【摘要】 针对当前管道失效分析的快速准确要求，提出了一种基于YOLO算法的裂纹扩展方向识别模型。选用开源数据集Crack-Seg和实际管道环焊缝裂纹数据作为模型训练数据集，初步获取YOLO算法框架下的识别模型。后通过对实际管道环焊缝裂纹形态和服役环境的观察研究，引入具有物理性约束的辅助特征以完善模型。经实例对比验证，引入特征的模型在裂纹源区和扩展方向的识别上具有更为出色的表现效果，表明特征引入对模型精度存有显著的改善作用。更多还原

【Abstract】 In response to the demand for rapid and accurate pipeline failure analysis,this paper proposed a recognition model based on the YOLO algorithm for identifying crack initiation zones and propagation directions. The open-source datasets( Crack-Seg) and actual pipeline girth weld crack data were selected as the model training dataset to preliminary obtain the recognition model based on the YOLO algorithm framework. Subsequently,by observation and analysis of the crack morphology and service environment of actual pipeline girth weld,auxiliary features with physical constraints were introduced to improve the model. Through comparative verification of examples,the model with introduced features has better performance in identifying crack initiation zones and propagation directions,indicating that feature introduction has a significant improvement effect on model accuracy.更多还原