基于多层特征融合目标检测网络的超声全聚焦钢板焊缝缺陷自动检测研究

doi:10.3901/JME.2025.04.055

机械工程学报 ›› 2025, Vol. 61 ›› Issue (4): 55-66.doi: 10.3901/JME.2025.04.055

• 仪器科学与技术 • 上一篇

扫码分享

基于多层特征融合目标检测网络的超声全聚焦钢板焊缝缺陷自动检测研究

秦飞红¹, 袁懋诞¹, 祝端阳¹, 刘晓睿², 李明³, 纪轩荣¹

1. 广东工业大学省部共建精密电子制造技术与装备国家重点实验室广州 510006;
2. 苏州热工研究院有限公司苏州 215004;
3. 中广核检测技术有限公司深圳 518026

收稿日期:2024-02-17 修回日期:2024-09-06 发布日期:2025-04-14
作者简介:秦飞红，男，1997年出生。主要研究方向为深度学习在超声检测中的应用。E-mail：Atanis-Qin@outlook.com
袁懋诞(通信作者)，男，1988年出生，博士，副教授，硕士研究生导师。主要研究方向为超声无损评价。E-mail：mdyuan@gdut.edu.cn
基金资助:
广东省基础与应用基础研究基金(2022A1515240040)、国家自然科学基金(51975131，U2133213)和广东省“珠江人才计划”引进创新创业团队(2016ZT06G375)资助项目。

Automatic Weld Defect Detection in Steel Plates Based on Ultrasonic Total Focusing Method and Multi-layer Feature Fusion Target Detection Network

QIN Feihong¹, YUAN Maodan¹, ZHU Duanyang¹, LIU Xiaorui², LI Ming³, JI Xuanrong¹

1. State Key Laboratory of Precision Electronic Manufacturing Technology and Equipment, Guangdong University of Technology, Guangzhou 510006;
2. Suzhou Nuclear Power Research Institute, Suzhou 215004;
3. CGN Inspection Technology of Company, Shenzhen 518026

Received:2024-02-17 Revised:2024-09-06 Published:2025-04-14

摘要/Abstract

摘要： 金属焊接结构的缺陷会导致关键设备在服役过程中产生严重危害，传统的缺陷识别主要以人工检测为主，但实际检测数据量庞大，而且缺陷形状和大小各异，导致检测结果容易受各种因素影响。提出基于多层特征融合目标检测网络对钢板焊缝的超声全聚焦图像进行智能识别。由于超声图像中缺陷的像素占比低且存在较多噪声和伪影等，使用具有特征复用且参数更少的Dense Net(Dense convolutional network)换原有的VGG(Visual geometry group)主干网络以优化模型的计算资源和速度。同时嵌入多层网络融合模块，将上下相邻网络特征图进行融合，充分结合大尺寸特征与小尺寸特征，提高模型的检测能力。此外，还嵌入空间注意力机制来增强缺陷信息并抑制背景噪声，最后将原有SSD(Single shot multibox detector)中的交并比(Intersection over union，IoU)非极大值抑制算法替换为距离交并比(Distance-IoU，DIoU)，以更好地检测缺陷图像中间距较小的缺陷。利用仿真和试验获取典型焊缝缺陷的超声全聚焦数据集进行模型测试，结果表明，该模型比原有SSD模型平均AP值(Mean average precision，mAP)提高了10.45%，检测帧率(Frames per second，FPS)提高了17%。提出的多层特征融合目标检测网络可以用于大批量焊接钢结构超声全聚焦检测结果的快速、准确、自动识别。

关键词: 全聚焦检测, 深度学习, 焊缝缺陷, 特征融合

Abstract: Defects in metal welding structures can cause serious harm to key equipment during service. Traditional defect identification is mainly based on manual inspection. However, the amount of actual inspection data is huge, and defects vary in shape and size, making the inspection results susceptible to various influencing factors. A multi-layer feature fusion target detection network is proposed to intelligently identify the ultrasonic total-focusing image of steel plate welds. Since the proportion of defective pixels in ultrasound images is low and there are many noises and artifacts, the original VGG backbone network is replaced with a dense convolutional network with feature reuse and fewer parameters to optimize the computing resources and speed of the model. At the same time, a multi-layer network fusion module is embedded to fuse the upper and lower adjacent network feature maps, fully combining large-size features and small-size features to improve the detection ability of the model. In addition, a spatial attention mechanism is embedded to enhance defect information and suppress background noise. Finally, the intersection over union ratio non-maximum suppression algorithm in the original SSD is replaced with a distance intersection over union ratio to better detect closely spaced defects in using simulation and experiments to obtain the ultrasonic total-focusing data set of typical weld defects for model testing, the results show that the average accuracy of the model is 10.45% higher than the original SSD model, and the detection frame rate is increased by 17%. The proposed multi-layer feature fusion target detection network can be used for fast, accurate and automatic identification of large-volume ultrasonic total-focusing inspection results of welded steel structures.

Key words: total focusing method, deep learning, weld defect, feature fusion

中图分类号:

TG156

秦飞红, 袁懋诞, 祝端阳, 刘晓睿, 李明, 纪轩荣. 基于多层特征融合目标检测网络的超声全聚焦钢板焊缝缺陷自动检测研究[J]. 机械工程学报, 2025, 61(4): 55-66.

QIN Feihong, YUAN Maodan, ZHU Duanyang, LIU Xiaorui, LI Ming, JI Xuanrong. Automatic Weld Defect Detection in Steel Plates Based on Ultrasonic Total Focusing Method and Multi-layer Feature Fusion Target Detection Network[J]. Journal of Mechanical Engineering, 2025, 61(4): 55-66.

参考文献

[1] VAN SLOUN R J G，COHEN R，ELDAR Y C. Deep learning in ultrasound imaging[J]. Proceedings of the IEEE，2019，108(1):11-29.
[2] HUANG W，KOVACEVIC R. Development of a real-time laser-based machine vision system to monitor and control welding processes[J]. The International Journal of Advanced Manufacturing Technology，2012，63:235-248.
[3] MEDAK D，POSILOVIC L，SUBASIC M，et al. Automated defect detection from ultrasonic images using deep learning[J]. IEEE Transactions on Ultrasonics，Ferroelectrics，and Frequency Control，2021，68(10):3126-3134.
[4] 李建乐，张佳奇，黄念，等. 基于深度学习的分布式光纤损伤识别方法[J]. 机械工程学报，2022，58(8):88-95. LI Jianle，ZHANG Jiaqi，HUANG Nian，et al. Distributed optical fiber damage identification method based on deep learning[J]. Journal of Mechanical Engineering，2022，58(8):88-95.
[5] 雷亚国，贾峰，周昕，等. 基于深度学习理论的机械装备大数据健康监测方法[J]. 机械工程学报，2015，51(21):49-56. LEI Yaguo，JIA Feng，ZHOU Xin，et al. A deep learning-based method for machinery health monitoring with big data[J]. Journal of Mechanical Engineering，2015，51(21):49-56.
[6] SLONSKI M，SCHABOWICZ K，KRAWCZYK E. Detection of flaws in concrete using ultrasonic tomography and convolutional neural networks[J]. Materials，2020，13(7):1557.
[7] VAN SLOUN R J G，COHEN R，ELDAR Y C. Deep learning in ultrasound imaging[J]. Proceedings of the IEEE，2019，108(1):11-29.
[8] WONG K C L，SINKOVSKAYA E S，ABUHAMAD A Z，et al. Multiview and multiclass image segmentation using deep learning in fetal echocardiography[C]//Medical Imaging 2021:Computer-Aided Diagnosis，Birmingham:Springer，2021:308-313.
[9] REDMON J，DIVVALA S，GIRSHICK R，et al. You only look once:Unified，real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas:IEEE，2016:779-788.
[10] REDMON J，FARHADI A. YOLO9000:better，faster，stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu:IEEE，2017:7263-7271.
[11] LIU W，ANGUELOV D，ERHAN D，et al. Ssd:single shot multibox detector[C]//Computer Vision–ECCV 2016:14th European Conference. Amsterdam:Springer International Publishing，2016:21-37.
[12] REN S，HE K，GIRSHICK R，et al. Faster r-cnn:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence，2017，39(6):1137-1149.
[13] HE Y，SONG K，MENG Q，et al. An end-to-end steel surface defect detection approach via fusing multiple hierarchical features[J]. IEEE Transactions on Instrumentation and Measurement，2019，69(4):1493-1504.
[14] SHARMA M，LIM J，LEE H. The amalgamation of the object detection and semantic segmentation for steel surface defect detection[J]. Applied Sciences，2022，12(12):6004.
[15] GUO Z，WANG C，YANG G，et al. Msft-yolo:Improved yolov5 based on transformer for detecting defects of steel surface[J]. Sensors，2022，22(9):3467.
[16] YANG L，WANG Z，GAO S. Pipeline magnetic flux leakage image detection algorithm based on multiscale SSD network[J]. IEEE Transactions on Industrial Informatics，2019，16(1):501-509.
[17] HUANG G，LIU Z，VAN D M L，et al. Densely connected convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu:IEEE，2017:4700-4708.
[18] HU J，SHEN L，SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City:IEEE，2018:7132-7141.
[19] ZHENG Z，WANG P，LIU W，et al. Distance-iou loss:faster and better learning for bounding box regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence. New York:AAAI，2020:12993-13000.
[20] HOLMES C，DRINKWATER B W，WILCOX P D. Post-processing of the full matrix of ultrasonic transmit–receive array data for non-destructive evaluation[J]. NDT & e International，2005，38(8):701-711.
[21] 吴斌，杨敬，焦敬品，等. 奥氏体不锈钢小径管焊缝缺陷多模式超声复合全聚焦成像研究[J]. 机械工程学报，2020，56(14):9-18. WU Bin，YANG Jing，JIAO Jingpin，et al. Study on ultrasonic multi-mode composite total focusing imaging method for welds defect inspection in small-diameter tube of austenitic stainless steel[J]. Journal of Mechanical Engineering，2020，56(14):9-18.
[22] VELICHKO A，WILCOX P D. Reversible back-propagation imaging algorithm for postprocessing of ultrasonic array data[J]. IEEE Trans. IEEE Transactions on Ultrasonics，Ferroelectrics，and Frequency Control，2009，56(11):2492-2503.
[23] HUNTER A J，DRINKWATER B W，WILCOX P D. Autofocusing ultrasonic imagery for non-destructive testing and evaluation of specimens with complicated geometries[J]. NDT & E International，2010，43(2):78-85.
[24] ZHANG J，DRINKWATER B W，WILCOX P D. Effects of array transducer inconsistencies on total focusing method imaging performance[J]. NDT & E International，2011，44(4):361-368.
[25] GUO X，HAN Y，NIE P. Ultrasound imaging algorithm:half-matrix focusing method based on reciprocity[J]. Mathematical Problems in Engineering，2021，2021:1-11.
[26] KIM J，LEE J K，LEE K M. Accurate image super-resolution using very deep convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas:IEEE，2016:1646-1654.
[27] LIN T Y，DOLLÁR P，GIRSHICK R，et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Amsterdam:IEEE，2017:2117-2125.
[28] LI C，HUANG Y，LI H，et al. A weak supervision machine vision detection method based on artificial defect simulation[J]. Knowledge-Based Systems，2020，208:106466.

基于多层特征融合目标检测网络的超声全聚焦钢板焊缝缺陷自动检测研究

Automatic Weld Defect Detection in Steel Plates Based on Ultrasonic Total Focusing Method and Multi-layer Feature Fusion Target Detection Network

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	姜洪权, 张毛杰, 程虎跃, 高建民, 姚欢, 高富超, 郭晓峰, 贺吉程. 融合动态评片知识AI模型的焊缝缺陷探测技术[J]. 机械工程学报, 2025, 61(6): 24-32.
[2]	黄思翰, 陈建鹏, 徐哲, 阎艳, 王国新. 基于大语言模型和机器视觉的智能制造系统人机自主协同作业方法[J]. 机械工程学报, 2025, 61(3): 130-141.
[3]	刘翰儒, 陈桂映, 许泽林, 彭世通, 郭嘉楠, 刘伟嵬, 王奉涛. 基于SERepVGG-A2的轻量化卷积神经网络在熔池状态识别中的应用[J]. 机械工程学报, 2025, 61(3): 440-448.
[4]	狄子钧, 袁东风, 李东阳, 梁道君, 周晓天, 信苗苗, 曹凤, 雷腾飞. 基于多尺度-高效通道注意力网络的刀具故障诊断方法[J]. 机械工程学报, 2024, 60(6): 82-90.
[5]	李春凯, 王嘉昕, 石玗, 代悦. GTAW熔池激光条纹动态行为与熔透预测方法研究[J]. 机械工程学报, 2024, 60(6): 236-244.
[6]	史丽晨, 史炜椿, 王海涛, 李金阳, 刘亚雄. 基于DRSN-BiLSTM模型的刀具磨损预测方法研究[J]. 机械工程学报, 2024, 60(24): 66-74.
[7]	赵德望, 姜超, 赵延广, 杨文平, 樊俊铃. 铝锂合金疲劳性能及预测研究-基于试验和“浅层”+“深度”混合神经网络方法[J]. 机械工程学报, 2024, 60(16): 190-199.
[8]	付玲, 佘玲娟, 颜镀镭, 张鹏, 龙湘云. 基于内嵌物理信息与注意力机制BiLSTM神经网络的臂架系统疲劳损伤预测模型[J]. 机械工程学报, 2024, 60(13): 205-215.
[9]	陈钱, 陈康康, 董兴建, 皇甫一樊, 彭志科, 孟光. 一种面向机械设备故障诊断的可解释卷积神经网络[J]. 机械工程学报, 2024, 60(12): 65-76.
[10]	刘岳开, 王天杨, 褚福磊. 基于低延迟可解释性深度学习的复杂旋转机械关键部件知识嵌入与诊断方法研究[J]. 机械工程学报, 2024, 60(12): 107-115.
[11]	赖旭伟, 丁昆, 张楷, 黄锋飞, 郑庆, 李致萱, 丁国富. 基于可解释物理引导空间注意力改进的跨工艺参数立铣刀磨损辨识[J]. 机械工程学报, 2024, 60(12): 147-157.
[12]	林昕, 毛杨坤, 傅盈西, 朱锟鹏. 基于图像语义分割与图卷积的选区激光熔融成形过程羽流运动特征分析[J]. 机械工程学报, 2024, 60(12): 168-182.
[13]	邵海东, 肖一鸣, 邓乾旺, 任颖莹, 韩特. 基于不确定性感知网络的可信机械故障诊断[J]. 机械工程学报, 2024, 60(12): 194-206.
[14]	李玖法, 邹博文, 任玥. 考虑车-路交互作用的车辆轨迹预测算法研究[J]. 机械工程学报, 2024, 60(10): 76-85.
[15]	许思远, 张卫东, 毛玉明, 牛斌. 数据驱动的指定变形非均质多胞结构优化设计[J]. 机械工程学报, 2024, 60(1): 85-95.