Evaluation of U- shaped weld prep identification and tracking
Title: Evaluation of U- shaped weld prep identification and tracking
Authors: D De Becker (Intelligent Automation Centre, Loughborough University); J Dobrzanski (Intelligent Automation Centre, Loughborough University); J Hodgson (Intelligent Automation Centre, Loughborough University); M Goh (Intelligent Automation Centre, Loughborough University); P Kinnell (Intelligent Automation Centre, Loughborough University); L Justham (Intelligent Automation Centre, Loughborough University);
Year: 2019
Citation: De Becker, D., Dobrzanski, J., Hodgson, J., Goh, M., Kinnell, P., Justham, L., (2019). Evaluation of U- shaped weld prep identification and tracking. UK-RAS19 Conference: “Embedded Intelligence: Enabling & Supporting RAS Technologies” Proceedings, 4-7. doi: 10.31256/UKRAS19.2
Abstract:
An autonomous welding system must be able to identify and extract the relevant features of the weld seam to generate an accurate weld path. Furthermore, the system must be able to adapt the weld torch position in real time during the weld. This has led to a two-stage approach, with the first stage identifying the weld path from a roughly scanned weld seam and the second stage adjusting the weld torch position in real time. In order to track weld seams, it has become popular to utilize a laser line scanner due to its versatility in measuring a wide range of materials and the non-contact nature.
Three methods were explored in extracting the shoulders of a U-shaped weld prep. This included a clustering method utilizing a density based spatial clustering approach, a line of best fit approach and an image processing approach utilizing Hough line transforms. Both the clustering and line of best fit approach use a spline fit to find the bottom of the weld prep. While the image processing approach uses a circular Hough transform to find the same position. Further testing, with real world data, showed that the clustering approach struggled when the weld prep was not perpendicular to the scanning axis. This issue was not observed in either the line of best fit method or the image processing method, however the image processing method often found multiple lines on the same shoulder of the weld prep. This led to more testing being carried out with the line of best fit method which tended to be the most robust method. The main drawback of this method was the higher computational requirement. However, during the real-time seam track testing it was found that the robot position could be updated at 30Hz without the use of buffers.