Can Transmission Line Inspection Be Completely Automated?

Two EPRI studies suggest that large-scale automation of transmission system inspection is feasible in the near future.

Increasingly, utilities inspect transmission systems by deploying drones to capture high-definition images of conductors, insulators, and other components. In most applications today, humans operate the drones and evaluate the images. By making drone-based inspections cheaper and less time-intensive, automation can drive widespread adoption.

In the first study, EPRI tested several low-cost, commercially available drones at a Georgia field site with energized and non-energized transmission structures. The drones autonomously followed complex, three-dimensional flight plans and captured images without striking the structures. Existing LIDAR data on the structures proved critical to successful flight planning. However, improvements are needed in image perspective and navigation.

In the second study, researchers examined the ability of nine machine learning technologies to evaluate thousands of images and identify defects in structures. Their performance was well above random prediction but well below human analysis. Larger image data sets are needed to train the technologies and improve their effectiveness, and researchers are collecting and curating additional images from utilities.

“To be successful, this research requires collaboration among utilities to pool and curate data and to test machine learning technologies,” said EPRI Senior Technical Leader Dexter Lewis. “Building an effective AI transmission and distribution inspection system is a challenge that is likely too large for any single utility, but together the industry can make a difference.”

EPRI plans to continue examining drone automation on energized transmission structures. EPRI has also established a broad AI initiative to facilitate collaboration among utilities, develop a research roadmap, collect and curate data sets for diverse applications, and educate electric power industry stakeholders.

---