AR Overlay for Rail Parts: Skip the Haul, Check In Place
Rail welded structures, frames and complex assemblies arrive by crane or forklift and wait in an incoming area. The familiar route is to move the part into an inspection room, queue for a slot and compare drawings by hand. Handling, queueing and manual comparison stack up, while the part itself is large enough that key areas fall outside a single inspector view.
The crane, not the caliper, sets the inspection cost
From unloading to the inspection room, a part is lifted, transferred, queued and lifted back. Measurement occupies a small share of that time. So the useful question is not how to inspect faster but how to leave the part where it stands and bring the inspector and the tablet to it.
A wrong CAD revision poisons every overlay
Tablet AR comparison assumes a 3D CAD model that matches the part number, model and design revision of the physical item. If the revision is wrong, every apparent difference is fiction. STEP, JT, CATIA and IGES files are imported on a PC, inspection zones are drawn around structurally risky areas, checkpoints and viewpoints are set, and the configuration is pushed to the tablet. Where engineering changes are frequent, this has to be a procedure rather than a habit.
Registration leans on the part, not on stickers
The system registers by matching the part's own 3D geometry to the CAD model; no markers are applied. Contours, mounting interfaces and stiffeners on large fabrications are the feature source. Whether they are strong enough to hold tracking is the first question in any evaluation. Long structures, large flat metal surfaces and parts with few visual features are better served by a tablet with LiDAR, where depth improves tracking stability.
Five categories, and only four get a verdict
- Overall contour
- Key mounting interfaces
- Bracket and connector completeness
- Holes, openings and reinforcement
- Regions reserved for later dimensional measurement
The first four are judged directly on the overlay. The fifth is only flagged, because a qualitative screen has no business producing a number. Complex assemblies take this further: several subpart models can be tracked at once, so the relative relationship between subparts is checked, not merely whether one part is present.
Evidence that survives the shift
When something looks wrong, the inspector marks the suspect region, adds a note, takes a photo and records the category. The report follows. Parts that pass move on; suspect parts go to dimensional measurement, which then handles only genuine questions instead of the whole population.
What keeps ten minutes from becoming forty
About ten minutes per part is a reference figure; actual time varies with structure and the number of check items, subject to site acceptance results. Three conditions decide whether that figure is reachable: the CAD and the part agree, the target areas can be reached and seen, and the geometry supports registration on site. Miss any one and the workflow degrades into repeated alignment attempts.
Questions that come up at the site visit
Can we run this without a 3D CAD model?
No. The overlay has a single baseline, so without a model there is nothing to compare against. No parameter setting works around that.
Why does a LiDAR tablet suit some parts?
Image features alone are unstable on long, flat or feature-poor surfaces. Depth data fills in the tracking part of the problem, and stability improves.
Does AR screening replace a CMM?
No. AR answers whether something is fitted and fitted correctly; a CMM answers by how much. Screening first narrows what metrology has to see.
Can an assembly be compared against several subparts in one pass?
Yes. Multi-model tracking compares subpart models simultaneously and checks their relative relationships.