Every day in the U.S., about 700 water pipes burst, causing damage and disruption and costing millions in repair and replacement. Utilities react quickly and effectively, but the damages to property, roads and the utility’s image can be significant. In addition, these disruptions often cause lost business in the area of the break. Worker productivity is frequently impacted by interferences in transportation.
Water utilities have effective and well-trained operational teams that respond quickly to pipe failures. The process is complex as well as urgent. Calls often come in the middle of the night or during a weekend, but being reactive is not enough.
All utilities have capital investment planning resources to replace their aging or problematic infrastructure. Proactive pipe replacement is employed by utilities with adequate capital budgets. The challenge is to select the right pipes to replace. Most engineers select these pipes based on prior break history, pipe age or material, and sometimes simply by using intuition (“those pipes by the high school have always been a problem”). All utilities share the frustration of digging up a pipe for replacement and finding it still has many good years of service remaining. They replace it anyway but lament its selection. Utilities report accuracy for proactive identification at around 20%.
Advances in Artificial Intelligence (A.I.) now offer a decision tool that looks at all the patterns that precede or surround previous pipe failures and predict the likelihood of failure for all pipes. Artificial Intelligence is software that can “learn” from data to predict future events. It offers utilities a powerful way to discover optimal solutions to the challenge of pipe failures and remaining life. Using available data, this software can find patterns and hidden insights to make informed, proactive choices to avoid costly bursts as well as premature pipe replacement. A common application of A.I. is financial credit scoring. Imagine a set of your water main pipes about to default on their responsibility to deliver water.
For water utilities, finding the likelihood of failure (LoF) for every active pipe in inventory is valuable. Pipes identified with a high LoF can be overlaid with consequence of failure (CoF) costs to help utilities make the best asset management decisions. These include inspection, monitoring, condition assessment or renewal as appropriate. The probability analysis enables effective prioritization that can avoid both catastrophic breaks and premature replacement.
The application of artificial intelligence and its offspring “machine learning” to predict likelihood of failure for pipes is now available for water utilities globally. Two companies (Voda and Fracta) compete in this space today. Both combine utility data on pipe inventory and failure history. Voda also uses publicly available data on soil, weather, seismic and other data to find hidden insights that project a likelihood of failure of each pipe in the next year. Combining these various data elements with proprietary “interaction variables” to build predictive models has demonstrated remarkable accuracy.
Artificial intelligence applications are ideal solutions for utilities trying to decide which water mains to monitor in the short term (within the next 12-24 months) in order to address the potential of catastrophic failures, and in the longer term (the next 3-10 years), for capital investment planning.
A case study validated their methodology and found the results could have helped prevent 10 out of the 18 breaks that took place in a pilot area. While capital investment is still required for pipe replacements, a more informed choice of which pipes to replace can help prolong the life of the system, avoid replacing pipes with remaining life and avoid catastrophic breaks.
While utilities could develop their own artificial intelligence capabilities internally, they will sacrifice a critical advantage of this application: combined data from multiple utilities. Artificial Intelligence benefits enormously from more and more data. It gets “smarter” as data from utilities is combined. This is not an easy technology for a utility to develop. It requires specialized knowledge and hard-to-find expertise. Companies that specialize in A.I. have the benefit of solving this problem for multiple customers and build deep experience, as well as combining data and learning from a multitude of regions and utilities.
The challenge of using A.I. is the quality and quantity of each utility’s digital pipe inventory and failure data. With new data science methods that can clean data and as more utilities embrace digitization and geospatial tracking, this challenge will be reduced. It is more important than ever for utilities to capture and store comprehensive digital information on their inventories and events.
Jim Fitchett is an entrepreneur, management consultant, international speaker and a longtime management professor at Harvard University. He can be reached at firstname.lastname@example.org. To learn more about this topic, visit voda.ai.