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

How machine learning models speed up and improve inspections in uBird

Specialized power line inspection platform for grid operators

Hepta Airborne’s power line inspection platform uBird is designed to ease the burden of grid operators ensuring a stable power connection to everyone connected to the grid. Instead of relying on the hard manual labour of on-foot patrols or imprecise and costly work of helicopter-based inspections, Hepta’s clients gather detailed information about their assets with drones. The collected data is then uploaded to uBird, which allows for comprehensive analysis, GPS based reporting and direct work assignments for work teams. The combination of highly detailed data gathering with drones and analysis in a specialized inspection platform allows grid operators to get an in-depth overview of the state of their assets and their most critical parts, while also making well-informed decisions based on the reporting.

GPS-based map overview of defects in uBird

Machine learning in uBird

One of the crucial parts of uBird are the machine learning models that help users automatically find defects in the gathered data. They enable users to conduct large scale analysis of data in minutes and detect more defects in the process. Machine learning models are always customer-specific and based on their asset data, thus ensuring a high degree of accuracy. Instead of teaching the machine learning models to detect general classification, such as insulators, they are focused on the exact model of insulators used by the customer. This helps to find more relevant defects while also avoiding any misses that could occur with general classification.

Machine learning algorithm annotation suggestions displayed in uBird

When running the machine learning models on the customers’ dataset, all images containing defects are marked by the models. After that, the user can see the images marked with defects and how sure the machine learning models are in the accuracy of those defect types. The machine learning models can analyse thousands of images in a matter of minutes, thus offering valuable insights quickly. They also improve the quality of inspections, constantly detecting defects that would have otherwise been missed.

Sample of defects found in overhead power lines by AI-models

Sample of defects found in overhead power lines by AI-models

Machine learning models influence on the speed and quality of inspections

As the machine learning models are used by Hepta’s clients on a daily basis and their performance is an integral part of the platform, their results are constantly monitored. Since implementing different machine learning models to uBird, we have seen significant changes in the analysis work conducted on the platform. Tens of thousands of defects have been detected by machine learning models, helping customers to take better care of their grid and avoid outages. Here’s a short overview of the most significant improvements seen in the platform and brought out by clients’ feedback, that stem from the use of machine learning models:

  • Up to 75% of defects are marked by customer-specific machine learning models
  • Customers using machine learning models speed up their analysis processes by 30%
  • Analysis work with the aid of machine learning models is less mentally straining for the users
  • The use of the machine learning models reduces manual mistakes significantly
  • Machine learning model supported analysis is the preferred way of work for Hepta’s customers, with multiple new models in development  
Power line inspection challenges

Power line inspection challenges and how to overcome them

Learn what are the most common power line inspection challenges grid operators face and how they are overcome with Hepta Airborne's help.

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

Get in touch