Automated quality inspection of die cast and hot-pressed metal components

Important results from the project

The project has given participating metal processing companies an increased understanding of current opportunities and challenges with AI for defect detection. The project has shown that AI-based machine vision can in principle be used for the detection of surface defects on metal components, but it has also identified a number of challenges linked to flexibility that must be solved before the technology can be widely disseminated in the Swedish metal processing industry, such as capacity of the AI models to be quickly recalibrated for inspection of new components and defect types.

Expected long term effects

The project focused on two defect types in particular, cold discharge and pores. AI models for both defect types were trained and could be shown to work in principle, but detection of pores was significantly easier than detection of cold flows, which were very sensitive to ambient light conditions. One effect is that the two camera companies have gained new experience in two new industries, which has the potential to accelerate the roll-out of AI in these industries. RISE has gained knowledge about defect detection that will most likely lead to a degree project and a new project application.

Approach and implementation

The first phase of the project, with laboratory work, went largely according to plan, but the second phase, when field tests were to be performed, was unfortunately stopped due to a number of unforeseen events, the single largest being the pandemic. Travel and visiting bans prevailed for a long time, followed by a prolonged restart period for some of the metal processing companies.