After all, automatic visual inspection is difficult...

We had hoped to have a break after the Sensing & Imaging Expo, but we have been quite busy with homework and inquiries that have been piling up on top of the homework.
Especially, we received an order for a project that FlexInspector could not handle.
The biggest selling point of FlexInspector is that it only accepts the same product as a good one.
The system can shut out defective products almost without fail, and its special algorithm greatly reduces false alarms (good products that are judged as NG), making it similar to visual inspections.
Therefore, "automatic visual inspection that can be trusted with confidence" has been realized.
However, there is a prerequisite: the inspection must be limited to "items with a small variation of good products.
This time, the premise is...
Therefore, we are working on "custom products" for the first time in a while.
So, once again, I feel that "automatic visual inspection is difficult.
(1) Basically, there should be no outflow of defective products.
(2) However, there are cases where it is not permissible to "punish the suspicious.
If this is the case, the result must be almost the same as the visual inspection...
#If it's a "poka-expulsion", that's fine.
# But that would limit the return on investment.
I can compete with the experience I've accumulated with FlexInspector and the powerful weapon that is HALCON, but it's still pretty tough.
But... the biggest problem is that my "eyes" are getting harder to see, lol.