Moving 81 Inspection Standards Into a Library
An engine block meets its appearance gate at the foundry: 36 inspection regions and more than ten defect types including blowholes, cracks and scratches. Manual inspection fails twice over. Coverage is incomplete, and the standard varies: the same blowhole is acceptable to an experienced inspector and a rework to someone else. Nor does it leave a queryable record.
The criterion moves from experience into a standard library
The core is not a machine but a library of 81 inspection standards covering defect type, position and size, compared item by item automatically on the image. That changes the nature of the judgment from whether a mark looks like a defect to whether this item passes this standard. Once the yardstick is fixed, conclusions are comparable across shifts and the results can feed process improvement. Maintaining the library is what keeps it valid, so new defect samples have to be added as they appear.
3D form puts depth into the verdict
A high-precision smart camera, 3D scanning equipment and multi-angle lighting simulation together produce a three-dimensional image of the block surface. Compared with a flat image, this brings form into the decision: defect depth and the height of a protrusion become usable criteria.
A blowhole and a stain can both look like a dark spot in 2D, while in 3D one is a depression with depth and the other is surface contamination.
How 36 regions are kept complete
Coverage has to be implemented region by region, and any region left out is a hole in the inspection. The method is a regional list: which standards apply to each region, from which viewpoint and under which lighting, all fixed at design stage. This differs from shooting a panorama and then hunting for defects, where coverage looks broad but regions without a matching standard are easily skipped. With a list, completeness can be verified rather than assumed.
Multi-angle light is aimed at the defect
Blowholes, cracks and scratches behave differently optically. A blowhole is a pit needing grazing light to cast a shadow; a crack is thin and long, and shows only at certain angles to the light; a scratch is shallow and visible through reflection difference. Multi-angle lighting is designed around these behaviours to separate defect from background. Calibrating the light parameters directly affects how well defects appear, so it is a deployment task, not ancillary tuning.
40 seconds against the conveyor
The cycle is 40 seconds, matching the foundry line rhythm, subject to site acceptance results. Blocks pass one by one, inspection software interlocks with conveyor control, good parts are released and exceptions diverted, and the flow does not stop. Imaging, 81 comparisons and the verdict fit inside the cycle, so imaging and comparison run in parallel rather than waiting for all images.
Blowhole incidence down 33 percent, and what that says
Interception above 99 percent, inspection accuracy rising from grade C to grade A, inspection time per part down 40 percent, labour cost per item down 66 percent and blowhole incidence down 33 percent, with full traceability, subject to site acceptance results.
The last figure matters most. Falling blowhole incidence is not the inspection station's own achievement; it comes from inspection data fed back into the casting process. Once defects are counted by type and region, the foundry knows what to change. Whether that loop forms depends on whether the data is structured and traceable.
Selection starts from defect types and region count, which size the scanning and lighting scheme and the library, with room reserved for new defects. Then look at maintenance: who updates the library, on what cycle. The first decides go-live, the second decides whether it is still useful in three years.
Keeping the library alive
Where do the 81 standards come from?
From that block's appearance requirements, decomposed by defect type, position and size into comparable items. A customised library, not a generic template.
What does 3D add over a flat image?
Depth in the verdict. Pits and protrusions are inferred from shading in 2D, measured directly in 3D.
Does the library go stale?
It does. When the process or the defect spectrum changes, the standards follow. Without maintenance, judgment drifts away from the actual quality requirement.