Forensic transient analyses of two pipeline failures

Forensic transient analyses of two pipeline failures

MARKO V. IVETIC*

Faculty of Civil Engineering, University of Belgrade, Bul. Revolucije 73 Belgrade, Yugoslavia

(Revised version received June 2004)

Details pertaining to two groups of preventable accidents, both caused by transient

events, are presented. The accidents could have been avoided if there had been better

coordination during the design process, during system construction and during

operation. Moreover, a number of common beliefs, inherited from the period when

computer simulations were rare, as well as ill-informed attempts to mimic ‘‘good

solutions’’, influenced design decisions and contributed to the performance difficulties.

The two examples illustrate transient analysis in complex systems and raise crucial issues

about the lines of responsibility; although such problems also exist in small systems, the

damage is typically proportionally smaller, and thus small systems have tended to receive

less attention. The first case presented involves the Abattemarco pipeline in Italy. In this

case, an unauthorized decision to marginally decrease capital costs by reducing the pipe’s

pressure class significantly compromised system reliability. In the second case, involving a

desalination plant, it was erroneously believed that such a low-pressure system would not

be susceptible to transients and that, in any case, a higher pressure class would invariably

cover all contingencies. The operation of this second system was further complicated by a

decision to oversize the pumps, a choice that increased both capital and running costs and

simultaneously diminished the reliability of the already vulnerable system.

Keywords: Hydraulic transients; Pipeline failure; Risk assessment

1. Introduction

In spite of theoretical and practical advances over many

years (e.g. Parmakian 1963, Wylie and Streeter 1978,

Thorley 1991), and even with continuing research into

numerical transient analysis, controlling transient events in

pipelines and water distribution networks is not easy

(Karney and McInnis 1990). In fact, it is difficult to assess

the reliability of transient control measures, since too few

detailed case studies have been presented. A detailed

analysis depends on the level of information available

regarding the performance of real systems. Such informa-

tion should include details of system layout, the conditions

of pipes and equipment, operational and control ap-

proaches, the state of system maintenance, and many

related factors.

Mathematical and numerical models based on the

method of characteristics have proven to be effective (Ivetic

1996). They have been successfully applied in highly

complex systems, even in cases involving transient cavita-

tion (Provoost 1976, Streeter 1983, Ivetic 2003). Moreover,

significant progress has been made in the use of transient

analysis to examine the conditions of distribution networks

and other complex systems. It is relatively easy to perform

simulations that correspond to measured values in an

existing system. On the other hand, there is much greater

uncertainty in defining control devices in systems that have

not yet been constructed. Certainly the equipment can be

specified, but it may change before being purchased and

commissioned, and may not operate as expected.

The main sources of uncertainty that remain unresolved

are:

*Corresponding author. Email: [email protected]

Urban Water Journal, Vol. 1, No. 2, June 2004, 85 – 95

Urban Water JournalISSN 1573-062X print/ISSN 1744-9006 online # 2004 Taylor & Francis Ltd

http://www.tandf.co.uk/journalsDOI: 10.1080/157306204112331289986

. For steady state conditions, friction coefficients can be

estimated within a margin of perhaps+ 10%. In the

majority of analyses the same expressions are used for

unsteady flow, giving a much wider margin of

uncertainty.. Pump characteristics are only known after delivery and

installation (and the actual pump characteristics in all

four quadrants will probably remain unknown).. Many small changes to the system take place in the later

phases of design, after the primary hydraulic analysis

has been completed. Even if these changes are small,

their cumulative effect can be significant. For example,

the pressure class of valves could change, and indeed it is

not uncommon to modify the size or configuration of a

valve.. Different contractors often complete different phases of

design and construction, and some may not treat the

whole system but only the part within their own ‘‘Terms

of Reference’’; this inevitably leads to variations in

system details.. The amount of free and dissolved air depends on local/

seasonal conditions and the design of the water intake.

The influence of free air on wave velocity is significant

(though, fortunately, usually conservative in terms of

peak pressures).. Although, some of the uncertainties may significantly

reduce the risk of transients and conservative safety

factors are used in the structural design of pipes,

accidents do happen (Almeida 1992, Tulis and Watkins

1992, Locher et al. 2000). The experience gathered from

such accidents is valuable, yet rarely shared.

The worrisome characteristic of the present design practice

is insufficient participation by transient analysis specialists

throughout design and construction. In a process that is

inherently iterative, where several interactive parameters of

the system have to be adjusted, all changes that may affect

system operation and reliability should be authorised by

the specialist. This observation has led more recently to the

idea of providing system operators with tools to assess the

risk of hydraulic transients, which can be used on a daily

basis and to share responsibility for system safety (Far et al.

2004).

In this paper, hydraulic analysis results from two groups

of accidents are summarized. The chief goal of both

analyses, performed by the author, was a search for

evidence to assign fault: to determine that something (or

someone) did not perform as expected. In a situation where

information might be unavailable for a variety of reasons,

the investigation assumed a forensic character.

2. Abattemarco pipeline, Italy

2.1 Accidents

The Abattemarco pipeline is a part of the water supply

system of Cosenza in Italy. The distance between the intake

and the main reservoir in Cosenza is 56 km, with diameters

ranging from 800 mm (upstream) to 550 mm (down-

stream), and a total head difference of 250 m (figure 1).

The flow capacity of the pipeline is approximately 500 l/s.

A pump with a capacity 100 l/s is connected to the main

pipeline (DN 700) via pipeline DN 400, a further 5 km

downstream (figure 2). The ground profile is highly

irregular with the maximum elevation being 615 m at the

water intake and a minimum along the line of 104 m. Over

the line’s first 16 km, there is an additional parallel segment

that locally reduces both velocities and head losses (figure

2). After more than 20 years of reliable operation, the

pipeline experienced two accidents in a short interval.

Figure 2 gives a perspective view and layout of the pipeline,

with an indication of the location of both the pump and the

Figure 1. The Abattemarco pipeline.

86 M. V. Ivetic

damaged valves; figures 3 and 4 show the damage to the

two valves themselves.

Unfortunately, no automatic recording devices or written

traces were available to determine in detail what actually

happened. Also, the available data relating to operating

conditions, valve characteristics, etc., were scarce and often

confusing. For example, according to the original design,

valves that were later damaged were specified for pressures

up to 40 bar (pressure rating, or nominal pressure, ISO PN

40), but during the site visit, it was discovered that the

valves were only PN 25. Both accidents occurred during

system restart after the system had been out of operation.

The first accident occurred during system restart after a

regular pipeline inspection, and the second accident also

happened during system restart but after the damaged valve

had been replaced. Operators of the systems claimed that

they had followed regular procedure and that filling

velocities were below the widely accepted safe practice

limit of 0.3 m/s.

The upstream part of the system, between the intake

(elevation 615 m) and the highest point in the longitudinal

profile (elevation 562 m, located 16 km downstream) is

filled first. When the total head reaches 562 m, the pump is

switched on and the filling of the downstream part of the

system is begun. Both accidents occurred after pump start-

up, and the first guess as to what went wrong was that the

pump was likely responsible for the accidents (a quite

common conclusion; see Ivetic 1996). Although not

required for the original design, construction of an air

chamber at the pump station was requested as a remedial

action to avoid repeated failures. Yet, the tentative

conclusion of the role of the pumps was not supported by

the fact that during pipeline filling no significant increase of

pressure in the pump house was noticed either before or

after pump start-up. In fact, the accidents were discovered

when the pressures unexpectedly and significantly dropped

after pump start-up, instead of tracking the expected

pressure rise.

Figure 3. Damaged valve No. 1.

Figure 2. Elevations of the upstream part of pipeline.

Figure 4. Damaged valve No. 2.

Forensic transient analysis of two pipeline failures 87

Both valves are located on pipe drains with diameters

400 mm (DN 400), close to the upstream end of the pipeline

(approx. 2 km), with around 200 m between them.

Operating pressures for both valves are equal to, or

perhaps just slightly below, 25 bar (their nominal pressure

ratings). Thus, the most likely source of the valve failure

was in the upstream part of the system, but the search for

explanations had to consider other possibilities as well.

2.2 Results of the analysis

The analysis started with the pump start-up simulation

under conditions explained by the system operator.

Simulation results for the upstream part of the pipeline,

as a snapshot of instantaneous wave front positions, are

shown in figure 5. A relatively small overpressure wave (of

the order of 2 bar with an air chamber, and 3 bar, without),

travelling upstream and downstream from the pump

connection can be observed. When added to the static

pressure of about 20 bar, the pump start-up pressures do

not exceed the design pressure of valves; in fact, these

values are less than working pressures for both valves.

Trapped air in the system would cause lower or the same

overpressures. The influence of filling velocity was exam-

ined by simulated opening values for the upstream control

valve. If the initial flow rate is 0.3 m3/s, instead of 0.17 m3/s

(which corresponds to 0.3 m/s), there will be a pressure

increase of about 2 bar, thus still causing a pressure less

than design values. Uncontrolled, the escape of air was not

simulated because there was no indication that something

like this had happened in the line (in particular, there was

no damage to any of the air valves).

Close inspection of the fracture surface of one valve (see

figure 4) showed a significant material imperfection (15 mm

by 25 mm) in its casting near the valve wedge seat.

Obviously, fracture would occur first at the weakest point.

The other valve was broken very near its flange, but the

failure surface here was uniform without obvious imperfec-

tions. However, the position of the fracture indicates that

the valve’s weight (of around 5000 N) was poorly

supported (figure 6, right), likely playing a role in the

failure. This structural load would cause an increase in

local stresses of 15 – 18 % over those caused by the internal

water pressure. Although not appearing to be a large

increase in itself, this additional load could well become

critical and lead to failure in the long run.

2.3 Responsibility

Technically, the original design, with a required pressure

class PN 40, was a correct and sensible choice. Working

pressures in steady state conditions were themselves close to

25 bar, leading to a nominal pressure class of PN 25;

however, taking into account pressure oscillations due to

transients, a pressure class selection of PN 40 is logical.

Although this is a predominately gravity pipeline, the

existence of its one pumping station led to the preliminary

assumption that the pumps were in some way responsible

for the failure of the valves. Also, this pumping station was

not protected by an air chamber; this is a very common

protection strategy for protecting large systems. Thus, there

was an immediate decision to install air chamber protec-

tion. Although this did not make operating conditions

worse, in this case it did not improve them either. After the

exploration and many combinations of possible causative

factors, it was concluded that pressure oscillations caused

by pump start-up will not significantly exceed the operating

pressures for valves. Thus, an air chamber at the pump

station was not required in this case.

It was originally believed that all valves were for PN 40,

but someone at some point decided that these two valves,

with working pressures around 25 bar, could have been

satisfactory at PN 25. One had an imperfection in its

casting (commissioning error) and the other had inade-

Figure 5. Captured frame at the moment of lateral pump start-up.

88 M. V. Ivetic

quate support (construction error). However, it is not clear

why these flaws were not triggered during field pressure

tests on the system, nor exactly why they failed when they

did. The poor decision to change the pressure class of likely

valves, probably motivated by saving a small amount of

money, was revealed when two hidden flaws were activated

in a short time span. The first accident (the sudden failure

at the valve) itself would cause significant pressure

oscillations that could well have worsened conditions at

the second hidden flaw, and likely contributed to the

second valve failure. Failures are expensive and trouble-

some, but rarely simple; attributing blame is typically

difficult and problematic, but often instructive.

3. Desalination plant

A huge pipeline system, within a single desalination plant,

was designed to collect and convey water from four

production blocks (C2, C3, C4 and C5) to seven reservoirs

(PWT 1 to PWT 7). The system with 40 pumps operating at

the same time (figure 7) had experienced operational

problems. The ground profile is flat and the piping is

mainly GRP (glass reinforced plastic). During hydraulic

transients caused by a cut in power to specific groups of

pumps, damage to several sections occurred. Although the

damage itself was relatively small, there was considerable

surprise that it took place at all. A sequence of these

accidents caused lengthy shut-downs of the system, causing

great complications and concern in a water system of this

size and importance. Not surprisingly, it was concluded

that system operational reliability was unacceptably low

and had to be improved.

The design plant capacity is 40 pumps at 330 l/s (for a

total flow rate of about 12 m3/s). The 40 pumps are

arranged in four production blocks (PB): C2, C3, C4 and

C5, with 10 pumps in each. Pipe diameters of the header

pipeline start from 350 mm (DN 350), for the most

upstream part of the PB C2, and increase gradually until

DN 1500 is reached just before the junction with the header

pipelines from PB C4 and C5. On the right-hand side of the

plant, diameters are larger, beginning from DN 350 and

increasing to a maximum of DN 1800. The common header

line from all PBs is DN 2100 with a total length of 2420 m

(i.e. 600 + 920 + 900 m); the pipes leading to production

water tanks (PWT1 to PWT7) are DN 1100. The water

level in PWT varies over about a 10 m range. The notation

of the production blocks suggests that there were plans to

expand the system. A separate power supply has been

provided to each block, so a simultaneous failure of all the

pumps is highly unlikely.

Although velocities in the system are high (over 3 m/s in

the pipe DN 2100), it was believed that in systems with low

operating pressures there was no danger from pressure

transients. This statement was supported by a transient

analysis performed on a simplified model, one that used

four equivalent pumps to represent the original 40 pumps.

Design operating pressures were only between 2 and 3 bars,

and the pressure class of the selected pipe and equipment

was much greater, being PN 09 (or 9 bar). For these

reasons, no specific surge control devices were considered

necessary. Nonetheless, transient-induced damage did in

fact occur.

3.1 Simulation results

The objective of hydraulic analysis was to explain causes of

damage and to recommend measures for transient control.

Hydraulic analysis started with a simulation of a transient

event caused by power failure for the 10 pumps in the PB

C2. Figure 8 depicts several characteristic snapshots,

obtained by numerical simulation. The vertical lines

indicated represent instantaneous elevations of hydraulic

Figure 6. Explanation of valve failures.

Forensic transient analysis of two pipeline failures 89

grade lines for three time snap-shots, with the pipe axis as

the reference level; circles indicate calculation nodes on the

pipe. The three snap-shots indicate (a) the initial steady

state conditions, (b) the minimum pressures when vaporous

cavitation is taking place in the collecting pipe within PB

C2, and (c) the moment when the largest vaporous cavity

has just collapsed. Fence diagram (c) indicates that a

pressure spike of around 15 bar occurs, certainly sufficient

to damage the pipe.

Simulated behaviour of the largest vaporous cavity is

shown in figure 9. Formation of the cavity (maximum size

0.4 m3) is relatively slow, lasting for about 10 s, but the

collapse is rapid, resulting in a sharp pressure rise. The

envelopes of extreme pressures in figure 10 are important

for the design and risk assessment of the existing system. It

can be seen that pressures over 15 bar occur in the first

100 m of the pipe, pressures well above the rating of the

pipeline. The pressure increase is caused first by cavity

collapse, but also because of wave amplification that takes

place in the progressively narrowing pipe. By contrast,

pressures in the whole length of the DN 2100 pipeline (from

900 m on) remain below allowed working pressures, and

remain positive. It again should be emphasized that an

over-simplified model, using four equivalent pumps, cannot

reproduce or explain the fluid behaviour in the pump

header lines.

A more serious situation occurs when power is cut to all

pumps in two PBs on one side, namely to C2 and C3, or C4

and C5. Minimum pressures in the DN 2100 pipeline are

below atmospheric pressure for the majority of the pipeline

(see figure 11), a result that may lead to pipe buckling.

Maximum pressures in the most upstream part may now

exceed 20 bar. The largest cavity in this case is approxi-

mately 1.4 m3, forming in about the same time as before,

but collapsing even faster.

3.2 Surge protection

The first decision regarding how to improve the operational

reliability of the system was to install a parallel pipe to the

existing one, thus reducing the maximum velocity and

enabling independent operation of each production block

in an emergency situation. Certainly this parallel connec-

tion increases the redundancy of the system, but it also

decreases the risk of damage due to hydraulic transients.

Another measure to lower the risk of pressure surge is to

install vacuum breaking valves at the connection to every

second pump (originally such valves on every pump were

suggested, but this was not implemented). Results of the

simulations shown in figures 12 and 13 indicate that this

Figure 7. Layout of the system—desalination plant.

Figure 8. Snapshots from the simulation of power cut to

pumps in the production block C2, (a) steady state, (b)

minimum pressures with vaporous cavitation, and (c)

maximum pressures after cavity collapse.

90 M. V. Ivetic

measure was effective. About 10 kg of air is introduced into

the pipeline and occupies a volume of up to 9 m3. (If air

valves were placed at each pump connection, the amount

would be 10% larger.) These valves have a much smaller

outflow capacity, so they cannot evacuate air efficiently.

Therefore, a certain delay for a safe system restart is

required. In the case of the power cut to pumps in PB C2

and C3, around 45 kg of air (occupying volume of 35 m3)

would be introduced.

This solution of installing double-action vacuum break-

ing valves at every other pump connection was relatively

inexpensive. Although negative pressures can reach – 0.5

bar between vacuum breakers, there is no risk of pipe

failure because of the relatively small pipe diameter (DN

Figure 9. Diagrams for the most upstream cavity after power cut to pumps in the production block C2.

Figure 10. Longitudinal profile with envelopes of extreme HGL after power cut to pumps in PB C2.

Forensic transient analysis of two pipeline failures 91

350 and DN 500) in the area affected. The risk of back-

flow, i.e. of introducing polluted water from outside, is

very low since the pipe runs above ground. The

disadvantage of this solution is that a significant amount

of air is introduced into the system, and the restart of the

system after pump power failure must be careful and

controlled.

3.3 Further analyses

After the recommended solution was implemented it was

realised that other solutions might also have been

considered. For example, a shift in operating approach

could be used to curtail the risk of pipe and equipment

damage without requiring any new equipment, even though

vaporous cavitation would still occur in parts of the system.

A solution which permits limited transient cavitation might

have been acceptable during the design phase, but is

unsuitable once accidents have occurred.

In the current design, a downstream valve controls the

flow through each pump. These valves are used to match

pump flow to the production of water, and to throttle the

pumps. Under regular operating conditions, the maximum

working pressures in the upstream part of the system within

PB C2, are between 2.0 and 3.2 bar, while the pump head is

around 48 m. In the case of vaporous cavitation (figure 9) it

can be seen that the net rate of cavity formation (the

difference between upstream and downstream flow rate) is

around 0.1 m3/s, while the discharge from the first pump is

about 0.3 m3/s. Thus, if the pumps were not throttled, the

greater flow of water would lead to less severe cavitation.

As a result, the most efficient measure to protect this system

could be to increase the pressure in the system by 2.0 bar,

and to have throttling at the entrances of PWTs.

Figure 11. Snapshot from simulation of power cut to pumps in the production blocks C2 and C3, minimum pressures with

vaporous cavitation.

Figure 12. Snapshots of simulation of power cut to pumps in the production block C2 with vacuum breakers, steady state

(upper) and minimum pressure (lower).

92 M. V. Ivetic

Results of the simulation related to power failure of the

pumps in PBs C2 and C3 (a more serious case than

considered previously) are given in figures 14 and 15. The

most interesting case is that although there is short-lived

vaporous cavitation in the first 500 m, pressures do not

increase above working pressures, and there is no risk of

pipe damage. Why is it possible to significantly reduce risk

by a relatively small change in operating schedule? The

point is that although localized cavitation does occur, the

cavities are small and well distributed along the pipe.

Figure 13. Longitudinal profiles with envelopes of extreme HGL after power cut to pumps in PB C2 with the vacuum

breakers in the adequately protected system.

Figure 14. Change of operating pressure as a measure to increase the system safety: (a) initial steady state, (b) minimum

pressures after power failure for pumps in blocks C2 and C3, (c) moment after collapse of the vaporous cavity.

Forensic transient analysis of two pipeline failures 93

Through an unrestricted pump, water can reach these

cavities more easily. The largest cavity is smaller than

0.02 m3, and its collapse no longer causes serious over-

pressures.

Another inexpensive solution would be to permit several

pipes to drain constantly and discharge a few litres of water

at each end of the header pipe. In the case of pump failure,

air would be introduced at these points. This is not a

popular measure, but less expensive than throttling of each

pump by 2.0 bar or more. Yet it should be noted that these

solutions do not necessarily apply to even apparently

similar cases. They are only attractive in a given situation

when poor decisions have already been implemented. This

kind of alternative can offer the advantage of requiring

little additional investment.

The best solutions are usually planned from the

beginning of the project, and yet flexible enough to be

adjusted as modifications to the system occur. Operation of

the system needs to be monitored and measurement results

must be analysed soon after they are collected. Any risk

assessment during system operation revealing that the

system has become vulnerable to transient flow accidents

would then encourage immediate action to prevent or

control damage.

3.4 Causes of failure

Compared to the remainder of the desalination plant, this

part of the system was given insufficient attention during

design. For example, a separate power supply for every

production block is provided, yet damage at any location of

the header pipeline requires the whole desalination plant to

be shut down. Moreover, the transient analysis performed

for this part was inadequate because the pipeline layout was

greatly oversimplified.

The lack of detailed transient analysis was masked by a

higher pressure class. GRP pipeline pressure class PN 09

is used for pressures up to 9 bar, although working

pressures were between 2 and 3 bar. Since these pipes

must be tested in the field at a pressure of 13.5 bar and in

the factory to a pressure of 18 bar, it seemed that the

design margin was more than sufficient to cover un-

expected situations.

However, although the pipes were probably tested in the

factory, they may not have been checked at the required

pressure in the field. In fact, it is difficult to conduct

pressure tests on a large pipeline laid above ground. GRP

pipelines have little ability to sustain high longitudinal

stress, so they need huge supports. For example, if the test

pressure is 13.5 bar, a 908 bend for DN 2100 pipeline needs

support to accommodate a force equal to 6600 kN, while

the maximum working pressures are about one quarter this

value. Moreover, under the worst conditions, such high

pressures can occur only in the most upstream part of the

pipeline where diameters are much smaller. The pipeline

DN 2100 was unnecessarily oversized. Since all pipes were

chosen of the same pressure class, the designer showed little

awareness of this fact. It is likely that any pressure test was

carried out with pressures say 50% higher than maximum

working pressures.

Figure 15. Longitudinal profiles with envelopes of extreme HGL after power cut to pumps in PB C2 and PB C3 in the

system with increased operating pressure.

94 M. V. Ivetic

In addition to all this, the pumps were also oversized. It

is quite common to choose larger pumps with the idea that

the system will be easier to control. With an increase of

capital costs, operational costs were also increased, because

it was necessary to throttle pumps.

4. Conclusions

These two case studies illustrate both the false sense of

security that can arise from remaining comfortable with

long held, but erroneous beliefs, and the consequences of

many ‘‘small’’ changes to systems that are not thoroughly

documented or checked. The consequences in these cases

were not that severe, but disruptions of operation were

unacceptable and expensive.

In the case of the Abbatemarco pipeline, transients were

probably not directly responsible for the damage to the

drain valves. However they aggravated two hidden flaws of

the system and one poor decision at the time of construc-

tion.

In the case of the desalination plant, one can identify

at least three common beliefs that compromised the

safety of the pipeline. First is the belief that low-pressure

systems are not susceptible to pressure surge. The second

is that a significant margin in design pressure class for

pipes will cover all unexpected events. The third

misconception was that by choosing larger pumps, the

system will be easier to control. Considerable throttling

prevented these pumps from acting as uncontrolled

inflows, and to some extent to decrease the risk of

column separation. Unexpectedly, however, the additional

capital expense not only increased operating costs, but

impaired the reliability of the system.

Common to both cases is that after the design had been

completed, no additional risk assessments were performed;

in fact, only one iteration of the design loop was

performed. Design procedures have to be improved in a

way that transient specialists are made aware of the

system layout and operating conditions at the inception

phase of the design, and then are called to verify that the

analysis that was performed is still suitable at the time the

system is commissioned. If this final check is not correctly

done, the accuracy of the numerical method is mean-

ingless.

Acknowledgements

The author wishes to thank Prof. Francesco Calomino,

University of Calabria, and Mr. Rale Markovic, from

Fichtner, Germany, who helped him in solving the

problems that are presented here. Snapshots of transients

are prepared with the software developed by Mr Ranko

Lazic. The paper has been prepared during the author’s

stay at University of Exeter as The Leverhulme Trust

Fellow. Financial help of The Leverhulme Trust and

collegial help of Prof. Dragan Savic from The University

of Exeter are greatly appreciated.

References

Almeida, A.B., Accidents and incidents: a harmful/powerful way to

develop expertise on pressure transients. In Hydraulic Transients with

Water Column Separation, 1992 (Universidad Politecnica de Valencia &

IAHR).

Far I., Stanic, M. and Ivetic, M.V., Integrated environment for analysis of

transients in water distribution systems, 6th International Conference on

Hydroinformatics, edited by Liong, Phoon and Babovic, 2004 (World

Scientific: Singapore).

Ivetic, M.V., Pumping stations and transient flow analysis. InWater Supply

Systems—New Technologies, NATO ASI Env. Series 2, No. 15, edited by

C. Maksimovic, F. Calomino and J. Snoxell, 1996 (Springer-Verlag:

Berlin).

Ivetic, M.V., The failure of a desalination plant header pipeline—causes

and proposed remedy, Proceedings of the International Conference on

CSHS03, 2003.

Karney, B.W. and McInnis, D., Transient analysis of water distribution

systems. J. Am. Wat. Works Ass., 1990, 82, 62 – 70.

Locher, F.A., Huntamer, J.B. and O’Sullivan, J.D., Caution—pressure

surges in process and industrial systems may be fatal, Proceedings of the

8th International Conference on Pressure Surges, 2000.

Parmakian, J., Water-Hammer Analysis, 1963 (Dover Publications: New

York).

Provoost, G.A., Investigation into cavitation in a prototype pipeline caused

by water hammer, 2nd International Conference on Pressure Surges, 1976.

Streeter, V.L., Transient cavitating pipe flow. J. Hydraulic Engng, 1983,

109.

Thorley, A.R.D., Fluid Transients in Pipeline Systems, 1991 (D. & L.

George: UK).

Tullis, J.P. and Watkins, R.K., Pipe collapse caused by a pipe rupture—a

case study. In Hydraulic Transients with Water Column Separation, 1992

(IAHR Working Group: Valencia).

Wylie, E.B. and Streeter V.L., Fluid Transients, 1978 (McGraw-Hill: New

York).

Forensic transient analysis of two pipeline failures 95