Computer vision system for weld bead geometric analysis

摘要

Welding is an important process in several industries and it is required precision and attention during the execution. In addition it is unhealthy work for the operator who is in direct contact with sparks, smoke and electricity. Thus, it is increasingly common to use robots in this type of task, reducing human intervention in the process. This work proposes an automated system to verify discontinuities in the weld beads, which may invalidate them. Through the use of a linear welding robot and a vision system, images of the pre-welding and post-welding metal plates are captured for future comparisons and evaluation metrics applied. The proposed method uses the combination of techniques such as: PCA, Gaussian Mixture and HMM with the objective of identifying the edges of the weld beads. Then, the proposed metrics are applied in order to assist the welding inspector to identify possible discontinuities in the weld.