Fraunhofer-Gesellschaft
Image based anomaly detection with artificial intelligence
Despite the ongoing corona pandemic, globalization continues to advance. German companies are increasingly finding themselves competing with rival companies on the world market. Many products manufactured in Germany are only competitive because they have outstanding quality.
If you can't produce the cheapest, you have to deliver the best quality.
Manufacturing companies are faced with the challenge of raising this quality standard even further in order to be able to compete on the world market. To ensure functional and aesthetic properties of a product, Fraunhofer IPK is developing novel approaches for data-reduced AI applications. Thus, it is possible to integrate the advantage of AI-based image processing into the inspection process without having to accept the resulting large effort of training data collection. This is possible by reformulating the inspection task: instead of explicitly detecting learned defects in images, any deviation (anomaly) from a defined quality standard is detected. For this purpose, the AI is trained only with defect-free products, which naturally exist in significantly larger numbers than defective products. This has two further advantages: 1. not all production-related defects have to be known in advance. 2. the high personnel costs for data collection and manual marking of each individual defect are eliminated. The technology of image-based anomaly detection thus offers great potential to revolutionize the application of automated optical inspection.
More about image-based anomaly detection by artificial intelligence