For entrepreneurs with a provocative vision for energy & IT; NGP is the only utility venture & innovation fund that taps into the global network of National Grid, accelerating companies with investment, innovation, incubation & biz dev; realizing the full potential of energy.

National Grid Partners

The integrity of gas distribution networks relies heavily on the fusions that join pipelines together. Internal defects during the fusion process may lead to pipeline failures that are slow to manifest but pose significant costs and safety risks. Ensuring fusion integrity requires time-consuming, expert work: visual inspections, preventative excavation, testing and, in some cases, replacing pipes that on further review turn out to be sound. Early detection of defects, therefore, is critical. 

New sensing technologies have potential to aid in these inspections by identifying issues not visible externally. One leading example is X-rays, which produce images capable of highlighting internal defects—but which still require a trained eye to identify them. 

The Challenge

By training artificial intelligence models on X-ray data reviewed by subject matter experts, we can turn a complex inspection into a simple pass-fail decision in the field—catching defects before gas ever enters a new pipeline.

That’s why National Grid launched a two-year research initiative leveraging a computer vision platform developed by Cogniac, a Silicon Valley startup that’s part of the National Grid Partners portfolio. The utility’s corporate venture capital unit identifies, invests in and helps introduce emerging technologies that can help National Grid boost efficiency and reliability while cutting costs and speeding service to customers. (As part of this initiative, National Grid collaborated with Cogniac both as an operational partner and as an equity investor.)

Knowing a wide range of sample data would best train Cogniac’s artificial intelligence model, Lessard reached out to NYSEARCH, a consortium of 25–30 utility companies who share expertise, project costs, and resources. For this initiative, National Grid led nine utilities to co-fund and collaborate—ensuring shared access to the research findings while respecting regulatory restrictions on data sharing. 

Operating Context 

The initial phase of research generated about 2,000 X-ray images, using samples with known defects collected from the utility partners. For this proof of concept, Cogniac trained separate AI models for each defect type. 

For example, one model is designed to detect grease contamination in images. Training the model took 19 days, after which it could identify defects in a validation set with 90 percent accuracy. The takeaway: With a relatively small number of images, the system can reliably spot defects in a laboratory setting.

Human feedback continued to train Cogniac’s model, so it quickly became even more accurate. Within six months, the platform could reliably recognize a range of specific defects—from “short stabs” (a gap between a pipe’s top end and the fitting) to “excessive peel” (a surface defect that weakens the pipe fusion).

The Innovation 

Once the system starts learning, the momentum is real. You quickly move from experimentation to meaningful impact. What’s exciting is not just the accuracy we’re seeing today but how rapidly the models continue to improve.

Human experts are providing critical guidance and review of the AI model’s outputs and will continue to supervise the process of defect detection and response. But Pattanayak and Lessard agree the project shows the potential value AI-enabled technologies like Cogniac’s can bring to utility inspections and field operations—while also noting independent validation remain critical. 

Later this spring, the team plans to kick off an 18- to 24-month second phase that will test the platform with other kinds of pipe fusions and under simulated field conditions. The eventual goal is to evaluate the potential for commercialization, subject to further testing and regulatory considerations.  

“Wide-scale use of the application is a ways off,” Lessard noted. But he’s excited about “the opportunity to build quality into the fusion process to ensure fewer failures.” 

National Grid, through its corporate venture capital arm National Grid Partners, holds a minority equity investment in Cogniac. The collaboration described above was conducted in accordance with applicable governance, regulatory, and data‑sharing requirements.

What’s Next 

Technology/Tool: Cogniac’s computer vision platform 

Application: Automated detection for defects in gas pipeline fusions using X- ray imaging 

Business Unit: National Grid’s Gas Innovation Group in collaboration with NYSEARCH, an R&D program serving gas utilities.  

Status: Phase 1 complete; Phase 2 planned 

How Cogniac’s Computer Vision Helps Utilities Improve Pipeline Reliability