Three-dimensional reconstruction of specular surface for a gas tungsten arc weld pool

Abstract

Observing the weld pool and measuring its geometrical parameters are key issues for developing the next generation intelligent welding machine and modeling the complex welding process. In the past few years, different techniques have been applied, but the dynamic specular weld pool surface and the strong weld arc complicate these approaches and make observation difficult. To resolve the problem, a new three-dimensional sensing system using structured light is proposed for a gas tungsten arc welding (GTAW) process. In the system, a dot-matrix laser pattern is projected on the specular weld pool surface, which can reflect light onto an imaging plane. The reflected images are captured by a high-speed camera and can successfully be processed by image processing algorithms developed. With the acquired information, a three-dimensional reconstruction scheme is proposed and discussed in this paper. A surface reconstruction method with several slope-based algorithms is first developed to rebuild the region of weld pool surface which reflects the laser pattern. Then a two-dimensional piecewise model is provided to calculate weld pool boundary by utilizing the edge condition. Finally the optimal estimate of the three-dimensional weld pool surface is synthesized. The acceptable accuracy of the results verified the effectiveness of the reconstruction scheme. List of notation and definitionsBase point (points). The reflection point (points), whose position(s) is (are) assumed and used to calculate the positions of other reflection points as a base, is (are) called base point(s).Column plane. For a point p i,j located in the ith row and the jth column of the projected dot matrix, the plane passing through all the dots in the jth column and the dot of laser diode is called the column plane of point p i,j .Center reference point. The center point in the projected dot matrix of structured light. For a 19-by-19 dot matrix, it is located in the 10th row and the 10th column. It is not an actual visible point but a reference position, which can be used to find the corresponding relationship between projected and reflected points.GTAW. Gas tungsten arc welding Normal. A normal to a flat surface is a three-dimensional vector which is perpendicular to that surface, and a normal to a non-flat surface at a point P on the surface is a vector which is perpendicular to the tangent plane to that surface at P.Reflected dots (points). The dots reflected on the imaging plane.Reflected image. The captured image including the reflected dots on the imaging plane.Reflection dots (points). The dots projected on the weld pool surface, also called projected dots.Row plane. For a point p i,j located in the ith row and the jth column of the projected dot matrix, the plane passing through all the dots in the ith row and the dot of laser diode is called the row plane of point p i,j .Tangent plane. P is a point on the surface S. If the tangent lines at P to all smooth curves on the surface S passing through P lie on a common plane, then that plane is called ...

A one-shot-projection method for measurement of specular surfaces

“…It is seen that the reconstructed points are close to the original points. The computed reconstruction MSE (Mean Squared Errors) are 0.30188 µm in xdirection, 0.59231 µm in y-direction, 2.3 µm in z-directionand 3.2 µm in total.The results are compared with those in [6], [7] and our previous work [22]. [6] reported 48.4 µm average squared error and [7] reported 28.9 µm average squared error.…”

“…The computed reconstruction MSE (Mean Squared Errors) are 0.30188 µm in xdirection, 0.59231 µm in y-direction, 2.3 µm in z-directionand 3.2 µm in total.The results are compared with those in [6], [7] and our previous work [22]. [6] reported 48.4 µm average squared error and [7] reported 28.9 µm average squared error. In our previous work [22], the reported measurement errors are 20.7 µm in x -direction, 43.3 µm in y-direction, 27.9 µm in zdirection and 91.9 µm in total.…”

“…The error induced in practical measurement is induced by the noise, which indicates that suppressing the noise could increase the accuracy of the proposed method further. The reconstructed weld pool by the proposed method and those from [6] and [7] are compared visually in Fig. 13(a)-13(c).…”

“…Huang [5] calculated the height distribution with an integration of gradient data obtained from one monoscopic fringe reflection, which could estimate the profile of the dynamic specular surface by one-shot projection. Similar estimation based one shot methods are proposed in [6] and [7] to measure the specular and dynamic weld pool However, none of them could calculate precise depth and lacks robustness for precision measurement. For stereo vision methods [8,9], the specular surfaces prevent the two cameras from obtaining two views directly.…”

A stereo vision method for tracking particle flow on the weld pool surface

“…It is then possible to obtain 3D surface profile measurements and flow velocity measurements for a weld pool under favorable cir- cumstances. In well shielded welding conditions, the surface remains quite clean, without any oxide particles; adding an additional structured illumination laser, as used by Song and Zhang,27,28 could offer an efficient way to reconstruct the weld pool surface and to track surface oscillations on both arc and laser generated weld pools. It is also possible to use this method when the object shape is a complex 3D component; however, the equipment arrangement is inherently more complex.…”