Porosity, Threads and a Barcode From One Robot Camera
After machining, an engine block gets one verification before assembly: whether cast porosity exceeds limits, whether the tapped holes have been cut, and whether the barcode and dot-peened characters read. The features sit on several faces, some inside holes, and checking by hand means turning between faces, slowly and inconsistently.
Three tasks, three imaging setups
Porosity is a surface defect judged on grey level and morphology; a tapped hole is judged on the thread inside it; a code or dot-peened character is a symbol to decode. The imaging conditions differ. Porosity needs even diffuse light to suppress the cast texture, the thread needs light that can reach into the hole, and the code needs enough resolution. Forcing them into one parameter set guarantees that one task loses. The cell presets imaging parameters per point and the robot captures and judges on arrival at each point.
The light will not go into a hole, so distance and angle must
The tapped holes sit inside the block where ambient light does not reach, and imaging is very sensitive to working distance. An 8.5 mm lens images close to the hole with flat diffuse light so the thread becomes visible. Close imaging costs depth of field and field of view, which means the robot's attitude at each point must be identical: shift a little and the camera sees a different region inside the hole. Judging whether the thread exists is not hard; capturing the same region every time is.
Codes and dot-peened characters need more resolution than porosity, so they carry their own parameters; their relief casts shadows, and high-resolution imaging with diffuse fill flattens it enough to read them.
The robot is a moving station here
The robot carries the smart camera along a preset path around the block, reaching side and bottom points that are awkward for a person. The camera processes and judges locally instead of shipping images to a host, which removes transmission delay and shortens the response at each point. An integrated unit takes little space, so the station does not need a bracket for every feature type.
Repeatability decides how tight the threshold can be
The thread imaged this cycle has to be comparable with the one imaged last cycle, or the threshold can never be tightened. Robot repeatability therefore sets the consistency of the imaging conditions: with a stable position, thresholds can sit close to the measured distribution and false calls stay low; with a drifting position, thresholds have to be loosened and escapes rise with them. When this item is neglected, the blame usually lands on the algorithm.
Changeover changes the path, not the hardware
The robot path is configured per block model; a changeover updates the point sequence while lens, light and camera stay put. If the inspection points keep growing, new capture moves can be inserted into the path without new hardware. In a shop with many models and mixed batches this matters: changeover cost drops from modifying equipment to modifying a program.
50 seconds, then release to assembly
The block arrives on a conveyor, the robot completes porosity checks, thread checks and code reading, and if every item passes the block is released to assembly. The cycle is 50 seconds, subject to site acceptance results. Multi-point inspection repeats better than manual viewing, judgments stay consistent across shifts and operators, and the record bound to the barcode shows the check status that block had before assembly.
Questions on deep holes and thresholds
Too many porosity false calls, is it the light?
Usually. Cast texture is random, and with uneven lighting texture and real porosity separate poorly in grey level. Even out the surface with flat diffuse light first, then discuss the threshold.
Is a robot slower than a fixed station?
Each point waits for the robot to arrive, but one unit covers many faces and feature types and removes transfer and re-clamping between stations. The more scattered the points, the better that trade.
Can this layout be reused on other parts?
Yes, on medium and large machined parts with many faces and mixed feature types, such as housings and cases, with a lens and light change and a rearranged path.