基于ResNet50-FPG模型在管道焊缝缺陷识别中的应用  

Application of ResNet50-FPG Model in Identification of Weld Defects in Pipelines

作　　者：卫小龙 余泽禹 WEI Xiaolong;YU Zeyu(Intelligent Manufacturing College of Jingzhou Institute of Technology,Jingzhou 434020,Hubei,China;Information and Communication Engineering College of Jingzhou Institute of Technology,Jingzhou 434020,Hubei,China)

机构地区：[1]荆州职业技术学院智能制造学院,湖北荆州434020 [2]荆州职业技术学院信息与通信工程学院,湖北荆州434020

出　　处：《焊管》2025年第1期55-60,71,共7页Welded Pipe and Tube

基　　金：荆州市2023年度科技计划基金项目“基于超声导波的管道焊缝缺陷检测与类型识别研究”(项目编号2023EC36);荆州职业技术学院重点科技创新成果培育工程项目“基于超声导波的焊接管道缺陷检测及定位研究”(项目编号jzzp202302)。

摘　　要：为了解决管道焊缝缺陷检测问题,提出了基于超声数据处理和残差神经网络(ResNet)优化的综合方案。介绍了超声数据采集和转换为图像的方法,详细阐述了ResNet网络模型的剪枝优化过程,以提高模型的推理速度。通过消融试验,验证各种改进方法对模型性能的影响。结果表明,ResNet50-small-FPGM FilterPruner模型在综合准确率、F1值和推理速度方面表现最优,适合实时管道缺陷检测。此研究为提高管道焊缝缺陷检测的效率和准确性提供了一种有效途径,具有重要的实际应用价值。In order to solve the problem of pipeline weld defect detection,a comprehensive scheme based on ultrasonic data processing and residual neural network(ResNet)optimization was proposed.The method of ultrasonic data acquisition and image conversion is introduced.The pruning optimization process of ResNet network model is described in detail to improve the inference speed of the model.The effect of various improved methods on the performance of the model was verified by the ablation test.The results show that the ResNet50-small-FPGM FilterPruner model has the best performance in terms of comprehensive accuracy,F1 value and inference speed,and is suitable for real-time pipeline defect detection.An effective way is provided to improve the efficiency and accuracy of pipeline weld defect detection,which has important practical application value.

关 键 词：管道焊缝 缺陷识别 残差神经网络 消融试验 

分 类 号：TG441.7[金属学及工艺—焊接]