Original title: Automated Weld Seam Recognition and 3D Mapping for Robotic Post Processing Using Photogrammetry and Semantic Segmentation
Augustin Raju, Abilash Madavath, Chandra Yuvesh Aubeeluck, Nicolas Pyschny +2 · September 2026
Robots need to know the exact shape and location of weld seams to grind, finish, and inspect metal parts. On large pieces, scanning every surface with high precision laser scanners or structured light sensors takes a lot of time. It also collects huge amounts of useless data. To solve this, researchers developed an experimental vision based pipeline to roughly find weld seams first. The system snaps photos of the workpiece from multiple viewpoints. Next, an image analysis method called semantic segmentation spots the weld seams in each picture. Then, a technique called photogrammetry uses the photos to build a digital shape of the workpiece. Finally, the system projects the detected seams onto that rebuilt model, saving overall scanning effort.
Imagine you need to clean stains on the walls of a very large warehouse. Walking around to scrub every inch would take days and waste effort on clean walls. Instead, you first do a quick walkthrough, marking only the stained spots with chalk. Then you bring your cleaning supplies and focus only on those marked areas. In this paper, the quick walkthrough is the image-based system that identifies weld seams, so the expensive scanning is done only where it matters.
For manufacturers working with large welded structures, this method cuts down the time and data needed to prepare robots for tasks like grinding and finishing. It makes automated post processing faster and less wasteful.
The method is experimental and provides only approximate localization of weld seams, intended as a preliminary step before precise measurement. It does not replace high-precision scanning but aims to reduce its scope.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Automated Weld Seam Recognition and 3D Mapping for Robotic Post Processing Using Photogrammetry and Semantic Segmentation — Augustin Raju, Abilash Madavath, Chandra Yuvesh Aubeeluck et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.