City Develops Predictive Modeling for Water Main Breaks

Syracuse, New York, partners with the University of Chicago on a data algorithm that predicts which pipes are at the highest risk of failure

City Develops Predictive Modeling for Water Main Breaks

(Photo from city of Syracuse)

Interested in Infrastructure?

Get Infrastructure articles, news and videos right in your inbox! Sign up now.

Infrastructure + Get Alerts

Where and when water mains will fail has long been the million-dollar question for municipalities — often in the most literal sense. Difficult to predict and expensive when they do occur, preventive maintenance and proactive water main replacement programs are becoming necessity for municipalities.

But that can frequently seem like an exercise in futility — or practice for the lottery. Often it’s a shot in the dark, success difficult to see even in an effective program. That is why many cities still fall back into a reactionary mindset, replacing mains and pipes as the break, constantly cleaning up messes instead of avoiding them altogether.

Syracuse, New York, was one such municipality, reacting to and dealing with about 332 water main breaks annually on average in its 550-mile system to the tune of $1 million.

“As is common with a lot of similar cities, the water system is at or past its useful life,” says Sam Edelstein, chief data officer for Syracuse. “We’re experiencing a lot of water main breaks; we’re typically very responsive to problems, rather than trying to address them more proactively.”

But that is gradually changing as the city embarks on an ambitious data analysis program targeted at predicting risk associated with water main breaks.

Putting Data to Work

Working with a University of Chicago program, Syracuse has developed a predictive algorithm for water main breaks based on data it’s collected over the years.

The University of Chicago program, Data Science for Social Good, is a fellowship aimed “to train aspiring data scientists to work on … data science projects with social impact,” according to the DSSG website. They work with governments and nonprofits. In the case of Syracuse, it meant providing services equaling about $100,000 toward tackling the water main break issue.

There were two major components to the project: collecting data across different departments, and analyzing that data to assess risk and predict which water mains are the most likely to break. Fortunately for Syracuse, its water department has long kept good records of breaks and repairs, going back decades. It was just a matter of combining that information with more anecdotal evidence sourced from current staff.

“Using all this data, we started to look through similar attributes over the history of the water main breaks that we’ve had. Based on that, where do we expect that there might be problems in the future?” Edelstein says. “We worked on building a model to help predict those things." 

The model considers a number of factors like age of the pipe, construction material, previous breaks, and pipe dimensions — applying it all to characteristics of Syracuse’s past water main breaks in order to then predict future breaks more accurately. The code behind the predictive modeling is open source and available on GitHub.

For Syracuse, one of the biggest benefits of the program is not only preventing water main breaks in general, but also using the data model to be more efficient in repair work across the city. Most water system projects involve tearing up and repaving roads, which can result in redundancy if a water main breaks under a road that was recently paved. 

“Paving a specific road that happens to be on top of a really risky water main, we may hold off on that, because the last thing we want to do is shut down a road to repave it, have it be nice and new, and then have a water main break underneath it,” Edelstein says. “Replacing water mains can be so expensive, what we have tried to do is find collaboration with either other departments or third parties that may be doing their own sort of repair work.

“We can then go to them and say, ‘We know that this water main is risky and likely to break. You’ve opened up the road for some reason to do construction on it, let us give you the materials to replace the water main, or we’ll go in and replace the water main when this road is opened.’ There can be some more cost sharing. Ultimately when the most expensive thing is to repair the road or to reinstall the road, we can at least share those costs.”

Moving Forward

The program began in January 2017 and is still in an experimental stage. Syracuse has used the model to collaborate on three projects so far, and though it hasn’t embarked on a full-scale replacement program, the data has been incorporated into the current maintenance schedule. In general, the algorithm has affirmed itself through the water main breaks that have occurred over the last year.

“One of the hard parts with the predictive modeling, the only way to really know if it’s working right is if there’s a water main break, which is ultimately what we don’t want to have happen,” Edelstein says. “As we’ve seen, because we haven’t replaced all of them, the model is operating sort of as expected and is predicting the most risky water mains. But we know that because the water mains are breaking.

“What we’d like to be able to do is have some sort of feedback mechanism, where the water main hasn’t broken, but maybe we see leaks that are springing up or some other way of us being able to assess the quality of that water main. Then that would feed back into the model to say, ‘Yep, you’re on the right track.’ So as you’re continuously looking at the analysis, that should help to inform where other risky mains are.”

Toward that end, the city is already incorporating new leak detection sensors into the distribution system. It’s early days for this arm of the project, still only testing the efficacy of various sensor options. If Syracuse decides to make the investment, the sensors will form the “feedback mechanism,” giving city leaders more information to decide on which projects receive priority.

“We’re trying to think about how all these different pieces play together, and use the data that we have at our fingertips now to be able to evaluate things, but then also look at where might we use additional technologies to help paint a more complete picture,” Edelstein says.

This chart shows how accurately the algorithm is able to predict future water main breaks compared to other methods.
This chart shows how accurately the algorithm is able to predict future water main breaks compared to other methods.

Early Success

“It does seem to be operating well,” Edelstein says of the algorithm. “The idea was basically if we at random picked out 50 water mains in the city to replace, without any other knowledge, 10 percent of them would likely have a water main break on them in the next three years. Using this system, it was essentially a sixfold increase (on pipe break prediction accuracy).

“We have then used that data to talk to partners and other third-party agencies throughout the city. It has ended up saving hundreds of thousands of dollars for the city, because we didn’t have to go and pay to replace a water main on our own, instead we were sharing the burden.”


Comments on this site are submitted by users and are not endorsed by nor do they reflect the views or opinions of COLE Publishing, Inc. Comments are moderated before being posted.