A vision welding robot does not follow a path taken from a drawing. It scans the real workpiece with 3D cameras, matches the scan to a joint template, generates the weld path itself and corrects for heat distortion with arc tracking while it welds. This article explains each link in that chain, using a JTCLASER 9-axis cantilever workstation in structural steel production as the example.
How does a vision welding robot work?
A vision welding robot first scans the actual workpiece with 3D cameras: a wide-view camera locates the part and a close-range camera picks precise points. The software applies a joint template to calculate the seams and generate the robot program. The robot then welds while arc tracking compensates for heat distortion. No point-by-point teaching is needed.
1What are the parts of a vision welding workstation?
Take a cantilever welding robot cell as the example. The robot arm hangs from a cantilever that travels along a ground rail (X axis), with cross travel (Y) and a vertical lift (Z). With the robot’s own six axes that makes nine axes, enough to reach seams across long beams and wide panels without moving the part. Two 3D cameras do the scanning. An industrial PC runs the welding software, and the robot controller and teach pendant hold the weld files.
On the JTCLASER 9-axis cantilever workstation the reference configuration lists a 0‚Äì1.8 m welding height, a 3.5 m single-position welding width, a global 3D camera working at 1.2‚Äì3 m and a local 3D camera scanning at 0.4‚Äì1 m. Rail length is configured to the customer’s longest parts.
2What are the steps from scan to weld?
The approach is sometimes called reverse modeling: instead of computing a path from the drawing, the system measures the real part and builds the path from what it finds.
3What is a joint template?
A joint template, called a node in the software, describes one type of welded structure, for example a double-sided section splice or a straight stiffener with wrap-around corners. Inside a template every seam has a name and a role: flat seams, vertical seams, start and end points. Each seam references a weld procedure package that sets joint type, leg size, wire type and weaving.
Because the template describes the structure rather than fixed coordinates, the same template works for parts of different lengths and sizes. The scan supplies the real dimensions; the template supplies the welding logic.
4Why not program the welding robot from drawings?
Offline programming from CAD works when parts are always identical to the model. In structural steel they rarely are. The drawing and the real part differ in three ways:
- Fit-up accuracy
Hand-tacked assemblies are never exactly at nominal dimensions, and gaps vary along the seam.
- Part placement
A long beam is never placed in exactly the same spot twice without expensive fixtures.
- Welding distortion
The part moves as it heats up; a path computed from the drawing does not.
Vision scanning solves the first two before welding starts, and arc tracking handles the third during welding. For a deeper comparison see CAD-based programming vs 3D vision.
5What does the operator still do?
| Task | Who or what does it |
|---|---|
| Place and tack the part | Operator, inside the marked working range |
| Choose the joint template | Operator, from the template library |
| Find seams and generate the path | Software, from the 3D scan |
| Check the path preview | Operator, before every new part type |
| Weld and follow distortion | Robot with arc tracking |
| Tune current, voltage and weaving | Operator or process engineer, in the weld files |
For repeat parts the routine becomes scan the first part ‚Üí check and save the path ‚Üí scan and run. The gain comes from cutting programming time on every new part and from stable cycle times on repeat parts. Where the limits of no-teach welding lie is covered in the technical boundaries of a no-teaching welding workstation.
6Which parts suit a vision welding robot?
Good candidates are structural assemblies with visible fillet and groove seams: H-beams, columns, stiffened panels, platforms, brackets and bridge or ship plate units. Two practical limits decide the layout: part height must stay within the welding height (0–1.8 m on the 9-axis cantilever), and the part must sit inside the travel range of the rail. Taller or wider parts call for a gantry layout; see welding robot working range and collision prevention and cantilever vs gantry for how to choose.
Frequently asked questions
Does a vision welding robot need teach programming?
No point-by-point teaching is needed. The robot scans the part, matches a joint template and generates the path. For repeat parts you check and save the path on the first part, then scan and run.
What is the difference between vision scanning and arc tracking?
Vision scanning finds the seams before welding starts. Arc tracking corrects the torch position in real time during welding as the part distorts. They work together.
What parts can a vision welding robot weld?
Structural steel assemblies with accessible fillet and groove seams, such as H-beams, columns, stiffened panels, platforms and brackets, within the cell's height and travel range.
See the workstation behind this article
The 9-axis cantilever workstation scans, generates paths and welds structural steel without teach programming. Send us a drawing of your typical part and we will check whether it fits the welding range.
- JTCLASER field commissioning and after-sales records, 2026
- JTCLASER 9-axis cantilever welding robot workstation, published reference configuration