Developing Drone-Based AI for Defect Detection in Transmission Tower Bolts-and-Nuts

To improve efficiency, we use drones to inspect about 400 towers per year. However, the captured images still require manual verification to ensure there are no defects, such as missing or loose bolts-and-nuts. This confirmation process demands significant time and effort.

To address this challenge, we have developed an AI that automatically detects defects—such as loose or missing bolts-and-nuts—from drone images. This AI introduces a new approach that eliminates the need for manual visual checks. To enhance accuracy and prevent oversight, we capture high-resolution grid images of towers from six different directions.

Currently, the AI achieves approximately 90% detection accuracy. About 10% of errors result from image deficiencies, including dark, blurry, or obscured areas, and poor quality from side angles. Introducing this AI is expected to significantly reduce costs, as image inspections previously performed by humans will now be automated.

For example, assuming that a human inspector spends one second inspecting each bolt, the expected time saving per tower is 140 to 475 minutes, depending on tower size. In addition, we are developing a “report generation function” to streamline documentation.