Unmanned Aerial Vehicles and Artificial Intelligence Technologies as a Tool for Automating of Thermal Power Plant Boiler Monitoring

Approaches to creating an automated system for monitoring the inner surface of a thermal power plant boiler using small-sized unmanned aerial vehicles, neural networks, and computer vision technologies are considered. Analysis of visual defects in boiler tubes made it possible to identify five main types of defects, for each of them datasets are created using augmentation and synthetic data generation procedures. Three neural networks (YOLOv4, DetectoRS, and DCN) are trained using the datasets. Their characteristics are determined experimentally, and a comparative analysis of the reliability and speed of defect detection is carried out.