News about digitizing infrastructures and automating utility inspection with drones and AI

Latest AI-models added to Hepta’s power line inspection platform uBird

For years, Hepta has developed and continuously improved its machine learning models in the uBird platform. The machine learning models are designed to ease the burden of grid operators’ inspection and analysis efforts. They allow grid operators to:

  • analyse thousands of images in a matter of minutes
  • improve the quality of inspections
  • find more relevant defects
  • increase the speed of the analysis process
  • and lower the overall cost of inspections

Today, machine learning algorithms are widely used by Hepta’s clients to inspect thousands of power line images from four continents and by constantly growing number of grid operators.


New AI-models

Due to the success of the machine learning models performance in power line inspections, new ones are constantly added to uBird. The latest additions are focused on detecting crooked cross-arms from power line images.

Normal cross-arm on a power line pole

Normal cross-arm on a power line pole

Cross-arms are used to fix the overhead line wires securely to the power line pole. The cross-arms get crooked most often either due to large branches or trees falling on lines or by corrosion effects on its structure. Crooked cross-arms might lead to wires coming into contact with each other or with the surrounding vegetation, thus posing both a blackout and a wildfire threat to the grid operator.

Example of acrooked cross-arm on a power line

Example of a crooked cross-arm on a power line

To help grid operators to detect crooked cross-arms two new machine learning models were created – one for normal, well-maintained cross-arms and one for crooked ones. Both models were extensively trained on a dataset with more than 10 000  images. The end result shown in testing for both of the models was exemplary:

  • 95.00% precision rate
  • F-score higher than 94%
  • and 95% recall rate

Yet these are just the first results of machine learning models, that were recently created. As time passes and the number of analysed images grows, the models performance will only become better. We fully expect the new models to become an integral part of analysis toolset in the future.

Should you have any questions about the machine learning models or are interested in applying them to your inspections, feel free to contact Hepta’s experts!


Myths and realities: precision and accuracy of drone-based power line inspection

Myths and realities: precision and accuracy of drone-based power line inspection   Hepta’s industry survey revealed that some DSOs and TSOs are wary of adopting drones for power line inspection purposes because they believe inspection with drones produces less precise and less accurate data compared to traditional inspection methods.    Feedback from Hepta’s existing customers […]

Contact us to learn more about digitizing infrastructures and conducting utility inspection with drones and AI

Get in touch