Optimal Intervention Planning: A Bottom-Up Approach to Renewing Aging Water Infrastructure

Kerwin, Sean; Adey, Bryan T.

doi:10.1061/(ASCE)WR.1943-5452.0001217

Open access

Technical Papers

Apr 30, 2020

Optimal Intervention Planning: A Bottom-Up Approach to Renewing Aging Water Infrastructure

Authors: Sean Kerwin https://orcid.org/0000-0001-5764-3742 [email protected] and Bryan T. Adey, Ph.D.Author Affiliations

Publication: Journal of Water Resources Planning and Management

Volume 146, Issue 7

https://doi.org/10.1061/(ASCE)WR.1943-5452.0001217

PDF

Abstract

Water distribution is an essential service provided in cities. To ensure that the public has access to high quality drinking water in sufficient quantity at all times, intervention programs must be determined to replace deteriorated objects in the network. These intervention programs have costs, cause temporary service interruptions, and impact traffic. Optimal intervention programs must be developed and the financial requirements of these programs must be communicated and justified to stakeholders. This is a difficult task at the scale of modern cities due to the extent of large water distribution networks, the variety of objects that must be considered, the number of constraints (i.e., financial and operational) that must be respected, the high level of uncertainty surrounding relevant data, and the reliance on nonformalized expert knowledge of utility workers for decision-making. In this paper, a novel methodology for determining intervention programs for an example water distribution network is presented. The approach is best suited to short-term intervention planning, is based on the existing network structure and components, and is reliant on the intervention strategies defined by the water utility. The methodology allows infrastructure managers to incorporate the expert knowledge of utility workers into the decision-making process, estimate the costs and benefits of executing preventive interventions on all objects (i.e., pipes and objects housed in facilities), and consider budget constraints. Intervention programs for a 5-year planning period for different budget scenarios are shown.

Introduction

Water managers are responsible for ensuring that their water distribution networks provide an adequate level of service (LOS) to clients. This includes managing all objects (i.e., pipes, pumps, hydrants, etc.) over all of their life cycle phases, i.e., construction, operation, maintenance, development, and demolition. The service provided by water distribution networks (WDNs) is of substantial value, as society could not continue to function as it does without it. From the societal viewpoint, service has a quality, quantity, economic, and sustainability aspect. More specifically, clients expect constant access to potable water in sufficient quantity to meet normal and extraordinary demands (e.g., firefighting); water fees should be kept at reasonable levels and fee increases justified; and water consumption should not cause unacceptable levels of environmental degradation nor affect the ability of future generations to consume water. These generally formulated ideals must of course be quantified using measurable performance indicators (Alegre et al. 2016).

With time, society’s requirements for service tend to increase, while various deterioration processes continuously lower the provided service, requiring infrastructure managers to plan and execute interventions as shown in Fig. 1. These interventions consist not only of replacing deteriorated objects. Inevitably, interventions will need to be planned and executed to expand the network (e.g., constructing an additional treatment center) or optimize network operation. For example, leakage is an issue that affects service in terms of economics (e.g., loss of treated water), water quality (e.g., increase of probability of contaminant entry), and quantity (e.g., decrease of pressure). Infrastructure managers can reduce leakage by replacing deteriorated pipes; however, pressure management, which involves analyzing network hydraulics and implementing measures to reduce pressure from unnecessarily high levels, may be more effective depending on the specific network (Pearson and Trow 2005). Nonetheless, interventions on deteriorated or obsolete objects represent a significant majority of all interventions for many utilities. As water infrastructure is largely hidden from society, the renewal of this infrastructure has often been historically neglected (Infraguide 2003b) and many utilities are now addressing this backlog. Such interventions are the focus of this paper.

Fig. 1. Temporal evolution of level of service.

Difficulty of Planning and Executing Interventions

The process of planning and executing preventive interventions is complex for a number of reasons. Firstly, these interventions are planned on specific objects (e.g., pipes, pumps, etc.); however, the service provided to users is measured at the network level (Alegre and Coelho 2012), thus infrastructure managers need adequate models to estimate the incremental improvement of level of service achieved by executing specific interventions. Additionally, service is often an abstract concept that for WDNs can be potentially measured using hundreds of different performance indicators (Alegre et al. 2016). Infrastructure managers need to decide, which specific performance indicators to use to quantify service. The complexity of the problem is increased by the variety of object types that exist in WDNs. Most research on intervention planning has focused on buried pipes; however, treatment plants, groundwater wells, pumping stations, reservoirs, and network surveillance centers are equally critical to the proper functioning of the network.

It is rarely possible to execute all desired interventions due to resource limitations such as an annual budget, availability of trained employees, or political constraints (e.g., interventions on utility networks cannot be executed if the corresponding road has been opened within the last 5 years). Due to such constraints, managers must decide between executing interventions on pipes, pumps, and objects of other types. In order to make such decisions, a cost-benefit analysis of possible interventions is needed. This analysis results in the net benefit of intervention execution and considers service improvement (e.g., reduction in pipe break frequency, improved firefighting ability) as well as all associated costs, e.g., owner costs (i.e., cost of intervention execution) and user costs (i.e., disruption of service, traffic interruption) due to the execution of preventive and corrective interventions. As pipes are typically buried under roads next to other utilities (Mair et al. 2017), this analysis must account for impacts to the public and other networks. Additionally, when other networks plan interventions, the costs and benefits of executing interventions change for the water utility. The combination of increased risk due to construction activities taking place in the immediate vicinity and the potential for cost sharing (i.e., excavation costs, mobilization and setup costs) increase the net benefit of intervention execution. The drawbacks are that such coordinated interventions are more complex and involve more planning to prepare common trenches, work schedules, and minimize impacts to the public. This intense level of coordination and planning must then be done at the city-scale. Modern WDNs consist of thousands of kilometers of buried pipeline and dozens of facilities where water is treated, pumped, stored, and monitored. Maintenance at this scale involves the planning and execution of hundreds of interventions annually. Ideally, the concept of economies of scale would be considered in the cost-benefit analysis, as longer pipe interventions are typically cheaper per unit length than smaller ones. Contractors will often offer discounts on large interventions and additionally setup and mobilization costs per unit length can be reduced. This should incentivize managers to identify groups of contiguous pipes that could be combined in an intervention. In reality, identifying such groupings is difficult due to the combinatorial explosion of ways contiguous pipes can be grouped in a large municipal network and the need to recalculate the intervention costs (IC) and benefits for each considered grouping. Thus, due to scale and complexity, a cost-benefit analysis of all possible interventions is rarely done.

Moreover, managers must make decisions on intervention execution despite high levels of uncertainty (Harvey et al. 2017). Poor data availability and quality, uncertain future level of service requirements, unclear budget needs, and the unknown intervention plans of other networks all contribute to uncertainty. Managers rely on experience and intuition to make the best decisions given the available information. This intuition is gained over decades and is often not formalized in any manner, which makes knowledge transfer difficult and leads to knowledge loss when workers leave or retire. In addition, this intuition is prone to error. As Van Riel et al. (2014) described for sewer networks, intuition in infrastructure management decision-making cannot be considered skilled because a relationship between the taken decision and resulting impact is often not observable, decisions are typically not recorded in a format that allows for periodic evaluation, and objects found in WDNs generally outlast the professional careers of infrastructure managers, which severely limits the learning opportunity needed to develop skilled intuition. Furthermore, the reliance on intuition can also make it difficult to communicate with nontechnical stakeholders (e.g., general public, financial accountants). Such stakeholders expect transparency regarding the spending decisions of public utilities. Basing such decisions entirely on experience or intuition as justification can lead to poor public perception and dissatisfaction.

Guidelines for Intervention Planning

There is increasing interest in the determination of intervention programs, as part of professional infrastructure management. This is perhaps best seen in the development of the ISO 55000 standards (ISO 2014a, b, c) for the management of physical assets, where general guidance is given as to how organizations should align the treatment of their assets with the goals of their organization, i.e., how to develop an asset management plan. Additional guidance can be found in many documents issued by many national organizations and expert groups. Examples of these for water utilities are shown in Table 1.

Table 1. Example guidelines for water utility asset management

Reference	Organization	Country of origin	Description
Infraguide (2003a)	InfraGuide	Canada	General guidelines for developing common intervention programs for municipalities (road, water, and sewer networks considered)
Infraguide (2003b)	InfraGuide	Canada	A number of guidelines for municipalities to aid in investment related decision-making for infrastructure networks including water distribution networks
Infraguide (2003c)	InfraGuide	Canada	Contains information on available rehabilitation and replacement technologies available for water mains and a decision tree on when to use which technology
NAMS Group (2006)	NAMS/IPWEA	New Zealand/Australia	Asset management reference manual used as standard by numerous water utilities worldwide
BSI (2008)	BSI	UK	Standard used for physical asset management of infrastructure
USEPA (2008)	USEPA	US	General introduction to asset management concepts aimed at water and wastewater utilities based on five key questions
DVGW (2010)	DVGW	Germany	Standard explains how to go from high-level maintenance strategies to short-term intervention programs using expected remaining lifetime. Examples are provided
SSIGE (2012)	SSIGE	Switzerland	General guideline covering short-term and long-term maintenance of water distribution systems. The standard contains an introduction to asset management concepts and a discussion of the data needed to make informed investment decisions as well as the rehabilitation technologies available
Alegre and Coelho (2012)	N/A	Portugal	Overview of infrastructure asset management for water utilities. Document outlines the levels of analysis that must be considered with examples from industry for context
Binggeli et al. (2014)	Kommunale Infrastruktur	Switzerland	A 13-step guideline for municipalities to develop asset management programs for municipalities
AWWA (2014)	AWWA	US	Contains decision trees for which replacement or rehabilitation technology should be used given a number of factors
ISO (2016)	ISO	N/A	Guideline document of managing physicals assets in a water distribution network on an operational level

Infrastructure Management Process

A high level view of the infrastructure management process, on which this work is based, is presented in Adey (2019). It is shown graphically in Fig. 2 using Business Process Model and Notation (Allweyer 2010). More information can be found in Adey (2019).

The work discussed in this paper is a proposed process for Step 3b in Fig. 2., Construct intervention programs for water distribution networks. Neither the ISO 55000-based guidelines in Table 1 nor the work by Adey (2019) provide specific information on determining intervention programs for infrastructure organizations like water utilities. However, in the literature, there exist numerous methodologies for various infrastructure networks. These are based on either prioritization, optimization, or multiobjective optimization (MOO). Prioritization models produce a ranking of infrastructure objects to help infrastructure managers decide where financial resources should be first invested. Optimization models determine the optimal solution, defined by the objective function and the considered constraints (e.g., budget). Examples are given in Table 2, which is composed of the following columns: Reference, i.e., the author and publication; Network, i.e., the type of infrastructure network; Type of objects, i.e., the type of objects considered in the methodology; Method of determination of intervention program, i.e., the model type used to determine the intervention program, e.g., optimization or prioritization; Type of interventions, i.e., the intervention types considered such as relining or replacement; Objective, i.e., the parameter to be maximized or minimized in an optimization model and in prioritization models the parameter used to rank candidate interventions; Algorithmic methods, i.e., the methods used to determine the intervention programs; Constraints, i.e., the resource limitations or minimum performance values, which must be achieved by the determined intervention program; Consideration of risk, i.e., whether risk is used in the methodology and if it is estimated qualitatively or quantitatively; Consideration of synergies from grouping interventions, i.e., whether the benefits of executing interventions on groupings of contiguous pipes simultaneously are considered; and Example, i.e., the specific network used in the publication to demonstrate the methodology. The comparison table focuses on methodologies published in the last 20 years. Older methodologies can be found in the comparison tables of Engelhardt et al. (2000) and Liu et al. (2012).

Table 2. Sampling of published intervention program methodologies

Reference	Type of network	Type of objects	Method of determination of intervention program	Type of interventions	Objective	Algorithmic methods	Constraints	Consideration of risk	Consideration of synergies from grouping interventions	Example
Halhal et al. (1997)	Water	Buried pipes	Multiobjective optimization	Do nothing, clean and line, replace with varying diameters	Maximize benefit and minimize cost	Structured messy genetic algorithm (SMGA)	Budget	No	No	167 pipe network in Morocco
Kleiner et al. (2001) and Kleiner et al. (1998)	Water	Buried pipes	Optimization	Reline, replace, replace with larger pipe	Minimize total lifecycle rehabilitation costs	Dynamic programming	Mass and energy conservation, min. supply pressure	No	No	Three pipe example network for 30 years
Dandy and Engelhardt (2001)	Water	Buried pipes	Optimization	Replace	Minimize total lifecycle costs	Three-step genetic algorithm (GA)	Annual budget	Monetized pipe failure costs	No	216 km pressure zone in Adelaide, AUS
Saegrov (2005) and Le Gauffre et al. (2007)	Water	Buried pipes	Prioritization	Replace	Multiple criteria	Multi criteria decision analysis (MCDA)	No	No	No	2,729 pipes in Reggio Emilia, Italy
Nafi et al. (2006)	Water	Buried pipes	MOO	Do nothing; rehabilitate; replace; reinforce	Minimize pressure and minimize cost	Proportional hazard model (PHM); GA; hydraulic simulation	Min and max pressure values; budget	No	No	12 pipe network for 5 years
Dandy and Engelhardt (2006)	Water	Buried pipes	MOO	Replace	Maximize reliability; minimize cost	GA	Budget	No	No	Time step of 5 years for 20 years for Adelaide, AUS
Halfawy et al. (2008)	Sewer	Buried pipes	MOO	Replacement, structural, semistructural, nonstructural lining	Minimize asset condition, life cycle cost, risk index	GA	Budget, minimum risk and condition values	Risk index	Homogenous groups based on proximity/physical properties	1 year I.P. of a 860 km network from Regina, CAN
Giustolisi and Berardi (2009)	Water	Buried pipes	MOO	Replacement	Minimize replacement cost, pipe break cost, pipe selection preferences, maximize reliability	GA	Budget	Yes, monetized risk of pipe breaks	No	UK network of 166 km
Nafi and Kleiner (2009) and Nafi and Kleiner (2010)	Water	Buried pipes	MOO	Replace	Minimum repair costs and replacement costs	GA, nonhomogenous Poisson	Annual budget and global budget	No	Road interventions, fixed and variable intervention costs, quantity discounts	5-year I.P for 152 cast iron pipes from Southern Ontario
Xu et al. (2013)	Water	Buried pipes	Optimization	Minimum pipe repair and pipe replacement	Optimal pipe replacement timing	Genetic programming	No	No	No	Part of network of Beijing
Zayed and Mohamed (2013)	Water	Buried pipes	Prioritization	Replacement, sliplining, cement lining and cathodic protection, do nothing	Asset priority index	Analytic hierarchy process (AHP)/multiattribute utility theory (MAUT)	No	No	No	Data from two cities
Carey and Lueke (2013)	Road, water, sewer	Buried water /wastewater pipes, roads	Optimization	Do nothing, maintenance, rehabilitation, renewal	Minimize component score	Evolutionary algorithm	Budget, political/exclusionary constraints	Criticality considered based on pipe diameter, road type	Quantity discounts, contiguity discounts, coordination	Pima county, AZ
Baah et al. (2015)	Sewer	Buried sewer pipes	Prioritization	No	Failure risk	Random forests	No	Risk matrix	No
Shahata and Zayed (2016)	Road, water, sewer	Buried water /wastewater pipes, roads	Prioritization	No	Determine risk index of infrastructure corridors	Delphi-AHP	No	Yes	Road, water and sewer segments grouped into corridors
Shin et al. (2016)	Water	Buried pipes	Optimization	Do nothing, renovation, replacement with various diameters	Minimize network life-cycle costs	GA	Annual budget, hydraulic performance criteria	Yes	No	South Korean network with 16.4 km of pipe
Tscheikner-Gratl et al. (2016) and Tscheikner-Gratl (2016)	Water, sewer, road	Buried pipes	Prioritization	No	Overall priority factor	Priority factors	No	Yes as vulnerability	Road, sewer, and water	245 km of sewer
Kielhauser et al. (2017b)	Navigable waterway	Ports, locks, junctions, measurement devices	Optimization	Repair, three types of replacements	Minimize total costs	GA	Budget (annual and global), failure probability limits	Yes	No	Network with 17 objects, 10 time steps
Saad et al. (2017)	Pavement	Road pavement	MOO	Five replacement types	Maximize utility	Linear integer program	Budget, one intervention/planning period	No	No	1,293 road sections for 5-year planning period
Lethanh et al. (2018)	Road	Roads, bridges, tunnels	Optimization	Do nothing, small maintenance, rehabilitation, renovation	Maximize net benefit	Integer linear program	Continuity, budget, max work zone, min distance between work zones	Yes	Fixed and variable costs	Wallis (Switz.) road network
Burkhalter et al. (2018)	Rail	Track, bridges, switches, signals	Optimization	Do nothing, rehabilitation, rebuild, rebuild at night	Maximize net benefit	Integer nonlinear program	Budget	Yes	No	Network with 8 track sections, a bridge, 2 switches and 2 signals
Kielhauser et al. (2017a) and Kielhauser and Adey (2019)	Water, sewer, gas, electricity, road	Roads, proximate buried pipes	Optimization/prioritization	Replace	Maximize risk reduction	Voronoi polygons to determine neighborhood	Budget	Risk failure units	Intervention grouping using voronoi polygons	Town of 30’000 ppl

Currently there is no methodology for determining intervention programs for WDNs, which includes all object categories and defined intervention strategies, quantifies the costs and benefits of executing preventive interventions while considering pertinent information such as the economies of scale gained from executing interventions on contiguous pipes, the intervention activities of other utility networks, and financial and operational constraints. This paper aims to address this gap. The use of the methodology is demonstrated on a real WDN (

pipe length = 4.6 km

) consisting of both facility and pipe objects over a period of 5 years.

Process

The process is shown using Business Process Model and Notation (Allweyer 2010). It consists of five main steps and is illustrated in Fig. 3. The steps are subsequently explained.

Step 1: Model Network

The first step is to define all models that will be used to describe the network and model failures and interventions. This involves cataloging all objects, modeling the deterioration and failure consequences of all objects, as well as the costs and benefits of executing the interventions of interest.

Catalogue Objects

As the goal of the methodology is to determine intervention programs, objects must be described at the level of detail where interventions are planned and executed. The attributes to consider are thus the object type (e.g., pipe, pump), its function (e.g., service line, transmission pipe) and relevant information on material, manufacturer, etc. The last attribute depends on the desired level of detail. If all centrifugal pumps can be treated the same (i.e., pumps from one manufacturer are not substantially different from the rest) than this attribute can be ignored. Additionally, it is important that the models used for estimating deterioration, failure consequence, intervention costs, and benefits be applicable to all objects in the same category.

Once categories have been defined, the objects in the network are catalogued. For facility objects (i.e., any object housed in a facility) this is straightforward. For pipes this involves a grouping algorithm. The algorithm first forms indivisible pipe objects based on physical properties (i.e., age, material, diameter) as illustrated in Figs. 4 and 5, differentiated by color. If an intervention is executed on a pipe object, the entire length of the pipe object will be included in the intervention. This is done to avoid executing interventions on very small objects (i.e., less than 15 m in length) alone. Other pipe cataloguing algorithms are possible. For example, considering the location of valves or road intersections as criteria for separating two pipe objects is useful for improving the estimate of the impact on users following a pipe rupture or the impact on traffic flow during intervention execution. More criteria will result in more pipe objects and an increase in computational complexity. The infrastructure manager must decide whether the increased model accuracy is worth the additional computational complexity.

Fig. 4. Example of initial collection of pipe segments.

Fig. 5. Example pipe objects formed by grouping initial pipe segments using grouping algorithm.

In addition, the modeler must decide whether to catalogue valves, joints, and pipe fittings separately or include them as part of the indivisible pipe objects (i.e., if a pipe object is replaced all corresponding valves, joints, and fittings would also be replaced). Considering such objects separately can substantially increase the computational complexity of the problem as WDNs can have tens of thousands of such objects. Due to the marginal additional cost of such objects compared to the actual pipe, it is recommended to incorporate them into the definition of the pipe object.

Furthermore, cataloguing considers operational status and ownership of objects. Objects that are out of service or not owned by the water utility (e.g., private service lines, private hydrants) can be excluded from the analysis, depending on the utility’s needs.

Model Object Failures

Next, failures are described. A failure is any event that results in a specific object no longer performing its intended function. Thus, it must be object-specific and observable. For example, if a specific pipe breaks or is known to be leaking then it has failed as the function of the pipe is to transport the full quantity of water, while maintaining the desired water quality and pressure. If instead it is simply known that there are leaky pipes in the network or water quality/hydraulic issues then there is no failure as these issues have not yet been linked to specific objects.

Risk,

r_{n}

, as defined as the product of failure probability and monetized failure consequence, is used to quantify failures

r_{i, n} = P_{f, i, n} (n, i, t, X) \cdot C_{f, i, n} (n, i, X)

(1)

r_{n} = \sum_{i = 1}^{I} r_{i, n}

(2)

where

r_{i, n}

= risk related to object

n

of failure mode

i

;

P_{f, i, n}

= failure probability; and

C_{f, i, n}

= consequence of failure. If failure data by mode does not exist than the modeler can treat all failure events of an object category the same.

The failure probability,

P_{f, i, n}

, models the object’s deterioration and estimates the probability of a given object failing within a specified time period. This depends on the object category,

n

; the failure mode,

i

; the object age,

t

; and a vector of influencing covariates,

X

, e.g., zoning, number of previous failures, presence of stray current, and traffic load. Further information on existing deterioration models for water distribution infrastructure focusing on pipes can be found in the scientific literature (Kleiner and Rajani 2001; Rajani and Kleiner 2001; Nishiyama and Filion 2013; Harvey et al. 2014; Scheidegger et al. 2015; Yuan 2016).

Failure consequences,

C_{f, i, n}

, is the expected amount of monetized damages or impacts resulting from a failure and depend on the object category,

n

; the failure mode,

i

; and a vector of influencing covariates,

X

, e.g., object dimension, object material, duration of service interruption, zoning, etc. In order to monetize failure consequences, the modeler must consider the specific resulting impacts, the affected stakeholders, the impact duration, and whether to employ simplifying assumptions. For example, in this publication, failure consequences were estimated assuming objects would fail independently. The possibility of cascading or multiple simultaneous failures were ignored.

Numerous studies have investigated models for monetizing the failures of objects in WDNs or proposed methods for estimating risk with a focus on pipes (Cromwell 2002; Clark et al. 2002; Damodaran et al. 2005; Gaewski and Blaha 2007; Davis and Marlow 2008; Grigg 2013; Kabir et al. 2015; Piratla et al. 2015; Elmasry et al. 2017). The impacts and relevant attributes considered in the evaluation of failure consequences are listed in Table 3.

Table 3. Impact types considered in the evaluation of failure consequence

Impact type	Stakeholder	Pipe/facility	Description	Relevant attributes
Repair cost	Owner	Both	Cost of executing repair intervention	—
Worker safety	Owner	Facility	Impact of exposing workers to dangerous environment	Facility objects: Object type, dimension, failure mode, expected number of occupants, occupancy time, and conditional probability of failure resulting in severe injury/fatality
Reduction in level of service	User	Both	Socioeconomic cost of temporary reduction of LOS	Pipes: Pipe type, zoning, dimension, expected response time, expected repair time, expected pressure drop, expected water demand, and value of lost service
Reduction in level of service	User	Both	Socioeconomic cost of temporary reduction of LOS	Facility objects: Object type, process type, consequences of object failure on process, consequences of process failure on facility, level of redundancy of process, response/repair time, expected pressure drop, expected water demand, and value of lost service
Flooding	Public	Pipe	Socioeconomic cost of water flooding surrounding area	Pipes: Pipe type, zoning, dimension, expected response time, expected repair time, conditional probability of failure resulting in flooding, and value of surrounding property
Reduced traffic capacity	Public	Pipe	Socioeconomic cost of traffic delay due to flooding	Pipes: Pipe type, dimension, expected response time, expected repair time, conditional probability of failure resulting in flooding, average traffic volume, and value of traffic interruption

Model Interventions

Subsequently, the models for calculating costs and benefits of preventive interventions are defined for all object categories. It is assumed that the execution of preventive interventions on facility objects only results in impacts to the owner (i.e., intervention cost) as it is possible to reconfigure the network hydraulics such that the level of service provided to users is not affected. There are many parameters to consider when estimating costs of executing preventive interventions and will vary widely from utility to utility depending on economic conditions. Publications exist on cost estimation models for pipe interventions to guide infrastructure managers (Clark et al. 2002; Zhao and Rajani 2002). The impacts of intervention execution and the relevant attributes are shown in Table 4.

Table 4. Impacts caused by execution of preventive interventions

Impact label	Stakeholder	Pipe/facility	Description	Relevant attributes
Intervention cost (IC)	Owner	Both	Cost of executing intervention	Object type, intervention type, dimension, length, zoning, local site conditions, and cost sharing due to coordination
Reduction in level of service (RLOS)	User	Pipe	Socioeconomic cost of service interruption	Pipe type, zoning, dimension, expected intervention duration, expected pressure drop, value of lost service, and expected water demand
Reduced traffic capacity (RTC)	User	Pipe	Socioeconomic cost of traffic delay/capacity reduction	Pipe type, diameter, length, expected intervention duration, expected intervention duration, intervention activities of other networks, average traffic volume, and value of traffic interruption/reduction

Total intervention costs are then estimated as shown in Eq. (3)

c_{n, k, t} = IC + RLOS + RTC

(3)

where RLOS stands for reduced level of service; and RTC stands for reduced traffic capacity. Intervention benefits are calculated as the reduction of risk associated with intervention execution as shown in Eq. (4)

b_{n, k, t} = F_{n, k} \cdot r_{n, t}

(4)

where

r_{n, t}

= risk related to object

n

in year

t

; and

F_{n, k}

= risk reduction factor of executing intervention

k

on object

n

, which are estimated based on the expected extension of the object’s service life. Thus, replacement interventions completely remove all of an object’s risk whereas a nonstructural relining only reduces one-third of its risk (Table 5). These factors are meant to be illustrative and are not meant to exactly state the effectiveness of the intervention. A process for determining intervention risk reduction factors is an area for future work. Table 5 contains the considered intervention types and default risk reduction factors. The infrastructure manager must decide the possible intervention types per object category. Information on the technical requirements and suitability of different pipe intervention types is readily available (Shahata and Zayed 2012; Infraguide 2003c; AWWA 2014).

Table 5. Risk reduction factors of intervention execution

Intervention type	Pipe/facility	Description	Default risk reduction factor, $F_{n, k}$
Do nothing	Both	Nothing is done	0
Nonstructural	Pipe (100–1,500 mm)	Interventions such as epoxy lining or cement mortar lining, where pipe interior is cleaned and relined with preserving substance. Intervention does not provide structural support	$\frac{1}{3}$
Structural relining	Pipe (50–1,000 mm)	Interventions such as close-fit, cured-in-place, or sliplining, which provide structural support to an existing pipe	$\frac{2}{3}$
Trenchless replacement	Pipe	Trenchless replacement techniques such as pipe bursting, pipe slicing, and pipe extraction where old pipe is replaced without having to dig an open trench	1
Open trench replacement	Pipe	Pipe is replaced with a new one. An open trench is dug along the length of the pipe and the new one installed	1
Minor rehabilitation	Facility	Inexpensive, superficial intervention is executed (e.g., cleaning and repainting of the object)	$\frac{1}{10}$
Major rehabilitation	Facility	Vulnerable/deteriorated parts are replaced with new parts and object is thoroughly cleaned	$\frac{2}{3}$
Replacement	Facility	Object is completely replaced with a new one	1

The consideration of potential cost savings and risk increase due to the intervention planning activities of other networks is considered using coordination factors. The variables

c_{n, k_{c o o r}, t}

and

b_{n, k_{c o o r}, t}

[Eq. (5)] are the resulting costs and benefits of executing preventive interventions on pipes that are located in the planned intervention areas of other networks (e.g., gas, sewer, road)

c_{n, k_{c o o r}, t} = F_{c_{c o o r}, t} \cdot c_{n, k, t} 0 \leq F_{c_{c o o r}, t} \leq 1 b_{n, k_{c o o r}, t} = F_{b_{c o o r}, t} \cdot b_{n, k, t} 1 \leq F_{b_{c o o r}, t} \leq 2 {[n, k, t]}_{c o o r} \in (N \cap N_{z})

(5)

where

F_{c_{c o o r}, t}

and

F_{b_{c o o r}, t}

are coordination factors used to modify the costs and benefits of executing an intervention on a pipe object when other interventions are to be executed, respectively. They are equal to 1 if they are not in coordination areas, otherwise they are modified as shown in Eq. (5) to take into consideration the potential intervention cost savings and increase in risk and subsequent benefit (i.e., risk reduction) due to interventions being executed on other networks. Variable

N

is the set of all objects in the water distribution network, and

N_{Z}

is the set of all objects in other networks where interventions are planned.

Table 6 lists example coordination factors, which are meant to illustrate the methodology and are not a definitive statement about the cost savings of coordinating interventions or the increase in risk due to other networks executing interventions. The burial depth of the pipes of the coordination partner relative to the water pipes must be considered when setting these factors. Sewer pipes for example are typically located beneath water pipes, thus the excavation work for sewer replacement will more likely adversely affect water pipes compared with road maintenance work but simultaneously represents a significant cost savings opportunity for the water utility (e.g., open trench pipe replacement).

Table 6. Example coordination factors

Coordination partner	Cost reduction factor, $F_{c_{c o o r}, t}$	Risk increase factor, $F_{b_{c o o r}, t}$
Stormwater, sewer	0.70	1.20
Gas, electricity	0.80	1.15
Road	0.90	1.05

This approach is similar to the approach used by Nafi and Kleiner (2009) to consider the cost savings of coordination with the road network, and answers the question from the viewpoint of the WDN infrastructure manager, namely how should the planning activities of other networks be accounted for in the cost-benefit analysis? In reality, municipal intervention planning is a multistakeholder negotiation and the WDN infrastructure manager can influence the intervention activities of other networks instead of simply accepting them as a fixed input. Studies have simulated this dynamic decision-making process using serious games (van Riel et al. 2017) and others have proposed methodologies for holistic municipal intervention planning and demonstrated the synergies that can be achieved (Carey and Lueke 2013; Tscheikner-Gratl et al. 2016; Kielhauser and Adey 2019). Although ideal, achieving this integrated level of planning involves overcoming several obstacles such as the organizational restructuring of the utilities that manage municipal infrastructure, and the development of a common financial framework, which will not happen without political will. In the meantime, WDN managers must consider the intervention activities of other networks and their implications for intervention planning.

Step 2: Define Intervention Strategies

Secondly, intervention strategies are defined for all object categories in the network. Strategies are based on scientific literature and the expert knowledge of the utility’s workers and describe how objects should ideally be maintained without regard to constraints. A strategy consists of an object category, optional additional attributes (e.g., zoning, facility type), a triggering condition (e.g., age threshold, risk, failure rate), and the intervention to be executed (e.g., rehabilitation, replacement). For pipes, a rehabilitation intervention refers to nonstructural and structural relining and for facility objects minor or major rehabilitation (Table 5). When a replacement intervention is selected, this means that at least a replacement must be executed and when rehabilitation is selected both rehabilitation and replacement interventions will be considered. There are three types (age, risk, and exclusion) used in this study as described in Table 7.

Table 7. Description of different types of strategies

Strategy type	Description	Example
Age	Use of age thresholds as the triggering condition for a preventive intervention	Replace centrifugal pumps that are older than 25 years
Risk	Use of thresholds of monetized risk as the triggering condition for a preventive intervention. The risk thresholds are defined by the infrastructure manager	Execute at least a rehabilitation intervention on cast iron distribution pipes if risk exceeds 2,000 MU
Exclusion	Exclude certain object types from consideration for preventive interventions. Age and risk thresholds can be used to exclude a subset of an object category	Exclude polyethylene service lines from consideration

Although other attributes could be used for defining the triggering condition (e.g., failure rate), age and risk were chosen because they complement each other well. Age is a simple but useful proxy for object condition as it is easy to determine and known with certainty, whereas risk is more uncertain but considers both failure probability and expected consequences.

As it is possible to define contradictory strategies (e.g., exclude pumps from consideration and replace pumps that are older than 30 years) a logical order must be established in case an object is selected by different strategies for different intervention types. In the provided example the following rules were used: replacement overrules rehabilitation, and replacement or rehabilitation overrules exclusion.

Step 3: Identify Candidate Interventions

The defined intervention strategies are then used to identify candidate interventions, i.e., the preventive interventions to be executed if each object is considered in isolation and constraints are not considered.

Step 4: Perform Cost-Benefit Analysis

The costs and benefits of executing all possible interventions are then determined using the models defined in Step 1. For pipe objects, potential intervention sites must first be identified. In order for the methodology to be applicable at the scale of real-world WDNs, it is necessary to make simplifications to avoid the combinatorial explosion of possible ways of grouping adjacent pipe objects. Thus, three types of possible intervention sites were considered: (1) individual pipe objects (service lines, distribution or transmission pipes); (2) two connected pipe objects (service lines, distribution or transmission pipes); and (3) distribution pipe objects with all connected service lines. This simplification results in a linear increase in potential intervention sites as the network size increases and greatly simplifies the complexity and required computational effort for the optimization. The intervention costs and benefits are then determined for every intervention site–intervention type combination of interest.

Similar to Nafi and Kleiner (2009), economies of scale are considered using quantity discounts on large interventions and reductions in site setup and mobilization costs. Table 8 shows the quantity discount rates used. The discounts for open trench replacement are greater because more contractors are able to execute this intervention type, leading to higher competition compared to other types.

Table 8. Quantity discounts used on pipe interventions

Intervention type	Intervention cost (MU)	Discount rate (%)
Open trench replacement	$\leq 50$	0
Open trench replacement	50–150	10
Open trench replacement	150–250	15
Open trench replacement	250–350	20
Open trench replacement	$\geq 350$	25
All other intervention types	$\leq 150$	0
All other intervention types	$> 150$	10

Savings on site setup and mobilization costs are considered by using an intervention cost model [Eq. (6)] that considers fixed costs (i.e., setup and mobilization costs) and variable costs (i.e., length dependent costs). The sum is multiplied by a coordination factor,

F_{c_{c o o r}}

, which accounts for cost sharing in a coordination area (i.e., coordinated intervention with road, gas, electricity network)

IC = F_{c_{c o o r}} \cdot (C_{fix.} + C_{var.})

(6)

Fixed costs are related to zoning and the intervention type. Variable costs are the costs per unit length to execute the intervention. Thus, the financial advantage of executing an intervention on multiple contiguous pipe objects simultaneously is that fixed costs are incurred only once per intervention site.

In order for interventions to be compared, the associated costs and benefits must be calculated using the same time reference. The approach calculates the present costs and benefits to the utility and public of executing the interventions during the 5-year planning period. The resulting net benefit is then used to determine which interventions should be executed as soon as possible and which ones postponed if there are insufficient financial resources. Future costs and benefits are not explicitly discounted because these rates would be applied to all considered interventions identically and the resulting intervention program would remain unchanged. Thus, the decision of whether to do an intervention or postpone it is not affected by the discount rate. For example, the costs and benefits of a trenchless replacement of Pipe X are assumed to be the same regardless of whether the intervention takes place in 2020 or 2021. Furthermore, if various discount rates were applied to the costs and benefits, the timing of the trenchless pipe replacement of Pipe X would be unaffected because the costs and benefits of all other possible interventions have also been identically discounted.

Step 5: Optimize within Constraints

A two-step approach is used for the optimization. This simplifies the computational effort required and results in a scalable algorithm.

Intervention Optimization

The first step answers the question, given a list of intervention candidates requiring an intervention and the results of the cost-benefit analysis, which intervention types (Table 5) should be executed on the intervention candidates in the next 5 years.

The intervention program is determined using the objective function of maximizing net benefit [Eq. (7)] and constraints

\max Z = \sum_{n = 1}^{N} \sum_{k = 1}^{K} (b_{n, k} - c_{n, k}) \cdot x_{n, k}

(7)

where

x

is a binary variable representing whether or not intervention type

k

has been executed on object

n

.

The object constraint ensures that an object is included only once in the intervention program

\sum_{k = 1}^{K} x_{k} \leq 1, \forall n

(8)

Exclusion constraints remove certain objects from consideration in the intervention program [Eq. (9)], whereas inclusion constraints force certain interventions to be executed on specific objects [Eq. (10)]. Inclusion constraints are used for ensuring intervention candidates are included in the program and for setting priority interventions, which are interventions that the infrastructure manager has determined must occur in a specified year

\sum_{k = 1}^{K} x_{n_{excluded}, k} = 0, \forall n_{excluded}

(9)

\sum_{k = 1}^{K} x_{n_{included}, k} = 1, \forall n_{included}

(10)

The binary optimization is done using the Simplex method with branch and bound and implemented in R Development Core Team (2013) using the lpsolveAPI package version 5.5. 2.0. The first step results in a list of interventions, which ideally should be executed in the 5-year period.

Temporal Allocation

Next, an allocation algorithm is used to determine the intervention year, taking into account an annual budget constraint,

β_{t}

[Eq. (11)]

\sum_{n = 1}^{N} \sum_{k = 1}^{K} i c_{n, k, t} \cdot x_{n, k, t} \leq β_{t}, \forall t

(11)

As certain interventions are time sensitive (i.e., priority and coordination), the allocation is done in steps. First, priority interventions are assigned to the year specified by the infrastructure manager. If there is insufficient budget for these interventions an error message is returned. Next coordination interventions are allocated to the expected year of coordinated intervention execution. If the sum of intervention costs exceeds the annual budget, the allocation is done using dynamic programming modeled on the binary knapsack optimization problem (Toth 1980). The allocation is then repeated for the remaining time insensitive interventions. If the budget is insufficient to execute all interventions then certain time insensitive interventions will be postponed or certain coordination interventions will not take place.

Example

Overview

The example network is a small subnetwork composed of 234 pipe objects and 104 facility objects. The facility serves as a reservoir and a pumping station. The network is summarized in Tables 9 and 10 and shown in Fig. 6. A simple network was chosen to illustrate the methodology so that certain features could be more easily explained such as defining priority interventions and coordination areas. The coordination interventions are with the road network (year 2019), the sewer network (year 2020), and the electricity network (year 2021). The pipes deemed problematic had failed recently several times and thus the infrastructure manager has decided to execute a priority intervention on them in 2019.

Table 9. Network dimensions

Item	Dimension
Buried pipe network	4.6 km
Total reservoir capacity	$23,200 m^{3}$
Number of reservoirs	2
Total pumping capacity	$45 L / s$
Number of pumps	4

Table 10. Network description

Item	Description
Zone	Residential and agricultural
Traffic	Low car traffic load, no truck traffic, no rail network present, no public transit network present
Soil	Clay and silt

Fig. 6. Pipe objects in coordination areas and problematic pipe objects.

Age- and risk-based intervention strategies were defined for all object categories. For all objects a rehabilitation intervention should be executed if the object’s age exceeds the expected service lifetime or if the calculated failure risk exceeds 20 monetary units (MU). In addition, pipes above the age of 50 that are located in coordination areas require a replacement intervention. Furthermore, exclusion strategies were defined for certain object categories (e.g., polyethylene service lines). Example strategies for some of the object categories are shown in Table 11.

Table 11. Example intervention strategies used to select candidate interventions

Object category	Object subcategory	In coordination area	Zoning	If age	If risk	Then execute at least intervention
Distribution pipe	Cast iron	Yes	All	$> 50$	—	Replacement
Distribution pipe	Cast iron	—	All	$> 100$	—	Rehabilitation
Distribution pipe	Cast iron	—	—	—	$> 20 MU$	Rehabilitation
Service line	Polyethylene	—	—	—	—	Do nothing
Hydraulic	Centrifugal pump	N/A	N/A	$> 30$	—	Rehabilitation
Hydraulic	Centrifugal pump	N/A	N/A	—	$> 20 MU$	Rehabilitation
Hydraulic	Centrifugal pump	N/A	N/A	$< 10$	—	Do nothing

The intervention strategies were then used to identify candidate interventions (i.e., possible interventions to execute prior to consideration of constraints.) as shown in Fig. 7. As illustrated in Figs. 4 and 5, 234 pipe objects were formed based on the same material, diameter, and age. These were in turn used to define 495 possible intervention sites using the criteria of Step 4 of the methodology. Four possible intervention types were considered per intervention site, which resulted in a total of 1,980 intervention site–intervention type combinations (see Step 4: Perform cost-benefit analysis). In the pumping station, there were 104 facility objects and three intervention types were considered per object, thus 312 facility object-intervention type combinations were included.

Fig. 7. Candidate interventions selected using intervention strategies.

The intervention costs and benefits were next calculated for all object-intervention combinations. The constraint matrix and objective function were then constructed. In total, the objective function contained 2,318 decision variables and the constraint matrix in the first step of the optimization contained 338 constraints.

Intervention Programs

Two budget scenarios were considered for the example network to illustrate the use of financial constraints in the methodology and the resulting effect on the intervention program. The first scenario had no budget limit and the second had an annual budget limit of 400 MU. The locations and types of interventions executed on pipe objects, as well as the intervention cost breakdown on all objects in both programs are shown in Figs. 8–11. The resulting intervention programs are summarized in Table 12.

Fig. 8. Pipe intervention types in time periods 1–5 with no budget limit.

Fig. 9. Intervention costs for objects with no budget limit.

Fig. 10. Pipe intervention types in time periods 1–5 with budget limit ( $400 MU / y e a r$ ).

Fig. 11. Intervention costs for objects with budget limit ( $400 MU / y e a r$ ).

Table 12. Intervention program summary table

Parameter	Intervention program
Parameter	1	2
Annual budget limit (MU)	No limit	400.00
Planning period (years)	5	5
All priority interventions executed (yes/no)	Yes	Yes
All candidate interventions executed (yes/no)	Yes	No
Costs (intervention execution, MU)	2,513.20	1,930.60
Costs (traffic disruption, MU)	71.10	2.50
Costs (service interruption, MU)	145.50	111.50
Risk reduction (MU)	2,849.90	2,327.90
Net benefit (MU)	120.10	283.30
Number of objects: Facility (building)	5	5
Number of objects: Facility (electrical supply)	8	8
Number of objects: Facility (hydraulics)	40	40
Number of objects: Facility (IT/surveillance)	10	10
Number of objects: Facility (measurements)	17	17
Number of objects: Facility (treatment)	2	2
Number of objects: Service lines	15	15
Number of objects: Distribution pipes	14	10
Number of objects: Transmission pipes	1	0
Length (km): Service lines	0.13	0.13
Length (km): Distribution pipes	1.68	0.83
Length (km): Transmission pipes	0	0
Figures	Figs. 8 and 9	Figs. 10 and 11

Discussion of Results

In Intervention program 1 (no budget limit), all interventions are executed in 2019 except those in coordination with other infrastructure networks taking place in other years, whereas in Intervention program 2 (budget limit) spending is spread across all 5 years. The intervention costs of Intervention program 2 add up to 96.5% of the 5-year allocated budget of 2,000 MU. As expected, the intervention costs and risk reduction of Intervention program 2 were respectively lower at 76.8% and 81.7% those of Intervention program 1; however, the total net benefit of Intervention program 2 was significantly higher (283.3 MU compared to 120.1 MU). This is due to the annual budget limit of the second scenario, which resulted in candidate interventions with negative net benefit being postponed to the next 5-year period. The interventions executed on facility objects were the same in both intervention programs, and both intervention cost breakdowns (Figs. 9 and 11) show that pipe interventions require significantly more financial resources than interventions on facility objects. Regarding coordination interventions, the objects in the 2020 and 2021 coordination areas were selected for open trench replacement but those in the 2019 coordination area (Figs. 6, 8, and 10) were not as these pipes had recently been installed and had a low level of risk and thus did not meet the criteria for selection as an intervention candidate. In both programs, a priority intervention (structural relining) was executed on the pipe objects providing an inadequate level of service in 2019 as specified.

Discussion

The example demonstrated the use of an intervention planning methodology, which allows infrastructure managers to consider both facility and pipe objects, cost savings from economies of scale, the intervention planning activities of other networks, and gives infrastructure managers the ability to set budget limits at each time step. Expert knowledge of the network is incorporated in this methodology in the defined intervention strategies, which are used to select candidate interventions. These strategies can be modified to make comparisons and communicate to stakeholders the budget needs of maintaining the required level of service.

There are several simplifications and assumptions to discuss. For example, the gradual loss of hydraulic capacity in the network was not explicitly modeled. Instead the methodology relies on intervention strategies defined by the infrastructure manager to ensure an adequate level of service. More precisely, the infrastructure manager must decide on the necessary age or risk-based intervention strategies that will ensure minimum hydraulic performance thresholds for all relevant network objects and the interventions that should be executed to rectify inadequate level of service. Detailed studies, experience, and expert knowledge is necessary to define these strategies. Other researchers have coupled hydraulic models with a multiobjective optimization model to ensure that selected network configurations met minimum distribution pressures in all parts of the network. Kleiner et al. (2001) determined optimal replacement timings of pipes by minimizing total lifetime cost of pipes. The example network consisted of 12 pipes and the optimization algorithm was linked to a hydraulic model to ensure that nodal residual pressure limits were upheld. Tolson et al. (2004) coupled a hydraulic network model with a genetic algorithm–based optimization algorithm to estimate the reliability of a 14-pipe example network to meet minimum distribution pressures. Nafi et al. (2006) presented a methodology for determining the optimal replacement schedules of pipes by combining a genetic algorithm with a hydraulic model of an example network composed of 12 pipes. One significant drawback to this approach is the significant computational effort required to run a hydraulic simulation for each tested network configuration. If desired, a detailed hydraulic model could be used within the proposed framework to improve estimates of the impact of failures or intervention execution on service or better estimate the gradual decrease in hydraulic capacity; however, the computational cost is prohibitive for large networks. Thus, the approach is taken to have infrastructure managers define criteria for triggering interventions on objects rather than coupling a hydraulic network solver with the optimization algorithm.

Another simplifying assumption is that multiple or cascading failures cannot occur. This is done to simplify the risk calculation, as estimating such additional consequences is complicated. For example, multiple pipe breaks in the same area might cause a sinkhole to develop, will increase the chance of contaminants entering the network, and will have a much larger impact on traffic circulation than if the failures occurred separately. Numerous local media sources have reported on the consequences of simultaneous water main failures. van der Kleut (2012) reported two water main failures resulted in a sinkhole forming in Redwood City, California, and Richards (2017) reported that a city-wide boil water advisory was issued in Kemp, Texas following four simultaneous water main breaks. The consideration of multiple object failures in the risk calculation was beyond the scope of this work.

The benefit of executing interventions is defined as the amount of risk reduced, which is calculated using several simplifications. Firstly, risk reduction factors are defined per intervention type based on the expected extension of the object’s service life. In reality, many additional parameters may influence the benefit of executing an intervention such as the quality of the installation, or the materials used to replace the object. These influencing parameters and others are not considered in the intervention benefit factors listed in Table 5. Furthermore, the risk reduction calculation is simplified by only calculating the short-term (5 year) risk reduction in the year of intervention execution instead of using the rest of the object’s service life. The uncertainty of the object’s service life makes this inherently difficult as this is a combination of the technical service life and the infrastructure manager’s level of risk aversion. In addition, the purpose of the resulting cost-benefit analysis is to determine which interventions should be executed in the next 5 years and which ones postponed if the defined annual budget limitation is insufficient for all planned interventions. Thus, it is important that the chosen time period for benefit estimation be the same for all object categories.

The methodology uses an algorithm to group pipes, as they appear in a GIS database, into pipe objects to reduce the computational complexity of the problem (i.e., fewer decision variables). Contiguous pipe objects are then further grouped to form potential intervention sites to consider the economies of scale gained by executing interventions simultaneously. In order to avoid a combinatorial explosion, only certain combinations are considered. This has the advantage of making the algorithm scalable and applicable to larger networks as the number of possible intervention sites increases linearly with the size of the network.

The potential cost savings and increase in risk due to other networks executing interventions has been estimated in this methodology using coordination factors. The true cost savings and risk increase due to intervention activities of other networks depends on the depth of the pipes in both networks, the lateral proximity of the pipes to one another, and the specific intervention type that the other network is planning (e.g., if sewer utility is executing a trenchless intervention instead of an open trench intervention, the cost savings of coordination may be minimal). The factors in Table 6 do not account for all these influencing parameters and are an approximation used to illustrate the methodology.

Many of the inputs used in the methodology are uncertain or not known by water utilities and require time and effort to determine. Despite the use of optimization in the methodology, generated solutions cannot be considered optimal without investigating this underlying input uncertainty. Nevertheless, this methodology, implemented in a decision support software, is a step toward optimal intervention planning. A detailed sensitivity analysis will allow infrastructure managers to identify inputs with a significant influence on the generated intervention program and subsequent studies can be done to obtain improved input estimates. For example, the estimate of intervention effectiveness could be greatly improved by recording executed interventions in a database with all relevant parameters and coupling this data with select performance indicators such as failure rate so that the intervention could be periodically evaluated.

This methodology is designed for short-term intervention planning (i.e., maximum of 5 years). This is sufficient time for project engineers to prepare the necessary steps for the planning and execution of these interventions. As the planning period increases, so does uncertainty related to the required level of service, the budgetary limits, the risk estimations, and the needed computation time. The advantage of longer planning periods is that infrastructure managers can potentially identify waves of required interventions early enough to spread the required spending over a longer period of time. This methodology could be applied to longer planning periods by taking the intervention program of the first 5 years and using it as an input to the subsequent 5-year period. The resulting intervention program will be useful to an extent; however, the associated increase in uncertainty and computational effort must be carefully considered. The next step in this study is to demonstrate the scalability of the methodology to a city-sized network composed of numerous facilities and subnetworks.

Conclusions

The presented methodology aids infrastructure managers with the planning and execution of interventions for WDNs by addressing many of the difficulties highlighted in the introduction. Past publications have focused on intervention planning for one object type (e.g., pipes) and for one aspect of service (hydraulics, economics, etc.). These studies have made valuable contributions but do not address the problem of intervention planning for WDNs at the municipal scale. This problem consists of developing intervention programs for large networks composed of a myriad of different object types while respecting all relevant constraints and maintaining a high level of decision transparency to stakeholders.

This methodology is a useful aid for infrastructure managers faced with this problem for the following reasons. Infrastructure managers can formalize their expert knowledge by defining intervention strategies needed to maintain an adequate level of service. The associated financial requirements can then be determined and the results communicated to stakeholders under different budget scenarios. Formalizing expert knowledge and recording executed interventions in an accessible format will have the additional benefit of reducing the utility’s knowledge loss following worker retirement and improves the learning opportunity needed in skilled intuitive decision-making. Furthermore, a detailed cost-benefit analysis allows the infrastructure manager to incorporate important information, such as the intervention planning activities of other networks and cost-saving considerations related to economies of scale, and directly compare the utility of executing interventions on different objects. Although demonstrated on a relatively small network, future work will demonstrate the applicability of the methodology to larger WDNs. As input uncertainty is not quantified, it is not possible to state that the resulting intervention programs are optimal (i.e., the execution of these interventions will result in the lowest life-cycle costs); however, it is a step toward determining optimality and subsequent studies (e.g., sensitivity analysis) can be performed to identify important model inputs and improve the accuracy of the estimates to ultimately develop optimal intervention programs.

Data Availability Statement

Some or all data, models, or code generated or used during the study are proprietary or confidential in nature and may only be provided with restrictions (e.g., anonymized data) and the written consent of the utility.

Acknowledgments

The authors thank the members of the water utility of Geneva (SIG) for their generous financial and logistical contributions, which have made this research possible. In particular, we would like to thank Mr. Yves de Siebenthal, Mr. Gérard Luyet, Dr. Stéphan Ramseier, and Mr. Eric Guillaume for their continued support and collaboration.

References

Adey, B. T. 2019. “A road infrastructure asset management process: Gains in efficiency and effectiveness.” Infrastruct. Asset Manage. 6 (1): 2–14. https://doi.org/10.1680/jinam.17.00018.

Abstract

Introduction

Difficulty of Planning and Executing Interventions

Guidelines for Intervention Planning

Infrastructure Management Process

Process

Step 1: Model Network

Catalogue Objects

Model Object Failures

Model Interventions

Step 2: Define Intervention Strategies

Step 3: Identify Candidate Interventions

Step 4: Perform Cost-Benefit Analysis

Step 5: Optimize within Constraints

Intervention Optimization

Temporal Allocation

Example

Overview

Intervention Programs

Discussion of Results

Discussion

Conclusions

Data Availability Statement

Acknowledgments

References

Information

Published In

Copyright

History

Authors

Affiliations

Metrics

Citations

Download citation

Cited by

Figures

Other

Share

Copy the content Link

Share with email

Share

Request Username

Create a new account

Change Password

Password Changed Successfully

Verify Phone

Congrats!