Hydraulic Analysis of Unsteady Flow in Pipe Networks

Hydraulic Analysis of Unsteady Flow in
Pipe Networks
Hydraulic Analysis
of Unsteady Flow
in Pipe Networks
J. A. FOX
Reader in Qvil Engineering
University of Leeds
M
J. A. Fox 1977
Softcover reprint of the hardcover 1st edition 1977 978-0-333-19142-2
All rights reserved. No part of this publication may
be reproduced or transmitted, in any form or by
any means, without permission
First published 1977 by

THE MACMILLAN PRESS LTD
London and Basingstoke
Associated companies in New York Dublin
Melbourne Johannesburg and Madras
ISBN 978-1-349-02792-7 ISBN 978-1-349-02790-3 (eBook)

DOI 10.1007/978-1-349-02790-3
This book is sold subject to the standard conditions of the

Net Book Agreement
Text set in 10/11 pt IBM Press Roman

Contents
Preface ix
Notation xi
1 Simple water hammer theory
1.1 Introduction 1
1.2 Rigid pipe-incompressible fluid theory 2
1.3 Sudden valve opening at the downstream end of a
pipeline 4
1.4 Slow valve closure 6
1.5 Distensible pipe-elastic fluid theory 9
1.6 Instantaneous valve closure 10
1.7 Separation 16
1.8 The calculation of the magnitude of the transient
caused by complete instantaneous valve closure at the
end of a simple pipeline 17
1.9 Pressure rise caused by instantaneous valve closure 21
1.10 Sudden valve closure 21
2 Analytic and graphical methods 23

2.1 Introduction 23
2.2 Analytic methods of solution 23
2.3 Stepwise valve closures at pipe period intervals 23
2.4 The Allievi interlocking equations 25
2.5 The Schnyder-Bergeron graphical method 36
3 Boundary conditions for use with graphical

methods 55
3.1 Introduction 55
3.2 Pumps 55
3.3 Four quadrant pump operation 60
3.4 Surge tanks 62
3.5 Types of surge tanks 63
3.6 Transient analysis of surge tanks 65
3.7 Mass oscillation of surge tanks 66
3.8 Pressurised surge tanks or air vessels 68
3.9 Methods of integrating the surge tank equations 70
v
vi Contents
4 The method of characteristics 72
4.1 Introduction 72
4.2 Method of deriving the characteristic forms of the water-
hammer equations 74
4.3 The characteristic forms of the waterhammer equations 77
4.4 The zone of influence and the domain of dependency 78
4.5 The zone of quiet 79
4.6 The integration of the characteristic equations 79
4.7 Boundary conditions 81
4.8 The method of the regular rectangular grid 82
4.9 Other finite difference methods 85
5 Variable parameters in unsteady flow 87

5.1 Variation of wavespeed 87
5.2 Gas evolution 88
5.3 The magnitude of variable wave speed and the inclusion
of gas release 90
5.4 The use of the variable wavespeed equation 94
5.5 Vaporous cavitation 94
5.6 Calculation of friction 95
5.7 The use of variable f values 96
5.8 Interpolation 97
5.9 The calculation of the free bubble content 98
5.10 Evaluation of velocities and potential heads at internal
points in a pipe length 99
6 Boundary conditions: pumps 100

6.1 Introduction 100
6.2 Pumps equipped with a nonreturn valve 100
6.3 The derivation of the pump's characteristic equation 101
6.4 Dynamometer/turbine operation of a pump with
forward flow 102
6.5 Pump efficiency 105
6.6 Pump power 105
6.7 Pump start up 106
6.8 Pump run down 107
6.9 The in-line pump boundary condition 108
6.10 Suction well pumps 110
6.11 Four quadrant pump operation Ill
6.12 The use of the Suter curves 118
6.13 Pump run up to steady speed of pumping 120
6.14 Pumps with by-pass valves 120
6.15 Pump stations 121
6.16 Surge suppression of transients generated by pump trip 124
6.17 Line pack and attenuation 127
6.18 Lock in 128
Contents vii
7 Other boundary conditions 130

7.1 Junctions 130
7.2 Joints 132
7.3 Air vessels 134
7.4 The motorised valve 136
7.5 Servocontrolled valves 142
7.6 Reservoirs 144
7.7 Bends 146
8 Unsteady flow in gas networks 147

8.2 Basic equations 147
8.3 Characteristic equations 149
8.4 The value of 1 150
9 Impedance methods of pipeline analysis 155

9.1 In traduction 155
9.2 The analogy between electrical and hydraulic impedance 156
9.3 The linearisation of the waterhammer equations 157
9.4 The solution of the linearised waterhammer equations 160
9.5 The evaluation of -y 162
9.6 The impedance concept 164
9.7 Receiving and sending ends 165
9.8 The equation of impedance 165
9.10 The impedance of a network 170
9.11 Harmonic analysis 172
9.12 The forcing oscillation 173
9.13 The oscillating valve 174
9.14 A network in which resonance can be excited by forcing
oscillations located at different points in the network 176
10 Unsteady flow in open channels 177

10.2 The equations of unsteady flow in open channels 178
10.3 The characteristic forms of the open channel equations 181
10.4 The travelling surge 192
10.5 The profile of a free surface flow when a travelling
surge is present 195
10.6 The method of analysis of an unsteady free surface
flow in which travelling surges are present 195
10.7 Other methods of analysis 199
viii Contents
11 Global programming 201
11.2 The route or link method of global programming 201
11.3 Pipe description 203
11.4 Longitudinal profiles 203
11.5 Upstream reservoirs 204
11.6 Downstream reservoirs 204
11.7 Pump description 204
11.8 Pipe longitudinal profiles at & intervals 205
11.9 Calls of procedures 205
11.10 Time level scanning 206
References 207
Bibliography 209
Index 211
Preface
The reader may be interested to know how this book came to be written.
The author has always found the subject of unsteady flow of great interest
and throughout his career has studied it with special application. As a con-
sequence most of his research effort has been in this area and he has guided
many of his Ph.D. students into this topic also.
In 1969 an engineer from a local Consulting Engineer's office approached
him requesting information concerning surge analysis methods which could
take into account variations in wavespeeds caused by free bubbles in the
fluid. At that moment in time the author had already developed a computer
program which could analyse surge in simple rising mains but had not in-
cluded this wavespeed effect. The effect was soon incorporated into the
program and it was used to analyse a main which had a history of bursting
to decide what was the main cause of the bursts. At the same time, un-
known to the author, one of his own ex-Ph.D. students had been employed
to take measurements of the pressure history of the main. When the analy-
tic results were compared with the measured results it was found that agree-
ment was extremely good, the only error of significance being in the timing
of the pressure peaks. The actual magnitudes of the maximum and mini-
mum pressures were excellently predicted.
Upon seeing these results the author and the engineer from the Consult-
ing Engineer's office, Bryan Smith, decided to open an office, Hydraulic
Analysis Ltd, Leeds, which would routinely undertake the analysis of pro-
posed or existing systems. This venture turned out quite successfully and
with the passage of time the firm has been called upon to analyse more
and more complex systems, ranging from simple rising mains delivering
sewage to a sewage works, to undersea oil pipelines such as that of the For-
ties field in the North Sea. The firm has been called in to analyse water
supply networks for various authorities throughout the world, oil pipelines
in the Middle East, most of the pipelines built or proposed for the North
Sea, water injection schemes to improve oil delivery from underground
strata, pipe networks in Condeeps and other complex networks such as
those found in oil refineries and gas liquefaction plants. The firm has had
to grow to handle this work, taking in a computer specialist, Andrew Keech,
ix
X Preface
as a partner and employing more staff.
Throughout this period it has been necessary to develop the original pro-
gram and now it has reached a considerable level of sophistication.
As an academic, the author feels that the essential material in this pro-
gram should be published and so he decided to write this book. It is not
possible to include within the confines of one book all of the material that
has gone into the program; there are many facilities which have not been
included but the main material on which the program is based has been
described.
The author would like to warn the reader that he has not tried to write
the definitive book of waterhammer. It probably could not be written at
present as the subject is still undergoing rapid development. Even so, this
book is an idiosyncratic view of waterhammer and many people who have
contributed greatly to the subject may feel slighted by the omission of
their material or by the failure even to mention its existence. The author
would like to apologise to such people and would plead, in advance, the
limitations of space.
The book is idiosyncratic in other ways, techniques of finite difference
integration such as those due to Lax, Wendroff and coauthors have only
been given passing mention and no mention at all has been made of what
the author believes is a potential technique for the future - the finite ele-
ment method. However, he has demonstrated, to his own satisfaction at
least, the complete adequacy of the method of characteristics and offers
this as partial justification for his limited presentation of a very large,
very complex subject.
Leeds, 1976 J.A. F.

Notation
Throughout this book, symbols are defined wherever they are used and
these are listed below. However, variables of local interests only are not
included in this list but are defined in the text.
A area of flow
A0 area of valve opening at time zero (chapter 1)
At area of valve opening at time t (chapter 1)
ap area of pipe (chapter 1)
av area of valve opening
A pump constant in equation H = AN 2 + BNQ- CQ 2 (chapter 6)
Ae exit area of pump impeller (chapter 6)
Asw plan area of a suction well (chapter 6)
ae effective valve area (chapter 9)
B pump constant in equation H = AN 2 + BNQ- CQ 2 (chapter 6)

B channel surface breadth (chapter I O)
b mean channel breadth (chapter 10)
Cct coefficient of discharge of a valve

c celerity of a small pressure wave
C constant in friction formula used in the surge tank analysis
(chapter 3)
C constant in pump equation H =AN 2 + BNQ- CQ 2 (chapter 6)
c5 coefficient in the stroke equation of a servocontrolled valve
(chapter 7)
Cv specific heat of gas at constant volume (chapter 8)
Cp specific heat of a gas at constant pressure (chapter 8)
C electrical capacitance/unit length of a transmission line (chapter 9)
c celerity of a small surface wave (chapter 10)
C Chezy C (chapter I 0)
Cct coefficient of discharge of a sluice gate (chapter I 0)
Cw celerity of a surge wave in an open channel (chapter I 0)
xi
xii Notation
dt time increment (infinitesimal)
dx distance increment (infinitesimal)
dp pressure increment (infinitesimal)
dp density increment (infinitesimal)
dv velocity increment (infinitesimal)
d pipe diameter
D pump impeller diameter
De constant in the pump efficiency equation
d sw depth in a suction well
d1 internal diameter of an air vessel
E Young modulus of elasticity

f: pump efficiency (chapter 3)
Ec constant in the pump efficiency equation (chapter 3)
e internal energy of gas/unit mass (chapter 8)
E1 \ constants in the characteristic forms of the unsteady gas equations

~:: J as defined in text (chapter 8)
E g(j-i) (chapter 10)
f Darcy fin the Darcy formula
hf=~y
_gm
2
(as defined in text)
fh hoop stress in pipe wall

f 'function or and wave height when wave is travelling downstream
(chapter 2)
F 'functiOit of' and wave height when wave is travelling upstream
(chapter 2)
Fe constant in the pump efficiency equation
F in chapter 8, the force acting upon the fluid/unit length of pipe
f frequency of applied head oscillation (chapter 9)
Fn Froude number (based on absolute velocity) (chapter 10)
g intensity of the local gravitational field throughout the text

Gr gradient of the pump's speed ~ time rundown curve (chapter 6)
h potential head - the sum of local pressure head and elevation of

the point above an arbitrary datum.
hr head lost due to friction
hs sometimes static head. sometimes head at pointS, according to
context
hn head immediately upstream of a valve or nozzle.
hi potential head change caused by momentum changes (note
Pi= whi)
Notation xiii
hair pressure head of air in an air vessel expressed as height of the
equivalent liquid column (chapter 3)
h3 atmospheric pressure head
hp height of the base of an air vessel above the pipe centre line
hw as hair above (chapter 7)
htr head sensed by a pressure transducer controlling a servo-operated
valve
hcnt critical head at which a servo-operated valve will start to move.
h steady state head (chapter 9)
h' unsteady head component (chapter 9)
H amplitude of pressure head wave (chapter 9)
hsp height of the reservoir surface above the spillway crest (chapter 10)
hw the height of a wave crest above channel bed level (chapter I 0)
I inertia of the rotating parts of a pump and motor set (chapter 6)

electrical current (chapter 9)
v'=J(chapter 9) (see context)
channel bed slope - taken positively downwards (chapter 10)
j frictional head loss/unit weight of fluid/unit length of channel

(chapter I 0)
v2
K constant in valve loss formula hr = K 2g
4[L
K =d + k (chapter 1 and chapter 5)
K bulk modulus of liquid

k constant, sometimes describing local losses,
2
i.e. ht = k ~g (due to bends. junctions etc)
In chapter I also used: k = voLh

gT s
k constant in head ~ q equation h = kq 2 ,

i.e. the friction formula used in the Schnyder-Bergeron method
(chapter 5)
k mean height of pipe roughnesses in the Colebrook-White formula
kr pump impeller head loss coefficient (chapter 6)
kv pump volute head loss coefficient (chapter 6)
k spillway constant (chapter I 0)
L length, usually pipe length L

L constant in the development of the characteristic form of the
differential equations of waterhammer (chapter 7)
/1 internal height of an air vessel (chapter 7)
L electrical inductance/unit length of a transmission line (chapter 9)
xiv Notation
m hydraulic mean radius =~(defined in text)

n area ratio~: (chapter I)
N running speed of a pump in rev rnin- 1 (chapter 3 and 6)
n polytropic index (chapter 8)
Pi pressure generated by momentum change

p wetted perimeter
p pump power in chapter 3 and chapter 6
Pair pressure of air in an air vessel (chapter 3)
Pa atmospheric pressure
Pwr pump power in chapter 6
q flow
qt flow at time t
q heat flow/unit area (chapter 8)
q steady flow rate (chapter 9)
q unsteady flow component (chapter 9)
Q amplitude of flow oscillation (chapter 9)
pvd vd
Re Reynolds number = - or-
IJ. ZJ
R universal gas constant (chapter 8)
R hydraulic resistance/unit length (chapter 9)
Rei electrical resistance/unit length of a transmission line (chapter 9)
Si valve stroke at beginning of a At period (chapter 7)

Sf valve stroke at end of a At period (chapter 7)
Sreq valve stroke required as defined by a pressure transducer
(chapter 7)
s an integer taking the value of+ I or -1 (chapters 7 and 10)
T pipe period 2L (chapters 2 and 3)

c
T pipe wall thickness (defined in text)
T torque applied in the pump equation (chapter 3)
T absolute temperature (chapter 8)
time
u velocity of impeller blade tips (chapter 6)
v mean flow velocity (chapter I)

V= velocity in pipe when t---* oo (chapter I)
v0 velocity at time zero (chapter I)
Vr velocity at time t (chapter I)
Vair air volume in an air vessel (chapter 3)
Vg volume of dissolved gas
Notation XV
velocity of whirl at exit from a pump impeller (chapter 6)

relative velocity at exit from a pump impeller (chapter 6)
absolute velocity at exit from a pump impeller (chapter 6)
velocity of flow at exit from a pump impeller (Chapter 6)
electrical voltage (chapter 9)
velocity of a surge wave in an open channel (Chapter I 0)
weight density of fluid

dimensionless heat parameter in the head Suter curve (chapter 6)
dimensionless torque parameter in the torque Suter curve (chap-
ter 6)
X distance along pipeline

X distance along pipeline
z elevation of the pipe centre line above datum

z elevation of surface in a surge tank above reservoir static surface
level (chapter 3)
elevation above a datum of a suction well base (Chapter 6)
the elevation above datum of the centre line of the pipeline at the
point of its junction with an air vessel
z hydraulic impedance (chapter 9)
Zc characteristic impedance (chapter 9)
z depth of the centroid of a channel cross section (chapter 10)
Q real component of 'Y (propagation constant) (chapter 9)

Q constant defining nature of a channel cross section (chapter 10)
CctA
the product ~y2g (chapter 2)
p
imaginary component of 'Y (propagation constant) (chapter 9)
pump blade angle (chapter 6)
propagation constant in chapter 9
ratio of specific heat of gas (chapter 8)
ratio of channel cross sectional area/surface breadth (chapter 10)
distance increment (finite)
time increment (finite)
volume increment (with subscripts to define which volume change
is intended) (chapter I)
pressure increment due to momentum change (finite) (chapter I)
potential head change: related to l:lpi by l:lpi = wl:lhi (chapter I)
fractional volume of free gas in liquid (chapter 5)
ratio of aefae, (chapter 9)
steady component of e (chapter 9)
xvi Notation
,
unsteady component of (chapter 9)
the square root of the head ratio: \ho
I h)OS (chapter 2)
the slope of the characteristic line
av
1/ fractional valve opening: av. (chapter 2)
e angle in Suter presentation of four quadrant pump characteristic
B=tan- 1 (~ gs)(chapter6)
constant in the characteristic formulation of the waterhammer equa
tions (chapter 4)
;>.. cxs
d aB
ax (chapter 10)
J1 dynamic viscosity of fluid
v kinematic viscosity of fluid =!:!.
p
7T 3.14159
p mass density
. . ha
All leVI . . CVo
p c ractenstlc: 2 gho (chapter 2)
T surface tension coefficient in chapter 5

T viscous sheer stress in chapter 2 and chapter 10
X the compound line produced by the summing of two eagres in
the Schnyder- Bergeron method (chapter 2)
l/1 the valve characteristic in the Schnyder - Bergeron method
(chapter 2)
l/1 phase angle (chapter 9)
D. angular velocity of pump impeller (chapter 3 and chapter 6)
D. angular velocity of the applied head oscillation (chapter 9)
1 Simple waterhammer theory
1.1 Introduction
The hydraulic analysis of flow in networks is usually based upon the con-
sideration of steady state conditions. This is due to historic reasons; the
analysis of unsteady state is an order of magnitude more difficult than
that of steady state and was only possible at all if grossly simplifying as-
sumptions were made. Until the relatively recent development of com-
puters the only methods available were graphical in type and these could
only be applied to simple networks in which the hydraulic controls were
of an elementary nature and in which the number of pipes was small.
Now that computers are available, a very great improvement has been
made to the quality of analytic techniques that can be used and it is no
longer necessary to confine the mathematical modelling of a network to
that of steady state. The analysis of unsteady state can include steady
state as a special case but it yields much more information than this. The
behaviour of the system during starting, its run up to steady state and
the transient phase that occurs after shut down can all be described with
considerable accuracy. It is usually found that the conditions occurring
during steady state operation are of only passing interest, what happens
during the starting and shut down phases being of much greater impor-
tance. The operation of complex hydraulic controls can be simulated and
the only limitation upon the size and complexity of the network is the
size of the computer store.
Waterhammer is the name commonly used for pressure transients. The
reason for the name is that when a steep pressure wave front passes through
a pipe it generates a sound that resembles the noise that occurs when a
pipe is struck by a hammer. In fact, all transients do not generate sound
but the name has gained such a wide acceptance that there is no point in
trying to change it. Wherever the word waterhammer is used in this text
it should be understood to include all pressure transients even if they are
not sufficiently steep fronted to cause noise.
In the usual Newtonian approach to the analysis of the motion of a
body it is usually assumed that a force causes an acceleration which is
2
2 Hydraulic analysis of unsteady flow in pipe networks
simultaneously applied to all particles within the body. In fact, when a
force is applied to a body those particles at the point of application of the
force are immediately accelerated. The movement of these particles rela-
tive to adjacent particles causes forces to be applied to them and they in
turn are accelerated. The process then operates upon the next layer of ad-
jacent particles and these are accelerated. Eventually all the particles in the
body will be accelerated. In effect, a wave of compressive stress has passed
through the body and this wave will have propagated at a speed that is
usually large but not infinite. Most bodies are not sufficiently long in the
direction of application of the force for the wave passage time to be in any
way significant but the effect is always present.
In the case of a long pipeline containing a fluid, the passage of a com-
pressional wave through the fluid can take a significant time and the pres-
sures caused by the compressional wave may be sufficiently large to burst
the pipe. In such a case an analysis which did not include the effect of such
transient behaviour would be of little value. However, if the pipeline were
short, and a pressure fluctuation were applied at one end of it over a time
which was much greater than the time taken for the compressional wave
to traverse the pipe, then an approach which assumed that all fluid particles
were being simultaneously accelerated would represent a reasonably accu-
rate model of the fluid's behaviour.
Two ways of predicting the behaviour of a fluid column when under the
action of a force are thus available:
(I) Rigid column theory which considers the entire fluid column to be
accelerating at the same value throughout its length, the wavespeed being
infinitely large.
(2) Elastic theory which considers any pressure change to be transmitted
through the fluid column at a large, but finite wavespeed.
Rigid column theory can only be used if the time of operation of the hy-
draulic control is considerably greater than the time taken for a wave of
pressure to pass through the fluid column. Elastic theory can always be
applied and gives more accurate results but it is usually more complex in
nature.
1.2 Rigid pipe-incompressible fluid theory
Historically, the development of waterhammer theory has followed a pat-

tern of increasing complexity. It started by using the solid body type analy-
sis which is now called 'rigid pipe-incompressible fluid' theory. Later, 'disten-
sible pipe-compressible fluid' theories were developed. The second category
of theories has been the basis of most of the work performed recently and
can now be considered to be in a high state of development. Rigid column
theory can be of considerable value, as situations arise in which pressure
transients are not of great interest but in which fluid movements are impor-
tant. Rigid column theory is capable of describing such motions moderately
accurately.
Simple waterhammer theory 3
A simplified form of the dynamic equation
First, one of the fundamental equations of waterhammer must be deve-
loped. Consider flow through a pipe of length L experiencing a pressure
gradient ~~which is decelerating the fluid.
Note Pressures are assumed to increase in the direction of x increasing. The
velocity vat time tis assumed to be the same at all points in the pipeline.
The fluid mass contained in the elemental length Ax is pA ~x.
The force decelerating the fluid is A ;~ Ax neglecting friction.
instantaneous
pressure grade
line
p
8x
Figure 1.1
By Newton's second Law of Motion

ap dv
A ax Ax + pA ~x dt = 0
so
ap dv
-+p-=0 (1.1)
ax dt
This is an extremely simplified version of the Euler equation
If dv is constant throughout the pipe length, and it is if the pipe is
dt
rigid and the fluid incompressible, the equation can be integrated to give
dv
~p=-pL (1.2)
dt
where ~p is the pressure difference over the pipe length L which it is neces-
sary to produce to generate the acceleration:~
Note that if the downstream pressure exceeds the upstream pressure by the
amount ~p then~~ will be negative, i.e. a deceleration.
As pressures and heads are related by the expression
p = wh = pgh
this result can be cast into the form
L dv
t:.h=--- (1.2a)
g dt
This solution is valid for frictionless flow but requires modification if
frictional effects are to be included.
1.3 Sudden valve opening at the downstream end of a pipeline
As an illustration of how the above simple theory can be used and how
frictional effects can be taken into account the case of a sudden down-
stre'lm valve opening will next be examined. It suffers from the defects of
all rigid pipe-incompressible fluid theories.
reservoir
lenglh L
diameter d
Figure 1.2
!::.hi in equation I .2a is the excess of head at the downstream end of

the pipe over that at the upstream end. At the downstream end the head
is atmospheric as the valve is full open (treated as zero head here) while
upstream it is hs so !::.hi = 0- hs if friction were not present, but as friction
is present hs must be reduced by the amount of hr so
( 1.3)
where hr is the friction head.
4fLv 2
Now hr = 2gd (I .4)
This is the Darcy-Weisbach equation (Fanning equation in USA); the fused

is one quarter that used in the USA version of the Darcy equation but is
the same as that used in the Fanning equation.
Therefore if local losses are also included
!::.hi=_ (h _ 4fLv 2 _ kv 2 ) = _f:dv

s 2gd 2g gdt
where!';; is the head losses caused by local losses such as those occurring
at bends, junctions etc,
h _ 4fL ~ _ kv2_!:_ dv
so s d 2g 2g- g dt
v2 Ldv (1.5)
or h -K-=--
s 2g g dt
where
dv 2gh 5 - Kv 2
- = -- ----
dt 2L
so
dt = '"lL(' _ _
d v___ _ )
- 2gh 5 - Kv
2
t = -J 2Lglz/( loge
( V2ifi s + -.JKV)
v'2ifis _ .JKij (1.6)
where vis the velocity at time t.

When t becomes int1nite V2ilzs V~ =YR.
where V~ denotes the asymptotic velocity as t tends to infinity
so V =
~~K
fiihs
a standard result from normal steady state theory as would be expected.
Rearranging equation 1.6
so t = y2g~sK loge ( :t~:)

. V ~ ~_v_ = e(2ghsK)o.s t/L
.. v~ -v
Rearranging
v
~ = e(2ghsK)o.st/L + 1 (1.7)
This can be plotted as shown in figure 1.3.

The dotted line shows how the velocity curve is affected by the elasti-
city of fluid and the elasticity of the pipeline wall material and demon-
strates how the rigid pipe-incompressible fluid theory predicts the mean
velocity changes. It is a commonplace experience that when a tap is sud-
denly opened the flow is very fast for an instant then drops and rises to a
steady state which is less than that of the first spurt. The high velocity
v
~ f\--- actual velocity history
t-~------
1 theoretical velocity history
(2qh, K )0 "5 /L
Figure 1.3
from the first spurt is approximately equal to the spouting velocity

Vfihs and its rapid fall off is due to the inability of the strain energy of
the fluid and pipe wall material to maintain this value for very long. A
more accurate solution of this problem is described in chapter 2.
1.4 Slow valve closure
A further example of the use of rigid pipe-incompressible fluid theory

is that of slow valve closure.
Assume that the general equation for the effective valve area is:
ae =aof(t) (1.8)
where C4J is the full open valve area and f(t) is some function of time. Effec-
tive valve area means the actual area multiplied by the coefficient of dis-
charge.
Also assume that the Bernoulli equation can be applied to the flow
through the valve even though it is in an unsteady state. This assumption
is always used and it has been experimentally justified on many occasions.
Then Qt = ae~ {1.9)
where Qt is the flow through the valve at time t and hn is the head immedi-
ately upstream of it at this time.
Therefore,
where A is the area of the pipe, so by differentiation with respect to
time dvt=dae ~+.!v'li(hro.saedhn

dt dt A 2 A dt
now ho _h _ 4 fLvl = L dvt

n 2gd g dt
where h 0 is the head in the supply reservoir.
This equation is derived from equation 1.2a. Substituting for ~Vt and
rearranging gives t
dhn = nVfihn (ho _

h _ 4[Lvt 2 ) _ 2h d (f(t)) (1.10)
dt L[(t) n 2gd f(t) dt
A
where n =-
ao
This equation can be integrated by finite difference methods. An esti-
mate of the maximum head can be made however. Multiply equation 1.10
by f(t). This gives
(fi( t)) dhn = n~ (h 0 _ h _ 4[Lvt )

2
_ 2 z. d (f(t))
dt L n 2gd "n dt
If the maximum head occurs at the time when the valve closes then
Vt = 0 and the frictional term will then have no effect. If the valve closure
is of a type which generates maximum head prior to the moment of incomp-
lete valve closure then Vt will not be zero at the instant when dd~n = 0.
If friction is important, a finite difference integration of equation 1.10 will
have to be undertaken to obtain the maximum head but if friction can be
neglected the following analysis can be used to obtain it.
If a maximum head, in the mathematical sense, occurs anywhere within
the closure period [(t) ~~will equal zero. If the head rises throughout the
valve's closure and reaches its largest value at the instant of closure without
producing a turning point then the expression f(t) d:r" will be zero because
the function[(t) must be zero when the valve is closed. So, irrespective of
whether a mathematically maximum value is produced at some time within
the valve closure, or a largest value at the end of the closure is produced,
setting the expression f(t) ~~n to zero will define the largest value.
0 = nVfihmax (ho- hmax ) - 2hmax -d (/(t ))

L dt
n2
2 2ghmax (ho- hmax) 2 = 4hinax - (f(t))
(d )2
L dt
Rearranging gives
hmax
( --
)2 - (2 + 2 L :t (f(t))\l)
~ h max + I = 0
~ ~2~ ~
Let
Then (1.11)
As an example, consider a valve closure in which the valve flow area reduces
linearly with time, i.e. ae =a0 (1 - ~) where Tis the time of valve closure.
t
Then f(t)=1--
T
d 1
and - (f(t)) = - -
dt T
Then
but 2g~o = v~ where v0 is the steady state velocity in the pipeline before
n
the commencement of valve closure.
Then k=-
voL
ghoT
so it is simple to evaluate hmax/h0 from equation 1.11.

Simple finite difference techniques can be used to integrate equation
1.10 to obtain the curve of hn ~ t if this should be required.
Consider the linear valve closure mentioned above in which the valve
area ae =a0 (I - t/T) and ignore friction:
when t =0, hn =hs
( dhn) = 2hs
dt t = o T
If the time Tis split up into m increments
T
t:J.t=-
m
so
The next step of integration can now be performed
h _h
nt=2 .:l.t- nt=.:l.t + dt
(dhn) t=.:l.t x
T
m
This process can be repeated until a sufficient time period has been ex-
plored. As the integration scheme is of an initial value type, m must be large
and so the process is best performed on a computer. The program required
can be written in a very short time and the run time will be very small even
when m is made large.
As stated already this analytic method is subject to very grave defects if
the focus of interest is the magnitude of pressure transients, as it omits all
elasticity effects and so only begins to approximate to reality when valve
closure times are large, in which case the problem becomes trivial. The
method of analysis can be useful when pressure transients are not of im-
portance, however.
1.5 Distensible pipe-elastic fluid theory
The remainder of this book will be concerned with elastic theory. The most
accurate of the methods, described later, requires the use of computers.
The various theories of waterhammer described will be presented in a
sequence of increasing complexity and the first three chapters will be de-
voted to describing what are considered by the author to be obsolescent
theories. The reader would be wise to master these obsolescent theories
in the sequence in which they are presented as it is the sequence in which
they were developed. He will then be able to assess the improvement in ac-
curacy attainable from the use of progressively more rigorous methods.
1.6 Instantaneous valve closure
This section is important, as it provides the reader with an insight into the
mechanism of wave propagation and reflection in pipelines.
Pressure rises generated by very rapid velocity fluctuations are known
as transients and have a period equal to four times the pipe length/wave
celerity, i.e. 4Ljc.
Instantaneous valve closure is a theoretical concept, as no valve can be
closed in zero time but the study of instantaneous closure leads to an
understanding which can be used to solve real problems.
When a valve at the end of a pipeline is closed in zero time the layer
of fluid immediately upstream of it is instantly brought to rest, and its
impact upon the valve will cause its pressure to rise. This increase in pres-
sure will cause the section of pipe containing the fluid layer to distend
and the fluid in the layer to compress. The fluid layer immediately up-
stream of the now stationary first layer will next be arrested a very short
time later. The time delay is caused by the second layer continuing
in motion for a small time while it moves forward to occupy the volume
made available by the pipe distension and fluid compression of the first
layer.
The third layer is brought to rest in the same way as the first and
second, its loss of momentum due to its impact upon the second layer
causing within it a pressure rise identical to that experienced by the
first and second layers. As the first and second layer cannot rebound
from the closed valve their pressure cannot fall and is maintained at its
initial impact value. Progressively, layer after layer of fluid is brought to
rest. The situation is depicted in figure 1.4a.
L
I
distended pipe
pressure wave
magnitude ~ f>;
wavespeed 'c
static head~ t>,
I< f- pressure head plat
(a)
Figure 1.4
Eventually, the entire pipe length is full of fluid which is at rest but at a
pressure head of hi + h 5 where hi is the head rise caused by the impact,
i.e. an inertia head, and h 5 is the static head of the fluid in the upstream
reservoir (neglecting local losses). The situation is then as depicted in
figure 1.4b.
The process of impaction of successive layers with the small time delays
involved in each layer's impact mentioned earlier is, in effect, the propaga-
tion of a wave of pressure hi at a velocity c. The time taken for this wave
to traverse the length of the pipe L is L/c. When the wave has traversed the
pipe the entire mass of fluid in the pipe is at rest but it is also at a pressure
hi + h 5 This situation is unstable as the reservoir is at a lower pressure h 5
The fluid therefore starts to flow out of the pipe in a direction towards
the reservoir. Successive layers of fluid move towards the reservoir at the
original velocity v, each layer of fluid expanding and its associated pipe
~ - 0
. 7.
"'reservoir d1stended p1pe
I=. pressure head plot

c
(b)
Figure 1.4 (continued)
section contracting back to its original diameter. Figure 1.4c depicts an

intermediate stage in this process.
Eventually the reflected wave arrives at the valve. Figure 1.4d depicts
the situation that then prevails.
The flow circumstance is now exactly the same as that which existed at
t = 0 except that the flow is now away from the valve instead of towards it.
Again this condition is unstable. As soon as it occurs, the fluid at the
valve end attempts to leave the closed valve and to move in an upstream
direction. It cannot do this, so its momentum is converted into a pressure
decrease. The fluid layer next to the valve is brought to rest and its pres-
sure is reduced by an amount equal to the original pressure rise, i.e. by
hi. Successive layers are brought to rest as before but this involves a pres-
sure drop as opposed to the original pressure rise. Figure 1.4e depicts an
intermediate stage in this process. Eventually the entire pipe length is filled
with fluid at rest but with a pressure head of hs- hi as shown in figure
1.4f.
Again this situation is unstable because fluid will flow in from the reser-
voir at the original velocity v. This will cause the pressure to rise back up
;
z
L
--- v .....,_v~o
l' .,
h,~
I lh, I
pressure head plat
0c_ >I> _1,_
c c
(c)
~ .....,;,-' X
pressure head plot
(d)
h,
(e)

J
L
:4
""reservo1r
~reduced pipe diameter X
pressure head plot
(f)
to its original pressure head h 5 and the velocity will revert to its original
value of v directed towards the valve. An intermediate stage of this process
is depicted in figure 1.4g.
The final consequence of this reversal of flow is shown in figure 1.4h.
This situation is exactly the same as the initial circumstance so the pro-
cess repeats endlessly. In fact, friction rapidly attenuates the transients
that have been described so the reflected waves are sequentially reduced in
magnitude. In fact, the above description has ignored the effect of friction
but this is described later in this section and in section 6.17.
Typically, five or six reflected waves of significant magnitude will be
seen. In the description of the mechanism of wave formation given above
it has been said that a wave reflects completely and negatively at a reser
voir and completely and positively at a closed end (see figures 1.4a, band
c, and figures 1.4e and f.). This means that a wave travelling over a fluid
at pressure h 5 with a magnitude /::;.h is reflected at a constant head point
(a reservoir) with a magnitude h 5 - /::;.hand at a zero velocity point (a closed
end or closed valve) at a head of h 5 + Ah. This is an automatic consequence
of the laws of conservation of energy. In a circumstance in which fluid has
v2 p2
velocity energy 2g Nm/N but no relative strain energy w~K Nm/N, i.e. at
~~=~~v==:;~r.....3E:o=T;~X
""original pipe diameter reduced pipe d1ameter
pressure head plot

~>!>~
c c
(g)
L
~ "reservoir
-
pressure head plot
I= 4L
c
(h)
a closed end, there is a direct conversion of kinetic energy to strain energy,

i.e. a positive wave reflection and vice versa at a reservoir. (Note K is the
bulk modulus of the fluid.) A major principle has been enunciated: 'a com-
plete positive reflection occurs at a closed end of a pipeline and a complete
negative reflection occurs at an open end'. This suggests that partial (posi-
tive or negative reflections) occur at ends which are not completely open
(constant head) or completely closed (zero velocity), i.e. at junctions.
This will be more definitively discussed in later chapters.
By careful consideration of figures 1.4a-1.4f, pressure plots at various
points can be produced.
At the downstream end of the pipeline a pressure-time plot can be
evolved, for example see figure l.Sa.
2L 2L 2L
c c c
I I_
h
I' I
hi
valve closure h;
occurs
....___;
(a)
Figure 1.5
At a point!' upstream from the valve the pressure-time diagram is as

shown in figure l.Sb, and at the reservoir end of the pipeline the pressure-
time diagram is as shown in figure l.Sc
valve closure
occurs
J/ L h,
_{
c
(b)
Note that although the wave shape changes greatly as the point of obser-
vation of the wave moves upstream there is no attenuation of the wave
magnitude.
The effect of friction on the wave is, in some ways, surprising. The
wave shape at the valve is as illustrated in figure 1.6. This diagram requires
explanation. At point A the valve has just closed, the head hi has been
generated because the velocity v has been destroyed. At point B the velocity
v has also been destroyed and an inertia head hi has been consequently
generated but the wave arriving at the valve at time B was generated by
stopping fluid moving at a point!' up the pipeline. At the instant when
this fluid was stopped the pressure head was greater than the original pres-
sure at the downstream end by an amount 4fl'v 2 /(2gd) and the fluid was
stopped at a time l'/c after the valve closure. The abrupt stopping of the
flow at a point l' upstream of the valve thus causes a total pressure rise of
hi+ 4fl'v 2 /(2gd) but this rise arrives at the valve at a time l'/c later. So, the
2L 2L
h c c c
I I_
h,
/
valve closure occurs
t
(c)

2L
T
h
h = 4fLv2
f 2qd
4flv 2 +h.
2qd '
Figure 1.6
pressure rise caused by stopping the flow /'upstream takes a further 1'/c to
be transmitted to the valve at wavespeed arriving there at a time 2l'lc after
the valve closure. Considering the circumstance at the reservoir end, the
liquid stops at a time L/c after the valve closure and the pressure rise is
communicated to the valve after a further L/c time interval. The pressure
rise is hi+ 4fLv 2 /(2gd). Immediately following this pressure wave there
will be a large negative pressure wave. This negative pressure wave will
be running through stationary liquid so the pressure will drop from
hi+ 4fLv 2 /(2gd) to -hi. Due to energy losses in friction the value ofv
will be less than the original v value so hi will be smaller than the original
hi and the friction head 4fLv 2/(2gd) will also be smaller than the original
value of the friction head. Thus the waves attenuate. (See section 6.17 for
further explanatory comments.)
1.7 Separation
If a negative wave created by a reflection at the reservoir end of the pipe

should attempt to reduce the pressure to a value less than vapour pressure
the liquid will boil at the ambient temperature and a hole will appear
within the liquid. The pressure will not be able to fall below the vapour
pressure.
The pressure trace will then appear as illustrated in figure I. 7. Because
an equivalent negative transient to the initial positive transient cannot oc-
cur (as the pressure is unable to fall below the vapour pressure) the fluid
moving away from the valve at a time a little greater than 2L/c cannot be
brought to rest quickly. Consequently a long delay occurs whilst the in-
adequate pressure difference operates to reverse the flow. The situation
may then repeat until the initial transient has been so attenuated as to be
unable to reduce the pressure during its negative phase to vapour pressure.
Once this occurs the transient behaves in an exactly similar manner to
that of any other transient. A similar phenomeon known as gas release
absolute zero head

Figure l. 7
can simulate a very similar condition to boiling. In water, if the pressure

falls below 2.4 m (8ft) absolute (approximately), air bubbles will evolve
from the air dissolved in all natural water. These bubbles reduce the rate
of reduction of water pressure whereas boiling prevents its reduction be-
low vapour pressure absolutely.
When the fluid column is at vapour pressure and a hole appears within
it the phenomenon is called column separation.
1.8 The calculation of the magnitude of the transient caused by complete

instantaneous valve closure at the end of a simple pipeline
Allievi expression
Consider a length of pipe ~x long through which a transient pressure of
magnitude ~Pi passes in time ~t, reducing the velocity from v to zero
(valve closure case).
~p = _w~x dv
I g dt
from equation 1.2.

~t is the time taken for the transient to traverse the ~x length reducing
the velocity by ~v and is equal to ~xjc where cis the celerity of the tran-
sient
w~x -~v
~pi=--- x--
g &jc
~Pi= ~h = C~V (1.12)
W I g
This is sometimes known as the Allievi expression but is also variously at-
tributed to Moens, Korteweg and Joukowsky.
As the ~v in the above expression is the velocity decrement occurring
in time ~t it can be replaced by v if the valve closure is total (but still
instant) and then
~h
I
=cgv (1.13)
The remaining problem is to calculate the value of c.

Wavespeed
The magnitude of the wavespeed in part depends on the bulk modulus of
the fluid and in part on the distensibility of the pipe. It can be calculated
quite simply if the pipe distensibility is definable.
A simple case will now be demonstrated and results for other cases
will be quoted.
Pipe fitted with expansion joints so that it can extend longitudinally with-
out generation of longitudinal stress and free to distend diametrally
When a pressure 11pi is applied to a pipe it will distend diametrally. The
fluid within it will compress. The pipe distension plus fluid compression
effects enable it to contain more fluid than it would do in its normal un-
pressurised condition.
This increase in volume can be calculated as follows.
Volume increase due to fluid compression= 11 Vp =~Pix~ d 2 L
where K is the fluid bulk modulus and dis the pipe diameter.
The hoop stress fh in the pipe wall = 11,{~d where Tis the wall thickness
of the pipe.
The circumferential strain in the pipe wall equals the diametral strain
ah =fh
E
E being Young's Modulus. (1.14)
The increment in pipe radius = ah x ~
The increment in pipe volume 11Vpipe =
circumference x length x radial increment
d 1
11 Vpipe = rrdL x ah 2 =2 ahrrd L
2
- 1 l1pjd d
so 11 Vpipe- E 2 T x 2 rrdL
The total volume now made available by pipe distension and fluid com
pression is therefore 11 Vtotal
and
(1.15)
Remembering that if the wave has not reached a section the fluid will
continue travelling at its original velocity v, then the time taken for the
continuing undisturbed flow to occupy this additional volume will be
so
This time is the same time as that required for the wave to traverse the
pipe compressing the fluid and distending the pipe.
From the Allievi equation (1.13)
D.pi =---g
wcv
=pcv
and equation 1.2
and as D.v = -v
PLv
pcv=-
D.t
so
1
c = D.pi
v K
('_!+ TE
.!!_)
but from equation 1.12
- - + d)
c=wc(1 -
g K TE
(1.16)
If the pipe had been infinitely rigid this result would have reduced to
c=-- 1
or c=Jf
Jrl
so the effect of distensibility can be imagined as reducing the effective
bulk modulus of the fluid from K to K' where
For the case of a lined tunnel in rock: steel pipe through rock tunnel
with concrete infilling between pipe and tunnel
c=
where
where ds external diameter of steel liner

de = external diameter of concrete pipe
Es = Young's modulus of steel
Ec = Young's modulus of concrete
ER = Young's modulus of rock
1
Poisson's ratio for rock
m
T = wall thickness of steel liner.
For a plain tunnel in rock
cJf!i=/w
-+-
I 2
K ER
For a thick walled pipe
where
d1 = external diameter of pipe
d 2 = internal diameter of pipe
K = 2.03067 x 10 9 N/m2 =42.336 x 10 6 lbf/fe for water
E = 2.10915 x IOn N/m 2 = 44.064 x 10 8 lbf/fe for steel
w "' 9810 N/m 3 = 62.4 lbf/ft 3 for water
g = 9.8lm/s2 = 32.2ft/s2
1.9 Pressure rise caused by instantaneous valve closure
The pressure rise can now be calculated from Allievi's formula

wcv
!::.pj=-g
wv
- g
-j;(k+ !e)
(1.17)
When using this formula it is vital to remember that units must be com-
pletely consistent, e.g. in SI units, g must be in m/s, w in N/m 3 , K and E
in N/m 2 and d and Tin m.
The results obtained predict very large pressure rises if a valve in a
pipeline is closed instantaneously, e.g. for a typical steel pipeline of nor-
mal dimensions an instantaneous valve closure will generate a pressure head
rise of 125 metres of working fluid for every metre per second of velocity
destroyed. As some pipelines are now working at high velocities (10 metres
per second is not an extreme value) great care must be taken to make either
the pipeline extremely strong or ensure that no instantaneous valve closures
can occur.
1.10 Sudden valve closure
The very large pressure rise created by an instantaneous valve closure can,
unfortunately, be generated by valve closures which are far from instan-
taneous. (This may not be the largest rise as line pack can produce even
higher pressures especially in long pipelines- see section 6.17.)
If a slow valve closure is thought of as occurring in a sequence of small
steps of closure, each step occurring instantaneously but separated in time
by a small time interval, then each step will generate a small velocity decre-
ment t::.v associated with a small pressure rise t:.p. t:.p will be given by the
Allievi equation
t::.p = wct::.v
g
and its wave form will resemble that described in figure !.Sa.
Each of the steps will produce such a wave but each will start a small
time interval after its predecessor. The waves so generated will superimpose
upon one another and the pressure at the valve will rise. If the last closing
motion of the valve is completed before the first wave's negative reflection
returns to it the sum total of the !lp values of all the waves will exactly
equal that produced by an instantaneous valve closure in which the same
initial velocity was destroyed. Such a valve closure is called 'sudden'. The
wave's shape will be different from an instantaneous closure but its peak
magnitude will be the same. It is produced if the closure occurs in a time
less than the pipe period 2L/c.
If the valve closure is slower than this, reflected expansion waves will
be returning while the later steps of valve closure are still occurring. These
pressure decrements superimposing upon pressure increments still being
generated by the continuing valve closure will cause a reduced rate of in-
crease of pressure or even a decrease to occur so ensuring that the pressure
rise generated by a valve closure which takes longer than 2L/c will produce
a peak pressure which is less than that generated by a sudden closure.
Pipelines are now being built which may be as long as I 00 kilometres
without any intermediate booster stations. The pipe period for such a line
could be as great as 200 seconds. The closure of a valve at the end of such
a pipeline in a time of 3 minutes 20 seconds might seem slow but in fact
it would be fast being a sudden closure and giving rise to transients of
maximum magnitude. It will be realised that it is not possible to discuss
valve closure rates in terms of being fast or slow without reference to the
period of the pipe (2L/c) to which the valve is fitted.
Note The pipe period 2L/c must not be confused with the period of oscill-
ation of the water hammer wave 4L/c.
2 Analytic and graphical
methods
2.1 Introduction
It is necessary to discuss the work that has been done in the past on the
analysis of transients generated by slow movements of hydraulic controls.
As this chapter is still concerned with providing a background to the more
modem techniques of analysis that will be presented in later chapters only
an outline will be provided.
2.2 Analytic methods of solution
Two analytical techniques for solving slow valve closures exist. Both of
them require the assumption that friction in the pipeline can be ignored.
The two methods are completely equivalent although this may not be
obvious upon a superficial examination.
The assumption that friction can be ignored is extremely dangerous in
the hands of an inexperienced analyst. It can lead not only to grossly wrong
solutions but to unsafe solutions. Therefore, before either of the following
techniques is used, the circumstance to which they are to be applied must
be very carefully examined bearing in mind the above comments.
2.3 Stepwise valve closure at pipe period intervals
The method is based upon the idea of considering the pressure and velo-
city conditions in the pipe at every pipe period (2L/c) interval throughout
the valve closure. It is necessary, of course, to know the position of the
valve at the end of each of these intervals. The first step of valve closure
will not have generated a negative reflection from the reservoir end of
the pipe at time zero so it will be dealt with separately.
A small time after the first closing movement of the valve has occurred
the situation in the pipeline will be as shown in figure 2.1.
23
Figure 2.1
There is a wave of magnitude ilhi, travelling up the pipeline at velocity

c. This wave is classifed as an F wave. Waves travelling down the pipe are
classified as/waves. The reason for this terminology will be explained in
the next section.
Initially, before any valve closure movement has occurred, the velocity
v0 exists throughout the pipe and the prevailing head is h 5
Therefore (2.1)
Where av o is the full open valve area and Cd 0 is the coefficient of dis-
charge.
Immediately after the first step of valve closure
- cd, avl
v---11/""g
l'j; (h
s
cilvl) 05
+-- (2.2)
ap g
where c is the wavespeed.
Cdavj2i
Denoting - 2gby 13
ap
rs
Then Vo = 13oh~ 5 (2.3)
and cilv
Vt =13. ( hs+T (2.4)
but ilv1 = Vo -v1
Vf = l3~ (hs +CVo; CV1)
v~ + l3~cv1 -13~ ( h5 + C:) =0
Solving this quadratic
vl =J3~c +_! 13jc2 +413lhs + 413g~cvo (2.5)

2g 2 IS
Analytic and graphical methods 25
13ic (l31c) 2 v0 c
vi=--+131 - +hs+- (2.6)
2g 2g g
In the foregoing, the assumption has been made that Bernoulli's equa-
tion can be applied at the partly open valve. As stated in section I .4 this
assumption has been shown to work well even in unsteady flow conditions.
Having calculated v1 the magnitude of the F wave: f1h 1 is readily ob-
tained from f1h 1 =c(v 0 -vi)/g.
After a period 2L/c a further closing valve movement occurs, the 13
value of the valve becoming:
~= cd 2 av v'fi/ap
l
However, an f wave will have been reflected from the reservoir at a

time Ljc after the initial step of closure and will be arriving back at the
valve at the time 2L/c just as the valve makes its next closing movement.
The magnitude of this F wave will equal -F because it was generated by a
negative reflection of the f wave at the reservoir. The head at the valve at
time 2L/c will thus be hs+ f. The velocity in the pipe will be v1 = v0 - f1v1
The head at the valve will rise from hs +/to hs + f + c(v1 - v2 )/g
V2 = 132 (hs + f + c(vi- V2 )/g) 0 "5
Now as
f1v1 =!..Fandf= -F
c
f1v 1 = -[gjc
so v1 - v2 = v0 - f1v 1 - v2= v0 +!..t- v2

c
v~ = 13~(hs + f + c~o + f- c;2)

Solving this quadratic gives
V
2
=- 13~c
2g
+ 13 rrl32c)2 + CVo + h + 2/
2,.f\2i g s (2.7)
As f equals -f1h 1 , v2 can be calculated and f1h 2 can then be evaluated from
f1h 2 =c(v 1 - v 2 )/g
The head h 2 = hs +f + f1h 2 The entire process can be repeated until the
valve closure has been explored.
2.4 The Allievi interlocking equation
Allievi1 developed the following analytic technique in 1903. It is more

complex than the method given in section 2.3 and it is more elegant to the
mathematically minded reader but it is no more accurate and in fact the
technique described above can be manipulated so as to produce the
Allievi interlocking equations quite easily.
Before the Allievi interlocking equations can be presented it is neces-
sary to develop the differential equations of waterhammer. These two equa-
tions are differential forms of the continuity and dynamic equations and
as they form the basis of all accurate analytic methods they will be devel-
oped here.
The continuity equation

In figure 2.2 a section of pipe is shown in which a wave is travelling in the
upstream direction. The mass of fluid entering the elemental length ox
in time otis pAv&t and the mass leaving it in time otis
(p + ~~ ox) (A + ~:ox) (v + ~~ ox) o t

The additional mass that can be accommodated due to fluid compression
and pipe distension within the ox length during the ot interval is due to
the increase in mean pressure that occurs over the 0t interval and is:
pAox ~:&t (k + ~) (see equation 1.15)
The net mass inflow to the ox length must equal the amount of mass that
the fluid compression and pipe distension can accommodate, so
ap aA av
pAv&t- ax oxAv&t- pv ax ox&t- pA ax ox&t- pAvot
ap
= pAox at ot
d)
( K1 + TE (2.8)
wave moving upstream

at velocity c
datum
Figure 2.2
neglecting second order small quantities,
so Av-
ax
aA
ap +pv-+pA
ax ax
ap ( -
av pA-
-+ 1+-
at ,K TE
d) = 0 (2.9)
g 1 d
but wc 2 =K+ TE (see equation 1.16)
so (2.10)
ap
but ap Po ax
-=--
ax K
~ aA=~ aA ap
and
A ax A ap ax
A pipe of circular cross section was assumed when developing equation
rr 2
1.16 on page 197 so A =4 d
~ aA = _v_ ap !1: 2d ad = 2v ip_ ad

A ax '!!: dz ax 4 ax d ax ap
4
but ad
ap
=~
ap
(ih d) _a (
E
pd
- ap 2TE x
d)
d2
2TE
~ aA = 2 ~ ap d 2 = v ap !i
so A ax d ax 2TE ax TE
As w(h - z) =p and w ~~ = ~equation 2.1 0 becomes
(2.11)
where p0 is the density at the origin pressure above which pressure is

measured.
But. = 1 + PK and asK is extremely large in comparison with any prac-
Po
tical p value pjp0 can be accurately approximated to unity.
:. ... ah + vw (ahax - ax~) (K_!_ + TE av =0

__) + ax
c2 at
but w(.!+
K TE
A.)= cK 2
so
ah ah c2 av az
-+v-+-- -v-=0 (2.12)
at ax g ax ax
This is the continuity equation of waterhammer in differential form.
The dynamic equation

A force balance equation can be written for the section of pipe shown in
figure 2.2.
Force acting L -+ R =
pA - (P +~~ox) (A +~:ox) +(P +~ ~~) (~:ox) - rPox- wAox ~~

(I) (2) (3) (4) (5)
Term (I) is the normal pressure force acting on the left end of the pipe
segment.
Term (2) is the normal pressure force acting on the right end of the pipe
segment.
Term (3) is the longitudinal component of the reaction of the mean pressure
force from the pipe wall upon the fluid.
Term (4) is the frictional force opposing flow.
Term (5) is the weight component acting along the pipe centreline opposing
the flow.
Note Pin term (4) is the mean wetted perimeter of the pipe segment and
T is the frictional shear stress between the fluid and the pipe wall.
Ignoring second order small quantities, force acting in the L -+ R direction is
ap dz
-A -ox-rPox-wAox-
ax dx
This force causes an acceleration of the fluid in the segment according to
Newton's Second Law.
ap dz dv
-A -ox -rPox -wAox- = pAox- (2.13)
ax dx dt
:. dividing through by Aox and rearranging
ap dz dV T _
ax+ w dx + p dt +A/P-O (2.14)
Now A/P = m, the hydraulic radius of the pipe, where A = cross sectional
area and P =the wetted perimeter of the flow, and dv can be expanded by
dt
the use of the definition of a total differential derivative, i.e.
dv av av
dt = v ax+ at
but
dz az
dx ax
a au au T
ax (p + wz) + pv ax + p at + m =0
Dividing through by w (=pg)
]__ (!!._ + z) + !:'_ av + .!_ av + __!___ = 0

ax w g ax gat pgm
From the Darcy-Weisbach formula (Fanning in USA)
_!__ =fvlvl
pgm 2gm
(This implies the use of a steady-state friction formula.)
Now.e_ + z
w
=h: the potential head
so ah +_!:'_ av +_!_ av +fvlvl =0 (2.15a)

ax g ax gat 2gm
If the pipe is of circular cross section this becomes
(2.15b)
as m = d/4.
Note These two equations, continuity and dynamic, are a pair of quasi-linear
hyperbolic partial differential equations and as such cannot be solved ana-
lytically. Together they represent the problem of transient propagation in
distensible pipes but, as they cannot be solved analytically, various simpli-
fications have been made to them in an attempt to obtain an analytic solu-
tion. The best known of these simplified theories is that due to Allievi and
this will be presented here. Allievi decided to ignore the nonlinear terms
and friction. This means that he ignored the v ~~ term in the continuity
.
equa t ton -aX
as v ah 1s
o f t h e ord er o f v ---a
ah an d so, m
v+ CX
li nes, 1s
some p1pe
small. He ignored the~g aavX term and the 2 fvdlvl term in the dynamic equation.
g,
The~g aavx term is of the order of(__!!___)
v+c~ t
aav so it is usually small but to
1gnore 2fvlvl term 1s

t h e --g;J" to 1gnore
fnction
1tse
If an d t h.IS can on Iy be d one
if frictional head losses are trivial fractions of static heads. This assumption
of no friction is critical; without it there is no chance of obtaining an ana-
lytic solution but with it the value of the analytic solution becomes severely
limited.
The simplified equations that Allievi used are:
av g ah
ax=- c 2 at continuity equation (2.16)
av ah
at= -g ax dynamic equation (2.17)
By differentiating the first with respect to t and the second with respect
to x the wave equation in classic form can be obtained, i.e.
av g a2 h
ax at - c2 3("2
and
av a2 h
at- -ax= -g ax 2
a 2h a2 h
= c2- (2.18)
at2 ax 2
Once the form has been recognised it will be realised that an analytic solu-
tion must be possible. The solution is usually ascribed to Riemann, i.e.
h= h0 + F(t +~)+ t(t -~) (2.19)
v=v0 -~ [F(t+~)-t~-~)] (2.20)
(It can be found in any mathematical textbook covering the solution of

partial differential equations.)
The reader may have noticed that in other texts the solution quoted
differs from that given above. It will be found that this is due to the author
adopting the convention in this book that xis measured from an upstream
origin (the reservoir) and vis assumed positive if the flow is in the direction
of x increasing, whereas other authors have chosen their x origin at the
downstream end of the pipe (i.e. at the valve) and taken their velocity as
positive if it is in a direction of x decreasing. This is mathematically in-
consistent so the author has not accepted it.
The symbols Fand f denote 'function of. Clearly F(t + ~) and f(t -~)
must have the dimensions of head and in fact must represent contributions
of head from waves; this will be more fully demonstrated in the next para-
graph.
Consider the Reimann head equation. It could be plotted on an x base
for an instant of time t =t 0 . For an observer travelling upstream at wave-
speed c and located at x =X at time t = t0 , the head seen will be hx (see
figure 2.3) and hx = ho + F (to +~),ignoring the f (t- :) term for the
X
..
X
Figure 2.3
moment, but the observer is travelling, so at time t 1 he will be located at

position X 1 obeying the equation X 1 ==X- (t1 - t 0 )c.
Now, according to the Reimann equation
hx, == ho +F(t, + ~1 )
but F(t 1 +~1 )==F(t 1 +X-(~-to)c) ==F(t0 +~)

i.e. the original unchanged value ofF at time t == t 6 at position x ==X.
If, to an observer travelling upstream at wavespeed, the F term does
not change then the F term can only be a wave travelling upstream at
velocity c.
Similarly it is argued that thefterm represents a wave travelling down-
stream at speed c.
Thus, like all equations, the Reimann head equation is a statement of
the obvious. The head at any point on a pipeline x at time t is made up of
the static head h 0 plus the head contributed at that point and time by a
wave travelling up the pipe (F wave) plus the head contributed by any
wave travelling down the pipe (!wave).
From these equations it is simple to deduce the reflection circumstances
at reservoirs and closed ends by mathematical means rather than by the
use of engineering insight as in chapter 1.
Consider a wave generated at the downstream end of a pipe travelling
upstream towards a reservoir. Consider a point on it generated at time
t == 0 at the downstream end (x ==L ). The wave magnitude will be F(o + ~).
As it travels upstream its shape will be unchanged (see before) so at time
t and position x
When this wave reaches the reservoir at time L/c it will still have the same
magnitude and the function will be F ({+~),i.e. F (~)At the reservoir the
head must remain constant at h 0 , so
h== ho == ho + F(~) + t(~ + o)

hence F(~) which is still equal to F(o +~)equals-[~+ o).

After a further period of!:. the f wave which has been generated will
c
travel back to the valve arriving at x = L at time 2L/c, i.e. j 2L
J \'c -!::.)
c ,
i.e.t(~), indicating that thefwave has travelled back to the valve with
unchanged magnitude. Thus the initial wave F(o +f) will travel upstream
to the reservoir with unchanged magnitude and shape at which it will be
reflected completely and negatively as an/wave, which will then travel
downstream to the valve with unchanged magnitude and shape arriving
there 2Ljc after the first F wave started out.
So, in a simple pipeline,[ waves arriving at the valve were generated
one pipe period earlier as F waves which were totally and negatively
reflected at the reservoir.
Similarly at a closed end F waves are totally and positively reflected.
This can be shown from the Reimann velocity equation. At a closed end
the velocity must always be zero, so at such an end v and v0 must always
be zero.
Thus o= o-~ [1~+ ~)-t(~- o)J

i.e. F(~) =t(~) (2.21)
Summarising: if time periods ofT( T= :L the pipe period) are considered

and any particular time is denoted by iT where i is any appropriate integer
number then:
for the case of a reflection at an open end (a reservoir)
/i =-Fi-1 (2.22)
and for the case of a reflection at a closed end
/i=Fi-1 (2.23)
Denoting the F(t +~)wave at a time iT by Fi and the t(t -~)wave at

time iT by fi the Reimann equations can be written as
(2.24)
(2.25)
Considering the sequence of events at a downstream valve caused by its
slow closure at pipe period intervals
at t = 0
ho = ho + Fo + fo (2.26)
Vo = Vo -~c (Fo- fo) (2.27)

at t= T
h1 =ho + Fi +ft (2.28)
V1 =vo-~(F, -[t) (2.29)

c
at t = 2T
h2 = ho + F2 +!2 (2.30)
V2 =Vo -~c (F2- f2) (2.31)
and so on, but / 0 = 0 as there cannot be an f wave until the initial F wave
has reflected at the reservoir and returned to the valve.
Now ft = -Fo (2.32)
!2=-F, (2.33)
[3 = -F2 (2.34)
and so on, so
h 0 =h 0 +F0 (2.35)
Vo=vo-~Fo (2.36)
c
h 1 = ho + F, - Fo (2.37)
v1 =v 0 -~(F1 +F0 ) (2.38)

c
h2 = h 0 + F2 - F 1 (2.39)
V2 =Vo - ~ (F2 + Ft) (2.40)

c
and so on.
Adding successive pairs of head equations and subtracting successive
pairs of velocity equations
h1 + h 0 = 2h 0 + F 1 (2.41)
h2 + h, = 2ho + F2- Fo (2.42)
h 3 + h 2 = 2h 0 + F3 - F1 (2.43)
etc
and
(2.44)
{2.45)
(2.46)
etc,
generally then
c
fl-fl-2 =- (Vj-1 -
g
VJ (2.47)
so
c
hi+ hi-t - 2h 0 =- (vi-t -Vi) (2.48)
g
and the F and f functions have been eliminated from the problem.
Nothing more can be done with these equations unless further informa-
tion is supplied. If Vi and Vi-t can be specified it is possible to solve
these equations sequentially by setting ito 1, 2, 3, etc, in turn.
The boundary conditions at the valve can next be introduced and this
leads to a solution.
Using the usual theory for defining conditions at a valve (see equation
1.9 et seq.)
(2.49)
or
(2.50)
When i= 0
so
{2.51)
Denote
and
TJ is greek letter eta

r is greek letter zeta
vi= voT/iri (2.52)
hi=horf (2.53)
Substituting back into equation 2.48
Denote VoC b 2
gho y p
then
(2.54)
The symbols used are those suggested by Allievi in his original paper.
The symbol p must not be confused with the symbol used elsewhere
to denote the specific mass of a fluid. Here it is called the Allievi pipe
. . d. l CVo
ch aractenshc an 1s equa to 2gho.
Equation 2.54 represents a family of equations obtained by making
i = I, 2, 3, etc, successively. They are known as the Allie vi interlocking
equations.
When i =I
(2.55)
~0 will be I since, up to the instant of valve closure commencement, heads
will be steady.71 0 will be I if the valve is fully open, so when i = I
~~ -1 = 2p(l -r~~~.) (2.56)
If the fractional valve opening 71 1 is known, ~ 1 can be calculated by the
usual quadratic solution.
Wheni = 2
(2.57)
~ 1 has just been calculated
711 and 71 2 must be specified, i.e. the valve closure pattern must be known,
~ 2 can therefore be calculated.
The entire valve closure period can be explored. As the solution from
one step is used in the next step, the solutions interlock, hence the name
of this technique.
Having obtained ~ 1 t 2 t 3 , etc, heads can be readily calculated
h1 =ho~~. h2 =hon,h3 =hon. etc
and velocities can be similarly acquired:
V1 =Vo111tt. V2 =Vo712~2 V3 =Vo113t3, etc
A complete solution has thus been obtained.
This method has been well worked out and techniques of applying it to
pipe networks as well as to simple pipes have been evolved. These methods
are well presented in books such as Hydraulic Transients by Rich and
Engineering Fluid Mechanics by Jaeger. However, the author feels that as
it is not possible accurately to include the effects of friction in this method
and also believes that it has probably reached the end of its useful life there
is little point in using space to describe such developments.
2.5 The Schnyder-Bergeron graphical method
The Allievi interlocking equations lead to an elegant solution of friction-

less waterhammer but become difficult and cumbersome to use if the pipe
network is not simple and if the hydraulic controls are in any way com-
plex.
Essentially the Schnyder2 -Bergeron 3 graphical method solves the same
fundamental equations as the Allievi method. The graphical method is
much easier to generalise than the Allievi method. It does not involve the
calculation of reflection and transmission coefficients if there is more
than one pipe in the network and is capable of making a moderately ac-
curate allowance for frictional effects in the system. It is able to deal with
quite difficult boundary conditions and has been used, by highly skilled
practitioners, to great effect. It suffers from a number of defects and
one of these is the level of skill needed to deal with anything other than
a simple situation. When considering most schemes it is necessary to in-
vestigate a large number of modes of operation of the scheme. As graphical
techniques are time consuming this can rarely be done.
Any graphical technique consists of the following:
(I) Graphically representing the boundary conditions present in the
network.
(2) Graphically representing the equations of waterhammer which de-
scribe the conditions within the pipe as the waves traverse it.
(3) Linking the events across the ends of two pipes (or more) at which
a boundary condition (i.e. a hydraulic control) is present.
In any waterhammer problem four variables are involved: head, velocity
(or flow), position in the pipe ~x, and time ~t. To represent these four
variables on a graph an ingenious device is used which employs the fact
that a transient is propagated at a speed c. In the graphical method this
speed is treated as being constant and this is not true in many circum-
stances. This is another defect of the graphical method.
In the early parts of this presentation friction will be ignored, but
methods of introducing friction will be described later.
The waterhammer or eagre lines

An eagre is a small wave that travels at constant velocity. In this context
the word is used to describe the small incremental waves that continually
traverse the pipe during and after the operation of a hydraulic control.
Rewriting the Reimann equations:
h=h 0 +F(t+~)+t(t-~) [2.18]*
[2.19]
Rearranging the velocity equation gives
i(v0- v) = F(t +n- t(t -n (2.58)
Writing them both for two points at x and x' and at times t and t'
hx,t -ho =F(t +n+ tt -~) (2.59)
hx,t-ho =F(t'+~')+ t(t'-~') (2.60)
and
~(vo- Vx,t) = F(t +~)-f~ -~) (2.61)
i(vo -vx,t')=F(t'+{)-t~'-~') (2.62)
Subtracting the head equations gives
hx,t -hx',t' =F(t +~) +t(t -~)

-F(t' +%)- t(t' -~') (2.63)
Subtracting the velocity equations gives
i(vx Vx,t) = F(t + ~)- t(t- ~)

,t' -
-F(t'+~') +t(t'-~') (2.64)
Up to this point these equations are applicable to any value of x, t, x'

and t'but if it is assumed that x, x', t and t' are related as follows
x = x' + c(t- t ')then a result of great importance emerges.
This relationship is that which would describe the motion of an observer
* Square brackets indicate that the equation with this number was introduced earlier.
travelling downstream in the direction of x increasing. This observer will be
called a waverider from now on.
For this waverider
x x'+ct-ct' , x1
t +- = t + = 2t - t +- (2.65)
c c c
and
x x' + ct- ct' , x'
t--=t- =t - - (2.66)
c c c
Thus for such a waverider travelling downstream F(t + xja) changes but
f(t -xja) does not- an idea that has been implied previously.
The head relationship for such a waverider becomes
hx,t-hx',t' = F(2t-t'+x'/c)+f(t'-x'fc)
- F(t'+x'fc)-f(t'-x'fc)
= F(2t- t'+ x'fc)- F(t'+ x'fc) (2.67)
The velocity expression becomes
g
~ (vx ' t'- Vx , t) = F(2t- t' + x'fc)- F(t' + x'fc) (2.68)
Therefore
c
h X, t-h X,'t'=--(v
g X, t-v X,'t') (2.69)
It must be emphasised that this result is only valid for a waverider moving
downstream.
For a waverider moving upstream obeying the equation x =x'- c(t- t')
it can be shown by an identical method that
c
h X, t- h X,, t' =-g (v X, t-v X,, t') (2.70)
As v = q/A where q is the flow and A the cross sectional area of the pipe
h x,t - h x,t
"= + .. (qx,t - qx,t' )
- Ag (2.71)
This equation describes two lines of equal but opposite slope in an

h ,._ q space. It must always be remembered that the line of positive slope
implies a waverider moving up the pipe whilst the line of negative slope
implies a waverider moving down the pipe. The line of positive slope is
called an eagre I and the line of negative slope is called an eagre II. Move-
ment along either of these lines implies movement along the pipe at velocity
c and hence movement in time also. The direction of increasing x in any
pipe must be chosen so as to be in the same direction as the initial steady
flow.
Boundary conditions
The events that occur at the end of a pipeline are defined by the interaction
of a wave with whatever hydraulic control is present. Many devices can be
present at the end of a pipe, for example, a reservoir, a valve, a pump, a
turbine, a surge tank, an air vessel, a dump tank, a junction. At this stage
only two simple cases will be considered, i.e. a reservoir and a valve.
(I) The reservoir is a particularly simple device as all it does it to pro
vide flow into the pipe at an (assumed) constant head if located upstream
or accept flow from it at a constant head if located downstream. Its graphi
cal representation is therefore a horizontal straight line on an h - q plot.
(2) Valve. At any instant in the valve closure
(2.72)
where ~ and av 0 are the fractional valve opening and the full-open valve
flow area respectively.
Note~ does not have the same meaning as it did when used earlier.
h= ( -q-
cd~avo
)2/ 2g (2.73)
or (2.74)
where (2.75)
2g(Cdav0 ) 2 ~ 2
Equation 2.74 is the equation of a family of parabolae, each member
of which is defined by the value of t/1 which in turn depends upon the cur
rent value of~- As a valve closes, ~ decreases, so t/1 increases. ~ ranges from
1.0 for a fully open valve to zero for a fully closed valve and takes a posi
tive value less than unity for a partial closure.
At times T, 2T, 3T, etc (Tis the pipe period 2L/c) the value of~ must
be known and hence values of t/1 for each step of closure can be calculated.
Thus for every time there will be a parabola which describes all possible h
values corresponding to all possible q values for the fractional valve opening
then current.
The graphical solution

For a simple pipeline connecting a reservoir to a valve there will be three
elements necessary to define the problem: the reservoir characteristic
line, the eagre lines of waterhammer and the valve's characteristic para
bolae. (Note The word characteristic is not used here to mean the charac
teristic p mentioned in connection with the Allievi interlocking equation.)
t/10 is the valve characteristic line for the fully open valve. The fact that
this line is not a horizontal straight line through the origin shows that the
full-open area of the valve is less than that of the pipe cross sectional area.
In figure 2.4 the valve is shown as closing fully, in a period of four pipe
periods (the subscripts to the t/1 symbol denote how many pipe periods
have elapsed since the commencement of closure). The t/14 parabola has
h
hs is the height of the reservoir static surface level above the valve
Figure 2.4
degenerated into a vertical straight line because l/1 4 = oo as~= 0 when

the valve is closed. At point A, see figures 2.5, the reservoir head equals
that behind the valve, so point A defines the steady state flow before
valve closure commences.
To solve the problem, the eagre lines must next be plotted.
Tis the pipeline period (2L/c ). At time 0 the flow through the valve
is defined as at A 0 . The flow at B (and throughout the pipe length) is
the same as that at A at time zero and the head at B is also the same as
the head at A as there is no friction in the pipeline. The head and flow
at B cannot alter until a wave arrives there so the point for Bo.ST is the
same as that for B0 and this in turn is coincident with A 0 as shown in
figure 2.5. If a wave rider is started from Bat time 0.5T he will travel down-
stream along an eagre II. Plotting this line onto the diagram produces a
line Bo5T to Al.OT i.e. as the waverider leaves Bat time 0.5T he will arrive
at time I.OT at the valve. At this instant the valve characteristic l/11.0T comes
into existence and the intersection of the valve characteristic and the eagre
h
t/lo
)
r------8 -----::~
plan of pipeline
Figure 2.5
II line defines the h ~ q conditions at the point A at time l.OT. Now ima-
gine a waverider travelling back up the pipe at wavespeed c travelling from
A to B. As he is travelling upstream he will move along an eagre I (positive
slope), i.e. line A LOT- B1.5T, and he will arrive at Bat time l.STwhen the
eagre I line will intersect the reservoir characteristic so defining the h - q
conditions at Bat time l.ST. Reversing the waverider will give an eagre II
joining B1.5T to A2.0T; the valve characteristic l/12.0T comes into existence
at the instant that the waverider arrives so defining the h - q conditions
at A at time 2.0T. The diagram can now be completed by exactly similar
methods. The diamond shape B3.5T "'* A4.0T "'* B4.5T "'* As.OT can only be
produced in the absence of friction and it represents the pendulation of flow
that occurs after valve closure is complete.
Plotting the head at A against time produces a curve of the type illus-
trated in figure 2.6.
h-ht
-- ----
I s
/ _.--......,
.......... /
\ 5T / I ..
Figure 2.6
The prediction of heads at time intervals other tlwn pipe periods

To calculate the pressure heads at time periods which are fractions of
a pipe period it is necessary to plot additional l/1 curves. As an example,
consider the case of the prediction of heads at times of half a pipe period
(see figure 2. 7).
The symbol Twill be omitted from the subscripts in the following sec-
tions, i.e. the subscript 1.5 should be read as l.ST.
At time 0, heads and flows at A will be defined by the point A 0 , the
conditions at C will be the same as those at A 0 until a wave reaches Cat
time 0.25Tand at B until a time O.STas before, so, A 0 , C0 .25 , B0 , B0 25 and
Bo.s will all be coincident.
Starting a waverider from Bat time O.STwill give an eagre II line
Bo.s -A l.O so conditions at the valve at time l.OT will be defined by the
intersection of this eagre and the valve characteristic for the time l.OT. A
waverider can be started from Cat time 0.25T and this will produce an
eagre II which will intersect the l/lo.s characteristic, so defining the heads
and flows at the valve at time O.ST, i.e. at A 0 5
A waverider starting off from A at time l.OT will reach Bat time l.ST,
i.e. at B~, 5 and similarly one starting from A at time O.STwill arrive at Bat
time l.OT, i.e. at B~, 0 . Additionally a waverider starting from A at time
l.OTwill arrive at the pipe's midpoint Cat time 1.25T and one starting
from Bat time l.OTwill arrive at Cat l.25T also. The intersection of the
/5
1/1.
)
c
-
~
Note C is the midpoint of the pipe
Figure 2.7
two eagres will define conditions at Cat time 1.25T. The rest of the dia-
gram shown in figure 2.7 can then be completed. Thus by inserting the
1/Jo.s, 1/1 15 , 1{12.5, family of valve characteristics the conditions at the valve
at mid-period values have been obtained, and the conditions at the pipe
midpoint have also been established.
Slow llOive opening

In this case the 1/10 line coincides with the zero flow ordinate because the
valve is initially closed. The valve is assumed to open over a 3Tperiod in
figure 2.8. Once it is full open the 1{13 characteristic is the characteristic
for all subsequent times because the valve does not move after it has
reached its full open state. At time zero the flow is zero and the head is
h5 The conditions at B do not change until a wave caused by the first step
of valve opening arrives, at time O.ST.
Starting a waverider from B at time O.ST moving towards A gives an
eagre II.
This eagre intersects the 1/1 1. 0 curve at A 1.0 . Reversing the waverider's
direction produces an eagre I which intersects the reservoir characteristic
at B1.5. The process can be repeated to complete the diagram. Depending
upon the slope of the eagre and the l/1 3 .oetc line two geometries can be
produced as shown in figures 2.9a and 2.9b.
Thus if ... is relatively small (i.e. in the case of a highly distensible pipe
gA
with a low wavespeed and/or a pipe of large cross sectional area) it is pos-
sible to produce transients which exceed the reservoir static head.
lj;20
lj;3 0
4.0 ere
Figure 2.8
A
-
..__
Slow partial valve closure

The closure is assumed to occur over two pipe periods and the !J; 2 line is
the lJ; line for the {3 2 .o value of the partial opening when the valve has
completed its movement. The !J;2 line is thus the !J;3 , !J;4 , !J;5 line also.
The analysis is performed as usual but the eagre lines eventually spiral in
upon the intersection of the !J;2 line and the reservoir characteristic. Thus,
even in the case of a frictionless flow, pressure transients attenuate for the
case of a partial valve closure; they also do so for the case of an opening
valve. See figure 2.11.
If the slope of the eagre line is relatively large, the spiral shape of the
eagres may be modified as shown in figure 2.1 0.
Plotting of eagre lines

The slope of the eagres is given by
tan a=!!...... (2.76)
Ag
The values so calculated may be very large, i.e. if c = 1000 metre/sec and
A= 0.5m 2 .
~:::>.200
Ag
(a) (b)
Figure 2.9
h V;z.o
q
Figure 2.10
Figure 2.11
100
Figure 2.12
B c Tz A
pipe 1 pipe 2 valve
Figure 2.13
As the h- q plot is not to natural scales the value of a calculated is not of
great use. The suggested method of obtaining the eagre slopes is illustrated
in figure 2.12.
If cfAg = 200, say, draw a line from the q = 0.3 point on the abscissa,
i.e. point D, to the h = 60 point on the ordinate B, i.e. BD. Complete the
rectangle ABCD. Draw in the other diagonal A C. By the use of a parallel
rule eagre II lines can be drawn parallel to BD and eagre I lines can be drawn
parallel to AC.
The method of dealing with joints

Consider a pipeline in which a pipe is joined to a smaller diameter pipe
as in figure 2.13.
The lengths of BC and CA must be adjusted so as to be in a ratio such
that the pipe periods T1 and T2 for the two pipes are in a relatively simple
ratio to one another, e.g.~~= 1 or 2 or 3 or~ or t
If this ratio is not simple, say 1.42, the graphical analysis will be com-
plicated and take a very long time to perform.
To illustrate the method, T 1 has been assumed to be equal to 2T2 and
arbitrary eagre slopes have been chosen. Note, the eagre slope of the up-
stream pipe of larger cross section will be less than that of the downstream
pipe. The closure of the valve at the downstream end of pipe II is to take
512 pipe periods so five 1/1 characteristic parabolae must be calculated and
plotted. (More than this number may be necessary if the TtfT2 ratio is
not a simple integer ratio.)
The following discussion applies to figure 2.14. The horizontal abscissa
of the diagram has been moved up from the true zero head to the h = h 1
head value to save space and this is a usual practice when performing
graphical analyses. Only the portions of 1/1 lines which lie above the reser-
voir line at h = hs need to be plotted.
At A 0 steady state is defined. A 0 also defines conditions at Cup until
the time at which a wave will arrive there, i.e. Co.s. It also defines condi-
tions at B up until time 1.5 T2 , i.e. Bu. Note, the subscripts express times
t
in multiples of T2 (also remember that T1 = T2 ).
Thus a waverider starting from Cat time 0.5T2 will arrive at A at time
T2 - on the diagram this gives an eagre II, i.e. from CO.s to At.o
Starting from Bat 0.5 T2 will give an eagre II arriving at Cat time 1.5 T2
and starting a waverider from A at time 1.0T2 will give an eagre I arriving
at Cat time 1.5 T2 Thus the intersection of these two eagres defines con-
ditions at Cat time 1.5 T2
Having obtained C1.5 two waveriders can be started, one travelling up-
stream and one downstream. The downstream eagre will be an eagre II and
the upstream eagre will be an eagre I. The downstream waverider will arrive
at A at time 2.0T2 and the associated eagre will intersect the 1/1 2 .0 line at
A 2 .0 The upstream eagre I will intersect the reservoir characteristic at
B2.S.
h-h, ~
The subscripts indicate multiples of 72

Figure 2.14
A waverider starting from B at time 1.5 T2 will have an associated eagre

II and he will arrive at Cat time 2.5T2 A waverider starting from A at
time 2.0T2 will move along an eagre I and will arrive at Cat time 2.5T2
The intersection of these two eagres will define C2.s
As B0 , Bo.s, B1 .0 and B1.s are coincident, eagres B05 -+ C15 and
Bl.S -+ c2.5 are colinear over part of their length.
After the time of 1.51; the origin of any waverider starting from B will
no longer be at the steady-state point but it will always be found that new
starting points forB will have been established from earlier stages in the
analysis, e.g. Bo.s, B~. 0 , and Bu are coincident at the steady state point but
Bz.s will have been established before it becomes necessary to use it. From
C2.s, A3.o and B3.5 can be located.
From A 3 .0 and /h.s the point C3.5 can be obtained and the rest of the
diagram can be completed similarly.
The method of dealing ~th junctions

At junction C in figure 2.15 the equation of continuity must always be
true, i.e.
ql = q3 -q2
C 1 denotes conditions at the junction end of pipe 1. C 2 denotes conditions

at the junction end of pipe 2. C 3 denotes conditions at the junction end of
pipe 3.
At the junction, neglecting local losses and kinetic energy changes
(2.77)
i.e. the heads at the junction ends of pipes joining at the junction at any
given time t must be equal.
A
8
Figure 2.15
(Note that the indices 1, 2 and 3 define locations, not the first, second
and third power; the subscript t denotes time.)
The method of analysing junctions given below is based on the method
of analysing surge tanks given by Hawkins and Zienkiewicz 4
Consider figure 2.15 and let the pipe periods of pipes 1, 2, and 3 be
T\ T 2 and T 3 . On an h ~ q graph (figure 2.16) assume that conditions
at A at time t - 0.5T 1, at Bat time t - 0.5T2 and at D at time t - 0.5T 3
are known and can be located at At-O.ST', Bt-O.ST, and Dt-0.5T3.
Through these points draw the eagre lines appropriate to waveriders
travelling towards the junction C. Now, at C 2 the head must equal that
at C 3 .
h eagre 12
eagre II 3
q
Figure 2.16
The flow into pipe 1, i.e. ql must equal q3 -q2 and the head at C1 ' h 1 '
must equal h 2 and also equal h 3 It is therefore possible to draw a line
upon the h ~ q graph which is a combination of the two eagres, eagre
1h and eagre 12 . Point X on eagre 11 3 and Point Yon eagre 12 have the
same head so the requirement that he = hc3 is fulfilled at all points on line
PQ. The abscissa of the required line must be given by qx- qy so if PSis
set off along line PQ and PS = XY, Swill define one point on the required
line obtained from eagre 11 3 and eagre 12 . This process can be repeated
for another pair of points, e.g. E and G and so point T can be obtained
(RT= EG). The required line is thus ST; it is called ax line (Greek letter
chi). This Xt line intersects the remaining eagre 11 line at Ct, and so condi-
tions at the junction end of pipe l have been obtained for time t.
By drawing a horizontal line through ct, points Cf and Cf are obtained.
This process is valid because of the requirement that hcf =hcf =hcl
Points cL Ci and c~ have now been obtained and new eagres can be
started off from these points to establish a new generation of points
At+O.ST' , Bt+0.5T' and Dt+0.5T3 ~nd the]e can be u~ed in turn to establish
a further generation of points, C t+T' C t+T' and C t+T' etc. These new
points are not shown on figure 2.16.
The construction of the x line can be simplified as shown in figure
2.17.
eagre 11
h
R
---- Xt line
eagre II3 -----

D, -0.5T'
Figure 2.17
By drawing the horizontal lineRS through S, point R on the Xt line is

established. At S, q 2 = q 3 so q 1 = q 3 - q 2 = 0 and this must define a zero
flow point on the required Xt line, i.e. at R. By drawing the horizontal line
PQ, Q is established. At P, q 2 is zero so at Q the flow q 1 in pipt 1 at C 1 is
equal to q 3 , i.e. q 1 equals the flow in pipe 3 at C 3 . Therefore Q is another
point on the Xt line. The Xt line must be straight so by joining RQ the
Xt line is obtained.
In some junctions q 1 =q 2 + q 3 . The combination process performed
above for the q 1 =q 3 - q 2 equation can be applied with a very simple modi-
fication.
An illustration of the analysis of a three way junction

This is shown in figure 2.18.
Ann way junction can be analysed by the methods given in the previous
section. The combination of the eagres of two pipes produces a x line,
by combining this x line with the eagre of another pipe a further line (say
a Aline) can be obtained. This A line can be combined with the eagre line
of the next pipe to produce a further line (say a J line) and, finally, after
all the eagre lines of all the n pipes except one have been combined into
one line, the operating conditions of the excepted pipe at its junction end
can be determined by the intersection point of the compounded line with
B pipe I
D
2 D' pipe 3 A
D 03
c pipe2
h T, o Tz o r3
h-h,
a, o 25 a 2 o 30 a 3 o 45
Note X lines are parallel
Figure 2.18
the eagre of the excepted pipe. The heads of all pipes (at their junction
ends) being equal, a horizontal line drawn through the intersection point
defines the flows of all the other pipes by its intersection with the other
pipes' eagres. An analysis such as this is difficult to perform as it requires
considerable concentration and is tedious to carry out.
Networks can be analysed in which there are a number of junctions.
The technique outlined above can be used over and over again if the
analyst can maintain his concentration for a long enough time. The many
construction lines involved in this method should be drawn very lightly
and erased when the required points have been established.
Inclusion of friction
Friction is a distributed phenomenon but it cannot be included as such
in a graphical analysis. Schnyder and Bergeron separately advanced the
suggestion that the frictional head losses could be concentrated at one
point in the pipe, for example as if the friction loss were caused by an ori-
fice. Such an imaginary orifice is called a choke or throttle. Schnyder
suggested placing the orifice at the reservoir end whilst Bergeron suggested
placing it at the downstream end just upstream of the valve. Both the
Schnyder and the Bergeron techniques produce relatively accurate
answers for one point on the pipeline - Schnyder's method gives correct
answers at the valve and Bergeron's at the reservoir. Modern techniques
permit the use of any number of throttles located at uniform spacings
along the pipe. They give more accurate results but require a very con-
siderably greater effort to perform. Systems in which most of the applied
head is used to overcome friction require the use of a relatively large num-
her of throttles if friction is to be well described but such analyses are
rarely performed adequately because of the heavy commitment of time
and effort involved. Knowledge of the pressure head at points along a pipe-
line's length may be very important as the largest and smallest pressures
may occur at points other than at the ends. The lowest pressures can, in
some circumstances, fall to vapour pressure {but not below it) and then
local boiling of the fluid occurs causing two phase flow. This phenomenon
is usually called column separation and this title suggests that at the point
at which vapour pressure occurs two vertical faces of fluid form which
move away from one another leaving a vapour filled space in between.
This picture of the events implies that the two vertical faces after a period
of separation will later accelerate towards one another, finally impinging
upon one another and so causing a very high pressure rise. The author
does not believe that this sequence occurs at highpoints in a pipeline con-
structed to an engineering scale although it may do so at closed valves and
other closed ends. Instead, he believes that a foaming mass of liquid is
generated when local pressures fall to vapour pressure and two-phase
flow then develops with the formation of a free surface. The cavity so
created does not usually occupy the diameter of the pipe and opens and
closes by surface wave action. No significant pressures are created by
the closure of the cavity as the interaction of free surface waves cannot
generate pressures of magnitudes of engineering importance. The author
has generated the 'column separation' phenomenon in his laboratory but
has been unable to detect any pressure transient that could be ascribed to
the collapse of the cavity. Graphical methods can predict the likelihood of
column separation but are not likely to produce accurate predictions of
the pressures that it causes.
The prediction of extremely low pressures in pipes may be as impor
tant as that of high pressures since a running buckle failure or simple col-
lapse of the pipe may occur in near-vacuum conditions
The method of modifying the analytical technique to include friction
is best demonstrated by describing Schnyder's solution (see figure 2.19).
\,......_:s_ _ _AC><]
.} Figure 2.19
The throttle is located at the reservoir end of the pipe just downstream of
the reservoir and B 1 is located just downstream of it. The orifice is assumed
to cause a head loss of 4{~~ , i.e. fLq 2 /(3d 5 ) where L and dare in metres
2
and q is in cubic metres/sec. This expression can be written as

h =kq 2 where k =fL/(3d 5 ) {2.78)
Thus at B 1 the potential head h = H5 - kq 2 {2.79)
H.~ is the height of the reservoir water surface level above the valve. By
h
characteristic of point B'
q
Figure 2.20
plotting kq 2 onto the h - q diagram as shown in figure 2.20 and subtracting

the ordinates of this graph from the reservoir characteristic the frictional
reservoir characteristic for point B' is obtained. The valve characteristic
1/J lines can now be plotted onto this graph and the analysis performed.
See figure 2.21.
q
Figure 2.21
Note This method of including friction is the analysis obtains the accurate
steady state flow and head (at points A and B) and also produces frictional
attenuation as shown by the spiral around the intersection of the reservoir
characteristic and the ordinate through the origin.
The Bergeron method will next be demonstrated. The throttle is now
located at the valve. At A the head "' quantity relationship is as defined
earlier and is described by the 1/J line. At A' the head is increased by the
friction loss of the pipeline which is assumed to be concentrated at the
downstream throttle. Therefore the A' characteristic must lie above the
1/J characteristic by the amount kq 2 So plotting the kq 2 graph onto the
h "' q graph, as for the Schnyder analysis, and then adding the kq 2 ordin-
_ B_ _ _ _ . _ , A ' A
) r--- : [><]
downstream throttle
Figure 2.22
Figure 2.23
ates to the t/1 characteristic ordinates gives the A' characteristic. This method
describes the steady state conditions accurately, i.e. at A and B but does
not produce frictional attenuation after valve closure is complete.
The use of multiple throttles

This technique is due to Angus 5 6 . It will be illustrated by the use of two
throttles. Assume two throttles located at B, the reservoir end, and at C,
the midpoint of the pipe. Half the pipe's friction loss is assumed to occur
at Band half at C. Throughout the section AC" and C'B' waveriders can
travel without experiencing friction and eagre lines can be drawn which
define conditions in these pipe segments. By plotting two friction curves,
c' c"
Figure 2.24
one for! kq 2 and the other for kq 2 the friction from one throttle is given
by the he= 1 kq 2 line and for both by the he= kq 2 line. The ordinates of
these two graphs can next be deducted from the Hs ordinate to give the
graphs of Hs- t kq 2 and Hs- kq 2 The valve characteristic lines must next
be drawn, i.e. t/lo, t/lo.s, t/lt.o, t/lt.s, t/12.0, t/12.s and t/13.0
The intersection of the full open valve characteristic t/1 0 with the h =Hs-
kq 2 line defmes steady flow conditions at the valve, i.e. A 0 and also defines
conditions just downstream of the midpoint throttle, i.e. C0 " and C~~s.
Just upstream of this throttle conditions are defined by the intersection of
a vertical line through A 0 with the h = Hs- 0.5 kq 2 line, i.e. at C~. The
conditions at Cwill not change until 0.25Tafter valve closure starts soC~
and C~.25 are coincident. Conditions at B' will be the same as those at C~
53
q
Figure 2.25
until a time of O.SThas elapsed soC~, C~_25 , B~ , B~.s are coincident. Con-
tinuation of the vertical line to intersect with the reservoir characteristic
will give conditions upstream of the upstream throttle so defining B 0 --+ Bo.s.
By starting a waverider off from C" at time 0.25T travelling downstream,
an eagre II will be obtained. The waverider will arrive at the valve at time
0.5T so the intersection of the eagre line with the 1/10 5 line defines A0 .5
Waveriders starting from A at time O.STand from B' at time O.STwill
meet at Cat time 0.75T. The head at c" must be less than that at C'by an
amount of! kq 2 so point c~: 75 on the eagre I coming from A0 .5 must be
so chosen that point Cd. 75 on the eagre II from B~. 5 is vertically above it
and the distance C~. 75 - cd:1 s isi kq 2 where q is the current q. This means
that the process is one of trial and error but in actuality the process of
selection of the two points is quite rapid.
Starting two eagres from C, i.e. an eagre I from Cd. 75 and an eagre II
:o
from c~:75 gives B 1 by the intersection of the eagre I and B' characteristic,
and A1.o by the intersections of the eagre II and the 1/11. 0 valve characteris-
tic.
Starting an eagre I from At.o and an eagre II from B{.0 produces results
for c,:25 and c;: 25 . As before, c;_ 25 must be vertically above Ct2sby the
amount! kq 2 , q being the relevant q value and the estimation process used
above must be used again to define c,:2s and c,:~s. The process can be con-
tinued to fill in the remainder of the diagram.
The process described above can be used to analyse as many throttles
as required but the higher the number the more the work involved. If
throttles are to be inserted at quarter points along the pipeline length there
must be a corresponding increase in the number of 1/1 curves and so the
work involved increases approximately with the square of the number of
throttles. In a very long pipeline four throttles will not be sufficient to
give an accurate description of frictional effects.
3 Boundary conditions for use
with graphical methods
3.1 Introduction
In chapter 2 the fundamental techniques of graphical analysis have been

demonstrated but, throughout, the downstream boundary condition was
always assumed to be that of a closing valve. In this chapter, methods of
dealing with other boundary conditions are described.
3.2 Pumps
When pumps start up, transients are generated, the pump operating point
moves along its characteristic curve attenuating these transients until even-
tually the steady state operating point is achieved. After a period of steady
state operation the pump will be switched off (pump trip) and this will
generate a negative transient which travels to the downstream end where
it will be negatively reflected, if the downstream boundary condition is
a reservoir, and return to the pump as a positive transient. If the pump is
fitted with a nonreturn (or reflux) valve the positive wave will try to gener-
ate a reverse flow which will close the reflux valve producing a closed end.
This will create a positive reflection and a high positive transient will be
transmitted back downstream.
As the starting transient cannot be larger than the no flow head (or
closed valve head) of the pump it is usual in graphical analysis to ignore
this transient and only consider the transient caused by pump trip.
The power supplied to a pump is used to accelerate the pump, to supply
the water power produced by the pump and the energy lost within it.
Eventually the water power generated by the pump plus the energy lost
within it equals the power supplied to it and then the pump ceases to acce-
lerate and steady state is reached. At pump trip the kinetic energy of the
rotating impeller and motor is all that is available to continue pumping. This
energy is thus rapidly reduced by the pumping of water and so the pump de-
celerates until finally it comes to rest. The rate at which energy is absorbed
55
from the rotational energy of the pump impeller, the motor armature and
other rotating masses is P and is given by
P=- wqH (3.1)

E
where His the head across the pump, q is the flow, E is the efficiency of
the pump. The negative sign indicates that power is being absorbed from
the pump.
Note wqH is the water power delivered by the pump but the power P ab-
sorbed from the pump and motor must be greater than this to supply losses
occurring within the pump, so it is necessary to divide by the efficiency
E, not multiply by it.
Now P=Tr?. (3.2)
where in this context T denotes torque and n denotes angular velocity in
radians per second,
but T=Ir?. (3.3)
wher~ I is the moment of inertia of the pump and motor's rotating parts
and n is the angular acceleration in_radians per second squared. (Note As
Pis negative, Twill be negative son will actually be a deceleration.)
Thus P=Inn (3.4)
so n=In
.!!._
Now ,. . = 2rrN, d A = 2rr (N2 - Nd

~~ 60 an H 60 At
where N 1 is the rotational speed in revolutions per minute of the pump at
the start of the At time interval and N 2 is the rotational speed in rev min- 1
at the end of the At time interval.
So (3.5)
PAt
N2 =N, + ( 4rr2 )
(3.6)
I 3600 N,
N =N _ 3600 wqHM
(3.7)
2 1 4rr 2 IN1E
where E is the pump efficiency at the flow q and speed N 1 Thus equation
3. 7 gives the pump running speed N 2 after the lapse of a time at the begin-
ning of which the running speed was N1 rev min- 1.
Consider the characteristic curve of a pump. At steady state the charac-
Boundary conditions for use with graphical methods 57
teristic curve is that given in figure 3.1, i.e. curve N0 where N 0 is the steady
state pump running speed.
After pump trip the pump slows down and finally stops. Consider
conditions at, say, pipe period intervals T, i.e. tJ. twill equal T. The running
speed at the end of any tJ.t interval will be given by equation 3. 7 where
M= T.
The dimensional analysis of rotodynamic machines gives the following
result:
.( q gH Nd 2 H k ) _
(3.8)
J\Nd 3 ' N 2 d 2 ' p---;;- 'd'd' E - O
where dis the diameter of the impeller, Jl is the dynamic viscosity of the
fluid, k is the roughness of the impeller surfaces and the other variables
are as already defined. This result is a standard result and is given in most
Figure 3.1 Q
textbooks on engineering fluid mechanics. In the author's book An Intro-

duction to Engineering Fluid Mechanics 14 , a complete derivation is given.
The k/d term has no effect because as the pump speed changes k and d re-
main constant. The term H/d is found to have a trivial effect in practice
and provided that the flow in the impeller remains turbulent, i.e. pNd 2 !JJ.
remains large whilst flow continues during the pump run down, the
pNd 2 !JJ. term has little effect also. As flow only continues during the
period that the running speed of the pump remains above that necessary
to cause pumping against the head applied to the pump and pNd 2/JJ. re-
mains large during this period, it is considered reasonable to ignore variations
in the value of pNd 2 /JJ. in this context.
The result therefore reduces to
q gH )-
[ (ii(j3 N2d2' E - 0 (3.9)
For a particular pump dis constant and g will not change so the following
result applies:
(3.10)
-- --
h
---......
system characteristiC
q
Figure 3.2
At full pump speed N 0 , the usual technique of steady state analysis can be
used- see figure 3.2.
From this graph hso and qso can be obtained.
It is necessary to know the efficiency -q curve of the pump (see figure
3.3). This curve can then be used to obtain theE- q/Nd curve which is
dimensionless presentation of theE- q curve but, as d is constant for a
given pump, theE- q/N plot can be used instead (figure 3.4).
Figure 3.3
After a time of one pipe period T the pump running speed will be given
by equation 3.7, i.e.,
N =N. _ 3600wqsohso T
(3.11)
' o 47T2 INoEso
Now qso =qTo (3.12)

No M
and -hso = hTo
- and E-r:0 =E (3.13)
Ni Nt so
so it is very easy to calculate qT0 and hTo and this establishes one point on
the pump characteristic for speed N 1 at time T. A second point can then
E
q
Figure 3.4 7V
be chosen on the steady state pump characteristic lzs1 , qs 1 and the corres-
ponding point on the N1 curve can be calculated, i.e.
qsl = qTl {3.14)

N0 N1
and
hsl - hTl
{3.15)
N~- Nt
and the process repeated for still further points to establish the pump
characteristic for speed N 1 at time T.
Mter the lapse of a further T period
{3.16)
where qT, hT and ET are the instantaneous flow head and efficiency values
at time T. The hT and qT values are obtained from the first step of the gra-
phical analysis, and the ET value from the efficiency -q/N curve.
The process can be used to develop the head - flow curve for the
second time step, i.e.
qs1 = qT1 _ q2T 1
(3.17)
No N1 - N2
and
hsl - hTl - h2Tl
{3.18)
NJ- N{- N?
q2T1 and h2T1 can be obtained. Similarly other points on the pump charac-
teristic

for speedN2 , e.g. q2T2 , h2T2 ;q2T 3 , h2T 3 ;q2T 4 , h2T 4 , etc, can be
obtatned. A complete set of pump characteristic curves for the necessary
number of speeds can thus be derived.
There are a number of assumptions made in this method.
(I) The time interval Tis assumed to be sufficiently small for an initial
value finite difference calculation to be adequately accurate. If Tis large
then O.STor 0.25Tvalues should be used instead.
(2) The efficiency ET taken from the efficiency -q/N curve is not com-
pletely accurate. The efficiency curve was obtained for speed N 0 and the
assumption is made that this curve remains unchanged as the speed changes.
This would be correct if the pNd 2 /JJ. group had no effect upon the efficiency
but in fact, of course, it has.
(3) The~ and~ groups which have been used to establish points on
the characteristic curve which is being developed can strictly only be used
if E is the same at both speeds but, as in assumption (2), this is not perfectly
true so the characteristic curves produced are also not perfectly accurate.
Figures 3.5a, 3.5b, 3.5c demonstrate how the graphical analysis is per-
formed. Knowing hT 1 and qT 1 the N2T curve can be obtained as already
discussed. The completed diagram is given in figure 3.5c.
The case illustrated could only arise in the circumstance of a pump set
having a relatively large inertia delivering to a short delivery pipe. In such a
case the pipe period Tis very small and the run down period of the pump is
relatively long.
If the pipe period Tis large in terms of the pump run down it will be
necessary to plot pump characteristics for fractions of the time period. It
will then be possible to obtain pressures and flows at intermediate pipe
points in the way which was illustrated in chapter 2. In certain texts the
series of pump h - q characteristics for various pump speeds are derived
using a technique which does not in any way depend upon the transient
heads in the pipeline. It seems to the author that at the instant when
pump trip occurs, the initial parameters of steady state- h 5, q 5 and 5 -
are available so it is possible to predict the rate of pump speed decrease
and so obtain the pump speed at a time !:1t seconds later. After !:1t seconds
the pump speed is therefore known but the head, flow and hence the ef-
ficiency are determined by the interaction of the eagre I line with the
pump characteristic, i.e. the precise flows, heads and efficiencies at the
instant when the next prediction of the pump speed (for At later) is to be
made, are dependent upon the waterhammer circumstances. Therefore it
would appear that a technique which does not take this waterhammer
circumstance into account cannot be correct. The method advanced in
this book is free from the defects listed above and it is simple and straight-
forward to use.
3.3 Four quadrant pump operation
If no reflux valve is fitted upstream of the pump it may happen that pres-
sures downstream of the pump may rise during transient operation to
such values that flow may be forced backwards through the pump even
though it is still running forwards. Again, the pressure may fall so low
that flow is passed forwards through the pump tending to speed it up or
steady state pump characteristic
h
throttle
(a)
coordinates q2T h 2 T
From q2r and h2 r the N 3 r curve can be established
(b)
(c)
Figure 3.5
reduce its rate of run down. In other words, the pump may operate in (a)
pumping mode, (b) dynamometer mode or (c) turbine mode. All three
modes may occur both when the flow is forward or reversed and when
the pump rotation is forward or reversed.
The number of pumps that have been tested in all modes of operation
during transient conditions is extremely limited and the analyst will not
be able to obtain the complete characteristic curves of the pump he is
wishing to analyse. It is possible to estimate suitable curves from the few
complete characteristic curves that are available but this is best done
using techniques that are not well suited to graphical analysis. For this
reason the graphical techniques are not presented here although, of
course, they exist; a technique of dealing with this problem is described
in chapter 5.
3.4 Surge tanks
Surge tanks are devices connected to pipelines to turn the high frequency,
high pressure transients into low frequency, low pressure, mass oscillations.
The simplest examples is that of a relatively large diameter vertical pipe
connected to the pipeline, situated near to the hydraulic control which is
generating the transients which it is wished to suppress. When a hydraulic
reservoir static water level
reservoir
Schematic layout of a typical hydroelectric power scheme
Figure 3.6
control (such as a valve, or turbine gate or pump) attempts to impose a

rapid velocity change upon the fluid, the fluid can enter the tank experienc-
ing almost no retardation. As more and more fluid enters the tank the level
within it rises, slowly applying a decelerating head to the fluid in the pipe-
line. As relatively large volumes of fluid must enter the surge tank to in-
crease its level, the period of the oscillatory surge pressures set up in the
pipeline is long. The decelerations are therefore low so the pressure mag-
nitudes are greatly reduced. However, to prevent completely pressure tran-
sients from passing up into the pipeline a surge tank of infinite cross sec-
tion is needed. Even so, surge tanks of normal dimensions are capable of
reducing pressure transients to very small values. A technique of estimating
the magnitudes of such transients is outlined later in this chapter.
When the hydraulic control in question is a turbine, a very important
function of a surge tank is to supply fluid during the early phases of a gate
opening operation. During the early part of an opening operation of the
turbine gate the fluid in the pipeline will be travelling too slowly to sup-
ply the demand made by the turbine but fluid drawn from the surge tank
can meet this demand with a consequent decrease in the level within it.
This decrease will cause the fluid in the pipeline to accelerate and eventu-
ally the flow in the pipeline will equal the turbine's demand.
The fluid level in the surge tank will be set into oscillation and it is
possible that its frequency or a sub or superharmonic of it may match
the natural frequency of the governor mechanism (or one of its harmonics).
If this should happen the governor may go into resonance and the system
will then be inoperable. For this reason variant forms of surge tanks have
been developed which are capable of damping any applied oscillation.
3.5 Types of surge tanks
Figures 3.7a, 3.7b, 3.7c, 3.7d, illustrate the four major types of surge tanks.
The simple surge tank has already been described. It has the following
advantages:
(I) It transmits very small transients.
{2) It behaves extremely well during opening phases of the downstream
control.
{3) It has a long period.
There is minimal attenuation of the mass oscillation in this type of tank.
The choke ring surge tank is equipped with a short riser entering the
main tank and the upper end of the riser is sealed. Fitted into its vertical
periphery are valves which are forced open during the rising surge providing
large orifices through which flow can enter the tank; during the falling
surge the valves close but in the valves small orifices are fitted which pro-
vide a frictional resistance to outflow from the tank. This type of surge
tank has the quality of attenuating any surge rapidly because of the turbu-
lent dissipation that occurs as flow passes through the orifices and generates
turbulence in the surrounding fluid masses. This attenuation has to be paid
for as the orifices cause transmission of transients up the pipeline.
The Johnson differential surge tank has a central riser of small cross
section. This means that flow up the riser causes a rapid but limited increase
of pressure in the pipeline when the turbine gates close. The top of the
riser is usually set at reservoir static level so the pressure rise is small. The
rapid rise causes faster retardation of flow in the pipeline than would occur
if a simple surge tank had been installed. Once the riser is overtopped a
(a) (b)
Simple surqe tank Choke ring surge tank
(c) (d)
'---- 0 ( r-=-
.k>hnson differential surge tank Pressurised surge tank or
air vessel
Figure 3.7
weir flow occurs over the lip of the riser filling the annular portion of the
tank. Some flow also enters the annular portion through the orifices at
the base of the riser. The pressure in the pipeline remains approximately
constant during this phase. Once the level in the annular portion reaches
the lip of the central riser, the surface rises uniformly and slowly across
the entire cross section of the tank. During a falling surge the level drops
until the level reaches that of the top of the central riser. The level in the
central riser then drops rapidly and flow occurs through the orifices in the
base of the riser. This causes the level in the annular space to drop relatively
slowly. The graph of pressure in the pipeline against time is thus a complex
shape made up of discontinuous portions.
This tank attenuates surges well, has a complex wave form which is less
likely to resonate with the turbine governor and has a shorter period than
the simple surge tank. It will transmit higher pressure transients to the
pipeline than will a simple surge tank and will not attenuate surges as well
as the choke ring surge tank. Like the choke ring surge tank it will not
behave as well as the simple surge tank in supplying flow during the
starting phase of a turbine's operation.
The pressurised surge tank is nothing more than an air vessel. It is used
when any other type of surge tank would have to be excessively high, or
when, for strategic reasons, the surge tank must be buried inside a moun-
tain. Water dissolves air; the higher the air pressure the more it can dissolve,
i.e. up to approximately 2% by volume per atmosphere. Consequently,
pressurised surge tanks need to be fitted with air compressors which can
maintain the necessary volume of air in the surge tank. Automatic equip-
ment is necessary to control the compressors and such systems need to be
duplicated to ensure certainty of operation. This is important because if
the surge tank loses all its air it ceases to be effective and full pressure
transients will be generated wltich may severely damage the pipeline.
Variations in design of surge tanks are frequently encountered. A com-
mon variation of the simple surge tank design is the use of variable cross
sections and horizontal galleries to increase the storage of the tank, as
shown in figure 3.8.
I
I
I
Figure 3.8
The analysis of surge tanks falls into two parts:

(1) The analysis of the ability of the surge tank to minimise the trans-
mission of transients into the upstream pipeline.
(2) The analysis of the mass oscillation of the fluid in the tank and
pipeline.
3.6 Transient analysis of surge tanks
At the instant that the turbine (or other hydraulic control) operates, the
situation will be as illustrated in figure 3.9.
During the very short time that pressure transients of significant size
exist, the fluid level in the tank will not alter very much so L 2 can be treated
as a constant without much error. The problem thus reduces to the analy-
sis of a three-way junction. The pressure at the top of the surge tank must
be constant at atmospheric pressure so it can be assumed that the surface
acts as a reservoir. The problem can then be analysed in the way described
for the three-way junction pp. 46 to 48 in chapter 2. Although pressure
transients so transmitted may be small it is possible that they may be large
enough to excite 'organ piping' or resonance in the upstream pipeline
which can cause high pressures at its nodes and these may be capable of
fracturing the pipe.
Figure 3.9
3.7 Mass oscillation of surge tanks
In this type of analysis it is assumed that velocity changes are so slow that
their effects are propagated throughout the pipe length in a relatively
negligible time. In other words the instantaneous velocities at all points
in the pipe are assumed to be the same, i.e. 'rigid column' theory applies
(chapter 1). In chapter 6 a technique of analysis will be described in which
this assumption need not be made but, as this depends upon the use of
the computer, the following analysis may be found useful when a computer
is not available.
't - ~~c-'- t -~ eswc
- - : .J L
Figure 3.10
At an instant t seconds after shut down of the downstream flow control

the situation is as illustrated in figure 3.1 0. The velocity in the pipeline is
then v, the level of the surface in the surge tank is Z below the reservoir
static water level (RSWL) and the friction head loss is (4fL/d)V.vlj(2g),
i.e. Ollvl where C = 4fL/(2gd), d being the diameter of the pipeline).
(Note Z is measured positively upwards.)
If the flow were steady at timet, at velocity v, the value of Z would
equal -Cv 2 but as the surface in the tank is higher than this, a deceleradve
head is applied to the pipeline, i.e. a head of Cvz- ( -Z) (remember, Z
has negative magnitude),
Ldv
so Cv2 +Z=--- (3.19)
g dt
so ~; = -f (Cvlvl + Z): the dynamic equation. (3.20)
The use of Cv lv I instead of Cv2 allows for the reversal of the frictional head
with flow reversal.
The continuity equation is equally simple to derive: let the flow through
the hydraulic control at time t be Q1
dZ
then av=Ad1 +Q1 (3.21)
dZ av-Q1 h . . .
so dt = A : t e conhnmty equation (3.22)
These two equations can be combined into a second order differential

equation but this equation has no analytic solution. If Q1 = 0 and friction
is ignored the equation of simple harmonic motion results, i.e.
~~ = -f Z from equation 3.19 (3.23)
and v = ~ ddZ from equation 3.22 (3.24)

a t
Differentiating equation 3.24 with respect to time and substituting for

dvjdt from equation 3.23 gives:
d2Z+ag Z=O (3.25)
dt 2 AL
The angular velocity il of this SHM isJII. so the period is:
T= 21TJ[f- (3.26)
The amplitude is then readily obtained as follows.

As ril = Vs1 where Vs is the initial steady state velocity in the pipeline
and r is the amplitude
r = Vsa/A (3.27)
~
~AL
(3.28)
The result for the period is very good and produces good results even when
friction is present. The value given for the amplitude is only approximately
correct and becomes less and less accurate with increasing friction. How-
ever, it overestimates the peak surge and can be used to obtain a quick ini-
tial estimate of its magnitude.
The equations for choke-ring surge tanks can be derived by modifying
the dynamic equation to include for local losses created at the choke ring's
orifices by increasing the friction term by kvrlvr lj2g, where Vr is the velo-
city in the riser and equal to (av- Qt)/ar If the choke ring is equipped
with orifices of different sizes, then the k value must be adjusted when the
surge direction changes. The continuity equation for the choke ring tank is
the same as that for simple tanks.
The Johnson differential tank has to be analysed differently for its six
different modes of operation during a surge. At first the water is rising in
the central riser and flow enters the annular tank through the orifices in
the base. When the level rises to the top of the riser, flow spills into the
annular tank as a weir flow. When the level in the annular tank rises to
equal that in the riser the level across the cross section rises uniformly
with almost negligible flow through the base orifices. Flow through the
base orifices in either direction involves local losses and these must be in-
cluded in the analysis.
When the surge starts to fall the level across the entire cross section falls
uniformly and during this phase no flow occurs through the base orifices.
When it reaches the level of the top of the riser, the level in the riser will
drop rapidly and flow out of the base orifices into the riser will occur,
dropping the level in the annular portion of the tank. Once this happens
the analyst will have to consider the annular section and the riser sepa-
rately.
The analysis of all types of surge tanks is based on finite difference
methods so it is simple to write the groups of dynamic and continuity
equations applicable to the various phases of the tank's operation. These
are used, as appropriate, as the analysis passes through each phase.
3.8 Pressurised surge tanks or air vessels
In steady state the value of the pressure in the tank
(3.29)
Pairs must be absolute so Pa- the atmospheric pressure- mvst be included.

In this case Z 5 i=- Cv5 lv5 l in steady state. Z 5 is measured relative to the
reservoir static water level and is taken as positive upwards.
After t seconds from the closure of the hydraulic control the level will
have risen in the tank. As Z increases, the air will be compressed to Pairt
s
z
Figure 3.11
according to a polytropic process and it is usually assumed that the index

is 1.2 so
Pairs v.airs12 -- pairt v.airt1.2 (3.30)
Thus (3.31)
where Vair denotes the volume of air in the tank.

If the tank has a uniform cross section Vair is proportional toy (see
figure 3.11 ).
:. Pairt = ( y:
y )1.2
Pairs (3.32)
P.. -P.
Now atrt a= the height of a water column equivalent to the gauge
w
pressure in the tank = hairt
whairt + Pa = (ysfYt { 2 Pairs (3.33)
:. hairt = (ysfYt) 12 ( -(Zs + CvsiVsi) + ha) -ha (3.34)
where ha is the head equivalent to atmospheric pressure. Remember that
in the convention used, Z is negative downwards (S is also taken negatively
downwards),
but Ys=S -Zs (3.35)
and Yt=S-Zt (3.36)
so hairt = G=~;) 12
(-(Zs + Cvslvsl) + ha)-ha (3.37)
The dynamic equation thus becomes
(3.38)
and the continuity equation (as before) is
dZ av-Qt
-=--- (3.39)
dt A
So the equations of mass oscillation of an air vessel are:
dZ=av-Qt
(3.40)
dt A
(3.41)
hairt = (~ =~:) 12
( -(Z5 + Cv5 lv5 1) + ha)- ha (3.42)
Although the index 1.2 is in common use, different analysts have suggested
slightly different values. The value 1.0 gives an isothermal process and 1.4
gives an isentropic process but the actual process must be polytropic.
3.9 Methods of integrating the surge tank equations

There is no analytic solution to the equations. Numerical methods are the
only available techniques. Such methods can be performed by hand or
preferably by computer.
Consider, for example, the integration of the pressurised surge tank
equations. At a time t the velocity v will be Vt, the Z value will be Zt, the
hair value will be hairt' etc, and these values will have been established from
previous steps of integrations
(3.43)
so Zt+~t = (avt~ Qt)~t + Zt (3.44)
~v g
At =- L( Zt + Cvtlvtl + hairt) (3.45)
At
so Vt+~t =- g L (Zt + Cvtlvtl + hairt) + Vt (3.46)
and
(3.47)
This method is the simplest form of initial value integration. There are
many more refined methods but this one will work quite well if ~tvalues
are made sufficiently small and a computer is used to perform the arith-
metic processes. Mid-value iterative methods can be used but although
these permit larger ~t intervals to be used they may involve as much com-
puter time.
Corrector-predictor methods and Runge-Kutta techniques can be ap-
plied to this problem. Such refinements are beyond the scope of this book
and the reader should study the mathematical literature if he wishes to
employ such techniques.
If the integration is to be performed by hand it is best done using a
tabular layout.
4 The method of characteristics
4.1 Introduction
The Schnyder-Bergeron technique has, until recently, been considered the

best method of solving transient problems. It is very limited and subject to
error, however. The work involved in analysing networks containing a sig-
nificant number of pipes, say more than eight or nine, is unacceptable in
any engineering circumstance. However, there are more complex reasons
for considering graphical methods to be inadequate.
(I) Almost all liquids encountered in engineering practice contain small
volumes of air in bubble form so, when pressures in the liquid increase,
the bubbles decrease in volume and when the pressures fall they expand.
In effect, the bulk modulus of the liquid changes with pressure and as the
wavespeed depends upon the bulk modulus the wavespeed varies with
pressure. Such variations in wavespeed can be very great. For example,
the wavespeed may fall from 1300 m s- 1 to as low as 100m s- 1 due to
the presence of an air bubble content of 0.01% solely due to pressure
changes. This means that the eagre lines on a graphical analysis instead of
being straight should really be complex curves. These curves cannot readily
be calculated so the author cannot see how the variable wavespeed effect
can be included in a graphical analysis.
(2) Not only do liquids contain free bubbles, but such liquids as oil
and water, can carry significant volumes of dissolved gases. Water can carry
2% by volume of dissolved air per atmosphere of pressure and oil can
contain very much larger volumes of dissolved gases depending upon its
geological place of origin. During the passage of a low pressure transient
this dissolved gas may come out of solution in the form of free bubbles so
greatly increasing the quantity of bubbles in free form and decreasing the
wavespeed even more. Again, the graphical method cannot include this
effect.
(3) At the present time complex servocontrolled valves are being fitted
into oil pipelines and designers of long water pipelines are also using such
devices. The description of the behaviour of such equipment in a graphical
72
The method of characteristics 73
method is extremely difficult due partly to the difference in time scale be-
tween the time of operation of such valves and the pipe period and partly
to the difficulty in satisfactorily describing the multi-variate nature
of such a valve's operation graphically.
( 4) The run up to steady state of a pumped network may not cause as
large transients as those created by pump trip but in a complex network in
which complicated valve operations may occur or in which dump tanks
empty or fill, the pump trip case may be of only relatively minor interest. In
the operation of oil pipelines pump trip is a rare event and the study of
events caused by changes of pump speed (including speed increases), opera-
tion of valves, permitting flow from or to dump tanks, etc, are the events
which are of importance. To study such pipelines many analyses are neces-
sary and the excessive labour involved in graphical analyses of so many
cases (even if possible at all) may well be totally unacceptable.
(5) Many types of boundary conditions can be mathematically modelled
but cannot be graphically represented - an example of such a boundary
condition is that of a sewage ejector which can be graphically represented
only if gross simplifications are made (see (3) above).
(6) The requirement that pipe lengths in a pipe network must all be in
relatively simple ratios is very limiting and can lead to considerable error.
(7) In complex networks there may be junctions at which as many as
fifteen pipes meet, i.e. a manifold. If the network is not to be simplified
the analysis of a fifteen-way junction must be faced. Very few graphical
analysts would be prepared to contemplate such a problem.
For the above reasons it has been necessary to find an analytic method of
solving the waterhammer problem which offers fewer constraints than the
graphical methods. To do this it is necessary to return to the fundamental
equations of water hammer which were presented on p. 26 ff.
They are requoted here:
continuity
equation
and
dynamic
equation
These equations can be integrated directly by finite difference methods

but great care must be taken in doing so as instability problems can arise.
The papers of Lax 7 and Lax and Wendrof 8 are relevant in this context and
the reader is recommended to them if he wishes to employ such methods.
The author is convinced that the finite difference integration of the charac-
teristic forms of these partial differential equations is preferable. This is
known as the method of characteristics. Before this can be demonstrated
it is necessary to derive these characteristic equations.
4.2 Method of deriving the characteristic forms of the waterhammer
equations
The waterhammer equations are a pair of quasi-linear hyperbolic partial

differential equations.
There are a variety of ways of obtaining the characteristic forms of
such a pair of equations. The method given here is a modification of a
method presented by Lister 9
A different method is used in chapter 10 of this book to illustrate
another technique.
Considering any pair of partial differential equations of the type
shown below:
(4.1)
and
au au av av
L2 =A2 ax+ B2 ay + C2 ax+ D2 ay + E2 = o (4.2)
where u and v are dependent variables and x andy are independent vari
abies; A" A 2, B,., B 2, C., C2, D" D 2, 1 and E 2 are all continuous known
functions ofu, v,x andy.
The condition that
A1 _ B1 _ C1 _ D1
A2 - B2 - C2 - D2
in part or in total is prohibited.
Consider a combination of L 1 and L 2 such that
L =L 1 + XL 2 (4.3)
Then
ov av
(C1 +X C2) ox+ (D1 + XD2) oy + 1 + XE2 (4.4)
Let y = y (x) be the equation of a curve of which : is the tangent slope.

If u = u(x, y) and v = v(x, y) and these are solutions ofL 1 and L 2 then
au au
~=~b+~~ ~~
and
ov ov (4.6)
dv=-8x +-8y
ox oy
Now,
au au (au B 1 + A.B2 au)
(A 1 + A.A2) ax+ (B1 + A.B2) ay =(At + A.A2) 3.x + A 1 + A.A 2 ay
(4.7a)
B1 + A.B2 _ Dt + A.D2 _ dy
so if (4.8)
At +A.A2- C1 +A.C2- dx
Rearranging equation 4.8 gives

A.=A 1 dy-B 1 dx =C1dy-D,dx ( 4 .IO)
B 2 dx -A 2 dy D 2 dx - C2 dy
hence p(dy) 2 + qdxdy + r(dx) 2 = 0 (4.11)

where p=A 1C2 -A2Ct (4.12)
q = A2D1 + B2C1- A1D2- B, C2 (4.13)
r = B 1D 2 - B2D1 (4.14)
If the roots of this quadratic equation are real and different the original
pair of partial differential equations are hyperbolic. If the roots are real
and equal the original equations are parabolic and if they are complex the
equations are elliptic,
i.e. q 2 -4pr >O hyperbolic
q 2 -4pr = 0 parabolic
q 2 -4pr<O elliptic
Thus by substituting the values of A" B" C1 , D" E" A 2, B 2, C2 , D2, 3 ,
and solving the quadratic, two values of dy/dx can be obtained if the equa-
tions are hyperbolic. Denote dy/dx by~ (Greek letter zeta).
The quadratic is then
p~ 2 +q~+r=O
and the two values of~ are~+ and~- Thus two directions are defined in an
x- y space.
So, at a point x, y in the x - y space there must be two lines of gradients
dy/dx =~+and dy/dx =~-passing through the point and these conform
to the differential equations. The lines tend to become colinear when
~+-+~_which arises in hypersonic flows in gases and in highly super criti-
cal cases of free surface flow, and then the equations tend to become para-
bolic.
Having calculated the values of~+ and~_ two total differential equations
have been established, i.e.
(4.15a)
and dy = ,_
dx )- (4.15b)
Back substitution of these values of~ into equation 4.10 gives
X= At~- Bt = Ct ~- Dt
B 2 -A 2 ~ D 2 -C 2 ~
Substitution for X in equation 4.9 gives
Simplifying gives:
(AtBz- AzBt)du +(CtBz- CzBt + (CzAt- CtAzn) dv
+ (BzEt- BtEz + (EzA 1 - E 1 Az)~) dx =0 (4.16)
Substituting C2 A 1 - C1 A 2 =p from equation 4.12 gives:
Fdu + (P~+- G) dv + (K~+- H) dx = 0 (4.17a)
Fdu + (P~-- G)dv+ (K~_-H)dx = 0 (4.17b)
where F= A 1Bz- A 2Bt ( 4.18)
G=B1C2 -BzCt (4.19)
K=A1E2 -AzEt {4.20)
H=B 1E 2 -B 2E1 (4.21)
Thus four equations are available:
Fdu + (P~- G) dv + (K~- H)dx = 0 [4.17a and b)*
and
[4.15a and b]
* Square brackets indicate that the equation with this number was introduced earlier.
The two lines in the x - y space specified by dyfdx =~are called charac-
teristic lines (hence the name method of characteristics) and along these
lines the equations
Fdu + (P~- G) dv + (K~- H) dx = 0
apply.
These equations are solved by finite difference methods.
4.3 The characteristic forms of the waterhammer equations
The equations are:

ah ah c 2av av az
v-+-+--+O x- - v - =0
ax at g ax at ax
and
(see chapter 2).

Comparing these with the general forms previously quoted, x in the general
case is equivalent to X, y tot, A, to v, B, to 1' c, to c2 /g, D, to 0, , to- v az
2ftilvl ax
A2 to 1, B 2 to 0, C2 to v/g, D 2 to 1/g, 2 to ---g;r- u to hand v to v.
So p =A, C2- A2C1 = (v 2 -c 2 )/g (4.22)

q = A 2 D 1 + B 2 C1 - A 1D 2 - B1 C2 =-2v/g (4.23)
r = B1D 2 - B 2 D 1 = 1/g (4.24)
As ~ =- ipJ(ipY -~
after substitution and reduction
1
~=- (4.25)
vc
F= A 1 B 2 - A 2 B 1 = -1 (4.26)
G = B1 C2 - B 2 C1 = vfg (4.27)
K = A 1E 2 A 2 E 1 = 2fti 2 lvl/(gd) + v ~
- (4.28)
ax
H =B1 E 2 - B 2 E 1 = 2ftilvlf(gd) (4.29)
P=(v 2 -c 2 )/g (4.30)
substituting these values and reducing gives
-dh +. dv + . ( 2fvlvl dx +gvdz ~) = 0 (4.31a and b)

g g d vc cdxvc
and
dx=(vc)dt (4.32a and b)
so g_dh + dv + 2jVivi + gv dz = 0 (4.33a and b)

edt dt d c dx
and
dx
-=vc (4.34a and b)
dt
The~ :. term is small and is usually omitted.
These are the characteristic forms of the water hammer equations.
4.4 The zone of influence and the domain of dependency

At point P(x 1 , t 1) assume that the velocity of the flow v and the celerity of
the wave c are both known. Through P two lines can be drawn for which
dt I dt
dx = -+- and -d = -I - Note that the supplement of tan_1( -I -) has a
v c x v-c v-c
value of tan- 1 (-I-)
c-v

These lines are tangent to the characteristic lines of course, and they are
drawn on figure 4.1, the forward characteristic C+, i.e. RPW and the back-
ward characteristic c_, i.e. SPV.
range of influence of P
domain of dependency of P
Figure 4.1
If at P a disturbance is initiated, i.e. at location x, and at timet, the
disturbance will travel as a wave downstream at velocity v and c and up-
stream at velocity c- v. In other words, in an x - t space a characteristic
t
line is a plot of the movement of a wave travelling either upstream or
downstream as ( : = v c.
Any point lying within the shaded area on figure 4.1 (i.e. VPW) will
have experienced the effect of the disturbance so this area is called the
zone of influence of P.
Consider what happens to disturbances which originate from points
lying between RandS at time zero: say X and Y. The forward characteris-
tic from X will intersect the backward characteristic from Y at Z. Point P,
which is located at a later time than Z, lies within the zone of influence of
Z so events occurring at Pare affected by events occurring at points lying
between R and S. Thus R -Sis called the domain of dependency of P.
Any events lying outside the R - S segment cannot affect events at Pin
any way.
4.5 The zone of quiet
The zone of quiet is that area in an x - t space in which no effects of any

disturbance can be experienced. If the disturbance of flow originates at
the upstream end the zone of quiet is as illustrated in figure 4.2a, if at the
downstream end it is as illustrated in figure 4.2b and if disturbances origin-
ate from both ends simultaneously it is as shown in figure 4.2c. The
characteristics may be curved or straight. They are straight if v and care
(a) X (b) X
zone of quiet
(c)
Figure 4.2
constant for all points on the characteristic and curved if they are not con-
stant. The concepts discussed above give considerable insight into the
nature of unsteady flow, and the reader is strongly recommended to con-
sider them very carefully.
4.6 The integration of the characteristic equations
Consider a forward characteristic of slope tan- 1( -1- ) originating from

v +c,
R (in figure 4.3) intersecting at P a backward characteristic of slope
1 -)originating from S. Assume that t:.t and t:.x are so small that
tan- 1( -
v-c
6t
ore ton (v~c)
X
6x
Figure 4.3
the short segments of the characteristics RP and SP can be treated as

straight. The lines RP and SP satisfy the equations
dx
dt=-- [4.34a and b]
v c
so RP and SP have been constructed so that along them the other two
characteristic equations will apply:
+! dh + dv + 2fvlvl = 0 [4.33a and b]

- c dt dt d
Therefore along RP equation
.!_dh + dv + 2[vlvl =0 [4.33a]

edt dt d
will apply, and along SP equation
_[dh + dv + 2[vlvl =0 [4.33b]
will apply. c dt dt d
The first of these two equations applies at both R and P and the second
applies at Sand P. Thus at P they both apply simultaneously.
The equations must be cast into finite difference form if anything fur-
ther is to be done with them
AlongRP
(4.35)
and alongSP
--
c
s + Vp - vs ) + 2fsvslvsl~t -_ 0
g(h P - h ) (
(4.36)
d
If hR, vR, hs and vs are known, vp and hp can be obtained by solving equa-
tions 4.35 and 4.36 simultaneously.
By this process a pair of values of head and velocity have been obtained
at a time ~t seconds after the last known value so this technique provides
a basis for calculating heads and velocities in unsteady flows at as many
points in a pipe as required including frictional effects, convective accelera-
tions and instantaneous head gradients all of which were omitted completely
or in part in earlier methods.
A complete technique can be built up on the basis described above.
Assume that the x axis lies along the centre line of a pipeline and that
at the point x = 0 an upstream control exists (see figure 4.4). Next, assume
that at & points along the pipe, heads and velocities are known at time
t = 0. Points R and Scan be established, R from X and Y and S from Y and
Z by methods already described. Then P can be established from points R
and S. This process can be generalised over the entire x - t space. At x = 0
values of hand v or a relationship between h and v must be known or spe-
cified in some way.
I 6x I 6x I 6x X
Figure 4.4
4.7 Boundary conditions
At a boundary there is only one characteristic available - a backward

characteristic at an upstream boundary and a forward characteristic at a
downstream boundary (see figure 4.5). It is from the boundaries that un-
steadiness is initiated so the analysis of boundary conditions is extremely
important.
As only one characteristic is available it is necessary to express the opera-
tion of the boundary condition in mathematical form so that it can be
solved simultaneously with the characteristic equation to give solutions
for hp( or hQ) and vp (or vQ). The details of this process will be discussed
more fully later in the chapter.
The method described above was developed long before modern com-
puting methods were available. It involved graphical constructions and
an unacceptably heavy computing load and consequently was rarely em-
ployed. Computer programs can be written which duplicate the graphical
process and perform the solution of the simultaneous equations rapidly
R s
X
Figure 4.5
so the method could be used today without difficulty. However, it is not

used because as velocities change with time and with distance, the slopes
of the characteristic lines change their slopes, and consequently the inter-
section points are not located at regular intervals of x and t so the results
obtained from the curved network require interpolation to give values on
a regular rectangular x - t grid. Such values can be interpreted by examin-
ing results at a fixed x distance and considering their variation with time
or alternatively examining results at a fixed time and considering their
variation with x. The original irregular grid is awkward to use for this
purpose. A modification of this method is now in common use.
4.8 The method of the regular rectangular grid
The method is based upon the concept of deciding initially that results
are to be obtained on a regular rectangular grid and to then devise a
method of obtaining values that apply at the nodes of the grid. An inter-
polative process is still necessary but this is much simpler than the one
required for the conversion of the results of the Hartree method to a rec-
tangular grid basis.
Choose a value of flt which is less than !lxf(v +c). In waterhammer
calculations the author has found that a value of flt equal to 0. 95 x !:J.xjc
works well. (cis the maximum wavespeed occurring anywhere in the net-
work). If a value of flt equal to flxjc were chosen, it would follow that ifv
were positive then R would lie upstream of M (see figure 4.6) and this could
lead to numerical instability. In the author's experience, the value of the
constant 0.95 has been found to provide an adequate margin to avoid this
happening.
The requirement that flt < flx/(v +c) is known as the Courant and
Lewy 10 stability criterion. The criterion so specified automatically satis-
fies the other required condition that flt < flx/(c- v).
Some authors of papers on this subject have suggested that the criterion
should be that flt = flx/c ignoring the v term as it is felt that vis much less
than c, so the excursion outside of the domain of dependency will be
small. For pure water with no contaminating air flowing at low velocities
in steel pipes this is true. Typical values of v and c are 2 m s- 1 and 1300
Figure 4.6
m s- 1 respectively. If vis as high as 15 m s- 1 which is common in the flexible

hoses of tanker loading facilities, and cis as low as 100 m s- 1 due to the
presence of air or gas bubbles in the flow or due to high distensibility of
the pipe itself this assumption could be dangerous. There is a considerable
incentive to place the intersection of the characteristic from P back to
the previous time level, as close to the grid point as possible, i.e. RM
should be made as small a fraction of OM as possible, consistent with R
not falling outside of MO with the consequent risk of numerical instability.
The reason for this is that if RM is a large fraction of ~x there will be a
tendency to disperse the wave front, steep waves will gradually flatten
and highly curved sections of the wave will become progressively less
curved. The author has not heard of any analysis in which this occurred
which produced erroneous maximum magnitudes of heads and velocities,
however, providing that a reasonably large number of ~x intervals was
used in the pipe length. He believes that the use of the constant 0.95 pro-
duces a satisfactory compromise between the need for the characteristic
to intersect the previous time line as near to a grid point as possible and
yet leave a margin for a v and c variation. If the ~t = ~xfc criterion is
used the interpolative procedure referred to earlier will not be necessary
but the author is persuaded that the use of interpolation is worthwhile.
Referring to figure 4.6, assume that the first time level of integration
has been completed for all points in the pipe length. Heads and velocities
at M, 0 and N will then be available. Construct two lines through P of
slopes dt/dx = - - 1 - and dtfdx = - 1-- respectively. These will pass

Vo +co Vo -co
through R and S. It may be thought that the slope values would have been
. . dt 1 dt 1 dt
better approximated by either dx = and-d =---or dx =
VR + CR X VS - CS
v
P
!c P
and dxdt = - -1-but at this stage none of the variables VR, vs, cR,
vp-cp
cs, vp or cp has yet been calculated. Next, linearly interpolate between
M and 0 to obtain VR and cR and between 0 and N to obtain vs and cs.
The characteristic equations can now be applied, i.e.
g (h p - h R) + Vp- VR 2fRvRivRIM _ O
- + - [4.15]
c0 d
so _ h R - c-0 ( Vp-VR + 2fRvRivRIM)

h p- (4.37)
g d
g (h p- h )
and -- s +vp-vs+ 2fsvsivsiM -_ 0 [4.36]
c0 d
so h _h
p- S+ gco ( Vp - Vs +
2/svslvsiM)
d (4.38)
co ( 2fRvRivRIM) _ h co ( 2/svslvsiM\
h R-g vp-vR+ d - s+g- Vp-vs+ d f
(4.39)
c0 (
h R- h S +- ) co (2fRvRivRIM + 2/svslvsiM)
VR + VS - - d
g g
so vp= (4.40)
2c0
g
vp =...L (hR _ hs) + vR + vs _ _!_ (2fRvRivRIM + 2/vslvsiM)

2c0 2 2 d d
(4.41)
hp can then be simply obtained from equation 4.37 or equation 4.38.
Elsewhere in this text cR has been used in place of CQ in equation 4.15
and cs in place of co in equation 4.36. Both techniques are valid and
produce only slightly different answers.
Remembering that the initial calculation of the position of R and S
was based upon the values of v0 and c0 it may be felt that these should
now be recalculated using slopes based upon mean values, i.e.
dt
dx VR + Vp CR + Cp
-2-+ 2
to establish a new position of R and

dt
dx VS + Vp _ CS + Cp
2 2
to establish S again and the entire process iterated until two successive
values of vp and cp respectively are insignificantly different. This is simple
to do but the author has not found it necessary, providing that a sufficient
number of Llx lengths has been used. The only way to be sure that this
number has been chosen sufficiently large is to run the computer program
over and over again using progressively smaller .::lx values. When two succes-
sive values of D.x are found to give acceptably similar answers, the larger of
the Llx values can be used. In fact, analysts gain considerable experience in
judging the value of D.x to be used and do not need to use this multi-run-
ning technique. It must always be borne in mind that for the case of a
simple pipe, making .::lx equal to the pipe length gives results which will
be at least as good as a Schnyder graphical analysis using one throttle at
the upstream end, so even if only a small number of D.x lengths are used
the result will be considerably better than that produced by a graphical
analysis. The author typically uses ten .::lx lengths but in long pipelines
he uses more. The problem arising in using a large number of D.x lengths
is that of the run time on the computer. The run time is approximately
proportional to the square of the number of D.x lengths so increasing this
number rapidly increases the cost of a run. Achieving run cost economy
whilst still maintaining a high level of accuracy is very much a matter of
experience.
Some analysts use a programmed version of the graphical technique.
What such analysts are doing is using a different presentation of the
waterhammer equations. The equations in use are the same as those used
in developing the graphical technique and these are limited versions of
the full water hammer equations and, as such, they are not completely correct.
The convective acceleration term v ~~ is omitted from the dynamic
equation and the v ~~term is omitted from the continuity equation. This
involves small errors if wavespeeds are high but these become larger as
wavespeeds fall. As wavespeeds vary and can become relatively small in
flows containing free gas bubbles this effect may not be negligible.
The argument that programmed Schnyder is equivalent to the method
of characteristics is true if the above limitations are accepted, as the prob-
lem is one in four dimensions, the variables being h, v, x and t. Schnyder-
Bergeron methods look at the problem on the h - v (or h - q) plane
whilst the method of characteristics looks at the problem on the x- t
plane. In this sense there is little difference between the two methods but
the method of characteristics is far simpler to use, facilitates the mathe-
matical description of boundary conditions and does not require the
elimination of any terms in the basic waterhammer equations.
For these reasons the author is convinced that the method of character-
istics is better than the programmed Schnyder method and he believes that
in the course of time programmed Schnyder methods will be superseded
by the method of characteristics.
4.9 Other fmite difference methods

It is possible to use direct finite difference methods of integration of the
fundamental partial differential equations of waterhammer.
Consider point E (see figure 4.7): examining the continuity and dynamic
equations
-f +
H
~
.f +E +
F
!:::.t
A 8 c
X
6.x I !:::.x
Figure 4.7
and
ah +!:!. av +_!_ av + 2f<Avl = 0
at g ax gat gd
( ah\ =hH- hB, {~~) =hr- ho

at }E 2!:::.t \ax E 2!:::.x
(av)
at E
=vH-vB
2 !:J '
(av)
ax E
=Vf-VD
2/:::,x
As ho, hr, vo, Vf, hB and VB are known, substitution of the above ex-
pressions for the partial differentials will provide two simultaneous expres-
sions for VH and hH.
The values of b.x and tlt, however, will still have to be related by the
Courant and Lewy criterion of stability, i.e.
dt
-=--
dx v +c
There are many variations of this method, such as the l..ax-Wendroff
method, the leap frog method, Amein's four point method, Liggett and
Woolhiser's method, all of which are better than the one illustrated above,
but the author is of the opinion that the regular grid characteristic method
previously illustrated is superior to all others for the waterhammer prob-
lem11.
5 Variable parameters in
unsteady flow
5.1 Variation of wavespeed
The variation of wavespeed has been mentioned in earlier chapters. Wave-

speed is dependent upon the effective bulk modulus of the fluid and this
has already been shown to be dependent upon the distensibility of the
pipe, i.e.
for bubble free liquid. Where K' is the effective bulk modulus, d, T and E
are the internal diameter, wall thickness and the Young modulus of the
pipe wall material respectively, and K is the bulk modulus of the fluid.
If gas bubbles are present in the liquid the effective bulk modulus is
reduced greatly, the effect of gas bubbles becoming greater than the pipe
distensibility term d/TE at low pressures. As the volume of gas in free
bubble form in any ax length depends upon the absolute pressure in
this length, the effective bulk modulus and hence the wavespeed must be
different for each and every Ax segment. Transients can generate very dif-
ferent pressures at different parts of a pipeline at the same time instant so it is
possible to have a waves peed of 1000 m s- 1 at one point of a pipeline and
at another a wavespeed as low as 10m s- 1
To ignore this effect can only lead to error and the author therefore
believes that its inclusion is vital if an analysis is to bear any similarity
to reality.
At the moment of writing, the subject of wavespeed variability is receiv-
ing a great deal of attention throughout the world and it is being researched
in many universities. The flow of gas/liquid mixtures is an example of two-
phase flow and little work has yet been done on the subject of unsteady
two-phase flow. The understanding of steady two-phase flow is still in-
complete although much effort has been expended upon research into it
so the problem of unsteady two-phase flow is even less adequately com-
prehended.
87
5.2 Gas evolution
The problem is further complicated by the evolution of additional gas bub-

bles from the dissolved gas present in all liquids during the passage of nega-
tive transients. When a liquid which was originally saturated with gas at
some relatively high pressure, i.e. atmospheric pressure for water, is sub-
jected to a low pressure for a long time some of its gas content will be re-
leased into bubble form and these bubbles will rise and escape from the
liquid at the free surface. The saturated gas content of the liquid depends
upon its absolute pressure. Henry's law states that the volume of dissolved
gas depends directly upon the pressure, i.e. Vg = kp where k is a constant
which reduces with temperature. For water the saturated air content is about
2% by volume at atmospheric pressure.
Thus if water at atmospheric pressure containing 2% by volume of air
is later subjected to a pressure of one half an atmosphere, I% by volume
reduced to atmospheric pressure will be released from the water in bubble
form, providing that sufficient time is allowed to elapse. This phenomenon
is a commonplace of everyday life. When a bottle of beer is first opened
it froths violently but gas (C0 2 ) bubbles continue to be evolved for a rela-
tively long time. When they finally cease being evolved the beer is said
to have 'gone flat'.
If a liquid is subjected to a high pressure it will absorb gas across its free
surface but, again, this process is highly time dependent. If the high pres-
sure is applied by bubbling gas into the liquid the process of absorption is
much faster because of the greatly increased surface area through which
gas can diffuse.
In the case of transient low pressures, the time available for gas release
and reabsorption is very much smaller than the time required to establish
equilibrium so the amount of gas generated by the passage of a low pressure
transient cannot be calculated by the use of Henry's law.
A method of modelling the process of bubble evolution will be suggested
here. It should be appreciated that the model proposed is not complete and
is based upon assumptions that cannot be fully validated.
Consider a bubble of radius r, as shown in figure 5.1.
The pressure inside the bubble will be greater than that outside of it by
an amount IJ.p because of the surface tension acting within the bubble,
and the excess pressure can be calculated by considering the force balance
across a bubble's diametral plane.
i.e. /J.prrr 2 = 2rrrr

IJ.p =-
2T
so (5.1)
T
where T is the coefficient of the liquid gas interfacial surface tension.

Thus, the excess pressure in a bubble increases hyperbolically as radius
decreases.
If it is assumed that the excess pressure /J.p causes gas to diffuse outwards
Variable parameters in unsteady flow 89
Figure 5.1
from the bubble and the magnitude of p decides how much gas must pass
into the bubbles present in the flow, it will be appreciated that at any particular
pressure (less than saturation pressure) and bubble radius, gas will be attempt
ing to diffuse into the bubble but, due to the high pressure within the bub-
ble, gas molecules will be moving so fast that many will be leaving the bub-
ble. At a critical bubble radius, gas will be leaving the bubble at a rate
equal to that at which it is diffusing into it. At any radius less than this
critical radius b.p will be even larger and gas will leave the bubble faster
than it is coming in so the bubble will diminish in radius; this will cause
b.p to increase still further and the bubble will implode. So, for every pres-
sure there will be a critical value of bubble radius below which no bubble
can exist.
If a critical radius exists below which bubbles implode, how can bubbles
ever come into existence? The answer to this question is simply that they
must come into existence with a radius greater than the critical radius. For
this to happen a mechanism must exist within the fluid which generates
such bubbles. Some authorities have hypothesised the existence within
the fluid of small particles called micronuclei upon which a film of gas is
adsorbed. If a micronucleus has a suitable geometry and size, bubbles can
form which are of a radius greater than the critical value.
Another competing hypothesis has recently been advanced which sug-
gests that either upon the surface of the containing vessel or upon the sur-
face of suspended particles there are fine cracks or surface roughnesses with-
in or upon which gas is trapped. Gas can cross the gas/liquid interface be-
cause the interface has a radius of curvature which is greater than the criti-
cal value. A bubble can generate at the crack and be released when it
reaches an adequate size for its buoyancy to tear it free (see figure 5.2).
The crack is then available to generate another bubble. This hypothesis is
attractive as any drinker of carbonated drinks will realise. In a glass of
liquid supersaturated with carbon dioxide streams of bubbles can be seen
rising, originating from defined sites on the glass surface.
It seems probable that both mechanisms operate, the micronucleus
mechanism during the initial frothing phase and the surface crack mecha-
nism continuing to operate when the degree of supersaturation has drop-
ped to a point at which the micronucleus mechanism has ceased to func-
tion. (Micronuclei are thought to be approximately eight microns in
diameter.)
For gas to evolve rapidly from a liquid, either micronuclei must operate
as coalescence points or microcracks must perform this function. In either
successive gas/liquid
interfacial shapes
Figure 5.2
case, such active devices will not be recruited until the pressure has fallen
significantly below the original saturation pressure as the unusually large
micronuclei (or microcracks) necessary to generate bubbles at pressures
just below the saturation pressure are unlikely to exist in normal liquids.
It seems probable that there will be a spectrum of particle/crack sizes pre-
sent and only a few of these will be very small and conversely only a few
of them will be large. The addition of sugar crystals to a solution saturated
with carbon dioxide (e.g. beer or soda water) produces rapid gas evolution
which seems to support this hypothesis.
The main band of particle/crack sizes present within a liquid will evolve
bubbles when the pressure falls to a value which lies within a band of pres-
sure values. This implies the existence of a 'gas release head' band which the
author suggests can be modelled as a unique 'gas release head'.
For water saturated with air at atmospheric pressure this gas release
head seems to be about 2.4 metres (8 feet) absolute. If this concept is
valid, a rapid reduction of the pressure of'a water mass below atmospheric
pressure will evolve negligible volumes of gas from solution until the pres-
sure drops below the gas release head. At this point most of the dissolved
gas will come out of solution.
Once the gas has come out of solution it is difficult to force it back into
solution by a subsequent pressure rise to atmospheric pressure but, as a
negative transient is followed by a corresponding positive pressure tran-
sient, the pressure usually rises far above atmospheric pressure and much
of the gas will return into solution. For this reason the author suggests
that, in the absence of a better theory, gas may be assumed to come out
of solution when the pressure falls below gas release head and re-dissolve
when it rises above it again. This assumption is not correct but it leads to
pressure predictions which are higher than those that will actually occur
and is therefore safe. Turbulence also has a powerful potentiating effect
upon bubble release. If a bottle of beer is shaken and then rapidly opened
its dissolved gas will be released in an almost explosive manner. Almost all
pipelines operate in turbulent flow so this effect is nearly always present.
It is fully appreciated that in many ways the foregoing concept is naive,
but it does permit an attempt to be made to describe the phenomenon.
S .3 The magnitude of variable wavespeeds and the inclusion of gas release
The effective bulk modulus is given by~, =} + :E as stated earlier in

this chapter.
Consider a mass of liquid containing a fractional volume of gas in free
bubble forme, the volume of the gas plus liquid being V.
The volume of liquid is (1- e) Vand the free gas volume Vg is eV.
Apply a pressure increment to the liquid !J.p.
The liquid volume will change to V1 where V1 = (1 - t:.pjK)(l- e)V.
The gas volume is assumed to be distributed in small bubble form and
to have an unchanging temperature equal to that of the water. Any gas
volume changes will therefore be isothermal
so peV= (p + Ap)V~ (5.2)
so Vg = p r tJ.p eV (5.3)
where v~ is the fractional gas volume at pressure p + t:.p.

The volume of the gas/liquid mixture will therefore become (approxi-
mating (1 + Apfpr 1 to (1- Apjp))
VT= (1-'i) (1-e)V+ (1-tJ.;)ev (5.4)
where VT is the total volume of gas plus liquid at the pressure p + Ap.
Expanding
tJ.p !J.pe !J.p) (5.5)
VT=V ( 1---e+ --+e-e -
K K p
The term !J.:e will be very small and can be ignored.
~ = (1- !J.p(~ +~)) (5.6)
The volumetric strain (V- VT )/V = I - VT/V= b.p(I/K + e/p) (5.7)
so K'- !J.p - 1 (5.8)

- l _ VT - _!_ + :.._
V K p
Including the distensibility of the pipe

1 1 e d
-,=-+- +- (5.9)
KT K p TE
Where KT' is the effective bulk modulus of the gas/liquid mixture including
the pipe distensibility effect.
So (5.1 0)
but w =we(! -E) neglecting the weight of the free gas so
(5.11)
c=Jwe(l-E) (___!_+~+~)
g K p TE
This result was published by Pearsall 12 .
The value ofp used must be the
absolute pressure.
Equation 5.11 is only valid for small values of as, when values of E are
large fractions of unity, the flow becomes a frothing flow which may sepa-
rate into open channel flow with a gas flow over the top of it, or it may
become a slug flow in which large bubbles possibly of diameter equal to
that of the pipe are interspersed with liquid. Again, such flows may con-
sist of a ftlm of fluid flowing around the periphery of the pipe with a cen-
tral core of gas flowing at high speed through it.
For this reason, the wavespeed equation quoted above should only be
applied to two-phase flows for which the assumption that gas bubbles con-
stitute a small fraction of the flow and are homogeneously distributed
throughout it is applicable.
The wavespeed is related to the gas/liquid volume ratio as shown in
figure 5.3. The curve shown is pressure dependent, a different curve would
c
1372m s'
0 /o gas
0 100
Figure 5.3
be produced for a different pressure, but the end values at 0% gas and
100% gas are not pressure dependent. Equation 5.11 is only valid for the
low end of the gas percentage range. It is obviously incorrect at the high
end as at 100% gas the equation predicts an infinite wavespeed.
As the wavespeed curve has a wide and relatively flat minimum it is
reasonable to suggest that once the value of c ralculated from equation
5.11 falls to this minimum value the minimum value is used. Figure 5.4
indicates how the minimum wavespeed varies with pressure (absolute).
The results are for air/water mixtures and are based upon results quoted
in a paper by Karplus 13 . No such similar results are known to the author
150
V>
E 100
"C
a.>
a.>
&1-
a.>
0
> 50
3:
250 500
pressure N cm- 2
Figure 5.4
for other gas/liquid mixtures but as the liquid's bulk modulus has almost
no effect upon the minimum wavespeed and as the gas bubble expansion/
compression is isothermal, results for air/water mixtures should be usable for
other gas/liquid mixtures without significant error. It is felt that the re-
sults obtained from an analysis in which local pressures fall well below
gas release head, so greatly expanding gas evolved from solution, will be
erroneous. The reason for such errors is that the 2% gas evolved will be
about 8% by volume at gas release head and correspondingly more at
lower pressures. Such large gas percentages invalidate a number of assump-
tions made in the derivation of the waterhammer equations and in the
development of the wavespeed equation, for example the four now listed.
(I) In the development of the waterhammer equation in chapter 2,
second order terms were ignored, in particular the term Au ap(ax ax at. How-
ever, ap(ax may not be negligible when large volumes of gas are present in
the flow.
(2) In the same development the term pfp 0 was approximated to unity
but this will not be true when large volumes of gas are present.
(3) When developing the waterhammer equation the substitution of h
for the pjw + z term was made. Normally w can be taken as the usual liquid
specific weight value without significant error but when E is large this as-
sumption is far from true.
(4) In the development of equation 5.11 the term t:.pE/K was neglected.
This term is undoubtedly small but its omission may produce some error
when Eisa large fraction of unity.
An analysis which produced results in which sections of pipe were at very
low pressures would be regarded by most engineers as suggesting that the
surge needed some type of suppression since such low pressure domains
cannot usually be accepted in engineering conditions. The analysis, al-
though inaccurate, would ihus be seen as of diagnostic value. Subsequent
analyses would then be performed to ensure that the surge suppressing
device operated to prevent the development of low pressure domains.
Near vacuum conditions in pipeline segments are not acceptable for
reasons mentioned earlier, i.e. high transmural pressures causing pipe
flattening, the possible development of running buckles, fatigue of the pipe
wall material, diametral pulsing causing high local stresses at stress raiser
points such as stones in the pipe trench or other pipes wrongly located in
the same trench. In water supply pipe networks it is not usual to accept
subatmospheric pressures in buried pipes because of the risk of pollution
from groundwater drawn into the pipeline through pipe joints.
5.4 The use of the variable wavespeed equation
A subroutine can be written to calculate wavespeeds at the pressures existing

at the interpolation points. If the wavespeed calculated from equation 5.11
should be less than the minimum wavespeed taken from figure 5.4 then it
must be assigned the valu~ of this minimum wave speed. Thus by using the
subroutine to calculate cR and cs, complete variability of wavespeed can be
included in the analysis.
S.S Vaporous cavitation
If the local pressure falls to vapour pressure the liquid will boil at the am-
bient pressure. Vapour bubbles will appear within the fluid and it will be
impossible for the local pressure to fall any further. If the pressure drop
is very rapid it may be that a large bubble may appear in the top section
of the pipe. This large bubble will be filled with a mixture of vapour and
gas. The bubble(s) will collapse when flow reverses and will disappear
leaving behind a faint haze of fine gas bubbles which will only return into
solution slowly and at relatively high pressure. This phenomenon is
thought, by some, to occur by the opening of two vertical liquid faces and
after their re-meeting it is thought that a very large pressure transient is ini-
tiated. This is called column separation. The formation of large long bubbles
can occur in long pipelines but, in the author's laboratory, the closure of
such bubbles has been shown to occur by a surface wave mechanism with-
out the development of pressure transients of any significant magnitude.
Even large bubbles close by a process of wave action on a free surface
flow, and the small pressure rises that may be generated are further mini-
mised by the suppressant effect of the compressibility of gas bubbles pre-
sent in the liquid which were previously recruited from the dissolved state
during the previous low pressure.
The process of vapour filled bubble flow, like gas filled bubble flow, is
that of a two-phase flow. One difference between vaporous and gaseous
cavitation is that whilst pressures can fall below the gas release head by
expansion of the limited quantity of gas present, this is not true of va-
porous cavitation. In vaporous cavitation the liquid's vapour is freely avail-
able to expand any bubble indefinitely without changing the pressure to a
measurable extent.
The phenomenon of vaporous cavitation can thus be dealt with by cal-
culating the local pressure head from the local absolute potential head. If
this is below the vapour head it must then be set to the vapour head and
the absolute potential head recalculated by adding the elevation of the
point above the datum level to the vapour head. This technique may not
model the behaviour of the bubble closure by surface wave action. At
a closed end or at a closed valve a bubble may form; this bubble may be
filled with vapour and gas or, if the pressure is not low enough, gas only.
If the valve closure causing the transient occurs in a very short time and the
2Ljc value is also small then such a bubble will not close by surface wave
action as one of the two necessary waves is not present. The changing
volume of such a bubble, and the pressure rise that results when it finally
closes must be calculated, the flow velocity in the liquid then being calcu-
lated (see J. Swaffieid 18 ).
The speed at which surface waves travel is relatively very small, being of
the order of I to 6 m s-1 , depending on the flow depth. An analysis by
Marsden and Fox 19 demonstrated that by combining the transient analysis
of the water hammer phenomenon with a surface wave analysis of the bubble
it was possible to obtain a highly accurate solution of the problem for bubble
closures occurring by surface wave action, i.e. in pipelines of normal engin-
eering size. The analysis takes a very long time on a computer and so is not
economic as a routine method. The use of the surface wave celerity of
approximately 1-6m s- 1 will also produce errors in the type of analysis
proposed here, however, for the following reasons. In this type of analysis
the formation of a bubble is assumed to occur over a complete .:ix length
when, in fact, it would be only a small fraction of such a lu in length,
perhaps a few metres long. Therefore, the closure of a bubble of .:ix length
at surface wavespeed would take an excessively long time and this time would
be completely erroneous. The author therefore suggests that the wavespeed
to be used when vaporous cavitation occurs should be the minimum value
relevant to the vapour pressure taken from figure 5.4, e.g. for water, the
vapour pressure of which is about 0.1 M absolute (i.e. about 10 N cm- 2 ),
the wavespeed should be 20 m s- 1 For very large .:ix values this value
would have to be considerably increased if the timing of the bubble closure
were to be accurately predicted, however. For such a case, much smaller
.:ix lengths should be used. Much more research into this subject is needed.
5.6 Calculation of friction

In the characteristic equation, the Darcy (Fanning)fvalue is required at
the interpolated points. Undoubtedly, the Colebrook-White formula is
the best available but unfortunately [is given in an implicit form, i.e.
_1 _ _
YJ-
[ 2.51 k/d
41ogto 2Y2JRe + 3.71
J
(Note that the f value commonly used in the USA is four times larger
than the Darcy [value used here.) To obtain an [value from the Cole-
brook-White equation an iterative method of solution must be used First
a reasonably good approximation for [must be substituted for fin the
right-hand side. A new value off can then be calculated. This new value
can be back substituted in the RHS and another [value obtained. The
process must then be repeated until two successively obtained values dif-
fer by an acceptably small value.
The 'reasonably good approximation' for [which is initially needed
can be obtained from the Moody formula for[, i.e.
t= o.ooi375 [1 + (20 oooj+ ~~ 6 ) ' 13] (5.12)
This is accurate to within 5% of the Colebrook-White f for Reynolds num-

bers between 4000 and 10 000 000 and k/d values up to 0.01, so it is cer-
tainly a good approximation. It may be thought that the Moody formula
is adequately accurate in its own right.
The iterative method should not need to be iterated more than five
times if a percentage difference between two successive iterations of 0.1%
is acceptable.
Of course, it is necessary to check that the Reynolds number is greater
than 2300. If it is below 2300 the laminar flow expression[= 16/Re
should be used instead.
It is fully accepted that steady state/values are not applicable to un-
steady flow but the errors produced by using steady state/values in un-
steady conditions do not appear to produce significant errors and no satis-
factory method of calculatingfvalues in unsteady flow is available.
In unsteady flow, boundary layers are thicker in the presence of adverse
pressure gradients than they are in steady flow at the same Reynolds num-
ber. If the adverse pressure gradient is sufficiently large the boundary layer
may separate from the boundary with a very large increase in energy loss.
Conversely, boundary layers are thinner in the presence of a negative pres-
sure gradient than they are in steady flow at the same Reynolds number.
This is the reason for the failure of steady state [values to accurately des-
cribe the frictional conditions of unsteady state. In gas flows this effect
can be large but in liquids the effect is usually small.
5.7 The use of variable f values
As with the variable wavespeed (5.4) a subroutine/procedure must be

written which evaluates[ at every interpolation point. The parameters
that must be supplied are the interpolated velocity, the relevant pipe
roughness and the relevant pipe diameter. The subroutine/procedure must
first calculate the Reynolds number. This must be tested to check whether
it is above or below the critical value of 2300. If the velocity is zero the
Reynolds number will be zero so the f value will become infinite, therefore
a cut off value must be supplied. It is suggested that if Re is less than 0.1
the/value is assigned the value of 160. If Re is greater than 0.1 and less
than 2300 then!= 16/Re. If it is greater than 2300 then a value [ 1 must
be calculated from the Moody formula (see equation 5.12). Usingfi then
f= [ 2.51 k/d 2
(~ 4
I
og 10 2V2!;Re + 3.71) (5.13)
A test must then be made for the difference between fi and f. If the dif-
ference is greater than the acceptable percentage error of 0.01% then fi
must be assigned the calculated value off, the program control returned
to the f calculation and the process repeated until the accuracy test is
satisfied.
5.8 Interpolation
Providing that the .:lx value is sufficiently small, a linear interpolative pro-
cess can be used.
A subroutine/procedure should be written to perform the interpolations
at every .:lx point and .:lt stage.
6x 6x
Figure 5.5
First, it is necessary to calculate RO and OS values (see figure 5.5) .

.:lt will have been calculated or read in and will be an available value. The
wavespeed at the pressure head current at 0 must next be evaluated (using
the subroutine/procedure previously developed for calculating wavespeeds ).
Then RO = (v0 + c0 ) M
and OS= (c 0 - Vo)M
At this point it is wise to check that RO and OS are both smaller than .:lx.
If either of them become greater than .:lx a suitable warning message should
be printed and the run terminated. If this occurs during a run then either
there must have been a slip in preparing data for the run or !:lt has been
assigned an excessively large value.
Then the velocities and the potential heads at R and S must be calcula-
ted.
RO
VR = v0 -(v 0 - vM) !:lx (5.14)
RO
and hR = ho - (ho - hM) !:lx (5.15)
OS
similarly vs = v0 - (v0 - vs) !:lx (5.16)
OS
and hs = ho- (ho- hs) !:lx (5.17)
As pointed out earlier in chapter 4 this process should, strictly, be iterated

once the velocity and head at P has been calculated, using a velocity and
wavespeed which is the mean between that at P and R to recalculate the
slope of the forward characteristic and to relocate R, and using a velocity
and wavespeed which is the mean between that at P and S, to recalculate
the slope of the backward characteristic and relocateS. However, this itera-
tion should not be necessary if the !:lx value chosen is sufficiently small.
S.9 The calculation of the free bubble content
The free bubble content at atmospheric head must be supplied initially

and the expansion or contraction of these bubbles together with that of
any bubbles evolved from solution during low pressure phases must then
be calculated. First, it is necessary to calculate the local pressure head.
The value of head in the characteristic equation is an absolute potential
head and consists of the local absolute pressure head plus the local ele-
vation of the pipe centre line. So the pressure head must be calculated by
subtracting the elevation from the potential head, i.e.
p = (h -z)w (5.18)
The value of the fractional bubble content e1 must then be increased by
the amount that comes out of solution, i.e. e 2 For water e2 is about
0.02 and a reasonably typical value for e1 is about 0.001. e 2 is only
evolved if the local head pfw falls below the gas release head hG.
The total gas in bubble form must then be obtained, i.e.
e = E1 + (if p ~ whG then e 2 else 0) (5.19)
so = 1 if p > WhG (5.20)
and = 1 + 2 if p ~ W X hG (5.21)
e must then be multiplied by whafp to obtain the actual fractional gas
volume expanded (or contracted) appropriately to match the local absolute
pressure p. The gas process is assumed to be isothermal.
(ha is the height of the liquid barometer or the atmospheric head, i.e.
I 0.3m or 34ft approximately for water.)
The way in which the total fractional gas volume is computed above in-
volves a number of assumptions.
(I) If the local absolute pressure falls below the gas release head all of
the dissolved gas is assumed to be evolved.
(2) If the local absolute pressure rises above the gas release head all that
portion of the gas that was originally dissolved but was evolved in bubble
form is assumed to redissolve.
The justification of these assumptions was given earlier on pp. 89-90.
The calculation of e is needed for use in the wavespeed calculation.
5.10 Evaluation of velocities and potential heads at internal points in a

pipe length
It is suggested that a subroutine/procedure called midstream be written to

calculate the potential heads and velocities at all internal points in each
pipeline. This can be called at every internal point.
The equations for velocity and head already derived are requoted be-
low.
[4.37]
Before these equations can be evaluated uR, us, hR and hs must be calculat-
ed from the interpolation subroutine/procedure, e must be calculated from
the fractional gas volume subroutine/procedure, c0 from the wavespeed
subroutine/procedure and fR and fs from the friction subroutine/procedure.
Up must then be evaluated and this value can be used to calculate hp.
6 Boundary conditions pumps
6.1 Introduction
The writing of a computer program to perform unsteady network analy-

sis depends upon an accurate description of wave transmission (described
in chapter 5) which is relatively simple to do, but the problem of describ-
ing the boundary conditions that either generate the unsteadiness or reflect
(totally or partially) pressure transients already generated is more difficult.
6.2 Pumps equipped with a nonreturn valve (forward flow only permitted)
The method of dealing with pumps differs from analyst to analyst. The
author has evolved a technique which has certain defects but has definite
advantages. It will be shown below that a centrifugal pump's characteristic
curve can be described by the equation
H = AN 2 + BNQ - CQ 2 (6.1)
where His the head developed by the pump, N is its running speed in rev
min- 1 and Q is the flow through it. A, Band Care constants applicable
to the particular pump. Only pumps with radial flow impellers can be ac-
curately described by equation 6.1 and as the impeller type becomes pro-
gressively more and more axial the equation becomes less and less accurate.
This is the major defect in the analytic technique to be described here.
In figure 6.1 a the characteristic curve designated A is for a pump equip-
ped with a highly efficient volute, e.g. a free vortex volute or a variable
velocity volute and curve B is for a pump with a less efficient (but much
cheaper) volute.
In figure 6.1 b curve Cis for a true axial flow pump and Dis the curve
which most closely approximates to it and which can be described by
equation 6.1. Providing that curveD is so chosen that it passes through
the expected steady state point of the network, only very trivial error will
result from its use as the starting transient will be only slightly wrong
100
Boundary conditions: pumps 101
h
q q
Pumps with 1adial flow type impellers Pumps with axial flow type impellers
(a) (b)
Figure 6.1
but some advantage can be gained as shown below. The derivation of the
H ~ Q (equation 6.1) is necessary as, from this derivation, the pump's
behaviour when operating as a turbine or dynamometer can be deduced.
The ability of this technique to describe turbining of the pump in the
case of forward flow is the advantage mentioned in section 6.2.
6.3 The derivation of the pump's characteristic equation
u is the peripheral velocity of the blades of the impeller at the outer

radius. Vis the absolute velocity of the fluid leaving the blades. V w is
the tangential component of the absolute velocity, i.e. the velocity of
whirl. Vr is the velocity of the fluid relative to the blading of the impel-
ler. Vr is the radial component of the absolute velocity, i.e. the velocity of
flow. Then assuming radial entry to the impeller
H = Vwu- k ~ - k Vr 2
g v2g r2g
Figure 6.2
where kv ~ is the energy loss due to friction and local turbulence in the
volute and kr ~2 is the friction loss between the fluid and the impeller
blading. See Fox 14 for a more detailed presentation of this theory.
rrDN Q
u = - - and Vr=-
60 Ae
where Ae is the flow area of the blading at exit, and Dis the external dia-
meter of the impeller.
Vw = u - V[cot "f
Vr = Vr cosec 'Y
V2 = v~ + v(
:. V 2 = (u- Vr cot r) 2 + V(
:. V 2 =u 2 - 2u Vr cot 'Y + V( cosec 2 'Y
H= (2 -kv)u 2 -(i- kv)2u Jifcot 'Y -kv Vr2 cosec 2 "f -krVl cosec 2 'Y
2g
so
H = (2- kv )u 2 + 2(kv- I)u Jif cot "f _ (kr + kr) V( cosec 2 "f
2g 2g 2g
substituting for u and Vr
H = AN 2 + BNQ - CQ 2
where
A= 2-kv
2g
(rrD)
60
2
(6.2)
B = (kv- I) cot "f....!!!!_ (6.3)

g 60Ae
2
C = (k + k ) cosec 'Y (6.4)
v r 2gAi
Hence equation 6.1 is valid.
6.4 Dynamometer/turbine operation of a pump with forward flow
It is possible for the flow through a pump to become so large that the head
across it becomes negative. In other words, the operating point on the
pump H- Q characteristic curve is far over to the right with a large flow
and negative head. In such a circumstance the pump is causing a reduction
in the energy of the flow and so is acting as a dynamometer rather than
as a pump. For the case of a pump in a normally designed network, a tran-
sient which could cause a negative head across it whilst the pump is still
being driven could only be generated by the operation of another hydrau-
lic control such as a valve. It is suggested that the head- flow equation
H =AN 2 + BNQ- CQ 2 is capable of describing the pump's behaviour
whilst operating in the dynamometer mode since, in this condition, the
only difference between its behaviour as a pump and as a dynamometer
is that at high delivery the friction loss in the volute and the impeller
becomes larger than the no flow head plus the head regained in the volute.
Thus the way in which it is working as a dynamometer does not differ essen-
tially from the way in which it is working as a pump. Almost certainly the
kv value will not be a constant and for an accurate representation of the
pump acting as a dynamometer greater knowledge of its variation would be
needed. Even so, the assumption of a constant kv value will give an adequate
representation of the pump as a dynamometer at a level of accuracy suffi-
cient for most engineering purposes. Research on this point is continuing
in the author's laboratory.
When pump trip occurs, the head across the pump may become negative,
i.e. the potential head upstream of the pump may become larger than that
downstream of it. This may be due to the development of transients in the
upstream and downstream pipes, or it may simply be due to the pump
being required to act as a booster to an existing flow in which case the
head upstream of the pump will be greater than that downstream once
the pump has been switched off. Under these conditions, with no power
being supplied to the pump from the motor, the pump will absorb energy
from the flow and act as a turbine. In effect, if negative heads occur across
the pump after pump trip the run down of pump speed will be slower and
in the case of the booster pump the pump may not come to rest if a
steady flow occurs through it. Of course, in the circumstance of a pump
which is acting as a turbine with forward flow through it, the flow is in
the opposite direction to that which is normal for an inward flow radial
flow turbine.
A simple analytic method can be developed for the case of the turbin-
ing pump.
Assume that the running speed of the pump at any instant in its run
down is N rev min- 1, that it is passing a forward flow of Q m 3 sec- 1 and
the head across the pump under these conditions isH where His negative.
The amount of energy transferred to the pump impeller per unit weight
of flow will be Vwu/g and kr ~2 units of energy will be lost in friction in
the impeller and kv ~g2 units of energy will be lost in turbulence in the
volute.
Then
also as before
Vw = u - Vr cot 'Y
v;. = Vjcot 'Y
V 2 = v~ + v?
Vwu _ kr Vl cosec 2 'Y k (u 2 - 2u Vj cot 'Y + V l cosec 2 -y)
g- -H---'--~--
2g v 2g
Vwu = _ H -(k + k ) V? cosec 2 'Y _ kv(u 2 - 2u Vj cot -y)

g r v 2g 2g
(6.5)
From equations 6.2 and 6.3
( 60
k =2-2g- )2 (6.6a)
v rrD A
and
cot 'Y _ 60gB
(6.6b)
~- (kv- l)rrD
The values of A, Band C will be available by direct calculation from the
H ~ Q curve supplied by the pump manufacturer. Using equations 6.5,
6.6a and 6.6b the value of Vwu/g can be obtained. The power supplied to
the impeller is partly consumed in overcoming bearing losses, disc friction
and windage losses in the motor. If P0 is the power consumed by the pump
at zero flow at steady running speed N~ then these bearing, disc friction
and windage losses at the current speed N can be represented by the term
P0 (N/N5 ) 3 . Thus the power P available to accelerate the impeller will be
wQVwu
g
-Po(!!_)
Ns
3
.
Then
H2 =P (6.7)
T= frl. (6.8)
where rl. is the angular velocity, rl. is the angular acceleration and I is the
moment of inertia
. T p
so rl. =I= rl.f (6.9)
2rr(N2 - Nd _ P
60~t -~
(6.10)
60NI
N _ N + 3600Pflt
2 - 1 4rr 2 N 1 I ( 6.11)
So N 2 can be calculated and may be larger or smaller than N 1 according to

the magnitudes of P0 (ZJ 3
and V;u, i.e. according to whether Pis positive
or negative.
6.5 Pump efficiency
The pump efficiency is involved in the problem during the pump run down
phase only when the pump is not turbining.
The output power of the pump or waterhorse power is simply
(6.12)
Assume that the input power can be described by the following equation:
~ = w(Def/ 2 Q + EeNQ 2 + f""efl 3 ) (6.13)
where De, Ee and Fe are constants. To justify this equation consider the
following argument. The input power is given by
Pi= wQ(V;u) +Po (6.14)
where P0 is the power required to overcome the zero flow torque, i.e. it is
the disc friction power. Now the disc friction, no flow power is propor-
tional to N 3 so
R = wQ (u 2 - uVrcot -y) +F. I\T 3
I g e''
Pi= w(DeN 2 Q + E"'ef/Q 2 + FeN 3 )

thus equation 6.8 is justified.
Therefore the efficiency of a pump can be calculated from the following
equation:
AN 2Q + BNQ 2 - CQ 3
Eff =DeNz Q + EeNQ2 + FeN3 (6.15)
A Band C can be calculated from the head~ flow curve and these
values can then be used to calculate the De E'e and Fe values from the
efficiency curve.
6.6 Pump power
The pump power calculation is needed in the calculation of pump run

down. The run down may occur in two phases as mentioned before:
(I) the normal pump run down and (2) in turbine mode. In both phases
the pump's power is needed to calculate the reduction or increase of the
pump speed occurring in the t:.t interval.
In phase 1 the pump's speed is reducing because the flow is still being
pumped and therefore energy is being absorbed from the rotational energy
of the pump.
The power absorbed by the flow in phase 1 is thus:
Pwr =- ~~~ (6.16)
(The negative sign implies that the pump is absorbing energy from the flow
rather than supplying energy to it. Remember also that when the pump is
operating as a dynamometer the value of Eff will be negativ~
The H value is given by the head equation and the efficiency by the
efficjency equation. Before this equation is used a check must be made
that the suction side head is less than the delivery head. In phase 2 power
will be supplied to the pump by the flow through it.
The power supplied will be given by
V. u
Pwr=wQ~
g
-P0 fNN3)
-.-
3
(6.17)
and the expression for Vwu given in equation 6.5 should be used here.
g
Again, before this expression is used a check must be made that the suction
head is greater than the delivery head.
6. 7 Pump start up
Pump start up is difficult to simulate because of the many and varied ways
in which different types of electrical starter gear start and run up the elec-
tric motor and pump to speed. If it were possible to specify the action of
the control gear it would be possible to simulate its action, but to make
provision for the large number of possible ways of arranging the starting
gear in a general program is virtually impossible (but see p. 120).
However, if the pump is assumed to start instantaneously at its full
running speed the starting transient will be the largest it can possibly be, but
even normal start up sequences can cause rapid run up to speed and may
well produce as large a start up transient as would be given by an instan-
taneous start up.
The start up transient can only reach a maximum equal to the closed
valve head (i.e. no flow head) of the pump and so will only rarely be dan-
gerous. However, by including a start up in the analysis it is possible to
run it on to steady state before tripping the pump. The check on steady
state that this gives is valuable to the designer. In the author's experience
steady states so obtained usually agree within 1% of that of a steady state
analysis and, as the frictional formula used in the program is better than
those in common use in most steady state programs, the unsteady result
for steady state can be considered to be the better value.
6.8 Pump run down
The method of analysing the pump's behaviour consequent upon cut off
of power is given on pp. 55-6. An outline only, without the discussion, will
be presented here. The methods used must be mathematically based if
they are to be used in a computer program employing the method of
characteristics, not graphically based as in chapter 3.
The pump run down is described by the equations used to describe
pump turbining, i.e. equations 6.7, 6.8, 6.9, 6.10 and 6.11.
The only difference is that the power P to be used in equation 6.11
should be the value of Pwr calculated in section 6.6.
In a computer program it is possible to employ a better method than
the simple initial value finite difference method of calculating N 2 given
previously. Even the simple predictor method described below will give
better results. Any rotating mass which is not driven will slow down ac-
cording to an exponential law. In the circumstance of a pump impeller
running down, yet still doing some pumping while it does so, the precise
nature of its run down curve will not be a perfect exponential, yet for
a short time interval t:.t, it will be far better approximated by a short
segment of an exponential curve than by a short straight line of gradient
equal to that prevailing at the speed at the beginning of the time interval.
At the beginning of the next l::.t interval a new exponential decay curve
can be calculated and the process repeated until the entire speed decay
curve has been developed.
This technique has been investigated and the author has found that
it gives results which are far superior to those obtainable from the initial
value finite difference method.
This method can be expressed mathematically as follows. Denote the
gradient of the pump speed- time curve by Gr. Then
G _ 3600 x Pwr
r- + 4rr2NI (6.18)
Assuming that N = ae bt- a general exponential law- then
dN=abebt
dt
dN 3600 x Pwr
but dt = + 4rr 2N/ Gr
dN
so dt abebt b=Gr
N =~=b N
Gr t
N
Now N 1 =ae 1
~r (t+At)
and N 2 =ae 1
(6.19)
This method has the advantage of calculating the speed relevant at the
time instant when the pump is running down or when it is turbining. It is
noniterative and is well adapted to the use of relatively large tl.t intervals.
However, if the pump speed is increasing, as it may in turbine mode, the
simple initial valve method outlined earlier is more suitable as the expo-
nential law does not then apply.
6.9 The in-line pump boundary condition
Points M and N denote the upstream and downstream tl.x points on the
suction and delivery pipes. a1 and a2 are the points immediately upstream
and downstream of the pump. R and S are interpolation points, R being
Figure 6.3
interpolated back from the P1 point based upon v0 and c 0 values, and S
being interpolated from~ based upon v02 and c02 vaiues (se~ figure 6.4).
As the heads across the pump (and the velocities also if pipe diameters are
not equal) are not the same, the points a and P have to be separately desig-
nated, i.e. a1 and a2 , }\ and ~-
Values of heads and velocities will be known at all grid points at the t
level and the interpolated values at RandS can be calculated from the
interpolation subroutine/procedure mentioned earlier.
. . 2/RvRivRIAt 2fsvslvsltl.t
The fncbon terms d and must then be calcu-
1 d2
lated; denote these terms by FR and Fs. Then writing the forward charac-
teristic equation:
(6.20)
M N
X
Figure 6.4
The backward characteristic equation is:
-L(hp2 - hs) + (Vp 2 - vs) + Fs = 0 (6.21)

cs
hp 2 - hp1 = AN2 + BNQ- CQ 2 (6.22)
(6.23)
where d 1 =diameter of suction pipe, d 2 =diameter of delivery pipe.

Adding the two characteristic equations
hp!- hp2- hR + hs + CR Vp + cs Vp - (CR VR + cs vs) +

g I g 2 g g
( ~R FR + ; Fs) =0
Substituting for hp1 - hp2 and equating Vp1 tog_ and Vp 2 tog_ where
ap1 ap 2
ap1 is the cross sectional area of the delivery pipe
then
-AN 2 - BNQ + CQ 2 -hR + hs + Q(~+___:_) -(~VR + cs vs)

ap1g ap~ g g
+ CR FR + csFs = O
g g
Let -AN2- (cR vR + cs vs) + (cRFR + csFs) -hR + hs

g g . g
="'
(6.24)
Let C=a (6.25)
Let (6.26)
The equation then reduces to
aQ 2 +/3Q+-y=O (6.27)
and so Q = t3 + v't32- 4a-y (6.28)

2a
If Q is less or equal to zero the nonreturn valve will close and negative
flows will therefore be prevented, so it is necessary to write an algolrithm
into the program to the effect that if Q ~ 0 then Q = 0.
Having calculated Q, Vp1 and Vp 2 can be simply calculated from
Vp
1
=JL
apl
and Vp =JL.
2 ap2
hp1 and hp 2 can then be calculated by back substitution into equations
6.20 and 6.21. The surd in equation 6.28 should always be positive and
failures in attempting to take the square root of a negative number should
not occur but it is wise to arrange to check the possibility. The author has
never known the value of /3 2 - 4a-y to become negative however.
If an account of time is kept sot= i x flt, i.e. i is a counter of the num-
ber of flt intervals that have elapsed since the commencement of pump
run, a test can be made to see if this time is less than the time at which
the pump is to be switched off. If it is less, then the pump speed to be
used in the next time level calculation must be the original pump steady
speed and, if it is larger than the time of pump trip initiation, then the
pump run down calculation must be used to obtain the pump speed ap-
plicable to the next time level.
6.10 Suction well pumps

Pumps, pumping straight from a suction well, can easily be analysed by.
setting the suction side potential head to the elevation of the suction
well level above the datum level of the scheme plus atmospheric head. It
is best to use an absolute potential as this simplifies the calculation of
the absolute pressures (by deducting the elevation of any point under
consideration above the datum level) which are needed in the calculation
of e bubble fractions and in checking whether local pressures have
fallen to vapour pressure or not.
As only the backward characteristic is needed to solve a suction well
pump the equations for the in-line pump can be redeveloped in a simpler
form. The suction well pump is an important case as it is in common use
in sewage pumping schemes. The pump equation is:
(6.29)
where Z'NI is the elevation of the base of the suction well above datum
level, dsw is the depth in the suction well, ha is the height of the liquid
barometer, i.e. atmospheric head. The Cvalue used should be the value
appropriate to the curve obtained by subtracting the pump station losses
(including losses in the suction pipe) from the ordinates of the pump curve.
The backward characteristic equation is:
-~ (h -hs)+ v -vs + 2fsvslvsl.1t = 0 (6.30)

CR p p d
As was done for the in-line pump, these two equations can be solved simul-
taneously for Q(vp =a~ where ap is the cross sectional area of the pipe).
A quadratic solution for Q results. Having obtained Vp the value of hp can
be found from the backward characteristic equation.
The value of dsw which is slowly changing, can then be updated. There
may be an inflow Qi to the suction well and there will be an outflow, i.e.
Q, so
(6.31)
where dsw, is the depth in the well at the end of the .1t time interval,
dsw 1 is the depth at the beginning and Asw is the cross sectional area of the
well. If the cross sectional area varies with depth this variation can be taken
into account but it may be thought that a mean area over the depth range
will produce a sufficiently accurate result. The variation of suction well
depth is important because in most sewage pumping schemes suction well
level switches are used to start and stop the pump.
The static head of a system may be a large or small fraction of the pump-
ing head. In a long flat main, friction head will be large and static head
will be small, and in such a case the variation in suction well level from
high to low suction well level will be a large fraction of the static head.
The pump delivery will be large at high suction well level, yet if power
failure occurred, the pump might turbine, the flow reduction would then
occur slowly and the consequent pressure transient would hence be mini-
mal. Conversely, flow delivered by the pump is minimal when the suction
well level is at its lowest value as then the static head is at its maximum
value.
Turbining is less likely to occur in such circumstances so a much larger
pressure transient may be generated by pump trip. It is not possible to pre-
dict what sort of pump behaviour will occur without performing a complete
analysis.
It is therefore important that pump switch on at a high electrode level,
followed by pump trip at a low electrode level, together with pump trip
at any level for the case of power failure, should all be simuiable in any
program.
6.11 Four quadrant pump operation
A pump is normally equipped with a nonreturn or reflux valve but for very
large installations it is not possible to construct an efficient reflux valve.
Such very large pumps are used to supply the cooling water of thermal
power stations and large scale gas liquefaction plants, for example. The
reflux valve is normally fitted to prevent reverse flow through the pump
when the power supply to the pump is cut off. If it is possible to fit are
flux valve it is advisable to do so. When the pump starts up after a stop
it is very desirable that it should deliver its discharge into an already full
pipeline since, if a pump has to deliver into an empty pipeline, it will
experience very little opposing head. It will therefore deliver a very large
flow and will swing down its characteristic curve, and its power demand
will be relatively very large. When the high speed flow that it generates
arrives at a hydraulic control a very large transient may be generated.
If the pipeline into which such a pump delivers has a sinuous longitudinal
profile, flow may occur through pipe segments of large negative gradients
in an intermittent manner because large air bubbles will rise upwards against
the flow in the pipe sections of negative slope and it may take a long time
to clear the pipeline of air. Empty pipelines of complex topography do
not fill in a simple manner. Long sections of pipe may contain air whilst
others run full. Air, entrapped in a long leg of a pipe network, can be com
pressed and so store a great amount of energy. Upon pump trip, such
large volumes of compressed air can rapidly expand causing liquid in other
sections of the network to move at high speeds. Such motions can interact
to generate pressure transients of surprisingly large magnitudes. Large
trapped bubbles can be set into vibration which resonate with flapping
reflux valves and then standing waves can be set up in the network. Stand-
ing waves of this type are known to have been the cause of catastrophic
pipe bursts.
It is therefore clear that pumping into empty or part-empty pipe net-
works should be avoided if at all possible unless allowance has been specific-
ally made for this event.
In large pump installations it is usual to trip the pump and then to shut
a downstream valve (usually ofbuttert1y type) to avoid emptying the system.
In cooling water systems it is not unusual to pump from a water reservoir
through heat exchangers and, after cooling, back into the original reservoir.
The static head on such a system will not be large but providing that the
local absolute pressure does not fall below the gas release head, there will
be no tendency to drain the system. However, if local pressures fall below
gas release head during the transient phase it is possible that gas will be
released and this will collect at high points. This gas will not redissolve
readily and will cause problems when the pump starts again.
A large pump delivering into a very large diameter main is unlikely to be
fitted with a reflux valve and so reverse flow back through the pump will
be permitted after pump trip, a downstream valve must then be slowly
closed and this will generate transients. Depending upon the time allowed
between the pump trip and the valve closure, the pump may continue run-
ning forward at a gradually reducing speed and may continue to deliver
flow in a forward direction or flow may reverse even while the pump is
still running forward. In one of these cases the pump is in its pumping
mode ~zone Pl (see figures 6.5a and b). With small negative head and
forward flow the pump acts as a dynamometer- zone Dl. If the pump
head becomes sufficiently negative the forward flow may start to drive
the pump and it then operates in a turbine mode - zone Tl. Once flow
has reversed, another series of operational modes arises. With reverse flow
and a forward rotation of the pump's impeller another dynamometer
phase occurs- zone D2. With reverse flow but with reversed pump direc-
tion and a higher downstream pressure than that in the suction pipe the
pump operates as a turbine- zone T2. With the same conditions but with
large negative speeds it operates as a dynamometer- zone D3. With for-
ward flow and large reversed pump speed the pump operates as a pump -
zone P2 and with forward flow but with rather smaller reversed pump
speed it operates as a dynamometer again - zone D4. It is most unusual
for a pump to operate in the fourth quadrant, however.
+N +ve torqu~ ~~
t \:~~ ~~:e
+ve head ~
+ve torque
normal - ve head
pumping
---r-~
Positive torque, positive head, positive Nand positive 0 are a II as for

normal pumping
(a)
Figure 6.5
There are eight different modes of behaviour that a pump can exhibit
spread over the four quadrants of theN- Q graph.
If a standard N- Q plot is used, four quadrant pump characteristics
are complex and difficult to use in any computer model of a pump's
behaviour. Suter 15 produced a dimensionless representation which is very
well suited to computer use. The following presentation is taken from
this paper.
0
Q*
200%
HIS head
0 IS flaw
N 1s speed
denotes pump duty values
(b)
If dimensionless variables WH and WT are used where WH is given by
H/H*
WH = sign(H) (N/N*)2 + (Q/Q*)2 (6.32)
and WT by
T/T*
WT = sign(T) (N/N*) 2 + (Q/Q*) 2 (6.33)
where affix * denotes steady state pumping conditions, His the head
across the pump, Q is the flow, N the rotational speed in rev min- 1 , Tis
the torque on the impeller shaft then graphs can be constructed of WH and
WT against 8 where 8 is given by
N Q*)
e = arctan (- - (6.34)
N* Q
i.e.
1.2
-1.2
Figure 6.6
These two graphs provide a complete representation of the very much

more complex complete set of characteristic curves of a pump over its
four quadrants of operation. It is relatively simple to read in to the com-
puter, say, 64 points on the WH and WT curves and to achieve a quite
accurate representation of them. By interpolation a particular value of
e
WT or WH corresponding to a particular value can be readily obtained.
From the equations quoted before (6.32 and 6.33) these values of
WT and WH can be converted into T and H values.
The torque value T can be used to calculate the running speed of the
pump at a D.t time period later and the value H, i.e. the head across the
pump, can be used in the boundary condition calculation representing
the pump. Before describing these calculations the problems a network
analyst will encounter in obtaining the necessary data concerning four
quadrant pump operation are discussed.
Very few pump manufacturers are prepared to supply the four quad-
rant characteristics of their pumps. It is most unlikely that anything other
than the normal H ~ Q, P ~ Q and E ~ Q curves of the pump in normal
pumping mode will be available, so it is necessary to devise a reasonable
approximation to the Suter curve for the other seven modes of pump
operation.
Three pumps were tested by Donsky (see. Bibliography (2)) in all modes
of operation. These three have a very wide range of specific speeds between
them so the analyst should find that the specific speed of the pump he is
proposing to use lies within the bracket of specific speeds covered by these
pumps. The specific speeds of the three pumps involved are 35 m 3 s- 1 units
(radial flow), 147m3 s- 1 units (mixed flow) and 261m 3 f 1 units (axial
flow).
The Suter diagrams for these pumps are given in figure 6.7 and the data
is shown in table 6.1.
The portion of the characteristic curves on the Suter diagram that will
be obtainable from the pump manufacturer's H ~ Q curve will lie in the
first quadrant, i.e. fore between rr/4 and rr/2.
If this very limited piece of information is plotted onto the Suter dia-
grams it will be found to lie between two of the specified WH curves.
Table 6.1
N 5 = 35 N 5= 147 N 5= 261
Radian Wfi WT WH WT WH WT
0 -0.728 -0.548 -1.249 -1.249 -0.707 -0.748
0.168 -0.639 -0.394 -1.048 -0.951 -0.935 -0.776
0.318 -0.445 +0.095 -0.789 -0.651 -0.828 -0.736
0.464 -0.179 +0.400 -0.529 -0.297 -0.632 -0.559
0.588 +0.398 +0.545 +0.186 +0.447 -0.276 +0.144
0.695 +0.576 +0.644 +0.555 +0.630 +0.468 +0.550
0.785 +0.707 +0.707 +0.707 +0.707 +0.707 +0.707
0.876 +0.806 +0.745 +0.791 +0.761 +0.896 +0.787
0.983 +0.904 +0.772 +0.881 +0.807 +1.043 +0.861
1.107 +0.992 +0.785 +0.984 +0.853 +1.187 +0.951
1.249 +1.069 +0.771 +1.094 +0.939 +1.348 +1.102
1.406 +1.120 +0.725 +1.216 +1.071 +1.506 + 1.275
1.571 +1.136 +0.663 +1.400 +1.217 +1.652 +1.400
1.736 +1.129 +0.608 +1.450 +1.240 +1.784 + 1.520
1.893 +1.102 +0.585 +1.479 +1.244 +1.864 +1.627
2.034 +1.107 +0.587 +1.505 +1.274 +1.891 +1.713
2.159 +1.039 +0.606 + 1.536 +1.308 +1.873 +1.741
2.266 +1.010 +0.661 +1.573 +1.381 +1.803 +1.716
2.356 +0.997 +0.721 +1.624 +1.442 +1.809 +1.660
2.447 +0.979 +0.777 +1.674 +1.535 +1.689 +1.596
2.554 +0.947 +0.831 + 1.703 + 1.594 +1.576 +1.477
2.678 +0.930 +0.885 +1.725 +1.650 +1.4 70 +1.342
2.820 +0.901 +0.926 +1.700 +1.658 +1.350 +1.201
2.976 +0.876 +0.940 + 1.620 + 1.580
I 3.142 +0.831 +0.927 +1.473 +1.450 +1.040 I +0.818 I
I 3.307 +0.789
+0.754
+0.887
+0.828
+1.247
+0.996
+1.235
+1.018
+0.887
+0.839
+0.646
+0.644
3.463
3.605 +0.727 +0.743 +0.785 +0.815 +0.785 +0.710
3.730 +0.710 +0.654 +0.644 +0.622 +0.680 +0.610
3.836 +0.709 +0.565 +0.528 +0.428 +0.510 +0.326
3.927 +0.711 +0.480 +0.624 0 +0.255 -0.274
4.018 +0.721 +0.376 +0.335 -0.414 -0.407 -0.570
4.124 +0.740 +0.263 +0.204 -0.564 -0.645 -0.763
4.249 +0.764 -0.155 -0.310 -0.709 -0.829 -0.938
4.391 +0.788 -0.379 -0.502 -0.843 -1.013 -1.082
4.547 +0.801 -0.600 -0.669 -1.030 -1.228 -1.240
4.712 +0.794 -0.819 -0.819 -1.225 -1.480 -1.526
2.0
-2.0
(a)
2.0
(b)
2.0

0+---~------~----------~--~~,--.--~
/ f N ~
~ -~
~H
Wr
-2.0
(c)
Figure 6.7
The torque - flow curve can be deduced from the power and efficiency
curves supplied by the manufacturer and this information can also be plot-
ted onto the Suter diagram and will be found to lie between the WT curves
of the specified pumps.
It is suggested that these short segments of the WH and WT curves of
the pump under examination can be completed for the entire 0 - 2n range
of e by interpolating between the two adjacent WH curves and WT curves.
This process is not free from risk. It is possible for two pumps to have
the same specific speeds yet have very different impeller geometries so
it would not be reasonable to expect the Suter diagrams of two such
pumps to be identical even though their specific speeds are the same. The
only way to be completely sure that the interpolated section of the Suter
diagram is correct is to obtain further information about the pump's
behaviour in another range of e but this may well be very difficult to do.
If no other information is available then the analyst may be able to
find a full set of characteristic curves for a different pump which has a
similar geometry and this information 'can be used to plot another set of
Suter diagrams. The shape of these curves can be used to guide the con
struction of the required Suter curves for the pump under examination.
Some sympathy is felt for the pump manufacturer in this situation. The
operation of pumps in four quadrant conditions usually only occurs in
the case of large pumps. The test of a pump operating in four quadrant
circumstances requires two identical pumps; flows and powers are
very large and the test rig is very big. For a manufacturer to supply a full
set of pump characteristics in such circumstances involves him in the
expenditure of a considerable amount of money and his reluctance to do
so is understandable. However, the analyst's position is not enviable. If the
manufacturer were prepared to perform four quadrant tests on model
pumps of the same specific speed as the prototype (which will probably
be included in his selling range anyway) and to then apply standard scaling
laws to the results of such tests, an acceptable solution to this problem
might be possible.
6.12 The use of the Suter curves

The way in which the Suter diagrams are used will now be described.
First the two curves, the WH and the WT against e curves, for the
pump must be represented in the computer by two arrays (sequences
of numbers). It is suggested that the coordinates of sixty-four points
on each curve should be used and a simple interpolation process be used to
establish intermediate values.
At steady stateN= Ns and Q = Qs so e =*,i.e. arctan( I). The then
current values of WH and WT can be obtained from the stored arrays.
Obviously H will equal Hs and T = Ts, the motor torque Tm will suddenly
become zero so in the torque equation
Tm- Ts = /Q Tm will equal 0 at pump trip
-Ts = JQ
so n = -TJI
21T
60 (~ -Ns) _Ts
b.t I
So N = N. _ 60 7'sb.t {6.35)
I s 21T I
At the beginning of the pump run down the head across the pump will
beH=H5
The forward characteristic equation can then be applied to the last
dx length of the suction pipe and the backward characteristic equation
to the first dx length of the delivery pipe
(h p1 - h) CR( ) 2fRvRivRIM_ 0 (6.36)

i.e. R +- Vp1 - VR + CR d -
g ~ su
2fsvslvsl~t 0
and
2
cs(
(h P - h s) -- Vp
g 2
-vs ) -cs gddel
(6.37)
where dsu = suction pipe diameter and dctel = delivery pipe diameter. Now
the known value of H for hp2 - hp1 , substituting Q 1/( *

hp 2 - hp 1 =H so by subtracting equation 6.36 from 6.3 7 then substituting
d~) for Vp1 and
Q 1I (~dJe1) for Vp 2 gives the value of Q 1
M R 0
5 N
pump
/':,.x
I /':,.x
Figure 6.8
Next (} can be calculated from

Nt
01 = arctan ( Ns Q 1
Qs) (6.38)
From the stored arrays of WT and WH the values of WH, and WT, relevant
to 81 can be obtained by interpolation.
From equations 6.32 and 6.33
11 = Ts w~ ( (Z~r + (g~r) (6.39)
and hl =Hs wJ, ((Z~r + ( g~r) (6.40)
Using 11
N2 = N1 _ 60 T1 ~t (6.41)
2rr I
Using /ft, the value of Q2 can be found and the above process repeated
as often as required.
6.13 Pump run up to steady pumping speed
The approach described in the preceding sections is applicable to the four

quadrant operation of a pump after its motor has been switched off (i.e.
pump trip). If it is wished to deal with the case of the run up to steady
speed of a pump starting from rest, it is necessary to possess the torque
(Tm) ~speed (N) curve of the motor. This torque~ speed curve must be
stored in the computer as an array and an interpolation technique used
to calculate the torque supplied by the motor at any particular speed.
The torque available to accelerate the pump impeller and the rotating
parts of the motor will then be Tm- Tp where Tm is the torque applied
by the motor and Tp is the torque of the pump at the particular running
speed. Tp is the torque obtained from the Suter diagram.
Then Tm- Tp =/D. (6.42)
so
n= Tm- Tp
I
2rr(N2 - N 1 ) = Tm- Tp
60.M I
30 (Tm- Tp)ill
N2 =N, +n I (6.43)
The method of dealing with the analysis is essentially the same as that used
for four-quadrant pump trip except for the substitution of equation 6.43
for equation 6.41.
6.14 Pumps with by-pass valves
In-line pumps are often equipped with a by-pass reflux valve (see figure
6.9).
The advantage of fitting such a by-pass reflux valve is that in a long
pipeline there may be a number of pump stations at roughly equidistant
points apart. When the demand on the pipeline is low some of the pumps
may not be used whilst the others remain operating to supply the required
flow. When such pumps are off-line the flow automatically passes through
the by-pass but the reflux valve is fitted to prevent backwards flow through
the by-pass when the pump is operational. Alternatively a gravity flow
system may require a booster pump to provide peak demand. Such a pump
will not be operational for much of the time and flow will then pass
through the by-pass under the gravity head.
The by-pass has a further advantage. When a pump trip occurs, there
will be a drop in head downstream of the pump. If this head falls below
the upstream head the by-pass valve will open and supply flow. The head
downstream of the pump will then not fall to such a low level as it would
have done if no by-pass had been present and the consequent waterhammer
- Figure 6.9
-
will be greatly reduced. The effect is similar to that which occurs when a
pump acts as a turbine during the low pressure phase following upon pump
trip.
The analysis of this circumstance is relatively simple. During the analy-
sis of the pump's behaviour (see section 6.12) a check should be made
that the upstream head remains less than the downstream head. If it be-
comes greater than the downstream head then, whilst the pump speed
run down calculations should be continued, the head- flow calculations
should be replaced by calculations which ignore the presence of the pump
and treat it as if it were a simple joint in the pipeline, i.e. the two charac-
teristic equations should be written and solved for head and velocity as if
no pump were present. If the full open reflux valve can create a local
energy loss this can be included by either lumping the loss into friction or,
more accurately, writing it as a separate head loss term, e.g.
(6.44)
_ _!_ (hp -hs) +(up -us)+ 2/suslusiM = 0 (6.45)

cs 2 2 ds
2
hp -hp =K~ (6.46)
1 2 2g
K% is the head loss caused by the open by-pass reflux valve.

2
Now
These equations can be reduced to a quadratic equation in Q which can

readily be solved. hp 1 and hp2 can then be obtained by back substitution
into equations 6.44 and 6.45. This approach neglects the flow through
the pump after pump trip but this is usually small.
6.15 Pump stations
It is unusual for only one pump to be installed in a pumping station. Usually

there are at least two, one being held in reserve for use if the other should
break down. More often there are more than two pumps, e.g. three or four,
and usually a number of these operate with, say, two on standby. When
modelling a group of pumps they can be treated as a single pump. The
characteristic of the group has to be calculated so that this can be used as
if it were that of a single pump.
Groups of pumps in a station usually operate in parallel and in this case
they should be identical if stable, satisfactory operation is to be obtained.
Pumps in parallel
Identical pumps operating at the same speed should have identical H ~ Q
and Eff ~ Q characteristics. The flow Q that the group produces is the sum
of that produced by the individual pumps.
If there are n pumps in the group then the equation of the equivalent
pump is:
He =A'N2 + B'NQ- C'Q 2 (6.47)
where A'=A
B'=B/n
C'= C/n 2
The inertia of the pump group can also be calculated. As far as an in-
dividual pump is concerned it is delivering its normal rated flow at steady
state and is unaffected by the operation of the other pumps in the group.
When the pump group is switched off the speed of the individual pump
will decay in exactly the same manner as if it were operating alone but
each one is only pumping 1/nth of the fluid. The inertia value to be used
to represent that of the group must, therefore, be that of an individual pump
multiplied by the number of pumps in the group.
The efficiency of the group will be the same as that of the individual
pumps and the efficiency equation quoted previously can be used but, for
the group, the flow Q will be replaced by Q/n, i.e.
_ A 'N 2 + B'NQ- C'Q 2
Effe - Q
DeN\1 +EeN
(Q)2 Q)3
n +Fe(n
(6.48)
so D~ =De/n
E; = Ee/n 2
and F; = Fe/n 3
where De, Ee, Fe are the constants for an individual pump and n;, E~ and
F; are the constants of the pump equivalent to the pump group.
Pumps in series
If the pumps in a pump group are arranged to operate in series the total
flow is the same as that through any one pump but the head is the sum
of the heads of all the pumps. The pumps do not need to be identical for
satisfactory series operation.
Thus the H- Q equation for a series arranged pump group is:
(6.49)
where
"f.ApNp2 =A 1 ~ 2 + A2Nl + A3Nl + ...
"f.BpNp =B1 N1 + B2N2 + B~3 + ...
r.cp = Ci + C2 + c3 + ...
If the pumps are running at different speeds (an unusual arrangement) the
equivalent pump which represents the pump group will store kinetic energy
within its equivalent rotating mass
so ! I'N2 = r,t:fpNp2
where r.tip~ =t(/,~2 + I2Nl + hNl + ... )
I'N 2 = "f.IpNp2
so I'= "f.Ipfp2 (6.50)

N
where I' is the equivalent pump's moment of inertia and N is the equivalent
pump's running speed.
A value of N can be chosen arbitrarily, say the speed of the fastest pump.
Then the equivalent head- Q equation will be:
He =A'N 2 +B'NQ-C'Q 2
where A'= "f.ApN~

N2
B' = "f.BpNp
N
and C'= "f.Cp
The efficiency of a series pump group is given by Effe =He L it

l->n P
Compound pump arrangement

It is possible to arrange a group of pumps in parallel and to then connect
this group to another group which is arranged in series.
By obtaining the equivalent pump to the parallel group and the equiva-
lent pump to the series group, and then combining these two groups in
series, an equivalent pump to the compound group can be obtained.
Tripping of a subgroup of pumps in a pumping station
It is possible to write a pump station subroutine/proc edure in which the
equivalent pump's characteristic curve, its inertia and its efficiency curve
are changed suddenly (when the time counter i in the expression t = i x dt
exceeds a preset value) due to the trip of a subgroup of pumps within the
main group. It will not be necessary to model the run down in speed of
the tripped subgroup because the continuing delivery from the untripped
pumps will prevent the development of transients of any significance.
Writing such a subroutine/proc edure is quite simple, the values of A', B',
C', n;, E;, F~ and!' being calculated for the entire group if i is less than
the preset value and for the group of running pumps only if i is greater
than the preset value.
Alternatively the group of pumps in the station can be divided into two
subgroups each of which is treated as a pump station, see figure 6.1 0.
The pipe network is slightly more complicated due to the introduction of
Figure 6.10
the additional .:::lx lengths. If station 1 represents the group of pumps

which will continue running and station 2 represents the group that is to
be tripped at a preset time then calculations of the characteristic curve
constants, etc, of the equivalent pumps can be performed without compli-
cation and station 2 can be tripped at the appropriate time. The run down
of the equivalent pump will be modelled by this method but it may be
thought that the introduction of the additional .:::lx lengths offsets the
advantage that this represents.
6.16 Surge suppression of transients generated by pump trip
Two main methods of surge suppression are available:

(1) Fitting flywheels to the pump.
(2) Fitting air vessels or surge tanks to the pipe just downstream of
the pump.
By fitting a flywheel onto the shaft of the pump, the inertia of the pump-
set can be greatly increased. This means that the pump run down in speed
will occur over a longer time so the decrease in the pump delivery will take
longer also. If this reduction in the pump's delivery can be made to occur
over a sufficiently long time without requiring an excessively large flywheel
a practical method of surge suppression is available. The pump delivers for
the period during which the closed valve head of the pump, AN 2 at the
current reducing value of N, is greater than the static head on the system.
If this period is greater than the pipe period of the delivery pipe (2L/c)
some surge suppression will be obtained. Obviously the larger the flywheel
the longer the delivery period and the greater the surge suppression result-
ing. Flywheels are expensive and there is a limit to how large they can be
made economically so this method of surge suppression is only useful when
pipes are relatively short with correspondingly short pipe periods. The
delivery period during the pump run down can thus be made large relative
to the pipe period, only if the pipe is short. To analyse a flywheel case is
easy however, as it is only necessary to increase the pump set's inertia by
that of the flywheel and then carry out another computer run.
It has been the author's experience that it is necessary to do about four
runs using an initial value of the flywheel's inertia of zero and increasing
this up to the largest practicable size over the next three runs.
By plotting the maximum (and, if critical, the minimum) pressure head
in the pipeline against the flywheel inertia the following graph results.
E
:J
~
"E
flywheel inertia
Figure 6.11
The horizontal section of the graph in figure 6.11 is due to the fact that
up to a certain critical flywheel inertia value the delivery period during
pump run down is Jess than the pipe period so it has no effect upon the
maximum transient pressure head. This is similar to the difference between
sudden and slow valve closures.
From figure 6.11 the maximum acceptable pressure head determines
the necessary flywheel inertia. The maximum pipe length which can be
economically surge suppressed by the use of flywheels is about 1-2 km,
depending upon the pipe's distensibility. For long pipelines it may be
found necessary to fit an air vessel or surge tank.
Surge tanks can only be used if the delivery main is relatively flat and
the peak head at the location of the tank is not excessively large. A rough
guide is that the top of a surge tank located just upstream of the pump
will have to be at a height above Ordnance datum equal to the height of
the liquid surface in the suction well above Ordnance datum plus the closed
valve head of the pump if the tank is not to spill. In practice this height can
be reduced somewhat but this rule allows the designer to determine if a
surge tank can be considered at all.
Air vessels have been discussed in previous chapters but in this context
it is necessary to point out that they are fairly expensive, being required
to withstand internal pressures which may be quite large and if located at
an elevated point they may also be required to withstand subatmospheric
pressures. An air vessel which is subjected to internal subatmospheric pres-
sures may need to be internally braced if it is not to fail in a buckling
mode. They must be equipped with compressors and the entire installation
must be regularly maintained so that air dissolved by the water is replaced
regularly. Whereas flywheel surge suppression always works, air vessel
surge suppression only operates if the liquid level in the vessel is set correctly
by automatic devices or at frequent maintenance inspections.
A third method of obtaining surge suppression is to fit a by-pass around
the pump. This by-pass pipe is usually arranged to permit water to flow
from a point just downstream of the pump via an electrically controlled
valve back into the suction well. Normally, the valve is closed during pump-
ing but when pump trip occurs a solenoid operates to open the valve. If
the pressure drops sufficiently after pump trip, water will be drawn through
the by-pass into the pipe so minimising the size of the pressure reduction.
The initial transient generated by pump trip will return to the pump after
reflection at the downstream end and this transient would normally reflect
positively if the by-pass valve were closed. If the valve is open, however, the
transient will be greatly reduced in magnitude and a reverse flow will be
generated; the slow closure of the valve will arrest this flow without develop-
ment of significant transients providing that the closure is slow enough. This
type of valve would be used more frequently if it were 'fail safe'. However,
depending as it does upon an electricity supply it is not 'fail safe' as pump
trip occurs due to a failure of the electricity supply as well as to normal
pipeline operating procedures. However, a valve of this type, actuated by
compressed air, is available and should be virtually 'fail safe' if regularly
maintained.
Subroutines/procedures written to describe the operation of hydraulic
controls can only be used at the beginning or end of a pipe length, the
smallest value of which is the ~x value chosen. Thus if a pump is to be
equipped with a surge tank or air vessel it would normally be necessary to
interpose a ~x length between the pump and the air vessel (see figure
6.12).
In many cases where the ~x value is small this method would be accept-
able but if the system is long, say 10 km, and the ~x length is to be 1 km,
the use of a length of 1 km to represent a distance of perhaps 10 m would
lead to very bad modelling.
air vessel
Actual pump/air vessel arrangement
Simulation of pump/air vessel arrangement
Figure 6.12
A case can therefore be made out for developing a subroutine/procedure

which would model the pump and air vessel as one unit.
The methods already described for modelling pumps can be combined
with the method given in the next chapter for modelling air vessels to give
a description of the behaviour of the pump/air vessel combination.
In long pipelines in which booster pumps are fitted it may be desirable
to fit a suction side and a delivery side air vessel. These can be modelled
by the method outlined in section 7.3 for the pump/delivery side air vessel.
Surge tanks are really only a special case of air vessels. As will be des-
cribed later, the gas compression/expansion process in the air vessel is usu-
ally assumed to be polytropic, the index being taken as 1.2. If this index is
given the value zero, the equation
p V 1.2 constant
becomes p constant
which is the case in the gas above the surface of a surge tank, the pressure
there being atmospheric pressure. If the height of the air vessel is made
very large no compression or expansion of the gas is possible so, either
making the polytropic index of the gas process zero, or making the air ves-
sel's height very large, or both, will give the required conditions for describ-
ing a surge tank of simple type.
6.17 Line pack and attenuation
When a hydraulic control such as a valve operates at the downstream end

of a pipeline causing the flow there to fall rapidly to zero, a steep wave
front travels in the upstream direction. As it reduces the velocity of suc-
cessive layers of fluid, the pressure of these layers rises and this rise is trans-
mitted downstream through the near-stationary fluid maintaining its eleva-
ted pressure. However, the pressure of the fluid before it experiences the
effect of the wave is higher than the initial pressure of layers of fluid down-
stream of it due to friction, and this higher pressure is also transmitted
downstream, together with the rise caused by the momentum effect of the
velocity change. Therefore, as the wave progresses upstream the pressure
at the downstream control continues to rise even after the initial steep
fronted rise has been generated. The fluid in the pipe between the wave
and the downstream control thus experiences a progressively increasing
pressure which causes it to compress and the pipe wall to distend, so, even
though the velocity immediately downstream of the wave is zero at the
control, it increases as the wave travels upstream in order to supply the
fluid that must continue to flow into the downstream section to fill the
space made available by the fluid compression and pipe distension. The
velocity change across the wave is therefore reduced so the wave height
itself decreases.
The phenomenon of pressure increase after the wave has passed is called
line pack and is commonly found in long pipelines, especially in oil pipe-
lines.
The phenomenon of the reduction of the pressure wave magnitude as
the wave travels upstream is called attenuation.
6.18 Lock in
If a pipeline has a valve at its downstream end and a pump equipped with
a reflux valve at its upstream end a phenomenon called lock in occurs
when the downstream valve is closed. The valve closure causes a positive
wave to travel upstream towards the pump. When it reaches the pump
the progressive flow reduction causes the pump's operating point to move
up along the pump's H- Q characteristic, the pressure rising as it does so.
A reflection will thus be generated due to the interaction of the pump
with the incident wave and this will travel downstream to the closed valve
where it will positively reflect and travel back upstream again to the pump.
Throughout the time of travel of the wave down to the valve and back
again the pump will still be delivering a reduced flow into the pipe which
is closed at its downstream end. Eventually, after one or more pipe periods
the magnitude of the pressure downstream of the pump will become
greater than the no-flow head of the pump and the reflux valve will then
close.
The complex wave system in the pipe will then travel backward and for-
ward along the pipe attenuating as it does so. During this period the pres-
sure at the pump may fall below its no-flow head for short times and the
pump will deliver more fluid into the pipe. Eventually flow will cease at
all points in the pipe and a pressure head will exist throughout the pipe
which may be significantly higher than the no-flow head of the pump. If
the reflux valve and the downstream valve do not leak, this pressure will
remain in the pipe and this is the lock-in pressure. The lock-in pressures
that can occur in long pipelines can be large. The author has encountered
lock-in heads as large as 1.5 times the no-flow head and in long pipelines
this can be a dangerously high value. If the downstream valve does not
seal perfectly, lock-in pressures of dangerous magnitude can be avoided as
only small quantities of fluid need be leaked to reduce pressures suffi-
ciently. Such a leaking valve can be modelled using methods illustrated
elsewhere in this text (see page 136) by reducing the K value of the closed
valve from a very high to a lower, but still high, value.
The phenomenon of lock in is automatically analysed by the method of
characteristics technique but not by the Schnyder Bergeron graphical
method.
These phenomena can all occur in any pipeline. Obviously, line pack
will only be detectable in relatively long pipelines and lock in is also un-
likely to be a matter of concern in short pipelines. They will be predicted
by the characteristic method even so.
7 Other boundary conditions
7 .1 Junctions
As stated earlier it is vitally necessary to have a subroutine/procedure

which is capable of describing an 'n way' junction if a program is to have
any generality.
An 'n way' junction is one in which n pipes join at a point in the net-
work. It may be thought that it is only possible physically to join a few
pipes at one point, say four or five, but often pipes are joined at a manifold.
The author has encountered a circumstance in which 22 such pipes joined
at a manifold. Also pipes may join at a simple junction, followed by another
junction connected to the first by a short length of pipe, and so on. None
of the short lengths between such junctions may be long enough to be
considered as a Llx length so it is logical to combine these distributed
junctions into one large junction and this will give the best modelling.
It is usual to ignore local losses at junctions. Their effect can be allowed
for by increasing pipe lengths slightly or by increasing pipe roughness but
in liquid flows their effect is generally small. If they are included, the
junction procedure becomes very much more complex and expensive in
computer run time.
At a junction, continuity of flow must be maintained and the head
must be the same for all pipes joining there.
The continuity equation can be simply written, i.e.
(7.1)
where aPa is the cross sectional area of the ath pipe and vPa is the velocity
at the junction end of the ath pipe at the end of the Llt interval, and n
is the number of pipes joining at the junction.
A convention has been adopted here, i.e. flow towards the junction is
positive and flow away from it is negative.
Next consider the dynamic circumstances at a junction. If the junction
130
Other boundary conditions 131
is located at the downstream end of a pipe (i.e. flow in the pipe is assumed
to be towards the junction) then a forward characteristic line can be drawn
from some point in the last ~x interval in the pipe length and, similarly,
for pipes for which the junction is located upstream, a backward charac-
teristic can be drawn towards the junction from some point in the down
stream ~x length (see figure 7 .I).
In figure 7 .I if the pipes associated with the junction are numbered with
positive sign if flow in the pipe has been assumed to be towards the junc-
tion and negative sign if flow has been assumed to be away from the june-
f axis
No pipe may be assigned a zero number
Figure 7.1
tion, then an integer numbers can be obtained from this numbering con-
vention which will either take the value +I if flow is towards the junction
or -1 if it is away from the junction; sis calculated from the following
equation
s = sign(a)
where a is the signed pipe number and s is the required integer number.
This s value can now be used to define the characteristic directions to
be used for each pipe i.e.
(7.2)
where a denotes the ath pipe and Va, Ca, ha, fa represents the interpolated
values in the ath pipe, vPa denotes the velocity at the junction end of the
ath pipe at the end of the ~t interval and dais the diameter of the ath
pipe.
From equation 7 .I LSaPa vPa = 0

(s has been introduced in this equation to preserve a positive value of
vPa in a pipe carrying fluid away from a junction, and remember that s 2 =1)
soL saPavPa = 0
1-->n
= L saPaVa- L~a aPahP + L;a aPaha
1-->n 1-->n 1-->11
As hp is the sar.~e for all pipes
(7.3)
This equation is very well suited to computer solution. All values on the
right hand side are known and hp can be readily calculated.
Then values of vPa- one for each pipe - can then be found by back
substitution into equation 7 .2. It can be seen that there is no limitation
on the number 11.
The junction has therefore been solved.
7.2 Joints
Joints consist of two pipes joining end to end. A joint is obviously a two-
way junction. However there is a good reason for writing a separate joint
subroutine/procedure. If two pipes meet end to end, 'two-way junction'
can be used but if there should be a reflux valve interposed between the
two ends, reverse or forward flows will be preverted according to the direc-
tion in which the reflux valve is fitted. Also the use of 'two-way junction'
would be rather more expensive in computer run time than a straightforward
'joint' procedure as it involves slow processes such as 'for' clauses (in Algol)
or 'do' loops (in Fortran). As there are many joints in most networks there
is a good case for using a joint procedure in any transient analysis program.
Assume that the reflux valve causes trivial losses when flow is occurring.
through it.
The forward characteristic equation describes conditions along RP
(7.4)
-N=*---
------kJ~--- M R JOint 5 N
Figure 7.2 Figure 7.3
and along SP the backward characteristic equation applies:
-L(h -h )+v -v + 2fsvs lvs l.0.t=O ( 7 _5 )

cs p, s p, s ds
Note vp, may not equal vp, if the pipe diameters are not equal, i.e.
dR =I= ds but hp is the same for both pipes if local losses can be ignored and
vp, is not zero.
From equation 7.4
CR CR 2fR VRivRiflt
hp =hR--(vp -vR)--x~"--d:.:-;:_:__-
' g I g R
and from equation 7.5
h p -_ h S + -cs ( Vp - VS ) + cs
- X
2fsvs ivs lilt
--=--==:.....:;-------"--
2 g , g ds
(7.6)
cs2fsvs I vs I .0.t
gds
CR 2fR VRIVRI!lt
gdR
h -h + cRvR + csvs -
2fsvs ivs llltcs 2fR vR lvR llltcR
-~~~~-~
R S g gds gdR
g (7.7)
At this point in the calculation it is necessary to check what type, if any,
of reflux valve is located at the joint.
If none the calculation of vp, stands.
If a reflux valve is fitted that only permits forward flow then a test
must be made, i.e. if Vp, < 0 then Vp, = 0.
lf a reflux valve that only permits reverse flow is fitted then the follow-
ing test must be made:
If vp, > 0 then vp = 0
After the value of vp, has been so adjusted Vp 2 can be calculated i.e.
a
v
p2
=
aP2p, v p,
Then h =h _ CR (v _ v ) _ 2fR VRivRILHcR (7.8)
p, R g p, R gdR
and h = h + cs(v _ v ) + 2fsvs Ius l.::ltcs (7.9)

P2 s g P2 s gds
Although hp1 will equal hp 2 if the reflux valve is open hp, will not equal
hp 2 if the valve IS closed so both values of hp, and hp 2 must be calculated
separately in the latter case.
7.3 Air vessels
A facility must be provided in the program that permits any number of

air vessels to be located anywhere in the pipe network.
The basic method of analysing an air vessel used in chapter 3 to demon-
strate how mass oscillation of air vessels is performed will be used here.
There is a difference, however, in that in this presentation an account of
water hammer will be kept in the pipes leading to and from the tank although
the propagation of waves up through the tank itself will not be included.
This limitation is unavoidable if run times are to be kept within reasonable
bounds. The path length of a wave from the junction of the air vessel with
the pipe to the free surface in the air vessel is very small, much smaller
than .::lx. If .::lx is reduced to approximately equal this distance the run
time of the program will become unacceptably large. The analytic tech-
nique offered below is thus a combination of quasi-steady analysis of the
tank with a full waterhammer treatment within the pipe network. The
passage of transients through the tank will not be demonstrated by this
method but these are heavily attenuated by the tank anyway and are tri-
vial in magnitude. However, they can be calculated using developments of
techniques already demonstrated.
In figure 7 .4/t is the longitudinal length of the tank, dt is its internal
diameter, hb is the height of the base of the tank above the centre line of
the pipe, Zt is the elevation above datum of the centre line of the main
pipeline at the point of its junction with the tank.
Figure 7.4
At the beginning of the analysis the initial height of the water in the
tank and the initial potential' head at the junction must be known. Denote
these two values by hq andhpj
The pressure in the air in the tank expressed in height of an equivalent
water column can be calculated. Let this equivalent water column height
be denoted by hwr
Then
hwi =hpj- Zt- hb- htj
This will be the absolute pressure head of the air because hPi is the
absolute potential head.
Applying a forward characteristic equation to the Ax segment upstream
of the tank
_!_(h p -h)+( _ VR ) + 2fRvRivRIA!_ 0

R Vpl d
CR I
and a backward characteristic equation to the Ax segment downstream

of the tank
g (h
-cs
- p
- h s ) + (v p 2 -vs ) + 2fsvsd2lvs IAt -_ 0
The subscripts R and Shave the significance ascribed to them through-
out this book. hp will be the same value for both the Ax segments but vp,
will not equal Vp 2 because of the flow into (or out of) the air vessel.
By continuity
so (7 .l 0)
where At is the cross sectional area of the air vessel and equals i" dt.
~~5
-~~'--------l~----""---::-"-'~-
pipe I ----:::-'-'
IDI
pipe 2
M -R-=D.-'--x--+1\+-------"D.='--x--------10 I
\
1--1
.
1uncton leading to air vessel
Figure 7.5
The depth of liquid in the air vessel at the end of the b.t interval therefore
is given by
ht2 = ht, + b.ht

where h 1 is the depth in air vessel at the end of the b.t interval and h 1, is
the deptfi in the air vessel at the beginning of the b.t interval.
The new pressure head in the air in the air vessel will be:
l - h
h w2 =( [ t
)n X h w,. (7 .11)
h t2lj.
t
where n is about 1.2

so (7 .12)
By back substitution of hp into the forward and backward characteristic
equations values of Vp and Vp 2 can be easily obtained.
From the values of Vp, and Vp 2 the flow into the air vessel at the end of
b.t interval is readily calculated, the potential head at the junction has al-
ready been obtained and the change in level in the air vessel has been pre-
dicted. The calculation is an initial value integration and could be improved
by iteration if this is felt necessary.
A horizontally positioned cylindrical air vessel can be analysed by a
similar method but the more complex geometry of the circular cross section
makes the problem marginally more difficult.
7.4 The motorised valve
A motorised valve may be fitted on any pipe in a network. To analyse such

a situation it is necessary to split the pipe into two sections one upstream
and one downstream of the valve. It may be assumed initially that the
valve is driven by an actuator which moves the valve spindle at a fixed
speed. The head loss across the valve can be taken as being given by the
equation
v2
h =K- (7 .13)
v 2g
where v is the velocity in the upstream pipeline. The value of K can be
obtained from manufacturers' catalogues.
The value of K so obtained is only valid for the circumstances in which
the manufacturer performed the test, i.e. in the situation of the valve fitted
into a pipe of a specified internal diameter. Before the valve K quoted can
be used in a pipe of different diameter it must be adjusted to suit the dif-
ferent conditions. The loss created by a valve is almost entirely caused by
the expansion of the flow downstream of the partly closed valve and is
caused by the boundary layer separation occurring there.
-
A, v,
Figure 7.6
The loss is closely approximated by

h =(vv-v2)2
v 2g (7.14)
There is a small loss occurring in the convergent section of the flow, i.e.
in the change of velocity from V1 in the pipe to Vv which occurs at the
point of the contracted stream emerging from the valve, but this has almost
no effect upon the validity of the argument that follows since this loss is
also very nearly proportional to vt
Now A 1 V1 = Av Vv= A2
V:
where A 1 =area of upstream pipe, A 2 =area of downstream pipe,
Av =area of contracted section.
Then
A.
Vv =-v.
Av
AI
and v2=A2vl
so h =(A1 _A1.) 2
~
v Av A2 2g
h =A2g2(A2-Av) 2 2
v A2Av v 1
The value of K is thus given by
(7.15a)
and if A2 =A1
(7.15b)
Denote this value of K by Km, t!1c subscript m denoting that Km is the

manufacturer's value.
If the valve were fitted into a different size of pipe from that used in
the manufacturer's test the following analysis shows how Km .should be
modified.
,----.
I I
l
Lll
y
I I actuator
.-lh
( \
Figure 7.7
Again the head loss can be regarded as being caused by the flow expan-
sion downstream of the valve.
h
- v4 )
=(vy 2g
2
So y
as before, but in this case v4 is much smaller than it was in the pipe used
in the manufacturer's test and hv is correspondingly larger. Vv is the same
as that obtained in the manufacturer's test for the same valve setting.
Now for the same flow
so hv = (~r(A4
A4 1 Av
-I) 2 v32
2g
using appropriate values: A 3 and A 4 in place of A 1 and A 2 in equation
7.15a.
Then
In all cases of which the author is aware the manufacturer tests his
valves in a pipeline of the same diameter upstream as that of the down-
stream pipe, so A 1 = A 2
hv = {_A3)2 A
\A4 (
4
Av- 1
A 2_ 1
~2 Km v/
2g
Av
so hv = (A3)2 {_A4 - Av)2 Km v3 2

A4 ~A2 - Av. 2g
The value of K for the modified pipe diameters is
But Av = (ffrn + 1) from equation 7 .15b
so
and then K= (A3) 2(ffrn + 1 - A2/A4) 2

Km (7.16)
A2 vKm
Now, usually the effect produced by a partly closed valve is only signifi-
cant if K (and hence Km) has become two orders of magnitude greater than
(A 2 /A 4 ) and at least one order of magnitude greater than unity so it may
well bt: thought that this result can be simplified to
K=(~:r Km
or K = ( d act ) 4 Km (7.17)
dtest
where dact is the diameter of the upstream pipe to be used and dtest is the
diameter of the pipe used in the manufacturer's test.
However, the result given in equation 7.16 could be used with no parti-
cular difficulty if great accuracy is required~
It may be argued that the treatment of the downstream enlargement
of pipe diameter as being of sudden type will introduce an error in cases
where a reversed taper section is used but it should be remembered that
the angle of divergence of the taper must be less than 30 before it will
reduce the loss by a factor of less than 15% for an area ratio of the taper
of 1 :4 and by a factor of 33% for an area ratio of the taper of 1 :9. As it
is unusual to fit valves of such relatively small size into pipelines the error
will be less than 10% in most cases and will tend to overestimate the K
value. It is possible to include the effect of a taper by introducing a taper
coefficient into the analysis, i.e.
hv = cT(AA4 (AAv 1)
3) 2 4- 2
!!:1__
2g
(7.18)
and for a particular taper CT can be obtained from manufacturer's cata-

logues.
The previous development will then be modified to
(7.19)
Having obtained the multiplier to be used to convert the manufacturer's

values, an array of K values can be obtained for stroke positions ranging
from fully open to fully closed. (If the valve is to be opened instead of
closed this array must be read in reverse order.) The stroke positions could
be at 1% or 5% of stroke intervals.
Consider the case of a valve closing (or opening) at a fixed speed, i.e.
the stroke increasing (or decreasing) at a constant rate with time. Initially
the valve is full open (or closed) and the stroke position stays at this value,
as does its associated K value, until the number of !:lts for which the inte-
gration process has been performed exceeds a previously chosen value. The
stroke then commences to decrease (or increase) at a previously chosen
rate. The stroke position is thus known at every subsequent f:lt value and
the associated K values for these !:lt values can be obtained by interpola-
tion between the appropriately arrayed K values already supplied. If a
sufficiently large array has been used, a linear interpolation process will
be found to work well. The author uses 21 values in the K array giving K
values at 5% intervals throughout the valve stroke.
If the valve is to close from an initially part-closed position the array
values supplied must run from the K value appropriate to the part-closed
valve position to the fully closed value. (Note that the fully closed valve
must always be infinity but as a computer cannot store such a value it
should be assigned a very large one such as 10 20 .)
The valve closure rate can be calculated by taking the fraction of the
stroke over which the closure is to occur and dividing this by the difference
between the time of closure completion and closure commencement.
The circumstance can arise in which a two or three-speed actuator is
used. Such actuators are employed to provide a rapid closure of the valve
over the first portion of the closure (say 80% of the stroke) and then to
provide a slow closure of the valve over the last portion of the stroke.
It should be remembered in this context that during a very large fraction
of the closure of a valve, relatively small reductions of flow occur with
the development of only small transients; the effective portion of the
closure occurs in the last 10% of the valve stroke.
It is possible to describe this circumstance by setting the K values ap-
propriate to 5% time increments, instead of 5% stroke increments, using
the K values corresponding to the stroke's increments which are current
at these time intervals. In effect this means that the multispeed actuator's
operation can be represented by a distorted presentation of the K- stroke
curve of the valve.
Similarly, the action of an actuator that produces a stepwise valve clo-
sure can be modelled by the same technique.
The representation of the last step of valve closure constitutes a special
problem. The value of Kin the penultimate position in the array will be
very large, say 10 000 or 100 000, but the last value will be extremely
large (I 0 20 ). This means that an attempt to linearly interpolate between
the penultimate and the final value will be extremely unrealistic. It is sug-
gested that for stroke values lying in this stroke range an exponential inter-
polation process should be used. At the penultimate value it is simple to
calculate the gradient of the K curve and this can be used to set one of
the constants in the exponential curve. The value of Kat its closed position
(10 20 ) can be used to calculate the other constant. The author's firm has
also used a logarithmic representation of the entire K array most success-
fully.
By careful choice of the K value at the valve's closed position a leaking
valve can be accurately represented. This facility can be important as a
leaking valve can be used to prevent lock in. (See p. 128 for a description
of this phenomenon.)
Having established a technique for finding K at any position of a closing
valve it is possible to calculate heads upstream and downstream of the
valve together with velocities upstream and downstream of it. These vela-
+ +
~
-+--..,R.&.~------1
---------"'::-5-+ -
N, N2
Figure 7.8
cities are not necessarily the same since pipe diameters either side of the
valve are not necessarily equal.
(7.20)
g 2f:svs Ius ldt =0 (7.21)

--(h -hs)+(v p, -vs)+
cs P2 ds
2
also hp, - h p2 -- K Vp, .
2g stgn (v )
11 (7.22)
but
- CR (
hp -hR-- Up
- ) CR (2/R VR I VR
VR . - -
I dt)
' g I g dR
and hp = hs + ~(v _ vs) + cs(2fsvs I vs I dt)

, g p, g ds .
_ 2:RfR vR I vR I dt _ 2csfsvs I vs I dt
gdR gds
2 .
Kvp
- -1 sign (
v )-h
- R - h s - (cRvp 1 +csvp)
1 + cRvR+csvs
__:_:____:_:______::::..__:::
2g PI g g
_ (2fR vR 1 vR 1 cRdt + 2/svs I vs I csdt) ( 7 .23 )

gdR gds
1r 2 1r 2
Now =4ds
t
4dR Vp 1 Vp 1
so cR " +csvp, +cs (~f)vp, (7.24)

g g
Avp, 2 + Bvp, + C =0 (7 .25)
where A= K sign (vp.)

2g
-B + yB'l- 4AC
Then v p, = (7.26)
2A
and dS Vp 1
Vp 1 = (dRf (7.27)
By back substitution into equations 7.20 and 7.21 hp and hp can be cal-
culated. VR, Vs, fR,fs, CR and cs are calculated by ~ethods ~lready
described.
7.5 Servocontrolled valves
This is a most important type of valve. As the current methods of operating

pipelines develop, this type of valve will be used more and more frequently.
It is a valve which is of exactly the same type as the motorised valve de-
scribed earlier but it is controlled by a servomechanism which is operated
by pressure transducers located elsewhere in the system. Frequently these
transducers are located immediately upstream or downstream of the valve
but quite often the transducer is located on another pipe and may be far
away from the valve it is controlling. Usually the valve is required to com-
mence closing/opening if the pre sure at the transducer rises above (or falls
below) a certain critical value. If the pressure continues to rise (or fall) the
valve will continue its movement in an attempt to regulate pressure fluctua-
tions but at another critical pressure the valve will be completely closed (or
full open). The valve thus opens or closes as the pressure at the transducer
fluctuates within the defined band of pressure. If the fluctuations are slow
the valve will move in sympathy with the pressure fluctuations but if the
pressure fluctuations are rapid it may not be able to do so and then the
valve will be altering its stroke in a manner which will be out of phase with
the pressure transient. (Note that such behaviour can cause resonance -
see chapter 9 for a description of this effect.)
Let the valve's stroke at any instant be Sj. Let the pressure head sensed
by the transducer at that instant be htr- The servomechanism will calculate
that the valve's stroke for such a pressure should be:
(7.28)
where Sreq is the required stroke, c 5 is a constant which converts the

amount by which the transducer head exceeds the lower critical pressure
head of the valve into the required valve stroke.
If Sreq is not the same as Sj the actuator will start and the valve stroke
will increase or reduce at the valve's stroke rate in an attempt to reach the
required stroke Sreq However, during the 11t interval it may or may not
not be possible for the valve to reach the required position so the stroke
at the end of the dt interval will become either Sreq or Si Sr x dt where sr
is the amount by which the stroke can be altered per second by the actua-
tor, i.e. the stroke rate. Denoting this value by Sf it is now possible to cal-
culate the relevant K for the valve and then to solve the head upstream and
downstream of it together with the velocities in exactly the same manner
as for the motorised valve. The sign in the expression quoted above for
the value of Sf appears because the valve might be opening in which case
the + sign is applicable, or closing, in which case the -sign applies. This
means that if the value of cs(htr - hcrit) is greater than Si the +sign should
be used and if it is less than si the negative sign should be employed.
So Sf= Sj +sign (htr- her it) x Sr x 11t (7.29)
unless this value is greater than Sreq if the valve is opening, or less than
Sreq if the valve is closing, in either of which cases Sf= Sreq = cs(htr- hcrit).
Of course, the value of Sf so calculated becomes the next value of si
for the subsequent 11t period and so on.
7.6 Reservoirs
The reservoir constitutes a very simple boundary condition but reservoirs

may be equipped with any of the following different types of valve:
(I) a control valve which will be either open or closed, (2) a reflux valve
which only permits flow into the reservoir, (3) a reflux valve which only
permits flow out of the reservoir, ( 4) a part-closed valve.
It is probably uneconomic to include the part-closed valve case in the
reservoir analysis as it can easily be represented by a motorised valve, for
which the associated K array has all its values set to the same value equal
to that corresponding to the part-closed valve position, and to insert this
value one 6-x away from the reservoir. The valve's operating times must
then be set at zero for the time of commencement of the valve's operation,
and a time greater than the simulated time of the computer run for the
time of finish of the valve's operation. This method of representing the
part-closed reservoir valve is slightly erroneous and a little expensive in
computer run time as it involves K array fetching at every time level, but
it does save program size and the associated file store costs. On balance
the author favours this type of approach.
For a reservoir at which no valve is fitted the potential head can be
treated as constant if its plan area is relatively large and slowly varying if
it is not.
Then if the reservoir is located at the upstream/downstream end of a
pipe a backward/forward characteristic must be used.
p~~--+-""'-----+p
N 5 R N
Figure 7.9
+ _.!_ (hp - hs/R) + (Up - US/R) + 2/s;R us;RI us;RI M = 0 (7 .30)

cs;R ds;R
As hp = hres: the reservoir head,
then _ g (h _h ) _ 2/s;R us;RI us;RI 6. t

Up - US/R - res S/R (7 .31)
cs;R ds;R
The minus sign is used for an upstream and the plus sign for a downstream
reservoir. If a forward reflux valve is fitted this calculation is correct and
can be accepted as it stands but if the flow should ever attempt to reverse
and Up take a negative value this answer will be wrong because the reflux
valve will close under these circumstances and Up must be zero.
Therefore it is now necessary to test what type of reflux valve, if any,
is present. If it is of forward type and Up is positive, then Up can be accep-
ted but if it is negative it must be set to zero. If the reflux valve is of back-
ward type (i.e. it only permits flow in the negative direction- an inflow
to an upstream reservoir and an outflow from a downstream reservoir)
then the calculated velocity can be accepted if it is negative but otherwise
must be set to zero. If the valve is of the control type it may be fully open
or fully closed. If it is fully open the calculation can be accepted but if it
is fully closed the value of Vp must be set to zero irrespective of its size
or direction.
It is next necessary to calculate hp again as, if the valve is closed, the
head at the valve is no longer equal to the reservoir head,
SO hp = hs(R cs/R 2fstR vs/RI vstRI A t~

- - [Vp- VS(R +-....c...._..:__.:___
g d~R
The minus sign is used for an upstream and the plus sign for a downstream
reservoir. Note that the subscript S/R means that Sis to be used for an up-
stream reservoir case and R for a downstream reservoir case.
If the transient case is the only case of interest the above approach is
adequate as reservoir levels cannot change significantly during the very
small periods of time in which transient conditions are present.
If it wished to study the cases that can arise when the downstream
boundary condition is not a simple reservoir but may be a sewer outfall to
a manhole, a reservoir of small area, a small reservoir equipped with an
overflow weir to a channel, a rock stratum into which water is to be in-
jected, a badly defined exit point from the network into another network,
in which there is storage capacity plus a great deal of looping, a further
facility can be built in to the reservoir analysis.
Assume that the exit condition at the reservoir can be represented as
shown in figure 7.10.
plan area A
l 00
\~
I
p1pe ex1f
Figure 7.10
Essentially this consists of a tank of plan area A into which the pipe
discharges. Once the level in this tank has risen to the sill height of the
weir an overflow occurs and discharge is then into a tank of infinite plan
area.
The depth in the first tank rises according to the equation
df= d + (Qp- Qo) At

I A
If the level of the liquid in the tank is higher than sill level Q 0 will be given
by Qo= kB(di- hsin)notherwise Q 0 = 0, where dris the depth in the first
tank at the end of the time interval D.t, di is the depth in the first tank at
the beginning of the time interval D.t, Qp is the flow in the pipe, hsmis the
height of the sill above the pipe centre line, A is the plan area of the first
tank, B is the breadth of the weir, k is the weir constant, n is the weir
power index. Then the absolute potential head of the flow at the pipe exit
will be
hp=dr+z+ha
where z is the elevation of the pipe centreline above datum and ha is the
atmospheric head in height of liquid.
This model is extraordinarily flexible. By making A very large a normal
reservoir circumstance can be modelled. By making A small, B large and
n large an initial rise in reservoir level followed by a fixed reservoir level
can be modelled. By choosing realistic values a reservoir with an overflow
weir can be modelled. By taking the head-flow characteristic curve of
almost any downstream boundary condition and employing a little
ingenuity in picking appropriate values of k, B, n, hsill and A, it is possible
to duplicate this curve almost exactly. A realistic representation of many
boundary conditions is possible using this device.
7.7 Bends
Fully fixed bends in a pipeline do not cause any reflection of an incident

transient. However, if the bend is not fully fixed, the increase in the force
acting on the bend caused by the passage of a pressure transient through
it will make the pipes leading to and from the bend extend and the bend
will move correspondingly. This movement will generate a partial negative
pressure reflection. As the amount of movement is controlled by the effec-
tiveness of the bend's anchor block it is not possible to calculate the mag-
nitude of this reflection without modelling the block's behaviour. To do
this, information is required about the dynamic forces created between
the anchor block and the soil and this will not usually be available. The
negative reflections created are of the order of 10% of the incident wave
but it must again be emphasised that the magnitude of this fraction depends
upon many factors. Usually, the effect of bends is to diminish the maxi-
mum pressures experienced by the section of the pipe on the side of the
bend opposite to that in which the transient was initiated.
8 Unsteady flow in gas networks
8.1 Introduction
The methods presented for analysing unsteady flow in pipe networks

transporting liquids need modification before they can be applied to
highly compressible fluids such as gases.
The fundamental equations are essentially similar to those applicable
to liquids.
8.2 Basic equations
The continuity equation

The rate of increase of mass of an element of gas of length Ax equals the
difference between the rates of entry of gas to the element and exit of
gas from the element.
So a~ (pAox) = pAv- (pA v +a: (pAv) ox)

a a
at (pA) +ax (pAv) = 0
ap aA ap aA av
Aa-r+par-+Avax +pvax +Apax=O
ap p aA ap pv aA av
at+ A at+ v ax+ A ax +p ax= O
In pipelines transporting gases the pressure fluctuations are not suf-
ficiently large to cause any pipe distension of significance so aAjat is
very small and can be neglected.
The continuity equation therefore reduces to
ap ap av pv aA
- + v - + p - + - -=0 {8.1)
at ax ax A ax
147
The dynamic equation
Applying Newton's second law of motion to an element of gas of length
ox the net force acting in the direction of flow (i.e. in the direction of x
increasing) equals the rate of change of mom en tum of the element.
a ) aA dv
pA- (pA +ax (pA) ox +pax ox- Fox= pAoxdt
The p ~:ox term is the longitudinal force acting upon the projection
of the increment of area in the longitudinal direction. F is the frictional
force acting per unit length of pipe,
a aA dv
-ax (pA) +pax- F= pA dt
ap av av
A ax + F + pAv ax+ pA at= 0
ap av av F
ax + pv ax + P at+ A= 0 (8.2)
The equation of state of the gas

It is not necessary to involve an equation of state for a liquid but, for a
gas, an equation of state is necessary because of the large changes of den-
sity and temperature caused by the compression or rarefaction of the gas
when a transient pressure passes through it.
It may be thought sufficient to use an adiabatic, an isothermal or poly-
tropic process according to circumstance but the best method is to use
the equation of energy.
This can be derived as follows:
For a perfect gas the internal energy per unit masse is
e=C T=-1-!!...
v r-Ip
where Cv is the specific heat at constant volume,
also
where Cp is the specific heat at constant pressure. Tis the absolute tem-
perature.
Let the perimeter of the gas element be sand the rate of heat inflow
to the system be q units of heat per unit area.
Then heat inflow to the element
=qsox
Unsteady flow in gas networks 149
The rate of increase of the energy of the element with time
The rate at which energy leaves the element axially
=a: (pAv(e + v; )) 5x
The rate of work done by the element against pressure forces

a
= - (pAv)5x
ax
By the first law of thermodynamics: rate of heat inflow= rate of increase
of stored energy +net rate of energy outflow+ net rate of work done by
the system. Combining this equation with the continuity and dynamic
equation:
(8.3)
8.3 Characteristic equations
By a similar method to that employed before, the characteristic equations

can be obtained.
dv c dp dx
-+--=1 along dt = v + c {8.4)
dt 'YP dt
dv c dp_E
dt-w dt- 2
dx
along dt = v - c (8.5)
dp -c2 dp= 3 dx
along dt = v (8.6)
dt dt
where
E = r-1 !1.. + .f_[v(r-1) 1] -!!.E.. aA (8.7)
1 c pm pA c A ax
vc aA
+-- (8.8)
A ax
(8.9)
m =hydraulic mean radius of the duct. aA{ax is the rate of increase of

area of the duct. In a parallel sided pipe ~~ = 0.
p
s
Figure 8.1
Along RP equation 8.4 applies and is a forward characteristic. Along

SP equation 8.5 applies and is a backward characteristic. MP is a particle
path and along the particle path equation 8.6 applies.
Combining the finite difference forms of equations 8.4 and 8.5 and
eliminating Vp gives:
L Vs +_!_(cR- c)~
Pp = ( 'Y )lvR- + (E1 - 2 ) dt (8.10)
CR +~ 'Y
PR Ps
Substitutingp p back into the finite difference form of equation 8.4
vp=vR+ cR (pR-pp)+E 1 dt (8.11)

'YPR
Usingpp in the finite difference form of equation 8.6
PP =PM+~ [pp- PM- E3dt) (8.12)

CM
Relevant values at R, M and S must be obtained by interpolation as
demonstrated in chapter 4.
Values at P can be calculated using equations 8.1 0, 8.11 and 8.12.
Any mid-pipe point can be established in this manner and the technique
is very similar to that illustrated earlier for liquids but because of the need
to use three characteristics the process takes rather longer and costs rather
more.
The above presentation follows closely the relevant portions of a paper
by Edgell 16 .
8.4 The value of q
q is the rate of heat inflow per unit area of pipe wall.

In gas mains it may be considered sufficient to calculate q on a quasi-
steady heat flow basis. This approach assumes that at any instant the
steady flow of heat through the pipe wall appropriate to the temperature
difference across it is a sufficiently accurate estimate of q. This ignores the
heat required to raise (or lower) the temperature of the pipe-wall material
itself.
Should it be felt that this quasi-steady heat flow method is acceptable
then q may be estimated as follows.
inside of external
p1pe environment
pipe wall
Temperature distribution across the pipe wall
Figure 8.2
Newton's law of heat transfer can be used, i.e.

q=CD.Te
where D. Te is the temperature difference.
In steady state
= G.v ( Tw2 2
- Tw 3 ) = Cw 3 ( Tw 3 - Te) (8.I3)
where Cr is the coefficient of heat transfer from the fluid to the wall,
Cw1 is the coefficient of heat transfer through the wall, Cw 2 is the coeffi-
cient of heat transfer through the lagging, and Cw is the coefficient of
heat transfer from the lagging to the external envi~onment.
Cqf =TJ.-TwI
q
- =Twl- Tw.
CwI
Cq = Tw,- Te
w,
q (....!.._ + _I_ + _I_ + _I_) = T1 - Te

Cr Cw I Cw 2 Cw 3
_ TJ.-Te
q- I I I I
-+-+-+--
Cr Cw1 Cw 2 Cw,
(8.I4)
where Cc is the compound coefficient of heat transfer, i.e.
q = Cc(T1- Te) (8.15)
It now only remains to calculate Cf, Cw,, Cw, and Cw 3
The transfer of heat from the gas to the pipe wall

In laminar flow:
K ( 11 ) o.14 (pvd C!J. d )o .33 (8.16)
Cf=1.86- - -- - -
d !lw , 11 K L
where Cr= heat transfer coefficient in watt per metre squared per kelvin.
K =thermal conductivity of fluid in W m- 1 K- 1
1K=1C
1 Js- 1 = 1 N m s- 1 = 1 W
c =specific heat per unit mass in J kg- 1 K- 1
!lw = dynamic viscosity of fluid in kg m -I s-1 units at the pipe wall tern
perature Tw
11 =dynamic viscosit'y of fluid in kg m- 1 s- 1
d =internal pipe diameter in m
L =pipe length in m
v = mean flow velocity
p =mass density of fluid in kg m- 3
In turbulent flow (Re > 2300):
K(dv) o.s (C!J.) 0.33 ( !1 ) 0.14

Cf= 0.027-- - - (8.17)
d V K !lw
where dis the internal diameter of the pipe, vis the kinematic viscosity of
the fluid at fluid temperature.
The transfer of heat through the pipe wall
C = 2rrK
w, loge(~::) heat units per second per metre per kelvin (8.18)
where K is the thermal conductivity of the pipe wall material in W m- 1 K- 1

dw, is the internal diameter of the pipe
dw, is the external diameter of the pipe.
The transfer of heat through the pipe insulation

This is essentially the same case as for the pipe wall.
C = 2rrK (8.19)
w, loge(~)
where K is the thermal conductivity of the insulation,
dw 3 is the external diameter of the insulation,
dw 2 is the internal diameter of the insulation.
Note that Cw and Cw are the heat losses in watt per metre length of pipe
per kelvin. They are different from Cr above and Cw 3 immediately below
which are heat losses in watt per square metre of pipe wall surface per
kelvin.
The transfer of heat to the external environment

If the outside environment is air:
(8.20)
where Cw 3 =heat transfer coefficient in W m- 2 K- 1

dw3 =outside diameter of the insulation
K= thermal conductivity of air in W m- 1 K- 1 .
Re = Reynolds number Vd/v where Vis the wind speed in m s-1 .
If the external environment is soil (i.e. a buried pipeline):
(8.21)
where Z is the depth in metres to the pipe centreline

dw is the ex temal diameter of the insulation.
K i~ the soil thermal conductivity in W m- 1 K- 1.
If flow in the pipe is laminar (a most unlikely circumstance in a gas
pipeline) the value of Cw, will depend upon the temperature difference
11 - Tw, and as Tw, is not known an iterative method of solution will have
to be used.
If this quasi-steady heat flow approach is not regarded as sufficiently
accurate then a more accurate method may be used. This is based on the
heat diffusion equation, i.e.
(8.22)
K
where ~=--
pC
e
and is the temperature difference. This equation can be solved by finite
difference methods but this will lead to a great increase in the program size
and complication and may well be regarded as an unacceptable method.
8. S Boundary conditions
Boundary conditions can be solved by methods similar to those described

for liquids but almost invariably the equations produced are of much
greater complexity and usually have to be solved by iterative methods such
as perturbation or Newton-Raphson techniques. The boundary conditions
that are of interest in gas pipelines include the following.
(1) Upstream or downstream reservoir: this is a tank from which gas
may flow or into which it is delivered.
(2) Turbo-blowers or compressors: these act in the same manner as
pumps in liquid networks and can be handled in a surprisingly similar way.
The energy per unit weight supplied to the gas is given by the equation
H =AN 2 + BNQ - CQ 2
where Q is the flow in m3 s- 1 at inlet pressure and temperature. The H- Q
curve is supplied by the blower manufacturer in the same way that the
manufacturers of pumps supply the pump curves.
Once H has been determined from the flow value the pressures can be
calculated from:
n-1
(8.23)
where n is the polytropic index, p2 is the discharge pressure and p 1 is the

inlet pressure. 1i is the temperature at inlet (absolute).
n can be estimated from:
{8.24)
where ep is the polytropic efficiency of the blower. The polytropic ef-

ficiency must be known or estimated from previous experience.
(3) Junctions: these can be analysed by the same method as that demon-
strated for the case of liquids but using the gas characteristic equations.
9 Impedance methods of pipeline
analysis
9.1 Introduction
It is possible for resonance to occur in pipe networks. Organ pipes and

other musical wind instruments operate by generating such a resonance.
Small pressure or flow fluctuations applied at one end of a pipeline can
superimpose upon one another if their frequency matches a simple mul-
tiple of the pipeline's period leading to the development of a standing
wave of considerable magnitude. This phenomenon has caused some major
catastrophes: pipelines have ruptured when the forcing vibration has been
of trivial magnitude. One of the circumstances which will generate small
fluctuations of pressure or flow is the oscillating valve. A very common
example of this is the vibrating ball valve in a domestic water supply. As
the level in the cistern rises it raises the ball which progressively closes
the valve. If the valve is very nearly closed, a small further closing move-
ment shuts off all flow causing a pressure wave to travel down the supply
pipe. The pressure rise tends to force the valve off its seat and to let flow
recommence so initiating a negative pressure wave which starts off down
the pipe following the initial positive wave.
The valve is forced back onto its seat by the upthrust upon the ball. The
rubber valve seat and the mass of the moving parts of the valve ball and its
arm constitute a mass and spring circumstance which has a natural frequency
of vibration. If the pipeline has a harmonic (i.e. a multiple of its funda-
mental frequency C/4L) which matches this mass-spring frequency, a reso-
nant vibration will be generated. This can be heard as a low thrumming
throughout the house and, if it is allowed to continue for a sufficiently
long time, the pipe may burst. To stop it, it is only necessary to arrest
the ball's vibration by touching it but to prevent it from occurring it is
necessary to either alter the hydraulic parameters or the mechanical para-
meters of the ball valve. Changing the valve's rubber seat alters its stiffness
so changing the spring constant in the mass-spring circumstance of the ball
valve. By thus changing the forcing frequency the resonance may be pre-
vented but usually the rubber alters its stiffness after some period of use
and the problem reappears. By ftxing a square plate to the base of the
155
ball in a horizontal position two mechanical parameters are changed:
(a) the mass of the ball and arm, (b) the frictional damping of its motion
is increased. This often solves the problem.
The pipe can be replaced with one of larger diameter in which velocities
and velocity variations are smaller. This involves the generation of smaller
pressure transients so reducing the magnitude of the forcing vibration.
Frictional damping already present in the system may then be sufficient
to prevent the development of resonance.
This simple example of the problem illustrates many of the features
that occur in very large scale pipelines. In hydroelectric installations, for
example, various mechanisms can generate resonance. In the reservoir
supplying the scheme, wind driven surface waves of amplitude 1.5-3
metres can be generated. These can have a frequency which may coincide
with a harmonic or the fundamental of the pipe system and resonance
can thus be generated. In large pipes it is usual to shut down a flow by the
use of a butterfly valve; such valves do not usually seal perfectly and a
rubber seal located around the periphery of the valve disc is inflated with
oil (or water from the pipeline) after closure has been completed. If for
any reason such a seal leaks, and hence the valve's sealing becomes imper-
fect, leakage over the seal may cause it to flutter, generating small pres-
sure waves travelling up the pipe. As described earlier this can cause
resonance and can cause pipe bursts under apparent no-flow conditions.
Servocontrolled valves may also cause resonance if they have a natural
frequency which matches that of the pipeline and are inadequately
damped. Governor controlled spear valves of Pelton Wheels and the gover-
nor controlled guide vanes of Francis Turbines are examples of such valves.
It must be emphasised that the assessment of a resonance risk is not an
academic exercise but is of real practice importance. If there is the
slightest possibility of a resonance occurring, a resonance analysis should
be performed to assess the risks involved and to ensure that palliative
measures are effective.
Resonance analysis can be performed by the characteristics method
as described in previous chapters and if no better method were available
this would be the one of choice. However, an analytic technique exists
which is adequately accurate and which can be used to provide a computer
solution in a very small fraction of the time taken by a characteristics
technique.
9.2 The analogy between electrical and hydraulic impedance
The method is based upon an analogy between electrical and hydraulic

flow. In the theory of the transmission of alternating current over
transmission lines two equations are used which bear considerable similarity
to those of waterhammer. This can be seen by comparison of equations
9.1 and 9.3 and equations 9.2 and 9.4.
Impedance methods of pipeline analysis 157
Transmission line:
(9.1)
av
-+L-+R
ai .
ax 1z=O (9.2)
at e
Pipeline:
ah ah c 2 au
-+u-+--= 0 (9.3)
at ax g ax
(9.4)
where C in the transmission line equation is the capacitance/unit length,

L is the inductance/unit length, Rei is the resistance/unit length of the
transmission line, Vis voltage and i is the current. The variables in the
waterhammer equation are as defined elsewhere in this book.
If the potential head is seen as analogous to the voltage V and the flow
ah
Av as analogous to the current i and, if the v term in equation 9.3 and
ax
the~aav
g X
term in equation 9.4 can be ignored, the analogy becomes exact
2jVIvl
prov1.d.1ng t h at t he - d - term can be 11neanse
. d.
The v ~~ term is small compared with the ~~term if the wavespeed is

large as ~~ ~ (v +c)~~ So if it is assumed that c is constant and large this
term can be neglected in equation 9.3; similarly in equation 9.4 the
E. aav term is small compared with the other terms if c is large and constant,
g X
so it can be neglected under such circumstances.
Equations 9.3 and 9.4 can now be rewritten:
ah+~ aq = 0 (9.5)
at gA ax
(9.6)
9.3 The linearisation of the waterhammer equations
The remaining question concerns the linearisation of the 2..:{ ~~I term.
This term cannot be linearised as it stands but if the flow q is regarded
as made up of a steady component plus an oscillating component the
term can be linearised.
Thus let q =q + q'
where q is the steady state flow and q' is the unsteady flow component.
aq aq aq'
Then -=-+-
ax ax ax
By definition~; is zero as lj is constant throughout the length of the

pipe so
aq _ aq'
ax- ax
aq aq aq'
Similarly at =-ar+ar
and ~; = 0 as the flow q, being steady, cannot vary with time,

aq- aq '
so a-r-ar-
Similarly the head h can be split up into a steady head and an oscillating
component, i.e.
h=h+h'
The value ~~is the hydraulic gradient and so equals- 2{d~1 1

The value~~= 0 as the steady head component cannot change with time.
Thus equation 9.5
a(Ji+h') c 2 a(q+q')
at + gA ax O
becomes
ah' c2 aq'
-+--=0 (9.7)
at gA ax
and equation 9.6
a(Ti. +h') 1 a(-+

---'-:,....----'-+- q q')1
q q')IC+
q q') + 2/C+ =0 (9.8)
ax gA at gdA
becomes
-2fqlql +~ + _!_ aq' + 2f(l! + q')l(q + q')l = 0 (9.9)

gdA 2 ax gA at gdA 2
If the analogy is to be drawn between friction and resistance the f value
must be treated as a constant / 0 and for this to be done the terms
2fqlql 2f(q + q')l(q + q')l
and
gdH 2 gdA
must be rewritten
2/oQn
-- and
gdAn
where n takes a value between 1. 7 5 and 2.0 according to the roughness
of the pipe.
Then the term 0
.
2!( (if+ ')n
g,
dAn
q can be expanded by the use of the Binomial
equation
2fo(q+q')n _ 2/o ('-n+ -n-1 '+n(n-1) n-2 2 )
gdA n - gdA n q nq q 1 2 q q
If q' is small compared with q, second order terms can be ignored so
2/o (q + q')n 2/oQn + 2nfoQn-lq 1
gdAn gdAn gdAn
so equation 9.9 becomes
-2~'-n ah' 1 a, 2~'-n 2nr-n-l ,
---'J:..::(O_,_Q_ + _ + _ _!!__ + ...l!!!L_ + JoQ q =0
gdAn ax gA at gdAn gdAn
Therefore it reduces to
a , 2nr-n-l ,
+ _1 _!]_+
-ah' JOQ q (9.10)
ax gA at gdAn
2nfoqn-l
Denote byR
gdAn
Then the hydraulic equations are
ah' c 2 aq'
-+--=0
at gA ax
[9. 7] *
~+_l_aq' +R '=o (9 .11)
ax gA at q
Comparing these with the electrical equations
av+_!_ai=o [9.1]
at cax
aV + Lj!_ + Reti = 0 [9.2]

ax at
*Square brackets indicate that the equation with this number was introduced earlier.
it can be seen that the hydraulic equivalent to the capacitance/unit length
Cis gA/c 2 , the hydraulic equivalent to the inductance/unit length L is
1/gA and the hydraulic equivalent to the resistance/unit length Rei is R
2nfolfn-l 32v
which equals gdA n If the flow is laminar R becomes gd 2 A
from the Hagen Poiseuille formula. The [ 0 value used in this development
is the British form not the American form although/0 is known in the
USA under the name of the Fanning[. The form commonly used in the
USA is equal to four times the British form as shown below:
British USA
hr= 4[Lv2 [Lv 2
hr=--
2gd 2gd
As a matter of interest, the reason that the British form retains the 4
in the numerator is that it is derived from the basic Darcy-Weisbach form,
[Lv 2
hr = '2gm
where m is the hydraulic mean radius. As m = d/4 for a circular pipe, the
British form results immediately and the f values apply to pipes of any
cross section.
9.4 The solution of the linearised waterhammer equations
First differentiate equation 9.7 with respect to time

a2hl c2 a2ql
at2 =- gA axat
a2
~--L
A a2 h
I
1
axat - c 2 at 2
Rearranging equation 9.11
a..!!...= -gA (ah
I
-+Rq
1
1)
at ax
Differentiating with respect to x
but a;a =-~at

I A i1h 1
from equation 9.7
a2 h' RA ah' 1 a2 h'
so ax 2 -g7at-c2 at 2 =O (9.12)
This is the wave equation ( cf equation 2.18 in which friction was ignored).
If the wavespeed is assumed to be constant then, although the wave
may attenuate with distance along the pipe, an oscillatory wave will not
alter its amplitude at any particular point on the pipeline. Thus, for the
case of a sinusoidal pressure head applied at the upstream end of a pipeline,
a solution of equation 9.12 must be:
h'=Heint
where His the amplitude of the wave
i.e. atx = 0
at x =x 1
h~~ = Hx,. eint
(Note eint =cos n t + i sin n t where i = v'-1
n = 2rrf where f is the frequency of the oscillation)
a h' a H
2
----e'.nt
2
ax 2
2 - ax
ah' -rm int
ar=' e
a2h' .
--=-n 2 He'.nt
at
2
Substituting these values into equation 9.12

a2H m -g-iDHe
--e' RA .n Q 2
1 t+-He 1.nt=o
ax2 2 c 2 c
so
(9.13)
denote
then (9.14)
In transmission line theory 'Y is called the propagation constant.

Equation 9.14 is well known and is called the harmonic equation.
A solution of equation 9.14 is of the form
H=C1emx
Then
a2
H
--=Cm2emx
3x2 I
So Cim 2emx = 'Y2 Cl emx

:. m2 =r2
:.m =r
The required solution is therefore H = ae -yx +be --yx where a and b are
constants.
So
h' = (ae'Yx + be-'Yx) ei.nt (9.15)
To obtain the solution for q' differentiate equation 9.15
ah' =in (ae'YX + be-'YX) ei.nt

at
but from equation 9.7
a ' A ah'
_!f_=_G!!__
ax c 2 at
q' = -~ inei.nt J (ae'Yx + be-'Yx)dx
q' = g~n eintlae-yx_ be_'YJ (9.16)

ZC 2 'Y ~ ]
This is the general solution of the wave equation for any sinusoidal oscil-
lation.
9.5 The evaluation of 'Y
'Y is complex as can be seen from its definition (equations 9.13 and
9.14) so it will have the form
r=a+i(3
Consider a de Moivre diagram (figure 9.1).
If a and (3 are both positive real numbers, 'Y must lie in the first quad-
rant.
Now
2_ n2
'Y - - - 2 +
iAgnR
2
c c
The real part of r 2 is thus negative and the imaginary part is positive so
r 2 must lie in the second quadrant.
Now r 2 =(a+ i(3) 2 = r'Y2 eiop'Y 2
Figure 9.1
and r = 0: + i(3 = r..,ei<P-y

r.., = (r..,) 2
and <P.., = 2<P-y from the simple theory of complex numbers
r.., =Jcx..,2+(3/
and r.., =JOG{ + (3~
n2
Now 0:-y2 = -~
and
so
r
'Y
= Ag.Qj(R)2
c2 gA
+ R2
and <P-y = rr- tan- 1 ((3-y\

0:-y .,
<P-y = rr- tan- 1 (R~)
Now <P-y=i<P-y =~-ttan-I(R~)

and r-y=~=H ((~f +R2Y'4
but
but cos (~-A) = sin A

a~ =R[(g~r +R2r14 sin(~tan-l(R;A)) (9.17)
(3~ = r~ sin 1{1~
f3~ =~ [(g~r + R 2 J 1
\os (~tan-l(R~A)) (9.18)
9.6 The impedance concept
The risk of resonance occurring in an electrical network can be assessed by

investigating the way in which its impedance varies with the frequency of
the voltage of the applied signal. This method can be applied to oscillatory
flow circumstances in a pipe network. The hydraulic concept of impedance
is based on the analogy between electrical and hydraulic variables, i.e.
voltage and head, current and flow. Electrical impedance is defined as the
ratio voltage/current so hydraulic impedance is defined ash' fq'. As in elec-
trical theory the D.C. components of voltage and current are ignored so,
in hydraulic impedance theory, the D.C. components of head h and flow
q are also ignored.
Thus
Z(x) = h'lq'
Z(x) =ic 2 -y (ae~x + be-~x) (9.19)

gAn ae~x-be ~x
-ic 2 -y
c 2 -y
The term
gAn = igAU
has the dimensions of impedance. It is called the characteristic impedance
and is denoted by Zc.
At the end of a very long line the values of h' and q' would both tend to
infinity unless a is zero.
Thus the impedance of an infinitely long line is given by
Zoo=-Zc
so -Zc is the impedance of an infinitely long pipeline extenC:ing in the x
direction and +Zc is the impedance of a pipeline extending in the -x direc-
tion.
In an infinite pipeline no reflected wave can be travelling in the oppo-
site direction to the incident wave so the characteristic impedance can only
be used in connection with heads and flows of a wave travelling in one
direction.
(9.20)
9.7 Receiving and sending ends
Following the practice used in transmission line theory the sending end will
be assumed to be at the x = 0 position and the receiving end at x = L posi-
tion. This assumes that the forcing oscillation of pressure head is located
at the x = 0 position.
In some other texts this practice is reversed and readers are warned of
this difference should they consult such texts.
Hs HR
L
s
I Q(x)
H(x) R
X
Figure 9.2
S denotes sending end and R denotes receiving end.
9.8 The equation of impedance
The change in impedance from one end of a pipe to the other will be de-
rived next.
Referring to figure 9.2, at x the oscillatory head component is given by
and the flow by q~ = Qxeintwhere Hx is the amplitude of the head oscil-

lation at x and Qx is the amplitude of the corresponding flow oscillation
atx.
At the sending end where x = 0
h~ =Hoeint=Hseint
q~ =Qoeint=Qseint
From equation 9.15

h~ =Hseint =(ae-rxo+be--rxo)eint
Hs =a+ b (9.21)
Similarly for q' from equation 9.16
q' =Q eint=~Ail (ae-rxo _ be--yxo) eint

0 S IC 2'Y
gAil
Qs=~(a-b) (9.22)
IC 'Y
Adding equation 9.22 to equation 9.21
ic2r
2a=Hs + - Qs
gAD
a= ! (Hs - ZcQs)
so b = ! (Hs + ZcQs)
h~ = eint (HHs- ZcQs)e'Yx + ~(Hs + ZcQs)e-'Yx
= eint (Hs (
e'Yx + e--yx) _
2 ZcQs
(c'Yx
2
-e--yx))
e'Yx +e--yx
but 2 =cosh (rx)
e'Yx - e--yx
and 2 sinh (rx)
h~ = eint (Hs cosh (rx)- ZcQs sinh (rx))

Similarly
q~ = eint (-~:sinh (rx) + Qs cosh (rx))
_ h~ _ Hs cosh (rx)- ZcQs sinh (rx)
Zx - ----, - ----:-;;---'----____::___ __
q -Hs
x -sinh (rx) + Qs cosh (rx)
Zc
Hs- ZcQs tanh ('yx)
Hs
Qs --tanh (rx)
Zc
Zs- Zc tanh (rx)
z X
=------
Zs
I - - tanh (rx)
Zc
At x = 0 tanh (rx) = 0 and the above result reduces to a simple equality
Zo = Zs
Atx =L
Zs- Zc tanh (rL)
ZL = ZR = Zs (9.23)
I --tanh (rL)
Zc
Re-expressing Zs in terms of ZR by manipulating equation (9.23)
z s =ZR
_:_:__+ Zc _ _____:(rL)
__::___tanh _:______.:.
ZR (9.24)
I +-tanh (rL)
Zc
From these results for Zs and ZR the change of impedance from one end
of a pipe to the other end can be calculated.
9.9 Boundary conditions
In previous chapters, boundary conditions were described and the way in

which they interact with incident waves producing reflected waves was
discussed. In a very similar way it is necessary to describe the boundary
conditions in a resonant network so that the impedance at a pipe end may
be specified and to link the impedances of constituent pipes in the network
to one another to obtain the impedance of the entire network.
Reservoirs
At a reservoir the head is rigidly controlled and must equal the height of
the reservoir surface above the datum. Thus h equals the height and h'
must be zero.
Therefore at a reservoir the ratio h'/q' = 0 irrespective of flow.
Z=O (9.25)
A blank end
At a blank end of a pipe the flow must be zero irrespective of head so 7j
and q' must both equal zero.
h'
Z=----;=oo (9.26)
q
A junction
For a junction the head is the same for all pipes joining there and also
~Qin = ~Qout
As the head is the same for all pipes joining at the junction
, ,
~ Qin= ~ Qout
h' h'
1 I
i.e.~-=~--
Zin Zout
For example consider a four-way junction (see figure 9.3).

Figure 9.3
Suppose that Zs 3 , Zs 4 and ZR 2 are known by calculation from their

other ends then
1 (
ZR = -
1
z;-+
1
l
Zs +Zs1)
I
4 3
A joint
This is a two way junction (see figure 9.4) and so
CD _ __ . r - -
Figure 9.4
Series sections of a network are merely a series of pipes separated by

joints and so are simply dealt with as illustrated in figure 9.5.
Figure 9.5
Now ZR 3 = 0 as R 3 is at a reservoir.
The characteristic impedance of pipe 3 is given by equation 9.20, re-
quoted here.
(9.27)
~3 and a 3 are calculated from equations 9 .I7 and 9 .I8 so Z c 3 can be evalu-
ated from equation 9.27
ZR = O =Zs 3 - Zc 3 tanh ('yL)

3 Zs
I - z 3
tanh ('yL)
c3
so
and Zs 3 can be evaluated.
From the joint boundary condition

ZR, = +Zs3
then
ZR = Zs, - Zc 2 tanh ("yL 2 )
2 z
I - -r:- tanh ('yL2)
c,
As ZR is known, Zs can be obtained once Zc has been calculated.
Again ZR 1 = Zs 2 an"J in the same manner as fo; pipe 2 Zs 1 can be found.
Loops
Zc can be calculated for every pipe in the network (see figure 9.6).
Figure 9.6
Then
so Zs 4 can be calculated.
At the junction
-I= ( -I+ -1 -)
Zs ZR 2 ZR 3
and
Z _ Zs 3 - Zc 3 tanh (Y3L3)
and z
-zc3s3 tanh (-y3L3)
R3
1
At the upstream junction
1 ( 1 1 )
ZRI = Zs2 + Zs3
Additionally by the reciprocal rule for adding impedances in parallel
1 1 1
--- = + (9.29)
Zs 4 -ZR I ZR 2 -Zs 2 ZR 3 -Zs 2
- Zc tanh (-y1Ld
and ZR =Zs I I
I z
1--j; tanh(-y 1Ld
cl
Sufficient equations have been obtained to solve all the impedance
values.
When two loops overlap as shown in figure 9.7 a circumstance called
second order looping exists.
1
1
' - - - -- -
1
I
First order looping Second order looping
Figure 9.7
The author is not aware of any method of analysing second order looped
networks.
9.10 The impedance of a network
By starting from a point in a network at which the impedance is known,

e.g. a reservoir or a blank end it is possible to work back up the network
calculating impedances en route until the sending end is reached.
blank end~
R3

53 s
Figure 9.8
The process is illustrated below by reference to figure 9.8.

zR. =zR. =o zR, =oo
ZR + Zc tanh (rsLs)
Zs = s
Then s
z
s
1 +2
c,
tanh(rsLs)
As
this reduces to:
Zs s = Zc s tanh (rsLs)
Similarly
and
Values of the impedance of the downstream network at every node have
now been calculated. It must be appreciated that all values of impedance
are complex. The programming language Fortran has the facility of specify-
ing variables as complex and has the necessary libraries for performing com-
plex algebra. The language Algol 60 does not possess these facilities so des-
pite its many advantages over Fortran it is not as well suited as Fortran to
perform resonance analyses and the programmer must be prepared to write
his own procedures for performing complex algebra - a comparatively
simple task.
9.11 Harmonic analysis
The result of performing the calculations for the network impedance illus-
trated above is a series of complex numbers, the magnitudes of which
depend upon the angular velocity n, so the moduli of these numbers varies
with the value of n.
By calculating the modulus of the value of Z for any point in the net-
work for a range of values of n, incrementing n in, say, 0.01 steps, a curve
such as that illustrated in figure 9.9 results.
z
f\
I I
......-
/
I I
\
-
Figure 9.9
From the diagram the risk of a resonant condition being developed in

the network, downstream of the point under consideration, by the applica-
tion of a forcing oscillation at that point can be assessed. If the frequencies
which cause resonance cannot occur, the network is safe but the likelihood
of their development must be assessed with great care. The actual magnitudes
of the amplitude of the oscillatory pressure depends upon the amplitude of
the forcing oscillation and unless this can be specified the magnitude of the
peak pressures cannot be calculated. However, it may be sufficient to esti-
mate dangerous frequencies and to ensure that they cannot occur. The
method of determining the magnitude of the forcing oscillation is discussed
in the next section.
It is possible to perform a harmonic analysis of the type described above
by using characteristic methods. Each run of a characteristic program would
cost more than a complete run of an impedance program. A harmonic ana-
lysis would require many runs so the characteristic approach is not suitable,
for economic reasons.
9.12 The forcing oscillation
Resonance occurs due to the interaction of waves generated by the forcing

oscillation with those reflected from distant parts of the network. For
resonance to develop, the frequency of the forcing oscillation must match
a harmonic or the fundamental of the network. The fundamental frequency
of the network cannot be easily specified unless it is of very simple type.
The frequency of natural vibration of the network is not simply related to
the pipe periods of the pipes making up the network, the situation being
greatly complicated by the partial transmission and reflection of waves at
junctions.
Even in a simple branched network no simple relationship between the
frequency of the fundamental and that of the pipes in the network exists
because different branches cause reflections to arrive back from their
downstream ends to their upstream junction at different times. The har-
monic analysis described in section 9.11 is the only economic way of cal-
culating the frequencies of the fundamental and its harmonics. To calculate
actual pressure heads in the network it is necessary to supply the amplitude
of the forcing oscillation. Sometimes this is very simple to do, as in the
case of a network connected to a reservoir upon the surface of which waves
are generated by wind and these waves constitute a forcing oscillation. The
wave forms may not be simple sine waves but may be made up of a num-
ber of sine waves of different frequencies, but these can be determined by
Fourier analysis of the wave form. By performing the calculations to obtain
the pressure head amplitude at various points in the network for each con-
stituent sine wave, superposition of the results will give the resultant wave
amplitude at all points required.
To demonstrate how the amplitude of a pressure wave is calculated,
consider a simple pipeline.
At the upstream (i.e. sending) end assume that a forcing oscillation is
present of known amplitude H 0 , i.e.
so
As Zs will have already been evaluated in the impedance analysis
1 h~
qo= Zs so Qs =Hs/Zs
At a point x downstream
h~ = e int (Hs cosh('yx)- ZcQs sinh( rx ))
so Hx = Hs cosh(r.x)- ZcQs sinh(r.x)

and similarly
Hs.
Qx= Qs cosh(r.x) --smh(r.x)
Zc
The pressure head amplitude and the amplitude of flow fluctuation can
thus be determined anywhere in the network.
As mentioned earlier, the forcing oscillation can be generated by a
variety of mechanisms. As an example of the method of dealing with such
a forcing oscillation the technique of analysing an oscillating valve will be
demonstrated in the next section.
9.13 The oscillating valve

The head drop across a valve due to flow expansion after passing the valve
orifice is
Kv 2
h=- (9.30)
2g
where K is a constant which varies from low values to high values as the
valve closes from its fully open to its fully closed position
ap
q=7KV2ih (9.31)
where ap is the cross sectional area of the pipe.

Denote the value }K
a
by ae the 'effective' valve area.
q =aeVfih (9.32)
As before
q= q + q' = ae ..}2g(Ti + h')
This equation must be linearised if it is to be used in impedance theory.
Consider the term (ii + h' ).
Now (l + h' )2 h' h' 2

2h = l + h + 4fi2
h'2
Providing that the term 47i 2 is small in comparison with the term h'jii,
which will be true if h' is small relative to h, then
(l + ~~r ~ h; h'
So
Then
Expanding and remembering that q ="iie .J2ih gives

_hi a~ _ h1
=aeV2ifl + q -+-:::-q-
I I
q
27i ae 2n
If ae0 is the effective area of the valve when fully open
ql =ae ~o~.J)ih +q[~ +a~ ~

ae ae 0 2h
hJ
ae0 ae 2hJ
ae ae
- I
r::;;y
Denote- by and -bye and remember that q = llev 2gh.
1
aeo aeo
Then ql=q h1 (I+-=~

[-+ h') I]
2Ti 2h
(9.33)
If the valve's oscillatory amplitude is small this approximates to
q
1 -[h
=q ili+"f
1 1
] (9.34)
Note denotes the ratio of the valve's steady flow area to its fully
open area and 1 denotes the ratio of the oscillatory component of the
valve's effective area to that of the full open valve.
Thus
z =--qr= -(h
h1 h1
1 1
(9.35)
)
q --=+-
2h
Usually a valve's position depends upon the head at the valve. Valves
which can be set into the oscillatory motion described here are controlled
by a servomotor which causes them to start opening further if the head
upstream becomes larger than a certain set value or to start closing if the
head downstream exceeds a certain set value according to which pipeline
(the upstream or the downstream one) is to be protected against excessive
pressure.
The precise way in which the valve behaves depends upon the servo-
mechanism used but usually the valve's movement will lag behind the ap
plied head by a phase angle <f!.
It is therefore possible to write an equation for the oscillatory motion
of the valve of the type:
where ~is the amplitude of the oscillation of the valve oscillation,

n = 2rrf where f is its frequency of oscillation and <{! is the phase angle of
the valve's motion. Assuming that these values <~;re known and h 1 is given
by the usual oscillatory head equation h 1 = H 0 e 1n.t
then
h' H eint
Z =q' = -(Ho_:i.nt + tei(.nt-.p))
q 2h t
z" ~r~+_;v"
~ Ho
J (9.36)
Note If t = 0, i.e. there is zero valve amplitude, the problem reduces
to that of a fixed valve opening or orifice. This case is thus defined by
Z= 2}j
q
(9.37)
Now Zs can be calculated from consideration of the downstream net-

work independently and when Z = Zs a matched impedance condition is
achieved, then H 0 can be solved and the forcing head amplitude generated
by the valve is obtained.
As before, this value can then be used to calculate pressure head am-
plitudes anywhere else in the network.
9.14 A network in which resonance can be excited by forcing oscillations

located at different points in the network
Examples of such networks occur when, for example, a network has in-
takes from two (or more) reservoirs upon the surfaces of which surface
waves - perhaps of different type and frequency - occur. Again, a group
of turbines in a hydroelectric scheme may be supplied by an asymmetric
pipe network, and each of these turbines could generate a forcing oscilla
tion. An in-line pump can cause a discharge side forcing oscillation, which
the downstream portion of the network will experience, and an inlet side
forcing oscillation, which the upstream side of the network will experience.
If the network is looped in a manner which joins the upstream and down-
stream portions of the network, the network will be experiencing two forc-
ing oscillations. Such examples occur in many circumstances.
In networks such as these there must be as many impedance analyses
performed as there are sources of forcing vibrations. Because the analysis
is of a linearised type, superposition is permissible and the risk to the
network can then be assessed from the graph resulting from the superposi-
tion of all the harmonic analyses performed.
Students of the phenomenon of resonance are strongly recommended
to study the relevant section of Streeter's book: Hydraulic Transients.
10 Unsteady flow in open channels
10.1 Introduction
Flow in open channels is normally considered in steady state conditions

only. The methods of analysis of unsteady flow conditions in pipe net-
works can be applied to open channel networks with only trivial modifi-
cation however.
Nonrectangular channel sections can be dealt with, the channel need
not be of prismoidal type, variable friction can be incorporated into the
analysis and the results obtained from such analyses are usually of high
accuracy. The main limitation upon the unidimensional method described
below is that no sudden significant changes of depth must occur within
the channel cross section.
A
A section suitable lor analysis A section unsuitable lor analysis
(a) (b)
Figure 10.1
The flow in the main channel to the left of line A -A in figure 10.1 b
will be at higher speed than that in the remainder of the cross section.
Consequently there will be a shear plane at A -A across which heavy
turbulence will be generated. The wavespeed in the left-hand part of the
section will be much higher than in the right-hand part and so a unidimen-
sional analysis cannot be accurate for such a cross section. Bi-characteristic
methods exist which could be used but they are considerably more com-
plex and involve much more computer run time so the analyst may well
decide to simplify the section by considering only the main channel and
ignoring the flood plain.
177
Unsteady flow occurs in almost every channel. A river is never in steady
state - it is either in flood or it is on a part of its recession curve. Flows in
sewers are constantly changing. Flows in the head and tail races of hydro-
power schemes are constantly changing as the load on the power station
changes. In some cases the consequential depth changes in the channel
are unimportant but in others unexpectedly large changes occur and,
frequently, travelling surges or bores are created which may overtop the
channel's sides. To ignore these effects can have most serious consequences.
A strong case can be made for investigating the unsteady flow conditions
of open channel networks and it is to be hoped that in the future such
investigations will be undertaken in the same way that unsteady flow
investigations in pipe networks are carried out routinely today.
10.2 The equations of unsteady flow in open channels
Consider the flow into and out of the element depicted in figure 10.2.
water surface
,,,.n~
Figure 10.2
Net volume entering the element in time ot =

[Av-(A+~~ox) (v+~~ox)]ot
Storage within the element =
ad
BX ox xatot
where B is the mean surface breadth of the element
l_ aAvox- A~ox]
ax
0t = B adoxot
at
[ ax
so v aA + A av + Bad= 0
ax ax at
This ignores fluid compression within the element and this is reasonable
in a free surface flow.
Now
aA aAad
ax ad ax
Unsteady flow in open channels 179
aA=B
and aa
vB ad+ A av + B ad= O (10.1)
ax ax at
Next consider the forces acting upon the element and the momentum
change that these cause.
Force in left to right direction=
wzA - w(:z +~! 8x) G+~~ 8x) +wA8x sin(i)- Pr8x
Tis the frictional shear stress acting over the perimeter of the flow, z is the
depth of the centroid of the upstream cross section of the element and P
is the mean wetted perimeter of the element.
Force= w (-A az- z aA + Ai- Pr\8x

w)
ax ax
where second order small quantities have been ignored. Also sine(i) has
been approximated to i so the usual limitation of small bed slope applies
to this analysis.
Now taking moments about the surface at the downstream end of the
element (see figure 10.3):
(A+~~8x) (z+~!8x) =A(:z+~:8x) +~(~:8xf

_ aA +A -;-8x aa +-
a:z =Az +A -;-8x B(aa
-8x)
2
Az_ +z~8x uX 2ax
uX uX
surface at x + 8x
aA 8x -r--.--------.----.;,-1
ax
Figure 10.3
ignoring second order small quantities.

. PT) 8x
ad Az--
Force= w (-A-+
ax w
The rate of change of momentum that this force causes =
dv (, av av)
pAfJx dt =pAfJx ~ ax + at
w(-A ad+ Ai- Pr\fJx =pAfJx lv av + av)

' ax w} ~ ox at
ad v av 1 av . PT
-+--+---z + -=0
ax gax gat wA
Pr _ ....!__ vlvl _.
Now wA - wm - C2 m- I
where Cis the Chezy C and j is the energy loss per unit wt of fluid per unit
length of channel.
ad v ov 1 ov . .
K' 'i
ox + ox + ot +1 -' = 0 (10.2)
Equations 10.1 and 10.2 are called the St Venant equations
i.e. v od +4 av + od= 0
ox B ax ot [10.1]
od VoV 1ov . .
K' 'i
ox + ox+ ot +1 -' = 0 [10.2]
In the development of the continuity equation an assumption is implied,

i.e. that the channel is prismoidal. This means that 6A = B6d. If the chan-
nel is nonprismoidal this is not true as can be seen from figure 10.4.
Nonprismoidal trian31e-like channel

8A~88d+ ll~x
8+88
Nonprismoidal rectangle-like channel
8A =88d+d88
Figure 10.4
so ~A = B~d prismoidal channel
~A :!:: BM +!!._ aaB ~x nonprismoidal triangle-like cross section

2 X
~A =B~d + d ~: ~z nonprismoidal rectangle-like cross section
If the channel is nonprismoidal an additional term will appear in the

continuity equation, i.e.
ad av ad
vB-+A- +B-=0 prismoidal
ax ax at
and
ad aB av ad . .
vB ax+ Otd ax v +A ax+ Bat= 0 nonpnsm01dal
where Ot lies between 0.5 and 1.0 depending upon the nature of the non-
prismoidality of the channel reach.
aB IS
. a constant 1"or . aB IS
Now ax a gtven reach so Ot ax . a constant. Le
t
daB
OtB ax= A
(Note that Ais a function of depth as B changes with depth.)
Then
ad A av ad
v-+Av+ --+-=0 {lO.la)
ax B ax at
10.3 The characteristic forms of the open channel equations
The equations can be cast into their characteristic forms in exactly the same
way used for the waterhammer equations, but a different, simpler technique
will be demonstrated here.
First re-express the equations replacing the depth d by the small wave
celerity c using the relating equation.
c2
d =- (see equation 10.24a)
g
Then
ad 2c ac
ax =-g ax
ad 2c ac
at =g-at
The characteristic forms of the unsteady flow equations for a nonpris-
moidal channel are more complicated than those of the prismoidal channel
and can be obtained using the Lister method presented in chapter 4. The
prismoidal forms will be developed here. Assuming that A/B = d which is
exactly true for a rectangular prismoidal channel, and closely approximately
true for a broad channel, then equation 10.1 becomes:
c
2cv ac + av + 2c ac = 0
g ax g ax g at
(10.1b)
and the dynamic equation becomes:
ac av av ( J-z
2c -+v-+-+g . ") = 0 (1 0.3)
ax ax at
Dividing equation 10.1 b by cjg and adding equation 10.3
ac ac av av . .
2(v +c) ax+ 2 at+ (v +c) ax+ at+ g(J -z) = 0
Then
2((v+c)a~ +;)c+ ((v+c)a~ +a~)v

+ g(j-i) = 0
[(v+c)a~ +:tJ(v+2c)+E=O (10.4)
where E = g(j - i).

Next, divide equation 10.1 b by cjg and then subtract equation 10.3
ac ac avav ..
2(v-c)ax +2at-(v-c)ax -at-g(J-z)=O
Multiply through by -1
r(v -c).._+ l_J(v-2c) + E= 0

L ax at
So [(v c) a~ + :t}v 2c) + E =0

Now, comparing this equation with the definition of a total differential
i.e. siP= alP ox+ a~Pst

ax at
Then
diP_ alP
- ---+-
ax alP
dt ax dt at
i.e. d'{J_(dx
dt- dt ax
a+a)
at '{J
If E is slowly varying and can be treated as a constant for the 8 t period
and if
dt 1
-=--
dx vc
then I(J=v 2c +Et
dt 1 . 1
Thus m an x-t space dx = -+- defines two hnes, one of slope - - and
0
v-c v+c
the other of slope - 1-; along these lines d'{J = 0 that is, '{J is a constant. In
v-c dt
fact, ddxt = - 1 - defines a positive characteristic line along which v + 2c + Et
v+c
0
ts dx = -1- defimes a negative
constant an d dt 0 ch aractensttc
0 0 10me aIong whotc h
v-c
v- 2c + Et is constant.
Summarising the characteristic equations of free surface flow:
dt 1
-=-- (lO.Sa and b)
dx v c
d
dt (v 2c + Et) = 0 (1 0.6a and b)
As will be seen there is a very great similarity between these equations

and those ofwaterhammer. There is also a great similarity in the way in
which they are manipulated.
The concepts of domain of dependency, zone of influence and zone of
quiet described in chapter 4 apply equally to unsteady flow in open chan-
nels.
The method of solving the value of c and v at any point on an x - t grid
is exactly similar to that used in chapter 4 for the calculations of h and v
values. Once c has been obtained, the depth d is immediately available, as
d = c 2/g (see figure 10.5).
Figure l0o5
As in chapter 4 Vx and Cx and Vy and Cy must be calculated by linear
interpolation of values at M and Nor Nand R, using the relation
and
Once XN and NY have been calculated, values of c and v at X and Y are

simple to evaluate.
Then Vx + 2cx +Ext= Vp + 2cp + Ep(t + Llt) (10.7a)
and Vy- 2cy + Eyt =Vp- 2cp + Ep(t + Llt) (10.7b)
Now Ep is unknown as it depends upon Vp and dp. It is given by
Ep = J.vP~vpl - i\
"\Cp mp J
However, E varies slowly and will change very little in a Llt time interval. It
is usual to approximate Ep to En (which is known of course). Some analysts
go even further and approximate Ex, Ey and Ep to EN and this does not
produce large errors if Llx and hence Llt are both small.
Solving equations 10.7a and 10.7b simultaneously gives values ofvp
and cp without difficulty and d is then given by dp = c~/g.
The choice of Llt is more difficult to make than in the case of water-
hammer analysis. As c = .Jid, and d can vary considerably during the
passage of a flood surge or other large wave, c can also vary significantly.
v can be a large fraction of c in subcritical flow and is greater than c in
supercritical flow. Ll Ll
Thus for a given value of Llx the Llt value is either~ or~ .
v+c v-c
If the negative characteristic were chosen the value of Llt could be very
large but would be unsuitable for use with the positive characteristic. Using
the positive characteristic the Llt value becomes Llx . However, if a
v+c
regular grid is to be used - as in the case of waterhammer - the Llt value
must be fixed. Both v and c vary throughout the duration of the analysis
so a Llt value must be chosen which will always ensure that points X and
Y lie within the segment MR. A suggested value for Llt is Llx/(2c) where
c in this instance equals the wave celerity calculated from the depth in the
channel at the commencement of the analysis. This depth must be known.
Either it may be assumed that the flow in the channel is in an initial steady
state or that the flow is zero, the velocity everywhere being zero and the
water surface being horizontal. The first of these alternatives requires the
calculation of the surface proflle in gradually varied steady state using the
usual equation, i.e.
where FN is the Froude number and equals
Q2B v2
A 3g or go
where 8 = ~ and B is the surface breadth.
Calculations of this type, as usually performed, are not really sufficiently
accurate to be used as an initial steady state but if the unsteady analytic
method is applied to this slightly erroneous steady state until it has become
truly steady, the required unsteady condition can be applied and a satisfac-
tory solution obtained. The second initial condition of zero flow avoids
the need to perform a gradually varied flow analysis. The steady condition
actually applicable to the circumstance must be applied as if it were an
unsteady state and the analysis then performed for a sufficient time for
the true steady state to develop. After this has been established, the re-
quired unsteady condition can be initiated and the unsteady analysis then
performed. The second initial condition of zero flow requires a longer run
time but saves the effort of calculating the initial gradually varied steady
state. If v and/or c change so as to cause the domain of dependency criterion
to be violated, the analysis becomes invalid. It is therefore necessary to pro-
vide an Algolrithm which outputs a warning message and aborts the run.
The program should then be re-run using a smaller !JJ value.
Unsteady supercritica/ flow

In supercritical flow the value of c is less than v. This implies that the slope
of the negative characteristic becomes positive as illustrated in figure 10.6.
In supercritical flow, information from downstream cannot affect con-
ditions upstream as the local velocity vis greater than the wavespeed c,
M~l---x------~~~x--------~
Figure 10.6
and information travelling at wavespeed cannot be carried upstream through

a fluid travelling downstream at a greater speed because waves travelling
upstream relative to the liquid are washed downstream by the current. Con-
sequently the downstream Ax length NR contains no information that
effects conditions at P and the domain of dependency XY lies entirely
within the upstream Ax length MN. Only one Ax length need be con-
sidered. Of course if supercritical flow occurs in the negative direction the
relevant Ax length becomes NR.
In the preceding paragraph it was stated that in supercritical flow a
wave cannot travel in the upstream direction. This is true for small waves.
If a large shock wave develops, its celerity may be larger than the local
velocity and under this condition it can travel upstream. The method of
dealing with this situation is described later in this chapter.
As flow becomes progressively more and more supercritical both the
positive and the negative characteristics slope more and more downstream,
tending to the horizontal and becoming progressively more parallel. This
implies that the basic partial differential equations are tending to become
more and more parabolic in type although they always, in fact, remain
hyperbolic.
However, the !::.t values that must be used become smaller and smaller
as the flow becomes more supercritical so the run time costs increase for
any given simulated time.
In conditions such as these it becomes possible for the chosen At value to
be too large. It is wise therefore to provide a facility for aborting a run and
outputting a warning message to the effect that the value of XN has be-
come greater than Ax (see figures 10.7a, 10.7b and 10.7c).
't ~~
I /::,.x X
Characteristics in a highly super critical flow
(a)
Crilical flows: wilh positive v wi lh negative v
(b) (c)
Figure 10.7
Boundary conditions
As with waterhammer, unsteadiness of flow is generated from a boundary
condition. There are a number of boundary conditions to be considered,
some of which are listed here.
(1) A downstream estuary in which the level fluctuates with the tides.
(2) An upstream catchment in which flows vary with rainfall producing
flood or storm surges.
(3) Upstream or downstream weirs.
(4) Upstream or downstream sluice gates.
(5) Downstrea m reservoirs in which the level fluctuates slowly.
(6) As (5) but with discharge over a spillway.
(7) An upstream spillway.
(8) A junction.
Tidal estuary
The level history in a tidal estuary can be measured and is often available
from records. Such a history can be read in as data and values of depth at
the end of the channel at At intervals can be interpolate d from the read-
in array. The tidal level variation occurs at the downstream end of the
channel so a forward characteris tic can be combined with tidal level dT at
any At time level to obtain a solution for the local velocity (see figure
10.8).
p
~M~-X~------~N
Figure 10.8
Cp = YidT
Vx + 2cx +Ext= Vp + 2cp + Ep(t +At)
Vx and Cx can be obtained by interpolatio n. Ex can be calculated and
Ep can be approxima ted to EN.
Then
Vp = Vx + 2cx +Ext- 2cp- EN(t +At)
Once up has been found the boundary condition is solved.
Flood or storm surge

The parameter that must be known in this circumstan ce is the flow at the
upstream end of the channel. It will vary with time and so a flow array
must be fJ.iled by read-in values, from which upstream flows at At time
intervals can be interpolate d.
Then if the channel is broad or rectangular in cross section
Qt+~t = bdt+~tVt+M
so Qt+~t
vt+~t = v = - - (as dp = dt+~t)
p bdp
as dp = c~fg
v =gQt+~t (10.8)
P be~
Combining this result with a backward characteristic equation (see
figure l 0.9)
Vy- 2cy + Eyt = Vp- 2cp + Ep(t + ilt)
Substituting for Vp from equation 10.8 and equating Ep to EN
Qt+~t--2cp+EN ( t+ ilt)
vy-2cy+Eyt=g--2
bcp
This gives a cubic equation in Cp which can be readily solved by, for
example, the Newton-Raphson method.
Once Cp has been obtained vp can be calculated by back-substitution of
cp into equation l 0.8.
This solves the boundary condition.
y R
Figure 10.9
Upstream or downstream weir

Consider the case of a downstream weir first.
The weir formula is:
(10.9)
2
where Eu = du + ~;which is the specific energy just upstream of the weir,
du is the depth, and Vu the velocity just upstream of the weir and hw is the
height of the weir crest above channel bed level. Use a positive characteris-
tic (a negative characteristic for an upstream weir)
Vx + 2cx +Ext= Vp+ 2cp + EN(t + ilt)
Now Q = bdpvp for a rectangular channel so
bdpvp =1.71b (dp- hw + ~J"s (10.10)
c2
As and d u =dp =~
g
it follows that
b
~
2
= 1.71 ~-h +2.
( 2 2)1.5
g g w 2g
2 2)1.5
1.71 (~ -hw +2_
Vx + 2cx +Ext= ~ 2 2g + 2cp +EN (t + At)
~
g
(I 0.11)
In this equation the term v~ appears. Upstream of a weir v is small and
v~ is therefore also small. Initially v~ can be ignored. This feaves a cubic
equation in Cp to be solved. Once this has been done v0 can be calculated.
By substituting this Vp value back into equation 10.11 a second estimate of
Cp can be obtained and the process repeated until two successive values of
Cp are insignificantly different.
In the case of an upstream weir a negative characteristic must be used
in place of the positive characteristic. The Q value will have been obtained
during the analysis of the weir as a downstream weir at the end of the
reach upstream of the reach under consideration. Q will thus be known.
The solution is thus identical to that for case 2, the flood or storm surge.
Sluice gates
Consider a sluice gate located at the downstream end of a reach. The sluice
gate equation is:
where cd is the coefficient of discharge, dco is the depth at the vena

contracta immediately downstream of the sluice and equals Cc x d 5 where
d 5 is the height of the opening below the sluice, Cc is the coefficient of
contraction of the sluice, and Eu is the specific energy of the flow just up
v2
stream of the sluice and equals du + 2;. Now du and Vu equal dp and Vp
respectively, so
(10.12)
If the sluice is undrowned dd = dco but if it is drowned dd is the depth

of the flow just downstream of the sluice.
Using a positive characteristic
Vx + 2cx +Ext= Vp+ 2cp+ EN(t +At)
and
bc 2 v I 2 v2 \
bdpvp=T=Cdbdcoy2g f+~-ddJ
(c
V =
Cdbdco~g(c~+
g
v~-dd\
g 1)
so
P bc_
_ P
2
(10.13)
As for the weir, the v~ term can be initially ignored and the resulting cubic
equation in cp solved. Vp can then be solved and the v~/2g term evaluated
and inserted into the equation which can then be solved again for cp, the
process then being repeated until two successive values are insignificantly
different.
The solution of an upstream sluice is handled similarly to that of an
upstream weir, the Q value having been obtained during the solution of the
sluice when it was considered as a downstream sluice in the upstream reach,
so the depth downstream of a sluice can be found by the method demon-
strated for the upstream weir. The method illustrated for the sluice can be
simply adapted to solve a Venturi-flume also.
A reservoir in which the level varies slowly

If the level fluctuation is very slow it will usually be possible to treat it
as a constant level point for the duration of the unsteady flow analysis,
i.e. dp = dreservoir so Cp = ..Jid;
will be a constant.
The solution is then trivial. Using a positive characteristic for a down-
stream reservoir (a negative characteristic for an upstream reservoir),
then
Vx + 2cx +Ext= Vp + 2cp+ EN(t + ~t)
and so Vp = Vx + 2cx +Ext- 2cp- EN (t + ~t) (10.14)
A downstream reservoir equipped with a spillway

The water surface in the reservoir is assumed to be flat and horizontal.
Assume that the plan area of the reservoir at a time t is At and the dis-
charge equation of the spillway is Q5 = kh~'.
Then let the flow from the channel entering the reservoir= Qc,
so dh
Qc=At dt + Qs
i.e. bdv =A{dp-dN)+kh 1"5

p p t\ ~~ sp
where hsp = height of the reservoir surface above the spillway crest.
Thus
bcJ,vP=A (c~ic~) +kht.s

g t g t sp
using a positive characteristic

Vx + 2cx +Ext= Vp + 2cp +EN (t + ~t)
A
t
(c~gM
- c ~) + kh t.s
sp
Now Vp bCp
g
This gives a cubic in Cp and once solved Vp can be obtained by back sub-
stitution. hsp must be increased by an amount of dp- dN before the next
time level is analysed.
The method of routing the flood through the reservoir assumes that no
wave action occurs on the reservoir and that the level increments occur at
the same time at all points on the reservoir surface. The method is com-
monly used for normal flood routing but this does not usually involve
the unsteady analysis of the flow in the channel as described here.
An upstream spillway
The upstream spillway case anises when a reservoir discharges into its
downstream river or channel. The Q of the spillway discharge will have
been calculated in the downstream reservoir example, so will be available,
and the method demonstrated in the flood or storm surge and upstream
weir can then be applied.
A junction
At a junction the depth is the same for all channels joining there. The in-
flow to the junction must equal the outflow from it. The situation is thus
the same as that described for waterhammer at pipe junctions. (For flows
at other than small Froude numbers it will be necessary to include local
losses and kinetic energy terms, however.
If the depth is constant for all channels joining at the junction the value
of the wave celerity Cp must also be the same for these channels.
Using the same methods as those used for the waterhammer case, con-
tinuity gives:
(10.16)
Where subscript a denotes the ath channel.

As in the section on junctions in chapter 7, define the flow direction in
any channel by the use of an integers which takes the value of+ 1 if the
channel is transporting water towards the junction and -1 if it is trans-
porting water away from it. Let the channel number 'a' be signed, 'a' being
positive if the channel is transporting water to the junction and negative
if away; then
s = sign(a)
Then for each channel joining at the junction, the characteristic equation
will be
(10.17)
where Va and Ca are the interpolated velocity and wavespeed respectively

on the t time level and vPa and cPa are the junction velocity and wavespeed
on the t + !::.t level respectively, appropriate to the channel numbered a.
(Note In equation 10.17 Ea is the value of Eat the junction end of
the !::.x length.)
Then
Vpa = Va- 2s(cpa- Ca)- Ea!::.t (10.18)
substituting this value into equation 10.16
c
Lsba ;\va- 2s(cPa- Ca)- Ea!::.t) =0
As cPa is the same for all channels, the c~afg term can be cancelled through.
Then
'E.sbaVz- 'E.2baCpa + 'E.2baCa- 'E.sbaEa!::.t =0
'E.sbaVa + 2 'E.baCa- ill 'E.sbaEa
so c = (10.19)
Pa Z 'E.ba
The right-hand side of equation 10.19 contains known values only so

cPa can be evaluated. dp at the junction then equals c ~a/g.
vPa can then be calculated for each channel using equation 10.18.
Boundary conditions other than those already described can be modelled
by similar methods.
10.4 The travelling surge
The unsteady analysis described up to this point is based on equations

which are accurate only when the flow is gradually varied and the stream-
lines are everywhere curved to only a small degree. If the small waves in
the characteristics approach can run over one another, the flow through
the resulting large wave will not be even approximately gradually varied
but will be rapidly varied. For such cases a separate approach must be used.
Consider a travelling surge, a large steep-fronted wave often with a
roller front (see figure IO.IO).
~v1
CD
~Vw
Figure 10.10
f
@
By applying a velocity Vw to the system as illustrated in figure IO.IO the

free surface boundary can be brought to rest (figure I O.II ).
Applying the continuity equation at sections I and 2
(v1 - Vw)At = (v2- Vw)A2
'1 t+-~ ~0 E--~

CD
Figure 10.11
Applying the force equation to sections I and 2
wA 1z 1 - wA 2z 2 = w A 2 (v 2 - Vw) 2 -~A 1 (v 1 - Vw) 2

g g
A.z. -A2z2 =: (cv.- vw)~~f

2 -: Cv1- vw) 2
=(A~_A 1) (vl-Vw) 2
A2 g
A2 (A1z1 -A2z2)
g- ~...,..:.----:--=---=:..:... (10.20)
A1 (A1- A2)
where z is the depth of the centroid of the flow cross section below the
free surface. A is the area of the cross section. This equation can be solved,
although sometimes with difficulty, for any regular cross section.
If v1 is zero Vw becomes the velocity of the wave through still water,
i.e. Cw, so
The plus sign applies if d1 > d 2 and the minus sign if d1 < d 2
Of course, this celerity is larger than it would be for a small wave travel-
ling over the fluid at the smaller depth, as will be seen from the following
sections.
A1 -A 2 ~-Bh
where h is the height of the wave
-
A tZt- - ~A tZt-
A 2Z2- - Bh 2 ~ A I h - Bh
A I (-Zt + h) -2-- 2
-2-
Also
(10.21)
Denote Atfb by 8 the mean depth based upon the surface breadth b.
(10.22)
If h is small compared with 8
Vw~Vt Jg((i +%h) (10.23)
and if it is very small

(10.24)
i.e. for a small wave c =.Ji8 and for a rectangular channel 8 =d so
c =.Jid (10.24a)
Of course, all of the above results are approximate compared with that
quoted in equation 10.20.
Commonly channels are of rectangular section or are effectively broad
and then equation 10.20 reduces to:
(10.25)
So
The result of this analysis of the travelling surge is exactly the same as
the dt = - 1 -result from the characteristics method and this is to be ex-

dx vc
pected as small and large waves must obey this relationship. In this result,
however, the value of cw is different from c in the characteristics equations.
10.5 The profJle of a free surface flow when a travelling surge is present
The phenomenon of the travelling surge is not as simple to analyse as the

simple equation derived above would appear to indicate. After a surge has
passed a point in the channel the depth at that point does not stay con-
stant. The proftle downstream of the surge changes with time due to storage
and frictional effect, as illustrated in figure I 0.12. (This corresponds to line
pack in pipelines as described earlier.)
Figure 10.12
Consider the sequence of events illustrated in figure I 0.12. If a travel-

ling surge moving in an upstream direction is generated by a closing move-
ment of the downstream sluice gate the wave will travel upstream as indic-
ated, i.e. the wave will move through positions I, 2, 3 and 4 sequentially.
As it does so the surface downstream of it will rise and progressively
flatten and the wave itself will diminish in height. The amount by which
the surface will rise may be much larger than the wave height so if this
effect is ignored the result could be very erroneous.
The analysis of the phenomenon is best performed by the characteristics
method but a separate account must be kept of the surge wave as the
characteristics method is not valid across a shock of significant size.
10.6 The method of analysis of an unsteady free surface flow in which

travelling surges are present
Various types of wave motion are possible as illustrated in figures I 0.13a,

b, c and d.
The negative wave illustrated in figure 10.13d cannot propagate with
unchanged shape. Such a wave will rapidly flatten because the deeper
portions of the wave will travel more rapidly than the shallow portions as
their celerities are greater.
The negative wave flattens as shown in figures I 0.14a, I 0.14b and
10.14c.
---- ----
~ ~
(a) (b)
---- ----
~
(c)
Figure 10.13
~
(d)
The type of wave illustrated in figure 10.13a is sometimes called a re-

jection surge as it is generated by a reduction of the downstream flow. That
illustrated in figure 10.13b is called a demand surge and is generated by
an increase in downstream flow.
The wave illustrated in figure 10.13c is called a flood surge and is caused
by an increase in upstream flow.
A wave that looks very similar to that in figure 10.13a is called an ebb
surge and is caused by a reduction in upstream flow.
It does not follow that increases or decreases in flows upstream or
downstream necessarily cause a travelling surge. For a travelling surge to
be generated it is necessary that the Froude number taken relative to the
surge, i.e. (v1 - Vw) 2 /{gd1 ), must be either greater than or less than unity,
whilst that downstream (v2 - Vw) 2 /{gd2 ) must be less than or greater than
unity ( Vw is taken positive in the direction in which x is increasing) respec-
tively. In other words the relative Froude number change across the surge
must be such that the two relative Froude numbers must span the value
of unity. If both relative Froude numbers are greater than unity or if both
are less than unity no surge will be generated.
Denoting the relative Froude numbers by FR 1 and FR 2 (the absolute
Froude numbers are denoted by FN 1 and FN;) then a suitable test for the
presence of a surge would be that (FR 1 - l)/~FR 2 -1) < 0.
(a) (b)
(c)
Figure 10.14
Now if a surge wave is present the absolute Froude number on one side
of the surge wave must be less than unity and that on the other may be
either less than or greater than unity.
If both absolute Froude numbers are less than unity, i.e. flow is sub
critical on both sides of the surge, both forward and backward characteris-
tics may be available for upstream and downstream points on either side
of the surge (see figure 10.15).
Figure 10.15
The velocity of the surge wave is given by
Vw=v1 gd2(d1+d2\ (10.26)

2d1 J
The plus sign is used when d1 > d 2 and the negative sign is used when
d1 <d'J..
The magnitude of this velocity Vw must lie between that of the small
wavespeed v 1 Yid,., on one side of the surge, and that on the other side,
v2 ..:;'id;. For example, v 1 - .Jid,. < Vw < v2 - ..Ji(i;. for the case illustra-
ted in figure l O.l3b and v 1 + Yid,. > Vw > v2 + ..../i(i; for the case illustrated
in figure 10.13c.
Therefore the surge always travels faster than a small wave on the shal-
low flow and slower than a small wave on the deep flow. In figure 10.13a
a small wave can run upstream on the deep flow, and overtake the surge
which is travelling upstream, becoming incorporated into it. In figure
1O.l3b a small wave can travel downstream on the shallow flow and will
catch up with the surge, whilst a wave travelling downstream on the deep
flow will travel faster than the surge and so can not be overtaken by it. In
figure 10.13c a wave on the deep flow will overtake the surge and become
incorporated into it and the surge will overtake small waves on the shallow
flow incorporating them into itself as it does so.
It should now be clear that the flow upstream of the surge may be super
or subcritical and that downstream the flow must be subcritical. Consider
the conditions at the point at the upstream end of the ..:lx length and the
conditions at the downstream end of the ..:lx length in which a surge wave
is present. In figure l 0.16b characteristic PS crosses the wavefront and this
invalidates the basic assumptions upon which the characteristic method
is premised.
Even so a simple technique can be evolved.
A
slope o __!_
v.
R
M N T v u w X
Upstream flow supercritical

Downstream flow subcril ical
(a)
x, x2
Upstream flow subcritical
Downstream flow subcritical
(b)
Figure10.16
The wavespeed is assumed to be known at X (from a previous step of

integration); see figure 10.17. Using this waves peed, point Y is readily loca-
ted. From Y, three characteristics can be constructed YA, YB and YC. A
must always lie upstream of Beven when the upstream flow is subcritical (as
illustrated). When it is supercritical the point A is even further upstream.
Depths (and hence celerities) and velocities at A, Band C can be obtained
by interpolating appropriately between M and Nor Nand 0.
Thus three characteristic equations can be written
AJongAY: VA+ 2c A+ EAt= VVl + 2cY1 + Ey1 (t + ~t) (10.27)
Along BY: VB+ 2cn + Ent = vy2 + 2cY2 + EY2 (t + M) (10.28)
AlongCY: vc- 2cc + Ect = vy2- 2cy2 + EY2(t + ~t) (10.29)
EYl can be approximated to EN and Ey2 to Eo.
The continuity equation can also be written for conditions across the
wave:
i.e. (10.30)
Solving equations 10.28 and 10.29 simultaneously gives vy2 and cy2.
wave path
/ _, L______ slope ..L

v,.
0 2
X
M
slope -f.-w,
Figure 10.17
The wave equation I 0.26 is
V. _
w - Vt -
+j!<~ 2
(dt + dz)
2dt '
As before, use the positive sign if d 1 > d 2 and the negative if d 1 < d 2
Cy22 Cy}2
Substituting-- for d 2 and-- for d 1
g g
gives
Vw y =Vy} (I 0.31)
substituting already calculated values for cy2 and vy2, substituting the ex-
pression for Vwy into equation I 0.30 and then solving the resulting equation
simultaneously with equation 10.27 gives solutions for vy1 and CYl Vwy
can then be calculated by substitution of the values ofvy1, cy1, and cy2
into equation I 0.31.
This value of Vwy2 gives the new slope of the wave path, i.e., I I Vwy
and using this the next time step can be solved similarly.
For a more extensive treatment of the topic of waves in open channels,
Water Waves by Stoker 17 is recommended.
If the backward characteristic toP crosses the wave path XY proceed as
follows. Calculate the length AR and linearly interpolate for the values of
VR and CR from VA and c A and vy and cy. Then use the forward character-
istic equation toP and the backward equation from R to solve Vp and cp.
10.7 Other methods of analysis
The method of characteristics has many adherents and it still is considered

by most analysts to be the best method. However, a number of different
methods have been evolved in recent years all of which are based upon
the integration of the St Venant equations using various finite difference
schemes.
The leading methods are listed below:
(l) The leap frog method.
(2) The two step Lax-Wendroff explicit method.
(3) Amein's four point implicit method.
(4) Six point implicit methods- used by Liggitt and Woolhiser, Abbot
and lonescu.
It is claimed that these methods permit increases in l:lt values without sig-
nificant increases in error or instability, giving a worthwhile reduction in
computer run time.
A paper by Price 11 outlines methods 1, 2 and 3 and describes the
technique ofthe fixed characteristic grid used in this chapter. The paper
suggests that Amein's four point method is superior to all others.
These methods can also be applied to the analysis of water hammer
but the author is not convinced that they offer any advantages over the
fiXed mesh characteristic method used for waterhammer analysis in this
book.
Lack of space precludes the presentation of these other methods but
the reader can study them in the papers listed in the Bibliography.
11 Global programming
11.1 Introduction
A global program is one which provides the facility of being able to des-
cribe the topology of any pipe network by the use of numbers, which are
read in as data. Additionally the presence, or absence, of any particular
hydraulic control, such as a pump, turbine, reservoir, air vessel at any loca-
tion must also be capable of specification by the use of numbers.
Such a program can be used to solve any network without amendment
and if the facility is not available a new program will have to be developed
for every network to be analysed. Obviously global programming increases
the value of the technique of analysis many times.
There are many ways of writing a global program but only one such
method is presented here. In the literature the nodal method is usually
described as being the method of choice but the author would prefer to
present his own method which he believes to possess significant advantages
over the nodal technique.
11.2 The route or link method of global programming
First a schematic of the network must be prepared. On this schematic all

hydraulic controls must be entered and all controls of the same type must
be numbered sequentially. For example, if the network contains ten pumps
each pump must be assigned a unique number between 1 and 10. Similarly
if there are 15 valves of the same type (but not necessarily possessing the
same parameters) then each valve must be sequentially numbered from I to
15. Reservoirs, air vessels, surge tanks, etc, must be treated in this manner also.
Next, arrows must be inserted onto the schematic to indicate the direc-
tion in which flow in each pipe is expected. If the arrow is inserted pointing
in the wrong direction the steady state velocity for such a pipe will be
found to be negative so it does not really matter which way the arrow points
as long as it is remembered that a negative steady state velocity only im
201
plies that the flow is in a direction opposite to that assigned by the arrow's
direction.
Next the pipes making up the network must be numbered. Here advan-
tage can be taken of the fact that if numbering is always sequential across
a joint, a pump, an air vessel, a surge tank, etc, less data need be supplied.
To retain complete flexibility, however, then way junction must retain
total freedom of numbering and no pre-determined sequences of numbering
should be employed around a junction (except for a two-way junction,
i.e. a joint).
An array called route should be created. It should be two dimensional
and should contain one row per pipe. Each row must contain n numbers
where n is the total of numbers necessary to specify the conditions at the
upstream end, the downstream end plus any other additional numbers as-
sociated with these boundary conditions.
Considering a hypothetical route array, each row could contain the
following numbers:
In the 1st position: 1 - defines an upstream reservoir.
(Upstream boundary 2 - defines a pump fitted with a free surface
condition) suction well.
3 - defines an in-line booster pump.
4 - defines a joint.
5 - defines an air vessel and also a surge tank.
In the 2nd position 6- defines a downstream reservoir.
(downstream boun- 7 - defines a motorised valve.
day condition) 8 - defines a servocontrolled valve.
9 - defines an n way junction.
10- etc
11 - etc
In the 3rd position: a - the number of the upstream pump. If no pump
is fitted then 0 must be supplied.
In the 4th position: b 1 - the number of the upstream reservoir. If no
reservoir exists on the upstream end of the pipe
then 0 must be supplied.
In the 5th position: b2- the number of the downstream reservoir. If
no downstream reservoir is connected to the pipe
than 0 must be supplied.
In the 6th position: c - the number of pipes joining at a downstream
c way junction, i.e. 0 must be supplied if no junc-
tion is present.
In the 7th position: + pipe number transporting fluid to the junction
7th+ 1 position: + pipe number transporting fluid to the junction
7th+ 2 position: + pipe number transporting fluid to the junction
7th+ 3 position: -pipe number transporting fluid from the junction
7th+ 4 position: -pipe number transporting fluid from the junction
7th + c position: etc

Global programming 203
If c is 0, the 8th position should be assigned the value 0 also.
In the 8th+ c position: d- the number of the motorised valve. It should
be 0 if no valve is present.
In the 9th+ c position: e - the number of the servocontrolled valve. It
should be 0 if no valve is present.
Any other boundary hydraulic control number can then be inserted in the
lOth+ c, 11th+ c etc, position.
It will now be realised that array 'route' must be dimensioned from 1
to 11 + c (if no other hydraulic controls other than those listed above are
present) and 1 to netno.
In Algol, dynamic arrays are permitted so n (where n = 11 + the largest
value of c in the network) and netno (netno = the number of pipes in the
network) can be read in and used to dimension the route array.
11.3 Pipe description
Next an array 'net' must be created and supplied with values. This array
must be dimensioned from 1 to 5 and 1 to netno (i.e. it is a two dimen-
sional array).
Each row of the array must carry five variables applicable to an indivi-
dual pipe in the network. These are:
(I) Pipe length
(2) Pipe diameter
(3) Pipe wall thickness
(4) Pipe roughness
(5) Elastic modulus of the pipe wall material
11.4 Longitudinal profiles
Next an array 'neti' must be created. This array must be of type integer.
In this array two numbers must be stored in each row and there must be
one row per pipe in the network. As it is not possible to model the pipe
length exactly it is necessary to calculate the number of Ax lengths that
most nearly represents the pipe length and store this number in the 1st
place in the row. In the 2nd place the number of points at which the eleva-
tion of the pipe centre line is to be supplied must be read in. (This infor-
mation is necessary as by deducting the elevation of any point from the
potential head at that point, the pressure head is obtained.)
Two arrays must next be created, one called Xz and the other called Z.
These arrays are two dimensional and are dimensioned from I to the lar-
gest value of neti [no, 2] (no is the pipe number) of any pipe in the net-
work. In the first of these two arrays the distances from the beginning of
each pipe to the points at which each elevation of the pipe centre line is
to be supplied, is stored. One complete set of elevation point distances
must be supplied for every pipe in the network. In array Z (dimensioned
identically to that of array Xz) the elevations of each point must be stored
similarly.
11.5 Upstream reservoirs
An array Ust (short for upstream) must be declared. In this array the level
of the water surface in each upstream reservoir is stored. This array is
dimensioned from I to the total number of upstream reservoirs present in
the network.
11.6 Downstream reservoirs
An array Dnst (short for downstream) must then be created. In this array
sets of five numbers must be stored: one set per downstream reservoir
in the network.
A set of five numbers consists of the following:
(1) The level of the water surface in the downstream reservoir, at the
beginning of a D.t interval.
(2) The area of the reser10ir free surface.
(3) The width of the weir at the end of the downstream reservoir (see
section 7 .6).
( 4) The weir constant of the weir in the downstream reservoir.
(5) The index in the weir equation.
11.7 Pump description
A further array called puma must be created. This is a two dimensional

array and there must be a row of 36 variables per pump in the network.
The 36 variables contain values describing the pumps' characteristic curves,
i.e. the H- Q curve and the Efficiency- Q curve, and allocated spaces
in which values calculated can be stored. By supplying heads and corres-
ponding flow values the A, B and C values in the H- Q equation can be
calculated. By supplying efficiency values corresponding to the flow rates
the constants in the efficiency - flow equation can also be calculated.
Once calculated the constants are then stored back into the unassigned
places in the puma array. To perform the computation of the H- Q equa-
tions' constants A, Band C and the Efficiency- Q equations' constants
it is probably best to write a subroutine or procedure to calculate them.
A call of this can be made which will perform the calculations and then
make the necessary assignments to the puma array for each pump present.
11.8 Pipe longitudinal profile at ~x intervals
An array called Zd is required. This is a two dimensional array dimensioned

from 1 to the largest value of neti (no, 1) of any pipe in the network and
from 1 to netno.
In this array, values of pipe centreline elevations at ~x intervals are
stored. These values are obtained by linear interpolation between appro-
priate values taken from the Xz and Z arrays.
11.9 Calls of procedures
Having read in all the necessary data and made their assignment to the
appropriate positions in the various arrays it becomes necessary to call
the relevant procedures or subroutines at the correct point in the control
sequence of the program run. This is done in the following way.
At any time level the entire network is scanned in pipe numbering se-
quence. At every ~x distance point the local condition is examined to see
what procedure/subroutine should be employed to calculate the potential
head and velocity at the point at a time a ~t interval later. To decide which
procedure/subroutine is to be used a simple algorithm is used, e.g. if the
position under examination is located at the upstream end of a pipe and
route(no, 1) = l then call the procedure/subroutine for the upstream reser-
voir. If route(no, 1) = 2 then call the subroutine for the suction well pump.
If route(no, 1) = 3 call the subroutine for the in-line booster pump. If the
point under examination is located at the downstream end of a pipe then:
if route(no, 2) = 6 call the subroutine for the downstream reservoir, if it
equals 7 call the motorised valve subroutine, if it equals 8 call the servo-
controlled valve subroutine and if it equals 9 call then way junction sub-
routine.
For midstream points the midstream subroutine should be used. The
subroutine, 'upstream reservoir' needs a parameter which specifies the
reservoir number- this it obtains from route(no, 4). It uses this to obtain
the relevant surface level from the array Ust.
Similarly the downstream reservoir subroutine requires the reservoir
number and it obtains this from route(no, 5). Using this number it obtains
the five relevant data items from array Dnst. Other subroutines obtain
their data similarly; in particular, subroutines 'suction well pump' and 'in-
line booster pump' obtain the parameters defining their numbers from
route(no, 3) and use this to obtain the parameters from array puma which
are needed in the computations that they perform. The pipe numbers
used in 'n way junction' are obtained from route(no, 7), route(no, 8) etc,
up to route(no, 7) + route(no, 6).
The method of turning a specific program into a 'global' program
should now be obvious.
The procedures/subroutines described in previous chapters should all
be included and suitable output sections should be added. The writer of
such a program is warned that iflarge amounts of output are expected
it is best to produce the output directly via a graph plotter. However, some
line-printer output is recommended as it is difficult to pinpoint errors from
graph plotter output alone.
11.1 0 Time level scanning
Having completed one scan of the network the counter i which determines
the simulated time T of the run obtained from T = i x tlt must be increased
by 1 and the scanning process repeated. This should be repeated until the
total simulated time equals or just exceeds the specified run time.
As tlt is usually small and tlx is usually between one tenth and one
twentieth of a pipe line length it may be thought desirable to inhibit output
by only outputting every 5th or lOth tlt level and every 2nd or 4th tlx
point. Integer type parameters can be read in as data to enable choice of
the amount of output to be made. Graph plotter output can also be con-
trolled similarly and this can be important as output storage must be kept
to reasonable proportions.
References
1. Allievi, L. Teoria generale del moto perturbato dell' acqua nei tubi in
pressione, Milan 1903. Translated into English by E. E. Halmos, The
Theory of waterhammer. Anz. Soc. Civil Eng., 1925.
2. Schnyder, 0. Druckstosse in Pumpensteigleitungen. Schweiz Bauztg.,
94, Nos. 22 and 23, 1929.
3. Bergeron, L. Etudes des variations de regime dans les conduites d'eau.
Rev. gen. Hydraulique, Nos. 1 and 2 (1935).
4. Zienkiewicz, 0. C. and Hawkins, P. Transmission of waterhammer pres-
sures through surge tanks. Proc. Inst. Mech. Eng., 68, No. 25 (1954).
5. Angus, R. W. Waterhammer in pipes, including those supplied by cen-
trifugal pumps; graphical treatment. Proc. Inst. Mech. Eng., pp. 136
and 245 (1937).
6. Angus, R. W. Waterhammer pressures in compound and branched pipes.
Trans. Anz. Soc. Civ. Eng., pp. 104 and 340 (1939).
7. Lax, P. and Wendroff, B. Systems of conservation laws, Comm. Pure
Appl. Maths, XII, 217-37 (1960).
8. Lax, P. Weak solutions of nonlinear hyperbolic equations and their
numerical computations, Comnz. Pure Appl. Maths, VII, 159-93
(1954).
9. Lister, M. The numerical solution of hyperbolic partial differential
equations by the method of characteristics, in Mathematical Methods
for Digital Computers (ed. Wilf, A and Ralston, H. S.) Wiley, New
York (1960).
10. Courant, R., Friedrichs, K. and Lewy, H. On the Partial Differential
Equations of Mathematical Physics, New York University Institute of
Mathematics, translated by P. Fox (1956).
11. Price, R. K. Comparison of four numerical methods for flood routing,
J. Hydr. Div., Anz. Soc. Civ. Eng., July (1974).
12. Pearsall, I. The velocity of waterhammer waves in Symposium on
Surges in Pipelines, Inst. Mech. Eng., 180 (1965-66).
13. Karplus, M. B. The velocity of sound in a liquid containing gas bubbles,
Armour Research Earth Foundation Report, June (1958).
14. Fox, J. A. An Introduction to Engineering Fluid Mechanics,
Macmillan, London (1974).
207
15. Suter, P. Representation of pump characteristics for calculation of
waterhammer, Sulzer Review, No. 1966.
16. Edgell, G. Pressure Transients in Tunnels; Extension of Theory to
!"eversible, Nonadiabatic Flow. Limited publication by Leeds Univer-
sity, Department of Civil Engineering, April (1974).
17. Stoker, J. J. Water Waves, Pure and Applied Mathematics, vol. 4,
The Institute of Mathematical Sciences, New York University (1957)
18. Swaffield, J. A. A Study of column separation following valve closure
in a pipeline carrying aviation kerosine, Proc. lnst. Mech Eng., No. 23
(1969).
19. Marsden, N. and Fox, J. A. An alternative approach to the problem of
column separation in an elevated section of pipeline, Proceedings of
2nd International Symposium on Pressure Surges, September 1976.
Bibliography
I. Angus, R. W. Simple graphical solution for pressure rise in pipes and

pump discharge lines, J. Inst. Ca1Ulda, February, 72-81 (1935).
2. Donsky, B. Complete pump characteristics and the effects of specific
speeds on hydraulic transients, J. Basic Eng., December, 685-99
(1961).
3. Fox, J. A. The use of the digital computer in the solution of water-
hammerproblems,Proc. Inst. Civ. Eng., 39,127-31 (1968)
4. Fox, J. A. and Henson, D. A. The prediction of the magnitudes of
pressure transients generated by a train entering a single tunnel, Proc.
/nst. Civ. Eng., 49, 53-69 (1971).
5. Henson, D. A. and Fox, J. A. Transient flows in tunnel complexes of
the type proposed for the channel tunnel, Proc. Inst. Mech. Eng.,
188, No. I 5, I 53-67 (1974). (Two papers).
6. Jaeger, C. Engineering Fluid Mechanics, Blackie, London (1956).
7. Knapp, R. T. Complete characteristics of centrifugal pumps and their
use in prediction of transient behaviour, Trans, Am Soc. Civ. Eng.,
59, 683-9 (I 939).
8. Lax, P. D. and Richtmyer, R. D. Survey of the stability of finite dif-
ference equations, Comm Pure. Appl. Maths, ix, 267-93 (I 956).
9. Parmakian, J. Waterhammer A1Ullysis, Dover, New York (1963).
I 0. Pickford, J. A1Ullysis of Surge, Macmillan, London ( 1969).
11. Rich, G. Hydraulic Transients, Dover, New York (1963)
12. Stepanoff, A. J. Centrifugal and Axial Flow Pumps, Wiley, New York
(1948).
13. Stoker, J. 1. Water Waves, Pure and Applied Mathematics, vol. 4,
The Institute of Mathematical Sciences, New York University (1957).
14. Streeter, V. and Wylie, E., Hydraulic Transients, McGraw-Hill, New
York (1967).
15. Symposium on surges in pipelines, Proc. Inst. Mech. Eng., 180, part
E (1965-66).
16. Wilf, H. S. and Ralston, A. Mathematical Methods for Digital Com-
puters, Wiley, New York (1960).
209
Index
Abbot and Ionescu method 200 Buckling mode I 25

Acceleration Bursts 112
angular 56
convective 81, 8 5
Actuators 136, 140
Capacitance 157, 160
Adiabatic 148 Catchments 186, 187
Air vessels 39, 64, 68, 124, I 25, Cavity 50
126, 127,134, 135, 136, Celerity 78, 181, 184, 186, 191,
201, 202 194, 195, 198
Allievi expression 17 Channels, open I 77
Allievi interlocking equation 21, 25, equations for unsteady flow in
29, 35, 36 178
Allievi pipe characteristic 35 nonprismoidal 180, 182
Amein's four-point method 86, 200 nonrectangular I 77
Analytical techniques I, 23 prismoidal 180, 182
Anchor block 146 rectangular I 77
Arrays 118,119,120,140,187, supercf"itical flow in I 85
202, 203 Characteristic 79, 80, 81
for global programming 203
Characteristic equations 73, 74, 80,
Attenuation 13, 15, 16, 63, 64, 127, 84,129,149,181
128,134,161 in gas flow 148
Characteristic line 7 8, 131, 13 2
Choke, see throttle
Bends 146 Coefficient of discharge 6, 24
Bernoulli equation 6, 25 Coefficient of heat transfer !51,
Bi-Characteristic method 177 !52, 153
Boiling I 7, 94 Colebrook-White equation 95, 96
Bore 178 Column separation 16, 17, 50, 95
Boundary conditions 36, 39, 55, 73, Compressor 65, 125, 154
81, 108, 130, 154, 167, Computerised methods 85, 201
186 Continuity equation 26, 29, 30, 46,
Boundary layer 96 66, 67, 68, 70, 73, 85, 147,
Bubble 17, 72, 87, 88, 89, 90, 91, 178, 179, 193, 198
93, 94, 95, 99,110, 112 Corrector-predictor method 71
evolution of 88 Courant and Lewy stability
Buckle, running 50, 94 criterion 82
211
212 Index
Crack, surface 89 Flood plain 177
Crack, micro- 89, 90 Flow, two phase 50, 87, 92
Current 157 Flywheels 124, 125, 126
Fourier analysis 173
Darcy-Weisbach equation 4, 29, 95, Francis turbine 156
Frequencies, dangerous 172
160
Differential equation, hyperbolic Frequency 173
partial quasi-linear, 29, 73, Friction 4, 13, 23, 28, 29, 36, 49,
66, 67, 68, 96, 99, 108,
74
Dimensional analysis of 121, 128,148
rotodynamic machines 57 variable 96
Frictional damping 156
Dispersion of wave 83
Froude number 185, 191, 197
Distensible pipe-elastic fluid
relative 196
theory 2, 9
Fundamental 156, 173
Domain of dependency 78, 183,
185 Galleries 65
Dump tank 39, 73 Gas 81, 89, 91,92
Dynamic equation 2, 3, 26, 28, 30, bubbles 72, 87, 88
67' 68, 69, 73, 85, 148, calculation of free content 98
179, 193 compressed 112
Dynamometer mode of pump evolution 88
operation 62, 100, 102, release 90
103, 113 release head 93, 95, 96, 98, 99,
112
Eagre lines 36, 38, 39, 40, 42, 45,
Governor and mechanism 63
46, 47, 48,49, 53
Graphical method 36, 50
plotting of 43
Graphical techniques 36, 55, 72
Edgell, G., 150
Grid, regular rectangular 82, 86
Efficiency, pump 56, 57, 60
Elasticity effects 9
Elasticity of fluid 6, 9 Hagen Poiseuille formula 160
Elasticity of pipe wall material 6 Harmonic 155, 17 3
Elliptic partial differential equations Harmonic analysis 172, 173
75 Hartree method 82
Ends, blank 167 Head
receiving 16 5 frictional 2 9, Ill
sending 165, 173 inertia 10, 15
Energy, conservation 13 maximum 56
kinetic 14, 46, 55, 191 no flow 55
strain 6, 13 pump 56
velocity 13 Heat
Equation of state 148 flow 150
Estuaries 186, 187 transfer 152
Euler equation 3 Henry's Law 88
Expansion joints in pipes 18 Hydraulic controls 1, 36, 37, 62,
112,201
Hydraulic radius 28
Fanning equation, see Darcy- Hydroelectric installations 156
Weisbach equation Hyperbolic partial differential
Finite difference methods 7, 9, 68, equations 7 5
85, 107, 199 Hypersonic 7 6
Index 213
Impedance 172, 173, 176 Modulus 172
characteristic 164, 16 8 bulk 18, 20, 72, 87, 90, 91
concept 164 Young's 18, 20, 87
equation 165 Moens 17
hydraulic 164 Momentum II
loops 169 Motor I 03, II 8, I 20
methods !55 Motor armature 56
networks 170
Impeller, pump 55, 56, 57, 100,
101, 103, 104, 117 Network 36, 73, 94, I 00, I 03, 112,
roughness 57 124, 130, 136, 145, 168,
Inductance !57, 160 169,172,173,176,178,
Inertia, moment of 56, 60, I 04, 201, 206
122, 123, 124, 125 gas 147
Interpolation 83, 97, 99, 115, 118 open channel 178
120,140,184,187,198 resonant I 6 7
Isothermal 148 Newtonian theory I
Iterative methods 70, 84, 96 Newton- Raphson 154, 188
Newton's 2nd law of motion 2, 28,
148
Nodal method 20 I
Joints 132, 202 Nodes 65
impedance theory of 168 Nonprismoidal channel 180, 181,
method of dealing with 45 182
Joukowsky 17
Junctions 14, 39, 46, 48, 65, 73,
167 Open channels 177
impedance theory of 167, 169, nonrectangular I 77
170, 173 prismoidal 177
in open channels 187, 191, 192 rectangular 177
n way, 130, 202 Organ pipes !55
Organ piping 65
Orifice 49, 63, 68, 176, see throttle
Korteweg 17 Oscillating component
of flow 158, 165
of head 158, 165
Oscillating valve 155, 174
Lax- Wendroff method 86, 199
Oscillation
Leap frog method 86, 199
forcing 166,172,173,174,176
Ligget and Woolhiser method 86, frequency of 161
200 mass 62, 65, 66, 134
Line pack 127, 128, 129 sinusoidal 162
Link method see Route method Oscillatory wave 161
Lister method 7 4, 18 2
Local losses 5, 46, 68, 130, 133,
154, 191 Parabolic partial differential
Lock in 128, 129 equations 7 5
Loops in networks 169, 170, 176 Pelton wheel !56
Perimeter, wetted 28
Period, pipe 22, 23, 25, 41, 60, 125
Method of characteristics 74, 77, 85, Pipe
107 bends 146
Micronuclei 89, 90 blank end of 167
214 Index
Pipe (contd.) suction-well 11 0
by-pass 126 torque 56
description for global trip 55, 56, 73, 103, 110, 111,
programming 203, 205 112, 121,124, 126
joints 45, 13 2, 168 turbine behaviour 1 03, 111, 113
junctions 46, 48, 73, 130, 154, Pumps 55, 56, 57, 100, 201, 202,
167 204, 205
lengths 73
loops 169 Reflection I 0, 13, 3 2, 3 6, 55, 1 00,
period 22, 23, 25, 41, 60, 125
128, 173
Poisson's ratio 20 negative 14, 22, 23, 25, 146
Polytropic efficiency !54 partial 14
Polytropic index 69, 127 Regular rectangular grid method 82,
Polytropic process 70, I 27, 148
184
Potential head equation 84, 99
Reimann equations 30, 32, 37
Power 55
Reservoir50
characteristic 39 42 45
Pressure head amplitude 173 , , ,
Pressure head, calculation at t:.t
Reservoirs 39, 40, 50, 55, 65, 144,
time intervals 41
Pressure rise, after instantaneous 145, 146, 154, 167, 173,
valve closure 21 187, 190, 191, 201,204,
Pressure transients, see Transients 205
and waterhammer static water level 62, 63, 66, 68
Price 200 Resistance, electrical 157, !59, 160
Prismoidal channels 180, 181, 182 Resonance 65, 15 5, 156, 164, 172,
Procedure calls, in global 173, 176
programming 205 Reynolds number 95, 96
Propagation constant 161 Rigid pipe-incompress ible fluid
evaluation of 162 theory 2, 66
Pump Riser 63, 64, 68
Route method, in global
by-pass 120, 1 26
programming 201, 202
characteristic equation 55, 56,
58, 59, 60, 62, 100, 101
102 , Schnyder-Berger on graphical
compound arrangement 123 method 36, 49, 72, 85,
description for global 129
programming 204 Separation 16
dynamometer behaviour 102 Shock wave 186
113 , Simple harmonic motion 67
efficiency 56, 58, 59, 60, 105, Sluice gates 186, 189, 190
122, 123, 124, 202 Solenoid 126
four quadrant operation 60, 111, Sound 1
113, 120 Specific energy 188, 189
impeller 55, 56, 57 Specific speed 115, 117, 118
in-line 108, 176 Speed, rotational 56
in parallel 1 22 Steady flow component 158
in series 122 Steady head component 158
power 104, I 05, 106, 107 Stoker, J. 199
run-down 107 Strain
run-up 120 circumferential 18
start-up 106 diametral 18
stations 120, 1 21 volumetric 91
Index 215
Strain energy 6 Unidimensional flow in channels
Streeter, V. 176 177
Stress, hoop 18
St. Venant equations 180, 199
Subcritical free surface flow 197, Vacuum 94
198 Valve 103, 129, 201
Supercritical free surface flow 76, ball 155
185, 186, 197, 198 butterfly 112, 156
Superposition 17 6 by-pass reflux 1 20, 126
Surface tension 88 characteristic 3 9, 40, 41, 4 2, 52
Surge 197 control 144
demand 196 downstream
ebb 196 closing 112
floodorstorm 186,187,189, opening 4
196 effective area 17 5
rejection 196 fractional opening 35, 39
suppression 124 instantaneous closure 10, 21
travelling 17 8, 192, 1 93, 19 5 linear closure 8
Surge tanks 47, 62, 68, 124, 125, motorised 136, 137, 138, 139,
126, 127,201,202 140, 141, 205
choke ring 63, 68 nonreturn or reflux 55, 100, 111,
equation integration 70 112,120,121,128,129,
Johnson differential 63, 68 132, 134, 144, 145
mass oscillation 66 part closed 144
pressurised 64, 65, 68, 125, 134 reflux, see nonreturn
simple 63 servocontrolled 72, 142, 156,
transient analysis 65 175
Suter diagrams 116, 117, 118 slow closure 6, 17, 22, 23, 45,
125
slow opening 42
Thermodynamics, first law of 149 slow partial closure 43
Throttles 85 stepwise closure 23
causing friction 49, 50, 52, 53, sudden closure 21, 125
54 Vapourous cavitation 94, 95
multiple 52 Vapour pressure 16, 50, 94, 110
Time level scanning, in global Venturi-flume 190
programming 206 Voltage 157
Torque, pump 56,115,117,118, Volute 100
120
Transducer, pressure 143
Transients I, 10, 16, 17, 22, 29, Waterhammer 1, 9, 28, 36, 60, 73,
36, 42, 50, 55, 60,63, 77, 78, 85, 86, 93, 121,
65, 72, 73, 87, 88, 90, 134, 156, 157, 160, 181,
100, 103, 106, 111, 112, 183, 184, 186
125, 146, 156 Waterhammer equations 73, 74, 77
Transmission line equation 157 integration of 79
Transmission line theory 161 linearisation of 157, 160
Turbine gate 62, 63 Waterhammer theory 1
Turbine mode of operation of a history of 2
pump 62, 100, 102, 103, Waves 10, 11, 12, 13, 14, 15, 16,
107, 111, 113 21, 173
Turbo-blowers 154 compressional 2
216 Index
Waves (contd.) 151,161,177,185,198,
[type 24, 25, 31, 32, 33 199
F type 24, 25, 31, 32, 33 equation 84, 99, 161
formation I 0 magnitude of variability 90
propagation I 0 minimum 93
shape 15 variation in 87,94
Wave form 64, 173 Weirs 64, 145, 146, 186, 188, 190
Waverider 38, 40, 41, 42, 45, 46,
47, 53 Zone, of influence 78, 183
Wavespeed 18, 42, 72, 85, 97 -99, of quiet 79, 183

Hydraulic Analysis of Unsteady Flow in Pipe Networks

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Hydraulic Analysis of Unsteady Flow in Pipe Networks

Hochgeladen von

Copyright:

Verfügbare Formate

Hydraulic Analysis of Unsteady Flow in

First published 1977 by

ISBN 978-1-349-02792-7 ISBN 978-1-349-02790-3 (eBook)

This book is sold subject to the standard conditions of the

Text set in 10/11 pt IBM Press Roman

2 Analytic and graphical methods 23

3 Boundary conditions for use with graphical

5 Variable parameters in unsteady flow 87

6 Boundary conditions: pumps 100

7 Other boundary conditions 130

8 Unsteady flow in gas networks 147

9 Impedance methods of pipeline analysis 155

10 Unsteady flow in open channels 177

Leeds, 1976 J.A. F.

B pump constant in equation H = AN 2 + BNQ- CQ 2 (chapter 6)

Cct coefficient of discharge of a valve

E Young modulus of elasticity

E1 \ constants in the characteristic forms of the unsteady gas equations

f Darcy fin the Darcy formula

fh hoop stress in pipe wall

g intensity of the local gravitational field throughout the text

h potential head - the sum of local pressure head and elevation of

I inertia of the rotating parts of a pump and motor set (chapter 6)

j frictional head loss/unit weight of fluid/unit length of channel

K bulk modulus of liquid

In chapter I also used: k = voLh

k constant in head ~ q equation h = kq 2 ,

L length, usually pipe length L

m hydraulic mean radius =~(defined in text)

Pi pressure generated by momentum change

Si valve stroke at beginning of a At period (chapter 7)

T pipe period 2L (chapters 2 and 3)

u velocity of impeller blade tips (chapter 6)

v mean flow velocity (chapter I)

velocity of whirl at exit from a pump impeller (chapter 6)

weight density of fluid

X distance along pipeline

z elevation of the pipe centre line above datum

Q real component of 'Y (propagation constant) (chapter 9)

T surface tension coefficient in chapter 5

1.2 Rigid pipe-incompressible fluid theory

Historically, the development of waterhammer theory has followed a pat-

By Newton's second Law of Motion

1.3 Sudden valve opening at the downstream end of a pipeline

!::.hi in equation I .2a is the excess of head at the downstream end of

This is the Darcy-Weisbach equation (Fanning equation in USA); the fused

!::.hi=_ (h _ 4fLv 2 _ kv 2 ) = _f:dv

where vis the velocity at time t.

so t = y2g~sK loge ( :t~:)

This can be plotted as shown in figure 1.3.

from the first spurt is approximately equal to the spouting velocity

1.4 Slow valve closure

A further example of the use of rigid pipe-incompressible fluid theory

where A is the area of the pipe, so by differentiation with respect to

time dvt=dae ~+.!v'li(hro.saedhn

now ho _h _ 4 fLvl = L dvt

dhn = nVfihn (ho _

(fi( t)) dhn = n~ (h 0 _ h _ 4[Lvt )

0 = nVfihmax (ho- hmax ) - 2hmax -d (/(t ))

so it is simple to evaluate hmax/h0 from equation 1.11.

If the time Tis split up into m increments

The next step of integration can now be performed

1.5 Distensible pipe-elastic fluid theory

static head~ t>,

]__ (!!._ + z) + !:'_ av + .!_ av + !_ = 0