Presentations

Abstract

Reliable inspection and maintenance of wind turbine blades (WTBs) are critical to ensuring turbine efficiency, preventing failures, and reducing downtime. Drone-based visual inspections provide large volumes of high-resolution imagery, making manual analysis costly and time-consuming. This work presents a new generative AI-based solution for automated WTB image segmentation—the process of automatically identifying and separating the blade region from the background—which enables subsequent assessment tasks such as blade defect detection and global damage localization. We leverage pretrained universal generative models—trained without labels on billions of natural images—and adapt them to the wind turbine domain with only limited annotated data. To further enhance performance, we introduce a dual-space augmentation strategy that enriches the dataset by combining and interpolating existing images in the input space while applying probabilistic augmentation in the model’s latent space. This approach improves robustness to variations in windfarms, lighting conditions, and drone configurations. Evaluated on multi-site datasets, our method achieves segmentation accuracy above 99%, consistently outperforming state-of-the-art models, including SAM2 from Meta. Accurate and automated identification of blade regions through image segmentation improves the reliability of defect detection and localization, reducing the risk of missed damages. Furthermore, its lightweight model solution could be incorporated into drone-based inspection pipelines, supporting real-time analysis in the field. Overall, the solution contributes to the automation of wind farm operations and the reduction of O&M costs, making large-scale deployment of drone-based assessments more feasible. This contribution demonstrates how general-purpose generative AI models can be effectively adapted to domain-specific challenges in wind energy operations, offering a scalable and practical tool to boost the safety, efficiency, and sustainability of wind turbine maintenance.