Beruflich Dokumente
Kultur Dokumente
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
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
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.
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)
hf=~y
_gm
2
(as defined in text)
4[L
K =d + k (chapter 1 and chapter 5)
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)
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.
instantaneous
pressure grade
line
p
8x
Figure 1.1
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
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)
so V =
~~K
fiihs
a standard result from normal steady state theory as would be expected.
Rearranging equation 1.6
v
~ f\--- actual velocity history
t-~------
1 theoretical velocity history
(2qh, K )0 "5 /L
Figure 1.3
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
Simple waterhammer theory 7
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,
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
( dhn) = 2hs
dt t = o T
T
t:J.t=-
m
so
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.
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.
10 Hydraulic analysis of unsteady flow in pipe networks
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
(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
Simple waterhammer theory 11
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
z
L
--- v .....,_v~o
l' .,
h,~
I lh, I
pressure head plat
0c_ >I> _1,_
c c
(c)
~ .....,;,-' X
(d)
h,
(e)
J
L
:4
""reservo1r
~reduced pipe diameter X
(f)
Figure 1.4 (continued)
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
~ "reservoir
-
pressure head plot
I= 4L
c
(h)
I I_
h
I' I
hi
valve closure h;
occurs
....___;
(a)
Figure 1.5
valve closure
occurs
J/ L h,
_{
c
(b)
Figure 1.5 (continued)
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)
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
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
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)
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
Simple waterhammer theory 19
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
20 Hydraulic analysis of unsteady flow in pipe networks
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
cJf!i=/w
-+-
I 2
K ER
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
Simple waterhammer theory 21
1.9 Pressure rise caused by instantaneous valve closure
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.
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
22 Hydraulic analysis of unsteady flow in pipe networks
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.
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.
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
24 Hydraulic analysis of unsteady flow in pipe networks
Figure 2.1
- 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)
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
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.
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)
datum
Figure 2.2
Analytic and graphical methods 27
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
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
(2.11)
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.
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.
Analytic and graphical methods 29
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)
(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
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-
30 Hydraulic analysis of unsteady flow in pipe networks
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.
figure 2.3) and hx = ho + F (to +~),ignoring the f (t- :) term for the
Analytic and graphical methods 31
X
..
X
Figure 2.3
hx, == ho +F(t, + ~1 )
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
/i =-Fi-1 (2.22)
and for the case of a reflection at a closed end
/i=Fi-1 (2.23)
(2.24)
(2.25)
Analytic and graphical methods 33
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)
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)
{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
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.
36 Hydraulic analysis of unsteady flow in pipe networks
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.
[2.19]
Writing them both for two points at x and x' and at times t and t'
and
* Square brackets indicate that the equation with this number was introduced earlier.
38 Hydraulic analysis of unsteady flow in pipe networks
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)
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.
hs is the height of the reservoir static surface level above the valve
Figure 2.4
t/lo
)
r------8 -----::~
plan of pipeline
Figure 2.5
Analytic and graphical methods 41
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
/5
1/1.
)
c
-
~
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.
lj;3 0
4.0 ere
Figure 2.8
A
-
..__
~:::>.200
Ag
(a) (b)
Figure 2.9
44 Hydraulic analysis of unsteady flow in pipe networks
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
Analytic and graphical methods 45
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.
(2.77)
i.e. the heads at the junction ends of pipes joining at the junction at any
given time t must be equal.
Analytic and graphical methods 47
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
48 Hydraulic analysis of unsteady flow in pipe networks
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
Figure 2.17
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-
50 Hydraulic analysis of unsteady flow in pipe networks
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
q
Figure 2.20
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
52 Hydraulic analysis of unsteady flow in pipe networks
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.
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
54 Hydraulic analysis of unsteady flow in pipe networks
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
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
56 Hydraulic analysis of unsteady flow in pipe networks
from the rotational energy of the pump impeller, the motor armature and
other rotating masses is P and is given by
so n=In
.!!._
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
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)
58 Hydraulic analysis of unsteady flow in pipe networks
-- --
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
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.
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.
60 Hydraulic analysis of unsteady flow in pipe networks
(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.
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
Boundary conditions for use with graphical methods 61
steady state pump characteristic
h
throttle
(a)
coordinates q2T h 2 T
(b)
(c)
Figure 3.5
62 Hydraulic analysis of unsteady flow in pipe networks
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.
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
Figure 3.6
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
64 Hydraulic analysis of unsteady flow in pipe networks
(a) (b)
(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
Boundary conditions for use with graphical methods 65
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
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.
66 Hydraulic analysis of unsteady flow in pipe networks
Figure 3.9
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.
- - : .J L
Figure 3.10
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)
T= 21TJ[f- (3.26)
(3.28)
68 Hydraulic analysis of unsteady flow in pipe networks
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.
(3.29)
s
z
Figure 3.11
Thus (3.31)
:. 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)
(3.38)
70 Hydraulic analysis of unsteady flow in pipe networks
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.
(3.43)
~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.
Boundary conditions for use with graphical methods 71
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
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
(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)
and
ov ov (4.6)
dv=-8x +-8y
ox oy
The method of characteristics 75
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
(4.15a)
and dy = ,_
dx )- (4.15b)
X= At~- Bt = Ct ~- Dt
B 2 -A 2 ~ D 2 -C 2 ~
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 method of characteristics 77
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.
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
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.
The method of characteristics 79
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.
(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.
6t
ore ton (v~c)
X
6x
Figure 4.3
(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
The method of characteristics 81
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
R s
X
Figure 4.5
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
The method of characteristics 83
Figure 4.6
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.
84 Hydraulic analysis of unsteady flow in pipe networks
The characteristic equations can now be applied, i.e.
g (h p - h R) + Vp- VR 2fRvRivRIM _ O
- + - [4.15]
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
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.
-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
(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
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
88 Hydraulic analysis of unsteady flow in pipe networks
5.2 Gas evolution
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
90 Hydraulic analysis of unsteady flow in pipe networks
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.
so Vg = p r tJ.p eV (5.3)
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
Where KT' is the effective bulk modulus of the gas/liquid mixture including
the pipe distensibility effect.
So (5.1 0)
92 Hydraulic analysis of unsteady flow in pipe networks
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
Variable parameters in unsteady flow 93
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
94 Hydraulic analysis of unsteady flow in pipe networks
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.
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-
Variable parameters in unsteady flow 95
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.
_1 _ _
YJ-
[ 2.51 k/d
41ogto 2Y2JRe + 3.71
J
96 Hydraulic analysis of unsteady flow in pipe networks
(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.
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
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
98 Hydraulic analysis of unsteady flow in pipe networks
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)
[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
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.
H = Vwu- k ~ - k Vr 2
g v2g r2g
Figure 6.2
102 Hydraulic analysis of unsteady flow in pipe networks
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)
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
Boundary conditions: pumps 103
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
104 Hydraulic analysis of unsteady flow in pipe networks
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
(6.5)
( 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
Boundary conditions: pumps 105
N _ N + 3600Pflt
2 - 1 4rr 2 N 1 I ( 6.11)
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
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''
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.
(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.
Boundary conditions: pumps 107
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)
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
108 Hydraulic analysis of unsteady flow in pipe networks
~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.
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)
Boundary conditions: pumps 109
M N
X
Figure 6.4
(6.23)
( ~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
+ CR FR + csFs = O
g g
Let (6.26)
110 Hydraulic analysis of unsteady flow in pipe networks
The equation then reduces to
aQ 2 +/3Q+-y=O (6.27)
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.
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
112 Hydraulic analysis of unsteady flow in pipe networks
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
Boundary conditions: pumps 113
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-~
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.
114 Hydraulic analysis of unsteady flow in pipe networks
0
Q*
200%
HIS head
0 IS flaw
N 1s speed
denotes pump duty values
(b)
Figure 6.5 (continued)
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
Boundary conditions: pumps 115
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
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
Boundary conditions: pumps 117
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
118 Hydraulic analysis of unsteady flow in pipe networks
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.
So N = N. _ 60 7'sb.t {6.35)
I s 21T I
Boundary conditions: pumps 119
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
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
M R 0
5 N
pump
/':,.x
I /':,.x
Figure 6.8
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
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.
120 Hydraulic analysis of unsteady flow in pipe networks
6.13 Pump run up to steady pumping speed
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.
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
Boundary conditions: pumps 121
- 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)
Now
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
Boundary conditions: pumps 123
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
B' = "f.BpNp
N
and C'= "f.Cp
Figure 6.10
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
126 Hydraulic analysis of unsteady flow in pipe networks
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.
Boundary conditions: pumps 127
air vessel
Figure 6.12
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
Boundary conditions: pumps 129
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
(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
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.
soL saPavPa = 0
1-->n
= L saPaVa- L~a aPahP + L;a aPaha
1-->n 1-->n 1-->11
(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)
Other boundary conditions 133
-N=*---
------kJ~--- M R JOint 5 N
Figure 7.2 Figure 7.3
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
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)
134 Hydraulic analysis of unsteady flow in pipe networks
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
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.
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
so (7 .l 0)
where At is the cross sectional area of the air vessel and equals i" dt.
136 Hydraulic analysis of unsteady flow in pipe networks
~~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
l - h
h w2 =( [ t
)n X h w,. (7 .11)
h t2lj.
t
-
A, v,
Figure 7.6
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
(7.15a)
138 Hydraulic analysis of unsteady flow in pipe networks
and if A2 =A1
(7.15b)
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
Other boundary conditions 139
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
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
140 Hydraulic analysis of unsteady flow in pipe networks
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)
(7.19)
+ +
~
-+--..,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)
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
142 Hydraulic analysis of unsteady flow in pipe networks
_ 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
t
4dR Vp 1 Vp 1
-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.
(7.28)
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.
144 Hydraulic analysis of unsteady flow in pipe networks
7.6 Reservoirs
p~~--+-""'-----+p
N 5 R N
Figure 7.9
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-
Other boundary conditions 145
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,
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
7.7 Bends
8.1 Introduction
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
148 Hydraulic analysis of unsteady flow in pipe networks
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)
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
=a: (pAv(e + v; )) 5x
(8.3)
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)
s
Figure 8.1
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
pipe wall
Figure 8.2
= 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,
(8.I4)
152 Hydraulic analysis of unsteady flow in pipe networks
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
C = 2rrK
w, loge(~::) heat units per second per metre per kelvin (8.18)
C = 2rrK (8.19)
w, loge(~)
Unsteady flow in gas networks 153
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.
(8.20)
(8.21)
(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.
154 Hydraulic analysis of unsteady flow in pipe networks
8. S Boundary conditions
(8.23)
{8.24)
9.1 Introduction
155
156 Hydraulic analysis of unsteady flow in pipe networks
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.
(9.1)
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)
2jVIvl
prov1.d.1ng t h at t he - d - term can be 11neanse
. d.
ah+~ aq = 0 (9.5)
at gA ax
(9.6)
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
158 Hydraulic analysis of unsteady flow in pipe networks
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
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
becomes
av+_!_ai=o [9.1]
at cax
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.
axat - c 2 at 2
Rearranging equation 9.11
a..!!...= -gA (ah
I
-+Rq
1
1)
at ax
Differentiating with respect to x
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
(9.13)
denote
then (9.14)
Then
a2
H
--=Cm2emx
3x2 I
'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
Impedance methods of pipeline analysis 163
Figure 9.1
and
so
r
'Y
= Ag.Qj(R)2
c2 gA
+ R2
f3~ =~ [(g~r + R 2 J 1
\os (~tan-l(R~A)) (9.18)
-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)
Impedance methods of pipeline analysis 165
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
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
Hs =a+ b (9.21)
Similarly for q' from equation 9.16
gAil
Qs=~(a-b) (9.22)
IC 'Y
166 Hydraulic analysis of unsteady flow in pipe networks
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)
z s =ZR
_:_:__+ Zc _ _____:(rL)
__::___tanh _:______.:.
ZR (9.24)
I +-tanh (rL)
Zc
Impedance methods of pipeline analysis 167
From these results for Zs and ZR the change of impedance from one end
of a pipe to the other end can be calculated.
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
Figure 9.3
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
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)
Impedance methods of pipeline analysis 169
~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
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
170 Hydraulic analysis of unsteady flow in pipe networks
Z _ Zs 3 - Zc 3 tanh (Y3L3)
and z
-zc3s3 tanh (-y3L3)
R3
1
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.
R3
53 s
Figure 9.8
Impedance methods of pipeline analysis 171
As
this reduces to:
Zs s = Zc s tanh (rsLs)
Similarly
and
172 Hydraulic analysis of unsteady flow in pipe networks
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.
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
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 ))
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' ).
(l + ~~r ~ h; h'
So
Then
Impedance methods of pipeline analysis 175
aeo aeo
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:
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)
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
(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
178 Hydraulic analysis of unsteady flow in pipe networks
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.
Consider the flow into and out of the element depicted in figure 10.2.
water surface
,,,.n~
Figure 10.2
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=
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.
surface at x + 8x
aA 8x -r--.--------.----.;,-1
ax
Figure 10.3
dv (, av av)
pAfJx dt =pAfJx ~ ax + at
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]
8+88
8A =88d+d88
Figure 10.4
Unsteady flow in open channels 181
so ~A = B~d prismoidal channel
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
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
182 Hydraulic analysis of unsteady flow in pipe networks
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)
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
Multiply through by -1
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)
Figure l0o5
184 Hydraulic analysis of unsteady flow in pipe networks
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
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.
Unsteady flow in open channels 185
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.
M~l---x------~~~x--------~
Figure 10.6
't ~~
I /::,.x X
Characteristics in a highly super critical flow
(a)
(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.
Unsteady flow in open channels 187
(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.
Qt+~t = bdt+~tVt+M
so Qt+~t
vt+~t = v = - - (as dp = dt+~t)
p bdp
188 Hydraulic analysis of unsteady flow in pipe networks
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
c2
As and d u =dp =~
g
Unsteady flow in open channels 189
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:
(10.12)
and
bc 2 v I 2 v2 \
bdpvp=T=Cdbdcoy2g f+~-ddJ
(c
190 Hydraulic analysis of unsteady flow in pipe networks
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.
so dh
Qc=At dt + Qs
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.
192 Hydraulic analysis of unsteady flow in pipe networks
Using the same methods as those used for the waterhammer case, con-
tinuity gives:
(10.16)
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
~v1
CD
~Vw
Figure 10.10
f
@
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.
194 Hydraulic analysis of unsteady flow in pipe networks
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
-
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)
(10.25)
So
The result of this analysis of the travelling surge is exactly the same as
Unsteady flow in open channels 195
10.5 The profJle of a free surface flow when a travelling surge is present
Figure 10.12
---- ----
~ ~
(a) (b)
---- ----
~
(c)
Figure 10.13
~
(d)
(a) (b)
(c)
Figure 10.14
Unsteady flow in open channels 197
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
A
slope o __!_
v.
R
M N T v u w X
x, x2
Upstream flow subcritical
Downstream flow subcritical
(b)
Figure10.16
i.e. (10.30)
Solving equations 10.28 and 10.29 simultaneously gives vy2 and cy2.
Unsteady flow in open channels 199
wave path
X
M
slope -f.-w,
Figure 10.17
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.
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.
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
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
204 Hydraulic analysis of unsteady flow in pipe networks
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.
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.
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.
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
206 Hydraulic analysis of unsteady flow in pipe networks
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.
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
208 Hydraulic analysis of unsteady flow in pipe networks
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