Together with Xactools GmbH, we have developed a solution for a manufacturer of metal components to identify even the smallest defects.
Xactools is a Swabian medium-sized company that offers automation solutions for feeding, measuring and palletizing processes.
Challenge: The different types of defects and lighting conditions
For this application, Xactools integrated a high-resolution line scan camera into the manufacturer's production process, which continuously captured images of the metal surfaces.
With our DENK VISION AI Hub, the AI was trained with the individual images by the customer (as an expert) himself. Only the customer himself can most reliably annotate all errors relevant to his company.
The customer chose “Instance Segmentation” as the network. In contrast to “object recognition”, the customer not only receives information about the location of a defect, but can also have the area content displayed, for example. This information could be important in downstream sorting or classification processes.
The AI's consistent learning process makes it possible to produce almost perfectly reliable results for all metal components and environmental conditions. Scratches are reported to the customer within milliseconds.
The early and reliable detection of errors has increased the efficiency of the production process itself and reduced reject rates.
