"We didn't even know this manhole was there, but it has now been detected in real time." (SKT researcher)
As an SK Telecom work vehicle drove along a road in Seongnam, Gyeonggi Province, its onboard camera spotted a manhole. A little further on, a construction truck and an excavator also appeared on the screen. This was the scene where SKT's AI-based communication network safety inspection solution 'Tokda', unveiled on the 1st, scanned the surroundings of the vehicle and identified potential hazards.
Tokda is a native Korean word meaning to thoroughly search. It is a technology that mounts an AI camera on work vehicles to inspect areas around communication networks. As the vehicle moves, it captures footage of its surroundings, which AI then analyzes. The system identifies 11 potentially problematic elements, ranging from heavy machinery like excavators (Gong (President)) to sagging communication lines.
During a pilot operation with six vehicles in the Seoul metropolitan area from May to August, the system inspected at least once over half of the major wired network sections. In this process, it discovered 17 risk factors that could lead to accidents and took preventive measures for three of them. SKT plans to expand its inspection coverage to 70% by July 2027 and over 90% by 2028 by increasing the number of vehicles.
The reason for utilizing AI is that there are simply too many communication facilities for humans to inspect one by one. SKT manages approximately 200,000 kilometers of optical fiber cables—a distance sufficient to circle the Earth five times. There are 2.35 million utility poles (Jeonju) with communication lines installed and 660,000 cable termination boxes connecting the optical fibers.
In particular, if an excavator cuts through underground optical cables during road construction, it can lead to large-scale communication disruptions. Lee Sang-jin, team leader of SKT's Wired Facility Solutions Team, explained, "If an excavator severs an optical cable even once, the damage can be quite significant." He added, "This solution reflects a sense of urgency to quickly identify risks and prevent single-point failure accidents."
During the day, as the vehicle traveled 3.06 kilometers over approximately 15 minutes, Tokda analyzed 2,959 captured images and identified 44 items, including manholes and heavy machinery. When an excavator is detected, the system checks whether it is actually digging. The excavator spotted that day was deemed not dangerous because it was not in operation. The presence of construction equipment does not automatically mean danger; the system also considers whether actual work is underway and the location of surrounding communication networks.
SKT trained its AI on 22,000 field photos collected over the past five years. Through this process, the accuracy of identifying risk factors increased from 65% in 2024 to 88% this year. The target for next year is 95%.
SKT also plans to apply Tokda to communication facilities inside tunnels that are dangerous for humans to inspect directly. It is also considering expanding the solution to other industrial facilities such as gas pipelines, power lines, and oil/gas transmission pipes. Bok Jae-won, SKT's Network Operations Manager, stated, "Our goal is not to replace people. Instead, we focus on having AI complement inspection tasks that are dangerous for humans or difficult to verify in detail."
in TECH
