Severstal continues to enhance the automation of quality control for rolled metal products, implementing 12 computer vision systems at the Cherepovets Metallurgical Plant (CherMK). These systems, developed by Severstal Digital’s AI team in collaboration with the Technical Development, Quality, and Repair Directorates, utilize neural networks and advanced surface inspection systems (SIS) based on proprietary designs.
By the end of 2024, four new SIS units were deployed on polymer coating units No. 1, 2, 3, and longitudinal cutting unit No. 8. Additionally, systems were successfully implemented on continuous hot-dip galvanizing units No. 1 and 4, and longitudinal cutting unit No. 4. These solutions can detect 40 to 100 defect classes in color images, aiding defect classification and quality control. Each system uses tailored computer vision models trained to identify defects specific to different metal types. When product types change, such as from cold-rolled steel to dynamo, the algorithm automatically adjusts to the appropriate model, ensuring precise defect classification.
The software also features adaptive lighting control, adjusting based on time of day and coating color, which enhances image quality for better defect recognition, particularly on polymer-coated metals, which can have dozens of color variants.