RCS-YOLOv8: an improved YOLOv8 for wind turbine blade defect detection

Area of Science:

• Renewable Energy Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Automated detection of wind turbine blade surface defects is crucial for operational safety.
• Challenges include illumination variations, complex backgrounds, and diverse defect scales in real-world wind farms.

Purpose of the Study:

• To develop an optimized model for accurate and robust automated detection of wind turbine blade surface defects.
• To address the limitations of existing methods in complex environmental conditions.

Main Methods:

• Proposed RCS-YOLOv8 model integrating RepViT backbone for enhanced feature perception.
• Incorporated Channel-Spatial Cooperative Attention Module (CAFM) and SlimNeck with VoVGSCSP for feature fusion.
• Utilized WIoU loss function for improved bounding box regression and recall of difficult defects.

Main Results:

• RCS-YOLOv8 achieved 90.3% accuracy on a wind turbine blade defect dataset with seven categories.
• Outperformed baseline YOLOv8n by 5.8% and RT-DETR by 3.0%, with the highest recall at 87.5%.
• Ablation studies and visualization confirmed module effectiveness and accurate defect localization with suppressed background interference.

Conclusions:

• RCS-YOLOv8 provides a reliable and effective technical solution for automated wind turbine blade defect detection.
• The model's multi-module collaborative optimization enhances robustness and accuracy in complex environments.
• This advancement contributes to ensuring the operational safety of wind power systems.