Real time optimization of robotic arc welding based on machine vision and neural networks

The successful application of welding robots relies on their abilities of automatic control of welding qualities. With the help of machine vision and neural networks, the authors developed an intelligent approach for real time optimization of the torch posture and welding parameters. The optimization is realized by neural networks which have been well trained beforehand with optimal welding samples. A double-eyes vision system accompanied by the newly developed Line-Point Matching Algorithm is adopted for determining the orientation of the weld seam. Investigations are also carried out in utilizing the neural networks for adaptive image processing and for producing the 3D coordinates of a point on the weld seam edges. The approach introduced in this paper is promising for attaining synthetic control of welding quality.