Variability of good products and inspection algorithms

In most cases, the applicable algorithm depends on the degree of variation in the good.

If there is almost no variation among good products, the images taken will all be almost the same. Therefore, it is not difficult to extract defects by comparing the images with those of good products.

On the other hand, if the variation is so great that the images cannot be expected to be the same, we have to extract information from the image and determine if the information is correct. For example, we can measure dimensions, count the number of pieces, or read the text. It is important to note that we are only judging whether some of the information is correct by allowing for variation in good products, and everything that is not informative is allowed. As a visual inspection, it is considered to be full of holes. Rather than spending money on visual inspection, it would be more effective to improve the process and reduce the variation of good products.

What is a little different is the method of detecting differences from "guessed good products" inferred from the surrounding circumstances, such as repetitive patterns. For example, a black spot on a blank piece of paper, a foreign substance on a glass plate, a scratch on a car body.... In most cases, "plain" is a prerequisite, but the system is overwhelmingly capable of detecting small scratches that are easy to guess good quality.

FlexInspector, in turn

1. comparison check
2. Length measurement inspection, character reading inspection, etc.
3. Touch-up inspection

The first two are carried by the
On the other hand, most commercially available image processing equipment is only 2. With the increase in manufacturing technology, the variation in good products has almost disappeared, and it is questionable whether this is really an effective inspection method.
FI users rarely use 2. Most of them end up with 1, and 3 is applied when the object is large and simple.

FlexInspector's comparative inspection is a method of determining a range of good parts by sampling a large number of good parts and performing a statistical process to absorb the variation in the good parts. This is a means of absorbing subtle variations in good parts, and when applied to items with large variations in good parts, it results in an extremely loose inspection. To improve detection capability, it is important to suppress the variation of good images.