You are on page 1of 8

Reliability Engineering and System Safety 100 (2012) 6774

Contents lists available at SciVerse ScienceDirect

Reliability Engineering and System Safety


journal homepage: www.elsevier.com/locate/ress

Risk-based prioritization and its application to inspection of valves in the water sector
David R. Marlow n, David J. Beale, John S. Mashford
Commonwealth Scientic and Industrial Research Organization (CSIRO), Land and Water, Victoria, Australia

a r t i c l e i n f o
Article history: Received 27 June 2011 Received in revised form 12 December 2011 Accepted 16 December 2011 Available online 28 December 2011 Keywords: Valves Water pipeline Risk-based inspection Analytical hierarchy process

abstract
Isolation valves facilitate the effective operation and maintenance of water supply networks, but their sheer number presents a signicant asset management challenge. If left unmanaged, valve reliability issues can become widespread. Inspections provide a means of increasing reliability, but a survey of industry practices indicated that some utilities did not have such a program in place. To improve asset management and reduce business risk exposure, such utilities need an effective means of commencing inspection programs. From a theoretical perspective, risk concepts provide a means of optimizing maintenance effort. However, in the face of poor data on reliability or condition, pragmatic approaches to risk-based prioritization are needed. One such approach, risk indexing, is considered in this paper. Background on the research is presented, including the application of risk-based inspection concepts within the water sector. The development of a risk indexing scheme is then investigated, drawing on two industry workshops in which the analytical hierarchy process was used to set relative weights. It is concluded that risk indexing provides the basis for a rational prioritization process in the absence of data on valve reliability or condition. Crown Copyright & 2012 Published by Elsevier Ltd. All rights reserved.

1. Introduction Water utilities are today faced with the challenge of delivering required levels of service with limited funds in the face of an ageing asset stock and increasingly stringent standards and customer expectations [1]. One aspect of this is to manage isolation valves. Isolation valves are installed to facilitate the operation and maintenance of a water network, but present signicant difculties because of the size of the asset stock. For example, data reported by the American Water Works Association indicates that 338 water utilities operate an aggregated 202,158 miles of pipe with 2,776,059 gate and 150,531 buttery valves. This equates to ca. 15 valves per mile of water main, with an average of approximately 9,000 valves per utility [25]. The challenge of managing these assets is increased by the spatial distribution of the valves and the heterogeneity of the asset stock [6]. Like all mechanical assets, the physical state or condition of valves changes over time as components deteriorate. The rate of deterioration is asset and context-specic, being inuenced by factors such as the type of valve, the quality of the installation, the external and internal environment the valve is exposed to and the operating and maintenance regime. As valve condition deteriorates, failures due to fatigue, corrosion and wear-out start to occur. There is thus a driver
n Correspondence to: Sustainable Infrastructure Asset Management, CSIRO Land and Water, P.O. Box 56, Highett, VIC 3190, Australia. Tel.: 61 3 9252 6614; fax: 61 3 9252 6244. E-mail address: david.marlow@csiro.au (D.R. Marlow).

for water utilities to understand valve condition at the individual asset level and across the network. Information on asset condition is collected by undertaking inspections, which is a key input into physical asset management [7,8]. Physical asset management aims to provide sustained services at a cost and level of risk that is acceptable to customers and the broader community [1,7,8]. In 2006, the Water Services Association of Australia (WSAA) held a workshop to formulate a program of research to address gaps in asset management capacity. One priority identied was the need to develop specic guidance on condition assessment and inspection processes for water main valves. The Water Research Foundation (Water RF) in the USA and WSAA subsequently co-sponsored a major research project aimed at meeting this need [6,9]. This paper draws from this research project, focusing specically on the challenge of prioritizing inspections within utilities that have poor data on valve condition and reliability. For example, one large utility who provided information for a case study had not undertaken maintenance on any isolation valves for over 15 years, and did not have a comprehensive database for these assets. Issues with data included that the inventory of valves was not complete, assessment of asset importance and risk had not been made, condition of valves was unknown and maintenance schedules were not allocated [6]. The context and focus of the work presented herein is signicantly different to that of related research previously described in the literature, including in this journal. For example, while Walski discussed the role that valves play in maintaining water distribution system reliability and emphasized the need to manage valves

0951-8320/$ - see front matter Crown Copyright & 2012 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.ress.2011.12.014

68

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

effectively [10,11], the practicalities of dealing with a backlog in maintenance was not considered. Similarly, risk-based approaches to the scheduling and prioritization of maintenance tasks, including inspections, have generally been considered within the context of relatively good estimates of reliability and costs [e.g. 12,13]. Various papers address risk-based inspection for assets within safety critical systems (e.g. reactors) [14,15], applications that imply a level of rigor that would not be justied for valving within a water supply system. In other studies, the systems analyzed have been simple enough to allow explicit analysis of all valves individually [13], which is not possible across a water supply network in the absence of good data. Prioritization procedures have also been developed that involve a detailed consideration of specic assets. For example, Stewart [16] presents a risk ranking procedure for bridges that uses the expected cost of failure as a risk measure. Other procedures have been developed that are applicable across a system of assets. For example, Brito and de Almeida developed a multi-attribute decision model for ranking sections of gas pipelines in terms of risk [17]. Their assessment procedure included a process to incorporate subjective perceptions of decision makers regarding risk, an approach that is paralleled in this study. However, their procedure again implies a more detailed consideration of assets than is possible given the context of this paper. In summary, while the literature provides useful insights into riskbased prioritization procedures, the focus and intended application of existing research is signicantly different to the context of the research described herein. More specically, an approach was needed that would allow utilities to commence inspection programs in the absence of data on asset reliability or condition. With this focus in mind, the remainder of this paper is structured as follows. Background information on the research context is rst given, followed by a mathematical representation of risk-based inspection and its practical interpretation. Pragmatic approaches to risk-based prioritization are then outlined, and the development of a risk indexing scheme investigated. The approach taken uses expert opinion mobilized through the application of the analytical hierarchy process. A synthesis of this approach is provided in the conclusion.

that some utilities have scaled back or abandoned their valve inspection and exercising programs. Furthermore, of 25 utilities who answered a web-based survey, seven indicated there was no routine inspection program in place for valves [6]. If left unmanaged in this way, reliability issues can become widespread across a network [20,21]. It can be anticipated that the more unreliable the valve stock, the higher the probability that additional effort will need to be expended to isolate any given burst or undertake maintenance on a pipeline. Given the additional costs and consequences involved, there is thus a driver for utilities to undertake inspection programs to improve overall reliability. When commencing such a program, it is highly desirable for utilities to target initial inspections where they are likely to provide the most valuable information and/or mitigate the greatest risks. Unfortunately, the number of valves, their spatial distribution and the differences in environmental and operational factors across a network all mean that it can be challenging to design an inspection program to achieve this end.

3. Theoretical basis of risk-based prioritization As part of effective asset management, a water utility must balance the cost of performing inspection and other maintenance activities against the cost and impact of not performing them [6,7]. Such decisions are often facilitated by adopting a risk-based framework [17,2224]. In asset management, risk is dened as the product of the likelihood (or probability) of an asset failure and the severity of the failure consequence [2527]. Risk dened in this way has been used to develop inspection programs for a range of purposes [2831]. More generally, risk-based inspection is a standard maintenance practice that focuses on failure modes initiated by material deterioration and controlled primarily through inspections [25,32]. In the context of water sector asset management, the basic elements that need to be addressed in the development of such a program are [7]: 1. The assessment of the risk introduced by potential asset failures. 2. The identication of the degradation mechanisms that can lead to failures. 3. The selection of effective inspection techniques that can detect the progression of degradation mechanisms. 4. The development of an optimized inspection plan. The later stage is important because inspection and subsequent repair/replacement of assets is carried out subject to budgetary constraints and the level of such activities inuences the total cost of operating the network. 3.1. Risk as a basis for optimization Ideally, an inspection program would be optimized to give the greatest benet relative to costs across the entire population of assets. The use of risk concepts for optimizing inspection programs can be justied from a theoretical basis if a number of simplications are made. To this end, let n be the number of assets and, for each i 1, y, n, let xi be a 01 valued decision variable which indicates whether or not asset i is to be inspected. Suppose that we have reasonable estimates of the probability pi of failure of asset i; the probability qi of failure of asset i given that it has been inspected; the consequence ci of failure of asset i; the cost di of inspecting asset i; the cost ei of repairing asset i and the expected benet bi of repairing asset i for each asset (i 1, y, n). Since we are assuming only probabilistic knowledge about some of the variables describing the system, we consider a sequence of trials over an ensemble of

2. Research context Isolation valves are generally only needed during specic events such as to isolate a pipe burst or undertake pipeline maintenance. As such, the failure of a valve may only become evident when there is an operational demand for its use. This increases the time taken to undertake planned and reactive maintenance, thereby increasing costs. In the case of a burst, failure of a valve to operate on demand can also increase failure consequences by prolonging social and commercial disruption, increasing damage to property and other assets and increasing the number of customers who lose water. Valves generally fail due to the corrosion of internal components, deterioration of seals and the internal build up of deposits [18]. A pragmatic approach to valve inspection is thus to undertake an inservice or functional test (commonly referred to as valve exercising) to determine if the valve is in an operational state [10]. Ideally, this involves operating the valve stop-to-stop, though some utilities only practice partial closure due to operational risks [6]. Periodic exercising of valves also ensures moving components remain free, so valve exercising also increases reliability [6,19]. 2.1. Industry practice A review of valve inspection practices was undertaken as part of the research presented in this paper. Information obtained from industry partners in the UK, Australia and the USA indicated

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

69

systems and let f i be a 01 variable that indicates whether asset i fails in trial j. In an ideal word, it can be shown that: Expected net benet of adopting an inspection strategy 0 2 31 n m X X 1 j @bi di ei xi ci 4lim as m-1 fxi g 1 f 5A mj1 i i1 If an asset is not inspected then it will fail with a long run frequency equal to its probability of failure. Therefore: xi 0 ) lim as m-1
m 1X j f pi , 8i 1, . . ., n: mj1 i

In practice, the data required for this optimization is collectable but, as noted above, can be expected to be absent for a signicant proportion of utilities. A more pragmatic application of risk concepts is therefore required to rank assets, as discussed in the next section. For completeness, prioritization of isolation valves, especially within the distribution part of the network, can also be undertaken using path dependant reliability analysis techniques whereby the impact of valve failures is assessed in terms of the additional number of customers that will be effected by a burst (due to the need to operate additional valves). Such techniques also allow water utilities to optimize the location and conguration of network valves [6,11].

In contrast, if an asset is inspected then it will fail with a long run frequency equal to its conditional probability of failure given that it has been inspected. Therefore: xi 1 ) lim as m-1
m 1X j f qi , mj1 i

4. Risk-based prioritization procedures Utility staff generally have signicant levels of knowledge with respect to factors that inuence asset-related risks. Prioritization can thus be undertaken by mobilizing relevant operational experience. A common technique applied to facilitate this is the application of what can be referred to as rapid ranking [7,9]. This qualitative approach to prioritization is used in various sectors [25,32], and involves undertaking a desktop study whereby assets are allocated a failure likelihood and consequence rank. This allows an analyst to obtain a broad prioritization of risks quickly, but it is difcult to take into account a comprehensive range of risk factors especially when factors are independent of one another. An alternative approach is to apply index-based risk assessment techniques. Risk indexes provide a measure of risk through simple scoring procedures, rather than complex risk calculations [33]. In comparison to rapid ranking, the scoring schemes used are more explicit in their treatment of individual risk factors in that the risk score are increased or decreased according to the presence or absence of individual factors. Weights are also allocated to reect the relative importance of the risk factors. In constructing an indexing scheme, the risk problem is rst decomposed into a risk-hierarchy that incorporates factors that inuence the level of risk, as indicated in Fig. 1 [7]. For convenience, risk factors can be grouped into categories and the hierarchy constructed with multiple levels. Mathematically, such a hierarchy can be represented by an acyclic directed graph H (N,A) where N is the set of nodes and A C N N is the set of arcs (or edges) in the graph. The roots of a hierarchy H (N, A) are the elements of the set R(H), as illustrated in Eq. (9). The roots are the nodes which have no arcs going into them. RH fx A N : $ (y A Ny, x A Ag, 9

8i 1, . . ., n:

From (2) and (3) it follows that: lim as m-1


m 1X j f pi 1xi qi xi , mj1 i

8i 1, . . ., n:

Thus, the expected net benet of adopting the inspection strategy fxi g
n X

bi di ei xi ci pi 1xi ci qi xi

n X

bi di ei r i si xi

n X i1

ri ,

i1

i1

5 where, ri pici, ; and si qici.. ri is called the risk of failure of asset i and is the product of the assets probability of failure and its consequence of failure. si is the risk of failure of asset i given that it has been inspected. The cost of carrying out inspections and associated repairs, not counting repairs associated with failures, is given by: Cost
n X i1

di ei xi :

Suppose that we have an inspection budget B and we are seeking an inspection strategy {xi} which maximizes the expected net benet such that the inspection cost does not exceed the inspection budget. This is equivalent to solving the following optimization problem: Max
n X i1

bi di ei r i si xi

such that
n X i1

di eixi r B:

The leaves of a hierarchy H (N,A) are the elements of the set L(H), as illustrated in Eq. (10). The leaves are the nodes, which have no arcs leaving them. LH fx A N : $ (y A Nx, y A Ag, 10

In the special case where di ei is the same (or approximately the same) for each asset i with di ei g, bi is the same for all i with bi b, b Z g and the risks si are negligible, then the optimal solution to this problem is obtained by listing all the assets in order of risk and choosing the top k of them to inspect where: k B=g greatest integer less than or equal to B=g : 8

Problem definition

In practice, the condition that di ei is the same (or approximately the same) for each asset applies for valves that are of a similar size and operational context. It can then be concluded that the optimal inspection strategy is to rank such assets in order of risk and to inspect those for which the benet of inspection outweighs the cost of inspection plus repair. If one assumes that the valve should operate on demand, the later criterion is generally met.

Likelihood
Factor 1 Factor 2

Consequence
Factor 3 Factor 4

Characterisitics

Characterisitics

Fig. 1. Decomposition of risk hierarchy.

70

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

Next, a stratied rooted tree is dened to be a quadruple H (x0,N,A,S), where (N,A) is a directed graph, R((N,A)) {x0} and there exists a partition {Li: i 0, y, k} of N such that: 8x, y A N x, y A A ) (i A f0, . . ., k1gx A Li &y A Li 1 : 11

Thus, a stratied rooted tree is a hierarchy which has precisely one root and which is divided up into a number of levels such that arcs only go from a level to the next highest level. In risk indexing, two nite stratied rooted trees are constructed, one for likelihood of failure and one for consequence of failure. Weights are assigned to the arcs of the trees (risk categories and factors) and values are given to the leaf nodes (characteristics). Thus we are given stratied rooted trees Hi (xi,Ni,Ai,Si) for i 1, 2; mappings wi: Ai-(0,1) and mappings Ci: X L(Hi)-R where X is the set of all assets. Given this information, we can calculate the risk index of each asset in X as follows. Compute the valuation functions Vi: X Ni-R recursively by Eqs. (12) and (13). V i x, x C i x, x, 8x A X , x A LHi , 8x A X , x A Ni $ LHi : 13 Then compute the risk function R: X-R following Eq. (14) Rx V 1 x, x1 V 2 x, x2 : The function R gives the risk R(x) of each asset xAX. 14 12 X fwi x, yV i x, y : x, y A Ai g, V i x, x

Size Number of customers impacted Type of customer impacted Potential for 3rd party losses Public relations Cost Potential access issues Other 0 0.2 0.4
AUS

0.6
USA

0.8

Normalized Score

Fig. 3. Relative importance of factors inuencing failure consequence.

5. Development of a risk indexing scheme To investigate the application of the risk indexing approach, a scheme was developed for large diameter (nominally 300 mm and above) valves using input from industry professionals gained in two workshops held in Australia. It should be noted that the intention was not to develop a generic scheme, but rather to determine the practicality or otherwise of applying risk indexing to valve prioritization. To streamline the workshops, the researchers developed an example risk hierarchy, selecting risk factors and characteristics based on information gained through a webbased survey, as illustrated in Figs. 2 and 3, supplemented by a review of available literature and industry practice. Examples of risk factors and characteristics included in the scheme are shown in Table 1. A full list of risk factors is provided later in Section 5.3. 5.1. Application of the analytical hierarchy process Given the risk hierarchy had been developed in advance, the focus of the workshops was to set the relative importance of the risk factors. In the absence of suitable data on condition and reliability, weighting of risk factors must be undertaken using expert opinion. Weights can be set relative to one another by
Appurtenance design Appurtenance operation Construction material Installation quality Maintenance procedures Operating environment Water quality Pipe failure Other 0 0.2 0.4
AUS

comparison. People generally nd it difcult to allocate relative weights to multiple factors simultaneously, but nd it relatively easy to rank the importance of just two factors. As such, the weighting process was facilitated by applying the Analytic Hierarchy Process (AHP). The AHP is a structured technique for organizing and analyzing complex decisions and provides a method for weighting criteria, aggregating scores and assessing the consistency of judgments [34]. In application, pairwise comparisons are undertaken at each level within the risk hierarchy according to the scale described in Table 2 [35,36]. The weights for each factor are then determined through calculation of normalized eigenvectors [34]. A key aspect of AHP is that it provides a check of the consistency of the expert judgements. More specically, the inconsistency of each pairwise comparison matrix is calculated and compared with an upper threshold of 10%. In practice, this means a small degree of inconsistency in the judgements of individuals can be tolerated. In this study, each pairwise comparison was facilitated by asking participants to consider which valve would be prioritized over another if only a limited piece of information was available for each. For example, Valve A is an old specication that would not be used today whereas Valve B is in a vulnerable location subject to vandalism. With such guidance in mind, each individual allocated weights in accordance with their own experience using the standardized scoring scheme shown in Table 2. Since it is based on experience and judgement, the weighting procedure does not quantify the contribution that risk factors make to overall risk; rather it quanties the opinion of this contribution. To achieve desired outcomes, the weights must be allocated by individuals with an appropriate knowledge of valve maintenance issues. Discussion of risk factors in light of business needs and valve maintenance issues can help achieve consensus on the relative importance of factors. Given that workshop are relatively time consuming, practitioners may be tempted to allocate weights to factors directly. However, the AHP provides both a systematic framework for capturing knowledge and, as noted above, allows the consistency of expert opinion to be assessed. 5.2. Results In summary, the workshops provided 25 different set of weights, 14 of which met the consistency requirement of the AHP (consistency ratio o 10%). Additional work would ideally be undertaken to reach consensus in the scoring but this was not possible given time and resource constraints. Figs. 4 and 5 graphically illustrate the range of weights derived during the workshops. Different means of combining the weights were identied [35,37]. Given that the workshops were one off events, the simplest of these was taken; the geometric average was calculated and weights then normalized [37]. The calculation of geometric average requires all individuals to have ranked each

0.6
USA

0.8

Normalized Score

Fig. 2. Relative importance of factors inuencing service life.

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

71

Table 1 Extract of the risk hierarchy. Likelihood 1st level factor Potential for deterioration 2nd level factor Age (asset) Characteristics New or young asset Ageing asset Old asset Not aggressive, not vulnerable Moderately aggressive Aggressive and vulnerable Not aggressive, not vulnerable Moderately aggressive Aggressive and vulnerable Consequence 1st level factor Response 2nd level Factor Isolation time Characteristics o1 h 12 h 42 h o 100 m 1001000 m Uncertain or 4 1000 m o 100 1001000 100010,000 4 10,000 None 12 42

External environment

Distance to back up assets

Internal environment

Service provision

Customers effected

Key customers

Table 2 Scoring for the analytical hierarchy process. Evaluation Factor 1 and Factor 2 equally important Factor 1 is slightly more important than Factor 2 Factor 1 is moderately more important than Factor 2 Factor1 is strongly more important than Factor 2 Factor 1 is absolutely more important than Factor 2 Score 1 3 5 7 9

Note: Where the judgement is reversed (i.e., Factor 2 is more important), the score is allocated as 1/3, 1/5 etc.

Fig. 5. Weighted likelihood factors.

Fig. 4. Weighted consequence factors.

item in the same sense. For example, all participants need to rank Factor 1 as more important than Factor 2 or vice versa, though the magnitude of the weight could vary. Tables 3 and 4 provides the geometric mean weights allocated through this process, along with the relative weight of the factors at both the rst and second level. In terms of the theoretical formulation of risk indexing given in Section 4, the likelihood of failure hierarchy has three levels L0, L1 and L2 where: L0 {Likelihood}, L1 {Asset Vulnerability, Operation & Maintenance, Asset State, Potential for Deterioration}, L2 {Overall Design, Emplacement, Exercising/use, Maintenance, Structural Condition, Performance, Age, External Environment, Internal Environment}. The arc set A1 in the likelihood of failure hierarchy can be determined from Table 3. For example (Likelihood, Asset Vulnerability)AA1 and (Asset Vulnerability, Overall Design)AA1.

The weight mapping w1: A1-[0,1] for the likelihood of failure hierarchy can also be read off from Table 3. For example w1(Likelihood, Asset Vulnerability) 0.0799 and w1(Asset Vulnerability, Overall Design) 0.0628. Similarly, the structure of the consequence of failure hierarchy H2, its arc set A2 and weight mapping w2: A2-R can be determined from Table 4. The relative importance of the characteristics (i.e. the leaf nodes) must also be dened in some way. To avoid the problem of double counting, these were not weighted explicitly. Instead, for each risk factor a score of 1 was allocated to the characteristic associated with the highest risk, 0 to the lowest risk characteristic, with the remaining scored proportionally. For example, from Table 1, the risk factor isolation time is considered to inuence consequence of failure. The measurable or assessable characteristic used to reect this risk factor was an estimate of time to isolate a burst, banded as o 1 h, 12 h and 4 2 h. In this case, the scores of 0, 0.5 and 1 were allocated to the characteristics o 1 h, 12 h and 4 2 h, respectively. 5.3. Practical application of the index scheme In application, data for each relevant asset characteristic needs to be collated to allow the risk score to be assigned. As a simple example, consider the two valves summarized in Table 5. Having collated the characteristics for each valve, the appropriate weighted score is allocated according to the risk indexing scheme, as per the examples shown in Tables 6 and 7. Since we are interested in the ordinal rank of the valves, prioritization can be assessed in terms of likelihood, consequence and overall risk scores. For this example, Valve 1 has characteristics that mean it is likely to be in poor condition (and by inference have a higher likelihood of failure), whereas Valve 2 is

72

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

Table 3 Likelihood weights. 1st Level Weight Normalized weight (%) 7.99 8.07 14.79 19.16 Second level Weight Normalized weight (%) 6.28 1.71 5.18 2.89 10.53 4.26 4.94 8.42 5.80 50.00

Asset vulnerability Operation and maintenance Asset state Potential for deterioration

10.3 10.4 19.0 24.6

Overall design Emplacement Exercising/use Maintenance Structural condition Performance Age External environment Internal environment Total

8.1 2.2 6.7 3.7 13.5 5.5 6.3 10.8 7.4 64.2

Total

64.2

50.00

Note: the judgements from 13 individuals from the workshops were used out of a possible 24.

Table 4 Consequence weights. 1st Level Weight Normalized weight (%) 3.79 20.96 14.95 Second level Weight Normalized weight (%) 3.12 0.67 6.38 14.58 10.56 4.39 1.46 3.45 1.70 1.76 1.93 50.00

Response Service provision Direct impact

5.8 32.2 22.9

Indirect impact

15.8

10.30

Isolation time Distance to backup Customers effected Key customers Regulatory contravention Direct costs Reputational Environmental impact Rail disruption Road disruption Infrastructure damage Total

4.8 1.0 9.8 22.4 16.2 6.7 2.2 5.3 2.6 2.7 3.0 76.7

Total

76.7

50.00

Note: the judgements from 13 individuals from the workshops were used out of a possible 24.

Table 5 Risk ranking of two valvesan example. Valve 1 Point of difference Questionable design or specication, is situated in a vulnerable location, is exercised/used opportunistically, not maintained, deteriorated and operation is unknown 37.60 6.78 254 Valve 2 Modern design, not vulnerable, exercised and maintained opportunistically with some deterioration. Valve operates. Relatively more important from the perspective of failure consequences 23.92 31.65 757

Due to the number of valves in the system, the risk indexing process will ideally be implemented in a database application linked to the asset inventory where the valve characteristics are stored. Risk ranking schemes can also be applied generically to identify different groups of valves that need to be inspected, which reduces the data needs.

6. On-going application of risk-based inspection As noted throughout this paper, the purpose of risk indexing is to provide effective prioritization in the absence of condition and reliability data. Once the inspection program has commenced, data will become available that can then be used to schedule and prioritize on-going inspections. For example, one case study partner operates approximately 1,800 large diameter valves. As recently as the late 1990 s, these valves were inspected every two to ve years. Cuts to maintenance budgets meant this practice was discontinued for some years. Over time, the condition of the valves thus became more and more uncertain, until a decision was made to address this through a comprehensive inspection program. As part of the inspections, the utility allocated a summary condition grade to each valve on a scale from 1 to 5 (1 indicating excellent condition, 5 indicating inoperable). Condition grades are often used in this way as a surrogate for the likelihood of failure [7]. For convenience, an equivalent grade can be specied to summarize failure consequences (1 being low consequences, 5 being high). As inspections proceeded, the utility was able to

Likelihood score Consequence score Total risk score

likely to be in better condition but still deteriorating. In terms of prioritization, Valve 2 would be prioritized for inspection on the basis of overall risk. The condition and potential failure consequences of Valve 2 might also imply a change to maintenance schedules, with planned exercising to ensure as far as is practicable the valve works on demand. Even though Valve 1 is anticipated to be in poorer condition, it would not be considered a priority for inspection based on overall risk due to the low potential failure impacts. It may, however, be considered a candidate for inspection under a policy to maintain all valves above a given condition threshold.

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

73

Table 6 Likelihood risk indexing scores: Valve 1. 1st level factor Asset vulnerability Operation and maintenance Asset state Potential for deterioration 2nd level factor Overall design Emplacement Exercising/use Maintenance Structural condition Performance Age (asset) External environment Internal environment Factor weight 6.28 1.71 5.18 2.89 10.53 4.26 4.94 8.42 5.80 Valve characteristic Questionable design or spec Vulnerable unmitigated Exercised/used opportunistically Not maintained Badly deteriorated, not known Inoperable, not known Ageing asset Moderately aggressive Aggressive and vulnerable Raw score 0.5 1.0 0.5 1.0 1.0 1.0 0.5 0.5 1.0 Total Weighted sore 3.14 1.71 2.59 2.89 10.53 4.26 2.47 4.21 5.80 37.60

Table 7 Consequence risk indexing scores: Valve 2. 1st order factor Response Service provision Direct impact Indirect impact 2nd level factor Isolation time Distance to back up assets Customers effected Key customers Likelihood of contravention Direct Cost Reputational Environmental impact Rail disruption Road disruption Infrastructure damage Factor weight 3.12 0.67 6.38 14.58 10.56 4.39 1.46 3.45 1.70 1.76 1.93 Valve characteristic 42 h 1001000 m 4 10,000 42 Low potential $50 k100 k High impact Low potential impact Moderate potential impact High potential impact Moderate cost Raw score 1.0 0.5 1.0 1.0 0.0 0.5 1.0 0.0 0.5 1.0 0.5 Total Weighted sore 3.12 0.33 6.38 14.58 0.0 2.2 1.46 0.0 0.85 1.76 0.97 31.65

Table 8 On-going prioritization using inspection data.

7. Conclusion This paper relates to a research project undertaken to develop guidance on condition assessment and inspection of water main valves. When starting an inspection program, it is highly desirable for initial effort to be targeted at assets that represent a high business risk. This paper has investigated issues related to the development of a risk indexing scheme for this purpose. To set the relative weights of risk factors within the indexing scheme, expert opinion was elicited at two workshops using the AHP. Such workshops provide an effective forum for collating expert opinion on relevant factors, thereby engendering buy-in from those who will use the prioritization process. The role of workshop facilitation is critical to obtaining useable results. In particular, it must be made clear what is being compared to ensure individuals address the comparisons in a consistent way. The AHP provides a means of structuring knowledge and checking the consistency of comparisons applied by each individual, but additional work must be undertaken to ensure consensus is achieved. As a result of the workshop experience outlined herein, the following suggestions are made to improve the process:

Note: valves at similar levels of risk are grouped in like-shaded cells, with higher risk cells represented by the darker shade. The highest risk valves were subject to corrective interventions; the next risk band were inspected annually (work queued as high priority); the next were inspected every three years; with the lowest risk band being subject to opportunistic inspections only.

group valves into risk-bands, as shown in Table 8. An on-going inspection regime could then be specied according to risk. In this case, the lowest risk valves were monitored opportunistically (with systems put in place to ensure relevant data was captured), whereas higher risk valves were placed on a program of routine inspections, with immediate corrective action being required for the highest risk valves. It is noteworthy that unless there is a major change in the use, service duty or service parameters for a valve or associated pipeline, the potential consequences of failure remain xed. From the perspective of the indexing scheme presented above, this allows the failure consequence score to be used as an on-going measure of asset importance. In contrast, condition changes over time, so the inspection regime for a given valve also changes as it shifts into the different risk bands shown in Table 8.

 An initial workshop should be arranged to discuss and agree


relevant risk categories, factors and associated characteristics;

 A second workshop should then be held wherein individuals


allocate weights according to their own experience; and

 A nal workshop should then be undertaken to discuss


differences in allocated weights and reach a consensus view. If this is not possible, then pragmatic approaches can be used whereby consistently allocated scores are combined to dene the overall scoring scheme. Overall, it can be concluded that in the face of poor data on asset condition and reliability, risk indexing schemes provide

74

D.R. Marlow et al. / Reliability Engineering and System Safety 100 (2012) 6774

a rational means of targeting initial inspection efforts. Eventually, information collected on asset condition can be used to augment the prioritization process and set on-going inspection schedules. The indexing scheme will, however, still provide information on asset importance, and thus inform on-going management and maintenance strategies. Improvements in data over time will also allow the weights used in the schemes to be calibrated against experience.

Acknowledgments The authors gratefully acknowledge the input and help of all industry professionals who gave their valuable time to contribute to this research. The funding provided by the Water RF, WSAA and CSIROs National Flagship Water for a Healthy Country is also gratefully acknowledged, as is the help by their staff. Finally, the CSIRO and external reviewers of this paper are thanked for their comments and input. The comments and views detailed in this paper may not necessarily reect the views of the Water Research Foundation, WSAA, CSIRO or any of their ofcers, directors, afliates or agents. References
[1] Marlow D, Beale D, Burn SA. Pathway to a more sustainable water sector: sustainability-based asset management. Water Science and Technology 2010;61:124555. [2] AWWA. Water:\stats 2002 distribution survey American Water Works Association. 2004. [3] Deb AK. Criteria for valve location and system reliability. Denver, Colo: American Water Works Association and Water Research Foundation; 2006. [4] MacLaren R, Keane K. Distribution System Valve Management: Leveraging Technology & Equipment to Deliver Results. AWWA ACE Toronto. Canada: American Water Works Association; 2007. [5] USEPA. Drinking Water Infrastructure Needs Survey and Assessment Third Report to Congress. 2003. [6] Water R.F. Condition Assessment of Water Main Appurtenances. Denver, Co.: Water Research Foundation; Accepted for publication. [7] WERF. Condition Assessment Strategies and Protocols for Water & Wastewater Utility Assets. Alexandria, VA: Water Environment Research Foundation; 2007. [8] Marlow DR, Burn S. effective use of condition assessment within asset management. Journal of the American Water Works Association 2008:100. [9] Marlow D.R., Beale D.J. Condition Assessment of Water Main Appurtenances: An Industry Perspective. Journal of the American Water Works Association. Accepted for publication. [10] Walski TM. Practical Aspects of Providing Reliability In Water Distribution Systems. Reliability Engineering and System Safety 1993;42(1):139. [11] Walski TM. Water Distribution Valve Topology for Reliability Analysis. Reliability Engineering and System Safety 1993;42(1):217. [12] Pascual R, Del Castillo G, Louit D, Knights P. business-oriented prioritization: a novel graphical technique. Reliability Engineering and System Safety 2009;94(8):130813. [13] Selvik JT, Aven T. A framework for reliability and risk centered maintenance. Reliability Engineering and System Safety 2011;96(2):32431.

[14] Vinod G, Bidhar SK, Kushwaha HS, Verma AK, Srividya A. A comprehensive framework for evaluation of piping reliability due to erosioncorrosion for risk-informed in-service inspection. Reliability Engineering and System Safety 2003;82(2):18793. [15] Simola K, Pulkkinen U, Talja H, Karjalainen-Roikonen P, Saarenheimo A. Comparison of approaches for estimating pipe rupture frequencies for riskinformed in-service inspections. Reliability Engineering and System Safety 2004;84(1):6574. [16] Stewart MG. Reliability-based assessment of ageing bridges using risk ranking and life cycle cost decision analyses. Reliability Engineering and System Safety 2001;74(3):26373. [17] Brito AJ, de Almeida AT. Multi-attribute risk assessment for risk ranking of natural gas pipelines. Reliability Engineering and System Safety 2009;94: 18798. [18] WRc. Guidance Manual for the Structural Condition Assessment of Trunk Mains. UK: Water Research Council; 1992. [19] Vinod G, Kushwaha HS, Verma AK, Srividya A. Importance measures in ranking piping components for risk informed in-service inspection. Reliability Engineering and System Safety 2003;80:10713. [20] Critchley RF, Aikman DI. Aqueduct management planning: thirlmere, haweswater and vyrnwy aqueducts. Water and Environment Journal 1994;8: 50212. [21] Schumi P. Criticality of Valve Operability. MI-AWWA/MWEA Annual Conference American Water Works Assocaition; 2009. [22] Marlow D, DeSilva D, Beale D, Marney D. Effective management of large diameter pressure pipes. Journal AWWA June 2010;102(6). [23] Chien C-H, Chen C-H, Chao YJ. A strategy for the risk-based inspection of pressure safety valves. Reliability Engineering and System Safety 2009;94: 8108. [24] Rogers PD, Grigg NS. Failure assessment modeling to prioritize water pipe renewal: Two case studies. Journal of Infrastructure Systems 2009;15: 16271. [25] ABS Guide for Surveys Using Risk-Based Inspection for the Offshore Industry, American Bureau of Shipping, December 2003. [26] Faber MH, Stewart MG. Risk assessment for civil engineering facilities: critical overview and discussion. Reliability Engineering and System Safety 2003;80:17384. [27] IPWEA. International Infrastructure Management Manual. Australia/NZ Edition ed: Institute of Public Works Engineering Australia; 2006. [28] Ridgeway M. Classifying medical devices according to their maintenance sensitivity: A practical risk-based approach to PM program management. Biomedical Instrumentation and Technology. 2001;May/June:16776. [29] Khan FI, Haddara MM. Risk-based maintenance (RBM): a quantitative approach for maintenance/inspection scheduling and planning. Journal of Loss Prevention in the Process Industries 2003;16:56173. [30] Lee AK, Serratella C, Wang G, Basu R, Spong R. Multilevel risk-based inspection scheme for FPSOs. Marine Technology 2007;44:11824. [31] Khan FI, Haddara MM, Bhattacharya SK. Risk-based integrity and inspection modeling (rbiim) of process components/system. Risk Analysis 2006;26: 20321. [32] ASME, Risk-based inspection-development of guidelines: general document: vol. 1, 1991, p. 172, ISBN: 0791806189. [33] Dey PK, Gupta SS. Risk-based model aids selection of pipeline inspection maintenance strategies. Oil and Gas Journal 2001:17384. [34] Saaty TL. The Analytical Hierarchy Process. New York: McGraw-Hill; 1980. [35] Basak I. Rank-based statistical procedures in the analytical hierachy process. European Journal of Operational Research 1997;101:3950. [36] Basak I, Saaty T. Group decision making using the analytic hierarchy process. Mathematical and Computer Modelling 1993;17:1019. [37] Forman E, Peniwati K. Aggregating individual judgments and priorities with the analytic hierarchy process. European Journal of Operational Research 1998;108:1659.