Study On Weld Vision Characteristic For All-position Autonomous Welding Robot With Visual Sensor

In order to adapt to the requirement of automation and intelligence for the manufacture of non-structural space welding of large-scale equipment,the main research direction for welding mobile robot at the present stage is to simulate welder in complex environments of work space and surface for welding task.This dissertation is based on the project of the national high technology research and development program(863 program)"all position autonomous welding robot system with passing obstacle ",having been independent developed for all position autonomous welding robot welding system vision sensor,to achieve weld image real-time acquisition and character processing.Mobile welding robot vision sensing system is one of the important ways to improve intelligent robotic welding.Based on the existing mobile welding robot,a visual sensing system had been reconstructed.On the aspect of hardware,selecting of more mature CCD camera and matching image acquisition card,to ensure the quality of images collecting for welding process.The image that acquired by visual sensing system for purpose of calculate the size of the actual object,calibration must be carried out.Use checkerboard method to create a mapping between the image coordinate system and the robot coordinate system,and determine the calibration points through the use of corner detection tool which Caltech developed,thus an effective coordinate mapping model established.Smart mobile welding robot welding power is an indispensable part of the system,so this experiment platform uses Aotai Pulse MAG / MIG 350 welding power.The welding power,using pulse waveform transition zone control and arc length closed-loop control technology,can effectively control the droplet and the molten pool.In addition,high reliability and all-position welding capabilities of the power meet the needs of intelligent robot welding.For the image features of MAG weld molten pool areas,the basic molten pool image approach such as filtering,image segmentation and image morphology,has been utilized to develop image processing method possessing of high adaptability and robustness.It is more suitable for the weld molten pool and the gap edge extraction by the method of directions wavelet and active contour model.These methods can filter out some disturbance in molten pool image and can effectively get the molten pool edge.Finally,visual information experiment of MAG welding molten pool conducted the welding robot test platform.Test objects include image processing and analysis between root welding and filling passes.Through experimental verification,the weld pool width of root weld penetration state is stable,with a minimum of 144 pixels and a maximum of 148 pixels,and comply with abnormal distribution with mean 145.96,the mean-square deviation of 1.2282.The statistics of filling passes width as follows: a minimum value of 153 pixels and a maximum of 147 pixels,which also comply with abnormal distribution with mean 157.66,the mean-square deviation of 3.0143.If the weld width in the above range,it can be considered that the filling welding is consistent.The experiment of mobile robot manifests that the visual sensing system is effective and reliable.