Visual inspection technology
Inspection technology for detecting diverse types of defects
Automated visual inspection process incorporating deep learning to overcome labor shortages and dependency on individual skills
Konica Minolta has developed a visual inspection technology for detecting defects in manufacturing processes and for other purposes. AI-enabled automation solves the problem of insufficient labor for inspections. Moreover, customized AI enables inspections that are difficult with the human eye.
Konica Minolta’s AI-based visual inspection technology enables your inspection process to identify the presence or absence of various defects, as well as the locations and categories of defects. Its highly customizable algorithm can be trained by customers and adapted to their respective inspection judgement needs. It can also be trained to judge defects exclusively with images of non-defective products.
This inspection technology is used in digital manufacturing for discriminating between non-defective and defective products and on automotive manufacturing lines for automatically identifying the types of coating defects.
Detection of defective products based on semi-supervised learning
To meet the need for automatic detection of defective products by means of images, we have developed an AI technology for detecting defects based on a small amount of training data. We realized that almost all data acquired on manufacturing sites were for non-defective products, whereas a large variety of defective products occurred. This AI technology features improved performance thanks to the combination of an image reconfiguration model that completes learning exclusively using images of non-defective products and optimization of parameters for degree-of-abnormality computation using a small number of images of defective products.
Detection of defect locations based on supervised learning
We have developed an AI technology for accurately detecting the locations of defects on inspected objects, by applying a technology used to detect target objects in images. This AI technology detects defect locations with high accuracy even on objects that have a complex texture. * Not yet incorporated in a product or service
Defect type classification based on supervised learning
We have developed an AI technology for classifying the types of defective products pictured in images. To prevent defects from escaping, many manufacturing sites tend to identify candidate defects as a sort of over-detection. This AI technology employs a multitask network structure that determines defects and non-defects independently from defect classification, in order to simultaneously eliminate detection errors and classify defect types with high accuracy. * Not yet incorporated in a product or service