How different regional approaches to the network design result in key differences in burst event severity and failure vulnerability

Though water distribution networks (WDNs) are critical to the functioning of urban centers, the public often takes them for granted. They provide water for drinking, business, and industrial applications and water for firefighting. The design of WDNs that often can involve thousands of pipes and uncertain demands presents a considerable challenge. A wide variety of techniques have sought to improve, or even to ‘optimize,’ design based on a range of assumptions. For example, maximizing flow entropy has been suggested to improve reliability (Liu et al. 2014), a Multiobjective Simulated Annealing Algorithm has been suggested for design optimization (Shokoohi et al. 2017), a Cost Gradient-Based Assessment has been suggested for design improvement under varying loads (Dziedzic & Karney 2016), and Agent Swarm Optimization has been suggested for design optimization (Montalvo et al. 2014). However, all these optimization approaches are subject to constraints, which can vary by the region. Common constraints in North America are that no pipes smaller than 150 mm in diameter are used and the flows needed for firefighting are large, often requiring a looped design to produce the required flow. This work examines two representative networks with different constraints based on regional preferences: a branched Dutch network with smaller pipes and lower needed fire flows is contrasted to a traditional North American style network.

The hydraulic design of WDNs is often constrained by the need to supply large fire flows. This constraint can dictate both pipe size and the topology of the network. In North America, in key urban areas, fire flows of 800 cubic meters per hour (CMH) or 3,500 gpm can be required (American Water Works Association 2008). In residential areas, flows of no less than 115 CMH (500 gpm) are required (Linder 1991; American Water Works Association 2008). In contrast, Dutch planners and engineers have worked with fire authorities to reduce the requirements in residential areas that meet modern building codes to just 30 CMH, or about a quarter of the American requirement (Vreeburg & Boxall 2007). In France and England, and many of its former colonies (except Canada), needed fire flows in residential areas are about half, which are approximately 60 CMH in residential areas (Benfer & Scheffey 2015). This reduced need for high fire flows provides more flexibility in pipe choice and network topology.

North American networks typically use large-diameter pipes (≥150 mm diameter) arranged in a looped grid. The gird design ensures that many flow paths are available in case of fire, with each pipe contributing a portion. In contrast, the Dutch design emphasizes water quality and relaxes the need for a larger hydraulic capacity, thus permitting smaller pipes (63 or 40 mm) in a branched structure. Other factors being equal, smaller diameter pipes have higher velocities than larger ones, thus reducing transit time and water age and providing less time for the formation of disinfection by-products or disinfectant decay. Also, such pipes may be self-cleaning, where the higher velocities help to remove sediments, or possibly condition the pipes, so that red water events are less likely (Vreeburg & Boxall 2007; Vreeburg et al. 2009; Abraham et al. 2018).

There are at least 240,000 pipe bursts that waste over two trillion gallons of treated drinking water in the USA each year (Baird & Wagner 2020). Pipe bursts release pressurized water, thus leading to transient flows and pressures within the pipe system. Water hammer is the best-known example of a transient pressure and can damage pipes. For example, Wang et al. (2021) used a model of transient pressures to estimate the probability of failure of prestressed concrete cylinder pipe segments. Similarly, Kwon & Lee (2011) report that transient effects significantly increase the probability of pipe breakage. Low-pressure transients can occur downstream of a closing valve or a stopped pump or from a sudden water release. The role of low-pressure transients introducing contaminants into leaking pipes by suction has been widely studied (Lindley & Buchberger 2002; Betanzo et al. 2008; Besner et al. 2010). Other recent WDN transient research includes an active controller to reduce transients from changing demands (Martin et al. 2022), a computationally efficient adaptive model (Nault & Karney 2020), and new modeling approaches (Andrade et al. 2022).

Despite considerable progress, previous research has rarely attempted to link different regional constraints to pipe burst behavior. This work compares how an innovative Dutch network design performs after a pipe burst relative to a typical North American network design, with an emphasis on transient pressures.

A pipe burst was simulated by connecting a pipe with negligible resistance to the test node indicated. This pipe terminates in an emitter that allows the pipe to discharge freely into the atmosphere. The steady-state solutions were determined using EPANET 2.0 (U.S. Environmental Protection Agency), and the transient analyses were done using TransAM (Northwest Hydraulic Consultants Inc., Pasadena, CA). The transient analysis consisted of 12 s model run where a pipe burst occurs over 1 s. The model accounts for cavitation when negative pressures were reached. The vapor pressure of water was assumed to be 0.24 m.

This initial analysis using EPANET considers the steady performance of the network several seconds after the pipe burst has occurred and until the pipe is isolated. The transient analysis of the pipe burst on shorter time scales is considered below.

All design and modeling choices require analysts to make assumptions and simplifications. Since the objective here is to compare different network design approaches, no surge protection or air valves were included. In actual networks, these devices have the potential, if well chosen and maintained, to dampen transient effects. Because both networks were assumed to have only plastic pipes, transients associated with sudden events were relatively mild; more severe transients would be expected for bursts associated with cast iron or other rigid pipe materials. For simplicity, the soil surrounding the pipe burst is assumed to be quickly washed away, and unrestricted discharge after the burst is the result. Overall, other than the choice of plastic pipes, the approach is conservative.

Design inevitably involves trade-offs depending on what aspects are prioritized. Thus, it is unsurprising that the two design approaches considered here have their advantages and disadvantages. The North American test network acts as an effective and robust fire extinguisher. When a burst occurs, the pipe contents are rapidly mobilized and negative transient pressures are expected. In this example, 50 m of the pipe upstream of the burst experiences pressures of −10 m. Also like a good fire extinguisher, the potential flow available is considerable. In this example, 930 CMH of water can be released after a burst. These high flows can result in water damage, product loss, local flooding, street damage and traffic interruption, pipe contamination, and disruption of other services like electricity (Gaewski et al. 2007; Pietrucha-Urbanik 2015; Karney & Gibson 2021). Much higher direct and indirect costs are expected after a pipe burst in the North American network.

In the Dutch design, the small diameter pipes act essentially as a limiter or governor, localizing the effects and restricting the flow that can be produced after a burst to just 50 CMH. In this case, there is a lower expected cost associated with a pipe burst since much less product loss and flooding are expected.

When not being used to fight fires, the contents of a fire extinguisher are at rest, a situation that is nearly true for the North American test network. The median flow velocity is a mere 0.04 m/s. Such a small velocity can lead to stagnation and degraded water quality, something the Dutch have sought to remedy in their design.

A potential disadvantage of the Dutch network, highlighted here for the first time, is that the Joukowsky equation predicts higher transient pressures during normal operation. In the top 10% of pipes of the test network, transient pressures of at least ±107 kPa (16 psi) would be experienced during the sudden switching of demands, even in normal operation. This is not severe enough to result in negative pressures and intrusion under most conditions. However, such routine transient events may affect the network in unexpected ways, finding and damaging weakened pipes for example. But these transients pressures also likely help to mobilize sediments and keep the pipes clean (Weston et al. 2021). The overall effect of these relatively low, but frequent, transients remains unclear.

Negative pressures can result in the ingress or intrusion of contaminants. Before the pipes are put back into service, they should be decontaminated where negative pressures are suspected. In the North American test network, there are three pipes that experience high velocities (>3 m/s) as they feed water to the burst pipe. Negative pressures can occur in any one of these three pipes. In addition, high velocities occur throughout the network due to its looped topology and high flows are produced. As such, all pipes adjacent to the burst should be decontaminated for approximately 50 m, which is a challenge in a grid network. The problem is simpler for the Dutch network. The pipe downstream of the burst should be decontaminated, which is easier to execute.

The WDNs studied here represent two different design philosophies. The key question is this: is it better to design networks so that their performance is excellent during normal conditions but merely adequate during emergencies, or is it better to design the network for best performance during emergencies and slightly compromised during normal conditions? Although there are difficult choices, it is worth noting that fires may represent just a few hours of operation in the 30-year service life of many pipes. Based on the results presented here, each of these designs has several advantages and disadvantages. For North American Networks:

• High fire flows are readily and reliably achieved due to large pipes (≥150 mm) and grid topology.

• Water is discharged to the environment at a very high rate after a burst, which can result in product loss and local flooding.

• All pipes near the burst should be decontaminated for approximately 50 m. Many parallel flow paths feeding the burst means negative pressures can occur in any one of them.

• Network demands can likely still be met during a pipe burst.

• Flow is nearly stagnant during normal operation.

For Dutch networks:

• Water demands cannot be met downstream of a burst pipe until repair is completed.

• The smaller pipes and branched structure act as flow limiters. Less water damage and flooding are expected after a burst, but also less fire flow is available.

• Lower water age and better water quality are expected. The median pipe velocity during normal operation was approximately 0.2 m/s, which was four times higher than the North American example.

• Only the pipe downstream of the pipe burst needs to be decontaminated.

• Transient pressures of approximately 100 kPa (16 psi) are expected in the top 10% of pipes during normal operation, likely mobilizing or conditioning sediments.

There are many advantages associated with the Dutch design, including better water quality, and lower costs and less damage expected after a pipe burst. However, it is not currently possible to implement this design, as shown here, in North America due to fire-flow requirements. Fundamentally, the controversial background question is how much water is really needed to provide fire protection in modern neighborhoods. The Dutch approach implies this need is much less than that currently provided in North America. This may suggest that it may be time to re-examine fire-flow requirements in North America that have remained largely unchanged for many decades.

The authors gratefully acknowledge the contribution of David Axworthy and Northwest Hydraulic Consultants Inc.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.