Weld bead recognition using laser vision with model-based classification. (August 2018)

• Record Type: Journal Article
• Authors: Ye, Guoliang; Guo, Jianwen; Sun, Zhenzhong; Li, Chuan; Zhong, Shouyan
• Is Part Of: Robotics and computer-integrated manufacturing. Volume 52(2018)
• Page Start: 9
• Page End: 16
• Publication Date: 2018-08
• Subjects: Bead geometry -- Segmentation -- Laser vision -- Expectation-Maximization algorithm; Robots, Industrial -- Periodicals; Computer integrated manufacturing systems -- Periodicals; Robotics -- Periodicals; Robots industriels -- Périodiques; Productique -- Périodiques; Robotique -- Périodiques; 670.285
• DOI: 10.1016/j.rcim.2018.01.006
• Languages: English
• ISSNs: 0736-5845

Abstract

Abstract: Weld bead recognition is important for providing information for automatic control of welding process and also for post-weld inspection. Bead geometry profiles can be measured by a laser-line profile sensor projecting a laser line onto the welding surface and receiving the reflection to calculate the height of the surface. As the bead is usually very small, it may be easily covered by the sensor noise or distorted by the bending of the work pieces after welding. In addition, it is difficult to estimate the parameters of bead width and bead height while moving the laser sensor to cover a 3D surface, since the movement may cause unstable measurements. This work investigates using a model-based classification method to automatically segment the bead from the welding surface regardless of the distance and the angle of the scanner to the welding surface. It firstly uses a polynomial model together with the Expectation-Maximisation (EM) algorithm fitting into the distance profiles measured by the laser sensor, and then automatically removes the surface curvature. Consequently the shape of the bead can be easily extracted from the welding surface by applying a small value of threshold depending on the noise level of the data. Experimental results show that the Butt welding bead parameters of bead width and height can be extracted successfully. The method is robust to the bending of the welding surface and the unstable movement of the laser sensor. Highlights: A novel model-based classification method with Expectation-Maximisation algorithm for weld bead recognition is proposed. The classification is fully unsupervised without training data sets. Parameters of bead width and bead height can be automatically measured. The method is robust to the bending of the base metal and the unstable movement of the laser sensor.