FIS-200

FINALLY, THE FIS WITH DEEP LEARNING "TOO" HAS STARTED WORKING.

First, as a reliable method, we adopted an approach that evolved from the conventional comparative inspection approach: registering only good products and using deep learning to learn only the fluctuations of good products.

The FlexInspector/FIS-100 comparative inspection system, of which more than 1,000 units are in operation, could only handle small fluctuations in good quality, and while it was reliable in that it would not give an OK judgment unless the object was almost identical to a good product, it tended to lower the percentage of good products. It also could not handle objects with large fluctuations in good quality.

This deep learning method can flexibly respond to fluctuations in good quality, which has the effect of improving the good quality rate. It is also expected to be able to handle objects with large fluctuations in good products, which could not be handled until now.

The system can be used in conjunction with the proven comparative inspection system, which is operated first to acquire image data. It is also possible to step up the evaluation of deep learning over time.

THIS VERSION OF FIS IS ALSO COMPATIBLE WITH KEYENCE'S VJ SERIES CAMERAS.

The challenge is,
The cost will increase by about 1 million yen. Depending on the specifications, the cost will be about 3 million yen and up for the camera, PC, and software. There is no annual renewal cost.
Compared to comparative inspections, where learning time takes only a few seconds, learning time is on the order of tens of minutes or hours. However, it is possible to continue inspecting using the "pre-training model" in the background during training.
Due to GPU memory limitations, it is difficult to process large images "as is. The size of the image will be downsized for inspection.

IF YOU WISH TO EVALUATE, PLEASE PREPARE ABOUT 100 GOOD IMAGES AND NG IMAGES TO BE DETECTED. WE WILL RESPOND AT NO CHARGE.