Smart image analysis for surface defects on cast components

Purpose and goal:

Swedish aluminum, iron and steel foundries want to become better at detecting and assessing defects and deviations on the surface of the cast component, in order to ensure a consistent and high quality of delivered products. The project will create the basic premises for a digital computer vision tool for the next-generation of quality control of surface defects on cast components.

Expected results and effects:

The foundries will be able to reduce discards and remelting of material and components and thereby become more resource efficient with regard to material use and energy consumption. The work environment and work content in the quality control station will be improved, as it currently includes physically demanding manual handling with heavy lifting and also entails considerable responsibility, often under great time pressure.

The project will also enable new and combined data flows between different systems in foundry production for new and expanded areas of use. Approach and implementation:

The idea of the project is to streamline the ongoing quality control of castings using computer vision systems and physics-based machine learning. The use of computer vision in Swedish foundries today faces two major challenges: the lack of relevant training data, and the demanding industrial environment in a component foundry. This project will develop hybrid methods with physics-based machine learning to produce relevant synthetic training data for the vision systems, and outline guidelines for the design and requirements of the digital tool.