A Camera Under the Assembly: Fixed-View EOL Checks
The last gate before an engine or transmission assembly ships is a missing-part confirmation: bolts, poly-V belt, nameplate, certificate, fan, filter, each checked. The unit is large and the points sit on the top, sides, end faces and bottom. This station is the goalkeeper of the line. Whatever passes here meets vehicle assembly, where rework is long and expensive, so the requirement is not only to judge everything but to explain the verdict afterwards.
End of line is the last place a miss is still cheap
A missing part gets more expensive with every meter it travels. Found at the assembly station it costs one operation; at end of line, one assembly; at the vehicle, schedule and reputation. Only at end of line is the unit complete enough to confirm assembly integrity in one pass. That sets the design priority: not the finest measurement, but a real pass over every listed item, with evidence.
The underside is an access problem before it is an optics problem
The underside is uneven and ground clearance is small. The difficulty is not resolution, it is getting a camera in there at all. The robot assigned to the bottom view carries its camera into the space beneath the assembly and images it from a fixed view, so nobody crawls under the unit. Fixing the view adds a second benefit: attitude and lighting at a point are identical every cycle, so the threshold is stable. Handheld inspection cannot offer that.
Two light layers, not one large lamp
The assembly is too large for an ordinary light to reach both brightness and evenness. Oversized high-brightness illumination lifts the whole unit consistently; high-brightness bar light is aimed at local detail such as bolts and hose ends to bring out contour and state. The division must not blur: the large source answers whether the whole surface is visible, the bar light whether a small feature is clear.
Split the checklist by view and a miss has an owner
Each robot owns only the points in its own view, and the checklist is split the same way. The benefit is accountability: any missed item traces back to a specific view, robot and parameter set. If the checklist is one table for the whole unit, a failure can only be described as the system not having judged it, and improvement has nowhere to start. Splitting also simplifies changeover: another model swaps the checklist and parameters while the light and camera layout stays locked per view.
Doubtful items go to a person; the machine does not force a call
Results come as pass or doubtful. Abnormal or ambiguous items are marked point by point with a hint about what to recheck, an operator confirms those points and releases the unit, and the conclusion is written back so the record closes. The machine handles volume and consistency, the person handles exceptions. One extra recheck is cheaper than releasing or scrapping a unit on an uncertain verdict.
Changeover switches lists, not cameras
The cycle is about 69 seconds per unit, subject to site acceptance results, and includes the barcode read, per-view imaging and judgment, the archived full frame and the hint shown for doubtful items. The barcode identifies the assembly at entry and pulls its checklist, so diesel engine, gasoline engine and transmission units share one line without manual configuration.
Points that come up in recheck discussions
Does anyone have to crawl under the assembly?
No. The bottom-view robot carries the camera into the space beneath the unit and images it from a fixed view, keeping the attitude consistent.
Does manual recheck turn into a rubber stamp?
Each item is marked at its own point with a hint, so the operator confirms a specific feature rather than glancing at the unit. The conclusion is written back, which makes the loop binding.
Why not use a single light type?
A large source is even but flat; bar light brings out detail but cannot cover a large area. An assembly has both, so the layers divide the job.