Checking a Welded Structure Without Any Markers at All
Large welded structures are classic oversized parts. Whether the product is construction machinery, agricultural equipment, rail rolling stock or heavy plant, moving the structure to a metrology room is never convenient. These parts carry frames and supports, so an assembly error affects the strength and service life of the whole machine. Conventional inspection depends on crane transport and metrology time, so check frequency drops and problems surface late. On-site comparison is the cheapest gate.
The problem is transport cost, not measurement precision
Every check needs a crane and a slot in metrology, and both are expensive, so inspections get fewer. Meanwhile a welded structure sits in storage for weeks, and a wrong sub-assembly is often discovered at the assembly line, where the rework cost far exceeds what interception at the factory gate would have cost. The binding constraint is not whether the part can be measured accurately but whether it can be checked without moving. That is a different requirement, and it leads to a different answer.
Tablet plus markerless tracking
The tablet camera recognises the part's 3D geometry to track it without markers: no targets or coded points are stuck on the workpiece. CAD is superimposed at the correct position, orientation and scale on the live view, and the inspector works through a fixed checklist.
Markerless is what makes this practical: welded geometry supports registration, which removes applying, recording, recovering and re-applying targets. The tablet form means the check opens wherever the part stands, without interrupting the flow.
Four steps and what they leave behind
| Step | What happens | Key point |
|---|---|---|
| CAD import | The structure model is loaded into the tablet software | Model revision must match the part |
| Markerless registration | 3D features are recognised and CAD overlaid | Anchored on the main structure, so the overlay holds as the inspector closes in |
| Overlay comparison | Missing, reversed, displaced parts and hole anomalies are checked item by item | Ticked on a fixed checklist, so unfinished items are visible |
| Exception record | Anomalies are photographed and bound to the part | Metrology receives a specific question |
How coverage stays complete
Two mechanisms work together. A check-step list is compiled from the risks common to that structure type, so different inspectors follow the same sequence, and a new structure type is set up by updating the list and the model. A fixed walk path adds the other half: main body before attachments, large faces before local details, moving region by region against the overall CAD reference and ticking as you go. On a large part, this is what stops one corner being checked for the whole.
Three constraints on site
- Oil and weld spatter disturb feature recognition; a simple clean of the feature areas before checking restores stable registration.
- Observation distance and angle: close up on a large part only shows a local area, so zoning and the walk path carry the coverage.
- Covered geometry: after repeated painting the features may be too faint, so tracking stability is verified before wider roll-out.
Where this belongs and where it does not
It fits cases where structural consistency matters and tolerance judgment does not: whether something is fitted and fitted correctly, not how many hundredths it is off. Parts that are awkward to move and awkward to tag are the typical objects. For precision dimensional judgment the scheme screens and intercepts on site, and quantitative work still belongs to metrology. Selection should look at whether the geometry supports tracking, how many check items there are, and the site lighting. Results are subject to site acceptance results.
What buyers ask
Why markerless, when targets are more stable?
Targets on a large part mean extra placement, recording and recovery, and the routine grows with the number of parts. Welded geometry is enough for registration, at the price of being more sensitive to surface condition.
Can AR comparison replace a CMM?
No. It answers whether a part is fitted and fitted correctly; a CMM answers how far off it is. This scheme lowers the cost of the first and reserves the second for doubtful parts.
How are results recorded?
Each item is ticked, anomalies are photographed and bound to the part, and records plus images are archived per workpiece.