Abstract
In recent years there has been an increased use of sensors to either enable the use of robots, or increase the productivity of robots for gas metal arc welding. Conventional robotic arc welding without the use of sensors can suffer from a number of possible process errors which, if not compensated, produce defective welds. Much work has already been carried out in developing sensors which address these process errors, but none can satisfy the requirements of all possible weld joint configurations and sizes which are fabricated. The majority of the previous work in this area has concentrated on providing seam tracking systems for the welding of sheet materials, typically used in the heavy and automotive industry, although some work is suitable for tubes between 50 and 100mm in diameter. This thesis presents the initial development stages of a novel fully automated computer vision based welding sensor for saddle type joints formed by small diameter intersecting tubing, less than 50mm in diameter. The sensor uses the active projection of multiple planes of light to highlight the weld joint features. The underlying image processing techniques which have been developed are based upon the local area analysis of an image of the illuminated weld joint. The algorithms extract each stripes feature points, reduce the information to a smaller set of weld feature points, and then generate the robot program path points. Calibration information within the image is used as part of the latter transformation process. These software based techniques may be implemented into specialised computer hardware to reduce the sensor cycle time, giving the capability of high speed operation. An application specific multistripe sensor has been validated through the integration of the sensor within an existing industrial robot workcell welding bicycle forks. Experimental results have shown the multistripe sensor to be accurate to within +/-0.4mm of the weld joint for 93% of the feature points. The use of the multistripe sensor for: weld process capability trials, weld parameter investigation, and on-line positional feedback are presented. The results when used for on-line weld path determination have shown the multistripe sensor to be both effective and reliable when performing the joint analysis on a non- optimal low-cost host platform.