See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/245327409

Analysis of Pipe Networks by the Finite Element

Method
Article in Water International January 1991
DOI: 10.1080/02508069108686103

CITATIONS

READS

196

3 authors, including:
Abdel Wahid Hago

Isam Mohammed Abdel-Magid

Sultan Qaboos University

University of Dammam

26 PUBLICATIONS 301 CITATIONS

168 PUBLICATIONS 241 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Isam Mohammed Abdel-Magid

Retrieved on: 21 October 2016

Analysis of Pipe Networks by the Finite Element Method1

By
Hassan Abdel Mageid , Abdel Wahid Hago3 And Isam Mohamed Abdel Magid4
2

Abstract
In this paper, the finite element technique used in the analysis and design of water distribution
networks will be presented. The method was applied successfully to the solution of three
different networks.
In order to demonstrate the feasibility of the approach and to show some merits and demerits of
the finite element method, a comparison was made on a digital computer with the standard Hardy
Cross method. In all cases the finite element method was shown to provide superior
performance.
Moreover, the programmes were run on three different machines {Wang vs 100, Apple IIe, and
IBM microcomputer} to show the effect of increasing storage capacity, machine accuracy, and
time saving

Introduction
The finite element method for the analysis of water distribution network has recently received
considerable attention {1,2}. This is almost entirely due to the existence of digital computers,
which make tedious, iterative calculation more amenable to quick solution. In addition, obtaining
a solution to these problems has important economic and design significance, especially in
developing countries.
There are several methods that may be used for the analysis of flow in complex networks,
varying and graphical techniques to the use of electrical analyzers. Also with the wide
distribution of fast computers the hydraulic analysis of pipe network has developed rapidly {35}.
The most widely used method analysis is that developed by Hardy Cross {3}. Although the
method is old and rather unsophisticated it is still a viable method of analysis. Actually, this
method is well suited for solution by hand; and is easily adapted for machine computation {6}.
However, there are several disadvantages to this method such as:
1. A lot of time and tedious work is exhausted in assuming initial flows:
2. Limitations in usage for large flows resulting in convergence problems;
3. An occasional incorrect assumption for direction of flow; and
4. Complications in the method for complex systems such as reservoirs, interior pumps,
valves etc.
Collins and Johnson {1} discussed two iterative procedures for solving a network that satisfied
the node and loop equilibrium. One procedure always converges, although slowly, from any
starting assumption. The other, using the Newton Raphson technique {7} converges rapidly from
a reasonable assumption of flow in each pipe but may not converge at all when initial assumption
of flow in unrealistic.
Jeppson and Davis {8} described the three alternative formulations of the governing equations in
which pipe flows, junction heads or corrective loop flows are the unknowns. They also described

First published in Water International Journal, 16 {1991}, pp 96-101

Civil engineer, Civil Aviation, Khartoum, Sudan
3
Civil Engineering Department, Sultan Qaboos University, P. O. Box 32483, Sultanate of Oman.
4
M. IWRA, Civil Engineering Department, Sultan Qaboos University, P. O. Box 32483, Sultanate of Oman.
2

96

the most commonly used solution methods {Hardy Cross, Newton Raphson, and linear theory}.
Computer programmes based on these methods were also described {5, 9, 10}.
Epp and Fowler {9} have described methods for improving the solution efficiency for some of
the methods when applied to large networks.
Liu {1} presented a Newton-Cross combined procedure that he compared with the Hardy Cross
Method alone, and later, separately with the Newton method. He concluded that the Hardy Cross
method corrects the heads one at a time, while the Newton-Cross combined method computes all
corrections at the same time. The latter is much easier to code than the former.
Wood and Charles {5} used linear network theory, modified to account for the nonlinear head
loss, to solve for the flow distribution in hydraulic networks.
Benjamin {12} presented a method for water distribution network based on three different
aspects of unsteady flow models, all related to, but not limited to, microcomputers. The method
of characteristics {MOC} was utilized as the numerical method.

General Description of the finite element method

Application of the finite element method to a structural problem demands the subdivision of the
structure into a number of discrete elements. Each of these elements must satisfy three conditions
{1}:
1. Equilibrium of forces;
2. Compatibility of strains; and
3. The force displacement relationship specified by the geometric and elastic properties of the
discrete element
An equivalent set of conditions for a pipe network exist; hence, the ability to draw the analogy:
1. The algebraic sum of the flows at any joint or node must be zero.
2. The value of the piezometric head at a joint or node is the same for all pipes connected to
that joint; and
3. The flow-head relationship {such as Darcy-Weisbach or Hazen-Williams} must be
satisfied for each element or pipe.
For Direct application of the finite element method involving a matrix solution, a linear
relationship is required to define the element or pipe. See Fig. 1.

Qj

Li

Qk

Figure (1) Pipe Element

Hence there is a relationship of the form:
q=ch
(1)
In which q = flow; h = head loss and c = the hydraulic properties of the pipe (to be assumed).
The solution technique can be subdivided into three steps:
1. An initial value of the pipe coefficient. c, is selected for each pipe and is then combined
to yield the system matrix coefficient {C}. The system matrix is then solved for the value
of piezometric head at each joint.
2. The individual pipe flows, q, are computed by means of Eq. (1) using the difference
between the determined piezometric heads. These flows are then substituted in the DarcyWeisbach equation to calculate the pipe head losses. If the pipe head losses obtained from
97

the Darcy-Weisbach equation correspond to those obtained from the matrix solution, then
the unique solution, satisfying both the Darcy-Weisbach equation and the linear equation
(1) has been found.
3. If there is a difference between the values of head loss calculated by the two methods, the
values of c are changed to cause the problem to converge to a solution.

Analysis of Pipe Network

Figure 2 is used to show the application of the finite element method. Nodes and pipes are
numbered for identification purposes. As a sign convention, any external flow at a joint will be
positive when fluid is input and negative when actual consumption occurs. Applying the
equilibrium flow criteria at each node then:
Q1 = q1 + q2
Q2 = q1 + q3 + q4 + q5
Q3 = q2 + q3 + q6 + q8
(2)
Q4 = q4 + q6 + q7 + q9
Q5 = q5 + q7 + q8
Q6 = q9
Q
6

6
2

q4

q9

4
q1
Q1

q3

q2

q7
q6

q5

q8
3

Q5

Figure (2) Pipe Network

But the pipe flow in individual elements are given by:
q1 = c1 (H1 H2)
q2 = c2 (H1 H3)
q3 = c3 (H2 H3)
q4 = c4 (H2 H4)
q5 = c5 (H5 H2)
(3)
q6 = c6 (H3 H4)
q7 = c7 (H4 H5)
q8 = c8 (H3 H5)
q9 = c9 (H4 H6).
Applying the sign convention on the flow in the element, it will be noted that while the flow at
one end, j. of the element will be input (i.e positive) it is an output (i.e negative) at the other end,
k, of the element. It follows that, if the flow in element m, whose ends are nodes j and k, is given
b y qm = cm (Hj Hk) at end j, the same flow when considering end k will be qm = cm (Hk- Hj).
Thus, substituting Eqs. 3 in Eqs.2, observing the sign concention, and writing in matrix form we
get:

98

 Q1 c1 c 2
Q c
1
2
Q 3 c 2

Q 4 0
Q5 0

Q6 0

c1

c3 c 4 c5
c3

c2

c3

c4

c5

c6

c8

c3 c 6 c8

c3 c 4 c5 c7

c4

c6

c5

c8

c7

c9

0 H1
0 H 2
0 H 3

c9 H 4
0 H 5

c 9 H 6

In compact form
Q=CH
(5)
In which C = the network characteristics matrix; H = the network head vector, and Q = the
network consumption vector.

Implementation of the Finite Element Procedure

Data input
The input data consists of the following:
a) The number of nodes, elements, known consumption node, known head nodes, and fluid
properties:
b) The element number, its diameter, its length, and the number of its first and second node
connectivity: and
c) The node number and its known head or consumption.

Computation of Pipe Coefficients

The Darcy-Weisbach equation states the relationship between the head loss and the flow in a
pipe as:
fL 2
q C1 q 2
(6)
h1
2
2
gDA

fL
With C1
(7)
2gDA2
in which h1 = head loss (m); f = friction factor; L = length of pipe (m); D = diameter of pipe (m);
A = cross sectional area of pipe (m2) ; g = acceleration of gravity (ms2); and q = flow discharge
(m3/s). To calculate the friction factor, f, the following expression was used {5,13}:
0.134
8
f 0.094 K 0.225 0.53 K 88.0 K 0.44 Re1.62k
In which K = e/D, e = roughness coefficient, and Re = Reynolds number
The initial value of pipe coefficient, Co, was chosen to correspond to Re = 200,000 in each pipe,
qD
a typical value for practical problems {1}. Accordingly, and since Re
, the initial flow in
A
the pipe becomes:
A
qo 200000
(9)
D
in which A = area of the pipe (m2), D = diameter (m), and = kinematic viscosity of the fluid
(m2/s).
The value of head loss, h, corresponding to the flow qo can be calculated from Eq.6. The pipe
coefficient is then found from Equation 1 as
q
co o
(10)
ho
99

This initial value of the pipe coefficient, co, for each pipe was then combined, according to the
geometry of the network, to obtain the initial network characteristic matrix, [Ci]o.

Assembly of the Matrix

Using a standard finite element procedure, the assembly process to form the global matrix can be
summarized. Starting with a zero matrix, the following operations are performed for each
element:
1. Add coefficient Ci to position (k, k) and (j, j); and
2. Add coefficient Ci to position (k, j) and (j, k), according to the connectivity table. Once
all elements are considered the [c] matrix is assembled.

Boundary Conditions
Before the total system of equations can be solved, it is necessary to introduce proper boundary
conditions for at least some of the nodes of the network. The two possible types of boundary
conditions involve specifying either head or consumption for any given node. The introduction
of the boundary conditions for prescribed heads can be implemented by performing the
following steps:
1. Add the contribution of the prescribed unknown Hj to the vector of nodal consumption
[Q].
2. Zero the jth column and jth row of the matrix [C].
3. Make the (j,j) coefficient of matrix [C] equal to 1.
4. Make the jth coefficient of the vector nodal consumption equal to Hj
A boundary condition of the second type, where the discharge rather than the head is prescribed,
is handled by simply placing the value of the prescribed discharge or consumption in the proper
position in the vector of nodal consumption.
Solution of the System of equations
The solution of the simultaneous algebraic equations was done using the Gauss elimination
technique {7}. The solution of the system of equations could provide the values of the previously
unknown nodal heads. With these, discharges could be computed for every element using
Equation 1.

Adjusting the value of c {1}

During the checking procedure, the flow qc, for each pipe calculated via Equation 1, and the
matrix solution were used to determine the head loss, hc, from the Darcy-Weisbach equation. The
first procedure used in the development of the programme was to obtain the correction of the c
value for each pipe by assuming that the point (hc, qc) was the unique solution. Thus, the correct
linear relationship was defined by the straight line joining this point to the origin and is defined
by the equation.
q
(11)
H c q
hc
q
The new value of c was then set equal to c . When all pipe coefficients were corrected in a
hc
similar way the flow distribution obviously was altered, and this method can be an overcorrection when the matrix was resolved. To dampen this over-correction effect, an averaging
technique was introduced. The corrected vale of c is taken to be the mean of the c value defined
by Equation 11 and the value of c used to obtain the matrix solution.

100

Application to Some Problems

In order to demonstrate the feasibility of the finite element approach, two different computer
programmes were developed and applied to three example networks.
The programme FEMNET, developed in this study, which was based on the finite element
technique, was applied to three water supply networks. Then to the same networks, the
programme HRDYX {also developed in this study}, based on the Hardy Cross technique of
balancing flow was also applied for comparison and checking of results.
The two programmes were run on three different machines {apple IIe, Wang vs 100, IBM microcomputer} to see the effect of storage capacity, time saving and to test machine accuracy.
Application to Network A
This simple network has been taken from Civil Engineering Handbook {15}. The network
consists of 10 elements and 8 nodes and is shown in Fig. 3 Input and withdrawals at the nodes
are shown there, while pipe characteristics are given in Table 1. The water level in the tank at
node 1 is 300 ft. The network was solved using the HARDYX programme in 6 iterations. When
the FEMNET programme was applied to the network the same solution was achieved in 16
iterations.
Q 22
1
5
1
2
7
3
3

9 0.44
5

2
4
8
6

Table (1) Pipe Characteristics for 10

Network A
Pipe
1
2
3
4
5
6
7
8
9
10

0.44
6
1.32
7

Diameter, ft
Length., ft
0.50
2000
Figure (3)
0.50Network 2000
0.50
3000
0.50
2500
0.50
5000
0.50
2000
0.50
2500
0.50
1500
0.50
4000
0.50
1500

Application to Network B
This network was taken from Lius's IAHR conference paper {11}. The network consists of 35 elements and 20
nodes, Fig. 4, and a complete listing of the characteristics of each pipe in the network is provided in Table 2. In
applying the HARDYX programme to this example network, the magnitude and direction of flow in each pipe
were assumed. When this programme ran on the Wang vs 100, the solution converged in 95 iterations in 1.23
minutes. On the apple IIE it converged, in the same number of iterations, but in 8 minutes. When the network
was solved using the FEMNET programme, the time and effort spent in assuming the initial flows was
eliminated and the same solution was achieved in only 23 iterations and in approximately 1.5 minutes when
ran on the Wang vs 100. When the programme was run on the Apple IIe, the solution converged in the same
number of iterations but in a longer time. This difference in time is mainly due to storage capacity and
accuracy of the machine used. Also this example indicated the feature of saving time and effort needed for
initial assumption of flows, and the ease of input data for the FEMNET programme over the HADYX
programme.

101

Table (2) Pipe Characteristics for Network B

Pipe
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35

Diameter, ft
5.00
5.00
1.67
2.50
2.50
1.67
2.50
2.50
2.50
2.50
0.83
1.67
3.00
1.67
2.50
1.67
1.67
1.60
2.50
1.67
1.67
3.50
2.50
2.50
2.50
2.00
3.33
4.00
2.50
2.00
1.67
2.50
1.67
2.00
1.67

36.2

6.0
29

17

25.0

Length, ft
5100
5900
3000
5600
4900
3000
4900
3100
6200
10200
6800
3000
5750
3000
3600
3000
5000
3400
6000
4000
2900
4000
3750
3000
4000
3000
7500
5500
5000
4000
3000
6500
4500
4000
3000
17.7
32

18

19

28
24
25
41.2

16
27

15

8.0

23

30

14

22

21

18

34

11

5.0

35
13

9
12

10

7.8
1.9

1.0

20
14

6.0

0.3

17

11

15

2.4

23.5

6.0

8
6

7.3

9.0

12
19

13

1
1
Reservoir
600' HIGH
Q = 214.30 FS

4.0
20

33

31

6
7

10

5
6.0

Figure 4. Network B

102

Application to Network C
This network is in Elobied Town, which lies in West Sudan {14}. This is a more practical
problem, due to its complexity and the increased size of the network. It consists of 137 elements
and 89 nodes and combines an existing network and its new extensions. Full details of the
network can be found in Ref. 14. It was assumed that the pipes were commercial steel, each with
a roughness element size of 0.03mm. When this network was solved using the HARDYX
programme, more than 7 days were spent in making initial assumptions to balance the network
before the solution began. The solution was achieved in 77 iterations.
In applying the FEMNET programme, the properties of the pipes and the configuration of the
network were the only requirements. In contrast with HARDYX the same solution was achieved
in only 17 iterations.
The two programmes were run on the Wang vs 100, Central Processing Unit {CPU}. The time
required for HARDYX was approximately 5 minutes, while that for FEMNET was 3.5 minutes

Conclusions
From the study of the various networks, it can be concluded that the major advantages of the
finite element method over the Hardy Cross method include:
i. The speed of convergence and the apparent lack of convergence problems
ii. No need for an initial guess of the flow distribution
iii. Flexibility in applying the boundary conditions
iv. Ease of modifying and extending the network without interrupting the whole system.
v. The absence of artificial loops
vi. The choice of flow-head loss relationship
vii. Ease of input data
viii.
The ability to account for temperature effects.
ix. The unlimited network size {depending only on computer storage capacity).
Although the effect of pumps, boosters, pressure reducing valves, non-return valves, etc. were
not included in this paper, they can easily be incorporated in the program if their actual head
flow relationships are known. This could be considered in subsequent research papers.
Finally, it can be said that the finite element method requires the formulation and solution of the
entire set of network equations. The large active computer space requirement for these
computations determines the maximum size of network, which can be handled by this method,
and this depends on the computer used.

References
1.
2.

3.

4.

Collins, A.G. & Johson, R.L. "Finite Element Method for Water Distribution Networks "
Journal of American Water Works Association, Vol. 67, No. 7, 1975 pp. 385 -389
Gientke, F.J. "Finite Element Solution for Flow in Noncircular conduits, "Journal of the
Hydraulics Division Proceedings of the American Society of Civil Engineers, Vol. 100, No
HY3. 1974, pp. 425 442
Lam, C.F. & Wolla, M.L., "Computer Analysis of Water Distribution Systems, Part 1
Formulation of Equations, Journal of the Hydraulics Division, Proceeding of the American
Society of Civil Engineers Vol. 98 No. HY2 , 1972, pp. 335-344.
Lam. C.F. & Wolla M. L. "Computer Analysis of Water Distribution Systems. Part II
Numerical Solution, Journal of the Hydraulics Division, Proceeding of the American
Society of Civil Engineers. Vol. 98. No HY3, 1972, pp. 447-460
103

5.

6.

7.
8.

9.

10.

11.
12.

13.
14.
15.

Wood, D.J. & Charles, Co.., "Hydraulic Network Analysis Using Linear Theory, Journal
of the Hydraulics Division, Proceedings of the American Society of Civil Engineers, Vol.
98. No HY7, 1972, pp. 1157-1170
Shamir, U. & Howard, C.D., "water Distribution Systems analysis. Journal of the
Hydraulics Division, Proceedings of the American Society of Civil Engineers, Vol. 94. No
HY1, 1968, pp. 219 -234
Henrici, P., Elements of Numerical Analysis, John Viley & Sons Inc., New York, New
York, U.S.A., 1967
Jeppson, R.W & Davis, A.L., "Pressure Reducing Valves in Pipe Network Analysis,
Journal of the Hydraulics Division, Proceedings of the American Society of Civil Engineers,
Vol. 102, No HY7, 1976, pp. 897 -1001
Epp, R. & Fowler, A.G., Efficient Code for Steady State Flows in Networks, Journal of
the Hydraulics Division, Proceedings of the American Society of Civil Engineers, Vol. 96,
No HY1, 1970, pp. 43 -56
Mc Cormich, M. and Bellay, C. J., "A Computer Programme for the Analysis of Networks
of Pipes and Pumps, Journal of the Hydraulics Division, Proceedings of the American
Society of Civil Engineers, Vol. 38, No 3, 1968, pp. 51 -58
Liu , K.T., "The Numerical Analysis of Water Supply Networks by a Digital Computer,
Proceeding IAHR 13th Congress, Kyoto, Japan, 31 Aug. - 5 Sept. 1969, pp. 35 42
Wylie, E.B., "The Micro Computer and Pipeline Transients, Journal of the Hydraulics
Division, Proceedings of the American Society of Civil Engineers, Vol. 109, No HY12,
1983, pp. 1723 -1739
Wood D.J. "Slurry Flow in Pipe Networks, Journal of the Hydraulics Division, Proceedings
of the American Society of Civil Engineers, Vol. 102, No HY1, 1980, pp. 57 -70
Abdel Magied, H., Analysis and Design of Networks using Finite Element Method, M.Sc.
Thesis, Department of Civil Engineering, University of Khartoum, 1987
Standard Handbook for Civil Engineers, McGraw Hill Inc, New York, New York U.S.A.,
2nd ed., 1976, pp 21-1 to 21-124.

104