On-line Laser Welding Monitoring System Design Based On Computer Vision Methods

In the laser welding process, the weld pool geometry can directly reflect the weldquality. And a vision system can detect the weld pool geometry in real time. It is the keyto achieve weld quality monitoring and automation of the laser welding. In practice, theimages of weld pool are overlapped by the interferences, which may influence the weldpool measurements. So it is very important to seek for an algorithm which can removethe interferences and get the accurate weld pool geometry.The dissertation established a computer vision system for deep penetration laserwelding, and proposed three image processing method which can remove theinterferences to achieve on-line laser welding monitoring of the weld pool geometry. First,the dissertation introduced the frameworks of the software and hardware of the systemand the display of the weld pool temperature field. Second, by analyzing the RGBchannel of the weld pool images, we get the R channel threshold method, which is moreeffectively on weld pool edge extraction. Third, the dissertation emphasized on someproblems which we may meet in practice, such like the real-time performance of thealgorithm and the interference on the weld pool contour. In practice, the interferenceshave big effects on the weld pool geometry. We introduced two methods which caneffectively remove the interferences to get the accurate weld contour. They are specificstructuring element based opening method and closed convex curve active contour.Through experiments and analysis, we can know that they are good at interferencesremoval to get the accurate weld pool geometry parameters.By experiments, the measurement results by the methods above in line with theresults of manual work, which shows the advantage of edge extraction and interferenceremoval. And the computing speed is fast to meet the real-time requirements. They can beused in a vision system to measure the weld pool geometry accurately and in real time.