Image: Dansk Industri.
Dansk Industri added Deepvis to its AI for alle case archive, as an example of how Vision AI can move quality control closer to the production line and give companies a better basis for managing quality.
The problem the case opens with
How do you assure the quality of a product when the line makes thousands of parts an hour? For many companies the answer is still spot checks, manual inspection and experienced employees who know what to look for.
That works in plenty of situations. But when volume is high, materials vary, and customers demand more documentation, spot checks quickly become a fragile method.
A startup built on experience from the production floor
At our father’s plastics factory, Carl worked with stock and tooling, Mads with automation systems and IT. The academic profiles differed too — Carl read business economics, Mads trained as a software engineer. That combination of business understanding, software skills and hands-on production experience is what made the idea workable.
We had tried to find a vision system that could catch faults on the plastic parts in production, and could not find one that met the need. So Mads took it on himself: bought a camera, started collecting images from the factory, and began training an AI vision model.
It took him about 14 days to get the prototype recognising several different fault types — specks, scratches, folds and general variation in cleanliness. That success is what pulled us both into the company full time.
Why the models were rebuilt from the ground up
We have gone further than the readily available path. In several cases we have trained models entirely from scratch and worked directly with the models’ underlying infrastructure.
The purpose was to be able to port the models so they run in deterministic software in real time — software that reacts predictably and immediately, every single time. That property is decisive when AI has to sit inside an industrial production line, where the decision to accept or reject a part is made in fractions of a second.
Source: Dansk Industri — “Deepvis sikrer produktionskvaliteten med Vision AI”, 3 July 2026. Includes a video conversation between DI’s Michael Øhlenschlæger and Carl Fagerlund. Quotes translated from Danish.
