作者：陈英红¹, 杜明坤²

作者单位：1. 广东省特种设备检测研究院珠海检测院, 广东 珠海 519000;
2. 珠海市安粤科技有限公司, 广东 珠海 519000

关键词：焊缝检测;机器视觉;深度学习;线结构光

摘要：

焊缝三维形态检测是提高焊接件安全运行的有效措施，该文提出一种基于Fast-RCNN与结构光的纵焊缝三维形态参数检测方法。首先应用Fast-RCNN卷积神经网络从视觉图像中检测并定位焊缝区域、测量焊缝宽度；其次采用颜色阈值分割与形态学处理相结合的方法提取激光中心线，从激光中心线沿焊缝方向的极大、极小值点计算出焊缝余高、咬边深度；最后，提取激光中心线处于焊缝热影响区部分，通过拟合圆弧曲线，检测焊缝错边量与棱角度。构造实验装置对输油管道的纵焊缝进行检测试验表明，能够一次性准确地检测出焊缝余高、宽度、咬边、错边量、棱角度5个参数，具有较低的测量不确定度，焊缝宽度、余高的测量不确定度较JJG704-2005说明的焊缝检验尺测量不确定度分别降低89%、85%。

A Fast-RCNN & structured light based three-dimensional shape parameter detection method for longitudinal welds

CHEN Yinghong¹, DU Mingkun²

1. Guangdong Special Equipment Testing and Research Institute Zhuhai Testing Institute, Zhuhai 519000, China;
2. Zhuhai An Yue Technology Co., Ltd., Zhuhai 519000, China

Abstract: Three-dimensional shape inspection of welds is an effective measure to improve the safe operation of welded parts. This paper proposes a three-dimensional shape parameter detection method based on the vertical weld of Fast-RCNN and structured light. Firstly, the method uses the Fast-RCNN convolutional neural network to detect and locate the weld area from the visual image and measure the weld width. Secondly, the color center segmentation and morphological processing are combined to extract the laser center line from the laser center line. The maximum and minimum values points of the weld direction are calculated to calculate the weld height and the undercut depth. Finally, the extracted laser center line is in the heat affected zone of the weld, the wrong side and edge angle of weld misalignment is detected by fitting the arc curve. The test of the longitudinal seam of the oil pipeline by the structural test device shows that the parameters of the weld height, width, undercut, wrong side and edge angle can be accurately detected at one time, and the measurement uncertainty is low. The measurement uncertainty of weld width and residual height is 89% and 85% lower than that of JJG704-2005.

Keywords: weld inspection;machine vision;deep learning;linear structured light

2018, 44(12): 85-90  收稿日期: 2018-09-02;收到修改稿日期: 2018-10-11

基金项目: 国家质量监督检验检疫总局科技计划项目（2017QK105）

作者简介: 陈英红(1974-),女,湖南郴州市人,高级工程师,研究方向为特种设备检验检测研发

参考文献

[1] 王冬林, 刘晶, 叶勇, 等. 钢制压力容器质量控制研究[J]. 新技术新工艺, 2015(10):110-112
[2] GUOLIANG Y, JIANWEN G, ZHENZHONG S, et al. Weld bead recognition using laser vision with model-based classification[J]. Robotics and Computer Integrated Manufacturing, 2018, 52:9-16
[3] 周济. 智能制造——"中国制造2025"的主攻方向[J]. 中国机械工程, 2015, 26(17):2273-2284
[4] 王豪, 魏昕, 林本宏, 等. 激光深熔焊接过程质量监测技术研究现状[J]. 应用激光, 2018, 38(1):123-129
[5] DAVIS T A, SHIN Y C. Vision-based clad height measurement[J]. Machine Vision & Applications, 2011, 22(1):129-136
[6] MAREK F, WOJCIECH J. Diagnostic method of welding process based on fused infrared and vision images[J]. Infrared Physics and Technology, 2013, 61:241-253
[7] 陈云霞, 李荣斌, 孙勤. 基于视觉传感的MAG焊图像采集及特征处理[J]. 焊接技术, 2018, 47(1):49-51
[8] CHANG D, SON D, LEE J, et al. A new seam-tracking algorithm through characteristic-point detection for a portable welding robot[J]. Robotics & Computer Integrated Manufacturing, 2012, 28(1):1-13
[9] 李琳, 林炳强, 邹焱飚. 基于条纹式激光传感器的机器人焊缝跟踪系统研究[J]. 中国激光, 2015, 42(5):34-41
[10] YAZID H, AROF H, YAZID H, et al. Discontinuities detection in welded joints based on inverse surface thresholding[J]. Ndt & E International, 2011, 44(7):563-570
[11] GB 150-2011, 压力容器[S]. 北京:中国质检出版社,2011.
[12] GIRSHICK R. Fast R-CNN[C]//IEEE International Conference on Computer Vision. IEEE Computer Society, 2015:1440-1448.
[13] 江永付, 江开勇, 林俊义. 线结构光光条中心亚像素精确提取方法[J]. 激光与光电子学进展, 2015, 52(7):185-191
[14] KHARE M, SRIVASTAVA R K, KHARE A. Object tracking using combination of daubechies complex wavelet transform and Zernike moment[J]. Multimedia Tools & Applications, 2015, 76(1):1-44
[15] 王纪兵. 压力容器检验检测[M]. 北京:化学工业出版社, 2016:90-91.