3D Reconstruction Of Weld Pool Surface Based On Global Stereo Matching Algorithm

Now, manufacturing industry of welding product is gradually towards automation andintelligent, It demands precise control of weld penetration with the widely used of weldingrobot. In the traditional manual welding, skilled welders observe the surface of weld poolwith eyes and adjust weld parameters according to penetration state information, visionsensors can obtain the3D topography information instead of human eyes to control theweld parameters. In this paper, the3D reconstruction of pulsed MIG welding (P-GMAW)pool surface, which during the base pulse period in Q235butt backing weld, wasconstructed by a biprism stereo camera vision system.A passive stereo vision system was designed with a biprism and a CCD camera, filtersystem were used to weaken the arc interference, which consist of a narrowband filter anda dimmer filter. The principle of system calibration was analyzed; the internal and externalparameters and the relative position of the virtual cameras were obtained by calibratingsystem with black and white checkerboard pattern. The stereo images of weld pool wererectificated in order to increase calibration precision.Two matching algorithms were applied with the study of the algorithm of stereomatching, they were the matching algorithm based on constant space belief propagationand the matching algorithm based on the growth of seed space. The construction processof P-GMAW backing weld pool surface under complete penetration by above twoalgorithms were described, the construction results of each algorithm were given andcompared, the reconstruction result from the second algorithm was better. So theconstruction precision of better algorithm were analyzed and proved by the images ofcalibration template, the maximum construction error is0.10714mm, the reliablility ofsystem was proved.Finally, the algorithm, which base on the growth of seed space, was used to reconstructthe pool surface under different penetration states, it can be conclude from the result ofreconstruction that the change of penetration and the maximum penetration increase withthe penetration increase.