KOMPASS brings weld seam inspection to every bead as parts leave the welding station. AI models trained on your welds find surface porosity, undercut, spatter, gaps and irregular beads, then flag the part before it moves on to assembly, coating or dispatch.
Welders and robots produce seams that look similar at a glance. Small surface faults, a short bead or a gap at the end of a seam are easy to miss, especially across shifts and part variants.
Our AI Product Inspection system, KOMPASS™, carries out weld seam inspection with cameras, 3D sensing and AI models trained on your welds, checking the visible bead surface and geometry on every part. Results reach the line controls with images of each seam.
Weld seam inspection depends on the view as much as the model, because weld surfaces are bright, textured and curved. KOMPASS combines lighting and 3D sensing suited to your welds, trained models and a direct link to the line.
Cameras and, where geometry matters, a 3D profile sensor capture each seam after welding, with lighting chosen to control glare.
The model checks seam continuity, bead shape and surface condition, then classifies porosity, undercut, spatter and gaps.
Failing parts are held for repair or scrap, defect trends point to the station and every seam image is stored.
Send us welded parts with typical defects, your acceptance criteria and the seams that matter most. We will assess the imaging your welds need.
Get In TouchVisual checks, NDT and inline weld seam inspection answer different questions. The difference is coverage and how quickly a surface fault reaches the line.
Lift hooks welded to exhaust components were checked by operator judgement, with variants, hook positions and weld appearance all changing. A two-camera KOMPASS station now makes the call.
Published results for the lift-hook weld seam inspection application.
Weld seam inspection with KOMPASS checks what is visible on the weld: surface porosity, undercut, spatter, cracks, end craters, gaps and irregular or short beads. It also confirms seam continuity from start to finish, so a weld that stopped early or skipped a section fails. Each defect is classified separately, which lets you apply different limits to a cosmetic mark and a structural gap. The result for each seam is stored with its image, so repairs and audits start from evidence.
Yes, where a 3D sensor is added. A laser profile sensor captures the cross-section of the weld, so bead width, height and shape can be measured along the seam. A weld bead profile that is too narrow, too tall or uneven often points to wrong travel speed or wire feed. Surface classification and profile measurement can run together at the same station. Profile data is also trended, so a drifting process shows up before beads fall outside the limit.
MIG and TIG welds, laser welds, brazed joints and robotic arc welds are common applications. Each process looks different: TIG beads are smooth and bright, while MIG beads carry more texture. Training includes the processes and materials you run, and lighting is adjusted for each, so the model learns what a good weld looks like in your plant. Different stations can run different models, so each one is judged against the right standard.
Welds are reflective and curved, which creates highlights that can look like defects. Diffuse lighting, polarisation and multiple viewing angles reduce glare, and training includes good welds with normal highlights so the model learns to ignore them. Where a seam wraps around a part, several cameras or a moving view capture the full length. Lighting is fixed once set, so every part is seen under the same conditions.
Yes. This fault usually appears as a groove along the weld toe, where the bead meets the base metal. It is easier to see in a 3D profile than in a flat image, so profile sensing is recommended where edge grooves are a concern. Burn-through and excessive penetration visible from the inspected side are classified in the same way. Images of each failure show the repair team exactly which part of the seam needs work.
The station signals the configured response, such as a hold, a repair lane or a reject. Each result records which seam failed and the defect type, so repair teams know exactly where to work. Trends by station and robot show whether a fault is random or linked to one torch, wire batch or fixture, which shortens the time to fix the cause. That turns each rejection into information the welding team can act on.
Validation uses a labelled set of good and defective welds from your production, including borderline cases agreed with your quality team. Missed defects and false rejects are measured separately by defect type and part variant. Automatic decisions are switched on only when results meet the targets you set, with a review lane during the first weeks. Results are reviewed by defect type, so a weak category gets more training before go-live.
Send welded parts with typical defects, your acceptance criteria and drawings showing the seams that matter. Add the welding processes, part variants, cycle time and where a failed part should go. With that, we can recommend cameras, lighting and any 3D sensing for weld bead inspection and seam checks on your line. Parts from different robots or welders show how much normal variation the model must accept.
Share welded parts, acceptance criteria, processes and cycle time. We will plan weld seam inspection at the right station and set out timing and cost. A KOMPASS deployment usually takes 6 to 8 weeks.
