Context

Johan Maricq

Artificial Intelligence

To ensure reliable transmission of electricity, Elia performs visual inspections of its infrastructure. When it comes to towers and overhead lines, Elia usually combines 3 types of inspections: patrolling on the ground, climbing the towers or using a helicopter to fly over the lines to obtain a full perspective on the asset’s conditions. Climbing, however, is a considerable safety risk for the person performing the inspection and is also a cumbersome activity. Using helicopters also does not provide the same detail one would get from a climbing inspection. While performing those inspections a quantity of pictures and data is collected that need to be analysed to initiate some actions.

Approach

With this project, Elia wants to test how drone technologies can improve the quality and efficiency of this very important task and at the same time avoid risk for our colleagues. Two aspects are being tested:

 

  • Generation of an auto flight plan: At present, a drone pilot needs to carefully operate the drone around the infrastructure while someone else is taking pictures from the camera attached to the drone. With this project, an application will automatically control the drone around the tower and will take pictures at fixed positions, with fixed angles.
  • Post-processing of the collected pictures: As mentioned above, inspections of our lines generate a number of pictures. To facilitate all this data, as well as optimizing its potential use, new post-processing techniques and data management systems will be needed. One of the most promising techniques available today is the use of Artificial Intelligence to automatically recognize defects.

Many companies have already demonstrated the technical feasibility of these processes. Our goal is to determine the maturity and scalability of the tools available on the market today.

Description

Collecting training data

Before a machine can recognise faults automatically, it needs to be trained based on examples of those faults. To this end, the first step of the proof of concept will be to capture training data. Elia’s trained drone pilots will perform the capture, assisted by an automated flight plan for the drone. Thereby, the image capturing will be highly accurate, standardised and it has the added benefit that the inspection will be performed even faster.

Labelling and identifying data

Afterwards, the collected data needs to be labelled before being fed into the machine. Present faults need to be highlighted and put into a category (e.g. corrosion). This allows us to train a neural network to perform the identification automatically in the future. This means that on the next flight, the pilot could upload the collected images into the machine, and these will automatically be classified according to the appropriate fault and linked with the corresponding asset. The entire process becomes more automated and eliminates tedium in the current processes.

Partners

Sterblue

Visit website

Sky-Futures

Visit website

Interested in becoming our next partner?

Partner with us