Our customized solution combined both physical and intangible technologies including high-resolution cameras, macro lenses, AI and robotics. We decided to start the automated control in the press shop, the starting point of production, where the early detection of defects proves to be the most economically advantageous.
Due to the microscopic nature of the defects, some as small as 10 μm, we used a high-resolution camera and a macro lens. Due to the limited field of view of this assembly (around 20x20mm), where the inspected products could be up to approximately 70 mm in size, we used motorized platforms that move the product under the camera and capture multiple overlapping images. The software then combines these images and gives one comprehensive insight.
The system’s core is image analysis based on AI deep learning, which effectively identifies product defects. This deep learning model offers unparalleled flexibility in conceptualizing and generalizing the appearance of different insert variants while covering our client's extensive product portfolio. Deep learning excels at handling natural variations in complex patterns, eliminating the need to define tolerance parameters for each variation.
A clear and intuitive user interface then provides operators with tools for quick and accurate analysis. They can mark and classify visual inspection results and provide feedback to the system, supporting a continuous learning process and contributing to model training and the ability to adapt to specific production characteristics. The inspection station communicates seamlessly with the robotic arm handling the products within the press. After the inspection is complete, the station signals the robot to remove a piece and insert another. In addition, the system informs the robot whether the inspected piece is considered good or defective.
Finally, the data management model enables the generation of detailed reports and statistics that provide valuable insight into the evolution of quality over time. Operators can easily access historical data, analyze visual inspection results and track defect trends on their production line.