Using the MOTOSense system, robotic welding is guided by 4 characteristic points on the front section of the laser line. A cross section of the joint was used with a V groove joint on a cylindrical workpiece. The geometry of the groove needs to be identified enabling guidance relative to the characteristic points, which have a trapezoidal shape.
The current method of characteristic point identification yields only the upper two points. The remaining points are obtained by derivative algorithms along the y axis and searching for the shortest distance from the centre line to each point on the line. The welding joint is often tack welded in advance and has an undefined weld joint, which makes identification of the characteristic points difficult.
In order to identify and analyse all characteristic points, a derivative algorithm along the x axis with a diagonal filter and a polynomial fitting algorithm was used. Software libraries supported by the Python programming language were used. The first algorithm searched for line breaks in the desired part of the image, which were then cross-referenced with a diagonal filter. The second algorithm, however, defined the changing parts of the line using first order polynomial fitting algorithms.
The second algorithm returned better results as it enabled the characteristic points to be identified with smaller absolute errors. Furthermore, greater precision and reliability was observed with the second algorithm over the first algorithm as well as the current method of identification. Identified characteristic points using the first algorithm were offset above or below the desired point.
Despite successful identification, the second algorithm needs to be further developed and tested on a wider set of images to ensure robust operation. There is potential for improvement in the identification of characteristic points by geometric mapping and alignment forms.
|
