‘Smart image database’ for energy asset fault recognition
Image: Elia/SID
The ‘Smart image database’ project is advancing to develop AI models for fault recognition on energy assets such as pylons.
The project, initiated within the Cross-Industry Innovators ecosystem of European TSOs, is aimed to pool images from the participating companies into a single repository for the training of AI models, with over 90% accuracy achieved to date.
Traditionally, monitoring of the state of the grid and assets was done manually, but increasingly today, drones and robots are being used to collect images for analysis.
By pooling these images, a more comprehensive database is created than is possible from individual collections, which also may lack images of particular assets, enabling greater accuracy in the training of the AI models.
Have you read?
SP Energy Networks to harness AI for weather-related fault prediction
AI in the energy sector – growing energy demand but potential to transform it
The project was launched in 2023, with the Hamburg-based innovation mediator Infront Consulting as the project coordinator.
To date with the first two phases complete, the initial dataset now contains 12,000 inspection images, from which a curated dataset including 3,000 labelled images representing 280,000 samples or labels has enabled comprehensive training of AI models.
As a result, the number of detection and classification models has increased to 25 detection models and 13 classification models, achieving an accuracy of over 90%.
In the final phase during 2025, the project is being launched on the Kubernetes open source platform to enhance its scalability and security.
Other plans include creating a new version of the dataset that includes more assets identified by AI models and exploring synthetic data, i.e. AI-generated images of damages, to further optimise model performance.
For the project consortium, the project is considered already to have yielded significant results and consideration is being given to how to sustain the outcomes beyond its conclusion.
The members of the project consortium are the Elia Group’s 50Hertz (Germany) and Elia (Belgium), Amprion, Netze BW and TransnetBW (Germany), Austrian Power Grid (Austria), Red Electric Elewit AI company unusuals (Spain), Swissgrid and the transport operator SBB (Switzerland) and TenneT (Netherlands).
