Abstract
In automatic welding of a 3-D weld seam, a vision sensor is applied to acquire image patterns at various sensor positions and orientations. To extract the proper features from the image data, it is required to identify the pattern categories. In this paper, typical image patterns acquired in the sensing of the 3-D weld seam are classified. To identify the pattern categories, several parameters of the 2-D image data were tested, and a neural network using two input parameters and two hidden layers was developed to find the pattern type as output. A laser vision sensor attached to a conceptual mobile platform-manipulator was considered for sensing simulations, and a back-propagation algorithm was adopted to teach the neural network by using the geometrical patterns acquired from those simulations. The classified image data are processed using the line segmentation method to determine the distinctive features, and these features can then be used to find the weld seam or its end position for the automatic welding system. Experimental results of sensing are also presented.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call