The use of drone technology to inspect power lines for damaged components has become increasingly popular. While drones take images efficiently, manual detection of damaged components in each image still requires laborious work. In this session, presenters will explain the potential of artificial intelligence (AI) to inspect damaged components of power lines. They will discuss how real images collected by Eversource Energy’s drone team were sorted by types of damage and divided into training, validation, and test datasets, and then used train an AI model. Test results provide numerous valuable insights to increase the accuracy of AI-enabled power line damage inspection.

Key Takeaways

• The potential of artificial intelligence (AI) to inspect damaged components of power lines
• How to train AI model using a GPU device running on Microsoft Azure.
• How to improve model performance from different aspects, including architecture, loss function, and data labeling.