PIPELINE SYSTEMS

FLUID MECHANICS AND ITS APPLICATIONS

Volume 7

Series Editor: R. MOREAU

MADYLAM
Ecole Nationale Superieure d' Hydraulique de Grenoble
Boite Postale 95
38402 Saint Martin d' Heres Cedex, France

Aims and Scope of the Series

The purpose of this series is to focus on subjects in which fluid mechanics plays a
fundamental role.
As well as the more traditional applications of aeronautics, hydraulics, heat and
mass transfer etc., books will be published dealing with topics which are currently
in a state of rapid development, such as turbulence, suspensions and multiphase
fluids, super and hypersonic. flows and numerical modelling techniques.
It is a widely held view that it is the interdisciplinary subjects that will receive
intense scientific attention, bringing them to the forefront of technological advance-
ment. Fluids have the ability to transport matter and its properties as well as
transmit force, therefore fluid mechanics is a subject that is particulary open to
cross fertilisation with other sciences and disciplines of engineering. The subject of
fluid mechanics will be highly relevant in domains such as chemical, metallurgical,
biological and ecological engineering. This series is particularly open to such new
multidisciplinary domains.
The median level of presentation is the first year graduate student. Some texts are
monographs defining the current state of a field; others are accessible to final year
undergraduates; but essentially the emphasis is on readability and clarity.

For a list of related meclwnics titles, see final pages.

Pipeline Systems

edited by

B. Coulbeck & E.Evans

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-90-481-4143-2 ISBN 978-94-017-2677-1 (eBook)
DOI 10.1007/978-94-017-2677-1

All Rights Reserved

© 1992 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1992
Softcover reprint of the hardcover I st edition 1992
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
This volume consists of papers presented at the International Conference on Pipeline Systems,
24-26 March 1992, Manchester, UK, organised by BHR Group Ltd.

Technical Advisory Committee

Mr E P Evans (Chairman) Sir William Halcrow & Partners Ltd
Mr K Austin Amstral UK Ltd
Mr I Bensted Thames Water
Emeritus Prof B Coulbeck Leicester Polytechnic
Mr D Crow Cranfield Institute of Technology
Mr C P Fairhurst BP Engineering
Dr R King BHR Group Ltd
Mr D Miller BHR Group Ltd
Dr D Muskett BP Chemicals
Mr M Papworth SFK Technology
Mr R Powell Brunei University
Prof AlanE Vardy University of Dundee

Overseas Corresponding Members

Ms M H Alegre National Laboratory of Civil Engineering, Portugal
Mr F Klingebiel Amtech (Deutschland) GmbH, Germany
Mr B Kusha Fluid Dynamic International, USA
Prof C S Martin Georgia Institute of Technology, USA
Dr Akihiko Udo Setsunan University

Organised and sponsored by BHR Group Ltd

Co-sponsored by the International Association for Hydraulic Research,
The Pipeline Industries Guild, The Institute of Petroleum and
The Institution of Chemical Engineers
Contents
Preface ix

SESSION A: ECONOMIC DESIGN

A review of pipe network optimization techniques 3

G A Walters, University of Exeter, UK

A generalized reduced gradient approach to expansion of water distribution networks 15

G Yu, R S Powell, Brunei University, UK

Optimal design of water distribution networks with multiple loadings 25

A Udo, Setsunan University, Japan

Design management of pipeline systems 29

D S Miller, BHR Group Ltd, UK

Economic pipe sizing- complex networks 37

B B Sharp, Burnell Research Laboratory, Australia

Cost reduction in irrigation networks by an efficient use of pressure reducing valves 43

R Perez, F Martinez, A Vela, Politechnical University of Valencia, Spain

Water supply systems optimization for developing countries 59

R G Cembrowicz, University of Karlsruhe, Germany

Time-step control in TMA for steady flows in large pipelines 77

M Shimada, National Research Institute of Agricultural Engineering, Japan

A system for the engineering design of transmission and distribution pipe networks 91
D G Papanikas, V Pantazis, P Papagiannidis, M Bitzas, A Protopsaltis, Alfapi SA, Greece

SESSION B: SAFE DESIGN

Measurements and computations of transients in pumped sewer plastic mains 117

T Larsen, University of Aalborg, Denmark; R Burrows, Liverpool University, UK

Surge protection on the Trimpley supply to Birmingham 125

T J Kingham, J M Drake, Sir William Halcrow & Partners Ltd, UK

Failure of a rural water system- a case study 141

A L Prasuhn, Lawrence Technological University; D Rollag, South Dakota State University,
USA

Fluid-structure interaction in non-rigid pipeline systems- large scale validation tests 151
A C H Kruisbrink, A G T J Heinsbroek, Delft Hydraulics, The Netherlands

Analysis of penstock fracture by water hammer 165

Y Kono, Y Sugai, Tohoku Gakuin University; T Sugano, Tohoku University, Japan

Experiences with surge protection devices 171

C S Martin, Georgia Institute of Technology; L Cobb, Christy Cobb Inc, USA
Pipeline systems for liquid sulfur 179
R Szczepanek, R Pichler, I-L-F Consulting Engineers, Germany

Characteristics of lignite ashes transport from steam power plants through pipelines 199
Z Vukelic, University of Zagreb, Yugoslavia

SESSION C: OPERATION

Network simplification by equivalent modelling for optimal operations 209

Y-C Chen, Thames Water Utilities Ltd; B Coulbeck, Leicester Polytechnic, UK

Knowledge based systems in operational scheduling of water distribution networks 231

M Likeman, SFK Technology Ltd, UK

Diagnosis of hydraulic performance of water supply systems 247

H Alegre, S T Coelho, National Laboratory of Civil Engineering, Portugal

On-line control of a city water supply and distribution system 261

B Coulbeck, Leicester Polytechnic, UK; C H Orr, E 0 Tech Inc, USA

SESSION D: MONITORING AND INSPECTION

Leak detection through hydraulic transient analysis 273

L Jonsson, M Larson, University of Lund, Sweden

State-of-the-art pipeline leak detection 287

P Black, Scientific Software Intercomp, UK

Flow measurement in large complex ductwork 299

A E Vardy, University of Dundee/Mott MacDonald, UK

Structural condition assessment of water trunk mains 311

R Oliphant, WRc pic, UK

Methods to determine parameters characterizing technical state of pipelines with pronounced 323
deposition
P V Davlianidze, L A Alekhin, T Tsamalashvili, I Zeikidze, Pilot Production
"SPETSTRUBTRANS", Republic of Georgia
Preface
This conference provides a forum for exchange of technical and operational information across
a wide range of pipeline activities. Various supply and distribution industries, and their service
organisations, have traditionally approached pipeline systems from many different perspec-
tives. The organisers believe that significant benefits can be gained by enabling representatives
from the oil, gas, water, chemical, power and related industries to present their latest ideas
and methods. An awareness of these alternative methodologies and technologies should result
in a more unified and coherent approach to each individual type of pipeline system.
The overall theme of the conference is the optimisation of pipeline systems, through design
analysis, component specification, operational strategies and performance evaluation, in order
to minimise both risk and the lifetime cost of ownership. Wherever possible emphasis is given
to important developing technologies with special consideration to use of computational equipment
and methods.

SYSTEMS APPROACH
For the major activities of design, operation and performance; pipeline systems can be
conveniently classified in terms of the systetV: components, constraints and objectives. These
are described using fluid terminology, to suit' the majority of conference participants, as given
below:
Components consist of pumps and valves (controls), pipe networks (transmission and distribu-
tion), reservoirs (storage) and consumer demands (disturbances). The arrangement of these
components, to form the system, must take into account the conflicting requirements of structural,
hydraulic, and cost, performance.
Constraints include the restrictions imposed, because of the physical construction of the
components, and because of operational necessities. For the present purpose the hydraulic
constraints, applying to pressures, flows, velocities and levels, are the most important. Both
steady-state and transient hydraulic performance must be considered to ensure adequate safety
margins.
Objectives encompass the distribution of source supplies to consumers as safely, securely and
as economically as possible. This implies optimisation, to ensure the best compromise between
these conflicting objectives, and also to achieve the best value of each objective criterion.
The above classification allows a systems-oriented approach, in general, and provides a sound
basis for optimisation considerations, in particular.

CONFERENCE PRESENTATIONS
The overall aims of the conference will be achieved by presentation of papers in the major
categories of system: economic design, safe design, operation and performance. In each case
the coverage will include aspects of safety, security and economy with a mix of analytical
techniques, computer implementations and practical validation of results, as outlined below:
Economic design covers methods and procedures to deduce the best configuration and layout
of pipe networks and components, for new systems, and also to cater for optimal expansion,
of existing systems. In each case the specified performance should be obtained for a minimum
investment cost whilst allowing for efficient operation over the expected range of operating
conditions.
 With a selection of papers covering: engineering design and management of pipe networks,
traditional and alternative steady-state analyses of pipe networks, review of some pipe network
optimisation techniques, and design and evaluation of optimal network configurations.
Safe design looks at some of the factors which must be taken into account to result in systems
which are intrinsically safe to operate. Consideration is largely restricted to the effects of fluid
transients. Uncontrolled pressure and flow surges can easily exceed the specified safety limits
of the network components. Such transient effects are important since they can be inadvertently
initiated by operator actions, or component failures, and can cause potentially hazardous
situations.
With a selection of papers covering: engineering and structural design activities, analyses
of network and component transients, effects of fluid transients on components and structures,
and systems failure modes and protective activities.
Operation concentrates upon recent developments for cost effective (optimal) control of some
specific water systems, but also introduces some of the related techniques used in other industries.
In essence these optimal control methods can be used to determine the best operating strategy
to satisfy the current operating requirements. In so doing they will take account of, and make
the most effective use of, the existing system design. However, practical applications of such
methods depend upon derivation of sufficiently simplified models and an adequate knowledge
of hydraulic performance and operating costs. Typically the supplies, consumptions, and
operating costs are time-varying and optimality will require time-varying control actions.
With a selection of papers covering: demand forecasting and prediction of consumption
patterns, evaluation of system performance using extended time simulations and performance
indicators, and evaluation and implementation of optimal control models and strategies.
Performance looks at methods for monitoring and inspection throughout the lifetime of some
common, and less common, types of pipeline systems. A knowledge of the degradation of the
system, particularly on account of changes in consumption patterns, component parameters,
and distribution losses, allows for corrective refurbishment and compensatory adjustment of
operating strategies.
With a selection of papers covering: inspection of pipes and other components, instrumentation
and measurement of systems data, and monitoring and detection of leaks and other abnormal
operating conditions.

ACKNOWLEDGEMENTS
The conference has been organised to suit a wide range of specialist staff including manage-
ment, computing, research, engineering and operations. While the conference will have special
appeal to delegates involved in fluid systems it will be of significant interest to all persons seeking
a systems-oriented approach to pipeline projects. In this respect the conference organisers are
pleased to acknowledge the response and active support of the many international experts in
the field,
Finally the organisers wish to thank all the individual authors for their important contributions.
Their presentations will help the participating delegates to focus on the key aspects of design,
operation and performance of pipeline systems
Bryan Coulbeck and Edward Evans, January 1992.
SESSION A: ECONOMIC DESIGN
 A REVIEW OF PIPE NETWORK OPTIMIZATION TECHNIQUES

GODFREY A. WALTERS
School of Engineering,
University of Exeter,
North Park Road, Exeter, EX4 4QF, U.K.

ABSTRACT

In the water, gas and oil industries, capital investment in pipe networks
is very high. Despite this, there has until recently been little effort to
adopt formal optimization techniques in pipe network design.

The case for using optimization is first presented. Then currently

available network optimization methods are outlined for various classes of
problem, with a review of their historical development and discussion of
their suitability and ease of use. These methods involve the use of Dynamic
Programming, Linear Programming, Non-Linear Programming and heuristic
techniques.
Networks of both fixed and variable plan layout are included. The
incorporation of reliability considerations into the optimal design process
is discussed, and the treatment of stochastic demands considered.

INTRODUCTION

Pipe networks are essential components of all systems involving the supply
of water and gas for domestic and industrial use and for the disposal of
waste water. They also feature in many irrigation schemes, in oil and gas
collection systems and throughout the process industries. Capital costs for
such networks are usually very high and it is therefore sensible that they
are designed to give maximum value for the money invested.
Historically most interest in pipe networks has focused on the
development of efficient algorithms for the analysis of flow, and there are
now very useful and efficient computer packages available for simulation of
existing and proposed new schemes. WATNET [1], for example, is widely used
in the U.K. for the analysis and simulation of water distribution systems.

In contrast, there has been comparatively little research interest

into the development of methodologies aimed at optimising .the design of

3
4

pipe networks, and there are very few computer packages available for
commercial use that help the designer to produce truly optimal pipe network
designs.

Traditional design methods

Design methods vary considerably between the different industries which use
pipe networks, and also between design groups within industries. However,
the process is likely to contain the following sequence of steps:

i) Determine likely demands/flows and assign these to nodes

ii) Choose the network layout
iii) Select pipe types and diameters
iv) Analyse system
v) Modify diameters (or layout) as indicated by analysis
vi) Repeat from (iv) until an acceptable design is produced

At some stage in the process the estimated cost of the system will be
evaluated and may be taken into account in deciding on the acceptable
design.

Stages (iv), (v) and (vi) may be omitted if pipe selection is on the
basis of an empirical guideline which uniquely determines the required pipe
size for a given flow. For instance, a uniform hydraulic gradient between
source and sink or a specified velocity in all pipes will fix the required
pipe diameters.

There may also be guidelines for the formation of the network layout,
either for the connectivity of a tree network, or for the provision of
"redundancy" in a looped system.

Although these guidelines have developed through the years from an

appreciation of what usually constitutes a good and economic design, there
is no guarantee that the designs produced will in any sense be optimal.
Even for a simple tree-like network there are a very large number of
possible designs that provide feasible solutions. For example Figure 1
shows a 6 x 6 grid of nodes to be connected by a tree network from a source
at one corner. The number of different possible layouts one could generate
is 22 5 or 33.5 million, each one of which could have a large number of
possibilities for pipe sizes.

It can be seen that the chances of hitting on the best solution are
very small. In the case of sewer system design the author has found that
most traditionally produced schemes based on standard guidelines are more
expensive than necessary by between 5% and 15%. Generally the larger and
more complex a system is, the larger the potential saving is likely to be.

Optimal design strategy

Adoption of optimization techniques in the design process aims to ensure
that design decisions are taken on a rational, usually minimum cost, basis.
Rules-of-thumb, such as specifying an "economic" pipe flow velocity,
therefore become redundant. Decisions on system layout, pipe type and pipe
diameter are taken on the basis of minimising the overall system cost, (or
maximising the benefit).

Where quantifiable, the predicted lifetime costs of the system

components should be used, but unless pumping is involved it is normal to
 5
source
• • • • • •
• • ••• •
• • ••• •
• • ••• •
• • ••• •
• • • •• •
nodes feasible pipes typical network
(1 of 33,554,432)

FIGURE 1

• •

•
•

•
(a) (b) (c)

FIGURE 2
6

use just the construction cost.

Any design produced must meet specified engineering and per·formance

criteria, and these therefore form a set of constraints on the optimization
process. The engineering criteria are relatively straight-forward.
Specified minimum and maximum velocities and pipe pressures are common, but
avoiding physical obstructions along the route and making use of
commercially available pipe sizes can also form constraints.
Reliability
Performance criteria are equally important and much less well defined. The
reliability of the system designed is clearly of fundamental importance
both to the system's owner and to any customers supplied by the network,
but perhaps for slightly different reasons.

Measures of network reliability are not defined on any universally

accepted basis, but various proposals have been formulated for water
distribution systems. For example, the author [2] has previously suggested
that from an individual customer's point of view the reliability is best
defined in terms of the availability of water at or above a specified
minimum pressure, (where availability is defined as the proportion of time
it is available), but that additional measures may also be desirable.

OPTIMIZATION OF TREE-LIKE NETWORKS

Tree-like or dendritic is the description given to networks with no flow
loops and only a single source (or sink). Flow diverges from the root
source, through a trunk main to branch mains which supply the demand nodes.

To the author's knowledge, all published optimization work for tree

networks has been based on minimising the network cost without specific
consideration of reliability issues.

Fixed Plan Networks

The optimization of fixed plan tree networks is the simplest class of pipe
optimization problem, and consists of choosing the set of pipe types and .
diameters that minimises the system cost.

For sewer systems and for gas and water distribution networks there
are usually a limited set of standard pipe types and diameters available
for the designer to specify. The head (or pressure) available between
source and sink is normally limited to a particular value.

The choice of diameter for a particular pipe can affect the design of
other pipes in the system. For example, use of too small ~pipe in one
branch of the network could result in a large head (pressure) loss in that
pipe, and require excessively large and expensive pipes elsewhere in the
network. Hence the choices of diameter for all pi~es in the system have to
be made simultaneously if they are to be made on an optimal basis.

Linear Programming Linear programming can be used to identify the

optimal pipe diameters provided certain conditions are met. This is
extremely useful as many highly efficient L.P. packages exist, greatly
reducing the need to develop special software.
 7

For L.P. to be applied, the following requirements must be met:

1) The cost of a pipeline of a particular diameter must be

proportional to its length, and must not depend on its
elevation or internal pressure. This is not valid for sewer
design, but is often applicable to pressurised pipe systems.

2) Pipes must be available in a limited number of discrete

types and sizes.

3) The design flow along each pipe length must be independent

of the design of the pipe system. As only tree networks are
being considered here, this will be true for design flows
which are based on steady state conditions.

The design can be optimised for several different flow conditions

simultaneously. Constraints on nodal pressures are readily incorporated
into the L.P. format.

The L.P. model is set up by specifying that each pipeline between

nodes consists of unknown lengths of each available pipe type and diameter
in series. The optimization will select the optimum lengths, most of which
will be zero. The optimum solution will use one or at most two different
pipe diameters along each pipeline length. A full description of the method
is given in several standard texts, for example Orth [3].

Dynamic Programming If condition (1) above is not met, a dynamic

programming approach can be adopted. This has been well developed for the
optimization of sewer system designs, in which the cost of a sewer is
clearly dependent on the sewer levels, which are in turn dependent on the
choice of pipe diameter. However the method can also be applied to those
pressurised pipe designs which cannot be handled by L.P.

Dynamic programming splits the design into a series of "stages", (the

design of a pipe between nodes), connected by a "state variable", (the
head, elevation or pressure at a node). A discrete formulation is used,
splitting the permissible range of the state variable into discrete values.
Working through the series of stages, assuming the cumulative cost of the
system is known for each of the discrete values of state variable at entry
to a stage, the cumulative cost can be found for all the states at exit by
considering all feasible designs and input states for that stage. Applying
the method recursively will generate the least overall cost of the system,
and will allow the optimum design to be identified.

The application of D.P. to sewer system design was proposed by Haith

[4] in 1966, and subsequently developed by Merrit and Bogan [5], Walsh and
Brown [6] and Walters and Templeman [7]. The method now appears in several
standard texts, for example Templeman [8].

In an attempt to reduce computational time, Discrete Differential

Dynamic Programming (D.D.D.P.) was introduced for sewer system design by
Mays and Yen [9] as a significant variation to the simple approach outlined
above.

Known also as Incremental Dynamic Programming, D.D.D.P. uses a very

limited number of discrete state variable values to define an initial
8

corridor, with successive iterations converging on the optimal solution.

One disadvantage with D.P. is that unlike L.P. the method is not
programmable as a standard algorithm, and so efficient software has to be
developed carefully for each type of application. Some design packages for
sewe~ systems do now incorporate D.P. optimization, for example DRENARP
[10].

Non-Linear Progr~~ming There have been several attempts at using

standard non-linear programming algorithms for diameter selection, for
example Lemieux et al [11] and Price [12]. However, these generally
require that the costs of the component pipelines are expressed as
continuous and preferably differentiable functions of pipe diameter or head
loss. As pipes are in reality usually available only in discrete sizes,
this makes cost functions discontinuous. To overcome this problem, a
continuous range of diameters is first assumed, and discrete diameters then
fitted in a post optimal process.

The results obtained cannot be regarded as necessarily optimal. This

together with high computational requirements and a lack of robustness
have seen non-linear programming algorithms largely abandoned for this
application.

Variable Plan Networks

The optimization becomes far more complex when the plan layout of the
network becomes part of the decision process. The plan layout problem can
take several different forms.

Spanning Tree Assume that a number of nodes (Figure 2a) are to be

connected to a single source (or sink) by a tree network. In general the
tree may be formed only from arcs between nodes as in Figure(2b), or may
use addi tiona! junction nodes, (Figure 2c) •. The former case is
superficially similar to the classic Operations Research problem of finding
the minimum spanning tree, but differs in one vital respect. This is that
the cost of a link is a non-linear function of the flow along the link, and
the flow depends on the connectivity of the network.

Barlow [13], looking at sewer design, and Rothfarb et al [14] dealing

with gas-field collection networks produced non-rigorous search techniques
for this type of problem.

Argaman et al [15] introduced the concept of "isonodal lines" to

define stages in a Dynamic Programming formulation of the sewer layout
problem, which however was unrealistically demanding in computational
effort. Work in this area was continued by Mays and Wenzel [1~] to produce
a more practicable model using Discrete Differentiable Dynamic Programming.
Severe ~estrictions in the model, however, make it of limited validity.
A radically different Dynamic Programming f·ormulation was used by
Walters and McKechnie [17] for pressurised pipeline networks, which
simultaneously optimises layout and pipe diameters. A directed graph is
first defined, this containing all the links that the designer wishes to
consider as candidates for the final tree network. The D.P. is so
structured that all possible trees that can be generated from this graph
are then considered, together with a set of possible pipe diameters for
each link. The method is successful for small to aedium sized networks, but
 9

computation is excessive for large systems.

Current interest focuses mainly on efficient heuristic algorithms

suitabl~ for large systems. The most promising of these seems to be the use
of Genetic Algorithms, as used by Goldberg and Kuo [18] for pipeline
optimization. Alternatively described as Evolution Theory the basic
technique was also used by Cembrowicz and Krauter [19] for sewer layout
design. The general principle is that subnetworks from a number of
relatively successful "'parent" designs are combined randomly to form a new
generation of "children". The children then form the parents of the next
generation, thus imitating the evolutionary process. New generations are
formed until no further improvement is detected.

Several other types of layout problem can be efficiently handled by

Dynamic Programming as described by Walters [20]. These include treating
the position of a junction node as a variable (as in Figure 2c), and the
siting of cross-connections between a parallel set of sewers.

OPTH!I ZATION OF LOOPED NETWORKS

Looped networks are designed for both water and gas distribution systems.
The flow along each link depends not only on the set of demands at the
nodes, but also on the design of each pipe in the system. There is a basic
paradox in the optimization of such networks, since if the optimization is
performed for a single set of design flows with cost minimisation the only
objective, then all duplicated paths from source to sink are removed and
the network becomes a tree, (Quindry et al [21]).

Even if the design is constrained to retain loops, the network will

be basically a tree with minimum sized links. forming the required loops,
but taking little flow.

To obtain optimal designs which use loops in an efficient manner, it

is necessary to know the purpose of the loops and to be able to quantify
their benefits. It appears that reliability is the main reason for
specifying loops. Hence quantifiable reliability constraints must somehow
be incorporated into the optimization process.

Looped Networks without Reliability

Fixed Layout If reliability issues are ignored, fixed layout optimization
can be performed by using the standard L.P. diameter selection technique
described earlier, but this time as a step in a gradient search method
which establishes the minimum cost flow pattern for the system, (Alperovits
and Shamir, [22]).

The results, however, will be similar to those obtained by using a

variable plan tree optimization method within the graph defined by the
looped network. Links not chosen for the optimum tree are then retained
as minimum diameter pipes.

Variable Layout For variable layout optimization, the optimal

network will again tend to be a tree system, but there is now additional
flexibility in the formation of the basic tree and in the selection of
links to form loops. As loop formation is part of the optimization, it is
10

essential to incorporate reliability constraints.

Looped Networks with Reliability

True assessment of network reliability, expressed in terms of the
availability of supply to the customers, must include the possibilities of
failure due to a number of different causes. While failure of the supply to
the system is the most serious, this is not within the control of the
network designer, This leaves the possibility of component failures (pipes,
valves, pumps etc.) and hydraulic failures due to extreme demands causing
low pressures. Pipe deterioration with time may also add to the problems.

The incorporation of reliability into the optimization process has

been attempted in a variety of ways, none of which cover all aspects of the
reliability issue.

At the simplest level, designing a network to handle a wide range of

possible flow patterns will increase the ability of the system to cope with
unusual circumstances, and hence its reliability. Indeed the minimum cost
network may become a looped system if widely differing sets of flows are
used as multiple design conditions.

Several authors have based design methods on identifying the

underlying optimum tree network as the basis for the looped system. For
example, Rowell and Barnes [23] used a non-linear programming method to
select the optimum tree, and then an integer programming model to choose
loop-forming links to provide emergency flow capacity when tree-links fail.

Goulter and Morgan [24] adopted a different two-stage approach. They

first select an optimal layout for a given set of nodal heads under a
constraint that there must be at least two pipe connections to each node.
They then select optimal pipe diameters and nodal heads, and repeat the
process until an optimum is reached.

A number of authors have used the concepts of "cut sets" and "nodal.
isolation" to deal with reliability, using probabilities of pipe failure to
calculate nodal isolation probabilities, (for example Tung [25]}. This idea
was incorporated into the optimization process by Kettler and Goulter [26]
and Goulter and Coals [27] to ensure that sufficient loops were
incorporated to provide all nodes with a specified probability of isolation
that should not be exceeded. However the concept has limited validity, as
supplies to a node may fail even if a link remains intact. This drawback
was partly overcome by Suet al 128], but pipe repair times were not
adequately modelled.
In the work so far mentioned nodal demands havP heen taken as fixed,
deterministic values. In reality, nodal demands are unsteady and will
follow some form of probability distribution. The implications of this on
the design of low pressure gas distribution networks was studied by Swan
and Walters [29] using Monte-Carlo simulation. Bao and Mays [30] used
similar simulation techniques to study the reliability of water networks
under variable demands. However, the incorporation of full simulation
techniques into the optimization problem is still some way off.
Tung et al [31] described a method for incorporating probabalistic
distributions for nodal demands, heads and pipe roughnesses into an
 11

optimization model, but did not consider pipe breakages.

The problems of incorporating both probabalistic demands and pipe

failures into an optimization model were discussed by Goulter and Bouchard
[32]. 'They proposed mathematical formulations of the problem, but did not
develop them into a practicable model.

CONCLUSIONS

Over the last two decades, considerable attention has been given to the
problem of developing suitable optimization models for pipe network design.
For tree networks with a fixed layout, standard methods exist both for
pressurised pipe flow systems and for free surface flow sewers. For
variable plan tree networks methods exist, but require further development
before being applied to large systems.

The development of methods for looped networks has been hampered by

lack of any clear definition of reliability for such systems. Reliability
constraints are required to ensure the preservation of loops which
would otherwise dissappear in the optimization process.

Full consideration of reliability in a practicable optimization model

has not yet been achieved, and may well be several years away from
fruition.
REFERENCES

1. Water Research Centre. WATNET (Water Network Simulation Model),

Swindon, U.K.

2. Walters G.A., Designing Water Distribution Systems for Reliability and

Economy, Proceedings of Conference on Computer Methods in Water
Resources II, Morocco, 1991, Vol.l, pp.217-228.

3. Orth H.M. Model Based Design of Water Distribution and Sewerage

Systems, John Wiley and Sons, 1986.

4. Haith A.D. Vertical Alignment of Sewer and Drainage Systems by Dynamic

Programming, Ph.D. thesis, M.I.T., Boston, 1966.

5. Merrit L.B. and Bogan R.H. Computer Based Optimal Dseign of Sewer
Systems, Pro~-~~~~-' 1973, EEl, pp.35-53.
6. Walsh S. and Brown L.C. Least Cost Method for Sewer Design, Proc. ASCE,
1973, EE3, pp.333-45.

7. Walters G.A. and Templeman A.B. Non-Optimal Dynamic Programming

Algorithms in the Design of Minimum Cost Drainage Systems. Engineeriqg
Qptimiz~~ion, 1979, Vol.4, pp.139-48.

8. Templeman A.B. Civil Engineering Systems, MacMillan, London, 1982.

9. Mays L.W. and Yen B.C. Optimal Cost Design of Branched Sewer Systems,
Water Resources Research, 1975, Vol.ll, pp.37-47.
12

10. Pereira D. and Camara A. Optimal Extension of Sewet' Networks, Proc.

Fifth Int. Conf. on Urban Storm Drainage, Osaka, 1990, Vo1.3,
pp.1:J9:J-8.

11. Lemieux P.F., Zech Y. and Delarue H. Desigu of a Stormwater Sewer by

Non-Linear Programming, £;?crr~J!ian J.gurnal__Q_f_ld_vil __ f:ngineering, 1976,
Vol.3, pp.83-89.

12. Price R.K. Design of Storm Sewers for Minimum Construction Cost, Proc.
First Int. Conf. on Urban Storm Drainage, Southampton, 1978,
pp.636-47.

l:J. Barlow J.F. Cost Optimization of Pipe Se•H•r Systems .!'I:Q_C__.__j__JL~h_Qf

(:_ty_il__Engineey..§, 1972, Pt.2, Vol.53, pp.57-64.

14. Rothfarb B. et al. Optimal Design of Offshore Natural Gas Pipeline

Systems. Qpera1;_j_Q_JLU_<;)_§.<l.~rch, 1970, Vol.18, pp.992-1020.

15. Argaman Y., Shamir U. and Spivak E. Design of Optimal Sewerage Systems.
p_roc. __ASCJi:, 1973, EE5, pp. 70:1-716.

16. Mays L.W. and Wenzel H.G. Optimal Design of Multi-Level Branching Sewer
Systems, Water Re~Q_~rce~~search, 1976, Vol.12, pp.913-17.

17. Walters G.A. and McKechnie S.J. Determining the Least Cost Spanning
Network for a System of Pipes by the Use of Dynamic Programming,
Qj_yil-Com_p_~-~. Proc. Second Int. Conf. on Civil and Structural
Engineering Computing, London, 1985, VoL 2, pp. 237-43.

18. Goldberg D.E. and Kuo C.H. Genetic Algorithms in Pipeline

Optimization. A.S.C.E. Journal of Computing in Civil Engineering,
1987, Vol.l, No.2, pp.128-41.

19. Cembrowicz R.G. and Krauter G.E. Design of Cost Optimal Sewer Networks,
Proc. Fourth Int. Conference on Urban Storm Drainage, Lausanne, 1987,
pp.367-72.

20. Walters G.A. The Design of the Optimal Layout for a Sewer Network.
-~_ng_~neering Optimization, 1985, Vol.9, pp. 37-50.

21. Quindry G.E., Brill E.D. and Liebman J.C. Optimization of Looped Water
Distribution Systems. ~_._Q.__j':__c, 1981, EE4, Vol.107, pp.665-679.

22. Alperovits E and Shamir U. Design of Optimal Water Distribution

Systems. Water Resources Research, 197~, Vol.13, No.6, pp.885-900.

2:1. Rowell W.F. and Barnes J.W. Obtaining Layout of Water Distribution
Systems. A-'Ji_,_Q~_j':_,,1982, HYl, Vo1.108, pp.137-148.

24. Goulter I. C. and Morgan D. R. An l ntegrated Approach to the Layout and

Design of Water Distribution Networks. Q_ivi.l_fu_tginee_r!!J~temJ;!,
1985, Vol.2, pp.104-ll:l.

25. Tung Y.K. Evaluation of Water Distribution Network Reliability.

Water Forum '86: World Water Issues in Evolution, ASCE, 1986, Vol.2,
pp.359-364.
 13

26. Kettler A.J. and Goulter I.C. Reliability Consideration in the Least
Cost Design of Looped Water Distribution Systems. Proceedings
International Symposium on Urban Hydrology, Hydraulics and Sediment
r:ontrol, Lexington, Kentucky, 19H3, pp.;~05-:l12.

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

Assessment in Pipe Networks. ~~~~~ Journal of Transportation
~_ngi,nee.r_il!g, Vol.l12, No.3, pp.287-301, 1986.

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

Optimization Model for Water Distribution Systems. A.S.C.E. Journal of
tl~raulic_Kngineerjng, Vol.l13, No.12, pp.1539-1555, Dec 1987.

29. Swan D.J. and Walters G.A. The Influence of Load Diversity on Gas
Network Design. Proc. Int. Symposium on Use of Computers for Gas
'Transmission and Distribution Systems, Budapest, Oct.l990.

30. Bao Y. and Mays L.W. Model for Water Distribution System Reliability,
ASCE Journal of___l:!x~auli~__]p_giaeering,, 1990, Vol.l16, No.9,
pp.1119-1137.

31. Tung Y-K., Lansey K. and Mays L.W. Water Distribution System Design by
Chance Constrained Model. Proceedings of 1987 National Conference on
Hydraulic Engineering, Williamsburg, Virginia, pp.588-593.

32. Goulter I.e. and Bouchart F. Joint Consideration of Pipe Breakage and
Pipe Flow Probabilities. Proceedings of 1987 National Conference on
Hydraulic Engineering, Williamsburg, Virginia, pp.469-474.
 A GENERALIZED REDUCED GRADIENT APPROACH TO
EXPANSION OF WATER DISTRffiUTION NETWORKS

GUOPING YU and ROGER POWELL

Control Engineering Centre, Brunei University
Uxbridge, UB8 3PH, UK

ABSTRACT

The paper explores the use of the generalized reduced gradient method(GRG) in
the design of expansion in water distribution networks. In the optimisation model,
the objective function includes capital cost of pipelines and operating cost of pumping
stations. The main constraints are a set of nonlinear hydraulic equations, upper and
low bounds of diameters, and a minimum requirement of pressure head at all nodes. A
modified Newton-Raphson technique is used in the hydraulic simulation to accelerate
calculation and the convergence and consistency of the algorithm is investigated. An
extension of the method that includes multiple demand loadings is also described.

INTRODUCTION

A considerable investment is made in water distribution systems every year. There-

fore, their optimal design has attracted much attention. A basic problem handled by
optimisation techniques is to derive the best solution for the minimization of both the
capital cost of pipe networks and the operating cost of energy under the constraints
of hydraulic factors and standards in service. Several approaches are available for op-
timisation of water distribution systems(1,2,3,4,5], but many techniques put emphasis
on the design of an entirely new system. These techniques, although suitable for the
design of new systems are not suitable for the expansion of existing water distribu-
tion systems[6,7,8,9,10]. Only a few models are reported to be able to handle optimal
expansion of an existing network[l1,12].

Water distribution systems are usually designed over a long period of time. Actu-
ally, the expansion of existing water distribution networks occurs more often than the
design of new networks. This is because there are some factors that make the systems
inadequate before the end of their design life. For example, the predicted demand
patterns to service consumers are impossible to define with any accuracy, accelerated
or uneven growth of consumers can occur. Furthermore the system capacity could be
affected by corrosion and deposition in pipes. Since one can not replace an existing

15
16

network with an entirely new network, attempts should be made to improve the de-
livery capability of the existing network by suitable additions. Such expansion and
modification may include:
1. Addition of new links connecting existing demand nodes.
2. Strengthening of some existing links by parallel connections or replacement of
these links by new ones.
3. Inclusion of additional demand nodes, source nodes and corresponding links to
expand the network to cover additional localities.
In producing the optimal design for an entirely new network, there are clearly less
constraints than in the design of the expansion of an existing network. Therefore,
the optimisation problem for the latter case is rather more complex. Essentially,
optimisation of new networks is just a subset of the optimal design of expansion of
existing systems.
Another challenge that few researchers have considered is to include multiple de-
mand loadings in the optimisation technique. In practice, the demand pattern is time-
varying and so it is necessary to consider a set of possible demands in the optimisation
process.
The objective of this paper is to present a methodology through the formulation
of an optimisation model for the optimal design of alterations to existing water dis-
tribution networks using multiple demand loadings. The model is based on nonlinear
programming and solved using the generalized reduced gradient method(GRG). De-
tails of the mathematical model are given in addition to an example to illustrate the
methodology.

OPTIMISATION MODEL

For the expansion of an existing water distribution network, consisting of m links

and n nodes, the basic optimisation model can be stated as:

Minimize COST= K 1 L L;Dt + K2 L Q;H; (1)

iEPn. iEp.

subject to

g(D,Q,H) =0 (2)

(3)

(4)

Where
D- diameter(m), D = [D 1 ,D2 ,···,Dm]T
Q- discharge supplied or withdrawn at nodes(l/s), Q = [Q1,Q2, · · · ,Qn]T
 17

H- pressure head at nodes(m), H = [Ht.H2 ,···,H,.]T

Dmln- permissible minimum diameter
Hmin- permissible minimum nodal head
g(D,Q,H)- vector of functions used in simulation
g(D, Q, H) = (gi(D, Q, H), 92(D, Q, H),···, g,.,.(D, Q, H)]T,
nn is the number of nodes at which the head is unknown.
L- length of pipe(m)
p,.- the set of new links
p.- the set of source nodes with pumping station
K 1- a constant related to cost of pipe and to present worth factor
e- exponent related to pipe constant, usually lying between 1 and 2
K 2 - a constant dependent on the unit cost of energy, time of pumping during a
year, efficiency of pump and motor, yearly average consumption factor
In the optimisation model, the decision variables are the diameters of new pipes
D,(i E p,.), the pressure head H will be solved in the simulation(Eq. (2)) if the diameter
D and loading Q are known. Thus the head H can be called state variable.
Because there is a nonlinear head loss relationship due to pipe flow, Eq. (2) is a
set of nonlinear algebraic equations which constitutes the most difficult constraint in
the model. Obviously, in the optimisation process, the flow simulation of the networks
will be performed many times, so a modified Newton-Raphson method is used to
accelerate the convergence rate. The formulation of the modified Newton-Raphson
method is stated as follows:

Hk,1 = Hk- <a':k)-1Vug(D, Q,Hk) (5)

Hk,, = Hk,1- <a':k)-1Vug(D, Q,Hk•1) (6)

Hk,m = Hk,m-1_ <a':k)-1Vug(D,Q,Hk,m-1) (7)

Hk+1 =Hk,m (8)

k =0,1,2,···

Where
(~)-Jacobian matrix at kth iteration in hydraulic simulation
m- the number of sub-iteration step
This method reduces the number of times the Jacobian matrix is re-calculated. If
two such intermediate iterations are carried out, ie m=2, then the convergence time is
reduced by about 15%.
18

GENERALIZED REDUCED GRADIENT METHOD

The generalized reduced gradient method(GRG) is an extension of the reduced

gradient method and is one of the most efficient techniques for dealing with nonlinear
programming with nonlinear constraints.
After defining the diameter D;(i E Pn) as nonbasic variables, the head H as basic
variables, the generalized reduced gradient R is given by

(9}

Where (~;;"")
Pn
is a sensitivity matrix, ~D:,
u '
the element in the matrix, means the sensi-
tivity of head H at node j to a change in diameter of ith pipe.
(~~)can be obtained from Eq. (2).

(10}

Where (~) is the Jacobian matrix resulted from hydraulic simulation in water distri-
bution networks.
The algorithm of GRG for the design of adjusted diameters in alteration of existing
water distribution networks is stated as follows:
1. Set k +- o, start with the initial feasible diameters Df(i E Pn), perform the
hydraulic simulation in networks, obtain head Hk.
2. At step k, let Dk be the current diameters, calculate the generalized reduced
gradient R

(11)

3. Determine the direction of moving for Df(l E Pn)·

d·-
'-
{-14,
(Dmin;- D;)14,
if 14 :S: Oj
if 14 > 0.
d; = o,
4. Find,\* by line search, such that

di -v;)l"·
'
Wh ere Amax = M.tn {(Dm, •. Ui < 0}

5. If max{,\*d;li E Pn} < e, stop; otherwise, Dk+ 1 = Dk + ,\*d, k +- k+l, go to step 2.

The main difficulty in the GRG algorithm is in step 4. There are no guidelines for
the search for >. to guarantee the head to be within the bound defined as Hk ~ Hmln.
In the situation where there is only one critical node, the node with the minimum
pressure can be found by simulation. Let the head at this node correspond to Hmin>
then the head at the other nodes will be within the constraint. A difficulty occurs if
 19

there are fixed head nodes in the networks. In this case the current ).. will be gradually
changed until the constraint is held.

TEST EXAMPLE

To illustrate the use of this technique, we will consider a simple hypothetical

network. The network, shown in Fig. 1, consists of eight nodes, nine pipes and two
loops. The solid lines represent existing links, the dotted lines represent planned links.
Table 1 contains the network parameters. In the optimisation, the following data are
used:
K1 = 67 .25; K 2 = 3.035; e = 1.57;"' = 2; fJ = 5.33

a,fJ are hydraulic exponents in the head loss relationship (Manning's Formula).

··Q}

...0
~ ( 2]

.....G) ----oldliDk
········-········ newlink

Figure 1 Test Network

TABLE 1
Data for Test Network

Pipe(Node) (N1,N2) Diameter(m) Length(m) Dmin(m) Load(I/s) Hmin(m)

1 1,2 unknown 450 0.005 -700 30
2 2,4 unknown 230 0.005 110 30
3 3,4 unknown 310 0.005 70 30
4 4,8 unknown 290 0.005 190 30
5 5,8 unknown 340 0.005 50 30
6 1,6 0.35 250 none 80 30
7 3,6 0.20 200 none 50 30
8 6,7 0.30 300 none 150 30
9 none 0.20 270 none none none

Throughout the test example the results of the GRG algorithm will be checked for
the global optimum and consistency.
The optimisation model (Eq.(1)-Eq.(4)) is composed of a nonlinear objective func-
tion and nonlinear constraint equations, unfortunately it is rather difficult to judge
whether or not these nonlinear functions are convex. However, it is well-known that
optimal design of water distribution networks results in a convex programming prob-
lem if the pipe flows at all links are assumed known. Because there exist old pipes,
20

the distribution of pipe flow is a more complicated factor in the design of expanding
networks than for entirely new networks. Different flow distribution in networks will
greatly influence system cost. Fig. 2 shows the results of optimisation using various
flow distributions[13].

(a) Cost= 122552 (b) Cost= 114527

(c) Cost= 108374 (d) Cost= 106525

~~
~ [5]
~0

:i>
~~~~
o,
~
0.236
- - - 4
[3] ~
:s,;:;
~ ~c:il'-'

- - - - -~:~ - - - -.....@
(e) Cost= 105483 flow (1/s} -head loss (m) ff) Cost= 106908
diameter (m)

Figure 2 Results of Optimisation Based on Different Network Flows

After analysing the results shown in Fig. 2, it is found that system cost drops when
the flows in the old pipes are reduced(Fig.2(a)-(e)), and the cost rises again if the flows
in old pipes are too small(Fig.2(f)). The solution of Fig. 2(e) is the best. This shows
 21

that deciding on the distribution of load carried by old and new links is the key to
the design of expansion in networks. Unlike the design of entirely new networks, the
idea of choosing hydraulic gradient level and discharge in links as decision variables
respectively can not be easily realized in expansion of existing systems. The GRG
algorithm which chooses the diameters of newly established links as decision variables
produces very good results. Table 2 shows the solutions of the test network solved by
GRG method under the input of different initial values of diameters.

TABLE 2
Results of GRG Algorithm Based on Different Initial Values of Diameters
TABLE 2.1

Pipe(Node) Diameter(initial value) Diameter(optimum) Flow Head of Node

1 0.2 0.5990 514.7 35.50
2 0.2 0.5619 404.7 32.79
3 0.8 0.2222 -33.7 30
4 0.8 0.4457 181.0 31.58
5 0.7 0.2687 -31.0 30.00

Initial Cost=337920 Optimal Cost=105549

TABLE 2.2

Pipe(Node) Diameter( initial value) Diameter( optimum) Flow Head of Node

1 0.4 0.6006 517.2 35.27
2 0.4 0.5784 407.2 32.57
3 0.4 0.2269 -35.0 30.01
4 0.4 0.4420 182.2 31.52
5 0.4 0.2868 -32.2 30

Initial Cost=133554 Optimal Cost=105715

TABLE 2.3

Pipe{ Node) Diameter(initial value) Diameter( optimum} Flow Head of Node

1 0.6 0.6001 515.5 35.44
2 0.6 0.5748 405.4 32.74
3 0.6 0.2208 -34.1 30.01
4 0.6 0.4413 181.3 31.67
5 0.6 0.2677 -31.3 30

Initial Cost=121192 Optimal Cost=105559

TABLE 2.4

Pipe(Node) Diameter(initial value) Diameter( optimum} Flow Head of Node

1 0.8 0.6013 514.6 35.50
2 0.7 0.5606 404.6 32.85
3 0.5 0.2210 -33.7 30.00
4 0.45 0.4434 180.9 31.63
5 0.3 0.2664 -30.9 30

Initial Cost=l15855 Optimal Cost=105515

Analysing the results in Table 2.1 to 2.4, it can be seen that there is a very slight
difference among the solutions, but considering the error resulting from the criterion of
22
convergence, these solutions are thought to be same. Looking at the extreme diversity
in the vaxious initial values of diameter, shown in Table 2.1, it is probable that the
GRG algorithm possesses a strong consistency. Comparing with the results in Fig. 2,
it is found that the flow distribution derived from GRG algorithm is very similar to
that of Fig. 2(e), which indicates that the algorithm has achieved the global optimum.
A significant phenomenon is that the assignment of flows and the choice of diameters
is such that there are two end nodes in network flow, node 3 and node 5, their pressure
being Hmin simultaneously.

EXTENSION OF THE MODEL

The demand pattern in a real water distribution system is time-varying. To get
satisfactory behavior of a system under different anticipated demand loading, the
model developed herein for a single loading must be adapted to deal with multiple
loadings. During the design period several typical loadings could be selected, these
loadings include peak-flow, average-flow, low-flow and other flow patterns which occur
more frequently. Denote Q 1(l = 1,2,···,n1) as selected loadings, n 1 is the number of
loadings.
The optimisation model can be reformulated as:

nr
Minhnize COST= K1 L L;Di + K2 L L QlH! (12)
iEp,.. 1=1 iEp~

subject to

(13)

(14)
(15)

The generalized reduced gradient R is rewritten by

(16)

Denote R; as a component of R,

R· _ aCOST ~"'""' aCOST aHf (17)

' - aD- + L...., L...., aH! aD-
' 1=1 iEp~ " 3

The generalized reduced gradient has a physical meaning. In Eq. (17), the first term
means the rate of variation in capital cost to a change in diameter of jth pipe, it's
value is a positive number; the second term means the sum of variation in energy
cost which is caused by the change in head at all pumping station nodes in different
demand loadings, going a step further, the change in head is due to the disturbance of
 23

diameter. ~ is the sensitivity of head at the ith pumping station node to a change
in diameter ~f jth pipe in lth loading. The decision of an increase or a decrease in
diameter depends on whether the sign in the sum of two terms is positive or negative.
aH'
The submatrix (~)can be solved from Eq. ( 18 ) .

l = 1,2,···,nl (18)

Where (i&r)- the Jacobian matrix in hydraulic simulation for lth demand loading
Eq. (18) shows in the algorithm extended to multiple loadings the simulation will
be performed for n1 times in each iteration.

SUMMARY AND CONCLUSIONS

A GRG algorithm is developed herein for the optimal expansion of existing water
distribution systems subjected to a single loading pattern. There are three advantages
in the use of the GRG method for the optimisation model. First, unlike other tech-
niques[6,7,8,9J, the algorithm can immediately treat diameters as decision variables
in complicated expanding systems. Second, through the Jacobian matrix the results
of hydraulic simulation can effectively be used to construct the vector of generalized
reduced gradient. Third, the algorithm can easily deal with the extension of multi-
ple demand loadings. The observation from the test example is that the algorithm
has good convergence and strong consistency; when the diameters are optimised, the
distribution of network flow and the selection of the nodes with permissible Hmin are
optimally determined automatically. To speed the convergence of iteration, a modified
Newton-Raphson method is used. This paper has explored the possibility of extend-
ing the optimisation model to include multiple loadings. The extension is rational and
realizable, although will result in longer convergence times.

References

[1] Jacoby, S.L.S., "Design of Optimal Hydraulic Networks," Journal of the Hy-
draulics Division, ASCE, 94(HY3): 641-661, May 1968.
[2] Liang, T., "Design of Conduit System by Dynamic Programming," Journal of
the Hydraulics Division, ASCE, 97(HY3): 383-393, Mar. 1971.
[3] Cembrowicz, R. G., and Harington, J. J., "Capital Cost Minimization of Hy-
draulic Networks," Journal of the Hydraulics Division, ASCE, 99(HY3): 431-440, Mar.
1973.
[4] Deb, A. K., "Optimization of water Distribution Network Systems," Journal of
the Environmental Engineering Division, ASCE, 102(EE4): 837-851, Aug. 1976.
[5] Cenedese, A., and Mele, P., "Optimal Design of water Distribution Networks,"
Journal of the Hydraulics Division, ASCE, 104(HY2): 237-247, Feb. 1978.
24

[6] Alperovits, G., and Shamir, U., "Design of Optimal Water Distribution Sys-
tems," Water Resources Research, Vol.13: 885-900, Dec. 1977.
[7] Quindry, G., Brill, E. D., and Liebman, J., "Optimization of Looped Wa-
ter Distribution System," Journal of the Environmental Engineering Division, ASCE,
107(EE4): 665-679, Aug. 1981.
[8] Rowell, W. F., and Barnes, J., "Obtaining Layout of Water Distribution Sys-
tems," Journal of the Hydraulics Division, ASCE, 108(HY1): 137-148, Jan. 1982.
[9] Kessler, A., and Shamir, U., "Analysis of The Linear- Programming-Gradient
Method for Optimal Design of Water Supply Networks," Report, Faculty of Civil
Engineering, Technion-Israel Institute of Technology, Haifa, Israel, Sep. 1988.
[10] Morgan, D. R., and Goulter, I. C., "Optimal Urban Water Distribution De-
sign," Water Resources Research, Vol.21: 642-652, May 1985.
[11] Shamir, U., "Optimal Design and Operation of Water Supply Systems," Water
Resources Research, Vol.10: 27-36, Feb. 1974.
[12] Bhave, P. R., "Optimal Expansion of Water Distribution Systems," Journal
of the Environmental Engineering Division, ASCE, 111(EE2): 177-197, April1985.
[13] Yu, G., "Optimal Design of The Expansion in Water Distribution Systems,"
Dissertation of Ph.D., Department of Environmental Engineering, Tongji University,
Shanghai, P. R. China, 1986.
 OPTIMAL DESIGN OF WATER DISTRIBUTION NETWORKS WITH MULTIPLE LOADINGS

AKIHIKO UDO
Dept. of Electrical Engineering, Faculty of Engineering
Setsunan University
Ikeda-Nakamachi 17-8, Neyagawa, Osaka 572, JAPAN

ABSTRACT
In water distribution networks, the demand rate becomes extremely high
locally and causally. Typical examples are the demand for fire fighting
and that for preparation of rice transplantation. Then, in the design of
water distribution networks, optimization has to be considered for
multiple loadings. This paper develops a new method of optimizing the
design of water distribution networks with multiple loadings. The problem
is formulated into a combinatorial optimization one, and efficient
algorithms to solve it are proposed.

INTRODUCTION
Various methods of designing water distribution networks have been
proposed to date. Among them, the method by Nishikawa and Udo [1] for
single loading has novel features: it can clarify the trade-off between
enhancement of the supply-reliability and reduction of the construction
cost; it is effective even for large-scale networks of 1.000 nodes, 1.500
links and/or 300 loops. Using the method as a basis, this paper develops
a new method of designing water distribution networks with multiple
loadings.

EMPIRICAL PROCEDURE FOR DESIGN OF NETWORKS WITH MULTIPLE LOADINGS

Design by Predetermining Flowrates in Each Pipe [1]
Design by the method developed in [1] for single loading (hereafter we
call the method Method-!) has two phases:
The first phase determines flowrates in each pipe-section so as to
minimize an objective function which is related to reliability of the
water-supply. The layout of mains determines an important characteristic
of the network obtained: highly reliable or economical.
The second phase finds out the minimum-cost combination of commercially
available pipes which realizes a flow pattern determined in the first
phase and satisfies the conditions on the pressure head and the flow
velocity.
25
26

Empirical Procedure for Design of Networks with Multiple Loadings

An alternative procedure which is now used for design of networks with
multiple loadings and is evaluated to be tolerably practical consists of
three steps:
Sl) Execute design by using Method- I for every loading.
Let the diameter of section i chosen for loading 1r be d"'.
S2) ~et d, = a,, where d, is the diameter of section i, and
d , = max. d. , .
S3) Modify d.'s so as to reduce the cost as far as possible within the
limits that the constraints on the pressure head and the flow velocity
are satisfied for all loadings.
This procedure is favorable in the following points:
(1) The layout of mains commonly specified in Sl) is transferred to the
network finally obtained.
(2) The diameter finally chosen for link i is d, or d ,, one rank
smaller value than a, commercially available. It is seldom that more
than one rank smaller values are chosen.
( 3) Flow velocity is fairly equalized in the whole network for all
loadings, as in the networks designed by using Method-! for each loading.
However, the work of S3) is a process of trial and error by iterating
steady-state flow calculation and requires significant time and expertise.

DESCRIPTION AS A COMBINATORIAL OPTIMIZATION PROBLEM

Being based on the forgoing discussion, a design problem of networks with
multiple loadings is described as follows:
[Object] Minimize the total cost of pipes under the following
constraints.
[Head Constraints] Heads at nodes are in an allowable range for all
loadings.
[Pipe-diameter Constraints] Candidate diameters of pipe-section i are
restricted to d, and d,. If a unique diameter is obtained for section
i, that is, d., 's are-the same for all 1r, d, is fixed to the value.
This set of constraints is effective for equalizing the flow velocity.

ALGORITHMS
Basic Algorithms
Forward Algorithm: Sl) Let the diameter of each section bed,, the
smaller one In the candidate list. Hereafter, let us call a set of
diameters of all sections be a state. In the first state, the head is
generally lower than the minimum allowable value at many nodes.
S2) Using an index introduced later, change the diameter of an appropriate
section for the larger one in the candidate list.
S3) As to this new state, execute steady-state flow calculation for every
loading. If the head constraints are not satisfied for some loading,
return to S2), otherwise stop.
AP, 'lAC, is used for the index of S2), where AP,' is the sum of the
effective parts of increase in heads at nodes, AC, an increment in the
cost, both by changing the diameter of section i from d i to a io
Mathematical expression of AP,' is
AP,' = 2:. 2:;.{min(p'.;, Pmin)-p.;} (1)
where P.; is the head at node j for state v and pattern 1r , p '.;
the head at node j when (_j_, is changed for d, from state v , Pmin
 27

the lowest allowable head, i* a set of nodes where p. 1 <Pmin. p, 1

and p',; differ by state, however, symbol v is omitted for brevity.
Calculate AP, 'lAC, of all section i whose diameter can be made
larger, and change the diameter of the section which gives the maximum
value of the index. If AP, "lAC, is large, it means that the heads at
nodes which are lower than Pmin at state v can be made higher
effectively with a lower increase in cost.
Backward Algorithm: The forward algorithm finally gives a solution
wh1ch sabshes the head constraints. However, this solution is not
necessarily optimum, and it is sometimes possible to restore some d,('s)
to d, ( 's), without violating head constraints. The backward algorithm
is used for this case in contrast to the forward algorithm.
Notes: Evidently, the basic algorithms requires a number of steady-
statetlow calculations. Therefore, for saving computation time, the
following measures are important.
(1) To avoid needless calculation- For example, if the head constraints
are satisfied for pattern n at state v , AP,' can be calculated without
executing steady-state calculations for the pattern.
(2) Use of good initial values- The values in state v become good
initial values for calculation of ( p '. ;)'s.
Modified Algorithm
For fast solution, the forward algorithm is modified as follows:
(1) Simultaneous enlargement of plural diameters- Based on the
calculation of a set of ( AP, 'lAC, )'s at state v, diameters of plural
sections are made larger at a time.
( 2) Partial omission in calculating ( AP, 'I AC,) 's - If AP, 'I AC, is
small for i at state v, the possibility that the diameter of section
i is enlarged at the next state is considered to be low. Therefore, the
value is not calculated at the next state. This omission is effective in
every other state.

Example
Topology of the network: Consider the network of 85 nodes and 99
sechons shown In Flg. I. Head at source S is 18m, the ground level Om
everywhere, and the minimum allowable head 5m.
Specification of mains: Mains are specified in three ways as shown in
Fig. 2. In the flgure, finally chosen diameters by the proposed method
are already shown; the diameters with character L are the smaller ones
in the candidate list, the others the larger ones.
ILoadings: Seven loading patterns are considered. In pattern-i( i=l.
2, · ... , 6), the demand rate is 0.0228m 3 Is at the nodes in region R i
shown in Fig. 1 and is 0.00384m 3 Is at the other nodes. In pattern-7, the
demand rate is 0.005376m 3 Is at all nodes. The demand rate is always zero
at the nodes where 0 is written. It is generally possible to omit a
pattern in which demands of high rate distribute along mains because large
diameters enough to draw out a large amount of demand are chosen for
mains. In our cases, pattern-3 and pattern-4 are not considered in (b),
the case of looped mains.
Available pipes: Table 1 shows available pipes. Hazen-Williams
coeff1c1ent 1s supposed to be 130 for all of them.
Computation time: A machine code generated by a BASIC compiler was
executed on a 16 b1 ts personal computer (Intel 80286, 12MHz). Each result
shown in Fig. 2 can be obtained in 2 hours.
28

s lo200ml 0 TABLE 1 TABLE 2

0 0

20ml
r ---,
I! r--, r
I
--,
I
I
Available pipes and
their costs
Distribution of the
flow velocity
I
I
I ! !
II Rl I
R3 R5
I
pipe cost• flow # of pipes
I I
I
II I
I iI i velosi ty
I (a) (b)
I
I
I
I
I I
I
II VP 75 5.89 (m/s) (c
I
I I I
I I
I
I
I I L VP100 6.77
r_._I -, r-t-I -,: rT VP125 7.73 0- OS 1* - -
--
I 1 I
I
I I I
II II I I 1: I: VP150 9.36 1.0 18 16 7
I! ! II II i_jI
I
I
I
I I
I I
I 28
I!I VP200 12.00

r'I
1.5 33 31
-
I I I
I I
35
I
VP250 15.80 2.0 31 32
I
L
t~
I ~ I
L VP300
DIP350
20.40
34.85
-- 2.5
3.0
16
-
15
4
21
7
I
II -
I

I I 1 DIP400 40.12 3.2 - 1 1

I
I I
l I
I
! DIP450 45.51
I
R2
I
I
I R4
I
I
I :R6 DIP500 51.22 * 0.184m/s
II II I
DIP600 63.78
i
I
i I

L ____ J
I L I I
I

*
L----.l L--.J
x 10' Yen/m
Figure 1. Sample network.

r200ml r-zooml 1200m1

zoo2n m) (m m)
bOO 600 m) 600

J J.
300, 300, !50 (m

J. I'""
400 350 200 200 ISO
400 300 250 zoo,. [450 300L !50
600 IZ5 125 150

2rlll I
300 2!l0 200t. 450 75 300, 125 500 100 100 125
400 2rm 240m
400 300 250 200 450 100 300, 125 75£ 100 125
500
400 300 250 200t. 250 200 IJttn 125 250 3001. 250 150 500 125 250 150 250 200
300t. 300 300 300 300 400 300 125 400 150 j500 250/. !Z5t. . lZOL 200
300£ 300 300 250{. 400 100 400 125 500 100 100 125
300, 300 300 250 400 lOOt. 400 125 500 IDOL 125 125
300 200l. 300 250 Z50 250 250 400 Z50 125t. 250 400 250 200 500 250,. IZ5 250, 150, zso ISO,
300 300 300 300, 3o0 150 350 200 500 . 125, . 150, 200
300 300 :mo 300 350 125 350 125 450 100 125 125
300 Z50 250 300 350 lOOt. 350 150 450 100 125 125£
300 150L Z50 zoo,_ 250 200L Z50 350 250 !50 200 350 zso, 200 450 250, IZ5L 250 150 250 200
250 250, 300 250 3uO,. 1,101. JUOL !50 4UU IZ5L 150 200
250 250 250 250 300 125 300, 150 400 100 125 150
250 250 250 250 300 125, 300L 150 400 125 125 125,
250 200 250 200 250 250 zoo, 300 250 150 250 300 300 200 400 300 !50 300 150 300 200
1zuu
'""'
200 250 250 250 250 2ou 200 250L . l5UL 200
200 250 200 200 zoo, 150 Z50 150 250L 100£ IZS 150
200 200 200 150£ zoo, 125L 250 l50t. 250£ lOOt. 125£ IZ5,
200 200 200 250 150, 250 200 Zl!!l£ zoo 150, zoo zoo 300 zoo 25!!J.~t.., J®L~OL 2Ql!L 250 200

(a) No main (b) Looped mains (c) Tree-like mains

( 144,389 X 10 3 ) ( 153,566 X 10 3 ) (140,805 X 10 3 )
Figure 2. Application results (Cost in Yen)
Flow velocity: Table 2 shows the distribution of the maximum flow
velocity 1n each pipe throughout all loading patterns. Except in a pipe
of case (a), where the maximum value is 0.184m/s, the velocity becomes
more than 0.5m/s for some loading, showing effient use of pipes.

REFERENCE
1. Y. Nishikawa and A. Udo: Modeling and Design of Large Water-
Distribution Networks for Irrigation, Proc. of IFAC/IFORS Symposium on
Large-Scale Systems, pp. 849-854, PERGAMON, 1986.
 MANAGING THE DESIGN OF PIPING SYSTEMS

DONALD S MILLER
BHR Group Limited
Cranfield, Bedford MK43 OAJ, UK

ABSTRACT

Piping systems are often designed without access to the best knowledge and data and to the
relevant plant operating experience. The result is high capital and operating costs, piping and
equipment failures, excessive maintenance, and down-time costs and serious safety and
environmental incidents. The knowledge and technology exists to develop decision support
systems which, combined with changes in the way projects are managed, could dramatically
improve the design and operation of piping systems. Work underway to produce decision
support systems for piping systems is outlined. How these systems relate to the development
of concurrent engineering within the fluid process based industries is discussed.

INTRODUCTION

Piping systems typically account for 20 to 40 percent of plant capital costs in the chemical,
power, oil and gas and related fluid process industries. Piping failures cause upwards of 40
percent of major insurance claims in the chemical industry. In the fluid process industries,
piping failures are the most common cause of catastrophic incidents leading to loss of life,
destruction of plant and environmental pollution. When companies have devoted attention and
resources to improving the design, reliability and safety of piping systems, dramatic
improvements have been achieved. These improvements need to be achieved as a matter of
routine. This requires changes in the socio-technical infrastructure of companies and the
development of a new generation of computer based decision support systems.

The knowledge and experience exists to design piping systems for good flow
performance, reliability and safety (1). There are many reasons why this knowledge and
experience is not accessible or used during the design of piping systems. This paper aims to:

29
30

• Show how the present methods of managing projects within the fluid process
industries contributes to poorly designed piping systems;

• Outline how concurrent engineering can provide the conditions for dramatically
improving the design of piping systems;

• Show the importance of considering the hydrodynamics of individual and assemblies

of piping system components throughout the design cycle;

• Detail the reasons why companies need to collaborate to speed up the production of
decision support systems for piping systems and to ensure that these systems meet
their company's requirements.

LIFE CYCLE METHODOLOGY FOR PLANT DESIGN

A better understanding of design processes has led to the realisation that, upwards of 80 per
cent of plant lifetime costs are committed before detailed design of the plant commences.
Typically, the conceptual and preliminary process design activities account for 1 to 2 percent
of overall design, construction and commissioning costs. There are few computer based tools
to support the conceptual design phase in comparison with the detailed phase of plant design.
Part of the reason for a lack of tools is that in the past it was not seen as cost effective to
invest in computer tools to support an activity that only accounted for at most two percent of
costs. In comparison, CAD and numerous analysis software packages are used for the
detailed design stage of process plants. The use of CAD and other software tools
substantially reduces detailed design costs but they add little value in terms of overall plant
lifetime costs.

Productivity and other benefits from using CAD and analysis software are gained by
design departments in plant owners and in plant contractors. It has, therefore, been relatively
easy for design departments and contractors to make the case to purchase software systems
since a direct relationship between investment and benefits could be demonstrated. A direct
relationship does not exist between investment in software tools to support the conceptual
design stage and the benefits this brings in reduced plant lifetime cost (cost of ownership).
Because benefits from improved conceptual design activities accrue to different parts of a
company, or even to a different company, it has proved virtually impossible, to date, to
 31

justify developing software tools for conceptual design, even though this is where 80 percent
of lifetime costs are committed.

The problems of justifying investment to improve the conceptual design process are
compounded by the way fluid processing based industries organise themselves. Typically the
design and building of a new piping transportation system or plant is carried out as a series
of phases. Each phase can be the responsibility of a different department or organisation.
This gives rise to many problems associated with the communication of information about the
design from one phase to the next. All too often part of the reasoning behind design intent
is lost from one phase to the next. Implementation of the conceptual design can be poorly
carried out because the reasoning behind the design is lost. Other problems arise because one
person does not have responsibility for a system from conception through to an operating
plant. This lack of continuity in responsibility was clearly demonstrated to the author's
organisation when it developed a suite of software, called FLOWMASTER (2) for piping
system analysis. Although the fluid process based industries have great need for such
software, its piping systems are often not analysed for dynamic events until problems are
encountered. The FLOWMASTER suite of computer programs were originally targeted at
the fluid process based industries but have found a much readier market in the aerospace,
automobile and defence industries. In these industries the team responsible for the design and
ongoing support of fluid systems are clearly identifiable. Within the fluid process based
industries responsibility is diffuse, ill defined and does not extent to the totallifecycle of the
plant. In these conditions no one can justify buying fluid flow analysis software for piping
systems or the manpower to utilise the software.

Worldwide there is much activity to develop electronic data interchange standards for
the process industries. These will have many benefits related to detailed drawings of the plant
and its material content. However, current work will have little benefit for the
communication of the reasoning and knowledge that goes into a design. Much research is
currently directed to how design intent can be carried through a project.

The literature shows there is an increasing awareness by plant owners and contractors
of the need to adapt a lifecycle approach to the design of production facilities, pipeline
transport systems and processing plants. Major consideration in a lifecycle approach are
factors such as cost of ownership, time to design, build and commission, minimal production
of waste product, inherent safety, low environmental impact. Concurrent
engineering/integrated process and plant design, is the name being given to a new approach
32

to the design of fluid processing plants. Concurrent engineering is a methodology that

incorporates from the outset all elements of a plant's lifecycle. In addition it provides a
design environment that supports many tasks being carried out concurrently across design
phases. Full implementation of concurrent engineering will take upward of ten years to
achieve in the fluid process based industries but companies can begin to achieve substantial
benefits soon after they begin to introduce concurrent engineering methods and support
systems. Through a better understanding what concurrent engineering involves companies
can:

• Make maximum use of their existing investment in CAD and other computer systems
whilst evolving towards concurrent engineering.

• Direct their R & D expenditure so as to ensure that the outcome of projects will
benefit future concurrent engineering operations.

• Collaborate with other companies in developing the core concurrent engineering

technology that is common across many companies and industries.

• Ensure that the development of computer tools for concurrent engineering support the
company in making best use of the technologies that give it competitive advantages.

PIPING SYSTEM SPECIFICATION AND SELECTION

Company managements, professional institutions, national and international standards

organisations and other bodies have a long history of developing standards, design manuals,
specifications and data sheets to support the specification and selection of equipment for
piping systems. The result is numerous codes of practice, company and international
standards, design guides, reports and other aids. However, for all the thousands of man
years of investment in such aids and the vast reservoir of data, experience and knowledge
created, the majority of designers have little effective support in carrying out specification and
selection tasks. There is an extensive literature which reflects the frustration felt by plant
owners, contractors, equipment manufacturers and others over our inability to solve the
problem of inadequate support for the specification and selection process. To many managers
specification and selection tasks are viewed as routine work and as such they feel providing
support should be a relatively easy task. The reality is that although the general task of
specification and selection is routine for an experienced designer, the knowledge content,
 33

individual application requirements and breadth and depth of the interactions required between
engineering disciplines, varies from task to task.

It is not unusual when specifying a component, such as a pump, for a fifty page
document to accompany the data sheets from an internationally accepted standard. The
document will typically deal with requirements like:

• compliance with government regulations;

• need to accommodate specific past experience of the plant owner or contractor;
• standardisation of spare parts across plants;
• utilisation of locally available components.

Strange as it may seem, no formal requirements analysis has been carried out on
specification and selection tasks in the fluid process based industries. Since the early eighties
at BHR Group, studies have been ongoing to understand the support companies need to carry
out the specification and selection of piping system components.

Using the findings about company needs, BHR Group, has been developing the
expertise, methodologies and tools to carry out large scale requirements analysis studies and
knowledge elicitation exercises. This work has involved industrial collaborative projects with
support from the Alvey and other DTI innovation schemes and a number of EC projects
under the framework programmes for R & D. BHR Group is now carrying out projects to
develop the requirements analysis for decision support systems for the specification and·
selection of piping system components. Extensive interaction with managers and specialists
in contractors, plant owners and equipment manufacturers is providing the basis for
developing requirements specifications. The extent of the understanding of requirements is
being investigated through the building of a demonstration system for the specification and
selection of control valves. Other applications scheduled for implementation are centrifugal
pumps and flowmeters. For these applications access is needed to managers, designers and
engineers in a broad spectrum of companies.

PIPING SYSTEM RELIABILITY

The scant information published on the cost of poor piping system reliability indicates that
these costs could be greater than 10 percent of the yearly profits of companies whose main
activities involve fluid based processes. With such a cost relative to profitability one would
34

expect companies to have strategies to tackle the reliability problem. In reality this is not
happening because the magnitude of the problem requires an evolutionary approach that
involves the industry as a whole and the industry is not organised to tackle such problems.
For the present, disasters like Piper Alpha and its subsequent enquiry, provide the main
impetus for change. Piper Alpha and other disasters bring out the fact that if piping system
components were much more reliable, maintenance would not have been necessary in the first
place removing the opportunity for human error that led to the accident.

Corrosion/erosion is the most common cause of piping system failures but piping
system reliability problems account for a majority of total plant maintenance costs. Valves,
centrifugal pumps and pipe joints are the three biggest offenders. BHR Group is investigating
how to support designers to improve the reliability of piping systems. This work is
concentrating on piping system hydrodynamics but will be capable of supporting other factors
such as corrosion and erosion.

As stressed in (1), serious plant problems are often the result of failing to provide
appropriate hydrodynamic conditions at inlet to equipment and processes. An all to common
problem is a well designed item of piping equipment suffering excessive vibration, loss of
performance and poor reliability because of bad inlet flow conditions. Similarly, numerous
process operations suffer from poor flow distribution, whilst nuclear and other plants have
been subjected to damaging vibrations because of failure to consider flow distribution and
flow stability.

A major difficulty is that engineers and designers do not have sufficient training or
background knowledge in hydrodynamics or the time to assess the hydrodynamic performance
of their designs. The question is how can they be helped to design piping systems for good
hydrodynamic conditions. As with the problem of specification and selection, the first task
is to understand the environment in which designers operate. Currently BHR Group is
carrying out a requirements analysis study of company needs to improve piping system
reliability. This will be followed by the development of the specification and production of
a software module for pumping installations. Companies are being encouraged to support the
project through the participation in a project steering group.
 35

CONCLUSION

The knowledge, data and technology exists to design piping systems that are efficient, reliable
and safe. To apply this knowledge and technology, within a lifecycle methodology for plant
design, requires the development of computer based decision support systems. Changes in
the socio-technical infrastructure of companies and their contractors and suppliers are also
necessary. The development of decision support systems and the bringing about of
organisational changes will take many years.

The move towards concurrent engineering, supported by appropriate computer based

decision support and analysis systems, will allow for smaller design teams. Within these
teams there will be much better communication of design intent leading to piping systems that
are robust, efficient, reliable and safe.

The most effective way for companies to acquire the benefits of concurrent
engineering is to participate in collaborative projects that develop the core technology and the
decision support systems to utilise the core systems. They can then customise these core
systems to meet their requirements.

REFERENCES

I. Miller, D.S., Internal flow systems. 2nd Edition, Elsevier, Barking 1990, pp. 33-49.

2. FLOWMASTER, Amtech UK Ltd, Amstral House, Featherstone Road, Milton

Keynes, 1991.
 ECONOMIC PIPE SIZING- COMPLEX NE1WORKS

Dr. BRUCE SHARP

Director, Burnell Research Laboratory
Hardidge Court, Doncaster, 3108 Victoria, Australia

ABSTRACf

The economic sizing of pipelines within complex networks may be accomplished by new
mathematical models (and enhanced computing facilities), but the best solutions may well
depend upon the judicious use of certain basic ideas derived from the study of simple
elements. Thus the notion that the existence of a degree of variability of flow demand should
lead to solutions where there should not be a gravity main, or that pumping by boosting to
cope with the peak flows is economical must be tested. Some examples will be illustrated to
demonstrate these ideas, particularly in the case of a 'tested' complex network.

INTRODUCI'ION

Whenever the optimum solution to a distribution problem is being examined, as well as the
reliability of supply, age-old principles need to be restated from time-to-time. Thus the notion
of a 'base demand (flow)' and the peak demands need to be put into perspective, exactly as in
the case of electrical networks where options combining coal or oil power stations and hydro-
electric often depend greatly on the peak load to base load ratio. There are supplies which are
essentially constant, as in the case of a direct industrial demand and where the utilization factor
is high and variability is low and the economical solution is not a question which depends on
flow (pipe line type is still an economical choice).
On the other hand, when there is variability of flow demand, some basic philosophies emerge
that are recent and perhaps they question some of the old entrenched habits of the planning and
design of supply systems. As a simple example, one learns at an early stage that it is
appropriate with an independent pumping (force) main to set a figure of 8-12 hours per day
pumping on the average to its associated storage. This allows both a labour supervised
system to work with approximately one shift a day and/or a peak pumping period of 14-16
hours with time in hand for maintenance or breakdowns. However, as shown by Sharp (1),
the simple force main operating as above is grossly under-utilised and costs maybe of the
order of 35% more than a system comprising two different pumping capacities so that there is
a 24 hour pumping per day with a smaller diameter pipcline as the appropriate economic
choice. Likewise the choice of a gravity main from a major supply storage can be disputed if
there is variability of the order of three times the flow requirements from a base (minimum)
flow to peak flow. This ratio is not unusual in water supplies taking into consideration daily,
monthly and seasonal variations. Thus, again, it has been proved by Sharp (1) that the
economic solution is a booster for the peak flows with a smaller pipeline for the gravity base
flows. This leads to the generalisation that given sufficient variability, there should never be a
gravity main.
37
38

With these generalisations as a starting point, one might turn ones attention to more complex
systems than those considered above, particularly if there is an announced variability of
demand, with base and peak flows.

MODERN OPTIMIZATION MODELS

Complex systems are now able to be analysed with the improved computing available and
network optimization models have emerged. There have been problems of efftciency of
network analysis with a large number of elements, whilst simultaneously developing
numerical schemes to handle the optimization aspects as indicated by Vigus (2) and Waters
(3). Some systems have had simplifying features which have made the analysis easy, such as
the New York City Water Supply studies reported by Quindry et al (4), and the current
literature is being directed towards reliability based optimization methods as in recent studies
by Duan et al (5) and Goulter and Coals (6).

If one takes the imposing study reported by Walski et al (7), there seemed to be a consensus
that of the many models applied to a test network, there was only of the order of 10% variation
in the solutions but all called for the exercise of a good deal of engineering judgement with
manual calculations to solve portions of the problem. Although the variability of the flow
demands is only of the order 2.2, it was decided to study that network with the above
philosophies in mind taking advantage of the trend of t.'le solutions already found by the
various optimization models.

BATILE OF THE NETWORK MODELS : EP1LOOUE

e booster

5km

-- Sharp
+ ------ Lee
Figure 1. Network and changes
 39

In Figure 1 above is shown the basic network tested in this exercise. It is first necessary to
emphasise the basic requirements laid down for a solution given that there was a growth
prediction to year 2005 and a number of options such as mains reconditioning, etc., that were
not taken up by the participants.

(1) A minimum pressure of 40 psi to be exceeded for 1.8*average flow.

(2) A minimum pressure of 20 psi to be exceeded for 1.3*average flow plus

designated frre loadings. (This was the more difficult to satisfy).

(3) The need to be able to fill elevated storages at off-peak flows from the
single pumped source.

A study of the solutions suggested some were able to satisfy (1) above but not (2). Using
the solution of Lee, et al, as a benchmark, which closely satisfied ( 1), but was slightly
deficient in (2), the optimum cost is reasonably of the order of $13 million U.S. Their
solution allowed for boosting into the elevated storages to enable (3) to be satisfied. As with
all solutions pipeline duplication was dominant

If one now adopts the basic concept that there is sufficient variability to suggest smaller or less
pipeline duplications and used boosting out of the elevated storages at peak flows, a solution is
possible by using only half the pipeline increases suggested by Lee, et al. In fact only one
booster is necessary to almost equate with that of Lee for the condition (1) above and two
boosters (from separate tanks) are able to satisfy condition (2), in all cases at year 2005. An
additional tank at node 140 as with Lee's solution is also used. In the table I below is listed
the costs, qualifying that some refinements might be necessary to exactly solve the critical
conditions. In Figure 1 are shown the pipeline additions for for both solutions.

TABLE I
Summary of costs in $ * 106
Description Lee.etal Sharp

Pipes 4.6 2.6

Pump equipment 1.8 1.8
Tank 0.5 0.5
Energy 6.0 6.6
Total 12.9 11.5

The presence of boosters at the two original tanks at nodes 65 and 165 provides for flexibility
as only one is absolutely necessary for condition (1) and in the costing the need for extra
boosting during the other condition is assumed not to incur significant energy costs as it is
largely for fire. Also the need to boost into the tanks during offpeak is acknowledged and is
accomplished with a relatively small amount of additional pipework: and valving using the
same boosters. The boosters had identical characteristics with normal discharge and head of
0.315m3 s-1 and 33.5m respectively. When one considers that the variability of flow is not as
great here as experienced by the author elsewhere, it is rewarding that there are real cost
benefits to be gained by boosting as indicated. Alternatively, one would have to express
scepticism for optimization models that do not seem to have incorporated as fundamental,
certain basic ideas which are proven for simple element systems.

With the implication that there might be benefit to be gained by more pumping than previously
envisaged. it is clearly important to develop reliability models coupled with optimization.
40

In the recent study by Duan et al (5), it was made clear that the role of pumping (and tanks)
had not been incorporated adequately in traditional optimization techniques and perhaps
confirms the need to recognise the findings of simple element systems. It should be noted that
the additional pumping and dynamic effects raises the possibility that water hammer is
significantly more serious, but the operation of boosters, generally in a byepass mode does
not usually lead to serious problems.

FURTHER COMMENT

The study by Walski et al (7), stated that the models could not size tanks or locate them and
hence greater consideration should be given in future to this requirement. This seems
particularly true as it has been suggested by Sharp (1), that it is perhaps conceptually wrong
to locate a storage (centrally or at some convenient high point), prior to consideration of how
much pumping is to be employed. The basic reason is that a storage acts as a buffer between
input and output and therefore always has the potential of reducing the variability of the flow
required as input to the system and variability is one of the major factors effecting the
economics of the system.

It need only be stated here that any optimization model should be able to assess the effect the
storage volume has on variability between the input and output flows generated for the system.

CONCLUSION

An additional evaluation of a 'tested' complex system suggests that further economic gains can
be made by incorporating the basic ideas derived from the economic analysis of simple element
systems.

In summary - In the presence of significant variability of flow demand there should be greater
recognition of the value of pumping (boosting), to cater for the higher peak flows and gravity
alone for the base (minimum) flows. If there is a simple direct pumping main, the idea of
limited (under utilisation) hours of pumping, could be abandoned in favour of continuous
pumping (at 2 rates), with economic advantage.

Finally, given the ability of storages to reduce the variability of flow input required, it may not
be realistic to place or size storages before considering the degree of pumping that may be
economically employed.

REFERENCES

1. Sharp, B.B., Economics of pumping and the utilization factor. Proc. ASCE., L
.!:WL. EnE.,1984, 111(11), pp. 1386-96.

2. Vigus, C.L., Mter computer water network analysis the next step is
optimization. Watercomp '89, Melbourne, 1989, pp. 165-8.

3. Waters, C., Pipe network optimization. Watercomp '89, Melbourne, 1989. pp.
169-72.

4. Quindry, G.E., Brill, E.D. and Liebmen, J.C., Optimization of looped water
distribution systems. Proc. ASCE., .!.~En&.. 1981, 107(4), pp. 665-79.
 41

5. Duan, N., Mays, L.W. and Lansey, K.E. Optimal reliability-based design of
pumping and distribution systems. Proc. L.Hxd..lm&.. 1990, 116(2), pp. 249-
68.

6. Goulter, I. C. and Coals, A. V. Quantitative approaches to reliability assessment in pipe

networks. Proc. ASCE., L.IrlmL~. 1986, 112(3), pp. 287-301.

7. Walski, T,M., Brill, E. Downey, JR., Gessler, J., Goulter, I.C., Jeppson,
R.M., Lansey, K., Lee, Han-Lin, Liebman, J.C., Mays, L., Morgan, D.R.
and Ormsbee, L. Battle of the network Models : Epilogue. Proc. ASCE., L.
~Ea. Plan. andMlm., 1987, 113(2), pp. 191-203.

APPENDIX - Network Model Data

Pipe No N N Dia. (m) Add Length (m)

1 2 7 .406 .406 3660

2 2 3 .305 .356 3660
3 2 11 .305 .457 3660
4 3 7 .305 2740
5 7 10 .305 1830
6 7 9 .254 .356 1830
7 6 7 .305 1830
8 6 9 .254 1830
9 6 8 .305 .152 1830
10 8 9 .254 1830
11 9 15 .254 1830
12 9 10 .254 1830
13 10 15 .305 1830
14 8 15 .254 1830
15 3 6 .254 1830
16 3 4 .254 1830
17 3 5 .254 .305 2740
18 4 5 .254 1830
19 5 8 .254 1830
20 8 140 .254 1830
21 15 140 .203 1830
22 15 16 .203 1830
23 10 16 .203 1830
24 10 11 .203 1830
25 11 16 .254 .254 1830
26 11 12 .203 .254 1830
27 12 16 .152 2740
28 12 13 .203 1830
29 13 16 .254 .254 1830
30 13 17 .203 .254 1830
31 16 140 .203 1830
32 17 140 .203 3660
33 5 140 .203 3660
34 140 21 .254 1830
35 140 20 .254 1830
36 20 21 .152 1830
37 5 18 .254 1830
38 18 20 .152 1830
39 6 65 .305 30
40 16 165 .305 30
41 25 140 .305 30
Tanks : 65,165 add 25
 COST REDUCTION IN IRRIGATION NE1WORKS BY AN
EFFICIENT USE OF PRESSURE REDUCING VALVES

Rafael Perez, Fernando Martinez, Antonio Vela.

Fluid Mechanics Group. Dept. of Hydraulic & Environmental Engineering
Politechnical University of Valencia, P.O. Box 22012, 46080-Valencia, SPAIN.

ABSTRACT

The cost effective design of hydraulic networks has been traditionally studied from the
point of view of the relationship between hydraulic variables and economic parameters,
with piping being the main element studied. The reason is clear: the piping is by far the
most costly item of a projected network. However it is not usual to find an explicit
consideration of the influence that pipe thickness has on the cost of the network because
of the added difficulty that this aspect poses to formulating the problem of optimization.
In irrigation networks, which are typically branched, it is advisable to place Pressure
Reducing Valves (PRY's) to fulfil three main goals: 1) to control the flow rate extracted
by users, 2) to ensure uniform irrigation and 3) to reduce the pipe pressure. In relation
to this last goal, we present a method for reducing network cost through the reduction
of the pipe pressure by means of an efficient use of PRY's. Its principal characteristics
are: 1) the method is based on the principles of Dynamic Programming and 2) it is
independent of the sizing of the diameters of the network.

INTRODUCTION

The economical design of hydraulic networks has long been a topic of study for
researchers. However the first precedents that we can find of the generalization of the
problem date from the sixties. In [1] we can find a schematic overview of the methods
used in the design of hydraulic networks. In the references we find different models based

43
44

on Linear Programming ([2], [3], [4] and [5]), Non-Linear Programming ([6], [7], [8] and
[9]), Dynamic Programming ([10] and [11]) and other heuristic methods [12]. Although
the models proposed differ, they all have in common the study of the relation between
pipe cost and hydraulic capacity (diameter).

The influence of source pressure on the network cost

The cost per linear metre of pipe does not only depend on the diameter, it also
depends on the material used and the pipe thickness (related with the maximum
allowable pressure). As a result, a reduction in the static pressure can lead to an
important saving in the cost of piping.
When the source head is a decision variable, the classes of pipe depend on this
head. It is usual to overcome this difficulty by proposing classes of pipe on the basis of
an initial estimate of the source head and then optimizing the diameters and that source
head; later, the classes of pipe affected by changing the initially proposed source head
will be changed if necessary.

Thousands S per year (pipes)

35 ---·--;- ;-··················· .
with speed
llmltatloa Cll
32 - .........................!................ , .

29 . . . . . . . i .................. !

20~J_~_L~-L4-~~~~~-+~L+J_L4-L~~~~~

72.5 80 87.5 95 102.5 110 117.5 125 132.5 140 147.5 155
Pumping Pressure (m.)
Figure 1- Variation of the cost of a branched network with source pressure.
 45

Fig. 1 shows how the cost of an optimized network varies with source head. The
continuous curve (1) represents the values obtained by constraining the flow velocity
values, while in the case represented by the discontinuous curve (2) the sizing has been
carried out without limiting allowable flow velocities.
In both curves, a relative minimum will be reached when the saving obtained by
reducing diameters is inferior to the increase in costs due to the change of pipe pressure
class, so that several relative minimums could be obtained.

The performance of Pressure Reducing Valves

Irrigation networks are designed to operate under more controlled conditions than
urban networks and the use of PRY's is advisable for different purposes: to limit the flow
rate available to the user, to keep the working pressure of irrigation devices within the
operating limits, etc... As regards the actual network, VRP's reduce the working pressure
of the pipes and allow thinner (and consequently more economical) pipe classes to be
used. Reducing the pressure levels also reduces danger of leaks.
PRY's are unidirectional devices that keep downstream pressure at a pressure
setting value, even without any flow, provided upstream pressure is greater than this
value. Otherwise the downstream pressure will be equal to the upstream pressure minus
the head loss caused by the valve itself. The cost of the units increases both with the
difference in pressure between its ends, and with the flow rate, but the cost of the PRY
is just a part of the total cost (that includes filters, relief and air release valves, ... ), so that
the difference in cost between two sizes of PRY is partially absorbed into the total cost
of the installation.

Criteria for placement of PRV's

Although the arrangement of PRY's in an irrigation network is a functional
requirement, it has not been considered as liable to optimization. The only exception that
we know is due to Robinson and Austin [14]. The main difficulty of their model lies in
the assumption that the pressure setting of a PRY can influence the hydraulic
performance of the network, and consequently, the choice of optimal diameters. This
forces the authors to make certain approximations on the cost of the PRY in order to
incorporate it in the linear model.
46

In this paper such difficulty is overcome by making the hydraulic requirements (pipe
diameters) independent of the mechanical requirements (pipe pressure class). It is
possible to do this as long as the setting pressure of the PRY would be greater than the
dynamic pressure of the point where it is placed. In fact, a pressure setting below this
value would not be cost-justified.
In effect, let us imagine a pipe Pt with diameter Dt\ with end nodes N. (upstream)
and N4 (downstream). The pipe has been sized in such a way that dynamic pressure Pd4
in the downstream node is equal to the minimum service pressure Pmind> and the pipe
thickness has been chosen according to the hydrostatic pressure HPt. If we locate a PRY
in the upstream node N., with a pressure setting PS lesser than HPk, according to the
value of PS we can reduce pipe thickness obtaining a saving with the change, but when
the pressure setting drops below the dynamic pressure Pd., it would be necessary to
change the diameter from Dk1 to Dk2 (Dk1 <Dk2) in order to maintain the minimum
service pressure Pda. This change of diameter means an increase on the pipe cost, that
is greater than the saving obtained with the reduction of thickness (this is always true, as
we can see in any list of pipe prices).
This conclusion is of fundamental importance for the optimization process, since it
establishes the optimal pressure setting value for a given PRY as the dynamic pressure
at the node where it is located, so that the flow rate being also known, its cost is
determinate. As a consequence, the decision variables would be binary: to install or not
a given PRY.
A very common functional constraint in the sizing of networks is the limiting of flow
velocity in the pipes to a maximum allowable value (normally 2-3 m/s) to avoid material
erosion and to limit the overpressure produced by the water hammer. In practice this
limitation means a minimum acceptable diameter in each pipe. As a result of this
constraint, on an optimized network we can find that some nodes reach dynamic
pressures above the required minimum simply because some diameters have reached
their minimum value in the optimization process. Since the optimal pressure setting will
be the minimum possible value compatible with the pressure constraints without
interfering with the sizing of the network, in this case PS = Pd. - ~Pd, where Pd. is the
dynamic pressure of the node where the PRY is located and ~Pd is the minimum value
of the pressure slack (Pdi- Pmini) for all the nodes i located downstream from the PRY.
 47
OPTIMIZATION MEniOD

In order to develop the optimization method, we will first consider a series of pipes,
and then, its application to branched networks.

Series of pipes
Let us assume a system constituted by n pipes arranged in series, and supplied from
a source head H,.. The system has n+l nodes whose numeration grows from 0 (source
node), ton (downstream end node). Numeration of pipes will be referred with the same
index as its downstream nodes. The system variables are :
- Node N;: Demand q;, ground elevation Z;, dynamic pressure Pd; and minimum
service pressure Pmin; .
- Pipe Pit: Length 4:, flow rate Q.., diameter D.., maximum hydrostatic pressure HP..,
maximum allowable pipe pressure CP.., flow velocity Vlt and head loss h~r

The system has been sized for one operational state in compliance with a number of
constraints, such as:
-Minimum service pressure: Pd; ;?; Pmini for all the nodes (1)
-Velocity constraints: Vmink ~ Vk ~ Vmaxk for all the pipes (2)
-Mechanical resistance of pipe: HPk ~ CPk for all the pipes (3)
The aim is to find an optimal set of PRY's that allow us to obtain maximum savings by
decreasing pipe class without modifying the pipe diameters. We shall consider the
possibility of placing a PRV at each of the nodes of the system, without including the
source node or the downstream node, since they have no influence on the possible
savings in class of pipe. The pressure setting PSi of PRVi will be, according to the former
considerations:

PSi = Pd, - 11Pd,, where 11Pd, = ~~ (P~ - Pm~) (4)

j>l

The saving obtained for the kth pipe if PRVi were installed in isolation, si,k> is obtained
simply by calculating the difference in cost of the pipe when its pressure class is changed
as a result of the performance of t~is PRY. The total saving in pipes that will produce
PRVi in isolation would be S; = L si,k" As a general rule, we can say that si,lt:~si.k if
1
k>j>i, because of pressure settfrig of PRVi is less than or equal to that of PRVi
(PSi~PSi).
48

The decision to place a given PRY or not will be determined by the savings
involved, and this, in turn, will depend on the placement of the other PRY's because of
if two PRY are located, PRY; and PRYi with j>i, the saving on a downstream pipe Pk
would be si.k instead of s~k + sj,k·
A posible method for optimizing the set of PRY's is to consider all possible
combinations; in the case of a number of PRY equal to n, there would be 2°-1
possibilities, and with a relatively small number of valves, the problem may become
unfeasible. As an alternative, the optimization process can be adapted to the sequential
structure of Dynamic Programming problems [15], considering as stages of the problem
the nodes selected to place PRY's. The state variable at i1h stage (ilh node)~. can, like
the decision variable d;, be described in a binary form (to locate or not a PRY), since the
optimal pressure setting takes on a known value. In such a way, the objective function
F";(XJ in the ith stage would be:
~*(X;) = max [ F;:t(X;.t) + C(X;.d;) ] (5)
~
where C(~,d;) is the contribution of stage i to the objective function.
It must be noted that the decision d; coincides with the state Xi+t of the following
stage. Thus the transformation to be applied between states at successive stages is:
~+1 = t(X;.d;) = di (6)

The function C(~,d;) should be described using only the variables of the stage in
question and to do this it is necessary to "isolate" the saving produced by a given PRY
from the rest. In order to achieve that, we define the residual saving rsi,k in the pipe k
due to the PRV; as rsi.k = si.k-si-l,k·
Considering then the residual savings rather than the absolute savings, the
contribution of stage i to the objective function will be:

where cvi+ 1 is the cost of PRYi+l• L;+ 1 represent the set of pipes downstream of the node
i + 1, and Lm the set of pipes downstream of the first PRYm located downstream of the
node i + 1, according to the optimal decisions taken until this stage.
 49
The saving that is exclusively attributable to PRYi+l is the sum of the residual
savings in the pipes situated downstream from it, minus the cost of the valve. If we
assume that PRY; has not been installed, then the saving must yet be increased in an
amount equal to the sum of the residual savings corresponding to PRY; in the pipes
downstream of PRYi+l• or if PRYi+l is not installed, downstream of the first node mat
which a PRY is located. In order to consider the last saving, a suboptimization of the
problem at each stage must be carried out, from the last stage to the first, as it will be
shown in the example.

Branched ne~orks

When applying the optimization process to a branched network, the only difference
to consider is the treatment of the branches. In a node where two pipes {A and B) branch
out, the recursive equation that we must formulate is:
F;*(X) = mdax (F;_,i+l(XA,i+l) + q_x;,dA.;)] + rr~r [F~,i+l(XB,i+l) + C(X:.da)] (8)
A,i B,i

where F;'(.X;) represents the optimal values for each state of the objective function in the
i1h stage, now composed of the contribution of each branch. The transformation between
states will be:
XA,i+l = dA,i ; XB,i+l = dB,i· (9)
If we define the degree of a node (D) as the number of pipes it has downstream, in
the previous case all the nodes have D=l or D=O {for the end node). In branched
networks it will be nodes with D> 1; in such a kind of nodes we can dispose different
arrangements of PRY's, as is shown in Figure 2.

01
PIPE I -

Om
~--~~. .~~~~m-

On
PIPE n

(a) (b)

Figure 2- Possible location of PRY's at a node with degree D=3.

50

For Figure 2.a, we would obtain the optimal pressure setting for the PRY by
calculating the minimum pressure slack (.dPd) in the nodes situated downstream from
node v, so that PSv=Pd,;.dPd. If instead of installing one PRY we install three, each one
heading pipes I, m and n, as in Figure 2.b, the saving that we can obtain will be different
for two reasons:
a) The total cost of PRY1, PRYm and PRY0 will be greater than that of PRYv due
to the fixed costs of installation (cost increasing factor).
b) The saving in pipes produced by the three valves will be greater than or equal
to that produced by PRY"' because the pressure setting PSv is equal to the
maximum of the pressure settings PS1, PSm and PS0 (cost decreasing factor).
Thus we cannot generally affirm that one situation is clearly better than another, and so,
we must consider all the PRY's presented in Figs. 2.a and 2.b in the optimization process.

RESULTS

Series of pipes
As an application of the method, we will first optimize the set of PRY's for a system
constituted by a series of pipes (see Fig. 4). The system is supplied by a reservoir with
an elevation head of 150 m. and the maximum permissible flow velocity is 2 m/s. The
sizing has been carried out with the program DIOPRAM, developed by our team for the
optimization of branched networks by Linear Programming. The working of the program
is described in [16] and its rationale in [17]. The aim of installing PRY's is to obtain
maximum savings by decreasing pipe class, without modifying the diameters of the
network or violating minimum service pressure head of 25 m. for all demand nodes.
The results of the optimized system are shown in Table 1. Column (5) indicates the
pipe class (max. allowable pressure), column (7) the maximum hydrostatic pressure and
column (8) the dynamic pressure at the downstream nodes. Table 2 shows the cost of
pipe (per metre), depending on diameter and pipe class.
 51

TABLE 1
Results for the optimized system
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Max. Allow. Head loss Max. Stati' Downstream node

Pipe Length Flow Diam. Pres. Head (m) Pressure Cost
(m) Rate (mm) (Class) Head (m) Ground Dynamic ($)
(1/s) (m) Elev. Pres. Head
(m) (m)
1 800 90 250 75 (C) 9.09 55 95 45.9 32,624
2 650 80 250 75 (C) 5.92 70 80 55.0 26,507
3 350 70 250 100 (D) 2.49 85 65 67.5 16,397
4 500 60 200 100 (D) 8.01 95 55 69.5 16,880
5 700 50 200 100 (D) 7.98 100 50 66.5 23,632
6 650 40 175 125 (E) 9.44 110 40 67.1 22,665
7 350 30 150 125(E) 6.37 120 30 70.7 8,935
8 450 20 125 125 (E) 9.46 125 25 66.2 8,284
9 200 10 100 150 (F) 3.49 130 20 67.8 2,616
Total 158,540

TABLE2
Cost of pipe ($/m.)
Max. allowable ressure head In m. (Class of pipe)
Diam 25 so 75 100 125 150
(mm) (A) (B) (C) (D) (E) (F)
80 --- --- --- 8.0 --- 9.7
100 --- --- 8.3 9.8 --- 11.9
125 --- --- 11.8 14.6 ---- 18.4
150 ---- 14.1 17.4 20.3 ---- 25.5
175 ---- 18.3 23.7 27.5 ---- 34.9
200 21.1 23.1 28.9 33.8 37.6 43.1
250 26.0 32.7 40.8 46.9 51.6 58.9
300 38.5 44.7 54.7 62.4 66.4 72.7
350 44.0 54.4 65.3 76.1 83.8 91.5
400 57.1 72.4 81.5 90.2 99.3 108.3
450 76.3 86.3 98.1 105.5 117.2 121.1

We shall first consider the possibility of placing a PRV at each of the nodes of the
system, without including the head node or the end node, since they have no influence
on the possible savings in class of pipe, as has been is explained.
Firstly the optimal pressure setting for each PRV is calculated, bearing in mind the
dynamic pressures and pressure slacks of the system. Table 3 shows the pressure setting
of the PRV, the flow rate, the valve cost, and then, the absolute and residual saving (in
bold face) obtained for each of the pipes.
52

TABLE3
Absolute and residual savings obtained by placing a PRY at system nodes ($).
PIPES
PRV Pres. Flow Valve P2 P3 P4 P5 P6 P7 P8 P9
Setting Rate Cost
(m.) (1/s) ($)
1 15.9 80 3000 5280 4970 2460 3440 7750 1810 1690 460
5280 4970 2460 3440 7750 1810 1690 460
2 13.7 70 3000 4970 5370 7510 7750 2850 2950 800
0 2910 4070 0 1040 1260 340
3 26.3 60 2500 5370 7510 7750 2850 2950 800
0 0 0 0 0 0
4 28.2 50 2500 7510 10800 2850 2950 800
0 3050 0 0 0
5 25.3 40 2500 10800 4050 2950 800
0 1200 0 0
6 25.8 30 2000 4050 3880 800
0 930 0
7 29.5 20 2000 3880 800
0 0
8 23.5 10 1800 800
0

The maximum saving that a PRY can produce occurs when it is placed in isolation;
thus all those PRY's which, if placed in isolation, produce a total saving that is less than
their own cost will never be chosen in the optimization process; to avoid unnecessary
calculations we do not take them into account. In the present example, that happens only
with PRY8 •
Figure 3 shows the optimization process in three of its phases, where a
suboptimization is applied at each stage in order to calculate the values of C(Xi,d),
labelled on the arrows.
The subset of PRY's that results in maximum saving is that formed by valves 1, 3,
and 5, with the net total saving being 33,730 $ (41,730 $saving in pipes, less 8,000 $,the
cost of the valves). Figure 4 shows the final pressures in the system once the optimal
PRY's have been installed.
 53

0 Stage / Decision
~ I Volvel Stotel /Optimal
F~ Value($) Decision

0 ® ® 0 0 0 CD @)

(a) D
L.,.
Direction of
optimization process

0 ® ® 0 0 0 CD @)

(b) D
L.,.

0 ® ® 0 0 0 CD @)

(c)

Figure 3 - Optimization process in three of its phases.

Elevation (m.)

100

75

so
nMin. ~
UPres. Node
25 Static head
·---------(before PRV)
(after PRV)

0 500 1000 1500 2000 2500 3000 3500 4000 4500

Distance to reservoir (m.)
Figure 4 - Final pressure state of the system.
54

Branchedne~ork

We shall now apply the extended method to optimize the location and setting of the
PRY's on the branched network shown in Figure 5, whose optimal sizing and hydraulic
parameters are presented in Table 4). The columns of the Table 4 contain, among others,
the following data: (5) class of pipe, (6) head loss, (7) maximum hydrostatic pressure, (8)
ground elevation of downstream node and {10) cost of pipe.

TABLE4
Results for the optimized branched network.
(1) (2) (3) (4) (S) (6) (7) (8) (9) (10)
Pipe Length Flow Diam. Max. Head Max. Cost
(m) Rate (mm) Allow. Loss Stat. Downstream ($)
(1/s) Pres. (m) Pres Node
(m) (m) Ground Dyn. Pres.
Class Elev. Head
(m) (m)
1 650 252 450 50 2,53 37 120 34,5 56,121
2 420 252 450 75 1,64 74 83 69,8 41,181
3 312 35 150 125 6,88 101 56 89,9 7,%5
4 401 25 150 125 4,72 124 33 108 10,237
5 563 202 400 100 2,59 89 68 82,2 50,760
6 427 76 250 125 3,22 112 45 102 22,008
7 363 42 175 125 5,25 120 37 105 12,658
8 400 30 150 150 6,61 129 28 107 10,212
9 920 70 250 100 5,94 95 62 82,3 43,102
10 530 30 150 100 8,76 97 60 75,5 10,796
11 412 10 80 100 19,7 97 62 53,9 3,304
Total 268,344

The network is supplied from a reservoir with an elevation head of 157 m., and has
been sized using the program DIOPRAM considering a maximum flow velocity of 2 m/s
and a minimum service pressure head at the demand nodes of 25 m. For the calculation,
the same costs per unit of pipe as in the previous example {shown in Table 2) have been
used.
The nodes 2 and 5 are the beginning of two branches, so that we must consider tree
possible PRY's at each of the nodes. The nomenclature we have followed is:
PRY; - A valve affecting both branches of the node i.
PRYia- A valve heading the branch with lower pressure slack (higher setting
pressure).
PRY;b- A valve heading the branch with higher pressure slack {lower setting
pressure).
 55

@) NODE
- DEMAND (1/s)
)-19- ELEVATION (m)

cfiJo PIPE LENGTH (m) 412m

650m

20

40
427m
312m
56
363m 37
~
34 7
400m

12
25

Figure 5 -Sketch of the branched network and location of optimal PRY's.

TABLE 5
Absolute and residual savings in pipes ($)due to placement of PRY's.

(1) (2) (3) (4) (5) PIPES

PRY Upstream PS Flow Valve P2 P3 P4 P5 P6 P7 P8 P9 PlO Pll
PRY (m) Rate Cost
<i7~) ($)
l 5.6 252 3,600 4918 2537 2069 4853 2003 2697 2064 5584 1574 0
4918 2537 2069 4853 2003 2697 2064 5584 1574 0
2 1 40.9 237 3,000 2537 2069 4853 2003 2697 2064 5584 1574 0
0 0 0 0 0 0 0 0 0
2a 2 40.9 202 3,000 4853 2003 2697 2064 5584 1574 0
0 0 0 0 0 0 0
2b 2 4.9 35 2,100 3607 3260
1070 1191
3 2b 6.7 25 1,800 4636
····················································································
....................... 1376 ···········
5 2a 53.4 146 2,600 2003 2697 2064 5584 1574 0
...................................•........... 0 0 0 0 0 0
Sa 5 53.4 70 1,900 5584 1574 0
............................................... 0 0 0
5b 5 5.2 76 2,800 8066 6029 4624
............................................... 6063 3332 2560 ....................................
6 5b 22.3 42 2,450 6029 4624
0 0
7 6 22.6 30 1,800
···············································-··········· 4624 ···········•···········•············
·······································································c..____!! ...........
9 Sa 53.4 30 1,400 1574 0
oooouooooo•lloouooooo•ououooou•oouo.,oooo•ooooooooooo•uoouoouo•••••••••••••••uououo
0 0
10 9 ...· 46,~ ., .. 10 •1,250 0
,,.i, I 0
56

At the sight of Table 5, we can eliminate the following PRY's from the optimization
process:
- PRV10 is not considered as its cost is greater than the saving it can result in.
- PRV2a is eliminated as its cost is equal to PRV2 and they have the same optimal
pressure setting.
The optimal situation consists in installing PRV1, PRV3 and PRVSb (see Fig. 5) and results
in a net saving of 34,621 $ (a saving of 42,821 $ in pipes less 8,200 $, the cost of PRY's).

CONCLUSIONS

We have described a method that allows important savings to be obtained in irrigation

networks through an efficient use of PRY's. The basis of the procedure is the conceptual
distinction between hydraulic and mechanical requirements. We can underline the
following major advantages.
-The optimization of the set of PRY's is carried out independently of the hydraulic
sizing of the network (diameters).
-The pressure setting and placing of the PRY's are two separate processes so that
the designer can choose any value for the pressure settings different from those
proposed.
-The procedure provides an optimal solution starting from any previous situation,
that is, the designer can impose any PRV (situation and setting) and optimize the
remaining set, without modifying the procedure.
-The sequential formulation of the problem requires a minimum storage space in
computer RAM for processing variables.

REFERENCES

1. Walters, G.A, Optimal Design of Pipe Networks: A Review. Invited paper in the
1'1 International Conference on Computer Methods and Water Resources, 14-18
March 1988, Rabat (Morocco).

2. Karmeli, D., Gadish, Y. Meyers, S., Design of Optimal Water Distribution

Networks. J. of the Pipeline Div. (ASCE), 1968, Vol. 94, PLl, pp. 1-10.
 57

3. Bhave, P.R., Selecting Pipe Sizes in Network Optimization by Linear Programming.

J. of the Hydraulics Div. (ASCE), 1979, Vol. 105, HY7, pp. 1019-25.

4. Morgan, D.R., Goulter, I.C., Optimal Urban Water Distribution Design. Water
Resources Research, Vol. 21, 5, pp. 642-52.

5. Martinez, F., Sanz, F., Garcfa-Serra, J., Cerrillo, J.L., Dimensionado Optimo de
Redes Ramificadas de Distribuci6n de Agua por Programaci6n Lineal. Tecnologia
del Agua, 1987, Num. 40, pp. 73-90.

6. Jacoby, S.L.S., Design of Optimal Hydraulic Networks. J. of the Hydraulics Div.

(ASCE), 1968, Vol. 94, HY3, pp. 641-61.

7. Cembrowicz, R.G., Harrington, J.J., Capital-Cost Minimization of Hydraulic

Networks. J. of the Hydraulics Div. (ASCE), 1973, Vol. 99, HY3, pp. 431-40.

8. Chiplunkar, A V., Khana, P., Optimal Design of Branched Water Supply Networks.
J. of Environmental Engineering, 1983, Vol. 109, 3, pp. 604-18.

9. Fujiwara, 0., Dey, D., Method for Optimal Design of Branched Networks on Flat
Terrain. J. of Environmental Engineering, 1988, Vol. 114, 6, pp. 1464-75.

10. Liang, T., Design Conduit System by Dynamic Programming. J. of the Hydraulics
Div. (ASCE), 1971, Vol. 97, HY3, pp. 383-93.

11. Yang, K., Liang, T., Wu, J., Design of Conduit System with Diverging Branches. J.
of the Hydraulics Div. (ASCE), 1975, Vol. 101, HY1, pp. 167-188.

12. Goldberg, D.E., Kuo, C.H., Genetic Algorithms in Pipeline Optimization. J. of

Computing in Civil Engineering, 1987, Vol. 1, 2, pp. 128-41.

13. Mateos, M., Valvulas para Abastecimientos de agu!!, Ed. Bellisco, Madrid 1990.

14. Robinson, R.B. and Austin, T.A, Cost Optimization of rural water systems. J. of the
Hydraulics Div. (ASCE), 1976, Vol. 102, HY8, pp. 1119-34.

15. Orth, H.M., Model-based Design of Water Distribution and Sewage Systems, John
Wiley & Sons, Great Britain, 1986.

16. Fluid Mechanics Group, DIOPRAM (Optimal Sizing of Branched Networks). User
Manual. Politechnical University of Valencia, 1989.

17. Martinez, F., Perez, R., Sanz, F., Programa de Calculo para el Dimensionado
Econ6mico de las Arterias Principales y Altura de Cabecera en Redes de Riego
Localizado. Riegos y Drenajes, 1990, 47, pp. 9-19.
 WATER SUPPLY SYSTEMS OPTIMIZATION FOR
DEVELOPING COUNTRIES

RALF G. CEMBROWICZ
Institut fiir Wasserbau und Kulturtechnik
Universitii.t Karlsruhe
Postfach 6980, 7500 Karlsruhe, F.R. Germany

ABSTRACT

With respect to water supply systems optimization, mathematical models of a vast

variety are available in the literature. It seems this had limited impact upon solving
the problem in practice. Sponsored by the German Science Foundation an attempt
was made for remedy. Cost optimal design models of urban and regional water
supply systems were devised for developing countries where they are most needed.
The programs were implemented and verified in planning situations. They include
both layout and design optimization. Different algorithms had to be employed in
conjunction, including graph theory, linear programming and a 'genetic' random search.
As a result, in the opinion of the author, water supply pipe line systems optimization
has been made available for practical application.

1 INTRODUCTION

Water supply systems are a vital part of the technical infrastructure. For numerous
expanding cities and rural communities of developing countries, though, water supply
is inadequate. It is estimated that most of the world population living in cities suffers
from insufficient supply [11]. The provision of safe and adequate drinking water is a
prerequisite for economic development, accruing health and welfare benefits are well-
known. However, the associated cost can be preventive. The United Nations Water
Supply and Sanitation Decade (1980-90) failed to improve the situation significantly.

59
60

Hence, there is need for planning methods that have the capacity for cost optimal
design taking into consideration local design criteria. The methods are also to facilitate
sensitivity analyses with respect to uncertain input data and parameter estimates. In
practice, trial and error methods are still employed for the layout and the design of
water supply networks without guarantee for economic and operational efficiency.
The following systems analysis approach uses graph theory, linear programming
and a 'genetic' algorithm to determine cost optimal layouts and designs of water supply
systems. The choice of the methods ensures numerical stability and allows for treatment
of large systems without resorting to approximations. The objective is to minimize
total investment and pumping cost. Arbitrary demand conditions, supply and pressure
requirements, standard diameters, maximum admissible velocities etc can be inposed
or are observed as constraints.
In general, there is the distinction between urban and regional water supply. This
will also be subsequently maintained.

2 URBAN SYSTEMS

2.1 Definitions

An urban water supply system is typically designed as a grid network or reticulate

system in accordance with the existing street pattern of the city. Its abstraction is a
graph whose nodes are the intersections of the lines .. The assumption that the water
demand is exerted at the nodes is an engineering convention, any desired degree of
exactness can be obtained by introducing intermediate nodes. A water supply network
must not necessarily consist of a grid containing 'loops'. In fact, any given set of
nodes can be connected by a branch system without any circuits or loops. This holds
for a sewer system as well as for water supply, where the reticulation of a grid is
chosen for safety to allow supply from alternative sides. In case of economic planning
or urgent remedy, a branch network may suffice. For example, with respect to the
optimal layout and design of an artery system of mains in developing countries, the
mains may be supplemented, even at a later point, by a 'secondary network' of reticulate
subdistribution consisting of desirable minimum diameters whose size is not subject to
 61

optimization.
The optimization concept must include the Hydraulic Balance. For the analysis it
proves convenient to denote the nodes by i = 1, ... , N;, the lines by j = 1, ... , N;, and
the loops by l = 1, ... , Nz and to introduce the following vectors:

q = {q;}, water demand or inflow h = {hj}, head loss

p = {p;}, pressure potential f = {!;}, flow

The graph of the network consisting of N; nodes and of N; lines is numerically

summarized by an incidence matrix A= [a;j], with AT= [aj;], where
a;j = +1 , if branch j leads to node i,
a;i = -1 , if branch j leads away from node i,
a;i = 0 , if branch j is not incident upon node i.

2.2 Hydraulic Balance

The Hydraulic Balance is described by a set of three simultaneous systems equations.

First, Kirchhoff's mass balance 'node law' stating that the inflow equals the outflow at
any node:
Af=q (1)

One redundant node equation being omitted (since the additional condition that
the total inflow equals the total demand is automatically fulfilled) the size of A in {1)
is (N;- 1) x (N;- 1). Second, Kirchhoff's energy balance 'loop law' follows the head
losses around:

j E Jz, l = 1, ... ,Nz, (2)

with ]z designating the set of branches contained in any loop.

The third set of equations are the system's functions relating flow f;, head loss h;,
diameter dj and length l; of a pressure conduit:

!~
h·-
J -
,..,.)._._L[.
I] J d~ J (3)
3
62

The number /j being a dimension factor, the friction characteristic:>..; representing

the material of the conduit and assumed to be a constant according to Hazen-Williams
(6 = 1.85,e = 4.85) or Darcy-Weisbach (6 = 2.0,e = 5.0). If friction characteristic,
diameter and flow are known, the hydraulic equation (3) reduces to the familiar linear
relationship between head loss h; and pipe length l;, with b; being a constant:

h; = b;l; (4)

The simultaneous solution of the linear equations (1 ), (2) and the nonlinear relation
(3) is called the Hydraulic Balance. The diameters, the water demand and inflow, the
pipe material (i.e. friction characteristics) are known. The 2 N; unknows elements of
flow f; and head loss h; are to be determined. The available number of equations in
(1), (2) and (3) equals the number of unknowns, 2 N;, according to Euler's theorem
[10]:
(N; -1) + N, + N; = 2 N; (5)

There is a unique solution to (1), (2), (3) corresponding to a system's minimum of

energy losses (1]. Because of the nonlinearity of (3) an explicit numerical solution does
not apply. The solution techniques are iterative, e.g. Hardy-Cross, Newton-Raphson
a.o. (4].

2.3 Graph Theory

The matrix A describing a reticulate network may be partitioned into an (N;- 1) x

(N;- 1) square matrix E and into an (N;- 1) x (NI) matrix G:

A=(EIG] (6)

E represents a tree if the determinant lEI= ±1, and E is not a tree if lEI= 0. A
tree E is a branch network connecting the entire set of nodes of the reticulate network
A without forming loops. Hence, the number of lines in a tree is (N;- 1). The total
number of trees contained in a reticulate system is (8]:

(7)

G is the associated co tree. The number of lines contained in the cotree equals the
number of loops of the reticulation system. The lines of the tree are called branche,,
 63

those of the cotree are called chords. Trees can be generated by allocating the position
of one chord in any loop. This concept will be used in the 'Genetic Algorithm' of
section 2.6. Combining equations (1) and (6) and partitioning f in accordance with
equation (6) into fl (flows in the branches of the tree), and !2 (flows in the chords of
the ·cotree), such that

it follows that:
(8)

Equation (8) states that the flows in a tree fl meeting given demand requirements
and inflow conditions q, are a linear function of the flows f2 in the chords. For example,
in a branch network or tree the flows in the chords are set zero. Since the chords
supplement any tree to become the original reticulate graph, any loop has one degree
of freedom to set a flow at random.

2.4 Nonlinear Optimization

The capital cost of a water supply pipe-line j, Ci including instalment, are generally
given as functions of the diameters. The unit cost c'j by production and instalment
refer to a diameter per unit length, hence, for a line j:

(9)

The exponent {3 is encountered in the order of 1 ~ {3 ~ 1.5. Eliminating the

diameters in (9) using equation (3) and taking the order of magnitude indicated above
of {3, 8, e into account, it follows that the cost are concave with respect to the flows
which is equivalent to stating 'economies of scale' with respect to the flows:

J {36/e
""ci(d·)- ""a (h· f·)- ""'c~l·("V·>.·l·)f3fc:!..i__ (10)
L c J - L c " J - L 1 ' '' ' ' h .f3fe
j j j J

Consequently, it is most economical for the design of a network to concentrate

the flows in as few mains as much as possible. The result is a tree. A tree or branch
network can be viewed as a reticulate system with zero flows in the chords, forcing an
utmost flow concentration into the remaining branches.
64

Mathematically, any tree of a reticulate system represents a local minimum of

the system's cost [3]. For a tree, the flows in (10) are given by (8), the corresponding
unknown head losses in (10) can be determined using Linear Programming (section 2.5).
The astronomical number of trees contained in a reticulate network according to
equation (7) usually prevents the total enumeration of all the trees, any of them
representing a local minimum. However, by engineering judgement computational
abbreviations are possible. For example, in a specific city some of the lines are always
chosen to be mains (branches), others are considered secondary (chords) according
to their function and location in the city. Thus the total number of trees T can be
somewhat reduced by an a priori branch and chord allocation. Still, on principal the
scanning of a space encompassing an enormous number of trees is required. In section
2.6 a Genetic Algorithm representing a 'systematic random scanning' is introduced to
facilitate the search for the global solution.
To formulate the complete nonlinear network optimization the objective function
(10) must be supplemented by the previous system's equations (1) to (3), and in
addition by inequalities to secure minimum pressure P at any node:

p?. p (11)

2.5 Linear Programming

Concerning the optimization of a branch or tree network the following method has
been adopted [6]. In a branch network the flows are known according to (8). The head
losses remain unknown in the objective function (10). Consider a line j oflength L;
consisting of k = 1, ... , K sections of standard size diameters d;1e in increasing order
k = 1, ... , K whose lengths l;1e are unknown. An appropriate choice of the smallest
and largest diameter admitted in k serves to control the largest and smallest admissible
velocity in a particular line. Also, a small range of k will reduce the number of variables,
l;le·
By this model, in essence the unknown head losses in the nonlinear objective
function (10) have been substituted by new variables l;1c introducing at the same time
standard size diameters.
In addition efficient Linear Programming will be employed. Since the capital cost
including instalment, ck, of a conduit are given per unit length, the total capital cost
 65

of a line j consisting of sections k amount to:

Cc(l) =L ckljk (12}

"
For an entire branch network consisting of branches j a Linear Program, which is
equivalent to the preceding nonlinear model of section 2.4, can be formulated as follows:

(13)

subject to Llik = Lj, Vj {14}

k

L L b;~ol;lc = o, j E J, vl (15)
i k
LLb;~el;~e ~H., j E J., Ve (16}
k

The constraint (1) is omitted since the flows are known in a branch network.
However, the self-explaining new condition (14) is now included. Constraint (15)
corresponds to previously (2), and constraint (16) is a convenient succint equivalent
to (11}: J. denotes the sequence of branches j leading from a defined input node to
an end node e of a tree. The maximum admissible head loss He in constraint (16)
is the difference between the given input node potential and the minimum potential
requirement at any end node. Equal pressure requirements of intermediate nodes are
automatically met. It can be shown that the optimal solution of the Linear Program
(13}-(16) may contain at most two diameters for any line (6]. Their sizes must be
adjacent.
The numerical advantages of the Linear Program (compared to the nonlinear
formulation) as to possible problem size, computation time, convergence etc need
not be emphasized. Also, pumping cost Cp(l) can be easily accommodated by the
objective function (13) since pumping energy cost are proportional to the product of
flow and head loss (f h) yielding, again, linear terms with respect to the section lengths
l;,., since, as stated above, the flows f are known in a branch network and the head
losses hare reduced to linear functions of l;1e as mentionned above by (4):
66
The tree, whose optimal pipe size diameters are now determined, is subsequently
supplemented by the original chords (choosing for them the smallest desirable or
admissible diameter) to recreate the graph of the original reticulate system. It remains
to evaluate the trees of the original graph, any of them constituting a local minimum,
in order to find the global. This is achieved by the following Genetic Algorithm.

2.6 Genetic Algorithm

The properties of a species are defined in nature by its chromosomes, consisting, in turn,
of genes. Any gene of a mammal is composed of"' 1 000 positions for a nucleotide base,
four of which occur. Hence, there are ~ 4 1 000 possibilities, theoretically, to 'construct'
a gene. The genes of a chromosome are arranged in a double helix called DNA as
discovered by Watson and Crick (9].
A chromosome set of a mammal consists of"' 3 x 10 6 genes. It follows that there are
"'43000000000 possibilities to 'build' synthetically a chromosome of a mammal. Five
main mechanisms of biological evolution are considered responsible to short-cut the
total enumeration of these possibilities, and to lead to the existing complexity of living
beings in a relative short period of time in the history of our planet: Gene mutation ,
chromosome mutation, recombination, selection, isolation.
Chromosome mutation and isolation will not be further considered here for
simplicity. Gene mutation refers to the variation of one or more genes by changing
the nucleotide base at several points of the DNA helix thus generating a new genotype.
Recombination (or mating) is responsible for the mixing and reinforcing of the heritage
of an existing population. One of the mechanisms of recombination is that the
chromosomes of the father and of the mother unite by random combinations of
chromosome sections. Gene mutation generally leads to hardly perceptible changes
of the properties of an individual compared to chromosome mutation. But also,
with low probability, gene mutation and recombination may sometimes cause relative
profound variations of the DNA chain, hence producing significant changes in the
ability of an individual. Rare successful results by superimposing changes may further
abbreviate the evolutionary process. Driven by the mechanisms of biological evolution
the population moves into the direction of increasing 'resilience', 'ability', 'economy',
etc.
 67

The comparison to trial and error steps imposed on 'technical species' (tools,
engines, cars) over generations of their technological development while searching for
the best 'performance', 'efficiency', 'structure' etc is obvious (7). For example, the cost
of a branch network solely depends on the topology of the graph. The transfer of the
previous principles to the search of finding trees of decreasing cost focusses on improving
the allocation of chords to different trees. In any computational 'generation', branch
networks will be generated by 'mutating' and 'recombining' chords (representing in a
biological sense a gene or a chromosome) as follows (5):
1. Choose randomly a population of n trees and optimize their network cost using
Linear Programming.
2. Choose randomly two tree networks (parents) from the population of n trees.
3. 'Mutate' both trees (see below).
4. 'Recombine' (unify) both trees generating a new tree (see below).
5. Add the resulting tree to the existing population, optimize the cost of the new
tree.
6. Delete the tree with the highest cost of the population of (n + 1) trees.
7. Go to 2, start a new generation.

Steps 3. and 4. contain in addition the following details:

3.1 Select randomly the number of loops in the network to impute changes.
3.2 Choose randomly the loops where changes shall occur.
3.3 Introduce randomly new chords in the designated loops.

4.1 Select randomly one of the parent trees.

4.2 For any loop of the selected tree: The existing chord is replaced by the chord
of the other parent tree with probability p.
The probability parameter p is to be derived from a suitable probability distribu-
tion in order to control desirable smaller or larger variations of the topology of the tree
and of the associated cost. This, usually, requires some experimental computational
work. The most efficient distribution proved to be the 'normal'. An increase of the rate
of convergence to a global minimum was observed by likewise applying the mutation-
selection mechanisms to the parameters of the probability distribution, again, in an
attempt to simulate nature. As the algorithm proceeds, rapid convergence towards
68

LEGEND
0 ~HNO.
131
@ ELEVATION M a.s.l.
F4Do OPTIMAL DIAMETERimm]- DAY/NIGHT OPERATION
1F>OOI OPTIMAL DIAMETERimml- PEAK HOUR

VERS LAC TANGANYIKA

Flg.1-0ptimal system Bujumbura,Zone 1-South

the global solution is usually encountered while the cost decrease from generation to
generation (Fig. 3).

2.7 Planning Example [2]

Fig. 1 displays the optimal design for part of the city of Bujumbura, capital of
Burundi, comprising about 120 000 inhabitants and a military camp. The graph has
34 nodes, 48 lines and 15 loops. The total number of trees is T = 711.83 x 10 6 •
The peak load at day time of 233.3 1/s is covered by a pumping station (node 174)
and simultaneous feeding of the network from an elevated reservoir (node 173). The
reservoir is replenished by the pumping station during the night. The pumping station
is supplied from a central treatment plant taking raw water from Lake Tanganyika.
Small diameters were required of PVC material (roughness k = 0.4 mm), larger
diameters of ductile iron (F, k = 3 mm), existing steel pipes (A, k = 1 mm) had to be
maintained. The investment cost including instalment were as follows (DM - Deutsche
 69

Mark- German currency unit, ND- nominal diameter in mm):

PVC F
ND 100 150 200 300 400 500 600
DM/m 170 210 250 380 500 700 800

The World Bank financing the scheme fixed the interest rate at 2% over a
planning horizon of 30 years implying a present value factor of 22.39. The loading
factor for pumping was assumed to 0.193, pumping efficiency to 0.7, energy cost to
0.1434 DM/kWh. Minimum supply pressure at any node was 20m (except node 171:
10m), flow velocities were limited to be smaller than 2 m/s. The minimum diameters
to be added as chord8 were chosen to be 100 mm.
The cost ofthe optimal design amounted to 7.33 X 106 DM, 6.12 X 106 DM of which
for investment (pipes including instalment) and 1.21 x 10 6 DM for operation (pumping
including equipment). By chance, an 'optimal' engineering design was developed
independently by an expert consultant using traditional methods and resulting in
8.06 x 10 6 DM (6.86 x 10 6 DM investment, 1.20 x 10 6 DM operation). The savings by
rigorous optimization were 9.9% of the total cost, 12% with respect to the investment.
However, the main artery between pumping station and reservoir was not amenable to
optimization since the pipe sizes derived directly from the flow requirement to replenish
the reservoir during the night. The optimal diameters derived for the day-time peak
load had to be adjusted accordingly. Disregarding the cost for this main artery, the
cost reduction achieved by optimization increased to 21%.

3 REGIONAL SYSTEMS

3.1 Optimization Model

The regional model is a special case of the urban one. Two differences may be stated.
First, a regional system consisting solely of arteries is a priori a branch network or tree.
Second, there is pressure discontinuity to be assumed at any node since pumping or
boosting may occur. These two conditions can be used to abbreviate the optimization.
70

Since the flows are known in a tree the problem reduces to determining the optimal
pumping pressure for any line j of the system. This is achieved by evaluating the
optimal trade-off between capital cjr. and pumping Sr. cost for any available diameter
k:
min { ... ,(cj1c + S~c), ... } , Vk (17)

The minimum value of (17) with respect to k yields the optimal diameter kj for
any line j taking into account the existing geodetic elevation between the end nodes
of line j. Given the flow and the optimal diameter, the corresponding pumping or
booster head is determined by (3). The optimization problem reduces to the simple
computation described by (17) for any tree. The optimal tree is obtained employing,
again, the Genetic Algorithm of chapter 2.6.
Often, a third property of the regional model is encountered differing from the
urban situation: Though the demand is given at any demand node (village, town,
industry, etc) and the resulting total demand must be supplied from the source nodes
(reservoir, treatment plant, wellfield, etc), the optimal yield from any individual source,
which usually has a capacity constraint, is not known and must be determined in
addition. This is conveniently achieved by performing inflow variations simultaneously
with the tree search of the Genetic Algorithm. Using as a ranking criterion the
production cost of the water at the available sources, the allocation of the water to
be produced at any source is marginally varied at random while generating another
tree. Thus the step-wise improvement at any generation towards the global optimum
implies simultaneously the optimal allocation of the. total demand with respect to the
available sources.
As a remark, the algorithm allows also to consider two planning phases in a given
time horizon (as in the subsequent planning example) yielding the optimal capacity
expansion for the second phase.

3.2 Planning Example

Fig. 2 displays the graph of the planned Rajasthan Regional Water Supply System
in India. It contains 44 nodes, 54 lines and 11 loops. The total number of trees is
T = 29.43 x 106 • The project area covers about 15 000 km2 of the three districts Sri
Gaganagar, Churu and Jhunjhunun. The area is ranked semi-desert with corresponding
 71

D SOURCE

@ ELEVATION M a.s.L

15.0 LENGTH KM

NO SCALE

Figure 2. Rajasthan System Graph

150

A
X Minimum
0 Mean
140
Maximum 1 -
Population Size: 20

~~ 130

.2:
120

J:
a..
0 \\
\\
~ 110
0

.1"A-
"'-._""-.
100

x-x'""- -o-~:::::$
x-r-x-x-x E:~-~-~= ~==~=~ ~ib-llll- 1-A-A-&-

20 30 50 60
Generation

Figure 3. Convergence of the minimal cost design for phase 1

72

water scarcity and har~h climatic conditions. The north-east is partially supplied by
the existing Gandelhi-Sahawa Scheme limited to 1.2 m 3 /s, most of the region relies
presently on saline groundwater. Mean annual rainfall varies statistically between 250
to 440 mm, but there may be no rain in several years. The only realiable source is
the Indira Gandhi Canal conveying about 600 m 3 /s from the Punjab. The project
area extends about 150 km south of the canal rising to about 150 m above the canal
elevation.

The water demand was estimated for the future population of 956 villages and
10 towns in the project area comprising a total of 2.16 million people in Phase 1 (year
2002) and 2.60 million in Phase 2 (year 2012), based on 35 litre/( capita day) -1/(cd)
rural and 70 1/( cd) urban water demand in Phase 1, respectively, 40 1/( cd) rural and
90 1/( cd) urban in Phase 2. In addition, 'cattle units' of 30 1/d were considered covering
a projection of cattle, camels, buffalos, goats. The total water demand was estimated
to 133 500m3 /d and 184 700m3 /d by the year 2002 and 2012, respectively.

For the design various constraints had to be taken into account. For example,
the construction of a distributary canal (Sahawa Lift Canal Project - SLCP) into the
centre of the project area was to provide a source point by the year 2002 since the
existing Gandelhi-Sahawa-Scheme would stay limited to 1.2 mfs. Also, pumping was
to be confined to 16 h/d due to irregular power supply. Maximum flow velocities in
the pipe lines was set at 1.0 m/s for water hammer protection, maximum head losses
in the lines were not to exceed 50 m. Prestressed concrete pipes (PSC 80, test pressure
80 m) were to be used at the following cost including instalment and appurtenances
(Rs- Indian Rupees):

Cost of PSC 80 pipes (roughness k = 0.1mm)

ND 300 400 500 600 700 800 900 11000 1100 1200 1300 1400 1500
Rs/m 725 876 1023 1344 1475 1744 2139 2439 2635 3143 3952 4348 4983

A cost function of the treatment plants (alum coagulation/rapid sand filtration)

was developed based on the following data:
 73

Cost of Treatment Plant

Cost, y [ 10 6 Rs] 27.5 37.1 40.5 57.5 96.5
Capacity, :c [ ms/h] 1600 2300 3200 4300 7490
Cost function y = 104 + 11.5 :c (10 3 Rs/(m3 /h))

In addition, local design criteria included:

- available service pressure: 30 m
- capital cost of pump power: 3 000.0 Rs/Hp, with Hp - 'Horse Power'
- efficiency of pump + motor: 65%
- energy cost: 1.15 Rs/kWh (regular tariff)
- annual effective growth rate of water demand 2.4% in both phases
- beginning of the system operation
phase 1: 1998
phase 2: 2003
- the operation time of phase 1 is 5 years and the growth rate of water demand is
12.6% over the period of 5 years
- the operation time of phase 2 is 10 years and the growth rate of water demand is
26.8% over the period of 10 years
- economic lifetime of pipes and pumps: 30 years
- raw water reserves at the treatment plants are available to meet the water demand
during the cleaning periods of the Indira Gandhi Canal
- pumping stations at existing pipelines are maintained
- economic time horizon: 30 years, interest rate: 5%, present value: PV = 15.372
The optimal designs for phase 1 and and the expansion to phase 2 are displayed in
Fig. 4. Isolated nodes in the plots are without demand, their purpose being to include
any possible routing of the mains (Fig. 2). The corresponding cost (10 6 Rs present
value) are summarized as follows:

Storage
Phase Mains Pumps Pumping Treatment Total
{fixed)
1 1 023.5 17.2 114.5 94.7 59.9 1 309.8
2 56.0 3.7 177.6 40.5 12.5 290.3
Grand Total {10 6 Rs): 1 600.1
74

PHASE

D SOURCE

-®- NODE WITH NO.

PIPE (WITH FLOW DIRECTION):

EXISTING

PLANNENO

15.0 FLOW LIS

(5001 DIAMETER MM

NO SCALE

PHASE 2

Fig. 4 • Optimal System \.ayout and Diameters

 75

Again, an alternative design was derived by an expert consultant resulting in

additional cost of 11.6% (1 526.1 x 10 6 Rs) after phase 1, and of 7.7% (1 722.8 x 106
Rs) after phase 2. However, as in the previous example, the cost savings by applying
optimization should not be overemphasized, since input data may be not exact, and
assumptions are taken anyhow. But it is valuable additional information, and it
increases the reliability of the design if the impact of 'soft' data can be determined
using the capacity of an optimization model for sensitivity analysis. Another advantage
proves to be the automatic display of subsequent second, third, etc best solutions.
The investigation for the Rajasthan system was supported by the German Bank
of Reconstruction (Kreditanstalt fiir Wiederaufbau). The pertaining engineering
investigations were performed by IGIP /KIRLOSKAR of Germany /India.

4 SUMMARY

The design of cost optimal urban and regional water supply systems is of eminent im-
portance for developing countries where financial resources are scarce and local criteria
must be considered. Hence, optimization models were devised utilizing operations re-
search techniques. Practical design examples show the results in urban and regional
planning. Data input requirements correspond to common engineering practice. The
model development was sponsored by the German Science Foundation (DFG). The
programming and modeling contribution of G.E. Krauter is gratefully acknowledged.

REFERENCES

1. Birkhoff, G. and Diaz, J.B., 'Non-linear Network Problems', Quart. Appl. Math.,
Vol. XIII, No. 4, 1956.

2. Cembrowicz, R.G., 'Siedlungswasserwirtschaftliche Planungsmodelle', Springer,

New York-Berlin-Heidelberg, 1988.

3. Cembrowicz, R.G. and Harrington, J., 'Capital-Cost Minimization of a Hydraulic

Network', ASCE, Hydraulics Div., Vol. 99, HY3, March 1973.

4. Chiplunkar, A.V. et al., 'Analysis of Looped Water Distribution Networks', Envi-

ronmental Software, Vol. 5, No. 4, 1990.
76

5. Krauter, G.E., personal communication.

6. Labye, Y., 'Etudes des Procedes de Calcul ayant pour But de Rendre Minimalle
Coiit d'un Reseau de Distribution d'Eau sous Pression', La Houille Blanche, No.5,
1966.

7. Rechenberg, I., 'Evolutionsstrategie', Frommann-Holzboog, Problemata 15, 1973.

8. Trent, H.M., 'A Note on the Enumeration and Listing of all Possible Trees in a
Connected Linear Graph', Proc. Math. Ac. Science, Vol. 40, No. 10, 1954.

9. Watson, J.D., 'The Double Helix', Atheneum, New York, 1980.

10. Weinberg, L., 'Network Analysis and Synthesis', McGraw-Hill, 1962.

11. WHO, 'International Water Supply and Sanitation Decade', Briefing Document,
Geneva, 1980.
 Tiae-Step Control in TMA for Steady Flows in Large Pipelines

!IIASASHI SHI!IIADA
Chf. Res.Engr., Dept. of Hydraul. Engrg.
Nat. Res. Inst. of Agric. Engrg.
2-1-2 Kannonndai, Tsukuba, Ibaraki, 305 Japan

ABSTRACT
The l\IOC-based time-marching approach (TM'A) is enhanced to have an
algorithm of adaptively controlling the time-step sizes. This method
gives a rapid assessment of steady flows in complex and large pipeline
systems over a wide range of initial guesses to the time step. The
time step is updated every given number of iterations to rapidly speed
up the convergence to a final steady state. The proposed T!IIA is evaluated
through comparisons between computed results of the T!IIAs and the mesh
method, which is one of the best network approaches of directly solving
steady flows.

IIITRODUCTION
The author [1,2] proposed the basic theory and practical solution
technique, called time-marching approach (TMA) [3,4] for finding steady
flows in pipelines with the MOC-based waterhammer analysis. TMA is based
on the fact that transients generally approach steady flows if a
properly coded computer program is running under well-posed boundary
conditions. Though TMA calculations require much time because of
residual pressure fluctuations, the convergence to steady flows can be
rapidly accelerated by some considerations: (1) steady flows are found
with the method of specified time intervals by setting the number of
reaches of a pipe to be one unit without interior grid points [3]; (2)
the linear second approximation to frictional terms in the characteristic
difference equations [5] assures stable computations for use of
arbitrarily large time-step sizes [2,4]; and (3) the best speed of
convergence is obtained by minimizing the spectral radius. To do this,
the actual pipeline systems can be changed into alternate ones, in which
the wavespeeds, the pipe lengths, and the friction factors are replaced
by artificial ones. If the energy and continuity relations are kept
constant, the steady flow solutions are all the same between the actual

77
78

and artificial systems justifying the transformation of systems [2).

This paper addresses further improvements useful for the application
of the TMA to computational analysis of steady flows in complex pipeline
systems. The objectives of this paper are: (1) to arrange the
theoretical background and computational algorithm of TMA using the
Hazen-Williams equation as the head-loss relation; (2) to find a
procedure for adjusting the time-step sizes near the optimum ones so that
the steady flows are quickly obtained for a wide range of initial time-
step size guesses; (3) to improve the existing criteria on the
convergence, [4) possibly estimating the accuracy of the solution; and
(4) to evaluate the TMA through comparisons between numerical examples
of the TMAs and the mesh analysis, which is one of the network approaches
for directly solving steady flows.

TIME-MARCHING APPROACH FOR STEADY FLOW ANALYSIS

Basic Equations for Water Ha..er Analysis [5,6)

Assuming that a>>IVI,
av aH fVIVI
+ g - - + - - - = 0 (Equation of motion) · ( 1J
at ax 2D
aH a" av
+ - · - - =0 (Continuity equation)· • • • • (2)
at u ax
where V= averaged velocity (m/sJ; H= potential head (m); a= wavespeed
(m/sJ; g= gravitational acceleration (m/s">; t= time (sJ; x= distance
coordinate measured downstream (mJ; D= pipe diameter (mJ; and f= friction
factor which is expressed using the Hazen-Williams formula [7),
133.7
f= • . • • . ( 3)
CH1.8~D0.1671VI0.148

in which CH= Hazen-Williams velocity coefficient.

MOC and Characteristic Difference Equations [5,6)
Characteristic form
fa vlvl
dx
c+: --=+a,
a
+dH + -dV + dt = 0 . • ( 4)
dt g 2gD

dx a fa VIVI
c-: -dH + -dV +
-a, dt = 0 • (5)
dt g 2gD
Characteristic difference equations: The linear second approximation to
the frictional terms is employed [6,8):
c+: (HpD-HAJ + B (VpD-VA) + RoA VPD = 0 (6)
c-: -(Hpu-HcJ + B (Vpu-VcJ +Roc Vpu = 0 (7)

where RoA=fAIVAia~t/(2gDJ; fA=133.1/(CH 1 · 8 ~D 0 · 167 IVAI 0 · 148 );

Roc=fciVcla~t /(2gD); fc=133.1/(CH 1 . 88 D0 . 167 IVcl 0 · 148 ); ~t= time-step
size; Hpu, VPu or HPD• VPD= unknown head and velocity at grid point Pu
or PD, respectively; and VA, HA, or Vc, He are known at grid point A or
C, respectively; and
B= ajg • • • • (8)
 79
Computations of TMA are done using the method of time specified
intervals without interior grid points (Fig.1). As shown later, the
time-step size, ~t. can take arbitrary values as the only optimization
parameter [2], since Eq.(16) satisfies Courant's stability condition.

Pu, A: upstream of a pipe.

PD,C: downstream of a pipe.

The whole length of a pipe

is taken as one reach in
computations using TMA.
A ~x=L c
FIGURE 1. Rectangular grid for a pipe.

Mathe•atical Background of Theory on TifA [1,2]

Speed of convergence and optimization problem: Eq.(9) yields a set of
linear characteristic difference equations perturbed near a steady flow
in a vectorized form,
• (9)

where IJ+l= unknown vector composed of perturbed velocity and potential

heads at time (J+l)*dt; IJ= known vector at time J*dt; and S= constant
matrix depending on both the dimensions of the pipeline system and the
steady state flows. The speed of convergence depends on Ps, namely, the
spectral radius of the matrix S [15]. To reduce the value of Ps, the
length of pipe, the friction factor, the pipe diameter, and the wavespeed
can be replaced by artificial quantities while keeping the energy and
continuity relations constant between the actual and artificial pipeline
systems. The best convergence is obtained when the objective function Ps
is minimized under the constraints of energy and mass conservation,
Eqs.(10) and (11), respectively:

(k=l, · · · · · ,Np) (head loss) • • • (10)

(k=l,·····,Np) · • · . (ll)

where subscript i denotes an artificial quantity; k= pipe index; Np=

number of pipes; Qk= discharge (m 3 /s); L,k= artificial length of pipes
(m); A,k= artificial cross sectional area of pipes (rnf); f,k= artificial
friction factor; and v,k= artificial velocity.
The optimization problem can be indirectly solved with the computations
of TMA having fixed time-steps, which gives a good approximate solution
[2].

Approxi•ate Solution Procedure

Use of artificial pipeline system: Assuming that Ak=A,k, and therefore
• • • • • • • • (12)
80

Eqs. (10), (11), and (12) give

• • • • • • (13)

Furthermore, assume that all the artificial length of pipes L,,., are
changed to the same length L,* without depending on the pipe index K, and
therefore we have
f ,,.,= f,.,L,.,fL.* (14)
Using the Hazen-Williams' formula, Eq.(14) is arranged as

. • . . . • ( 15)

Note that the friction factor f,., depends on the Reynolds number [9)
while the velocity coefficient CH,., does not, and that the Reynolds number
for each pipe varies in the calculation process. Hence, the use of
Eq.(15) leads to simpler iterative algorithms for computations of TMA.
The artificial wavespeed is also changed without depending on the index K
as
• • • • • • • (16)

TABLE 1
Algorithm for TMA using Hazen-Williams' formula.

Input LK, DK, CHK (K=l, · · · · · ·,Np)

~
Give L,*: CH,K=CHK[LK/L,*] 0 ' 54
~
M=O: IW=O: Give ~t(OJ and ~t(OJ: to=O.Ol
~
a,=L,*/~t(OJ: B,=a,/g
~
Input ICX, ICXN
~
Initial conditions
~
FOR IT=l TO NMAX: IW=IW+l
~
Calculations at Boundaries
~
If EQ. (20) is satisfied, then go to 100
~

time-step control lcsee Table 2)

~
200 NEXT IT
~
100 PRINT
~
STOP
END
 81

With this transformation of the system, the number of optimization

parameters can be reduced to only the time-step increment, ~t. since
L,• can be arbitrarily determined using the similarity law between Ps
and a, [2). Table 1 shows a calculation algorithm for TMA using Hazen-
Williams' formula.

Ti•e-step Control
As shown in Fig. 2, ~t and the increment, Bt, of ~t are updated every
ICX time steps (M·ICX; M=1,2,3,····; M=index) to reduce Ps and speed up
the convergence. Note that the values of a, and B, are also changed
following Eqs.(16) and (8)', respectively, every time ~tis updated.
B,= a,/g • ( 8).

0 ICX-ICXN ICX 2*ICX-ICXN 2*ICX 3$/CX-ICXN 3*ICX Ti• steps

t t t t t t
M= 1 2 3 Cou:nter
·····························
;Norml Norm2! iNorml Norm2 iNorml Norm2 i
'·················· p( 1) . '··· p(2) ;,............................ ~ p(3) .
llt (1) tJ.t(2) tJ.t ( 3)
Bt(1) Bt(2) Bt(3)
•... ............................. ~--·········--······ '············-·····-

FIGURE 2. Schematic Diagram for Time Step Control

Definition of increment vector and its norm: An increment vector of

variables and its vector norm are defined for numerically estimating the
spectral radius. An increment vector Y is defined in terms of the
difference of velocities, tJ.V, and heads, tJ.H, between two successive time
steps as follows:

in which B,=a,/g, tJ.V= increment vector of velocity, tJ.H= increment vector

of potential head; and superscript T denotes the transpose of the column
vector. The vector norm of the increment vector is defined as
II Yll = (yT· YV" ... . . . . . . . . . . (17)

The value in parenthesis denotes the inner product.

Approximate spectral radius Ps: The vector norms of two increment
vectors, Norml and Norm2, are calculated at the (M*ICX-ICXN) and M*ICX

l
time steps, respectively,

Norm1= II Yll M·rcx-rcxN

• • • • • • (18)
Norm2= II Y II M· rex

Thereby, an approximate spectral radius, Ps averaged over the ICXN

time steps is estimated every ICX time steps as
82

- Norm2 1/ICXN

Ps= • • • • • • (19)
Norm1

Eq.(l9) is associated with the fact that the element corresponding to the
maximum of the absolute values of eigenvalues of S dominates near the
steady flows at issue. For simplicity, let Ps be denoted by p in later
descriptions. -p,
Algorithm for time-step control: Depending on the variations of
the renewal of at and its increment at every ICX iterations is designed
to reduce the value of p. The basic target is to construct a robust
algorithm for quickly controlling at at about the same order of
magnitude as the best value, at*, even for initial guesses far from at*.
Table 2 shows the flow chart on updating the time-step sizes.

TABLE 2.
Algorithm and Flow Chart for Time-step Control .

.J,
If IW=ICX OR IW=ICX-ICXN then goto 10 else goto 200
.J,
10 If IW=ICX-ICXN then calculate NORM1: goto 200
.J,
M=M+1: IW=O: calculate NORM2
-p(M) +
=(NORM2/NORM1)'/ICXN
- - .J,
If p(M) < p(M-1) then at(MJ=ot(M-1) else ot(MJ=-0.5*ot(M-1)
.J,
~t(M+1)=~t(M)+at(M)
.J,
If M(M+l)~ 10 2 then ~t(M+1)=10": goto 300
.J,
If M(M+1)~50 AND M(M+1)<10" then ot(M)=40*SGN(ilt(M-1)): go to 300
.J,
If llt(M+1J~20 AND M(M+1J<50 then ot(MJ=10*SGN(ot(M-1)): go to 300
.J,
If tJt(M+lJ~ 5 AND ~t(M+1)<20 then ot(M)= 4*SGN(ot(M-1J): go to 300
.J,
If M(M+1)~ 2 AND M(M+1)< 5 then HOI)= l*SGN(H(M-1)): go to 300
.J,
If 4t(M+1J~ 0.5 AND 4t(M+1)<2 then ot(M)=0.4*SGN(ot(M-1)): go to 300
.J,
If ~t(M+1)~0.2 AND ~t(M+1J<0.5 then ot(M)=0.1*SGN(ot(M-1)): go to 300
.J,
If ~t(M+1)~0.05 AND ~t(M+1)<0.2 then ot(M)=0.04*SGN(ot(M-1)): goto 300
If
+
~t(M+1)<0.05 then ~t(MJ=0.05
.J,
300 a,=L,*/~t(M+1): B,=a.lg
.J,

a. Denote the increment of at by ot(MJ, and

if p(MJ> p(M-1), then ot(MJ=-ot(M-1)/2 else ot(M)= ot(M-1).
 83

b. Set dt(M+1)=df(M)+61(M).
c. Assume the upper and lower limits to the time-step sizes such as
dtMAx=lO~ and dtMIN=0.05 or dtHIN=O.l, respectively.
d. Depending on the magnitude of dt(M+l), reset 6t(MJ to quickly control
the time-step sizes near the same order as the best value such as
if 50~ M(M+lJ< 10:~~, then HCMJ= 40*sgn{6t(M-1JJ
if 5~ di(M+lJ< 10 , then 6t(MJ= 4*sgn{6f(M-1)}
in which sgn(·) denotes +1 if (·))0, -1 if (·)<0, or zero if (·J=O.
Criteria on convergence: The criteria Vardy employed [4] is modified
to possibly assess the relative error of the solution in terms of
velocity and head as follows:

6v=,Vu-Voi/VAv<€o
6h= 611-6lkl/6lk<€o • • • • (20)

in which 6H=IIIu-11ol: V,.v=IVu+Vol/2; subscripts U and D denote upstream

and downstream sections of a pipe, respectively; 611L= head lost through a
pipe; €o= tolerance parameter. Note that the criteria should not be
applied to pipes where the velocity is extremely slow, since the roundoff
error may affect the criteria so that the final steady flows may be
unobtainable.
EXAMINATION OF THEORY AND ESTIMATION OF TilE TMA
The TMA with time-step control (TC) is evaluated through comparisons of
numerical examples computed with other TMAs and the mesh method which is
a typical direct network approach.

FIGURE 3. Pipe Network Model.

84

Other TMAs: CD NOC: method of characteristics which directly uses

actual dimensions of pipelines; ~ CNOC: only the wavespeeds ak are
replaced by artificial ones, a,k· Set a,k·dt=L in Eqs.(6),(7), and it
is clear that the steady flows do not depend on the wavespeeds; and ~
PT: the TMA with fixed time-steps which uses the artificial dimensions of
pipe lengths, friction factors, and wavespeeds [1,2).
Network approach: The nodal [10) and loop methods [11,12,13) are
used for directly solving steady flows. The mesh method is the best
when pipelines can be modeled only by loops in a graph-theoretical
framework. In any procedure employed, sparse matrix techniques are
essential to quickly assess steady flows in large pipeline systems.
Exa•ple 1
Fig. 3 shows a pipeline model assumed for irrigation where the number
of pipes, NP, is 157, the number of nodes, Ns, is 88, and the number of
fundamental loops becomes 70. Assume that the water supply
Q4=0.56m 3 /s, Q4,=0.28m 3 /s, Q4e=0.5m 3 /S, and Qe4=0.46m 3 /s (net discharge
1.8m 3 /s), is poured through node number 4, 41, 48, and 84. At each of
the remaining nodes, the same consumption 0.02143m 3 /s is demanded.

10

DT0=100 DTT=10 PLIM=1000

ICX= 10 ICXN=5

0.1
I'U13ER a=
ITERATIONS

0 100 200 400 500 600 700

FIGURE 4a. 6t(M) versus number of iterations.

(dt(0)=10 2 , dt(0)=10, L,*=10 3 )

Fig. 4a. to 4d. (JCX=10, ICXN=5, L,*=10 3 m) shows dt(M) versus the
number of iterations (ITC). The fluid in the pipeline is initially still
and a given pressure head distribution is assumed. The time-steps are
initially given as dt(0)=0.1, 1, 10, and 102 sec, and the initial
increments of 6t are given as 6t(0)=0.04, 0.4, 4, and 10, respectively.
In the figure, L,*, dt(O), and 8t(0) are denoted by the letters PLIM,
DTO, and DTT, respectively. The symbol () is marked every time the
convergence condition (20) is satisfied, in other words, the tolerance
 85
parameter becomes more severe such as Eo=l0- 2 , 10- 3 , 10- 4 , • • • in turn as
the solution is more accurate. Clearly, the time-step sizes approach the
best value and oscillate about the optimum, which is about 10 sec in this
case.

H'l

DT0= 10 DTT= 4 PLIM= 1000

ICX= 10 ICXN= 5

0.1
NLJ13ER OF
ITERATIONS

100 200 400 700

FIGURE 4b. ~t(M) versus number of iterations.

(M(O)=JO, t.f(0)=4, L. .. =l0 3 )

10

DTI'l= 1 DTT= . 4 PLIM= 1000

ICX= 10 ICXN= 5
0.1
NLJ13ER OF
ITERATH.:l'lfS

100 300 400 700

FIGURE 4c. ~t(M) versus number of iterations

(~t(O)=J, ~t(OJ=0.4, L,*=l0 3 )
86
100 TIME-STEP SIZES
DT0~ .1 DTT~ .04 PLIM~ 1000
ICX~ 10 ICXN~ 5

1121

0.1

121 200 400 600 8121121 1000 1200 1400

FIGURE 4d. llt(M) versus number of iterations.

(llt(0)=0.1. ilt(0)=0.04, L;"'=JO·')

ITC

Log(O.l)

20

40

FIGURE 5. A bird's eye view of lTC in terms of ICX and Log{ilt(OJ}.

 87

The TC is a powerful solution procedure, but it is not necessarily

complete. The situation of convergence is minutely investigated by
changing ICX within a wide range of ICX from 4 to 40, in which ICXN is
assumed about half of ICX. Fig. 5 shows a bird's eye view of lTC (the
number of iterations required for convergence) in terms of ICX and
Log~I(OJ in which representative points of {ICX.~f(OJ,ITC} are, for
example, (40, 10 2 , 330), (40, 10, 180), (40, 1, 440), (40, 0.1, 1100),
(38, 0.1, 1150), (36, 0.1, 2378), (34, 0.1, 3000), and so on. The rate
of convergence slows when ICX becomes 14 and 16. This indicates that a
singularity occurs in the computations. The reason is not theoretically
clear, but one possible estimation is that the period averaged over ICX
time steps becomes synchronous with the characteristic response time of
the artificial pipeline system.
Fig. 6 shows the comparisons among three TMAs (TC, FT, and CIIOC) in
terms of the computational time required for obtaining the solution with
the tolerance parameter set at co=l0- 2 • The result using IIOC will be
nearly the same as that obtained with the smallest time-step size in
CIIOC. The number of Iterations with the TC can be reduced to a few
times the number of pipes, thus leading to economical computations of
large systems with a personal computer.

COMPUTATIONAL
TIME

MIN
CMOC
10 3

A CMOC
OFT
e TC

10 2
FT
0
0

0 0

• • 0

10
0. 1 1 10 10 2

FIGURE 6. Comparisons of computational time among TC, FT, and CIIOC .

Table 3 shows the eomparison of the computational time required

between the TMA and the 'mesh analysis', which becomes the best approach
for a network system as shown in Fig. 3. If sparse matrix techniques are
88

not applied (SYG), numerical results require much computational time and
have low accuracy. For the same problem the transient driven TMA
requires about 15 times the computational time required by the best
solution procedures based on sparse matrix techniques such as SPG using
the Gauss elimination method, SOR (Successive over relaxation method),
and CGM using conjugate gradient method [13, 14).

TABLE 3.
Comparison of Computational Time of TMA and Mesh Analysis

Method Number of Computational Use of sparse

1terations time matrix techniques
SPG 4 43 sec 0
SOR 17 1 min 7 sec (wc=1.2) 0
0 min 44 sec (wc=l.7)
CGM 42 2 min 25 sec 0
SYG 6 53 min 7 sec X
TMA about 350 about 14 min X

Initial values are calculated by solving linear equations given by

linear approximation to frictional terms.
we: relaxation parameter.

1000
TMA WITH TIME-STEP CONTROL

100
lTC= 572 EPS= . 01
DT= .05 PLIM= 250
10 ICX = 4 ICXN= 2

0.1

0.01 I'U'BER !J' ITERATIONS

0 200 400 800 Hl00

FIGURE 7a. 6t(M) versus ITC, and 6h versus ITC.

 89

1000

TMA WITH FIXED TIME-STEP

100 REUHIU: ERRffi
OF POTENTIAL HEAD lTC= 360 EPS= • 01
DT= 10 PLIM= 250

0.1

0.01 JIU13ER OF I TERAT I ()liS

0 100 200 300 400 500 600

FIGURE 7b. Ah versus ITC (TMA with a fixed time step, At= 10 sec)

A pipeline model (NP=307 and Ns=282}, planning for distributing water to

paddy fields for rice production, is used to test the TMA. The system
has a total length of pipes of about 90km, and consists of a main network
and nine subsystems. Fig. 7a shows (At(M) versus ITC) and (Ah versus
ITC) for ICX=4, ICXN=2, At(0)=0.05, 8t(0)=0.02, L,-=250 m, and €o=10- 2 •
The optimum time-step is about 10 sec. Fig. 7b shows the relative error
of potential head Ah versus ITC, which is obtained using TMA with a
fixed time-step At=10 (L,-=250 m, and €o=10- 2 ) .
The merits of transient driven TMA, are that TMA, based on simple
calculus, excludes sparse matrix calculations for large problems, can
cope well with difficult treatments for boundary conditions, and, after
solving steady flows, leads to waterhammer analysis using the same
program. If these merits are considered, the TMA can be more practical
in solving steady flows in large pipelines than the existing network
approaches.

CONCLUSIONS

The transient driven TMA automatically adjusting time-step sizes has been
developed to quickly assess steady flows in large pipeline systems for a
wide range of initial time-step size guesses. Though the problem that
the TMA does not work well for a few ICX is left theoretically obscure,
the TMA is useful alternative to the network approaches, as shown
through some examples calculated using personal computers. In
particular, the TMA can calculate steady flows of pipelines with complex
boundary conditions with ease.
90

ACKNOWLEDGEMENT

The author thanks Mr. Bryan Thoreson, University of Arizona, Dept.of

Agricul. and Biosystems Engrg., for his critical comments on the paper.

REFERENCES

1. Shimada,M.: Advances in numerical analysis using MOC, 5th Int. Conf.

on Pressure Surges, 1986, BHRA, Hanover.
2. Shimada,M.: Time-Marching Approach for Pipe Steady Flows, J. of
Hydraul. Engrg., 1988, ASCE, 114(11) --
3. Fox,J.A.: Pipe network analysis-novel steady state technique, J. of
Inst. for Water Engineering and Science, 1975, (29), 183-194 -----
4. Vardy,A.E.: Rapidly attenuated water hammer and steel hammer, 4th
Int. Conf. on Pressure Surges, Paper A1, 1983, BHRA, Bath.
5. Wylie,E.B., and Streeter,V.L.: Fluid Transients, FEB Press, Ann
Arbor, Mich., 1983
6. Chaudhry,M.H.: Applied hydraulic transients, 2nd ed., Van Nostrand
Reinhold, New York, N.Y., (1987)
7. Streeter,V.L. et al: Fluid mechanics, McGRAW-HILL International Book
Com., (1983)
8. Wylie,E.B.: The microcomputer and pipeline transients, J. of
Hydraul. Engrg., ASCE, 109(12), 1723-1739,(1983).
9. Colebrook,C.F.: Turbulent flow in pipes, with particular reference to
the transition region between the smooth and rough pipe laws, J. Inst.
Civ. Engrg., London, 133-156 (1939)
10. Chandrashekar,M.: Extended set of components in pipe networks.,
J. Hydr. Div., ASCE, 106(1), 133-149, (1980)
11. Epp,R. and Fowler,A.G.: Efficient code for steady state flows in
networks, J. Hydr. Div., ASCE, 96(1), 43-56, (1970)
12. Shimada,M.: Some theoretical problems on pipe network flow analysis,
Trans. JSIDRE, (116), 41-47, (1985) (in Japanese)
13. Shimada,M.:Optimization technique for solving steady flow in large
network pipe systems, Hydraulic and sanitary engineering, Proc. JSCE,
369/ IT -5, 253-259, (1986) (in Japanese)
14. Shimada,M.: Efficient loop analysis for steady flows in large network
pipelines, Bull. of Nat. Res. Inst. of Agricul. Engrg., 26, 183-212,
(1987) (in Japanese).
15. Wilkinson,J.H.: The algebraic eigenvalue problems, Oxford University
Press , ( 1965) .
 A system for the Engineering Design of Transmission and
Distribution Pipe Networks•.

Authors: D.G.Papanikas"* V.Pantazis" P.Papagiannidis"

M.Bitzas" - A.Protopsaltis"
**Prof. Dr. -Ing., "oipl. -Eng.
Institute of Computational Engineering - ALFAPI S.A.
304, Messoghion Ave, GR-15562 Athens

Abstract
This paper describes the coq>Utational systet11 ETNA which performs steady and I.I'ISteady
flow analysis and design of transmission and distribution pipe networks. The syst"'"
sirulates the significant elements of a network, carries out conceptual design and uses
optimization procedures with respect to installation and operational cost. Additionally
the system carries out engineering calculations such as design of ther11111l insulation and
performs standardization according to international standards.
The Newton·Raphson iterative procedure is used to solve the steady flow case, while the
method of characteristics is used for the transient flow solution.
The system is mainly oriented towards Hydrocarbons using a vast amount of data
structured in a database system concerning the pipe and transport properties.
The Coq>Utational system has been verified using existing data from real networks, while
it has already been used for the preliminary design of the Greek natural gas
transmission network.
NOMENCLATURE
a wave transmission velocity.
B1 ,B 3 polynomial coefficients.
{C} supplies/consumptions vector of network nodes.
caopt annual total operating cost of a piping system.
copm annual operation cost per unit length.
cophp annual operating cost per HP of compressors.
CCF capital cost of compressors.
CPF capital cost of piping system.
d,D pipe diameter.
DF depreciation factor.
f,l friction factor.
g gravity acceleration.
(G] ij characteristic matrix of an element.
[G(Pe)] characteristic matrix of the network.
hi node elevation.
H total elevation (=p/(p.g)+h]
HP. horsepower of the j-th compressor.
+ J
Partially supported by Directorate-General for Energy,(DG XVII) of the C011111ission of European
COIIIIUlities,Programme "Hydrocarbons Technology Development•.

91
92

K pipe installation cost.

1; total length of piping system.
m coefficient of pipe installation cost.
M Mach number.
m; mass flow rate of node i.
rnii mass flow rate in element ij.
n pump revolutions per unit time.
P; static pressure at node i.
Pe equals to P+p.g.h
Q pump discharge.
t time.
T total temperature.
To stagnation temperature.
u, v fluid velocity.
X direction along pipe center line.
z(mii) resistance function of ij element.
y ratio of specific heats (cpfcv) •
APe pressure correction vector.
APp pump head.
APvii pressure drop of ij element.
p fluid density.
cp element inclination.

1. INTRODUCTION

The current paper presents the principles of the Computer

Aided Engineering (CAE) system ETNA (Energy Transmission Network
Analysis) which is used in the analysis and design of complex
pipe networks.
The work which has been carried out focused on the Flow
Analysis of Hydrocarbon Transmission pipelines (Oil and Natural
Gas) and distribution pipe networks (city gas distribution,
hydraulic networks and oil-refinery piping systems). The ANSI
standards which cover the concept areas of Gas Transmission and
Distribution systems, Petroleum Refinery Piping systems and Oil
Transmission piping systems, have been incorporated within the
computational system.
 93

The general chart of Fig. 1 gives an overview of the

structure and organization as well as the interaction between the
system modules.
The code development as well as the computational techniques
have been based on engineering and modern calculation methods and
coding techniques.
Many stages should be completed in order to design a pipe
network. The first is the preliminary design of the piping system
and the definition of the application area (oil transportation,
natural gas distribution, steam plants, etc.). The estimation of
the Transport and Thermo-Physical properties of the fluid medium
(Hydrocarbons or not) is also of great importance.
The next significant step is the definition of pressure,
supplies and consumptions, velocity/pressure drop limitations
and requirements along with design rules and standards.
The Flow analysis is performed for the cases of:
-steady, Incompressible flow of gases or liquids.
-steady, Compressible flow of gases.
-Unsteady, Incompressible flow of liquids
-unsteady, Compressible flow of gases.
The network optimization and cost analysis have not been
neglected, neither has the standardization of the piping elements
according to ANSI or API data tables.

2. MATHEMATICAL MODELLING

The computational core of the Flow Analysis Processor has

been based on non-linear matrix analysis and the method of
characteristics.
The mathematical modelling of a piping system operation has
been developed for the cases of steady, incompressible flow of
gases and liquids, steady and unsteady compressible flow of gases
and unsteady (transient) flow in liquid piping systems.
94
Steady, Incompressible Flow

The simulation of the steady state flow in pipe networks is

based on the mathematical representation of the network
components and their connectivity throuqh an appropriate
topoloqical system of nodes and elements [2].
The behaviour of the network elements is simulated by
applyinq the enerqy equations while, in order to simulate the
interaction between the network elements, the continuity equation
has to be satisfied at any network node.
The aforementioned equations alonq with the boundary
conditions constitute a non-linear system of equations which
describes mathematically the behaviour of the whole pipinq
system.
The application of the enerqy-type equation for each network
element leads to a relationship between the nodal characteristics
of the element.
The enerqy equation which describes the flow between the
nodes i,j of a pipe element is (s. Fiq. 2):
(1)

The followinq equation, which depends on the mass flow rate

m1i, is formulated from the above expression:
(2)

or

(3)

After the flow reference direction has been defined, the

nodal flow rates m1 , mi (where m1=mfi and ml=-mfi) are determined
by the followinq equation:

(4)

so, the element "characteristic" matrix [G]ij is defined by

the followinq equation:
 95

[G) ij (5)

where [A]= [ 1
-1
-1]1 is the element direction.

The characteristic matrix defines the capacity of the ij

element. The flow ability is increased if the function z (m 11 )
decreases. So, the equation for the nodal flow rates becomes:

(6)

In order to simulate the behaviour of the various active

components which are commonly used in real applications, some
special element madels have been incorporated within the network
modelling procedure.
The study of the operation of these elements aimed first to
improve the simulation of their performance and second to include
their performance within the characteristic matrix.
For example, the operation of the pump element is simulated
(s. Fig. 3) using the following energy-type equation.
(7)

Hence the pump element characteristic matrix [G]ii becomes:

(8)

and the mass flow rates m1, m1 at the pump nodes are also given
by the Eq. (6).
The aforementioned flow rates represent the contribution of
the element to the nodes i,j respectively •
The application of the same principles, as well as of the
continuity equation, to network nodes, results to the following
non-linear system of equations:
96

[G(P8 )] • tFJ= {d (9)

This matrix equation describes the flow behaviour of the

pipe network and constitutes the core of the solution of the
network flow analysis in conjunction with the boundary conditions
and the required modifications for the non-pipe network elements
(pumps, valves, etc.) in the characteristic matrix [G(Pe)]. The
computational techniques which have been applied are similar to
the structure and methodology which are used by finite element
methods and the solution of the non-linear equations is
implemented by the use of the Newton-Raphson algorithm.
It is possible to identify two types of boundary conditions.
The first includes the existence of network nodes which have
prescribed consumption or supply and the second refers to the
network nodes which have prescribed pressures.

compressible Steady and Unsteady Flow of Gases

The problem of steady, compressible flow of gases has been

confronted in exactly the same way as mentioned for
incompressible flow. In order to calculate the pressure drop, the
Gas Pipeline Flow formula [4], [11], [18) has been taken into
consideration after its specific alteration. The input of the
boundary conditions has also remained unchanged.
The above alteration has been validated and the results were
encouraging enough. In addition, for pipelines only, the analysis
of the steady one-dimensional compressible gas flow with heat
loss computation was implemented by the aid of the following
classical aerodynamic differential equations [ 20) which have been
solved numerically by using the 4th order Runge-Kutta method:

dT (10)
dx

4f (11)
D
 97

The formulation of the above differential equations in pipe

network flow analysis is difficult enough and time consuming and
is one of the future objectives for the extension of the existing
flow analysis code.
The computational work has been extended to unsteady flow
too but only for a single pipe (pipeline) and it has been based
on the solution of the flow equations. To avoid mathematical
difficulties simplifying assumptions have been made. Except for
the fact that the flow is assumed one-dimensional and viscous,
the thermal conductivity effects on the flow are also neglected.
Since pressure waves are mostly transfered by the potential
flow part, the ~nsteady, compressible Euler equations (in
conservative form} have been numerically solved.
The validity of the code was attained by comparison with the
method of characteristics for hydraulic pipes. The comparison
showed an excellent agreement. The resulting code can be safely
used in the prediction of resonances in ducts, unsteady flow due
to downstream fluctuation of static pressure and shock wave
predictions. It can also be extended to two-dimensional flow,
where the flow in more complicated geometries could be
calculated.

unsteady Flow in Liquid Piping systems

With the objective of analysing unsteady (transient) flow

in hydraulic pipe networks a special processor has been developed
which simulates the performance of hydraulic pipe networks when
transient phenomena occur due to the transient operation of
special active elements, i.e. the sudden closure of a valve, the
shutdown of a pump, etc.
98

Specifically the following system of partial differential

equations from Fluid Dynamics is solved:

aHt
a + u· aH + ~-au = -u·sincj) (13)
ax g ax

au + u· au + g· aH = _ l·u·lul (U)
Tt ax ax 2·d

The above system has been solved under the assumptions that
the flow is one-dimensional and the friction factor calculation
can be treated as in steady flow conditions at the same velocity.
The fluid medium is assumed partially compressible and the
elastic properties of both the fluid and the pipe material obey
Hooke's law.
For the solution of the above equations (13) and (14) two
types of boundary conditions are defined:
a. Determination of the flow rate or the pressure at an
element node.
b. The flow rate and the pressure at a given position are
related by a function.
In the beginning, the analysis of transient flow requires
the piping system's initial conditions and at the same time takes
into consideration all the provided boundary conditions (s. Fig.
4). This means that before the transient analysis is carried out
it is necessary to know exactly the head of a tank on which a
pipe is connected, or the opening and pressure difference of a
gate valve, or head gain of a pump, etc.

3. OPTIMIZED DESIGN OF HYDRAULIC PIPE NETWORKS.

The design of a pipe network, under prescribed constraints

of supply/consumption and/or pressure, is a problem having an
infinite number of solutions. The other requirement which should
be satisfied is usually the minimization of the operational and
installation cost of a network. The problem is that the
relationship between pipe diameter, flowrate, pressure drop and
cost is non-linear while most of the applied optimization
techniques require linear relationships. For the case of an
 99

increase in consumption at some network nodes the design could

be optimized by the use of mathematical methods (6],[26].

Optimum Design Analysis

The installation cost (K) of the pipes is estimated by the

following equation:
(15)

This estimation includes the cost of provision of the pipes 1

the cost of transporting them to the installation location and
the necessary excavation and restoration. The coefficients ac 1 m
depend on the previous items and the local economic factors
concerning the application area.
The critical point that should be stated is that the
procedure aims to increase the consumption capacity at a network
node with the lowest possible alteration cost. The network of
Fig. 5 consists of N elements where M pipes are connected to node
P. The node consumption is a function of N pipe diameters and the
method which has been applied does not exclude the case M=N. The
procedure allows the computation of the optimum diameter of any
pipe of the network which may be changed according to the will
of the Engineer in order to reach the desired increase of node
capacity.
The installation cost equals:
(16)

The relation between a node flow rate and the installation

cost of two pipes connected to the node is shown in Fig. 6 1 where
the curves on K1-K2 plane are of constant flow rate Q.
The minimization of the installation cost is obtained by the
application of the steepest path ascent method.
For N pipes 1 the cost increase AK; is obtained from the
following equation:
100

Ao · (if;)
N(ao) 2
(17)

fu aK1

The aforementioned method does not impose restrictions to

the topology, scale and operational conditions of a network. The
results of the resulting procedure are the cost optimized
diameters of the selected pipes and the new operational
conditions which would occur after the network modification.

3. MAIN COMPUTATIONAL ASPECTS

The main computational aspects will now be presented for the

Network Flow Analysis regarding the items of the previous section
about physico-mathematical modelling.
Regarding the steady, incompressible flow analysis, Eq. (9)
describes the network flow condition. This system is non-linear.
For its solution an iterative method which is based on the
Newton-Raphson algorithm is applied to find the solution of the
system of equations:
(18)

An assumption of the initial values of the unknown variables

is initially required. In addition the friction factor, the local
loss coefficients as well as the fluid transport properties are
determined for each element of the network.
The derivative matrix [F)' of the above function {F(Pe)} at
the m-th step of the iterative procedure is found by the
following equation:

(19)

The correction vector {APe} is found after the solution of

the linear system of equations:
 101

[F] I.· {AP)III = -Will (20)

and the new pressure values at the m+1 step are defined by the
equation:
(21)

The above iterative procedure is completed when the

increment vector becomes less than a limiting case close to zero.
A very important factor which concerns the computation speed
is the bandwidth of the characteristic matrix [G]. The lower the
bandwidth then the faster the procedure converges •
. The bandwidth depends on the topological representation of
the network. For the network of Fig. 7 two different topological
descriptions are given which present different bandwidth in the
characteristic matrix hence resulting in different computing
time.
So, in the network analysis code in order to avoid time-
consuming computing procedures for large scale networks, a
bandwidth minimizing method has been developed and included to
optimize the network topology [25].
Regarding the transient flow, the partial differential
equations (13), (14) are of hyperbolic type and the method of
linear characteristics was used to solve them. The method was
chosen due to its simplicity and computing speed along with
satisfactory results compared to an implicit finite difference
scheme that was initially chosen.

4. CODE DESCRIPTION

All the computational work has been developed in such a way

to have the following general characteristics:
Free format input.
Solution procedures which are continuously controlled by the
user.
Step-by-step extensive error diagnostics.
Data management and overall modularity.
Comprehensive output and adaptation capabilities for
different output devices.
102

Interactive and friendly communication with the user.

The Flow Analysis Code consists of eight significant
modules.
The Input process module handles the main initial data and
information as given for the case under analysis. Included here
are the selection of the application area and the standard code
which covers it, the fluid definition and the average (and/or
reference) pressure-temperature conditions, the selection of the
pressure drop formula, the network geometry and topology, the
boundary conditions, the elements' data, the pressure and
consumption/supply requirements, the cost analysis parameters and
the pressure drop or velocity limitations for diameter
modification.
The Solution process module solves the steady and unsteady
flow equations (compressible or not) and is based on those
equations which are referred to in Sections 2 and 3.
The Transport Properties Estimation process module
calculates the required thermodynamical and physical properties
of the input fluid. The development of the module has been based
mainly on the compositional analysis and the use of engineering
formulae which are proposed by the American Petroleum Institute
(10],[12], (14],[17].
There is also the Optimization Analysis process module which
is based on the method discussed in section 3 and the
Standardization process module which standardizes the piping
system elements according the the design formulae of the selected
application area and the ANSI and/or API tables.
Finally, the Output process module presents all the general
and detailed information of the piping system characteristics and
analysis results. The presented output is divided into elemental
output (friction factor, volume and mass rate, pressure drop,
nominal pipe size, pipe thickness, mean fluid density, thermal
loss, local pressure loss, etc.) and nodal output (node pressure
head and supply or consumption). The cost analysis results are
also presented at the end of the output procedure. The ability
to obtain the output in table form and graphical form is also
included.
In short transmission gas pipelines and small scale networks
the operational cost estimation is calculated. The pipe cost
 103

should be balanced against those of required compressor

horsepower. system operating and maintenance expenses should be
added to the annual fixed charges on the capital investment.
The annual total operating expense Ceopt of a piping system
of n lines and k compressors is estimated by the following
equation:

(22)

According to the above relationship a gas piping system can

be designed for minimum capital cost or minimum annual operating
cost, but not for both.
At this point the Data Base Management System should be
mentioned. This is used to manipulate the vast amount of data
that is required for the execution of the system's modules. The
databank consists mainly of piping element (pipes and fittings)
standardization data and tables of the A. P. I. "Technical Data
Book - Petroleum Refinering" (10] which are needed for the
computation of the transport and thermodynamic properties of
Hydrocarbons (pure or mixtures in liquid or vapour phase).
Ashton-Tate's DBase III+ and IV have been used for the
manipulation of data due to their simplicity and the fact that
they are "popular" enough.

5. RESULTS AND CODE VALIDATION

The verification and validation of the ETNA code was

initially based on the literature, papers and other scientific
and engineering material that was selected during the code
development and on which the design procedures were based. All
the methods and techniques have been extensively checked
according to their sources and the overall response of the
processor has been valuated after each code part integration.
A great deal of effort has been taken in the analysis and
design of real applications such as the first one which is
presented below. The material for these applications has been
selected from specific literature and publications as well as
104

companies which are associated with piping installations and

facilities (i.e. the Athens Municipal Gas company).
The code has proven extremely accurate in designing piping
systems where the operational conditions, the requirements and
restrictions apply to its charcteristics and capabilities.

Water distribution pipe network, Warwick. R.I.

In the following application the water distribution network

of Warwick in the Rep. of Ireland (s. Fig. 8) is analysed in
order for the ETNA code to be validated by comparison with the
results of [1]. In [1] the water distribution system has been
analysed by applying the Hardy-Cross method.
The network is described by 36 nodes and 43 elements (Fig.
9). The network is supplied by a tank with total elevation of
68.5 m (node 25). In addition the change of the pressures at the
points A, B was examined when the consumption of node 6 is
altered. The results of the computational analysis are shown in
Table 1 and the diagram of Fig. 10. In Table 1 the differences
between the ETNA code and Hardy-cross are less than st. These
differences are mainly due to the application of different
methods for the computation of the element pressure drop. In ETNA
the White-Colebrook relation has been used to obtain the results
in contrast with the application of the Hazen-Williams relation
which has been applied in the Hardy-Cross method.

Optimized Network Design

The computational system was used for the optimized analysis

of the hydraulic network of Fig. 1L (26]. The pipelines diameters
and lengths are:

Pipeline Length(m) Diameter(m)

1 1219.20 0.5334
2 2133.60 0.7620
3 2133.60 0.6096
4 1524.00 0.6096
 105

Initially the network has the ability to supply the nodes

2, 3 and 4 with 816 m3 fh, 1046 m3 /h and 645 m3 /h respectively and
the total manometric difference between the nodes 1,3 is 6.1 m.
The diameters of the new pipelines 200, 201 are required
such that:
- the network supply ability to node 3 should reach 3800 m3 fh,
- the installation cost of the new pipelines should be minimal.
The lenqth of pipelines 200 and 201 are 1676.40 m and
2133.60 m respectively.
The results of the optimization network analysis are shown
in Table 2. The differences between the results of Table 2 and
(26] are less than 0.5%.
For the same network the diameters of the new pipelines have
been calculated so that the supply increase at node 3 will
become 8000 m3 fh. The results are shown qraphically in Fiq. 12.

6. CONCLUSION AND FURTHER REMARKS

The development of the ETNA system could not have been

implemented without the support of the Directorate General for
Enerqy of the Commission of the European communities (5]. The
development of the system is remarkable not only from an
enqineerinq point of view but for software and proqramminq
quality too. The system has been created in Standard FORTRAN 77
and it has the ability to use the minimum computer memory in
analysinq larqe scale networks even thouqh double precision
calculations are performed throuqhout. The system is user-
friendly and its modular structure allows further improvement and
extension of its capabilities.
Durinq the developme~t of the system further workinq
concerns emerqed which forced us to extend the network analysis
into two-phase and, qenerally, multiphase flow. Further work that
appears to be of qreat interest is the creation of a qeneral
Transport Properties Estimation Data Base which will be
independent but ready to be linked to any network flow analysis
code.
106

In the above areas of development two projects initiated in

January 1990 are currently beign carried out by ALFAPI S.A.
itself and a further project, in co-operation with Harwell
Laboratory of the U.K.A.E.A., commenced in July of 1991. These
projects are also partially supported by the C.E.c. THERMIE
programme and their results will be presented in scientific
journals and conferences after 1992.

REFERENCES

1. c. V. Davis, K. Sorensen, Handbook of Applied Hydraulics, 3rd

edition, McGraw-Hill, 1969.
2. C.A. Brebbia, A. Ferrante, Computational Hydraulics,
Butterworth & Co Ltd., 1983.
3. A. J. Osiadacz, Simulation and Analysis of Gas Networks, Gulf
Publ. Co., 1987.
4. R.C. King, s. Crocker, Piping Handbook, McGraw-Hill, 1983.
5. An Expert system for the Engineering Design of Hydrocarbons
Transmission and Distribution Pipe Networks, Project
TH/10059/86 partially supported by C.E.c., Final Technical
Report, 1991.
6. D. Stephenson, Pipeline Design for Water Engineers,
Elsevier Publ. Co., 1976.
7. v. Streeter, B. Wylie, Hydraulic Transients, McGraw-Hill,
1967.
8. I.J. Karassik, W.C. Kruzsck, W.H. Frazer, J.P. Messina, Pump
Handbook, McGraw-Hill, 1976.
9. IBM Application Program, system/360, Scientific Subroutine
Package.
10. American Petroleum Institute, Technical Data Book
Petroleum Refinering, Metric Edition, 1981.
11. Gas Engineers Handbook, 1st ed., Industrial Press Inc.,
1965.
12. W.L. Nelson, Petroleum Refinery Engineering, McGraw-Hill,
1985.
13. R.D. Blevins, Applied Fluid Dynamics Handbook, Van Nostrand
Reinhold co. Inc., 1984.
 107

14. R.C. Reid, J.M. Prausnitz, B.E. Poling, The properties of

gases and liquids, McGraw-Hill, 1987.
15. c. Yaws, Physical Properties, McGraw-Hill, 1977.
16. C.F. Beaton, G.F. Hewitt, Physical properties data for the
design engineer, Hemisphere Publ. Corp., 1989.
17. R.W. Gallant, J.M. Railey, Physical properties of
Hydrocarbons, Vol. 1,2, Gulf Publ. co., 1984.
18. D. L. Katz, D. Cornell et al, Handbook of Natural Gas
Engineering, McGraw-Hill, 1959.
19. J.P. Rollins, Compressed Air and Gas Handbook, Compressed
Air and Gas Institute, 1973.
20. A.H. Shapiro, The dynamics and thermodynamics of
compressible fluid flow, Vol. I,II, Ronald Press co., 1954.
21. J.S. Page, Cost Estimating Manual for Pipelines and Marine
Structures, Gulf Publ. Co., 1977.
22. D.N.W. Kentish, Industrial Pipework, McGraw-Hill, 1982.
23. D.N.W. Kentish, Pipework Design Data, McGraw-Hill, 1982.
24. P.W. Halbert, L.A. Lotito, The Simulation of Gas flow
dynamics in Pipeline networks, American Gas Association,
Inc., 1970.
25. A. Recuero, J.P. Gutierrez, An effective bandwidth reduction
algorithm for micro-computers, Institute Eduardo Torroja,
Spain, 1980.
26. D. Stephenson, Operations Research Techniques for Planning
and Operation of Pipeline Systems, Proc. Pipeline Eng.
Conf., Brintex Exib. c, London 1970, pp. 105-112.
108
FIGURES

fLIII~IHIJIIIO A~LYSIS
J)otort~i .. tion
or HultS and
Stan4ards
J covoring tilt
Concopt ANa
r·········· Working Prtsslll't
I J)oltl'lli .. tion

.
Hults or thwd>
for opti111111
=~~~·t l'rtSSIII't t
1entional
~~FT~m!~" con1i tY::f:~ Ow
Hotoork
JRult' or thwd>l. ••••••••••••••••••••••••.••••

Hults or thwd>
sizinf hg
:t~t~~~~ S.tcill !lutot
l'l'fSSIII't drop D.ta
ole.
rlrlmi :t~~'lcir
tilt currtol
piping siztl

C.lcul.tion or
fluid inltmll
lorcts
!Prtsslll't or
MOMIAIUM) lfll lossts
calculatin ...
liltl'llll
Insula tin
J)osigo

Fig. 1:General Chart of the piping design system ETNA.

 109

:P_+pgh_
1 1
=Pi+pghj

Fig. 2:Pipe element connecting the i,j nodes of a network.

'!GU , - - - - - - - 'r - - - - - - -
,

H
._
l
~
·e
200 g
0

100
fe
t.

L---4~0----~80__1L00--12~0--~1~60~~200°
Percentage of nominal Q

Fig. 3:Characteristic curves for centrifugal pumps.

TIME Comput1 ng Procedure
oJl
:;::
(J)
0
,_

I
;:;
....
0
:;::
0
'0 u
c
0 ::::7)
'.J L
::::7)
<0
'0
I-
<0
c
::l
'0 0
c .0
l max ::l

l
0
.0 E
<0
E Q)

<0
Q)
....
L

....
(J)
L c
(J) ~
Cl. 0
:;:) 0

lniliBl condition Bl l=O X

L
Fig. 4:Definition of computing area, initial and boundary
conditions.
110

rig. S:N-elements network where M pipes are connected at a

node.
Q

Projection triangle of the derivatives of the function

Q=F(Kl,Kzl on K1-Kz plain.

rig. §:Relation between flow rate and cost for two pipes.
 111

9 10 11 12
---.~----~6---~~--~~----~----~----~
14
____.
half bondwidth-9

/15 16 17 18 19 20 21
/

4 7 10 13 16 19

half bondw1dth=5

6 9 12 15 18 21

Fig. 7:Two topological descriptions of the same network

resulting to different bandwidth.

----,..----~

~\
0 I 2 3 4 5 6 1 8 9 10
Scale-thouse~nds of feet

Fig. 8:Water-distribu tion system, Warwick, R.I. [1).

112

Piq, 9:Pipe network of Warwick with 36 nodes and 43 elements.

7.00 ..-----------------~

6.00
,....
~
I()
0
~ 5.00

a. I
I

-----r-----r-----r-----r-----
I I I I
4.00
1 I I

0.02 0.04 0.06 0.08 0.10

Q6 (m3/sec)

Piq. lO:Diagram of the change of pressure at the points A,B of

Warwick network when the consumption of node 6 is
variable.
 113

Node Pressure Pressure Dill. Consumption Consumption DiU.

ETNA Hard,..cross % ETNA Hard,.. cross %
(Pa) (Pa) (m3fsec) (m3/sec)

2S 672769.750 672769.62$ 0.0 0.11577 0.11577 0.0

26 671393.375 670377.500 0.2 -0.00221 -0.00221 0.0

24 656796.12$ 643466.750 2.0 -0.00915 -0.00915 0.0

23 644504.000 621041.000 3.6 -0.00536 -0.00536 0.0

8 641826.375 615958.000 4.0 0.00000 0.00000 0.0

22 638605.250 609379.750 4.6 -0.01230 -0.01230 0.0

1 657145.375 643466.750 2.1 -0.00757 -0.00757 o.o

3 664420.2$0 657520.12$ 1.0 -0.00221 -0.00221 0.0

173 670206.750 668284.500 0.3 -0.00315 -0.00315 0.0

Table 1: Comparisen of the results of ETNA code with Hardy-

Cross method for the network of Warwick [1].

Fig. 11: Pipe network for optimized design [26].

114

a. BleaeJ~~tal Reait.

Pipe Node. Length DitUIJeter Plow rate Velocity

(m) (m) (m3/ll) (m/•)

1 1 2 1219.20 0.5334 1356.0165 1.69

2 2 3 2133.60 0.7620 540.3427 0.33

3 3 4 2133.60 0.6096 -1191.0823 -1.13

4 1 4 1524.00 0.6096 1035.3028 0.99

200 1 4 1676.40 0.9215 2889.51166 1.20

201 3 4 2133.60 0.7537 -2088.8660 -1.30

II. Nodal Ra:alb

Node Preuure Maunowrate Vol. now rate Volume now rate

(P•) (Xg/•) (m3/•) (m3/b)

1 59879.60 1460.639 1.467 5280.91

2 2798.15 -225.606 -0.227 -815.67

3 100.00 -1056.650 -1.061 -3820.29

4 39033.78 -178.383 -0.179 -644.94

Table 2: Elemental and nodal results of the optimized analysis

of the network of fig. 11.
1~~----------------~---------------,
I
I
1200 ------,-------r------,--
I I I

1000

'E
.§. 800
~
I I
~
E 600
0
l3
---L------J-------
2.01 I

1
I
I

I
I
I I I

~ ------ -------r------,-------
I

1
I

I
I

I
I I I
I I I
200 ------~-------~------~-------
1 I I
I

0
0 2000 ~0 6000 8000
Capacity of Node 3 (m3/h)

Fig. 12: Required diameters of 200, 201 pipelines of the

network of fig. 11 in order to increase the supply of
node 3.
SESSION B: SAFE DESIGN
 Measurements and Computations of Transients
in Pumped Sewer Plastic Mains
by
Torben Larsen
University of Aalborg, Department of Civil Engineering
Sohngaardsholmsvej 57, 9000 Aalborg, Denmark
and
Richard Burrows
Liverpool University, Department of Civil Engineering
Brownlow Street, P.O. BOX 157, Liverpool L69 3BX, U.K.

ABSTRACT
Measurements of pressure and flow in a number of pumped plastic sewer mains have been
carried out in order to compare the actual observed transients with computations. Primarily the
situation following pump run-down has been studied. The investigation covered pipelines with
steadily rising length profiles as well as pipelines with one or more high points. The comparisons
emphasize the effects of cavitation and air pockets.

INTRODUCTION
The last decades urbanization and sanitation developments have resulted in larger and more
centralized sewer systems, which have increased the need for distant transport of sewage. Ac-
cordingly a large number pumped sewer mains have been established. The new EC directive on
municipal sewage discharges is expected to stimulate this trend further.
Plastic pipelines of uPVC ( unplasticized polyvinycloride) for pumped sewer mains have
recently become popular. The investigation mentioned here covers only pipelines of uPVC, but
the principles described here are relevant to most other cases.
In contrast to traditional sewer pipeline practice, pumped plastic mains are often laid in
length profiles directly following terrain without special attention to high points. In some cases
manual or automatic air-valves are placed at the high point, in other cases not. The criterium
for the choise of length profile and air valve equipment in this respect seems to be more or less
based on individual experiences and judgement. The hydraulic design of such pipelines will,
beside the basic problem of the hydraulic capacity, also include an analysis of transients and
considerations on self cleaning of both sediments and air pockets.
The structural design of pumped sewer mains should consider the external soil pressure
and the internal loads. Of interest here is the internal loads from the pressure of the fluid. The
design practice differs slightly from country to country, but in relation to internal loads it is
normal to design for a maximum static pressure. Furthermore, the expected pressure surges (
from low peak to high peak ) has to be restricted to a limited percentage of maximum allowed

117
118

static pressure, e.g. 50 %, because of the risk of fatigue. Further details on fatigue can be found
in [1]. The pipe supplier can often advice on these points as the design codes and other rec-
ommendations do not cover all details.
Below is described measurements and computations for 3 different pumped sewer mains,
all of which are in service in North Jutland, Denmark. The investigation covered in total 6
pipelines and is described in detail in [2], but here only the 3 most typical examples are given.

NUMERICAL MODEL
The numerical model used in this investigation was based on standard procedures, which can
be found in the literature. It is not the point of this paper to present new ideas in the numerical
solutions of transient problems , so we refer to Wylie and Streeter, see [3], for all equations and
other details. The following points summarise the major principles of the models used:
1. Standard method of characteristics, without interpolation, for single pipeline was
applied for solution of the equations of motion of the fluid.
2. Cavitation ( or water column separation ) was included by assuming cavities to
develop at the node points when the absolute pressure drops close to zero. Transient
wave speed was not corrected for influence of cavitation.
3. Air pockets of pre selected sizes could be incorporated in chosen node points.
4. Pump inertia was included. Internal steel pipelines within the pumping station was
not included, explicity except for the effect of friction.
5. An air chamber could be included at the first node point, where the pump was
situated also ( equations were solved by iteration ).
The computer programme was written in Turbo Pascal. After the investigation a commercial
version of the programme called WHPS was made available through direct contact with the
authors [4].

BASIC RESULTS AND MODEL CALIBRATION

The method of characteristics is known to give accurate results in simple and well-defmed
situations. For each of the pumping mains tested some basic experiments were first completed
to ensure that combined effects could be identified separately later. The basic experiences were
as follows.
1. The wave speed of the transients seem only to vary within plus/minus 10 % of
standard value calculated from the pipe specifications.
2. The use of the steady state friction factor combined with the computation of the
friction at the old time step seems to be acceptable under the unsteady conditions
to describe the damping of the first one or two reflections of the wave following
the pump run-down. The peaks of the later reflections in the computations are
overestimated, but this is less important.
3. Most uncertainty in the simple cases were related to the uncertainty of the initial
flow conditions, especially the steady state flow before the pump run-down. Often
the flow was lower than expected by up to 25 %of the nominal capacity.
4. The function of air chambers can be simulated rather accurately, so no further on
this point will be discussed. The standard value of 1.2 for the polytropic exponent
was used in the equation describing the relation between air pressure and air volume.
 119

In the more complex cases described below the numerical model was first calibrated as well as
possible against measurements to give the correct steady state flow and the wave velocity of
simple ( small ) transients, where cavitation and effect of air pockets did not occur.

CAVITATION IN STEADILY RISING MAINS

It is often questioned what the real consequence of cavitaion is under different conditions. It is
sometimes said that the calculation of cavitation is so uncertain that the uncertainty itself is
argument enough to avoid it. In order to get some practical knowledge on this point it was
decided to provoke cavitation in a simple case, which is a steadily rising pipeline without high
points. The length profile and other data are seen in figure 1. It should be mentioned that cavitation
was prevented under normal duty by an air chamber placed in the pumping station, but the air
chamber was disconnected during the tests.

Level (m)

~~%?4~ ~L[m]
0 250 500 750 913

Figure 1. Length profile of sewer main in Voldsted, Denmark, 150 mm internal diameter,
length 913 m, geometric lift 12.2 m, flow 15 Vsec, class 6 Bar uPVC.
Pressure measurements in the pumping station was now compared with computations on 3levels
as shown in figures 2a, 2b and 2c.

H [mVs]
- ~1:easured
_Computed
70

60
50
40
30

20+--.-.--.-.--.-.--.-.-~
0 10 20 30 40 50 60 70 80 (sec)

Figure 2a. Measurements against simple method of characteristics, without cavitation and
without pump inertia. Computations underestimate the pressure peak.
120

H [mYs]
_ Measured

70 - Computed

60
50
40
30

10 20 30 40 50 60 70 80 (sec)

Figure 2b. Measurements against method of characteristics with cavitation included but
without pump inertia. Better agreement than figure 2a is seen, but then the calculations over-
estimate the pressure.

H [mYs]
- Heasured
- Computed
70
60
50
40
30
20+-~-~-r-.-.--.-.-~~
0 10 20 30 40 50 60 70 80 (sec)

Figure 2c. Measurements against method of characteristics with cavitation and pump inertia
included. An almost perfect agreement is seen.
The pump inertia itself has of course no direct effect on to the cavitation. But the pump inertia
controls the pump run-down and thereby the shape of the low pressure wave moving from the
pump out into the pipeline, so the pump inertia is affecting the location of the cavitaion in the
pipeline. The conclusion is that the boundary conditions have to be modelled carefully. As
cavitation in the calculation takes place at the node points it can also be concluded that a relatively
high number of node points are preferable in order to locate the cavitation correctly.
 121

RISING MAIN WI1H ONE IllGH POINT

Figure 3 shows the length profile of the pumped sewer in Oue, North Jutland, Denmark.
Level (m)
4-0

30
_,-1
20
~
10

0 L [m]
0 500 1000 1500
Figure 3. Length profile of sewer main in Oue, North Jutland, Denmark, 296 mm internal
diameter, length 1500 m, geometric lift 12.2 m, flow 54 Vsec, class 6 Bar uPVC.
In figures 4 and 5 are shown the pressure measurements and computations at a point within the
pumping stations on two different days. Figure no 4 was recorded during a rain period where
the pumps had run continously for hours. It can therefore be assumed that air pockets were
washed out. Figure no 5 was taken in a normal dry period and a significant difference in the
picture was observed. First it is obvious that the maximum peak was lower, but also a slow long
period oscillation occurred. A reasonable agreement between computations was achieved in the
f'rrst case, but in the second case agreement was first found after incorporating an air pocket as
large as 1350 1 at the high point.

H [mVs] Measured
40 Computed

30

20

10

0+-~--~--T-~--~--~~--~--_.
0 10 20 30 40 50 60 70 80 (sec)
Figure 4. Recorded pressure during rain period against computed pressure.

H [mVs]
t1easured
40
Computed
30

0+---~~---r--~--,---~~~~---r---.
0 10 20 30 40 50 60 70 80 90 (sec)
Figure 5. Recorded pressure during dry period against computed pressure where an air
pocket of 13501 was incorporated at the high point.
122

RISING MAIN WI1ll SEVERAL HIGH POINTS

The figure 6 shows the length profile of the pumped sewer in St!llvring, North Jutland Denmark.
Level (m)
40

30
I
20

io IV"--
0o
~ -
1000
.A
~ .........

2000
~
~
2924
L [m]

Figure 6. Length profile of sewer main in St!llvring, North Jutland, Denmark, 472 mm inter-
nal diameter, length 2924 m, geometric lift 13.2 m, flow 150 Vsec, class 6 Bar uPVC.
In figures 7 and 8 are shown the same pressure measurements compared with computations first
without and secondly with air pockets incorporated in the calculations.
H [mVs)
t Heasured
Computed

40

20

0~~--~~--~--~--~~--~-+
0 20 40 60 80 (sec)
Figure 7. Measured pressure against computations.
H [mVs]
-Measured
-Computed

40

20

0+-~--~~--~~--~~--~~--·
0 20 40 60 80 (sec)
Figure 8. Measured pressure against computation where 50 I of air was distributed equally in
the 5 high points.
It is remarkable from figure 7 that the measured pressure peak is significantly higher than the
computed value. In other word in this case the air pocket ( or pockets ) have amplified the
maximum peak. By trial and error in the computations assuming different sizes of air pockets
it was concluded, as seen in figure no 8, that if totally 50 I of air was distributed equally in the
 123

5 high points a reasonable agreement could be found. Although this was an uncertain de-
termination of the air pockets, it is clear that the pockets here must have been much smaller than
in the previous example from Oue ( figures 4 to 6 ).

CONCLUSIONS
The results presented here do not pretend to give full scientific evidence on the general aspects
of cavitation and air pockets in pipelines. The findings should be taken more as practical ex-
periences. From this viewpoint the conclusions are
The influence of cavitation seems to be computed acceptably by the simple assumption
of cavities in node points in the case of a steady ( monotonic ) rising main. It seems
acceptable to rely on such computation for the evaluation the effect of power failure and
other rare and extreme cases.
Air pockets seem more or less always to appear in pumped mains with high points during
dry periods where the pumps only runs for a short part of the time.
Air pockets can either damp or amplify the pressure peaks depending on their size and
the character of the transients. Accordingly one can expect that air pockets in special
cases can cause severe overload and even failure of the pipeline.

ACKNOWLEDGEMENTS
The former civil engineering students now practising engineers J. H. Christensen, J. Vollertsen
and A. Bruun are sincerely acknowledged for their excellent field measurements and compu-
tational work during this study.

REFERENCES
1. Stabel, J.J., Fatigue Properties ofUnplasticised PVC related to Actual Site Conditions in
Water Distribution Systems.~ .!illd Pipelines International, no 1 and 2, 1977
2. Christensen, J. H. and Vollertsen J., Waterhammer, cavitaion and air pockets in pumped
sewer mains, M.Sc. Thesis in Danish, unpublished, Department of Civil Engineering,
University of Aalborg, 1989.
3. Wylie, E. B. and Streeter, V. L., Fluid Transients, FEB Press, Ann Arbor, Michigan, USA,
1983.
4. Larsen, T., Waterhammer at Pump Shut-down ( WHPS ). Programme Use Guide, Torben
Larsen Hydraulics, Aalborg, Denmark
 SURGE PROTECTION ON THE TRIMPLKY SUPPLY TO BIRMINGHAM

TIMOTHY KINGHAM AND JOHN DRAKE

Sir William Halcrow & Partners Ltd
Burderop Park, Swindon, Wiltshire, SN4 OQD, UK

ABSTRACT

The Trimpley Scheme was constructed in the 1960s to augment the water supply
to Birmingham using water pumped from the River Severn. The main supply to
the city is derived from the Elan Valley in mid-Wales and conveyed via a 90
year old aqueduct part of which is also used for Trimpley Water.
Hydraulic analysis of the Trimpley system, carried out in conjunction
with a condition appraisal of the Elan Aqueduct, revealed that unacceptable
pressures, both negative and positive, could be experienced on some sections
of the old pipelines following sudden halting of the Trimpley pumps. Surge
protection options were evaluated and a feeder tank located on a high point
of the pipeline profile near the pumping station was adopted. The tank and
feeder pipelines were constructed in 1991.

INTRODUCTION AND BACKGROUND

Trimpley Water Supply Works

The Trimpley Water Supply Works comprise abstraction from the River Severn,
bankside storage, drawoff and treatment works and a high lift pumping station
delivering water to Frankley Waterworks in Birmingham. Trimpley Waterworks
is sited on the East bank of the River Severn about 3 km North of Bewdley.
The site was selected because of its close proximity to the Klan Aqueduct
which crosses the Severn Valley in a long inverted siphon, and the scheme uses
pipelines on the Elan Aqueduct to convey the Severn water to Birmingham
[Figure 1].
This paper is concerned with surge analysis and protection measures on
the pumping mains to Birmingham from the high lift pumping station at
Trimpley. The mains comprise pipelines and tunnels built as part of the
Trimpley scheme and also the pre-existing pipelines of the Severn Siphon on
the Elan Aqueduct.

The Klan and Severn (Trimpley) Aqueducts

Since the commissioning of the Elan works in 1904, Birmingham has received
almost all of its water supply from the Klan Valley in mid Wales via a gravity
aqueduct. The Klan Aqueduct has a capacity of about 370 Ml/d. There are
eleven inverted siphons which account for about half its 118 km length; the
remainder of the aqueduct consists of tunnels and conduits. Each siphon
comprises four parallel pipelines, two of 42 inch diameter, generally cast

125
126

..
( ........:
·· ...\
Elan Valley Elan Aaueduct __ _
AHAYADEA ,...--•----:---- ·
Reserv~irs •
,----.., _,, .' Severn Siphon

r
\

HEREFORD
WALES
J •
'·.
.....

SOkm
I

Figure 1. Location plan

 127

iron, and two of 60 inch diameter in steel. The Severn Siphon is the longest,
some 27 km in length, and it crosses the River Severn close to Trimpley
Waterworks. It is the eastern half of the Severn Siphon that is used for the
conveyance of water from Trimpley.
The high lift pumping station at Trimpley houses nine pumps. Seven of
these are dedicated to the Birmingham Supplies and the other two are used for
local supplies. The seven Birmingham pumps draw from a 54 inch suction main
from. the contact tanks at the treatment works and deliver via a 42 inch main
which rises from the waterworks to the junction with the Elan Aqueduct
pipelines some 2 km away. A branch off the delivery main near the pumping
station connects to an air vessel consisting of a 137 m length of 54 inch
diameter pipe.
Figure 2 shows the arrangement of pipelines used by the Trimpley
schemes. The pumping main connects to three of the four mains of the Severn
Siphon (Elan Aqueduct). Any of the three mains (North and South Mains, -
42 inch, and 3rd Main - 60 inch) can be used to carry Trimpley water, and
cross connections between the mains allow the 4th Main (60 inch) to be used
as well from a point about 1 km downstream of the junction. At the end of the
Severn Siphon, about 16 km from Trimpley, the pumped supply normally follows
the Severn Aqueduct. This is a 42 inch steel pipeline, 9 km long including
two short lengths of pressurised tunnel, which discharges directly onto
filters at Frankley Waterworks. However, all or part of the flow can be
allowed to pass over bellmouth weirs at the end of the Severn Siphon to reach
Frankley Reservoir via the Elan Aqueduct.

Development and Operation of the Trimpley Scheme

The Trimpley Scheme was promoted in the 1950s together with Clywedog reservoir
as an additional supply for Birmingham at a time when water demand was
projected to exceed the capacity of the Elan Supply. The proposed output of
the Trimpley works was 164 Ml/d in two instalments. The first instalment
(82 Ml/d) was commissioned in the 1960s. The second instalment was not built
because demand in Birmingham has not kept pace with the projections of the
1950s. However, in the 1970s additional pumping capacity was installed to
allow raw or part treated water to be pumped in emergency up to about
182 Mljd. This was intended as a temporary replacement for the Elan supply
if it were to be lost through some failure upstream.
The pumped Severn water from Trimpley is many times more expensive than
the gravity supply from the Elan Valley. Elan water will always be the first
choice for supply to Birmingham and Trimpley water is normally only pumped at
about 10 Ml/d, primarily to supply a borehole recharge scheme for which the
Elan water is unsuitable. Trimpley water has been pumped at rates up to about
70 Ml/d during dry summer weather when the Elan supply has been cut back
because of depleted reservoir storage. However, the emergency pumping
capacity has not been used in earnest or proved in field trials.

HYDRAULIC ANALYSIS ARD SURGE BEHAVIOUR

Scope of Studies
As part of a major engineering appraisal of the Elan Aqueduct the Consultants,
Halcrow, were asked to assess the safe pumping capacity of the Trimpley mains
under normal and emergency conditions. These initial studies indicated that
column separation, occurring after sudden halting of the pumps, limited the
safe capacity to unacceptably low levels. Various measures for surge
alleviation were discussed and further work was undertaken to investigate what
surge pressures could be safely resisted by the old pipelines of the Aqueduct.
Following a definition by Severn Trent of the pipeline configurations and safe
pumping capacities required for normal 'topping up' of the Elan flow and
128

!
Wolverley Stream=)

g. ~ Crossing

f. f.
t
:§

t
!l
~
!
0
River Stcu
Canal Cr:sslng
~·

i i
tI
:I !!"

I
:§

Monument Lane SVC

Figure 2. Pipeline configurations for analysis of surge protection options.

 129

emergency replacement of Elan supplies, surge protection options were

appraised and a scheme recommended.

Modelling
Surge analysis of the system was carried out with numerical models of the
pumping station and pipelines. The models were constructed using Halcrow's
COUP programme, part of the firm's in-house HYDRA suite of hydraulic modelling
software. COUP is a modular programme for the simulation of flows and
pressure transients in closed conduit networks employing the method of
characteristics for pressure wave propagation.
The input data for a model comprises a number of units describing
mathematically the physical features of the system; pumps, pipes, non-return
valves, air vessels, minor losses, junctions etc. The Trimpley models include
all seven pumps at the high lift pumping station each of which can be on, off,
starting or halting as required, together with pipeline units appropriate to
the configuration being modelled. Boundaries to the models are generally
constant head units. Figure 3 shows a simplified schematisation of the pumping
station and pumping mains.

T-.EY T-.EY MON.M:NT

.lJIICTION TOP LAN:
SVCHAMBER SVCHAMBER

rfl ....
~'!!-~!!----------------------------------.--+----·~•t-----4et---¥•

5~===~~~==~---~~=:: t
--- ~
All HAGLEY FRAN<I.EY SEVERN
VESSEL n.tte. 1\NIEI. FUEAS

SEVERN AQUEDUCT

~=
Check valve
90" bend
TIIM'LEY
PIM'ING fJU'Ill
90" bend
STATION

CONTACT
ll\NK

Figure 3. Schematic of Trimpley Model

The models were carefully constructed to define the system correctly.

Data was collected from drawings and other records, and the manufacturers were
consulted to confirm details of the pumps. Pipeline roughnesses were
calibrated with data from trials and pumping records, supplemented by results
from other hydraulic studies carried out on the Elan Aqueduct.
In the initial studies a number of different pipeline configurations
were investigated. Subsequently, the cases considered were limited to two;
130

the normal operation using the Elan Aqueduct South Main and the Severn
Aqueduct, and the emergency operation using the 3rd and 4th Mains with some
flow in the Severn Aqueduct and the remainder spilling over the siphon
bellmouths.
The principal surge condition analysed was the simultaneous halting of
all operating pumps caused by a failure of the electricity supply to the
works. Once the behaviour of the existing system had been established under
these conditions, the models were used to investigate options for protecting
against surge problems.

Behaviour of the Existing System

The modelling demonstrated that a surge problem can occur owing to the
topography of the Severn Siphon downstream of the Pumping Stations. From
Trimpley Waterworks on the River Severn the pumping main rises about 140 m to
the summit of a hill (Trimpley Top) some 2 km downstream. At this point the
pipeline is within about 20 m of the hydraulic grade line (depending on the
pumping rate). Further downstream the pipeline descends again such that the
static head is in excess of 100 m for several kilometres, before rising
towards the end of the siphon [Figure 4].

240,-------------------------------------------------------------------,
Rising
Main Severn Siphon (Elan Aquaduct)

200 Severn Outlet Frankley

Bellmouths
;--\Elan Aqueduct)
l:urv:i

v:J
'
Trimpley Top '
160

Filers

40 ri'npley
~
20 Station
0 4 8 12 18 20 24
~(km)

Figure 4. Longitudinal profile of aqueduct - Trimpley to Frankley.

On failure of the pumps, the delivery pressure drops rapidly and a low
pressure wave is propagated downstream. Even at fairly small pumping rates
this is sufficient to cause sub-atmospheric pressure and column separation
within the pipe at Trimpley Top. Both the low pressures and the subsequent
high pressure shock waves caused by the recombination of the water columns
give rise to concern.
 131

CONDITION OF THE AQUEDUCT PIPELINES

Condition Assessment
The inverted siphon pipeline sections of the Elan Aqueduct were examined in
1985 as part of the engineering appraisal. Examinations were carried out both
externally, by exposing the mains, and internally by draining down and man
entry.
The North and South cast iron Mains were constructed as part of the
original Elan Aqueduct and have been in service for nearly 90 years, whereas
the 3rd and 4th mains were constructed of steel protected with concrete
internally and externally, and laid approximately 60 and 40 years ago.
The old cast iron mains had a very poor serviceability record in early
life, but burst pipe failures are now infrequent and joint failures are
generally confined to a few areas of unstable ground. External corrosion was
found to be light and generally uniform along the length. Internally,
widespread broad pitting corrosion 2 to 3 mm deep was identified. The
remaining service life of the mains was deduced to be in excess of the normal
planning period adopted for new projects, but occasional stress induced
brittle fractures remain a possibility.
Of greater concern were the exposed pipes at stream crossings, generally
in steel with rivetted joints, where serious corrosion had occurred. All
these have now been replaced with buried pipes suitably protected, on the
section of the Severn Siphon used for pumping Trimpley water.
The steel mains were found to be well protected by their internal and
external concrete coatings, particularly on the English sections of the
aqueduct. Pipe joints are either spigot and socket or collars made up in each
case in-situ with molten run lead. They have performed well but occasionally
leaks occur which can usually be traced back to evidence of ground movement.
Valves and Specials were found to be in worse condition generally, than
the straight pipe sections. Radiographic· examination identified
concentrations of graphitisation on the cast iron bodies, particularly in gate
valves. These vital components are housed in deep chambers with little
ventilation and insufficient attention had been paid historically to their
maintenance including painting. A systematic valve refurbishment/replacement
programme is now underway.

Acceptable Operating Pressures

The maximum operating pressure in the Severn Siphon pipelines of the gravity
flow Elan system is limited by an upstream overflow to 17 bar at the lowest
point on the siphon, and prudence dictates that when the same mains are used
for the pumped Trimpley Supply, the maximum surge pressure should not exceed
this value. Hydraulic modelling revealed that pumping rates on the existing
Trimpley scheme would have to be severely limited in order to keep the maximum
pressure in the mains, following pump stoppage, below this figure.
It was considered that the steel pipes on the 3rd and 4th mains could
operate at pressures up to 20 bar, but the air vessel at Trimpley pumping
Station had been designed for a maximum of only 17 bar. It was therefore
decided that the maximum pressure in all Severn Siphon mains should not exceed
17 bar.
Sub-atmospheric pressures can cause thin walled pipes to collapse;
consequently the allowable minimum pressure for the steel mains was limited
to -0.4 bar, though this could be considered conservative due to the
stiffening effect of the concrete coatings. The cast iron mains, though less
susceptible to elastic failure were restricted to the same minimum pressure.
132

SURGE PROTECTION OPTIONS

General Principles
In order to prevent the damaging low pressures occurring at Trimpley Top, it
was found necessary to sustain the pressure either by introducing fluid at
Trimpley Top or by providing a continuing flow from the pumping station.
Furthermore, the chosen solution was required to have a high degree of
reliability in view of the importance of Trimpley as the only significant
alternative water source for Birmingham. In the event of a failure of the
Elan Aqueduct it would be necessary to pump water from Trimpley at close to
its full capacity until the Elan supply was restored. As the predicted surge
shocks following pump failure could well cause damage to the pipelines and
appurtenances, robustness and straightforward operation were considered
important requirements for the chosen scheme.
The major options for surge protection are discussed below. Other
methods considered briefly included pump flywheels and additional air vessels,
but these were found not to offer a solution to the problem.

Backup Generator
The surge analysis models were used to assess the effect of keeping one pump
running while the others cut out simultaneously. This would be achieved by
backup generators at the Waterworks or possibly an independent mains power
supply to the site. This option raised the safe pumping capacity close to the
required target, but was ruled out on grounds of cost. It also required a
complex electrical configuration because it was found necessary to operate
different pumps from the generator for the normal and emergency pumping cases.

Air Valves
The use of air valves at Trimpley Top to relieve the sub-atmospheric pressures
was considered in some detail. Such a scheme was found to require a number
of large air valves on the 3rd Main to admit air at a sufficient rate, the air
being retained in the pipe by suitable non-return devices on the air valves
to form a cushion to retard the returning water columns.
Air valves are frequently used in this way to alleviate surges in minor
pipelines. At Trimpley however, it was not considered an adequately reliable
method of surge protection for the exceptional conditions of the emergency
pumping case.

Feeder Tank
A feeder tank at Trimpley Top provides a means of introducing water rather
than air into the pipe to sustain the internal pressure. The surge analysis
models were used to determine the required tank elevation and volume and the
feeder pipe diameter for each operating case. This information enabled an
outline scheme to be developed and several sites at Trimpley Top were examined
as possible locations for the tank. Once its practical feasibility had been
established the feeder tank emerged as the preferred solution.

THE FEEDER TANK

Site Selection
Hydraulic constraints dictated that the tank be sited close to the summit of
the aqueduct at Trimpley Top. Several sites were examined and local impact
and the views of landowners were important considerations in the final
location of the tank.
The chosen site lies immediately north of the aqueduct corridor in the
corner of a field which also houses the Trimpley Top air valve chamber. The
site is a short distance downstream of the summit, the ground level being
 133

within about one metre of the level at the summit.

Access to the site is from an adjacent minor road and a small
watercourse a short distance away can be used to discharge water from the
tank.
The Severn Valley is an attractive area and negotiations with the local
planning authority resulted in substantial landscaping measures being
incorporated in the scheme.

Outline Design
The principal features of the tank - its elevation, capacity and the sizes of
the feeder connections - were determined directly from the surge model. Other
details of the tank were dictated by the constraints of the site.
Feeder connections are provided to the North, South and 3rd Mains,
reflecting the pipeline connections from the Trimpley Works. The North Main
is not normally used for pumping but a connection is provided for use in the
event that the South Main is out of service. The feeders pass below the
aqueduct mains, rather than above, because of the low hydrostatic pressure
when water is drawn from the tank. The location of the tank requires the
feeders to pass under the 4th Main of the aqueduct.
Each feeder is fitted with a non-return valve, with isolating valves
upstream and downstream. The non-return valves and upstream isolating valves
are located in a chamber adjacent to the feeder tank. The downstream
isolating valves are installed as close as possible to the aqueduct mains, in
deep valve chambers beneath the aqueduct.
Runs of the surge analysis model were used to determine the required
diameters of the feeder pipes and non-return valves. For the 3rd Main an
800 mm diameter feeder was found adequate for the design duty. Under the most
severe conditions velocities in the pipe will approach 5 mjsec, which is high
but acceptable in view of the short duration and very low frequency of use.
A larger diameter pipe would have been significantly more costly and bigger
valves and fittings would be difficult to accommodate in the available space.
The feeders to the North and South mains are 500 mm diameter, and were
designed on similar principles to the 3rd Main feeder. Figures 7 - 12 in the
Appendix show output from the hydraulic model demonstrating the performance
of the system under surge conditions.
For the most severe case, the 3rd Main feeder was found to draw a total
volume of about 70 cu m from the tank, at a maximum rate of about 2.4 cu
mjsec. A 6 m square tank was adopted. Pressure losses in the feeder pipe are
considerable at the highest flow rates, and in order to minimise the height
of the tank, sub-atmospheric pressures are permitted in the mains at peak
flows. In the 3rd Main the pressure is allowed to drop to 0.3 bar below
atmospheric for a few seconds; to avoid this the tank would have to be 3 m
higher.

Detail Design
The arrangement of the tank and feeder pipelines is shown on Figure 5. The tank
and adjacent valve chamber are of monolithic construction in reinforced
concrete designed in accordance with BS 8007. While the level of the tank
itself is fixed by hydraulic considerations, the roof of the adjacent valve
chamber is made lower to reduce the visual impact and earthfill volume.
Ventilation for the large outflow rate is provided by a louvred structure on
the side of the tank facing away from the road, which can also act as an
emergency overflow if a non-return valve fails to close properly.
The feeders are of fabricated steel and are fitted with pellmouth inlets
in the tank. The upstream isolating valves and non-return valves are located
together in the valve chamber, and a large opening in the chamber roof for
removal of the valves is covered with precast concrete planks. From the valve
chamber the feeder pipes incline downward to pass below the aqueduct 4th Main.
134

The feeders have flexible joints near the tank and welded joints below the
aqueduct.
The feeder connections to the North and South Mains of the aqueduct
require the removal of one length of cast iron pipe from each main. The new
pipe is in steel with the same outside diameter as the original pipe and is
fabricated in two lengths for ease of installation. As well as the offset
feeder branch a flanged branch for man access is provided. The two new pipe
lengths are joined with a site welded collar, and a lead run joint is made
into the adjacent socket on the original pipeline. The other end requires a
joint to the adjacent spigot and this is achieved with a special leadjweld
collar, site welded to the new pipe with a run lead joint to the old.

I• SevernElanSiphon ~s ~I
AqueOOct

0I 2 4 6 B 1pm
I

Scale SECTION N =North S =South

Feeder
Tank

valve
Chamber
PLAN

Figure 5. General arrangement of feeder tank and pipelines.

On the 3rd Main, the feeder connection is made by cutting out a part of
the old steel pipe length and installing a single replacement length of the
same diameter with site welded collars at each end. Again, the replacement
length has an offset branch for the feeder and a flanged access branch.
Figure 6 shows the site layout and landscaping scheme. The tank is
buried under an earth mound with shallow slopes and is provided with a screen
of trees.

Construction
The tank and valve chamber were built under contract by Biwater Construction
in advance of the installation of the feeders by direct labour. This
arrangement allowed the contract to proceed without programme constraints
arising from works on the aqueduct, and also made use of Severn Trent's
specialised experience in pipeline works. The valves and fabricated pipework
were ordered early by Severn Trent and the items to be fitted in the tank were
issued to the Contractor for installation. The benefits of this course of
 135

action were that it allowed the orders to be standardised and reduced the risk
of contract delays arising from long delivery periods.
The tank and the feeder pipeline works were constructed in Summer 1991,
at a cost of approximately £200,000. Field tests of the system are planned
to be carried out in May 1992 to confirm the response of the system to
simulated pump failures.

Earth mound
over tank

Feeder
Tank

Tree and shrub

planting between
hedgerow >-
w
andmoood. ...J
11.
::::E

Figure 6. Site layout and landscaping.

CONCLUSIONS

The Trimpley scheme has been in existence for nearly 30 years but during
that period has not been used to the extent envisaged by the promoters at the
time of its construction. It fulfills an essential emergency standby roll,
in case of serious failure of the Elan supply. Analysis of the scheme as
originally constructed identified potentially serious over pressurisation of
the mains with consequent risks of failure both in terms of the system and
danger to third parties. Construction of the feeder tank and improved
maintenance procedures on the pipeline system have reduced risks to acceptable
levels, and at very modest cost.
136
ACKNOWLEDGEMENT

The authors wish to acknowledge the support and encouragement given by

E Powell, Engineering Manager, Severn Trent Engineering who kindly gave
permission for this paper to be presented. The cooperation of operational
staff in Birmingham Water District is also acknowledged.
The views expressed are those of the authors and do not necessarily
reflect the policy of Severn Trent plc.

APPENDIX

Figures 7 to 12 on the following pages show plots from the computer model for
the selected scheme.
 137

~8
§.

40 Tri'f1lley
l'l.n1>
Station
20 to----.-----4.---~----,a-----.----,12-----r----1Ts----.-----~,----.----,24r---~
Cha~(km)

Figure 7. Surge envelope for emergency scheme, following failure of all pumps
(3rd and 4th mains. Flow 189 Mljd).
3.0

2.5

Non-return valve nnAn t_ressure peak

when check valve
closes
A

\
2. o~r----

1.5
('/ ~]\
~\ ~hilA/ ~~v\
v if ~J
I

~ ___, v y
v Non-relt.m I
valve reopens
5

-1.0
0 50 100 150 200
Time In seconds

Figure 8. Pressure fluctuations at Trimpley Top following failure of all

pumps (emergency scheme).
138

3.00

2.50

A
I\
2.00

I
.oo ~
Al881.01dercurwls
voklmedrawntmmtan k
(appotlk. 70cu. m)

L I \

I \
.50

0.00
('..
.50

-0.10
0.00 50.00 100.00 150.00 200.00
Time i1 seconds

Figure 9. Outflow from feeder tank following failure of all pumps

(emergency scheme).

a.._(km)

Figure 10. Surge envelope for normal operation following failure of all pumps
(South Main. Flow 82 Ml/d).
 139

4.0

3.0

~
2.0 \
Non-return valve open

------ ---- ..
I

--·-·

'- _./
r\
---~ .-----
J~
"Non return valve reopens /
/ --

·1.0
0 50 100 150 200
Time in seconds

Figure 11. Pressure fluctuations at Trimpley Top following failure of all

pumps (normal operation).
1.0 ..

I\
0.9

D.8

D.7 I \
D.8 I \
0.5 I ---\t••a under cll've Is YOiume drawn from tank
appmx. 40 cu. m)

( \
I
0.4

~ 0~ - · \ A
-~

~ D2 / \ /~
I \ I "\ """
I \ \v J\;t./"
D. 1
1/
D.0 /
v 'I

.0. 1
0. 50 100 150
Time in seconds

Figure 12. Outflow from feeder tank following failure of all pumps
(normal operation).
 FAILURE OF A RURAL WATER SYSTEM - A CASE STUDY

A L PRASUHN
Chairman, Civil Engineering Department
Lawrence Technological University
21000 West Ten Mile Road
Southfield, MI 48075 USA

D ROLLAG
Chairman, Civil Engineering Department
Cruthers Engineering Hall
South Dakota State University
Brookings, SD 57007 USA

ABSTRACT

A large rural water system constructed in South Dakota, USA

was involved in extensive litigation following poor quality
installation and subsequent failure to pass acceptance
testing. The resulting investigation assessed not only the
material handling, preparation, and installation, but also the
hydraulic performance of the system. In this case it was found
that the problems were due primarily to poor workmanship, and
that hydraulic considerations were at most a secondary effect.
After a successful litigative effort and the replacement of
significant portions of the system, it has functioned properly
for over four years.

INTRODUCTION

The establishment of rural water systems has become a

significant factor in the American rural landscape. Backed by
federal loans, farms have become increasingly connected to a
public water supply, thereby being released from a dependence
on potentially unsafe and/or unreliable individual wells.
Because of the low population density in much of rural America,
often as low as one or two homes per square mile, the cost of
supplying the water is quite high on a per user basis. While
the benefits are significant, the systems remain at the very
limits of cost effectiveness and absolute economy governs the
outlays for materials, design and installation.

141
142

The system in question [Figure 1) is located in two

counties in southern South Dakota. There are 954 users in a
service area which exceeds 2500 square miles. The source of
the water supply is the Missouri River. The system consists of
surface water treatment, pumping and storage facilities, along
with a distribution system composed of asbestos-cement (AC)
and polyvinyl chloride (PVC) pipe. There are 1.44 miles of
18-inch, 9.06 miles of 16-inch, 11.72 miles of 12-inch, and
3.34 miles of 10-inch for a total of 25.56 miles of AC pipe.
The remainder of the system consists of about 597 miles of PVC
pipe ranging from 10 inches to 1.5 inches in diameter. The
area is hilly to rolling with elevation changes of as much as
275 feet. The bid price for the distribution system
installation project was $5,627,210.

Figure 2 showing the profile for the pipeline north from

the treatment facility, includes the two larger sizes of AC
pipe as well as the start of the 12-inch pipe. This section
includes the largest elevation changes and reflects the
general rolling nature of the terrain. The stationing is in
1000's of feet.

Problems developed after partial completion of the work.

The system could not pass the pressure and leakage acceptance
tests and failed repeatedly when attempts were made to place

N
• Pump Sta.
o Reservoir

Lines 1 1-10 Mile..,l

Two County
Boundary

Water Treatment Plant

Figure 1. Location map of the rural water system.

 143

Stationing, 1000's ft.

Figure 2. Profile of AC pipeline leading north from the water
treatment plant.

the system in operation. When it was determined that these

problems could not be readily rectified, the water district
with the backing of the engineer, withheld payment and removed
the contractor, resulting in a lawsuit and counterclaim.
The arguments by the contractor included charges that the
system was improperly designed particularly with regard to the
placement of air relief valves and the pump arrangement, that
the inspection was inadequate and incompetent, and that the
pipe and gasket failures were due to hydraulic transients
which resulted from design and operational errors.

A team was assembled for the water district and charged

with the examination of the pipe placement and handling, tbe
hydraulic analysis of the system, and the overall quality of
the engineering design. Compliance with the American Water
Works Association (AWWA) Standards [1) was considered
essential.
ANALYSIS OF DISTRIBUTION SYSTEM AND COURTROOM FINDINGS

Installation of the Distribution System

In order to evaluate the handling of the materials, the
preparation of the trenches, and the installation of the
pipeline, a soils testing firm was employed. By carefully
excavating large portions of the system and documenting their
findings, they were able to present a picture of poorly
prepared trenches with improperly placed pipe. In addition to
poor bedding, pipe sections in many locations were actually
supported on small rocks or only on the couplings. Pipe joints
were frequently misaligned. Gaskets were frequently displaced
and the pipe ends actually broken.

Figure 3 shows one of the many field-end gaskets which

had been forced out of the groove, whereas Figure 4 shows the
resulting black stain left by the rolled gasket. The
manufacturers' installation guide [2] states "To make certain
that the field work is done properly, it is strongly
recommended that each and every ring position be checked after
field assembly with a feeler gage . Failure to gage every joint
is false economy." A typical joint detail is included in
Figure 5.

Other problems such as inadequate cover, improper

backfilling and large rock in the vicinity of the pipe were
also identified. Pipe sections were found with badly chipped
ends. A section of broken pipe is shown in Figure 6.
Specifically, on the 18-inch AC pipe alone, 569 lengths
were removed and it was found that 31% failed to meet the
bedding specifications. Similarly, 62% of the couplers which
were removed rested on unsuitable bedding conditions. On the
basis of 116 compaction tests, neither the backfill nor the
bedding compaction was in compliance with the contract

Figure 3. Gasket pushed out of groove over an 8 inch length in

the 18 inch AC pipe.
 145

Figure 4. Black stain left by rolled gasket.

documents. In addition, 40% of the joints exceeded the

allowable deflection of 1.75 degrees and 47% of the
field-assembled gaskets failed to meet specifications.

Broken pipe was undoubtedly due to improper handling,

bedding or insertion. The examination alone did not prove that
the displaced gaskets were the result of careless insertion,
but all displaced gaskets were displaced inward.

AC AC
Pipe Pipe

Figure 5. Typical joint detail.

146

Figure 6. Photograph of broken pipe section.

Hydraulic Analysis of Distribution System

Since the contractor credited hydraulic transients for most of

the problems, the pipeline and system hydraulics were given
considerable attention. It was argued that the displaced
gaskets were due to pressure surges. While high water hammer
pressures would more likely lead to outwardly displaced
gaskets, none were found. In actual fact, due to the very
limited clearances between the pipe surfaces, neither the
possible positive or negative surge pressures would be likely
to displace a gasket and the pipe would probably fail first.

The alleged surges were blamed on the starting and

stopping of the pumps as indicated by spikes on the pressure
and discharge recording charts. It was however demonstrated
that the spikes were simply electrical surges which could be
generated independently of the hydraulic system itself .
The engineering design was severely criticized for its
limited use of air-relief valves. Design guidelines generally
specify that air - relief valves be located at every high point
in the line [3). While this may be a valid requirement for
systems built to grade, it is not practical in the typical
built-to- cover rural water system . A comparison was made with
other similar rural water systems on the basis of the number
of downward-sloping sections and their average elevation
change per air-relief valve . It was determined that the valve
densities were equivalent. Calculation of air within the
 147

system indicated that it never exceeded more than a few

percent by volume. Finally, a case can be made that because of
freezing problems, the air-relief valves may potentially do
more harm than good in the harsh South Dakota environment.

The hydraulic analysis also included a consideration of

normal and surge pressures, problems of air binding, and
improper operation of the control valves. During filling and
pressure testing by the contractor there were unexplained pipe
failures (e.g. Figure 6) probably due to damaged pipe or
pressure surges. However, these were isolated events and the
many other factors completely overshadowed them.
The contractor claimed that it would not be possible to
operate the system as designed. The years of successful
operation would discredit that claim, although it was clearly
impossible to satisfactorily pressure test the system as
originally installed.

Engineering Design and Inspection

Although a sophisticated analysis was not performed by

the design engineer, involving such procedures as computer
modeling of the distribution system and possible hydraulic
transients, there was no indication of poor quality
engineering work. The pipe classes were determined on the
basis of static pressures which reached 140 psi. Maximum
velocities and water hammer pressures were analyzed on the
basis of peak flows, but provision for fire-demand flows was
n6t included in the system. There was no appearance of faulty
design procedures. Furthermore, the plans and specifications
did not differ materially from those of other similar properly
working rural water systems.

By owner choice, the on-site inspection was too limited

and possibly not fully qualified. This illustrates an
all-too-frequent problem presently associated with
construction within the United States. In the name of economy,
the owner is often unwilling to adequately compensate the
engineer for the necessary inspection throughout the
construction process. While not relieving the contractor of
his responsibility to provide quality workmanship, this
prevented a more timely discovery of the ultimate problems.

RESULTS

The ensuing litigation resulted in a settlement in favor of

the owner to the extent of $3,000,000. This was the largest
settlement in the history of the state of South Dakota.
Appeals ultimately lead to a retrial which had essentially the
same outcome.
Remedial action included removal and reinstallation of
the entire 1.44 miles of 18-inch AC pipe with the replacement
of all defective pipe sections. Based on their random
inspections, the testing firm had recommended that the entire
25.56 miles of AC pipe be removed and reinstalled. Instead,
148

the rural water system elected to isolate each of the

remaining 24 miles of AC pipeline and subject each section to
an individual pressure test. Based on leaks disclosed by the
pressure tests, one mile of 12-inch AC pipe was completely
removed and reinstalled, as were portions of the remaining 23
miles. Because the leaks were primarily the result of
poorly-supported andjor misaligned pipe, the sections where
leaks were observed were all completely removed and properly
reinstalled, rather than following the usual in-place
leak-repair procedures. Altogether, about 13 miles of the
25.56-mile AC pipeline were removed and reinstalled. The funds
remaining after this work was completed have been placed on
reserve to cover the costs of future leaks that may develop in
portions of the line that were not repaired.

The system has been operating for approximately four

years since the remedial actions were completed. Two
additional leaks have developed during this period. Currently,
plans are being made to expand the system to accommodate a
total of over 2000 users residing in a service area covering
approximately 3900 square miles. In all subsequent work, the
water district has provided for adequate inspection by the
engineer.

CONCLUSIONS AND RECOMMENDATIONS

The experiences discussed herein demonstrate the fragility of
a water system of this nature. It clearly emphasizes the
necessity for careful engineering design and inspection. The
need to stress economy at every point places additional strain
on the engineer; this must be reflected in the design,
materials, choice of contractor, installation, inspection,
acceptance testing, start-up, and operation.

Specifically, the low user density demands that the

design must be bare-bones and the materials selected for
maximum economy. This adds to the need for precise contract
specifications.

Installation of the system proved to be the primary

source of problems. Consequently, the choice of contractor and
continual competent inspection of the work are vital
responsibilities. Installation must closely adhere to accepted
procedures. The handling of the material, preparation of the
trench, and the placement of the pipe are perhaps the critical
steps.
In addition to the monitering of acceptance tests, the
engineer must provide clear instructions for the start-up and
operation of the system.

In summary, the more economy required of a rural water

system, the more skill is required of the engineer. Certainly,
there are other types of systems where strict economy is also
the driving consideration. If there is one single lesson to be
learned by this experience, it is the absolute necessity for
adequate inspection.
 149

REFERENCES

1. AWWA standard for Installatio n of Asbestos-C ement Pressure

Pipe, American Water Works Associatio n, Denver, 1978.
2. Installatio n Guide for Transite Ring-Tite Pressure Pipe,
Johns-Man ville, 1969.
3. Concepts of Water Hammer & Air Entrainmen t in the Filling &
Testing of Pipelines, The Johns-Man ville Corporatio n, 1977.
 FLUID-STRUCTURE INTERACTION IN NON-RIGID PIPELINE SYSTEMS
- LARGE SCALE VALIDATION TESTS -
(EUREKA PROJECT 274)

A.C.H. KRUISBRINK and A.G.T.J. HEINSBROEK

Industrial Technology Division
Delft Hydraulics
P.O. Box 177, 2600 MH Delft, The Netherlands

ABSTRACT

To validate the fluid-structure interaction computer code FLUSTRIN, developed by

DELFT HYDRAULICS, experiments are performed in a large scale 3D test facility. The
test facility consists of a flexible pipeline system which is suspended by wires. Pressure
surges, which excite the system, are generated by a fast acting shut off valve. Dynamic
pressures, structural displacements and strains (in total 70 signals) are measured under
well known initial and boundary conditions.
The experiments are simulated with FLUSTRIN, which solves the acoustic
equations using the method of characteristics (fluid) and the finite element method
(structure).
The agreement between experiments and simulations is shown to be good.
The FLUSTRIN computer code enables the user to determine dynamic fluid
pressures, structural stresses and displacements in a liquid filled pipeline system under
transient conditions. As such, the code may be a useful tool to process and mechanical
engineers in the design and operation of pipeline systems.

NOMENCLATURE

A1 = cross-sectional discharge area u, = vertical lateral displacement

A, = cross-sectional pipe wall area uy = horizontal lateral displacement
c1 = pressure wave speed u, = axial displacement
D = internal diameter V = fluid velocity
E =Young's modulus of pipe material V, = relative fluid velocity
e = pipe wall thickness z = distance along pipe axis
f = Darcy-Weisbach friction factor 'Y = pipe elevation angle
G = shear modulus e 1•2•3 = strain
151
152

g = gravitational acceleration 8, = torsional rotation

H = fluid pressure head Jl = Poisson's ratio
11 = moment of inertia Pt = fluid mass density
J = polar moment of inertia P, = pipe mass density
K1 = fluid bulk modulus 0"1,3 = normal stresses
L = pipe length u, = axial stress
P = fluid pressure Tl3 = shear stress
t =time ~ = loss coefficient
INTRODUCTION

In general the transient behaviour of fluid-filled pipelines is determined by both the

hydraulic and structural conditions of the system. Hydraulic conditions like flow and
pressure changes, caused by e.g. valve closure, will change structural conditions like pipe
strains and displacements. On the other hand structural conditions like pipe expansion or
contraction and bend motion will change the hydraulic conditions. This interaction
between hydraulic and structural conditions is referred to as fluid-structure interaction
(FSI).
For a long time, the transient flow in pipeline systems has been described by the
classical water hammer theory, in which the structure is assumed to be rigidly supported.
Structural conditions like the pipe elasticity, the wall thickness and the way the pipeline is
supported can be found only in the propagation speed of pressure waves [27]. Structural
dynamics are not considered in this approach.
In the second half of this century, the water hammer theory is extended by taking
into account 1) pipe wall inertia effects related to the expansion and contraction of the
pipe under pressure changes (the so-called Poisson effect), 2) the pipe motion caused by
friction (friction effect) and 3) the motion caused at pipe ends, elbows or tees Gunction
effect). These structural motions influence the hydraulic conditions. The interactions
between hydraulic and structural conditions are referred to as Poisson, friction and
junction coupling respectively. In most cases the junction coupling is dominant.
Up to now the phenomenon of fluid-structure interaction has been studied by many
researchers [1-3, 5-14, 16-26]. However, the number of experimental studies is restricted
to a few tests at a rather small scale [7, 8, 18, 19, 21, 24, 25, 26]. The experiments
described in [18, 19, 21, 24, 26] have already been simulated with previous versions of
the fluid-structure interaction computer code FLUSTRIN [10, 11, 16, 17].
In the present paper experiments are described which are performed in a large scale
test facility at DELFT HYDRAULICS. The 3D test loop is especially designed to obtain
significant FSI effects. The relatively flexible pipeline system allows axial, lateral and
torsional motion. The results of the experiments are used to validate the FLUSTRIN
computer code.

FLUID-STRUCTURE INTERACTION MODEL

The transient behaviour of fluid-filled pipeline systems is governed by acoustic waves

propagating in both fluid and pipe. Pressure waves in the fluid, referred to as water
hammer, coexist with axial, lateral and torsional stress waves in the pipe wall. Due to
radial expansion and contraction of the pipe wall, fluid and pipe interact at the fronts of
axial waves. Interactions between all four types of waves may take place at pipe junc-
tions. The behaviour of the fluid is governed by extended water hammer equations [10,
13, 21], whereas the dynamics of the pipe is modelled by standard beam theory [4].
 153

The theory is applied under the assumption that the pipe is thin-walled and linearly
elastic. The radial inertia of the pipe wall is neglected. The basic equations are:

Fluid Equations
The fluid behaviour is described by extended equations of momentum and mass conserva-
tion:

av aH f v, I v, I = 0 (1)
'T""" + g "'!l +
ut uZ
2D

1 aH + .!. av _ 2v auz = 0 (2)

cf2 Tt g oz gE at

The extended equation of momentum conservation is equal to its classical equivalent, with
the exception of the friction term, in which the fluid velocity V is replaced by the relative
fluid velocity V, = V - liz (liz = axial velocity of the pipe wall). In the extended
equation of mass conservation an extra term is added to account for the Poisson effect.

The pressure wave speed c1 is defmed as:

(3)

Structural Equations
The structural behaviour is described by the equations of motion, applied in axial, lateral
and torsional direction. The local coordinate system used is given in figure 1.

fx

Figure 1. Local coordinate system.

154

Axial motion: The equation of motion is given by:

EPu EPu vDA, ap (4)

P,
A
I 7ii2'-EA--'
azz
I
2e az
+ p, A, g sin'Y

The first term at the left hand side represents the axial inertia of the pipe. Note that
the fluid mass density is not included. The second term accounts for the axial stiffness of
the pipe. At the right hand side the various external loads on the pipe in axial direction
are given. The first term represents the force due to the fluid pressure. In this term the
Poisson's ratio is present. The second term represents the force due to fluid friction.
Again the relative fluid velocity V, is used. The third term describes the axial component
of the gravitational force on the pipe. Note that also here the fluid mass density is not
present.

Lateral motion: The Bernoulli-Euler beam theory has been applied. The equations of
motion in the x- and y-direction (see figure 1) are respectively:

u a2 u
(p, A, + piAI)--x + £[ __x
a4 -(p, A, + p1 A1 ) g (5)
at2 Iaz4 COS')'

a2u a4u
+pi AI) at;
+ £[ __Y = 0 (6)
(p, A, t az4
The first term at the left hand side represents the lateral inertia of the pipe. The fluid
inertia is included. The second term describes the bending stiffness to which the fluid
does not contribute. At the right hand side, the lateral component of the gravitational
force is given, only for the non-horizontal x-direction.

Torsional motion: The equation of motion is given by:

(7)

The first and second term represent torsional inertia and stiffness respectively.

Interaction
Both the fluid and the structural equations contain terms with fluid as well as structural
quantities. The axial equations (2) and (4) are coupled via the Poisson's ratio, which is
referred to as Poisson coupling. The equations (1) and (4) are coupled via the friction
coefficient, which is referred to as friction coupling. The Poisson and friction effects are
caused by distributed loads which are modelled in the differential equations. Junction
effects are caused by concentrated loads which are modelled in the boundary conditions.
They couple all equations.

Strains
Internal forces and moments are derived from structural displacements, according to the
standard FEM. Normal and shear stresses are derived from the internal forces and
 155

moments and the fluid pressure under the assumptions: 1) plane stress conditions, 2) cross
sections remain plane and perpendicular to the neutral axes, 3) effective shear area can be
applied for shear forces. From the axial stress u1 , hoop stress u3 and shear stress T 13 (see
figure 2) the strains in the directions 1 to 3 are computed:

(8)

(9)

(10)

Figure 2. Stresses

FLUSTRIN COMPUTER CODE

The FLUSTRIN computer code is suitable to solve FSI problems in serial pipeline
systems in a convenient way. The hydraulic components presently incorporated in the
code are: Pumps, control and check valves, air vessels, surge towers, reservoirs,
resistances and pipe rupture. The structural components are: Rigid supports, springs,
hangers and dampers. Fluid-structure interaction takes place along the pipes (Poisson and
friction coupling) and at elbows, dead ends and axially moving components (junction
coupling).
The numerical procedures in FLUSTRIN, used to solve the basic equations together
with their initial and boundary conditions, are described in detail in [10]. The fluid
equations (1) and (2) are solved by the method of characteristics (MOC). The structural
equations (4) to (7) are treated by the finite element method (FEM). For the time
integration the Newmark {3 = 1/4 method is applied.
An iteration process takes care of the FSI coupling mechanisms.
156

TEST FACILITY

The test facility is a water-filled closed loop consisting of a variable speed pump, an air
vessel, a welded pipe with six elbows (square bends), a fast acting shut off valve and a
control valve (see figure 3). A flexible hose closes the loop between the control valve and
the pump. The structural boundary conditions of the system are:
- Rigid supports at the locations A and H. These supports allow neither translation
nor rotation.
- Bend supports at the locations B and G. These supports only allow translation and
rotation around the Xcdirection.
- Suspension wires located at about every 6 m along the pipe, which allow translation
in the horizontal plane (bouncefree) and rotation in all directions.
- An adjustable spring at location E. The spring can be mounted in the Xc or X3-
direction.
The stiffness of the "rigid" supports was measured under static conditions. The
relationship between force and displacement appeared to be linear, indicating that the
supports can be considered as springs with a constant stiffness. The axial stiffness of the
supports A and H is 316778 kN/m and 214307 kN/m repectively. The stiffness of the
suspension wires is specified by the manufacturer. The stiffness of the adjustable spring
varies from 30 to 100 kN/m.

1\
Adjustable spring 1:!:,

I
D

Bend support

Figure 3. The tested and simulated pipeline system.

EXPERIMENTS
The experiments were carried out starting from steady state conditions. Transients were
generated by closing the fast acting shut off valve at the downstream end. Different initial
and boundary conditions were obtained by varying the initial flow rate, the closure time
 157

of the shut off valve and the stiffness and direction of the adjustable spring. During an
experiment, the following signals were measured:
- The steady state flow rate using an electromagnetic flowmeter.
- 2 steady state fluid pressures using static pressure transducers.
- 6 dynamic fluid pressures using piezo-electric transducers.
- 9 structural displacements using inductive transducers. Eight displacements of
elbows were measured in different directions. The displacement of the shut off valve
was measured in axial direction.
- 3 forces in suspension wires using load cells.
- 48 pipe wall strains at four locations along the pipe. At each location four three-way
strain gauges (rosettes) were used (see figure 4).
- The valve disc position using an inductive transducer.
The signals were recorded simultaneously with a sample rate of 800 Hz during 5 seconds.

I
I
I
I
\
1\;l\
}_;:::) '\.
'-

Figure 4. Strain gauge arrangement.

COMPUTER SIMULATIONS

The test facility as shown in figure 3 is simulated with the FLUSTRIN computer code.
The pipeline between locations A and H is divided into 7 straight pipe sections,
connected by 6 elbows'(B to G). The pipe length between the rigid supports A and His
76 m with an additional length of 1.5 m between the airvessel and rigid support A. This
leads to a total length of L is 77.5 m, corresponding to 51 elements of 1.5 m. The
internal pipe diameter D is 108.7 mm, the wall thickness e is 3.07 mm, the effective
friction factor f (including bend losses) is 0.031, the mass density p1 is 8000 kg/m3 , the
Young's modulus E is 2.00 * 1011 N/m2 and the Poisson's ratio "is 0.3.
The fluid is water with a density p1 of 998.23 kg/m3 and bulk modulus ~of 2.19 *
109 N/m2 (temperature: 20 °C}. The pressure wave speed c1 according to equation (3) is
1257 m/s.
Hydraulic boundary conditions: The upstream boundary consists of a pump and an
air vessel. The pump is modelled by its head and efficiency characteristics (see table 1)
with constant suction head. The area of the air vessel is 1.109 m2 and the height is 2.865
m. Air expansion is assumed to be adiabatic. The downstream boundary consists of a shut
off valve and a control valve. The shut off valve is modelled using its pressure loss
characteristic (see table 2) with constant downstream head. The control valve was not
used during the experiments, and is not modelled.
158

Table 1 Table 2
Pump characteristics Pressure loss characteristic
(speed of 960 rpm) Shut off valve
(diameter D = 107 mm)
discharge head efficiency (D.P = ~ 2.pV 2 )
2
[m3/h] [m] [-]

0.0 8.66 0.000 valve position ~

10.0 8.65 0.250 [%open] [ -]
20.0 8.60 0.370
30.0 8.50 0.465 20.0 340.17
40.0 8.30 0.570 40.5 45.03
50.0 7.90 0.630 59.1 22.19
60.0 7.40 0.670 80.6 11.82
70.0 6.75 0.680 89.5 10.03
80.0 5.80 0.660 100.0 8.80
90.0 4.50 0.560

Structural boundary conditions: The displacements of the supports A and H are

assumed to be zero. The motion of the elbows is not restrained, with the exception of
elbows B and G where only axial displacement and rotation are allowed. The suspension
wires are modelled as springs with a stiffness of 285 kN/m. Since the wires may assume
non-vertical positions, the springs have both vertical and horizontal stiffness components.
The horizontal component (.243 kN/m) can not be disregarded, since it is relatively large
compared to the overall horizontal component of the structural stiffness. Concentrated
masses like rigid and bend supports, and attached equipment (e.g. strain gauge cable
boxes and suspension clamps) are neglected.
Initial conditions: The measured steady state flow rate and air vessel pressure
prescribe the hydraulic initial conditions, whereas the structural initial conditions consist
of computed pipe displacements due to hydraulic and gravitational loads.

RESULTS

Results are presented of a representative (cavitation free) experiment and corresponding

simulation. The initial flow rate is 0.3 m/s (pump speed 169 rpm). The initial pressure in
the air vessel is 7 bar (pump suction head 59.70 m and downstream head 59.79 m) and its
initial water level is 1. 732 m. The closure time of the shut off valve is 10 ms. The adjust-
able spring is not used here.

Pressures
Figures 5 and 6 show measured and computed dynamic pressures near the locations E and
H respectively. The agreement in amplitude and frequency is very good, although the
measured extreme values are slightly exceeded by the simulation.
The pressure signals show a more or less triangular shape where the classical theory
would predict a block shape. The maximum pressure at the shut off valve is about 5 bar
where the classical theory would predict 3.76 bar (Joukowsky: D.P = p1 c1 D.V =
998.2 X 1257 X 0.3 = 3. 76 X 105 N/m 2). The basic period of pressure waves is about
0.20 s, corresponding to a virtual wave speed of 1550 m/s, where the classical theory
 159

would predict 0.247 s (4Lic1 = 4 x 77.5 I 1257). Since Poisson coupling only slightly
changes wave speeds [15], this virtual increase of the pressure wave speed is attributed to
junction coupling [9].
A wave with a period of approximately 0.04 s is superimposed on the basic wave.

----Experiment - - - - - Simulation
5.0 ------------r----------- -----r--------- -------~

2.5 --
(._
ro
B
0.0
(I)
(I)
UJ
!f -2.5--------
z
>-
0
- 5 ~.~o~o----~o~.~2~o------o~.4~o~----~o-.6~o~----~o~.~a~o------1~.o·o
TIME ( s ) -
Figure 5. Measured and computed dynamic pressure 3 m upstream of location E.

5.0 - -- -------r------------,------------r----- ------,------------~

I I I I

I I
I

2.5-
(._
ro
B
0.0
(I)
(I)
UJ
!f -2.5 --------
z
>-
0
- 5 ~.~o~o----~o~.~2~o------~o~.4~o~----~o~.6~o~----~o~.~a~o------1~.oo
TIME (s) __________...
Figure 6. Measured and computed dynamic pressure at the shut off valve.

Displacements
In figures 7, 8 and 9 dynamic displacements are presented at elbows D and E.
Amplitudes are predicted quite well by the simulation, whereas the basic periods
tend to be too small. This is attributed to ignoring, among others, the motion of "rigid"
supports and the inertia of concentrated masses.
The three signals differ in basic period, since the ratio of effective mass and
stiffness varies with direction and location. However, the period of 0.20 s, effected by
pressure waves, is equal in all three cases.
160

----Experiment - - - - - Simulation
25. 0 ------ ------- -~---- ---- ------,-------- --------~---- --
1 '
' I

r 12.5 ----- ----------~--------

I

----, I

....J
0.0
a.
([)
.....
0 -12.5 ---- _ _ _ _ _ _ _ _ .L _ _

I
z I
>-
0
- 25 ·cP.oo 0.50 1. 50
1.00 2.00
TIME (s) -
Figure 7. Measured and computed dynamic displacement in X3-direction at location D.

100.0 ----------- ----r------- ---~----

r
I
I 'I
50.0 --------- -+--- -------1
' I
I '
....J
0.0 -------------~---------------~
I
a.
([)
.....
0 -50.0 ----------------~---------------l
' I
_________ _ 4------ --------~
I ' I
z I
>- I
0
- 100 ·cP. +:--oo-=------,:o-.-=-5--=-o----1-.+-o--=-o----1-.-=5,.,.-o_ _ _ _2=----'.oo
TIME ( s ) -
Figure 8. Measured and computed dynamic displacement in X1-direction at location E.

50.0 ---------------~-------------- ,-- ------------ -~---- ------------1

1 I
I I
I
25.0 --------~----------
I'
I
_________ I
0.0
....J I'
a.
([) I
..... I
' I
0 -25.0 ----------------~------ ----4---------------~
' I
z I'
>-
0
- 50 ·Jl. -~::--oo=---------=o:-.-=5-=-o----1,-.*o-=-o----1:-.-=5:-:::o----2:;;:--J.oo
TIME (s) -----a-

Figure 9. Measured and computed dynamic displacement in Xrdirection at location E.

 161

Strains
Figures 10, 11, 12 and 13 show dynamic strains near the locations E and H.
In figure 10 an axial strain 1.5 m upstream of location E is presented. The overall
tendency in amplitudes agrees, although the predicted frequencies are too high.
The strains shown in figures 11, 12 and 13 are measured 0.3 m upstream of the
valve, whereas they are computed at the valve. The axial strain is shown in figure 11.
Compared to the other strains, this strain is relatively small in magnitude and of less
importance. The measured and computed amplitudes differ significantly, due to the fact
that the axial motion of the valve (measured displacements in the order of 0.1 mm) is not
simulated. The pressure wave period of 0.20 s can be recognized. The shear strain is
given in figure 12. The agreement in amplitudes and frequencies is good, apart from
some drift. The hoop strain is shown in figure 13. The agreement between the measured
and computed values is excellent, which was expected because the hoop strain is
dominated by the (well predicted) pressures.

----Experiment -----Simulation
100.0 ------------r---
1

r 50.0

0.0
z
1-1
<(

g:
U)
-50.0 --------- ---<-------- ------<
I

z
iS - 100 ·<9. '=-o-=-o~~-----=o=-.--'::2'"""o~~~o=---'.4-:-:o,---~--=-o~.6=--o=----~~--=o-.-'="a-=-o~~-1~.o·o
TIME (s)-------<>-
Figure 10. Measured and computed dynamic axial strain 1.5 m upstream of location E.

100.0 ------------r------------,------------r------------,---------- --1

I I I I I
I I I t

r I
I
I
I
•
I I
I
I
I
I
'
50.0 ------------~-------------;------------~-------------1------------~
II I I I
I- ' '
tg_ I
'
I
I
'
I
0.0
z
1-1
<( I
g:
U)
-50.0
I I I
-------- -----1-------------'--------- ---1-------------'- -----------1
• I
I
I
I I

z
iS - 100 ·c9.~o-=-o~~-------,o=-.-':2=-=o--~-=-o--~.4-:-:o,-------~~-=-o-.6~o=--~~--=o-.~a-=-o~~~1~.oo
TIME (s)-
Figure 11. Measured and computed dynamic axial strain at the shut off valve.
162

- - - - - Simulation

r 50.0

0.0
z
.....
<(

g:
(()
-50.0 --------

z
~ - 100 ·cP. <::-oo-=----...,o-.'::-2o=----..,.o-.4"--o:------o~.6--,0,------.,..o----~.e,.--o---~1.oo
TIME (s)
Figure 12. Measured and computed dynamic shear strain at the shut off valve.

100.0 ------------r------------,------------r------------,------------1
I I

r
I I
I • I •
'
I
'
I
'
I
I I I I I

~
50.0 ·------------~------------~------------~------------~------------~
I I I I I

I • I
1- I
'g_ I
'
0.0
z
.....
<( I' I
a: I I '
-50.0
I I

1- ·------ ------ ~-- ----------_I_-----------~-------------'------------ _J

I I I I I
(f)
' I '
I I
' '
z I

~- 100 ·cP.~oo-=----~o~.~2~o---o~.44~o---~o~.6~o=----~o-.~e~o---1~.oo
TIME (s)
Figure 13. Measured and computed dynamic hoop strain at the shut off valve.

CONCLUSIONS

The fluid-structure interaction (FSD computer code FLUSTRIN has been validated by
means of experiments in a large scale test facility. Two important differences between FSI
and classical theory, already known from literature, are actually measured: 1) Pressures
exceeding Joukowsky's classical value (by more than 30%), 2) An increase of the fre-
quency of the pressure waves due to FSI.
The test facility is modelled in a rather simple way by ignoring details like e.g.
concentrated masses and displacements of "rigid" supports. In this way a first impression
of the usability of the computer code is obtained. However, to get more insight into the
influence of modelling on accuracy and computational effort, a further sensitivity analysis
is needed.
Some results of a representative experiment and simulation are compared. The
overall agreement between measured and computed fluid pressures, structural displace-
ments and strains is good in amplitudes and reasonably well in frequencies.
It is demonstrated that the classical theory is inadequate to describe the transient
 163

behaviour of the flexible pipeline system considered here. In this case and for a number
of practical pipeline systems the more sophisticated FSI approach is necessary. To judge
in which cases FSI is of importance, guidelines as proposed in [10] are needed. The
validated computer code FLUSTRIN offers possibilities to develop these guidelines.

ACKNOWLEDGEMENTS

The work done has been part of phase 3 of the FLUSTRIN project, initiated by DELFT
HYDRAULICS, The Netherlands. The FLUSTRIN project is financially supported by:
(France) Bergeron Rateau; Elf Aquitaine; (Germany) Rheinisch-Westfalischer TUV; (The
Netherlands) Dutch State Mines Research; Elf Petroland; Ministry of Social Affairs and
Employment, Nuclear Department and Pressure Vessel Division; Nucon Nuclear
Technology; Shell Intemationale Petroleum Maatschappij B. V.; National Foundation for
the Coordination of Maritime Research; (United Kingdom) ICI PLC; Nuclear Electric
PLC; Powergen.

REFERENCES

l. Belytschko, T., Karabin, M., Lin, J.I., Fluid-structure interaction in waterhammer

response of flexible piping, Journal of Pressure Vessel Technology, Vol. 108, pp.
249-255, August 1986.
2. Biirmann, W., Druckstosse in koaxialen Rohrsystemen, Dissertation, Universitiit
Karlsruhe, 0. Berenz, Karlsruhe, Germany (in German), 1974.
3. Ciccotelli, M., Mazza, G., Micelotta, A., Molinaro, P., Transient analysis in
piping networks including fluid-structure interaction and cavitation effects, Transac-
tions of the 11th Int. Conf. on Structural Mechanics in Reactor Technology, Tokyo,
Japan, August 1991, Paper K35/5.
4. Craig, R.R. Jr., Structural dynamics, John Wiley & Sons, New York, 1981.
5. Giesecke, H.D., Calculation of piping response to fluid transients including effects
of fluid/structure interaction, Transactions of the 6th Int. Conf. on Structural
Mechanics in Reactor Technology, Paris, France, August 1981.
6. Hatfield, F.J., Wiggert, D.C., Otwell, R.S., Fluid structure interaction in piping by
component synthesis, ASME Journal of Fluids Engineering, Vol. 104, pp. 318-325,
September 1982.
7. Jones, S.E., Wood, D.J., The effect of axial boundary motion on pressure surge
generation, Journal of Basic Engineering, Vol. 94, Nr. 2, pp. 441-446, June 1972.
8. Kellner, A., Voss, J., Schonfelder, C., Fluid/structure-interaction in piping
systems: theory and experiment, Transactions of the 7th Int. Conf. on Structural
Mechanics in Reactor Technology, Chicago, USA, August 1983.
9. Kuiken, G.D.C., Amplification of pressure fluctuations due to fluid-structure
interaction, Journal of Fluids and Structures, No. 2, pp. 425-435, 1988.
10. Lavooij, C.S.W., Tijsseling, A.S., Fluid-structure interaction in compliant piping
systems, Proceedings of the 6th International Conference on Pressure Surges,
BHRA, Cambridge, UK, pp. 85-100, 1989.
11. Lavooij, C.S. W., Tijsseling, A.S., Fluid-structure interaction in liquid-filled piping
systems, Journal of Fluids and Structures, No. 5, pp. 573-595, 1991.
12. Obradovic, P., Fluid-structure interactions: an accident which has demonstrated the
necessity for FSI analysis, IAHR Symposium on Hydraulic Machinery and
Cavitation, Belgrade, Yugoslavia, September 1990, Paper J2.
164

13. Schwarz, W., Druckstossberechnung unter Berucksichtigung der Radial- und

Liingsverschiebungen der Rohrwandung, Dissertation, Universitat Stuttgart, Institut
fiir Wasserbau, Mitteilungsheft 43, Stuttgart, Germany (in German), 1978.
14. Schwirian, R.E., Karabin, M.E., Use of spar elements to simulate fluid-solid-
interaction in the finite element analysis of piping system dynamics, ASME Sympo-
sium on Fluid Transients and Structural Interactions in Piping Systems, Boulder,
USA, pp. 1-11, June 1981.
15. Stuckenbruck, S., Wiggert, D.C., Otwell, R.S., The influence of pipe motion on
acoustic wave propagation, ASME Journal of Fluids Engineering, Vol. 107, No. 3,
pp. 518-522, 1985.
16. Tijsse1ing, A.S., Lavooij, C.S.W., Fluid-structure interaction and column separ-
ation in a straight elastic pipe, Proceedings of the 6th International Conference on
Pressure Surges, Cambridge, UK, BHRA, pp.27-41, 1989.
17. Tijsseling, A.S., Lavooij, C.S.W., Waterhammer with fluid-structure interaction,
Applied Scientific Research, Vol. 47, No. 3, pp. 273-285, 1990.
18. Vardy, A.E., Fan, D., Water hammer in a closed tube, Proceedings of the 5th
International Conference on Pressure Surges, BHRA, Hannover, F.R. Germany, pp.
123-137, 1986.
19. Vardy, A.E., Fan, D., Flexural waves in a closed tube, Proceedings of the 6th
International Conference on Pressure Surges, BHRA, Cambridge, UK, pp. 43-57,
1989.
20. Walker, J.S., Phillips, J.W., Pulse propagation in fluid-filled tubes, ASME Journal
of Applied Mechanics, pp. 31-35, 1977.
21. Wiggert, D.C., Otwell, R.S., Hatfield, F.J., The effect of elbow restraint on
pressure transients, ASME Journal of Fluids Engineering, Vol. 107, No. 3, pp.
402-406, 1985.
22. Wiggert, D.C., Hatfield, F.J., Stuckenbruck, S., Analysis of liquid and structural
transients by the method of characteristics, ASME Journal of Fluids Engineering,
Vol. 109, No. 2, pp. 161-165, 1987.
23. Wilkinson, D.H., Dynamic response of pipework systems to water hammer,
Proceedings of the 3rd International Conference on Pressure Surges, BHRA,
Canterbury, UK, pp. 185-202, 1980.
24. Wilkinson, D.H., Curtis, E.M., Water hammer in a thin-walled pipe, Proceedings
of the 3rd International Conference on Pressure Surges, BHRA, Canterbury, UK,
pp. 221-240, 1980.
25. Williams, D.J., Waterhammer in non-rigid pipes: precursor waves and mechanical
damping, Journal of Mechanical Engineering Science, Institute of Mechanical
Engineers, Vol. 19, No. 6, pp. 237-242, 1977.
26. Wood, D.J., Chao, S.P., Effect of pipeline junctions on water hammer surges,
Transportation Engineering Journal, Proc. ASCE, Vol. 97, pp. 441-456, August
1971.
27. Wylie, E.B., Streeter, V.L., Fluid transients, McGraw-Hill, New York, 1978.
 ANALYSIS OF PENSTOCK FRACTURE BY WATER HAMMER

Yukio Kono 0, Takahiro Sugano2) and Yukihito Sugai 3)

1), 3) Faculty of Engineering, Department of Civil Engineering at Tohoku Gakuin University

Tagajo, Miyagi-Ken 985 Japan
2) Faculty of Engineering, Department of Civil Engineering at Tohoku University
Aoba Aramaki Aoba-Ku Sendai, Miyagi-Ken 980 Japan

ABSTRACT

It is presumed from this study that the breaking pattern of pipes after the sudden closure of the
valve at stage by stage by assuming initial velocity and specify the yielding stress from elastic
condition to plastic condition. The results of analysis are compared with the condition of
actually-braked penstock. The water hammer waves acting on each section of the longitudinal
pipes are analyzed by the method of characteristics to find the high peaks of the surge
pressure. The yielding point and the changes from elastic to plastic condition was determined
by a two dimensional fmite element elastic-plastic analysis.

INTRODUCTION

Recently, the use of electricity has increased rapidly. Most of the demand is supplied by
nuclear power, thermal power, and hydropower plants. However, nuclear power plants and
thermal power plants cannot be stopped quickly after they have started to operate. Hydropower
plants are therefore used to control the supply to satisfy the demand for electricity because of
their ease of control in comparison with other two systems.

Many hydropower plants were built about 50 years ago at the time of the building rash of
hydropower plants. These are getting old and have many problems besides the case mentioned
above. The worker at the hydropower plants are from the next generation. There are less
people and they have less experience.

165
166

When hydropower plant is operated discontinuously, continual closing of the valve can cause
penstock water hammer. If the valve is closed suddenly, a tremendous high pressure of water
hammer can occur and can destroy the penstock connected facilities and even kill people. Once
an accident has occurred, it is very difficult to determine the main cause of the accident.

One possible cause of an accident is described herein for a particular penstock failure case that
occurred in a hydropower plant. The analysis is applied to a penstock fracture which has actu-
ally occurred, and it describes the process of penstock fracture in a stepby step manner.

The water hammer waves in each section of the longitudinal pipes are analyzed by the method
of characteristics to find the high peaks of the surge pressure. The yielding point and the
changes from elastic to plastic condition was determined by a two dimensional finite element
elastic-plastic analysis.

WATER HAMMER ANALYSIS

Simple Numerical Model

Figure I is a simplified model of the hy-
dropower plant. It is assumed that the
height of dam is 45 meters, the single steel
45 m
pipeline, 800 m in length, is connected to
the valve and turbine, and the valve was Pipeline Valve
closed suddenly. The upper boundary
800m
condition is the reservoir of the dam, so
the head, H, does not vary with time. The
lower boundary condition is assumed to Turbine
be a linearly closing valve. This pipeline
Figure 1. Simpilfied model of hydropower plant
is divided into 10 parts to calculate the
water hammer and its section area is de-
fined by the diameter of the steel pipe, 3.6
m, its thickness, 8 em, and the thickness
of the covering soil, 3 m, as Figure 2.

Wave Speed
The wave speed in a tunnel or pipeline
covered by a thick layer of soil can be
found by the thick-walled pipe formula
and letting the thickness go to infinity Figure 2. Pipe and covered soil
and it is given by following equation;
 167

p
a= (l)
1+2(t)(l+~)
If the diameter of pipe, 3.6 m, the bulk modulus of water, 2.07xl0 4 kgti'cm 2 (2.03xl0 9 Pa), the
modulus of elasticity of pipe, 2.lxl0 5 kgti'cm 2 (2.06xl0 10 Pa), and the Poisson's ratio 0.3 are
substituted into the equation (1), the wave speed is found to be 1271 mls.

Theory of Water Hammer

The theory of water hammer is generally based on one dimensional unsteady flow. The equa-
tion of motion and the continuity equation are:

oV V+oV +goH +-f-V2= O (2) 2SO,..._----....------,

ax ot OX 2D

Jl2 oV + ilHV+
8 ox ox
oH _~ V=O
ot ox
( 3)
"zoo--
B
- - - - 1-

171.2m 1

A linear combination of the equation (2)

A
I
D 150 - - ,.._.......... ""' -
and (3) gives C+ and C- equations, which
are the equations (4) and (5). They are
[m]
i
i I,~
I
It
II

solved numerically by the method of char- 100 -t-h-.-

acteristics. i• I: ~I \
OS
-=V±a
Ot
(4) 50 - t' - :.
I
~-1 -~ - 1\

·-·-1 : L_ -·-'\
L.,~.~_.J ... l
I I
~ dlly+!JY. +..__i_v I VI ~V sin a= 0 (5)
a dt dt "'2D a I

The finite differences of dS,dt,dH, and dV 0 I

become dS=SP-SL, dt=l). t, dV=VP-VL, TIME (sec.]
and dH=HP-HL on C+ equation and
Figure 3. Water hammer pressure
dS=SP-SR, dt= l). t, dV=VP-VR, and dii=HP
-HR on C- equation. VL, HL, VR, HR, VP, and HP can be solved by substituting characteristic
equations and the relationship (6) into the equations (4) and (5).

(6)

These data are input into the water hammer program and the results become as Figure 3. The
maximum pressure 17.12 kgti'cm 2 (l.679x10 6 Pa) occurs in front of the valve as soon as it
closes. The elastic-plastic analysis is applied for the section area of the pipe where the maxi-
mum water hammer pressure occws.
168

ELASTIC-PLASTIC ANALYSIS

The sections of the pipes in which the water

hammer acts are treated as a two dimensional
strain problem by the finite element analysis. It
is assumed that there is no strain toward x di-
rection, but that there is strain in the other di-
rections.

The internal virtual work become as follows ;

where h is the thickness of the triangular ele-

ment and it takes unit thickness h= 1 in case of
plane strain analysis.
By the external forces acting on nodes the ex-
ternal virtual work 15W 0 is found by multipli-
cation of virtual displacement and external
forces as follows; Figure 4. Division of elements

(8)
SOIL
The external virtual work and the internal vir-
tual work are equal each other and the equation
(9) is given by setting the equation (7) equal the
equation (8);

{f} =A h [BJI [D] [B] {d} = [KJ {d} (9) _Lj ELASTIC CONDITION

where [K] is called the stiffness matrix.

..dill NEW PLASTIC CONDITION

The unknowns of nodal displacement and re- ...ollllllll PLASTIC CONDITION

acting forces are found by substituting bound-

ary conditions into this equation. The stress is
then found from equation (10) by using the
nodal displacement {d} which is just been cal-
culated.
SOIL

(~;) =[D] [B] {d} (lO)

Figure S. Stage 1
 169

Stiffuess Equation of Elastic-Plastic Problem

An elastic-plastic is used as soon as any element SOIL
reaches its elastic limit under the action of the ex-
ternal forces (fluid pressure and soil pressure).
In this case,
{dF Ji =[Kep] 1 {dli }I (11)

.L:l
where [Kep] 1 is the elastic -plastic stiffness rna- ELASTIC CONDITION

trix. Using an incremental strain approach, the ..d NEW PLASTIC CONDITION
solutions {l:} 1 can be found by adding {clih pro- ..oilllllll
PLASTIC CONDITION
duced with the external loads {dF} i to the solu-
tions {li} 1•1 which were given at each previous
incremental stage. That is:

{F Ji = {F li.t + {d F Ji
(12)
{l: l 1 = {li li. 1 + {d li L
The method to evaluate {dF}i relating the un-
known elastic-plastic boundary is to limit the in-
cremental loads so that only one new element
becomes plastic at each incremental stage . The Figure 6. Stage 3
constitutive law adopted was the Von-Mises
equationP>
SOIL

Numerical Elastic-Plastic Analysis

The section area of the pipes is divided into num-
bers of small elements as shown as Figure 4. The
maximum water hammer pressure 17.12 kgt7cm2
(1.679xl0 6 Pa) acts inside of the pipes. .Ll ELASTIC CONDITION

~ NEW PLASTIC CONDITION

The characteristic values of the steel pipes are ..oillllllll PLASTIC coNDITION
the modulus of elasticity 2.lx lOS kgf/cml
(2.06xl010 Pa), the Poisson's ratio 0.3, the yield-
ing stress 420 kgf!cm2 (4.12xl07Pa), and the
characteristic value of hardening c=210 and n=l.
The chamcteristic values of soil are the modulus SOIL
of elasticity 5xl0 3 kgt7cm2 (4.9xl0 8Pa), the
Poisson's ratio 0.3, the yielding stress 105.4 kgt7
cm2 (1.034xi07Pa), and the chamcteristic value
of hardening c=52.7 and n=l. Figure 7. Stage 5
170

Result of Elastic-Plastic Analysis

As shown on Figure 5 to Figure 8 the plastic SOIL
condition starts approximately 55 degrees up
from the horizontal and moves to both the
lower and upper parts of the pipes. Finally,
the entire top part becomes plastic condition.
The profile of the penstock failure greatly re-
sembles the simulation by this analysis. ..L:J ELASTic coNDITION

..d!ild NEW PLASTIC CONDITION

CONCLUSION
PLASTIC CONDITION

The initial velocity is assumed as 1.0 m/s

when the Johnson valve closed suddenly.
The wave speed became 1271 m/s when the
modulus of elasticity is assumed as 2.1xl05
SOIL
kgflcm2 (2.06x10 10 Pa). The maximum surge
pressure that occurred in front of the closing
valve is 17.12 kgflcm 2 (1.679x10 6 Pa).

The plastic condition starts approximately 55 Figure 8. Stage 10

degrees up from the horizontal and moves to
both the lower and upper parts of the pipes. Finally, the entire top part becomes plastic condi-
tion. This changing from elastic to plastic condition can be observed in the action of the pipe
failure as the breakage began at the side of the pipe and the broken plates are bent upwards. It is
conjectured that the breakage of pipe occurred in a process similar to the one outlined in this
paper even though the breakage occurred very quickly.

If the assumption of the initial velocity of water 1.0 mls and the maximum surge pressure 17.12
kgfi'cm2 (1.679x106 Pa) are accurate, the steel pipe which has a modulus of elasticity 2.lx106
kgflcm2 (2.06x1011 Pa) and yielding stress 2400 kgflcm2 (2.353x10 8 Pa) remains elastic. Ac-
cording to calculations which assign increasingly lower values to the elasticity and yield stress
of the steel pipe, the pipe will not become plastic before the modulus of elasticity is 2.1x1 0 5 kgfl
cm2 (2.06x1010 Pa) and yielding stress is 420 kgfi'cm2 (4.12x107 Pa), which is as soft as copper
or brass.
REFERENCES

1. Britto, A.M. and Gunn, MJ., Critical State Soil Mechanics via Finite Elements, JohnWiley
& Sons,1987.
2. Chen, W.F. and Mizuno, D., Nonlinear Analysis in Soil Mechanics, Elsevier, Development
in Geotechnical Engineering Vol. 53, 1990, p132-139.
3. Kono, Y. and Sugai, Y., Water hammer analysis by method of characteristics and statical
stress Analysis of tunnels by fmite element method, In Proc.29th Japanese Conference on
Hydraulics,1985, pp.657-662.
4. Merrill, L., The Herald Journal, Dec. 5, 6 and 7, 1984 and Mar. 17 and Apr. 7, 1985.
 EXPERIENCES WITH SURGE PROTECTION DEVICES

C. SAMUEL MARTIN
Professor
School of Civil Engineering
Georgia Institute of Technology
Atlanta, Georgia 30332 USA

LEN COBB
Christy Cobb, Inc.
Birmingham, Alabama 35205 USA

ABSTRACT

For a large water pumping system with major surge protection devices, waterhammer analyses
and site measurements have been conducted. Experiences have been obtained with respect to
on-line and one-way surge tanks. The efficacy of these various surge protection devices has been
assessed from site measurements. The large pumping system consisted of a 30-kilometer steel
pipeline with four surge tanks and two one-way surge tanks. For this three-pump installation,
measurements of pump speed, discharge valve position, pump flow rate, and pressure at seven
locations were conducted under various transient test conditions. The site measurements under
three-pump operation allowed for improvement of hydraulic transient calculations for future
expansion to four and five pumps.

INTRODUCTION

A very large pumping station has been installed and commissioned to deliver water over a
distance of over 30 kilometers. The three pumps, which run at a synchronous speed of 720 rpm,
have individual rated capacities of 1.14 m3fsec, rated heads of 165m, and power of 2090 kw.
Initial surge analysis by others via the graphical method indicated potential water-column
separation. The surge protection system was then designed with one-way surge tanks as well as
air-vacuum valves strategically located.

171
172

Following design of the piping system with respect to surge protection and installation of the
pumping station and piping system, the authors were asked to check the analysis and the degree
of surge protection initially determined by the original engineer, who had died in the interim.
An extensive test program was initiated In 1989 in order to ascertain the steady-state flows,
pump and valve characteristics, and the level of surge protection. In particular, it was desired
to investigate the efficiency of four one-way surge tanks and two on-line surge tanks. The test
data were also useful in order to know if the transient analysis could predict with a reasonable
accuracy the performance of the future projected pump station configuration of four and then,
possibly, five pump operation. Figure 1 illustrates the profile of the ground and the location of
the three pairs of surge tanks.

Profile of Ground ond locotion of Surge Tonks

250,----------------------------------------------------,
Simple
Surge Tonka

150 One-Way
Surge Tonka

100

100

50 ~o------~~------~,o~------~,5~----~2~o~----~2~5~----~Jo·
Oistonce along Pipe in km

Figure 1. Location of Surge Tanks and Ground Proflle from Pumping Station

The frrst and second pair of surge tanks are of the one-way (feed tank) variety, while the
third pair are simple open on-line tanks. At each ofthe three locations there are two tanks 12.2
m (40ft) in diameter for purposes of redundancy.

At the pump house water is delivered from river elevation 77.7 m through the vertical
pumps to a 762 mm (30 inch) cone valve into a manifold which leads to the 30-km 1.83 m (72
inch) steel pipe located at essentially the ground level profile shown in Figure 1.
 173

MEASUREMENT PROGRAM

The measurement or data acquisition program was designed to monitor a total of 15 signals
-- three from each unit. For each unit the following quantities were measured: pump speed,
pump discharge pressure, cone valve position, pressure difference across the cone valve, and
flow from the signal from a flow transmitter in the manifold piping. For all quantities except
the pump rotation the signals were fed into a 12-bit PC based A-D converter. The software
used to drive the A-D card was QuickBASIC. Pressures were monitored by pressure
transducers, flow by a 4-20 milliampere commercial flow transmitter, and cone valve position
by a precision resistor connected to a regulate power supply. Finally, for the transient tests the
rotational speed of one of the motors was determined by a video camera. The resolution of the
latter was on the order of a fraction of a second. All instruments were calibrated prior to any
tests.

TEST PROGRAM

Although the principal objective of the test program was the determination of the efficacy
of the surge protection and the accuracy of the transient analysis, in addition the characteristics
of the pumps and cone valves were also determined. Following the steady-state tests for the
pump and valve characteristics as well as the ascertainment of pipe system head losses, the
transient tests were conducted. Subsequent to pump startup tests the transient testing was
conducted. Only the cases of complete pump tripout will be presented here inasmuch as one
pump out of two or three pump operation is not as severe with regard to surge protection.

SURGE RESULTS

Pump trip tests were conducted for single, two, and three pump operation. In all instances
the cone valve was actuated by the control logic by the loss of motor power. After a 14-second
unseating motion the cone valve began to close, first rapidly, and then at a slower motion at
the end of the stroke, which was approximately 294 seconds. The data acquisition system was
begun just before motor shutdown and stopped after 400-600 seconds.

For numerical analysis a standard computer program applying the method of characteristics
was employed to simulate the transient events. Care was taken to vary the wave speed with the
wall thickness of the main pipe. The pump characteristics provided by the pump manufacturer
for both normal operation and in the reverse flow and reverse rotation quadrants were utilized
directly. The four quadrant data were generally from a similar pump, but a single stage model
provided runaway data. Initially the analysis was based upon cone valve data furnished by the
valve manufacturer. Subsequent to steady-state flow tests the valve data was modified for
transients analysis due to considerable discrepancy between furnished data and measured.
174

Single Pump Trip

The comparisons of transient pressures at both the pump manifold and at the first air and
vacuum valve (890 m from the pump) with predicted values are presented in Figure 2. The
pressures at the pump were determined by the data acquisition system while a circular chart
recorder was employed at the high point. Figure 3 is a comparison of steady-state and minimum
hydraulic grade line (HGL) values with predicted values along the system. The calculated pump
speeds are compared with those recorded from a video camera on Figure 4.

. .
~E i
E
.s 40 .s

j 20 j 20

00 40 80 160 00 40 80 120 160

Time in seconds Time in seconds

Figure 2. Comparison of Measured and Predicted Pressures for One Pump Trip

250,,_--------~P~r~o~f~i~le~o~f~G~r~o~un~d~o~n~d~HG~L~o~l~o~n~g~P~ip~•~---------------,

200

50~
0--------~--------~1~0--------~1~5--------~2~0--------~2~5--------~,o
Distance olong Pipe in km

Figure 3. Comparison of Measured and Predicted HGL's along System for One Pump Trip
 175

Recorded ond Predicted Pump Speed

One Pump Trip
800

600 ' ' '

------------------~--------------------~--------------------1--------------------
'' ' '
' '' ''
''' ''' '''
400 ------------------~--------------------~--------------------t--------------------
'' '' ''
'' '' ''
E 200 ''
-----------------.--------------------
'
..''' ------------------- ''.. -------------------
'
e- '
''
''
'
''
'
.<= ''
-----------------.-------------------- ..'' ------------------- .'' -------------------
".,.,
Q.
Vl
-200
-----------------f-------------------1~~~~~~-~~d--------;
-----------------r-------------------r------ ---- -------r-------------------
~

0 -400
..<:
Vl

-600 ----------------r-------------------i-- -- ------------r-------------------

-800 ---------------r-------------------r-------------------

-IOOQQL-----~5LQ______
10L0----~15LQ-----2~0L0-----2~5L0-----3~0L0-----3~5L0-----4~00
Time in seconds

Figure 4. Comparison of Measured and Predicted Pump Speeds for One Pump Trip

Two Pump Trip

For two pump trip first one pump was started and allowed to run for 30 minutes to an hour
before both motors were suddenly cut off. The pressure comparisons are shown in Figure 5 for
the pump manifold and high point location. Figure 6 illustrates the minimum HGL's along the
piping system and the measured steady-state values at several locations. As illustrated by Figure
7 the pump speeds are well predicted by the analysis, particularly during the initial speed drop.

Recorded and Predicted Pressures at Pump Manifold Recorded and Predicted Pressures 890 m from Pump
Two Pump Trip
160 ,_-------.----'~·~o~P~"~m~p~T~rip~~~--------, 160

140 140

120 ~ 120
8
100 t 100
~
0
~
60
]
.< 40
5
zor------------------~------------------·--~--------------------1--------------------l ] 20

0 o~--~--~4~o--~L_~.~o~--'--~1~2~0--~--~ 00 40 80 120 160

160 Time in seconds
Time in seconds

Figure 5. Comparison of Measured and Predicted Pressures for Two Pump Trip
176

2 5 O,------'-P-'-r"-o-'-1'--i:_:1•:__::oc:f__::Gc_r::._0":::"c.:d:__::ac:o~d__:_:H::._GL'---'a'-'IC:oC.:.oCC!g-'-P-'-iP"-''----------------,

50 ~o-----}----,,~o----,,~s----c2~0~---~2~5-----i,.o
Distance olon9 Pipe in km

Figure 6. Comparison of Measured and Predicted HGL's along System for Two Pump Trip

Recorded ond Predicted Pump Speed

Two Pump Trip
800
'
__________________ .,. ___________________ .,. ___________________ .,. __________________ _
600 ' ' '
'' '' ''
''' ''' '''
400 ' ' '
------------------t--------------------t--------------------t--------------------
'
0
'' '''
'

''' '''
200 ----------------- '... -------------------r--------------------
' '' ... -------------------
E 0 ' '
' '
0.
.... '' '' ''
'' '' '''
-----------------.--------------------.--------------------.--------------------
! !
.!: 0
'0 Predicted '
"'"'
0.
Vl
-200
-----------------:-------------------r---------- -,-------------------
~
0
.c
Vl
-400 ----------------r-------------------r------ -- ---------r-------------------
-600

-800
---------r-- :::::::::::;::: _________ [-- -
-1000
0 50 100 150 200 250 300 350 400
Time in seconds

Figure 7. Comparison of Measured and Predicted Pump Speeds for Two Pump Trip
 177

Three Pump Trip

Finally, Figures 8 shows the transient pressures for the worst condition -- three pump power
failure. Again the transient pressures agree reasonably well for the first 80 seconds. The
minimum HGL's in Figure 9 also show good agreement, as well as the comparison of measured
and calculated pump speeds in Figure 10.

Recorded and Predicted Pressures 890 m from Pump

Three Pump Trip
160
!

! j Predicted
'
140 ------------------L-------------------~-------------------L------------------
I
l l •
~ 120 -------- '----~------------------- ~ 120 ------------------t----··············-t··········· ------}-
8 8
I ecorded! : : '
100 --------------t------------------- )100 ------------------~------------------~------- ..... ---~-----------------
:
1
! g.
-- ! ! --- !
- - _ _ _ _ _ _ [~~~~~~~~~~~~~~~~~~J~~~~~~~~~~~~~~~~~~I~~~~~~~~~~~~~~~~~~ 0 80 -------
!
-----~-------------------~- ------------~------------------
! :
i
Recorded
------------ - r·----------------·-r·-----------------r-·-----------------
.. 1 ! 1
--------------~-------------------~-------------------}------------------- ..
E : : :
··········-··[--------·---·--·-·-[---------------·-·-[-·-·-··-·-·-····--·
1
0.
20 _________________ J__________________J__________________ J__________________ _ j
..
----------·····+·--··-···---·------t-----------------·-t------------------·
l ! l
: : : i ! !
0 o~--~--~4~o~--~--.t.o~--~--1~2~o--~--~160 00 40 80 120 160
Time in seconds Time in seconds

Figure 8. Comparison of Measured and Predicted Pressures for Three Pump Trip

250 r----------p~,o~f~;~le~o~f~G~'~o~u~n~d~a~n~d~HG~L~a~lo~n~g~P~;~p~e------------------,

50.~o~------~~------~1~0~------~1~,--------~2~o~-------.2~,---------,do
Distance along Pipe in km

Figure 9. Comparison of Measured and Predicted HGL's along System for Three Pump Trip
178

Recorded and Predicted Pump Speed

Three Pump Trip
800

'
600 ' ' '
------------------~--------------------1--------------------l--------------------
'' '' ''
''' ''
'' '''
'
400 '
------------------t--------------------1--------------------t--------------------
'' '' '
''
'' ''' ''
200
''
-----------------.--------------------r--------------------r--------------------
' '
E
e-
' '
''
0
'' ''
'' '' ''
.£ 0
i'
'
!'
'
·--------=-1
----------------.--------------------,.-------------------.--------------------
'

"'a.
<I)
<I)
-200
Predicted
- ________________ ! ___________________ !__________ ----
(/)

-----------------r------------------r------ -----------r-------------------
~
0 -400
.<:

---l : : : :
(/)

-600

-800
--r --- -: :: :::; __
-1000
0 50 100 150 200 250 300 350 400
Time in seconds

Figure 10. Comparison of Measured and Predicted Pump Speeds for Three Pump Trip

CONCLUSIONS

The test program and transient analysis clearly indicated that the piping system was
adequately protected by the array of surge tanks inasmuch as there were no negative pressures.
Steady-state pipe friction and pump characteristics were essentially validated by the
measurement. The cone valve data did deviate considerably from the manufacturer's curve,
however. Moreover, the deviation of the calculated transient pressures from the measured
during the latter phase of the event is probably caused by the fact the four-quadrant pump data
used did not completely represent the pump, as evidenced by differences between steady-state
runaway and model data.

The WR2 values for the pump-motor assembly were apparently quite accurate as witnessed
by the close agreement between theory and measurement immediately subsequent to power
outage.
 PIPELINE SYSTEMS FOR LIQUID SULFUR

R. Szczepanek, R.Pichler
1-L-F Consulting Engineers, Innsbruck- Munich.

ABSTRACT

This paper deals with the technology of storage and the conveying of liquid sulfur by
pipeline. Due to its allotropic characteristic sulfur is a medium which cannot be compared
to any other element. This results in tight specifications for all the components of the con-
veying and storage system for the liquid phase. Constructive solutions are represented by
examples of two pipelines and a tank storage facility. One of the pipeline systems, which
is 4 112", 7 km long, is laid underground and the second, 2 112", 6.5 km long pipeline is
a surface installation. The tank storage facility consists of two 10,000 cubic meter tanks
and it is equipped with corresponding handling facilities for storage and intermediate
storage. Particular attention is given to the special features of the support, the heating and
the design of the pumps.

INTRODUCTION

Sulfur is beeing produced in large quantities all over the world. Traditionally it occurs in
natural deposits in its elementary state as sulfide in the combined state with heavy metals.
A great quantity also results from gas and oil purification processes. Only in the recent
years a new source of sulfur has come up with the flue gas purification plants. For long
distances liquid sulfur is transported by ship or railway, in some cases also by road trucks.
For medium range distances the transport via pipelines was first performed in Mexico [1]
during the 1950s. Although since then several liquid sulfur pipelines have been con-
structed, this way of transportation is still considered an exceptional case. This may be
because of the small range of temperature required or the specific physical qualities of li-
quid sulfur.

179
180

SULFUR PROPERTIES

With respect to its allotropy sulfur is very different from other elements. In the various
phases from the solid to the gaseous state, it undergoes a number of modifications which
are dependent on the temperature. The melting points of initially rhombic "a"-sulfur,
which transforms into monoclinic "8"-sulfur above 96°C (203°F) are ll3°C (235°F) and
l19°C (246°F), respectively. Sulfur is then a thin yellow liquid.
At a temperature of l60°C (320°F) it abruptly becomes viscous and dark while at a
higher temperature of about 250°C (48rF) it becomes thin again to vaporize at 445°C
(833°F).

In the thin molten state below l60°C (320°F) the predominant form is "Lambda" sulfur.
Viscosity as well as other physical properties of sulfur are illustrated in Fig.l.
It is evident that the range between ll5°C (239°F) and l60°C (320°F)has to be con-
sidered for the conveying of liquid sulfur by pipeline. This range is shown scaled-up in
Fig.2

--
0

>-
!:;133
u 8
w
e;0.700 "' 0
>

SULPHUR

Fig. I: Viscosity as well as other physical properties of sulfur in the range

of ooc to 300°C.
 181

1.55 10

u 1.40
0
u7.
Ill
Ill
Ol
'55.0
N
;:.1.25
.....
...
<[ >-
!i!UO ~2.
u
w
[J
u u
w
e; 0.95 "'> 0 e; 1.77+--+---4,.._--+--+--+--+---41----+--4
115 120 125 130 135 140 145 150 155 160
TEMPERATURE [ "Cl

Fig.2: Viscosity as well as other physical properties of sulfur in the range

of l15°C to 160°C.

LIQUID SULFUR PIPELINE VOIGTEI-BARENBURG

The sulfur pipeline Voigtei-Barenburg became already operational in 1976 and it has been
in operation without any form of interruption since then, namely for 16 (sixteen) years
now. At that time ILF was commissioned by NEAG (Norddeutsche Erdgasauf-
bereitungsgesellschaft) a subsidiary of Mobil Oil and BEB, with the planning of a convey-
ing system for 1,000 t/day of liquid sulfur. The entire system is shown in Fig 3.

TC
LOADING

THERMAL
DIL F'URNACE

PIG RECEIVER PIG LAUNCHER

Fig 3: Flowsheet Voigtei-Barenburg pipeline

182

The liquid sulfur is pumped at a temperature of approximately 140°C with the help of
steam-heated centrifugal pumps from the stack storage tanks at NEAG with an over-
pressure of approximately 2 bar to the head station of the 7 km-long pipeline. The pipeline
is installed underground and it is provided with a cover which is one meter thick. Steam-
heated booster pumps bring it up to the corresponding pressure level to facilitate it being
conveyed through the pipeline. Thermal oil is heated to a temperature of 160°C in a gas-
heated boiler and the temperature of the sulfur is increased to 155°C in a heat exchanger.
Then the sulfur enters the pipeline via a flowmeter and a quantity control facility. With a
planned quantity of 1,000t/day= 23.3 cubic m/hour and a 4 112" product pipeline the
flow velocities are 0. 7 m/sec. and the calculated pressure drop is 11 bar.

The system was designed in PN 25 (25 bar design pressure) taking the pressure surge
(which occurs when starting-up or with the sudden closing of a valve) into consideration.
A steam-heated heat exchanger is integrated at the end of the pipeline and its purpose is to
increase the temperature of the sulfur from 125°C to the loading temperature of 145°C.

Pipeline construction, heating and insulation

Whereas the pipes at the two stations are case heated and insulated, the conveying pipeline
is provided with electrical ancillary heating and is insulated in the steel-in-steel
system(Fig.4). The product pipeline, which is used to convey the sulfur, runs centrally in-
side a steel protecting conduit which has a plastic sheat. First of all the thermal conductor
is attached by means of circlips to the product pipeline. In order to improve the contact
the thermal conductors are additionally anchored by means of wire mesh over which there
is an aluminium foil which functions as a radiation shield. Next comes the actual insulat-
ing material which is 60 mm thick. An aluminium foil is also applied in order to
counteract radiation losses in the outer pipeline. The remaining annular gap is kept to a
mimimum in order to avoid convection. In addition to this subsequent to its having been
purged of any existing dampness and air, it is filled with an inert insulating gas e.g
nitrogen or freon at a low overpressure of 0.5 bar. The theoretically-established tempera-
ture curve over the cross section of the pipeline is shown in Fig.5. The theoretically-
established thermal losses with the temperature of the medium at 140°C are approximately
33 W/m and they are approximately 36 W/m at a temperature of approximately 150°C.
 183

CD Produktenrohr ® Mineralwolie
Product pipe Mineralwool
® Helzleitung @ Aluminiumfolie
Heating cable Aluminium sheath
® Orahtgeflecht ® Zentrier-Einrichtung
Wire mesh Centralizer
@ Alumlniumfolie ® Schutzrohr
Aluminium sheath Protection pipe
® PE-Isolierung
PE-Insulation

Fig.4: The design of the Voigtei-Barenburg pipeline

GROUND
PE-INSULATION
PROTECTION-PIPE - - - - - - - - .
GAS
MINERAL IJOOL
PRODUCTPIPE
SULPHUR

114.3
{ll244
{ll263
{ll273

150

p 100
L...l
c.:
....
:::1
<
b.t=16.411C
c.: 50 A=0.0322 IJ/MK
L...l
~
....
L...l
b.t=12.411C
0
). =1.74 IJ/MK

Fig 5: The temperature curve in the insulation

184

The pipeline is equipped with two different heating-systems, each having a heat capacity
of 50 W/m. The one system use Teflon-insulated thermal conductors, the other one use
mineral-insulated (Pyrotenax) thermal conductors. The whole heating system is divided in
six single heating circles at a voltage of 850 V.

Catering for thermal expansions

Special measures have to be taken in order to cater for changes in the length of
the product pipeline due to thermal expansion. Deliberations as to whether the forces
could be absorbed by pre-tensioning the piping train had to be dispensed with for struc-
tural/design reasons and the quality of the material. For similar reasons axial com-
pensators could not be used for this project. The most favourable and most economical
solution was the integration of a total number of thirty-two expansion elbows at intervals
of approximately 300 meters. Due to the fact that the protecting pipe remains at ground
temperature and thus does not need to be compensated for, the movement of the product
pipeline occurs in the elbows inside the protecting pipe which results in an increase in the
diameter of the protecting pipe. This increase results in a simultaneous heat loss due to the
convection in the gas chamber.

A compromise regarding the contradictory conditions was found whereby approximately

50% of the maximum expansion length was catered for by pre-tensioning which meant
that the diameter of the protecting conduit in the elbows only needed to be increased to
24". The thermal losses which can still be anticipated were compensated for by increasing
the heating capacity in the elbows.

Commissioning and Operating

For the commissioning the initial requirement was to evacuate the dampness which was
present in the annular gap due to the pressure test in the pipeline and due to the construc-
tion. Dried nitrogen was blown through the line in order to dry the annular gap and the
entire heating system was checked simultaneously. In addition to a number of faulty
sleeves which were directly due to the installing work, short circuits occured with the
teflon thermal conductors after having been drawn-on several times, also on the thermal
conductors themselves. Apparently this cable could not cope with the prevailing conditions
in 1976. Therefore a switch was made to pyrotenax cable whereby there were no further
difficulties apart from two faulty sleeves. The provision of a temperature profile over the
entire pipeline clearly indicates that a "frozen" line cannot be thawed-out again by means
of the electric ancillary heating facility.
 185

The uneven heat distribution meant that the melting point of the sulfur could not be
reached simultaneously over the entire pipeline. However due to the fact that the thermal
coefficient of expansion in the liquid phase is approximately double that of the solid state,
pressures could develop which could cause the line to burst. It is therefore necessary to
"pig" the line clear in the event of a longer downtime (i.e. for more than seven hours).
Subsequent to the pipeline having been dried the sulfur could be introduced into it
without any problem: Fig.6 illustrates a temperature profile for various throughput rates.
With the nominal throughput rate of 24 cubic m/hour the loss in temperature from NEAG
to Barenburg is 23.5°C, the thermal loss is approximately 44 W/m without ancillary heat-
ing. The values are approximately 15-20 % higher compared to the theoretically-
established values. The measured pressure drop of 10.7 bar on the line concides well with
the calculated values. With rates of between 75 % and 100 % of the normal capacity sup-
plementary heating is not required with a temperature drop of approximately 30°C to
20°C. With reduced throughput rates the end of the pipeline has to be heated periodically.

16

G 15
c
....
L.J
14
IX
,_
:::;)
13
<[
IX
L.J 12
a..
::E:
,_
..... 110
NEAG BEB
S1 S2 S3 S4 S5

Fig.6: The temperature profll.e for the sulfur pipeline for various throughput rates

There were no essential problems during the sixteen years of operation. In summer
1991 there first time was a heating defect at the Pyrotenax thermal conductors, which
could be repared by local measures.
186

THE LIQUID SULFUR PIPELINE- BUSCHHAUS-ALVERSDORF

The Buschhaus-Alversdorf liquid sulfur conveying pipeline is a partial system of the com-
plex flue gas sulfur extraction installation of the lignite-fired power stations Buschhaus-
Offleben. In 1987 flue gas extraction installations became operational both in the
Buschhaus power station as well as in the Offleben power station which is located ap-
proximately 6.5 km away.
These installations employ the Wellmann-Lord method. The end product is pure sul-
fur. Fresh sodium sulfite solution is enrichened with S02 in the Buschhaus and Offleben
absorbers. Sodium hydrogen sulfite is produced. The thus-charged solution is returned to
the absorber whereas the S02 rich gas is converted in a modified Claus installation to li-
quid sulfur which has the highest degree of purity. The annual sulfur production can
amount to 80,000 tons.
The last link in this process chain is the conveying pipeline and the loading station for
the disposal of the liquid sulfur which has been produced. In Buschhaus this is extracted
from a steam-heated 2,000 cubic m. storage tank and conveyed via a 4.5 km.long pipeline
into a heated 2,000 cubic m. storage tank in Alversdorf.

Planning the Pipeline

ILF in cooperation with Mannesmann Anlagenbau, Berlin, was commissioned with the
turnkey construction of the complete pipeline system.
It was decided that the pipeline should be a surface installation. The route runs for ap-
proximately 4 km together with the Offleben-Buschhaus sodium sulfite line and after
branches-off to Alversdorf.
The insulation is fully adequate to ensure that the pipeline can be operated without
having to enlist any supplementary heating although a heating facility is provided. In this
case the sulfur is heated up to 158°C by means of a heat exchanger and, with a normal
rate of flow of 5.8 cubic m/hour cools-off in the conveying pipeline to a minimum of
125°C. Attention was paid to the greatest possible redundance with regard to all the com-
ponents of the installation such as the supplementary heating for the pipeline, the station
heating or the pumps.

Construction Features of the Pipeline

Fig 7 shows the flow sheet for the conveying line. The liquid sulfur is extracted at a
temperature of approximately 135°C from the steam-heated storage tank (1) and pumped
by means of a controlled speed centrifugal pump (2) (a reserve) through a steam-heated
heat exchanger (3) in the conveying pipeline (5). The pumps as well as the entire station
piping are provided with thermal jacketing, insulation and they are heated by steam.
 187

A planned quantity of 5.8 cubic m/hour and a pipeline diameter of 73 mm results in flow
velocities of 0.5 m/sec. and a calculated pressure drop of 5.1 bar.

F"RDM
CLAUS
PLANT

THERMAL
'---r--OIL

~ ~
4 5 4

1. STORAGE TANK BUSCHHAUS 6. HEAT-EXCHANGER ALVERSDDRF"

2. PIPELINE-PUMPS 7. STORAGE TANK ALVERSDORF"
3. HEAT-EXCHANGER BUSCHHAUS 8 RECIRCULATION-PUMPS
4. PIG TRAP 9. RECIRCULATION-HEAT EXCHANGER
5. PIPELINE 10. LOADING PUMPS

Fig 7: The Buschhaus-Alversdorf Pipeline flowsheet

In the Alversdorf station a thermo-oil heated heat exchanger is integrated at the end of
the pipeline and it guarantees a loading temperature of 145°C. The sulfur is pumped into a
thermo-oil heated 2,000 cubic m storage tank (7) from where it is pumped for loading into
railway goods waggons with centrifugal pumps(lO).
The contents of the tank at Alversdorf can also be heated by means of the recircula-
tion pumps (8) by the second heat exchanger (9).
The entire station piping and the pumps in Alversdorf are equiped with heating
jackets, insulation and they are heated by thermo-oil.
188

Heating and Insulating the Conveying Pipeline

Whereas in both stations the piping is provided with the usual jacket heating and outer in-
sulation the conveying pipeline is provided with supplementary electrical heating. The
length is subdivided into four equal sections each of which is 1,150 meters in length. Op-
timum length of sections and optimum voltage level have been the result of extensive
studies. Two independent heating circuits are provided for each of these sections.
A mineral-fiber insulated single-core cable with a copper conductor and a copper-
nickel outer sheath is used as the termal conductor. The heating rating is 2 x 30 W/m dis-
tributed over three cables.
The pipeline is insulated by means of 25mm thick mineral wool and 90 mm thick
PUR shells. The outer protection is in the form of an aluminium casing (Fig.8).

1. PRODUCTPIPE

2. HEATING CABLE

3. MINERAL 'w'OOL

4. PUR-FOAM

Fig.8: The Design of the Insulation of the Conveying Pipeline

The sliding bearings and the fixed bearings which are necessary for supporting the
pipeline are special designs (which are patented by ILF) whereby careful attention was
paid to the minimizing of the thermal losses in order to avoid the formation of thermal
bridges (Figs. 9 and 10). As one can see there is no direct metal connection between the
product pipe and the support structure. The forces are transmitted via the insulation. The
nevertheless unavoidable additional thermal losses at the fixed points were compensated
for by winding the heating cables.
 189

1. PRODUCT PIPE
2. HEATING CABLE
3. RANGE SPACER
4. STEEL PIPE
5. FOAM GLAS
6. PUR-FOAM
7. PE-PIPE
8. CLIP
9. SHOE

Fig. 9: A sliding bearing

f::-:·.:·\·.:·J PUR-FOAM

~~~~ FOAM GLAS

• MINERAL -\,/DOL

.·.·.·.·.·.·.·::.·:.·:.·.·.·::::.1
.....................•
.·::: :.·.·:.·: :::.·:.·.·:.·.·: l
·····················
. . . . . . . . . . ·:::::::::.].

.·.·.·.·::::: :.·:: :.·.·:: :. 'J

................

Fig 10: A fixed-point bearing

190

Control and Supervision

The pipeline system has been equipped with a SCADA system. Besides standard function
a numerical model simulating the transient thermodynamic behaviour of the pipeline
system on-line has been implemented. This thermodynamic model calculates on-line and
in real time

the actual temperature profile and viscosity profile for the pipeline with
adaptive node resolution
the setpoints for the temperature controllers for each heating section
the survival time until the pipeline is to be flushed
monitors the insulation quality.

The model itself is adaptive thus eliminating automatically environmental influence.

Setpoint control by the model enables optimized operation with minimum energy con-
sumption. If the telecontrol limit between the model and the local controllers fails, the lo-
cal controllers automatically switch to local setpoint control with maximum thermal safety
margin.

Catering for the thermal expansions of the conveying pipeline

Expansion elbows were employed in order to cater for the changes in the length of the
product pipeline subsequent to thermic influences. The deformation is shown in Fig. 11.

LENGTH-SCALE 1: SO
OEFORMA liON-SCALE 1: 1 0

Fig.ll: Deformation at expansion elbows

 191

The Sulfur Pumps

Centrifugal pumps equipped with hydro-dynamic shaft seals are used for pumping the sul-
fur (Fig.12).
The pressure drop at the shaft penetration point decrease during the operation of the
pump by the back buckets of the impeller wheel and the two or three additional relief im-
pellers. During downtime the penetration point of the shaft is sealed by means of a ring
valve. This valve is opened automatically by means of a centrifugal-force elevating facility
when the pump starts-up and it is closed again when the pump is switched-off. The elevat-
ing facility pushes the shaft in the axial direction together with all the components which
are mounted on it. An additional gas seal has been provided in order to prevent gases and
steam from escaping. Contrary to a sliding ring seal the sealing faces of the ring valve do
not slide on top of each other during the time the pump is in operation. The sealing faces
are off-set from each other by from 0.3 to 0.4 mm and are thus protected against mechani-
cal abrasion.
The result of this is that small amounts of leckage can occur during the time the pump
is being either started-up or stopped.

Fig 12. A cross-section through the pump [2]

192

Experiences gained through the operation

The system has been in operation for five years now without any signs of bigger
problems. Pipeline downtimes are avoided for the most part by the forwards and reverse
pumping in order to avoid local overheating. In one instance the leckage which occurred
when starting-up the pumps resulted in larger quantities of sulfur escaping because the
pumps did not attain sufficient operating speed.
Another problem was caused initially with the heating of the sulfur tank at Al-
versdorf. Here the tank lost more heat than was intended because the bottom of the tank
had not been adequately insulated. Here the intended possibility of heating the contents of
the tank by means of a heat exchanger in order to reach the necessary loading temperature
of 145° C if necessary proved itself.

THE SULFUR STORAGE FACILITY AT BRAKE

NEAG operates a liquid sulfur tank storage facility in Brake near Bremen and which is
provided with subsequent facilities for solidifying the sulfur and for marketing it in either
a solid or a liquid state. Initially the storage location was laid-out in 1980 to cater for a
10,000 cubic m storage tank. In 199011991 the installation was expanded by an additional
10,000 cubic m storage tank. The planning of the entire expansion was carried-out by
ILF. The system is shown in a simplified form in Fig 13; The liquid sulfur from the
natural gas purification phase is supplied by railway goods waggons and it is pumped in a
liquid state into the storage tanks. Each storage tank is equipped with its own group of
pumps so that the liquid sulfur can be pumped to a processing installation (rotoformer) or
a ship's loading station.

The 10,000 cubic m.Storage Tank

The subsoil in the alluvial land area is not particularly suitable for supporting such a heavy
structure as a 20 meter-high, 25 meter-diameter sulfur storage tank. The compressive load
per unit area of 3,6 · lOS N/m2 is comparitively high at a density of 1800 kg/cubic m.
For this reason the tank was mounted on 76 (seventy six) concrete piles each measuring 28
meters and provided with a continuous foundation plate (Fig 14).
 193

1 STORAGE: TANK 10.000 " 3 4 ROTOFORMER

2 RAIL UNLOADING PUMPS 5 SHIP LOADING PUMPS
3 ROTOFORHER PUMPS

SULFUR SHIP
PE:LE:TS LOADING IPIPE-
STORAGE FACILITIES !BRIDGE
I
i

Fig 13.: The flowsheet of the sulfur handling installation at Brake

SULFUR
STORAGE TANK

Ill IZI61CM Ill Ill

lfl ~ lfl
025.56M
PILE LENGTH• 28M
PILE RASTI!:Ro 2.8x2.8M

Fig 14: The tank foundation

194

Special attention was given to the build-up of the insulation of the bottom of the tank.
Whereas the wall and the top of the tank were fitted with 160 mm thick mineral wool the
bottom was built-up as follows (Fig 15):

- concrete foundation
- compensating (levelling) layer
- adhesive (bonding)
-foam glass 160 mm (two layers) with intermediate bonding layer
- sand compensating (levelling) layer
- the bottom of the tank

TANK BOTTON

ADHESIVE <BONDING>

f:<:>l SAND

~ FOAM GLAS

[II[[] MINERAL WOOL

Fig. 15: The insulation of the tank bottom

As one can see the load exerted by the tank is transmitted directly via the foamed
glass to the foundation. The sensitivity of the foamed glass called for extreme care with
regard to the design and the installing of the bottom of the tank.
 195

The design provides for a thermal transmittance of 55 W/m 2 - a comparatively very low
value. The internal temperature of the tank is 140°C. The total heat loss of the tank is 160
kW. The tank is heated internally by means of heating pipes. The top is additionally
heated by means of a heating coil and the tank wall remains unheated.
The entire heating of the tank and the piping is provided by means of thermo-oil at a
temperature of l60°C.

The Pipes
The two sulfur storage tanks are about 200 meters apart. All the pipes between the two
tanks run on a piping duct over railway lines and dyke installations. The sulfur pipes have
casings and they are heated by means of thermo-oil. The construction of the 250 m long
piping duct required very high demands on everyone involved. The piping duct was fabri-
cated section-by-section on the ground (from elbow to elbow) and the pipes were fitted.
Several cranes were used to position the partial sections onto the supporting columns after
which they were bolted together. This demanded precise measurements and tight manufac-
turing tolerances. Then the final step consisted of welding the elbows in the knee points
(Fig 16.)

Fig 16: The Piping Duct while being positioned on the supporting columns
196

The Pumps
Pump development took a further step forward with the development of magnetically-
coupled pumps for pumping liquid sulfur. This type of pump has a completely sealed
pump body. (Fig. 17).
The pump shaft is equipped with permanent magnets. The end of the shaft is encased
in a stuffmg box which completely seals the body of the pump. The counter magnets at-
tached to the coupling shaft are located on the outside of the stuffing box. Thus the driv-
ing power is transmitted exclusively by means of magnetic force. There is a considerable
development of heat in the stuffing boxes due to the effects of scatter (approximately 10%
of the pump efficiency). This heat has to be dissipated by a corresponding rinsing with the
medium.

Fig 17. A cross section through the body of the pump (without heating jacket) [3).

Commissioning
The commissioning of the installation in the summer of 1991 was an unexpectedly tedious
affair.
 197

The main problem was caused by the sulfur pumps. The sliding bearings on the end of the
pump shafts are lubricated with the medium, i.e sulfur from the stuffing box. Undersized
lubricating grooves resulted in damage being caused to the bearings and in the complete
breaking down of all the new pumps.
The pumps have been running perfectly since then subsequent to the manufacturer
having replaced the bearings and after the diameters of the oil lubricating holes were in-
creased.
The operating staff were shocked by the loud sliding/slipping noises when the tank
was initially filled. The base of the tank moved by about one centimeter radially outward
on the foundation due to the heating-up of the bottom plate and the wall of the tank. The
empty tank, as a resonance element, produced a most impressive thunder! But the tank is
still in order.
Some sulfur leaks at the flange connections were corrected by exchanging the seals.
The originally intended alumimium-asbestos seals displayed too little elasticity in order to
be able to compensate for the tensions due to thermal expansion which is inherent in
heavily-branched pipe systems. Help was obtained here in the form of a more elastic It-
seal (elastomer-asbestos).
Apart from these initial difficulties the installation has operated without any problem
for about half a year now.

REFERENCES
[1] Planung und Bau der Fliissigschwefel-Transportleitung Voigtei-Barenburg, S.
Lechler, D. Kossert, K. Zehl; Erdoel-Erdgas-Zeitschrift, 93 Jg., Juli 1977
[2] Chemie - Kreiselpumpe Typ RCE, Fa. Rheinhiitte
[3] Chemie - Kreiselpumpe Typ RMKN, Fa. Rheinhiitte
 CHARACTERISTICS OF LIGNITE ASHES TRANSPORT FROM STEAM
POWER PLANTS THROUGH PIPELINES

ZVONIMIR VUKELIC
University of Zagreb
Faculty of Civil Engineering
Kaciceva 26, Zagreb 41000, Yugoslavia

ABSTRACT

Lignite ashes from a steam power plant is not an useful raw

material and the cheapest way of its removal is to deposit it
into some natural depression or abonded river bed. Hydraulic
transport of such ashes has its specific characteristics and
therefore it is in a certain extent different from the
hydraulic transport of similar materials. For instance, a
hydraulic transport of ashes is performed through pipes from a
steam power plant to a depositary where ash deposits and the
cleared up water does not flow to some river but over a
pumping station and pipes flows to the basin of plant and
again serves as a medium for the transport of ashes. There is
a supposition that ash belongs to the group which contains
under (10%) of (CaO) and (MgO) and that there is no deposit of
limestone in the pipes.

INTRODUCTION

Regarding the taken protective measures of river courses which

used to be undertaken at the flows of waste waters from the

199
 200
objects of steam power plants one may say that the problem is
very complex because the treatment plants of those waters are
very expensive and produce nothing. It is a reason that most
of different technical solutions are incomplete. Concentration
of one steam power plant of 1,000 MW and more on one place
near a natural recipient may have many bed effects on the
natural characteristics of that recipient. For an illustration
in this paper is taken the solution of hydraulic transport of
lignite ashes from the objects of steam power plant near
Prishtina. Lignite is from the adjacent dig and it is a very
basic material with about (47%) of (CaO).
The previous drainage of ashes was solved as a hydraulic
transport through the pipeline of (D = 250 mm). After the
deposition of ashes the cleaned up water was flowing to
Sitnica river. During the flowing of water the hydroxides
Ca(OH) 2 and Mg(OH) 2 were flowing into the river, too. The
discharge of river is about (15.00 m3 /s) but in the summer
period, practically there is no flowing of water in the
riverbed.
This problem of water drainage from the steam power plant is
successfuly solved by a closed circuit as shown in Fig. 1. The
waste water does not flow to the river now but it returns into
the basin of the plant with water pumps and there it serves
again as a means of transportation of ashes.

STEAM POWER
PLANT

PUMPING
STATION

Figure 1. New solution of hydraulic transport of lignite ashes

 201

PROBLEM OF LAYERS IN THE PIPELINE

Ash reacts on sour components of flowing water and an effect

of that is a deposit of different chemical compounds on the wals
of pipeline. One of the reactions is according to the known
formula

(1)

and the made limestone separetes on all solides.

The closed circuit improved the situation about the deposit
even though (pH) factor is increased a little because there is
a mixture of operating water and raw ash. The change of (pH)
value is spproximately from 6 to 10 according to the
corresponding values of ash from 1 to (20%).
Observations show that deposit was less than before and that
the maximal thicknees of deposit was (5 mm). Besides, the
structure of new deposit was quite different from the previous
one.
During the winter days, once or twice a year, the pipeline is
empty in order to clean its walls. The deposit on the walls is
broken because of the changing of temperature and for their
removal it is enough to use a hammer and water. There is an
opinion, the cleaning may be with the composed air, water,
balls and jute as shown in Fig. 2.

VALVE

I JUTE
AIR PRESSURE i RUBBER SPONGE
PIPE BALL
Figure 2. Cleaning of the pipeline
202

The bad cleaned pipeline makes difficulties about the

beginning of repeated work because the laged deposit brakes
the flow of fresh mixture and it may be accompanied by
increasing of pressure in the system.

NECESSARY QUANTITY OF WATER AS A FUNCTION OF THE PROBABILITY

One may suppose that the maximal discharge is a product of an

average discharge and a common coefficient of unequality of
the drainage

( 2)

If (qm) is a maximal discharge per unit of the transported

material and (T) time of the pouring, a probability of the
pouring of water and ashes may be

(3)

The probable number of the quantity units of ashes is equal to

a product of the number of units and the probability of one
unit

( 4)

Depending on the production and organization of the work there

are some deviations of this number and a real number (k) will
be

k =s Pe ::!: r

where (r) is a dimensionless apsolute deviation.

Due to relation between (r) and the reduced dimensionless
deviation (t) in the further composing the following may be
obtained
r
t = ( 5)
 203

qe 1 - Pe ( 6)

Qn = k qm (s pe + r) qm ( 7)

Qn = s qf + r qm ( 8)

Qn = s qf + q m t v2 s p e qe ( 9)

Qn qf (s + t0 qm
s ( - - 1))
qf
(10)

K = 1 +
n
t
0Ys qm
( - - 1)
qf
( 11)

Parameters (qf) and (qm) may be derived as a function of the

production and type of the drainage. Parameter (t) is, in the
other hand, a quality of the function of the system as shown
in Table 1.
TABLE 1
Some characteristic values of the reduced deviation (t)

t 1,30 1,50 1,70 2,00 3,00 3,80

F(t) 0,9032 0,9332 0,9554 0,9772 0,9986 0,9999

Curves of (Qn) and (Kn) are shown in Fig. 3.

Qn = f(s)

~~=-------------------------------------~>$
Figure 3. Curves of the maximal probable discharge and the common
coefficient of unequality in the function of number (s)
204
DISCHARGE OF MIXTURE

Mixtures of ash and water are anomalous liquids which have

certain characteristics between plastic bodies and liquids and
are in accordance with the laws of anomalous liquids. Energy
losses along the length of transport (L) in the turbulent
regime of flow may be derived by the modified pattern of
Darcy-Weissbach

(12)

Coefficient of friction (~ ) may be expressed by a sum of the

p
coefficient of water friction (~ 0 ) and deviation from (k·f)

1\.p =1\.0 + k f (13)

where (k) - apsolute roughness and (f) is given by the quotient

1p -'"'(
(14)
f = Ta -T
where (T) - volume weight of water, CTa) - volume weight of
ashes and CTP) - volume weight of mixture.
Using the formula of Chezy the discharge of mixture (Q) is

Q = A v = A cp '(Hi; = A Vii; VR i 0 ~P (15)

where (A) - inside surface of pipes, (V) - velocity of flow,

(C ) - coefficient of Chezy, (R) - hydraulic radius, (g) -
p
gravitational constant, (i 0 ) - slope of energy grade line for
water and (ip) - energy grade line .for mixture.
If one is introduced into Eq. (15) the known formulas

and R = 1j"d (16)

discharge (Q) may be expressed by

Q = K' Fo (17)
 205

where ( K' ) is modulus of mixture which is equal

K'=
12 d2
(18)
125 n2 "Yp -I
+ k
dl/3 "Ya -T
In this equation (n) is coefficient of roughness and (d) is
inside diameter of pipe.
The quantity of ashes from two phases of the plant amounts to
(380 10 6 kg/year) or about (18 kg/s for 6,DOO hours) of the
work during one year.
Results of the laboratory experiments show that the
corresponding quantity of water for the hydraulic transport
is (0.112 m3 /s). The built in pump has discharge (Q = O.llm 3 /s),
about (0.092 m3 /s) of water and (0.018 m3 /s) of ashes for the
flow through one pipeline (d = 250 mm).
Total losses that are necessary for a connection of water and
ashes and for an evaporation amounts to (0.017 m3 /s) in the
winter and up to (0.030 m3 /s) of water in the summer period.
This quantity of water is compensated on a part from the spoil
area to the steam power plant.
For the return of water to the basin of plant there are two
pumps of (0,06 m3 /s) and two pipelines (0 200) and (0 250).
The average diameter of ashes is (0.066 mm) and volume weight
(19900 N/m 3 ) . Values of (k) and (n) are (0.025 mm) and
(0,01 m- 113 s).

CONCLUSION

It has been demonstrated that some technical solutions may

improve function of the hydraulic transport of lignite ashes
through the drain system. It is also of interest to say that
the developed functions may be used with succes for the
determination of necessary quantity of water at the
dimensioning of such drain systems. The analysis showed good
agreements with the observed losses of energy and slopes of
energy grade line for different velocities and discharges in
the pipelines.
206

REFERENCES

1. Gancev, I. and Kalcev, I., Hydraulics and Hydraulic

Transport, Tehnika, Sofia, 1964.

2. Jonuzi, A., Extraction of Lignite Ashes with Organic

Solvents, Collection of Papers of Electrical Industry,
Prishtina, 1967.

3. Popovic, N., Characteristics of the Gas Spoil Area of

Lignite, Collection of Papers of Electrical Industry,
Prishtina, 1967.

4. Radovanovic, M., Waste Waters and Deposit of Ashes from

Steam Power Plants, Journal Vodoprivreda, Belgrade, 1972.

5. Sewing, T. T., Pumping Concrete and Concrete Pumps.

6. Vasilev, K. S., Hydraulic Transport, Tehnika, Sofia, 1976.

7. Vukelic, Z., Maximal Flow of Productive Waters in Pipes

of Sewerage Calculated by a Method of the Theory of
Probability, Journal Water and Sanitary Technique, Belgrade,
1979' 4' p. 35-41.
8. Vukelic, Z., Thermal Pollution of Recipients and Protective
Measures, Sisak, 1982, Bulletin of the Symposium, p. 144-
-158.

9. Vukelic, Z., Hydraulic Transport of Lignite Ashes from

Steam Power Plants as a Function of the Drain System, IAHR,
XX th Congress, Moscow, 1983, Volume VI, p. 424-532.

lO.Vukelic, Z., Modulus of Discharge of the Hydraulic Transport

of Ashes from Steam Power Plants, Journal Gradjevinar,
Zagreb, 1988, 40, 8, p. 377-380.

ll.Vuletic, B., Complex Treatment of Waste Waters of Steam

Power Plants, Collection of Papers of Electrical Industry,
Prishtina, 1967.
SESSION C: OPERATION
 NETWORK SIMPLIFICATION BY EQUIVALENT
MODELLING FOR OPTIMAL OPERATIONS

DR YUE-CHUN CHEN
Network Engineer, Thames Water Utilities Ltd
&

DR BRYAN COULBECK
Emeritus Professor, Leicester Polytechnic

ABSTRACT
The paper gives a detailed description of an equivalent network modelling
procedure. Evaluation of the model uses statistical analyses based on the
well-known nodal equations. Validation of the model. for different
operational conditions, shows that the simplified equivalent network can
replace the original network for simulation and control purposes. The
paper also indicates the significant reduction in system complexity as a
result of the equivalent network modelling procedure.

INTRODUCTION
In order to achieve efficient operations of water supply distribution
systems of ever increasing complexities and power demands, it is
considered inevitable to have to adopt optimal control using computers.
Generally, a water distribution network is composed of hundreds or even
thousands of pipes and thus is a very complicated system. Moreover,
time-varying consumers demands occur at different points throughout the
network, few of which are measured hourly or even daily. As discussed in
[10] , it is impractical to evaluate optimal controls using a detailed
model of the actual network. Therefore, certain reasonable and effective
simplifications are necessary.
Typical aggregation involves the cancellation of small diameter pipes and
replacement of parallel or serial pipes by single equivalent elements, to
form a skeletonized network. However, this would still contain a large
number of pipes and nodes and a single network "balance" would require the

209
210
calculation of all pipe flows and nodal pressures: this would be difficult
or even impossible for optimization purposes. Consequently some
complicated networks have been simplified by eliminating a large part of
the network to leave only a few pipes. In this case the characteristics of
the original network might not be properly reflected and customers service
requirements as well as some important components in the network cannot be
taken into accound directly. Therefore, through the presentation of the
well-known nodal equations and based on statistical analyses, an
equivalent network modelling procedure is developed in this paper.

METHODOLOGY
The essence of equivalent network modelling is to deal only with major
components in a network (such as reservoir nodes, pressure monitoring
nodes and nodes which link a booster or a control valve) and to connect
them with each other and to pump stations by fictitious pipes in order to
construct a simplified equivalent network from the original detailed
network. The constructed equivalent network should be able to replace the
main aspects of the original network for simulation or optimization
purpose.
To further explain the concepts of the equivalent network modelling
precedure, take a simple example as shown in Fig. 1, [1] . For the
original network, drawn with solid lines, the well-known nodal equation
will be:
(1)

where
N = total number of nodes in the network
N(i) = subset of j connections to node i
q11 = pipe flow from node i to j (positive flow)
u1 pump flow into node i
y1 = consumption flow out of node i
PI

,------- -------- ------- ,. ..;,./. Tl

~
I
I
....
.,. ""/
' .... ' , .,. ""
~
.,. -
""
,
/
/

"
I /
I
.,.
", ,
I I
I
' .... ~
~ /

I '~
,.''
"PP2,"
-
/
I
I ~-
~/ /
I, ~
~
I
,..; ::;1 ~~
' /
PPI Pl,l'2 --water sour ces
', I /
Tl --storage tank
--
I I

'' I /
PPI ,PP2 pressure
O:/r2
I /

'~ monitoring points

Fig. 1 A Water Supply Network

 211

In Fig. 1 all the dashed lines which link the main variables of the water
supply distribution system will constitute its equivalent network, which
can be further simplified. The equivalent network will be determined by
the statistical analysis of correlation, ie, if two variables of the
systems are correlated, a dashed line will connect these two variables,
otherwise there will be no dashed line between them.
The procedure of correlation analysis is stated as follows:
Since a water supply distribution system operates somewhat periodically,
nodal pressures will fluctuate around their average values. In the long
ter, nodal pressures can be assumed to follow a Guassian distribution.
Therefore, nodal pressures can be selected as statistical variables of a
Guassian distribution. If any pair of the two statistical variables are
not correlated, this means they are independent from each other as well.
From the above discussion the correlation analysis can be performed as
follows.
Suppose r is a sample estimate of the population parameters [8], then:
p (2)

r = (3)

L
where 0';, 1/ (L-1) 'Z (u, - u> (v, - v>. sample covariance.
i=l
L
o-;
2
1/ (L-1) :l:. (u, - u)z' sample variance.
i=l
2.. L
frv 1/(L-1) ~ (vi - v), sample variance.
i=l

L
u 1/(L) :I: ui
i=1

L
v = 1/ (L) l:. vi
i=l

L length of sample
212
pressure value of node u (i 1, 2, ... L)
pressure value of node v (i 1, 2, ... L)

To test hypothesis H: 1' =0 versus the alternative A: 7 1 0

It is necessary to calculate

t r-0/ (S,)= r(L-2)'· 1 / ( (1-r 1 ) 1 · 1 ) (4)

and reject H if t~ t 11 .oi.JzHHl

or if t:E- t 0 -ClfzHHl

t 11 -.I./Zl\L-2! can be found in the t - distribution table; where ~100

is the significance level.
This correlation analysis will be performed between each pair of nodes.
At the same time, a practical engineering understanding of the particular
network should be taken into account in the final construction of the
equivalent network.
After the correlation analysis, assume the final equivalent network is as
shown in Fig. 2.

~~----------------~
I'
I
//
//
Tl

I I //
I I ///
\ I ////
I I // /
I \ PP2 ; /
I \ ~ //
I \ I /
I \ I //
• \ I /
' \ I /
PPl ' , \ I //
' \ I/
' I I/
(j'r2

Fig. 2 Equivalent Network of Fig. 1.

Using the operating data of a given water supply distribution system, it
 213

is possible to obtain the nodal consumptions and pipe resistances of its

equivalent network so that the equivalent network can be dealt with as a
real network. This will be demonstrated through the following deduction.
For the equivalent network, the nodal equations will still be:
1: q 11 = u1 - Y1 tfiE N (5)
jeN(i)
But where
N = total number of nodes in the equivalent network
N(i) = subset of node j connections to node i.
q11 = the "flow" through an equivalent pipe from node i to node
(positive flow).
u1 = pump flow into node i.
y1 equivalent consumption flow out of node i.
{y1 i = 1, ..• N-Ns
Y, =
0 i = N - Ns + 1, .. N
where Ns is number of source nodes
Using Manning's formula [9]:

Ah 11 = h1 - h1 = R11 q111
then

(6)

where
= head drop from node i to node j
=head at node i and node j, respectively
resistance coefficient of an equivalent pipe from node
ito node j.

The N nodal equations in eqn. (5) can be combined into a vector equation
as follows:

l7)

where
~ = connection matrix of dimension N x Np (whose elements are 0
or 1)
!;[ vector of dimension Np with elements q11
~ = vector of dimension N with elements u1
214

y = vector of dimension N with eleqdqds y1

Np = number of "pipes" in the equivalent network

For M sets of measurements, there are M vector equations:

!I(') - yi•J (8)

If Rij changes slowly, the following approximations may be valid:

then

q (1)
II

k = 2,3, ... M

or

(9)

in which

substituting eqn. (9) into eqn. (8), whjch can be rewritten as:

(10)

!!:(') - yi•J
 215

where ~!tl = ~; ~' 11 = ~ . lY 1J!kl], k = 2,3, ... M

l'f;/ 11] = diagonal matrix of dimension N, x N, with elements Y;J!tl.

Defining a1 as the fractional consumption rate at node i, such that:

N-Ns
a1 = y 1 /Y0 and I a1 1 (11)
i=1
or

N-Ns
~ (a 1 X Y0 ) = Y1
i=1
where

Y0 = total demand flow.

Combining eqn. (11) with eqn. (10), leads to the partitioned
matrix-vectorequations as follows:

(12)

where

~ = [1 1 .... 1] a row vector of dimension N-Ns

a,_.,] r

y (kl
0
D (l)
Y,
y (k)
~
y (k)
D

0 ... 0.- ...... 0

0 .. 0 0 N x (N-Ns)
Finally, the parameters to be estimated are:
216
~ can easily be estimated by using the method of least-squares [5] :

Let

r-ACkl
-----1------
0-
I y Ckl
1-'-1

1
I
B
-
]

k = 1, 2, .... , M

then

K' l&' • &' • ... • &'1

and let

k 1,2, .... , M

then

Eqn. (12) becomes

Kt ~
& ~

~ =
~. ~.

or K~ =~ (13)

Define an error vector £= (2: 1 ,~ 1 . . . . , ~.>'and let

The estimate ~ of & will be chosen in such a way that the criterion, J,
given by:

M
J =I~t 1 =s.'! = (~- K ~)'(~- ~. ~)
i=1

(14)

is minimized.
Differentiate J with respect to ~ and equate the result to zero to
determine the conditions on the estimate~ that minimizes J. Thus
 217

aJ 1
----1
~~ I ~

This yields

!! ! i = !! ~
from which~ can be solved for as
(15)

A program coding EQUNET was produced in Fortran 77 to implement this

algorithm.
However, in some cases of the study, some "flows" occur from lower
pressure nodes to higher pressure nodes without the existence of pumping,
and also some negative consumptions appear on non-source nodes. To avoid
these problems a constrained least-squares formulation needs to be solved.
The problem is:

min J =~ 1 ~
e
(~ - ! ~)I {~ - ! ~) (16)

s.t. i = 1,2, ... N, + N-Ns

But note that in problem (16), e 1 ~1 ~ is a scalar, then ~~~~~ =

(~1 !1 ~) 1 = ~1 Xe and ~~~ is a constant. Therefore, problem (16) is
equivalent to

min J 1 = ~~ ~ + -i- ~~ !! ~
e
(17)
1,2, ... N, + N-Ns

where C1 = -~1 !.

In fact, problem (17) is a Linear-Quadratic Programming problem, which can

be solved with the routine E04NAF' in the NAG FORTRAN LIBRARY [7].
Therefore, a program coding was produced in Fortran 77, which employed
E04NAF, to cater for this algorithm.
With the estimate ~. substituting q111 into eqn. (6), the "resistance" R1J of
the equivalent network can be computed; substituting ~ into eqn. (11), the
distributed consumption y1 can also be computed.
So far, the equivalent network has been thus constructed. Note that the
218
parameters in the model have a clear and definite physical meaning; and
are consistent with the conventional theory of water network analysis. In
particular the method does not rely on the assumption of proportional
loading as compared with the macroscopic model [4] or the piece-wise
macroscopic model [3], which cannot be guaranteed to apply to most water
supply systems.

APPLICATION
The model validations were performed on a network as shown in Fig. 3, l3].
The 48 half-hourly sets of data simulated with GINAS [2] are used for the
parameter estimation of its equivalent network.
The results of the nodal correlation analysis are listed in Table 1 and
the equivalent network is shown in Fig. 4.
Using the program coding EQUNET for the unconstrained least-squares
procedure, the computed results are as summarized in Table 2.
From Table 2., it may be noticed that the flow of pipe 8 and the flow of
pipe 10 are negative, where the pressures at node 7 and node 3 are always
higher than those at node 5 and node 1, respectively. This means that the
flows are from lower pressure nodes to higher pressure nodes, without the
existence of pumping, which are therefore not acceptable.
Consequently, the program coding ENCQP for the constrained least- squares
procedure has been employed.
 219

TABLE 1 RESULTS OF CORRELATION

Analysis between Nodes (~=0.01)

NODE PAIRS t, _,./1 Computed t Correlation Result

------------------------------------------------------------------------

Node1/Node2 2.6868 36.5346 YES

Node1/Node3 2.6868 3.4001 YES
Node1/Node4 2.6868 -1.4525 NO
Node1/Node5 2.6868 3.1621 YES
Node1/Node6 2.6868 4.8032 YES
Node1/Node7 2.6868 -1.4525 NO
Node2/Node3 2.6868 4.2317 YES
Node2/Node4 2.6868 -0.8713 NO
Node2/Node5 2.6868 3.9889 YES
Node2/Node6 2.6868 4.6607 YES
Node2/Node7 2.6868 -0.8717 NO
Node3/Node4 2.6868 -3.6431 YES
Node3/Node5 2.6868 -0.5052 NO
Node3/Node6 2.6868 7.9354 YES
Node3/Node7 2.6868 -3.6456 YES
Node4/Node5 2.6868 8.7352 YES
Node4/Node6 2.6868 -6.5673 YES
Node4/Node7 2.6868 3186.6274 YES
Node5/Node6 2.6868 -1.7295 NO
Node5/Node7 2.6868 8.7337 YES
Node6/Node7 2.6868 -6.5712 YES
-------------------------------------------------------------------------
220

The computed results are summarized in Table 3. and Table 4, respectively

and their related equivalent networks are shown in Fig 5. and Fig. 6,
respectively. The c-values used in Hazen-Williams relationships in Fig.5
and Fig. 6 are computed by assuming all pipe lengths to be equal to 1 m,
all pipe diameters to be equal to 100mm and by use of Hazen-Williams
relationships [9].
The results in Table 3. are obtained by using the simulation data from
0. 00hr to 8. 00hrs, when all pumps are in use. Whereas the results in
Table 4 are obtained by using the simulation data from 8.30hrs to 23.00hrs
when one pump is off (the lower pump between node 122 and node 6).
In order to validate the equivalent network model, GINAS has been employed
to perform simulations over the equivalent network under the same
operational conditions as were performed over the original network. Both
of these results are listed in Table 5 to give a comparison; this shows
that the accuracy of the equivalent network model is satisfactory.
Validation of the suitability of the equivalent network for different
operating conditions is investigated by changing the initial head of
reservoir 4 from 182.00m to 183.00m and scaling the total demand by a
factor of 1.3. The simulation results, performed for the original network
and the equivalent network, respectively, are listed in Table 6. The
largest difference in Table 6 is 2.1m, the relative error being 1.17%,
which shows that the equivalent network is suitable for simulation and
control purposes.
Further, using the parameters for the equivalent network obtained for the
period of 8. 30hr to 23. 00hrs to simulate over the period of 0. 00hr to
8.00hrs, give the results as tabulated in Table 7. The largest difference
is 2.89m, the relative error being 1.63% which is still acceptable. It
seems, at least for the network studied, that the equivalent network is
not too dependent on the variation of pump combinations when used for
simulations. In other words, it is only necessary to use one set of
parameters of the equivalent network regardless of the variations of pump
combinations.
There are only 10 nodes and 13 pipes in the equivalent network. Whereas
there are 36 nodes and 47 pipes in the original network. The reduction is
about 72%, which is very effective in such a type of a non-linear system.
The reduction would be much more in more complicated systems.

CONCLUSIONS
It has been shown that the equivalent modelling procedure is theoretically
strong and is consistent with the conventional theory of water network
analysis. This should lead to a reliable model. Model validation results
show that the method suitable for practical simulation and control
purposes and that the reductions in system complexity is quite effective.
 221

ACKNOWLEDGEMENTS
One of the authors is grateful to his previous PhD supervisor, the late
Professor Q. Yang, since part of the work was done under his guidance at
Tongji University, China. This paper is published to memorize the late
honourable professor.

REFERENCES
1. CHEN, Y. c. (1988} Simplifications of Wat.er Supply and Distribution
Systems for the Optimal Operations, in Coulbeck, B. and ORR, C.H.
(Editors}: Computer Applications in Water Supply, Vol. 2, Systems
Optimization and Control, Research Studies Press, U.K.
2. COULBECK, B. and ORR, C.H. (1985) Computer Control of Water Supply:
Applications Documentation for GINAS (A Graphical Interactive Network
Analysis and Simulation Program), Research Report No. 16, Leicester
Polytechnic, UK.
3. COULBECK, B. and ORR, C.H. (1986) Computer Control of Water Supply,
GINAS Applications Workshop Training Guide, Research Report No. 40,
Leicester Polytechnic, UK.
4. DEMOYER, R., Jr. and HORWITZ, L.B. (1975} A System Approach to Water
Distribution Modelling and Control, Lexington Books, USA.
5. HSIA, T.C. (1977} System Identification,Least-Squares Method,
Lexington Books, USA.
6. LIU, S.R. and DUAN, w.z. (1986}, A Study of Modelling Technique on
Water Supply Systems, Proc. of the Conference for Applications of
Automatic Technology, China.
7. NAG FORTRAN Library Routine Document (1983}, NAGFL11: 2072/0. MK11.
8. OSTLE, B. (1974} Statistics in Research, Second Edition, The Iowa
State University Press, USA.
9. YANG, Q. (1957} Design and Computation of Water Distribution systems,
Constructional Engineering Press, China.
10. CHEN, Y.C. (1989}, Analysis, Modelling and Optimal Control of Water
Supply and Distribution Systems,PhD thesis, Leicester Polytechnic, U.K.
222

::
g "c
.....
u

.. .-·.•.
~

~«:'~
..... u
~~~
..... u - -c
"~
..... u

..... c •

..- .,------
~
w

_...
..-

_...
... t .
s •
.. ;l'!
~

' J

......' -
0
"~N
H 0

J Q u
""'ex:G
....
~t;o ~
E-

.''
~

.,~ .....
~ o.__ ~Q~ z
E-
~ ·~ tr.

...
;:.LI
E-
!I: «:
•'...:!· o•,g "'AU
.. NO r:...
~;£: .:~- 0
,HI \I ~4 ~ ,u
!l"' z
0

...
H
E-
«:
~l:: ex:
:::>
"'r:...
H

z
0
u
a...
'!!J~
~~~
"'r:...
..
H

.
l
i'
i•
I . .~·~=
.
~

i 1-0:!i
I '!: .. .....
:::: I
I

I
rlst
~

0 0 1.. .s--· D ...

 223

,.,
~·

...
0

'-
0

t
i
....,
Gl
z
u ....,
c
."
N
~

N
.....,Gl
.....>
:: ;_
=
'"":

..;:.::-. .!.
J; ,. I

....! i
!

iI i
i. , ;;:l ":i
 N
N
.j::o.

TABLE 2 RESULTS FOR EQUIVALENT NETWORK MODELl.ING

OBTAINED FROM PROGRAM EQUNET

Kqivalent Pipe Plow Qp(i) 1/s

Qp(1) Qp(2) Qp(3) Qp(4) Qp{5) Qp(6) Qp{7) Qp(8) Qp(9) Qp{lO)

1.061 0.428 18.531 16.323 2.563 8.083 117.60 -4.782 3.587 -1.805

TABLE 2 (Continued)

Kqivalent Pipe Plow Qp(i) 1/s I Fractional Consu•ption Rate a(i)

Qp{11) Qp{12) Qp{13) Qp(14) Qp{15) a( I) a(2) a(J) a(4) a(5)

0.103 1.577 3.916 2.257 186.441 0.05600 0.08063 0.09825 0.79862 0.02902
 TABLE '3

Eqivalent Pipe Plow Qp(i) 1/s

Qp(1) Qp(2) Qp(3) Qp(4) Qp(5) Qp(6) Qp(7) Qp(8) Qp(9) Qp( 10)

o.o 0.415 48.983 2.494 o.o o.o 115.966 o.o o.o 0.705

Eqivalent Pipe Resistance RE(i) •/(l/s)2

RE(l) RE(2) RE(3) RE(4) RE(5) RE(6) RE(7) RE(8) RE(9) RE(lO)
Oo 3.2487 0.0091 0.0275 00 00 0.0007 00 00 2.3209

TABLE 3 (Continued)

lqivalent Pipe Plow Qp(i) 1/s Fractional Consu•ption Rate s{i)

Qp(l1) Qp(l2) Qp(l3) Qp(l4) Qp(15) a(l) a(2} a(J) a(4) a(5)

o.o o.o o.o 3.757 161.366 0.00379 0.00237 0.24869 0.72037 0.01321
---------------------------------------------
Equivalent Pipe Resistance RE(i) •/(l/s)2
---------------------------------------------
RE( 11) RE(l2} RE(l3) RE(l4) RE(l5)
00 00 00 0.0196 0.0008

N
N
VI
 [:3
0'1

TABLE 4 RESULTS FOR EQUIVALENT NETWORK MODELLING

FROM PROGRAM ENCQP (8.30 IIRS TO 23.00 IIRS)

Eqivalent Pipe Flow Qp{i) 1/s

Qp(1) Qp(2) Qp{3) Qp(4) Qp(5) Qp(6) Qp(7) Qp(8) Qp(9) Qp(lO)

7.811 2.099 38.755 29.103 24.680 o.o 115.783 o.o o.o 5.017

Eqivalent Pipe Resistance RE(l) m/(l/s) 2

RE(l) RE(2) RE{3) RE(4) RE(5) RE(6) RE( 7) RE( 8) RE(9) RE(lO)

0.0234 1.7843 0.0124 0.0004 0.0029 00 0.0007 00 00 0.2556

TABLE 4 (Continued)

Eqivalent Pipe Flow Qp(i) 1/s I Fractional Consu•ption Rate a(i)

Qp( 11) Qp(l2) Qp( 13) Qp( 14) Qp(l5) a(l) a(2) a(3) a(4) a(5)

o.o o.o o.o 53.609 110.228 0.03778 0.05587 0.22549 0.66376 0.01537
---------------------------------------------
Equivalent Pipe Resistance RE(i) •/(l/s)2
-----------------------------
RE(11) RE(12) RE( 13)
----------------
RE(14) RE( 15)
00 00 00 0.0035 0.0007
 I."' I
D-100
_!_ ;;oJe nunber in Fi.a 4,4(.- 11 1 24 1 52 etc.) c-o
(p) Pii'e nur:Dar (n•l,2,) etc:.)
(I)
~-.oJal conawtptlon (i•l,2,l,4,S) ~1
••
Yz yl

( 10)
L•l
J•IOO
C•l,O
Exi.uina Pump
·)
A -1. IJxiO
8 -2.72:r.IO-l
c 122.148 1,•1
n.. IOO

L!•IOil Y;
c-9.91

Yi,

\'SP
t. -I.OE-1
~ lg)~w

Fig S Equivalent Network of Fig 3 (0.0 hr to 8.0 hra)

N
N
-:1
 N
N
00

.fod• number in ri;:;4.4(n-17,24.52 etc.) L•l

..!. D-100
(n) 11 1[)• numlhH (n•1,2,3 l:fte.) ..:•'l.R'> I
1--------"(•:.c>_ - --------
yi .lodal consumj)tion (i•I,:!,1,4,S) -' 12

yI
( 10)
L•l
n-100
C:•7. ROP.
"
Exiuins PI!IJP
" -l.llxiO
L•l
/ 8 -2.717:10-J ll•IOO
\ C L!2.148

\ '!.4

11-10[)
Exi.stine ru"!!p(orf) '··
C•Rl.71'J

Q- 1>•1011 '1:.
l 7 f C•)O.'l77
\\..!_) teo
\:::::::..____vsr(!.1dablc ~pccJ J_:unp)
A -I.Of.-J
iJ -0.90f>2
c 2.JJ. 79

Fig 6 Equivalent Netvor•k of Fig 3 (9.0 hro to 23.00 hro)

 229
TABlE 6
VALID\TIOO OF F9JIVAlml' t£1\olJRK lfiEILII'C

tO>AL HfAllS (H)

nxlet nxle2 .-) nxle4 .rn.s

v~ vSENt vs:~~ v~ vsm V!Bf v~ vsm v~ v~ v~ v~ vs.t.~ VSI·li v~H v~ vsrw VSI-li V9.N vsm
---------------------- -------------------
o too.s7 too.s7 tst.t6 tat.J6 tat.n tst.89 182.00 t82.oo IBI.SJ IBJ.BJ zm.'i6 2tn.% tt~2.1l6 Pn.n 56.67 56.86 69.33 69.J1 96.ts 96.18
1 180.88 t8t.l9 181.38 181.69 182.85 182.27 182.27 182.27 181.82 181.92 21n.tJJ 2114.01 J92.1J l'n.I9 56.88 56.89 69.33 69.33 96.18 96.18
2 181.37 181.74 181.69 182.05 184.76 184.40 182.17 182.17 182.04 182.00 21¥~.25 2<Yt.23 192.21 11Jl.28 56.89 56.90 69.33 69.33 96.18 96.18
3 181.00 181.90 181.99 181.90 186.50 186.94 182.28 182.28 182.19 182.2] 204.12 :J)4.'l6 192."JI 192.18 56.90 56.91 69.33 69.13 %.18 %.18
4 181.98 182.37 182.15 l82.75 186.91 188.44 182.41 182.41 182.12 182.17 20',,H() 2(Y,.68 192.42 1'12.50 'j(J,'JI 56.92 69,]] tf.I.H 96.18 <J6,J8
5 I82.Q4 182.83 182.24 182.79 186.70 187.77 182.53 182.53 182.44 182.48 204.50 2UI1.6S 192.53 192.60 56.92 ~.92 69.33 69.1J 96.18 96.18
6 182.03 182.50 182.28 182.81 186.10 186.47 182.65 182.65 182.54 182.59 20'1.66 2!Yi.56 192.61 1'12.11 56.91 56.9) 69.]) 69.33 96.18 96.18
1 181.50 182.02 182.20 182.57 184.50 183.85 182.76 182.76 182.62 182.64 204.2H 204.J'J 192.71 IIJ2.fl] 56.'11 56.93 69.33 69.33 96.18 96.18
8 un.9J 100.11 t81.66 18t.lo 179.os 179.01 182.82 182.82 182.61 tR2.so 201.21 201.% t92.H6 t9z.n %.91• 56.94 69.33 69.13 96.18 96.18
9 179.28 178.75 (80.71 179.90 172.85 113.45 182.78 182.78 182.45 182.18 191.47 192.18 192.75 192.77 56.9"1 56.94 69.33 69.1196.18 96.18
10 176.20 175.85 179.66 179.17 175.50 176.14 182.73 182.73 IR2.26 182.18 191.79 192.22 192.71 192.71 56.9"1 56.94 69.3)69,JJ 96.18 96.18
ll 172.70 173.17 178.86 179.05 179.24 180.12 182.70 162.70 J62.l3 182.29 l'J2,!i5 192,1•3 I'J2.f!H I'J:l.10 ~IL.•n 56.93 69.33 fl9.)} 96.16 96.16
12 171.12 l7J.79 179.19 181.60 IM.65 1Jf.I.J7 162.69 162.69 182.69 182.17 192.HI 192.58 11Jl.b1 192.69 56.93 ')6.93 69.33 69.31 96.18 96.18
13 174.52 17~.52 179.13 180.03 181.44 180.99 182.70 182.70 182.26 182.44 191.05 192.}() 192.()8 192.70 56.9] 56.93 69.)] 69.]3 96.18 %.18
14 t74.oo 174.62 179.ro too.so 181.54 181.29 182.71 to2.71 182.25 182.45 19"1.09 JCJ2,70 192.h9 1•n.11 56.91 56.93 69.JJ 69.11 96.18 96.18
15 174.19 174.51 179.63 180.49 181.37 180.69 182.72 182.72 182.26 182,1+5 19"1.0'• 11.12.6'1 192.611 1'12.72 56.'f") 56.9} 69.1'1 69.")1 %.18 %.18
16 177.70 178.12 100.65 181.:1J 181.48 UIJ.84 182.13 182.1) 182.40 182.51 19).(~ I'Jl,IH I'Jl,JI 1'12,/] 56.93 56.94 69.31 6'J,JJ 96.18%.18
17 180.22 l&l.57 181,46 181.55 181.46 181.33 182.76 182.76 182.52 182.57 I'H.OI I'Jl.81"1 J'J2,7J IIJZ,75 5(,,9] 56.94 69.)3 69."11 %.18 96.18
18 179.98 tOO.I)j 181.32 181.29 Ult.81 UJJ,75 182.78 182.78 182.52 182.55 t'J2.00 l92.tll 19:.!.76 1'1:!..71 56.93 56.94 69.1169.11 96.18%.18
19 179.44 179.48 181.09 l8J.oo tao.n 180.t2 182.79 182.79 182.49 182.51 192.h7 1•n.n I'Jl.Jll 1'12.78 56.91+ 56.9/t 69."Jl M.n %.18 96.tfl
20 100.45 179.98 181.53 181.26 179.96 UJ>.67 182.M 182.00 182.57 182.56 1'12.f.IJ l92.fU 1'.12.71 19l.19 56. 114 56.94 69.)") 69.".ll 96.18%.18
21 181.36 180.67 181.96 181.64 180.13 180.73 182.83 182.81 182.65 182.M 192.78 t•J2,92 1'12.79 I•Ji.RI 5(,.94 S(,,qt, 61),11 69.TI %.18 9f~ol8
22 lfll.64 181.12 182.13 181.89 100.85 181.29 182.fl6 IH2.81 IR2.70 IH2.6a 111:!.911 lll!. 1Jtl I'J:!.H! 192.iH 'lh.'l'' 'X~o9t, W.11 69.H %.18 96.18
2l 181.77 181.50 182.24 l82.11 181.56 182.26 182.91 J82. 4H 182.7& 182.7':1 l'l"l.21t I'J],IIH 192.lill 1'11../i'J ~..c,.9') ~>f,.IJS (Jt).l) 69."B 96.18 96.18

* V~ - ~lue ~tnuL"ltt..>d over Q_riginal ~twork

+ V~ - !_alue ~inulate::S over the .§_quivalent _!!!twork

TABlE 6
VALID\TIOO OF Fgl!VAIENT NIDOlK lfiEILJK;
FOR DIFFERENf OPERATIOOAL (lH)lTIOOS

row. JF.AOO CH)

-----------------------
nxlel nxle4
-- .-) .- 5

0 180.73 179.37 181.70 180.35 179.78 179.53 183.00 183.00 182.78 182.69 21H.M 2f)l',.tl') I'J2,1Jl• 191.02 ~••% 56.96 69.31 69.11 96.18 96.18

2 181.85 182.22 182.39 182,70 183.96 183.27 183.11 l8l.ll 182.95 182.96 'l.()f,,J.l 2CVi.41t 193,(~ 19"l.ll 56.91 56.97 69.33 69.3196.18 96.18

4 182.66 183.63 182.94 184.83 186,86 187.07 183.12 183.12 183.21 18).26 2U4.•Y~ 204,81 191.21 19).:kJ 56.96 56.99 69.1169.13 96.18 96.18

6 182.57 184.aJ 182,97 184.52 185,63 184.93 183.52 181.52 183.39 11::11.41 21Yt.!Jb 2U4.fl1 I'JJ.41 I'J1,4J 'll.fl.l 51.00 6'J,J1 &9.)] %.18 96.18

8 1&;.61 179.23 181.82 180.89 175.56 171.86 183.65 183.65 183.35 181.12 21H.OI 21JJ.64 l'H.l2 191.59 S7.01 57.01 69.31 69.33 96.18 9&.18

10 172.82 171.25 178.45 177.34 J7t.l1171.25 181.44 183.44 182.68 182.~ 191.61! 192.15 191.J!t J'J"I..J6 56.9':1 56.99 69.]] 69.33 96.18 96.18

12 16"1.77 169.33 177.62 179.20 179.00 177.59 183.31 181.31 182.46 182.75 192.71 192.h"l 19"1.22 191.24 56.98 56.98 69.33 69.31 96.18 %.18

14 169.00 171.28 178.19 11:1).33 100.11 178.03 183.25 183.25 182.50 l82.tlJ I'U.'Jr> l'J2,76 191.11 liJ"J,I'J %.98 56.98 69.31 69.B 96.18 %.18

16 175.00 176,47 179.81 UIJ.70 180.00 178.95 183.19 183 .. 19 182.65 182.HII 192.81J lq2,H"I 191.11 I'IJ,Jl 56.97 56.97 69.1] 69.:H 96.18 96.18

18 178.64 178.23 U!0.81 Ull.57 178.96 178.69 161.16 181.17 182.76 182.77 142.61 1'12.78 t•n.IO I'Jl.l2 56.9} 56.97 69.:n 69.3196.18 96.18

20179.32 178.02 181.07 lfll.43 177.66 178.46 186.11 18).11 182.17182.(19 19.!.)4 1'12.72 I'J.I.U'l I'J:I.Uh ~lfJ,'U 56.'JI 69."11 b9.1l 96.18 96.18

22 18L.J1 179.97 181.98 181.40 178.83 lfll.62 183.11 183.10 182.1J(J 182.84 I'Jl.hl 1')2.98 I'JJ.lV! l'f).U') 'll.,,tJ/ 56.% f>'L11 hi),J) %.18 96.18

-~------------------·------··--·------- . ·-·----··------------

+ VSEN- _yalue _!lmllated aver- the _!lui valent ~work

230

TABlE 7
VALIDI\TION OF Fq.IIVAUNI' I'£'I.W)RK lflEil.Itr.
roR DIFffiRENf OPERATIONAL amrl'IONS
(the influence of JUI1P cond>imtions Is lt;nored)

I<DAL lllAffi (M)

- - - - - - - - - - - - - - ----------------- - - - - - - - - - - - - - - -
1u.le l ••xJe 4 axle 5

o uo.n 179.96 tst.7o tat.J2 179.78 tat.48 taJ.oo taJ.oo 182.78 taz.l'l Ztn-'+4 201.19 J9z.w, 192.97 56.% s6.% 69.11 &q,J) 96.ts 96.ts
2 181.85 181.84 182.39 182 .. 38 183.96 183.09 183.11 JSJ.ll 182.95 182.99 204.2l 201.55 t9J.lV+ !9J.06 56.97 56.97 b9.JJ 69.JJ 96.16 96.18

4 182.66 182.92 182.94 183.07 186.86 183.97 183.32 183.32 183.21 181.21'! 2lYi.94 201.fKI 191.·11 l'll.Z'i 56.98 56.98 69.31 &9;J) 96.18 96.18

6 182.57 182.67 182.97 183.03 185.63 183.64 183.52 183.52 18].39 183.44 2()l!.h6 210.72 193.41 19}."1 ')7,0] 57.(X) 69.1169.)] 96.18 96.]8

8 18J.6l 177.8) 181.82 U0.47 175.46 179.14 18J.6S 183.65 163.15 UB.IS 201.01 2m.41, JIJ].S2 11H.'>4 ':17.01 57.01 69.31 69."13 %.18 96.18

* V:ni- y_alue l!_lnulato.l over .2_rlglnal _!!eLWOI"k

+ VSEN- yalue ~IIIIJIJJted over Lhe §9ulvalenl ~twork

KNOWLEDGE BASED SYSTEMS IN OPERATIONAL SCHEDULING OF WATER
DISTRffiUTION NETWORKS

MARTIN LIKEMAN
SFK Technology Ltd.
Amstral House, Mill Court, Wolverton Mill, Milton Keynes.
MK12 5QP. United Kingdom.

ABSTRACT

An aspect of operating water distribution networks that has received much attention in the
past is pump scheduling. This paper reviews the existing work in pump scheduling and
discusses its merits and drawbacks. As a result of the limitations of the existing methods, a
new approach to pump scheduling has been developed at SFK Technology that is founded on
Knowledge Based Systems (KBS) techniques. The KBS approach has been incorporated into
a piece of scheduling software called PUMPLAN. The KBS approach and its benefits are
discussed with reference to studies performed using PUMPLAN. SFK Technology is leading
a SPRINT collaborative project funded by the European Commission to develop sophisticated
operational control schemes and install them in European water undertakings. A general
framework for operational control schemes of this type is presented and the role of KBS
techniques in the framework is considered.

INTRODUCTION

Until recently, the water industry lagged behind the other major distribution utilities
(electricity and gas) in the sophistication of its operations scheduling and control. Amongst
the reasons suggested for this are the complexity of water networks, the difficulty of
measuring certain parameters accurately and the high cost of telemetry equipment [3]. In the
past, operating procedures have incorporated large safety margins and have been concerned
mainly with the maintenance of security of supply to customers rather than the economic
operation of plant.

231
232

Over the last few years, a number of specific scheduling techniques for water network
operation have been devised. In certain instances, these techniques have enabled operational
cost savings of over 10% to be realised. As it has been estimated that the UK water industry
is spending in excess of £100 million per annum on electricity for pumping water, it is
unsurprising that these scheduling techniques have aroused considerable interest.
The recent privatisation of the 10 former water authorities along with the introduction
of stringent regulatory legislation has added further impetus to the efforts of the water
industry to rationalise their business. The requirement to meet tougher limits (for example
on water quality) under increasing pressure for economic network operation highlights the
need for sophisticated scheduling software. Scheduling software is also needed for situations
when maintaining water quality at acceptable levels or controlling leakage to avoid over-
abstraction rather than reducing pumping cost is of prime importance. Pump scheduling
software has mainly been used in a standalone environment to date. The software is run at
infrequent intervals and the schedules used to support the operators in controlling networks.
Advances in telemetry and sensor technology along with the positive experiences gained with
the early pump scheduling software has lead to a general interest within the industry to
linking pump scheduling software directly to the telemetry and control systems and using it
for the day to day control of the network.

Pump Scheduling
Pump scheduling is the process of deciding how to operate the pumps in a water distribution
system so as to satisfy all the constraints on the system in a cost effective manner. The most
common system constraints include:
choosing pumps with sufficient driving head to overcome the hydraulic resistance of
the system and supply the consumer demand of water.
maintaining water levels in storage reservoirs within specified upper and lower limits.
The upper limits are imposed to avoid the danger of flooding and the lower limits to
ensure there is sufficient reserve of water in the reservoirs to maintain supply for a
reasonable time if the pumps malfunction.
pressures at critical points in the system are sometimes constrained to remain below
certain values to reduce loss of water from the system as a result of leakage. At other
 233

points, pressures are constrained to remain above certain values to ensure that there
is sufficient head to supply isolated consumers.
water supplied to consumers is often blended from a number of sources of differing
quality. It is necessary to ensure that water supplied to consumers satisfies the criteria
laid down by the European Commission and the Drinking Water Inspectorate.
environmental and climatic changes have lead to the introduction of strict limits on
source abstraction in certain areas. Pump operating schedules must comply with these
regulations.

One of the simplest and most effective ways of pump scheduling is to organise the
pumping and use of reservoir storage to utilise the structure of the electricity tariff. Current
industrial electricity tariffs in the UK typically comprise a fixed component for rental of
metering and generating equipment, a charge based on the number of electricity units used
during the charging period (typically one month) and a charge for the maximum half-hourly
demand during the middle of the day over the charging period. The unit consumption charge
is calculated at a different rate during the night and the day, with units used in the night
costing about a third as much as those used in the day. By pumping heavily during the night,
one is able to fill the storage reservoirs inexpensively. This allows demand to be partially
met from the reservoirs during the day, reducing the expenditure on day time units and
maximum demand charges.
Other methods that may lead to more cost effective network operation include:
ensuring that the reservoir level and pressure constraints correspond to the operational
needs of the network. Lower level reservoir bounds are traditionally set very
conservatively, and a risk analysis might identify that they can be lowered without
jeopardising system security. This might lead to further savings from an empty-fill
scheduling strategy.
optimising source usage so that the highest quality sources are used within the
abstraction limits in order to reduce marginal costs of treatment.
using up to date site test data to assess relative pump performance rather than nominal
manufacturers' curves for plant which has been operating for a number of years.
in a system with flow control valves, operating the valves in such a way that throttling
234

losses are minimised and the water is routed away from high loss parts of the system
where possible.
using the most suitable tariffs for the operational needs of the network. As a result
of the privatisation of the UK electrical power generation industry, electricity
companies are offering flexible and competitive tariff arrangements to their large
consumers, such as water undertakings. A choice between tariffs can be made by
assessing the effect of different tariffs on pumping costs using scheduling software

In addition to choosing an efficient pump schedule that meets the system constraints,
there are a number of operational criteria that should be satisfied by the chosen schedule.
Many of these criteria are specific to the network being scheduled. Some generally applicable
criteria are:
avoiding switching pumps on and off very frequently as this would accelerate the wear
on the switch gear.
distributing the water evenly across the reservoirs as if one reservoir becomes depleted
and then there is a pump breakdown, it may be impossible to route water from other
parts of the system to the vicinity of the depleted reservoir.

Traditional Methods of Pump Scheduling

Before the benefits of pump scheduling became widely known, water networks were operated
with security of supply assuming prominent importance and little consideration being given
to the economics of operation. A typical method of control was to turn pumps on and off
when reservoir levels attained preset values. In this way, the pumping could be controlled
by how depleted of water the system was. More sophisticated manual scheduling methods
that incorporate economic considerations have been subsequently devised, but for networks
of a realistic size with a number of constraints, it is impossible to produce effective schedules
without the aid of a computer. Two main categories of computational pump scheduling
algorithms have been produced - those based on derivatives of linear programming and those
based on dynamic programming.

Linear Programming Techniques: As a consequence of the non-linear nature of the

hydraulics of a water network, one cannot directly apply linear programming techniques to
 235

the pump scheduling problem. Linear programming schedulers [5], [8] usually involve a
preliminary stage where a number of hydraulic analyses are performed using network analysis
software to determine pump outputs at anticipated demand conditions. This enables a bulk
flow transfer model to be constructed from which a linear programming problem in terms of
pump operation times may be assembled. After the problem has been solved, the schedule
suggested may be checked for hydraulic consistency using network analysis.
Linear programming approaches are well suited to networks in which the non-linear
nature of the hydraulics does not predominate (for example due to the topography of the
network). Linear programming can converge quite rapidly for large problems, and there are
a number of robust linear programming packages on the market. For problems where
variations in demand and reservoir level have significant effect on the network hydraulics, an
unanticipated change in demand is likely to invalidate the assumptions made in setting up the
linear programming problem. In addition, the process of preliminary network analysis can
be time consuming and require a great deal of expertise to decide which pump options should
be considered. Pure linear programming is unable to schedule problems where the tariff
contains a maximum demand component. The techniques required for coping with a
maximum demand charge (such as setting up a mixed integer problem and solving it using
branch and bound search) are generally considerably slower than pure linear programming.

Dvnamic Programming Techniques: Dynamic programming techniques are concerned

with problems where a sequence of decisions has to be made, and as such appear well suited
to pump scheduling. A typical approach used in pump scheduling [4], [10] is to precalculate
estimated reservoir level profiles based on predicted demand and the structure of the
electricity tariff. The problem is discretised by dividing the day into a number of time
intervals and problem states are defined by splitting up the allowable range of level for each
reservoir into several discrete values. At each time interval the results of applying each pump
option to each possible problem state are evaluated and, using weighting functions based on
the deviations of the reservoir level states achieved from the target profiles, transition costs
between the states are calculated. The problem is then solved using standard dynamic
programming techniques.
Dynamic programming schedulers are in principle more widely applicable than linear
programming schedulers as there is no inbuilt restriction in the form of the hydraulics that can
236

be considered in calculating the transition costs. However, in order to gain sufficient

accuracy, it is often necessary to divide the reservoirs into a large number of discrete states
which can cause excessively long run times. Solution time is also exponential in the number
of reservoirs in the system and experience has shown that systems with more than one or two
reservoirs cannot be practically scheduled using these techniques. In order to combat these
problems, a number of decomposition algorithms (e.g. [2]) have been devised that split up a
large scheduling problem into a number of smaller ones that are tractable to dynamic
programming techniques. The decomposition algorithms are difficult to set up and are
restricted in applicability to a limited range of systems.

KNOWLEDGE BASED PUMP SCHEDULING

Although some worthwhile results have been obtained with the Operations Research based
pump schedulers described above, it is apparent that they suffer from a number of drawbacks
which limit their applicability. The major problem with these techniques is that they are only
suited to problems posed in a specific mathematical form. This makes it necessary to
translate the hydraulic equations and data for the pump scheduling problem into the required
form, which is time consuming and requires various simplifications and assumptions to be
made. It is particularly difficult to include qualitative information on operating the network
from the experience of the operators. This information is often neglected, which can lead to
results being produced that the operators do not trust. As a result, scheduling software has
not fulfilled its potential in a decision support role.
The assumptions made in the Operations Research schedulers only hold over a limited
range of operating conditions. In the face of unexpected events, such as a pipe burst or a
pump malfunction, the assumptions can break down. New assumptions then have to be made
and the problem reconfigured which is time consuming and disruptive.
In an attempt to overcome these problems, a new approach to pump scheduling has
been developed at SFK Technology which uses Knowledge Based Systems (KBS) search and
scheduling techniques. The philosophy behind the KBS approach is that the scheduler needs
to be able to access all the pertinent information about the water network in an explicit form.
Rather than trying to "optimise" a coarse model of the network which may bear little
resemblance to the current network status, the KBS strategy aims to use full and accurate
 237

information to find a very efficient schedule in an acceptable run time. It is unwise to spend
a great deal of time trying to optimise a mathematical model that has considerable
discrepancies from the real-life scheduling problem due to assumptions made in the
modelling, the inaccuracy of data measurements and the continuously changing state of the
environment.
The KBS strategy has been implemented in a piece of PC software called PUMPLAN.
The aims of PUMPLAN were to demonstrate the viability of KBS scheduling on a range of
small to medium sized networks, to allow off-line scheduling of both pumps and flow control
valves and to allow the system data (such as pump characteristics and demand profiles) to be
updated easily to reflect changes in the network.

The PUMPLAN Search Strategy

The search strategy of PUMPLAN is based on a technique called heuristic search. There are
initially a vast number of possible schedules to choose between, and it is clearly impractical
to investigate all of these. A number of guidelines called heuristics are used to direct which
schedules should be chosen for further investigation, and which should be neglected. These
heuristics are based on general problem solving knowledge, system constraints and operational
experience. An example of such a heuristic is that 4 changes of pumping a day at a pumping
station should be allowed in the scheduling. Analysis has shown that fewer than 4 changes
a day allows insufficient flexibility for calculating efficient schedules, whereas allowing more
than 4 changes a day makes the scheduling problem considerably more complex and does not
cause significantly cheaper schedules to be chosen. Having used a number of heuristics to
reduce the number of plausible options to a few schedules, each of these schedules is costed
accurately using a full hydraulic model of the network and the least cost schedule that obeys
the system constraints is chosen.

A Case Study Using PUMPLAN

One of the studies undertaken with PUMPLAN was on a distribution system in Oxfordshire
depicted schematically in figure 1. The current system supplies 50,000 domestic consumers
and several industrial demands and is being extended by building a new pumping station
(pumping station 1). The extended system will contain 4 reservoirs, 4 remotely controlled
flow control valves and 2 pumping stations. The study involved calculating pump and valve
238

schedules for the extended system under predicted demand conditions.

The first part of the study was to construct and calibrate a suitable scheduling model
for the system. Although a reasonably accurate hydraulic model of the network is at the
foundation of PUMPLAN, it would be inadvisable to include needless information as this
would slow down the search without any benefits of further accuracy in the scheduling. As
the demand can only be predicted to limited accuracy, it is reasonable to make some
simplifications to the network model used. These simplifications involved aggregating local
demands and pipes into single components, and verifying the accuracy of the simplified model
using the Flowrnaster network analysis package (Flowrnaster is a suite of hydraulic analysis
programs that has been developed at BHR Group Ltd. and is marketed by Amstral (UK) Ltd.).
The scheduling model was then entered into PUMPLAN and schedules produced under
a number of different operating scenarios. These were compared with equivalent results
obtained by operating the network under reservoir level control. The schedules obtained from
PUMPLAN cost on average 17% (£15,000 p.a.) less than those using reservoir level control.
Studies on other systems investigated with PUMPLAN yielded similar savings in comparison
with current operating costs (where available) or simulated operating costs under traditional
methods of control, with one study predicting savings of 36%. It is dangerous to draw
sweeping conclusions from these results as the level of savings obtainable depend on how
much excess pumping capacity and reservoir storage is available in the network and upon how
efficiently the network was previously operated. However, these results are very encouraging
in comparison with the 10% typically quoted for the Operations Research methods.
 239

L::::. Pumping Station

1::::1- Reservoir
*Flow Control Valve P.S.l
1>t- Non Return Valve
0 Demand

Figure 1. PUMPLAN Scheduling Model of Distribution System in

Oxfordshire.

An Appraisal of KBS Scheduling

The experience with PUMPLAN indicates that the KBS scheduling approach has a number
of advantages over the Operations Research techniques.
The KBS scheduling approach is not tied to a set mathematical representation of the
network. This allows realistic models of the network to be used which produce
reliable and cost effective schedules.
a detailed hydraulic model allows the explicit representation of flow control valves,
and a number of constraints (such as pressure and quality constraints) that would
involve a lot of manipulation to incorporate into the Operations Research schedulers.
operational experience may be incorporated into the scheduling, which makes the
software more readily acceptable to the operators. Using the software enables one to
investigate the different "rules of thumb" used by the operators and see which ones
are useful and which have no sound engineering basis.
in the event of a schedule being invalidated by unexpected system conditions, the KBS
approach is suitably robust to produce a new schedule rapidly as it has reference to
240

all the network information. In the Operations Research techniques it is normally

necessary to reconfigure the scheduling model from scratch which can be time
consuming.

PUMPLAN has proved the viability of KBS scheduling for small systems. For larger
systems, the techniques used in PUMPLAN will need some modification to avoid excessively
long run-times. The field of Distributed Artificial Intelligence [1] has yielded a number of
useful techniques for decomposing large KBS problems into smaller and more manageable
ones. These techniques appear to be conceptually simpler and applicable to a wider range of
pump scheduling problems than the decomposition methods associated with dynamic
programming schedulers. In addition, a number of KBS techniques such as Constraint
Directed Search [6] have been developed for the manufacturing industry that enhance the
power of heuristic search in scheduling. These support many types of constraint and facilitate
the generation of heuristics by providing a structure to the problem based on the constraints.
Although it was easy to generate heuristics for the relatively simple cases considered with
PUMPLAN, it would be much more difficult to find suitable scheduling heuristics for larger
and more complex networks and constraint sets if the problem is not structured formally.
As a result of the increasing pressure on the water industry throughout the European
Community to maintain high standards of operation, the European Commission has decided
to fund a project under the SPRINT initiative to further develop operational scheduling by
transferring KBS scheduling technology from other sectors such as the manufacturing
industry, and to disseminate the results throughout the Community. The project contains
partners from Denmark, Portugal and the UK and is being lead by SFK Technology. The
project will involve a formal requirements analysis to determine the scheduling needs of a
range of Community undertakings. It is anticipated that these requirements will lead to the
specification and production of a scheduler based on heuristic search, but utilising constraint
based reasoning and Distributed Artificial Intelligence techniques to make it applicable to
complex networks.
 241

THE FUTURE OF OPERATIONAL CONTROL IN THE WATER INDUSTRY

The majority of the scheduling systems produced to date have concentrated on off-line
scheduling and relatively simple models of the network and constraint sets. The water
industry's requirements are now moving towards on-line control with stricter legal regulations
necessitating a more extensive list of constraints to be considered. A possible framework for
a sophisticated on-line control scheme is depicted in figure 2. The constituent parts are as
follows.
a framework within which all the functional modules will fit. It is suggested that this
should be Knowledge Based, due to the great flexibility in terms of the type of
knowledge that can be represented and control structures offered.
a core scheduler to perform scheduling under "normal" operating conditions. This will
need to be able to handle sophisticated, multi-constraint systems and produce accurate
schedules rapidly. The scheduler must be robust under a wide range of operating
conditions and be able to incorporate local operating knowledge and expertise for
individual networks. The scheduler may need to be able to schedule not only single
speed pump operation, but also other plant such as variable speed pumps and remotely
controlled valves (flow control and pressure reducing). The scheduler must be able
to react rapidly to changes in network conditions to enable the schedules to be kept
up to date. It has previously been argued that a KBS approach should be taken in the
scheduling, although it is possible that using linear programming in conjunction with
KBS scheduling could be of benefit for certain types of network.
in order to perform scheduling over a period in the future, it is necessary to predict
the consumer demand over that period. Current research [7] indicates that hybrid
methods incorporating moving average trend analysis and KBS rule inferencing
techniques can give fast and accurate predictions for "normal" and "unusual" days
(such as bank holidays). The trend analysis is fast and accurate for "normal" days,
but loses accuracy on "unusual" days. The rules are able to incorporate complex
information, especially about the "unusual" days to improve the accuracy of the trend
analysis.
to enable the scheduler to operate properly, it needs to know up-to-date and accurate
information about the system. This requires sensors to be distributed throughout the
242

network, and information to be transmitted to the scheduler via a telemetry system.

Due to the cost of sensors, it is currently impractical to place them at every point
where measurements are desired and instead a few critical values are measured and
the rest of the values are inferred from these. Current work on state estimation has
relied on minimum norm approaches such as least squares approximation [9]. A
particular problem has proved to be the detection and remedying of false sensor
readings. This diagnosis problem will require the use of qualitative knowledge to
identify typical errors and cures, and will need mechanisms to reason with inaccurate
and uncertain data. A number of Knowledge Based techniques have been specifically
developed for these sort of applications and combining these with existing minimum
norm state estimation methods will be of benefit.
there will be certain emergency situations that lie outside the "normal" conditions that
the core scheduler will be able to react to. An example might be a major pipe burst.
The system will need to be able to detect these situations and take appropriate action.
The action will probably consist of alerting the operators with an alarm that identifies
the fault and suggests remedial action that should be taken. Once again, qualitative
reasoning is needed so Knowledge Based diagnosis techniques are suited to this task.
They will need to infer the type of fault from the input data, to determine the severity
of the fault and hence the time available for remedying it (some faults will be so
severe that economic scheduling around the fault is impractical and immediate
measures must be taken; others allow limited time for scheduling meaning that instead
of taking panic measures with little regard for the economics of operation, a more
sober analysis is possible which combines normal cost criteria of operation with the
need for rapidly repairing the fault). A suitable remedy for the fault must then be
inferred, which in the case of a severe fault would probably require consultation of a
rule base of faults and associated remedies.
a number of off-line facilities will be needed such as tariff bargaining, risk analysis
(e.g. to determine suitable reservoir level bounds) and "mimic" runs for testing out
various scenarios (for example for training new operators in network operation or
testing the validity of existing operational criteria). These will be based on calls to
the core scheduler, but will need a knowledge based control strategy for organising
these calls (based on tariff bargaining and other application-specific expertise).
 243

data and knowledge bases for the network model, operational information, archives
of previous conditions, emergency actions etc. will be needed to support the functional
modules. The underlying Knowledge Based strategy allows information to be
represented in a suitable, explicit form to allow information to be readily browsed and
updated and used by the software effectively.
a graphical user interface will be needed to allow the software to be used confidently
by staff who are unfamiliar with the underlying engineering or software details.
Whatever the technical merits of a piece of software, it is little more than useless if
the users lose patience with a poor user interface and revert to manual methods of
operation. The user interface will have to be tailored to the specific needs of the
client. In the immediate future, it is likely that most on-line schedulers will include
operator intervention to check the schedules. Once confidence is achieved, then the
operation can be made more automatic.

Telemetry
ITHE NETWORK I
Interface

~lState
stimator
J
~Demand
Predictor

I CORE Emergency
Handler
SCHEDULER
(OH·line
Analysis

User
Interlace
ITHE USER
Figure 2. Modules in an On-line Operational Control System for Water
Networks.
244

CONCLUSIONS

In the current economic and regulatory climate, ad hoc methods of water network operation
are no longer acceptable. A number of Operations Research techniques have proved their
worth for scheduling of simple systems (in terms of network complexity or types of constraint
and network information considered). These techniques are unlikely to be able to cope on
their own with the increasing intricacy of water scheduling problems. Knowledge Based
scheduling techniques appear to overcome the most obvious drawbacks in the Operations
Research techniques in that they are able to schedule complex multi-constraint networks with
reference to a full hydraulic model and detailed system information. The benefits to the water
industry from Knowledge Based scheduling include reducing electrical charges for pumping,
improving leakage control to avoid over-abstraction, maintaining the water supply at
acceptable quality levels, obtaining better control over emergencies and encapsulating and
appraising operational knowledge. Knowledge Based scheduling has been proved for small
networks with the SFK Technology software PUMPLAN, which has predicted savings
averaging over 15% in studies on real systems.

REFERENCES

1. Bond A., Gasser L., Readings in Distributed Artificial Intelligence, Morgan Kaufmann,
San Mateo, 1988.

2. Brdys M., Coulbeck B. and Orr C., A Method for Scheduling of Multi-Source, Multi-
Reservoir Water Supply Systems Containing only Fixed Speed Pumps. In Proc. lEE
Conference on Control, 1988.

3. Coulbeck B. and Orr C., Computer Applications in Water Supply, Volume 2, Systems
Operation and Control, Research Studies Press, Letchworth, 1988, pp. vii-x.

4. Coulbeck B., A Review of Methodologies for Modelling and Control of Water Supply.
In Computer Applications in Water Supply, Volume 2, Systems Operation and
Control, ed. B. Coulbeck and C. Orr, Research Studies Press, Letchworth, 1988, pp.
80-109.

5. Creasey J., Pump Scheduling in Water Supply: More than a Mathematical Problem.
In Computer Applications in Water Supply, Volume 2, Systems Operation and
Control, ed. B. Coulbeck and C. Orr, Research Studies Press, Letchworth, 1988, pp.
279-89.
 245

6. Fox M. and Smith S., ISIS - a Knowledge Based System for Factory Scheduling. In
J. Expert Systems, 1, 1984.

7. Hartley J., Powell R., A Combined Approach to Demand Prediction. In Digest of lEE
Colloquium on Control and Optimisation Techniques in the Water Industry, 1990, pp.
2/1 - 2/3.

8. Jowitt P., Garrett R., Cook S. and Germanopoulos G., Real-Time Forecasting and
Control for Water Distribution. In Computer Applications in Water Supply, Volume
2, Systems Operation and Control, ed. B. Coulbeck and C. Orr, Research Studies
Press, Letchworth, 1988, pp. 329-55.

9. Sterling. M and Bargiela A., Minimum Norm State Estimation for Control of Water
Distribution Systems. In lEE Proceedings, 131, 1984.

10. Zessler U. and Shamir U., Optimal Operation of Water Distribution Systems, J. Water
Resources Planning and Management, 1989, 115, 735-52.
 DIAGNOSIS OF HYDRAULIC PERFORMANCE OF WATER SUPPLY SYSTEMS

HELENA ALEGRE
SERGIO TEIXEIRA COELHO
LNEC - National Laboratory of Civil Engineering
Av.BrasillOl, 1799 Lisboa Codex, Portugal

ABSTR~,.cr

Hydraulic simulation and demand analysis have been, for some years now, key research areas
for the water supply team at the National Laboratory of Civil Engineering of Portugal. It is
felt that the planning, design, operation and control of water supply and distribution sysWI:.ls
are greatly improved by the use of adequate analysis and processing tools.
Special attention has recently been paid to the assessment of the hydraulic performance
of the system, which can greatly assist in establishing a useful diagnosis and in comparing
different operational alternatives of upgrading or expansion solutions, in a clear and quanti-
fied basis. This paper describes the algorithms developed for the purpose, presenting a set of
results for one of the case studies where the techniques have already been applied.
This type of methodologies can be envisaged as decision support tools, and are easily
applicable so long as demand data and a hydraulic simulator are available. The scope of such
tools can be seen as a first step towards the definition of easily implemented decision rules
that may eventually integrate a Decision Support System.

INTRODUCTION

The main goals taken into account in the traditional approach to the design of water supply
and distribution systems were the minimisation of initial investment and operational costs.
The hydraulic performance was often relegated to a secondary role especially in terms of
exploring how it can be affected by varying conditions throughout the useful life of the
system. The tendency is nowadays clearly reversed with a new emphasis being placed on
reliability and the way in which the system performs its water distribution task. These are

247
248

particularly important points in the case of urban distribution networks, often characterised by
simplified operation of frequently complicated layouts with myriad supply points.
This type of problem is not exclusive to the design stage, since it is common to find
existing systems with under-design problems or functional and operational difficulties. With
the current rapid growth of some regions in Portugal, there is a strong need for the type of
tools that will allow the engineer and designer to evaluate the hydraulic performance of a
system in order to facilitate diagnose and decisions, without having to rely totally on the
empirical insight of the experienced decision maker.
The systematic use of network analysis models is certainly a correct path towards the
solution of the problem. But they still return the type of result that is a long way from the
final decision needed, or even from an easy to understand diagnosis.

PERFORMANCE EVALUATION INDEXES

Within this framework, a group of three indexes was developed in order to provide an
evaluation of a system's performance in a rapid and simplified manner. They were named
pressure index (PI), head fluctuation index (HPJ) and velocity index (VI).
The rationale for the first two is based on the assertion that, for a network to perform
well from a hydraulic point of view, the pressure at every supply point must fall between a
maximum and a minimum requirement, and that the head surface across the network should
not be subject to great fluctuations over time. The maximum pressure is set up according to
the structural capabilities of pipes and other network elements, so as to avoid leakage.
Excessive pressure also makes water difficult to use at the domestic tap. The minimum
pressure requirement is there obviously in order to ensure supply but also to avoid sub-
pressures across the network. Significant fluctuations of the head surface are unwanted both
because they mean inconvenience for the user and because they are generally associated to a
greater sensitivity of the system to small changes in the operational scenario.
The approximately quadratic relationship between velocity and headloss means that
variations in the head surface are associated with flow velocities. The third index is based on
a velocity measure, largely derived from the fact that one of the main design criteria used in
Portugal is the limitation of pipe velocities - a system without "excessive" velocities is
deemed to perform well hydraulically.
The following sections describe the three performance measures in some detail. It must
be born in mind that the type of analysis required is based on hydraulic simulation models
and makes use of all the inherent simplifications. The definition of network nodes at pipe
intersections, network boundaries, connections to large consumers, devices, changes in pipe
 249

size or material, or generally at any point where a discretisation is required by the analysis, is
especially noteworthy.

The Pressure Index (PI)

The global pressure index of a network is calculated from a weighed average of the hydraulic
performance with respect to nodal pressures. To every single operational scenario or
snapshot corresponds a value of this index.
Each node of the network is graded from 0 to 4 (4 is optimum, 0 is failure to perform)
according to the following classification, which is also depicted in Fig.l:

* The optimal nodal pressure is considered to be the one that equals the minimum
pressure requirement Pmin , since it is the best compromise between satisfying demand,
minimising pumping costs and controlling leakage. Therefore, a node with such a
pressure is graded at 4, i.e., Pl=4.
* A pressure value equalling the maximum allowable Pmax is still meeting demand
appropriately and is not expected to cause damage or a breakdown. Pl=3.
* Nodes where the pressure has exceeded in more than 50% the upper limit are thought to
provide generally unacceptable service, and are penalised with a Pl=l. The value of 2
will correspond to the intermediate stage (see Fig.l).
* Finally, when the nodal pressure has fallen bellow 70% of the minimum requirement,
there is no supply and the index takes the null value.

0 c c :.! :.!
o..e o..e
...0 D..E D..E
"!

Figure 1 - The pressure index

The indexes corresponding to intermediate situations are calculated by linear

interpolation. After each node has been graded, the global value is obtained by a weighed
average. The weighing is used to make the calculation independent from the number and
placement of the nodes taken into consideration. Two possible alternatives for the weighing
factors would be the number of consumers inside the nodal area of influence, or the demand
250

associated with the node. The latter was preferred since it ensures a better description of
large consumers.

The Head Fluctuation Index (HFI)

In a similar way to the previous index, the head fluctuation index measures the hydraulic
performance of a system from the point of view of the daily fluctuations in the head surface.
As before, to each scenario corresponds a particular value of the index, in which the nodal
head fluctuations are calculated in relation to the maximum head that can occur at each node,
usually corresponding to the hours of nocturnal minimum demand.
The calculation is equally done on a nodal basis and weighed across the network using
the nodal demands.
The classification used for the head fluctuation index is as follows (see also Fig.2):

* The optimal state would be to keep a node always at constant pressure, regardless of
demand. This case would be graded at 4.
* Each additional 10 metres of head fluctuation dH will be penalised with 1 negative
point. Hence, 20 metres are still acceptable and HFI=2, while 30 metres would already
be unacceptable, hence HFI=l. HFI=O is achieved at 40 metres or more of fluctuation.

X 3
Q)
-g 2

0
0 10 20 30 40
Ll. h (m)

Figure 2 - The head fluctuation index

This classification is in accordance with the relevant Portuguese standard [1] which sets
the maximum daily pressure fluctuation at 300 kPa.
The head fluctuation index depends on the energy loss across the system. However,
two systems with the same overall energy loss and similar demand scenarios will still have
different indexes if the topology is different. A system with an elongated layout supplied
only at one end will have a much worse performance form this point of view than a circular
one supplied radially from its centre of gravity. Although there is usually little that can be
 251

done about the layout of a particular system, it is rather crucial to keep in mind that the
tolerable headlosses decrease as the shape differs from the ideal configuration.

The Velocity Index (VI)

The velocity index is an attempt to measure the proximity to the reference value of the flow
velocities in all links across the network. A link is defined as any stretch of pipe between two
consecutive nodes. The reference velocity Vref in a link of diameter D can be given by the
following expression (Baptista, [2]):

Vref (m/s) =0.1274 oo.4 (mm) (1)

The index is based on the following classification, also depicted in Fig.3:

* The optimal situation is when the velocity on a link equals its reference value, since it
means the maximum flow capacity has been reached. Therefore, Vl=4.
* Both situations when the velocity is null or double the reference value are considered to
be unacceptable, and get a VI of 1.
* The remaining values are obtained by linear interpolation (see Fig.3), and the cut-off
null value of the index is arbitrarily set up at 2.33 Vref.

0
1.oo vrel 2.00 vrel 2.33 vrer
Figure 3 - The velocity index

This index is calculated for each link, and the global value is averaged out across the
network using the link volumes as weighing factors. The weighing factor must make the
result independent from the choice and placement of nodes. This would be achieved by using
the pipe length. However, pipe size (cross-section) must be taken into consideration,
otherwise the smallest pipe in the network will have the same importance as the largest main.
To incorporate both the length and the cross-section, the pipe volume was chosen.
The velocity index must be analysed in conjunction with the other two, since if taken
alone its significance is relatively poor. The index was calibrated through experimentation in
252

over 15 networks, trying that, on average, it would match the behaviour of the head
fluctuation index. This way, comparing the two can give a rough idea of how favourable or
unfavourable is the topology of the system on average terms. Indeed, velocity indexes below
their head fluctuation counterparts may indicate a bad configuration, and vice-versa.
However, since the networks used for calibration are far from representing a true random
sample of the existing population, the authors are only too aware of the fragility of such
calibration. Only the continued application to real case studies will allow for more reliable
values.

INDEX CURVES

The factors that can most influence the hydraulic performance of a system and make it vary
along time are the system's physical characteristics, operational conditions and the demand
loading. Of the three, demand loading has the most significant variability over time, and it is
the only variable that lends itself to a systematic and general analysis of its influence on the
hydraulic performance.
The range of demand loads over which the analysis is to be performed is the first
decision to be made. If the diagnosis is made in terms of the daily operation of the system,
then the domain to be analysed will range from minimum night flow to diurnal peak. If on
the other hand it is necessary to look into medium or long-term periods, the range must be
extended accordingly.
Having decided on the range of demand loads, it is necessary to calculate the indexes
and depict their variation in a convenient format. In order to standardise results, it is useful to
divide demand values by the correspondent current daily average, making the problem
dimensionless and independent from network type and size.
The curves thus created by plotting any of the three indexes against a load factor will be
termed index-curves of a system. Studying alternative configurations of the same system or
different operational scenarios can be done by analysing the index-curves calculated for each
of the new sets of conditions.
Fig.4 shows the index-curves for a case study, a network in the city of Abrantes in
Portugal. They show a system with pressure insufficiency across the whole range of load
factors, sometimes coinciding with excess pressure situations. The system can probably be
divided in two different zones, judging from the change of slope on the head fluctuation
index curve, and the steep slopes also mean increased sensitivity to the slightest variations in
the demand load. The system would seem to be the product of a somewhat patchy design,
since the velocity index does not venture above 3, the likely result of a mix between very low
and very high velocities for the same load scenario. Apart from the various problems shown
 253

for all the demand domain, the main conclusion is that the system's full capacity is reached at
a load factor of 1.5, which is less than the average daily peale

HEAD FLUCTUATION
PRESSURE INDEX VELOCITY INDEX
INDEX
VI 4 .--~-----...,

3 3 3

2 2 2

. . . .. .. ... .. . . . . . .. .
~

0 L-----1.--'--~----l 0 '---'---'------'.--'
1.0 2.0 3.0 4.0 0.0 1.0 2.0 3.0 4.0 0.0 1.0 2.0 3.0 4.0
Load factor Load factor Load factor

Figure 4 - The index curves

Dispersion Limits
The performance indexes are calculated from weighed averages of nodal values. It is crucial
however to be able to differentiate between two systems with the same average indexes but
different dispersion in the local (nodal) values, since the two situations may have totally
different diagnoses and solutions.

HEAD FLUCTUATION
PRESSURE INDEX VELOCITY INDEX
INDEX
PI 4 HFI 4 VI 4

3 3 3

2 2 2

1.0 2.0 3.0 4.0 1.0 2.0 3.0 4.0 1.0 2.0 3.0 4.0
Load factor Load factor Load factor

.........,... Average curve • 0-25% • 25-50% • 50-75% ITN1 75-100%

l:J2.I

Figure 5 - Index curves and dispersion curves

254

Dispersion can be depicted in a number of different ways, and perhaps the most obvious
choice is to calculate the extreme curves (absolute lower and upper boundaries of the
indexes). This may still not be good enough, since when the boundaries are too far apart it is
hard to tell whether such dispersion corresponds to a few insignificant outliers or to entire
zones with separate behaviours. It is convenient therefore to define intermediate curves,
preferably with a direct physical meaning. This is provided by percentile curves of the total
demand. These should be read as follows: if (x,y) are the coordinates of a given point in the
P% percentile curve, it means for a load factor of x, P% of the total demand are being
supplied with a performance index smaller or equal toy.
Fig.5 shows the dispersion curves as well as the index curves for the same case of Fig.4.
The system reveals itself to be rather heterogenous, and the information we had previously
drawn about the likely existence of two different zones is confirmed by the fact that the 0-
25% band is narrower and the 25-50% band broader than average.

Penalty Curves vs Behaviour of Physical Variables

The index-curves are the result of applying a penalty every time the performance is not
optimal with respect to a particular aspect such as nodal pressure. They allow for global
conclusions to be drawn, such as "the system performs well". However, and since two of the
penalty rules (Figs. 1 and 3) have an ascending branch followed by a peak and a descending
branch, they give no indication as to whether, for example, the penalty is due to too low or
too high velocity. It may prove useful in those cases to analyse the behaviour of the physical
variable that was used to calculate the penalties, over the same demand domain and using the
same ;,.-.,;&ling for computing global averages (nodal consumption for the first two indexes
and flow velocity for the third).
The pressure index is calculated by penalising the discrepancy between nodal pressure P
and the minimum pressure requirement Pmin for each demand loading. The physical variable
involved is therefore P- Pmin. When these are weighed by the nodal demands, we are multi-
plying differences in pressure by flows and consequently obtaining power values. If P - Pmin
is the excess or deficit of pressure relative to the minimum requirement, the sum of all the
nodal values of this variable weighed by the correspondent demands represents the excess
power supplied to the system, or if divided by the total demand, the excess power by unit
flow.
In the case of the head fluctuation index, the relevant physical variable is ~ax - H,
where H is nodal head and ~ax is maximum nodal head, usually corresponding to the lowest
demand loading (night flows). Ignoring the energy dissipated in the system for this minimum
flow scenario, Hmax - H represents the headloss along the path(s) to the node concerned.
Multiplying by the nodal demand will yield the dissipated power along the path. The sum for
 255
all the nodes represents the power dissipated in the system, and dividing by the total demand,
the power dissipated by unit flow.
The significance of these two quantities, P - Pmin and Hmax - H, somehow transcends the
mere supporting role to the index curves. The total power to be supplied to a distribution
network through one or more supply points must be such that all the demand points are fed
with enough pressure (i.e., not less than the minimum requirement). This way, the total
power is constrained by the minimum head surface, which is not something that can be easily
modified or shaped in existing networks. Consequently, the minimum power to be made
available to then nodes at timet, Pwmin(t), is given by:

(2)
i=l

Where Qi(t) is the consumption on the ilh node at time t. Unless the demand and supply
points coincide, there will also be inevitable power losses along the way, by energy
dissipation:

~ i i
Pwdiss (t) = LJ ~p (t) Q (t) (3)
i=l

Where Pwd;..(t) is the power dissipated in the entire network, and Mi(t) is the headloss
between supply source and the i1h node at time t. This quantity can be controlled by the
engineer to a certain extent, since the headloss is a function of pipe size and length, network
layout and supply location. Without financial and technical constraints, this type of reasoning
would lead to a multitude of supply points and the use of large pipe sizes with negligible
headlosses. This of course would be too expensive, difficult to run and would probably imply
water quality problems due to excessively slow flows. It is the engineer's task to find the
appropriate balance.
Finally, there is a third member in the power equation, Pwcomp , reflecting the
compensation that must be allowed for at the supply points because of the topographical
restrictions imposed by the nodes:

Pwcomp =~ [Pi (t)- P~in] d (t) (4)

i=l
256

It is a problem of geometric compatibility, due to the fact that all of the flow supplied at
a given moment by any particular source has the same head. For the most adverse node to be
properly supplied, all the other nodes will receive excess energy. Once again it is the
engineer's task to minimise this quantity.
In short, the excess power supplied to the system and the dissipated power along the
path are the two variables over which it is possible to actuate, and it is important to take them
into account when analysing alternative designs.
It will be useful to put some numbers into the above considerations. Assuming the
power variables mentioned above can be directly or indirectly associated with energy
consumption, the related costs can be easily computed. Estimating the average electrical
motor efficiency at 90% and the pump efficiency at 60%, and assuming an average cost of
13$00/Kwh (1), the total energy cost in a distribution network is in the order of 24$00/Kwh.
For the example in Fig.6, with an average demand of 62 1/s, about 150$00 are being wasted
per hour in energy dissipation (24$00/Kwh x 100 kPa x 0.062 m3/s). At peak hour, with a
load factor of 2, this value will increase to 890$00 per hour, six times higher.

EXCESS POWER DISSIPATED POWER REF. VELOCITY - VELOCITY

60 150 3 .------~---,

40 130 2 ... ........... . .. .

20 110
I 0 I 90
J: 70
-20
a.' 40 50
'
Q. :x:"
~ 30
-2
~

-100 -3 L-.-l.---'
0.0 1.0 2.0 3.0 4.0 1.0 2.0 3.0 4.0 0.0 1.0 2.0 3.0 4.0
Load factor Load factor Load factor

' - " Average curve • Q-25% • 25-50% • 5D-75% • 75-100%

Figure 6 - Index curves and physical variables

This type of calculation would certainly be worth considering, · for example, when
scheduling reinforcement works with the purpose of reducing the total power needed by the
network.

1 The costs shown are in Portuguese &cudos (255$00--.£1.00 at current rates) and the example used is a
relatively small distribution system.
 257

In the case of velocities, the relevant physical variable is the diference between the
reference velocity and the velocity in the pipe, v,.r- V. The three physical variables and their
relationships with the indexes are depicted in Fig.7.

Figure 7 - Indexes and the physical variables

Program SIDHRA
A Fortran 77 program, SIDHRA, has been developed to perform systematic calculation of the
index curves over the results of hydraulic simulation. It works as a post processor to LNEC's
network analysis package, SIMALl, all running under MS-DOS. SIDHRA calculates all the
index and dispersion curves, total and dissipated power, and it lists all the nodal values of the
performance indexes. Graphical representation such as shown in Fig.4 can be obtained by
coupling SIDHRA to a graphics package.

PRACTICAL APPLICATIONS

Continued use of the index curves over case studies has allowed for the build up of valuable
experience on how to read them correctly. It must be said that their interpretation can be a
trying task, and would greatly benefit from some means of automated pattern recognition, a
current research concern at LNEC for this particular problem.
The information perceived from the curves may relate to their shape, slope and absolute
value, as well as to the width of the dispersion bands. It is relatively straightforward to see,
for example, that an HFI curve with significant plateaus indicates the existence of networks
258

zones with distinct behaviours, while steep slopes in the PI curve correspond to loading
ranges for which the system is very sensitive to even the smallest changes, and when
combined with HF'I's below 2.5, we have a deficient working situation. If the VI curve is
steadily increasing, then the system's pipes are oversized, and so on.
The index curves can be used in various different ways. So far the case studies where
the methodologies have been used and tested were primarily concerned with urban network
upgrades and expansions. The systems analysed are parts of the distribution networks of the
cities of Maputo, in Mozambique, and of Abrantes and Matosinhos, in Portugal. The index
curves were employed mainly as an aid to the following tasks:

* Performance diagnosis
* Support to the definition of calibration tolerances
* Analysis of the influence of imprecise data in the simulation results
* Comparison of different design and operation solutions
* Support to works scheduling

The point about calibration is worth looking into. Calibration tolerances must always
be established in terms of the particular system to be analysed. The tolerance for a stable
system should definitely be lower than that for an unsteady one. The pressure and head
fluctuation indexes have proved useful for sensitivity analysis. If the indexes calculated over
a particular simulation are not significantly affected by changes, for example, in the pipe
roughnesses, then it is perhaps not worth it investing much time and money in calibration
campaigns in order to improve that parameter's accuracy. Moreover, the index curves can be
drawn not only for load factors but with any of the other relevant parameters on the X-axis,
which will allow those parameters to be tested across their variation ranges. This can provide
some very solid ground on which to decide how far to go in terms of calibration investment,
in order to achieve a given predetermined precision level. On the other hand, it also gives a
much better picture of the degree of confidence deserved by the results of simulation, when
the input data's precision can be estimated.
Comparing between alternative solutions, another of the listed uses, allowed for the
economic and hydraulic objectives to be balanced out in a more quantifiable manner.
Finally, the use of the index curves to help scheduling upgrade and expansion works.
An example is shown in Fig.8. From the starting point of the current configuration of the
system, the pipes to be upgraded and the new diameters were identified. Since there was
pressure from the financial side to hold off at least part of the investment, and given that the
network had various degrees of problematic zones, it was decided that the upgrade would be
 259

carried in two stages. The pipes to be upgraded during the first stage were identified. In
order to schedule the two stages, the following procedure was adopted:

* Calculate the index curves for the current configuration, after the first upgrade and after
the second upgrade, using the complete range of load factors from current minimum to
peak at design horizon.
* Change the X-axis variable from load factors to time, through an appropriate demand
projection.
* Define the threshold of acceptable performance for each of the configurations, based on
the interpretation rules mentioned previously.
* Schedule the implementation of the upgrades so that the system is kept above the
acceptable threshold.

HEAD FLUCTUATION
PRESSURE INDEX VELOCITY INDEX
INDEX

3 3 3

2 2 2

0 l..,_.....J.._...l...._ _,___. 0 L--l....L-.J.....l-L.:..=:::J 0 L...---'---"---"---'

~ 0.0 1.0 2.0 3.0 4.0 ~ 0.0 1.0 2.0 3.0 4.0 1 0.0 1.0 2.0 3.0 4.0
\ Load factor time (years) ) \ Load factor time (years) ) \ Load factor time (years) )

0 10 20 0 10 20 0 10 20
current • phase II
• phase I
configuration

Figure 8 - Scheduling an upgrade

In the example shown, only the three index curves and the 0% percentile curves were
depicted to avoid overloading the graph. It can be seen that the best option would be to
schedule the first upgrade for as soon as possible, and the second in 10 years' time, when the
performance drops to minimum permissible levels.

CONCLUSIONS

The assessment of hydraulic performance in water supply systems is an increasingly

important topic in an industry progressively driven by a need to deliver competent levels of
service.
260

This paper proposes a set of tools which can greatly assist in establishing a useful
diagnosis and in comparing different operational alternatives of upgrading or expansion
solutions, in a clear and quantified basis.
The results presented for a case study clearly show that it is possible to manipulate the
information produced by current network analysis to capture a surprising understanding on
some aspects of the system, allowing such tasks as calibration or expansion planning to be
clearly improved.

REFERENCES

1. Conselho Superior de Obras Publicas e Transportes (National Council of Public Works

and Transport, Portugal), New regulation on water distribution and wastewater drainage,
Proceedings of Seminar, National Laboratory of Civil Engineering, Portugal, October
1991.
2. Baptista, J.M., Economical design of water supply systems. Thesis, Hydraulics Dept.,
National Laboratory of Civil Engineering, Portugal, May 1983.
3. Alegre, H., Coelho, S.T., Demand analysis and the quantification of levels of service in
water supply systems, Proceedings, IWSA Workshop on Future Water Demand, Basel,
Switzerland, February 1990.
 ON-LINE CONTROL OF A CITY WATER SUPPLY AND DISTRIBUTION SYSTEM

B COULBECK and C H ORR*

Water Contro\ Unit, Leicester Polytechnic, Leicester, LEl 9BH, UK
E 0 Tech., Inc., Arcadia, CA91006, USA

ABSTRACT

This paper outlines the design and implementation of an on-line, computer

control scheme for a city water supply system. Descriptions are given of
the data management and processing activities leading to cost optimal
control of the system. The project has been undertaken by the Leicester
Polytechnic Water Control Unit in collaboration with Severn Trent Water.

INTRODUCTION

One of the projects being pursued at the Water Control Unit concerns the
design and implementation of an on-line, optimal control scheme for a city
water supply and distribution system. This project, currently approaching
completion, forms part of a major modernization program which was
undertaken by Severn Trent Water for the city of Wolverhampton. The
installation of a modern telemetry system, as part of the initial
redevelopment phase, has provided the basic facilities for implementation
of such an advanced control scheme [1].
The Wolverhampton water supply and distribution system includes the
city central region and extends westwards to encompass several villages.
See Figure 1 for an overriew of the major features. In total this covers
an area of about 400 km and the water network supplies a population of
400,000 with an average daily quantity of 112 megalitres (1988 values).
About 62% of the water is supplied for domestic consumption, and the
remaining 38% is used for industrial and trade purposes. Underground
borehole sources, which are located within the city's supply area, provide
60% of the total water requirement, while 40% is abstracted from a nearby
river source.
Several on-line computer control schemes for water supply systems
have been reported [2,3,4], however these are largely concerned with the
optimal control of bulk water supplies. This paper describes a fully
integrated, and optimal, scheme for control of a city water system,
including its distribution network.
Overall the paper reviews the objectives for advanced control of
water systems and provides an overview of the implemented control system

261
 ",, ........................-... -"'f
I \..,J"' N
0'1
,..... ,/ "" .... ., .....\ N
I ..... ~ 1 ' ~~ I
/ ' ! ,...'
(,, ""' ......... ___,I ("'"''
\
.•.-..... .... ___ , ... ..... ,,'
( ,\
,'
·..' '...... (J~
....... I -N-
NEACHLEY '\\. , ....... _
\ I ... ,
r \ / ....... \:('
·--.... '\,. ~~
) ........... / ............ _.... __ , ....
COSFORD WELL!
..
~
(
I
;
I
I
: KEY

!:::. BOREHOLE SOURCE

t:J SERVICE RESERVOIR
~ RIVER SOURCE
0 BOOSTER STATION
SUPPLY AREA BOUNDARY
* INTERZONAL CONTROL VALVE

0 t 1 ) :: 5 t\kt:l
SCALE

FIGURE I: WOLVERHAMPTON WATER SUPPLY AND DISTRIBUTION SYSTEM

 263

which meets these objectives. Sections on overall control aspects, the

telemetry data management, and the data processing modules, indicate the
important characteristics of the on-line control scheme.

OBJECTIVES FOR OPTIMAL CONTROL OF WATER SYSTEMS

The following key objectives have been identified and are briefly
discussed:

Provision of Adequate Supplies to Consumers

The level of service for consumers is of primary importance and must be

maintained at reasonable standards at all times. These standards entail
the delivery of the required quantity of water at acceptable pressure
levels and water qualities. Contingency plans must be devised to cater for
any emergency conditions outside the normal operating regime of automatic
controls.

Minimization of System Operating Costs

The increasing use of water network analysis and simulation methods has
demonstrated that traditional, manual, methods of operation can often be
significantly improved upon. ·Dynamic simulation studies, in particular,
have indicated that many water networks are inherently capable of more
efficient operation, without adversely affecting the level of service
provided to consumers. For water supply and distribution systems the
variable operating costs mainly consist of electricity charges for pump
usage. Since these charges are based on time-varying unit and maximum
demand tariffs, it is necessary to consider the system hydraulic and cost
performance over a minimum period of one day, in order to derive least-cost
schedules.

Compliance with System Constraints

System constraints usually fall into two categories. Firstly, there are
the absolute physical limitations of the water system components, ie., pump
availabilities, station flow capacities, reservoir storage capacities, etc.
Secondly, there are the constraints on the operational use of the
components, typically, allowable pump speed ranges, control valve ranges
and reservoir level variations. Other factors which may need to be
considered include the limitations on pipeline pressures and rapidity of
pump on-off switching. In all cases, and particularly for minimization of
system operating costs, the control scheme must be designed to take account
of these limits.

OVERVIEW OP THB CONTROL SYSTEM

Figure 2 illustrates the general configuration of the on-line control

scheme for the Wolverhampton system. For this implementation, the
telemetry computer collects and sends live system data, at 15 minute
intervals, to the control computer. The received data, which are held in
an ASCII formatted file, consist of standard header information,
(identification, date and time), together with the data readings associated
with a set of meter references. These readings include measurements of
pipe flows, reservoir levels, pressures and ·pump speeds; any values that
are in an alarm state are flagged. Transmission can take place from the
 264

Optimal Controls
CONTROL
COMPUTER

MicroVAX II PDP 11124

Telemetry Data Pump Speeds
Pipe Flows
Valve S.lllngs
Requests Reservoir
Levels

OPERATOR
COntrol CONSOLE
System
Status FIGURE 2.· sySTEM CONFIGURATION
Status

DEMAND
PREDICTOR

R 0

y c

I
N PUMP
p OPTIMIZER

u
T

FIQURfi ~ PRINCipAl MQOULES IN AN QN ..LJNE CONTROL SQHEME

 265

control computer to the telemetry computer, also every 15 minutes, for

returning the pump schedules, reporting predicted responses and activating
or de-activating alarms.
The first task of the control system involves collation of all of the
input telemetered data, ready for processing during, and at the end of,
each operational control period. The control period for this scheme is 24
hours, starting at 08:00 hours (as decided for operational convenience).
The collated live (time-varying) data is also archived off-line for
subsequent retrieval and use. The other main data requirement consists of
the longer-term static data, such as the network configuration, system
component characteristics, etc., which are usually system and application
specific and are initially available from off-line sources.
The next task involves extraction of the system data, from the
various sources, for the individual control modules. This sequence
involves updating the inputs for the demand prediction module followed by
those for the pump optimization module. The predicted demand profiles must
then be related to the control model to give the individual nodal demands
which are required by the pump optimization module. Optimized pump
schedules are then calculated, for the following control period, based upon
these predicted demands and the current system status.
The penultimate task of the control system is to send out the
resultant pump schedules and predicted system responses to the telemetry
computer, while the final task is to activate the real-time monitoring and
regulation module. The latter module automatically computes any on-going
major discrepancies between predicted and actual values of critical system
responses. Remedial measures, which may include pump re-scheduling during
the current control period, can be activated if required.

TELEMETRY DATA MANAGEMENT

Access to good quality telemetered data is a primary requirement for the

successful implementation of an on-line control scheme; the following data
handling facilities are required:

Management of Incoming Data from Telemetry

The incoming live telemetry data are transmitted via an inter-processor

software link, and the telemetry data management and coordination modules
are designed to handle this intermittent inflow without affecting other on-
going control processes. Principal activities at this stage include:
identification of the type and format of the data being received and
decisions on the interpretation requirements for the data. The types of
data that could be received consist of standard network measurements, alarm
conditions, control acknowledgements, etc.

Archival and Processing of Telemetry Data

In this case the archival method can optionally use specially created and
formatted files, a dedicated database, or a combination of the two. Once
the data has been archived it can then be retrieved and processed to suit
specific applications. Since the telemetry data is archived exactly as
received by the control computer, including any spurious data or null
values, it is imperative that this further processing be carried out in a
reliable manner, with adequate checks to ensure the integrity and accuracy
of the refined data.
266

Management of Outgoing Data to Telemetry

After the optimal control schedules have been computed, these must be
transmitted to the telemetry outstations, adjacent to the supply system
components. In general, the output data to telemetry consists of time and
control settings for pumps and valves. However, computed reservoir levels,
flows and pressures, with associated alarm flags, may also be required in
order to complete the control instructions. All of this data requires
conversion into a format suitable for direct transmission to the
outstations, eg., pump speed expressed in percentage rather than rpm, etc.
Acknowledgement of successful transmission is necessary, since the data is
used by the outstations for direct operational control of the system.

CONTROL MODULES AND PROCESSING

Four principal control activities are required for the reliable operation
of the control scheme. These activities are described below and their
corresponding module interconnections are shown on Figure 3.

Demand Prediction Module

The demand prediction module used in this on-line control scheme is GIDAP
[5,6]. GIDAP provides facilities for data screening, data smoothing, and
profile prediction.
Data screening is used to eliminate any missing or spurious data
values within the incoming telemetry data. Firstly, missing and extreme
values are filtered by using pre-defined minimum and maximum thresholds.
Secondly, adjacent data values are tested against a first difference
threshold. Finally, sudden peaks and troughs in the data can be detected,
and rejected, by using a second difference threshold. Any missing or
rejected values are replaced by interpolation or by use of previously
predicted values for the corresponding period.
The data smoothing process is based on Fourier analysis of the
screened data. In essence the screened data is separated into all of its
constituent harmonic components but only the significant harmonics are then
reconstituted to produce a smoothed version of the original profile. This
process thus yields the underlying demand pattern by removing any random
noise components which may be present.
Profile prediction uses Triple Exponential Smoothing which maintains
estimates of position, velocity and acceleration of each data point. These
estimates are continuously updated using the prediction errors; this occurs
as the actual demands become available to allow the prediction errors to be
evaluated. The updated estimates are then used to obtain the predicted
demand profiles for each following time period.

Pump Optimization Module

The main pump optimization module used for this scheme is GIMPOS [5,6].
This module is applicable to systems with multiple sources supplying water
to a controlled service reservoir via multiple pump stations. Each of the
pump stations are deemed to contain a pre-defined number of pumps,
including at least one of variable speed. Lengths of equivalent direct
pipe lines, with demands at intermediate and final nodes, connect the
sources to the reservoir.
The GIMPOS two stage algorithm uses Forward Dynamic Programming, for
optimization of the time-varying reservoir storage, together with
 267
derivative methods, for optimization of individual pump stations. During
the computations, all feasible pump combinations, as well as pump speeds
for variable speed pumps, are checked. Also full account is taken of the
efficiency of the individual pump groups and their associated electricity
tariffs. The algorithm requires information on the hydraulic pressures and
flows; these are calculated from an approximation of the pipe network,
together with the pump characteristics and reservoir levels. The overall
computation is carried out, within a set of defined operational conditions,
to obtain optimal schedules which define the pumps in use and their
associated speeds over the specified operational control period. These
schedules can now be applied to control the system and will yield the
cheapest operational cost (under the assumed conditions).

Network Simulation Module

Network simulation is a desirable, and sometimes essential, activity for

on-line control of water systems. This application uses GINAS5 [5,6],
which has been enhanced to operate in two modes. An Off-Line mode provides
for background analyses of both the water and the control system
operations, and an On-Line mode provides direct evaluations of the adequacy
and accuracy of the control system.
In the off-line mode GINAS5 is mainly used to derive simplified control
models to suit each specific optimization module. These particular control
models are typically derived through a series of dynamic simulations, of
the full network, using historic or predicted data. This mode can also be
useful in establishing operator responses to emergency conditions.
In the on-line mode the simulator is commonly used to assess the
recommended pump schedules, before and during their actual use, on a more
detailed network configuration. This particular mode of operation can make
automatic use of an appropriate mix of live and predicted data; thus
providing added system security by allowing the operators to examine the
predicted optimal schedules and make any necessary adjustments to them
before they are implemented.

Monitoring and Regulation Module

The monitoring and regulation module contains all the relevant

information and procedures for monitoring and display of the critical water
system parameters, and for regulation of the entire system. Figure 4 gives
an indication of the graphical displays presented to the operator. For
effective monitoring of time-varying parameters, such as reservoir levels,
system demands, station flows and pump speeds, the module requires current
instantaneous values as well as the values since the beginning of the
operational control period. The regulatory procedures, for maintaining a
stable operating regime, can be activated either manually, or automatically
from violated discrepancy limits.

CONTROL BXBCUTIVB AND ACTIVITIES

The control executive sequences and oversees the complete operation of the
on-line control scheme; with data handling routines, control modules, and
other general procedures providing solutions to specific tasks. The
control executive is essentially a real-time program loop which is
activated upon initial entry. Telemetry data, comprising demand flows,
reservoir levels, flows and pump speeds, are received, and processed, by
the control executive. The processed data is subsequently passed on, as
 N
0\
00

X10 2 TOTAL Demand Cl/s) Time : 11 :53

Curr.u let tiiJe t.otal (Ml)
J
~ - - Aclua I ----Pr edicted
..., 4

i....
11
o-l

~
~
~
0
0~~~~~~~~~~~~~~~~~~~~~
0 14 16 18 20 22 24 26 28 30 32 34 36 38
~

~
~
..., Ht- Ac t. Pr·e d Hr· ~ct Pred Hr ~ct Pt~ed Stati.J s 1:! Acl:. Pr·ed
1 1:5 3 : 0 0
~ 14 377.0 228.0 22 292.0 237.0 6 158.0 197.0 1

0 LtJSt l!s 337.0 347.0

15 306.0 289.0 23 249.0 229.0 7 245.0 24 7. 0 '
§ 11G 328.0 283.0 :24 196.0 211.13 g 362.0 307.0 . 11ean l / s 269.9 254.4
=
~ 17 344.0 287.0 1 155.0 186.0 9 380.0 361 . 0 Pea k l/s 390.0 383.0
N 171.0 1 0 384.0 383.0
18 352.0 292.13 2 140.0 P~; . time~.09:53 10: 08
~ 19 352.0 291 .0 ·-:.-' 131.0 162.0 1 1 357.0 371 . 0
T•J ta 1 Ml 2 1 .37 20.15
20 342.0 270.0 4 154.0 160.0 1 2 339.0
170.0 13 307.0 11 iss i ng None None I
2 1 269.0 249.0 5 135.0
 269
and when required, to the various· modules residing within the control
executive.
Throughout each operational control period the monitoring and
regulation module provides a real-time check of· all important water system
parameters. Discrepancies between predicted and actual parameters are
compared against pre-set limits and appropriate actions initiated. The
regulatory procedures may involve complete or partial re-computation and
scheduling, depending upon the severity of the original discrepancies.
Towards the end of each operational control period, at a
pre-determined time of day, the control executive automatically sets up a
sequence for determining the new optimal schedules for the following
control period. This sequence begins with the processing of metered data
for the demand predictor. After completing the demand prediction for
all pre-defined nodes, the new demand data and other time-varying data,
such as initial reservoir levels, are updated for the pump optimization
module. This module is then brought on-line to compute the optimal pump
schedules for the following control period. Limited operator intervention
is allowed for at this stage, primarily to enable the operator to verify
the proposed pump schedules.
Satisfactory results are sent to the pump outstations, via the
telemetry system, using separate files to transmit pump schedules,
predicted reservoir levels and alarm conditions. The control executive
then produces a standard daily report which summarizes the predicted
operational performance. This report contains the pump schedules, consumer
demands, estimated operating costs, station flows and reservoir levels.
After completing all of the above tasks, the control executive
returns to the data archival process and, subsequently, to the monitoring
and regulatory activities for the next operational control perio~. For the
current implementation, the network simulator can be used during this stage
to perform system simulations for on-line assessment purposes.

CONCLUSIONS

An automated on-line control scheme has been developed for a major city
water network. Preliminary results from the implementation have been
encouraging. Confidence has increased about the reliability of the
predicted demands. Under normal operating conditions, the accuracies have
been around 5% for the daily average, and 10% for the instantaneous values.
This has led, in turn, to increased confidence in the optimality of the
pump schedules which, on application, have led to reduced operational
costs. Typical cost savings of 15% have been reported, with occasional
savings of up to 27%. System monitoring has proved to be of vital
importance, giving early warning of unexpectedly high or low demands and
reservoir levels, together with pump failures and other unpredictable
events. The control scheme has been applied to several parts of the city
supply system and is readily extendable to cover the whole system, subject
to the on-going engineering works. While important advances, in the
techniques for on-line control of water supply and distribution systems,
have been made; the project cost, in terms of finance and effort, has been
significant.

Severn Trent Water are to be congratulated on their foresight in launching

and sponsoring this project which should prove of ultimate benefit to the
overall water industry. The authors wish to acknowledge the full support
270

of Severn-Trent Water, with particular thanks to M. Williams, A. Elton and

L. Brammer for their significant contributions to the overall project. The
authors are also grateful to M.A. Parkar and S.T. Tennant who provided most
of the computer programming effort.

REFERENCES

1. Williams, M. and Coulbeck, B., Dynamic Modelling and Control

Optimization in the Development of the Wolverhampton Supply and
Distribution System. J. Inst. Wtr. Poll. Contr •• 1987, 86, 272-91.

2. Alla, P. and Jarrige, P.A., Optimal Control of the West Parisian Area
Water Supply Network. In Water Resources Planning and Management.
ed. S.C. Harris, ASCE, New York, 1989, pp. 661-4.

3. Pallside, P., Computer Techniques for On-Line Control of Water Supply

Networks. In Computer Applications in Water Supply. Vol. 2. eds. B.
Coulbeck and C.H. Orr, Research Studies Press, John Wiley & Sons,
1988, pp 313-28.

4. Jowitt, P.W., Garrett, R.T., Cook, S.C. and Germanopoulos, G., Real-
Time Forecasting and Control for Water Distribution. In Computer
Applications in Water SuDDlv. Vol 2. eds. B. Coulbeck and C.H. Orr,
Research Studies Press, John Wiley & Sons, 1988, pp. 328-55.

5. Coulbeck,B. and Orr, C.H., Control of Water Distribution Systems:

Practical Considerations. In Computerized Decision Support Systems
for Water Managers. eds. J.W. Labadie, L.E. Brazil, I. Corbu and I.E.
Johnson, ASCE, New York, 1989.

6. Orr, C.H. and Coulbeck,B., Control of Water Distribution Systems:

Program Descriptions. In Computerized Decision Support Systems for
Water Managers. eds. J.W. Labadie, L.E. Brazil, I. Corbu and I.E.
Johnson, ASCE, New York, 1989, pp. 774-87.
SESSION D: MONITORING AND INSPECTION
 LEAK DETECTION THROUGH HYDRAULIC TRANSIENT ANALYSIS

LENNART JONSSON and MAGNUS LARSON

Department of Water Resources Engineering, University of Lund,
Institute of Technology, Box 118, S-221 00 Lund, Sweden

ABSTRACT

A leak in a hydraulic system generates reflected waves when a pressure wave is initiated in the
system, making it possible to detect leaks by measuring and analyzing the pressure variation in
time at a point. The paper begins by discussing the effect of a pressure wave propagating past a
leak from a theoretical point of view, including the derivation of an analytic solution for the
magnitude of the reflected wave for a simple configuration. Numerical simulations using the
method of characteristics are then employed to determine the effect of a leak on the pressure
variation in simple hydraulic systems. Spectral analysis is used to analyze time series of pressure
in order to detect reflected waves originating from a leak. Finally, the presented technique for
leak detection is employed to field measurements carried out in a hydraulic system where a leak
could be simulated.

INTRODUCTION

Pressure transients occur in a hydraulic system when there is a change in the flow conditions,
causing pressure waves to propagate from the source of disturbance through the system until a
new steady-state is attained. The geometric properties of the system, such as network
configuration and the existence of hydraulic components, and the characteristics of the flow
change determine the conditions for wave propagation. Furthermore, reflected pressure waves
may arise wherever there is an abrupt change in the geometric or hydraulic properties (Contractor
1965). For example, if a pressure wave encounters a series or branch connection, the incident
wave generates a reflected wave that depends on the ratio between cross-sectional area and wave
speed for respective pipe in the joint (Wood and Chao 1971). Thus, a complex pipe configuration

273
274

causes a complicated pattern of waves that propagate in the system and that are superimposed on
one another. A leak in a hydraulic system also gives rise to a reflected wave with a magnitude
which depends on the leakage flow and the ratio between the incident pressure wave and the local
pressure at the leak.
By identifying waves propagating in a hydraulic system from recorded time series of
pressure, it may be possible to determine the location of the reflection sources. The frequency of
a reflected wave is given by a/41, where a is the wave speed in the pipe and I is the distance from
the source of reflection to the point where the pressure is recorded, if the pressure is measured
where complete reflection of the wave occurs such as at a closed valve. The wave speed is
computed from the measured frequency of a pressure wave where the location of the reflection
source is known. A leak may be detected as an unexplainable reflection source in the system, and
the location of the leak is determined from knowledge of the system configuration. Thus, if the
pressure is measured at a point in the hydraulic system and then analyzed with respect to
propagating waves, additional waves that are not due to the system configuration will reveal
possible leaks.
If a leak is small in comparison with the flow through the pipe it might be difficult to
identify reflected waves by means of a visual inspection of the measurements, but instead spectral
analysis could be employed to distinguish the presence of such waves. The measured time series
of pressure is transformed into the frequency plane to display any additional harmonics resulting
from reflected waves.

THEORETICAL CONSIDERATIONS

A pressure wave propagating past a leak in a single pipe generates a reflected wave with a
magnitude that is related to the leakage flow, and that has an opposite sign to the incident wave.
The leakage flow, q, is approximately proportional to the square root of the pressure head, H,
at the point of leaking according to,

q- K{il (1)

where K denotes a constant that includes properties such as the cross-sectional area of the leak
and a discharge coefficient. It is instructive to derive an analytic solution for the case of reflection
at a leak due to an incident pressure wave under idealized conditions. A pressure wave with a
 275

magnitude ~H, resulting from for example an instantaneous stop in the flow, is propagating along
a pipe towards a leak where the initial pressure head is H 0 • The leakage flow before the pressure
wave reaches the leak constitutes a portion a of the total flow upstream the leak. Friction in the
pipe is neglected and it is assumed that no minor losses occur at the leak. Under these conditions,
the relationship between the magnitude of the reflected wave (HR) and the incident pressure wave
(~H) is expressed as,

(2)

where:

(3)

ll.H
X-- (4)
H.

a:
a- 4(1-a:)
(5)

As Equation 2 shows, the portion of the incident pressure wave that is reflected depends upon the
leakage factor (a) and the rate between the magnitude of the incident pressure wave and the local
pressure at the leak (X = ~H!HJ. For a specific value on ~H/H0 , a larger leakage factor implies
that a larger portion of the incident wave is reflected. Similarly, a specific leakage factor causes
a larger portion of the incident wave to be reflected for a larger value on the ratio ~H/H 0 • In
Figure 1 is Equation 2 displayed for different values on a and ~H/H 0 •

If a pressure wave propagates back and forth in a closed hydraulic system with a leak,
reflected waves are generated every time the pressure wave passes the leak. A complex wave
pattern consisting of superimposed reflected waves arises depending on the position of the leak
and the system configuration. In order to investigate the effect of a leak, a simple hydraulic
system was studied composed of a straight pipe with a reservoir at the upstream and a valve at
276
the downstream end (see Figure 2). A leak occurs at a position nL from the downstream end of
the pipe, where L is the pipe length and 0 < n < 1, and the valve is instantaneously closed at time
t =0. Immediately after closure a pressure wave (primary) is generated at the valve that propagates
upstream towards the reservoir.

~ c.H
.3 AH Ho
2.0

1.5

.2
1.0

0.5
.1

•0

•0 ·1 •2 .3 .4 •5

Figure 1. The relationship between the magnitude of a reflected and incident pressure wave as
a function of leakage factor ex and aH/H 0 .

When the pressure wave passes the leak, a reflected wave arises that returns to the valve
with a negative sign. The wave reflected from the leak arrives at the valve at time t=2nL/a,
reducing the local pressure head at the valve. The reflected wave, generated by the primary
pressure wave arriving at the leak from the valve side, will always lag the primary wave in time
by a factor 2nL!a. Every time a reflected wave passes the leak it will give rise to new reflected
waves, although these will be of a secondary order in comparison with the reflection from the
primary wave. Next time the primary pressure wave arrives at the leak from the valve side (after
reflection against the reservoir and the valve), a new reflected wave is generated. This reflected
wave coincides with the firstly generated reflected wave which returns after reflection against the
reservoir, and the two waves are superimposed on each other, both having the same sign.
 277

L• 6000 m
n. 1/6

11-n). L n·L

Q,• 80 v.-,.-------1
I
Q•= 7 1/s

Figure 2. Definition sketch of a leaking pipe with a reservoir at the upstream end and a
valve at the downstream end.

Thus, every pressure wave arising from reflection on the valve side of the leak is
superimposed on previously generated waves, all having the same sign. The reflected pressure
waves generated by the primary wave arriving at the leak from the reservoir side are super-
imposed in the same manner. In summary, energy is constantly being transferred from the
primary wave to the reflected waves, which are slowly built up by superposition. These reflected
waves also loose energy when passing the leak; however, the net contribution from the primary
wave is initially large enough to allow for a growth in the pressure of the reflected wave. The
primary pressure wave is attenuated according to the size of the leak, where a large leak reduces
the consecutive pressure peaks faster than a smaller one. In Figure 3 is the pressure variation at
the valve shown for instantaneous closure when a leak exists in the pipe. The leak was located
l/6L upstream the valve, and the calculations were carried out disregarding friction (parameter
values for the calculation are given in Figure 2).
The reduction of the primary pressure wave depends, apart from the leakage factor, on
the local pressure at the leak. Therefore, the continuing decrease in the peak pressure of the
primary wave is dependent upon all reflected waves present in the system. However, a rough idea
of the reduction in the magnitude of a pressure wave due to a leak may be obtained by
investigating a wave passing a consecutive number of leaks. By using Equation 2 it is possible
to estimate the reduction of the pressure wave by determining the magnitude of the wave
transmitted (HT) passed the leaks. The pressure head at a specific leak is H0 before the wave
arrives at the leak, and the flow behind the wave is taken as zero. From Figure 4 the influence
of the leakage factor on the transmitted pressure wave may be estimated for different number of
consecutive leaks (figure valid for .1.H/H0 = 1.0). Thus, the larger the leak is, the faster the wave
energy is transferred to the reflected waves.
278

Pressure
bm H20)
10

80

60

40

20

Time
0 (sec)
0 20 40 60 80

Figure 3. Pressure variation at a valve after instantaneous closure with a leak located 1/6L from
the downstream end.

RESULTS

Numerically Simulated Data

In order to investigate the possibility of detecting a leak in a hydraulic system, numerical
simulations were carried out for simple network configurations using the method of characteristics
(Wylie and Streeter 1978). The chosen hydraulic system consisted of two reservoirs connected
by a pipe with a valve at the downstream end and a leak located at a distance of 1/3L upstream
the valve, where Lis the length of the pipe. Values for the different hydraulic parameters were
selected to agree with data from a field study discussed later in this paper, in order to assess the
possibility of detecting a leak in the system. Figure 5 presents a definition sketch for the
hypothetical case studied and the chosen parameter values. The reservoir levels were set high
enough to avoid cavitation in the pipe during calculation. Both linear and quadratic valve closure,
 279

that is, the reduction of the valve opening area was linearly respectively quadratically decreasing
functions of time, were employed to introduce a pressure wave in the system.

1D

o(

.a
0.1

.6
0.2

O.l
.4

0.4

0.5
.2

N.M!ER OF
.o ,_------r------,-------r------,-----~~~T06
12 15

Figure 4. The magnitude of a transmitted pressure wave (.c1HT) after passing a number of
consecutive leaks as a function of the leakage factor a (AH/H0 = 1.0).

2000 ..

=0.401 f: 0.02
ll.= 80 1/S --r--------J'------------1
~
Q0 = 51/s

L• 6000 II

Figure 5. Definition sketch for sample calculation with two reservoirs connected by a pipe
containing a leak.

The time for valve closure was set to 80 sec which allowed the primary pressure wave to
return to the valve after reflection at the downstream reservoir before closure, thus reducing the
280

magnitude of the pressure rise at the valve. For the case of linear valve closure, reflection from
the leak is soon noticeable even at the first pressure peak at the valve (see Figure 6). The waves
reflected from the leak are built up simultaneously as the primary pressure wave is experiencing
a reduction in magnitude. A quadratic valve closure implies a smaller pressure rise, and the
reflected waves from the leak are not seen until a certain number of oscillations have passed (see
Figure 7). However, the attenuation of the primary pressure wave is almost as strong as in the
case of linear valve closure. The leakage flow in the calculation examples presented in Figures
6 and 7 was about 6% (5 1/s) of the flow upstream the leak. Simulations with larger leaks were
also carried out, implying a more rapid attenuation of the main pressure wave.

PRESSURE
BO lm HzOI

60

40

zo

+------,,------.-----.--------.n~t~
100 200 300 400

Figure 6. Pressure at the valve for the case of linear valve closure with a leak located
upstream the valve.

As can be seen from Figures 6 and 7, a more distinct primary pressure wave in the system
implies more pronounced reflected waves originating from the leak that are easier to identify. It
could be quite difficult to distinguish a wave reflected from a small leak by visual inspection only,
especially if the primary pressure wave is not distinct. Also, in the field the measured attenuation
due to friction is often higher than the calculated attenuation from steady-state friction models
 281

(Jonsson 1986), implying that the number of pressure cycles in reality is lower than the number
obtained by calculation. Thus, attenuation due to friction in the field may exceed the growth of
waves reflected from a leak and complicate the analysis. In this context, spectral analysis may
prove a useful method in discerning any reflected waves in a recorded time series of pressure.

PRESSURE
JO lmHzOI

60

so

40

30

20

10

TIME
+-----.------.-----.-----.------.----.(5~1
100 200 300 400 500 600

Figure 7. Pressure at the valve for the case of quadratic valve closure with a leak located
upstream the valve.

Spectral Analysis For Detecting Reflected Waves

A time series of pressure may be represented by a number of periodic functions with varying
amplitude, frequency and phase (Jenkins and Watts 1969). By applying such a technique, it is
possible to distinguish predominant wave frequencies in the signal and thus to identify reflected
waves. The values in the pressure time series are transformed into the frequency plane by means
of the Fourier transform technique. The amplitude associated with a certain frequency represents
the contribution to the total variance from that frequency, which may also be interpreted as the
average power of that frequency.
282

The fast Fourier transform technique (FFT) was used to numerically compute the discrete
Fourier transform for time series of pressure. As an example, spectral analysis was performed
for the calculated pressure presented in Figure 7, and Figure 8 shows the calculated spectrum.
As can be seen in Figure 8, the predominant frequency is the fundamental pipe frequency a/4L.
However, it is possible to distinguish the result of waves reflected from the leak at a frequency
3a/4L in the spectrum corresponding to the location of the leak.

AVERAGE POWER
10

zo 40 60 80. 100

Figure 8. Spectrum for the sample calculation displayed in Figure 7 (fundamental

frequency: 1/658 Hz).

Measured Field Data

Field measurements were carried out in order to investigate the possibility of finding leaks in
hydraulic systems by analyzing pressure transients. The measurement site is described in detail
in JOnsson (1984). A system of pumps, connected in parallel, raises water to a reservoir from
which the water is artificially infiltrated into the ground to be used as drinking water. The static
head is 35 m H20 and the pipe has a total length of 5040 m. A tap is located at a distance of
1450 m from the reservoir, making it possible to simulate a leak with a maximum steady-state
flow of 5 IIsee out from the leak (the flow upstream of the leak is 75 IIsee). The pressure was
 283

recorded immediately downstream a check valve, located close to the pump station. Pressure
measurements were carried out after pump stop both with and without the leak, to quantify the
effect of the leak on pressure variation. Two representative examples of pressure measurements
with and without a leak are shown in Figures 9 and 10, respectively.

Pressul'l!

Time (sec)
0 10 20 30 40 50 60 70 80 90 100

Figure 9. Pressure at the check valve after pump stop without a leak in the system.

It is difficult to distinguish any waves reflected from the leak if Figure 9 and Figure 10
are only visually compared. Furthermore, cavitation occurred at a high point located about
1000 m from the check valve, influencing the pressure measurements at the pump station.
Initially, therefore a number of reflected waves appear between the check valve and the cavitation
pocket, which arises when the pressure drops at the high point after a pump stop. However, these
reflected waves almost disappear after the check valve has closed completely. The attenuation of
the primary pressure wave is about the same in the Figures 9 and 10.
FFT-analysis was carried out for the two measured pressure time series displayed in
Figures 9 and 10, and the result is shown in Figures 11 and 12. Unfortunately, the measurements
only encompassed a few cycles of the primary pressure-wave, which impaired the resolution of
the spectrum. If a time series only comprises a limited number of cycles in respect of a certain
284

frequency, the FFT-analysis is quite sensitive to the number of samples used in the analysis. In
order to identify a wave frequency clearly, this frequency should be a multiple of the fundamental
frequency (equal to 1/82.5 Hz in this case) used in the spectral analysis. In Figure 11, the basic
pipe frequency a/4L associated with the primary pressure wave is predominant. However, it is
also possible to distinguish the effect of some other waves at higher frequency in the spectrum,
originating from reflection against the cavitation pocket.
Pres9.Jre

o ~~~"''l-r'~'"l''-,'''-l"',-'''l''"''-rl''.-r''l-r'.-'"1''-r''-l"',-'''l-'"'''1 Time (sec)

0 10 20 30 40 50 60 70 80 90 100

Figure 10. Pressure at the check valve after pump stop with a leak located 1450 m
upstream the reservoir.

In Figure 12, the two frequencies immediately above the basic pipe frequency are much
more pronounced than in Figure 11. This is caused by reflections against the leak, which should
generate a wave with a frequency between 4/82.5 and 5/82.5 Hz. Thus, the average power from
this wave is distributed between the frequencies 4/82.5 and 5/82.5. A number of identical
measurements to the ones presented in Figures 9 and 10 were carried out to verify the presence
of reflected wave energy in the spectrum when a leak was present. The general tendency of an
increase in power within the range corresponding to the above-mentioned frequencies for cases
with a leak was confirmed by these measurements.
 285

AVERAfiE POWER

150
BASIC PIPE
FREQl£NCY

125

100

15

50

25

CAVIl ATl ON
MULTIPLE OF
FUNDAMENTAl
4-LJ_L~~-L~~~----~----~-rJ-~--;-~--~FR~EOOENCY
10 15 20 25 30

Figure 11. Spectrum calculated for a pressure time series measured upstream the check valve at
pump stop without a leak (fundamental frequency: 1/82.5 Hz).

AVERAGE POWER

150 BASIC PIPE

FREQUENCY

125

100

75

25 REFLECTION
WAVES AGAINST
THE LEAK
MULTIPLE OF
FUNDAMENTAL
f-~-L~~-L~--r------1-------r-------r----~F~REOOENCY
10 15 20 25 30

Figure 12. Spectrum calculated for a pressure time series measured upstream the check valve at
pump stop with a leak 1450 m from the reservoir (fundamental frequency: 1/82.5 Hz).
286

CONCLUDING REMARKS

By analyzing pressure transient measurements in a hydraulic system, it may be possible to detect

leaks. The leak generates reflected waves that are superimposed upon the primary pressure wave
and the reflected waves are dependent upon the leakage flow and the relationship between the
incident pressure wave and the local pressure at the leak. In a straight pipe, with a reservoir
upstream and a valve downstream, the primary pressure wave is significantly attenuated if a leak
exists in the system. Energy is transferred from the primary pressure wave to the waves reflected
from the leak every time the primary wave passes the leak. Reflected waves from the leak are
superimposed upon each other, which lead to an increase in the reflected wave pressure. Under
field conditions this effect may be difficult to distinguish, as frictional attenuation of the transient
pressure tends to exceed the friction estimated from steady-state models. By applying spectral
analysis to the measured pressure time series, it is easier to identify reflected waves in a hydraulic
system. In the paper a number of examples of are presented where this technique is employed,
encompassing both numerical simulations for sample cases as well as field measurements.

REFERENCES

Contractor, D. N., The reflection ofwaterhammer pressure waves from minor losses. Journal of
Basic Engineering. 1965, June, 445-452.

Jenkins, G. and Watts, D., Soectral Analysis and its Applications. Holden-Day, San Francisco,
1968.

JOnsson, L., Tryck:transientmiitningar i Renvattenledning - Alma An. Institutionen for Teknisk

Vattenresursliira, Lunds Tekniska Hogskola/Lunds Universitet, Rapport Nr 3089, Lund, 1984 (in
Swedish).

JOnsson, L. 1986. "Flow Structure of Transient Pressure Flow," 5th International Conference on
Pressure Surges, Hannover, 22-24 Sept, 1986.

Wood, D. J. and Chao, S. P., Effect of pipeline junctions on waterhammer surges. Transportation
Engineering Journal. 1971, 97, 441-457.

Wylie, B. and Streeter, V., Fluid Transients. McGraw-Hill, 1978.

 A REVIEW OF PIPELINE LEAK DETECTION TECHNOLOGY

PETER BLACK
Pipeline and Facilities Division
Scientific Software-Intercomp
Monarch House, Crabtree Office Village, Egham, Surrey TW20 8RY

ABSTRACT

There has long been a recognition of the importance of pipeline leak detection in oil and gas
transportation, an importance not unconnected with the economic realities of capacity loss
in an industry which moves billions of dollars of fluids every year. Such valid concerns have
been reinforced by the ever-present spectre of releases resulting in environmental damage
and potentially massive impairment of personnel safety. Recent legislative changes world-
wide have already provided considerable impetus for the adoption of enhanced monitoring
technology, and this trend is likely to continue. In view of this, this paper attempts to review
the options for leak detection, concentrating on model based techniques which offer the
potential for management as well as monitoring. Advantages and disadvantages will be
highlighted, emphasizing the selection of the appropriate solution for a particular pipeline
system. New developments in the field will be summarised, and particular reference will be
made to the expanding interest in leak detection for multiphase flow.

INTRODUCTION

The attitude of pipeline operators to leak detection is varied and unpredictable, and depends
on factors as diverse as individual experience, system reliability, product costs, internally
and externally imposed safety requirements, environmental sensitivity, and company policy.
Where policy bodies, such as the UK Health and Safety Executive, do not intervene it is
reasonable to describe current attitudes as pragmatic. However, it is clear that changing
conditions in the industry mandate a review of these attitudes, focussing particularly on the
effectiveness of measures implemented to detect leaks. It is important that the limitations
as well as the benefits of such systems are understood, in order to avoid unrealistically high
performance expectations. Real systems invariably represent a compromise between

287
288

accuracy, false alarm rate, and cost. This paper will attempt to provide broad guidelines on
different options that are available and an indication of their capabilities.

Each year oil and gas with a value measured in millions of dollars is transported both on
land and sub-sea, and even a small percentage loss can represent events with considerable
economic impact. In terms of the overall cost of construction and installation of pipelines
the expenditure on even the most sophisticated leak detection is comparatively insignificant,
though operators are correctly more concerned with value rather than absolute cost. To
represent value it is imperative that a detection system is sensitive enough to afford fast and
accurate notification of significant leaks while maintaining a frequency of false alarms below
that which would alienate control room staff.

However, it would be wrong to suggest that monitoring is solely driven by consideration of

financial loss. Concern has also been evident in the past in relation to environmental impact,
as was evident in assessment of proposals for development of Morecambe Bay and Poole
Harbour. Such concerns have been accentuated by increased awareness of the implications
of uncontrolled emissions, not least because of recent tanker incidents. Of equal, if not
greater, importance than environmental protection are considerations of safety, also an area
in which awareness has been dramatically heightened by recent events.

The context for the evaluation of the benefits of improved pipeline monitoring techniques
(including leak detection) is provided by the figures produced by groups such as the
European gas pipeline incident data group. As an example, the CONCAWE Oil Pipelines
Management Group's Special Task Force on Pipeline Spillages [1] reports that during 1989
there were thirteen reportable spillage incidents resulting in gross spillage of 2184 m3 which,
after clean-up, resulted in a net loss of oil into the environment of 1291 m3 • In an oil
pipeline network of nearly nineteen million metres this represents only 0.00024% of the total
volume transported, though the total cost of repairs and clean-up stood at 8.7 million ECU.
Seven of the leaks were classified as severe in one or more respects, and in one incidence
three fatalities ensued. While it is correctly identified that these incidents are
unrepresentative of the safe and reliable record that pipelines have established over many
years, nor is there any suggestion that leak detection would have reduced their severity, the
industry cannot afford complacency.

On this final point it is telling to consider the report on the offshore fire research programme
produced by Frank Knight in 1984 [2]. He pointed out that a disastrous event, involving the
loss of a platform and associated multiple fatalities, had an estimated event probability per
UK platform-year of 10·5 , implying that such a catastrophy had a 1% chance of occurring
on some North Sea platform during a ten year period. On 6 July 1988 the Piper Alpha
disaster provided a terrible vindication of Knight's foresight.

In the USSR and Eastern Europe the problems are undoubtedly greater, as underlined by
Alex Ustraykh from Bankers Trust during a London SPE Meeting. He is reported [3] as
saying that in the USSR "existing pipelines are frequently in desperate need of repair and
the average age is 12 years old with 20 per cent actually over 20 years old. Many were built
with insufficient insulating polymers and leakages cause accidents as well as representing
an enormous hazard to the environment, a growing concern among the increasingly vocal,
local population."
 289
It is interesting to consider, in view of the above, why leak detection systems are not
routinely employed on all pipelines. The reality of today's industry is that there is a great
deal of suspicion of, and a lack of faith in, the vendors of integrity monitoring technology
in general, and leak detection software and hardware in particular. This is undoubtedly due
to the exaggerated nature of many of the claims made for the performance of leak detection
systems, with sales literature that states that "...any size of leak detected"! Installed systems
have often been tuned to give the high sensitivity promised to the customer, at the cost of
a false alarm rate that drove operators to disable the system altogether.

This paper starts with a brief overview of various leak detection techniques, followed by a
more detailed discussion of systems which use a model of the pipeline network to compare
with field measurements. Modelling of the pipeline offers the potential for management
functions, and this possibility will also be reviewed. Hardware and software environments
will be considered, including display technology, and the paper will close with a view on
future directions.

OVERVIEW

This section provides a brief overview of most of the available methods for pipeline leak
detection, none of which is without its own particular advantages and disadvantages. There
is no intention to be exhaustive, but the list should be comprehensive enough to give a
flavour of the world of leak detection.

Observation

One of the simplest ways to detect leaks is to patrol the pipeline on a regular basis, making
visual checks on the integrity of the line. For short lines containing relatively innocuous
product this is ostensibly attractive, though the requirement for absolute regularity is onerous
and the method works less well for sub-sea lines.

In a similar vein, an odour may be provided for non-noxious gases through the injection of
chemical additives. Thus, the occurrence of a leak may be established during regular
inspections, though the effectiveness of this method may deteriorate depending on the
weather conditions and wind direction.

The disadvantages of such methods are clear, except in special circumstances, and they offer
little promise of prompt detection and minimal reassurance in terms of reliability.

Sensor-based methods

Into this category fall the general class of acoustic monitors which attempt to detect the
rarefaction pressure wave that is generated at the instant a rupture occurs. This negative
pressure wave can travel great distances with only moderate attenuation, and is particularly
useful for the rapid detection of large line breaks. However, Furness and Van Reet [4] point
out that problems may be encountered in "... setting the background threshold correctly as
this may be affected by the location of the instrument in relation to bends, valves, pumps,
290

regulators etc." They add that experience has shown that the number of false alarms is
affected by the installation and that reliable performance is also dependent on the Reynolds
number of the flow.

Instruments are also available which rely on the detection of the noise created by a leak,
generally produced in the ultrasonic range of frequencies. The associated software must be
carefully developed to ensure adequate filtering of extraneous signals.

Finally, evidence of escaping product may be deduced from changes in the electromagnetic
properties of buried cables following liquid contact, or from hydrocarbon gas detectors.

Aside from their own intrinsic problems, all of these methods can be very expensive except
when used for small lengths of line in highly sensitive regions. It should be noted that the
output from these sensors can be used in combination with a model-based system to permit
a small number of measurements to be used highly effectively.

Intelligent pigs

It is possible to use pigs for the monitoring of internal conditions in the line, and British
Gas in particular has widely publicised its developments in this area. Certainly, the
advertised performance of the intelligent pig is impressive, though it is clearly a regular
rather than continuous monitoring method. Its particular forte is the assessment of corrosion
damage and pitting, and the investigation of weld characteristics.

Since pig launches are a regular feature of off-shore operation there is clearly an attraction
to increased utilisation of intelligent pigs in this arena.

Single measurement analysis

SCAD A vendors have often provided leak detection based on the analysis of data from a
single pressure or flow measurement. Indeed, considerable resources have been devoted to
the development of so-called "pressure point analysis" (PPA) techniques which use statistical
methods to evaluate first and higher order derivatives of the pressure. The appeal of such
an approach, and its apparent simplicity, is undeniable: it is difficult to imagine a pipeline
operation without at least one pressure measurement.

However, such methods suffer from the important defect that they detect transients in
general, rather than leaks in particular. This is crucial, and has led certain vendors to supply
systems which are disabled for several minutes following an operating change (such as a
valve closure) in the line.

Uncompensated volume balance

Uncompensated volume balance is, in common with single measurement analysis, frequently
offered by SCADA vendors since it is simple to implement. As the name suggests, a
straightforward summation of the flows in and out of the network is performed, ignoring line
pack, recognising that conservation of mass is assured in the absence of a leak. Some
systems are slightly more sophisticated, in that they attempt to deduce line pack based on
 291

the limited available measurements combined with some form of interpolation, but this is
not always entirely successful. Thus, we have

where ..1. Vis the leakage volume, Vin is the metered inlet flow, V0 ut is the metered outlet
flow, and vj is the pipeline fluid inventory.

To avoid identification of changing line pack as a leak, uncompensated volume balance

techniques must compromise sensitivity or accumulate errors over a very long time period.
Neither of these options is particularly desirable, and dynamic model based volume balance
offers significant advantages.

Model based leak detection

The modelling of single phase pipeline flow is now a mature science, and large networks
can be solved extremely quickly with high accuracy. It has long been recognised that it
should be possible to utilise certain SCADA measurement values as boundary conditions for
simulations of flow conditions in the line, and using these predictions to accurately
determine the line packing in the volume balance equation discussed above. Such is the
elegance and apparent simplicity of this concept that work carried out in the early eighties
was undertaken with the promise that extraordinary results could be achieved. Unfortunately
it was not long before reality, and a rising tide of disappointment, became evident.

The development and implementation of a good model based system is a non-trivial

exercise, and requires more than a good simulation tool. For example, the selection of
boundary conditions, or the response to loss of instrument data or intermittent bad
measurement are crucial to the performance of the leak detection system. In addition,
properties of the fluids such as viscosity may influence sensitivity, as may operating
conditions such as the range of line flow rates.

An increasing number of highly successful installations of model-based leak detection

systems can now be identified, and there has been a significant and favourable change in
the perception of these techniques. The major disadvantages of such systems now lies in
their intrinsic complexity, which renders it difficult to evaluate vendors competing claims,
and cost. This last point has become less of an issue with the introduction of standard
software packages.

THE LEAK SENSITIVITY STUDY

The leak sensitivity study should be at the heart of any leak detection system installation.
Its existence reflects the fact that each pipeline network is essentially unique, with its own
mix of geometry, fluid properties, operating conditions and instrumentation: with this in
mind, it is clearly impossible to accurately predict the performance of any method for leak
292

detection without the performance of a sensitivity study. Without this, only very general
statements on sensitivity can be made.

When purchasing leak detection it is normally the case that the sensitivity study is carried
out prior to the final selection of a system. This allows the operator to evaluate competing
methods, and ensures that his expectations are commensurate with what is physically
reasonable and practically achievable. It is incumbent on the operator to expect and demand
that such a study is carried out.

MODEL BASED LEAK DETECTION

Recognising the predilections and affiliations of the author, the following sections will
discuss in greater detail the basis for model based leak detection. The first section will
review the basic flow equations, and associated issues such as real-time modelling, boundary
conditions, and equipment simulation. This will be followed by a discussion of methods for
leak detection.

Basic equations: the real-time model

While several different formulations exist, the essence of most modelling systems is
essentially the same and consists of the solution of the following equations for conservation
of mass, momentum and energy:

A. Mass Balance Equation

(Ap) 1 + (Apv)x =0
0 ~X~ L; t ~ 0

B. Momentum Balance Equation

p" (j)
v, + w .. + + gh" + - - v 1v 1 = o
p (2DJ

0 ~X ~ L; t ~ 0

C. Energy Balance Equation

pc,. (T, + vT..) -r{ap} CJT P

v
x
+ p (j)
2(D1)
lvl 3 - ( 4 Uw)
(D1)
(T- T)
6

Where:

L length of the pipe

X position along the pipe
 293

t time
A internal pipe area
p density of the fluid
p pressure
v velocity of the fluid
h elevation
g gravitational constant
f Moody friction factor
Di internal diameter of pipe
T temperature of the fluid
uw overall heat transfer coefficient
Tg ground temperature
Cv heat capacity of the fluid

As they stand, these equations do not represent a closed system and must be supplemented
by an equation of state (such as BWRS) and a friction factor formulation (such as the
Colebrook correlation). A variety of numerical methods may be used to solve this set of
coupled, non-linear equations, though the choice of implicit, unconditionally stable schemes
is fairly wide-spread. The resulting predictive tool is known as the real-time model (RTM).

It is worth noting the distinction between an RTM, and a model which is described as on-
line. Given a transient event, the RTM must be able to predict the system response before
it occurs in the real pipeline, and therefore must continuously be running faster than real
time. The designation "on-line" simply implies that the model takes a snapshot of the state
of the pipeline as the initial conditions for its predictions, but there is no particular
constraint on the speed of the subsequent simulation. Clearly, being on-line is necessary but
not sufficient to provide leak detection capability.

Boundary conditions may be determined rather easily for a well-defined set of equations
such as those considered above. Where fluid enters the pipeline network is temperature, fluid
type or composition, and either a flow or pressure is required. For exits from the system
only the fluid pressure or flow need be specified, though boundary conditions will also have
to be specified at internal points where equipment affects or controls pipeline pressure or
flow. While these boundary conditions must be supplied from measured SCADA data, it is
generally the case that additional (or redundant) measurements exist with which it is possible
to check consistency of the simulation. This consistency check is the basis of both the leak
detection methods discussed below.

Generally, the RTM is combined with an automatic look ahead model (ALAM), which runs
at a pre-specified frequency and analyses events in the line for a given period in the future.
Since this is done without operator intervention, this can run silently until preset limits (on
gas quality or pressure, for example) are violated and an alarm is raised. Thus, it is a
powerful early-warning device. To complete the trio, a predictive modelling (PM) capability
may be added to give the planner the chance to consider future operating scenarios based
on the current state of the network.
294

Model based volume balance

The accurate prediction of line pack and the effect of transients on volume balance can only
be captured if a model based system is employed.

In essence, the fundamental concept is identical to that in uncompensated volume balance,

in which the flow out of the network must equal the flow in or there is a leak (assuming no
changes in line pack). The significant advantage that modelling the pipe system brings lies
in the ability to predict changes in composition and fluid density arising from the transient
events that are an inescapable part of normal operations. This gives additional sensitivity at
the same time as minimising false alarms.

Generally, the volume balance is averaged over a number of time intervals, both short-term
(to give fast response for large leaks) and long term (to give sensitivity to large leaks). In
addition, threshold settings allow instrument inaccuracies to be accounted for.

The choice of boundary conditions for the model is extremely important in realising good
results with volume balance. However, there are advantages and disadvantages to all
selections, particularly when this method is used in conjunction with deviations analysis
(discussed below). It is commonly stated that volume balance has a low speed of response
to leaks, but this is not necessarily true and can be dramatically affected by the choice of
boundary conditions. Similarly, the selection of averaging periods can strongly influence the
performance quality.

Finally, contrary to popular belief it is practical to develop volume balance systems that will
permit leak location, and which do not significantly degrade as flow decreases.

Deviations analysis

Deviations analysis is based on the idea of redundant measurements, these being measured
data which are not used as boundary conditions for model calculations. It is clearly possible
to take any of these redundant measurements (pred' say) and compare it with the simulation
result (p•lm' say) to form a deviation
5p=p red-p slm"

Theoretically, when this deviation is non-zero it indicates a leak in the system, but the SSI
deviations analysis package (DV AP) also attempts to compute a range of normal deviation
based on known measurement errors. The addition of deviations analysis provides a
significant augmentation of the functionality available with volume balance.

While DV AP will not alarm on normal transient events, such as valve closure or flow rate
changes, the package does not model the pressure variations due to pig motions, for
example. Thus, leak detection capability will be significantly degraded during pigging runs.
This is not normally a problem, since volume balance is still enabled.

Tuning and instrument analysis

These two functions, while distinct, share a common foundation.

 295

Tuning is present due to the fact that the simulation results will never be identical with the
real pipeline pressure drop. This is a result of slight modelling imperfections, changing pipe
wall conditions, variations in ambient temperature, and other small but important factors
which influence the predicted flows. To avoid this the wall roughness and ambient
temperature are tuned slowly over a long time period to ensure that the model is close
enough to reality to avoid false alarms. Leak events are not masked by this tuning, since
their effect is seen on a much shorter time scale.

Instrument analysis recognises that one of the reasons for gradual loss of modelling accuracy
is instrument performance degradation (such as drift). The software detection algorithms
should be able to detect movement of measurement values outwith valid ranges and warn
the system engineer, though leak detection will continue. In addition, intermittent loss of
instrument data, due to SCADA or RTU failures, should not cause leak detection to be
disabled.

The paper by Van Reet and Skogman [5] discusses in greater detail the importance of
measurement uncertainty in real time modelling.

PIPELINE MANAGEMENT

It is not necessarily the case that the existence of a model implies that accurate predictions
of pipeline operation, including equipment performance, can be made. This apparent
contradiction arises because certain elements of the pipe network may safely be neglected
if leak detection is the only requirement. However, the SSI packages are based on their full
transient network simulation tools, and this permits access to a number of additional
applications. Among these are batch or composition tracking, pig tracking, pipeline inventory
analysis, over- and under-pressure prediction, instrument validation, compressor performance
monitoring, and predictive modelling.

As an example of the power which this affords, it is possible to consider a complex gas
network in which the operator has to guarantee certain levels of deliveries with a particular
heating value. Using the same technology that underlies his leak detection system, he can
guard against failure to meet his contractual obligations, and ensure efficient use of his line
capacity.

HARDWARE AND SOFTWARE

Leak detection software is available on a range of hardware platforms, including PCs,

workstations (primarily SUN and DEC) and VAX minicomputers. Often the choice for such
systems is limited by the selection of SCAD A hardware, and many vendors have a marked
preference for VAX VMS solutions.

The leaning towards VMS is not unreasonable, since it is an excellent real-time operating
system: its advantages in this respect are clear when it is compared with other operating
296
systems, including UNIX. However, the industry is moving rapidly towards open systems
and distributed computing, and this increasingly implies UNIX, X-Windows and TCP/IP.
This point is cogently argued for the pipeline information management and control systems
industry by Nagala [6]. Further, graphics products such as the X-Windows based Sammi
offer the possibility of full data distribution and exchange, with graphical displays of
SCADA and model information in the control room and in the head office. Each user can
see a different selection of a vast pool of data, including networked databases, displayed in
the format which is of greatest use to him. This technology is available now.

FUTURE DIRECTIONS

It is difficult in a paper of this length to give anything other than an indication of some of
the highly innovative developments in the industry. However, one area that is of increasing
concern, and requires immediate attention, is the application of leak detection to multiphase
flow.

The flow of oil and gas mixtures in pipelines is increasingly a feature of the offshore
industry, and it has been predicted that by the year 2000 almost all of the North Sea reserves
will be accessible by sub-sea tie-backs to existing facilities. This demands a significant rise
in the level of multiphase transportation. Unfortunately, the inherently transient nature of
multiphase flow, in which behaviour such as slug flow is possible even under steady flow
conditions, means that measured pressure and flow may fluctuate dramatically. This renders
most methods for leak detection, aside from model-based techniques, of dubious value.

To consider modelling of transient multiphase flow is to enter into a number of areas which
are still the subject of research. However, a small number of working simulators do exist,
the most widely used and heavily funded of these being the OLGA model from Norway (see
[7]).

This code was used as the basis for a computerised monitoring system for the Troll Oseberg
Gas Injection (TOGI) pipeline in the Norwegian sector of the North Sea. This pipeline takes
unprocessed gas from the Troll to the Oseberg fields to provide gas injection to aid oil
recovery. Ek et al. [8] state that the fluid has a liquid drop out of approximately 4.5% (wt)
at typical pipeline conditions. Boundary conditions are provided by the well head pressure
together with reservoir pressure: it is argued that the flow changes in the well are very fast
compared with those in the pipeline, and therefore that steady state data can be used to
provide well head temperature and flow as a function of well head pressure. On-line tuning
is accomplished using wall roughness and ambient temperature, as was discussed previously.
Ek et al. discuss leakage and blockage detection in greater detail in their paper.

It is clear that the use of OLGA and similar simulators provides the only clear route to leak
detection on multiphase pipelines, and SSI and Scandpower a.s (the latter are responsible
for the marketing, sale and support of OLGA) are actively involved in analysing the options
for a full commercial system. This is an exciting area but one that has many difficulties.
 297

CONCLUSIONS

In this paper there has been no attempt to recommend a particular method for leak detection,
concentrating instead on a summary of the available options and their particular advantages
and disadvantages.

Leak detection is an issue of undeniable importance, and it is likely that impending changes
world"wide will increase pressure to implement systems and to demonstrate effectiveness
of those systems. Particular attention in this paper has been given to model-based systems,
and difficulties with early installations have been discussed. The benefits of such an
installation are, however, very real, as is evident in the paper by Mears [8] of the Williams
Pipeline Company. The technology is now robust, and standard systems and applications are
readily available.

Ultimately, it is important to choose the method that is appropriate for the particular pipeline
and its operation, using the leak sensitivity study as a guide to the likely effectiveness of
any particular choice. Good advice is readily available, but is rarely entirely unbiased.

ACKNOWLEDGEMENTS

The author would like to gratefully acknowledgement the significant influence of J. Van
Reet and R.E. Nicholas on the contents of this paper. However, any errors that may be
identified are clearly my responsibility.

REFERENCES

[1] I:.yons, D., Beuillard, C., Meriggi, C.R. and Pries, P., Performance of oil industry
cross-country pipelines in western europe (statistical summary of reported spillages-
1989). Prepared by the CONCA WE Oil Pipelines Management Group's Special Task
Force on Pipeline Spillages (OP/STF-1), November 1990.

[2] Knight, F.I., Review of the Offshore Fire Research Programme, October 1984.

[3] SPE Review, European Edition 27, December 1991.

[4] Furness, R.A., and Van Reet, J.D., Pipeline leak detection techniques, Pipeline
Integrity Monitoring Conference, September 1986.

[5] Van Reet, J.D., and Skogman, K.D., The effect of measurement uncertainty on real
time pipeline modelling applications, ASME ETCE Conference, 1987.

[6] Nagala, D.W., Solving data availability problems in pipeline information

management and control systems, Pipeline, February 1988.
298

[7] Bendiksen, K.H., Espedal, M., and Maines, D., Physical and numerical simulation
of dynamic two-phase flow in pipelines with applications to existing oil-gas field
lines. Presented at the Conference on multiphase flow in industrial plants,
Bologna, Italy, September 1988.

[8] Mears, M.N., Major pipeline installs system to pinpoint leak size, location. Oil &
Gas Journal, April 1988.
 FLOW MEASUREMENT IN LARGE COMPLEX DUCTWORK

Alan Vardy
Royal Society/SERe Industrial Fellow
University of Dundee and Mott MacDonald

ABSTRACT

Field measurements of air flow rates in a road tunnel are used to illustrate difficulties
encountered by engineers attempting to assess the performance of complex piping systems.
At most times of day, direct measurements in the roadway are inhibited by vehicles. At all
times, the flows are spatially varied. Simpler conditions exist in some ducts, but few
accessible sections are more than one duct diameter from a major discontinuity.
In any piping system, failure to obtain accurate flow estimates can have serious
consequences - consider leak-detection, for instance - so methods must be devised to
overcome deficiencies. The method used herein is specific to its application, but it
nevertheless has features of general validity.
Analogous difficulties with numerical analyses are described and consequential
problems with the validation of computer codes are raised. It is concluded that no technique
should be based on an assumption of an exact knowledge of conservation of mass, either from
measurements or from numerical computations.

INTRODUCTION

Accurate flow measurement is a prerequisite of many aspects of flows in pipes and ducts,
including system control, leak detection, invoicing and the validation of computer software.
Often, high accuracy is attainable with off-the-shelf equipment needing little maintenance, but
situations also exist in which much lower accuracy has to be accepted. Typically, these
involve either unusually large ductwork or unusually complex ductwork. This paper deals
with an example involving both of these complications. In this particular case, some flow
rates are believed to have been determined with an accuracy of about ± 10%, but others are
reliable only within ±30%. This is an extreme example in the context of general piping

299
300
systems, but it is a valid illustration of a common problem, namely flow measurement at
large scale.
The particular example discussed herein concerns flow measurements made during a
major field study in the Tyne Tunnel near Newcastle-upon-Tyne in the UK. The
measurements were only one part of a wide investigation and did not merit the allocation of
extensive resources, either for equipment or for manpower. The problem was loosely
analogous to flow measurement in a large diameter water pipeline without electro-magnetic
flowmeters.
The most obvious complication in a vehicle tunnel - road or rail - is the presence of
vehicles. These greatly restrict the types of measurements that can be carried out if the
traffic must pass unimpeded through the duct (tunnel). Traffic also causes great local
disruption to the flow, thereby reducing or even eliminating the number of regions where
reliable assumptions can be made about velocity distributions. Flow measurement can be
undertaken in ventilation ducts supplying the main roadway, but even these can be far from
simple. In the Tyne Tunnel, for example, the main supply duct is intentionally leaky (see
below) and some ducts involve no regions of straight, parallel flow.
One of the secondary purposes of the investigation in the Tyne Tunnel was to assess
the validity of newly developed computer software capable of predicting the distribution of
pollution along a tunnel. Briefly, the program first predicts flow rates throughout the tunnel
and its ventilation system, and then tracks pollutants released by vehicle exhausts. The air
flows are influenced by vehicle drag and by ventilation fans. The rates of emission of
pollutants depend upon local conditions such as speed and gradient. Since many factors
contribute to the distribution of pollutants, it is desirable to measure more basic parameters
such as flow rates when attempting to validate software.
Two specific sources of potential error in the software should be mentioned. Firstly,
the tunnel ventilation system involves spatially varied flow between the roadway and the main
air supply duct. This is closely analogous to free-surface flow over a side weir between two
channels; neither case can be represented exactly by one-dimensional flow approximations.
Secondly, the particular numerical scheme used in the program is inherently incapable of
ensuring mass conservation exactly. This is also true of many computer programs used in
the oil, water and gas industries, although it might not be so widely recognised. For present
purposes, the importance of these two effects is that they dictate the necessary degree of
accuracy of any experimental measurements designed to assess the capability of the program.
 301

In summary, the purposes of this paper are (i) to highlight the general problem of flow
measurement and prediction in large and/or complex pipework and (ii) to describe the
methods used to measure flow rates in the Tyne Tunnel and to invite discussion on alternative
methods.

TYNETUNNEL

The field tests were undertaken during 1989 and 1990. The principal objective was to assess
the extent and significance of thermally stratified flows in the tunnel, but an important
secondary purpose was to improve understanding of the causes of a maldistribution of exhaust
fume concentrations along the tunnel - with a view to improving the conditions if possible.
The remainder of this Section is devoted to a description of the tunnel ductwork and
to the methods used to assess the rates of flow. Thereafter, attention is focused on action
taken after it became clear that measurements of the desired accuracy would not be obtained
with the available resources. Some tentative inferences are made about flow measurement
in more general pipe systems.
The tunnel and the ventilation ducts are illustrated schematically in Figures 1 and 2.
The space below the roadway is used as an air supply duct which receives air from two fan
shafts, one at its northern end and one about two-thirds of the way to its (blocked) southern
end. Air leaks into the main roadway through kerb-level "windows" and exhausts to
atmosphere either through the portals or through exhaust shafts. When the rate of flow is
enhanced by fans in the exhaust shafts, the direction of flow through one or both tunnel
portals can be inwards. This type of ventilation system with a lateral air supply to the
roadway, but longitudinal exhaust, is known as semi-transverse. 1
To obtain a good knowledge of air flows throughout the system, it would be
preferable to measure flows along the main roadway, along the supply duct, and within all
four fan shafts. Alternatively, if sufficiently accurate measurements could be made, it would
be possible to deduce some of these parameters from the others, but there is relatively little
scope for this. For example, an accurate knowledge of all flows in the tunnel and in the

Some tunnels have fully transverse systems where the exhaust as well as the supply
is lateral; in these cases, the vitiated air is removed along a separate exhaust duct.
Many modern tunnels have longitudinal ventilation systems in which there are no
independent supply or exhaust ducts; in these cases, all air travels along the same duct
as the vehicles.
302

NORTH SOUTH

Jl roadway \/
~----~/------~~----~

Figure 1. Tyne Tunnel: general arrangement

supply duct would enable the rates of flow to be deduced in all four shafts. Unfortunately,
the converse does not hold because the flows along the tunnel and the supply duct are
spatially varied.

Roadway
Measurements in the roadway were deemed impracticable at the times of day when they are
most important, namely when traffic densities are high. Traffic plays a significant r6le in the
development of the air flows, but its presence causes large local disturbances, especially in
a tunnel carrying two-way traffic. No method of measurement involving obstructions in the
roadway space was acceptable. Methods involving no obstructions are possible in principle,
but most of them are extremely difficult to calibrate. It was therefore decided at an early
stage that flow measurements would not be made in the roadway itself.
A seemingly inevitable consequence of this decision was that the air flows had to be
measured as accurately as possible in all shafts and along the whole length of the supply duct.
None of this information is redundant in the absence of measurements in the roadway.

Supply Duct
The most accurate flow measurement was undertaken in the supply duct, using hand-held
anemometers. Independent readings were obtained with (i) a hot wire probe, (ii) a 15 rom
diameter propeller meter and (iii) a 100 rom propeller meter. The frrst two of these were
calibrated in a wind tunnel and the third was calibrated in-situ by comparison with the others.
Spot velocity measurements were taken at regular intervals along the axis of the duct
and traverses were undertaken over a few cross-sections, mainly close to the bases of the
supply shafts where the velocities were greatest. The traverses were not carried out in
accordance with BS 1042 or even with the less demanding BS 848. Instead, just sufficient
 303

t A

Figure 2. Supply and exhaust shafts at northern end of tunnel

measurements were taken in the cross-section, especially close to the "windows", to give
reasonable confidence in the assumed velocity profile. This methodology cannot be justified
rigorously, but it is easily explained. Firstly, the air flows in the duct are not steady; they
are influenced quite strongly by vehicles in the main tunnel. Measurements permitting
averages over long times were judged more important than spatial averages. Secondly,
conditions in the supply duct are very uncomfortable close to the shafts, with air velocities
approaching 70 km/h.
With the benefit of hindsight, it would have been sensible to obtain instruments such
as those used in coal mines and to follow the procedures described in the relevant National
Coal Board Handbook (1979). Nevertheless, reasonable agreement was obtained between
different operators on different days, thereby justifying a fair degree of confidence in
measurements along the duct.

Supply Shafts
The North Supply Shaft is illustrated schematically in Figure 2. The South Supply Shaft is
slightly more complex, but it need not be described herein. The main regions are:
304

A: an inlet region, square in plan (4.9m x 4.9m) with a slightly streamlined junction to:
B: the fan in a circular section housing;
C: a vertical diffuser of constant outer diameter, merging with:
D: an approximately conical tube, with a streamlined, vaned elbow to:
E: an approximately rectangular horizontal section, with another streamlined, vaned
elbow to:
F: the main supply duct beneath the roadway (R).

No flow section is much more than one shaft diameter from a major discontinuity. Every
flow section is therefore disqualified from use for flow measurement in BS 848 even though
this deals specifically with fans and recognises that many installations are complex.
Nevertheless, some attempt had to be made to estimate the flow rate, and this was achieved
with velocity traverses at several sections and with an ultrasonic flow meter. The flow rates
obtained by the various methods were 189 m3/s, 214 m3/s, 169m3/sand 216 m3/s.
A weighted average of these values, allowing for engineering judgment about the
difficulty of obtaining and interpreting the measurements led to an estimated flow rate of
205 m3/s ±10%. Similar accuracy is claimed for the south supply shaft.

Exhaust Shafts
The North Exhaust Shaft is illustrated in Figure 2. At first sight, this is a more favourable
arrangement than the supply shaft. Unfortunately, however, there is no safe means of access
to the parallel flow region below (upstream of) the fan. Once again, therefore, readings were
obtained by velocity traverses and by an ultrasonic flow meter. The apparent flow rates were
177 m3 /s, 133 m3/s and 110 m3 /s and a suitably weighted is 128 m3 /s ±30%. Similar
accuracy is claimed for the South exhaust shaft.

ENGINEERING SOLUTION

The poor levels of accuracy in the estimation of flow rates led to several reappraisals of the
methods of measurement. Some alternative methods are discussed briefly below, but none
seems certain to lead to major improvements in a system of this type. As so often in
engineering, a solution to the main problem therefore had to be obtained on the basis of
inadequate knowledge and, for the sake of completeness, the chosen method is described
briefly.
 305

The most accurate flow measurements were those in the supply duct. These enabled
the rates of flow through the windows into the roadway duct to be estimated quite well. Thus
axial variations in the roadway flow rates were known. Absolute flow rates were then
deduced by locating null points (Figure 3) in the regions of tunnel (a) between the shafts and
(b) near the South portal.

null point
~.......----- .......- ~ ....- - ~
.......... - _..., __.
rt-t-t-t-t-t-t-t-t-t-t-t---,

Figure 3. Null point in a spatially varied flow

(When the lateral flows and the location of the null point are known,
the flow distribution along the main duct can be deduced)

This procedure enabled a good qualitative understanding to be built up of the flow

distribution along the tunnel. The most important features of the behaviour were found to
agree well with numerical predictions from the newly developed computer program and the
latter was used to explore a wide range of alternative fan settings. Eventually, a very simple
control system was devised, based on pollution measurements at two locations in the tunnel.
The control system does not attempt to predict flow rates; it simply determines which fans
should be speeded up and which should be slowed down in order to maintain predetermined
standards of air quality in the tunnel.
Unfortunately, the need to measure flow rates accurately cannot always be
circumvented in this way, even in this particular tunnel. The computer program has also
been used to assist in the selection of fan settings in the event of fire or other serious
incidents. In these cases, a knowledge of absolute velocities at key locations in the tunnel
is necessary.

ALTERNATIVE METHODS OF FLOW MEASUREMENT

In this investigation, the main method of flow measurement was by velocity traverses over
a flow section. This was reasonably successful in the supply duct, even though the flow is
spatially varied, but it was very unreliable in the shafts. Ideally, a more direct method of
flow measurement should be used instead of velocity measurement. The main possibilities
are now discussed.
306

Ultrasonic Meters
The particular ultrasonic meter used in this study was initially faulty, but this was not easily
deduced because so many other factors introduced uncertainty. Even after repair, the
readings could not be relied upon absolutely because the instrument was unavoidably mounted
in non-ideal locations. In the case of the exhaust shafts, for example, its line of sight was
in a conical tube and passed through the wake region above (downstream of) the fan hub.
Even if its reading is assumed to be a true measure of the mean velocity along its line of
sight, the estimation of the shaft flow rate is strongly dependent on the assumed velocity
distribution in this complex location. It is known that ultrasonic meters can give reliable
readings in suitable locations, but they are not well suited to installations with contorted
geometries.
Ultrasonic meters essentially measure a Doppler shift. This gives them a valuable
advantage over most other forms of flow measurement, namely an insensitivity to swirl
components of velocity, provided that these are normal to the line of sight of the instrument.
They are influenced by radial components of flow, but these are usually small.

Orifice Meters
In a laboratory and in many pipelines conveying homogeneous fluids, venturi meters or
orifice meters are highly reliable. Consideration was given to the construction of orifice
meters in the Tyne Tunnel, but no accessible location could be found where the flow
disturbance would be acceptable. In common with ultrasonic meters, they could have a place
in flow measurement in some tunnels, but they will usually have too great an influence on
the overall resistance to flow. That is, they will force the fans to operate away from their
normal range on their characteristics performance curves. If installed anywhere other than
in a fan shaft, they could even change the behaviour of the system itself.

Tracer Gases
The most promising method of flow measurement is by the use of tracer gases. A tracer is
released continuously at a known rate at a suitable upstream location. It mixes with the air
and samples of the mixture are collected at suitable downstream locations. The
concentrations of the tracer in the samples give an accurate measure of the rates of air flow.
In principle, this method is simple and yet it is potentially accurate. The key
requirements for its successful use are:
 307

1. the tracer gas must be non toxic and easily measurable;

2. it must be released at an accurately known rate;
3. mixing must be thorough;
4. the sampling process must be leakproof;
5. the measurement of concentrations must be accurate.

Dayman (1982) used this method to measure air flows induced along a tunnel by a
fleet of buses. Mixing was thorough, partly because the buses created considerable
turbulence. In that study, greatest difficulty was found in Step 5, namely the measurement
of concentrations in the mixture. The tracer was SF6 and the gas chromatograph was highly
non linear at the low concentrations involved. Nevertheless, careful calibration led to
reproducible results which are believed to have been accurate.
Despite this success, the method was not adopted in the present study. There was
doubt about the degree of mixing of the tracer in the airstream, notwithstanding the intended
passage through a fan, and there was doubt about the sampling process which would have had
to be more elaborate than in Dayman's study. The latter problem could be overcome, but
the former might not have been. Looking back, however, this method would probably have
been more accurate than the velocity traverses that were undertaken instead.

FLOW MEASUREMENT IN OTHER PIPING SYSTEMS

What are the implications of this work for designers and operators of piping systems in
general?
Firstly, consider the accuracy of individual flow measurements at any particular
section. The problems in the Tyne Tunnel stemmed from (i) a lack of accessible sections
sufficiciently remote from discontinuities for well developed, parallel flows to exist and (ii)
a lack of reliable methods of flow measurement even at suitable flow sections. These
problems should be less common in pipelines, but they certainly exist occasionally. Few
suitable flow measurement sections exist in the neighbourhood of some pumping stations, for
instance. Likewise, even where suitable sections are accessible, accurate measurements may
be possible only with prohibitively expensive equipment, especially in large diameter
pipelines. They may be unattainable at any price in pipelines conveying liquid/gas or
liquid/solid mixtures or free-surface flows.
Secondly, consider the system as a whole. In the tunnel, the spatially-varied nature
of the flows along the main ducts made it difficult to infer flowrates at one location from
308

known flowrates at adjacent locations. This would have been possible if (i) the distribution
of the spatially varied flows and (ii) the rates of flow at the adjacent locations had both been
known exactly, but neither of these conditions was satisfied in practice. Instead, an
approximate method of deducing flow rates was developed, using null points which could be
located with reasonable accuracy. The analogy with water supply systems is leakage. In
some ways, this is more difficult to assess than in the tunnel where the distribution of leakage
holes is known a priori. In other ways, however, it is easier to assess because the leakage
is a true loss, not simply a transfer to another pipe. The point is that regions of relatively
high leakage might be identified more easily by deductive reasoning using remote flow
measurements than by detailed local measurements. The more remote the measurement sites,
the less accurate the readings need to be.
It can be useful to extrapolate the concept of remote measurements much further. The
gas supply industry has adopted a delightfully effective approach to leak detection, namely
added "smell". In effect, human noses are used as an extensive supply of leak detectors. It
is not obvious how the oil or water industries could mimic this approach, but there may be
some cases where the detection of tracer elements is feasible after leaking from a pipe.
Tracer elements could also be used for the indirect detection of leaks. For example,
the flow could be seeded in a similar manner to that described above for air flow
measurements using tracer gases. Known concentrations at upstream locations should result
in predictable concentrations further downstream. If they do not, either the predictions are
false or there is a leak.

Numerical leakage
As mentioned briefly in the Introduction, leakage is not confined to physical artefacts.
A great deal of numerical analysis is undertaken for pipeline systems, sometimes with the
objective of providing benchmarks against which to assess actual performance. A surprising
number of such analyses "leak" like their physical counterparts, especially along pipelines
containing zones of compressible flow, two-phase flow, free surface flow or transient flow.
The computer program referred to above is no exception.
In the case of unsteady flows, the only numerical schemes that do not suffer from this
problem are those based on the so-called 'conservation' form of the continuity equation. In
practice, such methods also use the conservation forms of the momentum and (if applicable)
energy equations so they also conserve other important physical quantities along any
 309

particular pipe. If this were the only criterion on which the analyses should be judged, these
forms of the equations would be used almost exclusively. In practice, however,
complications arise when these methods are applied to flows in pipe networks, especially in
the representation of boundary conditions. As a result, although the various quantities are
conserved along individual pipelines, they are not necessarily conserved for the system as a
whole. As a consequence, conservation of mass and momentum should not be taken for
granted even in computer programs.
There are obvious implications for the vaidation of computer software. If neither
measured nor predicted flow rates satisfy conservation of mass, how can the accuracy of
software be assessed? The answer, of course, is that validation is rarely as precise as users
would wish. Instead, engineering judgement is necessary and the best checks are those which
test the particular features that really matter in any particular installation. The important
thing is to be aware of likely limitations when devising the checks.

CONCLUSIONS

6.1 Flow measurement can be exceedingly demanding in complex ductwork at large scale.
In the recent investigation in the Tyne Tunnel, some measurements of flow rates were
estimated to be subject to a tolerance of ±30%.
6.2 In the tunnel investigation, the most effective solution to this problem was to identify
other parameters whose measurement would obviate the need to improve the poor
levels of accuracy in difficult locations.
6.3 It is speculated, by analogy, that efforts to circumvent metering difficulties in other
industries - eg. water supply - might be more productive than efforts to improve or
extend metering.
6.4 Numerical analyses of pipeline flows can leak, too.

ACKNOWLEDGEMENTS

The author would like to thank Bain Dayman, Jet Propulsion Laboratory, California and Ken
Evans, manager of the Tyne Tunnel, and his staff for extensive help in obtaining and
interpreting the measurements described herein.
310

REFERENCES

BS 848 (1987) "Fans for general purpose, Part 1: Methods of testing performance",
Amendment 2, British Standards Institution.

BS 1042 (1984) "Measurement of fluid flow in closed conduits, Part 2: Velocity

measurements", British Standards Institution.

Dayman, B (1982) "Small-scale aerodynamic testing for highway tunnels", Proc 4th int symp
on the Aerodynamics and Ventilation of Vehicle Tunnels, BHRA Fluid Engrg, York, UK,
483-504.

National Coal Board (1979) "Ventilation in coal mines: a handbook for colliery ventilation
officers", NCB Mining Department, London.

Sanderson, M L (1982 "Electromagnetic and ultrasonic flowmeters: their present states and
future possibilities", Electronics and Power, Institution of Electrical Engineers, February,
161-164.
 STRUCTURAL CONDITION ASSESSMENT OF WATER TRUNK MAINS

ROGER OLIPHANT
Engineering and Structural Performance Group
WRc, PO Box 85, Frankland Road, Blagrove
Swindon, Wilts, SNS 8YR, UK

ABSTRACT
As part of the exercise to quantify appropriate levels of
investment, the Water Industry has been investigating the
condition of its under ground assets. Many of its trunk main
systems, which form a critical initial link in the
distribution of water, were constructed around the turn of
the century. Over the past six years WRc has developed a
methodology and adapted various measurement techniques to
carry out structural condition assessment of these systems.
The approach has now been described and illustrated with
actual case histories in a guidance manual which was
produced with the financial support of various water
utilities. This paper outlines the methodology and provides
details of its application.

INTRODUCTION

As part of the privatisation process, each prospective new

water company had to quantify its future investment needs to
maintain and improve the performance of its asset base. These
Asset Management Plans are subject to scrutiny by the water
industry regulator, the Director General of Water Services.
The preparation of these plans necessitated an assessment,
often for the first time, of the structural condition and
operational performance of trunk mains. Prior to this, and the
original reason for developing the structural condition
assessment methodology, trunk main systems were only examined
to establish the engineering reliability of a particularly
311
312

strategic asset or where a major change in operational practice

and pressure regimes were contemplated.

THE METHODOLOGY OF CONDITION ASSESSMENT

The basic skill in a structural condition assessment is that

from a limited number of inspections, discrete in both space
and time, conclusions are drawn about the current condition of
the system as a whole and the way this will change over time.
Obviously for this to be achieved successfully the inspections
made have to be collectively representative of the system and
some knowledge is required of the mode and rates of
deterioration of the pipes and other structures of which it is
made up.

In order to obtain a representative set of inspections the

system has to be analysed into a series of categories that have
been found from previous studies to be significant from the
corrosion point of view. These include the environmental
factors of soil and water types, and there are the pipe
parameters of material type, diameter, presence of any
protective coatings, time since laid etc. Then there are the
different structures in the system, pipes and valves are
obviously the chief of these but pipe bridges and various
headworks at the beginning and end of siphon sections may also
be present.

Another way of analysing the system is into 'strategic

categories' by considering the consequence of failures. There
may well be sections of the system where failures would be
particularly undesirable eg where mains pass under motorways,
across electrified railway lines on major inter-city routes,
through the grounds of hospitals or schools or near the house
of the local MP. It may well be prudent to include inspections
for some of these sections into any programme for reassurance
if nothing else. Of course the two methods of categorisation
are not mutually exclusive, at worst they may be complementary
 313

and often they will overlap so that examination of a particular

strategic section will cover a particular set of environmental
categories of interest as well.

It can be seen from the above that many of the categories

identified will have a series of sub-divisions or levels. The
extent to which all these have to be covered in the inspection
programme will be determined by the precise objectives of the
study and the extent to which the different levels turn out to
have real significance in the particular study. On both counts
the inspection programme needs to be carried out in a
structured manner so that the field work can be terminated at
the earliest opportunity. This entails starting with what are
expected to be the worst levels in each category; if these show
the system to be in good condition, little further work, other
than of strategic sections, would be justified. If, however,
significant corrosion is discovered, further inspection of the
less corrosive levels would be required to define the extent of
the problem. Inspections would be continued in this way until
there was data of sufficient quantity and quality to allow the
objectives of the study to be achieved, or the new information
was not refining the picture any further eg not reducing the
scatter in the results any further. In the latter instance,
where a definitive conclusion cannot be reached, engineering
judgement has to be applied based on the data available.

ASSESSMENT METHODS

The techniques used to assess structural condition have been

adapted from traditional corrosion methods.

Soils

The assessment of soils involves using soil maps, to identify

the different types the pipes pass through from which any
tendency to waterlog can be gauged. Apart from that, soil
resistivity and pH are the only parameters routinely measured.
314

The former is determined on site using the Wenner four pin

method and in the laboratory in a soil-box. An approximate
correlation exists between resistivity and pitting of iron
pipes below a value of 3,000 Qcm. However, as with most
factors involved in this type of corrosion, resistivity gives
only a guide to pipe condition, it is not a predictor.

pH of the soil determined in the laboratory according to

BS 1377 by mixing a weighed sample with deionised water and
carrying out the measurement on this extract. Obviously the pH
value of soil is not only relevant to ferrous pipes,
cementitious materials are also affected. Other soil parameters
have been suggested as a guide to corrosivity eg redox
potential, presence of sulphide ion etc. But the correlation
between these and pipe condition is even more tenuous and so
are not routinely measured.

Ferrous Pipes

The techniques used to assess the structural condition of

pipelines are obviously dependent to some extent on the
material type. For ferrous pipes the external corrosion is
measured directly, after exposing the pipe and shotblasting
back to sound metal. The pipe is marked up into a grid so as
to correlate the number of measurements made to surface area
examined. The five deepest pits in each grid square are
measured by a simple pit depth gauge. The internal corrosion
can then be determined by measuring the remaining wall
thickness using an ultrasonics probe or by direct measurement
where access is available. Internally the corrosion is more
uniformly spread than externally, which may be a reflection of
the more homogeneous water environment compared to the soil.
 315

Figure 1. Cast iron pipe with grid system and external

corrosion highlighted

Another technique that has been used with ferrous mains is

measurement of pipe to soil electrochemi cal potential. This
involves making an electrical connection to the pipe at some
convenient point eg an air-valve and connecting the wire via a
high impedance voltmeter to a Cu/CuS0 4 reference electrode.
Placing the reference electrode in contact with the ground
above the pipe completes the circuit and potential drop at the
pipe/ soil interface is displayed on the voltmeter. A detailed
potential survey of the whole main can be made by moving the
reference electrode along the line of the pipe.

The difficulty in the technique is in the interpretatio n of the

results. This is because, although the electrochemi cal
potential, unlike other parameters such a resistivity, is a
direct measurement of the interaction between the pipe and
316

soil, it measures the thermodynamic driving force of the

chemical reactions involved and not the corrosion rate. It is
the corrosion current, generated by the electrochemical
interaction between pipe and soil, that is directly
proportional to the corrosion rate and this is related to the
electrochemical potential by the polarisation resistance of the
particular soil. Unfortunately, attempts to measure this
resistance directly on a sample of soil in the laboratory have
so far failed. Thus at present the only way of determining the
significance of electrochemical readings is direct calibration
by condition assessment of pipes in each soil type.

Electrochemical potential measurements have been used to locate

corrosion hotspots on mains and to detect deterioration of high
quality anticorrosion coatings.

Cementitious Materials

For cementitious materials eg asbestos cement, reinforced

concrete etc, the extent of deterioration can be quantified by
the phenophthalein test. Unfortunately this requires a ring or
core sample to be taken, which may not be convenient.
Phenolphthalein is an acid/base indicator that turns from
colourless to bright pink above a pH of 8.3. As the original
hydration reaction of cement produces free lime, cementitious
materials should stain bright pink. Cementitious material
deteriorates through the loss of this lime, either by direct
leaching or by reaction with carbon dioxide, so that the
affected areas fail to stain pink when sprayed.

Valves

A radiographic technique has been developed for the inspection

of valves and rebars in reinforced concrete. This method uses a
y ray source, iridium 192, which is placed one side of the
object being radiographed with a photographic plate on the
other. Where there are two walls between the source and the
 317

plate eg when shooting through the body of a valve, the

features on the wall nearest the source get obliterated,
because of the intensity of the beam, and only the features of
the wall nearest the photographic plate are recorded. Of
course, by switching the apparatus around, and shooting from
the opposite direction, the other wall can also be examined on
its own, hence the name of technique 'double wall, single image
approach'.

Figure 2. Removing the corroded gate from a 27" valve

uPVC

For uPVC pipes the tests applied are those in BS 3505:1986 ie

immersion in methylene chloride to determine the degree of
gelation and fracture toughness measurements.
318

The degree of gelation is a measure of the homogeneity of the

polymer melt prior to extrusion. Poor homogenation results in a
pipe with poor gelation which is strongly attacked by the
methylene chloride. The attack turns the pipe white and may
roughen the surface. The degree of gelation is believed to be
related to the fracture toughness of the pipe ie its ability to
withstand mechanical shock.

Fracture toughness is quantified directly by hanging a weight

on the lower limb of a notched c-ring sample and measuring the
time to failure. The weight used is calculated from the
dimensions of the sample being tested. Unfortunately, the
present state of knowledge does not allow these measurements to
be projected into the future ie there is no established mode of
deterioration for which a rate can be calculated. Attempts to
obtain such information, by exhuming pipes with a range of
ages, have always been frustrated by the significant
improvement in the mechanical properties of the starting
material over the same period. Thus a structural assessment of
a uPVC main can determine the current quality of pipe material
and that of the installation and from these express an opinion
as to whether its future performance is likely to be similar to
its past history, but it cannot be more quantitative than that.

WHAT THE RESULTS DO/DO NOT MEAN

The final outcome of a structural condition assessment is to

place the various parts of a system into one of three
structural condition categories. These are based on the time
remaining before some form of remedial work is required. Where
possible this is calculated in terms of remaining life of which
three types can be distinguished ie:

1 remaining asset life

2 remaining operational life

 319

3 time to next failure

Remaining asset life is the time left before corrosion has so

destroyed the fabric of the system that complete replacement is
required.

Remaining operational life is the time left before the

performance of the system becomes unsatisfactory in some way eg
excessive leakage from joints or unreliability of operation in
valves.

Time to next failure is what it says; this may arise from

brittle fracture in cast iron pipe or third party damage etc.

Remaining asset life is mainly determined by the general

condition of the system and it is this type of information that
the structural condition assessment supplies. The remaining
operational life is more difficult to assess. The process
whereby a run lead joint deteriorates to the point when it
leaks is not known, flexing of the joint by ground movements
may be the major factor. Without this understanding it is
difficult to know how to assess joints and the extent to which
it is possible to generalise about their condition from the
examination of a few examples.

The concept of operational life iq more useful for valves;

valves, being substantial castings, rarely fail because of
through wall penetration due to corrosion. Instead, by
affecting the load bearing surfaces, the corrosion makes the
valve operationally unreliable. As this process occurs in a
fairly consistent fashion, it is moderately predictable and so
remaining operational life is the concept most useful for
valves.

Time to next failure is a largely unpredictable parameter.

Pipes are subjected to longitudinal stresses arising from
320

differential ground movements which can produce brittle

fractures. WRc commissioned the University of Wales at Cardiff,
who have done extensive work in this area, to summarise the
current state of knowledge. The major factors involved are
well established. It is not clay soils as a class that
experience large volume changes with variation of moisture
content, they have to contain the fine particled mineral
montmorillinite. Also it is not every moisture increase that
produces significant soil heave eg an intrusion of groundwater
which contains dissolved salts has much less effect than the
equivalent amount of rainwater. After an average summer in the
UK, the soil moisture deficit of exposed soil does not even
extend to the normal pipe depths. It is the presence of deep
rooted trees that can create moisture deficits down to 2 - 3 m,
that will produce the annual variation in soil volume and hence
the stress on pipes.

However, although all these risk factors can be identified,

this is still a long way short of a predictive model of where
and when failure will occur. The unpredictability of the time
to next failure makes the operational reliability valves so
crucial if the the consequence of failures are to be minimised.

Calculating the Remaining Asset Life

To carry out a remaining asset life calculation requires

knowledge of the current rate of corrosion, the change in its
value over future years and the point in the deterioration
process when the pipe actually fails. In order to obtain
values for all three of these factors certain assumptions have
to be made.

The current corrosion rate is calculated using the pit depths

and the age of the pipe. However, it is a matter of choice
whether the sum of the deepest internal and external pits is
used or some other less extreme value, such as sum of the mean
or median of the deepest internal and external pits measured.
 321

Given the limited data points on the time axis, ie the year
when the pipe was laid (perhaps not even that!) and the year of
the condition assessment, it is almost unavoidable that a
linear rate of corrosion is assumed ie that the current rate of
corrosion will remain the same in the future. In studies where
samples have been exhumed over a series of years the corrosion
rate has been found to decrease with time in an exponential
manner; the value of the exponent has been found to vary
between 0.8 and 0.4 (an exponent of 1.0 corresponds to a linear
rate of course). Fortunately, when the remaining asset life is
short, the value calculated is not very sensitive to the rate
assumption used.

The other assumption implicit in using zero time as a point on

the graph, ie the date when the pipe was laid, is that
corrosion initiates as soon as the pipe is placed in the
ground. For many cases, where the protective coating was
damaged during the installation, this is indeed the case.
However, it need not always be so and a pipe may be protected
for may years even by a simple bitumen coating.

The other area where assumptions have to be made is the point

in the deterioration process at which the pipe 'fails'.
Through-wall penetration by the corrosion has the advantage of
being a readily defined point and for steel and ductile iron
pipe is a good definition of failure. For flake graphite
irons, ie vertical cast and spun-grey, a pipe can continue to
function satisfactorily, ie take full mains pressure, even with
very large areas of through wall corrosion. As a way of trying
to accommodate this difference in realistic end-points for
flake graphite irons, the mean or median of the deepest pits
could be used to calculate the corrosion rate. By the time the
'median pit' is breaking through the wall of the pipe other
areas will already have been penetrated. Consequently there
will be a more realistic chance of pipe failure than is the
case when the very first pit breaks through.
322
Of course, as flake graphite irons can also fail by brittle
fracture, due to differential stresses, even when little damage
from corrosion has occurred, the actual condition when a pipe
will fail is impossible to predict for these materials.

Guidance Manual for Structural Condition Assessment

At the request and with the financial support of various water

utilities, WRc have now produced a guidance manual on
structural condition assessment of trunk mains. This describes
the approach in more detail than has been possible in this
paper and illustrates its application with various actual case
histories.
 METHODS TO DETERMINE PARAMETERS CHARACfERIZING
TECHNICAL STATE OF PIPELINES WITH PRONOUNCED
DEPOSITION

PRIDON DAVLIANIDZE, LEVAN ALEKHIN,

TENGIZ TSAMALASHVILI, IOSEB ZEIKIDZE
Pilot Production "Spetstrubtrans", Stansia Veli,
380068, Tbilisi, Republic of Georgia

ABSTRACf

Methods to determine parameters, such as diameter D, Nik:uradse roughness size K of pipelines,

pipe friction coefficient A, characterizing the technical state of used pipelines with abundant
deposits are considered. On the basis of a study of the variation dynamics of these parameters,
it is shown how to determine the share of each of the parameters D and K, individually, in the
pipeline capacity loss. A simplified method to determine the coefficient A., based on the
measurement of the outflow time of a predetermined volume discharged from the pipeline under
study, for which the head loss is a priori defined, is considered. A specially developed device is
described, by means of which A can be found both for a steady and unsteady (pulsating) flow.
A possibility is shown for checking the precision of the parameters A., D and K determined by
the above mentioned device under laboratory conditions.
The importance of the proposed methods is discussed to establish optimal operating
conditions ofhydrotransport systems and to choose the appropriate time, kind and extent of repair
and maintenance measures to be taken for pipelines with pronounced deposition.

INTRODUCTION

The technical state of hydrotransport systems, deteriorates in use due to deposition on the walls
of the pipelines. This is usually accompanied by changes both in the inner diameter and the
quality of the inner surface of pipelines. As a result, the pipeline capacity reduces and, hence,
the hydrosystem operating conditions are violated [1-4].
To maintain or re-establish hydrosystem optimal operating conditions it is necessary: to
introduce additional capacities, to lay new pipelines instead of worn-out ones or to do the repair
and maintenance work. From the economical point of view the latter is preferable. Taking into
account the fact that there are various methods and deviCes for such kind of work nowadays, the
following problems become most urgent: an appropriate choice of the time, period and type of
work. Besides it is equally important to check the quality of the performed work.

323
324

For a successful solution of these problems it is necessary to study the variation dynamics
of the parameters D and K and to determine, on the basis, their individual share in the pipeline
capacity drop in the process of pipeline use. The main drawbacks of the existing techniques for
the determination of the above parameters are the following: necessity of direct measurement of
pipeline diameter; impossibility of application of these methods to all types of deposition
materials; the constructions are sophisticated and the measuring devices are expensive; hydrosys-
tems are forced to stand idle, etc.
Our methods for the determination of the pipeline technical state are only based on head loss
and flow rate measurements and avoid inconveniences associated with a direct measurement of
pipeline diameter. Besides, the advantage of our methods is that they are true for pipelines with
any kind of deposits.

METHODS TO DETERMINE THE PARAMETERS D, KANDA.

Consider the methods to determine the parameters characterizing the technical state of a
pipeline with abundant deposits in the general case when D=D(t); K=K(t); A.=A.(t).
~!.At the beginning a laminar flow regime is created in the pipeline and by measuring
the head loss per unit lenght i and the corresponding flow rate q the equivalent diameter of the.
pipeline is obtained from the following expression

(1)

Then a turbulent movement regime is created; the relative head loss per unit lenght I and the
corresponding flow rate Q are measured again and under a fixed value of D the remaining two
parameters are found from the following relation

a s
~<s-Dl
a G.*
t (2)

K -?>
- I
..,D(
'J" l
1 ~ 211!_ 0 625 T!D'V
IQJI).i )

where i is relative head loss per unit lenght with the flow rate q for laminar flow conditions; I is
the relative head loss per unit lenght for the flow rate Q under turbulent flow conditions; v is
the kinematic viscosity; g is the gravitational acceleration.
The fact that roughness under laminar flow conditions in no way affects the coefficient A.
has, certainly, been considered in the above method.
Method 2. A turbulent flow is created in the pipeline under conditions A.=f(K) and A.=f(Re,K),
( R~id's number). Under regime I for some flow rate ~the corresponding relative head
 325

loss per unit lenght I,is measured and regime II for two different flow rates ~and QJlead losses
I 1and I,_are measured. The pipe friction coefficients ("s,"A,,f.J are found for the above regimes by
means of expression (2). On the other hand, the same coefficients for the mentioned flow rates
(Q5,Q1,Q.J are obtained from the well-known formulae

0,2. 5
A=

(3)

Transforming system (3) we obtain the following equation:

(4)

which uniquely def"mes D and then the remaining ("A and K) unknown parameters.
Method 3. A turbulent flow is created in the pipeline under regimes N=f(K) and N=f(Re,K).
By ~~ement of the values of flow rates Q and the corresponding head losses per unit
lenght I the ratios Q./..ff' are found separately for both regimes. Then the flow rate satisfying
the following condition

(5)

is found and the desired parameters are obtained from the system (2) and the following expression
326

(6)

The proposed methods make it possible to determine, on the basis of the pipeline capacity
loss, the share of the reduced cross-section of the pipe and that due to the increased roughness
at any moment of pipeline use.
The relative resistance A of the pipeline at the beginning of its use (K=K.,D=D,,A.=A.,) is
defined in the following manner

(7)

and after deposition, say, at the time ts, when K>Ko,D<Do, 'J..>'J... o

(8)

it is implied that there are turbulent flow under the conditions A.=f(K).
The relative resistance for the same moment ts under the condition that the decline in the
technical state of the pipeline is only due to the reduction of the cross-section of the pipe, i.e.
K=K.,D<D,,i..=A.ois defmed by

(9)

Then the. share of the reduced capacity (Q0-Q) due to the reduced cross-section of the pipe
is

(10)

and that due to the increased roughness is

(11)
 327
The values Ooand Q are obtained from expressions (7)-(9) where A. and D parameters are defined
by formulae (1)-(6).

A SIMPLIFIED METHOD FOR THE DETERMINATION

OF THE COEFFICIENT A.

Steady flow. Consider a simplified method for the determination of the coefficient A., wich makes
it possible to verify the validity of the above methods under laboratory conditions. The method for
the determination of A. [5] based on the discharge time of a known liquid volume from a pipeline
section under consideration, where the head loss is predetermined, has been improved by us, and a
special laboratory device has been developed, the scheme of which is given in Fig. 1.
The device has a transparent vessel! to which pipeline 2 under test is attached with plug 3 at the
end. At the end of the pipe an electric balance 4 measuring vessel 5 of a predetermined capacity is
adjusted. The electric balance is connected to printer 6. On the upper wall of vessel·! transparent
drain-pipes 7 and 8 are located, the lower end sections of which are inside the vessel and the upper
ones are outside the vessel. Both drain-pipes can vertically move. On the upper parts of drain- pipes
7 and 8 there are gate valves 9 and 10, respectively, gate valve 10 being electromagnetic {;Onnected
to timer 11. On the upper wall of vessel! gate valve 12 is located and to the lower part of it (at the
zero level) transparent graded tube 13 is attached by gate valve 14.

Figure 1. Schematic diagram of the laboratory device. The starting time of work.

The technique is the following: at the beginning gate valves 9, 10, 12 and 14 are open and
pipe 2 is closed with plug 3. Vessel 1 if filled with liquid through the upper ends of drain- pipes
328
7 and 8 up to the level Hll> after which gate valves 10 and 12 are closed. The liquid in vessel 1,
in pipes 8 and 13 is at the same level H 0 (Fig. 1). Then plug 3 is removed and since the upper
sections of pipes 8 and 13 are open the liquid begins to flow out of pipeline 2. The levels of the
liquid in pipes 8 and 13 gradually go down till level ho, corresponding to the lower section of
pipe 8, is reached. During this time the level of the liquid in vessel! remains constant. This fact
can be explained by the following: in the upper part of vessel 1 during all this time vacuum is
created just enough to hold the part of the liquid column in vessel 1 between levels hoand lfo.
When pipe 8 is drained, air in the form of bubbles enters vessel 1 and then the level of the
liquid in vessel 1 gradually goes down. Till the level of the liquid in vessel 1 reaches the level
of the lower section of drain-pipe 8, the liquid in the graded pipe 13 remains at the same level
(Fig. 2). This fact indicates that during timeT, while the level of the liquid in vessell goes down
from H o to h 0 the flows out of the pipeline at a constant velocity and with a constant head.

Figure 2. Schematic diagram of the laboratory device. The intermediate time of work.

As it has been noted, pipe 8 can vertically move wich enables one to adjust the lower section
of the drain-pipe at different levels. This makes it possible to obtain different flow-rates (in a
wide range of Reynolds numbers) in the pipeline under test.
The coefficient :\. can be found by the following formula:

(12)
 329

where h 0 is the head shown by graded tube 13; Q.=W../tois the flow rate through pipeline 2; Wo
is the capacity of vessel 5; t 0 is the filling time of vessel 5; L is the lenght of the pipeline under
test.
Pulsating flow. By means of the device (Fig. 1) a pulsating flow can be created in the pipeline
under test. --
At the starting moment the lower sections 7 and 8 of the pipeline are set at different levels,
e.g. the lower section of 8 is adjusted at h,, that of 7- at h1 (~>ha). The device is prepared and
starts operating similar to the above one. Since moment t0 , when the liquid in pipes 8, 13, is at
the level h., by means of timer 7 gate valve 10 is periodically opened or closed. When gate valve
10 is closed the air enters vessel 1 only through drain-pipe 8, the liquid in pipe 13 is at the level
ho. and when gate valve 10 is open and the liquid flows in only through pipe 7, the liquid in pipe
13 is at the level ~. This indicates that liquid flows out of the pipeline under test with a different
head, which i:. confmned by liquid level variation from hJ:o htin pipe 13, i.e. the liquid in pipeline
2 has a pulsating flow whit a predetermined amplitude and frequency. It is expedient to keep a
pulsating flow till the moment t*, which corresponds to a level drop in vessel! from Hoto l;(hz.> 11,).
At the moment yt is necessary to close gate valves 9 and 10 simultaneously.
It is possible, for instance, to find the coefficients \average for the pulsation period To in
the following maner:

- hp
I . -T
(13)
(\\lf..,. -- WP
-~;,
:.

RESULTS

Methods to determine such parameters as the diameter, Nikuradse roughness size of the
pipline and the pipe friction coefficient, characterizing the technical state of a pipeline with
abundant deposits are developed. Relations are found to determine the values of the desired
parameters. The considered methods allow the determination of the avarage radial thickness of
roughnesses A and the volume W of the deposits in the pipeline

The application of the developed methods is possible for pipelines with any kind of deposits.
Besides they exclude inconveniences associated with direct measurement of the pipeline
diameter in the process of its use. The proposed methods can be also applied to inaccessible
330

sections of pipelines. A spesial device is developed to determine the above parameters under
laboratory conditions.

DISCUSSION

The authors believe it is expedient to stady "quasistationary" condition of the pulsating flow
and pulsation effect on the capacity of pipelines with deposits.
A further fundamental study ofthe above problems is desirable, in particular, by fundamental
experimental investigations.

CONCLUSIONS

The obtained results are of a practical importance, when it is necessary to determine the
actual parameters of pipelines with a detonated technical state due to deposition. The methods
are true for pipelines with any kind of deposits. This makes their application natural for
circulating hydrotransport systems of various materials; hot water systems; oil pipelines with
paraffm deposits, etc.
The application of the proposed methods provides great opportunities for an appropriate
choice of the starting time, duration, periodicity, type of repair and maintenance work to be done
and also helps to check the quality of the performed work.

REFERENCES

1. Chekanov G.S., Zorin V.A., Deposition and deposit removal in hydraulic ash disposal
systems. Energoizdat, Moscow, 1987, pp. 176.

2. Chistyakov N.N., How to improve the effeciency of hot water system operation, Stroiizdat,
Moscow, 1988, pp. 314.

3. Wilfer H.P., Theoretical studies on the motional behavior of fluids in piping with
incrustations and deposits,~ Int., 1990, 29, No 7-8, pp. 389-409.

4. Smoldyrev A.B., On the reduction of pipe clogging in hydrotransport, .J:. Non-Ferrous

~ 1986, No 5, pp. 107-109.

5. Author's Certificate of the USSR No 573482, Int Cl., F17D 1/10, 1973.
 Mechanics
FLUID MECHANICS AND ITS APPLICATIONS
Series Editor: R. Moreau
Aims and Scope of the Series
The purpose of this series is to focus on subjects in which fluid mechanics plays a fundamental
role. As well as the more traditional applications of aeronautics, hydraulics, heat and mass transfer
etc., books will be published dealing with topics which are currently in a state of rapid develop-
ment, such as turbulence, suspensions and multiphase fluids, super and hypersonic flows and
numerical modelling techniques. It is a widely held view that it is the interdisciplinary subjects that
will receive intense scientific attention, bringing them to the forefront of technoldgical advance-
ment. Fluids have the ability to transport matter and its properties as well as transmit force,
therefore fluid mechanics is a subject that is particularly open to cross fertilisation with other
sciences and disciplines of engineering. The subject of fluid mechanics will be highly relevant in
domains such as chemical, metallurgical, biological and ecological engineering. This series is
particularly open to such new multidisciplinary domains.
1. M. Lesieur: Turbulence in Fluids. 2nd rev. ed., 1990 ISBN 0-7923-0645-7
2. 0. Metais and M. Lesieur (eds.): Turbulence and Coherent Structures. 1991
ISBN 0-7923-0646-5
3. R. Moreau: Magnetohydrodynamics. 1990 ISBN 0-7923-0937-5
4. E. Coustols (ed.): Turbulence Control by Passive Means. 1990 ISBN 0-7923-1020-9
5. A. A. Borissov (ed.): Dynamic Structure of Detonation in Gaseous and Dispersed Media.
1991 ISBN 0-7923-1340-2
6. K.-S. Choi (ed.): Recent Developments in Turbulence Management. 1991
ISBN 0-7923-1477-8
7. BHR Group (ed.): Pipeline Systems. 1992 ISBN 0-7923-1668-1
8. BHR Group (ed.): Fluid Sealing. 1992 ISBN 0-7923-1669-X
9. T.K.S. Murthy (ed.): Computational Methods in Hypersonic Aerodynamics. 1992
ISBN 0-7923-1673-8

Kluwer Academic Publishers - Dordrecht I Boston I London

 Mechanics
SOLID MECHANICS AND ITS APPLICATIONS
Series Editor: G.M.L. Gladwell
Aims and Scope of the Series
The fundamental questions arising in mechanics are: Why?, How?, and How much? The aim of this
series is to provide lucid accounts written by authoritative researchers giving vision and insight in
answering these questions on the subject of mechanics as it relates to solids. The scope of the series
covers the entire spectrum of solid mechanics. Thus it includes the foundation of mechanics;
variational formulations; computational mechanics; statics, kinematics and dynamics of rigid and
elastic bodies; vibrations of solids and structures; dynamical systems and chaos; the theories of
elasticity, plasticity and viscoelasticity; composite materials; rods, beams, shells and membranes;
structural control and stability; soils, rocks and geomechanics; fracture; tribology; experimental
mechanics; biomechanics and machine design.
I. R.T. Haftka, Z. GUrdal and M.P. Kamat: Elements of Structural Optimization. 2nd rev.ed.,
1990 ISBN 0-7923-0608-2
2. J.J. Kalker: Three-Dimensional Elastic Bodies in Rolling Contact. 1990
ISBN 0-7923-0712-7
3. P. Karasudhi: Foundations of Solid Mechanics. 1991 ISBN 0-7923-0772-0
4. N. Kikuchi: Computational Methods in Contact Mechanics. (forthcoming)
ISBN 0-7923-0773-9
5. Y.K. Cheung and A.Y.T. Leung: Finite Element Methods in Dynamics. (forthcoming)
ISBN 0-7923-1313-5
6. J.P. Doyle: Static and Dynamic Analysis of Structures. With an Emphasis on Mechanics and
Computer Matrix Methods. 1991 ISBN 0-7923-1124-8; Pb 0-7923-1208-2
7. 0.0. Ochoa and J.N. Reddy: Finite Element Modelling of Composite Structures.
(forthcoming) ISBN 0-7923-1125-6
8. M.H. Aliabadi and D.P. Rooke: Numerical Fracture Mechanics. ISBN 0-7923-1175-2
9. J. Angeles and C.S. L6pez-Cajun: Optimization of Cam Mechanisms. 1991
ISBN 0-7923-1355-0
10. D.E. Grierson, A. Franchi and P. Riva: Progress in Structural Engineering. 1991
ISBN 0-7923-1396-8
11. R.T. Haftka and Z. GUrdal: Elements of Structural Optimization. 3rd rev. and exp. ed. 1992
ISBN 0-7923-1504-9; Pb 0-7923-1505-7
12. J.R. Barber: Elasticity. 1992 ISBN 0-7923-1609-6

Kluwer Academic Publishers - Dordrecht I Boston I London

 Mechanics
From 1990, books on the subject of mechanics will be published under two series:
FLUID MECHANICS AND ITS APPLICATIONS
Series Editor: R.J. Moreau
SOLID MECHANICS AND ITS APPLICATIONS
Series Editor: G.M.L. Gladwell
Prior to 1990, the books listed below were published in the respective series indicated below.

MECHANICS: DYNAMICAL SYSTEMS

Editors: L. Meirovitch and G.JE. Oravas

1. E.H. Dowell: Aeroelasticity of Plates and Shells. 1915 ISBN 90-286-0404-9

2. D.G.B. Edelen: Lagrangian Mechanics of Nonconservative Nonholonomic Systems.
1977 ISBN 90-286-0077-9
3. J.L. Junkins: An Introduction to Optimal Estimation of Dynamical Systems. 1978
ISBN 90-286-0067-1
4. E.H. Dowell (ed.), H.C. Curtiss Jr., R.H. Scanlan and F. Sisto: A Modern Cowse in
Aeroelasticity. Revised and enlarged edition see under Volume 11
5. L. Meirovitch: Computational Methods in Structural Dynamics. 1980
ISBN 90-286-0580-0
6. B. Skalmierski and A. Tylikowski: Stochastic Processes in Dynamics. Revised and
enlarged translation. 1982 ISBN 90-247-2686-7
7. P.C. MUller and W.O. Schiehlen: Linear Vibrations. A Theoretical Treatment of Multi-
degree-of-freedom Vibrating Systems. I985 ISBN 90-247-2983-1
8. Gh. Buzdugan, E. MiMilescu and M. Rade§: Vibration Measurement. 1986
ISBN 90-247-3111-9
9. G.M.L. Gladwell: Inverse Problems in Vibration. 1987 ISBN 90-247-3408-8
10. G.I. Schut\ller and M. Shinozuka: Stochastic Methods in Structural Dynamics. 1987
ISBN 90-247-36I1-0
II. E.H. Dowell (ed.), H.C. Curtiss Jr., R.H. Scanlan and F. Sisto: A Modern Course in
Aeroelasticity. Second revised and enlarged edition (of Volume 4). I989
ISBN Hb 0-7923-0062-9; Pb 0-7923-0I85-4
I2. W. Szempliriska-Stupnicka: The Behavior of Nonlinear Vibrating Systems. Volume I:
Fundamental Concepts and Methods: Applications to Single-Degree-of-Freedom
Systems. I990 ISBN 0-7923-0368-7
13. W. Szempliriska-Stupnicka: The Behavior of Nonlinear Vibrating Systems. Volume II:
Advanced Concepts and Applications to Multi-Degree-of-Freedom Systems. I990
ISBN 0-7923-0369-5
Set ISBN (Vols. I2-13) 0-7923-0370-9

MECHANICS OF STRUCTURAL SYSTEMS

Editors: J.S. Przemieniecki and G.JE. Oravas

I. L. Frjba: Vibration of Solids and Structures under Moving Loads. I970

ISBN 90-01-32420-2
2. K. Marguerre and K. Wolfe!: Mechanics of Vibration. 1979 ISBN 90-286-0086-8
 Mechanics
3. E.B. Magrab: Vibrations of Elastic Structural Members. 1979 ISBN 90-286-0207-0
4. R.T. Haftka and M.P. Kamat: Elements of Structural Optimization. 1985
Revised and enlarged edition see under Solid Mechanics and Its Applications, Volume 1
5. J.R. Vinson and R.L. Sierakowski: The Behavior of Structures Composed of Composite
Materials. 1986 ISBN Hb 90-247-3125-9; Pb 90-247-3578-5
6. B.E. Gatewood: Virtual Principles in Aircraft Structures. Volume 1: Analysis. 1989
ISBN 90-247-3754-0
7. B.E. Gatewood: Virtual Principles in Aircraft Structures. Volume 2: Design, Plates,
Finite Elements. 1989 ISBN 90-247-3755-9
Set (Gatewood 1 + 2) ISBN 90-247-3753-2

MECHANICS OF ELASTIC AND INELASTIC SOLIDS

Editors: S. Nemat-Nasser and G..tE. Oravas

1. G.M.L. Gladwell: Contact Problems in' the Classical Theory of Elasticity. 1980
ISBN Hb 90-286-0440-5; Pb 90-286-0760-9
2. G. Wempner: Mechanics of Solids with Applications to Thin Bodies. 1981
ISBN 90-286-0880-X
3. T. Mura: Micromechanics of Defects in Solids. 2nd revised edition, 1987
ISBN 90-247-3343-X
4. R.G. Payton: Elastic Wave Propagation in Transversely Isotropic Media. 1983
ISBN 90-247-2843-6
5. S. Nemat-Nasser, H. Abe and S. Hirakawa (eds.): Hydraulic Fracturing and Geother-
mal Energy. 1983 ISBN 90-247-2855-X
6. S. Nemat-Nasser, R.J. Asaro and G.A. Hegemier (eds.): Theoretical Foundation for
Large-scale Computations ofNonlinear Material Behavior. 1984 ISBN 90-247-3092-9
7. N. Cristescu: Rock Rheology. 1988 ISBN 90-247-3660-9
8. G.I.N. Rozvany: Structural Design via Optimality Criteria. The Prager Approach to
Structural Optimization. 1989 ISBN 90-247-3613-7

MECHANICS OF SURFACE STRUCTURES

Editors: W.A. Nash and G.JE. Oravas

1. P. Seide: Small Elastic Deformations of Thin Shells. 1975 ISBN 90-286-0064-7

2. V. Pane: Theories of Elastic Plates. 1975 ISBN 90-286-0104-X
3. J.L. Nowinski: Theory ofThermoelasticity with Applications. 1978
ISBN 90-286-0457-X
4. S. Lukasiewicz: Local Loads in Plates and Shells. 1979 ISBN 90-286-0047-7
5. C. Fii't: Statics, Formfinding and Dynamics of Air-supported Membrane Structures.
1983 ISBN 90-247-2672-7
6. Y. Kai-yuan (ed.): Progress in Applied Mechanics. The Chien Wei-zang Anniversary
Volume. 1987 ISBN 90-247-3249-2
7. R. Negrutiu: Elastic Analysis of Slab Structures. 1987 ISBN 90-247-3367-7
8. J.R. Vinson: The Behavior of Thin Walled Structures. Beams, Plates, and Shells. 1988
ISBN Hb 90-247-3663-3; Pb 90-247-3664-1
 Mechanics
MECHANICS OF FLUIDS AND TRANSPORT PROCESSES
Editors: R.J. Moreau and G.JE. Oravas

1. J. Happel and H. Brenner: Low Reynolds Number Hydrodynamics. With Special

Applications to Particular Media. 1983 ISBN Hb 90-01-37115-9; Pb 90-247-2877-0
2. S. Zahorski: Mechanics of Viscoelastic Fluids. 1982 ISBN 90-247-2687-5
3. J.A. Sparenberg: Elements of Hydrodynamics Propulsion. 1984 ISBN 90-247-2871-1
4. B.K. Shivamoggi: Theoretical Fluid Dynamics. 1984 ISBN 90-247-2999-8
5. R. Timman, A.J. Hermans and G.C. Hsiao: Water Waves and Ship Hydrodynamics. An
Introduction. 1985 ISBN 90-247-3218-2
6. M. Lesieur: Turbulence in Fluids. Stochastic and Numerical Modelling. 1987
ISBN 90-247-3470-3
7. L.A. Lliboutry: Very Slow Flows of Solids. Basics of Modeling in Geodynamics and
Glaciology. 1987 ISBN 90-247-3482-7
8. B.K. Shivamoggi: Introduction to Nonlinear Fluid-Plasma Waves. 1988
ISBN 90-247-3662-5
9. V. Bojarevics, Ya. Freibergs, E.I. Shilova and E.V. Shcherbinin: Electrically Induced
Vortical Flows. 1989 ISBN 90-247-3712-5
10. J. Lielpeteris and R. Moreau (eds.): Liquid Metal Magnetohydrodynamics. 1989
ISBN 0-7923-0344-X

MECHANICS OF ELASTIC STABILITY

Editors: H. Leipholz and G.JE. Oravas

1. H. Leipholz: Theory of Elasticity. 1974 ISBN 90-286-0193-7

2. L. Librescu: Elastostatics and Kinetics of Aniosotropic and Heterogeneous Shell-type
Structures. 1975 ISBN 90-286-0035-3
3. C.L. Dym: Stability Theory and Its Applications to Structural Mechanics. 1974
ISBN 90-286-0094-9
4. K. Huseyin: Nonlinear Theory of Elastic Stability. 1975 ISBN 90-286-0344-1
5. H. Leipho1z: Direct Variational Methods and Eigenvalue Problems in Engineering.
1977 ISBN 90-286-0106-6
6. K. Huseyin: Vibrations and Stability of Multiple Parameter Systems. 1978
ISBN 90-286-0136-8
7. H. Leipholz: Stability of Elastic Systems. 1980 ISBN 90-286-0050-7
8. V.V. Bolotin: Random Vibrations of Elastic Systems. 1984 ISBN 90-247-2981-5
9. D. Bushnell: Computerized Buckling Analysis of Shells. 1985 ISBN 90-247-3099-6
10. L.M. Kachanov: Introduction to Continuum Damage Mechanics. 1986
ISBN 90-247-3319-7
11. H.H.E. Leipholz and M. Abdel-Rohman: Control of Structures. 1986
ISBN 90-247-3321-9
12. H.E. Lindberg and A.L. Florence: Dynamic Pulse Buckling. Theory and Experiment.
1987 ISBN 90-247-3566-1
13. A. Gajewski and M. Zyczkowski: Optimal Structural Design under Stability Con-
straints. 1988 ISBN 90-247-3612-9
 Mechanics
MECHANICS: ANALYSIS
Editors: V.J. Mizel and G.JE. Oravas

I. M.A. Krasnoselskii, P.P. Zabreiko, E.l. Pustylnik and P.E. Sbolevskii: Integral
Operators in Spaces ofSummable Functions. 1976 ISBN 90-286-0294-1
2. V.V. Ivanov: The Theory of Approximate Methods and Their Application to the
Numerical Solution of Singular Integral Equations. 1976 ISBN 90-286-0036-1
3. A. Kufner, 0. John and S. Pucik: Function Spaces. 1977 ISBN 90-286-0015-9
4. S.G. Mikhlin: Approximation on a Rectangular Grid. With Application to Finite
Element Methods and Other Problems. 1979 ISBN 90-286-0008-6
5. D.G.B. Edelen: Isovector Methods for Equations of Balance. With Programs for
Computer Assistance in Operator Calculations and an Exposition of Practical Topics of
the Exterior Calculus. 1980 ISBN 90-286-0420-0
6. R.S. Anderssen, P.R. de Hoog and M.A. Lukas (eds.): The Application and Numerical
Solution of Integral Equations. 1980 ISBN 90-286-0450-2
7. R.Z. Has 'minskil: Stochastic Stability of Differential Equations. 1980
ISBN 90-286-0100-7
8. A.I. Vo1'pert and S.I. Hudjaev: Analysis in Classes of Discontinuous Functions and
Equations of Mathematical Physics. 1985 ISBN 90-247-3109-7
9. A. Georgescu: Hydrodynamic Stability Theory. 1985 ISBN 90-247-3120-8
10. W. Noll: Finite-dimensional Spaces. Algebra, Geometry and Analysis. Volume I. 1987
ISBN Hb 90-247-3581-5; Pb 90-247-3582-3

MECHANICS: COMPUTATIONAL MECHANICS

Editors: M. Stem and G.JE. Oravas

I. T.A. Cruse: Boundary Element Analysis in Computational Fracture Mechanics. 1988

ISBN 90-247-3614-5

MECHANICS: GENESIS AND METHOD

Editor: G.JE. Oravas

1. P.-M.-M. Duhem: The Evolution of Mechanics. 1980 ISBN 90-286-0688-2

MECHANICS OF CONTINUA
Editors: W.O. Williams and G.JE. Oravas

1. C.-C. Wang and C. Truesdell: Introduction to Rational Elasticity. 1973

ISBN 90-01-93710-1
2. P.J. Chen: Selected Topics in Wave Propagation. 1976 ISBN 90-286-0515-0
3. P. Villaggio: Qualitative Methods in Elasticity. 1977 ISBN 90-286-0007-8
 Mechanics
MECHANICS OF FRACTURE
Editors: G.C. Sih

1. G.C. Sih (ed.): Methods of Analysis and Solutions of Crack Problems. 1973
ISBN 90-01-79860-8
2. M.K. Kassir and G.C. Sih (eds.): Three-dimensional Crack Problems. A New Solution
of Crack Solutions in Three-dimensional Elasticity. 1975 ISBN 90-286-0414-6
3. G.C. Sih (ed.): Plates and Shells with Cracks. 1977 ISBN 90-286-0146-5
4. G.C. Sih (ed.): Elastodynamic Crack Problems. 1977 ISBN 90-286-0156-2
5. G.C. Sih (ed.): Stress Analysis of Notch Problems. Stress Solutions to a Variety of
Notch Geometries used in Engineering Design. 1978 ISBN 90-286-0166-X
6. G.C. Sih and E.P. Chen (eds.): Cracks in Composite Materials. A Compilation of Stress
Solutions for Composite System with Cracks. 1981 ISBN 90-247-2559-3
7. G.C. Sih (ed.): Experimental Evaluation of Stress Concentration and Intensity Factors.
Useful Methods and Solutions to Experimentalists in Fracture Mechanics. 1981
ISBN 90-247-2558-5

MECHANICS OF PLASTIC SOLIDS

Editors: J. Schroeder and G.JE. Oravas

1. A. Sawczuk (ed.): Foundations of Plasticity. 1973 ISBN 90-01-77570-5

2. A. Sawczuk (ed.): Problems of Plasticity. 1974 ISBN 90-286-0233-X
3. W. Szczepiflski: Introduction to the Mechanics of Plastic Forming of Metals. 1979
ISBN 90-286-0126-0
4. D.A. Gokhfeld and O.F. Chemiavsky: Limit Analysis of Structures at Thermal Cycling.
1980 ISBN 90-286-0455-3
5. N. Cristescu and I. Suliciu: Viscoplasticity. 1982 ISBN 90-247-2777-4

Kluwer Academic Publishers - Dordrecht I Boston I London