Why is visual inspection by image processing so difficult?

Here is a story we heard from a user we recently visited.

• The production process is automated, and if only the materials are fed in, it can be done rapidly. However, inspections are a challenge because of the "how much per piece" cost. As the production volume increases...
• We have tried to put in image processing equipment from various manufacturers. However, only "positioning" and "dimensional measurement (of a part)" are in operation. Visual inspection" is handled visually, since it is eventually leaked out.

These are not "just a few" stories; most of the sites I spoke with were similar, and it is likely that "inspection man-hours" account for a large portion of parts costs in the manufacturing industry.

A manufacturer claims "25 years of image processing," but if it is really worth it, it should be more widely used. In reality, however, this is not the case. Appearance inspection" is not so easy.

The biggest difference between "positioning" and "dimensional measurement," which are relatively easy to accomplish, and "visual inspection" is that "everything must be seen. Even if there is an "unexpected defect" that is not mentioned in the specifications, if it cannot be detected, it will be "leaked. In visual inspection, "inspection items" are meaningless. Everything must be inspected.

The standard practice has been to do this by "each inspection item," "use this lighting," "create the algorithm this way," and "limit the tolerance to this level," which does not work. This is not only time-consuming, but also results in leakage.

In the end, visual inspection comes down to "if it is the same as a good product, it is OK; if it differs from a good product, it is NG.
Based on this concept

1. Only "good products" are sorted through strict inspection. These are shipped as is.
2. Only those items that are not considered "good" are visually inspected, and those that are acceptable as "good" are sorted out by human judgment.

I believe that there is no other way to ensure quality and reduce inspection man-hours at the same time.

If "only good products" cannot be sorted out, or if they can be sorted out, they are too few, it is safer not to try to automate the visual inspection process. It would be more meaningful to spend money on improving the manufacturing process, since it can be said that there is a large variation in products.