Drone and helicopter inspections are already a reality when Energinet checks electric components for rust and other possible defects. The pictures are examined to assess whether there could be defects in, for example, electricity pylons.

The need to make the process of manually checking pictures easier nevertheless arises now because a large number of cables and pylons from the 1960s and 1970s must soon be checked thoroughly for any need for renovation. This means that there will be a huge number of pictures that need to be scrutinised, and it will thus be highly useful to automate the process, if possible.

Digital model can ‘detect’ defects

Pictures are data, and data can be put into algorithms. There is therefore a very promising potential in examining how pattern recognition software can be used to ‘detect’ defects or repair needs in the various electric components.

On a trial basis, Energinet is therefore supplying a large quantity of data – pictures of components with rust or other defects – to a Norwegian company, which then works on creating the digital model that, in the future, will hopefully be able to ‘see’ when there is a need for maintenance. The pictures have been chosen by Energinet on the basis that it must not be too easy for the algorithm – for example, there are also pictures of components on which it is difficult to see the rust because there are trees and leaves in the background.

Technology not completely ready

The idea is that, in the future, it will only be necessary to look at a much smaller number of pictures based on a selection made by the sorting algorithm. Optimally, the trial will make it possible to create an intelligent electricity grid that can auto-report the need for maintenance of a pylon or cable. 

“The challenge is that the pictures must be sharp and the areas with rust must be marked for the digital algorithm to be able to learn and, in the long term, assess whether a cable is rusty,” says Lars Rasmussen, Senior Engineer in Energinet.

“At the same time, the construction of both the algorithm and the database which will form the basis from which the algorithm can learn to ‘see’ rust, takes a long time. The technology is not completely ready yet. But the method can potentially be used to check all our material to provide us with knowledge about the need for repairs, both quicker and easier.”

This picture is an example of how an algorithm will deliver a picture to our technical experts, who can assess whether there is a defect. The algorithm can thus both show where there are defects and deselect the pictures which do not show any defects.