A resilience-based prioritization scheme for water main rehabilitation

There is a great dependability of modern urban societies on lifeline systems such as drinking water distribution networks (WDNs) for basic survival and economic prosperity. Therefore, reliability of supply both in terms of quantity and quality is one of the top priorities for any water utility. Supply interruptions could also impede firefighting capabilities, potentially resulting in loss of lives and property. The American Society for Civil Engineers (ASCE) in their latest infrastructure report card gave a near-failing ‘D’ grade for drinking water infrastructure (ASCE 2013). Our WDNs, currently in a deteriorated state, need huge investments that are beyond the capabilities of local municipalities and governments. For example, the estimated 20-year capital investment need for revamping deteriorated pipelines in the United States is about $334.8 billion (ASCE 2013). This poor state can be primarily attributed to the fact that a majority of pipelines have been in service for considerably longer than their intended use. Sub-optimal design techniques at the time of their installation combined with lack of adequate maintenance exacerbated this problem.

Due to the lack of economically feasible and reliable condition assessment techniques, a majority of pipelines are not adequately monitored in order to identify and fix the defects before they grow into catastrophic failures. With about 240,000 water main breaks reported annually in the USA, consequences of the current state of WDNs include decreased reliability, supply interruptions, and other societal inconveniences (ASCE 2013). These consequences could be prohibitively expensive, depending primarily on the size of the failed pipeline and its importance in the overall WDN functioning. For example, emergency repairs to large diameter water mains in critical locations have proven to be expensive compared to planned rehabilitation (ASCE 2013). Consequently, effective asset management and subsequent prioritization of critical pipelines for planned rehabilitation is a timely need for water utilities. This paper presents a novel prioritization framework for rehabilitation of critical pipelines. The novelty lies in the modeling of time taken to detect and repair failed pipelines. The proposed framework is demonstrated on a section of real WDN to highlight its advantages compared to the more traditional risk-based framework for rehabilitation prioritization.

Several researchers have proposed schemes in the past for rehabilitation prioritization of WDN pipelines based on their failure criticality (Moglia et al. 2006; Rogers & Grigg 2009; Studzinski & Pietrucha-Urbanik 2012; Grigg et al. 2013; Rahman et al. 2014; Yoo et al. 2014). Failure criticality in a majority of these past studies was evaluated based on the respective pipeline's condition and the estimated impact of its failure. As water pipelines age, they deteriorate due to the combination of several factors that include natural material degradation and subsequent loss of structural integrity, lack of proper maintenance, fatigue loading and subsequent localized damage, design defects or construction errors that weaken the system over time, adverse operating schedules, and adverse environments (Piratla et al. 2014). Pipeline condition upon investigation is usually translated into the probability of its failure. Impacts of water pipeline failures in some cases are felt only locally in the form of flooding of streets, damage to roads and traffic restrictions, and in other cases are system-wide in the form of supply interruptions, low pressures, and potential contamination. The aggregate failure impacts are usually represented using a normalized index. Approaches adopted in the literature for estimating failure probability and failure impacts are briefly reviewed in the following paragraphs.

Previous researchers employed different approaches for estimating the failure probability of water supply pipelines. Moglia et al. (2006) used a non-homogeneous Poisson process (NHPP) model by considering pipe length, size, type, age, operating pressure, and soil type as influential variables, and also modified this using the best linear unbiased predictor for better failure probability prediction. Rogers & Grigg (2009) used NHPP and multi-criteria decision analysis (MCDA) for estimating pipeline failure probability. Rogers & Grigg's (2009) NHPP model is based on pipe age, condition, and historical failure information, and was found to be applicable only to pipelines with three or more previous breaks. The MCDA model is based on the ratings given to several influential factors along with their respective user-assigned weights, and it was found to be applicable to pipelines with one or two previous breaks. Studzinski & Pietrucha-Urbanik (2012) estimated failure probability based on the ratio of average pipeline suspension time due to repair to the average operational time without a failure. Grigg et al. (2013) calculated the pipeline failure likelihood index using the weighted sum of ratings given to variables such as age, break rate, and service conditions (i.e., traffic load, pressure zone, and soil corrosivity), where weights are defined by the user. Rahman et al. (2014) proposed a probability failure score based on the remaining useful life of a pipeline, which is estimated by considering age, material, and failure history. Yoo et al. (2014) estimated failure probability using a probabilistic neural network (PNN) model based on pipeline deterioration rate, which was calculated by considering internal and external factors that influence pipeline failure. Internal factors included several pipeline attributes, failure history, failure type, water quality, and operating pressure. External factors included corrosion rate, backfilled soil type, and road width. In addition to the studies reviewed in this paper, there are several other published studies that evaluated failure probability of pipelines; however, a major limitation has been the availability of quality data that enhances the accuracy of the model output.

Previous researchers also employed different approaches for estimating the failure consequences of water supply pipelines. Moglia et al. (2006) estimated failure consequences based on the costs of pipeline renewal, valve insertions, pipeline repairs, supply interruption, and physical damages caused by failures. Berardi et al. (2008) estimated failure consequences based on direct costs of pipe replacement, risk of pipe breaks, system reliability, work allocation, and leakage reduction using the pressure-driven model for simulating the hydraulic performance of the system. These failure consequences are employed as objectives in a multi-objective genetic algorithm (MOGA) in order to identify a set of optimal pipeline replacement schemes. Berardi et al. (2009) added a sixth objective, namely, preferential pipe selection, to their previous MOGA framework for rehabilitation decision-making in WDNs. Rogers & Grigg (2009) estimated monetary failure consequences based on economic, environmental, and social factors. Studzinski & Pietrucha-Urbanik (2012) estimated failure consequences as the ratio of volume of water not delivered to the required water volume. Grigg et al. (2013) used a weighted sum of the ratings given to repair cost, pipe size, and failure location, in which the weights are defined by the user. Rahman et al. (2014) used a degree of impact which is based on water service demand, number of critical consumers, density of customers, land use, traffic impact, pipe materials, and repair cost. Yoo et al. (2014) used a hydraulic importance formula, which was calculated as the ratio of delivered water to required water for both single and multiple pipeline failure scenarios. Although many researchers have presented approaches for assessing water main failure consequences, due to the difficulty associated with predicting the true failure impact in the form of interruption duration, notifications of interruptions, time and day of interruption, number of interruptions per year and the resulting environmental and societal costs, estimating failure consequences has been challenging and less accurate.

In summary, several studies in the past developed risk-based prioritization models for WDNs; however, many limitations led to over-simplification of risk calculations, especially while estimating the consequences. Not much consideration has been given to estimating the time taken by a water utility operator in responding to a pipeline failure. Clearly, failure consequences depend on how fast and efficiently a water utility operator responds. This paper proposes and demonstrates a modified risk assessment model, named as resilience assessment, which is based on pipeline failure probability, consequences, and restoration capability, which are separately estimated and integrated into one metric.

The ability to resist failure without loss of functionality is a combined effect of the inherent system strength and the type and extent of stresses imposed upon it. For example, deterioration due to wall corrosion in metallic pipelines will degrade the structural capacity to perform under a variety of loads. Consequently, the failure probability will be higher in such pipelines, making them less resilient to failures. This ability to resist failures is expected to decrease with increased deterioration, which is a currently prevailing trend in the USA's WDN pipelines. As illustrated in Figure 1, if a pipeline fails at time ‘T_F’ due to its reduced ability to resist failures, overall system performance drops from its original performance (C_O) to performance in failed state (C_F). The extent of performance reduction (C_O-C_F) depends on the failure severity, the overall WDN's reliance on the failed pipeline, and the amount of redundancy available in the system. Once a failure is detected, the corresponding pipeline will be isolated at time ‘T_D’ for repair. At this time, the overall system performance changes from C_F to C_D, where C_D denotes WDN performance when the failed pipe is isolated for repair. Emergency crews are sent to repair the pipeline, after which full functionality (C_O) is recovered at time ‘T_R.’ The loss in WDN resilience for a given pipeline failure can be characterized by the hashed area in Figure 1, which is approximately equal to [{(C_O-C_F)*(T_D-T_F)}+{(C_O-C_D)*(T_R-T_D)}].

The pipelines are appropriately prioritized for rehabilitation or replacement based on the corresponding resilience values. The five parameters, namely, P_j, C_O-C_F, T_D-T_F, C_O-C_D, and T_R-T_D are modeled in this study in such a way that their values range between 0 and 1. The individual parametric models are further explained in the following sub-sections.

Failure consequences are the losses incurred as a result of water pipeline failure. Depending on topology, design, redundancy, and demand patterns, failure consequences vary across a WDN. Consequences can be categorized into ‘local impacts' and ‘system-wide impacts.’ Failure consequences of a large diameter pipeline that forms a critical link to a section of WDN could have system-wide impacts, rendering a significant portion of WDN dysfunctional. On the other hand, there will be some pipelines where failure will result in mere local impacts. System-wide consequences can be quantified by estimating the shortage of supply at all demand nodes in the WDN for a pipeline failure scenario. Consequences in the following two sequential stages of pipeline failure are separately estimated in this study: (a) from failure occurrence until the pipeline is isolated for repair (i.e., C_O-C_F in Figure 1) and (b) from pipeline isolation until it is repaired and re-commissioned (i.e., C_O-C_D in Figure 1).

Failure consequence in the first sequential stage is the reduction in system pressure due to energy loss from the failure. Only after the problem is detected and located is the pipeline isolated for repair by closing the isolation valves nearest to either side of the failure location.

Consequences of water pipeline failure in the second sequential stage include possible system-wide pressure reduction and supply outages. These consequences (C’_j) are estimated following the same procedure described for C_j using Equation (4), except there are no simulated orifice flows but instead failed pipelines are isolated using the same integrated EPANET and pressure-driven simulation model used for estimating C_j. Similar to C_j, C’_j is a metric that ranges between 0 and 1, and represents C_O-C_D in Figure 1.

A water utility's response to a pipeline failure begins with detection, followed by repair. Pipeline failures could be detected from an unusual pressure drop at the supply source as well as at other locations in the WDN where pressure is continuously monitored. Otherwise, it is only realized when the effects are felt above the ground in the form of water flooding streets, sidewalks, or even houses. Upon realizing the presence of a failure and detecting its location, water utility crews isolate the failed component for repair work. Minimizing the time between the break occurrence and its isolation (i.e., T_D-T_F), referred to as detection time (TD) in this study, is critical because much energy is wasted through pressure loss in addition to significant amounts of water losses. Repair work begins after the failed section has been isolated. Emergency pipeline repairs could take a few hours to days depending on location, site access, failure severity, soil cover depth, pipe material, pipe diameter, surrounding utilities, and soil conditions. Minimizing the time between isolation of a failed section and completing the repair (i.e., T_R-T_D), referred to as repair time (TR) in this study, is also crucial because closing of a WDN section for repair may lead to supply outages at several critical locations in the WDN, in addition to traffic-related woes. Both TD and TR are estimated in this study to range between 0 and 1, and together they epitomize the response time of a water utility to pipeline failure.

As can be seen from Figure 3, detection time (TD) is considered to be dependent on four factors, namely, frequency (D₁₁) and type (D₁₂) of monitoring administered by the water utility, techniques used for precisely locating failures (D₂₁), and the expertise of inspecting field crews (D₂₂). Table 1 presents the criteria developed in this study for rating these four factors for a given pipeline on a scale of 0.1 to 1. The rating scale is chosen in such a way that TD when calculated will range between 0 and 1. Rating criteria for different categories of these four factors are devised, as shown in Table 1, based on the influence of individual categories on TD. For example, the frequency of monitoring (D₁₁) factor is classified into ‘passive,’ ‘periodic,’ and ‘continuous’ categories with corresponding ratings of 1, 0.55, and 0.1. The smaller the rating, the sooner a failure is detected. Passive monitoring refers to the practice where pipelines are not pro-actively monitored but only assessed upon a failure. Periodic monitoring refers to the practice where pipelines are monitored periodically (e.g., annually). Continuous monitoring refers to the practice where pipelines are continuously monitored, for example, using smart sensor networks that are embedded in WDNs. Continuous monitoring of WDNs will enable water utilities to detect failures as soon as they occur and, therefore, reduce the detection time. Similarly, periodic monitoring is better than passive monitoring. Passive, and periodic monitoring to some extent, are common practices followed by water utilities. It should be noted that ratings of 0.1, 0.55, and 1 are given to the D₁₁ factor only based on the relative influence of different D₁₁ categories on TD and not their absolute influence. Ratings based on their proportional absolute influence will require further empirical research in the future. Similarly, categories and corresponding rating criteria for the other three factors, namely, D₁₂, D₂₁, and D₂₂, are formulated and presented in Table 1.

As seen in Figure 3, repair time (TR) is considered to be dependent on several factors, such as pipeline diameter (R₁), pipeline material (R₂), type of soil (R₃), presence of surrounding utilities (R₄), ground water table (R₅), working space (R₆), reachability of the work site (R₇), and labor skills (R₈). Larger diameter pipelines are expected to take a longer time for repair than smaller diameter pipelines (Hartley 2013). R₁ is classified into three categories and ratings assigned accordingly, as shown in Table 2. Concrete pipes were reported to have taken more time for repair followed by metal pipes and then plastic pipes (Bueno 2010); rating criteria R₂ factor is accordingly formulated and presented in Table 2. Rating criteria for R₃, R₅, R₇, and R₈ factors are appropriately formulated based on evidence from the literature (Doloi et al. 2012; Weir & Cullen 2014), whereas rating criteria for R₄ and R₆ are intuitively formulated, as shown in Table 2.

Out of the 12 factors listed in Figure 3 that influence TD and TR, R₁, R₂, and R₃ are rated, as per Table 2, based on available GIS data for CWDN. Ratings for D₁₁, D₁₂, and D₂₁ are considered to be dependent on pipe sizes, an assumption that is consistent with most common asset management practices. Pipes less than 24 inches in diameter are considered to be ‘passively’ monitored using ‘visual’ techniques for failure detection and location. Pipes with a diameter greater than 24 inches are considered to be ‘periodically’ monitored using ad hoc techniques for failure detection and location. R₅ was given a rating of 0.1 for all pipelines, as per Table 2, because the ground water level in the CWDN region is in the range of 5 to 15 feet (Walter 1971) and the typical soil cover depth for CWDN pipelines is only about 4 feet. R₆ and R₇ factors are rated, as per Table 2, based on road types and distances of respective pipelines from the water utility office location.

Due to the lack of available data, ratings for the remaining three factors, namely, D₂₂, R₄, and R₈, are simulated using the Monte Carlo technique. One hundred simulations were carried out using random ratings for these three factors (i.e., D₂₂, R₄, and R₈) and estimated ratings for the other nine factors are discussed in the preceding paragraph. The resultant TD_j and TR_j are combined with P_j, C_j, and C’_j of the respective CWDN pipelines using Equation (15) to obtain resilience values (R_j) and subsequent priority rankings. The pipeline corresponding to the least CWDN resilience is given the highest priority for rehabilitation.

The overall CWDN resilience calculated using Equation (2) is found to vary in a small range of 0.76 to 0.77 over the 100 simulations, with an average value of about 0.766.

Upon further investigation of the results, it was observed that some time factors were largely uniform across CWDN, thereby contributing little to the variation in detection and repair times. For example, the criteria of pipe size less than, equal to, or greater than 24 inches in diameter influences four of the twelve time factors, namely D₁₁, D₁₂, D₂₁, and R₁, and it was observed that 85.1% of CWDN pipelines fall in the ‘pipe size less than 24 inches’ category. Similarly, 98.6, 100, 100, and 100% of CWDN pipelines fall in the ‘metal’ – (R₂), ‘sandy facies’ – (R₃), ‘above ground water table’ – (R₅), and ‘less than one mile from water utility location’ – (R₇) categories, respectively (1 mile = 1.609 km). CWDN is a small 72-pipeline section considered for demonstrating the proposed approach in this study, and it definitely lacks the kind of variation in pipe material, soil type, and accessibility that is usually observed in a typical large-scale WDN. Clearly, the percent change, h (R, R’), would have been greater in a large-scale WDN, thereby highlighting the greater significance of incorporating response time into prioritization schemes.

It is also suggested in this study that WDN resilience can be enhanced by not only improving the physical infrastructure but also the ability of a utility operator to respond quickly to a failed pipeline. The cost–benefit analyses of such improvements should be investigated in the future using a robust optimization algorithm.

WDNs are currently in need of immediate attention in the form of technological innovation, financial resources, and planning and management tools. This study presents a resilience-based prioritization approach for pipeline rehabilitation planning. Resilience is quantified in this study as a combination of estimated failure probability, failure consequences, and failure response time. To the best knowledge of the authors, there have not been any previous studies that explicitly considered response time in the evaluation of pipeline failure risk. Utility response time is estimated in this study based on the outcomes of a survey on several factors that influence both failure detection and repair times. A section of a large WDN with about 4.35 miles of pipeline is used to demonstrate the proposed resilience-based prioritization approach. The WDN used for demonstration contained ductile iron and cast iron pipe materials for the most part, and has diameters ranging from 4 inches to 24 inches. WDN pipelines are prioritized in the increasing order of their resilience values, and the overall system resilience for the demonstrative WDN was calculated to be 0.766.

The results from the resilience-based prioritization framework are compared with those obtained from a more traditional risk-based approach to evaluate the merits of incorporating response time. A relative pair-wise comparison metric was employed for a meaningful comparison of pipeline priorities from resilience- and risk-based approaches. The comparison revealed that the average percent change in relative pipeline priorities for the top one, three, five, seven, and 10 critical pipelines from the risk-based approach are about 12, 15.3, 13.6, 13.2, and 16% when compared to the resilience-based approach. With an overall average percent change in pair-wise relative rankings of 9.2%, the comparison highlighted reasonably significant differences in pipeline priorities. This overall percent change is expected to be higher in the case of a large-scale WDN, highlighting the greater significance of incorporating response time into prioritization schemes. It needs to be further investigated as to what set of rehabilitation actions will result in the greatest increase in WDN resilience as defined in this study. An optimization algorithm subjected to budget constraints needs to be developed and used in the future for effective rehabilitation decision-making. A more practical way of rating the detection time and repair time factors may also be pursued in the future for enabling easier adoption of the proposed approach by water utilities.

The authors greatly appreciate the participation of the 40 utilities in the survey administered as part of this study. These utilities span across 30 different states in the United States. The authors also thank the officers at the water utility that provided GIS data for this study. Finally, the authors thank the reviewers for their many valuable comments.