Wind Turbine Blade Defect Detection Based on the Genetic Algorithm-Enhanced YOLOv5 Algorithm Using Synthetic Data

Abstract

Regular inspection and maintenance of wind turbine blades are crucial to effectively avoid potential structural failures. By utilizing drone inspection shots, a substantial number of high-resolution images of wind turbines can be obtained. This experiment involves data preprocessing, including image enhancement and manual annotation of wind turbine blade defects in these images. Wind turbine blade defect detection is then performed using YOLOv5. The experimental results demonstrate that the model can accurately predict the location and class of blade defects with nearly human-level accuracy. To further augment the model's capabilities, Genetic Algorithm was applied to fine-tune YOLOv5’s hyperparameters. The original YOLOv5 version achieved an accuracy of 85%, while our method achieved 89%. Additionally, the Unity engine was leveraged to simulate real-world environments and create a synthetic dataset impervious to variations in weather, lighting, and camera angles, thereby enhancing data diversity and quantity. Our method achieved an accuracy of 88%, compared to 85% when using real datasets alone. These innovative approaches significantly enhance the precision and robustness of wind turbine blade defect detection.