Direct answer: Teach-free welding is a repeat-location method for one class of geometry, not automatic programming for any shape. A scan returns where this particular part is, so part-to-part variation is tolerated while a change in the geometry itself is not. A closed regular seam such as a flange ring can be computed from the scan result. What the method removes is the teaching; every part is still located, by the equipment instead of by a person.
Who this is for: Written for engineers and buyers evaluating scan-based welding cells for medium-batch fabricated assemblies.
Scope: This EVST guide covers the preconditions a product has to meet for scan-based, teach-free robotic welding to work: seam regularity, what the scan actually returns, what is saved and what is not, and the conditions that break it. It does not quote scan accuracy, compensation ranges, first-pass yield or cycle times; those come from test records for your own parts.
Measure first, then weld
In the reference footage a red laser line crosses the lid of a transformer tank and the flange opening in it before any arc is struck. Only after that pass does the robot start welding around the flange ring.
That order is the method. The scan is not a check performed on a program that already exists; it is the input the path is built from, and the sequence repeats for each part.
It is worth naming what this is not. It is not offline programming, which builds a path from a model, and it is not seam tracking, which corrects a taught path during welding. Those three answer different questions and are often specified as if they were interchangeable.
What teach-free robotic welding needs from the seam
A flange ring is a closed regular seam lying in a plane. Given a scan that locates it, the path around it can be computed, because the shape is described by a small number of parameters rather than by an arbitrary curve.
That is the geometry the method suits. As the seam becomes less regular – a space curve, an intersection of two cylinders, a joint whose shape changes along its length – the scan still returns a position, but a path can no longer be derived from it without a model.
The practical test is simple to state: can the seam be described well enough that locating it is sufficient to generate it? If yes, teach-free is a candidate. If the shape itself has to be described, the question is an offline programming question.
The order EVST works in is the same each time: seam geometry first, then what the scan can reach, then fit-up, and only then the cell itself.
According to ISO 9692-1:2013, joint preparations for fusion welding are defined by named geometry and dimensions, which is what makes a regular joint describable in the first place.
It saves teaching, not locating
In the footage the same tank carries more than one weld location, and the equipment scans again before it works the next one. Nobody re-teaches anything, but the position is taken again every time.
The distinction matters for the cycle-time argument. Teaching is a one-off setup cost; scanning is a per-part cost. The case therefore improves as batches get smaller and variety rises, and weakens as batches get long and parts get consistent.
It also matters for expectations at commissioning. A cell that scans every part is doing work that a taught cell does not do, and that work has to appear somewhere in the cycle.
| Situation | Usual method | What to verify first |
|---|---|---|
| Regular closed seams, small batches, position varies | Scan-based teach-free | Whether the scan reaches every seam, and gap range |
| Complex or free-form seams, model available | Offline programming from the model | Model fidelity against the real assembly |
| Long runs of one consistent part | Taught program, possibly with tracking | How much position varies part to part |
| Seam position drifts during welding | Seam tracking on a taught path | Whether drift is from distortion or from fit-up |
Where the method breaks
The scan has a line of sight. A seam hidden behind structure, pressed under a clamp or approached at an angle the sensor cannot take is a seam the method cannot locate, and no amount of processing recovers it.
Reflective surfaces are the second common failure. Polished, plated or wet surfaces disturb the reading from a line-structured light sensor, and the practical fixes are optical and mechanical – incidence angle, fixture design, surface condition – rather than software.
The third is fit-up. A scan can measure a gap; it cannot weld one that falls outside the process window. According to ISO 13920:1996, general tolerances for welded constructions are specified as tolerance classes, and a gap that exceeds the class the procedure assumes is an assembly problem rather than a welding equipment problem.
That last one is worth stating plainly because it is regularly filed as an equipment fault during commissioning, and it is not one.
What belongs to test records, and safety
Scan accuracy, the compensation range actually applied and first-pass yield belong to test records produced on your own parts. None of them can be read from footage, and none are quoted here.
According to ISO 9283:1998, pose accuracy and repeatability are measured under stated load and speed conditions, so any positioning figure for a scanning cell has to name the sensor, the standoff and the part it assumes.
According to ISO 10218-2:2025, the safeguarded space follows from the movement of the whole integrated system, which for a gantry-mounted cell includes the gantry travel and not only the arm.
Every EVST application review opens from the same project inputs – drawing, joint preparation and current assembly tolerance – so selection rests on evidence and acceptance is agreed against your own parts rather than a datasheet.
Frequently asked questions
Is teach-free welding the same as offline programming?
No. Offline programming builds a path from a model; teach-free builds it from a scan of the actual part. Offline programming can handle shapes a scan cannot describe, and teach-free can handle part-to-part variation an offline path does not know about. They solve different problems and are often combined.
Which seams suit the method?
Regular, describable seams that a scan can locate – a flange ring, a straight fillet, a bolt circle. As the seam becomes free-form, locating it stops being enough to generate it and the question becomes an offline programming one.
What breaks it in practice?
Line of sight, reflection and fit-up. A seam the sensor cannot see cannot be located; polished, plated or wet surfaces disturb the reading; and a gap outside the process window is an assembly issue that scanning will measure but not fix.
What should we send to get a useful answer?
The part drawing, the joint preparation, the assembly tolerance class you work to, and the range of gaps you actually see on incoming parts.
Project inputs for an application review
To have this checked against your own product rather than a generic cell, send:
- part drawing and the seams to be welded
- joint preparation and the tolerance class you work to
- the range of assembly gaps seen on incoming parts
- batch size and how many variants share the station
Send the drawing and the joint preparation and whether teach-free holds on this product can be worked through against them. Related reading: automatic welding system solutions, robotic welding path boundaries.