AI & Cross Bores – How PG&E Leverages AI in Sewer Cross Bore Safety Program
For more than 40 years, horizontal directional drilling (HDD) has been a valuable trenchless technology for installing small diameter underground utilities in densely populated areas with minimal disruption. However, because this method involves drilling and installing hundreds of feet of underground pipe at a time, it can lead to accidental conflicts and damage to existing utilities, also known as “cross bores.” In North America, side sewer laterals are one of the primary types of utilities at risk of cross bore from HDD.
To address safety concerns caused by utility cross bores, most gas utilities have dedicated significant resources to CCTV sewer inspection programs over the past 10 years, and Pacific Gas and Electric (PG&E), headquartered in California, which owns approximately 43,700 miles of gas distribution facilities, has been at the forefront of cross bore safety, having completed over 250,000 inspections since 2013, when it began its Cross Bore Mitigation Program.
“The program uses video camera inspections to verify that no damage has occurred to sewer lines when using trenchless construction methods. The goal of the program is to identify cross bores by completing inspections of potential conflict locations and repairing all occurrences as they are discovered,” says Aaron Rezendez, manager of Gas R&D, Utility Partnerships & Innovation at PG&E.
According to its 2022 Gas Safety Plan, PG&E has averaged more than 28,000 inspections completed throughout California each year for the previous nine years, finding and addressing cross bore conflicts in just under 0.3 percent of sewers. The amount of CCTV data generated in this project has significant, consisting of more than 30,000 hours of video, and exceeding 22.5 TB of storage space. This is a similar amount of video to the entire catalog of the numerous movies, TV shows, and documentaries currently hosted on Netflix, and to put this in perspective, it would take one person over 15 years to watch this amount of video (if working standard 40-hour weeks).
One of the biggest problems faced by PG&E and other gas utilities making good-faith efforts to mitigate potential risk to communities posed by HDD construction activities is ensuring review of the CCTV data is performed accurately, consistently, and at the lowest possible cost.
To put it mildly, this is a difficult task when such a critical part of the project involves people needing to view, interpret, and annotate each inspection. People in the CCTV inspection space have long been aware of challenges in manual sewer assessment in regards to the tedious nature of the work, the risk of technicians missing important defects in pipes, and the overall subjectivity involved in the process.
As its Cross Bore Mitigation Program continued to grow, many folks at PG&E started to look at innovations in the AI computer vision space as a potential solution to reduce risks posed by manual data review.
Enter SewerAI, a Walnut Creek, California-based technology company that has been applying AI computer vision and cloud computing tools to solve problems in the sewer inspection space since 2019. Its AI tool, called AutoCode, automatically detects conditions in sewer inspection videos at a rate of tens of thousands of feet per hour while adding a level of quality assurance never before seen, while also enabling CCTV operators in the field to conduct their inspections far more efficiently, and at less cost.
Since late 2022, a groundbreaking project has been under way where SewerAI has partnered with PG&E to assist in its cross bore safety program through a multi-year contract conducting AI-powered “cross bore safety audits” of approximately 10,000 lateral CCTV inspections per year, focusing at this initial stage on PG&E’s projects in Northern California.
These audits involve analyzing video footage of sewer laterals using its AutoCode technology for signs of potential cross bore hazards. The auditing process is then followed up with an automated alert system from its cloud-based platform, PIONEER, notifying project stakeholders of any issues with the data, and any potential utility conflicts.
This project is being carried out in coordination with four different sewer services contractors, making it a complex undertaking, but already showing promising results. For example, prior to this, PG&E’s data management process required contractors to burn DVDs of all inspections for storage at a physical PG&E location. However, with SewerAI’s help, the utility has adopted a more efficient and scalable approach utilizing cloud-based storage, active alerts, and AI-powered analysis.
This project is the result of a two-plus-year evaluation process, where PG&E tested and measured the accuracy of SewerAI’s tools and processes.
In January 2021, SewerAI analyzed a dataset of 1,040 lateral inspection videos to see how well its AutoCode AI model could detect cross bores conflicts between gas distribution lines and side sewer laterals. The dataset contained 40 videos with known cross bores, with a total playback time of 64 hours.
The AI tool, combined with review by data technicians, was able to correctly identify all 40 cross bores. To boot, the use of AI reduced the review time from 64 to 19 hours, or approximately 30 percent of the original time. Manual review of the 19 hours of footage took about 32 hours, so using the AutoCode tools reduced overall review time by approximately 70 percent compared to manual review.
By leveraging the power of AI and big data analytics, SewerAI is helping utilities like PG&E take a much more proactive approach to safety and infrastructure management.
“Our goal is to help our clients mitigate risks and save lives. By combining the latest advances in AI and machine learning with expert knowledge of sewer infrastructure, we’re able to provide insights that can make a real difference in people’s lives,” says SewerAI CEO Matt Rosenthal.
Eric Sullivan is director of business development at SewerAI.