基于深度学习模型的油气管道焊缝缺陷智能识别  

Weld defect intelligent identification for oil and gas pipelines based on the deep learning models

作　　者：罗仁泽[1,2,3] 王磊[1] LUO Renze;WANG Lei(School of Computer Science and Software Engineering,Southwest Petroleum University,Chengdu,Sichuan 610500,China;State Key laboratory of Oil&Gas Reservoir Geology and Exploitation//Southwest Petroleum University,Chengdu,Sichuan 610500,China;School of Electrical Engineering and Information,Southwest Petroleum University,Chengdu,Sichuan 610500,China)

机构地区：[1]西南石油大学计算机与软件学院,610500 [2]油气藏地质及开发工程全国重点实验室·西南石油大学 [3]西南石油大学电气信息学院

出　　处：《天然气工业》2024年第9期199-208,共10页Natural Gas Industry

摘　　要：焊接技术在油气管道连接时广泛应用,确保焊缝区域可靠是保障油气管道安全运行的关键。受工艺和技术制约,油气管道焊接过程中可能出现不同类型的焊缝缺陷。针对油气管道焊缝部分缺陷尺寸小、缺陷与背景差异性较小导致焊缝缺陷识别效果不理想、人工识别工作量大等问题,提出了基于SCT-ResNet50模型的管道焊缝缺陷智能识别新方法。首先将焊缝区域图像输入特征提取网络;然后在特征提取的浅层使用SCC(Spatial Channel Context)进行局部空间和通道信息融合,在特征提取较深的层次使用ECA-MHSA(Efficient Channel Attention-Multi-Head Self-Attention)来捕捉长程依赖和上下文信息;最后通过全连接层和Softmax得到最终的缺陷识别结果。研究结果表明:(1)该新方法在油气管道X射线图像焊缝缺陷数据集上缺陷识别准确率达到98.28%;(2)相较于ResNet50、VGG16、DenseNet121、MobileNetv3和EfficientNetv2分类方法,其准确率分别提高了3.05%、46.05%、28.99%、15.95%和18.84%;(3)在缺陷尺寸小、缺陷和背景差异较小的场景下,该新方法在油气管道焊缝缺陷识别中具有更高的准确率。结论认为,该新方法的优势在于结合SCC模块与ECA-MHSA模块学习图像的局部信息和全局信息,能较好解决油气管道焊缝缺陷分类效果不理想的问题,为保障油气管道安全运输提供了技术支撑。Welding technology is widely used in the connection of oil and gas pipelines,so ensuring the reliability of weld areas is crucial for the safe operation of oil and gas pipelines.Due to the limitation of process and technology,different types of weld defects may occur in the process of oil and gas pipeline welding.To address the problems such as small difference between defect and background leading to poor defect identification result and large manual identification workload,this paper proposes a novel method of intelligent identification of pipeline weld defects based on the SCT-ResNet50 model.This newly proposed method involves firstly inputting weld area images into a feature extraction network,then applying the SCC(Spatial Channel Context)in the shallow layers of feature extraction to perform local spatial and channel information fusion and the ECA-MHSA in the deep layers to capture long-range dependencies and contextual information,and finally obtaining the ultimate defect identification results through fully connected layers and Softmax.The following research results are obtained.First,the novel method achieves a defect identification accuracy of 98.28%on a data set of weld defects from X-ray images of oil and gas pipelines.Second,compared with the classification methods such as ResNet50,VGG16,DenseNet121,MobileNetv3 and EfficientNetv2,its accuracy is 3.05%,46.05%,28.99%,15.95%and 18.84%higher,respectively.Third,in the scenarios with small sizes of pipeline weld defects and small differences between defects and backgrounds,this novel method exhibits a higher accuracy in identifying weld defects in oil and gas pipelines.In conclusion,the advantage of the algorithm of this novel method lies in the combination of SCC module with the local information and global information of ECA-MHSA module learning image.This novel method effectively improves the classification of weld defects in oil and gas pipelines,providing technical support for the safe operation of oil and gas pipelines.

关 键 词：深度学习 图像处理 油气管道焊缝 缺陷智能识别 注意力机制 

分 类 号：TE88[石油与天然气工程—油气储运工程]