How to handle the gray zone
ovitWhen inspecting with images, ideally the results of visual inspection and the results of image inspection should be in perfect agreement, but we have not yet technically reached that point. How should this "gray zone" be handled?
If this gray zone is set to the side of good products, it is called "defective product outflow. If the gray zone is shifted to the side of defective products, it will result in "yield loss. Which is better?
If the former is the case, then the outflow of defective products cannot be avoided unless all products are visually inspected after the image inspection. In the end, there is no reduction in man-hours. On the other hand, if the latter is the case, visual re-inspection is not necessary for those items that are deemed "good" by the "strict" image inspection, and only those items that are "defective or in the gray zone" need to be visually re-inspected.
When you think about it
With image inspection equipment, it is extremely difficult to achieve a 100% reduction in man-hours.
(In the final visual inspection), the "gray zone" should not be assigned to the "good" side. Only absolutely good products must be judged as "good products.
The results of image inspection equipment that has items that cannot be inspected are all in the "gray zone. The results of image inspection equipment that has items that cannot be inspected are all in the "gray zone" and should not be used for final visual inspection.
We believe that
This is the biggest difference between FlexInspector and other image inspection systems. Basically, FlexInspector judges only those items that are "almost the same" as good items and can find even the "slightest difference". On the other hand, the biggest weakness of FlexInspector is that it cannot fully tolerate such "slight differences". However, in this case, I think it is because the good products are not uniform and it cannot be said that it is a problem of FI.
In the gray zone, the accuracy of visual inspections by humans is also vague. For this reason, it is impossible for the results of image and visual inspections to be in perfect agreement. The original limits for visual inspection were not rigidly set, but were often based on the subjective judgment of the workers on site. Therefore, when introducing image inspection, the inspection standards must be basically reviewed because it is a new thing to do, and unless we do it with the intention of creating a new standard that matches the image inspection, it will be endless, and it will be difficult to achieve the desired effect.
It is a very rare case (although it is not related to image inspection) that a defect that can be seen visually is NG, and if it cannot be seen visually, it is OK. However, there are some patterns in which defects are inherent, and if such defects are to be detected by, for example, ultrasonic inspection, the criteria must be completely redefined. Well, it is only natural to replace visual inspection (images) with ultrasonic inspection (nondestructive inspection).
Certainly, we should make standards that are tailored to imaging inspections.
In many cases, the story is complicated by the forced application of visual inspection standards. In the first place, the standards for visual inspection are often lax, and it seems that in many cases, decisions are made on the basis of "eiyo-ya". As a result, product functionality and manufacturing process capability are not taken into account.
The FI method, which statistically processes good products by sampling them and inspecting whether they are within the distribution of good products, is considered to be more effective in terms of early detection of abnormalities.
Actually, I think this story is not limited to "images. FI can also be thought of as "software for judging the quality of 2-dimensional array data.