Wind turbine blade damage identification using lightweight YOLO11 with multi-scale dilated attention

In recent years, companies such as Siemens and GE have frequently experienced wind turbine shutdowns due to blade damage and fractures, resulting in annual losses of up to $4.6 billion. As a critical component of wind turbines, blades have a high failure rate, which directly affects the safety and reliability of wind turbine operations. However, accurately identifying wind turbine blade damage is a significant challenge, mainly due to the scarcity of damage samples, the small scale of blade damage which is difficult to detect, and the interference from external environmental factors. To address these issues, this paper proposes a high-accuracy and low-consumption damage identification method using SlimNeck-structured YOLO11 with Multi-Scale Dilated Attention (SNMSDA-YOLO11). First, to overcome the issue of limited damage samples from wind turbines, three data augmentation methods are used, including random noise augmentation, image sharpening augmentation, and saturation adjustment augmentation. Second, to solve the problem of adaptive feature extraction for damage at various scales, the multi-scale dilated attention mechanism is integrated into the YOLO11 algorithm. This enhances its ability to extract damage features across different forms of blade damage, thereby improving detection performance. Finally, a slim-neck structure is employed to optimize the model, ensuring accurate damage identification while significantly improving the computational speed and efficiency. Experimental results demonstrate that the proposed SNMSDA-YOLO11 effectively utilizes attention mechanisms to learn multi-scale damage features of wind turbine blades, achieving an accuracy of 94.9 %. Compared to existing methods, this method holds significant potential for large-scale applications in wind turbine blade damage identification. • The proposed SNMSDA-YOLO11 method effectively identifies blade damage. • Data augmentation algorithm effectively expands blade damage samples. • The MSDA module enhances feature representation of blade damage. • The SlimNeck structure improves inference speed and computational efficiency. • Experimental results show that the proposed method outperforms other methods.