20-40 Cameras and 10-30 s Cover a 5 m Dashboard Bracket
The dashboard bracket is one of the large irregular metal parts in a vehicle, supported up to the 5000 mm class and produced by welding, stamping, casting or die casting. Several questions land on the same workpiece: are the assembled parts present, are the screws and nuts fitted, has the residual film been removed from the holes, is anything wrongly or doubly fitted, and are the form and position dimensions correct. Offline CMM sampling cannot cover every unit, so the line runs 20 to 40 cameras at 20 MP, cross-covering the part in 10 to 30 seconds at 99.9 percent accuracy, subject to site acceptance results.
Five defect families, one fixture
Presence, fasteners, hole film, wrong or missing assembly and form-position dimensions are five different judgments. Split across five stations, the part is loaded and unloaded repeatedly. Put in one station, a single system has to carry all five criteria at once, which is a camera layout problem before it is a software problem.
Camera count follows geometry and point density
The 20 to 40 range is not arbitrary. Part size, the distribution of inspection points and the required cycle together set how many cameras are needed and where they go. Cross-coverage means every feature is seen from at least two directions, so a reflection or an occlusion in one view leaves another path. Whether a project lands near the lower or upper bound depends on the bracket variant and how densely the points are packed.
A thin film in a hole is a lighting problem before an algorithm problem
A residual film in a hole is thin and weak in contrast. A 20 MP sensor keeps the physical size represented by one pixel small enough for the feature to hold, and dual bar lights provide even illumination. The two belong together: raise pixel count without fixing the light and the edges stay soft.
Where ten seconds ends and thirty begins
The range comes from product variation. A larger bracket with more inspection points and more cameras takes the longer end. It is a boundary, not a promise, and a selection study should confirm where the target variant sits inside it.
A human recheck lane is not an admission of failure
The cell reads a code to identify each part and outputs SPC data and a traceability report. Suspect items go to a manual recheck lane. Keeping that lane moves human attention to genuinely uncertain samples instead of asking people to hunt for screws one by one.
Online 100 percent does not retire the CMM
This cell replaces offline CMM sampling with an online full check. The two roles differ: the line cell confirms that every unit was seen, while high-accuracy metrology remains available where a real number is required. Claiming that an online cell replaces a metrology lab would be dishonest.
Four questions worth settling before purchase
How is a large part kept free of blind spots?
By cross-covering with 20 to 40 cameras so each feature has at least two sight lines. The exact count and angles are engineered to the part, not fixed.
Which items can be inspected?
Assembled parts, screws and nuts, hole film removal, wrong or missing components, and form and position dimensions of machined features. All results come from one pass.
How should the 10 to 30 second range be read?
As a spread across variants, driven by size, point count and camera configuration. Confirm it on the target part; the final figure is subject to site acceptance results.
What happens to the data?
Code reading, manual recheck and traceability feed SPC output and batch records, with a human lane retained for suspect parts.