针对采用工业内窥镜对燃气轮机涡轮叶片进行表面缺陷检测效率较低的问题,提出一种基于改进YOLOv8的涡轮叶片表面缺陷检测算法。首先,在主干网络中引入可变形卷积,设计了C2f_DCNv3模块,以提高模型的特征提取能力；其次,加入Shuffle Attention注意力模块,进一步提升模型检测精度；此外,在特征融合部分设计了GSCSP模块,降低模型的计算量并提高对小目标的检测能力；最后,建立了涡轮叶片表面缺陷数据集,用于模型的训练与测试。实验结果表明,改进后的模型对于缺陷检测的平均精度均值达到92.3%,比YOLOv8原始模型提升了2.1%,且参数量和计算量分别下降了13%和17%。在与其他算法的比较中,本研究的算法也取得了更好的检测效果,更适用于对涡轮叶片表面缺陷的检测。

In order to solve the issue of low efficiency in surface defect detection of gas turbine blades by using industrial endoscope, an algorithm based on improved YOLOv8 is proposed for turbine blade surface defect detection. Firstly, deformable convolutions are introduced into the backbone network, and the C2f_DCNv3 module is designed to enhance the model’s feature extraction capability. Secondly, the Shuffle Attention module is added to further improve the detection accuracy of the model. In addition, the GSCSP module is designed in the feature fusion part to reduce the model’s computation and enhance its detection capability for small targets. Finally, a dataset for turbine blade surface defects is established for model training and testing. Experimental results show that the improved model achieves a mean average precision of 92.3% for defect detection, which is a 2.1% improvement over the original YOLOv8 model. Moreover, the number of parameters and computation are reduced by 13% and 17% respectively. In comparison with other algorithms, the proposed algorithm in this study achieves better detection results and is more suitable for detecting surface defects on turbine blades.