Composite Vision Method for Measurement of 3D Bead Profile, Part 1 — Principles

Identifying the as-welded bead 3D profile during multi-layer and multi-pass (MLMP) welding is challenging since the as-welded bead is easily disturbed and similar to the surrounding weld beads. However, the profile of the as-welded bead determines whether the preset path planning and the welding parameters need to be adjusted to avoid incomplete fusion. In MLMP welding, passive vision sensors can distinguish weld beads but struggle to extract height information, while active vision sensors excel at obtaining height details. Thus, in this study, an innovative composite vision method integrating a passive vision sensor and an active vision sensor was proposed for the as-welded bead profile measurement in MLMP welding. A deep-learning model based on U-Net was used to extract the weld bead features in the passive vision image, and coordinate transformation was used to correspond the coordinates of passive vision images to those of active vision images. The experiment’s results showed that the method can precisely measure the profile dimensions of the as-welded bead during MLMP welding, with an error in the width direction of less than 5% and an error in the height direction of less than 7%, meeting the actual MLMP welding requirements.