CHAPTER FIVE

Network Hydraulics

The fundamental relationships of conservation of mass and

energy mathematically describe the flow and pressure distribution within
a pipe network under steady state conditions. To begin, parallel and
series pipe systems are considered in Section 5.1. The basic concepts
applied in these simple systems to determine the pipe flow rates and nodal
pressure heads are then extended to full networks. Four mathematical
formulations are discussed (Sections 5.2 and 5.3). Quasi-dynamic flow for
tank modeling under time varying conditions is then presented. Finally, flow
conditions that change in the short term are analyzed using conservation of
mass with conservation of momentum to include the impact of dynamic
changes.
Notation for this section can become confusing since a number
of subscripts and counters are used. To summarize, the subscripts i and l
identify nodes and pipes, respectively. The pipe flows are typically
summed over the sets of pipes providing flow to or carrying flow from
a node. To avoid confusion with pipe lengths, L, the set of incoming and
outgoing pipes are defined as Jin and Jout, respectively. Depending upon the
conservation of energy formulation, head losses in pipes are summed over the
set of pipes in a loop, lloop, or path, lpath. The notation with the letter n
followed by an identifier is used for the number of components such as nnode,
npipe, nloop, and nploop for the number of network nodes, pipes, closed
loops, and pseudo-loops, respectively. The subscript j is used as an
identifier in summations and defined for the specific equation including an
identifier for a node at one end of a pipe. To compute the nodal heads and
pipe flows a set of nonlinear equations is solved by an iterative process.
The counter m is used to define the iteration number and added to the
unknown variable as a superscript in parentheses, i.e., Q(m). Lastly, variables are
shown in italics and matrices and vectors are denoted by bold characters.

5.1 SIMPLE PIPING SYSTEMS

Simple piping systems provide insight into the understanding of pipe

networks. Variation of total head through a network is seen in a set of
pipes in series. Analysis of pipes in parallel is the first application of the
conservation of mass at a junction and conservation of energy around a
loop. In addition to the physical understanding that they offer,
simplification of these systems to equivalent pipes reduces computations
during analysis.

5-1
5-2 CHAPTER FIVE

5.1.1 Pipes in Series

As shown in Figure 5-1, a series of pipes may have different pipe

diameters and/or roughness parameters. The total head loss is equal to the sum
of the head loss in each pipe segment or:
lpath lpath

h K
n
hL L ,l l Ql (5-1)
l 1 l
1
where lpath is the number of pipes in series, Kl is the coefficient for
pipe l containing information about the diameter, length, and pipe roughness,
n is the exponent from the head loss equation, and Ql is the flow rate in pipe l.
If no withdrawals occur along the pipe, each pipe carries the same flow rate
but the rate of head loss in each pipe may be different. If we use the same head
loss equation for all pipes (i.e., the same n), we can take Q out of the
summation or:
lpath lpath lpath

n n n s n
hL Kl l K eq Q (5-2)
l 1
1
Q
l Kl Q Q
Kl
where eq
s
is the equivalent K coefficient forl 1series of pipes. If flow is
K
turbulent, the Kls are constant and a single equivalent K eqs can be computed for
all turbulent flows.

A B C D

D1 = 30 cm D2 = 20 cm D3 = 40 cm
L1 = 2000 m L2 = 1000 m L3 = 2000 m
f1 = 0.022 f2 = 0.025 f3 = 0.021
zA = 20 m, zB = 25 m, zC = 32.5 m, zD = 37.5 m
Figure 5-1: Pipes in series with data for Example 5.1.

Example 5.1

Problem: For a flow rate of 0.04 m3/s, determine the pressure and total heads at
points A, B, C, and D for series pipes shown in Figure 5-1. Assume
fully turbulent flow for all cases and the pressure head at point A is 40 m.
 NETWORK HYDRAULICS

5-3

Solution: The pressure and total heads are computed using the energy equation
along the path beginning at point A. Given the pressure head and elevation, the
total head at point A is: p
H A A z A 40 20 60
m

Note the velocity and

velocity head are:
V1
Q
0.04
0.57 m /
A (0.3) s
2
4
and
1 0.57
V2
2 0.017
2g 2 m
(9.81)
The velocity head is four orders of magnitude less than the static head so it
can be neglected. Neglecting velocity head is a common assumption in
pipe network analysis.
All energy loss in the system is due to friction. So following the
path of
flow the total heads at A, B, C, and D are:
2
L V 0.57
H H h A B 60 f 1 1 2000 2
B A f D1 2 g 60 0.022
1 0.3 2
60 2.4 H B 57.6 (9.81)
m 2
0.04

L (0.2) 2 4
H C H B h fB C 57.6 2 2 V 22 1000

57.6 0.025 2
f D2 2g
0.2
57.6 10.3 H C 47.3 (9.81)

m
2

0.04
L (0.4) 2
H H hC D 47.3 f 3 V 22 2000

D C f D3 2g 47.3 0.021 4
0.4 2
3
47.3 0.5 H D 46.8 (9.81)
m

The pressure heads are:

at point B pB pB pB
HB z B 57.6 25 32.6

m
5-4 CHAPTER FIVE

at point C
pC pC pC
HC z C 47.3 32.5 14.8

and at point D m
p p p
H D D z D 46.8 D 37.5 D 9.3
m

Example 5.2

Problem: For the series pipe system in Example 5.1, find the equivalent
roughness coefficient and the total head at point D for a flow rate of 0.03 m3/s.

Solution: By Eq. 5-2, the equivalent pipe loss coefficient is equal to the sum of
the pipe coefficients or:
3 3
s 8 fl l
eq l
gL D
2 5
K
l K l l
1 1
For this problem:

s
8 f 1 L1 8 f 2 L2 8 f3
K eq K1 K 2 K3 2 5 L 2 5 2 5
g D1 g 3 D 2 g D3
8 (0.022) (2000) 8 (0.025) (1000) 8 (0.021)
(2000) 2
9.81 (0.3) 5 9.81 2 (0.2) 5 9.81 2
(0.4) 5 1496 6455 339 K s
eq
8290
Note that pipe 2 has the largest loss coefficient since it has the
smallest diameter and highest flow velocity. As seen in Example 5.1, although
it has the shortest length, most of the head loss occurs in this section.
The head loss between nodes A and D for Q = 0.03 m3/s is then:

h A D K s Q 2 K Q 2 K Q 2 K Q 2 8290 (0.03) 2 h A D
7.5 m
f eq 1 2 3 f

We can also confirm the result in Example 5.1 by substituting Q =

0.04 m3/s in which case:

h A D K s Q 2 8290 (0.04) 2 13.26 m

f eq

and
 NETWORK HYDRAULICS 5-5

H D H A h AD
f 60 13.26 H D 46.74
m
that would be equivalent to the earlier result if Example 5.1 was carried to 2
decimal places.

5.1.2 Pipes in Parallel

When one or more pipes connect the same locations (junctions), the hydraulics
are much more interesting. The relationships in these small networks lead to the
fundamental relationships for full network modeling. Locations A and B
in Figure 5-2 are described as nodes or junctions of several pipes. As in
Example 2.1, conservation of mass must be preserved at these locations.
That is, in steady state the known incoming flow at node A must
balance with the outgoing flows in pipes 1, 2, and 3. Similarly, the incoming
flows to node B in the incoming pipes 1, 2, and 3 must equal the known
withdrawal at node B.

qA = qB = Q1 + Q2 + Q3

(5-3)

where Ql and qj define the flow rate in pipe l and the nodal withdrawal/supply
at node j, respectively. The mass balance for node B provides the same
information as the above and is redundant.
(L2 = 200 ft, D2 = 16 in, CHW, 2 = 130)
A (L1 = 300 ft, D1 = 14 in, CHW, 1 = 120) B
qA = 10 cfs qB = 10 cfs
HA = 80 ft HB = ?
(L3 = 400 ft, D3 = 18 in, CHW, 3 = 110)

Figure 5-2: Pipes in parallel with data for Example 5.3.

The second relationship that must hold is that the head loss in pipes 1, 2,
and 3 must be the same. Since all begin at a single node (A) and all end at a
single node (B) and the difference in head between those two nodes is unique,
regardless of the pipe characteristics the head loss in the pipes is the same or:

H A H B hL,1 hL,2 hL,3 hL (5-4)

where HA and HB are the total heads at nodes A and B, respectively, hL,l is the
head loss in pipe l, and hL is the single value of head loss between nodes A and
5-6 CHAPTER FIVE

B. Eq. 5-4 is a statement of conservation of energy for a pipe and is used

in several formulations for solving for flows and heads in a general network.
In other network solution methods, we write conservation of energy
for
closed loops. A closed loop is a path of pipes that begins and ends at the
same node. Pipes 1 and 2 form a closed loop beginning and ending at
node A. Starting at node A around the path, energy, hL,1, is lost as water flows
from A to
B. As we follow the path back to node A to close the loop, we gain energy,
hL,2,
(5-5)
A
since we Hare hL,1 hin
moving Hdirection
L,2the A hL,1 hL,2 to0 the
opposite hflow.
L,1 hWe
L,2 can write the
path equation around the loop and manipulate it to show:
Now using Eq. 5-3 and either Eqs. 5-4 or 5-5, we can determine the head
loss and flow for each pipe and an equivalent pipe coefficient, K eq p
. In any pipe
network system the nodal inflows and outflows (qA and qB) and at least
one nodal heads total energy (HA in this case) must be known to provide a
datum for the pressure head. For steady flow conditions in the network in
Figure 5-2, we have a total of seven unknowns, node Bs total energy (HB),
three pipe flows (Q1, Q2, and Q3) and three head losses (hL,1, hL,2, and hL,3).
Eq. 5-4 provides two independent equations relating the head losses (hL,1 =
hL,2, and hL,1, = hL,3). The third equation is that the head loss in any pipe equals
the difference in head between nodes A and B (the first part of Eq. 5-4).
Conservation of mass at node A (Eq. 5-3) is the fourth relationship. The final
three equations are the head loss versus discharge equations:
1

h
h K Q n or Q L,l (5-6)
L,l l l l K
l
n
We can substitute Eq. 5-6 in the mass balance equations (Eq. 5-3) with hL
equal to each pipes head loss or:
1 1 1

hL h hL
L qA (5-7)
K
1

K K
2 3
n n n
In this equation, all terms except for hL are known. After solving for hL, the
unknown pipe flows can be computed by Eq. 5-6 and HB can be determined in
Eq. 5-4.
Like pipes in series, an equivalent pipe coefficient can be computed
for parallel pipes. In Eq. 5-7, hL can be pulled from each term on the left hand
side or for a general discharge and three parallel pipes:
 NETWORK HYDRAULICS 5-7

1 1 1
n n 1
1 n
(5-8)
2 h n
3 L
1


1 q

K 1 K K
The equivalent coefficient is then:

1 1 1 1 1
1
1 n
1 lp
1 (5-9)
Kn p
1 K12
K
n eq
K


l


1
K l
n

3
n
where K eqp is the equivalent pipe coefficient for parallel pipes. As shown in the
last term, Eq. 5-9 can be generalized for lp parallel pipes.
The head loss between the two end nodes is:
h K p (Q (5-10)
)eqn L
total

Example 5.3

Problem: Given the data for the three parallel pipes in Figure 5-2, compute (1)
the equivalent parallel pipe coefficient, (2) the head loss between nodes A and
B, (3) the flow rates in each pipe, and (4) the total head at node B.

Solution: (1) The equivalent parallel pipe coefficient allows us to determine the
head loss that can then be used to disaggregate the flow between pipes. The loss
coefficient for the Hazen-Williams equation for pipe 1 with English units is:
4.73 L1 4.73 (300)
K1 0.0954
1.85 4.87 1.85 4.87
1 1
C D 120 (14 /12)
Similarly, K2 and K3 equal 0.0286 and 0.0439, respectively.
The equivalent loss coefficient is:
1 1 1 0.54 0.54
1 1 n 1 0.54


K n 1
1 Kn 2 K 3 1

1
0.54 0.0286

3.557 6.817 5.408 1 0.0439
0.0954 K eqp 0.00607

K p
eq

(2 and 4) The head loss between nodes A and B is then:
5-8 CHAPTER FIVE

hL K eq
p
(Q total ) n 0.00607 (10)1.85 0.43
ft
(4) So the head at node B, HB, is:

HA HB = hL = 80 HB = 0.43 ft HB = 79.57 ft

(3) The flow in each pipe can be computed from the individual pipe head loss
equations since the head loss is known for each pipe (hL = 0.43 ft).
1 0.54
1
1 n 1 0.54
Q1 hL (0.43) 2.26 cfs
K1n

The flows in pipes 2 and 3 can be0.0954 by the same equation and are 4.32
computed

and 3.43 cfs, respectively. The sum of the three pipe flows equals 10 cfs, which
is same as inflow to node A.

5.2 SYSTEM OF EQUATIONS FOR STEADY FLOW

Conservation of mass at a junction node (Eq. 5-3) and conservation of energy

(Eq. 5-1) can be extended from parallel pipes to general networks for
steady state hydraulic conditions. The resulting set of simultaneous
quasi-linear equations can be solved for the pipe flows and nodal heads for
steady state and step-wise (quasi) dynamic (known as extended period
simulation or EPS) analyses. EPS analysis requires an additional relationship
describing changes in tank levels due to inflow/outflows and is discussed
in later sections. Only steady state hydraulics is considered in Sections 5.2 and
5.3.
5.2.1 Conservation of Mass

As defined earlier, a junction node is a connection of two or more

pipes. Although demands are distributed along pipes, these demands are
lumped at junctions and defined as qnode. Conservation of mass at a node was
presented in Section 2.1.2.1. For a junction node i, conservation of mass can be
written as:
(5-11)
l l i
Q Q q
l J in l J out
where qi is the external demand (withdrawal), Jin,i and Jout,i are the set of pipes
supplying and carrying flow from node i, respectively, and l J in denotes that l
is in the set of pipes in Jin. This equation can be written for every junction node
in the system.
 NETWORK HYDRAULICS 5-9

5.2.2 Conservation of Energy

The second governing law is that energy must be conserved between any two
points. Along the path between nodes A and B that only includes pipes,
conservation of energy is written as:

n
HA HB (5-12)
h L,l K Q
l l

where HA and HB are the total energy at nodes A and B, hL,l, Kl, Ql are the head
loss, loss equation coefficient and iflow
lpath
rate ini pipe
lpath
l and n is the exponent from
the head loss equation.

The sign for the flow rate is defined using the Ql symbol and this
symbol
does not have its conventional meaning. This symbol is intended as a
short form notation and reminder of how the signs of this relationship
should be interpreted. The absolute value of Q is raised to the power of n and
the sign of the pipe term is based on the flow direction. If flow is
moving from node A toward node B then the sign should be taken
positive and a negative sign is used if flow is away from B toward A. Eq.
5-12 can be written for a closed or pseudo-loop or a single pipe. lpath defines
the set of pipes in the path.
A closed loop is one that begins and ends at the same node. Since
each location in the networkH Ahas
closed loop is zero. For a loop
Ha Aunique
0 energy h L,l the
0 net energy loss around
(5-13) a
beginning and ending at node A:
l lloop

where lloop is the set of pipes in the closed loop.

A pseudo-loop is a path of pipes between two points of known energy such
as two tanks or reservoirs. Eq. 5-12 applies directly to pseudo-loops. Pseudo-
loop equations include additional information regarding the flow
distribution and are needed for some solution methods. Finally, Eq. 5-12
also applies directly for individual pipes with HA and HB being the total
heads at the two ends of the single pipe or:

H A H B Kl
Ql (5-14)

5.2.3 Systems of Equations n

The unknowns in a steady state hydraulic analysis are the flows in each pipe, Q,
and the total energy head at each junction node, H. In a system with
nnode
5-10 CHAPTER FIVE

nodes and npipe pipes, the total number of unknowns is nnode + npipe. Four
equation formulations can be developed to solve for these unknowns. They can
be expressed in terms of unknown pipe flows or nodal heads. All sets
are nonlinear due to the energy loss relationships and require iterative
solutions. The Newton-Raphson method is the most widely used
iterative solution procedure in network analysis. Its convergence properties
have been studied in detail by Altman and Boulos (1995). At least one
point of known energy is required to provide a datum (or root) for the
nodal heads. The four solution approaches are summarized below and
mathematical details are presented in Section 5.3.

1. Loop Equations

The smallest set of equations is the loop equations that include one equation for
each closed loop and pseudo-loop for a total of nloop + nploop equations where
nloop and nploop are the number of closed and pseudo-loops, respectively. The
unknowns in the loop equations are Qs that are defined as the corrections to
the flow rate around each loop. Beginning with a flow distribution that satisfies
conservation of mass, the corrections maintain those relationships. When zero
corrections are needed in all loops, the flow rates in each loop and each pipe
has been found. After the flows have been determined, Eq. 5-12 is
applied beginning at a location of known total energy (e.g., root) to determine
the nodal heads.
The Hardy Cross method is one approach to solve the loop equations. This
method first determines corrections for each loop independently then
applies the corrections to compute the new pipe flows. With the new flow
distribution,
another set of corrections is computed. Hardy Cross introduced this
method in 1936 and, although amendable to hand calculations, it is not efficient
compared to methods that consider the entire system simultaneously. Epp
and Fowler
(1970) presented a more efficient method that simultaneously solves
for all loop corrections using the Newton-Raphson method with the
corrections as the unknowns.

2. Node-Loop Equations

Wood and Rayes (1981) compared a number of solution algorithms with their
modified linear theory (flow adjustment) method and showed that this approach
was efficient and robust. Modified linear theory solves directly for the
pipe flow rates rather than the loop equation approach of loop flow corrections.
The nloop + nploop loop equations (Eq. 5-12) incorporate the concept of
energy conservation and nnode node equations (Eq. 5-11) introduce
conservation of mass. The total number of independent equations is
nnode + nloop + nploop that number is equal to the number of unknown pipe
flow rates,
 NETWORK HYDRAULICS 5-11

npipe. A Newtons type method is used to solve for the Qs directly rather than
the loop flow rate corrections, Qs. As in the loop equations, after the
pipe flow rates are found, they are substituted in Eq. 5-12 beginning at a
point of known energy to compute the nodal heads.

5.2.3.3 Node Equations

The node equations can be rewritten in terms of the nodal heads by writing Eq.
5-14 for pipe l that connects nodes j and i as:
1

Ql H j H i (5-15)
Kl

These terms are substituted for the flow
n
rate in Eq. 5-11 for each pipe and
one equation of the form of Eq. 5-15 is written for each node. This substitution
combines the conservation of energy and mass relationships resulting in nnode
equations in terms of the nnode unknown nodal heads, H. Shamir and Howard
(1968) solved these equations using the Newton-Raphson method. After
the nodal heads are computed, they can be substituted in Eq. 5-15 to compute
the pipe flow rates.

5.2.3.4 Pipe Equations

The previous methods solve for the pipe flows, Q, or nodal heads,
H, in a nonlinear solution scheme then use conservation of energy to
determine the other set of unknowns. Haman and Brameller (1972) and
Todini and Pilati (1987) devised a method to solve for Q and H
simultaneously. They wrote the node equations (Eq. 5-11) with respect to the
pipe flows and Eq. 5-14 for each pipe including both the pipe flows and the
nodal heads. Although the number of equations (nnode + npipe) is larger than
the other methods, the solution times and the convergence to the true solution
are similar or better. In addition, the algorithm does not require defining loops
that may be a time consuming task.

5.3 SOLUTION ALGORITHMS FOR STEADY FLOW

5.3.1 Solution of the Loop Equations

5.3.1.1 Hardy Cross Method (Single Loop Adjustment Algorithm)

The oldest and probably best known solution method for pipe networks is the
Hardy Cross method that is found in most textbooks and taught in
undergraduate hydraulics courses (Cross 1936). As noted in Section 5.2.3.1, the
5-12 CHAPTER FIVE

method solves the energy equations for loops and pseudo-loops for a loop flow
correction. Although a set of loop equations must be solved for the system, this
algorithm was developed for hand calculations and solves one loop at a time.
One closed loop equation is written (Eq. 5-12) for each loop. For closed loops
that contain only pipes, the loop equation for loop LP by Eq. 5-13 is:

FLP (Q) K l Ql

(5-16)
l lloop
n

In this equation, the sign of Ql is applied to the term and the absolute value
is raised to the exponent n. The sign is based on the flow direction relative to
loop LP and discussed in more detail below.
Since the flow rates that satisfy the set of loop equations are not known, a
loop equation is expanded to a Taylor series truncated at the first order term or:

dFLP
FLP ( ) FLP ( )
Q (m) Q (m Q (m) Q (m dQl
) l (1)
1) lloop (5-17)
n
K (Q dFLP
(m
l
Q1) (m)
Q ( m 1)
)
l lloop l dQl
lloop

where Q(m-1) is the estimate of the flow at iteration m-1 and FLP / Ql is the
derivative of the LPth loop equation with respect to the lth pipe flow, Ql.
Defining Q = Q(m) Q(m-1) and substituting in Eq. 5-17:

(m1)
FLP (Q ( m 1) ) Q F (Q )
0 Q
FLP Ql (m

Q 1)

FLP Ql

(m
Q 1)
l lloop
l lloop (5-18)

The development of Eq. 5-18 is equivalent to the Newton-Raphson method

except that Q is computed rather than the updated flow Q(m). In addition, the
Hardy Cross method simplifies the determination of the correction term
by considering each loop independently rather than all loops simultaneously.
Since all pipes in a loop will have the same correction, a single Q is
determined by Eq. 5-18. The numerator of Eq. 5-18 is computed from
Eq. 5-16 with appropriate signs for flow directions. Standard
convention is to define clockwise flow in each loop as positive. If F(Q)
equals zero, the equation has been satisfied.
The denominator is the sum of the absolute values of the derivative terms
of Eq. 5-16 evaluated at Q(m-1). The individual gradient terms are:
 NETWORK HYDRAULICS 5-13

n
F ( K (Q Q) Q ( m1)
n K Q n 1 n hL / (5-19)
) Q Q
Q
For loop LP, Eq. 5-18 is then:

n l l l
n l nl l

QLP K Q K Q K Q
l lloop l lloop l lloop

n K Q n K Q
n1 n
l l l l
/ l L,l l
l lloop
Q
l lloop
n
l lloop h /
Q (5-20)

where the denominator becomes an absolute value because the signs of hL

and
Ql are the same.
A similar equation can be written for each loop in the network.
Since a first-order Taylor series is used to approximate a nonlinear equation,
a single set of corrections is likely insufficient to converge to the true
flows and the process must be repeated until all loops equations are satisfied
within a desired tolerance.
In summary, the Hardy Cross algorithm consists of the following steps.

1) Define loops and set m = 0. Assume an initial set of pipe flows that
satisfy conservation of mass at all nodes. Note that the loop
corrections will maintain conservation of mass after this initial step.
2) Update m = m + 1
3) Compute the sum of head losses around a loop by solving Eq. 5-16
for each loop substituting Q(m-1) for Q. This sum is the numerator of
Eq. 5-20.
4) Compute the denominator of Eq. 5-20 for each loop. Note that the
denominator is the sum of the absolute values of n hL /Q over the set
of pipes in the loop, lloop.
5) Compute the loop correction, QLP, by solving Eq. 5-20.
6) Repeat steps 3-5 for each loop.
7) Apply correction factors
Ql(m) toQall
l
l pipes or: Q
(m1) lp lp
ncp(l )

where ncp(l) is the set of one or two loops in the network that contain
pipe l.
8) Check if all Qs are less than specified small tolerance. If so, stop.
If not, go to step 2.
5-14 CHAPTER FIVE

The equations above account for the flow direction change without
modification. In the numerator, the appropriate sign is taken from the flow rate
as defined by the initially assumed flow rate and relative to the loop
being considered. Absolute values are always taken in the denominator. A
negative flow relative to the initially defined loops denotes that the flow
is in the opposite direction of the original assumption with the computed
magnitude.
In step 7, the corrections are applied with similar logic. A positive
correction implies a larger flow in the clockwise direction. Thus, Q is
added to pipes with assumed positive flow directions. If the assumed
direction is counter-clockwise, the correction is subtracted from the Q(m-1). This
convention is also applied if the actual flow directions is opposite of the
assumed direction. In the approach described above, the loop corrections are
applied after all corrections have been computed. It is possible to be more
sophisticated by, for example, applying the corrections as the method proceeds
through an iteration. However, although the Hardy Cross method is acceptable
for hand calculation, it is not efficient for or applied to large systems so these
improvements are not
considered here.

Example 5.4

Problem: Determine the flow rates in the pipes in the three loop
network in Figure E5-4a and the nodal heads at all nodes using the Hardy
Cross method and the Hazen-Williams equation. With the flow rates, compute
the energy at node 5.

Solution: Step 1: m = 0. Following the procedure outlined above, the first

step is to define a set of loops. Four loops are identified in Figure E5-4b. The
head loss around the closed loops I, II and III is zero since the loops begin and
end at the same node and the node has a unique total head (as in the
parallel pipe analysis in Section 5.1.2). A pseudo-loop is defined between the
two reservoirs and the difference in energy between the locations is 200
ft. The positive direction is defined in the clockwise direction for all loops.
Assuming the flow directions shown on Figure E5-4b, the loop equations
are shown below. For Loop P (the pseudo-loop) starting at reservoir 1
and Loop
continuing
P : along
K1 Q1a1.85
path to
K2reservoir
Q 2 1.85 2:
K 9 Q9 1.85 (240 0.9376 QP2 )
H
res. 2 Hres. 1 0 200 200

We have added the sign convention relative to the loop and the
initially assumed flow directions. Pipe 1 is assumed to flow in the
counter-clockwise
 NETWORK HYDRAULICS 5-15

direction relative to loop P and its term is given a negative sign. A positive flow
in pipe 1 denotes that the flow direction is from the tank to node 1. If flow is
assumed incorrectly and flow is actually from node 1 toward the tank in pipe 1,
the sign is switched to a negative. In this case, pipe 1s head loss term in Loop
P will be positive (the initially assumed negative sign times the negative sign
from the flow term).

P
1 2
1 2

9
3 4 P
I
7
6 7 III 3

5 6
II 10

8
5 4

Figure E5-4b: Example hydraulic analysis pipe network with defined loops
and assumed flow directions.

Pipe 9 is assumed to flow from node 3 to node 2 that is positive relative to

loop P and its head loss term is positive. The direction of flow through
the pump is also clockwise but the pump provides a head gain so it is
given a negative sign. No sign convention is applied to the pump since it can
only be non-negative, that is, a positive flow or zero.
For loop III beginning and ending at node 3:
1.85 1.85 1.85
1.85 H3 H3
0
K 10 Qpositive
Again, 10 K 6 Qare
signs 6 given Qflow
K 4for 4 moving
K 9 clockwise
Q9 relative to the
loop. Thus, pipe 9 is positive relative to loop P but negative in loop III.
For loop II (beginning and ending at node 6):
1.85 1.85 1.85
1.85 H6 H6
0
K7 Q 7 K6 Q 6 K8 Q8 K5 Q5
5-16 CHAPTER FIVE

Hres.1 = 200 ft
(1, 1000, 30, 140)

(2, 2000, 24, 100)

1 2 [4]

Hres. 2 = 0 ft
(9, 3000, 6, 100)
(3, 2000, 24, 100) (4, 1500, 16, 100)

(Pump, H=240-0.9376 Q 2)
(7, 2000, 6, 100)
[5] 6 7 [3] 3

[6]

(5, 1000, 16, 100) (6, 1500, 12, 100)

(10, 3000, 6, 100)

(8, 2000, 8, 100)

[5] 5 4 [5] Legend
Pipe number, Length (ft), Diameter (inch), C-value
[Nodal demand (cfs)]

Figure E5-4a: Example hydraulic analysis pipe network.

 NETWORK HYDRAULICS 5-17

Finally, loop I begins and ends at node 1:

1.85 1.85 1.85 1.85
H1 H1 0
K2 Q 2 K4 Q 4 K7 Q 7 K3 Q3
Any pipe that appears in two loops has a negative sign in one loop equation
and
a positive in the other equation. This convention must also be considered
when updating flows.
The K for each pipe is given in Table E5-4a using:
L
K 4.73 4.87 1.85
D C HW

with Q in cfs, D and L in ft. Table E5-4a also gives a set of pipe flow rates that
satisfy conservation of mass. Initial pipe flows were determined for the
sequence of nodes 1, 3, 2, 6, 7, and 5. The last node is then checked to confirm
system mass balance. Since the total network demand is 28 cfs, the
flows in pipe 1 and the pump must equal 28 cfs. To begin, the flow in pipe 1 is
assumed to be 20 cfs. At a given node, all but one outflow pipe flows are
assumed and the last value is computed by the nodes mass balance equation.
Step 2: Set m = m + 1 = 1

Table E5-4a: Initial data and K coefficients for Example 5.4.

Pipe 1 2 3 4 5
Unode Tank 1 1 2 6
Dnode 1 2 6 7 5
K 0.00584 0.0645 0.0645 0.349 0.233
Q(0) 20 9 11 6 5.5
hL 1.49 3.76 5.45 9.60 5.45
nhL/Q 0.138 0.774 0.918 2.96 1.83

Pipe 6 7 8 9 10 P
Unode 7 6 5 3 3 Tank
Dnode 4 7 4 2 4 3
K 1.42 55.2 13.6 82.8 82.8

Q(0) 3.5 0.5 0.5 1 1 8

hL 14.37 15.31 3.77 82.79 82.79 180
* Unode and Dnode
nhLare
/Q the initially
7.60 assumed
56.7 upstream and downstream
14.0 153.33 153.33 nodes,
15.0respectively.

Step 3: Compute sum of the head losses around each loop using the values
listed in Table E5-4a for the assumed flow conditions and the loop equations
defined in Step 1. The sums are listed in Table E5-4b. Using the initial guess,
the residual of loop Ps energy equation, Fp, is:
5-18 CHAPTER FIVE

1.85 1.85 1.85

FP (Q ) K1 1 K 2
Q2 K9 Q9
(0) 2
(240 0.9376 QP )
Q 1.85 200
0.00584 (20) 0.0645 (9) 82.8 (1) (240 0.9376 (8) 2 ) 200
1.85 1.85

1.49 3.76 82.8 180. 200 97.55

Step 4: The computed sums of absolute values of the derivatives are also
listed in Table E5-4b. For loop P:

n h
l lloop
L,l
/ Ql n L,1

h

h
Q
Q1 2
L,2

h 2 (0.9376 Q )
L,9
Q9
P

0.138 0.774 153.33 2 (0.9376( 8)) 169.24

Note that last term is the derivative of the pump equation with respect to Q.
Step 5 and 6: Compute the correction for each loop using Eq. 5-20.
For loop P:
1.85
97.55
KQ 169.24
Q P nh /QL
0.576
Table E5-4b gives the values for all loops.

Table E5-4b: Loop corrections for iteration 1 of Hardy Cross

method example.
Loop P I II III

hL 97.55 -7.40 20.46 -23.97

n | hL/Q| 169.25 61.37 80.12 317.23
Ql -0.576 0.121 -0.255 0.076

Step 7: The loop corrections are applied to each pipe as follows. For
pipe 1, the pseudo-loops correction is applied with a negative sign
since the correction is in the clockwise direction and pipe 1 is assumed
to flow in the counter-clockwise direction or:

20 (0.576)
Q (1) Q (0) Q
1 1 20.576
Corrections for loops P Pand I are applied to pipe 2 since it is located in
both loops. Loop Ps correction is negative since pipe 2 flow is assumed to
flow in the counter-clockwise direction for that loop (negative) and
the loop I correction is positive since pipe 2s flow is clockwise relative to
that loop.
 NETWORK HYDRAULICS 5-19

QI 9 (0.576) 0.121
Q (1) Q (0) Q
2 9.697
2

P
(1)
11 0.121 10.879
Q(0) Q (0) Q
Similarly for the Qpipes:
Q4other (1)
43
Q I3 Q III 6 0.121 0.076
I
6.045
5(1) 5
(0) II 5.5 (0.255) 5.755
Q Q Q
6(1) 6
(0) II Q
III 3.5 (0.255) 0.076
Q Q Q
Q7 Q7
(1) (0) 3.169 (0.255)
Q I Q II 0.5 0.121
0.124
8 8 II 0.5 (0.255) 0.755
Q (1) Q (0) Q
9 9 P Q
III 1 (0.576) 0.076
Q (1) Q (0) Q
0.348
1 0.076
10 10 III
Q (1) Q (0) Q
P 81.076
P (0.576) P
Q (1) Q (0) Q
7.424
Nodal flow balances continue to be conserved and can be verified. It
is worthwhile to compute those balances to check if a computational or sign
error has been introduced. In this case, pipe 10 is rounded down to preserve the
mass balance for node 10.
Step 8: The maximum correction is 0.576 cfs so iterations continue. Go to
step 2.
To provide some insight into the sign convention, consider pipe 1 during
the first iteration. If we had assumed that pipe 1s flow was from node 1 to
the tank and the flow was exiting the tank, the sign on Q1 would be negative
(i.e.,
Q1 = -20). In calculating the correction, the denominator of QP would be
the same since the absolute value of the individual terms are summed.
The resulting numerator would also be the same because a different sign
would be applied to the pipe 1 term1.85
or:
1.85 1.85
FP (Q ) K1 1 2 2 K9
(0) 2
KQ Q9 (240 0.9376 QP )
Q 0.00584 (1) (20)1.85 0.0645 (9)1.85200 82.8 (1)1.85 (240 0.9376 (8) 2 ) 200

1.49 3.76 82.8 180. 200 97.55

Pipe 1 would be positive since the assumed flow from node 1 to the tank is a
clockwise flow relative to loop P. However, the actual flow would carry
a negative sign (shown in parenthesis) since flow was opposite of the
assumed direction. Thus, the numerator and correction term would not change.
Based on the flow direction assumption, the flow is in the positive
(clockwise) direction for loop P so the correction would be added to Q1 or:
5-20 CHAPTER FIVE

20 (0.576)
Q (1) Q (0) Q
1 1
20.576
Thus, the magnitude for P the next iteration would be same as above and the
negative sign would denote that the flow was in the opposite direction of the
assumption (node 1 to tank).
A number of iterations are required for the Hardy Cross method to
converge to the solution. Flow values and the loop corrections are given
in Table E5-4c and d, respectively, for 11 additional iterations until the
largest
loop correction is less than 0.02 cfs. Values in the tables are computed
without rounding in the spreadsheet that performed the calculations.
The total head at node 5 can be computed by beginning a path at
either
reservoir 1 or 2. Starting at reservoir 1 with a head of 200 ft, a path to node
5 is pipes 1, 3, and 5. Signs on the energy loss terms are based on flow
directions in the path. Since flow goes from the tank to node 1, energy is lost
as the water travels through pipe 1.85
1. Similarly,
1.85
energy
1.85
is lost as flow moves
from node 1 to 6 inH res.1
pipe 3K 1and
Q1 from Knode
3 Q3 6 toK55 in H5. The overall energy
Q5 pipe
equation
200 is0.00584
then: (21.27) 1.85 0.06455 (11.40) 1.85 0.233 (6.06) 1.85 185.98 ft

Table E5-4c: Pipe flow values for Hardy Cross iterations.

m Pipe Flow (cfs)
1 2 3 4 5 6 7 8 9 10 P
2 20.80 9.58 11.22 5.96 5.74 3.45 0.49 0.74 0.38 0.82 7.20
3 21.01 9.89 11.11 6.16 5.91 3.37 0.20 0.91 0.27 0.72 6.99
4 21.12 9.80 11.31 6.05 5.89 3.47 0.43 0.89 0.24 0.64 6.88
5 21.19 9.93 11.26 6.14 5.99 3.41 0.27 0.99 0.21 0.60 6.81
6 21.23 9.86 11.36 6.06 5.97 3.45 0.39 0.97 0.20 0.57 6.78
7 21.25 9.91 11.34 6.10 6.03 3.41 0.31 1.03 0.19 0.56 6.75
8 21.26 9.86 11.39 6.05 6.02 3.43 0.38 1.02 0.19 0.56 6.75
9 21.26 9.88 11.38 6.07 6.05 3.40 0.33 1.05 0.19 0.55 6.74
10 21.27 9.86 11.41 6.04 6.04 3.41 0.37 1.04 0.19 0.55 6.74
11 21.27 9.87 11.40 6.05 6.06 3.40 0.35 1.06 0.19 0.55 6.73

Node 5s head can also be computed starting at reservoir 2. In this

path, energy is gained as water is lifted by the pump and lost as water moves
from node 3 to 4 in pipe 10. The energy at node 5 is greater than at
node 4 since water is flowing from node 5 to 4 in pipe 8. The head loss in
pipe 8 therefore must be added in the path equation from reservoir 2 to node 5
or:

H (240 0.9376 Q 2 ) K Q 1.85 K Q 1.85 H

res. 2 P 10 10 8 8 5

0 (240 0.9376 (6.73) ) 82.8 (0.55)

2 1.85
13.6 (1.06) 1.85
185.28 ft
 NETWORK HYDRAULICS 5-21

Slight differences result from the two paths since the Hardy Cross
iterations were stopped before full convergence.

Table E5-4d: Loop corrections for Hardy Cross iterations.

Iteration Loop P Loop I Loop II Loop III
2 -0.227 -0.342 0.020 -0.260
3 -0.203 0.108 -0.177 -0.093
4 -0.111 -0.200 0.026 -0.081
5 -0.073 0.055 -0.104 -0.040
6 -0.035 -0.106 0.019 -0.027
7 -0.022 0.026 -0.056 -0.011
8 -0.008 -0.054 0.010 -0.009
9 -0.007 0.012 -0.029 -0.003
10 -0.002 -0.027 0.006 -0.003
11 -0.003 0.006 -0.015 -0.001

5.3.1.2 Simultaneous Loop Equation Solution (Simultaneous Loop

Flow Adjustment Algorithm)

In the Hardy Cross method, each loop correction is determined independently

of the other loops. As seen in Figure E5-4b, several loops may have common
pipes so corrections to those loops will impact energy losses around more than
one loop. Epp and Fowler (1970) developed a more efficient approach by
simultaneously computing corrections for all loops. As in the Hardy
Cross method, an initial solution that satisfies continuity at all nodes is
required. For a simultaneous loop equation solution, Eq. 5-16 for loop LP
becomes:
n
m1
LP l lp (5-21)

l
F Q llloop Q
lp
ncp(l )
K Q 0
where ncp(l) is the set of one or two loops in the network that contain pipe l

(e.g., loops P and I for pipe 2 in Example 5.4). The sign convention on the pipe
flow relative to the loop is the same as for the Hardy Cross method.
The Newton-Raphson method is then used to solve Eq. 5-21 for the Qs. A
system of linear equations must now be iteratively solved rather than the
single equations of the Hardy Cross method.
A first order Taylor series approximation of Eq. 5-21 for loop LP is:

FLP m1

FLP
Q
m 1 LP
Q lp ncl
Qlp ) Q m Q lp
F
LP

)
l lloop ( Q
( LP )
1
(Q
)
lp ( (5-22)
5-22 CHAPTER FIVE

where ncl(LP) is the set of loops that have a common pipe with loop LP (e.g.,
loops I and III for loop II in Example 5.4). In vector form for all loops
simultaneously Eq. 5-22 can be written as:

JL Q = -F(Q(m-1)) (5-23)

where Q(m-1) is the vector of npipe pipe flow, Q is the [1 x (nloop+nploop)]

vector of loop flow corrections and F(Q(m-1)) is the [1 x (nloop+nploop)] vector
of residuals of the loop conservation of energy equations (Eq. 5-14) evaluated
at Q(m-1). Residuals are the values of the right hand side at the trial values of Q.
The objective is for all of those terms to equal zero such that all loop equations
are satisfied.
JL equals F/ (Q) and is the Jacobian matrix of first derivatives of
looptheequations evaluated JL is square [(nloop+nploop)
Q(m-1).
(nloop+nploop)],
at symmetric and positive definite.
x The rows in JL correspond to
the loop equations and the columns are related to the loop corrections.
The LPth diagonal term of the Jacobian is the sum of the first derivatives of
the lloop pipes in loop LP or the summation in the first term in Eq. 5-22. This
term is also equivalent to the denominator of Eq. 5-20.
The difference between the simultaneous loop method and the Hardy Cross
method is that some of the off-diagonal terms are non-zero. The Jacobian term
in row LP (loop LPs conservation of energy equation) and column lp (common
loop) corresponds to the gradient term for the pipe that is common to
loops LP and lp. These terms are equal to zero if the loops do not have
common pipes. If the loops have common pipes, these terms are the negative
absolute
of the sum values of F Q n hL / for pipes that appear in loop LP and
of the
Q The negative sign results because the flow direction in loop lp is opposite to
lp.
loop LP. An example of forming Eq. 5-23 is given in Example 5.5.
Once the matrices are formed, Eq. 5-23 can be solved by any linear
equation solver for Q. The pipe flows are updated by the loop corrections as
in the Hardy Cross method (i.e., Q(m) = Q(m-1) +/- Q). The solution algorithm is
the same as the Hardy Cross method except steps 5 and 6 are reduced
to a single step and all corrections are computed simultaneously.
Since the equations are nonlinear, several iterations may be necessary to
converge to the solution, like the Hardy Cross method. To end the
algorithm, one of several stopping criteria can be applied: (1) number of
iterations allowed or (2) the magnitude of the change in loop flow rates, Q.
As in the Hardy Cross method, if a pipes flow direction changes from the
assumed value, the signs for that pipe head loss terms are switched for all loops
containing the pipe during the next iteration in the loop equations using the sign
convention noted above. The direction change does not alter the
coefficient matrix, JL, or the signs on the individual terms. The diagonal terms
are the sum
 NETWORK HYDRAULICS 5-23

of the absolute values of the gradients and the off-diagonal terms are
always negative.

Example 5.5

Problem: Determine the flow rates in the pipes in the three loop
network in Example 5.4 and the nodal heads at all nodes using the
simultaneous loop method and the Hazen-Williams equation.

Solution: The Example 5.4 starting point is used again in this example.
Equation 5-23 is solved to provide the simultaneous loop corrections. For the
example network, the four loop equations were developed in Example 5.4 for
loops P, I, II, and III as:
1.85 1.85 1.85 2
P : K1 Q1 K 2 Q2 K 9 Q9 (240 0.9376 Q P ) H res 2 H res1
200 I : K Q1.85 K Q1.85 K 1.85 K Q1.85 H H 0
2 2 4 4 7 3 3 1 1
Q 1.85 1.85
II : K 7 Q7 K 6 Q6 K 8
7
1.85 1.85
K 5 Q5 H 6 H 6
Q
III8: K Q K Q K0 Q1.85 K Q 1.85 H H 0
1.85 1.85

10 10 6 6 4 4 9 9 3
3

The coefficient matrix terms are the gradients of the loop equations
with respect to each loop flow correction or:
FP FP FP FP
(Q I ) (Q II ) (Q III )
(Q
FI FI FI
P )
(Q I ) (Q
FI II ) (Q III )
JL

FII FII FII
(Q P)
(Q (Q I ) (Q II ) (Q III )
F
F F III FIIIII
P) F
(Q I ) (Q II ) III
III

The diagonal terms are identical to the denominator of the Hardy

Cross correction: (Q III )
(Q P )
FLP h L,l
n Ql
(QLP ) l
lloop
5-24 CHAPTER FIVE

where lloop is the number of pipes in loop LP. From the values in Table E5-4a
and b, a symmetric coefficient matrix can be filled. The rows 1-4 correspond to
loops P, I, II, and III. As shows in step 4 of Example 5.4:

FLP

(QLP ) l n h L,l n h,1L hL ,2 hL ,9 2 (0.9376P
Ql Q 1 Q 2 Q9
Q )
lloop
0.138 0.774 153.33 2 (0.9376) ( 8) 169.24

The off-diagonal terms are the gradients for pipes that appear in loop LP
and another loop, lp or:

Flp

FLP h L,l
Ql (Q LP )
lncpipe( LP,lp) n
(Qlp )
where ncpipe (LP,lp) is the set of pipes that are common to loops LP and lp
(e.g., pipe 2 for loops I and P). The value in row 1 (corresponding to
loop equation P) and column 2 (corresponding to loop I) is the gradient of the
only common pipe 2 with respect to the flow correction in loop I or:

FP

h L,l h L, 2 3.76

(Q I ) Ql n Q2 1.852 9
l ncpipe( P,I ) n
0.774
As noted above, this gradient is also FI / (Q P ) .
The remaining diagonal terms are given in Table E5-4b and the off-
diagonals correspond to pipes common to two loops (Table E5-4a). Pipe
2 appears in loops P and I, pipe 9 appears in P and II, pipe 4 is located in I and
II and pipe 6 is in loops II and III. No pipe is common to loops P and II so
the coefficients are zero in those locations (column 3-row 1 and column 1-row
3). The JL matrix is then:

0.774 0 153.33
169.25 61.37 56.71 2.96
JL
56.71 80.12

0.774
0
2.96 7.60
7.60
153.33 value of the equation
The right hand side of the equation is the computed
with the current flow estimates that were computed for the Hardy Cross method
317.23
and listed in the second row of Table E5-4b.

F(Q(0))T = [97.55. -7.40, 20.46, -23.97]

 NETWORK HYDRAULICS 5-25

For example, row 3 corresponds to loop II:

F K Q 1.85 K Q 1.85 K Q 1.85 K Q 1.85

II 7 7 6 6 8 8 5 5

55.2 (0.5) 1.85

1.42 (3.5) 1.85
13.6 (0.5) 1.85
0.233 (5.5) 1.85

15.31 14.37 3.77 5.45 20.46

The system of linear equations, JL Q = -F, is solved for the unknown flow
corrections.
0.774 0 153.33

169.25 61.37 56.71
J L Q 2.96
Q P
0
0.774
56.71 80.12 7.60 II
2.96 7.60 317.23Q I
Q
153.33 Q III
7.40
F 97.55



20.46

Resulting in:

QT = [ QP, QI QII QIII ]T23.97

= [-0.942, -0.525, -0.665, -0.404]T

The corrections are applied using the equations shown in Example 5.4, step
7. For example, the flow rates in pipes 1 and 2 become:

Q (1) Q (0) Q Q 9 0.942 0.525 9.416

2 2 P I

The loop flow corrections are still large so additional iterations are needed
to converge to the solution. Since flow corrections are made on all loops
simultaneously, this method converges in four iterations (to an absolute change
in Q of 0.001 cfs) compared to 11 for the Hardy Cross method (with a larger
tolerance). The results are summarized in Tables E5-5a and b.
Since the flow rates from the Hardy Cross and simultaneous loop methods
are the same the nodal heads will be the same for both results.
5.3.2 Solution of the Node-Loop Equations (Flow Adjustment Algorithm)

Using the loop equations to represent conservation of energy, Wood and

Charles (1972) developed the linear theory (Flow Adjustment) method by
5-26 CHAPTER FIVE

coupling the loop equations with the node equations. Wood and Rayes (1981)
later showed that a modified linear theory (presented here) exhibits
superior convergence characteristics compared with the original linear theory
method.

Table E5-5a: Pipe flow values for the simultaneous loop

Iter. Pipe Flow (cfs)
correction method.
1 2 3 4 5 6 7 8 9 10 P
m

2 20.94 9.42 11.53 5.88 6.17 3.24 0.36 1.17 0.46 0.60 7.06
3 21.21 9.79 11.43 6.04 6.07 3.39 0.36 1.07 0.25 0.54 6.79
4 21.27 9.84 11.43 6.03 6.07 3.39 0.36 1.07 0.19 0.54 6.73
5 21.27 9.85 11.43 6.03 6.07 3.39 0.36 1.07 0.19 0.54 6.73
Table E5-5b: Loop corrections for simultaneous loop
correction iterations.
Iteration Loop P Loop I Loop II Loop III
1 -0.942 -0.525 0.053 -0.404
2 -0.271 0.100 0.097 0.060
3 -0.053 0.001 0.002 0.004
4 -0.005 0. 0. 0.001

In the modified method, rather than solve for loop corrections and be
required to provide a feasible initial solution, conservation of energy around a
loop (Eq. 5-16) is written directly in terms of the pipe flow rates or for a closed
loop:
n
l l (5-24)
K Q
F (Q) l lloop
However, the number of 0 unknown pipe flows is equal to the number
of pipes (np) but only nloop + nploop equations of the form of Eq. 5-16
are available. Therefore, these equations are coupled with the nodal
conservation of mass equations (Eq. 5-11):

Q
l J in
l lJ Q
out
l q

With conservation of mass, the number of equations is nnode

node equations plus nloop closed loop equations and nploop pseudo-loop
equations or a total of np equations written in terms of the np unknown pipe
flow rates.
These nonlinear equations are also solved iteratively by applying the
Newton-Raphson method. Taking a Taylor series expansion of a loop
equation results in:
 NETWORK HYDRAULICS 5-27

) LP (m1)
(m) (m1)
(Ql Ql
l lloop

l Q( m F

(Q )
1)

where F is theloop dQ
dFLP equation (Eq. 5-16) and Q(m-1) are the known pipe flows for
the previous iteration and Q(m) is the unknown flow rates at iteration m.
This equation can be rearranged with the known terms on the right hand side as:

F (Q )
( m1)
(5-25)
Q l(m) Q l(m1)
l lloop l lloop
l Q ( m 1) l Q ( m 1)


dF
dQ dF
dQ
A similar relationship can be written for conservation of mass. It is linear
with respect to the unknowns, Q(m) since the gradient terms and the functions F
can be evaluated at Q(m-1). Eq. 5-25 can be written in matrix form as:
JNL Q(m) = FNL = - F + JNL Q(m-1) (5-26)

where JNL is the Jacobian of the node-loop equations and FNL is the vector of
functions of known values from the previous iteration. JNL and FNL vary
between iterations as Q moves toward the solution. F is the vector of residuals
computed by substituting Q(m-1) in the node-loop equations.
The rows in JNL correspond to the conservation of mass and energy
equations and the columns relate to the unknown pipe flow rates. For the
conservation of mass equation for node i, the terms in the corresponding row in
JNL will be 0 if the pipe is not connected to node i, +1 if the pipe is carrying
flow to the node (i.e., the pipe is in the set Jin,i, e.g., pipes 6 and 8 for node 4 in
Example 5.4) and 1 if the pipe is in set Jout,i and carries water from node i (e.g.,
pipes 2 and 3 for node 1 in Ex. 5.4). For the conservation of energy equations,
the gradient terms are the same as the Hardy Cross terms (Eq. 5-21)
(i.e., |n hL/Q| = |nKQ n-1|), if the pipe appears in the loop, and zero, otherwise.
The full term becomes more complex when a pump appears in the loop
(Example 5.6).
JNL and FNL are evaluated at Q(m-1) and Eqs. 5-26 are solved for the new
pipe flows, Q(m). This iterative process continues until a defined stopping
criteria is met, such as when the absolute or percentage difference between two
iterations flows, Q(m) and Q(m-1), is less than a tolerance for all pipes or a
limiting number iterations are completed. Since conservation of mass is solved
as part of Eq. 5-26, the initial solution does not have to satisfy this condition.

Example 5.6

Problem: Determine the flow rates in the pipes in the three loop
network in Example 5.4 and the nodal heads at all nodes using the modified
linear theory
5-28 CHAPTER FIVE

method and the Hazen-Williams equation. Use the same starting point as
Example 5.4.

Solution: The node-loop equations consist of the node equations written with
respect to the pipe flow (Eq. 5-11) and the loop equations (Eq. 5-16).

Node 1: Q1 Q2 Q3

q1 0 Node 2 : Q2 Q4

Q9 q 2 4 Node 3 : Q9

Q10 Q P q 3 6 Node 4 :

Q6 Q8 Q10 q 4
1.85 1.85 1.85
Loop P : K1 1 5 Node 5
K 2Q : 2
Q5 9 89
K Q Q q 5 (240
0.9376 QP2 )
Q H H 5 0 200
1.85
res.2
1.85
res.1
1.85 1.85
Node 6 : Q3 Q5 Q7 q 6 5 H1 H1 0
Loop I : K 2 Q2 1.85 K 4 Q 4
1.85 K 7 Q7 1.85
K 3 Q31.85
Node 7 : Q4 Q7 Q6 q 7 3 H6 H6
Loop II : K 7 Q7 K 6 Q6 K 8 Q8 K 5 Q5
0
Loop III : K10 Q10 1.85 K 6Q6 K 4 Q4 1.8 K 9Q9 1.8
0
These 11 equations can for the 10 pipe5 flows and 15 pump flow.
be solved
An arbitrary positive flow direction has been assigned to each pipe that
is consistently applied in1.85
the conservation of mass and energy equations.
For example, pipe 2 is positive when water flow from node 1 to node 2. Thus,
it is an outflow from node 1 and is given a negative sign in that
conservation of mass equation. It is an inflow to node 2 and given a positive
sign in that nodes mass balance equation. In addition, choosing clockwise
as positive in all conservation of energy equations, flow from node 1 to
2 is counterclockwise (negative) in loop P and clockwise (positive) in loop
I. Using our convention for taking the sign of the flow rate, flows that
are opposite of the assumed direction become negative and change the signs
on the terms. For example, if Q2 became 2 with n = 1.85 then:

1.85
( Q2 ) 1.85
1.85

Q2 3.61.
2
Thus, negative flows
are possible and distinguish the proper magnitude and that
the assumed direction was incorrect.
 NETWORK HYDRAULICS 5-29

As noted above, the columns in the coefficient matrix, JNL, for these
equations correspond to pipes and the rows correspond to the nodes plus loop
equations. The first seven (nnode) rows of the right hand side vector are nodal
demands (Eq. 5-11). For node 6, the non-zero Jacobian terms correspond
to pipes 3, 5 and 7. For that node, pipe 3 is in inflow pipe and 5 and 7
outflow pipes so the gradients for these pipes are:

FN ,6 FN ,6 FN ,6

Q3 1; Q 5 1; Q 7 1

where FN,6 denotes the node equation for node 6. The sixth row in JNL is:

0 1 0 1 0 1 0 0 0
J NL, row 6 0
0
The loop equations are linearized by a Taylor series expansion and
the terms in the gradient matrix, FLP Ql , are the derivatives of loop equation
for
loop LP with respect to flow in pipe l or n hL,l /Ql for the pipes in the loop
and zero otherwise. The sign for the term relates if the assumed flow is
clockwise (+) or counter-clockwise (-) relative to the loop. Therefore, values
of the last four rows of JNL corresponding to the loop equations come
directly from the last row of Table E5-4a. Based on the assumed flow
directions, the signs are: (Loop P = [-1, -2, 9, -Pump]), (Loop I = [2, -3, 4, -7]),
(Loop II = [-5, 6, 7, -8]),
and (Loop III = [-4, -6, -9, 10]).
1 the 1
For 1 0 0 0 0 0 0
initial point used in the Hardy Cross method, JNL is: 0
0
1 0 1 0 0 0 0 1
0 0
0
0 0 0 0 0 0 0 1 1 1
0
J NL
0 0 0 0 1 0 1 0
1 0
0
0
0 0 0 1 0 0 1 0 0
0
0 1 0 1 0 1 0 0 0
0
0
0 0 1 0 1 1 0 0
The right
0 hand side of Eq. 5-24 for the node equations is equal to 0the nodal
0

demand, q, demonstrating that the linearization of the node results does
0.774 0 0 0 0 0 0 153.33 0
not alter the equation. For node 6 with the assumed flows the RHS is:
0.138 15.0
2.96 0 0 0 0 0


0 0.774 0.918 56.7 0

0 0 0 0 1.83 7.60 56.7 13.97 0 0 0
0 2.96 0 7.60 0 0 153.33
153.33 0
0 0
5-30 CHAPTER FIVE

FN ,6
FN ,6 (Q (0) )


l Jin , Jout Ql Q
l (0)
Q () 0

N ,6 (0) N ,6 (0) N ,6 (0)

(Q3 Q5 Q7 q6 ) F Q3 F Q Q 5 F Q7
Q 3 5

(11 5.5 0.5 5)Q7 (1)(11) (1)(5.5) (1)(0.5) 5

The reader can confirm that an unbalanced assumption of pipe flows gives the
same results (e.g., Q3 = +10, Q5 = -5.5 and Q7 = -0.5).
The last four rows are more than the deviations in the energy balances (i.e.,
-F(Q)) as in the simultaneous loop solution. The additional terms are the
gradients of the loop equations (the last four rows of JNL) times the present flow
estimates (Q(m-1)). For loop I that contains only pipes, the right hand side is:

F
Q m1 h n (h / Q (0) ) Q
L L
F (Q m 1 ) Q
Q m 1 (0)

hL n hL (n 1) hL 0.85 (7.40) 6.29

The resulting relationship is different

for loop P that contains a pump.
F (Q m1 ) m1
Q
F Q Q m 1

(h h E) n (h / Q (0) ) Q (0) (2 (0.9376) Q (0) )Q (0)

L P L
P P

(97.55) 1.85 (77.55) (2(0.9376) 8) 8 165.93

The full loop equation is included in the first term including the
energy difference for pseudo-loops. The gradient of the energy difference is
appear in the second 0term.
zero so it does not 0
iteration, the resulting
For the first 4 right hand side is:
6
5 4
5
FNL (Q )
(0)
5

3

6
(97.55) 1.85 (1.49 3.76 82.8)
(7.40) 1.85 (3.76 9.60 15.31 120.01

165.93
5.45) 5
(20.46) 1.85 (15.31 14.37 3.77 17.39
6.29
(23.97) 1.85 (82.79 14.37 9.60 82.79)
5.45)

20.37 5



5

3

 NETWORK HYDRAULICS 5-31

The system of linear equations (JNL Q(1) = FNL, Eq. 5-24) are then solved
for the unknown Q(1):

Q (1) T [20.94, 9.42, 9.42, 11.53, 5.88, 6.17, 3.24, 0.36, 1.17, 0.46, 0.600,
7.06]

Updating JNL and the RHS values, the method converges to the solution
in 3 iterations as shown in Table E5-6.

Table E5-6: Pipe flows for 4 iterations of the modified linear

Pipe
theory method for the pipe equations.

m 1 2 3 4 5 6 7 8 9 10 P
0 20 9 11 6 5.5 3.5 0.5 0.5 1 1 8
1 20.94 9.42 11.53 5.88 6.17 3.24 0.36 1.17 0.46 0.60 7.06
2 21.21 9.79 11.43 6.04 6.07 3.39 0.36 1.07 0.25 0.54 6.79
3 21.27 9.84 11.43 6.03 6.07 3.39 0.36 1.07 0.19 0.54 6.73
4 21.27 9.84 11.43 6.03 6.07 3.39 0.36 1.07 0.19 0.54 6.73

5.3.3 Solution of the Node Equations (Simultaneous Node Adjustment

Algorithm)

The pipe head loss equation for pipe l that connects nodes i and j:

hL,l H j H i K l
Ql (5-27)

can be transformed to nodal head equation as: n
1/ n

Ql H jH i (5-28)
Kl

Shamir and Howard (1968) solved used this transformation to form nnode
node equations for the nodal heads using the Newton-Raphson
method. Substituting Eq. 5-28 in Eq. 5-24 for a general node i gives:
1/ n
H j H i
FN ,i :

l Jin , J out K l
qi
(5-29)

where the summation is over the pipes entering or leaving the node. In the head
difference, the node for which the mass balance is written is always the second

term. The sign notation, H j H i , provides the flow direction. If Hi is
greater
5-32 CHAPTER FIVE

than Hj, flow is from node i towards j (an outflow) and the sign is
negative. When Hj exceeds Hi, the sign is positive and flow is supplied to
node i from node j (an inflow).
For node i, application of the Newton-Raphson method yields:

N ,i
H i F (H (5-30)
FHi
i ncnode(i( H 0 ( m1)
)

where H(m-1) are the nodal heads for the previous and present iterations
and ncnode(i) is the set of nodes that are connected by pipes to node i and
node i (e.g., nodes 2, 4, 6, and 7 for node 7). In matrix form for all
equations and nodes:

JN H = - FN

(5-31)

where JN is the Jacobian matrix of the node equations with respect to

the changes in nodal heads and FN is the residuals of the node equations.
Both JN and FN are evaluated at H equal to zero or the present
iterations head estimates. It should be noted that the square Jacobian matrix
(nnode x nnode ) is symmetric and positive definite.
For pipe l that connects nodes i and j, the Jacobian terms gradients for
the conservation of mass at node i are: 1
FN ,i 1 H j H i 1 n Ql
(5-32a)
n K l K l
n (H j i
j H m 1
(H ) H )
1
FN ,i 1 H j H i n Ql
(H i )

K l
1
n (H j H i )
(5-32b)
H m 1 l ncp(i) n K l l ncp(i)

Regardless of the flow direction, the gradient sign for the flow balance at
node i are all positive for Hi terms while the terms for the connecting nodes j
are all negative.
After Eq. 5-31 are solved for H, the heads are updated by subtracting the
nodal corrections or:

(5-33)
H (m) H (m1) H
i i
i
 NETWORK HYDRAULICS 5-33

The process is completed iteratively until the changes in nodal heads for all
nodes are less than a tolerance or a desired number of iterations are completed.
The overall process is:
1) Initialize m = 0 and define starting set of nodal heads, H(0).
2) Set m = m + 1
3) Compute nodal balances using Eq. 5-29 and gradients using Eq. 5-32
4) Solve the system of equations 5-31 for H
5) Update nodal heads using Eq. 5-33
6) Check stopping criteria. If satisfied, stop. If not satisfied, go to step 2.

Example 5.7

Problem: Determine the nodal heads at all nodes in the three loop network in
Example 5.4 using the modified linear theory method and the Hazen-Williams
equation. Assume an initial head vector,
H(0),T = [198, 193, 195, 175, 188, 190, 184]
Compute the flow rate in pipe 4.
Solution: Given the initial head distribution (step 1), we can now update the
values at m=1 (step 2).
Step 3: As listed in Example 5.6, the nodal mass balance equations are:

Node 1: Q1 Q2 Q3 q1 0
Node 2 : Q2 Q4 Q9 q 2 4
Node 3 : Q9 Q10 Q P q 3
6 Node 4 : Q6 Q8 Q10 q 4
5 Node 5 : Q5 Q8 q 5 5
Node 6 : Q3 Q5 Q7 q 6 5
Node 7 : Q4 Q7 Q6 q 7 3

For the node equation solution, these seven equations are written in terms
of the seven nodal heads as: 0.54 0.54 0.54

FN ,1 H res.1 H 2 H 1 H q1 0
K1 K2 K3
H 6 H 1 0.54
0.54 0.54
1 H H 2 H 2
1 2 7
3 q2 0
F N ,2
H K 2 H K 4 H K 9

5-34 CHAPTER FIVE

0.54 0.54 0.5

res.2 H 3
4H
FN ,3 2H q3
3
3
240
H K 9

H K 10
H 0.9376
0

0.54 0.54 0.54
7 H 4 H 5 H 4 H 3 H 4 H

FN ,4
K K8 q4
K 10 0
6
0.54
H
FN ,5 H6 5 4 5 0.54 q5

H K 5 K8 0
H
0.54 0.54 0.54
1 H 6 5 H 6 7 H 6
FN ,6 q6
H K K K 0
H 0.54
H 0.54 0.54
3 5 7

FN ,7 2 H 7 H
6 H 7 H 4 H 7 q
K4 K7 K 7
H 0
6

Since this form is nonlinear, all nnode equations are linearized about the nnode
corrections.
To set up the linear equations, the pipe flows are computed using the
previously computed Ks (Table E5-4a) and the initial nodal heads and
are listed in Table E5-7a. For pipe 1 and node 1:
0.54 0.54

Q 1(0)

H res.1 H 200
23.368 cfs
K1 0.00584
198

1
The gradient of the flow rate in pipe 1 with respect to a change in head at
node one is given by Eq. 5-32b:

F (H (H
H 1) 0.46 H 1) 0.54
H11 (0) res.1
1
res.1
n (H res.1 1 1
H1 n K 0.54 H ) K 0.54
Q1 23.368

n ( H res.1 H1 )
1.85 (2)
6.316
The gradient term for each pipe is listed in the last row of Table E5-7a.

Table E5-7a: Pipe flows computed with the defined initial nodal
Pipe 1 2
heads. 3 4 5
K 0.00584 0.0645 0.0645 0.349 0.233
(Hi - Hj) 2 5 8 9 2
23.368 10.474 13.50 5.785 3.20
Q(0)
Q/n (Hi - Hj) 6.316 1.131 0.911 0.347 0.863
 NETWORK HYDRAULICS 5-35

Table E5-7a (cont.): Pipe flows computed with the defined initial nodal
heads.
Pipe 6 7 8 9 10
K 1.416 55.20 13.60 82.79 82.79
(Hi - Hj) 9 6 13 2 20
2.715 0.302 0.976 0.134 0.464
Q(0)
Q/n (Hi - Hj) 0.163 0.027 0.041 0.036 0.013

For the first iteration, the nnode x nnode coefficient matrix, JN, is:

8.352 1.131 0 0 0 0.911 0

0.347

0 1.514 0.036 0 0 0 0

1.131
0 0.036 0.126 0.013 0
JN 0 0.163
0 0
0 0.013 0.216
0.041 0 0.027
0
0.911 0 0 0.041 0.537
0.904 0.863
The diagonal terms are the sum of the gradients for 0that node.
0 0 For node 2,
(0)
0.863 (0)
1.802
(0)
F2 Q2
0.347 Q09 0.163 Q04

H 2
(0) 0.027
n (H H ) n (H (0) H )
(0) (0)
n (H (0) H (0) )
1 1.131
2 0.347 0.036
3 2 1.514 2

7
where the individual terms were taken from Table E5-7a. The off-
diagonal terms are equal to the gradients of the node equation with
respect to the adjacent node and are always negative.
For example, the change in node 2s mass balance due to a change in head
at node 3 is:

F2 9 F3
Q (0) 0.036
( H ( H
n (H (0) H (0) )
3 2
3 ) 2)

This term is placed in row two for mass balance at node 2-column 3
for connecting node 3. A change in the head at node 2 causes an equal
change in the mass balance at node 3. So the term also appears in row
(node equation) three column (connecting node) two.
The right hand side of the system of equations is FN, the residuals of the
nodal balance equations. They are computed by substituting the summed head
5-36 CHAPTER FIVE

values in the mass balance equations listed above. For example for node 6 at
iteration 0 the residual is:
0.54 0.54 0.54

FN ,6 H 1 H H 5 H 6 H q6
K3 K5 K7
6 0.54 6 H 7
0.54
0.54
198 190 188 190 190 5
0.0645 1840.233
5.002
55.2


Their values are shown in first row of Table E5-7b. The signs of
these terms are then changed in the solution of Eq. 5-31. The set of equations
5-31 is then:

8.352 1.131 0 0 0 0.911 0

H 1 0.347

1.131
1.514 0.036 0 0 H0 2
0.163
0 0.036 0.126 0.013 0 0 00

H 4
J N H 0 0.013 0.216 0.041 H
0 0 0.027
H 3
5
0
0.911 0.537
0 0 0 H 6
0.041 0.904 0.863

0 0 0 0.863 1.802 H 7
0.606

0.347 0 0.163 0 0.027
0.822

0.845
FN

2.779
0.330


5.002
Step 4: The first iteration solution for HT = [-0.333, -0.921, -2.520,
3.684,

0.782, -2.575, -0.301].
0.372are
Step 5: The updated nodal heads listed in Table E5-7c where H (1) = H(0)
- H. For node 1:

H (1) H (0) H 198.0 (0.333) 198.333

1 1 1

Step 6: A large change in nodal heads was applied so return to step

2 (m=2). The nodal mass balance equations are evaluated again and mass
balance
 NETWORK HYDRAULICS 5-37

has not been achieved (Table E5-7b). The nodal heads are updated two more
times and the values converge as seen in Table E5-7c.
After the flows have converged after iteration 3, the flow rate in pipe 4 can
be computed by:
0.54 0.54
H 2 H 193.863 184.145 6.028
Q
4
K 0.349
cfs
( final )
4
7
Table E5-7b: Residuals of the nodal balance equations, FN, computed using
the nodal heads at the beginning of iteration. To solve the set of equations
these residuals are multiplied by 1 in Equation 5-31.
Iteration Node

1 2 3 4 5 6 7

0 -0.606 0.822 0.330 -0.845 -2.779 5.002 0.372

1 0.085 -0.024 0.010 -0.0.65 -0.646 0.503 0.047
2 -0.001 0 0 -0.001 -0.028 0.027 0.001
3 0 0 0 0 0 0 0

Table E5-7c: Nodal heads for the three iterations.

Iteration Node

1 2 3 4 5 6 7

0 198 193 195 175 188 190 184

1 198.333 193.921 197.52 171.316 187.218 192.575 184.301
2 198.320 193.864 197.523 170.603 185.948 192.445 184.154
3 198.320 193.863 197.522 170.584 185.860 192.446 184.145

5.3.4 Solution of the Pipe Equations

In the loop equation formulation, head losses were balanced around a series of
pipes between points with a known difference in energy. Hamam and Brameller
(1972) for gas networks and Todini and Pilati (1987) for water networks wrote
conservation of energy for each pipe (Eq. 5-14) resulting in a set of
npipe equations with the npipe pipe flows and the nnode nodal heads as
unknown. They coupled these equations with the node equations written in
terms of the pipe flows (Eq. 5-11) to form a set of npipe plus nnode equations
for an equal number of unknowns. The method is also known as the
hybrid or gradient approach. For additional background on the method, see
Osiadacz (1991).
5-38 CHAPTER FIVE

For the four-pipe network shown in Figure 5-3, the pipe equations include
one equation for each node and each pipe. With the assumed set of pipe flow
directions, the node equations are the conservation of mass relationships or:

FQ2: + Q1 Q2 Q3 q2 = 0 {node 2}
FQ3: + Q2 Q4 q3 = 0 {node 3}
FQ4: + Q3 + Q4 q4 = 0 {node 4}

The pipe equations are written for each pipe with the nodal head and pipe
flow on the left-hand side of the equation. The friction loss equation is given a
positive sign, so the upstream source node head has a negative sign and
the downstream head takes a positive sign. The equations for pipes 1 to 4
in the four-pipe network are:
n
Q + H2 - H 1 = 0 {pipe
FP,1: K1 1} 1
n
Q - H2 + H 3 = 0 {pipe 2}
FP,2: K2 2
n
Q - H2 + H4 = 0 {pipe
F
P,3 : K 3 3} 3 n
Q - H3 + H 4 = 0 {pipe 4}
F
P,4 : K 4 4

where H1 and all Ks are

known.

H1 = 50

(1)

(2) (3)

q3 = 2.0 (4) q4 = 2.5

3 4

Figure 5-3: Four pipe example network for pipe equations formulation.
 NETWORK HYDRAULICS 5-39

n
The notation, Q1 , represents that the absolute value of the pipe flow 1
is to the power
raised n and the sign of the pipe flow is applied to the head loss
equation term. The flows sign also is applied in the node equations. Thus, a
negative flow is acceptable and defines a flow that is in the opposite direction
from the initial assumption. In our example, conservation of mass and energy
comprise seven equations written with respect to four pipe flows and
three nodal heads. Applying the Newton-Raphson method (but solving
for the changes in flow and head) for pipe l that connects nodes i and j gives:

Q F (Q ( m1) )

(m)
Ql Q (5-34)
F

l J in , J out
Q
l Q0
FP FP FP
H j m
(m)
H i l P
Hi Hj Ql Q (m) F (Q (m1) , H (m1) )
H 0 Q0
H 0
(5-35)
The derivatives of the mass balance equations (FQ) are 1 (outflow pipe), -1
(inflow pipe) or 0 (not connected to node). The right-hand side is calculated by
substituting in the present estimates of the flow rates and defined as dq.
For node 2, Eq. 5-34 becomes:

FQ (m ) FQ (m ) FQ (m) (m1)
Q1 Q 2 Q3 FQ1 (Q )
Q1 Q 0
Q2 Q 0
3 Q
0 Q
(m-1) (m-1) (m-1)
(+1) Q1 + (-1)Q2 +(-1)Q3 = -(+Q1 - Q2 - Q3 - q2) = - dq2
{FQ,2: node 2} (5-36a)

Similarly for nodes 3 and 4:

+Q2 -Q4 = -(+Q2(m-1) -Q4(m-1) - q3) =- dq3 {FQ,3: node 3} (5-36b)

+Q3 +Q4 = -(+Q (m-1)

+Q (m-1)
-q )=- dq {FQ,4: node 4} (5-36c)
3

For the energy balances,

4 the derivatives with respect to the nodal heads are
1 (sink node), -1 (source 4node), and 0 (not connected to pipe). The derivatives
n1
with respect to pipe flow
4 are n K Q. The right-hand side is computed
using
the present iterations flows and nodal heads and defined as dE. For pipe 1,
Eq. 5-35 is:
5-40 CHAPTER FIVE

n1 n
n K 1 Q1 Q1 +(+1) H2 = -(H2 + K1 Q1 -H1 ) = -dE1 {FP1 : pipe 1} (5-37a)

The gradient for the head at node one is zero since it is a fixed
head reservoir. Node 2 is the downstream (sink) node so its gradient is
positive and the pipe flow term is evaluated at the present Q(m-1). For clarity the
m-1 iteration counter is not included in all pipe equations.
For pipes 2 to 4, the gradient equations are:
n1 n
{FP2 : pipe 2}
n K 2 Q2 Q2 - H2 + H3 = -(-H2 + H3 + K2 Q2 )
= - dE2 (5-37b)
n1 n
n K3 Q 3 Q3 - H2 + H4 = -(-H2 +H4 +K3 Q3 )= - 3 {FP3 : pipe 3}
dE (5-37c)
n1
n K 4 Q4 Q4 - H3 + H4 = -(+H3 - H4 + K4 Q4 n)= - 4 {FP4 : pipe 4}
dE (5-37d)

These seven equations (Eq. 5-36 a-c and 5-37 a-d) can be solved for
the changes in nodal head and flow. The new iterations values are then
computed by:

(5-38)
H(m) = H(m-1) + H(m)
(5-39)
Q(m) = Q(m-1) + Q(m)
Todini and Pilati (1987) generalized this formulation in matrix form.
Conservation of energy (in the pipes) and mass (at the nodes) equations (Eq. 5-
16 and 5-24) can be written in matrix form as:

FP(Q,H) = A11 Q(m) + A12 H(m)= 0 (5-40)

FQ(Q,H) = A 21 Q(m) q= 0 (5-41)

respectively.
Looking back at the node and pipe equations written at the beginning
of this section, you will note that the coefficients on the flows entering and
leaving node 2 are 1, -1, -1, and 0 for pipe flows 1-4, respectively. These
values are identical to the coefficients in the column related to nodal head H2
in the pipe
 NETWORK HYDRAULICS 5-41

equations. These coefficients comprise the matrices, A12 and A21. A21 is
T
the connectivity, also known as the topological, matrix and A21 = A12 . The
terms in the A12 matrix identify the network connections and take on values of
1, -1 and
0. Each column corresponds to a pipe and values of -1 are assigned to
the upstream node for the pipe, 1 to outlet node of the pipe and 0 if the pipe is
not connected to the node. For the four pipe network,
0 0
1 1 0
1
11
T
A 211 1 0 1 and12A A21
0
0 .
0 0 1 1 1 0 1
A11 is defined as:

n1 0 1 1
Q1 0
1

A 11 K
O (5-42)
0 n 1

K npipe Q npipe
Note that the rows in A21 correspond to nodes 2 4.

Applying the Newton-type solution scheme to the system of equations
gives:
dFP(Q,H) = n A11 Q(m) + A12 H(m) = - dE (5-43)

dFQ(Q,H) = A21 Q(m) = - dq (5-44)

The resulting equations for the four-pipe network are given in Eqs. 5-37a-d
and 5-36a-c. The right-hand side terms are shown for the four-pipe network as
the residuals in the mass and energy balance equations at iteration m-1. This set
of equations is solved for Q and H and H(m) and Q(m) are updated by:

(5-45)
H(m) = H(m-1) + H(m)
(5-46)
Q(m) = Q(m-1) + Q(m)
As in the earlier methods, the absolute or relative changes in flow or head
or the number of iterations can be used as stopping criteria.
The overall procedure for solving the pipe equations is:
5-42 CHAPTER FIVE

1) Initialize m = 0 and define starting set of nodal heads, H(0), and pipe
flows, Q(0)
2) Form matrix A12
3) Set m = m + 1
4) Form matrix nA11 using Eq. 5-42.
5) Compute nodal balance error (dFQ = -dq) using Eq. 5-16 and pipe
balance error (dFH = -dE) with Eq. 5-24
6) Solve system of equations (Eq. 5-43 and 5-44) for H and Q
7) Update nodal heads and pipe flows using Eqs. 5-45 and 5-46.
8) Check stopping criteria. If satisfied, stop. If not satisfied, go to step
3.

Assume that all pipes are identical in the four-pipe network and the Hazen-
Williams equation is applied with K = 0.935 and n = 1.852. Also assume that
the flows in the four pipes are [4.5, 2, 2, 0.5] and the nodal heads are [40, 35,
30]. The head in the reservoir, H1, is 50. The A21 is given above and A11
diagonal matrix is computed by KQ n-1 = KQ0.852. For pipe 1, K|Q|0.852= 0.935
(|4.5|)0.852 = 3.37 and the term in the solution matrix is 1.852 K|Q|0.852 = 6.24.
The overall left hand side matrix is:

63.13
. 2 40 0 1 1 00 0

n A 12 0 3.13 0 1 0 0 1

A 11 0 0 1 0 0
A 21 00 1 1 0 0 0 0
0 0 0.96 0 1
1


1 0 11 0 0 0

where the upper left portion 0is nK|Q|n-10, the upper

0 1 is1A12, 0the 0lower left is
right
A21, and the lower right are zeros corresponding to the node
0
equation coefficients on the nodal heads.
The right hand side of the equations is the errors in the equations. For pipe
1, dE1 is computed from Eq 5-37a:

-(K1 Q1n+ H2 - H1) =- dE1 =-(0.935 (4.5)1.852 + 40 50) = - 5.16

For node 2 (Eq. 5-36a):

-(+Q1 - Q2 - Q3 - q2) = -dq2 = -(4.5- 2 - 2 - 0) = - 0.5

After computing values for each equation, the transpose of the right hand
side vector is:
 NETWORK HYDRAULICS 5-43

dq 5.16 1.62 6.62 4.74 0.5

T 0.5
dE 0 T
and the full matrix equation is:

n A 12 Q

A 11 A 0 dE H


0 21 0 1 0 0 0 Q1

6.2 dq

4
0 3.13 0 0 1
1
5.16 0

Q2 1 Q 3


1.62
Q
16.62 4
01 1 03.13 0 0 0 1 0
0 H
0 0 4.74
1 10 0 10 00.96 200 1 0.5


0.5
0 0
0 H H
3 4

0 1 1 0

Solving 0for Q and
0 H gives:

0
Q H T 0 0.247 0.253 0.253 5.16 4.30 0.68 T

Substituting this vector in Eq. 5-45 and 5-46 gives the next iterations pipe
flow and nodal heads or:

Q H T
4.5 2.247 2.253 0.247
34.84 30.70 30.68 T

Repeating the process for another iteration gives the final solution of:

Q H T
4.5 2.24 2.26 0.24 34.85 30.69
30.62 T

Example 5.8

Problem: For the Example 5.4 network, determine the flow rates in the
pipes and the nodal heads at all nodes using the gradient method and the
Hazen- Williams equation. Assume an initial head vector, H(0) T = [198, 193,
195, 175, 188, 190, 184] and an initial pipe flow vector, Q(0) T = [20, 9, 11, 6,
5.5, 3.5, 0.5,
0.5, 1, 1, 8].

Solution: Given the initial head and flow distribution (step 1), we
can now update the values at m=1 (step 2). The set of equations for pipe
equations are comprised of the head loss relationship for each pipe in
terms of the nodal heads and pipe flows (conservation of energy):
5-44 CHAPTER FIVE

n n
H res 1 H 1 0
Pipe 1: hL,1 H res 1 H 1 K1 Q1 n K 1 Q1 n
K Q2
H1 H 2 0
Pipe 2: hL,2 H 1 H 2 K 2 Q2n 2 n
H1 H 6 0
n n
K 4 Q4 H 2 H 7 0
n n
Pipe 3: h L,3 H 1 H 6 K 3 Q3 K 3 Q3 H 6 H 5 0
n n
H7 H4 0
n n
K 5 Q5
Pipe 4: h L,4 H 2 H 7 K 4 H6 H7 0
n n

Q4 H5 H4 0
n n

Pipe 5: h L,5 H 6 H 5 K 5 n K 6 Q6 n H 3 H 2 0
H3 H4
Pipe hL,10 H 3 H 4 K10 Q10 K 10 Q10
Q10:
5
0.9376 Qp 0
h p H res 2 H 3 (240
Pipe Pump:
6: h L,6 )
2
H 7 H 4 K 6 2
(240 0.9376QKp ) HQres 2 H 3
7 7
Q
6 Pipe 7: h 0 H H
L,7 6 7
and the conservation of mass at each node in terms of the pipe flows:
K 7 Q7 Pipe 8: h L,8 H 5 H
Node 1: Q1 Q2 QK3 8 q1Q8 0

4 K 8 Q8 Node 2 : Q2 Q4 Q9 q 2 0
Pipe 9: hL,9 HNode 3 : Q Q QP q3
3 H 2 9K 9 10
K
0 Node 4 : Q Q Q9 q9
Q
Q9 6 8 10 4

0 Node 5 : Q5 Q
8 q5 0
Node 6 : Q3 Q5 Q7 q 6 0
Node 7 : Q4 Q6 Q7 q 7 0

From the node equations, A21 is:

1 1 0 0 0 0 0 0 0
0
1 0 1 0 0 0 0 1 0
1
0 0
0 0 0 0 1 1
0 0 0
0 0 0 0 0 0 1 1
1 0 1
A 21 0 0 0 1 1 0 0 0 0
00
0 1 0 1 0 1 0 0 0
00
0 0 1 0 1 1 0 0 0
0

0 0

0
 NETWORK HYDRAULICS 5-45

The rows of A21 represent nodes and the columns of the pipes. For example,
row 1 corresponds to node 1. The values in the matrix equal the coefficients on
the pipe flows in the node equation. Row 1 is then 1 for pipe 1, -1 for pipes 2
and 3, and 0 for all others. Column 11 corresponds to the pump flow and only
has a non-zero for node (row) 3. Examining the pipe equations and the
coefficients on the nodal heads we can confirm that A21 = A12 T.
Step 3: m=1
Step 4: At iteration 1, the initial flow values are used to compute
matrix
A11. For pipe 1, the diagonal term is:
(0) n 1 0.85
nK1 Q1 1.85 (0.00584 20 )
0.138
The full nA11 matrix is:

0.13
0.774 L 0
8
0.917

2.959 M
1.832
7.596

56.651
M 13.956
nA11 153.17

0 L 153.17

15.002



Step 5: It is
unlikely that the initial guess will be the correct flow and nodal
head values.
Therefore, the conservation laws (Eq. 5-16 and 5-24 and written
above for the example network) will not be satisfied. To determine the residuals
(dE and dq), the assumed values H(0) and Q(0) are substituted. Since we
developed the flows so that they balance at node, dq equals zero but dE does
not. For pipe 1:

H res.1 H 1 0.00584 (20) 200 198

n 1.85
dE1 K 1 Q1
0.509

The full vector is:

dET = [-0.509, -1.240, -2.549, -0.576, 3.447, 5.371, 9.311, -9.228, 80.794,
62.794, 15.006]T
5-46 CHAPTER FIVE

and

dqT = [0, 0, 0, 0, 0, 0, 0]T

Step 6: The vector computed in step 5 is the right hand side of the system of
equations.

n 11
A A12 Q

A 0
21
The unknowns are the vectors, Hand Q.dq
H dE The coefficient matrix is (with
truncated nA11 terms):

nA
11 A12

0

A21 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0.1 0
0.77 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0
4
0
0 0.92 0 0 0 0 0 0 0 0 1 0 0 0 0 1

0 0 0 2.96 0 0 0 0 0 0 0 0 0 0 0 0
1 0
0 1
0 0 0 1.8 0 0 0 0 0 0 0 0 0 0 1
1
0 0
0 0 0 0 0 7.6 0 0 0 0 0 0 0 0 1 0 0

0 0 0 0 0 56.7 0 0 0 0 0 0 0 0 0 1
0 1
1
0 0 0 0 0 0 0 14.0 0 0 0 0 0 0 1 1 0
0
0 0 0 0 0 0 0 153.2 0 0 0 1 1 0 0 0
0
0
0 0 0 0 0 0 0 0 153.2 0 0 0 1 1 0 0
0 0 0 0 0 0 0 0 0 0 15.0 0 0 1 0 0 0 0
0
1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0
1 1 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0
0 0
0 0 0 0 0 1 0 1 0 1 0 0 0 0 0 0 0

0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0
0 0
0 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0
0
0 0 0 1 0 1 1 0 0 0 0 0 0 0 0 0 0
0 0

Solving for H and Q gives HT = [0.38, 1.30, -0.87, -2.28, -2.22, 2.45,
1.07] and QT = [0.94, 0.42, 0.53, -0.13, 0.67, -0.27, -0.14, 0.67, -0.54, -0.40, -
0.94].
Step 7: The new nodal heads and pipe flows are then computed by Eqs. 5-
45 and 5-46:
 NETWORK HYDRAULICS 5-47

H(1) = H(0) + H(1)

Q(1) = Q(0) + Q(1)

resulting in:

HT = [198.38, 194.30, 194.13, 172.72, 185.78, 192.45, 185.07]

and
QT = [20.94, 9.42, 11.53, 5.87, 6.17, 3.23, 0.36, 1.17, 0.46, 0.60, 7.06]

Step 8: The changes during iteration were large so at least one additional
iteration is needed. Return to step 2. The results for the remaining three
iterations are given in Tables E5-8a and b.
Iteration NodeConvergence
Table E5-8a: 1 2 of 3nodal head
4 with iterations
5 6of the pipe
7
1 198 193 method
195 solution.
175 188 190 184
2 198.38 194.30 194.13 172.72 185.78 192.45 185.07
3 198.34 193.95 196.88 170.71 185.96 192.49 184.26
4 198.33 193.89 197.48 170.66 185.96 192.48 184.20
5 198.33 193.89 197.54 170.66 185.96 192.48 184.20

Table E5-8b: Convergence of pipe flow with iterations of the

pipe method solution.
Iteration Pipe 1 2 3 4 5 6 7 8 9 10 Pump
1 20 9 11 6 5.5 3.5 0.5 0.5 1.0 1.0 8.0
2 20.94 9.42 11.53 5.87 6.17 3.23 0.36 1.17 0.46 0.60 7.06
3 21.21 9.79 11.43 6.04 6.07 3.39 0.36 1.07 0.25 0.54 6.88
4 21.27 9.84 11.43 6.03 6.07 3.39 0.36 1.07 0.19 0.54 6.73
5 21.27 9.85 11.43 6.03 6.07 3.39 0.36 1.07 0.19 0.54 6.73

5.4 FIRE FLOW ANALYSIS

A critical factor in water distribution design is the ability to supply

adequate pressure during extreme conditions such as fires. This design
criterion is based on the fact that fire flow requirements often exceed the
normal domestic, industrial, and other demands imposed on the water system.
Fire flow is defined as the rate of water flow at a specified residual pressure and
for a specified duration that is necessary to control a major fire in a specific
structure (AWWA M31). This localized demand can be estimated as discussed
in Chapters 4 and 7. Pressure requirements vary between 20 and 40 psi
(138 275 kPa) during fire and
5-48 CHAPTER FIVE

normal demand conditions. Minimum pressures are selected based on

local regulations and general guidelines. If the pressures are not
acceptable, the design or operations must be modified to meet the pressure
requirements.
Four approaches can be used to determine if pressure requirements
are satisfied during a fire demand. The first is to define the fire demand at
one or
more nodes and the demands at other nodes and solve the system of
equations for that condition. All hydraulic analysis methods discussed in
this chapter assume that the specified nodal demands are satisfied regardless of
the pressure
that would result in meeting the demand. Although the equations may
converge to a valid mathematical solution, the resulting pressures may not
be feasible from a design perspective or from a cavitation point of view.
The alternative approaches determine the fire flow that can be provided and
insure that the pressure is satisfied at the fire location (Figure 5-5). One
technique is to add an emitter with a large discharge coefficient (e.g., 10,000)
just downstream of the fire node (Fig. 5-5b). The emitter elevation is set to
the desired total head (the actual elevation plus the required pressure
head). The numerical solution determines the maximum flow that can be
supplied with the
nodal head greater than the desired total head and can be compared
to the desired fire supply. The emitter discharge coefficient must be
sufficiently large
so that the computed total head is (nearly) equal to the input junction
elevation (elevation plus pressure requirement). A smaller emitter
coefficient will not give the maximum achievable fire flow.
The third technique is to add a short pipe to the fire flow node that
is connected to a reservoir with total head equal to the elevation plus
desired pressure head at the fire flow node (Figure 5-5c). The short pipe
should have a
negligible head loss (resistance) 1 by using a large diameter1 and low
roughness coefficient.
s p ap The c (flow
f p ainp the pipe
n to sthe
H reservoir
aH c equals
f(H a the
H demand at
qa q f node. This
the junction configuration ) results
q in the same system as the
(5-47)
emitter.
ps p f f Hs f
) ntechnique,
In the fourth
the mathematical relation
H between the flow in the

where c (qpipe
connecting s
/ qand
) n the
, qs reservoir
designates
f head
the can
staticbedemand
expressed in terms
at the of node,
junction the target
ps
pressure and available flow. Under this condition, the fire flow available qa at a
is
the static pressure, qf is the normal fire flow demand, pf is the pressure at the
target
normalpafire
candemand,
be computed
and nfrom (Boulos
is a flow et al, 1997):
exponent that is dependent on the head loss
expression used. The above expression represents the exact analytical solution
of the basic pressure-flow equilibrium relationship and is applicable to
any system of consistent units.
 NETWORK HYDRAULICS5-49

A
(a)
qA
A Hemit = zA + pmin /

(b) q A,fire

qA
Hres = zA + pmin /
A
q A,fire
(c)

Figure 5-5: Representations to determine fire flow capacity of a system at

node A (a). An emitter is added adjacent to the original node (b) and the
fire flow is passed through the emitter. Alternatively, a reservoir is
attached by a short pipe (c) and flow to the reservoir is the maximum
fire flow that can be satisfied while meeting the pressure requirements at
node
A. In both cases, the total head at the downstream location is set equal to
node As elevation plus the desired pressure head.

Care must be taken when analyzing the results from these configurations.
The additional flow coming to the fire flow node causes higher pipe flows (and
head losses) throughout the system. Although the fire locations pressure head
is maintained by the emitter or reservoir, pressures at other nodes may not be
acceptable. For example, if the fire is near the source, downstream nodes
at higher elevations may have low (or negative) pressure heads. Thus, the
full system results, not only the fire nodes pressure head, must be
reviewed to determine if the results are satisfactory.

Example 5.9

Problem: Use the emitter approach to determine the maximum flow that
can be provided at node 3 under the average demand condition in the
network considered in Example 5.4. Assume that the required total head at
the node is 290 ft (221 ft node elevation plus 30 psi (69 ft)).

Solution: Table E5-9 lists the emitter flow for a range of emitter coefficients
and the computed total heads as computed by a hydraulic analysis model. Note
5-50 CHAPTER FIVE

that the 4 cfs nodal demand is supplied in addition to the emitter discharge. In
this case, an emitter coefficient over about 50 provides essentially the
same result.
Table E5-9: Node data for various emitter coefficients.
Emitter Total head Emitter discharge Total node withdrawal
coefficient (ft) (cfs) (cfs)

1 292.01 0.93 4.93

10 290.16 2.60 6.60
20 290.04 2.70 6.70
100 290.00 2.73 6.73

5.5 UNSTEADY FLOW CONDITIONS

Steady flow hydraulic modeling provides a snapshot of the conditions in

a distribution system assuming that the hydraulic conditions have
reached equilibrium. In general, however, demands vary over time, pump
operations are altered or pumps may fail in a sudden event perhaps due
to a power failure. These temporal variations cause the pressure and flow
distributions to change and can be modeled by three approaches.
The first and most common approach is known as extended
period simulation (EPS). An EPS is a series of steady state simulations
and is an available option in most hydraulic analysis software. An EPS
begins with an initial set of tank levels, a given demand distribution and
duration and a set of operation decisions. A steady state simulation is
completed for the initial set of demands to determine the pressure and flow
distribution including flow rates into/out of tanks. Using the tank flow rates
and the demand duration, a mass balance calculation is completed to update
the tank levels. The new tank levels are then used as the fixed grade node
elevations for the next steady state hydraulic analysis and time step. The
demands may be changed between time steps. Many hydraulic analysis models
allow operation conditions to be altered based upon the hydraulic condition,
such as a pump being turned on/off as a function of a tanks water level.
The resulting tank flows from a second steady hydraulic analysis are used to
update the tank water levels for the third time period. This process is
repeated until the entire simulation duration is completed.
The time step between tank level changes in an EPS is typically on
the order of hours. This increment is acceptable under most normal
operating conditions, particularly for a well-designed system. However,
demands may change dramatically at shorter time steps, or operation
conditions may be altered rapidly. These changes can cause problems such as
low pressure areas that may result in backflows into the system. More
detailed modeling may be