means of Determining Means of Affecting

Measure Which Link to Change Improvement

A Link with aCOST/aRHS Decrease the value of Ri for

value, i.e., link minimum link selected
dual variable for equation
(14.3.6)

B Link with minimum value Decrease the value of Ri for

of link selected

BR, WB,

where BRi = actual number

of breaks per unit time in
links (= RHSi when equation
(14.3.6) is binding for that link)

C Link with minimum value Decrease the value of Ri for

of link selected
aCOST/aRHS.1

where a(Relk)/ aRHSi =

rate of change in probability
of isolation of node k with
change is RHSi for link i
(Goulter and Coals, 1986,
measure)

D N.A. Increase confidence interval

of demand at node by
decreasing the value of bk

519
SOURCE

LEGEND:

-
A Link Number
-0- Node Number
Direction of f l o w in Link

Figure 14.3.1 Layout of Example Problem

 Table 14.3.2 Demand Characteristics of Each Node

Elevation of Minimum Average1 Standard Deviation

Node Node Pressure Head Demand of Demand
(m> (m) (m3/hr) (m3/hr)
1(source) 380.0 - - -
2 320.0 30.0 150.0 178.1
3 300.0 30.0 50.0 59.4
4 330.0 30.0 50.0 59.4
5 310.0 30.0 50.0 59.4
6 290.0 30.0 50.0 59.4
7 320.0 30.0 50.0 59.4
8 300.0 30.0 150.0 178.1
9 290.0 30.0 100.0 118.8

1Demand is assumed to be normally distributed.

Table 14.3.3 Link Data

Initial Flow1 Link Weighting

Link in Link (WBi)
(m3/hr)

1 600.0 0.462
2 200.0 0.134
3 700.0 0.538
4 100.0 0.077
5 100.0 0.077
6 200.0 0.154
7 100.0 0.077
8 400.0 0.308
9 100.0 0.077
10 100.0 0.077
11 300.0 0.231
12 100.0 0.077
IThese flows correspond to a demand confi-
dence of 80% for all nodes, i.e., Pk = 0.20 for all k
= 1,2, ...,9

521
 Table 14.3.4 Pipe Cost and Breakage Data

Diameterl Cost Average Number of Variance of Breaks

(mm) ($/m) Breaks per yr per km per yr per km

25 (1.0) 2.0 1.552 0.0262

51 (2.0) 5.0 1.so 0.023

76 (3.0) 8.0 1.45 0.023

102 (4.0) 11.0 1.35 0.020

152 (6.0) 16.0 1.05 0.012

203 (8.0) 23.0 0.75 0.006

254 (10.0) 32.0 0.40 0.014

305 (12.0) 50.0 0.10 0.014

356 (14.0) 60.0 0.082 0.0142

406 (16.0) 90.0 0.062 0.0172

457 (18.0) 130.0 0.052 0.0172

508 (20.0) 170.0 0.042 0.0202

559 (22.0) 300.0 0.042 0.0232

610 (24.0) 550.0 0.042 0.0262

IDiameters are given in millimeter equivalents of pipes whose diameters

are given in inches in the parentheses alongside each value.

2These values are estimates, other values are taken from the work of
Kettler and Goulter (1985).

522
modifications to confidence level of demand P k at a particular node k, the
following successively tighter confidence levels were used: 80% (initial
value), 90%, 95%, 97.5%, 99%, 99.5%, and 99.9%. Whenever it was neces-
sary to change the demand at a particular node to improve network
reliability, for example, by using measure D, the resulting design flow in
each link was calculated using the formulae listed in Table 14.3.5.

A summary of the results obtained from each measure is given in

Table 14.3.6. The lengths and diameters of the pipes selected using each
measure are described in Tabie 14.3.7. The base solutions in Tables 14.3.6
and 14.3.7 correspond to the solution with Pk = 0.20 and no pipe breakage
constraints (equation 14.3.6) binding. The termination criterion for each
measure, i.e., the point where the network was considered to have satis-
factory reliability, was maximum probability of no node failure 2 65%. It
was not possible to get the probabilities of no node failure to be exactly
65%. With the exception of measure XI@), the values shown in Table
14.3.6 represent the closest that the probabilities could be matched without
using unrealistically high numbers of runs with minor modifications to
the appropriate parameters.

The different results shown for D(a) and D(b) in Table 14.3.6 arise
from use of a different stopping criterion. The criterion for D(a) was the
same as for Measures A, B, and C, namely the probability of no node failure
2 65%. The D(b) case arose from examination of the results for D(c),,which,
relative to the results for the other measures, showed a higher number of
expected breaks per year. With Db),an attempt was made to see if a
solution with a greater cost than that obtained for Measures A, B, and C
would result in a smaller number of expected breaks per year. It can be
seen that such a result was not obtainable under the speafied conditions.

Discussion of Results

Examination of the results show that there is very little difference

between the solutions obtained by measures directed at improving the
reliability by modification to links, Lea,for Measures A, B, and C. The dif-
ferences between the results is within the accuracy of the solution given
the estimates of the parameters used in the formulation. The overall
similarity of the results for the three measures suggests that if "link
improvement" methods are to be used, the simplest measure, namely
Measure A, is most appropriate.

Both applications of Measure D, which is the only measure directed

at flow rather than component failure, produced improvement to the
reliability at less cost than either of Measures A, B, or C (see Col. 3 of Table
14.56). This observation suggests that Measure D is the most efficient

523
 Table 14.3.5 Formulae to Determine Design
Flows in Each Link

Link Flow In Link

Table 14.3.6 Summary of Results

Cost Worst P (no AP(no "node ExpectedNo.

Solution ($ x 103) "node failure") failure") of Breaks per
(%I A Cost Year in Network

Base solution 337.5 55.7 - 7.621

(80% flow
confidence level
and pipe breakage
constraints not
binding)

Measure A 359.1 65.4 2.72 '6.755

Measure B 359.7 65.5 2.62 6.744
Measure C 360.4 65.8 2.66 6.775
Measure D(a) 354.4 67.7 4.30 7.111
Measure D(b) 363.9 70.0 3.29 6.901

524
 1

Table 14.3.7 Lengths and Diameters of Pipes Selected

1 Measure
Base Solution A B C
Diam.1 Length Diam.1 Length
~-
(mm) (m> (mm) (m)
254 563 254 563 254 563 254 563 254 320 305 1000
305 437 305 437 305 437 305 437 305 680
203 1000 152 71 152 66 254 862 203 427 203 415
203 461 254 934 305 138 254 573 305 585
254 468
356 1000 356 1000 356 1000 356 1000 356 1000 356 1000
152 491 152 491 152 431 152 308 152 185 152 167
203 231 203 231 203 343 203 571 203 799 203 833
254 278 254 278 254 226 254 121 254 16
152 107 203 323 152 265 152 479 203 1000 203 1000
203 893 254 677 203 376 203 203
254 359 254 319
152 320 152 320 152 304 152 270 152 152 152 52
203 548 203 548 203 579 203 642 203 848 203 948
254 132 254 132 254 117 254 88
 Table 14.3.7 (cont'd)

Measure
Link Base Solution A 13
Diam.1 Length Diam.1 Length Diam.1 Length
(mm) (m) (mm) (m) (mm) (m)
7 152 612 152 167 152 167 152 167 152 253 152 140
203 7 203 833 203 833 203 833 203 673 203 860
254 381 254 74
8 254 1000 254 1000 254 1000 254 1000 254 763 305 663
305 237 356 337
LJl
9 152 433 152 433 152 493 152 615 152 185 152 242
rJ 203
m 203 339 203 339 203 228 254 285 799 203 694
254 228 254 228 254 279 254 16 305 64
10 152 167 152 42 152 282 152 225 203 1000 203 1000
203 833 203 836 254 718 203 497
254 122 254 278
11 203 400 203 400 203 400 203 400 203 282 203 200
254 600 254 600 254 600 254 600 254 718 254 800
12 152 346 203 72 203 357 203 72 152 40 203 1000
203 501 254 928 254 643 254 928 203 960
254 153

IThe actual pipe diameters used were in Imperial units (in), reflecting the availability of break records for these pipes. The
values given in this table are the "rounded-off'metric equivalents of those diameters.
measure to use to improve node reliability. Its use appears to provide
more flexibility in the choice of pipe diameters. Part of this flexibility is
due to the fact that the improvement is not directed at a single link
connected to the node of interest,'but rather to a region of the network
where the reliability problem exists. As described in Table 14.3.5, increases
in the design demand affect flows and subsequent pipe choices throughout
the network. This "broad-scale" approach seems to permit the optimiza-
tion model to adjust large sectors of the network resulting in cheaper
networks for the same cost. This "broad-scale" approach is also intuitively
more attractive, as it recognizes relationships within the network as a
whole.

It is also interesting to note that both applications of Measure D do

not reduce the expected number of breaks per year to the same extent as
the first three measures. If reduction in number of breaks per year is an
underlying objective of the problem, this feature may be considered a
weakness. The smaller reduction in breaks required to achieve the better
reliability does, however, support the general conclusions in the previous
paragraph. Improvement in network reliability has not been achieved
simpiy by the "mechanical" improvement used by the other three
methods. The improvement in reliability is achieved by a more compre-
hensive consideration of network performance.

Based on the above features of the four measures, it appears that

Measure D is the best approach to use to improve network reliability.
These conclusions are tentative, however, as they may be a function of the
network chosen and the heuristic reliability measure chosen. The relia-
bility measure of probability of "no node isolation" is, however, one of the
most comprehensive measures for pipe network reliability currently
available in that it explicitly recognizes both the pipe breakage and
demand exceedance issues of the reliability question.

14.3.6 Summarv

A general philosophical framework for approaching the reliability

issue is the design of optimization-based network design models. The
framework uses a general optimization model which includes considera-
tion of the probability of component, in a particular pipe, failure and the
probability that the actual demand at a node will exceed the design value.
The component failure issue is addressed through a set of chance con-
straints which defines acceptable levels for probability of pipe breakage.
The probability of demand at node being greater than the network capacity
to supply is addressed through the use of probabilistic-based flow values
from which the model parameters, such as hydraulic gradient, are derived.
A new "measure" describing the reliability of the networks in terms of the

52 7
probability of "no node failure" has been developed and included in the
overall approach. Although the measure is heuristic, it explicitly com-
bines mechanical, e.g., pipe, failure probabilities and the probability of a
"failure" caused by demand being greater than the design value.

A series of measures which can be employed to guide modification

to network design to achieve improvements in reliability at least-cost were
investigated. Based upon the results of application to a sample network, it
appears that improving the probability that design capacity is greater than
or equal to actual demand, as opposed to reducing the probability of com-
ponent failure, provides the most improvement in reliability at least cost.
If, however, reduction in the expected number of pipe breaks within a
network over a given time period is a critical issue, measures which
address the probability of pipe failure appear to be more appropriate.

These conclusions are preliminary, however, and further work on a

wider range of networks is needed before any general recommendations
can be made.

REFERENCES

Alperovits, E. and U. Shamir, "Design of optimal water distribution

systems," Water Resources Research, Vol. 13, No. 6, pp. 885-890, 1977.

Billinton, R and R.N. Allan, Reliabilitv Evaluation of Enqineering

Systems: Concepts and Techniaues, Pitman, London, 1983.

Charnes, A. and W. Cooper, Deterministic Equivalents for Optimizing and

Satisfying Under Chance Constraints, Operations Research, Vol. 11, No. I,
pp. 18-39,1963.

Chames, A. and A. C. Sterdy, "A Chance-Constrained Model for Real-

Time Control in Research and Development Management," Management
Science, Vol, 12, No. 8, pp. B-353 to B-342, April 1966.

Ciottoni, A.S., "Computerized data management in determining causes of

water main breaks: The Philadelphia case study," Proceedings of the 1983
International Symposium on Urban Hydrology, Hydraulics, and Sediment
Control, University of Kentucky, Lexington, Kentucky, pp. 323-329, 1983.

City of St. Louis, "1985 Main Break Report,'' Water Division, City of St.
Louis, Missouri.

528
Clark, R.M., C.L. Stafford and J.A. Goodrich, "Water distribution systems-
A spatial and cost evaluation," journal of the Water Resources Planning
and Management Division, ASCE, Vol. 100, No. WR3, pp. 243-256, 1982.

Coals, A. and I.C. Goulter, Approaches to the Consideration of Reliability

in Water Distribution Networks, Proceedings, 1985 International
Symposium on Urban Hydrology, Hydraulic Infrastructures and Water
Quality Control, University of Kentucky, Lexington, Kentucky, pp. 287-295,
July 23-25,1985.

Cullinane, M.J., "Hydraulic reliability of urban water distribution

systems," Proceedings of ASCE Conference, Water Forum '86: World
Water Issues in Evolution; Volume 2, Edited by M. Karamouz, G. Baumli,
and W. Brick, AXE, New York, New York, pp. 1264-1271,1986.

Damelin, E., U. Shamir and N. Arad, "Engineering & Economic Evalua-

tion of the Reliability of Water Supply," Water Resources Research, AGU,
V01.8, NO.4, pp.861-877, August 1972.

Germanopoulis, G., P.W. Jowitt and J.P. Lumber, "Assessing the reliability
of supply and level of service for water distribution systems," Proceedings
of the Institution of Civil Engineers, Part I, 80, pp. 414428, 1986.

Goulter, I. D. and A. V. Coals, "Approaches to the Consideration of

Reliability in Water Distribution Networks," Proceedings of the Twelfth
International Symposium on Urban Hydrology, Hydraulics Infrastructure
and Water Quality Control, Lexington, Kentucky, pp. 287-295, 1985.

Goulter, I. D. and D.R.Morgan, "An Integrated Approach to the Layout

and Design of Water Distribution Systems," Civil Engineering Systems,
V01.2, No.2, pp. 104-113,1985.

Goulter, I. D, and A.V. Coals, "Quantitative Approaches to Reliability

Assessment in Pipe Networks," Journal of Transportation Engineering,
AXE, Vol. 112, No.3, pp. 287-301, May 1986.

Goulter, I.C. and A.V. Coals, "Quantitative Approaches to Reliability

Assessment in Pipe Networks, " Journal of Transportation Engineering,
AXE, Vol. 112, No.3, pp. 104413,1986.

Goulter, I.C., B.M. Lussier and D.R. Morgan, "Technical note: Implica-
tions of head loss path choice in the optimization of water distribution
networks," Water Resources Research, Vol. 22, No. 5, pp. 819-822, 1986.

529
Hobbs, B.F., "Reliability analysis of urban water supply," Proceedings of
ASCE Specialty Conference, Computer Applications in Water Resources,
Edited by H. Torno, ASCE, New York, New York, pp. 1129-1238,1985.

Hobbs, B.F. and G.K. Beim, "Verification of a supply reliability model,"

Proceedings of ASCE Conference, Water Forum '86: World Water Issues
in Evolution; Volume 2, Edited by M. Karamouz, G. Baumli, and W.
Brick, ASCE, New York, New York, pp. 1248-1255,1986.

Hudson, W.D., "Studies of Distribution System Capacity in Seven Cities,"

Journal of American Water Works Association, pp. 157-164, February 1966.

Kettler, A. J. and I.C. Goulter, "Reliability Consideration in the Least Cost

Design of Looped Water Distribution Networks," Proceedings of the Tenth
International Symposium on Urban Hydrology, Hydraulics and Sediment
Control, Lexington, Kentucky, pp. 305-312,1983.

Kettler, A.J. and I.C. Godter, "An analysis of pipe breakage in urban water
distribution networks, " Canadian Journal of Civil Engineering, Vol. 12,
NO. 2, pp. 285-293,1985.

Lansey, K.E.and L.W.Mays, "A Methodology for Optimal Network

Design," Computer Applications in Water Resources, H. C. Torno, Ed.,
AXE, New York, N.Y., pp. 7320738,1985.

Lansey, K.E. and L.W. Mays, "Optimal Design of Large Scale Water Distri-
bution Systems," Proceedings, 1987 National Conference on Hydraulic
Engineering, ASCE, Williamsburg, Virginia, pp. 475-480, August 1987.

Lasdon, L. S. and A.D. Waren, "Generalized Reduced Software for Linearly

and Nonlinearly Constrained Problems," Design and Implementation of
Optimization Software, H. Greenberg, ed., Sigthoff and Noordoff, pubs.,
1979.

Lasdon, L. S., A.D. Waren and M.S. Ratner, "GRG2 Users Guide," The
University of Texas at Austin, November 1984.

Mays, L.M., N. Duan and Y-C. Su, "Modeling reliability in water

distribution network design," Proceedings of ASCE Conference, Water
Forum '86: World Water Issues in Evolution; Volume 2, Edited by M.
Karamouz, G. Baumli, and W. Brick, ASCE, New York, New York, pp.
1272-1279,1986.

Mays, L. W. and M. J. Cullinane, "A Review and Evaluation of Reliability

Concepts for Design of Water Distribution Systems," Miscellaneous Paper
5 3c
 EL-86-1, U.S. Army Engineer Waterways Experiment Station, Vicksburg,
MS, January 1986.

Morgan, D.R. and I.C. Goulter, "Optimal urban water distribution design,"
Water Resources Research, Vol. 21, No. 5, pp. 642-652, 1985.

O'Day, D.K., "Organizing and Analyzing Break Data for Making Water
Main Replacement Decisions," Journal of the American Water Works
Association, Vol. 74, No. 11, pp. 589-596, 1982.
Quindry, G.E.,E.D. Brill and J.C. Liebman. "Optimization of looped water
distribution systems, Journal of the Environmental Engineering
I'

Division, ASCE, Vol. 74, No. 11, pp. 589-596, 1981.

Quindry, G., E. D. Brill and J.C.

Liebman, "Optimization of Looped Water
Distribution Systems," Journal of fhe Environmental Engineering
Division, ASCE, Vol. 107, No. EE4, pp. 665-679, August 1981.

Ross, S. M., "Introduction to Probabilitv Models," Academic Press Inc.,

Orlando, Florida, 1985.

Shamir, U. and C.D. Howard, "An Analytic Approach to Scheduling Pipe

Replacement," Journal of the American Water Works Association, pp.
248-258, May 1979.

Shamir, U. and C.D. Howard, "Reliability and risk assessment for water
supply systems," Proceedings of ASCE Speaalty Conference, Computer
Applications in Water Resources, Edited by H. Torno, ASCE, New York,
New York, pp. 1218-1228,1985.

Su, Y.-C.,L.W. Mays, N. Duan and K.E. Lansey, "Reliability-Based Opti-

mization Model for Water Distribution Systems,'' Journal of Hydraulic
Engineering, ASCE, Vol. 114, No. 2, pp. 1539-1556, December 1987.

Sullivan, J.P., Jr., "Maintaining Aging Systems - Boston's Approach,"

Journal of the American Water Works Association, pp. 555-559,
November 1982.

Templeman, A.G, "Discussion of 'Optimization of looped water

distribution systems,' by G.E. Quindry, E.D. Brill and J.C. Liebman,"
Journal of the Environmental Engineering Division, ASCE, Vol. 107, No.
3, pp. 599402,1982.

531
Tung, Y. K., "Evaluation of Water Distribution Network Reliability,"
Proceedings, ASCE Hydraulics Division Specialty Conference, Orlando,
Florida, August 13-16, 1985.

Tung, Y-K.,"Model for optimal risk-based water distribution network

design," Proceedings of ASCE Conference, Water Forum '86: World
Water Issues in Evolution; Volume 2, Edited by M. Karamouz, G. Baumli,
and W. Brick, AXE, New York, New York, pp. 1280-1284,1986.

U.S. Army Corps of Engineers, "Methodology for Areawide Planning

Studies," Engineer Manual 111-2-502, Washington, D.C.,1980.

Wagner, J.M., U. Shamir and D.H. Marks, Reliability of Water Distribu-

tion Systems, Report No. 312, Ralph M. Parsons Laboratory, M.I.T.,
Cambridge, Massachusetts, 150 pp, June 1986.

Wood, D. J., "Computer Analysis of Flow in Pipe Networks Including

Extended Period Simulations," Reference Material for a Three-Day Short
Course and Correspondence Course, Office of Continuing Education and
Extension of the College of Engineering of the University of Kentucky,
Lexington, Kentucky, September 1980.

532