Objective At present, pipeline automatic welding is generally equipped with a structural light sensor for seam tracking. And the detection of welding quality is one of the key points in robotic welding. Therefore, it has great practical value for the research on the detection of welding quality monitored by a linear structured light sensor. However, the existing weld forming quality detection based on structured light is mostly carried out on each contour line, and the change of adjacent contour lines is not properly utilized. For pipeline welds, it is difficult to guarantee the interval of adjacent contour lines when reconstructing the three-dimensional (3D) morphology of the weld surface by line structured light scanning. The weld edge, width and reinforcement extracted from each single contour also exist deviations due to the positioning error of the detection platform. Aiming at these problems, this paper presents a weld forming quality inspection method based on 3D reconstruction. And we hope that our findings can be helpful for pipeline automatic welding production.

Methods In order to improve the speed of extracting central-lines from structured light, the region of interest (ROI) should be extracted from the first and the last frames of the original N-frame images. Then the ROI is corrected by random sampling to ensure that the feature region is always within the ROI in the acquisition process. After Gaussian filtering, median filtering, image binarization, and morphological processing, the edge of structured light is extracted by the Laplace operator, and the sub-pixel centerline of structured light is obtained by the gray centroid method. Then the average value of each column is taken as the actual centerline position (Fig. 3). To calculate the deviation between the axis of the pipeline to be measured and the rotation center of the profiler, the feature points of each central-line are extracted by calculating the changing rate and piecewise fitting, according to the measured results of the profilometer in both the horizontal and vertical states. After position and attitude of the platform are corrected, the information on weld surface is obtained. The 3D morphology of the pipeline weld surface is reconstructed by numerical calculation and point cloud processing. Aiming at the point cloud outliers, the filtering algorithm is improved by setting the height threshold and the outliers are reset according to the formulas (8) and (9). In order to facilitate the follow-up process, the circular seam is treated as a plane in the 3D reconstruction. After the base metal plane is fitted by the improved least squares algorithm, the weld edge is extracted by calculating the deviation and removing the contour in which weld width is greater than W and smaller than W-2v where W is ideal weld width and v is weld broadening. The central-line of weld is fitted according to the median value of the left and right weld toe coordinates. Then it is moved along the base metal plane to obtain the allowable boundary of the weld edge. Finally, the idealized model is established according to the width and reinforcement of seams to detect the weld forming quality and to identify the defects, such as excessive weld metals, undercut, lack of fusion (Fig. 13).

Results and Discussions The difficult problem of contour stitching can be solved in the reconstruction of the pipeline weld surface based on line structure light by correcting the position-posture of the platform and perfecting the detection process. The closer the fitting parameters of the left and right base metal surfaces of the pipeline are, the better the butt joint will be. Compared with that of the weld edge extraction from each contour, the distribution of weld toe extracted from the 3D reconstruction model is closer to the same plane and more accurate (Fig. 12). The closer the actual weld surface contour is to the ideal model, the better the weld forming quality will be (Tables 1 and 2). When the weld edge is located within the maximum and minimum boundaries and the actual height is lower than the minimum surface height of the idealized model, it is undercut. In the weld zone, the actual height is larger than the maximum surface height of the idealized model, means over reinforcement (Fig. 14). At the same time, it should be noted that after repeated filtering of the data, the abnormal points caused by small cracks and porosity or other defects on the weld surface are also removed, making them almost impossible to detect. Detection of weld defects according to the spatial threshold, such as crater, spatter, and overlap, should be further studied.

Conclusions Idealized model established based on the measured results of 3D reconstruction of pipeline weld surfaces can be used to make a general assessment of the weld surface forming quality and identify reinforcement, undercut, etc. Compared with that based on the contour of each frame, forming quality measured based on the reconstruction results performs better.