An ice cream maker: testing X-ray AI on open and weak seals of flowpack packs
Our client is one of the world’s largest ice cream makers, and at its factory in Poland ice cream leaves the line wrapped in flowpack film. If the film is not heated enough where it is sealed, the seam does not close properly and lets air and moisture into the pack. In 2024 the client asked us whether AI reading X-ray images could find open and weak seals on the line.
The challenge
A flowpack seal is a thin strip of welded film. Once the pack is closed, an under-heated seam looks almost the same from outside as one that holds, and the line runs too fast for a person to check every pack.
X-ray sees through the film and the product, so the question was whether a model could tell a good seal from a bad one on the images, product by product.
What we did
- June 2024: a visit to the factory and the line where the packs are sealed.
- Samples of six kinds of ice cream scanned in an X-ray lab at 30 to 60 kV, to find the settings that show the seal best.
- A test application and a report on the first images.
- Late July and early August 2024: a trial on the factory line.
How a test works
You send samples, including packs with the defect you want to catch. We scan them, label the images and train a model, then report what it found and what it missed, before anything is installed on a line. That is the first step of a two-week proof of concept.
The same approach runs in production at Milarex, where Firecrux X-Ray QC checks every salmon tray on its lines.
The ice cream line during the trial.