Urban Water 2 (2000) 153±170

www.elsevier.com/locate/urbwat

Rehabilitation strategies for water distribution networks: a literature

review with a UK perspective
M.O. Engelhardt a, P.J. Skipworth b, D.A. Savic a,*, A.J. Saul b, G.A. Walters a
a
Centre for Water Systems, Department of Engineering, School of Engineering and Computer Science, University of Exeter, Exeter, Devon EX4 4QF,
UK
b
Department of Civil Engineering, University of Sheeld, UK
Received 7 January 2000; received in revised form 16 May 2000; accepted 11 September 2000

Abstract
Primarily, a rehabilitation strategy should aim to satisfy the regulatory requirements set down in respect of water distribution
network operation. However, water companies in the UK have come to recognise that the business needs associated with the
improvement of the deteriorating fabric of their distribution networks extend beyond these requirements. Extra economy can be
gained by operating the networks eciently based on a rehabilitation strategy which considers the associated costs over an extended
period. Economic, hydraulic, reliability and water quality performance criteria must be optimised as part of an e€ective strategy.
Numerous rehabilitation decision making approaches have been presented. However, many have adopted ¯awed economic ap-
proaches and have been based inadequately on one or two selected performance criteria. Few models have considered the extended
planning horizons associated with a whole-life costing approach to this problem. However, the multi-objective optimisation ap-
proaches which have been developed recently have the potential to be developed into the required whole-life costing model based on
the appropriate economic model and performance criteria. Ó 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Rehabilitation; Decision models; Water distribution networks

1. Introduction by a sound understanding of the interactive economic

impacts of system performance.
A rehabilitation strategy needs to be in place to en- Much has been written on system performance and
sure that a water distribution system continues to op- rehabilitation strategies. It is appropriate that, before
erate eciently and economically within de®ned any attempt is made to derive an approach to optimise
operating requirements over an extended period. At- rehabilitation strategy, relevant literature that describes
tention is now moving away from reactionary strategies, the ®ndings and opinions of policy makers, researchers,
which involve little long-term planning, towards pro- industrialists and others is evaluated. The current review
active approaches based on predictive analyses to aims to address this need.
achieve longer-term economic eciency. These predic- Many facets of the water supply discipline have a
tive analyses must be based on a thorough under- bearing on the holistic approach to distribution network
standing of how the various and numerous system management, and the volume of relevant literature is
descriptive parameters impact on the process of system therefore large. An attempt has been made to concen-
deterioration and, in turn, on the di€erent aspects of trate on more recent literature that is closely related to
system performance. In addition, the need to optimise the core subjects of mains deterioration and its conse-
the economic eciency of the rehabilitation strategy quences, and the development of rehabilitation strate-
within operational requirements must be complemented gies. It is emphasised, however, that an awareness of
related areas is necessary in the development of a whole-
life costing approach to distribution network manage-
ment.
*
Corresponding author. Tel.: +44-1392-263-637; fax: +44-1392-
The primary aim of the review is to identify the var-
217-965. ious elements that a rehabilitation decision model
E-mail address: [email protected] (D.A. Savic). should encompass. The compromises that are inherent

1462-0758/00/$ - see front matter Ó 2000 Elsevier Science Ltd. All rights reserved.
PII: S 1 4 6 2 - 0 7 5 8 ( 0 0 ) 0 0 0 5 3 - 4
154 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

in existing models have been identi®ed in order that a must be achieved by the essentially ®xed assets that
more equitable approach be derived. make up the distribution network.
Added to the changing requirements of the distribu-
tion system is the in¯uence of the surrounding envi-
1.1. Rehabilitation strategies
ronment and its role in the deterioration of the system's
performance. This is not limited to the forces attribut-
In the UK, minimum performance levels associated
able to the external environment (e.g., ground heave,
with the delivery of potable water to the customer have
trac loading, soil corrosivity) but extends to the in-
been de®ned at the regulatory level. A rehabilitation
ternal interaction between the mains and the transported
strategy must ensure that these performance require-
water (water pressure, corrosivity of water). The par-
ments are met. In addition, the strategy should aim to
ticular mechanism of mains deterioration through in-
maximise the economic eciency of the water company
ternal and external corrosion is largely dependent on the
in operating its distribution networks.
type of main and water quality.
Numerous discussion papers (Satcha, 1978; Andreou
Iron-based mains have inherently high structural
& Marks, 1986; Walski, 1987b; Evins, Stephenson,
strengths but are prone to destructive corrosion pro-
Warren, & Williams, 1989; Deb, 1994; Slipper & Whipp,
cesses. Corrosion is an electrochemical reaction between
1994; Wright, 1994; AWWA, 1996; Madiec, Botzung,
two materials in contact with each other that have a
Bremond, & Eisenbeis, 1996; Skarda, 1996; Sgrov et
di€erence in potential. This allows for a transfer of
al., 1999) have identi®ed the components that a reha-
electrons from one to the other, resulting in the loss of
bilitation decision model should ideally comprise. Of
parent material, in this case the main. In respect of ex-
paramount importance is a cost model which aims to
ternal corrosion, the interface is between the main and
minimise the costs borne by the water company. The
the surrounding soil. The properties of the soil, in par-
strategy must ensure that the hydraulic performance of
ticular its pH, redox potential and the existence of
the rehabilitated system is within regulatory stipula-
sulphides, a€ect the rate of corrosion. External corro-
tions. In addition, the water delivered to the customer
sion can also result where stray currents, most com-
must meet the quality guidelines that are placed on the
monly from transport systems, are present.
®nal product. The service must also be reliable with a
In iron-based mains, internal corrosion has limited
minimum of interruptions; hence the need for a reli-
structural e€ects but instead impacts on water quality
ability component to the decision model. The ability to
and the hydraulic performance of the system. The type
carry out the scheduled work forms what is often re-
and consequences of the reaction which occurs in the
ferred to as the operational component.
electrochemical cell, which is set up between the mains
It has been common in the past for rehabilitation
water and the internal bore of the pipe, is dependent
decision models to consider separately the elements
on water quality. An ideal combination of water pH
outlined above. Each of these components, therefore,
and alkalinity results in a thin protective layer of cor-
has been examined in turn in the early sections of this
rosion product, which inhibits further corrosion.
report. The various rehabilitation decision models that
However, if this balance does not exist, then the cor-
have been reported are introduced later. These models
rosion reaction continues, resulting in the creation of
are grouped in terms of the ®nal form of the decision
voluminous hydroxide and oxide by-products from
model and the components considered by each. This
small amounts of parent metal. These by-products in-
has allowed the limitations of each to be easily iden-
crease the roughness and reduce the e€ective diameter
ti®ed.
of the mains, thus reducing their hydraulic eciency.
The presence of these by-products may also result in
water quality problems. If, upon formation, the prod-
2. Deterioration of water distribution system performance ucts are taken up in solution, they will form red-col-
oured carbonates and hydroxides that contribute to
It has been identi®ed that the water distribution discoloured water events.
network must convey water of the quantity and quality Cement-based mains may also undergo a chemical
set out in the required levels of service. These require- reaction internally and externally. If the carbon dioxide/
ments change with time. In addition, the pattern of de- calcium carbonate content of the mains water is unbal-
mand changes with the growth in per capita anced, then the calcium carbonate content of the cement
consumption and demographic changes associated with, will be leached out, with an accompanying reduction in
for example, population increase and changes in occu- the structural strength of the main. This mechanism can
pancy rate. Water quality requirements are becoming also increase the pH and alkalinity of the water to un-
ever more stringent, not least because the understanding acceptable levels. This calcium oxide leaching can also
of associated health e€ects is ever increasing. These occur externally, and to a much higher degree, in high
factors have resulted in a changing de®nition of what sulphide content soils.
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 155

3. Statutory and mandatory requirements in the UK water leakage targets have been introduced together with strict
industry enforcement measures to ensure compliance. Such tar-
gets are set annually after consultation with the indi-
The UK water industry was privatised in 1989 with vidual water companies, the EA and the Secretary of
the then government-run water utilities split into 26 State.
private water companies. Due to the monopolistic na-
ture of the water industry, the UK Government eco-
3.3. Water quality
nomically regulates these companies through the Oce
of Water Services (Ofwat). A primary function of Ofwat
In terms of water quality, Ofwat is purely a reporter.
is to ensure that the water companies carry out their
The Drinking Water Inspectorate (DWI) is responsible
functions in line with the Water Industry Act, 1991, and
for the sampling and testing of the water supplied to the
within the framework set out by their company licenses.
customer. All rehabilitation decisions concerning water
Further, Ofwat must ensure that the companies are able
quality fall under the purview of the DWI. These deci-
to ®nance their operations, in particular by securing a
sions are relayed to Ofwat, who put them into context
reasonable rate of return on their capital.
with the other regulatory issues.
3.1. Levels of service
3.4. Serviceability
A major component of the regulatory regime imposed
by Ofwat is comparative competition. This involves As the economic regulator, Ofwat make the ®nal
cross-comparison of the performance of the 26 water determination on the price that the water companies can
companies. To aid comparison, Ofwat has de®ned level- charge their customers. These prices determine the
of-service measures that the water companies are re- capital raised by the water companies and, hence, de-
quired to report annually to Ofwat. These are known as termine rehabilitation expenditure. To aid in their de-
Director General (DG) standards. Those which are most terminations, Ofwat use the term serviceability to
directly relevant to water distribution system operation measure the performance of water assets. Ofwat (1999)
are the DG2 and DG3 levels of service measures. de®ned serviceability as ``a measure of the ability of the
The DG2 measure corresponds to the number of company assets to provide the service required by cus-
properties which generally receive or are at risk of re- tomers''. Arguably, this de®nition encompasses the hy-
ceiving mains water at a pressure below 10 m head draulic, reliability and water quality performance
(except in periods of high demand) and are therefore criteria that have been introduced. However, the de®-
placed on a register by the water company. The DG3 nition also encompasses the customer's expectations
standard corresponds to the number of interruptions to and, in turn, their willingness and ability to pay. These
supply, with a distinction between planned and un- factors are dicult to determine, and therefore, under
planned interruptions and their duration (less than 6, 12 this de®nition, serviceability is perhaps unquanti®able.
or 24 h). The other levels of service correspond to the The value of the approach outlined by Ofwat (1999),
availability of water resources (DG1), the number and which identi®ed four key elements (the DG2 and DG3
level of water restrictions imposed (DG4), the number of measures, the number of bursts and the level of quality
sewer ¯oodings (DG5), billing contact (DG6), written compliance) for consideration in determining a main's
complaints (DG7), number of meter readings (DG8) serviceability, is therefore uncertain.
and ease of telephone contact (DG9). UKWIR (1998) identi®ed that, in general, measures
An indication has been given by Ofwat (1998) of how of serviceability could fall into one of three categories:
these individual measures are to be incorporated into a a measure of quality output, a measure of company
single level-of-service indicator. Ofwat proposed the use input or an estimate of the operating capability of the
of a weighted sum of the level-of-service measures, to company's infrastructure. The serviceability measure
indicate the current state of the system, and a further which is incorporated as part of the Ofwat regulatory
measure to capture improvements in the levels of service regime is a measure of quality output. This is em-
over a single year. phasised in Ofwat (1999), which indicated that
changes in levels of capital maintenance would be tied
3.2. Mandatory leakage targets to changes observed in the levels of serviceability. The
rearward view to serviceability taken by Ofwat has
Water companies, Ofwat and the UK Environment serious disadvantages. The approach results in no in-
Agency (EA) all support the concept of reducing dis- dication being given as to why past levels were expe-
tribution losses to their economic level. To promote and rienced. Another disadvantage of a rearward view is
sustain a reduction in company leakage levels, in addi- that no consideration is given to the levels that a
tion to the levels of service requirements, mandatory company can or should aspire to.
156 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

4. Modelling water distribution system performance out at Severn Trent Water, estimated that growth rates
in distribution losses varied between 1 and 4 l/prop/h per
4.1. The economics of water distribution systems year, with the lower end of the range tending to re¯ect
the presence of newer pipelines.
In terms of their regulated business, the primary aim
of water companies in England and Wales is to maxi- 4.1.1.1. The economics of leakage and leakage control.
mise pro®ts within the terms of their license and the Report 26 (NWC/DoE, 1980) stated that it was clearly
constraints imposed by their government-appointed uneconomic to ensure that pipelines and reservoirs will
regulatory bodies. The revenue generated through the never leak but also that there was an economic limit to
supply of potable water, and collection and treatment of the extent of water loss. This highlighted the need to
sewage, is essentially outside the control of the water control leakage and to achieve a balance in the cost of
company. To maximise pro®ts, therefore, eciency in the e€ort expended with the costs associated with the
delivering the service to the required standard must be loss of water. This prompted practitioners to consider
increased. This includes eciency with regard to the the economics of di€erent approaches to controlling
operation and maintenance of the water distribution leakage at di€erent geographical levels.
system. Operating costs are incurred in the day-to-day Report 26 recommended that some degree of active
delivery of water through the assets, which comprise the leakage control would be economical in all cases. Active
water distribution network, for example those associat- leakage control would involve the disaggregation of
ed with pumping. Maintenance costs are those borne in large networks into district meter areas in order to
maintaining the water distribution system such that it promote easier management, primarily with regard to
can deliver the service within the required standards. leakage. In combination with this re-con®guration,
With respect to maintenance costs, the e€ects of which has been underway since the early 1980s, pressure
leakage have been quanti®ed to represent the deterio- reduction schemes were recommended as providing a
ration of the distribution network in two di€erent ways. good return in the amount of water saved for modest
Firstly, the quantity of water lost in distribution is installation costs. Pressure reduction schemes have been
considered. Although driven by cost, this approach is optimised in recent years to the point where concerns
primarily undertaken to limit the water required to have arisen regarding the ability of networks to provide
supply the particular network, and in view of mandatory adequate ¯ows for ®re®ghting purposes.
leakage targets. Secondly, burst prediction models can The economic level of leakage for a particular net-
be used to model the costs associated with the future work has commonly been considered in terms of the unit
operation of the network to maintain the continuation cost of leakage (see Fig. 1). The costs of producing and
of supply. distributing water and of reducing leakage will di€er
between networks. For example, where water is fed
4.1.1. Distribution losses through systems from high ground via gravity, the
Distribution losses can be split into the two basic marginal cost of the extra water, which provides the
components of bursts and background leakage. Losses incentive to reduce leakage, is more than that of water
associated with bursts are characterised by the water lost which has had value added to it through being pumped
during the time that reported and unreported bursts are (POST, 1995). There are other considerations that a€ect
allowed to run, including those associated with cus-
tomers' service pipes. Background losses are character-
ised by the seeping of water from ®ttings on mains and
mains which are cracked or perforated through corro-
sion. Lambert (1994) distinguished between background
losses and bursts with an arbitrary ¯owrate value of 500
l/h. System age tends to aggravate the problem of
leakage as the condition of pipes deteriorates. By
avoiding carrying out work in some areas of South-West
Water's undertaking, and assessing the change in dis-
tribution losses, Arscott and Grimshaw (1996) observed
an average rise of 2 l/prop/h per year. The credibility of
this ®gure depends on the choice of a representative
sample of zones on which to carry out such a study. This
®gure will be truly dependent on a large number of
physically and operationally derived variables, special to
each individual zone, and should therefore be treated
with caution. Butler and West (1987), in a study carried Fig. 1. The identi®cation of an optimum leakage level.
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 157

the optimum level. Reducing overall demand through and Sgrov (1998) as aggregated, multiple regression
leakage control activity introduces the possibility of and probabilistic predictive.
deferring demand-related capital investment associated, Aggregated type analyses group together mains that
for example, with the expansion of a treatment work's have the same intrinsic properties. Using linear regres-
capacity. This possibility will increase the value of the sion, a relationship is obtained for each group between
leaked water and make increased leakage control ac- time and the number of bursts. The primary ¯aw in this
tivity more economical. type of analysis is that there commonly exist little or no
Welsh Water developed the bursts and background data for a large percentage of these homogenous groups.
estimates (BABE ± Lambert, 1994) concept for leakage However, Shamir and Howard (1979) proposed an ex-
management which provides a component-based esti- ponential increase with time, of the form
mate of distribution losses. BABE calculates 15 com-
k t† ˆ k t0 †eA tÿt0 † ; 1†
ponents of loss from ®ve parts of the distribution system
(trunk mains, service reservoirs, distribution mains, where k t† is the number of bursts/yr/1000 ft at time t, t0
company service pipes and customer service pipes) the base year of the analysis, k t0 † the number of bursts/
through the three basic mechanisms of background yr/1000 ft at time t0 and A is the growth rate coecient
losses, reported and unreported bursts. The component- (1/yr), with A in the range of 0.05±0.15 depending on the
based model uses a combination of information from pipe material and diameter.
three distinct sources of data: standard components Based on the exponential equation of Shamir and
(e.g., pressure correction factors, average burst ¯ow- Howard (1979), a number of researchers have used
rates); auditable local data (e.g., infrastructure data, multiple regression type models. These models (Clark,
recorded frequencies of bursts); and, company policy Sta€ord, & Goodrich, 1982; Walski & Pelliccia, 1982;
(standard of service, leakage control method, waste Walski, Wade, Sharp, Sjostrom, & Schlessinger, 1986),
notices, etc.). This information is expressed in terms of which are again based on aggregated data, relate the
its in¯uence on the average number of days for which environmental and intrinsic properties of the mains with
bursts will run. The BABE model is calibrated using the burst rate as a function of time. Recently, two
locally auditable data, and once calibrated, it is possible studies have investigated the implementation of these on
to examine alternative strategies and associated cost an individual main basis. The ®rst of these was where
implications. This allows the locally based appraisal of the entire burst history was known (Mavin, 1996) and
leakage policy and performance, which can then be re- the second where there was only partial knowledge
lated to economic theory. (Constantine & Darroch, 1995). The increases in bursts
Company estimates of their overall economic level of with time were modelled using exponential and power
leakage are considered by the regulators as part of the functions, respectively.
derivation of mandatory leakage targets. These esti- Probabilistic predictive type models predict the prob-
mates must encompass an economic assessment of the ability that a pipe will burst at a particular time, the time
environmental impact of leakage. This may include to ®rst failure and the useful life of a main. Andreou,
groundwater levels being reduced further than would be Marks, and Clark (1987) used two separate models for
necessary with increased leakage control e€ort, or in- the entire burst history of a main. A Cox proportional
creased river abstraction. The e€ect on ecosystems and hazard model (Cox, 1972), relating the hazard function
in terms of value to the stakeholder is dicult to esti- with a set of co-variate variables (environmental and
mate. Inclusion of a monetary component in consider- intrinsic properties), was used to model what is termed
ation of these factors in an economic analysis may be the early stages of deterioration (up to the third burst).
considered dicult, inappropriate or impossible. The basic form of this hazard model is
In summary, leakage models consider the structural h t : z† ˆ h0 t†ezb ; 2†
failure of mains in terms of leakage levels. These ap-
proaches have been developed to increase the eciency where, h t : z† is the hazard function or the failure rate,
of water distribution systems through decreasing overall h0 t† is some unspeci®ed baseline hazard function, z is a
demand. The economic decision reduces to the question vector of co-variates describing the system's character-
of what leakage levels are economically sustainable. istics (i.e., burst history, intrinsic properties and exoge-
Generally, this enables systemwide decisions to be made, nous variables), and b is a vector of coecients
for example, in respect of pressure reduction schemes or estimated by a regression analysis. After this early stage
the wholesale replacement of parts of a network which of deterioration, a Poisson-type model, assuming a
experience high levels of leakage. constant break rate, was used. The basic form of the
Poisson equation is

4.1.1.2. Maintenance events. Three predominant model- kt†x eÿkt

P x† ˆ ; 3†
ling types for water main bursts were identi®ed by Lei x!
158 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

where P x† is the probability of x failures, t the time posed by Tillman, Larsen, Pahl-Wostl, and Gujer (1999)
interval, and k is the estimated average break rate ob- attempted to include these various social interactions to
tained through a regression analysis of the data. model demand increases into the future. Models fol-
Two other approaches provided probabilistic models lowing such an approach are in their infancy, as Tillman
for the useful life of a main. Lei and Sgrov (1998) et al. (1999) concentrated purely on one interaction, that
assumed that the useful life of a main followed the of the engineer's reaction to increases in demands. This
Weibull distribution. An accelerated lifetime model was is only one of the social phenomena that can readily be
used to relate a set of co-variate variables to the useful identi®ed as part of operating a water distribution sys-
life of a main. Herz (1996) di€ered in that a new prob- tem. Therefore, signi®cant developments are required
ability distribution, the Herz distribution, was assumed before such a model would be of any practical use to the
to model the useful life of a main. This allowed the water industry.
application of a cohort survival model that enabled the The deterioration of ferrous mains through internal
average age of the mains in the network to be forecast corrosion results in decreased hydraulic eciency due to
given a replacement rate. Therefore, the number of increased pipe roughness and reduction in pipe diame-
mains that were beyond their useful life given varying ter. Networks must be maintained at higher pressures in
renewal rates could be estimated. order for increased hydraulic losses to be overcome such
In summary, the development of the maintenance that adequate pressures remain at the extremities of the
event models was initially driven by the consideration of network. A number of studies have investigated this loss
the economic savings associated with the minimisation in hydraulic capacity with time. Generally, these studies
of the number of failures. They allowed rehabilitation have been limited to supplying an empirical equation
decisions to be made on an asset-by-asset basis. This relating pipe roughness to time (Lamont, 1981; Sharp &
reasoning changed with the advent of probabilistic-type Walski, 1988; Dean & Koch, 1993). The generalised
models, which attempted to assign a useful life to a form of the equation for the corrosion of cast iron (as
main, thereby obtaining a time of replacement. These given in Dean & Koch, 1993) is
models allowed decisions to be made on the renewal p ˆ ktn ; 4†
rates required for the overall system and, in this way, are
similar to the leakage models. where p is the pit depth in time t, and k and n are
constants.
4.2. Hydraulic performance
4.3. Reliability performance
In terms of Ofwat's levels-of-service measures, the
hydraulic performance of a distribution network is re- Distribution networks have inherent reliability due to
¯ected in the DG2 measure. The future hydraulic per- their grid-like layout, which is usually a result of the
formance of a water distribution network and the road layout, and the use of service reservoirs throughout
impact on the DG2 measure will be a€ected by increases the system. However, a large number of factors a€ect
in demand and the deterioration of ferrous mains. the reliability of a particular network. These can be split
Based on companies' strategic business plans, Ofwat into design, operational and maintenance considerations
(1994) anticipated that, within a 20-year planning hori- (Walski, 1993). Design includes the choice of materials,
zon, there would be virtually no growth in the overall sizes and location of the components. Operational fac-
demand for water. Reductions in company leakage tors are associated with the monitoring of failures, surge
levels and falling industrial demand would largely o€set minimisation and personnel training. The last consid-
the expected growth in demand from households due to eration, maintenance, is ensuring the proper function-
population increase, a falling occupancy rate and ality of the components and the proper choice and
growth in per capita consumption. The scope for con- implementation of rehabilitation works. Although ef-
sidering the impact of increased demand as part of the forts have been made to include these variables in reli-
justi®cation for increased expenditure on rehabilitation, ability models, there are signi®cant diculties associated
therefore, is currently limited. However, in the medium with their quanti®cation.
to long term, when there is no longer scope for reducing There is no universal agreement of how to de®ne or
overall demand through leakage reduction, the magni- measure the reliability of a water distribution system or
tude of overall demand may increase with increased per as to what should be included in this measure. This has
capita consumption and demographic changes. led to variations in the reported techniques, calculation
Modelling demand increases is fundamentally a social methods and justi®cation for problem formulation.
study. The factors that in¯uence it include changing Diculties arise from the de®nition of failure. A water
regulatory requirements, their associated increases in distribution system is fundamentally a capacitated net-
costs, the customer's willingness to pay and the long- work where connection is insucient without the ability
term e€ect on the customer's demand. A model pro- to supply the required quantity and quality of water.
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 159

The ability to supply the required water to all locations of their techniques from reliability theory to approxi-
is made more dicult in light of the spatial and tem- mate the behaviour of a water distribution system.
poral variability that exists in the various network de- Simulation techniques utilise a hydraulic solver to ac-
scriptors, for example, those associated with component curately model the interactions of the various compo-
deterioration. nents in a system. Inevitably, there is a trade-o€ between
The reliability of a water distribution system can be the computational time and the accuracy of the mea-
de®ned in terms of the probability that the system is sure.
operational (reliability), the percentage of time that the
system is operational (availability), or in terms of indices 4.3.2.1. Path enumeration. Path enumeration techniques
or surrogate measures that are determined to re¯ect the cover di€erent levels of complexity. The simplest of
operational requirements of the system (serviceability). these consider only the connectivity of the network,
Reliability measures have been de®ned as the probabil- disregarding the ability to supply the required demand
ity that all nodes are connected to a source (Goulter & (Goulter & Coals, 1986; Wagner et al., 1988a,b; Kansal
Coals, 1986; Wagner, Shamir, & Marks, 1988a; Wagner, et al., 1995; Yang et al., 1996a,b). More complex ap-
Shamir, & Marks, 1988b; Kansal, Kumar, & Sharma, proaches utilise analytical methods to ensure that the
1995; Yang, Hsu, Louie, & Yeh, 1996a; Yang, Hsu, supply is of sucient magnitude (Wagner et al., 1988a,b;
Louie, & Yeh, 1996b), or the probability that the de- Fujiwara & De Silva, 1990; Quimpo & Shamsi, 1991).
mand is met (Beim & Hobbs, 1988; Hobbs & Beim, Another path enumeration technique utilised a simula-
1988; Lansey, Duan, Mays, & Tung, 1989; Duan & tion-based approach (Su, Mays, Duan, & Lansey, 1987).
Mays, 1990; Duan, Mays, & Lansey, 1990; Yang et al., A slight variation of these techniques were those that
1996a,b; Xu & Goulter, 1998). Some of the surrogate ensured that the system contained a given level of re-
measures that have been utilised include the number of dundancy (Kessler, Ormsbee, & Shamir, 1990; Ormsbee
days outage (Walski, 1987a), the volume of unserved & Kessler, 1990; Ost®eld & Shamir, 1993). Redundancy
demand (Wagner et al., 1988a,b; Bouchart & Goulter, is achieved by ensuring that two paths exist to every
1991; Jowitt & Xu, 1993; Tanyimboh & Templeman, network node. This approach has been extended further
1995) and the number of customers interrupted (Engel- using entropy-based equations (Awumah, Goulter, &
hardt, 1999). Index-based measures have included the Bhatt, 1991; Awumah & Goulter, 1992) to ensure that
ratio of the served demand to the total demand (Fu- the capacity of the links going into a node is sucient.
jiwara & Tung, 1992; Fujiwara & Ganesharajah, 1993;
Gupta & Bhave, 1994), and the complement of the ratio 4.3.2.2. State enumeration. State enumeration ap-
of the unserved demand over the total demand (Fujiw- proaches vary from those based on path enumeration in
ara & De Silva, 1990). Other measures have been sug- that the system failure states are de®ned and then
gested, including the number of failures and the checked to ascertain the reliability of the system in these
duration in the failure state. states. State enumeration techniques can be used to in-
The choice of which measure of reliability to utilise vestigate uncertainty in variables such as demand and
determines the approach used for its calculation. Three pipe roughness (Lansey et al., 1989; Bao & Mays, 1990).
main approaches have been reported: Monte Carlo The minimum cut set approach of Su et al. (1987)
simulation, path enumeration and state enumeration. highlighted the extensive computational time required to
investigate each of the combinations of mains failures.
4.3.1. Monte Carlo simulation To overcome this constraint, in consideration of the low
Monte-Carlo-simulation-based approaches are re- probability that any two mains would fail at the same
garded as the superior technique, as they allow for any time, state enumeration techniques generally only con-
reliability measure to be calculated. However, there is a sider states where one component is in the failure state.
large computational cost, as simulations are carried out State enumeration lends itself to simulation-type ap-
over the entire lifetime of a water network (Beim & proaches. However, the analytical approaches of fre-
Hobbs, 1988; Wagner et al., 1988a,b). This is a major quency duration analysis (Hobbs & Beim, 1988) and
concern in determining the reliability of a system when Markov models (Beim & Hobbs, 1988) were used to
incorporating this measure in an optimisation frame- calculate the reliability of bulk supply. These models
work where a large number of solutions require inves- were extended to include pumping (Duan & Mays, 1990;
tigation. This led to the introduction of the two distinct Duan et al., 1990) and then the distribution mains
approaches of path enumeration and state enumeration. (Fujiwara & Ganesharajah, 1993), with the latter in-
corporating a hydraulic simulator. The approaches of
4.3.2. Path and state enumeration Goulter and Bouchart (1990), Bouchart and Goulter
These techniques reduce computational requirements (1991), Fujiwara and Tung (1992), Tanyimboh and
and can be further di€erentiated into analytic and sim- Templeman (1995), Khomsi, Walters, Thorley, and
ulation approaches. Analytic approaches borrow many Ouazar (1996), Xu and Goulter (1998) and Engelhardt
160 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

(1999) utilised state enumeration, simulation-based ap- Deb (1994) sought performance indicators for the
proaches. These methods di€ered in the variables that standard of water quality in distribution. A bacterio-
were used to de®ne the states of the system, which in- logical measure was presented which indicated the levels
cluded the design demands, uncertainty in variables and of bacteria present. An aesthetic measure referred to the
component failures. An increased number of variables, occurrences of discoloured water, taste and odour. A
however, leads to the consideration of more system further measure dealt with lead, copper and other metals
states and greater complexity and computational re- concentration and disinfection by-products.
quirements. Many problems associated with water quality at the
customer's tap can be recti®ed to a large extent at the
4.4. Water quality performance water treatment stage. In terms of the bene®t of pipe
rehabilitation on water quality, there is a requirement in
Increasingly, the distribution network must perform a decision model for distribution network management
the task of conveying water from local storage areas to to assess the impact of each individual pipe on bacteri-
the customer with as little impact on water quality as ological and aesthetic water quality. However, the dif-
possible. Detrimental e€ects on water quality imposed ®culty in deriving the cost of the problems associated
by the fabric of the distribution network include those with poor water quality must be overcome if this is to be
which are visible to the customer in the form of in- included in a cost±bene®t analysis.
creased turbidity or increased colour. These e€ects are
predominantly caused by corrosion products from un- 4.4.1. Bacteriological water quality
lined ferrous pipes passing into the water. The rate of Levels of bacteria are generally checked through
corrosion is a€ected by the chemical qualities of the routine sampling at locations around the network using,
water. The mobilisation of the associated products de- for example, measures of total coliforms present or
pends on the hydraulic regime within the particular pipe. disinfection levels. Set requirements are generally
The distribution network can also harbour microbial achieved through the dosing of chlorine at the treatment
growth which can cause bacteriological failure and works. The chlorine o€ers a disinfection facility, oxi-
consume free chlorine (Biswas, Lu, & Clark, 1993; dising any potentially harmful bacteria as the water is
Clarke, Grayman, & Deininger, 1994). The residence transported through the distribution network. Free
time in each pipe and in the network (i.e., the age of chlorine is consumed in the oxidation of bacteria and by
water) can be important factors in encouraging micro- heavily corroded mains, resulting in a reduction in its
bial growth. residual as the water passes through the network.
The need to ensure higher water quality standards at Quantities of chlorine are added at the treatment works
the customers' taps and the increase in available com- such that at the extremities of the network, the residual
puting power has led to an extension of the capabilities can provide the disinfection requirement to eliminate
of hydraulic network simulation software. For example, bacteria and bacterial re-growth. However, there is a
it has become common for software to be able to cal- requirement to limit the amount of chlorine added, not
culate the age of water and the propagation and only in respect of its cost but because, at the upstream
blending of conservative substances in networks fed by end of the network, high levels of chlorine can cause
one or more sources (Skov, Hess, & Smith, 1991; taste and odour problems. Chlorine can also react with
Gwynne & Brammer, 1994). In addition, models have impurities to produce trihalomethanes, which can be
been developed to simulate the change in water quality damaging to health at high levels. Heavily corroded
as water passes through a network. The impact of each mains which consume large amounts of free chlorine or
pipe will di€er based on material, diameter, hydraulic mains which are likely to harbour or encourage micro-
roughness, the hydraulic regime within the pipe and the bial growth, therefore, con¯ict with the requirement to
chemical make-up of the water. The accumulation of limit the amount of chlorine added to drinking water.
these a€ects, for example with respect to pH or redox
potential, is modelled with each time step. 4.4.2. Aesthetic water quality
Customers have been shown to be more interested in Events which result in aesthetic water quality prob-
aesthetic water quality problems than the health risks lems occur sporadically, depending on the operation of
associated with water supply. Gilbert (1994) referred to the system, and are therefore dicult to capture through
the compromise between what was technically correct routine sampling. They are often reported, however, by
and what customers commonly perceived to be accept- customers, and these complaints are logged by the water
able or correct. Gilbert showed that the customer was companies for reporting purposes. Continuous mea-
willing to pay for increases in water quality. AWWA surement of turbidity within the distribution network,
(1996) concluded that customers were more willing to however, can indicate the cause of aesthetic problems
pay for decreases in the occurrence of red water events (van den Hoven, van der Kooij, & Brink, 1994). High
than for bacteriological improvements. turbidity levels can result through the mobilisation of
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 161

corrosion products by the ¯ow or through high resi- for trenching. These no-dig techniques include molep-
dence times that increase the possibility of internal loughing, pipe bursting, pipe eating, microtunneling,
corrosion. If turbidity increases during times of low directional drilling, impact moling, pipe ramming and
consumption, then this is an indication that a main jacking. Excavations are required to disconnect and re-
would bene®t from being cleaned and re-lined in terms connect all of the services and at either end of the section
of its impact on water quality. Alternatively, measures of main being installed.
could be taken to reduce residence time in the main. If
turbidity increases are observed during periods of high 5.2. Relining
demand then, van den Hoven et al. (1994) suggested that
the removal of by-products by scraping would be su- The lining of a main protects the inner wall from
cient to limit the impact of the main on this aspect of corrosion and its accompanying negative e€ects. Lining
water quality. However, locating the source of aesthetic after pipe cleaning of deposition can be achieved most
water quality problems in terms of the o€ending main cheaply through the use of cement-based or epoxy resin
may be dicult as its e€ects may be manifested down- that has no structural impact on the main and, there-
stream of its location. fore, no positive e€ect on its leakage performance. The
more expensive option of structural relining has the
4.4.3. Chemical quantities additional e€ect of increasing the strength of the main
Numerous chemical quantities are required to be and impacts positively on leakage performance. Plastic
checked. A number of these are a€ected by the uptake pipes can be inserted in the structural relining of a main
by the ¯ow of the material from which the distribution via sliplining or its derivatives of swagelining, rolldown
mains and service pipes are constructed. Unlined mains or a ``U'' liner. The major cured-in-place technique is
and service pipes contribute to increased concentrations hoselining.
of manganese and iron. Lead is also passed into solution
from old service pipes. Polyaromatic hydrocarbons
5.3. Cleaning
(PAH) are released from coal tar linings which were
used in the 1970s.
Cleaning is normally undertaken to remove the build-
The detection of these types of water quality failures
up of corrosion products from mains. Non-aggressive
is again a function of the sampling regime. The best
cleaning techniques (¯ushing, swabbing and air scour-
rehabilitation option from the point of view of pro-
ing) remove the soft deposits without dislodging the
moting improvements in water quality may be to replace
harder encrustations. The more aggressive cleaning
those mains which have the potential to cause water
techniques of pressure jetting, scraping, pigging and
quality failures. However, the need to replace must also
boring remove the hard encrustations as well. The re-
be considered in terms of other performance criteria.
moval of all the deposits will expose the mains to further
corrosion and, therefore, a thin layer of these deposits is
generally allowed to remain or relining is recommended.
5. The operational criterion

A number of options exist in the type of rehabilita- 5.4. Other techniques

tion that can be applied to a water main. In any par-
ticular case, the choice of which to use will be based on A number of other actions can be taken to retard the
the cost and bene®t of each particular rehabilitation rate of deterioration and therefore stem its conse-
option. The number of choices is ever increasing but can quences. For example, the rate of increase of leakage
be split into the four categories listed and discussed can be reduced through the application of detection
below. techniques and pressure reduction schemes. External
corrosion can be prevented through the installation of
5.1. Replacement sacri®cial anodes, and internal corrosion can be inhib-
ited through stabilisation of water during the treatment
The replacement of water mains has traditionally process.
consisted of creating a trench that uncovers the main
allowing the necessary work to be undertaken. This
method has a number of negative e€ects, which include 6. Rehabilitation decision models
the sound, noise and visual pollution, the cost of rein-
stating the road surface, and the disruption to the trac. Numerous decision models of varying complexities
This has spurred two variations, narrow-trenching have been developed to solve for the rehabilitation
methods that minimise the width of the required problem. These are reviewed in the following sections of
trenching, and no-dig techniques that eliminate the need this report. Section 6.1 is an attempt to classify the
162 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

models based on the di€erent approaches which have oritise the mains requiring rehabilitation such that, for
been utilised. a given budget, the extent of the works that are
Attempts to model the economics, hydraulics, water possible can be identi®ed. Since the determination of
quality and reliability of the water distribution system the budget is outside the decision making process, the
were reviewed in Section 4 of this paper. All of these performance of the ®nal rehabilitated system in terms
factors should be included in a rehabilitation decision of hydraulics, reliability or water quality is never
model. As such, a discussion is presented in Sections considered. As the budget has no role in the priori-
6.2±6.6 of how the various rehabilitation decision tisation process, the models cannot be used in long-
models integrate the consideration of these performance term planning. The prioritisation models consider each
criteria. A summary of these points raised in Sections component in isolation, and the performance of the
6.1±6.6 is provided in Table 1. An overall discussion entire rehabilitated system, therefore, is outside the
presented in Section 6.7 outlines how progress might be decision making process.
made in addressing any de®ciencies that may have been
highlighted in the existing models. Current research into 6.1.2.1. Models for prioritising component rehabilitation.
this ®eld is summarised in Section 6.8. In common with the general rehabilitation models dis-
cussed in Section 6.1.1, the prioritisation models often
concentrated on the consideration of a single perfor-
6.1. Decision model types
mance measure. For example, Quimpo and Shamsi
(1991) used a minimum cut set reliability approach to
6.1.1. General rehabilitation guides
place an importance factor on each main. This approach
Early attempts at developing rehabilitation strategies
is one of the path enumeration analytic approaches used
resulted in the formulation of decision models which
in system reliability theory that were discussed in Sec-
considered each main separately and in isolation. There
tion 4.3. A minimum cut set is de®ned as the smallest set
was no attempt made within these models to prioritise
of pipes that causes the system to fail. In this case,
the rehabilitation requirements. Furthermore, it was
failure is de®ned as the loss of connectivity of the supply
common for assessment to be centred around a single
to one of the demand nodes.
water distribution system performance measure, with
Like Quimpo and Shamsi (1991), Schneiter et al.
the ®nal decision tempered by the implicit consideration
(1996) used a capacity-based reliability approach which
of the other performance measures.
gave priority to those mains which would provide the
Shamir and Howard (1979) used a net present value
most improvement in systemwide capacity reliability.
analysis of the future burst costs and the cost of re-
The capacity of the mains was considered as a stochastic
placing the main to decide whether replacement would
variable based on the uncertainty in the roughness and
be economically bene®cial. In the same vein, Walski and
e€ective diameters of the mains in the system. The as-
Pelliccia (1982) de®ned a critical break rate which, if
signment of an upper and lower band to each pipe's
exceeded, would indicate that the main should be re-
capacity allowed an estimate to be made of the upper
placed. Walski (1982) guided the relining decision
and lower bands to describe the capacity reliability of
through consideration of the associated cost savings in
the system as a whole. This led to the ability to prioritise
pumping and energy. Walski (1985) extended this con-
the replacement of each main based on the improvement
cept by providing an equation to identify the length of
that would be o€ered in the system capacity reliability.
main that needed to be lined or, if necessary, whether a
Schneiter et al. (1996) applied this prioritisation model
parallel main was required.
to a simpli®ed network and acknowledged that the re-
Wright (1994) proposed a more comprehensive guide
quired computing time would prohibit its use on a large
for the identi®cation of mains that required rehabilita-
network.
tion. Wright recommended steps which included an as-
To prioritise mains rehabilitation, the approach de-
sessment of water quality surveys, a hydraulic study
veloped in Waterfowl by WRc (Anon., 1997b) utilised a
extended over a 20-year planning horizon with expected
whole-life costing approach to identify all of the costs of
increases in demand, a reliability study which identi®ed
operating and maintaining a main over its entire service
components whose failure would cause excessive prob-
life. Quantifying the operational and rehabilitation costs
lems, the identi®cation of sensitive customers and the
over the lifetime of the system enables prioritising both
best rehabilitation method. How these investigations
the time and method of rehabilitation for each compo-
were incorporated in a decision framework was not
nent.
discussed.
The approach encompassed by UTILNETS (Anon.,
1997c) assigned a reliability index to each main based on
6.1.2. Prioritisation models hydraulic and structural reliability and the e€ect of the
In addition to the functions of the general guides main on water quality. The strength of a main and the
discussed above, prioritisation models attempt to pri- loads imposed on it, both considered as stochastic
 i-
r-

a-
a-
v-
Table 1
The available rehabilitation decision models in the literature

Authors Model type Economic eciencies Rehab. action Hydraulic Network Water quality
reliability
Burst events Leakage

Shamir and Howard (1979) General guidea Regression ) Replace ) ) )

Walski and Pelliccia (1982) General guide Regression ) Replace ) ) )
Walski (1982) General guide ) ) Lining Capacity ) )
deterioration
Walski (1985) General guide ) ) Parallel mains Capacity ) )
deterioration
Walski (1987b) Optimisation Regression ) Replace/reline Capacity ) )
deterioration
Quimpo and Shamsi (1991) Prioritisation ) ) Replace ) Path enumeration )
Lansey, Basnet, Mays, and Optimisation Regression ) Replace/reline Increased demands ) )
Woodburn (1992)
Li and Haimes (1992b) Optimisation ) ) Replace ) State )
enumeration
Kane (1994)b Prioritisation of No. of failures ) ) No. of failures No. of failures No. of failures
systems
Kim and Mays (1994) Optimisation Regression ) Replace/reline Pumping costs ) )
Wright (1994) General guide ± ) ) ) Increased demands State Survey
systemwide enumeration assessment
Schneiter, Haimes, Li, and Prioritisation ) ) Replace Capacity State )
Lambert (1996) deterioration enumeration
Anderson, Scott, and Prioritisation Weighted score ) ) Graded score ) Graded score
McManamon (1997)
Halhal, Walters, Ouzar, and Optimisation ) ) Replace/parallel/ Capacity Flexibilityc Simple
Savic (1997) reline deterioration descriptived
Waterfowl (Anon., 1997b) Prioritisation Probability model BABE model Replace/reline/ Capacity ) No. of failures
clean deterioration
UTILNETS (Anon., 1997c) Prioritisation Probability model ) Replace/reline/ Capacity Path Simple descriptive
M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

clean deterioration enumeratione

Schaetzen, Randall-Smith, Optimisation ) ) Replace/reline Capacity ) )
Savic, and Walters (1998) deterioration
Deb, Hasit, Grablutz, and Prioritisation ± Probability model ) Renewal rates ) ) )
Herz (1998) systemwide
Kleiner, Adams, and Optimisation Probability model ) ) Capacity ) )
Rogers (1998) deterioration
Engelhardt (1999) Optimisation Regression ) Replace Increased State )
demands enumerationf
a
Unless otherwise stipulated, decision models are asset-by-asset types.
b
Prioritisation is achieved through utilising reported failure events.
c
Halhal et al. (1997) comment on the increased ¯exibility in providing parallel mains.
d
Nominal ratings were given to each main based on material type.
e
This reliability assessment is outside the decision model.
f
This model consists of two steps; therefore, either the hydraulic or reliability components are present.
163
164 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

bles, were calculated explicitly to ascertain its structural

reliability. Although, in principle, this type of calcula-
tion is sound, the information necessary to gain reliable
values is not available, and signi®cant assumptions,
therefore, have to be made. The hydraulic reliability of
the system was calculated by assuming each pipe's
roughness as a stochastic variable and observing its ef-
fect on ¯ow and pressure in the main. UTILNETS
considered water quality in terms of the reliability of the
main as quanti®ed by its probability to cause problems.
This probability was assigned subjectively based pri-
marily on whether or not the main was cast iron and on
the chemical properties of the water. A lower probability
of 0.2 signi®ed that the main had a high contribution to
poor water, whereas a probability of 0.8 meant no or
Fig. 2. Criticality-based approach.
very little in¯uence on water quality. A probability of
0.5 signi®ed no result due to missing data. The three
values pertaining to structural, hydraulic and water
quality reliability were combined into an overall value in
order to prioritise mains replacement. It is implied that, 6.1.3. Optimisation models
as part of the UTILNETS rehabilitation decision mak- Optimisation techniques consider the interaction of
ing procedure, the hydraulic performance of the reno- each main with the system as a whole. They allow for
vated system is checked as the ®nal stage. However, it is both the performance and cost of the rehabilitated sys-
unclear how this hydraulic analysis a€ects the decision tem to play a role in the formulation of the rehabilita-
making process. tion programme. The approaches which have been
reported include the optimisation of performance given
6.1.2.2. Prioritising systemwide rehabilitation decisions. a cost constraint (Li & Haimes, 1992b) and the mini-
Kane (1994) used a prioritisation approach to allocate misation of cost given a performance constraint (Walski,
funds by district. A subjective point scoring system was 1987b; Lansey et al., 1992; Kim & Mays, 1994; de
utilised which was based on the failures that had oc- Schaetzen et al., 1998; Kleiner et al., 1998). These op-
curred in the system in the previous year. As this deci- timisation techniques can also be used to consider net-
sion model was developed for a system in the UK, the work rehabilitation as a multi-objective problem
failures were primarily de®ned by the levels-of-service (Halhal et al., 1997; Engelhardt, 1999). This approach
measures proposed by Ofwat. Although not strictly a allows for the trade-o€ between system performance and
prioritisation model, Deb et al. (1998) provided software cost of rehabilitation. However, such techniques require
(KANEW) that made systemwide renewal rate decisions large numbers of trial evaluations to obtain near-global
for di€erent main categories. The model utilised the optimal solutions. To create a practical formulation in
cohort survival model that Herz (1996) reviewed in terms of computational requirements, therefore, many
Section 4.1.1.2 to provide both an upper and lower limit of the models to date only consider one or two perfor-
to the required renewal rate. The output from this model mance measures. Where there is more than one, sim-
compared favourably against the renewal rates of four pli®ed models or measures are utilised.
water companies in the US.
6.1.3.1. Optimising the availability of a rehabilitated
6.1.2.3. Criticality models. A number of authors have system. A Markov model is a step-like analytical ap-
presented what are termed criticality models (PPK, proach that has been used extensively in reliability the-
1993; Anderson et al., 1997; Louws, 1997; Vincent, ory to model any system whose states can be de®ned by
Read, & McManamon, 1997). These models are essen- discrete parameters. It is assumed that the parameters
tially prioritisation models, utilising the basic decision that describe the next system state are dependent on the
framework illustrated in Fig. 2. The decision to replace current parameters. This assumption is paramount to
and the urgency of replacement are based on the con- the use of Markov model theory, which allows for the
dition of the main, or more precisely the risk of it failing, availability of the system and its components to be
and its criticality as a component to the distribution calculated. Li and Haimes (1992b) used the system
network. This conceptual decision framework can also availability as their performance measure and utilised
be visualised in terms of the equation below. the Markov model developed by Li and Haimes (1992a).
They modelled the states of deterioration of a main,
Loss potential ˆ risk  consequences: following the probabilistic model of Andreou et al.
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 165

(1987), and included the state of planned rehabilitation. Kleiner et al. resulted in a time line of the required reha-
Therefore, the decision of whether to replace or repair bilitation actions for the entire life of the system. It should
could be made to maximise the probability that the main be noted that the model was only demonstrated on a small
was operational at any time in its deterioration. Li and network, and its practical application to a large network,
Haimes (1992b) applied this Markov-based deteriora- therefore, remains unproven. Further, there was a the lack
tion model on a simpli®ed network, maximising the of any budgetary constraint in the model, which is a
system availability given a cost constraint. As with the practical consideration of all water companies.
earlier reliability work of Beim and Hobbs (1988), a
Markov-based model required large computational
6.1.3.3. Multi-objective optimisation. In a multi-objective
times, which was re¯ected in the application of the
approach, Halhal et al. (1997) used the rehabilitation
proposed model to a simpli®ed network.
cost as a minimisation objective and the maximisation of
the bene®ts of the rehabilitation schedule as a further
6.1.3.2. Optimising the economics of a rehabilitated sys-
objective. These bene®ts included the improvement in
tem. Lansey et al. (1992) minimised rehabilitation cost
hydraulic performance of the rehabilitated system, its
subject to a hydraulic performance constraint. Two time
increased ¯exibility provided by including parallel
steps were considered such that works were scheduled
mains, the economic savings of replacing mains that
for the current time and 10 years hence given that there
would experience bursts and the water quality bene®ts
would be an increase in the demand on the system. de
associated with replacing old mains. Except for savings
Schaetzen et al. (1998) used a similar approach restricted
associated with reduced numbers of bursts, the estima-
to a single time step. Instead of minimising the reha-
tion of these bene®ts is very subjective.
bilitation cost, Kim and Mays (1994) and Engelhardt
Engelhardt (1999) also reported a multi-objective
(1999) used slightly di€erent approaches and minimised
approach. The ®rst objective considered was the mini-
the operating cost to include rehabilitation, pumping
misation of operating cost. A second objective was
and maintenance costs. Engelhardt (1999) extended this
maximising reliability, which was represented by a
approach by allowing the replacements to be scheduled
surrogate measure based on the number of customer
over a 20-year period. The 20-year period was split into
interruptions. An example of the curves that are ob-
four ®ve-year time periods. The model allowed
tained from a multi-objective approach is seen in Fig. 3,
for mains replacements to be scheduled in any of
which provides the expected customer interruptions
these time intervals, with the expenditure in each period
(TENCI) for the next 20 years given a range of ex-
constrained by the available funds. This ability
penditure levels. To simplify and make practical this
to schedule replacements over an extended period al-
calculation, an assumption was made that mains were
lowed for the various time-dependent parameters, for
replaced by those of the same size. The systemwide
example demand increases, to be included as part of the
e€ect on the surrogate reliability measure, therefore,
model.
had only to be calculated once, and did not change
Kleiner et al. (1998) used an extended planning horizon
after a main had been replaced. The requirement to
to identify the time to the next rehabilitation and a cycle
simulate the e€ect of taking each main out of operation
time between replacements for each main in the network.
at each step of the optimisation process was therefore
The model made the assumption that mains currently in
eliminated. This illustrates the trade-o€s which must be
service were unlined metallic, whereas the mains they
made for the practical implementation of these opti-
would be replaced with would be either lined or non-
misation techniques.
metallic. The time to ®rst replacement was thus a function
of structural deterioration through corrosion and in-
crease in hydraulic roughness. The time between future
replacements was purely a function of structural deteri-
oration. Structural deterioration was considered as part
of an economic analysis of future maintenance costs and,
as in earlier models (Shamir & Howard, 1979; Walski &
Pelliccia, 1982), provided an optimum time of replace-
ment. The deterioration in the hydraulic eciency of the
original mains was modelled using the empirical hydraulic
roughness model of Sharp and Walski (1988) in con-
junction with the hydraulic solver EPANET to ensure
that the pressures in the system remained above the
minimum required. The model attempted to extend the
useful economic service life of the existing main by con- Fig. 3. The reliability±economics trade-o€ for a water distribution
sidering lining as opposed to replacement. The model of system (from Engelhardt, 1999).
166 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

6.2. Economic models occur in a main's hydraulic capacity. Most of the opti-
misation models utilise a hydraulic network model to
A great diversity exists in the costs associated with the de®ne the mains that require rehabilitation. As such,
operation and maintenance of a distribution system. As measurement of hydraulic performance is easily in-
such, the variety of costs and their roles di€er in the cluded, for both the current system and the rehabilitated
rehabilitation decision models considered in Section 6.1. system. An advantage of this approach is that it allows
The optimisation techniques o€er signi®cant advantages the e€ects to be gauged of upsizing and downsizing
over other models. Lansey et al. (1992) and Engelhardt mains on the performance of the network. The use of the
(1999) included multiple time steps, which allowed hydraulic model also allows for a variety of factors to be
spending to be planned over a period of time. considered such as pumping regimes (Lansey et al.,
Many authors (Slipper, 1994; Anon., 1997b; Conroy, 1992; Kim & Mays, 1994), increases in demand (Lansey
1997; Conroy & Hughes, 1997) have recommended a et al., 1992; Engelhardt, 1999) and a variety of demand
whole-life costing approach to distribution network scenarios (Halhal et al., 1997).
management. Such approaches aim at a comprehensive
quanti®cation of the cost associated with operating a 6.4. Reliability models
main throughout its useful service life. For example, the
costs associated with a mains burst include those at- The need to consider the hydraulic reliability of the
tributable to the lost water, the repair, damage to the water distribution system has been well documented,
overlying road surface and surrounding services, dis- although this has been excluded from many of the ap-
ruption to trac, disruption and damage to consumers, proaches discussed in Section 6.1. The prioritisation
loss of ®re®ghting capabilities and possible contamina- models (Quimpo & Shamsi, 1991; Schneiter et al., 1996)
tion of the water supply. However, as shown in PPK considered system reliability but at the expense of ex-
(1993), to quantify just the third-party costs associated cluding other performance measures. In terms of hy-
with a mains burst is too resource intensive. This prob- draulic reliability of the system, a facility is available in
lem identi®ed the role of criticality. In their consideration UTILNETS to carry out this type of assessment.
of criticality, Walski and Pelliccia (1982) and Engelhardt However, the reliability of the system plays no role in
(1999) used a damage factor based on overlying land use. the rehabilitation prioritisation procedure.
Other studies have factored average burst cost based on Two of the optimisation techniques considered reli-
the surrounding infrastructure, overlying land use and ability explicitly. Li and Haimes (1992b) considered
the type of customer being served (Anon., 1997c; Louws, system reliability with their Markov model, but this
1997; Vincent et al., 1997). The application of these came at a signi®cant computational cost which limited
factors, however, is very subjective. This highlights a its applicability in terms of network size and complexity.
need to assign actual costs in a more objective approach. Engelhardt (1999) used a surrogate measure based on
An example of this is found in WRc's Waterfowl model the risk of failure and surrounding land use. To make
(Anon., 1997b), which includes monetary costs to re¯ect the technique practical in computing terms, Engelhardt
the level of customer complaints. (1999) assumed that mains were replaced with mains of
In terms of economic optimisation, it is prudent for an equivalent diameter.
any rehabilitation decision to take into account all the Clearly, there is great scope for further research to
costs that will be borne throughout a water distribution address the inclusion of system reliability in a practical
system's lifetime. Hence the need for a whole-life costing rehabilitation model.
approach to distribution network management. The time
horizons in the model proposed by Engelhardt (1999) 6.5. Water quality models
allow the changing costs of replacement and non-re-
placement to be assessed. The model o€ers the ¯exibility The most complex of the water quality analyses to be
necessary to assess the whole-life costs associated with included in a rehabilitation decision making tool was
the long-term operation of a water distribution network. that of Kane (1994). The ®nal measure used by Kane
reduced to a weighted sum of the number of complaints
6.3. Hydraulic models and failed water quality tests. This provided a measure
of the state of the system but could not be used to de-
The hydraulic performance of a system has been termine rehabilitation actions. Due to this inability, the
considered in a number of rehabilitation models. The consideration of water quality was limited.
general guides and prioritisation models were limited to Regardless of cost, the best policy in terms of im-
modelling the deterioration of the hydraulic perfor- proving water quality, is to replace as great lengths of
mance and possible improvement in individual assets. mains as possible. In view of this, Halhal et al. (1997)
Schneiter et al. (1996) and UTILNETS extended this used a simplistic approach which considered the length
concept to a stochastic analysis of the variation that can of mains replaced as a measure of the overall improve-
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 167

ment in water quality. UTILNETS took an approach that they fail to incorporate whole-life costing. However,
based on the subjective assignment of an index which these approaches can be developed to incorporate a
re¯ected the likelihood that a main may cause water whole-life costing ideology. The multi-objective criteria
quality problems. can be extended to include systemwide reliability, leakage
and water quality analyses. However, the computational
6.6. Operational models requirements expand signi®cantly with each criterion
included in this multi-objective approach. The trade-o€s
The various optimisation-based rehabilitation deci- between computational requirement and model accura-
sion making models decide on the mains to replace or cy, therefore, need to be understood.
reline. UTILNETS and Apollo (Anon., 1997a) went a
step further and considered which was the most appro- 6.8. Current research
priate of the trenching or no-dig options for the re-
placement or rehabilitation of a main based on factors Substantial work is underway in the ®eld of rehabil-
which include, for example, the type of road surface. In itation strategy. WRc are currently investigating per-
addition, through soil classi®cation and surge analysis, formance indicators for water supply system
Apollo (Anon., 1997a) provided a routine to decide on serviceability. They are also investigating the assessment
the material of the main. Some guidance was also given of pipe failures and burst clusters to optimise capital
on the required bedding and back®ll. expenditure. The authors of this review are undertaking
an EPSRC/WITE project addressing many of the fac-
6.7. Discussion tors addressed in this paper.
In the US, AWWARF are sponsoring four projects
In formulating a rehabilitation model, there is a need to investigating the ®nancial planning and prioritisation of
identify both the performance requirements and the im- rehabilitation decision models. They have a further four
provements that are made to these through the rehabili- projects underway which are associated with the oper-
tation of individual components. The optimum economic ational aspects of rehabilitation techniques. A further
solution for the water company to reach the required two projects are being carried out by AWWARF in
performance standards is then central to the process. It respect of the assignment of failure rates and lifetimes to
may be considered that the prioritisation models o€er the cast iron water mains, and some eight studies are con-
best option as far as whole-life costing ideology is con- sidering various aspects of water quality.
cerned. However, they do not account explicitly for the
budget and are not able to consider extended planning
horizons. A further disadvantage is that the prioritisation 7. Conclusions
models cannot predict explicitly the levels of service that
will be provided by the rehabilitated system. This review has highlighted the requirements of a
The various optimisation procedures overcome most rehabilitation strategy. It has been identi®ed that a de-
of the inabilities of the prioritisation models. The in- cision model must incorporate economic, hydraulic, re-
clusion of a hydraulic model of the network allows for liability and water quality criteria. Numerous measures
the rehabilitated system to be analysed in terms of its and models have been reported in the literature. Clearly,
performance. A number of the optimisation approaches there exists no universal agreement on how performance
are multi-objective in that they consider more than one should be measured in a rehabilitation decision model.
performance criterion. In addition, the ability to con- This review does not attempt to identify the most suit-
sider a number of time horizons allows for spending to able of the presented measures; rather, it identi®es the
be planned for an extended period and allows for ob- advantages and disadvantages of each measure. The
servation of the e€ect that no action will have on future most appropriate measures and models may be deter-
budget requirements. mined by the nature of the network, company policies
Regardless of the advantages that the optimisation and the regulatory regime.
models present, it is clear that the models discussed do not There has existed a wide gulf between practice and
fully satisfy the requirements of a rehabilitation model. the research literature. Until recently, only the general
Because of constraints associated with computational guidelines have been applied to actual systems. This gap
power, none of the models considers all of the important is closing in the UK with the increased use of prioriti-
performance criteria as part of an exhaustive analysis. sation models to help meet the regulatory requirements.
This is particularly so in respect of performance criteria Special consideration has been given in this review to
associated with water quality and reliability. Although the current regulatory regime in the UK. Ofwat have
the optimisation models can be said to represent the best proposed some performance requirements through their
attempt at satisfying the requirements of a rehabilitation serviceability measures. These measures, which are uti-
model, they are lacking in their economic approach in lised to determine future capital investment require-
168 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

ments, currently take a rearward view of the perfor- indicate, that an approach is the ``best'' given the knowl-
mance of the system. It is questionable whether a rear- edge available.
ward outlook provides a good indication of the state of
deterioration in a system's assets. It could be argued that
the short-term, rearward serviceability measures play no
role in the long-term economic decisions made in respect Acknowledgements
of underground pipework assets. So soon after privati-
sation, the UK regulatory regime must be considered as This work was supported by the UK Engineering
evolutionary, and the regulatory requirements, there- and Physical Sciences Research Council, grant GR/
fore, cannot be regarded as ®xed in the longer term. The M16115.
regulatory regime in the UK, therefore, presents a
quandary when formulating a rehabilitation strategy.
If required, a whole-life costing approach to distri- References
bution network management can provide a framework
to include regulatory guidelines into the decision process Anderson, M. J., Scott, R., & McManamon, P. (1997). Applying
at a number of levels. It would be possible to allocate condition and criticality to develop ecient inspection, mainte-
monetary amounts to each breach of the levels-of-ser- nance and renewal programs. In 17th AWWA federal convention,
vice requirements. Alternatively, the monetary require- Melbourne, Australia (pp. 674±679).
Andreou, S. A., & Marks, D. H. (1986). Maintenance planning
ments to achieve a given improvement on the required strategies for large diameter cast-iron mains. AWWA Conference
levels of service could be calculated. The impacts on Synopsis, 365±372.
performance of not spending can be assessed as well as Andreou, S. A., Marks, D. H., & Clark, R. M. (1987). A new
the impacts of given levels of spending. This allows for methodology for modelling break failure patterns in deteriorating
the decision model to minimise a system's operating and water distribution systems: Theory. Advanced Water Resources,
10(March), 2±10.
maintenance costs over its entire life. It is therefore an Anon. (1997a). Apollo: mains rehabilitation optimization. Construc-
ideal approach for long-term planning. It must be borne tion and Operation of Underground Utilities, February, 1±7.
in mind, however, that such an approach must be based Anon. (1997b). Optimizing investment in pipeline systems. Construc-
on a well-founded knowledge of underground asset be- tion and Operation of Underground Utilities, February, 1±14.
haviour over extended periods. Anon. (1997c). UTILNETS: system guide and user manual. Brite/
Euram 11, project no. 7210, May, 46 pp.
The role of optimisation in the water industry has Arscott, A., & Grimshaw, F. (1996). Evaluating investment in the
long been in contention and its use, therefore, has been water network on leakage detection. In IIR conference on cost
limited. Optimisation, like all modelling techniques, will e€ective management of water pipelines and networks, London, 24±
fall prey to the old adage of ``garbage in-garbage out'' 25 January 1996.
where input data sets and parameters are incorrect, in- Awumah, K., & Goulter, I. C. (1992). Maximising entropy de®ned
reliability of water distribution networks. Engineering Optimiza-
complete or corrupted in some way. Thus, any research tion, 2, 57±80.
in developing further models in this ®eld must be for- Awumah, K., Goulter, I. C., & Bhatt, S. K. (1991). Entropy based
mulated for real systems and industry data. redundancy measures in water distribution networks. Journal of
This review has identi®ed the advantages of imple- Water Resources Planning and Management Division ASCE, 117(5),
menting an optimisation-based, holistic approach to 595±614.
AWWA (1996). Committee report: water accountability. AWWA
distribution network rehabilitation. Such an approach Journal, 88(7).
allows for the performance of the network as a whole to Bao, Y., & Mays, L. W. (1990). Model for water distribution system
be considered in the decision making process. Literature reliability. Journal of Hydraulic Engineering Division ASCE, 116(9),
indicates how these approaches allow for expenditure to 1119±1137.
be forecast over an extended period in order to meet a Beim, G. K., & Hobbs, B. F. (1988). Analytical simulation of water
system capacity reliability ± 2. A Markov chain approach and
range of performance requirements. These optimisation veri®cation of the models. Water Resources Research, 24(9), 1445±
approaches have been advanced in recent years such 1458.
that the problem can be de®ned as multi-objective. This Biswas, P., Lu, B., & Clark, R. M. (1993). A model for chlorine
allows for the decision maker to consider the trade-o€s concentration decay in pipes. Water Resources, 27(12), 1715±1724.
necessary in making decisions for water distribution Bouchart, F., & Goulter, I. C. (1991). Reliability improvement in
design networks recognising valve location. Water Resources
network rehabilitation and management. Research, 27(12), 3029±3040.
Some recently developed optimisation models are Butler, M., & West, J. (1987). Leakage prevention and system renewal.
being used in the decision making process of the UK In Pipeline management seminar, 1987. Pipeline Industries Guild.
water industry. Once any decision making process Clark, R. M., Sta€ord, C. L., & Goodrich, J. A. (1982). Water
becomes economically auditable, models are required distribution systems: A spatial cost evaluation. Journal of Water
Resources Planning and Management Division ASCE, 108(WR3),
to demonstrate that the proposed alternative is 243±255.
the best of those available. Clearly, it is here that opti- Clarke, R. M., Grayman, W. M., & Deininger, R. A. (1994).
misation models can be used to obtain, or at least Measuring and modeling chlorine propagation in water distribu-
 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170 169

tion systems. Journal of Water Resources Planning and Manage- Hobbs, B. F., & Beim, G. K. (1988). Analytical simulation of water
ment Division ASCE, 120(6). system capacity reliability ± 1. Modi®ed frequency duration
Conroy, P. J. (1997). Achieving a cost e€ective rehabilitation solution. analysis. Water Resources Research, 24(9), 1431±1444.
In CIWEM meeting, November (pp. 1±7). Jowitt, P. W., & Xu, C. (1993). Predicting pipe failure e€ects in water
Conroy, P. J., & Hughes, D. M. (1997). Cost e€ective system distribution networks. Journal of Water Resources Planning and
rehabilitation through whole life costing analysis. In AWWA Management Division ASCE, 119(1), 18±31.
conference proceedings: Vol. E. Engineering and operations, Atlanta, Kane, M. J. (1994). Database to prioritise main rehabilitation. In
Georgia, June (pp. 137±144). Hydrotop '94, Marseille, France (pp. 201±213).
Constantine, G., & Darroch, J. (1995). Predicting underground Kansal, M. L., Kumar, A., & Sharma, P. B. (1995). Reliability analysis
pipeline failure. Water, 2(2), 9±10. of water distribution systems under uncertainty. Reliability Engi-
Cox, D. R. (1972). Regression models and life tables. Journal of the neering and Safety Systems, 50, 51±59.
Royal Statistical Society, 34(Series B), 187±220. Kessler, A., Ormsbee, L., & Shamir, U. (1990). A methodology for
de Schaetzen, W., Randall-Smith, M. J., Savic, D., & Walters, G. A. least cost design of invulnerable water distribution networks. Civil
(1998). A genetic algorithm approach for rehabilitation in water Engineering Systems, 7(1), 20±28.
supply systems. In International conference on rehabilitation tech- Khomsi, D., Walters, G. A., Thorley, A. R. D., & Ouazar, D. (1996).
nology for the water industry, Lille, March (pp. 1±11). Reliability tester for water distribution networks. Journal of
Dean, A. M., & Koch, T. (1993). Assessment of corrosion trends in a Computing in Civil Engineering Division ASCE, 10(1), 10±19.
potable water distribution system including an automatic analysis Kim, J. H., & Mays, L. W. (1994). Optimal rehabilitation model for
technique. In Corrosion and related aspects of materials for potable water distribution systems. Journal of Water Resources Planning
water supplies (pp. 172±194). and Management Division ASCE, 120(5), 674±691.
Deb, A. K. (1994). Water distribution system performance indicators. Kleiner, Y., Adams, B. J., & Rogers, J. S. (1998). Selection and
Water Supply IWSA (Zurich), 12(3/4), 11±20. scheduling of rehabilitation alternatives for water distribution
Deb, A. K., Hasit, Y. J., Grablutz, F. M., & Herz, R. K. (1998). systems. Water Resources Research, 34(8), 2053±2061.
Quantifying future rehabilitation and replacement needs of water Lambert, A. (1994). Accounting for losses: Burst and background
mains. AWWA Research Foundation, 156 pp. concept. Journal of the Institution of Water Environment Manage-
Duan, N., & Mays, L. W. (1990). Reliability of pumping stations. ment, 8(April), 205±214.
Journal of Hydraulic Engineering Division ASCE, 116(2), 230±247. Lamont, P. A. (1981). Common pipe ¯ow formulas compared with the
Duan, N., Mays, L. W., & Lansey, K. E. (1990). Optimal reliability theory of roughness. AWWA Journal, 73, 274±280.
based design of pumping and distribution systems. Journal of Lansey, K. E., Basnet, C., Mays, L. W., & Woodburn, J. (1992).
Hydraulic Engineering Division ASCE, 116(2), 249±268. Optimal maintenance scheduling for water distribution systems.
Engelhardt, M. O. (1999). Development of a strategy for the optimum Civil Engineering Systems, 9, 211±226.
replacement of water mains. Ph.D., Department of Civil and Lansey, K. E., Duan, N., Mays, L. W., & Tung, T. K. (1989). Water
Environmental Engineering, University of Adelaide, 514 pp. distribution system design under uncertainties. Journal of Water
Evins, C., Stephenson, G., Warren, I. C., & Williams, S. M. (1989). Resources Planning and Management Division ASCE, 155(5), 630±
Planning the rehabilitation of water distribution systems, principal 645.
document of the water mains rehabilitation manual. Water Lei, J., & Sñgrov, S. (1998). Statistical approach for describing failures
Research Centre plc. and lifetimes of water mains. Water Science and Technology, 38(6),
Fujiwara, O., & De Silva, A. U. (1990). Algorithm for reliability based 209±217.
optimal design of water networks. Journal of Environmental Li, D., & Haimes, Y. Y. (1992a). Optimal maintenance related decision
Engineering Division ASCE, 116(3), 575±587. making for deteriorating water distribution systems ± 1. Semi-
Fujiwara, O., & Ganesharajah, T. (1993). Reliability assessment of Markovian model for a water main. Water Resources Research,
water supply systems with storage and distribution networks. 28(4), 1053±1061.
Water Resources Research, 29(8), 2917±2924. Li, D., & Haimes, Y. Y. (1992b). Optimal maintenance related decision
Fujiwara, O., & Tung, D. H. (1992). Reliability improvement for water making for deteriorating water distribution systems ± 2. Multilevel
distribution through increasing the pipe size. Water Resources decomposition approach. Water Resources Research, 28(4), 1053±
Journal, 28(3), 29±36. 1061.
Gilbert, J. B. (1994). The management perspectives on rehabilitaion. Louws, P. (1997). Risk/consequence approach to capital program
Water Supply IWSA (Zurich), 12(3/4), 1±10. prioritisation at City West Water. In 17th AWWA Federal
Goulter, I. C., & Bouchart, F. (1990). Reliability-constrained pipe Convention, Melbourne, Australia (pp. 780±783).
network model. Journal of Hydraulic Engineering Division ASCE, Madiec, H., Botzung, P., Bremond, B., & Eisenbeis, P. (1996).
116(2), 211±229. Implementation of a probability model for renewal of drinking
Goulter, I. C., & Coals, V. (1986). Quantitative approaches to water networks. Water Supply IWSA (Durban), 14(3/4), 347±351.
reliability assessment in pipe networks. Journal of Transport Mavin, K. (1996). Predicting pipe failure performance of individual
Engineering Division ASCE, 112(3), 287±301. water mains (p. 189). UWRAA, research report no. 114, July.
Gupta, R., & Bhave, P. R. (1994). Reliability analysis of water NWC/DoE (1980). Leakage control policy and practice. Standing
distribution systems. Journal of Environmental Engineering Division technical committee report 26.
ASCE, 120(2), 447±460. Ofwat (1994). Future levels of demand and supply for water. Ofwat
Gwynne, M. R., & Brammer, L. F. (1994). Extending the bene®ts of occasional paper 1, November.
water distribution modelling in Severn Trent Water. In Hydrotop Ofwat (1998). A proposed approach to assessing overall service to
94, Marseille, France. customers ± a technical paper. Oce for Water Services, 22 March
Halhal, D., Walters, G. A., Ouzar, D., & Savic, D. A. (1997). Water 1998, http://www.open.gov.uk/ofwat/performa.htm.
network rehabilitation with a structured messy genetic algorithm. Ofwat (1999). Serviceability of the water main and sewer networks in
Journal of Water Resources Planning and Management Division England and Wales up to the March 1998. Ofwat information note
ASCE, 123(3), 137±146. 35a, http://www.open.gov.uk/ofwat/info35a.htm.
Herz, R. K. (1996). Ageing processes and rehabilitation needs of Ormsbee, L., & Kessler, A. (1990). Optimal upgrading of hydraulic-
drinking water distribution networks. Journal of Water SRT ± network reliability. Journal of Water Resources Planning and
Aqua, 45(5), 221±231. Management Division ASCE, 116(6), 784±801.
170 M.O. Engelhardt et al. / Urban Water 2 (2000) 153±170

Ost®eld, A., & Shamir, U. (1993). Incorporating reliability in optimal van den Hoven, T., van der Kooij, D., & Brink, H. (1994). Methods to
design of water distribution networks ± review and new concepts. analyse and to cure water quality problems in distribution systems.
Reliability Engineering and System Safety, 42, 5±11. Water Supply IWSA (Zurich), 12(3/4), 151±159.
POST (1995). The 1995 drought. Parliamentary Oce of Science and Vincent, D., Read, M., & McManamon, P. (1997). Strategic water
Technology (POST) technical report 71. supply project evaluation utilising a practical assessment of risk. In
PPK (1993). Identi®cation of critical water supply assets. UWRAA, 17th AWWA federal convention, Melbourne, Australia, 16±21
research report no. 57, June, 131 pp. March (pp. 731±736).
Quimpo, R. G., & Shamsi, U. M. (1991). Reliability based distribution Wagner, J. M., Shamir, U., & Marks, D. H. (1988a). Water
system maintenance. Journal of Water Resources Planning and distribution reliability: analytical methods. Journal of Water
Management Division ASCE, 117(3), 321±339. Resources Planning and Management Division ASCE, 114(3), 253±
Sñgrov, S., Melo, J. F., Conroy, P., Herz, R. K., Legrau€e, P., Moss, 275.
G., Oddevald, J. E., Rajani, B., & Schiatti, M. (1999). Rehabili- Wagner, J. M., Shamir, U., & Marks, D. H. (1988b). Water
tation of water networks: Survey of research needs and on-going distribution reliability: simulation methods. Journal of Water
e€orts. Urban Water, 1, 15±22. Resources Planning and Management Division ASCE, 114(3), 276±
Satcha, J. H. (1978). Criteria for pipeline replacement. AWWA 293.
Journal, 70(5), 256±259. Walski, T. M. (1982). Economic analysis for rehabilitation of water
Schneiter, C. R., Haimes, Y. Y., Li, D., & Lambert, J. H. (1996). mains. Journal of Water Resources Planning and Management
Capacity reliability of water distribution networks and optimum Division ASCE, 108(WR3), 296±307.
rehabilitation decision making. Water Resources Research, 32(7), Walski, T. M. (1985). Cleaning and lining versus parallel mains.
2271±2278. Journal of Water Resources Planning and Management Division
Shamir, U., & Howard, C. D. D. (1979). An analytic approach to ASCE, 111(1), 43±53.
scheduling pipe replacement. AWWA Journal, 71(5), 248±258. Walski, T. M. (1987a). Discussion on: Multi-objective optimisation of
Sharp, W. W., & Walski, T. M. (1988). Predicting internal roughness water distribution networks by I. C. Goulter. Civil Engineering
in water mains. AWWA Journal, 80, 34±40. Systems, 4(4), 215±217.
Skarda, B. C. (1996). Methods of diagnosis and performance indica- Walski, T. M. (1987b). Water supply system rehabilitation. New York:
tors for rehabilitation policies. A Swiss point of view: Pipeline Task Committee on Water Supply System Rehabilitation, ASCE.
networks. Water Supply IWSA (Durban), 14(3/4), 347±358. Walski, T. M. (1993). Practical aspects of providing reliability in water
Skov, K. R., Hess, A. F., & Smith, B. (1991). Field sampling distribution systems. Reliability Engineering and System Safety, 42,
procedures for calibration of a water distribution system water 13±19.
quality model. In AWWA/USEPA technology transfer conference Walski, T. M., & Pelliccia, A. (1982). Economic analysis of water main
on water quality modelling in distribution systems, Cincinatti, OH, breaks. AWWA Journal, 74(March), 140±147.
February 1991. Walski, T. M., Wade, R., Sharp, W. W., Sjostrom, J. W., &
Slipper, M. J. (1994). Pipeline rehabilitation ± challenges and innova- Schlessinger, D. (1986). Conducting a pipe break analysis for a
tions. Water Supply IWSA (Zurich), 12(3/4), 59±69. large city. In AWWA conference symposium (pp. 387±402).
Slipper, M. J. Whipp, S. Integrated rehabilitation of water distribution Wright, W. G. (1994). Network simulation modelling ± a tool in
systems. Water Supply IWSA (Budapest) 12 (1/2) (1994) 6-1, 6-3 . rehabilitation planning. In second international conference on water
Su, Y. C., Mays, L. W., Duan, N., & Lansey, K. E. (1987). Reliability pipeline systems, Edinburgh, Scotland (pp. 299±308).
based optimisation model for water distribution systems. Journal of Xu, C., & Goulter, I. C. (1998). Probabilistic model for water
Hydraulic Engineering Division ASCE, 114(12), 1539±1556. distribution reliability. Journal of Water Resources Planning and
Tanyimboh, T. T., & Templeman, A. B. (1995). A new method for Management Division ASCE, 124(4), 218±228.
calculating the reliability of single source networks. Developments Yang, S. L., Hsu, N. S., Louie, P. W. F., & Yeh, W. W. G. (1996a).
in Computational Techniques for Civil Engineering, 1±9. Water distribution network reliability: Connectivity analysis.
Tillman, D., Larsen, T. A., Pahl-Wostl, C., & Gujer, W. (1999). Journal of Infrastructure Systems Division ASCE, 2(2), 54±64.
Modeling the actors in water supply systems. Water Science and Yang, S. L., Hsu, N. S., Louie, P. W. F., & Yeh, W. W. G.
Technology, 39(4), 203±211. (1996b). Water distribution network reliability: Stochastic simu-
UKWIR (1998). Quanti®cation of serviceability (p. 96). UKWIR, lation. Journal of Infrastructure Systems Division ASCE, 2(2),
research ref. no. 98/RG/01/1. 65±72.