AI in Action

Using artificial intelligence to predict likelihood of failure.

Machine learning is used to rank pipe segments based on likelihood of failure within five years. Red indicates greatest risk; dark blue, least risk. The company’s marketing material focuses on cast iron because the material comprises 30% to 50% of installed pipeline nationwide and is involved in 75% of leaks, but the underlying technology applies to any pipe material.
Fracta Machine learning is used to rank pipe segments based on likelihood of failure within five years. Red indicates greatest risk; dark blue, least risk. The company’s marketing material focuses on cast iron because the material comprises 30% to 50% of installed pipeline nationwide and is involved in 75% of leaks, but the underlying technology applies to any pipe material.

If you use age, break history, and soil type to build a model that projects condition assessment results of a pipe segment onto segments nearby, you still lack variables – such as proximity to train tracks – required to confidently predict their likelihood of failure (LOF) within the next five years. Fracta Inc. (formerly HiBot USA) software fills in knowledge gaps by adding 800 other variables to the equation. The result is a systemwide replacement prioritization solution delivered in less than two months.

This is an example of machine learning: a field of artificial intelligence that uses algorithms to identify and analyze patterns in “big data” and learn from it to predict the future. Historical data is broken into a training group used to create formulas and weightings, which are then validated against a group of known data. The cycle is repeated thousands of times until the algorithm is able to predict validation group results with a high degree of accuracy.

In addition to a utility’s age, break history, location, material, size, and location data, the algorithm incorporates data about a particular location gleaned from national sources like the U.S. Census Bureau and U.S. Geological Survey street slope data. For example, it considers average pH for a neighborhood and the city as well as around the pipe to look for broader correlations to failure.

Imposing Order on Chaos


Assembling a database that provides a big-picture overview.

A Netherlands water utility with 2.5 million customers restored service within two hours of a 630 mm-diameter pipe break. A sensor that picked up the sudden pressure drop sent a signal to the PI System. In addition to alerting emergency crews, the database enabled the utility to estimate the spreading size of the pond and identify ways to cut off the pipe and reroute service.
evides A Netherlands water utility with 2.5 million customers restored service within two hours of a 630 mm-diameter pipe break. A sensor that picked up the sudden pressure drop sent a signal to the PI System. In addition to alerting emergency crews, the database enabled the utility to estimate the spreading size of the pond and identify ways to cut off the pipe and reroute service.

A couple crucial details most vendors won’t tell you about the Internet of Things. Sensors generate an overwhelming volume of incoherent information and speak several different languages. Much of the data comes in out of order. For example, the 3:00 reading from pump 1250 might come in at 3:17 – and must be reconfigured to generate accurate real-time operational analyses. Someone has to do the unglamorous job of putting it all in order.

OSIsoft’s PI System is a database product that captures, scrubs, structures, and presents values, sensors on pipes, vibration signals on pumps, energy meters, etc., in a way that quickly highlights problems. Adding this information to what’s coming in through a SCADA system enables a utility to go beyond monitoring operations to saving money and predicting failures.

Spotting Leaks from Space


Using soil information gathered from 400 miles above Earth.

Using a unique wavelength in the radar spectrum, Utilis Inc. discovered that treated water reflects differently than rain and sewage. Because radar is susceptible to “noise” caused by vegetation, buildings, metal, and the atmosphere, the company’s algorithm removes these obstructions while preparing raw data for analysis. A sieve then extrapolates leaks that correspond to treated water’s spectral signature.
Utilis Using a unique wavelength in the radar spectrum, Utilis Inc. discovered that treated water reflects differently than rain and sewage. Because radar is susceptible to “noise” caused by vegetation, buildings, metal, and the atmosphere, the company’s algorithm removes these obstructions while preparing raw data for analysis. A sieve then extrapolates leaks that correspond to treated water’s spectral signature.

X-ray machines use electromagnetic radiation to "see" through skin and muscle and zero in on the calcium in bones. Similarly, Utilis Inc. uses L-band microwaves to "see" through trees, pavement, and grass to find soil that's saturated with treated water.

The company's algorithms crunch millions of satellite images for the spectral “signature” of treated drinking water and overlay the results on a water utility's GIS. Thousands of square miles of territory are analyzed to identify leaks of one-half to one liter per hour to within 150 feet.

Instead of going to the same location every few years to assess pipe condition, a utility’s entire pipeline network can be scanned in seconds. Doing this every few weeks and comparing the results to a baseline analysis reveals leaks as they evolve. Managers can prioritize rehabilitation or replacement budgets to specific leaks or problematic areas and confirm that a repair was made and is performing satisfactorily.

Utilities receive graphic and/or tabular report containing location and estimated size of suspected leaks. Crews can verify in the field. Reports are provided monthly, quarterly, or semi-yearly. Benefits include early detection and repair verification. Repeatedly updating the reports improves accuracy of findings, enabling utilities to refine operational work-plans.