MORE FLEXIBLE HANDLING OF VARIATION IN GOOD PRODUCTS USING AI
If the variation of good products is zero, the easiest and most reliable inspection is to check whether they are the same as good products. However, there are always variations among good products.
The new version of FIS-200DL uses AI (deep learning) to learn these variations and flexibly process variations in good products that cannot be absorbed by "whether or not they are the same as good products" to improve the good product rate.
Conventional inspection methods use statistical processing to deal with variation in good products.
The conventional FIS-200's comparative inspection is an algorithm that uses statistical processing to create a range of good products, loosely inspecting areas of large variation and strictly inspecting areas of small variation.
This method works very well for objects with high product precision and small variations, and has been used for more than 10 years for a large number of precision parts.
On the other hand, the tolerance for variation of good products is low, resulting in a low percentage of good products, and fine defects cannot be detected in areas where the variation is large.
TWO AI VISUAL INSPECTION ALGORITHMS
AI COMPARISON INSPECTION
AI COMPARATIVE INSPECTION (ANOMALY INSPECTION) IS AN ALGORITHM THAT USES AI (DEEP LEARNING) TO LEARN ABOUT VARIATIONS IN GOOD PRODUCTS TO MORE FLEXIBLY RESPOND TO VARIATIONS IN GOOD PRODUCTS.
The advantage is that only the variation of good products is subject to learning, so the amount of learning is small, and because no extra learning is required, the tolerance is not unnecessarily wide.
Compared to conventional comparative inspections,
- Greatly improved yield
- Fine defects can be detected even in areas with large variation.
- Expanding applicability to objects that could not be inspected due to too much variation in good quality
The following are some of the advantages of this system.
AI DEFECT DETECTION
THE WEAK POINT OF COMPARATIVE INSPECTION/AI COMPARATIVE INSPECTION IS THAT WHEN SEGMENTED IMAGING IS REQUIRED FOR LARGE OBJECTS OR THREE-DIMENSIONAL OBJECTS, IT IS NECESSARY TO LEARN A GOOD PRODUCT FOR EACH SEGMENTED FIELD OF VIEW.
For example, even if one field of view can be set up in 3 minutes, it will take 30 minutes for 10 fields of view, and 5 hours for 100 fields of view. In addition, the amount of data increases, which also reduces operational response time. Therefore, it is not an inspection system that can withstand daily operation.
AI DEFECT DETECTION (SEMANTIC SEGMENTATION) DEFINES "DEFECTS" BY LEARNING THEM WITH AI (DEEP LEARNING) AND INSPECTING THE TARGET IMAGE FOR "DEFECTS".
SINCE "DEFECTS" CAN BE DEFINED IN THE SAME FIELD OF VIEW, IT IS POSSIBLE TO REALIZE COMMON TYPE DATA FOR ALL FIELDS OF VIEW AND EVEN COMMON TYPE DATA FOR ALL TYPES, GREATLY IMPROVING THE OPERABILITY OF THE INSPECTION SYSTEM. HOWEVER, IT IS INFERIOR TO COMPARATIVE INSPECTION/AI COMPARATIVE INSPECTION IN TERMS OF INSPECTION CERTAINTY.
INSPECTION EXAMPLE OF AI VISUAL INSPECTION SENSOR FIS-200DL
DIFFERENCES BETWEEN AI VISUAL INSPECTION SENSOR FIS-200DL AND OTHER COMPANIES' AI VISUAL INSPECTION
Easy to evaluate and deploy on the actual line
FIS-200DL IS AN AI FUNCTION-ENHANCED VERSION OF FIS-200, A CONVENTIONAL VISUAL INSPECTION SENSOR WITH A PROVEN TRACK RECORD. FUNCTIONS SUCH AS CONNECTION TO A CAMERA AND SIGNAL TRANSFER TO/FROM CONVEYING EQUIPMENT HAVE ALREADY BEEN IMPLEMENTED.
THEREFORE, IT IS POSSIBLE TO EVALUATE AI FUNCTIONS ON THE ACTUAL LINE EVEN IN THE STUDY PHASE, AND IT IS ALSO EASY TO DEPLOY AI FUNCTIONS ON THE ACTUAL LINE AFTER EVALUATING THEM OFFLINE.
CURRENTLY, SEVERAL FIS-100 USERS HAVE STARTED COLLECTING AND EVALUATING SEVERAL THOUSAND IMAGES OF GOOD PRODUCTS FOR THE PURPOSE OF IMPROVING THE GOOD PRODUCT RATE. THE SYSTEM IS ALREADY IN OPERATION ON THE ACTUAL LINE USING THE CONVENTIONAL "COMPARATIVE INSPECTION" METHOD, SO IMAGES CAN BE EASILY COLLECTED AND EVALUATION IS PROGRESSING SMOOTHLY.
cost performance
THE INITIAL COST OF FIS-200DL IS APPROXIMATELY 3,300,000 YEN AND UP (AI-ENABLED VISUAL INSPECTION CONTROLLER + 1 X 2M CAMERA). THERE IS NO MONTHLY FEE.
IN ADDITION, DEPENDING ON THE PRODUCT TO BE INSPECTED, A CONVENTIONAL COMPARATIVE INSPECTION MAY BE SUFFICIENT IN MANY CASES; THE VERSION WITHOUT AI FUNCTIONALITY COSTS APPROXIMATELY 1,750,000 YEN AND UP (APPEARANCE INSPECTION CONTROLLER + 1 X 2M CAMERA).
USE OF COMMERCIAL AI (HALCON)
FIS-200DL USES COMMERCIAL AI (HALCON'S DEEP LEARNING FUNCTION) RATHER THAN OPEN SOURCE AI. IT IS EXPECTED TO ENHANCE THE SPEED OF AI FUNCTIONS AND SUPPORT THE LATEST HARDWARE IN THE FUTURE.
WE WILL CONTINUE TO MAKE IMPROVEMENTS TO AI TO MAKE IT EASIER TO USE.