Research on Detecting Bearing Defects Based on Improved YOLOv8

To improve the speed and accuracy of wind turbine bearing end-face defect detection, this paper proposes an improved CREL-YOLOv8 algorithm. The algorithm introduces the RepViT-EMA module into the YOLOv8 backbone network to optimize the C2f structure and adopts the Large Separable Kernel Attention (LSKA) module in the SPPF. Comparative experiments show that the improved algorithm increases mAP by $\mathbf{1. 7 \%}$ in wind turbine bearing defect detection, validating the effectiveness of the improvements. This research helps enhance the reliability and efficiency of wind power equipment and is of significant importance to the energy and power field.