Vision sensors using optical triangulation have been widely used for automatic welding in various ways, but they have some limitations such as shadow effect, specular reflection, and arc noise. If they are adopted to the automation of welding of linear weldments with height variation, their limitations will cause some critical problems. These limitations were categorized into two types. The one is the data deficiency and the other is the reliability of the vision sensor. The data deficiency was analyzed and the criteria for its occurrence were established, then the algorithm for simulating the data deficiency was developed. The proposed algorithm was applied to two welding geometries with height variation and was found to be able to efficiently predict the occurrence of the data deficiency. The reliability of the vision sensor was analyzed for the variation of the arc noise and reflected laser beam. First, the property of reflection of the base metal was medelled using BRDF, then the variations of brightness of the reflected arc and laser beam were formulated by assuming the arc light as a point light source for the different inclination of the base metal. The experimental data of laser light reflection were obtained for three materials which are frequently used for linear weldments with height variation. The data of mild steel were fitted to the model of laser beam reflection and the results were in a good agreement. The experimental data of the reflected arc light were obtained for two materials and compared with the models. The results showed that the assumption of considering the arc light as a point light source is valid for the distance which is larger than 45 mm from the arc.
